moebiusband/MD-Bench
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Move analysis and result files to log directory

Browse Source 
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
This commit is contained in:
Rafael Ravedutti 2021-06-11 02:16:10 +02:00
parent 0482e4f09a
commit b8b364d265
36 changed files with 0 additions and 59965 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

90
arch_analysis/avx2/force_aos_lt600_markers.txt
View File

@ -1,90 +0,0 @@
iwia021h@testfront1:~/MD-Bench$ /home/hpc/iwia/iwia021h/.local/bin/osaca --ignore-unknown --arch=BDW asm/avx2/force_aos_lt600_markers.s
Open Source Architecture Code Analyzer (OSACA) - 0.3.14
Analyzed file: asm/avx2/force_aos_lt600_markers.s
Architecture: BDW
Timestamp: 2021-05-07 23:01:31
P - Throughput of LOAD operation can be hidden behind a past or future STORE instruction
* - Instruction micro-ops not bound to a port
X - No throughput/latency information for this instruction in data file
Combined Analysis Report
------------------------
Port pressure in cycles
| 0 - 0DV | 1 | 2 - 2D | 3 - 3D | 4 | 5 | 6 | 7 || CP | LCD |
----------------------------------------------------------------------------------------------------
12567 | | | 0.50 0.50 | 0.50 0.50 | | | | || 4.0 | | vmovdqu (%rax,%r8,4), %xmm2 #60.25
12568 | 1.00 | | | | | | | || 1.0 | | vmovd %xmm2, %rdx #60.25
12569 | | | | | | 1.00 | | || | | vpunpckhqdq %xmm2, %xmm2, %xmm8 #60.25
12570 | 1.00 | | | | | | | || | | vmovd %xmm8, %r9 #60.25
12571 | 0.00 | 0.00 | | | | 0.00 | 1.00 | || 1.0 | | movl %edx, %ebx #60.25
12572 | 0.00 | | | | | | 1.00 | || | | shrq $32, %rdx #60.25
12573 | | 0.50 | | | | 0.50 | | || | | lea (%rdx,%rdx,2), %esi #61.40
12574 | | 0.50 | | | | 0.50 | | || 1.0 | | lea (%rbx,%rbx,2), %ebx #61.40
12575 | 0.00 | 0.00 | | | | 0.00 | 1.00 | || 1.0 | | movslq %ebx, %rdx #61.40
12576 | 0.00 | 0.00 | | | | 0.00 | 1.00 | || | | movslq %esi, %rbx #61.40
12577 | 0.00 | 0.00 | | | | 0.00 | 1.00 | || | | movl %r9d, %esi #60.25
12578 | 0.00 | | | | | | 1.00 | || | | shrq $32, %r9 #60.25
12579 | | | 0.50 0.50 | 0.50 0.50 | | | | || 4.0 | | vmovups (%rcx,%rdx,8), %xmm13 #61.40
12580 | | | 0.50 0.50 | 0.50 0.50 | | | | || | | vmovups (%rcx,%rbx,8), %xmm0 #61.40
12581 | | | 0.50 0.50 | 0.50 0.50 | | | | || | | vmovq 16(%rcx,%rdx,8), %xmm14 #61.40
12582 | | 0.00 | | | | 1.00 | | || | | lea (%rsi,%rsi,2), %esi #61.40
12583 | 0.00 | 0.00 | | | | 0.00 | 1.00 | || | | movslq %esi, %rsi #61.40
12584 | | 0.00 | | | | 1.00 | | || | | lea (%r9,%r9,2), %r9d #61.40
12585 | 0.00 | 0.00 | | | | 0.00 | 1.00 | || | | movslq %r9d, %r9 #61.40
12586 | | | 0.50 0.50 | 0.50 0.50 | | 1.00 | | || | | vmovhpd 16(%rcx,%rbx,8), %xmm14, %xmm15 #61.40
12587 | 0.50 | 0.50 | 0.50 0.50 | 0.50 0.50 | | | | || 1.0 | | vinsertf128 $1, (%rcx,%rsi,8), %ymm13, %ymm7 #61.40
12588 | | | 0.50 0.50 | 0.50 0.50 | | | | || | | vmovq 16(%rcx,%rsi,8), %xmm2 #61.40
12589 | 0.50 | 0.50 | 0.50 0.50 | 0.50 0.50 | | | | || | | vinsertf128 $1, (%rcx,%r9,8), %ymm0, %ymm0 #61.40
12590 | | | 0.50 0.50 | 0.50 0.50 | | 1.00 | | || | | vmovhpd 16(%rcx,%r9,8), %xmm2, %xmm8 #61.40
12591 | | | | | | 1.00 | | || | | vunpcklpd %ymm0, %ymm7, %ymm14 #61.40
12592 | | | | | | 1.00 | | || 1.0 | | vunpckhpd %ymm0, %ymm7, %ymm7 #61.40
12593 | | 1.00 | | | | | | || | | vsubpd %ymm14, %ymm5, %ymm0 #61.40
12594 | | | | | | 1.00 | | || | | vinsertf128 $1, %xmm8, %ymm15, %ymm13 #61.40
12595 | | 1.00 | | | | | | || 3.0 | | vsubpd %ymm7, %ymm4, %ymm8 #62.40
12596 | | 1.00 | | | | | | || | | vsubpd %ymm13, %ymm6, %ymm2 #63.40
12597 | 0.50 | 0.50 | | | | | | || 3.0 | | vmulpd %ymm8, %ymm8, %ymm14 #64.53
12598 | 0.50 | 0.50 | | | | | | || 5.0 | | vfmadd231pd %ymm0, %ymm0, %ymm14 #64.53
12599 | 0.25 | 0.75 | | | | | | || 5.0 | | vfmadd231pd %ymm2, %ymm2, %ymm14 #64.67
12600 | | | | | | | | || | | X vcmpltpd %ymm11, %ymm14, %ymm7 #66.26
12601 | | 0.00 | | | | 1.00 | | || | | vpcmpeqd %ymm13, %ymm13, %ymm13 #66.26
12602 | 1.00 | | | | | 1.00 | | || | | vptest %ymm13, %ymm7 #66.26
12603 | | | | | | | | || | | X je ..B1.26 # Prob 50% #66.26
12604 | | | | | | | | || | | # LOE rax rcx rdi r8 r14 r10d r11d r12d r13d r15d ymm0 ymm1 ymm2 ymm3 ymm4 ymm5 ymm6 ymm7 ymm8 ymm9 ymm10 ymm11 ymm12 ymm14
12605 | | | | | | | | || | | ..B1.25: # Preds ..B1.24
12606 | | | | | | | | || | | # Execution count [1.25e+04]
12607 | 2.50 16.00 | 0.50 | | | | | | || 23.0 | | vdivpd %ymm14, %ymm10, %ymm13 #67.42
12608 | 0.50 | 0.50 | 0.50 0.50 | 0.50 0.50 | | | | || 3.0 | | vmulpd 352(%rsp), %ymm13, %ymm14 #68.42[spill]
12609 | 0.50 | 0.50 | | | | | | || 3.0 | | vmulpd %ymm14, %ymm13, %ymm14 #68.48
12610 | 0.50 | 0.50 | | | | | | || 3.0 | | vmulpd %ymm14, %ymm13, %ymm15 #68.54
12611 | 0.50 | 0.50 | | | | | | || | | vfmsub213pd %ymm9, %ymm13, %ymm14 #69.58
12612 | 0.50 | 0.50 | 0.50 0.50 | 0.50 0.50 | | | | || | | vmulpd 320(%rsp), %ymm13, %ymm13 #69.58[spill]
12613 | 0.50 | 0.50 | | | | | | || 3.0 | | vmulpd %ymm13, %ymm15, %ymm15 #69.65
12614 | 0.50 | 0.50 | | | | | | || 3.0 | | vmulpd %ymm14, %ymm15, %ymm13 #69.71
12615 | 0.75 | 0.25 | | | | | | || | | vmulpd %ymm13, %ymm0, %ymm0 #70.35
12616 | 1.00 | 0.00 | | | | | | || | | vmulpd %ymm13, %ymm8, %ymm8 #71.35
12617 | | | | | | 1.00 | | || | | vandpd %ymm0, %ymm7, %ymm0 #70.35
12618 | | 1.00 | | | | | | || | 3.0 | vaddpd %ymm0, %ymm12, %ymm12 #70.21
12619 | 1.00 | 0.00 | | | | | | || 3.0 | | vmulpd %ymm13, %ymm2, %ymm0 #72.35
12620 | | | | | | 1.00 | | || | | vandpd %ymm8, %ymm7, %ymm2 #71.35
12621 | | | | | | 1.00 | | || 1.0 | | vandpd %ymm0, %ymm7, %ymm7 #72.35
12622 | | 1.00 | | | | | | || | | vaddpd %ymm2, %ymm1, %ymm1 #71.21
12623 | | 1.00 | | | | | | || 3.0 | | vaddpd %ymm7, %ymm3, %ymm3 #72.21
12624 | | | | | | | | || | | # LOE rax rcx rdi r8 r14 r10d r11d r12d r13d r15d ymm1 ymm3 ymm4 ymm5 ymm6 ymm9 ymm10 ymm11 ymm12
12625 | | | | | | | | || | | ..B1.26: # Preds ..B1.25 ..B1.24
12626 | | | | | | | | || | | # Execution count [2.50e+04]
12627 | 0.00 | 0.00 | | | | 0.00 | 1.00 | || | | addq $4, %r8 #59.13
12628 | 0.00 | 0.00 | | | | 0.00 | 1.00 | || | | cmpq %rdi, %r8 #59.13
12629 | | | | | | | | || | | * jb ..B1.24 # Prob 82% #59.13
14.0 16.00 14.0 5.50 5.50 5.50 5.50 14.0 10.0 75.0 3
Loop-Carried Dependencies Analysis Report
-----------------------------------------
12618 | 3.0 | vaddpd %ymm0, %ymm12, %ymm12 #70.21| [12618]
12622 | 3.0 | vaddpd %ymm2, %ymm1, %ymm1 #71.21| [12622]
12623 | 3.0 | vaddpd %ymm7, %ymm3, %ymm3 #72.21| [12623]
12627 | 1.0 | addq $4, %r8 #59.13| [12627]

74
arch_analysis/avx2/force_soa_lt600_markers.txt
View File

@ -1,74 +0,0 @@
iwia021h@testfront1:~/MD-Bench/asm/avx2$ /home/hpc/iwia/iwia021h/.local/bin/osaca --ignore-unknown --arch=BDW force_soa_lt600_markers.s
Open Source Architecture Code Analyzer (OSACA) - 0.3.14
Analyzed file: force_soa_lt600_markers.s
Architecture: BDW
Timestamp: 2021-05-06 15:40:52
P - Throughput of LOAD operation can be hidden behind a past or future STORE instruction
* - Instruction micro-ops not bound to a port
X - No throughput/latency information for this instruction in data file
Combined Analysis Report
------------------------
Port pressure in cycles
| 0 - 0DV | 1 | 2 - 2D | 3 - 3D | 4 | 5 | 6 | 7 || CP | LCD |
----------------------------------------------------------------------------------------------------
12343 | | | 0.50 0.50 | 0.50 0.50 | | | | || 4.0 | | vmovups (%rax,%rdx,4), %xmm2 #60.25
12344 | | 1.00 | | | | 0.00 | | || | 1.0 | vpcmpeqd %ymm13, %ymm13, %ymm13 #61.40
12345 | | | | | | 1.00 | | || | | vxorpd %ymm8, %ymm8, %ymm8 #61.40
12346 | | | | | | | | || | | * vmovdqa %ymm13, %ymm0 #61.40
12347 | | | | | | 1.00 | | || | | vxorpd %ymm14, %ymm14, %ymm14 #62.40
12348 | | | | | | | | || | | * vmovdqa %ymm13, %ymm7 #62.40
12349 | 0.00 | 0.00 | 2.00 | 2.00 | 1.00 | 3.00 | 1.00 | || | | vgatherdpd %ymm0, (%r8,%xmm2,8), %ymm8 #61.40
12350 | 0.00 | 0.00 | 2.00 | 2.00 | 1.00 | 3.00 | 1.00 | || 5.0 | | vgatherdpd %ymm7, (%rdi,%xmm2,8), %ymm14 #62.40
12351 | | 1.00 | | | | | | || | | vsubpd %ymm8, %ymm5, %ymm0 #61.40
12352 | | 1.00 | | | | | | || 3.0 | | vsubpd %ymm14, %ymm4, %ymm8 #62.40
12353 | | | | | | 1.00 | | || | | vxorpd %ymm7, %ymm7, %ymm7 #63.40
12354 | | | | | | | | || | 0.0 | * vmovdqa %ymm13, %ymm15 #63.40
12355 | 0.00 | 0.00 | 2.00 | 2.00 | 1.00 | 3.00 | 1.00 | || | 5.0 | vgatherdpd %ymm15, (%rsi,%xmm2,8), %ymm7 #63.40
12356 | 0.50 | 0.50 | | | | | | || 3.0 | | vmulpd %ymm8, %ymm8, %ymm14 #64.53
12357 | | 1.00 | | | | | | || | 3.0 | vsubpd %ymm7, %ymm6, %ymm2 #63.40
12358 | 0.50 | 0.50 | | | | | | || 5.0 | | vfmadd231pd %ymm0, %ymm0, %ymm14 #64.53
12359 | 0.50 | 0.50 | | | | | | || 5.0 | 5.0 | vfmadd231pd %ymm2, %ymm2, %ymm14 #64.67
12360 | | | | | | | | || | | X vcmpltpd %ymm11, %ymm14, %ymm7 #66.26
12361 | 1.00 | | | | | 1.00 | | || | | vptest %ymm13, %ymm7 #66.26
12362 | | | | | | | | || | | X je ..B1.26 # Prob 50% #66.26
12363 | | | | | | | | || | | # LOE rax rdx rsi rdi r8 r9 r12 r14 ecx ebx r10d r13d r15d ymm0 ymm1 ymm2 ymm3 ymm4 ymm5 ymm6 ymm7 ymm8 ymm9 ymm10 ymm11 ymm12 ymm14
12364 | | | | | | | | || | | ..B1.25: # Preds ..B1.24
12365 | | | | | | | | || | | # Execution count [1.25e+04]
12366 | 2.50 16.00 | 0.50 | | | | | | || 23.0 | 23.0 | vdivpd %ymm14, %ymm10, %ymm13 #67.42
12367 | 0.50 | 0.50 | 0.50 0.50 | 0.50 0.50 | | | | || 3.0 | | vmulpd 352(%rsp), %ymm13, %ymm14 #68.42[spill]
12368 | 0.50 | 0.50 | | | | | | || 3.0 | | vmulpd %ymm14, %ymm13, %ymm14 #68.48
12369 | 0.50 | 0.50 | | | | | | || 3.0 | | vmulpd %ymm14, %ymm13, %ymm15 #68.54
12370 | 0.50 | 0.50 | | | | | | || | | vfmsub213pd %ymm9, %ymm13, %ymm14 #69.58
12371 | 0.50 | 0.50 | 0.50 0.50 | 0.50 0.50 | | | | || | 3.0 | vmulpd 320(%rsp), %ymm13, %ymm13 #69.58[spill]
12372 | 0.50 | 0.50 | | | | | | || 3.0 | 3.0 | vmulpd %ymm13, %ymm15, %ymm15 #69.65
12373 | 0.75 | 0.25 | | | | | | || 3.0 | 3.0 | vmulpd %ymm14, %ymm15, %ymm13 #69.71
12374 | 1.00 | 0.00 | | | | | | || | | vmulpd %ymm13, %ymm0, %ymm0 #70.35
12375 | 1.00 | 0.00 | | | | | | || | | vmulpd %ymm13, %ymm8, %ymm8 #71.35
12376 | | | | | | 1.00 | | || | | vandpd %ymm0, %ymm7, %ymm0 #70.35
12377 | | 1.00 | | | | | | || | | vaddpd %ymm0, %ymm12, %ymm12 #70.21
12378 | 1.00 | 0.00 | | | | | | || 3.0 | | vmulpd %ymm13, %ymm2, %ymm0 #72.35
12379 | | | | | | 1.00 | | || | | vandpd %ymm8, %ymm7, %ymm2 #71.35
12380 | | | | | | 1.00 | | || 1.0 | | vandpd %ymm0, %ymm7, %ymm7 #72.35
12381 | | 1.00 | | | | | | || | | vaddpd %ymm2, %ymm1, %ymm1 #71.21
12382 | | 1.00 | | | | | | || 3.0 | | vaddpd %ymm7, %ymm3, %ymm3 #72.21
12383 | | | | | | | | || | | # LOE rax rdx rsi rdi r8 r9 r12 r14 ecx ebx r10d r13d r15d ymm1 ymm3 ymm4 ymm5 ymm6 ymm9 ymm10 ymm11 ymm12
12384 | | | | | | | | || | | ..B1.26: # Preds ..B1.25 ..B1.24
12385 | | | | | | | | || | | # Execution count [2.50e+04]
12386 | 0.00 | 0.00 | | | | 0.00 | 1.00 | || | | addq $4, %rdx #59.13
12387 | 0.00 | 0.00 | | | | 0.00 | 1.00 | || | | cmpq %r14, %rdx #59.13
12388 | | | | | | | | || | | * jb ..B1.24 # Prob 82% #59.13
11.8 16.00 12.2 7.50 1.50 7.50 1.50 3.00 16.0 5.00 70 46
Loop-Carried Dependencies Analysis Report
-----------------------------------------
12373 | 46.0 | vmulpd %ymm14, %ymm15, %ymm13 #69.71| [12344, 12354, 12355, 12357, 12359, 12366, 12371, 12372, 12373]
12377 | 3.0 | vaddpd %ymm0, %ymm12, %ymm12 #70.21| [12377]
12381 | 3.0 | vaddpd %ymm2, %ymm1, %ymm1 #71.21| [12381]
12382 | 3.0 | vaddpd %ymm7, %ymm3, %ymm3 #72.21| [12382]
12386 | 1.0 | addq $4, %rdx #59.13| [12386]

79
arch_analysis/avx512/iaca_force_aos_geq1200.txt
View File

@ -1,79 +0,0 @@
iwia021h@testfront1:~/MD-Bench/asm$ iaca -arch SKX force_aos_geq1200_markers.o
Intel(R) Architecture Code Analyzer Version - v3.0-28-g1ba2cbb build date: 2017-10-23;16:42:45
Analyzed File - force_aos_geq1200_markers.o
Binary Format - 64Bit
Architecture - SKX
Analysis Type - Throughput
Throughput Analysis Report
--------------------------
Block Throughput: 33.05 Cycles Throughput Bottleneck: Backend
Loop Count: 22
Port Binding In Cycles Per Iteration:
--------------------------------------------------------------------------------------------------
| Port | 0 - DV | 1 | 2 - D | 3 - D | 4 | 5 | 6 | 7 |
--------------------------------------------------------------------------------------------------
| Cycles | 20.0 0.0 | 4.5 | 13.0 13.0 | 13.0 13.0 | 0.0 | 18.0 | 4.5 | 0.0 |
--------------------------------------------------------------------------------------------------
DV - Divider pipe (on port 0)
D - Data fetch pipe (on ports 2 and 3)
F - Macro Fusion with the previous instruction occurred
* - instruction micro-ops not bound to a port
^ - Micro Fusion occurred
# - ESP Tracking sync uop was issued
@ - SSE instruction followed an AVX256/AVX512 instruction, dozens of cycles penalty is expected
X - instruction not supported, was not accounted in Analysis
| Num Of | Ports pressure in cycles | |
| Uops | 0 - DV | 1 | 2 - D | 3 - D | 4 | 5 | 6 | 7 |
-----------------------------------------------------------------------------------------
| 1 | | | | | | 1.0 | | | vpcmpgtd k3, ymm2, ymm3
| 2 | | 1.0 | 1.0 1.0 | | | | | | vmovdqu32 ymm17{k3}{z}, ymmword ptr [r15+r13*4]
| 1 | 1.0 | | | | | | | | kmovw r9d, k3
| 1 | | 1.0 | | | | | | | vpaddd ymm18, ymm17, ymm17
| 1 | | 1.0 | | | | | | | vpaddd ymm17, ymm17, ymm18
| 1 | 1.0 | | | | | | | | kmovw k1, k3
| 1 | 1.0 | | | | | | | | kmovw k2, k3
| 1* | | | | | | | | | vpxord zmm18, zmm18, zmm18
| 1* | | | | | | | | | vpxord zmm19, zmm19, zmm19
| 1* | | | | | | | | | vpxord zmm20, zmm20, zmm20
| 5^ | 1.0 | | 4.0 4.0 | 4.0 4.0 | | 1.0 | 1.0 | | vgatherdpd zmm18, k1, zmmword ptr [rdi+ymm17*8+0x10]
| 5^ | 1.0 | | 4.0 4.0 | 4.0 4.0 | | 1.0 | 1.0 | | vgatherdpd zmm19, k2, zmmword ptr [rdi+ymm17*8+0x8]
| 5^ | 1.0 | | 4.0 4.0 | 4.0 4.0 | | 1.0 | 1.0 | | vgatherdpd zmm20, k3, zmmword ptr [rdi+ymm17*8]
| 1 | | 0.5 | | | | | 0.5 | | add r13, 0x8
| 1 | | 1.0 | | | | | | | vpaddd ymm3, ymm3, ymm16
| 1 | 0.5 | | | | | 0.5 | | | vsubpd zmm29, zmm4, zmm18
| 1 | 0.5 | | | | | 0.5 | | | vsubpd zmm27, zmm0, zmm19
| 1 | 0.5 | | | | | 0.5 | | | vsubpd zmm26, zmm1, zmm20
| 1 | 0.5 | | | | | 0.5 | | | vmulpd zmm25, zmm27, zmm27
| 1 | 0.5 | | | | | 0.5 | | | vfmadd231pd zmm25, zmm26, zmm26
| 1 | 0.5 | | | | | 0.5 | | | vfmadd231pd zmm25, zmm29, zmm29
| 3 | 2.0 | | | | | 1.0 | | | vrcp14pd zmm24, zmm25
| 1 | | | | | | 1.0 | | | vcmppd k2, zmm25, zmm14, 0x1
| 1 | | | | | | 1.0 | | | vfpclasspd k0, zmm24, 0x1e
| 1 | 1.0 | | | | | | | | kmovw edx, k2
| 1 | 1.0 | | | | | | | | knotw k1, k0
| 1* | | | | | | | | | vmovaps zmm17, zmm25
| 1 | | | | | | | 1.0 | | and r9d, edx
| 2^ | | | | 1.0 1.0 | | 1.0 | | | vfnmadd213pd zmm17, zmm24, qword ptr [rip]{1to8}
| 1 | | | | | | 1.0 | | | kmovw k3, r9d
| 1 | 1.0 | | | | | | | | vmulpd zmm18, zmm17, zmm17
| 1 | 0.5 | | | | | 0.5 | | | vfmadd213pd zmm24{k1}, zmm17, zmm24
| 1 | 0.5 | | | | | 0.5 | | | vfmadd213pd zmm24{k1}, zmm18, zmm24
| 1 | 0.5 | | | | | 0.5 | | | vmulpd zmm19, zmm24, zmm13
| 1 | 0.5 | | | | | 0.5 | | | vmulpd zmm21, zmm24, zmm10
| 1 | 0.5 | | | | | 0.5 | | | vmulpd zmm22, zmm24, zmm19
| 1 | 0.5 | | | | | 0.5 | | | vmulpd zmm20, zmm24, zmm22
| 1 | 0.5 | | | | | 0.5 | | | vfmsub213pd zmm24, zmm22, zmm5
| 1 | 0.5 | | | | | 0.5 | | | vmulpd zmm23, zmm20, zmm21
| 1 | 0.5 | | | | | 0.5 | | | vmulpd zmm28, zmm23, zmm24
| 1 | 0.5 | | | | | 0.5 | | | vfmadd231pd zmm9{k3}, zmm28, zmm26
| 1 | 0.5 | | | | | 0.5 | | | vfmadd231pd zmm8{k3}, zmm28, zmm27
| 1 | 0.5 | | | | | 0.5 | | | vfmadd231pd zmm11{k3}, zmm28, zmm29
| 1* | | | | | | | | | cmp r13, rbx
| 0*F | | | | | | | | | jb 0xfffffffffffffef7
Total Num Of Uops: 60
Analysis Notes:
Backend allocation was stalled due to unavailable allocation resources.
There were bubbles in the frontend.

104
arch_analysis/avx512/iaca_force_aos_lt1200.txt
View File

@ -1,104 +0,0 @@
iwia021h@testfront1:~/MD-Bench/ICC$ iaca -arch SKX force.o
Intel(R) Architecture Code Analyzer Version - v3.0-28-g1ba2cbb build date: 2017-10-23;16:42:45
Analyzed File - force.o
Binary Format - 64Bit
Architecture - SKX
Analysis Type - Throughput
Throughput Analysis Report
--------------------------
Block Throughput: 36.70 Cycles Throughput Bottleneck: Backend
Loop Count: 23
Port Binding In Cycles Per Iteration:
--------------------------------------------------------------------------------------------------
| Port | 0 - DV | 1 | 2 - D | 3 - D | 4 | 5 | 6 | 7 |
--------------------------------------------------------------------------------------------------
| Cycles | 17.5 0.0 | 11.0 | 20.5 17.0 | 20.5 17.0 | 7.0 | 20.5 | 7.0 | 0.0 |
--------------------------------------------------------------------------------------------------
DV - Divider pipe (on port 0)
D - Data fetch pipe (on ports 2 and 3)
F - Macro Fusion with the previous instruction occurred
* - instruction micro-ops not bound to a port
^ - Micro Fusion occurred
# - ESP Tracking sync uop was issued
@ - SSE instruction followed an AVX256/AVX512 instruction, dozens of cycles penalty is expected
X - instruction not supported, was not accounted in Analysis
| Num Of | Ports pressure in cycles | |
| Uops | 0 - DV | 1 | 2 - D | 3 - D | 4 | 5 | 6 | 7 |
-----------------------------------------------------------------------------------------
| 1* | | | | | | | | | mov r13, r8
| 1 | | 1.0 | | | | | | | imul r13, rcx
| 1 | | | | | | 1.0 | | | vbroadcastsd zmm2, xmm6
| 1 | | | | | | 1.0 | | | vbroadcastsd zmm1, xmm7
| 1 | | | | | | 1.0 | | | vbroadcastsd zmm0, xmm12
| 1 | | | | | | | 1.0 | | movsxd rbx, r12d
| 1 | | | | | | | 1.0 | | add r13, r10
| 2^ | | | 0.5 | 0.5 | 1.0 | | | | mov qword ptr [rsp-0x40], rax
| 2^ | | | 0.5 | 0.5 | 1.0 | | | | mov qword ptr [rsp-0x38], r8
| 2^ | | | 0.5 | 0.5 | 1.0 | | | | mov qword ptr [rsp-0x30], r10
| 2^ | | | 0.5 | 0.5 | 1.0 | | | | mov qword ptr [rsp-0x28], rsi
| 2^ | | | 0.5 | 0.5 | 1.0 | | | | mov qword ptr [rsp-0x20], rcx
| 2^ | | | 0.5 | 0.5 | 1.0 | | | | mov qword ptr [rsp-0x50], r9
| 2^ | | | 0.5 | 0.5 | 1.0 | | | | mov qword ptr [rsp-0x48], rdx
| 1 | | | 0.5 0.5 | 0.5 0.5 | | | | | vmovdqu ymm3, ymmword ptr [r13+rbx*4]
| 1 | | 1.0 | | | | | | | vpaddd ymm4, ymm3, ymm3
| 1 | | 1.0 | | | | | | | vpaddd ymm3, ymm3, ymm4
| 1 | | | 0.5 0.5 | 0.5 0.5 | | | | | mov r10d, dword ptr [r13+rbx*4]
| 1 | | | 0.5 0.5 | 0.5 0.5 | | | | | mov r9d, dword ptr [r13+rbx*4+0x4]
| 1 | | | 0.5 0.5 | 0.5 0.5 | | | | | mov r8d, dword ptr [r13+rbx*4+0x8]
| 1 | | | 0.5 0.5 | 0.5 0.5 | | | | | mov esi, dword ptr [r13+rbx*4+0xc]
| 1 | | 1.0 | | | | | | | lea r10d, ptr [r10+r10*2]
| 1 | | | 0.5 0.5 | 0.5 0.5 | | | | | mov ecx, dword ptr [r13+rbx*4+0x10]
| 1 | | 1.0 | | | | | | | lea r9d, ptr [r9+r9*2]
| 1 | | | 0.5 0.5 | 0.5 0.5 | | | | | mov edx, dword ptr [r13+rbx*4+0x14]
| 1 | | 1.0 | | | | | | | lea r8d, ptr [r8+r8*2]
| 1 | | | 0.5 0.5 | 0.5 0.5 | | | | | mov eax, dword ptr [r13+rbx*4+0x18]
| 1 | | 1.0 | | | | | | | lea esi, ptr [rsi+rsi*2]
| 1 | | | 0.5 0.5 | 0.5 0.5 | | | | | mov r15d, dword ptr [r13+rbx*4+0x1c]
| 1 | | 1.0 | | | | | | | lea ecx, ptr [rcx+rcx*2]
| 1 | | 1.0 | | | | | | | lea edx, ptr [rdx+rdx*2]
| 1 | | 1.0 | | | | | | | lea eax, ptr [rax+rax*2]
| 1 | | 1.0 | | | | | | | lea r15d, ptr [r15+r15*2]
| 1 | | | | | | 1.0 | | | vpcmpeqb k1, xmm0, xmm0
| 1 | | | | | | 1.0 | | | vpcmpeqb k2, xmm0, xmm0
| 1 | | | | | | 1.0 | | | vpcmpeqb k3, xmm0, xmm0
| 1* | | | | | | | | | vpxord zmm4, zmm4, zmm4
| 1* | | | | | | | | | vpxord zmm17, zmm17, zmm17
| 1* | | | | | | | | | vpxord zmm18, zmm18, zmm18
| 5^ | 2.0 | | 4.0 4.0 | 4.0 4.0 | | | 1.0 | | vgatherdpd zmm4, k1, zmmword ptr [rdi+ymm3*8+0x10]
| 5^ | 1.5 | | 4.0 4.0 | 4.0 4.0 | | 0.5 | 1.0 | | vgatherdpd zmm17, k2, zmmword ptr [rdi+ymm3*8+0x8]
| 5^ | 1.0 | | 4.0 4.0 | 4.0 4.0 | | 1.0 | 1.0 | | vgatherdpd zmm18, k3, zmmword ptr [rdi+ymm3*8]
| 1 | | | | | | | 1.0 | | add r12d, 0x8
| 1 | | | | | | | 1.0 | | add rbx, 0x8
| 1 | 0.5 | | | | | 0.5 | | | vsubpd zmm26, zmm0, zmm4
| 1 | 0.5 | | | | | 0.5 | | | vsubpd zmm24, zmm1, zmm17
| 1 | 0.5 | | | | | 0.5 | | | vsubpd zmm23, zmm2, zmm18
| 1 | 0.5 | | | | | 0.5 | | | vmulpd zmm3, zmm24, zmm24
| 1 | 0.5 | | | | | 0.5 | | | vfmadd231pd zmm3, zmm23, zmm23
| 1 | 0.5 | | | | | 0.5 | | | vfmadd231pd zmm3, zmm26, zmm26
| 3 | 2.0 | | | | | 1.0 | | | vrcp14pd zmm22, zmm3
| 1 | | | | | | 1.0 | | | vcmppd k2, zmm3, zmm14, 0x1
| 1 | | | | | | 1.0 | | | vfpclasspd k0, zmm22, 0x1e
| 2^ | 1.0 | | 0.5 0.5 | 0.5 0.5 | | | | | vfnmadd213pd zmm3, zmm22, qword ptr [rip]{1to8}
| 1 | 1.0 | | | | | | | | knotw k1, k0
| 1 | | | | | | 1.0 | | | vmulpd zmm4, zmm3, zmm3
| 1 | 0.5 | | | | | 0.5 | | | vfmadd213pd zmm22{k1}, zmm3, zmm22
| 1 | 0.5 | | | | | 0.5 | | | vfmadd213pd zmm22{k1}, zmm4, zmm22
| 1 | 0.5 | | | | | 0.5 | | | vmulpd zmm17, zmm22, zmm13
| 1 | 0.5 | | | | | 0.5 | | | vmulpd zmm19, zmm22, zmm10
| 1 | 0.5 | | | | | 0.5 | | | vmulpd zmm20, zmm22, zmm17
| 1 | 0.5 | | | | | 0.5 | | | vmulpd zmm18, zmm22, zmm20
| 1 | 0.5 | | | | | 0.5 | | | vfmsub213pd zmm22, zmm20, zmm5
| 1 | 0.5 | | | | | 0.5 | | | vmulpd zmm21, zmm18, zmm19
| 1 | 0.5 | | | | | 0.5 | | | vmulpd zmm25, zmm21, zmm22
| 1 | 0.5 | | | | | 0.5 | | | vfmadd231pd zmm9{k2}, zmm25, zmm23
| 1 | 0.5 | | | | | 0.5 | | | vfmadd231pd zmm8{k2}, zmm25, zmm24
| 1 | 0.5 | | | | | 0.5 | | | vfmadd231pd zmm11{k2}, zmm25, zmm26
| 1* | | | | | | | | | cmp r12d, r14d
| 0*F | | | | | | | | | jb 0xfffffffffffffed3
Total Num Of Uops: 91
Analysis Notes:
Backend allocation was stalled due to unavailable allocation resources.
There were bubbles in the frontend.

82
arch_analysis/avx512/iaca_force_aos_lt8.txt
View File

@ -1,82 +0,0 @@
Intel(R) Architecture Code Analyzer Version - v3.0-28-g1ba2cbb build date: 2017-10-23;16:42:45
Analyzed File - force_aos_lt8_markers.o
Binary Format - 64Bit
Architecture - SKX
Analysis Type - Throughput
Throughput Analysis Report
--------------------------
Block Throughput: 69.79 Cycles Throughput Bottleneck: Backend
Loop Count: 22
Port Binding In Cycles Per Iteration:
--------------------------------------------------------------------------------------------------
| Port | 0 - DV | 1 | 2 - D | 3 - D | 4 | 5 | 6 | 7 |
--------------------------------------------------------------------------------------------------
| Cycles | 21.0 0.0 | 5.5 | 13.0 13.0 | 13.0 13.0 | 0.0 | 21.0 | 5.5 | 0.0 |
--------------------------------------------------------------------------------------------------
DV - Divider pipe (on port 0)
D - Data fetch pipe (on ports 2 and 3)
F - Macro Fusion with the previous instruction occurred
* - instruction micro-ops not bound to a port
^ - Micro Fusion occurred
# - ESP Tracking sync uop was issued
@ - SSE instruction followed an AVX256/AVX512 instruction, dozens of cycles penalty is expected
X - instruction not supported, was not accounted in Analysis
| Num Of | Ports pressure in cycles | |
| Uops | 0 - DV | 1 | 2 - D | 3 - D | 4 | 5 | 6 | 7 |
-----------------------------------------------------------------------------------------
| 1 | | 1.0 | | | | | | | imul rcx, r8
| 1 | | | | | | 1.0 | | | vbroadcastsd zmm4, xmm6
| 1 | | | | | | | 1.0 | | sub r11d, r14d
| 1 | | 0.5 | | | | | 0.5 | | add rcx, r10
| 1 | | | | | | 1.0 | | | vpbroadcastd ymm0, r11d
| 1 | | | | | | 1.0 | | | vpcmpgtd k3, ymm0, ymm15
| 1 | | 0.5 | | | | | 0.5 | | movsxd r14, r14d
| 1 | 1.0 | | | | | | | | kmovw ebx, k3
| 2 | | 1.0 | 1.0 1.0 | | | | | | vmovdqu32 ymm1{k3}{z}, ymmword ptr [rcx+r14*4]
| 1 | | 1.0 | | | | | | | vpaddd ymm2, ymm1, ymm1
| 1 | | 1.0 | | | | | | | vpaddd ymm0, ymm1, ymm2
| 1 | 1.0 | | | | | | | | kmovw k1, k3
| 1 | 1.0 | | | | | | | | kmovw k2, k3
| 1* | | | | | | | | | vpxord zmm1, zmm1, zmm1
| 1* | | | | | | | | | vpxord zmm2, zmm2, zmm2
| 1* | | | | | | | | | vpxord zmm3, zmm3, zmm3
| 5^ | 1.0 | | 4.0 4.0 | 4.0 4.0 | | 1.0 | 1.0 | | vgatherdpd zmm1, k1, zmmword ptr [rdi+ymm0*8+0x10]
| 5^ | 1.0 | | 4.0 4.0 | 4.0 4.0 | | 1.0 | 1.0 | | vgatherdpd zmm2, k2, zmmword ptr [rdi+ymm0*8+0x8]
| 5^ | 1.0 | | 4.0 4.0 | 4.0 4.0 | | 1.0 | 1.0 | | vgatherdpd zmm3, k3, zmmword ptr [rdi+ymm0*8]
| 1 | | | | | | 1.0 | | | vbroadcastsd zmm7, xmm7
| 1 | | | | | | 1.0 | | | vbroadcastsd zmm12, xmm12
| 1 | 1.0 | | | | | | | | vsubpd zmm23, zmm12, zmm1
| 1 | 1.0 | | | | | | | | vsubpd zmm21, zmm7, zmm2
| 1 | 0.5 | | | | | 0.5 | | | vsubpd zmm20, zmm4, zmm3
| 1 | 0.5 | | | | | 0.5 | | | vmulpd zmm19, zmm21, zmm21
| 1 | 0.5 | | | | | 0.5 | | | vfmadd231pd zmm19, zmm20, zmm20
| 1 | 0.5 | | | | | 0.5 | | | vfmadd231pd zmm19, zmm23, zmm23
| 3 | 2.0 | | | | | 1.0 | | | vrcp14pd zmm18, zmm19
| 1 | | | | | | 1.0 | | | vcmppd k2, zmm19, zmm14, 0x1
| 1 | | | | | | 1.0 | | | vfpclasspd k0, zmm18, 0x1e
| 1 | 1.0 | | | | | | | | kmovw ecx, k2
| 1 | 1.0 | | | | | | | | knotw k1, k0
| 1* | | | | | | | | | vmovaps zmm0, zmm19
| 1 | | 0.5 | | | | | 0.5 | | and ebx, ecx
| 2^ | | | | 1.0 1.0 | | 1.0 | | | vfnmadd213pd zmm0, zmm18, qword ptr [rip]{1to8}
| 1 | | | | | | 1.0 | | | kmovw k3, ebx
| 1 | 1.0 | | | | | | | | vmulpd zmm1, zmm0, zmm0
| 1 | 0.5 | | | | | 0.5 | | | vfmadd213pd zmm18{k1}, zmm0, zmm18
| 1 | 0.5 | | | | | 0.5 | | | vfmadd213pd zmm18{k1}, zmm1, zmm18
| 1 | 0.5 | | | | | 0.5 | | | vmulpd zmm2, zmm18, zmm13
| 1 | 0.5 | | | | | 0.5 | | | vmulpd zmm4, zmm18, zmm10
| 1 | 0.5 | | | | | 0.5 | | | vmulpd zmm6, zmm18, zmm2
| 1 | 0.5 | | | | | 0.5 | | | vmulpd zmm3, zmm18, zmm6
| 1 | 0.5 | | | | | 0.5 | | | vfmsub213pd zmm18, zmm6, zmm5
| 1 | 0.5 | | | | | 0.5 | | | vmulpd zmm17, zmm3, zmm4
| 1 | 0.5 | | | | | 0.5 | | | vmulpd zmm22, zmm17, zmm18
| 1 | 0.5 | | | | | 0.5 | | | vfmadd231pd zmm9{k3}, zmm22, zmm20
| 1 | 0.5 | | | | | 0.5 | | | vfmadd231pd zmm8{k3}, zmm22, zmm21
| 1 | 0.5 | | | | | 0.5 | | | vfmadd231pd zmm11{k3}, zmm22, zmm23
Total Num Of Uops: 65
Analysis Notes:
Backend allocation was stalled due to unavailable allocation resources.
There were bubbles in the frontend.

74
arch_analysis/avx512/iaca_force_soa_geq1200.txt
View File

@ -1,74 +0,0 @@
iwia021h@testfront1:~/MD-Bench/asm$ iaca -arch SKX force_soa_geq1200_markers.o
Intel(R) Architecture Code Analyzer Version - v3.0-28-g1ba2cbb build date: 2017-10-23;16:42:45
Analyzed File - force_soa_geq1200_markers.o
Binary Format - 64Bit
Architecture - SKX
Analysis Type - Throughput
Throughput Analysis Report
--------------------------
Block Throughput: 31.47 Cycles Throughput Bottleneck: Backend
Loop Count: 22
Port Binding In Cycles Per Iteration:
--------------------------------------------------------------------------------------------------
| Port | 0 - DV | 1 | 2 - D | 3 - D | 4 | 5 | 6 | 7 |
--------------------------------------------------------------------------------------------------
| Cycles | 18.0 0.0 | 3.0 | 13.0 13.0 | 13.0 13.0 | 0.0 | 18.0 | 3.0 | 0.0 |
--------------------------------------------------------------------------------------------------
DV - Divider pipe (on port 0)
D - Data fetch pipe (on ports 2 and 3)
F - Macro Fusion with the previous instruction occurred
* - instruction micro-ops not bound to a port
^ - Micro Fusion occurred
# - ESP Tracking sync uop was issued
@ - SSE instruction followed an AVX256/AVX512 instruction, dozens of cycles penalty is expected
X - instruction not supported, was not accounted in Analysis
| Num Of | Ports pressure in cycles | |
| Uops | 0 - DV | 1 | 2 - D | 3 - D | 4 | 5 | 6 | 7 |
-----------------------------------------------------------------------------------------
| 1 | | | | | | 1.0 | | | vpcmpgtd k5, ymm3, ymm4
| 1 | | 1.0 | | | | | | | vpaddd ymm4, ymm4, ymm18
| 2 | | 1.0 | 1.0 1.0 | | | | | | vmovdqu32 ymm20{k5}{z}, ymmword ptr [rcx+r15*4]
| 1* | | | | | | | | | vmovaps zmm22, zmm19
| 1 | | 1.0 | | | | | | | add r15, 0x8
| 1 | 1.0 | | | | | | | | kmovw k2, k5
| 1* | | | | | | | | | vmovaps zmm21, zmm19
| 1 | 1.0 | | | | | | | | kmovw k1, k5
| 1* | | | | | | | | | vmovaps zmm23, zmm19
| 1 | 1.0 | | | | | | | | kmovw k3, k5
| 5^ | 1.0 | | 4.0 4.0 | 4.0 4.0 | | 1.0 | 1.0 | | vgatherdpd zmm23, k3, zmmword ptr [rsi+ymm20*8]
| 5^ | 1.0 | | 4.0 4.0 | 4.0 4.0 | | 1.0 | 1.0 | | vgatherdpd zmm22, k2, zmmword ptr [rax+ymm20*8]
| 5^ | 1.0 | | 4.0 4.0 | 4.0 4.0 | | 1.0 | 1.0 | | vgatherdpd zmm21, k1, zmmword ptr [rdx+ymm20*8]
| 1 | | | | | | 1.0 | | | vsubpd zmm0, zmm5, zmm22
| 1 | | | | | | 1.0 | | | vsubpd zmm1, zmm2, zmm21
| 1 | | | | | | 1.0 | | | vsubpd zmm21, zmm6, zmm23
| 1 | 1.0 | | | | | | | | vmulpd zmm20, zmm0, zmm0
| 1 | | | | | | 1.0 | | | vfmadd231pd zmm20, zmm1, zmm1
| 1 | 1.0 | | | | | | | | vfmadd231pd zmm20, zmm21, zmm21
| 3 | 2.0 | | | | | 1.0 | | | vrcp14pd zmm31, zmm20
| 1 | | | | | | 1.0 | | | vcmppd k6{k5}, zmm20, zmm16, 0x1
| 1 | | | | | | 1.0 | | | vfpclasspd k0, zmm31, 0x1e
| 1* | | | | | | | | | vmovaps zmm24, zmm20
| 2^ | 1.0 | | | 1.0 1.0 | | | | | vfnmadd213pd zmm24, zmm31, qword ptr [rip]{1to8}
| 1 | 1.0 | | | | | | | | knotw k4, k0
| 1 | | | | | | 1.0 | | | vmulpd zmm25, zmm24, zmm24
| 1 | | | | | | 1.0 | | | vfmadd213pd zmm31{k4}, zmm24, zmm31
| 1 | 1.0 | | | | | | | | vfmadd213pd zmm31{k4}, zmm25, zmm31
| 1 | | | | | | 1.0 | | | vmulpd zmm26, zmm31, zmm15
| 1 | 1.0 | | | | | | | | vmulpd zmm28, zmm31, zmm14
| 1 | | | | | | 1.0 | | | vmulpd zmm29, zmm31, zmm26
| 1 | 1.0 | | | | | | | | vmulpd zmm27, zmm31, zmm29
| 1 | | | | | | 1.0 | | | vfmsub213pd zmm31, zmm29, zmm7
| 1 | 1.0 | | | | | | | | vmulpd zmm30, zmm27, zmm28
| 1 | | | | | | 1.0 | | | vmulpd zmm24, zmm30, zmm31
| 1 | 1.0 | | | | | | | | vfmadd231pd zmm13{k6}, zmm24, zmm1
| 1 | | | | | | 1.0 | | | vfmadd231pd zmm12{k6}, zmm24, zmm0
| 1 | 1.0 | | | | | | | | vfmadd231pd zmm11{k6}, zmm24, zmm21
| 1* | | | | | | | | | cmp r15, r14
| 0*F | | | | | | | | | jb 0xffffffffffffff19
Total Num Of Uops: 55
Analysis Notes:
Backend allocation was stalled due to unavailable allocation resources.
There were bubbles in the frontend.

72
arch_analysis/avx512/iaca_force_soa_lt1200.txt
View File

@ -1,72 +0,0 @@
iwia021h@testfront1:~/MD-Bench/asm$ iaca -arch SKX force_soa_lt1200_markers.o
Intel(R) Architecture Code Analyzer Version - v3.0-28-g1ba2cbb build date: 2017-10-23;16:42:45
Analyzed File - force_soa_lt1200_markers.o
Binary Format - 64Bit
Architecture - SKX
Analysis Type - Throughput
Throughput Analysis Report
--------------------------
Block Throughput: 30.25 Cycles Throughput Bottleneck: Backend
Loop Count: 23
Port Binding In Cycles Per Iteration:
--------------------------------------------------------------------------------------------------
| Port | 0 - DV | 1 | 2 - D | 3 - D | 4 | 5 | 6 | 7 |
--------------------------------------------------------------------------------------------------
| Cycles | 16.0 0.0 | 2.0 | 13.0 13.0 | 13.0 13.0 | 0.0 | 19.0 | 3.0 | 0.0 |
--------------------------------------------------------------------------------------------------
DV - Divider pipe (on port 0)
D - Data fetch pipe (on ports 2 and 3)
F - Macro Fusion with the previous instruction occurred
* - instruction micro-ops not bound to a port
^ - Micro Fusion occurred
# - ESP Tracking sync uop was issued
@ - SSE instruction followed an AVX256/AVX512 instruction, dozens of cycles penalty is expected
X - instruction not supported, was not accounted in Analysis
| Num Of | Ports pressure in cycles | |
| Uops | 0 - DV | 1 | 2 - D | 3 - D | 4 | 5 | 6 | 7 |
-----------------------------------------------------------------------------------------
| 1 | | | | | | 1.0 | | | vpcmpeqb k2, xmm0, xmm0
| 1 | | 1.0 | | | | | | | add r9d, 0x8
| 1 | | | | | | 1.0 | | | vpcmpeqb k1, xmm0, xmm0
| 1 | | | | | | 1.0 | | | vpcmpeqb k3, xmm0, xmm0
| 1 | | | 1.0 1.0 | | | | | | vmovdqu ymm3, ymmword ptr [rcx+r14*4]
| 1 | | 1.0 | | | | | | | add r14, 0x8
| 1* | | | | | | | | | vpxord zmm5, zmm5, zmm5
| 1* | | | | | | | | | vpxord zmm4, zmm4, zmm4
| 1* | | | | | | | | | vpxord zmm6, zmm6, zmm6
| 5^ | 1.0 | | 4.0 4.0 | 4.0 4.0 | | 1.0 | 1.0 | | vgatherdpd zmm5, k2, zmmword ptr [rax+ymm3*8]
| 5^ | 1.0 | | 4.0 4.0 | 4.0 4.0 | | 1.0 | 1.0 | | vgatherdpd zmm4, k1, zmmword ptr [rdx+ymm3*8]
| 5^ | 1.0 | | 4.0 4.0 | 4.0 4.0 | | 1.0 | 1.0 | | vgatherdpd zmm6, k3, zmmword ptr [rsi+ymm3*8]
| 1 | | | | | | 1.0 | | | vsubpd zmm29, zmm1, zmm5
| 1 | 1.0 | | | | | | | | vsubpd zmm28, zmm0, zmm4
| 1 | | | | | | 1.0 | | | vsubpd zmm31, zmm2, zmm6
| 1 | 1.0 | | | | | | | | vmulpd zmm20, zmm29, zmm29
| 1 | | | | | | 1.0 | | | vfmadd231pd zmm20, zmm28, zmm28
| 1 | 1.0 | | | | | | | | vfmadd231pd zmm20, zmm31, zmm31
| 3 | 2.0 | | | | | 1.0 | | | vrcp14pd zmm27, zmm20
| 1 | | | | | | 1.0 | | | vcmppd k5, zmm20, zmm16, 0x1
| 1 | | | | | | 1.0 | | | vfpclasspd k0, zmm27, 0x1e
| 2^ | 1.0 | | | 1.0 1.0 | | | | | vfnmadd213pd zmm20, zmm27, qword ptr [rip]{1to8}
| 1 | 1.0 | | | | | | | | knotw k4, k0
| 1 | | | | | | 1.0 | | | vmulpd zmm21, zmm20, zmm20
| 1 | | | | | | 1.0 | | | vfmadd213pd zmm27{k4}, zmm20, zmm27
| 1 | 1.0 | | | | | | | | vfmadd213pd zmm27{k4}, zmm21, zmm27
| 1 | | | | | | 1.0 | | | vmulpd zmm22, zmm27, zmm15
| 1 | 1.0 | | | | | | | | vmulpd zmm24, zmm27, zmm14
| 1 | | | | | | 1.0 | | | vmulpd zmm25, zmm27, zmm22
| 1 | 1.0 | | | | | | | | vmulpd zmm23, zmm27, zmm25
| 1 | | | | | | 1.0 | | | vfmsub213pd zmm27, zmm25, zmm7
| 1 | 1.0 | | | | | | | | vmulpd zmm26, zmm23, zmm24
| 1 | | | | | | 1.0 | | | vmulpd zmm30, zmm26, zmm27
| 1 | 1.0 | | | | | | | | vfmadd231pd zmm13{k5}, zmm30, zmm28
| 1 | | | | | | 1.0 | | | vfmadd231pd zmm12{k5}, zmm30, zmm29
| 1 | 1.0 | | | | | | | | vfmadd231pd zmm11{k5}, zmm30, zmm31
| 1* | | | | | | | | | cmp r9d, ebx
| 0*F | | | | | | | | | jb 0xffffffffffffff22
Total Num Of Uops: 52
Analysis Notes:
Backend allocation was stalled due to unavailable allocation resources.
There were bubbles in the frontend.

78
arch_analysis/avx512/iaca_force_soa_lt8.txt
View File

@ -1,78 +0,0 @@
iwia021h@testfront1:~/MD-Bench/asm$ iaca -arch SKX force_soa_lt8_markers.o
Intel(R) Architecture Code Analyzer Version - v3.0-28-g1ba2cbb build date: 2017-10-23;16:42:45
Analyzed File - force_soa_lt8_markers.o
Binary Format - 64Bit
Architecture - SKX
Analysis Type - Throughput
Throughput Analysis Report
--------------------------
Block Throughput: 35.00 Cycles Throughput Bottleneck: Backend
Loop Count: 22
Port Binding In Cycles Per Iteration:
--------------------------------------------------------------------------------------------------
| Port | 0 - DV | 1 | 2 - D | 3 - D | 4 | 5 | 6 | 7 |
--------------------------------------------------------------------------------------------------
| Cycles | 20.0 0.0 | 4.0 | 13.0 13.0 | 13.0 13.0 | 0.0 | 20.0 | 4.0 | 0.0 |
--------------------------------------------------------------------------------------------------
DV - Divider pipe (on port 0)
D - Data fetch pipe (on ports 2 and 3)
F - Macro Fusion with the previous instruction occurred
* - instruction micro-ops not bound to a port
^ - Micro Fusion occurred
# - ESP Tracking sync uop was issued
@ - SSE instruction followed an AVX256/AVX512 instruction, dozens of cycles penalty is expected
X - instruction not supported, was not accounted in Analysis
| Num Of | Ports pressure in cycles | |
| Uops | 0 - DV | 1 | 2 - D | 3 - D | 4 | 5 | 6 | 7 |
-----------------------------------------------------------------------------------------
| 1 | | 1.0 | | | | | | | imul r8, r12
| 1 | | | | | | 1.0 | | | vbroadcastsd zmm9, xmm9
| 1 | | | | | | 1.0 | | | vbroadcastsd zmm2, xmm8
| 1 | | | | | | 1.0 | | | vbroadcastsd zmm10, xmm10
| 1 | | 1.0 | | | | | | | sub r13d, ebx
| 1 | | 1.0 | | | | | | | add r8, r11
| 1 | | | | | | 1.0 | | | vpbroadcastd ymm0, r13d
| 1 | | | | | | 1.0 | | | vpcmpgtd k5, ymm0, ymm17
| 1 | | | | | | | 1.0 | | movsxd rbx, ebx
| 1* | | | | | | | | | vmovaps zmm4, zmm19
| 1 | 1.0 | | | | | | | | kmovw k2, k5
| 1* | | | | | | | | | vmovaps zmm3, zmm19
| 2 | | 1.0 | 1.0 1.0 | | | | | | vmovdqu32 ymm1{k5}{z}, ymmword ptr [r8+rbx*4]
| 1 | 1.0 | | | | | | | | kmovw k1, k5
| 1* | | | | | | | | | vmovaps zmm5, zmm19
| 1 | 1.0 | | | | | | | | kmovw k3, k5
| 5^ | 2.0 | | 4.0 4.0 | 4.0 4.0 | | | 1.0 | | vgatherdpd zmm5, k3, zmmword ptr [rsi+ymm1*8]
| 5^ | 1.0 | | 4.0 4.0 | 4.0 4.0 | | 1.0 | 1.0 | | vgatherdpd zmm4, k2, zmmword ptr [rax+ymm1*8]
| 5^ | 1.0 | | 4.0 4.0 | 4.0 4.0 | | 1.0 | 1.0 | | vgatherdpd zmm3, k1, zmmword ptr [rdx+ymm1*8]
| 1 | 1.0 | | | | | | | | vsubpd zmm30, zmm10, zmm5
| 1 | | | | | | 1.0 | | | vsubpd zmm28, zmm9, zmm4
| 1 | 1.0 | | | | | | | | vsubpd zmm27, zmm2, zmm3
| 1 | | | | | | 1.0 | | | vmulpd zmm26, zmm28, zmm28
| 1 | 1.0 | | | | | | | | vfmadd231pd zmm26, zmm27, zmm27
| 1 | | | | | | 1.0 | | | vfmadd231pd zmm26, zmm30, zmm30
| 3 | 2.0 | | | | | 1.0 | | | vrcp14pd zmm25, zmm26
| 1 | | | | | | 1.0 | | | vcmppd k6{k5}, zmm26, zmm16, 0x1
| 1 | | | | | | 1.0 | | | vfpclasspd k0, zmm25, 0x1e
| 1* | | | | | | | | | vmovaps zmm6, zmm26
| 2^ | 1.0 | | | 1.0 1.0 | | | | | vfnmadd213pd zmm6, zmm25, qword ptr [rip]{1to8}
| 1 | 1.0 | | | | | | | | knotw k4, k0
| 1 | | | | | | 1.0 | | | vmulpd zmm8, zmm6, zmm6
| 1 | 1.0 | | | | | | | | vfmadd213pd zmm25{k4}, zmm6, zmm25
| 1 | | | | | | 1.0 | | | vfmadd213pd zmm25{k4}, zmm8, zmm25
| 1 | 1.0 | | | | | | | | vmulpd zmm20, zmm25, zmm15
| 1 | | | | | | 1.0 | | | vmulpd zmm22, zmm25, zmm14
| 1 | 1.0 | | | | | | | | vmulpd zmm23, zmm25, zmm20
| 1 | | | | | | 1.0 | | | vmulpd zmm21, zmm25, zmm23
| 1 | 1.0 | | | | | | | | vfmsub213pd zmm25, zmm23, zmm7
| 1 | | | | | | 1.0 | | | vmulpd zmm24, zmm21, zmm22
| 1 | 1.0 | | | | | | | | vmulpd zmm29, zmm24, zmm25
| 1 | | | | | | 1.0 | | | vfmadd231pd zmm13{k6}, zmm29, zmm27
| 1 | 1.0 | | | | | | | | vfmadd231pd zmm12{k6}, zmm29, zmm28
| 1 | | | | | | 1.0 | | | vfmadd231pd zmm11{k6}, zmm29, zmm30
Total Num Of Uops: 60
Analysis Notes:
Backend allocation was stalled due to unavailable allocation resources.
There were bubbles in the frontend.

80
arch_analysis/avx512/osaca_force_aos_geq1200.txt
View File

@ -1,80 +0,0 @@
iwia021h@testfront1:~/MD-Bench/asm$ /home/hpc/iwia/iwia021h/.local/bin/osaca --ignore-unknown --arch=CSX force_aos_geq1200_markers.s
Open Source Architecture Code Analyzer (OSACA) - 0.3.14
Analyzed file: force_aos_geq1200_markers.s
Architecture: CSX
Timestamp: 2021-04-29 15:53:50
P - Throughput of LOAD operation can be hidden behind a past or future STORE instruction
* - Instruction micro-ops not bound to a port
X - No throughput/latency information for this instruction in data file
Combined Analysis Report
------------------------
Port pressure in cycles
| 0 - 0DV | 1 | 2 - 2D | 3 - 3D | 4 | 5 | 6 | 7 || CP | LCD |
-------------------------------------------------------------------------------------------------
196 | | | | | | 1.00 | | || | | vpcmpgtd %ymm3, %ymm2, %k3 #67.9
197 | 0.00 | 1.00 | 0.50 0.50 | 0.50 0.50 | | 0.00 | | || 4.0 | | vmovdqu32 (%r15,%r13,4), %ymm17{%k3}{z} #68.21
198 | 1.00 | | | | | | | || | | kmovw %k3, %r9d #67.9
199 | 0.00 | 1.00 | | | | 0.00 | | || 1.0 | | vpaddd %ymm17, %ymm17, %ymm18 #69.36
200 | 0.00 | 1.00 | | | | 0.00 | | || 1.0 | | vpaddd %ymm18, %ymm17, %ymm17 #69.36
201 | | | | | | | | || | | # LOE rax rcx rbx rbp rsi rdi r8 r10 r13 r15 r9d r11d r12d r14d xmm6 xmm7 xmm12 ymm2 ymm3 ymm15 ymm16 ymm17 zmm0 zmm1 zmm4 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14 k3
202 | | | | | | | | || | | ..B1.21: # Preds ..B1.18
203 | | | | | | | | || | | # Execution count [1.25e+01]
204 | 1.00 | | | | | | | || | | kmovw %k3, %k1 #69.36
205 | 1.00 | | | | | | | || | | kmovw %k3, %k2 #69.36
206 | 0.50 | | | | | 0.50 | | || | | vpxord %zmm18, %zmm18, %zmm18 #69.36
207 | 0.50 | | | | | 0.50 | | || | | vpxord %zmm19, %zmm19, %zmm19 #69.36
208 | 0.50 | | | | | 0.50 | | || | | vpxord %zmm20, %zmm20, %zmm20 #69.36
209 | 1.50 | 0.50 | 4.00 0.50 | 4.00 0.50 | | 0.50 | 0.50 | || | | vgatherdpd 16(%rdi,%ymm17,8), %zmm18{%k1} #69.36
210 | 1.50 | 0.50 | 4.00 0.50 | 4.00 0.50 | | 0.50 | 0.50 | || 4.0 | | vgatherdpd 8(%rdi,%ymm17,8), %zmm19{%k2} #69.36
211 | 1.50 | 0.50 | 4.00 0.50 | 4.00 0.50 | | 0.50 | 0.50 | || | | vgatherdpd (%rdi,%ymm17,8), %zmm20{%k3} #69.36
212 | | | | | | | | || | | # LOE rax rcx rbx rbp rsi rdi r8 r10 r13 r15 r9d r11d r12d r14d xmm6 xmm7 xmm12 ymm2 ymm3 ymm15 ymm16 zmm0 zmm1 zmm4 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14 zmm18 zmm19 zmm20
213 | | | | | | | | || | | ..B1.22: # Preds ..B1.21
214 | | | | | | | | || | | # Execution count [2.50e+01]
215 | 0.00 | 0.16 | | | | 0.00 | 0.84 | || | | addq $8, %r13 #67.9
216 | 0.00 | 1.00 | | | | 0.00 | | || | | vpaddd %ymm16, %ymm3, %ymm3 #67.9
217 | 0.50 | | | | | 0.50 | | || | | vsubpd %zmm18, %zmm4, %zmm29 #71.36
218 | 0.50 | | | | | 0.50 | | || 4.0 | | vsubpd %zmm19, %zmm0, %zmm27 #70.36
219 | 0.50 | | | | | 0.50 | | || | | vsubpd %zmm20, %zmm1, %zmm26 #69.36
220 | 0.50 | | | | | 0.50 | | || 4.0 | | vmulpd %zmm27, %zmm27, %zmm25 #72.49
221 | 0.50 | | | | | 0.50 | | || 4.0 | | vfmadd231pd %zmm26, %zmm26, %zmm25 #72.49
222 | 0.50 | | | | | 0.50 | | || 4.0 | | vfmadd231pd %zmm29, %zmm29, %zmm25 #72.63
223 | 2.50 | | | | | 0.50 | | || 8.0 | | vrcp14pd %zmm25, %zmm24 #75.38
224 | | | | | | 1.00 | | || | | vcmppd $1, %zmm14, %zmm25, %k2 #74.22
225 | | | | | | 1.00 | | || | | vfpclasspd $30, %zmm24, %k0 #75.38
226 | 1.00 | | | | | | | || | | kmovw %k2, %edx #74.22
227 | 1.00 | | | | | | | || | | knotw %k0, %k1 #75.38
228 | | | | | | | | || | | * vmovaps %zmm25, %zmm17 #75.38
229 | 0.00 | 0.34 | | | | 0.00 | 0.66 | || | | andl %edx, %r9d #74.22
230 | 0.50 | | 0.50 0.50 | 0.50 0.50 | | 0.50 | | || 4.0 | | vfnmadd213pd .L_2il0floatpacket.9(%rip){1to8}, %zmm24, %zmm17 #75.38
231 | 1.00 | | | | | | | || | | kmovw %r9d, %k3 #78.17
232 | 0.50 | | | | | 0.50 | | || 4.0 | | vmulpd %zmm17, %zmm17, %zmm18 #75.38
233 | 0.50 | | | | | 0.50 | | || | | vfmadd213pd %zmm24, %zmm17, %zmm24{%k1} #75.38
234 | 0.50 | | | | | 0.50 | | || 4.0 | | vfmadd213pd %zmm24, %zmm18, %zmm24{%k1} #75.38
235 | 0.50 | | | | | 0.50 | | || 4.0 | | vmulpd %zmm13, %zmm24, %zmm19 #76.38
236 | 0.50 | | | | | 0.50 | | || | | vmulpd %zmm10, %zmm24, %zmm21 #77.54
237 | 0.00 | | | | | 1.00 | | || 4.0 | | vmulpd %zmm19, %zmm24, %zmm22 #76.44
238 | 0.00 | | | | | 1.00 | | || 4.0 | | vmulpd %zmm22, %zmm24, %zmm20 #76.50
239 | 0.00 | | | | | 1.00 | | || | | vfmsub213pd %zmm5, %zmm22, %zmm24 #77.54
240 | 0.00 | | | | | 1.00 | | || 4.0 | | vmulpd %zmm21, %zmm20, %zmm23 #77.61
241 | 0.00 | | | | | 1.00 | | || 4.0 | | vmulpd %zmm24, %zmm23, %zmm28 #77.67
242 | 0.00 | | | | | 1.00 | | || | | vfmadd231pd %zmm26, %zmm28, %zmm9{%k3} #78.17
243 | 0.00 | | | | | 1.00 | | || | | vfmadd231pd %zmm27, %zmm28, %zmm8{%k3} #79.17
244 | 0.00 | | | | | 1.00 | | || 4.0 | 4.0 | vfmadd231pd %zmm29, %zmm28, %zmm11{%k3} #80.17
245 | 0.00 | 0.00 | | | | 0.00 | 1.00 | || | | cmpq %rbx, %r13 #67.9
246 | | | | | | | | || | | * jb ..B1.18 # Prob 82% #67.9
20.5 6.00 13.0 2.50 13.0 2.50 20.5 4.00 70.0 4
Loop-Carried Dependencies Analysis Report
-----------------------------------------
215 | 1.0 | addq $8, %r13 #67.9| [215]
216 | 1.0 | vpaddd %ymm16, %ymm3, %ymm3 #67.9| [216]
244 | 4.0 | vfmadd231pd %zmm29, %zmm28, %zmm11{%k3} #80.17| [244]
243 | 4.0 | vfmadd231pd %zmm27, %zmm28, %zmm8{%k3} #79.17| [243]
242 | 4.0 | vfmadd231pd %zmm26, %zmm28, %zmm9{%k3} #78.17| [242]

112
arch_analysis/avx512/osaca_force_aos_lt1200.txt
View File

@ -1,112 +0,0 @@
iwia021h@testfront1:~/MD-Bench/ICC$ /home/hpc/iwia/iwia021h/.local/bin/osaca --ignore-unknown --arch=CSX force.s
Open Source Architecture Code Analyzer (OSACA) - 0.3.14
Analyzed file: force.s
Architecture: CSX
Timestamp: 2021-04-26 22:33:06
P - Throughput of LOAD operation can be hidden behind a past or future STORE instruction
* - Instruction micro-ops not bound to a port
X - No throughput/latency information for this instruction in data file
Combined Analysis Report
------------------------
Port pressure in cycles
| 0 - 0DV | 1 | 2 - 2D | 3 - 3D | 4 | 5 | 6 | 7 || CP | LCD |
-------------------------------------------------------------------------------------------------
261 | | | | | | | | || | | ..B1.25: # Preds ..B1.24
262 | | | | | | | | || | | # Execution count [4.50e+00]
263 | 0.00 | 0.00 | | | | 0.00 | 1.00 | || 1.0 | 1.0 | movq %r8, %r13 #56.43
264 | | 1.00 | | | | | | || 3.0 | 3.0 | imulq %rcx, %r13 #56.43
265 | | | | | | 1.00 | | || | | vbroadcastsd %xmm6, %zmm2 #58.23
266 | | | | | | 1.00 | | || | | vbroadcastsd %xmm7, %zmm1 #59.23
267 | | | | | | 1.00 | | || | | vbroadcastsd %xmm12, %zmm0 #60.23
268 | 0.00 | 0.00 | | | | 0.00 | 1.00 | || | | movslq %r12d, %rbx #67.9
269 | 0.00 | 0.00 | | | | 0.00 | 1.00 | || 1.0 | 1.0 | addq %r10, %r13 #37.5
270 | | | 0.00 | 0.00 | 1.00 | | | 1.00 || | | movq %rax, -64(%rsp) #37.5[spill]
271 | | | 0.00 | 0.00 | 1.00 | | | 1.00 || | | movq %r8, -56(%rsp) #37.5[spill]
272 | | | 0.00 | 0.00 | 1.00 | | | 1.00 || | | movq %r10, -48(%rsp) #37.5[spill]
273 | | | 0.00 | 0.00 | 1.00 | | | 1.00 || | | movq %rsi, -40(%rsp) #37.5[spill]
274 | | | 0.00 | 0.00 | 1.00 | | | 1.00 || | | movq %rcx, -32(%rsp) #37.5[spill]
275 | | | 0.00 | 0.00 | 1.00 | | | 1.00 || | | movq %r9, -80(%rsp) #37.5[spill]
276 | | | 0.00 | 0.00 | 1.00 | | | 1.00 || | | movq %rdx, -72(%rsp) #37.5[spill]
277 | | | | | | | | || | | # LOE rbx rbp rdi r13 r11d r12d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm0 zmm1 zmm2 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
278 | | | | | | | | || | | ..B1.26: # Preds ..B1.30 ..B1.25
279 | | | | | | | | || | | # Execution count [2.50e+01]
280 | | | 0.50 0.50 | 0.50 0.50 | | | | || 4.0 | | vmovdqu (%r13,%rbx,4), %ymm3 #68.21
281 | 0.00 | 1.00 | | | | 0.00 | | || 1.0 | | vpaddd %ymm3, %ymm3, %ymm4 #69.36
282 | 0.00 | 1.00 | | | | 0.00 | | || 1.0 | | vpaddd %ymm4, %ymm3, %ymm3 #69.36
283 | | | 0.50 0.50 | 0.50 0.50 | | | | || | | movl (%r13,%rbx,4), %r10d #68.21
284 | | | 0.50 0.50 | 0.50 0.50 | | | | || | | movl 4(%r13,%rbx,4), %r9d #68.21
285 | | | 0.50 0.50 | 0.50 0.50 | | | | || | 4.0 | movl 8(%r13,%rbx,4), %r8d #68.21
286 | | | 0.50 0.50 | 0.50 0.50 | | | | || | | movl 12(%r13,%rbx,4), %esi #68.21
287 | | 1.00 | | | | 0.00 | | || | | lea (%r10,%r10,2), %r10d #69.36
288 | | | 0.50 0.50 | 0.50 0.50 | | | | || | | movl 16(%r13,%rbx,4), %ecx #68.21
289 | | 1.00 | | | | 0.00 | | || | | lea (%r9,%r9,2), %r9d #69.36
290 | | | 0.50 0.50 | 0.50 0.50 | | | | || | | movl 20(%r13,%rbx,4), %edx #68.21
291 | | 1.00 | | | | 0.00 | | || | 1.0 | lea (%r8,%r8,2), %r8d #69.36
292 | | | 0.50 0.50 | 0.50 0.50 | | | | || | | movl 24(%r13,%rbx,4), %eax #68.21
293 | | 1.00 | | | | 0.00 | | || | | lea (%rsi,%rsi,2), %esi #69.36
294 | | | 0.50 0.50 | 0.50 0.50 | | | | || | | movl 28(%r13,%rbx,4), %r15d #68.21
295 | | 1.00 | | | | 0.00 | | || | | lea (%rcx,%rcx,2), %ecx #69.36
296 | | 1.00 | | | | 0.00 | | || | | lea (%rdx,%rdx,2), %edx #69.36
297 | | 1.00 | | | | 0.00 | | || | | lea (%rax,%rax,2), %eax #69.36
298 | | 1.00 | | | | 0.00 | | || | | lea (%r15,%r15,2), %r15d #69.36
299 | | | | | | | | || | | # LOE rbx rbp rdi r13 eax edx ecx esi r8d r9d r10d r11d r12d r14d r15d xmm6 xmm7 xmm12 ymm3 ymm15 ymm16 zmm0 zmm1 zmm2 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
300 | | | | | | | | || | | ..B1.29: # Preds ..B1.26
301 | | | | | | | | || | | # Execution count [1.25e+01]
302 | | | | | | | | || | | X vpcmpeqb %xmm0, %xmm0, %k1 #69.36
303 | | | | | | | | || | | X vpcmpeqb %xmm0, %xmm0, %k2 #69.36
304 | | | | | | | | || | | X vpcmpeqb %xmm0, %xmm0, %k3 #69.36
305 | 0.50 | | | | | 0.50 | | || | | vpxord %zmm4, %zmm4, %zmm4 #69.36
306 | 0.50 | | | | | 0.50 | | || | | vpxord %zmm17, %zmm17, %zmm17 #69.36
307 | 0.50 | | | | | 0.50 | | || | | vpxord %zmm18, %zmm18, %zmm18 #69.36
308 | 1.50 | 0.17 | 4.00 0.50 | 4.00 0.50 | | 0.50 | 0.83 | || | | vgatherdpd 16(%rdi,%ymm3,8), %zmm4{%k1} #69.36
309 | 1.50 | 0.00 | 4.00 0.50 | 4.00 0.50 | | 0.50 | 1.00 | || 4.0 | | vgatherdpd 8(%rdi,%ymm3,8), %zmm17{%k2} #69.36
310 | 1.50 | 0.00 | 4.00 0.50 | 4.00 0.50 | | 0.50 | 1.00 | || | | vgatherdpd (%rdi,%ymm3,8), %zmm18{%k3} #69.36
311 | | | | | | | | || | | # LOE rbx rbp rdi r13 r11d r12d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm0 zmm1 zmm2 zmm4 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14 zmm17 zmm18
312 | | | | | | | | || | | ..B1.30: # Preds ..B1.29
313 | | | | | | | | || | | # Execution count [2.50e+01]
314 | 0.00 | 0.00 | | | | 0.00 | 1.00 | || | | addl $8, %r12d #67.9
315 | 0.00 | 0.00 | | | | 0.00 | 1.00 | || | | addq $8, %rbx #67.9
316 | 0.50 | | | | | 0.50 | | || | | vsubpd %zmm4, %zmm0, %zmm26 #71.36
317 | 0.50 | | | | | 0.50 | | || 4.0 | | vsubpd %zmm17, %zmm1, %zmm24 #70.36
318 | 0.50 | | | | | 0.50 | | || | | vsubpd %zmm18, %zmm2, %zmm23 #69.36
319 | 0.50 | | | | | 0.50 | | || 4.0 | | vmulpd %zmm24, %zmm24, %zmm3 #72.49
320 | 0.50 | | | | | 0.50 | | || 4.0 | | vfmadd231pd %zmm23, %zmm23, %zmm3 #72.49
321 | 0.50 | | | | | 0.50 | | || 4.0 | | vfmadd231pd %zmm26, %zmm26, %zmm3 #72.63
322 | 2.50 | | | | | 0.50 | | || 8.0 | | vrcp14pd %zmm3, %zmm22 #75.38
323 | | | | | | 1.00 | | || | | vcmppd $1, %zmm14, %zmm3, %k2 #74.22
324 | | | | | | 1.00 | | || | | vfpclasspd $30, %zmm22, %k0 #75.38
325 | 0.50 | | 0.50 0.50 | 0.50 0.50 | | 0.50 | | || 4.0 | | vfnmadd213pd .L_2il0floatpacket.9(%rip){1to8}, %zmm22, %zmm3 #75.38
326 | 1.00 | | | | | | | || | | knotw %k0, %k1 #75.38
327 | 0.50 | | | | | 0.50 | | || 4.0 | | vmulpd %zmm3, %zmm3, %zmm4 #75.38
328 | 0.50 | | | | | 0.50 | | || | | vfmadd213pd %zmm22, %zmm3, %zmm22{%k1} #75.38
329 | 0.50 | | | | | 0.50 | | || 4.0 | | vfmadd213pd %zmm22, %zmm4, %zmm22{%k1} #75.38
330 | 0.50 | | | | | 0.50 | | || 4.0 | | vmulpd %zmm13, %zmm22, %zmm17 #76.38
331 | 0.50 | | | | | 0.50 | | || | | vmulpd %zmm10, %zmm22, %zmm19 #77.54
332 | 0.50 | | | | | 0.50 | | || 4.0 | | vmulpd %zmm17, %zmm22, %zmm20 #76.44
333 | 0.50 | | | | | 0.50 | | || 4.0 | | vmulpd %zmm20, %zmm22, %zmm18 #76.50
334 | 0.50 | | | | | 0.50 | | || | | vfmsub213pd %zmm5, %zmm20, %zmm22 #77.54
335 | 0.50 | | | | | 0.50 | | || 4.0 | | vmulpd %zmm19, %zmm18, %zmm21 #77.61
336 | 0.50 | | | | | 0.50 | | || 4.0 | | vmulpd %zmm22, %zmm21, %zmm25 #77.67
337 | 1.00 | | | | | 0.00 | | || | | vfmadd231pd %zmm23, %zmm25, %zmm9{%k2} #78.17
338 | 1.00 | | | | | 0.00 | | || | | vfmadd231pd %zmm24, %zmm25, %zmm8{%k2} #79.17
339 | 1.00 | | | | | 0.00 | | || 4.0 | | vfmadd231pd %zmm26, %zmm25, %zmm11{%k2} #80.17
340 | 0.00 | 0.00 | | | | 0.00 | 1.00 | || | | cmpl %r14d, %r12d #67.9
341 | | | | | | | | || | | * jb ..B1.26 # Prob 82% #67.9
342 | | | | | | | | || | | # LOE rbx rbp rdi r13 r11d r12d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm0 zmm1 zmm2 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
21.0 11.2 17.0 6.50 17.0 6.50 7.00 17.0 8.83 7.00 75.0 10.0
Loop-Carried Dependencies Analysis Report
-----------------------------------------
287 | 6.0 | lea (%r10,%r10,2), %r10d #69.36| [269, 283, 287]
291 | 10.0 | lea (%r8,%r8,2), %r8d #69.36| [263, 264, 269, 285, 291]
295 | 9.0 | lea (%rcx,%rcx,2), %ecx #69.36| [264, 269, 288, 295]
314 | 1.0 | addl $8, %r12d #67.9| [314]
339 | 4.0 | vfmadd231pd %zmm26, %zmm25, %zmm11{%k2} #80.17| [339]
338 | 4.0 | vfmadd231pd %zmm24, %zmm25, %zmm8{%k2} #79.17| [338]
337 | 4.0 | vfmadd231pd %zmm23, %zmm25, %zmm9{%k2} #78.17| [337]

91
arch_analysis/avx512/osaca_force_aos_lt8.txt
View File

@ -1,91 +0,0 @@
iwia021h@testfront1:~/MD-Bench/asm$ /home/hpc/iwia/iwia021h/.local/bin/osaca --ignore-unknown --arch=CSX force_aos_lt8_markers.s
Open Source Architecture Code Analyzer (OSACA) - 0.3.14
Analyzed file: force_aos_lt8_markers.s
Architecture: CSX
Timestamp: 2021-04-29 15:49:27
P - Throughput of LOAD operation can be hidden behind a past or future STORE instruction
* - Instruction micro-ops not bound to a port
X - No throughput/latency information for this instruction in data file
Combined Analysis Report
------------------------
Port pressure in cycles
| 0 - 0DV | 1 | 2 - 2D | 3 - 3D | 4 | 5 | 6 | 7 || CP | LCD |
-------------------------------------------------------------------------------------------------
358 | | | | | | | | || | | # LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
359 | | | | | | | | || | | ..B1.33: # Preds ..B1.32
360 | | | | | | | | || | | # Execution count [2.50e+01]
361 | | 1.00 | | | | | | || 3.0 | | imulq %r8, %rcx #56.43
362 | | | | | | 1.00 | | || | 3.0 | vbroadcastsd %xmm6, %zmm4 #58.23
363 | | | | | | | | || | | X subl %r14d, %r11d #67.9
364 | 0.00 | 0.00 | | | | 0.00 | 1.00 | || 1.0 | | addq %r10, %rcx #37.5
365 | | | | | | | | || | | X vpbroadcastd %r11d, %ymm0 #67.9
366 | | | | | | 1.00 | | || | | vpcmpgtd %ymm15, %ymm0, %k3 #67.9
367 | 0.00 | 0.00 | | | | 0.00 | 1.00 | || | | movslq %r14d, %r14 #67.9
368 | 1.00 | | | | | | | || | | kmovw %k3, %ebx #67.9
369 | 0.00 | 1.00 | 0.50 0.50 | 0.50 0.50 | | 0.00 | | || 0.0 | | vmovdqu32 (%rcx,%r14,4), %ymm1{%k3}{z} #68.21
370 | 0.00 | 1.00 | | | | 0.00 | | || 1.0 | | vpaddd %ymm1, %ymm1, %ymm2 #69.36
371 | 0.00 | 1.00 | | | | 0.00 | | || 1.0 | | vpaddd %ymm2, %ymm1, %ymm0 #69.36
372 | | | | | | | | || | | # LOE rax rdx rbp rsi rdi r8 r9 r10 ebx xmm7 xmm12 ymm0 ymm15 ymm16 zmm4 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14 k3
373 | | | | | | | | || | | ..B1.36: # Preds ..B1.33
374 | | | | | | | | || | | # Execution count [1.25e+01]
375 | 1.00 | | | | | | | || | | kmovw %k3, %k1 #69.36
376 | 1.00 | | | | | | | || | | kmovw %k3, %k2 #69.36
377 | 0.50 | | | | | 0.50 | | || | | vpxord %zmm1, %zmm1, %zmm1 #69.36
378 | 0.50 | | | | | 0.50 | | || | | vpxord %zmm2, %zmm2, %zmm2 #69.36
379 | 0.50 | | | | | 0.50 | | || | | vpxord %zmm3, %zmm3, %zmm3 #69.36
380 | 1.50 | 0.50 | 4.00 0.50 | 4.00 0.50 | | 0.50 | 0.50 | || | | vgatherdpd 16(%rdi,%ymm0,8), %zmm1{%k1} #69.36
381 | 1.50 | 0.50 | 4.00 0.50 | 4.00 0.50 | | 0.50 | 0.50 | || 4.0 | | vgatherdpd 8(%rdi,%ymm0,8), %zmm2{%k2} #69.36
382 | 1.50 | 0.00 | 4.00 0.50 | 4.00 0.50 | | 0.50 | 1.00 | || | | vgatherdpd (%rdi,%ymm0,8), %zmm3{%k3} #69.36
383 | | | | | | | | || | | # LOE rax rdx rbp rsi rdi r8 r9 r10 ebx xmm7 xmm12 ymm15 ymm16 zmm1 zmm2 zmm3 zmm4 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
384 | | | | | | | | || | | ..B1.37: # Preds ..B1.36
385 | | | | | | | | || | | # Execution count [2.50e+01]
386 | | | | | | 1.00 | | || | | vbroadcastsd %xmm7, %zmm7 #59.23
387 | | | | | | 1.00 | | || | | vbroadcastsd %xmm12, %zmm12 #60.23
388 | 0.50 | | | | | 0.50 | | || | | vsubpd %zmm1, %zmm12, %zmm23 #71.36
389 | 0.50 | | | | | 0.50 | | || 4.0 | | vsubpd %zmm2, %zmm7, %zmm21 #70.36
390 | 0.50 | | | | | 0.50 | | || | 4.0 | vsubpd %zmm3, %zmm4, %zmm20 #69.36
391 | 0.50 | | | | | 0.50 | | || 4.0 | | vmulpd %zmm21, %zmm21, %zmm19 #72.49
392 | 0.50 | | | | | 0.50 | | || 4.0 | 4.0 | vfmadd231pd %zmm20, %zmm20, %zmm19 #72.49
393 | 0.50 | | | | | 0.50 | | || 4.0 | 4.0 | vfmadd231pd %zmm23, %zmm23, %zmm19 #72.63
394 | 2.50 | | | | | 0.50 | | || 8.0 | 8.0 | vrcp14pd %zmm19, %zmm18 #75.38
395 | | | | | | 1.00 | | || | | vcmppd $1, %zmm14, %zmm19, %k2 #74.22
396 | | | | | | 1.00 | | || | | vfpclasspd $30, %zmm18, %k0 #75.38
397 | 1.00 | | | | | | | || | | kmovw %k2, %ecx #74.22
398 | 1.00 | | | | | | | || | | knotw %k0, %k1 #75.38
399 | | | | | | | | || | | * vmovaps %zmm19, %zmm0 #75.38
400 | 0.00 | 0.00 | | | | 0.00 | 1.00 | || | | andl %ecx, %ebx #74.22
401 | 0.50 | | 0.50 0.50 | 0.50 0.50 | | 0.50 | | || 4.0 | | vfnmadd213pd .L_2il0floatpacket.9(%rip){1to8}, %zmm18, %zmm0 #75.38
402 | 1.00 | | | | | | | || | | kmovw %ebx, %k3 #78.17
403 | 0.50 | | | | | 0.50 | | || 4.0 | | vmulpd %zmm0, %zmm0, %zmm1 #75.38
404 | 0.50 | | | | | 0.50 | | || | 4.0 | vfmadd213pd %zmm18, %zmm0, %zmm18{%k1} #75.38
405 | 0.50 | | | | | 0.50 | | || 4.0 | 4.0 | vfmadd213pd %zmm18, %zmm1, %zmm18{%k1} #75.38
406 | 0.50 | | | | | 0.50 | | || 4.0 | | vmulpd %zmm13, %zmm18, %zmm2 #76.38
407 | 0.50 | | | | | 0.50 | | || | | vmulpd %zmm10, %zmm18, %zmm4 #77.54
408 | 0.50 | | | | | 0.50 | | || 4.0 | 4.0 | vmulpd %zmm2, %zmm18, %zmm6 #76.44
409 | 0.50 | | | | | 0.50 | | || 4.0 | | vmulpd %zmm6, %zmm18, %zmm3 #76.50
410 | 0.50 | | | | | 0.50 | | || | | vfmsub213pd %zmm5, %zmm6, %zmm18 #77.54
411 | 0.00 | | | | | 1.00 | | || 4.0 | | vmulpd %zmm4, %zmm3, %zmm17 #77.61
412 | 0.00 | | | | | 1.00 | | || 4.0 | | vmulpd %zmm18, %zmm17, %zmm22 #77.67
413 | 0.00 | | | | | 1.00 | | || | | vfmadd231pd %zmm20, %zmm22, %zmm9{%k3} #78.17
414 | 0.00 | | | | | 1.00 | | || | | vfmadd231pd %zmm21, %zmm22, %zmm8{%k3} #79.17
415 | 0.00 | | | | | 1.00 | | || 4.0 | | vfmadd231pd %zmm23, %zmm22, %zmm11{%k3} #80.17
22.0 5.00 13.0 2.50 13.0 2.50 22.0 5.00 70.0 35.0
Loop-Carried Dependencies Analysis Report
-----------------------------------------
363 | 0.0 | subl %r14d, %r11d #67.9| [363]
367 | 1.0 | movslq %r14d, %r14 #67.9| [367]
386 | 3.0 | vbroadcastsd %xmm7, %zmm7 #59.23| [386]
387 | 3.0 | vbroadcastsd %xmm12, %zmm12 #60.23| [387]
415 | 4.0 | vfmadd231pd %zmm23, %zmm22, %zmm11{%k3} #80.17| [415]
414 | 4.0 | vfmadd231pd %zmm21, %zmm22, %zmm8{%k3} #79.17| [414]
413 | 4.0 | vfmadd231pd %zmm20, %zmm22, %zmm9{%k3} #78.17| [413]
397 | 28.0 | kmovw %k2, %ecx #74.22| [361, 364, 369, 371, 382, 390, 392, 393, 395, 397]
408 | 35.0 | vmulpd %zmm2, %zmm18, %zmm6 #76.44| [362, 390, 392, 393, 394, 404, 405, 408]

71
arch_analysis/avx512/osaca_force_soa_geq1200.txt
View File

@ -1,71 +0,0 @@
iwia021h@testfront1:~/MD-Bench/asm$ /home/hpc/iwia/iwia021h/.local/bin/osaca --ignore-unknown --arch=CSX force_soa_geq1200_markers.s
Open Source Architecture Code Analyzer (OSACA) - 0.3.14
Analyzed file: force_soa_geq1200_markers.s
Architecture: CSX
Timestamp: 2021-04-29 15:54:23
P - Throughput of LOAD operation can be hidden behind a past or future STORE instruction
* - Instruction micro-ops not bound to a port
X - No throughput/latency information for this instruction in data file
Combined Analysis Report
------------------------
Port pressure in cycles
| 0 - 0DV | 1 | 2 - 2D | 3 - 3D | 4 | 5 | 6 | 7 || CP | LCD |
-------------------------------------------------------------------------------------------------
189 | | | | | | | | || | | # LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 r14 r15 ebx r9d r13d xmm8 xmm9 xmm10 ymm3 ymm4 ymm17 ymm18 zmm2 zmm5 zmm6 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
190 | | | | | | | | || | | ..B1.18: # Preds ..B1.18 ..B1.17
191 | | | | | | | | || | | # Execution count [2.50e+01]
192 | | | | | | 1.00 | | || | | vpcmpgtd %ymm4, %ymm3, %k5 #67.9
193 | 0.00 | 1.00 | | | | 0.00 | | || | | vpaddd %ymm18, %ymm4, %ymm4 #67.9
194 | 0.00 | 1.00 | 0.50 0.50 | 0.50 0.50 | | 0.00 | | || 4.0 | | vmovdqu32 (%rcx,%r15,4), %ymm20{%k5}{z} #68.21
195 | | | | | | | | || | | * vmovaps %zmm19, %zmm22 #70.36
196 | 0.00 | 0.50 | | | | 0.00 | 0.50 | || | | addq $8, %r15 #67.9
197 | 1.00 | | | | | | | || | | kmovw %k5, %k2 #70.36
198 | | | | | | | | || | | * vmovaps %zmm19, %zmm21 #69.36
199 | 1.00 | | | | | | | || | | kmovw %k5, %k1 #69.36
200 | | | | | | | | || | | * vmovaps %zmm19, %zmm23 #71.36
201 | 1.00 | | | | | | | || | | kmovw %k5, %k3 #71.36
202 | 1.50 | 0.50 | 4.00 0.50 | 4.00 0.50 | | 0.50 | 0.50 | || | | vgatherdpd (%rsi,%ymm20,8), %zmm23{%k3} #71.36
203 | 1.50 | 0.50 | 4.00 0.50 | 4.00 0.50 | | 0.50 | 0.50 | || 4.0 | | vgatherdpd (%rax,%ymm20,8), %zmm22{%k2} #70.36
204 | 1.50 | 0.50 | 4.00 0.50 | 4.00 0.50 | | 0.50 | 0.50 | || | | vgatherdpd (%rdx,%ymm20,8), %zmm21{%k1} #69.36
205 | 0.50 | | | | | 0.50 | | || 4.0 | | vsubpd %zmm22, %zmm5, %zmm0 #70.36
206 | 0.50 | | | | | 0.50 | | || | | vsubpd %zmm21, %zmm2, %zmm1 #69.36
207 | 0.50 | | | | | 0.50 | | || | | vsubpd %zmm23, %zmm6, %zmm21 #71.36
208 | 0.50 | | | | | 0.50 | | || 4.0 | | vmulpd %zmm0, %zmm0, %zmm20 #72.49
209 | 0.50 | | | | | 0.50 | | || 4.0 | | vfmadd231pd %zmm1, %zmm1, %zmm20 #72.49
210 | 0.50 | | | | | 0.50 | | || 4.0 | | vfmadd231pd %zmm21, %zmm21, %zmm20 #72.63
211 | 2.50 | | | | | 0.50 | | || 8.0 | | vrcp14pd %zmm20, %zmm31 #75.38
212 | | | | | | 1.00 | | || | | vcmppd $1, %zmm16, %zmm20, %k6{%k5} #74.22
213 | | | | | | 1.00 | | || | | vfpclasspd $30, %zmm31, %k0 #75.38
214 | | | | | | | | || | | * vmovaps %zmm20, %zmm24 #75.38
215 | 0.50 | | 0.50 0.50 | 0.50 0.50 | | 0.50 | | || 4.0 | | vfnmadd213pd .L_2il0floatpacket.5(%rip){1to8}, %zmm31, %zmm24 #75.38
216 | 1.00 | | | | | | | || | | knotw %k0, %k4 #75.38
217 | 0.50 | | | | | 0.50 | | || 4.0 | | vmulpd %zmm24, %zmm24, %zmm25 #75.38
218 | 0.50 | | | | | 0.50 | | || | | vfmadd213pd %zmm31, %zmm24, %zmm31{%k4} #75.38
219 | 0.50 | | | | | 0.50 | | || 4.0 | | vfmadd213pd %zmm31, %zmm25, %zmm31{%k4} #75.38
220 | 0.50 | | | | | 0.50 | | || 4.0 | | vmulpd %zmm15, %zmm31, %zmm26 #76.38
221 | 0.50 | | | | | 0.50 | | || | | vmulpd %zmm14, %zmm31, %zmm28 #77.54
222 | 0.50 | | | | | 0.50 | | || 4.0 | | vmulpd %zmm26, %zmm31, %zmm29 #76.44
223 | 0.50 | | | | | 0.50 | | || 4.0 | | vmulpd %zmm29, %zmm31, %zmm27 #76.50
224 | 0.00 | | | | | 1.00 | | || | | vfmsub213pd %zmm7, %zmm29, %zmm31 #77.54
225 | 0.00 | | | | | 1.00 | | || 4.0 | | vmulpd %zmm28, %zmm27, %zmm30 #77.61
226 | 0.00 | | | | | 1.00 | | || 4.0 | | vmulpd %zmm31, %zmm30, %zmm24 #77.67
227 | 0.00 | | | | | 1.00 | | || | | vfmadd231pd %zmm1, %zmm24, %zmm13{%k6} #78.17
228 | 0.00 | | | | | 1.00 | | || | 4.0 | vfmadd231pd %zmm0, %zmm24, %zmm12{%k6} #79.17
229 | 0.00 | | | | | 1.00 | | || 4.0 | | vfmadd231pd %zmm21, %zmm24, %zmm11{%k6} #80.17
230 | 0.00 | 0.17 | | | | 0.00 | 0.83 | || | | cmpq %r14, %r15 #67.9
231 | | | | | | | | || | | * jb ..B1.18 # Prob 82% #67.9
18.0 4.17 13.0 2.50 13.0 2.50 18.0 2.83 68.0 4
Loop-Carried Dependencies Analysis Report
-----------------------------------------
193 | 1.0 | vpaddd %ymm18, %ymm4, %ymm4 #67.9| [193]
196 | 1.0 | addq $8, %r15 #67.9| [196]
228 | 4.0 | vfmadd231pd %zmm0, %zmm24, %zmm12{%k6} #79.17| [228]
227 | 4.0 | vfmadd231pd %zmm1, %zmm24, %zmm13{%k6} #78.17| [227]
229 | 4.0 | vfmadd231pd %zmm21, %zmm24, %zmm11{%k6} #80.17| [229]

69
arch_analysis/avx512/osaca_force_soa_lt1200.txt
View File

@ -1,69 +0,0 @@
iwia021h@testfront1:~/MD-Bench/asm$ /home/hpc/iwia/iwia021h/.local/bin/osaca --ignore-unknown --arch=CSX force_soa_lt1200_markers.s
Open Source Architecture Code Analyzer (OSACA) - 0.3.14
Analyzed file: force_soa_lt1200_markers.s
Architecture: CSX
Timestamp: 2021-04-29 15:39:58
P - Throughput of LOAD operation can be hidden behind a past or future STORE instruction
* - Instruction micro-ops not bound to a port
X - No throughput/latency information for this instruction in data file
Combined Analysis Report
------------------------
Port pressure in cycles
| 0 - 0DV | 1 | 2 - 2D | 3 - 3D | 4 | 5 | 6 | 7 || CP | LCD |
-------------------------------------------------------------------------------------------------
253 | | | | | | | | || | | # LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 r14 ebx r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm0 zmm1 zmm2 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
254 | | | | | | | | || | | ..B1.22: # Preds ..B1.22 ..B1.21
255 | | | | | | | | || | | # Execution count [2.50e+01]
256 | | | | | | | | || | | X vpcmpeqb %xmm0, %xmm0, %k2 #70.36
257 | 0.00 | 0.50 | | | | 0.00 | 0.50 | || | | addl $8, %r9d #67.9
258 | | | | | | | | || | | X vpcmpeqb %xmm0, %xmm0, %k1 #69.36
259 | | | | | | | | || | | X vpcmpeqb %xmm0, %xmm0, %k3 #71.36
260 | | | 0.50 0.50 | 0.50 0.50 | | | | || 4.0 | | vmovdqu (%rcx,%r14,4), %ymm3 #68.21
261 | 0.00 | 0.50 | | | | 0.00 | 0.50 | || | | addq $8, %r14 #67.9
262 | 0.50 | | | | | 0.50 | | || | | vpxord %zmm5, %zmm5, %zmm5 #70.36
263 | 0.50 | | | | | 0.50 | | || | | vpxord %zmm4, %zmm4, %zmm4 #69.36
264 | 0.50 | | | | | 0.50 | | || | | vpxord %zmm6, %zmm6, %zmm6 #71.36
265 | 1.50 | 0.50 | 4.00 0.50 | 4.00 0.50 | | 0.50 | 0.50 | || 4.0 | | vgatherdpd (%rax,%ymm3,8), %zmm5{%k2} #70.36
266 | 1.50 | 0.50 | 4.00 0.50 | 4.00 0.50 | | 0.50 | 0.50 | || | | vgatherdpd (%rdx,%ymm3,8), %zmm4{%k1} #69.36
267 | 1.50 | 0.50 | 4.00 0.50 | 4.00 0.50 | | 0.50 | 0.50 | || | | vgatherdpd (%rsi,%ymm3,8), %zmm6{%k3} #71.36
268 | 0.50 | | | | | 0.50 | | || 4.0 | | vsubpd %zmm5, %zmm1, %zmm29 #70.36
269 | 0.50 | | | | | 0.50 | | || | | vsubpd %zmm4, %zmm0, %zmm28 #69.36
270 | 0.50 | | | | | 0.50 | | || | | vsubpd %zmm6, %zmm2, %zmm31 #71.36
271 | 0.50 | | | | | 0.50 | | || 4.0 | | vmulpd %zmm29, %zmm29, %zmm20 #72.49
272 | 0.50 | | | | | 0.50 | | || 4.0 | | vfmadd231pd %zmm28, %zmm28, %zmm20 #72.49
273 | 0.50 | | | | | 0.50 | | || 4.0 | | vfmadd231pd %zmm31, %zmm31, %zmm20 #72.63
274 | 2.50 | | | | | 0.50 | | || 8.0 | | vrcp14pd %zmm20, %zmm27 #75.38
275 | | | | | | 1.00 | | || | | vcmppd $1, %zmm16, %zmm20, %k5 #74.22
276 | | | | | | 1.00 | | || | | vfpclasspd $30, %zmm27, %k0 #75.38
277 | 0.50 | | 0.50 0.50 | 0.50 0.50 | | 0.50 | | || 4.0 | | vfnmadd213pd .L_2il0floatpacket.5(%rip){1to8}, %zmm27, %zmm20 #75.38
278 | 1.00 | | | | | | | || | | knotw %k0, %k4 #75.38
279 | 0.50 | | | | | 0.50 | | || 4.0 | | vmulpd %zmm20, %zmm20, %zmm21 #75.38
280 | 0.50 | | | | | 0.50 | | || | | vfmadd213pd %zmm27, %zmm20, %zmm27{%k4} #75.38
281 | 0.50 | | | | | 0.50 | | || 4.0 | | vfmadd213pd %zmm27, %zmm21, %zmm27{%k4} #75.38
282 | 0.50 | | | | | 0.50 | | || 4.0 | | vmulpd %zmm15, %zmm27, %zmm22 #76.38
283 | 0.50 | | | | | 0.50 | | || | | vmulpd %zmm14, %zmm27, %zmm24 #77.54
284 | 0.50 | | | | | 0.50 | | || 4.0 | | vmulpd %zmm22, %zmm27, %zmm25 #76.44
285 | 0.50 | | | | | 0.50 | | || 4.0 | | vmulpd %zmm25, %zmm27, %zmm23 #76.50
286 | 0.50 | | | | | 0.50 | | || | | vfmsub213pd %zmm7, %zmm25, %zmm27 #77.54
287 | 0.50 | | | | | 0.50 | | || 4.0 | | vmulpd %zmm24, %zmm23, %zmm26 #77.61
288 | 0.00 | | | | | 1.00 | | || 4.0 | | vmulpd %zmm27, %zmm26, %zmm30 #77.67
289 | 0.00 | | | | | 1.00 | | || | | vfmadd231pd %zmm28, %zmm30, %zmm13{%k5} #78.17
290 | 0.00 | | | | | 1.00 | | || | 4.0 | vfmadd231pd %zmm29, %zmm30, %zmm12{%k5} #79.17
291 | 0.00 | | | | | 1.00 | | || 4.0 | | vfmadd231pd %zmm31, %zmm30, %zmm11{%k5} #80.17
292 | 0.00 | 0.50 | | | | 0.00 | 0.50 | || | | cmpl %ebx, %r9d #67.9
293 | | | | | | | | || | | * jb ..B1.22 # Prob 82% #67.9
17.5 3.00 13.0 2.50 13.0 2.50 17.5 3.00 68.0 4
Loop-Carried Dependencies Analysis Report
-----------------------------------------
257 | 1.0 | addl $8, %r9d #67.9| [257]
261 | 1.0 | addq $8, %r14 #67.9| [261]
290 | 4.0 | vfmadd231pd %zmm29, %zmm30, %zmm12{%k5} #79.17| [290]
289 | 4.0 | vfmadd231pd %zmm28, %zmm30, %zmm13{%k5} #78.17| [289]
291 | 4.0 | vfmadd231pd %zmm31, %zmm30, %zmm11{%k5} #80.17| [291]

79
arch_analysis/avx512/osaca_force_soa_lt8.txt
View File

@ -1,79 +0,0 @@
iwia021h@testfront1:~/MD-Bench/asm$ /home/hpc/iwia/iwia021h/.local/bin/osaca --ignore-unknown --arch=CSX force_soa_lt8_markers.s
Open Source Architecture Code Analyzer (OSACA) - 0.3.14
Analyzed file: force_soa_lt8_markers.s
Architecture: CSX
Timestamp: 2021-04-29 15:52:48
P - Throughput of LOAD operation can be hidden behind a past or future STORE instruction
* - Instruction micro-ops not bound to a port
X - No throughput/latency information for this instruction in data file
Combined Analysis Report
------------------------
Port pressure in cycles
| 0 - 0DV | 1 | 2 - 2D | 3 - 3D | 4 | 5 | 6 | 7 || CP | LCD |
-------------------------------------------------------------------------------------------------
300 | | | | | | | | || | | # LOE rax rdx rbp rsi rdi r8 r10 r11 r12 ebx r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
301 | | | | | | | | || | | ..B1.25: # Preds ..B1.24
302 | | | | | | | | || | | # Execution count [2.50e+01]
303 | | 1.00 | | | | | | || 3.0 | | imulq %r12, %r8 #56.43
304 | | | | | | 1.00 | | || | | vbroadcastsd %xmm9, %zmm9 #59.23
305 | | | | | | 1.00 | | || | 3.0 | vbroadcastsd %xmm8, %zmm2 #58.23
306 | | | | | | 1.00 | | || | | vbroadcastsd %xmm10, %zmm10 #60.23
307 | | | | | | | | || | | X subl %ebx, %r13d #67.9
308 | 0.00 | 0.50 | | | | 0.00 | 0.50 | || 1.0 | | addq %r11, %r8 #37.5
309 | | | | | | | | || | | X vpbroadcastd %r13d, %ymm0 #67.9
310 | | | | | | 1.00 | | || | | vpcmpgtd %ymm17, %ymm0, %k5 #67.9
311 | 0.00 | 0.16 | | | | 0.00 | 0.84 | || | | movslq %ebx, %rbx #67.9
312 | | | | | | | | || | | * vmovaps %zmm19, %zmm4 #70.36
313 | 1.00 | | | | | | | || | | kmovw %k5, %k2 #70.36
314 | | | | | | | | || | | * vmovaps %zmm19, %zmm3 #69.36
315 | 0.00 | 1.00 | 0.50 0.50 | 0.50 0.50 | | 0.00 | | || 0.0 | | vmovdqu32 (%r8,%rbx,4), %ymm1{%k5}{z} #68.21
316 | 1.00 | | | | | | | || | | kmovw %k5, %k1 #69.36
317 | | | | | | | | || | | * vmovaps %zmm19, %zmm5 #71.36
318 | 1.00 | | | | | | | || | | kmovw %k5, %k3 #71.36
319 | 1.50 | 0.50 | 4.00 0.50 | 4.00 0.50 | | 0.50 | 0.50 | || | | vgatherdpd (%rsi,%ymm1,8), %zmm5{%k3} #71.36
320 | 1.50 | 0.34 | 4.00 0.50 | 4.00 0.50 | | 0.50 | 0.66 | || 4.0 | | vgatherdpd (%rax,%ymm1,8), %zmm4{%k2} #70.36
321 | 1.50 | 0.00 | 4.00 0.50 | 4.00 0.50 | | 0.50 | 1.00 | || | | vgatherdpd (%rdx,%ymm1,8), %zmm3{%k1} #69.36
322 | 0.50 | | | | | 0.50 | | || | | vsubpd %zmm5, %zmm10, %zmm30 #71.36
323 | 0.50 | | | | | 0.50 | | || 4.0 | | vsubpd %zmm4, %zmm9, %zmm28 #70.36
324 | 0.50 | | | | | 0.50 | | || | 4.0 | vsubpd %zmm3, %zmm2, %zmm27 #69.36
325 | 0.50 | | | | | 0.50 | | || 4.0 | | vmulpd %zmm28, %zmm28, %zmm26 #72.49
326 | 0.50 | | | | | 0.50 | | || 4.0 | 4.0 | vfmadd231pd %zmm27, %zmm27, %zmm26 #72.49
327 | 0.50 | | | | | 0.50 | | || 4.0 | 4.0 | vfmadd231pd %zmm30, %zmm30, %zmm26 #72.63
328 | 2.50 | | | | | 0.50 | | || 8.0 | 8.0 | vrcp14pd %zmm26, %zmm25 #75.38
329 | | | | | | 1.00 | | || | | vcmppd $1, %zmm16, %zmm26, %k6{%k5} #74.22
330 | | | | | | 1.00 | | || | | vfpclasspd $30, %zmm25, %k0 #75.38
331 | | | | | | | | || | | * vmovaps %zmm26, %zmm6 #75.38
332 | 0.50 | | 0.50 0.50 | 0.50 0.50 | | 0.50 | | || 4.0 | 4.0 | vfnmadd213pd .L_2il0floatpacket.5(%rip){1to8}, %zmm25, %zmm6 #75.38
333 | 1.00 | | | | | | | || | | knotw %k0, %k4 #75.38
334 | 0.50 | | | | | 0.50 | | || 4.0 | 4.0 | vmulpd %zmm6, %zmm6, %zmm8 #75.38
335 | 0.50 | | | | | 0.50 | | || | | vfmadd213pd %zmm25, %zmm6, %zmm25{%k4} #75.38
336 | 0.50 | | | | | 0.50 | | || 4.0 | | vfmadd213pd %zmm25, %zmm8, %zmm25{%k4} #75.38
337 | 0.50 | | | | | 0.50 | | || 4.0 | | vmulpd %zmm15, %zmm25, %zmm20 #76.38
338 | 0.50 | | | | | 0.50 | | || | | vmulpd %zmm14, %zmm25, %zmm22 #77.54
339 | 0.50 | | | | | 0.50 | | || 4.0 | | vmulpd %zmm20, %zmm25, %zmm23 #76.44
340 | 0.50 | | | | | 0.50 | | || 4.0 | | vmulpd %zmm23, %zmm25, %zmm21 #76.50
341 | 0.50 | | | | | 0.50 | | || | | vfmsub213pd %zmm7, %zmm23, %zmm25 #77.54
342 | 0.50 | | | | | 0.50 | | || 4.0 | | vmulpd %zmm22, %zmm21, %zmm24 #77.61
343 | 0.50 | | | | | 0.50 | | || 4.0 | | vmulpd %zmm25, %zmm24, %zmm29 #77.67
344 | 0.00 | | | | | 1.00 | | || | | vfmadd231pd %zmm27, %zmm29, %zmm13{%k6} #78.17
345 | 0.00 | | | | | 1.00 | | || | | vfmadd231pd %zmm28, %zmm29, %zmm12{%k6} #79.17
346 | 0.00 | | | | | 1.00 | | || 4.0 | | vfmadd231pd %zmm30, %zmm29, %zmm11{%k6} #80.17
19.5 3.50 13.0 2.50 13.0 2.50 19.5 3.50 68.0 31.0
Loop-Carried Dependencies Analysis Report
-----------------------------------------
308 | 4.0 | addq %r11, %r8 #37.5| [303, 308]
304 | 3.0 | vbroadcastsd %xmm9, %zmm9 #59.23| [304]
306 | 3.0 | vbroadcastsd %xmm10, %zmm10 #60.23| [306]
307 | 0.0 | subl %ebx, %r13d #67.9| [307]
311 | 1.0 | movslq %ebx, %rbx #67.9| [311]
346 | 4.0 | vfmadd231pd %zmm30, %zmm29, %zmm11{%k6} #80.17| [346]
345 | 4.0 | vfmadd231pd %zmm28, %zmm29, %zmm12{%k6} #79.17| [345]
344 | 4.0 | vfmadd231pd %zmm27, %zmm29, %zmm13{%k6} #78.17| [344]
334 | 31.0 | vmulpd %zmm6, %zmm6, %zmm8 #75.38| [305, 324, 326, 327, 328, 332, 334]

70
arch_analysis/avx512/osaca_force_soa_lt8_iln1000.txt
View File

@ -1,70 +0,0 @@
iwia021h@testfront1:~/MD-Bench$ /home/hpc/iwia/iwia021h/.local/bin/osaca --ignore-unknown --arch=CSX ICC/force.s
Open Source Architecture Code Analyzer (OSACA) - 0.3.14
Analyzed file: ICC/force.s
Architecture: CSX
Timestamp: 2021-04-30 16:08:44
P - Throughput of LOAD operation can be hidden behind a past or future STORE instruction
* - Instruction micro-ops not bound to a port
X - No throughput/latency information for this instruction in data file
Combined Analysis Report
------------------------
Port pressure in cycles
| 0 - 0DV | 1 | 2 - 2D | 3 - 3D | 4 | 5 | 6 | 7 || CP | LCD |
-------------------------------------------------------------------------------------------------
306 | | | | | | | | || | | # LOE rbp rdi r8 r9 r10 edx ecx r11d r12d r13d r14d r15d ymm13 ymm14 zmm2 zmm3 zmm4 zmm5 zmm6 zmm7 zmm8 zmm9 zmm10 zmm11 zmm12 zmm15
307 | | | | | | | | || | | ..B1.29: # Preds ..B1.28
308 | | | | | | | | || | | # Execution count [2.50e+04]
309 | 0.00 | 0.50 | | | | 0.00 | 0.50 | || | | movl %r14d, %eax #64.13
310 | | | | | | | | || | | X subl %ecx, %eax #64.13
311 | | | | | | | | || | | X vpbroadcastd %eax, %ymm0 #64.13
312 | | | | | | 1.00 | | || | | vpcmpgtd %ymm14, %ymm0, %k5 #64.13
313 | 0.00 | 0.16 | | | | 0.00 | 0.84 | || | | movslq %ecx, %rcx #64.13
314 | | | | | | | | || | | * vmovaps %zmm15, %zmm17 #67.40
315 | 1.00 | | | | | | | || | | kmovw %k5, %k2 #67.40
316 | | | | | | | | || | | * vmovaps %zmm15, %zmm16 #66.40
317 | 0.00 | 1.00 | 0.50 0.50 | 0.50 0.50 | | 0.00 | | || 4.0 | | vmovdqu32 (%rdi,%rcx,4), %ymm1{%k5}{z} #65.25
318 | 1.00 | | | | | | | || | | kmovw %k5, %k1 #66.40
319 | | | | | | | | || | | * vmovaps %zmm15, %zmm18 #68.40
320 | 1.00 | | | | | | | || | | kmovw %k5, %k3 #68.40
321 | 1.50 | 0.50 | 4.00 0.50 | 4.00 0.50 | | 0.50 | 0.50 | || | | vgatherdpd (%r8,%ymm1,8), %zmm18{%k3} #68.40
322 | 1.50 | 0.50 | 4.00 0.50 | 4.00 0.50 | | 0.50 | 0.50 | || 4.0 | | vgatherdpd (%r9,%ymm1,8), %zmm17{%k2} #67.40
323 | 1.50 | 0.34 | 4.00 0.50 | 4.00 0.50 | | 0.50 | 0.66 | || | | vgatherdpd (%r10,%ymm1,8), %zmm16{%k1} #66.40
324 | 0.50 | | | | | 0.50 | | || | | vsubpd %zmm18, %zmm3, %zmm31 #68.40
325 | 0.50 | | | | | 0.50 | | || 4.0 | | vsubpd %zmm17, %zmm2, %zmm29 #67.40
326 | 0.50 | | | | | 0.50 | | || | | vsubpd %zmm16, %zmm4, %zmm28 #66.40
327 | 0.50 | | | | | 0.50 | | || 4.0 | | vmulpd %zmm29, %zmm29, %zmm27 #69.53
328 | 0.50 | | | | | 0.50 | | || 4.0 | | vfmadd231pd %zmm28, %zmm28, %zmm27 #69.53
329 | 0.50 | | | | | 0.50 | | || 4.0 | | vfmadd231pd %zmm31, %zmm31, %zmm27 #69.67
330 | 2.50 | | | | | 0.50 | | || 8.0 | | vrcp14pd %zmm27, %zmm26 #72.42
331 | | | | | | 1.00 | | || | | vcmppd $1, %zmm12, %zmm27, %k6{%k5} #71.26
332 | | | | | | 1.00 | | || | | vfpclasspd $30, %zmm26, %k0 #72.42
333 | | | | | | | | || | | * vmovaps %zmm27, %zmm19 #72.42
334 | 0.50 | | 0.50 0.50 | 0.50 0.50 | | 0.50 | | || 4.0 | | vfnmadd213pd .L_2il0floatpacket.5(%rip){1to8}, %zmm26, %zmm19 #72.42
335 | 1.00 | | | | | | | || | | knotw %k0, %k4 #72.42
336 | 0.50 | | | | | 0.50 | | || 4.0 | | vmulpd %zmm19, %zmm19, %zmm20 #72.42
337 | 0.50 | | | | | 0.50 | | || | | vfmadd213pd %zmm26, %zmm19, %zmm26{%k4} #72.42
338 | 0.50 | | | | | 0.50 | | || 4.0 | | vfmadd213pd %zmm26, %zmm20, %zmm26{%k4} #72.42
339 | 0.50 | | | | | 0.50 | | || 4.0 | | vmulpd %zmm11, %zmm26, %zmm21 #73.42
340 | 0.50 | | | | | 0.50 | | || | | vmulpd %zmm9, %zmm26, %zmm23 #74.58
341 | 0.50 | | | | | 0.50 | | || 4.0 | | vmulpd %zmm21, %zmm26, %zmm24 #73.48
342 | 0.50 | | | | | 0.50 | | || 4.0 | | vmulpd %zmm24, %zmm26, %zmm22 #73.54
343 | 0.00 | | | | | 1.00 | | || | | vfmsub213pd %zmm10, %zmm24, %zmm26 #74.58
344 | 0.00 | | | | | 1.00 | | || 4.0 | | vmulpd %zmm23, %zmm22, %zmm25 #74.65
345 | 0.00 | | | | | 1.00 | | || 4.0 | | vmulpd %zmm26, %zmm25, %zmm30 #74.71
346 | 0.00 | | | | | 1.00 | | || | | vfmadd231pd %zmm28, %zmm30, %zmm7{%k6} #75.21
347 | 0.00 | | | | | 1.00 | | || | | vfmadd231pd %zmm29, %zmm30, %zmm6{%k6} #76.21
348 | 0.00 | | | | | 1.00 | | || 4.0 | 4.0 | vfmadd231pd %zmm31, %zmm30, %zmm5{%k6} #77.21
18.0 3.00 13.0 2.50 13.0 2.50 18.0 3.00 68.0 4
Loop-Carried Dependencies Analysis Report
-----------------------------------------
313 | 1.0 | movslq %ecx, %rcx #64.13| [313]
348 | 4.0 | vfmadd231pd %zmm31, %zmm30, %zmm5{%k6} #77.21| [348]
347 | 4.0 | vfmadd231pd %zmm29, %zmm30, %zmm6{%k6} #76.21| [347]
346 | 4.0 | vfmadd231pd %zmm28, %zmm30, %zmm7{%k6} #75.21| [346]

12817
asm/avx2/force_aos_lt600_markers.s
View File

File diff suppressed because it is too large Load Diff

12567
asm/avx2/force_soa_att.s
View File

File diff suppressed because it is too large Load Diff

12571
asm/avx2/force_soa_lt600_markers.s
View File

File diff suppressed because it is too large Load Diff

563
asm/avx512/force_aos_geq1200_markers.s
View File

@ -1,563 +0,0 @@
# mark_description "Intel(R) C Intel(R) 64 Compiler for applications running on Intel(R) 64, Version 19.0.5.281 Build 20190815";
# mark_description "-I./src/includes -S -D_GNU_SOURCE -DAOS -DPRECISION=2 -DALIGNMENT=64 -restrict -Ofast -xCORE-AVX512 -qopt-zm";
# mark_description "m-usage=high -o ICC/force.s";
.file "force.c"
.text
..TXTST0:
.L_2__routine_start_computeForce_0:
# -- Begin computeForce
.text
# mark_begin;
.align 16,0x90
.globl computeForce
# --- computeForce(Parameter *, Atom *, Neighbor *, int)
computeForce:
# parameter 1: %rdi
# parameter 2: %rsi
# parameter 3: %rdx
# parameter 4: %ecx
..B1.1: # Preds ..B1.0
# Execution count [1.00e+00]
.cfi_startproc
..___tag_value_computeForce.1:
..L2:
#35.1
pushq %r12 #35.1
.cfi_def_cfa_offset 16
.cfi_offset 12, -16
pushq %r13 #35.1
.cfi_def_cfa_offset 24
.cfi_offset 13, -24
movq %rsi, %r8 #35.1
vmovsd 72(%rdi), %xmm2 #38.27
movq %rdx, %rax #35.1
vmovsd 8(%rdi), %xmm1 #39.23
vmovsd (%rdi), %xmm0 #40.24
movl 4(%r8), %r12d #36.18
movq 64(%r8), %r11 #41.20
movq 72(%r8), %rdx #41.45
movq 80(%r8), %r9 #41.70
testl %r12d, %r12d #44.24
jle ..B1.42 # Prob 50% #44.24
# LOE rax rdx rbx rbp r8 r9 r11 r14 r15 r12d xmm0 xmm1 xmm2
..B1.2: # Preds ..B1.1
# Execution count [1.00e+00]
movl %r12d, %edi #44.5
xorl %esi, %esi #44.5
movl $1, %r10d #44.5
xorl %ecx, %ecx #44.5
shrl $1, %edi #44.5
je ..B1.6 # Prob 9% #44.5
# LOE rax rdx rcx rbx rbp rsi rdi r8 r9 r11 r14 r15 r10d r12d xmm0 xmm1 xmm2
..B1.3: # Preds ..B1.2
# Execution count [9.00e-01]
xorl %r10d, %r10d #44.5
# LOE rax rdx rcx rbx rbp rsi rdi r8 r9 r10 r11 r14 r15 r12d xmm0 xmm1 xmm2
..B1.4: # Preds ..B1.4 ..B1.3
# Execution count [2.50e+00]
movq %r10, (%rcx,%r11) #45.9
incq %rsi #44.5
movq %r10, (%rcx,%rdx) #46.9
movq %r10, (%rcx,%r9) #47.9
movq %r10, 8(%rcx,%r11) #45.9
movq %r10, 8(%rcx,%rdx) #46.9
movq %r10, 8(%rcx,%r9) #47.9
addq $16, %rcx #44.5
cmpq %rdi, %rsi #44.5
jb ..B1.4 # Prob 63% #44.5
# LOE rax rdx rcx rbx rbp rsi rdi r8 r9 r10 r11 r14 r15 r12d xmm0 xmm1 xmm2
..B1.5: # Preds ..B1.4
# Execution count [9.00e-01]
lea 1(%rsi,%rsi), %r10d #45.9
# LOE rax rdx rbx rbp r8 r9 r11 r14 r15 r10d r12d xmm0 xmm1 xmm2
..B1.6: # Preds ..B1.2 ..B1.5
# Execution count [1.00e+00]
lea -1(%r10), %ecx #44.5
cmpl %r12d, %ecx #44.5
jae ..B1.8 # Prob 9% #44.5
# LOE rax rdx rbx rbp r8 r9 r11 r14 r15 r10d r12d xmm0 xmm1 xmm2
..B1.7: # Preds ..B1.6
# Execution count [9.00e-01]
movslq %r10d, %r10 #44.5
xorl %ecx, %ecx #45.9
movq %rcx, -8(%r11,%r10,8) #45.9
movq %rcx, -8(%rdx,%r10,8) #46.9
movq %rcx, -8(%r9,%r10,8) #47.9
# LOE rax rdx rbx rbp r8 r9 r11 r14 r15 r12d xmm0 xmm1 xmm2
..B1.8: # Preds ..B1.6 ..B1.7
# Execution count [9.00e-01]
vmulsd %xmm2, %xmm2, %xmm13 #38.45
xorl %ecx, %ecx #55.15
vmovdqu32 .L_2il0floatpacket.0(%rip), %ymm16 #67.9
vmulsd .L_2il0floatpacket.3(%rip), %xmm0, %xmm0 #77.41
vmovdqu .L_2il0floatpacket.1(%rip), %ymm15 #67.9
vmovups .L_2il0floatpacket.4(%rip), %zmm5 #77.54
vbroadcastsd %xmm13, %zmm14 #38.25
vbroadcastsd %xmm1, %zmm13 #39.21
vbroadcastsd %xmm0, %zmm10 #77.41
movslq %r12d, %r12 #55.5
xorl %esi, %esi #55.5
movq 24(%rax), %r13 #57.25
movq 16(%r8), %rdi #58.25
movslq 16(%rax), %r8 #56.43
movq 8(%rax), %r10 #56.19
xorl %eax, %eax #55.5
shlq $2, %r8 #37.5
movq %r12, -24(%rsp) #55.5[spill]
movq %r13, -16(%rsp) #55.5[spill]
movq %r11, -8(%rsp) #55.5[spill]
movq %r14, -88(%rsp) #55.5[spill]
movq %r15, -96(%rsp) #55.5[spill]
movq %rbx, -104(%rsp) #55.5[spill]
.cfi_offset 3, -128
.cfi_offset 14, -112
.cfi_offset 15, -120
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 ymm15 ymm16 zmm5 zmm10 zmm13 zmm14
..B1.9: # Preds ..B1.39 ..B1.8
# Execution count [5.00e+00]
movq -16(%rsp), %rbx #57.25[spill]
vxorpd %xmm24, %xmm24, %xmm24 #61.22
vmovapd %xmm24, %xmm18 #62.22
movl (%rbx,%rsi,4), %r11d #57.25
vmovapd %xmm18, %xmm4 #63.22
vmovsd (%rax,%rdi), %xmm6 #58.25
vmovsd 8(%rax,%rdi), %xmm7 #59.25
vmovsd 16(%rax,%rdi), %xmm12 #60.25
testl %r11d, %r11d #67.28
jle ..B1.39 # Prob 50% #67.28
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d xmm4 xmm6 xmm7 xmm12 xmm18 xmm24 ymm15 ymm16 zmm5 zmm10 zmm13 zmm14
..B1.10: # Preds ..B1.9
# Execution count [4.50e+00]
vpxord %zmm9, %zmm9, %zmm9 #61.22
vmovaps %zmm9, %zmm8 #62.22
vmovaps %zmm8, %zmm11 #63.22
cmpl $8, %r11d #67.9
jl ..B1.44 # Prob 10% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.11: # Preds ..B1.10
# Execution count [4.50e+00]
cmpl $1200, %r11d #67.9
jl ..B1.43 # Prob 10% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.12: # Preds ..B1.11
# Execution count [4.50e+00]
movq %r8, %r15 #56.43
imulq %rcx, %r15 #56.43
addq %r10, %r15 #37.5
movq %r15, %r12 #67.9
andq $63, %r12 #67.9
testl $3, %r12d #67.9
je ..B1.14 # Prob 50% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r15 r11d r12d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.13: # Preds ..B1.12
# Execution count [2.25e+00]
xorl %r12d, %r12d #67.9
jmp ..B1.16 # Prob 100% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r15 r11d r12d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.14: # Preds ..B1.12
# Execution count [2.25e+00]
testl %r12d, %r12d #67.9
je ..B1.16 # Prob 50% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r15 r11d r12d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.15: # Preds ..B1.14
# Execution count [2.50e+01]
negl %r12d #67.9
addl $64, %r12d #67.9
shrl $2, %r12d #67.9
cmpl %r12d, %r11d #67.9
cmovl %r11d, %r12d #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r15 r11d r12d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.16: # Preds ..B1.13 ..B1.15 ..B1.14
# Execution count [5.00e+00]
movl %r11d, %r14d #67.9
subl %r12d, %r14d #67.9
andl $7, %r14d #67.9
negl %r14d #67.9
addl %r11d, %r14d #67.9
cmpl $1, %r12d #67.9
jb ..B1.24 # Prob 50% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r15 r11d r12d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.17: # Preds ..B1.16
# Execution count [4.50e+00]
vmovdqa %ymm15, %ymm3 #67.9
xorl %r13d, %r13d #67.9
vpbroadcastd %r12d, %ymm2 #67.9
vbroadcastsd %xmm6, %zmm1 #58.23
vbroadcastsd %xmm7, %zmm0 #59.23
vbroadcastsd %xmm12, %zmm4 #60.23
movslq %r12d, %rbx #67.9
movq %r9, -80(%rsp) #67.9[spill]
movq %rdx, -72(%rsp) #67.9[spill]
# LOE rax rcx rbx rbp rsi rdi r8 r10 r13 r15 r11d r12d r14d xmm6 xmm7 xmm12 ymm2 ymm3 ymm15 ymm16 zmm0 zmm1 zmm4 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.18: # Preds ..B1.22 ..B1.17
# Execution count [2.50e+01]
movl $111,%ebx #IACA/OSACA START MARKER
.byte 100,103,144 #IACA/OSACA START MARKER
vpcmpgtd %ymm3, %ymm2, %k3 #67.9
vmovdqu32 (%r15,%r13,4), %ymm17{%k3}{z} #68.21
kmovw %k3, %r9d #67.9
vpaddd %ymm17, %ymm17, %ymm18 #69.36
vpaddd %ymm18, %ymm17, %ymm17 #69.36
# LOE rax rcx rbx rbp rsi rdi r8 r10 r13 r15 r9d r11d r12d r14d xmm6 xmm7 xmm12 ymm2 ymm3 ymm15 ymm16 ymm17 zmm0 zmm1 zmm4 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14 k3
..B1.21: # Preds ..B1.18
# Execution count [1.25e+01]
kmovw %k3, %k1 #69.36
kmovw %k3, %k2 #69.36
vpxord %zmm18, %zmm18, %zmm18 #69.36
vpxord %zmm19, %zmm19, %zmm19 #69.36
vpxord %zmm20, %zmm20, %zmm20 #69.36
vgatherdpd 16(%rdi,%ymm17,8), %zmm18{%k1} #69.36
vgatherdpd 8(%rdi,%ymm17,8), %zmm19{%k2} #69.36
vgatherdpd (%rdi,%ymm17,8), %zmm20{%k3} #69.36
# LOE rax rcx rbx rbp rsi rdi r8 r10 r13 r15 r9d r11d r12d r14d xmm6 xmm7 xmm12 ymm2 ymm3 ymm15 ymm16 zmm0 zmm1 zmm4 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14 zmm18 zmm19 zmm20
..B1.22: # Preds ..B1.21
# Execution count [2.50e+01]
addq $8, %r13 #67.9
vpaddd %ymm16, %ymm3, %ymm3 #67.9
vsubpd %zmm18, %zmm4, %zmm29 #71.36
vsubpd %zmm19, %zmm0, %zmm27 #70.36
vsubpd %zmm20, %zmm1, %zmm26 #69.36
vmulpd %zmm27, %zmm27, %zmm25 #72.49
vfmadd231pd %zmm26, %zmm26, %zmm25 #72.49
vfmadd231pd %zmm29, %zmm29, %zmm25 #72.63
vrcp14pd %zmm25, %zmm24 #75.38
vcmppd $1, %zmm14, %zmm25, %k2 #74.22
vfpclasspd $30, %zmm24, %k0 #75.38
kmovw %k2, %edx #74.22
knotw %k0, %k1 #75.38
vmovaps %zmm25, %zmm17 #75.38
andl %edx, %r9d #74.22
vfnmadd213pd .L_2il0floatpacket.9(%rip){1to8}, %zmm24, %zmm17 #75.38
kmovw %r9d, %k3 #78.17
vmulpd %zmm17, %zmm17, %zmm18 #75.38
vfmadd213pd %zmm24, %zmm17, %zmm24{%k1} #75.38
vfmadd213pd %zmm24, %zmm18, %zmm24{%k1} #75.38
vmulpd %zmm13, %zmm24, %zmm19 #76.38
vmulpd %zmm10, %zmm24, %zmm21 #77.54
vmulpd %zmm19, %zmm24, %zmm22 #76.44
vmulpd %zmm22, %zmm24, %zmm20 #76.50
vfmsub213pd %zmm5, %zmm22, %zmm24 #77.54
vmulpd %zmm21, %zmm20, %zmm23 #77.61
vmulpd %zmm24, %zmm23, %zmm28 #77.67
vfmadd231pd %zmm26, %zmm28, %zmm9{%k3} #78.17
vfmadd231pd %zmm27, %zmm28, %zmm8{%k3} #79.17
vfmadd231pd %zmm29, %zmm28, %zmm11{%k3} #80.17
cmpq %rbx, %r13 #67.9
jb ..B1.18 # Prob 82% #67.9
movl $222,%ebx #IACA/OSACA END MARKER
.byte 100,103,144 #IACA/OSACA END MARKER
# LOE rax rcx rbx rbp rsi rdi r8 r10 r13 r15 r11d r12d r14d xmm6 xmm7 xmm12 ymm2 ymm3 ymm15 ymm16 zmm0 zmm1 zmm4 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.23: # Preds ..B1.22
# Execution count [4.50e+00]
movq -80(%rsp), %r9 #[spill]
movq -72(%rsp), %rdx #[spill]
cmpl %r12d, %r11d #67.9
je ..B1.38 # Prob 10% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d r12d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.24: # Preds ..B1.23 ..B1.16 ..B1.43
# Execution count [2.50e+01]
lea 8(%r12), %ebx #67.9
cmpl %ebx, %r14d #67.9
jl ..B1.32 # Prob 50% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d r12d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.25: # Preds ..B1.24
# Execution count [4.50e+00]
movq %r8, %r13 #56.43
imulq %rcx, %r13 #56.43
vbroadcastsd %xmm6, %zmm2 #58.23
vbroadcastsd %xmm7, %zmm1 #59.23
vbroadcastsd %xmm12, %zmm0 #60.23
movslq %r12d, %rbx #67.9
addq %r10, %r13 #37.5
movq %rax, -64(%rsp) #37.5[spill]
movq %r8, -56(%rsp) #37.5[spill]
movq %r10, -48(%rsp) #37.5[spill]
movq %rsi, -40(%rsp) #37.5[spill]
movq %rcx, -32(%rsp) #37.5[spill]
movq %r9, -80(%rsp) #37.5[spill]
movq %rdx, -72(%rsp) #37.5[spill]
# LOE rbx rbp rdi r13 r11d r12d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm0 zmm1 zmm2 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.26: # Preds ..B1.30 ..B1.25
# Execution count [2.50e+01]
vmovdqu (%r13,%rbx,4), %ymm3 #68.21
vpaddd %ymm3, %ymm3, %ymm4 #69.36
vpaddd %ymm4, %ymm3, %ymm3 #69.36
movl (%r13,%rbx,4), %r10d #68.21
movl 4(%r13,%rbx,4), %r9d #68.21
movl 8(%r13,%rbx,4), %r8d #68.21
movl 12(%r13,%rbx,4), %esi #68.21
lea (%r10,%r10,2), %r10d #69.36
movl 16(%r13,%rbx,4), %ecx #68.21
lea (%r9,%r9,2), %r9d #69.36
movl 20(%r13,%rbx,4), %edx #68.21
lea (%r8,%r8,2), %r8d #69.36
movl 24(%r13,%rbx,4), %eax #68.21
lea (%rsi,%rsi,2), %esi #69.36
movl 28(%r13,%rbx,4), %r15d #68.21
lea (%rcx,%rcx,2), %ecx #69.36
lea (%rdx,%rdx,2), %edx #69.36
lea (%rax,%rax,2), %eax #69.36
lea (%r15,%r15,2), %r15d #69.36
# LOE rbx rbp rdi r13 eax edx ecx esi r8d r9d r10d r11d r12d r14d r15d xmm6 xmm7 xmm12 ymm3 ymm15 ymm16 zmm0 zmm1 zmm2 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.29: # Preds ..B1.26
# Execution count [1.25e+01]
vpcmpeqb %xmm0, %xmm0, %k1 #69.36
vpcmpeqb %xmm0, %xmm0, %k2 #69.36
vpcmpeqb %xmm0, %xmm0, %k3 #69.36
vpxord %zmm4, %zmm4, %zmm4 #69.36
vpxord %zmm17, %zmm17, %zmm17 #69.36
vpxord %zmm18, %zmm18, %zmm18 #69.36
vgatherdpd 16(%rdi,%ymm3,8), %zmm4{%k1} #69.36
vgatherdpd 8(%rdi,%ymm3,8), %zmm17{%k2} #69.36
vgatherdpd (%rdi,%ymm3,8), %zmm18{%k3} #69.36
# LOE rbx rbp rdi r13 r11d r12d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm0 zmm1 zmm2 zmm4 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14 zmm17 zmm18
..B1.30: # Preds ..B1.29
# Execution count [2.50e+01]
addl $8, %r12d #67.9
addq $8, %rbx #67.9
vsubpd %zmm4, %zmm0, %zmm26 #71.36
vsubpd %zmm17, %zmm1, %zmm24 #70.36
vsubpd %zmm18, %zmm2, %zmm23 #69.36
vmulpd %zmm24, %zmm24, %zmm3 #72.49
vfmadd231pd %zmm23, %zmm23, %zmm3 #72.49
vfmadd231pd %zmm26, %zmm26, %zmm3 #72.63
vrcp14pd %zmm3, %zmm22 #75.38
vcmppd $1, %zmm14, %zmm3, %k2 #74.22
vfpclasspd $30, %zmm22, %k0 #75.38
vfnmadd213pd .L_2il0floatpacket.9(%rip){1to8}, %zmm22, %zmm3 #75.38
knotw %k0, %k1 #75.38
vmulpd %zmm3, %zmm3, %zmm4 #75.38
vfmadd213pd %zmm22, %zmm3, %zmm22{%k1} #75.38
vfmadd213pd %zmm22, %zmm4, %zmm22{%k1} #75.38
vmulpd %zmm13, %zmm22, %zmm17 #76.38
vmulpd %zmm10, %zmm22, %zmm19 #77.54
vmulpd %zmm17, %zmm22, %zmm20 #76.44
vmulpd %zmm20, %zmm22, %zmm18 #76.50
vfmsub213pd %zmm5, %zmm20, %zmm22 #77.54
vmulpd %zmm19, %zmm18, %zmm21 #77.61
vmulpd %zmm22, %zmm21, %zmm25 #77.67
vfmadd231pd %zmm23, %zmm25, %zmm9{%k2} #78.17
vfmadd231pd %zmm24, %zmm25, %zmm8{%k2} #79.17
vfmadd231pd %zmm26, %zmm25, %zmm11{%k2} #80.17
cmpl %r14d, %r12d #67.9
jb ..B1.26 # Prob 82% #67.9
# LOE rbx rbp rdi r13 r11d r12d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm0 zmm1 zmm2 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.31: # Preds ..B1.30
# Execution count [4.50e+00]
movq -64(%rsp), %rax #[spill]
movq -56(%rsp), %r8 #[spill]
movq -48(%rsp), %r10 #[spill]
movq -40(%rsp), %rsi #[spill]
movq -32(%rsp), %rcx #[spill]
movq -80(%rsp), %r9 #[spill]
movq -72(%rsp), %rdx #[spill]
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.32: # Preds ..B1.31 ..B1.24 ..B1.44
# Execution count [5.00e+00]
lea 1(%r14), %ebx #67.9
cmpl %r11d, %ebx #67.9
ja ..B1.38 # Prob 50% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.33: # Preds ..B1.32
# Execution count [2.50e+01]
imulq %r8, %rcx #56.43
vbroadcastsd %xmm6, %zmm4 #58.23
subl %r14d, %r11d #67.9
addq %r10, %rcx #37.5
vpbroadcastd %r11d, %ymm0 #67.9
vpcmpgtd %ymm15, %ymm0, %k3 #67.9
movslq %r14d, %r14 #67.9
kmovw %k3, %ebx #67.9
vmovdqu32 (%rcx,%r14,4), %ymm1{%k3}{z} #68.21
vpaddd %ymm1, %ymm1, %ymm2 #69.36
vpaddd %ymm2, %ymm1, %ymm0 #69.36
# LOE rax rdx rbp rsi rdi r8 r9 r10 ebx xmm7 xmm12 ymm0 ymm15 ymm16 zmm4 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14 k3
..B1.36: # Preds ..B1.33
# Execution count [1.25e+01]
kmovw %k3, %k1 #69.36
kmovw %k3, %k2 #69.36
vpxord %zmm1, %zmm1, %zmm1 #69.36
vpxord %zmm2, %zmm2, %zmm2 #69.36
vpxord %zmm3, %zmm3, %zmm3 #69.36
vgatherdpd 16(%rdi,%ymm0,8), %zmm1{%k1} #69.36
vgatherdpd 8(%rdi,%ymm0,8), %zmm2{%k2} #69.36
vgatherdpd (%rdi,%ymm0,8), %zmm3{%k3} #69.36
# LOE rax rdx rbp rsi rdi r8 r9 r10 ebx xmm7 xmm12 ymm15 ymm16 zmm1 zmm2 zmm3 zmm4 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.37: # Preds ..B1.36
# Execution count [2.50e+01]
vbroadcastsd %xmm7, %zmm7 #59.23
vbroadcastsd %xmm12, %zmm12 #60.23
vsubpd %zmm1, %zmm12, %zmm23 #71.36
vsubpd %zmm2, %zmm7, %zmm21 #70.36
vsubpd %zmm3, %zmm4, %zmm20 #69.36
vmulpd %zmm21, %zmm21, %zmm19 #72.49
vfmadd231pd %zmm20, %zmm20, %zmm19 #72.49
vfmadd231pd %zmm23, %zmm23, %zmm19 #72.63
vrcp14pd %zmm19, %zmm18 #75.38
vcmppd $1, %zmm14, %zmm19, %k2 #74.22
vfpclasspd $30, %zmm18, %k0 #75.38
kmovw %k2, %ecx #74.22
knotw %k0, %k1 #75.38
vmovaps %zmm19, %zmm0 #75.38
andl %ecx, %ebx #74.22
vfnmadd213pd .L_2il0floatpacket.9(%rip){1to8}, %zmm18, %zmm0 #75.38
kmovw %ebx, %k3 #78.17
vmulpd %zmm0, %zmm0, %zmm1 #75.38
vfmadd213pd %zmm18, %zmm0, %zmm18{%k1} #75.38
vfmadd213pd %zmm18, %zmm1, %zmm18{%k1} #75.38
vmulpd %zmm13, %zmm18, %zmm2 #76.38
vmulpd %zmm10, %zmm18, %zmm4 #77.54
vmulpd %zmm2, %zmm18, %zmm6 #76.44
vmulpd %zmm6, %zmm18, %zmm3 #76.50
vfmsub213pd %zmm5, %zmm6, %zmm18 #77.54
vmulpd %zmm4, %zmm3, %zmm17 #77.61
vmulpd %zmm18, %zmm17, %zmm22 #77.67
vfmadd231pd %zmm20, %zmm22, %zmm9{%k3} #78.17
vfmadd231pd %zmm21, %zmm22, %zmm8{%k3} #79.17
vfmadd231pd %zmm23, %zmm22, %zmm11{%k3} #80.17
# LOE rax rdx rbp rsi rdi r8 r9 r10 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.38: # Preds ..B1.23 ..B1.37 ..B1.32
# Execution count [4.50e+00]
vmovups .L_2il0floatpacket.10(%rip), %zmm19 #63.22
vpermd %zmm11, %zmm19, %zmm0 #63.22
vpermd %zmm8, %zmm19, %zmm6 #62.22
vpermd %zmm9, %zmm19, %zmm20 #61.22
vaddpd %zmm11, %zmm0, %zmm11 #63.22
vaddpd %zmm8, %zmm6, %zmm8 #62.22
vaddpd %zmm9, %zmm20, %zmm9 #61.22
vpermpd $78, %zmm11, %zmm1 #63.22
vpermpd $78, %zmm8, %zmm7 #62.22
vpermpd $78, %zmm9, %zmm21 #61.22
vaddpd %zmm1, %zmm11, %zmm2 #63.22
vaddpd %zmm7, %zmm8, %zmm12 #62.22
vaddpd %zmm21, %zmm9, %zmm22 #61.22
vpermpd $177, %zmm2, %zmm3 #63.22
vpermpd $177, %zmm12, %zmm17 #62.22
vpermpd $177, %zmm22, %zmm23 #61.22
vaddpd %zmm3, %zmm2, %zmm4 #63.22
vaddpd %zmm17, %zmm12, %zmm18 #62.22
vaddpd %zmm23, %zmm22, %zmm24 #61.22
# LOE rax rdx rbp rsi rdi r8 r9 r10 xmm4 xmm18 xmm24 ymm15 ymm16 zmm5 zmm10 zmm13 zmm14
..B1.39: # Preds ..B1.38 ..B1.9
# Execution count [5.00e+00]
movq -8(%rsp), %rcx #84.9[spill]
addq $24, %rax #55.5
vaddsd (%rcx,%rsi,8), %xmm24, %xmm0 #84.9
vmovsd %xmm0, (%rcx,%rsi,8) #84.9
movslq %esi, %rcx #55.32
vaddsd (%rdx,%rsi,8), %xmm18, %xmm1 #85.9
vmovsd %xmm1, (%rdx,%rsi,8) #85.9
incq %rcx #55.32
vaddsd (%r9,%rsi,8), %xmm4, %xmm2 #86.9
vmovsd %xmm2, (%r9,%rsi,8) #86.9
incq %rsi #55.5
cmpq -24(%rsp), %rsi #55.5[spill]
jb ..B1.9 # Prob 82% #55.5
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 ymm15 ymm16 zmm5 zmm10 zmm13 zmm14
..B1.40: # Preds ..B1.39
# Execution count [9.00e-01]
movq -88(%rsp), %r14 #[spill]
.cfi_restore 14
movq -96(%rsp), %r15 #[spill]
.cfi_restore 15
movq -104(%rsp), %rbx #[spill]
.cfi_restore 3
# LOE rbx rbp r14 r15
..B1.42: # Preds ..B1.1 ..B1.40
# Execution count [1.00e+00]
vzeroupper #93.12
vxorpd %xmm0, %xmm0, %xmm0 #93.12
.cfi_restore 13
popq %r13 #93.12
.cfi_def_cfa_offset 16
.cfi_restore 12
popq %r12 #93.12
.cfi_def_cfa_offset 8
ret #93.12
.cfi_def_cfa_offset 24
.cfi_offset 3, -128
.cfi_offset 12, -16
.cfi_offset 13, -24
.cfi_offset 14, -112
.cfi_offset 15, -120
# LOE
..B1.43: # Preds ..B1.11
# Execution count [4.50e-01]: Infreq
movl %r11d, %r14d #67.9
xorl %r12d, %r12d #67.9
andl $-8, %r14d #67.9
jmp ..B1.24 # Prob 100% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d r12d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.44: # Preds ..B1.10
# Execution count [4.50e-01]: Infreq
xorl %r14d, %r14d #67.9
jmp ..B1.32 # Prob 100% #67.9
.align 16,0x90
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
.cfi_endproc
# mark_end;
.type computeForce,@function
.size computeForce,.-computeForce
..LNcomputeForce.0:
.data
# -- End computeForce
.section .rodata, "a"
.align 64
.align 64
.L_2il0floatpacket.2:
.long 0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000
.type .L_2il0floatpacket.2,@object
.size .L_2il0floatpacket.2,64
.align 64
.L_2il0floatpacket.4:
.long 0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000
.type .L_2il0floatpacket.4,@object
.size .L_2il0floatpacket.4,64
.align 64
.L_2il0floatpacket.5:
.long 0x02010101,0x04040202,0x08080804,0x20101010,0x40402020,0x80808040,0x00000000,0x00000000,0x00000000,0x00000000,0x00000000,0x00000000,0x00000000,0x00000000,0x00000000,0x00000000
.type .L_2il0floatpacket.5,@object
.size .L_2il0floatpacket.5,64
.align 64
.L_2il0floatpacket.6:
.long 0x00000000,0x00000000,0x00000004,0x00000000,0x00000008,0x00000000,0x0000000c,0x00000000,0x00000001,0x00000000,0x00000005,0x00000000,0x00000009,0x00000000,0x0000000d,0x00000000
.type .L_2il0floatpacket.6,@object
.size .L_2il0floatpacket.6,64
.align 64
.L_2il0floatpacket.7:
.long 0x00000001,0x00000000,0x00000005,0x00000000,0x00000009,0x00000000,0x0000000d,0x00000000,0x00000000,0x00000000,0x00000004,0x00000000,0x00000008,0x00000000,0x0000000c,0x00000000
.type .L_2il0floatpacket.7,@object
.size .L_2il0floatpacket.7,64
.align 64
.L_2il0floatpacket.8:
.long 0x00000002,0x00000000,0x00000006,0x00000000,0x0000000a,0x00000000,0x0000000e,0x00000000,0x00000002,0x00000000,0x00000006,0x00000000,0x0000000a,0x00000000,0x0000000e,0x00000000
.type .L_2il0floatpacket.8,@object
.size .L_2il0floatpacket.8,64
.align 64
.L_2il0floatpacket.10:
.long 0x00000008,0x00000009,0x0000000a,0x0000000b,0x0000000c,0x0000000d,0x0000000e,0x0000000f,0x00000008,0x00000009,0x0000000a,0x0000000b,0x0000000c,0x0000000d,0x0000000e,0x0000000f
.type .L_2il0floatpacket.10,@object
.size .L_2il0floatpacket.10,64
.align 32
.L_2il0floatpacket.0:
.long 0x00000008,0x00000008,0x00000008,0x00000008,0x00000008,0x00000008,0x00000008,0x00000008
.type .L_2il0floatpacket.0,@object
.size .L_2il0floatpacket.0,32
.align 32
.L_2il0floatpacket.1:
.long 0x00000000,0x00000001,0x00000002,0x00000003,0x00000004,0x00000005,0x00000006,0x00000007
.type .L_2il0floatpacket.1,@object
.size .L_2il0floatpacket.1,32
.align 8
.L_2il0floatpacket.3:
.long 0x00000000,0x40480000
.type .L_2il0floatpacket.3,@object
.size .L_2il0floatpacket.3,8
.align 8
.L_2il0floatpacket.9:
.long 0x00000000,0x3ff00000
.type .L_2il0floatpacket.9,@object
.size .L_2il0floatpacket.9,8
.data
.section .note.GNU-stack, ""
# End

976
asm/avx512/force_aos_intel.s
View File

@ -1,976 +0,0 @@
# mark_description "Intel(R) C Intel(R) 64 Compiler for applications running on Intel(R) 64, Version 19.0.5.281 Build 20190815";
# mark_description "-I./src/includes -S -masm=intel -D_GNU_SOURCE -DAOS -DPRECISION=2 -DALIGNMENT=64 -restrict -Ofast -xCORE-AVX";
# mark_description "512 -qopt-zmm-usage=high -o ICC/force.s";
.intel_syntax noprefix
.file "force.c"
.text
..TXTST0:
.L_2__routine_start_computeForce_0:
# -- Begin computeForce
.text
# mark_begin;
.align 16,0x90
.globl computeForce
# --- computeForce(Parameter *, Atom *, Neighbor *, int)
computeForce:
# parameter 1: rdi
# parameter 2: rsi
# parameter 3: rdx
# parameter 4: ecx
..B1.1: # Preds ..B1.0
# Execution count [1.00e+00]
.cfi_startproc
..___tag_value_computeForce.1:
..L2:
#35.1
push r12 #35.1
.cfi_def_cfa_offset 16
.cfi_offset 12, -16
push r13 #35.1
.cfi_def_cfa_offset 24
.cfi_offset 13, -24
# r8 <- atom
mov r8, rsi #35.1
# xmm2 <- param->cutforce
vmovsd xmm2, QWORD PTR [72+rdi] #38.27
# rax <- neighbors
mov rax, rdx #35.1
# xmm1 <- param->sigma6
vmovsd xmm1, QWORD PTR [8+rdi] #39.23
# xmm0 <- param->epsilon
vmovsd xmm0, QWORD PTR [rdi] #40.24
# r12d <- atom->Nlocal
mov r12d, DWORD PTR [4+r8] #36.18
# r11 <- atom->fx
mov r11, QWORD PTR [64+r8] #41.20
# rdx <- atom->fy
mov rdx, QWORD PTR [72+r8] #41.45
# r9 <- atom->fz
mov r9, QWORD PTR [80+r8] #41.70
# atom->Nlocal > 0
test r12d, r12d #44.24
jle ..B1.42 # Prob 50% #44.24
# LOE rax rdx rbx rbp r8 r9 r11 r14 r15 r12d xmm0 xmm1 xmm2
..B1.2: # Preds ..B1.1
# Execution count [1.00e+00]
# edi <- atom->Nlocal
mov edi, r12d #44.5
# esi <- 0
xor esi, esi #44.5
# r10d <- 1
mov r10d, 1 #44.5
# ecx <- 0
xor ecx, ecx #44.5
# edi <- atom->Nlocal >> 1
shr edi, 1 #44.5
je ..B1.6 # Prob 9% #44.5
# LOE rax rdx rcx rbx rbp rsi rdi r8 r9 r11 r14 r15 r10d r12d xmm0 xmm1 xmm2
..B1.3: # Preds ..B1.2
# Execution count [9.00e-01]
# r10d <- 0
xor r10d, r10d #44.5
# LOE rax rdx rcx rbx rbp rsi rdi r8 r9 r10 r11 r14 r15 r12d xmm0 xmm1 xmm2
..B1.4: # Preds ..B1.4 ..B1.3
# Execution count [2.50e+00]
# atom->fx[i] <- 0.0
mov QWORD PTR [rcx+r11], r10 #45.9
# i++
inc rsi #44.5
# atom->fy[i] <- 0.0
mov QWORD PTR [rcx+rdx], r10 #46.9
# atom->fz[i] <- 0.0
mov QWORD PTR [rcx+r9], r10 #47.9
mov QWORD PTR [8+rcx+r11], r10 #45.9
mov QWORD PTR [8+rcx+rdx], r10 #46.9
mov QWORD PTR [8+rcx+r9], r10 #47.9
add rcx, 16 #44.5
cmp rsi, rdi #44.5
jb ..B1.4 # Prob 63% #44.5
# LOE rax rdx rcx rbx rbp rsi rdi r8 r9 r10 r11 r14 r15 r12d xmm0 xmm1 xmm2
..B1.5: # Preds ..B1.4
# Execution count [9.00e-01]
lea r10d, DWORD PTR [1+rsi+rsi] #45.9
# LOE rax rdx rbx rbp r8 r9 r11 r14 r15 r10d r12d xmm0 xmm1 xmm2
..B1.6: # Preds ..B1.2 ..B1.5
# Execution count [1.00e+00]
lea ecx, DWORD PTR [-1+r10] #44.5
cmp ecx, r12d #44.5
jae ..B1.8 # Prob 9% #44.5
# LOE rax rdx rbx rbp r8 r9 r11 r14 r15 r10d r12d xmm0 xmm1 xmm2
..B1.7: # Preds ..B1.6
# Execution count [9.00e-01]
movsxd r10, r10d #44.5
xor ecx, ecx #45.9
mov QWORD PTR [-8+r11+r10*8], rcx #45.9
mov QWORD PTR [-8+rdx+r10*8], rcx #46.9
mov QWORD PTR [-8+r9+r10*8], rcx #47.9
# LOE rax rdx rbx rbp r8 r9 r11 r14 r15 r12d xmm0 xmm1 xmm2
..B1.8: # Preds ..B1.6 ..B1.7
# Execution count [9.00e-01]
# xmm13 <- cutforcesq
vmulsd xmm13, xmm2, xmm2 #38.45
# ecx <- i
xor ecx, ecx #55.15
# ymm16 <- 8 (vector width / sizeof(double) ?)
vmovdqu32 ymm16, YMMWORD PTR .L_2il0floatpacket.0[rip] #67.9
# xmm0 <- 48 * sr6
vmulsd xmm0, xmm0, QWORD PTR .L_2il0floatpacket.3[rip] #77.41
# ymm15 <- [0..7]
vmovdqu ymm15, YMMWORD PTR .L_2il0floatpacket.1[rip] #67.9
# zmm5 <- 0.5
vmovups zmm5, ZMMWORD PTR .L_2il0floatpacket.4[rip] #77.54
# zmm14 <- cutforcesq
vbroadcastsd zmm14, xmm13 #38.25
# zmm13 <- param->sigma6
vbroadcastsd zmm13, xmm1 #39.21
# zmm10 <- param->epsilon
vbroadcastsd zmm10, xmm0 #77.41
# r12 <- atom->nlocal
movsxd r12, r12d #55.5
# esi <- 0 (i)
xor esi, esi #55.5
# r13 <- neighbor->numneighs
mov r13, QWORD PTR [24+rax] #57.25
# rdi <- atom->x
mov rdi, QWORD PTR [16+r8] #58.25
# r8 <- neighbor->maxneighs
movsxd r8, DWORD PTR [16+rax] #56.43
# r10 <- neighbor->neighbors
mov r10, QWORD PTR [8+rax] #56.19
# eax <- 0 (i * 3)
xor eax, eax #55.5
# r8 <- neighbor->maxneighs << 2
shl r8, 2 #37.5
# [-24+rsp] <- atom->nlocal
mov QWORD PTR [-24+rsp], r12 #55.5[spill]
# [-16+rsp] <- neighbor->numneigh
mov QWORD PTR [-16+rsp], r13 #55.5[spill]
# [-8+rsp] <- atom->fx
mov QWORD PTR [-8+rsp], r11 #55.5[spill]
# [-88+rsp] <- r14 (unused so far)
mov QWORD PTR [-88+rsp], r14 #55.5[spill]
# [-96+rsp] <- r15 (unused so far)
mov QWORD PTR [-96+rsp], r15 #55.5[spill]
# [-104+rsp] <- rbx (unused so far)
mov QWORD PTR [-104+rsp], rbx #55.5[spill]
.cfi_offset 3, -128
.cfi_offset 14, -112
.cfi_offset 15, -120
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 ymm15 ymm16 zmm5 zmm10 zmm13 zmm14
..B1.9: # Preds ..B1.39 ..B1.8
# Execution count [5.00e+00]
# rbx <- neighbor->numneigh
mov rbx, QWORD PTR [-16+rsp] #57.25[spill]
# xmm24 <- 0 (fix)
vxorpd xmm24, xmm24, xmm24 #61.22
# xmm18 <- 0 (fiy)
vmovapd xmm18, xmm24 #62.22
# r11d <- neighbor->numneigh[i]
mov r11d, DWORD PTR [rbx+rsi*4] #57.25
# xmm4 <- 0 (fiz)
vmovapd xmm4, xmm18 #63.22
# xmm6 <- atom->x[i * 3]
vmovsd xmm6, QWORD PTR [rax+rdi] #58.25
# xmm6 <- atom->x[i * 3 + 1]
vmovsd xmm7, QWORD PTR [8+rax+rdi] #59.25
# xmm6 <- atom->x[i * 3 + 2]
vmovsd xmm12, QWORD PTR [16+rax+rdi] #60.25
# neighbor->numneigh[i] > 0
test r11d, r11d #67.28
jle ..B1.39 # Prob 50% #67.28
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d xmm4 xmm6 xmm7 xmm12 xmm18 xmm24 ymm15 ymm16 zmm5 zmm10 zmm13 zmm14
..B1.10: # Preds ..B1.9
# Execution count [4.50e+00]
# zmm9 <- 0 (fix)
vpxord zmm9, zmm9, zmm9 #61.22
# zmm8 <- 0 (fiy)
vmovaps zmm8, zmm9 #62.22
# zmm11 <- 0 (fiz)
vmovaps zmm11, zmm8 #63.22
# neighbor->numneigh[i] < 8
cmp r11d, 8 #67.9
jl ..B1.44 # Prob 10% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.11: # Preds ..B1.10
# Execution count [4.50e+00]
# neighbor->numneigh[i] < 1200
# L1? 1200 * 3 * 8 = 28800 = 28.8kB
cmp r11d, 1200 #67.9
jl ..B1.43 # Prob 10% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.12: # Preds ..B1.11
# Execution count [4.50e+00]
# r15 <- neighbor->maxneighs
mov r15, r8 #56.43
# r15 <- neighbor->maxneighs * i
imul r15, rcx #56.43
# r15 <- neighbor->neighbors + neighbor->maxneighs * i
add r15, r10 #37.5
# r12 <- neighbor->neighbors + neighbor->maxneighs * i
mov r12, r15 #67.9
# r12 <- (neighbor->neighbors + neighbor->maxneighs * i) & 63 -- 0b00111111
and r12, 63 #67.9
# (neighbor->neighbors + neighbor->maxneighs * i) & 63 == 3 -- 0b00000011
test r12d, 3 #67.9
je ..B1.14 # Prob 50% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r15 r11d r12d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.13: # Preds ..B1.12
# Execution count [2.25e+00]
# r12d <- 0
xor r12d, r12d #67.9
jmp ..B1.16 # Prob 100% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r15 r11d r12d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.14: # Preds ..B1.12
# Execution count [2.25e+00]
# r12d == 0
test r12d, r12d #67.9
je ..B1.16 # Prob 50% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r15 r11d r12d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.15: # Preds ..B1.14
# Execution count [2.50e+01]
# r12d <- ~r12d + 1
neg r12d #67.9
# r12d <- (~r12d + 1) + 64
add r12d, 64 #67.9
# r12d <- ((~r12d + 1) + 64) >> 2
shr r12d, 2 #67.9
# neighbor->numneigh[i] < r12d
cmp r11d, r12d #67.9
# r12d <- min(r12d, neighbor->numneigh[i])
cmovl r12d, r11d #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r15 r11d r12d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.16: # Preds ..B1.13 ..B1.15 ..B1.14
# Execution count [5.00e+00]
# r14d <- neighbor->numneigh[i]
mov r14d, r11d #67.9
# r14d <- neighbor->numneigh[i] - r12d
sub r14d, r12d #67.9
# r14d <- (neighbor->numneigh[i] - r12d) & 7 (0b0111)
and r14d, 7 #67.9
# r14d <- ~r14d + 1
neg r14d #67.9
# r14d <- r14d + neighbor->numneigh[i]
add r14d, r11d #67.9
# r12d < 1
cmp r12d, 1 #67.9
jb ..B1.24 # Prob 50% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r15 r11d r12d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.17: # Preds ..B1.16
# Execution count [4.50e+00]
# ymm3 <- [0..7]
vmovdqa ymm3, ymm15 #67.9
# r13d <- 0
xor r13d, r13d #67.9
# ymm2 <- r12d (k? amount of neighs?)
vpbroadcastd ymm2, r12d #67.9
# zmm1 <- atom->x[i * 3]
vbroadcastsd zmm1, xmm6 #58.23
# zmm0 <- atom->x[i * 3 + 1]
vbroadcastsd zmm0, xmm7 #59.23
# zmm4 <- atom->x[i * 3 + 2]
vbroadcastsd zmm4, xmm12 #60.23
# rbx <- r12d (k? amount of neighs?)
movsxd rbx, r12d #67.9
# [-80+rsp] <- atom->fz
mov QWORD PTR [-80+rsp], r9 #67.9[spill]
# [-72+rdx] <- atom->fy
mov QWORD PTR [-72+rsp], rdx #67.9[spill]
# LOE rax rcx rbx rbp rsi rdi r8 r10 r13 r15 r11d r12d r14d xmm6 xmm7 xmm12 ymm2 ymm3 ymm15 ymm16 zmm0 zmm1 zmm4 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.18: # Preds ..B1.22 ..B1.17
# Execution count [2.50e+01]
# k3 <- [x when amount_of_neighs < [0..7](x)] (i.e. mask of available neighbors)
vpcmpgtd k3, ymm2, ymm3 #67.9
# ymm17 <- neighs[k]
vmovdqu32 ymm17{k3}{z}, YMMWORD PTR [r15+r13*4] #68.21
# r9d <- k3
kmovw r9d, k3 #67.9
# ymm18 <- neighs[k] * 2
vpaddd ymm18, ymm17, ymm17 #69.36
# ymm17 <- neighs[k] * 3
vpaddd ymm17, ymm17, ymm18 #69.36
# LOE rax rcx rbx rbp rsi rdi r8 r10 r13 r15 r9d r11d r12d r14d xmm6 xmm7 xmm12 ymm2 ymm3 ymm15 ymm16 ymm17 zmm0 zmm1 zmm4 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14 k3
..B1.21: # Preds ..B1.18
# Execution count [1.25e+01]
# k1 <- k3
kmovw k1, k3 #69.36
# k2 <- k3
kmovw k2, k3 #69.36
# zmm18 <- 0.0
vpxord zmm18, zmm18, zmm18 #69.36
# zmm19 <- 0.0
vpxord zmm19, zmm19, zmm19 #69.36
# zmm20 <- 0.0
vpxord zmm20, zmm20, zmm20 #69.36
# zmm18 <- atom->x[j * 3 + 2]
vgatherdpd zmm18{k1}, QWORD PTR [16+rdi+ymm17*8] #69.36
# zmm19 <- atom->x[j * 3 + 1]
vgatherdpd zmm19{k2}, QWORD PTR [8+rdi+ymm17*8] #69.36
# zmm20 <- atom->x[j * 3]
vgatherdpd zmm20{k3}, QWORD PTR [rdi+ymm17*8] #69.36
# LOE rax rcx rbx rbp rsi rdi r8 r10 r13 r15 r9d r11d r12d r14d xmm6 xmm7 xmm12 ymm2 ymm3 ymm15 ymm16 zmm0 zmm1 zmm4 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14 zmm18 zmm19 zmm20
..B1.22: # Preds ..B1.21
# Execution count [2.50e+01]
# k++ (+= 8)
add r13, 8 #67.9
# ymm3 <- [0..7] + 8
vpaddd ymm3, ymm3, ymm16 #67.9
# zmm29 <- atom->x[i * 3 + 2] - atom->x[j * 3 + 2]
vsubpd zmm29, zmm4, zmm18 #71.36
# zmm27 <- atom->x[i * 3 + 1] - atom->x[j * 3 + 1]
vsubpd zmm27, zmm0, zmm19 #70.36
# zmm26 <- atom->x[i * 3] - atom->x[j * 3]
vsubpd zmm26, zmm1, zmm20 #69.36
# zmm25 <- dely * dely
vmulpd zmm25, zmm27, zmm27 #72.49
# zmm25 <- delx * delx + dely * dely
vfmadd231pd zmm25, zmm26, zmm26 #72.49
# zmm25 <- rsq (delz * delz + delx * delx + dely * dely)
vfmadd231pd zmm25, zmm29, zmm29 #72.63
# zmm24 <- sr2 (1.0 / rsq -- compute reciprocal)
vrcp14pd zmm24, zmm25 #75.38
# k2 <- [rsq < cutforcesq]
vcmppd k2, zmm25, zmm14, 1 #74.22
# k0 <- [true when +0, Neg. 0, +inf, -inf]
vfpclasspd k0, zmm24, 30 #75.38
# edx <- k2
kmovw edx, k2 #74.22
# k1 <- [false when +0, Neg. 0, +inf, -inf]
knotw k1, k0 #75.38
# zmm17 <- rsq
vmovaps zmm17, zmm25 #75.38
# r9d <- [true if k < numneighs and rsq < cutforcesq]
and r9d, edx #74.22
# zmm17 <- [-(rsq * sr2) + 1.0] -- check if 1.0 / rsq is valid! -- call it error?
vfnmadd213pd zmm17, zmm24, QWORD BCST .L_2il0floatpacket.9[rip] #75.38
# k3 <- [true if k < numneighs and rsq < cutforcesq]
kmovw k3, r9d #78.17
# zmm18 <- [(-rsq * sr2 + 1.0) * (-rsq * sr2 + 1.0)] -- errorsq
vmulpd zmm18, zmm17, zmm17 #75.38
# zmm24 <- [sr2 * error + sr2]
vfmadd213pd zmm24{k1}, zmm17, zmm24 #75.38
# zmm24 <- [(sr2 * error + sr2) * errorsq + (sr2 * error + sr2)]
vfmadd213pd zmm24{k1}, zmm18, zmm24 #75.38
# zmm19 <- sr2 * sigma6
vmulpd zmm19, zmm24, zmm13 #76.38
# zmm21 <- sr2 * epsilon
vmulpd zmm21, zmm24, zmm10 #77.54
# zmm22 <- sr2 * sr2 * sigma6
vmulpd zmm22, zmm24, zmm19 #76.44
# zmm20 <- sr6 (sr2 * sr2 * sr2 * sigma6)
vmulpd zmm20, zmm24, zmm22 #76.50
# zmm24 <- sr2 * (sr2 * sr2 * sigma) - 0.5 (why not vsubpd zmm24, zmm20, zmm5 ?)
vfmsub213pd zmm24, zmm22, zmm5 #77.54
# zmm23 <- (sr6) * (sr2 * epsilon)
vmulpd zmm23, zmm20, zmm21 #77.61
# zmm28 <- force -- (sr6 * sr2 * epsilon) * (sr6 - 0.5)
vmulpd zmm28, zmm23, zmm24 #77.67
# zmm9 <- force * delx + fix
vfmadd231pd zmm9{k3}, zmm28, zmm26 #78.17
# zmm8 <- force * dely + fiy
vfmadd231pd zmm8{k3}, zmm28, zmm27 #79.17
# zmm11 <- force * delz + fiz
vfmadd231pd zmm11{k3}, zmm28, zmm29 #80.17
# k < numneighs
cmp r13, rbx #67.9
jb ..B1.18 # Prob 82% #67.9
# LOE rax rcx rbx rbp rsi rdi r8 r10 r13 r15 r11d r12d r14d xmm6 xmm7 xmm12 ymm2 ymm3 ymm15 ymm16 zmm0 zmm1 zmm4 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.23: # Preds ..B1.22
# Execution count [4.50e+00]
# r9 <- atom->fz
mov r9, QWORD PTR [-80+rsp] #[spill]
# rdx <- atom->fy
mov rdx, QWORD PTR [-72+rsp] #[spill]
# numneighs == amount_of_neighs
cmp r11d, r12d #67.9
je ..B1.38 # Prob 10% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d r12d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.24: # Preds ..B1.23 ..B1.16 ..B1.43
# Execution count [2.50e+01]
# COND: neighbor->numneigh[i] < 1200
# ebx <- k + 8
lea ebx, DWORD PTR [8+r12] #67.9
# k - (k % 8) < k + 8 TODO: this may be wrong, but probably checks if the number of remaining neighbors is less than 8
cmp r14d, ebx #67.9
jl ..B1.32 # Prob 50% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d r12d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.25: # Preds ..B1.24
# Execution count [4.50e+00]
# r13 <- neighbor->maxneighs << 2
mov r13, r8 #56.43
# r13 <- neighbor->maxneighs * 4 * i
imul r13, rcx #56.43
# zmm2 <- atom->x[i * 3]
vbroadcastsd zmm2, xmm6 #58.23
# zmm1 <- atom->x[i * 3 + 1]
vbroadcastsd zmm1, xmm7 #59.23
# zmm0 <- atom->x[i * 3 + 2]
vbroadcastsd zmm0, xmm12 #60.23
# rbx <- r12d (k? amount of neighs?)
movsxd rbx, r12d #67.9
# r13 <- neighs (neighbor->neighbors + neighbor->maxneighs * 4 * i)
add r13, r10 #37.5
mov QWORD PTR [-64+rsp], rax #37.5[spill]
mov QWORD PTR [-56+rsp], r8 #37.5[spill]
mov QWORD PTR [-48+rsp], r10 #37.5[spill]
mov QWORD PTR [-40+rsp], rsi #37.5[spill]
mov QWORD PTR [-32+rsp], rcx #37.5[spill]
mov QWORD PTR [-80+rsp], r9 #37.5[spill]
mov QWORD PTR [-72+rsp], rdx #37.5[spill]
# LOE rbx rbp rdi r13 r11d r12d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm0 zmm1 zmm2 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.26: # Preds ..B1.30 ..B1.25
# Execution count [2.50e+01]
# ymm3 <- neighs[k]
vmovdqu ymm3, YMMWORD PTR [r13+rbx*4] #68.21
# ymm4 <- neighs[k] * 2
vpaddd ymm4, ymm3, ymm3 #69.36
# ymm3 <- neighs[k] * 3
vpaddd ymm3, ymm3, ymm4 #69.36
# Prefetching Instructions????????????
# r10d <- neighs[k]
mov r10d, DWORD PTR [r13+rbx*4] #68.21
# r9d <- neighs[k + 1]
mov r9d, DWORD PTR [4+r13+rbx*4] #68.21
# r8d <- neighs[k + 2]
mov r8d, DWORD PTR [8+r13+rbx*4] #68.21
# esi <- neighs[k + 3]
mov esi, DWORD PTR [12+r13+rbx*4] #68.21
# r10d <- neighs[k] * 3
lea r10d, DWORD PTR [r10+r10*2] #69.36
# ecx <- neighs[k + 4]
mov ecx, DWORD PTR [16+r13+rbx*4] #68.21
# r9d <- neighs[k + 1] * 3
lea r9d, DWORD PTR [r9+r9*2] #69.36
# edx <- neighs[k + 5]
mov edx, DWORD PTR [20+r13+rbx*4] #68.21
# r8d <- neighs[k + 2] * 3
lea r8d, DWORD PTR [r8+r8*2] #69.36
# edx <- neighs[k + 6]
mov eax, DWORD PTR [24+r13+rbx*4] #68.21
# esi <- neighs[k + 3] * 3
lea esi, DWORD PTR [rsi+rsi*2] #69.36
# edx <- neighs[k + 7]
mov r15d, DWORD PTR [28+r13+rbx*4] #68.21
# ecx <- neighs[k + 4] * 3
lea ecx, DWORD PTR [rcx+rcx*2] #69.36
# ecx <- neighs[k + 5] * 3
lea edx, DWORD PTR [rdx+rdx*2] #69.36
# ecx <- neighs[k + 6] * 3
lea eax, DWORD PTR [rax+rax*2] #69.36
# ecx <- neighs[k + 7] * 3
lea r15d, DWORD PTR [r15+r15*2] #69.36
# LOE rbx rbp rdi r13 eax edx ecx esi r8d r9d r10d r11d r12d r14d r15d xmm6 xmm7 xmm12 ymm3 ymm15 ymm16 zmm0 zmm1 zmm2 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.29: # Preds ..B1.26
# Execution count [1.25e+01]
# k1 <- [true for all elements]
vpcmpeqb k1, xmm0, xmm0 #69.36
# k2 <- [true for all elements]
vpcmpeqb k2, xmm0, xmm0 #69.36
# k3 <- [true for all elements]
vpcmpeqb k3, xmm0, xmm0 #69.36
# zmm4 <- 0.0
vpxord zmm4, zmm4, zmm4 #69.36
# zmm17 <- 0.0
vpxord zmm17, zmm17, zmm17 #69.36
# zmm18 <- 0.0
vpxord zmm18, zmm18, zmm18 #69.36
# zmm4 <- atom->x[j * 3 + 2]
vgatherdpd zmm4{k1}, QWORD PTR [16+rdi+ymm3*8] #69.36
# zmm17 <- atom->x[j * 3 + 1]
vgatherdpd zmm17{k2}, QWORD PTR [8+rdi+ymm3*8] #69.36
# zmm18 <- atom->x[j * 3]
vgatherdpd zmm18{k3}, QWORD PTR [rdi+ymm3*8] #69.36
# LOE rbx rbp rdi r13 r11d r12d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm0 zmm1 zmm2 zmm4 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14 zmm17 zmm18
..B1.30: # Preds ..B1.29
# Execution count [2.50e+01]
# k++
add r12d, 8 #67.9
# k++
add rbx, 8 #67.9
# zmm26 <- atom->x[i * 3 + 2] - atom->x[j * 3 + 2]
vsubpd zmm26, zmm0, zmm4 #71.36
# zmm24 <- atom->x[i * 3 + 1] - atom->x[j * 3 + 1]
vsubpd zmm24, zmm1, zmm17 #70.36
# zmm23 <- atom->x[i * 3] - atom->x[j * 3]
vsubpd zmm23, zmm2, zmm18 #69.36
# zmm3 <- dely * dely
vmulpd zmm3, zmm24, zmm24 #72.49
# zmm3 <- delx * delx + dely * dely
vfmadd231pd zmm3, zmm23, zmm23 #72.49
# zmm3 <- rsq (delz * delz + delx * delx + dely * dely)
vfmadd231pd zmm3, zmm26, zmm26 #72.63
# zmm22 <- sr2 (1.0 / rsq -- compute reciprocal)
vrcp14pd zmm22, zmm3 #75.38
# k2 <- [rsq < cutforcesq]
vcmppd k2, zmm3, zmm14, 1 #74.22
# k0 <- [true when +0, Neg. 0, +inf, -inf]
vfpclasspd k0, zmm22, 30 #75.38
# zmm3 <- [-(rsq * sr2) + 1.0] -- check if 1.0 / rsq is valid! -- call it error?
vfnmadd213pd zmm3, zmm22, QWORD BCST .L_2il0floatpacket.9[rip] #75.38
# k1 <- [false when +0, Neg. 0, +inf, -inf]
knotw k1, k0 #75.38
# zmm4 <- [(-rsq * sr2 + 1.0) * (-rsq * sr2 + 1.0)] -- errorsq
vmulpd zmm4, zmm3, zmm3 #75.38
# zmm22 <- [sr2 * error + sr2]
vfmadd213pd zmm22{k1}, zmm3, zmm22 #75.38
# zmm22 <- [(sr2 * error + sr2) * errorsq + (sr2 * error + sr2)]
vfmadd213pd zmm22{k1}, zmm4, zmm22 #75.38
vmulpd zmm17, zmm22, zmm13 #76.38
vmulpd zmm19, zmm22, zmm10 #77.54
vmulpd zmm20, zmm22, zmm17 #76.44
vmulpd zmm18, zmm22, zmm20 #76.50
vfmsub213pd zmm22, zmm20, zmm5 #77.54
vmulpd zmm21, zmm18, zmm19 #77.61
vmulpd zmm25, zmm21, zmm22 #77.67
vfmadd231pd zmm9{k2}, zmm25, zmm23 #78.17
vfmadd231pd zmm8{k2}, zmm25, zmm24 #79.17
vfmadd231pd zmm11{k2}, zmm25, zmm26 #80.17
# k < numneighs
cmp r12d, r14d #67.9
jb ..B1.26 # Prob 82% #67.9
# LOE rbx rbp rdi r13 r11d r12d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm0 zmm1 zmm2 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.31: # Preds ..B1.30
# Execution count [4.50e+00]
mov rax, QWORD PTR [-64+rsp] #[spill]
mov r8, QWORD PTR [-56+rsp] #[spill]
mov r10, QWORD PTR [-48+rsp] #[spill]
mov rsi, QWORD PTR [-40+rsp] #[spill]
mov rcx, QWORD PTR [-32+rsp] #[spill]
mov r9, QWORD PTR [-80+rsp] #[spill]
mov rdx, QWORD PTR [-72+rsp] #[spill]
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.32: # Preds ..B1.31 ..B1.24 ..B1.44
# Execution count [5.00e+00]
# COND: neighbor->numneigh[i] < 8
lea ebx, DWORD PTR [1+r14] #67.9
cmp ebx, r11d #67.9
ja ..B1.38 # Prob 50% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.33: # Preds ..B1.32
# Execution count [2.50e+01]
imul rcx, r8 #56.43
vbroadcastsd zmm4, xmm6 #58.23
sub r11d, r14d #67.9
add rcx, r10 #37.5
vpbroadcastd ymm0, r11d #67.9
vpcmpgtd k3, ymm0, ymm15 #67.9
movsxd r14, r14d #67.9
kmovw ebx, k3 #67.9
vmovdqu32 ymm1{k3}{z}, YMMWORD PTR [rcx+r14*4] #68.21
vpaddd ymm2, ymm1, ymm1 #69.36
vpaddd ymm0, ymm1, ymm2 #69.36
# LOE rax rdx rbp rsi rdi r8 r9 r10 ebx xmm7 xmm12 ymm0 ymm15 ymm16 zmm4 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14 k3
..B1.36: # Preds ..B1.33
# Execution count [1.25e+01]
kmovw k1, k3 #69.36
kmovw k2, k3 #69.36
vpxord zmm1, zmm1, zmm1 #69.36
vpxord zmm2, zmm2, zmm2 #69.36
vpxord zmm3, zmm3, zmm3 #69.36
vgatherdpd zmm1{k1}, QWORD PTR [16+rdi+ymm0*8] #69.36
vgatherdpd zmm2{k2}, QWORD PTR [8+rdi+ymm0*8] #69.36
vgatherdpd zmm3{k3}, QWORD PTR [rdi+ymm0*8] #69.36
# LOE rax rdx rbp rsi rdi r8 r9 r10 ebx xmm7 xmm12 ymm15 ymm16 zmm1 zmm2 zmm3 zmm4 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.37: # Preds ..B1.36
# Execution count [2.50e+01]
vbroadcastsd zmm7, xmm7 #59.23
vbroadcastsd zmm12, xmm12 #60.23
vsubpd zmm23, zmm12, zmm1 #71.36
vsubpd zmm21, zmm7, zmm2 #70.36
vsubpd zmm20, zmm4, zmm3 #69.36
vmulpd zmm19, zmm21, zmm21 #72.49
vfmadd231pd zmm19, zmm20, zmm20 #72.49
vfmadd231pd zmm19, zmm23, zmm23 #72.63
vrcp14pd zmm18, zmm19 #75.38
vcmppd k2, zmm19, zmm14, 1 #74.22
vfpclasspd k0, zmm18, 30 #75.38
kmovw ecx, k2 #74.22
knotw k1, k0 #75.38
vmovaps zmm0, zmm19 #75.38
and ebx, ecx #74.22
vfnmadd213pd zmm0, zmm18, QWORD BCST .L_2il0floatpacket.9[rip] #75.38
kmovw k3, ebx #78.17
vmulpd zmm1, zmm0, zmm0 #75.38
vfmadd213pd zmm18{k1}, zmm0, zmm18 #75.38
vfmadd213pd zmm18{k1}, zmm1, zmm18 #75.38
vmulpd zmm2, zmm18, zmm13 #76.38
vmulpd zmm4, zmm18, zmm10 #77.54
vmulpd zmm6, zmm18, zmm2 #76.44
vmulpd zmm3, zmm18, zmm6 #76.50
vfmsub213pd zmm18, zmm6, zmm5 #77.54
vmulpd zmm17, zmm3, zmm4 #77.61
vmulpd zmm22, zmm17, zmm18 #77.67
vfmadd231pd zmm9{k3}, zmm22, zmm20 #78.17
vfmadd231pd zmm8{k3}, zmm22, zmm21 #79.17
vfmadd231pd zmm11{k3}, zmm22, zmm23 #80.17
# LOE rax rdx rbp rsi rdi r8 r9 r10 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.38: # Preds ..B1.23 ..B1.37 ..B1.32
# Execution count [4.50e+00]
vmovups zmm19, ZMMWORD PTR .L_2il0floatpacket.10[rip] #63.22
vpermd zmm0, zmm19, zmm11 #63.22
vpermd zmm6, zmm19, zmm8 #62.22
vpermd zmm20, zmm19, zmm9 #61.22
vaddpd zmm11, zmm0, zmm11 #63.22
vaddpd zmm8, zmm6, zmm8 #62.22
vaddpd zmm9, zmm20, zmm9 #61.22
vpermpd zmm1, zmm11, 78 #63.22
vpermpd zmm7, zmm8, 78 #62.22
vpermpd zmm21, zmm9, 78 #61.22
vaddpd zmm2, zmm11, zmm1 #63.22
vaddpd zmm12, zmm8, zmm7 #62.22
vaddpd zmm22, zmm9, zmm21 #61.22
vpermpd zmm3, zmm2, 177 #63.22
vpermpd zmm17, zmm12, 177 #62.22
vpermpd zmm23, zmm22, 177 #61.22
vaddpd zmm4, zmm2, zmm3 #63.22
vaddpd zmm18, zmm12, zmm17 #62.22
vaddpd zmm24, zmm22, zmm23 #61.22
# LOE rax rdx rbp rsi rdi r8 r9 r10 xmm4 xmm18 xmm24 ymm15 ymm16 zmm5 zmm10 zmm13 zmm14
..B1.39: # Preds ..B1.38 ..B1.9
# Execution count [5.00e+00]
mov rcx, QWORD PTR [-8+rsp] #84.9[spill]
add rax, 24 #55.5
vaddsd xmm0, xmm24, QWORD PTR [rcx+rsi*8] #84.9
vmovsd QWORD PTR [rcx+rsi*8], xmm0 #84.9
movsxd rcx, esi #55.32
vaddsd xmm1, xmm18, QWORD PTR [rdx+rsi*8] #85.9
vmovsd QWORD PTR [rdx+rsi*8], xmm1 #85.9
inc rcx #55.32
vaddsd xmm2, xmm4, QWORD PTR [r9+rsi*8] #86.9
vmovsd QWORD PTR [r9+rsi*8], xmm2 #86.9
inc rsi #55.5
cmp rsi, QWORD PTR [-24+rsp] #55.5[spill]
jb ..B1.9 # Prob 82% #55.5
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 ymm15 ymm16 zmm5 zmm10 zmm13 zmm14
..B1.40: # Preds ..B1.39
# Execution count [9.00e-01]
mov r14, QWORD PTR [-88+rsp] #[spill]
.cfi_restore 14
mov r15, QWORD PTR [-96+rsp] #[spill]
.cfi_restore 15
mov rbx, QWORD PTR [-104+rsp] #[spill]
.cfi_restore 3
# LOE rbx rbp r14 r15
..B1.42: # Preds ..B1.1 ..B1.40
# Execution count [1.00e+00]
vzeroupper #93.12
vxorpd xmm0, xmm0, xmm0 #93.12
.cfi_restore 13
pop r13 #93.12
.cfi_def_cfa_offset 16
.cfi_restore 12
pop r12 #93.12
.cfi_def_cfa_offset 8
ret #93.12
.cfi_def_cfa_offset 24
.cfi_offset 3, -128
.cfi_offset 12, -16
.cfi_offset 13, -24
.cfi_offset 14, -112
.cfi_offset 15, -120
# LOE
..B1.43: # Preds ..B1.11
# Execution count [4.50e-01]: Infreq
# COND: neighbor->numneigh[i] < 1200
# r14d <- numneighs
mov r14d, r11d #67.9
# r12d <- 0 (k)
xor r12d, r12d #67.9
# r14d <- numneighs & -8 (0b1111...111000), 32bits -- k - (k % 8)
and r14d, -8 #67.9
jmp ..B1.24 # Prob 100% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d r12d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.44: # Preds ..B1.10
# Execution count [4.50e-01]: Infreq
# COND: neighbor->numneigh[i] < 8
xor r14d, r14d #67.9
jmp ..B1.32 # Prob 100% #67.9
.align 16,0x90
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
.cfi_endproc
# mark_end;
.type computeForce,@function
.size computeForce,.-computeForce
..LNcomputeForce.0:
.data
# -- End computeForce
.section .rodata, "a"
.align 64
.align 64
.L_2il0floatpacket.2:
.long 0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000
.type .L_2il0floatpacket.2,@object
.size .L_2il0floatpacket.2,64
.align 64
.L_2il0floatpacket.4:
.long 0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000
.type .L_2il0floatpacket.4,@object
.size .L_2il0floatpacket.4,64
.align 64
.L_2il0floatpacket.5:
.long 0x02010101,0x04040202,0x08080804,0x20101010,0x40402020,0x80808040,0x00000000,0x00000000,0x00000000,0x00000000,0x00000000,0x00000000,0x00000000,0x00000000,0x00000000,0x00000000
.type .L_2il0floatpacket.5,@object
.size .L_2il0floatpacket.5,64
.align 64
.L_2il0floatpacket.6:
.long 0x00000000,0x00000000,0x00000004,0x00000000,0x00000008,0x00000000,0x0000000c,0x00000000,0x00000001,0x00000000,0x00000005,0x00000000,0x00000009,0x00000000,0x0000000d,0x00000000
.type .L_2il0floatpacket.6,@object
.size .L_2il0floatpacket.6,64
.align 64
.L_2il0floatpacket.7:
.long 0x00000001,0x00000000,0x00000005,0x00000000,0x00000009,0x00000000,0x0000000d,0x00000000,0x00000000,0x00000000,0x00000004,0x00000000,0x00000008,0x00000000,0x0000000c,0x00000000
.type .L_2il0floatpacket.7,@object
.size .L_2il0floatpacket.7,64
.align 64
.L_2il0floatpacket.8:
.long 0x00000002,0x00000000,0x00000006,0x00000000,0x0000000a,0x00000000,0x0000000e,0x00000000,0x00000002,0x00000000,0x00000006,0x00000000,0x0000000a,0x00000000,0x0000000e,0x00000000
.type .L_2il0floatpacket.8,@object
.size .L_2il0floatpacket.8,64
.align 64
.L_2il0floatpacket.10:
.long 0x00000008,0x00000009,0x0000000a,0x0000000b,0x0000000c,0x0000000d,0x0000000e,0x0000000f,0x00000008,0x00000009,0x0000000a,0x0000000b,0x0000000c,0x0000000d,0x0000000e,0x0000000f
.type .L_2il0floatpacket.10,@object
.size .L_2il0floatpacket.10,64
.align 32
.L_2il0floatpacket.0:
.long 0x00000008,0x00000008,0x00000008,0x00000008,0x00000008,0x00000008,0x00000008,0x00000008
.type .L_2il0floatpacket.0,@object
.size .L_2il0floatpacket.0,32
.align 32
.L_2il0floatpacket.1:
.long 0x00000000,0x00000001,0x00000002,0x00000003,0x00000004,0x00000005,0x00000006,0x00000007
.type .L_2il0floatpacket.1,@object
.size .L_2il0floatpacket.1,32
.align 8
.L_2il0floatpacket.3:
.long 0x00000000,0x40480000
.type .L_2il0floatpacket.3,@object
.size .L_2il0floatpacket.3,8
.align 8
.L_2il0floatpacket.9:
.long 0x00000000,0x3ff00000
.type .L_2il0floatpacket.9,@object
.size .L_2il0floatpacket.9,8
.data
.section .note.GNU-stack, ""
# End

563
asm/avx512/force_aos_lt1200_markers.s
View File

@ -1,563 +0,0 @@
# mark_description "Intel(R) C Intel(R) 64 Compiler for applications running on Intel(R) 64, Version 19.0.5.281 Build 20190815";
# mark_description "-I./src/includes -S -D_GNU_SOURCE -DAOS -DPRECISION=2 -DALIGNMENT=64 -restrict -Ofast -xCORE-AVX512 -qopt-zm";
# mark_description "m-usage=high -o ICC/force.s";
.file "force.c"
.text
..TXTST0:
.L_2__routine_start_computeForce_0:
# -- Begin computeForce
.text
# mark_begin;
.align 16,0x90
.globl computeForce
# --- computeForce(Parameter *, Atom *, Neighbor *, int)
computeForce:
# parameter 1: %rdi
# parameter 2: %rsi
# parameter 3: %rdx
# parameter 4: %ecx
..B1.1: # Preds ..B1.0
# Execution count [1.00e+00]
.cfi_startproc
..___tag_value_computeForce.1:
..L2:
#35.1
pushq %r12 #35.1
.cfi_def_cfa_offset 16
.cfi_offset 12, -16
pushq %r13 #35.1
.cfi_def_cfa_offset 24
.cfi_offset 13, -24
movq %rsi, %r8 #35.1
vmovsd 72(%rdi), %xmm2 #38.27
movq %rdx, %rax #35.1
vmovsd 8(%rdi), %xmm1 #39.23
vmovsd (%rdi), %xmm0 #40.24
movl 4(%r8), %r12d #36.18
movq 64(%r8), %r11 #41.20
movq 72(%r8), %rdx #41.45
movq 80(%r8), %r9 #41.70
testl %r12d, %r12d #44.24
jle ..B1.42 # Prob 50% #44.24
# LOE rax rdx rbx rbp r8 r9 r11 r14 r15 r12d xmm0 xmm1 xmm2
..B1.2: # Preds ..B1.1
# Execution count [1.00e+00]
movl %r12d, %edi #44.5
xorl %esi, %esi #44.5
movl $1, %r10d #44.5
xorl %ecx, %ecx #44.5
shrl $1, %edi #44.5
je ..B1.6 # Prob 9% #44.5
# LOE rax rdx rcx rbx rbp rsi rdi r8 r9 r11 r14 r15 r10d r12d xmm0 xmm1 xmm2
..B1.3: # Preds ..B1.2
# Execution count [9.00e-01]
xorl %r10d, %r10d #44.5
# LOE rax rdx rcx rbx rbp rsi rdi r8 r9 r10 r11 r14 r15 r12d xmm0 xmm1 xmm2
..B1.4: # Preds ..B1.4 ..B1.3
# Execution count [2.50e+00]
movq %r10, (%rcx,%r11) #45.9
incq %rsi #44.5
movq %r10, (%rcx,%rdx) #46.9
movq %r10, (%rcx,%r9) #47.9
movq %r10, 8(%rcx,%r11) #45.9
movq %r10, 8(%rcx,%rdx) #46.9
movq %r10, 8(%rcx,%r9) #47.9
addq $16, %rcx #44.5
cmpq %rdi, %rsi #44.5
jb ..B1.4 # Prob 63% #44.5
# LOE rax rdx rcx rbx rbp rsi rdi r8 r9 r10 r11 r14 r15 r12d xmm0 xmm1 xmm2
..B1.5: # Preds ..B1.4
# Execution count [9.00e-01]
lea 1(%rsi,%rsi), %r10d #45.9
# LOE rax rdx rbx rbp r8 r9 r11 r14 r15 r10d r12d xmm0 xmm1 xmm2
..B1.6: # Preds ..B1.2 ..B1.5
# Execution count [1.00e+00]
lea -1(%r10), %ecx #44.5
cmpl %r12d, %ecx #44.5
jae ..B1.8 # Prob 9% #44.5
# LOE rax rdx rbx rbp r8 r9 r11 r14 r15 r10d r12d xmm0 xmm1 xmm2
..B1.7: # Preds ..B1.6
# Execution count [9.00e-01]
movslq %r10d, %r10 #44.5
xorl %ecx, %ecx #45.9
movq %rcx, -8(%r11,%r10,8) #45.9
movq %rcx, -8(%rdx,%r10,8) #46.9
movq %rcx, -8(%r9,%r10,8) #47.9
# LOE rax rdx rbx rbp r8 r9 r11 r14 r15 r12d xmm0 xmm1 xmm2
..B1.8: # Preds ..B1.6 ..B1.7
# Execution count [9.00e-01]
vmulsd %xmm2, %xmm2, %xmm13 #38.45
xorl %ecx, %ecx #55.15
vmovdqu32 .L_2il0floatpacket.0(%rip), %ymm16 #67.9
vmulsd .L_2il0floatpacket.3(%rip), %xmm0, %xmm0 #77.41
vmovdqu .L_2il0floatpacket.1(%rip), %ymm15 #67.9
vmovups .L_2il0floatpacket.4(%rip), %zmm5 #77.54
vbroadcastsd %xmm13, %zmm14 #38.25
vbroadcastsd %xmm1, %zmm13 #39.21
vbroadcastsd %xmm0, %zmm10 #77.41
movslq %r12d, %r12 #55.5
xorl %esi, %esi #55.5
movq 24(%rax), %r13 #57.25
movq 16(%r8), %rdi #58.25
movslq 16(%rax), %r8 #56.43
movq 8(%rax), %r10 #56.19
xorl %eax, %eax #55.5
shlq $2, %r8 #37.5
movq %r12, -24(%rsp) #55.5[spill]
movq %r13, -16(%rsp) #55.5[spill]
movq %r11, -8(%rsp) #55.5[spill]
movq %r14, -88(%rsp) #55.5[spill]
movq %r15, -96(%rsp) #55.5[spill]
movq %rbx, -104(%rsp) #55.5[spill]
.cfi_offset 3, -128
.cfi_offset 14, -112
.cfi_offset 15, -120
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 ymm15 ymm16 zmm5 zmm10 zmm13 zmm14
..B1.9: # Preds ..B1.39 ..B1.8
# Execution count [5.00e+00]
movq -16(%rsp), %rbx #57.25[spill]
vxorpd %xmm24, %xmm24, %xmm24 #61.22
vmovapd %xmm24, %xmm18 #62.22
movl (%rbx,%rsi,4), %r11d #57.25
vmovapd %xmm18, %xmm4 #63.22
vmovsd (%rax,%rdi), %xmm6 #58.25
vmovsd 8(%rax,%rdi), %xmm7 #59.25
vmovsd 16(%rax,%rdi), %xmm12 #60.25
testl %r11d, %r11d #67.28
jle ..B1.39 # Prob 50% #67.28
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d xmm4 xmm6 xmm7 xmm12 xmm18 xmm24 ymm15 ymm16 zmm5 zmm10 zmm13 zmm14
..B1.10: # Preds ..B1.9
# Execution count [4.50e+00]
vpxord %zmm9, %zmm9, %zmm9 #61.22
vmovaps %zmm9, %zmm8 #62.22
vmovaps %zmm8, %zmm11 #63.22
cmpl $8, %r11d #67.9
jl ..B1.44 # Prob 10% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.11: # Preds ..B1.10
# Execution count [4.50e+00]
cmpl $1200, %r11d #67.9
jl ..B1.43 # Prob 10% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.12: # Preds ..B1.11
# Execution count [4.50e+00]
movq %r8, %r15 #56.43
imulq %rcx, %r15 #56.43
addq %r10, %r15 #37.5
movq %r15, %r12 #67.9
andq $63, %r12 #67.9
testl $3, %r12d #67.9
je ..B1.14 # Prob 50% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r15 r11d r12d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.13: # Preds ..B1.12
# Execution count [2.25e+00]
xorl %r12d, %r12d #67.9
jmp ..B1.16 # Prob 100% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r15 r11d r12d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.14: # Preds ..B1.12
# Execution count [2.25e+00]
testl %r12d, %r12d #67.9
je ..B1.16 # Prob 50% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r15 r11d r12d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.15: # Preds ..B1.14
# Execution count [2.50e+01]
negl %r12d #67.9
addl $64, %r12d #67.9
shrl $2, %r12d #67.9
cmpl %r12d, %r11d #67.9
cmovl %r11d, %r12d #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r15 r11d r12d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.16: # Preds ..B1.13 ..B1.15 ..B1.14
# Execution count [5.00e+00]
movl %r11d, %r14d #67.9
subl %r12d, %r14d #67.9
andl $7, %r14d #67.9
negl %r14d #67.9
addl %r11d, %r14d #67.9
cmpl $1, %r12d #67.9
jb ..B1.24 # Prob 50% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r15 r11d r12d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.17: # Preds ..B1.16
# Execution count [4.50e+00]
vmovdqa %ymm15, %ymm3 #67.9
xorl %r13d, %r13d #67.9
vpbroadcastd %r12d, %ymm2 #67.9
vbroadcastsd %xmm6, %zmm1 #58.23
vbroadcastsd %xmm7, %zmm0 #59.23
vbroadcastsd %xmm12, %zmm4 #60.23
movslq %r12d, %rbx #67.9
movq %r9, -80(%rsp) #67.9[spill]
movq %rdx, -72(%rsp) #67.9[spill]
# LOE rax rcx rbx rbp rsi rdi r8 r10 r13 r15 r11d r12d r14d xmm6 xmm7 xmm12 ymm2 ymm3 ymm15 ymm16 zmm0 zmm1 zmm4 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.18: # Preds ..B1.22 ..B1.17
# Execution count [2.50e+01]
vpcmpgtd %ymm3, %ymm2, %k3 #67.9
vmovdqu32 (%r15,%r13,4), %ymm17{%k3}{z} #68.21
kmovw %k3, %r9d #67.9
vpaddd %ymm17, %ymm17, %ymm18 #69.36
vpaddd %ymm18, %ymm17, %ymm17 #69.36
# LOE rax rcx rbx rbp rsi rdi r8 r10 r13 r15 r9d r11d r12d r14d xmm6 xmm7 xmm12 ymm2 ymm3 ymm15 ymm16 ymm17 zmm0 zmm1 zmm4 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14 k3
..B1.21: # Preds ..B1.18
# Execution count [1.25e+01]
kmovw %k3, %k1 #69.36
kmovw %k3, %k2 #69.36
vpxord %zmm18, %zmm18, %zmm18 #69.36
vpxord %zmm19, %zmm19, %zmm19 #69.36
vpxord %zmm20, %zmm20, %zmm20 #69.36
vgatherdpd 16(%rdi,%ymm17,8), %zmm18{%k1} #69.36
vgatherdpd 8(%rdi,%ymm17,8), %zmm19{%k2} #69.36
vgatherdpd (%rdi,%ymm17,8), %zmm20{%k3} #69.36
# LOE rax rcx rbx rbp rsi rdi r8 r10 r13 r15 r9d r11d r12d r14d xmm6 xmm7 xmm12 ymm2 ymm3 ymm15 ymm16 zmm0 zmm1 zmm4 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14 zmm18 zmm19 zmm20
..B1.22: # Preds ..B1.21
# Execution count [2.50e+01]
addq $8, %r13 #67.9
vpaddd %ymm16, %ymm3, %ymm3 #67.9
vsubpd %zmm18, %zmm4, %zmm29 #71.36
vsubpd %zmm19, %zmm0, %zmm27 #70.36
vsubpd %zmm20, %zmm1, %zmm26 #69.36
vmulpd %zmm27, %zmm27, %zmm25 #72.49
vfmadd231pd %zmm26, %zmm26, %zmm25 #72.49
vfmadd231pd %zmm29, %zmm29, %zmm25 #72.63
vrcp14pd %zmm25, %zmm24 #75.38
vcmppd $1, %zmm14, %zmm25, %k2 #74.22
vfpclasspd $30, %zmm24, %k0 #75.38
kmovw %k2, %edx #74.22
knotw %k0, %k1 #75.38
vmovaps %zmm25, %zmm17 #75.38
andl %edx, %r9d #74.22
vfnmadd213pd .L_2il0floatpacket.9(%rip){1to8}, %zmm24, %zmm17 #75.38
kmovw %r9d, %k3 #78.17
vmulpd %zmm17, %zmm17, %zmm18 #75.38
vfmadd213pd %zmm24, %zmm17, %zmm24{%k1} #75.38
vfmadd213pd %zmm24, %zmm18, %zmm24{%k1} #75.38
vmulpd %zmm13, %zmm24, %zmm19 #76.38
vmulpd %zmm10, %zmm24, %zmm21 #77.54
vmulpd %zmm19, %zmm24, %zmm22 #76.44
vmulpd %zmm22, %zmm24, %zmm20 #76.50
vfmsub213pd %zmm5, %zmm22, %zmm24 #77.54
vmulpd %zmm21, %zmm20, %zmm23 #77.61
vmulpd %zmm24, %zmm23, %zmm28 #77.67
vfmadd231pd %zmm26, %zmm28, %zmm9{%k3} #78.17
vfmadd231pd %zmm27, %zmm28, %zmm8{%k3} #79.17
vfmadd231pd %zmm29, %zmm28, %zmm11{%k3} #80.17
cmpq %rbx, %r13 #67.9
jb ..B1.18 # Prob 82% #67.9
# LOE rax rcx rbx rbp rsi rdi r8 r10 r13 r15 r11d r12d r14d xmm6 xmm7 xmm12 ymm2 ymm3 ymm15 ymm16 zmm0 zmm1 zmm4 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.23: # Preds ..B1.22
# Execution count [4.50e+00]
movq -80(%rsp), %r9 #[spill]
movq -72(%rsp), %rdx #[spill]
cmpl %r12d, %r11d #67.9
je ..B1.38 # Prob 10% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d r12d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.24: # Preds ..B1.23 ..B1.16 ..B1.43
# Execution count [2.50e+01]
lea 8(%r12), %ebx #67.9
cmpl %ebx, %r14d #67.9
jl ..B1.32 # Prob 50% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d r12d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
movl $111,%ebx #IACA/OSACA START MARKER
.byte 100,103,144 #IACA/OSACA START MARKER
..B1.25: # Preds ..B1.24
# Execution count [4.50e+00]
movq %r8, %r13 #56.43
imulq %rcx, %r13 #56.43
vbroadcastsd %xmm6, %zmm2 #58.23
vbroadcastsd %xmm7, %zmm1 #59.23
vbroadcastsd %xmm12, %zmm0 #60.23
movslq %r12d, %rbx #67.9
addq %r10, %r13 #37.5
movq %rax, -64(%rsp) #37.5[spill]
movq %r8, -56(%rsp) #37.5[spill]
movq %r10, -48(%rsp) #37.5[spill]
movq %rsi, -40(%rsp) #37.5[spill]
movq %rcx, -32(%rsp) #37.5[spill]
movq %r9, -80(%rsp) #37.5[spill]
movq %rdx, -72(%rsp) #37.5[spill]
# LOE rbx rbp rdi r13 r11d r12d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm0 zmm1 zmm2 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.26: # Preds ..B1.30 ..B1.25
# Execution count [2.50e+01]
vmovdqu (%r13,%rbx,4), %ymm3 #68.21
vpaddd %ymm3, %ymm3, %ymm4 #69.36
vpaddd %ymm4, %ymm3, %ymm3 #69.36
movl (%r13,%rbx,4), %r10d #68.21
movl 4(%r13,%rbx,4), %r9d #68.21
movl 8(%r13,%rbx,4), %r8d #68.21
movl 12(%r13,%rbx,4), %esi #68.21
lea (%r10,%r10,2), %r10d #69.36
movl 16(%r13,%rbx,4), %ecx #68.21
lea (%r9,%r9,2), %r9d #69.36
movl 20(%r13,%rbx,4), %edx #68.21
lea (%r8,%r8,2), %r8d #69.36
movl 24(%r13,%rbx,4), %eax #68.21
lea (%rsi,%rsi,2), %esi #69.36
movl 28(%r13,%rbx,4), %r15d #68.21
lea (%rcx,%rcx,2), %ecx #69.36
lea (%rdx,%rdx,2), %edx #69.36
lea (%rax,%rax,2), %eax #69.36
lea (%r15,%r15,2), %r15d #69.36
# LOE rbx rbp rdi r13 eax edx ecx esi r8d r9d r10d r11d r12d r14d r15d xmm6 xmm7 xmm12 ymm3 ymm15 ymm16 zmm0 zmm1 zmm2 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.29: # Preds ..B1.26
# Execution count [1.25e+01]
vpcmpeqb %xmm0, %xmm0, %k1 #69.36
vpcmpeqb %xmm0, %xmm0, %k2 #69.36
vpcmpeqb %xmm0, %xmm0, %k3 #69.36
vpxord %zmm4, %zmm4, %zmm4 #69.36
vpxord %zmm17, %zmm17, %zmm17 #69.36
vpxord %zmm18, %zmm18, %zmm18 #69.36
vgatherdpd 16(%rdi,%ymm3,8), %zmm4{%k1} #69.36
vgatherdpd 8(%rdi,%ymm3,8), %zmm17{%k2} #69.36
vgatherdpd (%rdi,%ymm3,8), %zmm18{%k3} #69.36
# LOE rbx rbp rdi r13 r11d r12d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm0 zmm1 zmm2 zmm4 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14 zmm17 zmm18
..B1.30: # Preds ..B1.29
# Execution count [2.50e+01]
addl $8, %r12d #67.9
addq $8, %rbx #67.9
vsubpd %zmm4, %zmm0, %zmm26 #71.36
vsubpd %zmm17, %zmm1, %zmm24 #70.36
vsubpd %zmm18, %zmm2, %zmm23 #69.36
vmulpd %zmm24, %zmm24, %zmm3 #72.49
vfmadd231pd %zmm23, %zmm23, %zmm3 #72.49
vfmadd231pd %zmm26, %zmm26, %zmm3 #72.63
vrcp14pd %zmm3, %zmm22 #75.38
vcmppd $1, %zmm14, %zmm3, %k2 #74.22
vfpclasspd $30, %zmm22, %k0 #75.38
vfnmadd213pd .L_2il0floatpacket.9(%rip){1to8}, %zmm22, %zmm3 #75.38
knotw %k0, %k1 #75.38
vmulpd %zmm3, %zmm3, %zmm4 #75.38
vfmadd213pd %zmm22, %zmm3, %zmm22{%k1} #75.38
vfmadd213pd %zmm22, %zmm4, %zmm22{%k1} #75.38
vmulpd %zmm13, %zmm22, %zmm17 #76.38
vmulpd %zmm10, %zmm22, %zmm19 #77.54
vmulpd %zmm17, %zmm22, %zmm20 #76.44
vmulpd %zmm20, %zmm22, %zmm18 #76.50
vfmsub213pd %zmm5, %zmm20, %zmm22 #77.54
vmulpd %zmm19, %zmm18, %zmm21 #77.61
vmulpd %zmm22, %zmm21, %zmm25 #77.67
vfmadd231pd %zmm23, %zmm25, %zmm9{%k2} #78.17
vfmadd231pd %zmm24, %zmm25, %zmm8{%k2} #79.17
vfmadd231pd %zmm26, %zmm25, %zmm11{%k2} #80.17
cmpl %r14d, %r12d #67.9
jb ..B1.26 # Prob 82% #67.9
# LOE rbx rbp rdi r13 r11d r12d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm0 zmm1 zmm2 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
movl $222,%ebx #IACA/OSACA END MARKER
.byte 100,103,144 #IACA/OSACA END MARKER
..B1.31: # Preds ..B1.30
# Execution count [4.50e+00]
movq -64(%rsp), %rax #[spill]
movq -56(%rsp), %r8 #[spill]
movq -48(%rsp), %r10 #[spill]
movq -40(%rsp), %rsi #[spill]
movq -32(%rsp), %rcx #[spill]
movq -80(%rsp), %r9 #[spill]
movq -72(%rsp), %rdx #[spill]
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.32: # Preds ..B1.31 ..B1.24 ..B1.44
# Execution count [5.00e+00]
lea 1(%r14), %ebx #67.9
cmpl %r11d, %ebx #67.9
ja ..B1.38 # Prob 50% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.33: # Preds ..B1.32
# Execution count [2.50e+01]
imulq %r8, %rcx #56.43
vbroadcastsd %xmm6, %zmm4 #58.23
subl %r14d, %r11d #67.9
addq %r10, %rcx #37.5
vpbroadcastd %r11d, %ymm0 #67.9
vpcmpgtd %ymm15, %ymm0, %k3 #67.9
movslq %r14d, %r14 #67.9
kmovw %k3, %ebx #67.9
vmovdqu32 (%rcx,%r14,4), %ymm1{%k3}{z} #68.21
vpaddd %ymm1, %ymm1, %ymm2 #69.36
vpaddd %ymm2, %ymm1, %ymm0 #69.36
# LOE rax rdx rbp rsi rdi r8 r9 r10 ebx xmm7 xmm12 ymm0 ymm15 ymm16 zmm4 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14 k3
..B1.36: # Preds ..B1.33
# Execution count [1.25e+01]
kmovw %k3, %k1 #69.36
kmovw %k3, %k2 #69.36
vpxord %zmm1, %zmm1, %zmm1 #69.36
vpxord %zmm2, %zmm2, %zmm2 #69.36
vpxord %zmm3, %zmm3, %zmm3 #69.36
vgatherdpd 16(%rdi,%ymm0,8), %zmm1{%k1} #69.36
vgatherdpd 8(%rdi,%ymm0,8), %zmm2{%k2} #69.36
vgatherdpd (%rdi,%ymm0,8), %zmm3{%k3} #69.36
# LOE rax rdx rbp rsi rdi r8 r9 r10 ebx xmm7 xmm12 ymm15 ymm16 zmm1 zmm2 zmm3 zmm4 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.37: # Preds ..B1.36
# Execution count [2.50e+01]
vbroadcastsd %xmm7, %zmm7 #59.23
vbroadcastsd %xmm12, %zmm12 #60.23
vsubpd %zmm1, %zmm12, %zmm23 #71.36
vsubpd %zmm2, %zmm7, %zmm21 #70.36
vsubpd %zmm3, %zmm4, %zmm20 #69.36
vmulpd %zmm21, %zmm21, %zmm19 #72.49
vfmadd231pd %zmm20, %zmm20, %zmm19 #72.49
vfmadd231pd %zmm23, %zmm23, %zmm19 #72.63
vrcp14pd %zmm19, %zmm18 #75.38
vcmppd $1, %zmm14, %zmm19, %k2 #74.22
vfpclasspd $30, %zmm18, %k0 #75.38
kmovw %k2, %ecx #74.22
knotw %k0, %k1 #75.38
vmovaps %zmm19, %zmm0 #75.38
andl %ecx, %ebx #74.22
vfnmadd213pd .L_2il0floatpacket.9(%rip){1to8}, %zmm18, %zmm0 #75.38
kmovw %ebx, %k3 #78.17
vmulpd %zmm0, %zmm0, %zmm1 #75.38
vfmadd213pd %zmm18, %zmm0, %zmm18{%k1} #75.38
vfmadd213pd %zmm18, %zmm1, %zmm18{%k1} #75.38
vmulpd %zmm13, %zmm18, %zmm2 #76.38
vmulpd %zmm10, %zmm18, %zmm4 #77.54
vmulpd %zmm2, %zmm18, %zmm6 #76.44
vmulpd %zmm6, %zmm18, %zmm3 #76.50
vfmsub213pd %zmm5, %zmm6, %zmm18 #77.54
vmulpd %zmm4, %zmm3, %zmm17 #77.61
vmulpd %zmm18, %zmm17, %zmm22 #77.67
vfmadd231pd %zmm20, %zmm22, %zmm9{%k3} #78.17
vfmadd231pd %zmm21, %zmm22, %zmm8{%k3} #79.17
vfmadd231pd %zmm23, %zmm22, %zmm11{%k3} #80.17
# LOE rax rdx rbp rsi rdi r8 r9 r10 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.38: # Preds ..B1.23 ..B1.37 ..B1.32
# Execution count [4.50e+00]
vmovups .L_2il0floatpacket.10(%rip), %zmm19 #63.22
vpermd %zmm11, %zmm19, %zmm0 #63.22
vpermd %zmm8, %zmm19, %zmm6 #62.22
vpermd %zmm9, %zmm19, %zmm20 #61.22
vaddpd %zmm11, %zmm0, %zmm11 #63.22
vaddpd %zmm8, %zmm6, %zmm8 #62.22
vaddpd %zmm9, %zmm20, %zmm9 #61.22
vpermpd $78, %zmm11, %zmm1 #63.22
vpermpd $78, %zmm8, %zmm7 #62.22
vpermpd $78, %zmm9, %zmm21 #61.22
vaddpd %zmm1, %zmm11, %zmm2 #63.22
vaddpd %zmm7, %zmm8, %zmm12 #62.22
vaddpd %zmm21, %zmm9, %zmm22 #61.22
vpermpd $177, %zmm2, %zmm3 #63.22
vpermpd $177, %zmm12, %zmm17 #62.22
vpermpd $177, %zmm22, %zmm23 #61.22
vaddpd %zmm3, %zmm2, %zmm4 #63.22
vaddpd %zmm17, %zmm12, %zmm18 #62.22
vaddpd %zmm23, %zmm22, %zmm24 #61.22
# LOE rax rdx rbp rsi rdi r8 r9 r10 xmm4 xmm18 xmm24 ymm15 ymm16 zmm5 zmm10 zmm13 zmm14
..B1.39: # Preds ..B1.38 ..B1.9
# Execution count [5.00e+00]
movq -8(%rsp), %rcx #84.9[spill]
addq $24, %rax #55.5
vaddsd (%rcx,%rsi,8), %xmm24, %xmm0 #84.9
vmovsd %xmm0, (%rcx,%rsi,8) #84.9
movslq %esi, %rcx #55.32
vaddsd (%rdx,%rsi,8), %xmm18, %xmm1 #85.9
vmovsd %xmm1, (%rdx,%rsi,8) #85.9
incq %rcx #55.32
vaddsd (%r9,%rsi,8), %xmm4, %xmm2 #86.9
vmovsd %xmm2, (%r9,%rsi,8) #86.9
incq %rsi #55.5
cmpq -24(%rsp), %rsi #55.5[spill]
jb ..B1.9 # Prob 82% #55.5
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 ymm15 ymm16 zmm5 zmm10 zmm13 zmm14
..B1.40: # Preds ..B1.39
# Execution count [9.00e-01]
movq -88(%rsp), %r14 #[spill]
.cfi_restore 14
movq -96(%rsp), %r15 #[spill]
.cfi_restore 15
movq -104(%rsp), %rbx #[spill]
.cfi_restore 3
# LOE rbx rbp r14 r15
..B1.42: # Preds ..B1.1 ..B1.40
# Execution count [1.00e+00]
vzeroupper #93.12
vxorpd %xmm0, %xmm0, %xmm0 #93.12
.cfi_restore 13
popq %r13 #93.12
.cfi_def_cfa_offset 16
.cfi_restore 12
popq %r12 #93.12
.cfi_def_cfa_offset 8
ret #93.12
.cfi_def_cfa_offset 24
.cfi_offset 3, -128
.cfi_offset 12, -16
.cfi_offset 13, -24
.cfi_offset 14, -112
.cfi_offset 15, -120
# LOE
..B1.43: # Preds ..B1.11
# Execution count [4.50e-01]: Infreq
movl %r11d, %r14d #67.9
xorl %r12d, %r12d #67.9
andl $-8, %r14d #67.9
jmp ..B1.24 # Prob 100% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d r12d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.44: # Preds ..B1.10
# Execution count [4.50e-01]: Infreq
xorl %r14d, %r14d #67.9
jmp ..B1.32 # Prob 100% #67.9
.align 16,0x90
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
.cfi_endproc
# mark_end;
.type computeForce,@function
.size computeForce,.-computeForce
..LNcomputeForce.0:
.data
# -- End computeForce
.section .rodata, "a"
.align 64
.align 64
.L_2il0floatpacket.2:
.long 0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000
.type .L_2il0floatpacket.2,@object
.size .L_2il0floatpacket.2,64
.align 64
.L_2il0floatpacket.4:
.long 0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000
.type .L_2il0floatpacket.4,@object
.size .L_2il0floatpacket.4,64
.align 64
.L_2il0floatpacket.5:
.long 0x02010101,0x04040202,0x08080804,0x20101010,0x40402020,0x80808040,0x00000000,0x00000000,0x00000000,0x00000000,0x00000000,0x00000000,0x00000000,0x00000000,0x00000000,0x00000000
.type .L_2il0floatpacket.5,@object
.size .L_2il0floatpacket.5,64
.align 64
.L_2il0floatpacket.6:
.long 0x00000000,0x00000000,0x00000004,0x00000000,0x00000008,0x00000000,0x0000000c,0x00000000,0x00000001,0x00000000,0x00000005,0x00000000,0x00000009,0x00000000,0x0000000d,0x00000000
.type .L_2il0floatpacket.6,@object
.size .L_2il0floatpacket.6,64
.align 64
.L_2il0floatpacket.7:
.long 0x00000001,0x00000000,0x00000005,0x00000000,0x00000009,0x00000000,0x0000000d,0x00000000,0x00000000,0x00000000,0x00000004,0x00000000,0x00000008,0x00000000,0x0000000c,0x00000000
.type .L_2il0floatpacket.7,@object
.size .L_2il0floatpacket.7,64
.align 64
.L_2il0floatpacket.8:
.long 0x00000002,0x00000000,0x00000006,0x00000000,0x0000000a,0x00000000,0x0000000e,0x00000000,0x00000002,0x00000000,0x00000006,0x00000000,0x0000000a,0x00000000,0x0000000e,0x00000000
.type .L_2il0floatpacket.8,@object
.size .L_2il0floatpacket.8,64
.align 64
.L_2il0floatpacket.10:
.long 0x00000008,0x00000009,0x0000000a,0x0000000b,0x0000000c,0x0000000d,0x0000000e,0x0000000f,0x00000008,0x00000009,0x0000000a,0x0000000b,0x0000000c,0x0000000d,0x0000000e,0x0000000f
.type .L_2il0floatpacket.10,@object
.size .L_2il0floatpacket.10,64
.align 32
.L_2il0floatpacket.0:
.long 0x00000008,0x00000008,0x00000008,0x00000008,0x00000008,0x00000008,0x00000008,0x00000008
.type .L_2il0floatpacket.0,@object
.size .L_2il0floatpacket.0,32
.align 32
.L_2il0floatpacket.1:
.long 0x00000000,0x00000001,0x00000002,0x00000003,0x00000004,0x00000005,0x00000006,0x00000007
.type .L_2il0floatpacket.1,@object
.size .L_2il0floatpacket.1,32
.align 8
.L_2il0floatpacket.3:
.long 0x00000000,0x40480000
.type .L_2il0floatpacket.3,@object
.size .L_2il0floatpacket.3,8
.align 8
.L_2il0floatpacket.9:
.long 0x00000000,0x3ff00000
.type .L_2il0floatpacket.9,@object
.size .L_2il0floatpacket.9,8
.data
.section .note.GNU-stack, ""
# End

565
asm/avx512/force_aos_lt8_markers.s
View File

@ -1,565 +0,0 @@
# mark_description "Intel(R) C Intel(R) 64 Compiler for applications running on Intel(R) 64, Version 19.0.5.281 Build 20190815";
# mark_description "-I./src/includes -S -D_GNU_SOURCE -DAOS -DPRECISION=2 -DALIGNMENT=64 -restrict -Ofast -xCORE-AVX512 -qopt-zm";
# mark_description "m-usage=high -o ICC/force.s";
.file "force.c"
.text
..TXTST0:
.L_2__routine_start_computeForce_0:
# -- Begin computeForce
.text
# mark_begin;
.align 16,0x90
.globl computeForce
# --- computeForce(Parameter *, Atom *, Neighbor *, int)
computeForce:
# parameter 1: %rdi
# parameter 2: %rsi
# parameter 3: %rdx
# parameter 4: %ecx
..B1.1: # Preds ..B1.0
# Execution count [1.00e+00]
.cfi_startproc
..___tag_value_computeForce.1:
..L2:
#35.1
pushq %r12 #35.1
.cfi_def_cfa_offset 16
.cfi_offset 12, -16
pushq %r13 #35.1
.cfi_def_cfa_offset 24
.cfi_offset 13, -24
movq %rsi, %r8 #35.1
vmovsd 72(%rdi), %xmm2 #38.27
movq %rdx, %rax #35.1
vmovsd 8(%rdi), %xmm1 #39.23
vmovsd (%rdi), %xmm0 #40.24
movl 4(%r8), %r12d #36.18
movq 64(%r8), %r11 #41.20
movq 72(%r8), %rdx #41.45
movq 80(%r8), %r9 #41.70
testl %r12d, %r12d #44.24
jle ..B1.42 # Prob 50% #44.24
# LOE rax rdx rbx rbp r8 r9 r11 r14 r15 r12d xmm0 xmm1 xmm2
..B1.2: # Preds ..B1.1
# Execution count [1.00e+00]
movl %r12d, %edi #44.5
xorl %esi, %esi #44.5
movl $1, %r10d #44.5
xorl %ecx, %ecx #44.5
shrl $1, %edi #44.5
je ..B1.6 # Prob 9% #44.5
# LOE rax rdx rcx rbx rbp rsi rdi r8 r9 r11 r14 r15 r10d r12d xmm0 xmm1 xmm2
..B1.3: # Preds ..B1.2
# Execution count [9.00e-01]
xorl %r10d, %r10d #44.5
# LOE rax rdx rcx rbx rbp rsi rdi r8 r9 r10 r11 r14 r15 r12d xmm0 xmm1 xmm2
..B1.4: # Preds ..B1.4 ..B1.3
# Execution count [2.50e+00]
movq %r10, (%rcx,%r11) #45.9
incq %rsi #44.5
movq %r10, (%rcx,%rdx) #46.9
movq %r10, (%rcx,%r9) #47.9
movq %r10, 8(%rcx,%r11) #45.9
movq %r10, 8(%rcx,%rdx) #46.9
movq %r10, 8(%rcx,%r9) #47.9
addq $16, %rcx #44.5
cmpq %rdi, %rsi #44.5
jb ..B1.4 # Prob 63% #44.5
# LOE rax rdx rcx rbx rbp rsi rdi r8 r9 r10 r11 r14 r15 r12d xmm0 xmm1 xmm2
..B1.5: # Preds ..B1.4
# Execution count [9.00e-01]
lea 1(%rsi,%rsi), %r10d #45.9
# LOE rax rdx rbx rbp r8 r9 r11 r14 r15 r10d r12d xmm0 xmm1 xmm2
..B1.6: # Preds ..B1.2 ..B1.5
# Execution count [1.00e+00]
lea -1(%r10), %ecx #44.5
cmpl %r12d, %ecx #44.5
jae ..B1.8 # Prob 9% #44.5
# LOE rax rdx rbx rbp r8 r9 r11 r14 r15 r10d r12d xmm0 xmm1 xmm2
..B1.7: # Preds ..B1.6
# Execution count [9.00e-01]
movslq %r10d, %r10 #44.5
xorl %ecx, %ecx #45.9
movq %rcx, -8(%r11,%r10,8) #45.9
movq %rcx, -8(%rdx,%r10,8) #46.9
movq %rcx, -8(%r9,%r10,8) #47.9
# LOE rax rdx rbx rbp r8 r9 r11 r14 r15 r12d xmm0 xmm1 xmm2
..B1.8: # Preds ..B1.6 ..B1.7
# Execution count [9.00e-01]
vmulsd %xmm2, %xmm2, %xmm13 #38.45
xorl %ecx, %ecx #55.15
vmovdqu32 .L_2il0floatpacket.0(%rip), %ymm16 #67.9
vmulsd .L_2il0floatpacket.3(%rip), %xmm0, %xmm0 #77.41
vmovdqu .L_2il0floatpacket.1(%rip), %ymm15 #67.9
vmovups .L_2il0floatpacket.4(%rip), %zmm5 #77.54
vbroadcastsd %xmm13, %zmm14 #38.25
vbroadcastsd %xmm1, %zmm13 #39.21
vbroadcastsd %xmm0, %zmm10 #77.41
movslq %r12d, %r12 #55.5
xorl %esi, %esi #55.5
movq 24(%rax), %r13 #57.25
movq 16(%r8), %rdi #58.25
movslq 16(%rax), %r8 #56.43
movq 8(%rax), %r10 #56.19
xorl %eax, %eax #55.5
shlq $2, %r8 #37.5
movq %r12, -24(%rsp) #55.5[spill]
movq %r13, -16(%rsp) #55.5[spill]
movq %r11, -8(%rsp) #55.5[spill]
movq %r14, -88(%rsp) #55.5[spill]
movq %r15, -96(%rsp) #55.5[spill]
movq %rbx, -104(%rsp) #55.5[spill]
.cfi_offset 3, -128
.cfi_offset 14, -112
.cfi_offset 15, -120
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 ymm15 ymm16 zmm5 zmm10 zmm13 zmm14
..B1.9: # Preds ..B1.39 ..B1.8
# Execution count [5.00e+00]
movq -16(%rsp), %rbx #57.25[spill]
vxorpd %xmm24, %xmm24, %xmm24 #61.22
vmovapd %xmm24, %xmm18 #62.22
movl (%rbx,%rsi,4), %r11d #57.25
vmovapd %xmm18, %xmm4 #63.22
vmovsd (%rax,%rdi), %xmm6 #58.25
vmovsd 8(%rax,%rdi), %xmm7 #59.25
vmovsd 16(%rax,%rdi), %xmm12 #60.25
testl %r11d, %r11d #67.28
jle ..B1.39 # Prob 50% #67.28
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d xmm4 xmm6 xmm7 xmm12 xmm18 xmm24 ymm15 ymm16 zmm5 zmm10 zmm13 zmm14
..B1.10: # Preds ..B1.9
# Execution count [4.50e+00]
vpxord %zmm9, %zmm9, %zmm9 #61.22
vmovaps %zmm9, %zmm8 #62.22
vmovaps %zmm8, %zmm11 #63.22
cmpl $8, %r11d #67.9
jl ..B1.44 # Prob 10% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.11: # Preds ..B1.10
# Execution count [4.50e+00]
cmpl $1200, %r11d #67.9
jl ..B1.43 # Prob 10% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.12: # Preds ..B1.11
# Execution count [4.50e+00]
movq %r8, %r15 #56.43
imulq %rcx, %r15 #56.43
addq %r10, %r15 #37.5
movq %r15, %r12 #67.9
andq $63, %r12 #67.9
testl $3, %r12d #67.9
je ..B1.14 # Prob 50% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r15 r11d r12d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.13: # Preds ..B1.12
# Execution count [2.25e+00]
xorl %r12d, %r12d #67.9
jmp ..B1.16 # Prob 100% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r15 r11d r12d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.14: # Preds ..B1.12
# Execution count [2.25e+00]
testl %r12d, %r12d #67.9
je ..B1.16 # Prob 50% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r15 r11d r12d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.15: # Preds ..B1.14
# Execution count [2.50e+01]
negl %r12d #67.9
addl $64, %r12d #67.9
shrl $2, %r12d #67.9
cmpl %r12d, %r11d #67.9
cmovl %r11d, %r12d #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r15 r11d r12d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.16: # Preds ..B1.13 ..B1.15 ..B1.14
# Execution count [5.00e+00]
movl %r11d, %r14d #67.9
subl %r12d, %r14d #67.9
andl $7, %r14d #67.9
negl %r14d #67.9
addl %r11d, %r14d #67.9
cmpl $1, %r12d #67.9
jb ..B1.24 # Prob 50% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r15 r11d r12d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.17: # Preds ..B1.16
# Execution count [4.50e+00]
vmovdqa %ymm15, %ymm3 #67.9
xorl %r13d, %r13d #67.9
vpbroadcastd %r12d, %ymm2 #67.9
vbroadcastsd %xmm6, %zmm1 #58.23
vbroadcastsd %xmm7, %zmm0 #59.23
vbroadcastsd %xmm12, %zmm4 #60.23
movslq %r12d, %rbx #67.9
movq %r9, -80(%rsp) #67.9[spill]
movq %rdx, -72(%rsp) #67.9[spill]
# LOE rax rcx rbx rbp rsi rdi r8 r10 r13 r15 r11d r12d r14d xmm6 xmm7 xmm12 ymm2 ymm3 ymm15 ymm16 zmm0 zmm1 zmm4 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.18: # Preds ..B1.22 ..B1.17
# Execution count [2.50e+01]
vpcmpgtd %ymm3, %ymm2, %k3 #67.9
vmovdqu32 (%r15,%r13,4), %ymm17{%k3}{z} #68.21
kmovw %k3, %r9d #67.9
vpaddd %ymm17, %ymm17, %ymm18 #69.36
vpaddd %ymm18, %ymm17, %ymm17 #69.36
# LOE rax rcx rbx rbp rsi rdi r8 r10 r13 r15 r9d r11d r12d r14d xmm6 xmm7 xmm12 ymm2 ymm3 ymm15 ymm16 ymm17 zmm0 zmm1 zmm4 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14 k3
..B1.21: # Preds ..B1.18
# Execution count [1.25e+01]
kmovw %k3, %k1 #69.36
kmovw %k3, %k2 #69.36
vpxord %zmm18, %zmm18, %zmm18 #69.36
vpxord %zmm19, %zmm19, %zmm19 #69.36
vpxord %zmm20, %zmm20, %zmm20 #69.36
vgatherdpd 16(%rdi,%ymm17,8), %zmm18{%k1} #69.36
vgatherdpd 8(%rdi,%ymm17,8), %zmm19{%k2} #69.36
vgatherdpd (%rdi,%ymm17,8), %zmm20{%k3} #69.36
# LOE rax rcx rbx rbp rsi rdi r8 r10 r13 r15 r9d r11d r12d r14d xmm6 xmm7 xmm12 ymm2 ymm3 ymm15 ymm16 zmm0 zmm1 zmm4 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14 zmm18 zmm19 zmm20
..B1.22: # Preds ..B1.21
# Execution count [2.50e+01]
addq $8, %r13 #67.9
vpaddd %ymm16, %ymm3, %ymm3 #67.9
vsubpd %zmm18, %zmm4, %zmm29 #71.36
vsubpd %zmm19, %zmm0, %zmm27 #70.36
vsubpd %zmm20, %zmm1, %zmm26 #69.36
vmulpd %zmm27, %zmm27, %zmm25 #72.49
vfmadd231pd %zmm26, %zmm26, %zmm25 #72.49
vfmadd231pd %zmm29, %zmm29, %zmm25 #72.63
vrcp14pd %zmm25, %zmm24 #75.38
vcmppd $1, %zmm14, %zmm25, %k2 #74.22
vfpclasspd $30, %zmm24, %k0 #75.38
kmovw %k2, %edx #74.22
knotw %k0, %k1 #75.38
vmovaps %zmm25, %zmm17 #75.38
andl %edx, %r9d #74.22
vfnmadd213pd .L_2il0floatpacket.9(%rip){1to8}, %zmm24, %zmm17 #75.38
kmovw %r9d, %k3 #78.17
vmulpd %zmm17, %zmm17, %zmm18 #75.38
vfmadd213pd %zmm24, %zmm17, %zmm24{%k1} #75.38
vfmadd213pd %zmm24, %zmm18, %zmm24{%k1} #75.38
vmulpd %zmm13, %zmm24, %zmm19 #76.38
vmulpd %zmm10, %zmm24, %zmm21 #77.54
vmulpd %zmm19, %zmm24, %zmm22 #76.44
vmulpd %zmm22, %zmm24, %zmm20 #76.50
vfmsub213pd %zmm5, %zmm22, %zmm24 #77.54
vmulpd %zmm21, %zmm20, %zmm23 #77.61
vmulpd %zmm24, %zmm23, %zmm28 #77.67
vfmadd231pd %zmm26, %zmm28, %zmm9{%k3} #78.17
vfmadd231pd %zmm27, %zmm28, %zmm8{%k3} #79.17
vfmadd231pd %zmm29, %zmm28, %zmm11{%k3} #80.17
cmpq %rbx, %r13 #67.9
jb ..B1.18 # Prob 82% #67.9
# LOE rax rcx rbx rbp rsi rdi r8 r10 r13 r15 r11d r12d r14d xmm6 xmm7 xmm12 ymm2 ymm3 ymm15 ymm16 zmm0 zmm1 zmm4 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.23: # Preds ..B1.22
# Execution count [4.50e+00]
movq -80(%rsp), %r9 #[spill]
movq -72(%rsp), %rdx #[spill]
cmpl %r12d, %r11d #67.9
je ..B1.38 # Prob 10% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d r12d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.24: # Preds ..B1.23 ..B1.16 ..B1.43
# Execution count [2.50e+01]
lea 8(%r12), %ebx #67.9
cmpl %ebx, %r14d #67.9
jl ..B1.32 # Prob 50% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d r12d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.25: # Preds ..B1.24
# Execution count [4.50e+00]
movq %r8, %r13 #56.43
imulq %rcx, %r13 #56.43
vbroadcastsd %xmm6, %zmm2 #58.23
vbroadcastsd %xmm7, %zmm1 #59.23
vbroadcastsd %xmm12, %zmm0 #60.23
movslq %r12d, %rbx #67.9
addq %r10, %r13 #37.5
movq %rax, -64(%rsp) #37.5[spill]
movq %r8, -56(%rsp) #37.5[spill]
movq %r10, -48(%rsp) #37.5[spill]
movq %rsi, -40(%rsp) #37.5[spill]
movq %rcx, -32(%rsp) #37.5[spill]
movq %r9, -80(%rsp) #37.5[spill]
movq %rdx, -72(%rsp) #37.5[spill]
# LOE rbx rbp rdi r13 r11d r12d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm0 zmm1 zmm2 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.26: # Preds ..B1.30 ..B1.25
# Execution count [2.50e+01]
vmovdqu (%r13,%rbx,4), %ymm3 #68.21
vpaddd %ymm3, %ymm3, %ymm4 #69.36
vpaddd %ymm4, %ymm3, %ymm3 #69.36
movl (%r13,%rbx,4), %r10d #68.21
movl 4(%r13,%rbx,4), %r9d #68.21
movl 8(%r13,%rbx,4), %r8d #68.21
movl 12(%r13,%rbx,4), %esi #68.21
lea (%r10,%r10,2), %r10d #69.36
movl 16(%r13,%rbx,4), %ecx #68.21
lea (%r9,%r9,2), %r9d #69.36
movl 20(%r13,%rbx,4), %edx #68.21
lea (%r8,%r8,2), %r8d #69.36
movl 24(%r13,%rbx,4), %eax #68.21
lea (%rsi,%rsi,2), %esi #69.36
movl 28(%r13,%rbx,4), %r15d #68.21
lea (%rcx,%rcx,2), %ecx #69.36
lea (%rdx,%rdx,2), %edx #69.36
lea (%rax,%rax,2), %eax #69.36
lea (%r15,%r15,2), %r15d #69.36
# LOE rbx rbp rdi r13 eax edx ecx esi r8d r9d r10d r11d r12d r14d r15d xmm6 xmm7 xmm12 ymm3 ymm15 ymm16 zmm0 zmm1 zmm2 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.29: # Preds ..B1.26
# Execution count [1.25e+01]
vpcmpeqb %xmm0, %xmm0, %k1 #69.36
vpcmpeqb %xmm0, %xmm0, %k2 #69.36
vpcmpeqb %xmm0, %xmm0, %k3 #69.36
vpxord %zmm4, %zmm4, %zmm4 #69.36
vpxord %zmm17, %zmm17, %zmm17 #69.36
vpxord %zmm18, %zmm18, %zmm18 #69.36
vgatherdpd 16(%rdi,%ymm3,8), %zmm4{%k1} #69.36
vgatherdpd 8(%rdi,%ymm3,8), %zmm17{%k2} #69.36
vgatherdpd (%rdi,%ymm3,8), %zmm18{%k3} #69.36
# LOE rbx rbp rdi r13 r11d r12d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm0 zmm1 zmm2 zmm4 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14 zmm17 zmm18
..B1.30: # Preds ..B1.29
# Execution count [2.50e+01]
addl $8, %r12d #67.9
addq $8, %rbx #67.9
vsubpd %zmm4, %zmm0, %zmm26 #71.36
vsubpd %zmm17, %zmm1, %zmm24 #70.36
vsubpd %zmm18, %zmm2, %zmm23 #69.36
vmulpd %zmm24, %zmm24, %zmm3 #72.49
vfmadd231pd %zmm23, %zmm23, %zmm3 #72.49
vfmadd231pd %zmm26, %zmm26, %zmm3 #72.63
vrcp14pd %zmm3, %zmm22 #75.38
vcmppd $1, %zmm14, %zmm3, %k2 #74.22
vfpclasspd $30, %zmm22, %k0 #75.38
vfnmadd213pd .L_2il0floatpacket.9(%rip){1to8}, %zmm22, %zmm3 #75.38
knotw %k0, %k1 #75.38
vmulpd %zmm3, %zmm3, %zmm4 #75.38
vfmadd213pd %zmm22, %zmm3, %zmm22{%k1} #75.38
vfmadd213pd %zmm22, %zmm4, %zmm22{%k1} #75.38
vmulpd %zmm13, %zmm22, %zmm17 #76.38
vmulpd %zmm10, %zmm22, %zmm19 #77.54
vmulpd %zmm17, %zmm22, %zmm20 #76.44
vmulpd %zmm20, %zmm22, %zmm18 #76.50
vfmsub213pd %zmm5, %zmm20, %zmm22 #77.54
vmulpd %zmm19, %zmm18, %zmm21 #77.61
vmulpd %zmm22, %zmm21, %zmm25 #77.67
vfmadd231pd %zmm23, %zmm25, %zmm9{%k2} #78.17
vfmadd231pd %zmm24, %zmm25, %zmm8{%k2} #79.17
vfmadd231pd %zmm26, %zmm25, %zmm11{%k2} #80.17
cmpl %r14d, %r12d #67.9
jb ..B1.26 # Prob 82% #67.9
# LOE rbx rbp rdi r13 r11d r12d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm0 zmm1 zmm2 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.31: # Preds ..B1.30
# Execution count [4.50e+00]
movq -64(%rsp), %rax #[spill]
movq -56(%rsp), %r8 #[spill]
movq -48(%rsp), %r10 #[spill]
movq -40(%rsp), %rsi #[spill]
movq -32(%rsp), %rcx #[spill]
movq -80(%rsp), %r9 #[spill]
movq -72(%rsp), %rdx #[spill]
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.32: # Preds ..B1.31 ..B1.24 ..B1.44
# Execution count [5.00e+00]
lea 1(%r14), %ebx #67.9
cmpl %r11d, %ebx #67.9
ja ..B1.38 # Prob 50% #67.9
movl $111,%ebx #IACA/OSACA START MARKER
.byte 100,103,144 #IACA/OSACA START MARKER
# LLVM-MCA-BEGIN
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.33: # Preds ..B1.32
# Execution count [2.50e+01]
imulq %r8, %rcx #56.43
vbroadcastsd %xmm6, %zmm4 #58.23
subl %r14d, %r11d #67.9
addq %r10, %rcx #37.5
vpbroadcastd %r11d, %ymm0 #67.9
vpcmpgtd %ymm15, %ymm0, %k3 #67.9
movslq %r14d, %r14 #67.9
kmovw %k3, %ebx #67.9
vmovdqu32 (%rcx,%r14,4), %ymm1{%k3}{z} #68.21
vpaddd %ymm1, %ymm1, %ymm2 #69.36
vpaddd %ymm2, %ymm1, %ymm0 #69.36
# LOE rax rdx rbp rsi rdi r8 r9 r10 ebx xmm7 xmm12 ymm0 ymm15 ymm16 zmm4 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14 k3
..B1.36: # Preds ..B1.33
# Execution count [1.25e+01]
kmovw %k3, %k1 #69.36
kmovw %k3, %k2 #69.36
vpxord %zmm1, %zmm1, %zmm1 #69.36
vpxord %zmm2, %zmm2, %zmm2 #69.36
vpxord %zmm3, %zmm3, %zmm3 #69.36
vgatherdpd 16(%rdi,%ymm0,8), %zmm1{%k1} #69.36
vgatherdpd 8(%rdi,%ymm0,8), %zmm2{%k2} #69.36
vgatherdpd (%rdi,%ymm0,8), %zmm3{%k3} #69.36
# LOE rax rdx rbp rsi rdi r8 r9 r10 ebx xmm7 xmm12 ymm15 ymm16 zmm1 zmm2 zmm3 zmm4 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.37: # Preds ..B1.36
# Execution count [2.50e+01]
vbroadcastsd %xmm7, %zmm7 #59.23
vbroadcastsd %xmm12, %zmm12 #60.23
vsubpd %zmm1, %zmm12, %zmm23 #71.36
vsubpd %zmm2, %zmm7, %zmm21 #70.36
vsubpd %zmm3, %zmm4, %zmm20 #69.36
vmulpd %zmm21, %zmm21, %zmm19 #72.49
vfmadd231pd %zmm20, %zmm20, %zmm19 #72.49
vfmadd231pd %zmm23, %zmm23, %zmm19 #72.63
vrcp14pd %zmm19, %zmm18 #75.38
vcmppd $1, %zmm14, %zmm19, %k2 #74.22
vfpclasspd $30, %zmm18, %k0 #75.38
kmovw %k2, %ecx #74.22
knotw %k0, %k1 #75.38
vmovaps %zmm19, %zmm0 #75.38
andl %ecx, %ebx #74.22
vfnmadd213pd .L_2il0floatpacket.9(%rip){1to8}, %zmm18, %zmm0 #75.38
kmovw %ebx, %k3 #78.17
vmulpd %zmm0, %zmm0, %zmm1 #75.38
vfmadd213pd %zmm18, %zmm0, %zmm18{%k1} #75.38
vfmadd213pd %zmm18, %zmm1, %zmm18{%k1} #75.38
vmulpd %zmm13, %zmm18, %zmm2 #76.38
vmulpd %zmm10, %zmm18, %zmm4 #77.54
vmulpd %zmm2, %zmm18, %zmm6 #76.44
vmulpd %zmm6, %zmm18, %zmm3 #76.50
vfmsub213pd %zmm5, %zmm6, %zmm18 #77.54
vmulpd %zmm4, %zmm3, %zmm17 #77.61
vmulpd %zmm18, %zmm17, %zmm22 #77.67
vfmadd231pd %zmm20, %zmm22, %zmm9{%k3} #78.17
vfmadd231pd %zmm21, %zmm22, %zmm8{%k3} #79.17
vfmadd231pd %zmm23, %zmm22, %zmm11{%k3} #80.17
movl $222,%ebx #IACA/OSACA END MARKER
.byte 100,103,144 #IACA/OSACA END MARKER
# LLVM-MCA-END
# LOE rax rdx rbp rsi rdi r8 r9 r10 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.38: # Preds ..B1.23 ..B1.37 ..B1.32
# Execution count [4.50e+00]
vmovups .L_2il0floatpacket.10(%rip), %zmm19 #63.22
vpermd %zmm11, %zmm19, %zmm0 #63.22
vpermd %zmm8, %zmm19, %zmm6 #62.22
vpermd %zmm9, %zmm19, %zmm20 #61.22
vaddpd %zmm11, %zmm0, %zmm11 #63.22
vaddpd %zmm8, %zmm6, %zmm8 #62.22
vaddpd %zmm9, %zmm20, %zmm9 #61.22
vpermpd $78, %zmm11, %zmm1 #63.22
vpermpd $78, %zmm8, %zmm7 #62.22
vpermpd $78, %zmm9, %zmm21 #61.22
vaddpd %zmm1, %zmm11, %zmm2 #63.22
vaddpd %zmm7, %zmm8, %zmm12 #62.22
vaddpd %zmm21, %zmm9, %zmm22 #61.22
vpermpd $177, %zmm2, %zmm3 #63.22
vpermpd $177, %zmm12, %zmm17 #62.22
vpermpd $177, %zmm22, %zmm23 #61.22
vaddpd %zmm3, %zmm2, %zmm4 #63.22
vaddpd %zmm17, %zmm12, %zmm18 #62.22
vaddpd %zmm23, %zmm22, %zmm24 #61.22
# LOE rax rdx rbp rsi rdi r8 r9 r10 xmm4 xmm18 xmm24 ymm15 ymm16 zmm5 zmm10 zmm13 zmm14
..B1.39: # Preds ..B1.38 ..B1.9
# Execution count [5.00e+00]
movq -8(%rsp), %rcx #84.9[spill]
addq $24, %rax #55.5
vaddsd (%rcx,%rsi,8), %xmm24, %xmm0 #84.9
vmovsd %xmm0, (%rcx,%rsi,8) #84.9
movslq %esi, %rcx #55.32
vaddsd (%rdx,%rsi,8), %xmm18, %xmm1 #85.9
vmovsd %xmm1, (%rdx,%rsi,8) #85.9
incq %rcx #55.32
vaddsd (%r9,%rsi,8), %xmm4, %xmm2 #86.9
vmovsd %xmm2, (%r9,%rsi,8) #86.9
incq %rsi #55.5
cmpq -24(%rsp), %rsi #55.5[spill]
jb ..B1.9 # Prob 82% #55.5
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 ymm15 ymm16 zmm5 zmm10 zmm13 zmm14
..B1.40: # Preds ..B1.39
# Execution count [9.00e-01]
movq -88(%rsp), %r14 #[spill]
.cfi_restore 14
movq -96(%rsp), %r15 #[spill]
.cfi_restore 15
movq -104(%rsp), %rbx #[spill]
.cfi_restore 3
# LOE rbx rbp r14 r15
..B1.42: # Preds ..B1.1 ..B1.40
# Execution count [1.00e+00]
vzeroupper #93.12
vxorpd %xmm0, %xmm0, %xmm0 #93.12
.cfi_restore 13
popq %r13 #93.12
.cfi_def_cfa_offset 16
.cfi_restore 12
popq %r12 #93.12
.cfi_def_cfa_offset 8
ret #93.12
.cfi_def_cfa_offset 24
.cfi_offset 3, -128
.cfi_offset 12, -16
.cfi_offset 13, -24
.cfi_offset 14, -112
.cfi_offset 15, -120
# LOE
..B1.43: # Preds ..B1.11
# Execution count [4.50e-01]: Infreq
movl %r11d, %r14d #67.9
xorl %r12d, %r12d #67.9
andl $-8, %r14d #67.9
jmp ..B1.24 # Prob 100% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d r12d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
..B1.44: # Preds ..B1.10
# Execution count [4.50e-01]: Infreq
xorl %r14d, %r14d #67.9
jmp ..B1.32 # Prob 100% #67.9
.align 16,0x90
# LOE rax rdx rcx rbp rsi rdi r8 r9 r10 r11d r14d xmm6 xmm7 xmm12 ymm15 ymm16 zmm5 zmm8 zmm9 zmm10 zmm11 zmm13 zmm14
.cfi_endproc
# mark_end;
.type computeForce,@function
.size computeForce,.-computeForce
..LNcomputeForce.0:
.data
# -- End computeForce
.section .rodata, "a"
.align 64
.align 64
.L_2il0floatpacket.2:
.long 0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000
.type .L_2il0floatpacket.2,@object
.size .L_2il0floatpacket.2,64
.align 64
.L_2il0floatpacket.4:
.long 0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000
.type .L_2il0floatpacket.4,@object
.size .L_2il0floatpacket.4,64
.align 64
.L_2il0floatpacket.5:
.long 0x02010101,0x04040202,0x08080804,0x20101010,0x40402020,0x80808040,0x00000000,0x00000000,0x00000000,0x00000000,0x00000000,0x00000000,0x00000000,0x00000000,0x00000000,0x00000000
.type .L_2il0floatpacket.5,@object
.size .L_2il0floatpacket.5,64
.align 64
.L_2il0floatpacket.6:
.long 0x00000000,0x00000000,0x00000004,0x00000000,0x00000008,0x00000000,0x0000000c,0x00000000,0x00000001,0x00000000,0x00000005,0x00000000,0x00000009,0x00000000,0x0000000d,0x00000000
.type .L_2il0floatpacket.6,@object
.size .L_2il0floatpacket.6,64
.align 64
.L_2il0floatpacket.7:
.long 0x00000001,0x00000000,0x00000005,0x00000000,0x00000009,0x00000000,0x0000000d,0x00000000,0x00000000,0x00000000,0x00000004,0x00000000,0x00000008,0x00000000,0x0000000c,0x00000000
.type .L_2il0floatpacket.7,@object
.size .L_2il0floatpacket.7,64
.align 64
.L_2il0floatpacket.8:
.long 0x00000002,0x00000000,0x00000006,0x00000000,0x0000000a,0x00000000,0x0000000e,0x00000000,0x00000002,0x00000000,0x00000006,0x00000000,0x0000000a,0x00000000,0x0000000e,0x00000000
.type .L_2il0floatpacket.8,@object
.size .L_2il0floatpacket.8,64
.align 64
.L_2il0floatpacket.10:
.long 0x00000008,0x00000009,0x0000000a,0x0000000b,0x0000000c,0x0000000d,0x0000000e,0x0000000f,0x00000008,0x00000009,0x0000000a,0x0000000b,0x0000000c,0x0000000d,0x0000000e,0x0000000f
.type .L_2il0floatpacket.10,@object
.size .L_2il0floatpacket.10,64
.align 32
.L_2il0floatpacket.0:
.long 0x00000008,0x00000008,0x00000008,0x00000008,0x00000008,0x00000008,0x00000008,0x00000008
.type .L_2il0floatpacket.0,@object
.size .L_2il0floatpacket.0,32
.align 32
.L_2il0floatpacket.1:
.long 0x00000000,0x00000001,0x00000002,0x00000003,0x00000004,0x00000005,0x00000006,0x00000007
.type .L_2il0floatpacket.1,@object
.size .L_2il0floatpacket.1,32
.align 8
.L_2il0floatpacket.3:
.long 0x00000000,0x40480000
.type .L_2il0floatpacket.3,@object
.size .L_2il0floatpacket.3,8
.align 8
.L_2il0floatpacket.9:
.long 0x00000000,0x3ff00000
.type .L_2il0floatpacket.9,@object
.size .L_2il0floatpacket.9,8
.data
.section .note.GNU-stack, ""
# End

12730
asm/avx512/force_soa_att.s
View File

File diff suppressed because it is too large Load Diff

475
asm/avx512/force_soa_geq1200_markers.s
View File

@ -1,475 +0,0 @@
# mark_description "Intel(R) C Intel(R) 64 Compiler for applications running on Intel(R) 64, Version 19.0.5.281 Build 20190815";
# mark_description "-I./src/includes -S -D_GNU_SOURCE -DPRECISION=2 -DALIGNMENT=64 -restrict -Ofast -xCORE-AVX512 -qopt-zmm-usag";
# mark_description "e=high -o ICC/force.s";
.file "force.c"
.text
..TXTST0:
.L_2__routine_start_computeForce_0:
# -- Begin computeForce
.text
# mark_begin;
.align 16,0x90
.globl computeForce
# --- computeForce(Parameter *, Atom *, Neighbor *, int)
computeForce:
# parameter 1: %rdi
# parameter 2: %rsi
# parameter 3: %rdx
# parameter 4: %ecx
..B1.1: # Preds ..B1.0
# Execution count [1.00e+00]
.cfi_startproc
..___tag_value_computeForce.1:
..L2:
#35.1
pushq %r12 #35.1
.cfi_def_cfa_offset 16
.cfi_offset 12, -16
pushq %r13 #35.1
.cfi_def_cfa_offset 24
.cfi_offset 13, -24
pushq %r14 #35.1
.cfi_def_cfa_offset 32
.cfi_offset 14, -32
movl 4(%rsi), %r9d #36.18
vmovsd 72(%rdi), %xmm2 #38.27
vmovsd 8(%rdi), %xmm1 #39.23
vmovsd (%rdi), %xmm0 #40.24
movq 64(%rsi), %r13 #41.20
movq 72(%rsi), %r14 #41.45
movq 80(%rsi), %rdi #41.70
testl %r9d, %r9d #44.24
jle ..B1.30 # Prob 50% #44.24
# LOE rdx rbx rbp rsi rdi r13 r14 r15 r9d xmm0 xmm1 xmm2
..B1.2: # Preds ..B1.1
# Execution count [1.00e+00]
xorl %r10d, %r10d #44.5
movl %r9d, %ecx #44.5
xorl %r8d, %r8d #44.5
movl $1, %r11d #44.5
xorl %eax, %eax #45.17
shrl $1, %ecx #44.5
je ..B1.6 # Prob 9% #44.5
# LOE rax rdx rcx rbx rbp rsi rdi r8 r10 r13 r14 r15 r9d r11d xmm0 xmm1 xmm2
..B1.4: # Preds ..B1.2 ..B1.4
# Execution count [2.50e+00]
movq %rax, (%r8,%r13) #45.9
incq %r10 #44.5
movq %rax, (%r8,%r14) #46.9
movq %rax, (%r8,%rdi) #47.9
movq %rax, 8(%r8,%r13) #45.9
movq %rax, 8(%r8,%r14) #46.9
movq %rax, 8(%r8,%rdi) #47.9
addq $16, %r8 #44.5
cmpq %rcx, %r10 #44.5
jb ..B1.4 # Prob 63% #44.5
# LOE rax rdx rcx rbx rbp rsi rdi r8 r10 r13 r14 r15 r9d xmm0 xmm1 xmm2
..B1.5: # Preds ..B1.4
# Execution count [9.00e-01]
lea 1(%r10,%r10), %r11d #45.9
# LOE rax rdx rbx rbp rsi rdi r13 r14 r15 r9d r11d xmm0 xmm1 xmm2
..B1.6: # Preds ..B1.2 ..B1.5
# Execution count [1.00e+00]
lea -1(%r11), %ecx #44.5
cmpl %r9d, %ecx #44.5
jae ..B1.8 # Prob 9% #44.5
# LOE rax rdx rbx rbp rsi rdi r13 r14 r15 r9d r11d xmm0 xmm1 xmm2
..B1.7: # Preds ..B1.6
# Execution count [9.00e-01]
movslq %r11d, %r11 #44.5
movq %rax, -8(%r13,%r11,8) #45.9
movq %rax, -8(%r14,%r11,8) #46.9
movq %rax, -8(%rdi,%r11,8) #47.9
# LOE rdx rbx rbp rsi rdi r13 r14 r15 r9d xmm0 xmm1 xmm2
..B1.8: # Preds ..B1.6 ..B1.7
# Execution count [9.00e-01]
vmulsd %xmm2, %xmm2, %xmm15 #38.45
xorl %r8d, %r8d #55.15
vmovdqu32 .L_2il0floatpacket.0(%rip), %ymm18 #67.9
vmulsd .L_2il0floatpacket.3(%rip), %xmm0, %xmm0 #77.41
vmovdqu32 .L_2il0floatpacket.1(%rip), %ymm17 #67.9
vmovups .L_2il0floatpacket.4(%rip), %zmm7 #77.54
vbroadcastsd %xmm15, %zmm16 #38.25
vbroadcastsd %xmm1, %zmm15 #39.21
vbroadcastsd %xmm0, %zmm14 #77.41
movslq %r9d, %r9 #55.5
xorl %r10d, %r10d #55.5
movq 24(%rdx), %rcx #57.25
movq 8(%rdx), %r11 #56.19
movslq 16(%rdx), %r12 #56.43
movq 16(%rsi), %rdx #58.25
movq 24(%rsi), %rax #59.25
movq 32(%rsi), %rsi #60.25
shlq $2, %r12 #37.5
movq %r9, -32(%rsp) #60.25[spill]
movq %rcx, -24(%rsp) #60.25[spill]
movq %r14, -16(%rsp) #60.25[spill]
movq %r13, -8(%rsp) #60.25[spill]
movq %r15, -40(%rsp) #60.25[spill]
movq %rbx, -48(%rsp) #60.25[spill]
vpxord %zmm19, %zmm19, %zmm19 #61.22
.cfi_offset 3, -80
.cfi_offset 15, -72
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 ymm17 ymm18 zmm7 zmm14 zmm15 zmm16 zmm19
..B1.9: # Preds ..B1.27 ..B1.8
# Execution count [5.00e+00]
movq -24(%rsp), %rcx #57.25[spill]
vxorpd %xmm25, %xmm25, %xmm25 #61.22
vmovapd %xmm25, %xmm20 #62.22
movl (%rcx,%r10,4), %r13d #57.25
vmovapd %xmm20, %xmm4 #63.22
vmovsd (%rdx,%r10,8), %xmm8 #58.25
vmovsd (%rax,%r10,8), %xmm9 #59.25
vmovsd (%rsi,%r10,8), %xmm10 #60.25
testl %r13d, %r13d #67.28
jle ..B1.27 # Prob 50% #67.28
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 r13d xmm4 xmm8 xmm9 xmm10 xmm20 xmm25 ymm17 ymm18 zmm7 zmm14 zmm15 zmm16 zmm19
..B1.10: # Preds ..B1.9
# Execution count [4.50e+00]
vmovaps %zmm19, %zmm13 #61.22
vmovaps %zmm13, %zmm12 #62.22
vmovaps %zmm12, %zmm11 #63.22
cmpl $8, %r13d #67.9
jl ..B1.32 # Prob 10% #67.9
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.11: # Preds ..B1.10
# Execution count [4.50e+00]
cmpl $1200, %r13d #67.9
jl ..B1.31 # Prob 10% #67.9
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.12: # Preds ..B1.11
# Execution count [4.50e+00]
movq %r12, %rcx #56.43
imulq %r8, %rcx #56.43
addq %r11, %rcx #37.5
movq %rcx, %r9 #67.9
andq $63, %r9 #67.9
testl $3, %r9d #67.9
je ..B1.14 # Prob 50% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.13: # Preds ..B1.12
# Execution count [2.25e+00]
xorl %r9d, %r9d #67.9
jmp ..B1.16 # Prob 100% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.14: # Preds ..B1.12
# Execution count [2.25e+00]
testl %r9d, %r9d #67.9
je ..B1.16 # Prob 50% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.15: # Preds ..B1.14
# Execution count [2.50e+01]
negl %r9d #67.9
addl $64, %r9d #67.9
shrl $2, %r9d #67.9
cmpl %r9d, %r13d #67.9
cmovl %r13d, %r9d #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.16: # Preds ..B1.13 ..B1.15 ..B1.14
# Execution count [5.00e+00]
movl %r13d, %ebx #67.9
subl %r9d, %ebx #67.9
andl $7, %ebx #67.9
negl %ebx #67.9
addl %r13d, %ebx #67.9
cmpl $1, %r9d #67.9
jb ..B1.20 # Prob 50% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 ebx r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.17: # Preds ..B1.16
# Execution count [4.50e+00]
vmovdqa32 %ymm17, %ymm4 #67.9
xorl %r15d, %r15d #67.9
vpbroadcastd %r9d, %ymm3 #67.9
vbroadcastsd %xmm8, %zmm2 #58.23
vbroadcastsd %xmm9, %zmm5 #59.23
vbroadcastsd %xmm10, %zmm6 #60.23
movslq %r9d, %r14 #67.9
movl $111,%ebx #IACA/OSACA START MARKER
.byte 100,103,144 #IACA/OSACA START MARKER
# LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 r14 r15 ebx r9d r13d xmm8 xmm9 xmm10 ymm3 ymm4 ymm17 ymm18 zmm2 zmm5 zmm6 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.18: # Preds ..B1.18 ..B1.17
# Execution count [2.50e+01]
vpcmpgtd %ymm4, %ymm3, %k5 #67.9
vpaddd %ymm18, %ymm4, %ymm4 #67.9
vmovdqu32 (%rcx,%r15,4), %ymm20{%k5}{z} #68.21
vmovaps %zmm19, %zmm22 #70.36
addq $8, %r15 #67.9
kmovw %k5, %k2 #70.36
vmovaps %zmm19, %zmm21 #69.36
kmovw %k5, %k1 #69.36
vmovaps %zmm19, %zmm23 #71.36
kmovw %k5, %k3 #71.36
vgatherdpd (%rsi,%ymm20,8), %zmm23{%k3} #71.36
vgatherdpd (%rax,%ymm20,8), %zmm22{%k2} #70.36
vgatherdpd (%rdx,%ymm20,8), %zmm21{%k1} #69.36
vsubpd %zmm22, %zmm5, %zmm0 #70.36
vsubpd %zmm21, %zmm2, %zmm1 #69.36
vsubpd %zmm23, %zmm6, %zmm21 #71.36
vmulpd %zmm0, %zmm0, %zmm20 #72.49
vfmadd231pd %zmm1, %zmm1, %zmm20 #72.49
vfmadd231pd %zmm21, %zmm21, %zmm20 #72.63
vrcp14pd %zmm20, %zmm31 #75.38
vcmppd $1, %zmm16, %zmm20, %k6{%k5} #74.22
vfpclasspd $30, %zmm31, %k0 #75.38
vmovaps %zmm20, %zmm24 #75.38
vfnmadd213pd .L_2il0floatpacket.5(%rip){1to8}, %zmm31, %zmm24 #75.38
knotw %k0, %k4 #75.38
vmulpd %zmm24, %zmm24, %zmm25 #75.38
vfmadd213pd %zmm31, %zmm24, %zmm31{%k4} #75.38
vfmadd213pd %zmm31, %zmm25, %zmm31{%k4} #75.38
vmulpd %zmm15, %zmm31, %zmm26 #76.38
vmulpd %zmm14, %zmm31, %zmm28 #77.54
vmulpd %zmm26, %zmm31, %zmm29 #76.44
vmulpd %zmm29, %zmm31, %zmm27 #76.50
vfmsub213pd %zmm7, %zmm29, %zmm31 #77.54
vmulpd %zmm28, %zmm27, %zmm30 #77.61
vmulpd %zmm31, %zmm30, %zmm24 #77.67
vfmadd231pd %zmm1, %zmm24, %zmm13{%k6} #78.17
vfmadd231pd %zmm0, %zmm24, %zmm12{%k6} #79.17
vfmadd231pd %zmm21, %zmm24, %zmm11{%k6} #80.17
cmpq %r14, %r15 #67.9
jb ..B1.18 # Prob 82% #67.9
movl $222,%ebx #IACA/OSACA END MARKER
.byte 100,103,144 #IACA/OSACA END MARKER
# LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 r14 r15 ebx r9d r13d xmm8 xmm9 xmm10 ymm3 ymm4 ymm17 ymm18 zmm2 zmm5 zmm6 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.19: # Preds ..B1.18
# Execution count [4.50e+00]
cmpl %r9d, %r13d #67.9
je ..B1.26 # Prob 10% #67.9
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 ebx r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.20: # Preds ..B1.19 ..B1.16 ..B1.31
# Execution count [2.50e+01]
lea 8(%r9), %ecx #67.9
cmpl %ecx, %ebx #67.9
jl ..B1.24 # Prob 50% #67.9
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 ebx r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.21: # Preds ..B1.20
# Execution count [4.50e+00]
movq %r12, %rcx #56.43
imulq %r8, %rcx #56.43
vbroadcastsd %xmm8, %zmm0 #58.23
vbroadcastsd %xmm9, %zmm1 #59.23
vbroadcastsd %xmm10, %zmm2 #60.23
movslq %r9d, %r14 #67.9
addq %r11, %rcx #37.5
# LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 r14 ebx r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm0 zmm1 zmm2 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.22: # Preds ..B1.22 ..B1.21
# Execution count [2.50e+01]
vpcmpeqb %xmm0, %xmm0, %k2 #70.36
addl $8, %r9d #67.9
vpcmpeqb %xmm0, %xmm0, %k1 #69.36
vpcmpeqb %xmm0, %xmm0, %k3 #71.36
vmovdqu (%rcx,%r14,4), %ymm3 #68.21
addq $8, %r14 #67.9
vpxord %zmm5, %zmm5, %zmm5 #70.36
vpxord %zmm4, %zmm4, %zmm4 #69.36
vpxord %zmm6, %zmm6, %zmm6 #71.36
vgatherdpd (%rax,%ymm3,8), %zmm5{%k2} #70.36
vgatherdpd (%rdx,%ymm3,8), %zmm4{%k1} #69.36
vgatherdpd (%rsi,%ymm3,8), %zmm6{%k3} #71.36
vsubpd %zmm5, %zmm1, %zmm29 #70.36
vsubpd %zmm4, %zmm0, %zmm28 #69.36
vsubpd %zmm6, %zmm2, %zmm31 #71.36
vmulpd %zmm29, %zmm29, %zmm20 #72.49
vfmadd231pd %zmm28, %zmm28, %zmm20 #72.49
vfmadd231pd %zmm31, %zmm31, %zmm20 #72.63
vrcp14pd %zmm20, %zmm27 #75.38
vcmppd $1, %zmm16, %zmm20, %k5 #74.22
vfpclasspd $30, %zmm27, %k0 #75.38
vfnmadd213pd .L_2il0floatpacket.5(%rip){1to8}, %zmm27, %zmm20 #75.38
knotw %k0, %k4 #75.38
vmulpd %zmm20, %zmm20, %zmm21 #75.38
vfmadd213pd %zmm27, %zmm20, %zmm27{%k4} #75.38
vfmadd213pd %zmm27, %zmm21, %zmm27{%k4} #75.38
vmulpd %zmm15, %zmm27, %zmm22 #76.38
vmulpd %zmm14, %zmm27, %zmm24 #77.54
vmulpd %zmm22, %zmm27, %zmm25 #76.44
vmulpd %zmm25, %zmm27, %zmm23 #76.50
vfmsub213pd %zmm7, %zmm25, %zmm27 #77.54
vmulpd %zmm24, %zmm23, %zmm26 #77.61
vmulpd %zmm27, %zmm26, %zmm30 #77.67
vfmadd231pd %zmm28, %zmm30, %zmm13{%k5} #78.17
vfmadd231pd %zmm29, %zmm30, %zmm12{%k5} #79.17
vfmadd231pd %zmm31, %zmm30, %zmm11{%k5} #80.17
cmpl %ebx, %r9d #67.9
jb ..B1.22 # Prob 82% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 r14 ebx r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm0 zmm1 zmm2 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.24: # Preds ..B1.22 ..B1.20 ..B1.32
# Execution count [5.00e+00]
lea 1(%rbx), %ecx #67.9
cmpl %r13d, %ecx #67.9
ja ..B1.26 # Prob 50% #67.9
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 ebx r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.25: # Preds ..B1.24
# Execution count [2.50e+01]
imulq %r12, %r8 #56.43
vbroadcastsd %xmm9, %zmm9 #59.23
vbroadcastsd %xmm8, %zmm2 #58.23
vbroadcastsd %xmm10, %zmm10 #60.23
subl %ebx, %r13d #67.9
addq %r11, %r8 #37.5
vpbroadcastd %r13d, %ymm0 #67.9
vpcmpgtd %ymm17, %ymm0, %k5 #67.9
movslq %ebx, %rbx #67.9
vmovaps %zmm19, %zmm4 #70.36
kmovw %k5, %k2 #70.36
vmovaps %zmm19, %zmm3 #69.36
vmovdqu32 (%r8,%rbx,4), %ymm1{%k5}{z} #68.21
kmovw %k5, %k1 #69.36
vmovaps %zmm19, %zmm5 #71.36
kmovw %k5, %k3 #71.36
vgatherdpd (%rsi,%ymm1,8), %zmm5{%k3} #71.36
vgatherdpd (%rax,%ymm1,8), %zmm4{%k2} #70.36
vgatherdpd (%rdx,%ymm1,8), %zmm3{%k1} #69.36
vsubpd %zmm5, %zmm10, %zmm30 #71.36
vsubpd %zmm4, %zmm9, %zmm28 #70.36
vsubpd %zmm3, %zmm2, %zmm27 #69.36
vmulpd %zmm28, %zmm28, %zmm26 #72.49
vfmadd231pd %zmm27, %zmm27, %zmm26 #72.49
vfmadd231pd %zmm30, %zmm30, %zmm26 #72.63
vrcp14pd %zmm26, %zmm25 #75.38
vcmppd $1, %zmm16, %zmm26, %k6{%k5} #74.22
vfpclasspd $30, %zmm25, %k0 #75.38
vmovaps %zmm26, %zmm6 #75.38
vfnmadd213pd .L_2il0floatpacket.5(%rip){1to8}, %zmm25, %zmm6 #75.38
knotw %k0, %k4 #75.38
vmulpd %zmm6, %zmm6, %zmm8 #75.38
vfmadd213pd %zmm25, %zmm6, %zmm25{%k4} #75.38
vfmadd213pd %zmm25, %zmm8, %zmm25{%k4} #75.38
vmulpd %zmm15, %zmm25, %zmm20 #76.38
vmulpd %zmm14, %zmm25, %zmm22 #77.54
vmulpd %zmm20, %zmm25, %zmm23 #76.44
vmulpd %zmm23, %zmm25, %zmm21 #76.50
vfmsub213pd %zmm7, %zmm23, %zmm25 #77.54
vmulpd %zmm22, %zmm21, %zmm24 #77.61
vmulpd %zmm25, %zmm24, %zmm29 #77.67
vfmadd231pd %zmm27, %zmm29, %zmm13{%k6} #78.17
vfmadd231pd %zmm28, %zmm29, %zmm12{%k6} #79.17
vfmadd231pd %zmm30, %zmm29, %zmm11{%k6} #80.17
# LOE rax rdx rbp rsi rdi r10 r11 r12 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.26: # Preds ..B1.19 ..B1.25 ..B1.24
# Execution count [4.50e+00]
vmovups .L_2il0floatpacket.6(%rip), %zmm10 #63.22
vpermd %zmm11, %zmm10, %zmm0 #63.22
vpermd %zmm12, %zmm10, %zmm5 #62.22
vpermd %zmm13, %zmm10, %zmm21 #61.22
vaddpd %zmm11, %zmm0, %zmm11 #63.22
vaddpd %zmm12, %zmm5, %zmm12 #62.22
vaddpd %zmm13, %zmm21, %zmm13 #61.22
vpermpd $78, %zmm11, %zmm1 #63.22
vpermpd $78, %zmm12, %zmm6 #62.22
vpermpd $78, %zmm13, %zmm22 #61.22
vaddpd %zmm1, %zmm11, %zmm2 #63.22
vaddpd %zmm6, %zmm12, %zmm8 #62.22
vaddpd %zmm22, %zmm13, %zmm23 #61.22
vpermpd $177, %zmm2, %zmm3 #63.22
vpermpd $177, %zmm8, %zmm9 #62.22
vpermpd $177, %zmm23, %zmm24 #61.22
vaddpd %zmm3, %zmm2, %zmm4 #63.22
vaddpd %zmm9, %zmm8, %zmm20 #62.22
vaddpd %zmm24, %zmm23, %zmm25 #61.22
# LOE rax rdx rbp rsi rdi r10 r11 r12 xmm4 xmm20 xmm25 ymm17 ymm18 zmm7 zmm14 zmm15 zmm16 zmm19
..B1.27: # Preds ..B1.26 ..B1.9
# Execution count [5.00e+00]
movq -8(%rsp), %rcx #84.9[spill]
movq -16(%rsp), %rbx #85.9[spill]
movslq %r10d, %r8 #55.32
incq %r8 #55.32
vaddsd (%rcx,%r10,8), %xmm25, %xmm0 #84.9
vmovsd %xmm0, (%rcx,%r10,8) #84.9
vaddsd (%rbx,%r10,8), %xmm20, %xmm1 #85.9
vmovsd %xmm1, (%rbx,%r10,8) #85.9
vaddsd (%rdi,%r10,8), %xmm4, %xmm2 #86.9
vmovsd %xmm2, (%rdi,%r10,8) #86.9
incq %r10 #55.5
cmpq -32(%rsp), %r10 #55.5[spill]
jb ..B1.9 # Prob 82% #55.5
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 ymm17 ymm18 zmm7 zmm14 zmm15 zmm16 zmm19
..B1.28: # Preds ..B1.27
# Execution count [9.00e-01]
movq -40(%rsp), %r15 #[spill]
.cfi_restore 15
movq -48(%rsp), %rbx #[spill]
.cfi_restore 3
# LOE rbx rbp r15
..B1.30: # Preds ..B1.1 ..B1.28
# Execution count [1.00e+00]
vzeroupper #93.12
vxorpd %xmm0, %xmm0, %xmm0 #93.12
.cfi_restore 14
popq %r14 #93.12
.cfi_def_cfa_offset 24
.cfi_restore 13
popq %r13 #93.12
.cfi_def_cfa_offset 16
.cfi_restore 12
popq %r12 #93.12
.cfi_def_cfa_offset 8
ret #93.12
.cfi_def_cfa_offset 32
.cfi_offset 3, -80
.cfi_offset 12, -16
.cfi_offset 13, -24
.cfi_offset 14, -32
.cfi_offset 15, -72
# LOE
..B1.31: # Preds ..B1.11
# Execution count [4.50e-01]: Infreq
movl %r13d, %ebx #67.9
xorl %r9d, %r9d #67.9
andl $-8, %ebx #67.9
jmp ..B1.20 # Prob 100% #67.9
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 ebx r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.32: # Preds ..B1.10
# Execution count [4.50e-01]: Infreq
xorl %ebx, %ebx #67.9
jmp ..B1.24 # Prob 100% #67.9
.align 16,0x90
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 ebx r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
.cfi_endproc
# mark_end;
.type computeForce,@function
.size computeForce,.-computeForce
..LNcomputeForce.0:
.data
# -- End computeForce
.section .rodata, "a"
.align 64
.align 64
.L_2il0floatpacket.2:
.long 0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000
.type .L_2il0floatpacket.2,@object
.size .L_2il0floatpacket.2,64
.align 64
.L_2il0floatpacket.4:
.long 0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000
.type .L_2il0floatpacket.4,@object
.size .L_2il0floatpacket.4,64
.align 64
.L_2il0floatpacket.6:
.long 0x00000008,0x00000009,0x0000000a,0x0000000b,0x0000000c,0x0000000d,0x0000000e,0x0000000f,0x00000008,0x00000009,0x0000000a,0x0000000b,0x0000000c,0x0000000d,0x0000000e,0x0000000f
.type .L_2il0floatpacket.6,@object
.size .L_2il0floatpacket.6,64
.align 32
.L_2il0floatpacket.0:
.long 0x00000008,0x00000008,0x00000008,0x00000008,0x00000008,0x00000008,0x00000008,0x00000008
.type .L_2il0floatpacket.0,@object
.size .L_2il0floatpacket.0,32
.align 32
.L_2il0floatpacket.1:
.long 0x00000000,0x00000001,0x00000002,0x00000003,0x00000004,0x00000005,0x00000006,0x00000007
.type .L_2il0floatpacket.1,@object
.size .L_2il0floatpacket.1,32
.align 8
.L_2il0floatpacket.3:
.long 0x00000000,0x40480000
.type .L_2il0floatpacket.3,@object
.size .L_2il0floatpacket.3,8
.align 8
.L_2il0floatpacket.5:
.long 0x00000000,0x3ff00000
.type .L_2il0floatpacket.5,@object
.size .L_2il0floatpacket.5,8
.data
.section .note.GNU-stack, ""
# End

541
asm/avx512/force_soa_intel.s
View File

@ -1,541 +0,0 @@
# mark_description "Intel(R) C Intel(R) 64 Compiler for applications running on Intel(R) 64, Version 19.0.5.281 Build 20190815";
# mark_description "-I./src/includes -S -masm=intel -D_GNU_SOURCE -DPRECISION=2 -DALIGNMENT=64 -restrict -Ofast -xCORE-AVX512 -q";
# mark_description "opt-zmm-usage=high -o ICC/force.s";
.intel_syntax noprefix
.file "force.c"
.text
..TXTST0:
.L_2__routine_start_computeForce_0:
# -- Begin computeForce
.text
# mark_begin;
.align 16,0x90
.globl computeForce
# --- computeForce(Parameter *, Atom *, Neighbor *, int)
computeForce:
# parameter 1: rdi
# parameter 2: rsi
# parameter 3: rdx
# parameter 4: ecx
..B1.1: # Preds ..B1.0
# Execution count [1.00e+00]
.cfi_startproc
..___tag_value_computeForce.1:
..L2:
#35.1
push r12 #35.1
.cfi_def_cfa_offset 16
.cfi_offset 12, -16
push r13 #35.1
.cfi_def_cfa_offset 24
.cfi_offset 13, -24
push r14 #35.1
.cfi_def_cfa_offset 32
.cfi_offset 14, -32
# r9d <- atom->Nlocal
mov r9d, DWORD PTR [4+rsi] #36.18
# xmm2 <- param->cutforce
vmovsd xmm2, QWORD PTR [72+rdi] #38.27
# xmm1 <- param->sigma6
vmovsd xmm1, QWORD PTR [8+rdi] #39.23
# xmm0 <- param->epsilon
vmovsd xmm0, QWORD PTR [rdi] #40.24
# r13 <- atom->fx
mov r13, QWORD PTR [64+rsi] #41.20
# r14 <- atom->fy
mov r14, QWORD PTR [72+rsi] #41.45
# rdi <- atom->fz
mov rdi, QWORD PTR [80+rsi] #41.70
# atom->Nlocal <= 0
test r9d, r9d #44.24
jle ..B1.30 # Prob 50% #44.24
# LOE rdx rbx rbp rsi rdi r13 r14 r15 r9d xmm0 xmm1 xmm2
..B1.2: # Preds ..B1.1
# Execution count [1.00e+00]
# r10d <- 0
xor r10d, r10d #44.5
# ecx <- atom->Nlocal
mov ecx, r9d #44.5
# r8d <- 0
xor r8d, r8d #44.5
# r11d <- 1
mov r11d, 1 #44.5
# eax <- 0
xor eax, eax #45.17
# ecx <- atom->Nlocal >> 1
shr ecx, 1 #44.5
je ..B1.6 # Prob 9% #44.5
# LOE rax rdx rcx rbx rbp rsi rdi r8 r10 r13 r14 r15 r9d r11d xmm0 xmm1 xmm2
..B1.4: # Preds ..B1.2 ..B1.4
# Execution count [2.50e+00]
# fx[i] <- 0
mov QWORD PTR [r8+r13], rax #45.9
# i++
inc r10 #44.5
# fy[i] <- 0
mov QWORD PTR [r8+r14], rax #46.9
# fz[i] <- 0
mov QWORD PTR [r8+rdi], rax #47.9
# fx[i] <- 0
mov QWORD PTR [8+r8+r13], rax #45.9
# fy[i] <- 0
mov QWORD PTR [8+r8+r14], rax #46.9
# fz[i] <- 0
mov QWORD PTR [8+r8+rdi], rax #47.9
# i++
add r8, 16 #44.5
# i < Nlocal
cmp r10, rcx #44.5
jb ..B1.4 # Prob 63% #44.5
# LOE rax rdx rcx rbx rbp rsi rdi r8 r10 r13 r14 r15 r9d xmm0 xmm1 xmm2
..B1.5: # Preds ..B1.4
# Execution count [9.00e-01]
# r11d <- i * 2 + 1
lea r11d, DWORD PTR [1+r10+r10] #45.9
# LOE rax rdx rbx rbp rsi rdi r13 r14 r15 r9d r11d xmm0 xmm1 xmm2
..B1.6: # Preds ..B1.2 ..B1.5
# Execution count [1.00e+00]
# r11d <- i * 2
lea ecx, DWORD PTR [-1+r11] #44.5
# i < Nlocal
cmp ecx, r9d #44.5
jae ..B1.8 # Prob 9% #44.5
# LOE rax rdx rbx rbp rsi rdi r13 r14 r15 r9d r11d xmm0 xmm1 xmm2
..B1.7: # Preds ..B1.6
# Execution count [9.00e-01]
# r11 <- i * 2
movsxd r11, r11d #44.5
# fx[i] <- 0
mov QWORD PTR [-8+r13+r11*8], rax #45.9
# fy[i] <- 0
mov QWORD PTR [-8+r14+r11*8], rax #46.9
# fz[i] <- 0
mov QWORD PTR [-8+rdi+r11*8], rax #47.9
# LOE rdx rbx rbp rsi rdi r13 r14 r15 r9d xmm0 xmm1 xmm2
..B1.8: # Preds ..B1.6 ..B1.7
# Execution count [9.00e-01]
# xmm15 <- cutforcesq
vmulsd xmm15, xmm2, xmm2 #38.45
# r8d <- 0
xor r8d, r8d #55.15
# ymm18 <- 8
vmovdqu32 ymm18, YMMWORD PTR .L_2il0floatpacket.0[rip] #67.9
# xmm0 <- 48 * sr6 * epsilon
vmulsd xmm0, xmm0, QWORD PTR .L_2il0floatpacket.3[rip] #77.41
# ymm17 <- [0..7]
vmovdqu32 ymm17, YMMWORD PTR .L_2il0floatpacket.1[rip] #67.9
# zmm7 <- 0.5
vmovups zmm7, ZMMWORD PTR .L_2il0floatpacket.4[rip] #77.54
# zmm16 <- cutforcesq
vbroadcastsd zmm16, xmm15 #38.25
# zmm15 <- param->sigma6
vbroadcastsd zmm15, xmm1 #39.21
# zmm16 <- 48 * sr6 * epsilon
vbroadcastsd zmm14, xmm0 #77.41
# r9 <- atom->Nlocal
movsxd r9, r9d #55.5
# r10d <- 0 (i)
xor r10d, r10d #55.5
# rcx <- neighbor->numneigh
mov rcx, QWORD PTR [24+rdx] #57.25
# r11 <- neighbor->neighbors
mov r11, QWORD PTR [8+rdx] #56.19
# r12 <- neighbor->maxneighs
movsxd r12, DWORD PTR [16+rdx] #56.43
# rdx <- atom->x
mov rdx, QWORD PTR [16+rsi] #58.25
# rax <- atom->y
mov rax, QWORD PTR [24+rsi] #59.25
# rsi <- atom->z
mov rsi, QWORD PTR [32+rsi] #60.25
# r12 <- neighbor->maxneighs * 4
shl r12, 2 #37.5
# [-32+rsp] <- atom->Nlocal
mov QWORD PTR [-32+rsp], r9 #60.25[spill]
# [-24+rsp] <- neighbor->numneigh
mov QWORD PTR [-24+rsp], rcx #60.25[spill]
# [-16+rsp] <- atom->fy
mov QWORD PTR [-16+rsp], r14 #60.25[spill]
# [-8+rsp] <- atom->fx
mov QWORD PTR [-8+rsp], r13 #60.25[spill]
# [-40+rsp] <- r15
mov QWORD PTR [-40+rsp], r15 #60.25[spill]
# [-48+rsp] <- rbx
mov QWORD PTR [-48+rsp], rbx #60.25[spill]
# zmm19 <- 0
vpxord zmm19, zmm19, zmm19 #61.22
.cfi_offset 3, -80
.cfi_offset 15, -72
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 ymm17 ymm18 zmm7 zmm14 zmm15 zmm16 zmm19
..B1.9: # Preds ..B1.27 ..B1.8
# Execution count [5.00e+00]
# rcx <- neighbor->numneigh
mov rcx, QWORD PTR [-24+rsp] #57.25[spill]
# xmm25 <- 0
vxorpd xmm25, xmm25, xmm25 #61.22
# xmm20 <- 0
vmovapd xmm20, xmm25 #62.22
# r13d <- neighbor->numneigh[i] (numneighs)
mov r13d, DWORD PTR [rcx+r10*4] #57.25
# xmm4 <- 0
vmovapd xmm4, xmm20 #63.22
# xmm8 <- atom->x[i]
vmovsd xmm8, QWORD PTR [rdx+r10*8] #58.25
# xmm9 <- atom->y[i]
vmovsd xmm9, QWORD PTR [rax+r10*8] #59.25
# xmm9 <- atom->z[i]
vmovsd xmm10, QWORD PTR [rsi+r10*8] #60.25
# numneighs <= 0
test r13d, r13d #67.28
jle ..B1.27 # Prob 50% #67.28
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 r13d xmm4 xmm8 xmm9 xmm10 xmm20 xmm25 ymm17 ymm18 zmm7 zmm14 zmm15 zmm16 zmm19
..B1.10: # Preds ..B1.9
# Execution count [4.50e+00]
# zmm13 <- 0
vmovaps zmm13, zmm19 #61.22
# zmm12 <- 0
vmovaps zmm12, zmm13 #62.22
# zmm11 <- 0
vmovaps zmm11, zmm12 #63.22
# numneighs < 8
cmp r13d, 8 #67.9
jl ..B1.32 # Prob 10% #67.9
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.11: # Preds ..B1.10
# Execution count [4.50e+00]
# numneighs < 1200
cmp r13d, 1200 #67.9
jl ..B1.31 # Prob 10% #67.9
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.12: # Preds ..B1.11
# Execution count [4.50e+00]
mov rcx, r12 #56.43
imul rcx, r8 #56.43
add rcx, r11 #37.5
mov r9, rcx #67.9
and r9, 63 #67.9
test r9d, 3 #67.9
je ..B1.14 # Prob 50% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.13: # Preds ..B1.12
# Execution count [2.25e+00]
xor r9d, r9d #67.9
jmp ..B1.16 # Prob 100% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.14: # Preds ..B1.12
# Execution count [2.25e+00]
test r9d, r9d #67.9
je ..B1.16 # Prob 50% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.15: # Preds ..B1.14
# Execution count [2.50e+01]
neg r9d #67.9
add r9d, 64 #67.9
shr r9d, 2 #67.9
cmp r13d, r9d #67.9
cmovl r9d, r13d #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.16: # Preds ..B1.13 ..B1.15 ..B1.14
# Execution count [5.00e+00]
mov ebx, r13d #67.9
sub ebx, r9d #67.9
and ebx, 7 #67.9
neg ebx #67.9
add ebx, r13d #67.9
cmp r9d, 1 #67.9
jb ..B1.20 # Prob 50% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 ebx r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.17: # Preds ..B1.16
# Execution count [4.50e+00]
vmovdqa32 ymm4, ymm17 #67.9
xor r15d, r15d #67.9
vpbroadcastd ymm3, r9d #67.9
vbroadcastsd zmm2, xmm8 #58.23
vbroadcastsd zmm5, xmm9 #59.23
vbroadcastsd zmm6, xmm10 #60.23
movsxd r14, r9d #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 r14 r15 ebx r9d r13d xmm8 xmm9 xmm10 ymm3 ymm4 ymm17 ymm18 zmm2 zmm5 zmm6 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.18: # Preds ..B1.18 ..B1.17
# Execution count [2.50e+01]
vpcmpgtd k5, ymm3, ymm4 #67.9
vpaddd ymm4, ymm4, ymm18 #67.9
vmovdqu32 ymm20{k5}{z}, YMMWORD PTR [rcx+r15*4] #68.21
vmovaps zmm22, zmm19 #70.36
add r15, 8 #67.9
kmovw k2, k5 #70.36
vmovaps zmm21, zmm19 #69.36
kmovw k1, k5 #69.36
vmovaps zmm23, zmm19 #71.36
kmovw k3, k5 #71.36
vgatherdpd zmm23{k3}, QWORD PTR [rsi+ymm20*8] #71.36
vgatherdpd zmm22{k2}, QWORD PTR [rax+ymm20*8] #70.36
vgatherdpd zmm21{k1}, QWORD PTR [rdx+ymm20*8] #69.36
vsubpd zmm0, zmm5, zmm22 #70.36
vsubpd zmm1, zmm2, zmm21 #69.36
vsubpd zmm21, zmm6, zmm23 #71.36
vmulpd zmm20, zmm0, zmm0 #72.49
vfmadd231pd zmm20, zmm1, zmm1 #72.49
vfmadd231pd zmm20, zmm21, zmm21 #72.63
vrcp14pd zmm31, zmm20 #75.38
vcmppd k6{k5}, zmm20, zmm16, 1 #74.22
vfpclasspd k0, zmm31, 30 #75.38
vmovaps zmm24, zmm20 #75.38
vfnmadd213pd zmm24, zmm31, QWORD BCST .L_2il0floatpacket.5[rip] #75.38
knotw k4, k0 #75.38
vmulpd zmm25, zmm24, zmm24 #75.38
vfmadd213pd zmm31{k4}, zmm24, zmm31 #75.38
vfmadd213pd zmm31{k4}, zmm25, zmm31 #75.38
vmulpd zmm26, zmm31, zmm15 #76.38
vmulpd zmm28, zmm31, zmm14 #77.54
vmulpd zmm29, zmm31, zmm26 #76.44
vmulpd zmm27, zmm31, zmm29 #76.50
vfmsub213pd zmm31, zmm29, zmm7 #77.54
vmulpd zmm30, zmm27, zmm28 #77.61
vmulpd zmm24, zmm30, zmm31 #77.67
vfmadd231pd zmm13{k6}, zmm24, zmm1 #78.17
vfmadd231pd zmm12{k6}, zmm24, zmm0 #79.17
vfmadd231pd zmm11{k6}, zmm24, zmm21 #80.17
cmp r15, r14 #67.9
jb ..B1.18 # Prob 82% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 r14 r15 ebx r9d r13d xmm8 xmm9 xmm10 ymm3 ymm4 ymm17 ymm18 zmm2 zmm5 zmm6 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.19: # Preds ..B1.18
# Execution count [4.50e+00]
cmp r13d, r9d #67.9
je ..B1.26 # Prob 10% #67.9
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 ebx r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.20: # Preds ..B1.19 ..B1.16 ..B1.31
# Execution count [2.50e+01]
lea ecx, DWORD PTR [8+r9] #67.9
cmp ebx, ecx #67.9
jl ..B1.24 # Prob 50% #67.9
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 ebx r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.21: # Preds ..B1.20
# Execution count [4.50e+00]
mov rcx, r12 #56.43
imul rcx, r8 #56.43
vbroadcastsd zmm0, xmm8 #58.23
vbroadcastsd zmm1, xmm9 #59.23
vbroadcastsd zmm2, xmm10 #60.23
movsxd r14, r9d #67.9
add rcx, r11 #37.5
# LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 r14 ebx r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm0 zmm1 zmm2 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.22: # Preds ..B1.22 ..B1.21
# Execution count [2.50e+01]
vpcmpeqb k2, xmm0, xmm0 #70.36
add r9d, 8 #67.9
vpcmpeqb k1, xmm0, xmm0 #69.36
vpcmpeqb k3, xmm0, xmm0 #71.36
vmovdqu ymm3, YMMWORD PTR [rcx+r14*4] #68.21
add r14, 8 #67.9
vpxord zmm5, zmm5, zmm5 #70.36
vpxord zmm4, zmm4, zmm4 #69.36
vpxord zmm6, zmm6, zmm6 #71.36
vgatherdpd zmm5{k2}, QWORD PTR [rax+ymm3*8] #70.36
vgatherdpd zmm4{k1}, QWORD PTR [rdx+ymm3*8] #69.36
vgatherdpd zmm6{k3}, QWORD PTR [rsi+ymm3*8] #71.36
vsubpd zmm29, zmm1, zmm5 #70.36
vsubpd zmm28, zmm0, zmm4 #69.36
vsubpd zmm31, zmm2, zmm6 #71.36
vmulpd zmm20, zmm29, zmm29 #72.49
vfmadd231pd zmm20, zmm28, zmm28 #72.49
vfmadd231pd zmm20, zmm31, zmm31 #72.63
vrcp14pd zmm27, zmm20 #75.38
vcmppd k5, zmm20, zmm16, 1 #74.22
vfpclasspd k0, zmm27, 30 #75.38
vfnmadd213pd zmm20, zmm27, QWORD BCST .L_2il0floatpacket.5[rip] #75.38
knotw k4, k0 #75.38
vmulpd zmm21, zmm20, zmm20 #75.38
vfmadd213pd zmm27{k4}, zmm20, zmm27 #75.38
vfmadd213pd zmm27{k4}, zmm21, zmm27 #75.38
vmulpd zmm22, zmm27, zmm15 #76.38
vmulpd zmm24, zmm27, zmm14 #77.54
vmulpd zmm25, zmm27, zmm22 #76.44
vmulpd zmm23, zmm27, zmm25 #76.50
vfmsub213pd zmm27, zmm25, zmm7 #77.54
vmulpd zmm26, zmm23, zmm24 #77.61
vmulpd zmm30, zmm26, zmm27 #77.67
vfmadd231pd zmm13{k5}, zmm30, zmm28 #78.17
vfmadd231pd zmm12{k5}, zmm30, zmm29 #79.17
vfmadd231pd zmm11{k5}, zmm30, zmm31 #80.17
cmp r9d, ebx #67.9
jb ..B1.22 # Prob 82% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 r14 ebx r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm0 zmm1 zmm2 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.24: # Preds ..B1.22 ..B1.20 ..B1.32
# Execution count [5.00e+00]
lea ecx, DWORD PTR [1+rbx] #67.9
cmp ecx, r13d #67.9
ja ..B1.26 # Prob 50% #67.9
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 ebx r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.25: # Preds ..B1.24
# Execution count [2.50e+01]
imul r8, r12 #56.43
vbroadcastsd zmm9, xmm9 #59.23
vbroadcastsd zmm2, xmm8 #58.23
vbroadcastsd zmm10, xmm10 #60.23
sub r13d, ebx #67.9
add r8, r11 #37.5
vpbroadcastd ymm0, r13d #67.9
vpcmpgtd k5, ymm0, ymm17 #67.9
movsxd rbx, ebx #67.9
vmovaps zmm4, zmm19 #70.36
kmovw k2, k5 #70.36
vmovaps zmm3, zmm19 #69.36
vmovdqu32 ymm1{k5}{z}, YMMWORD PTR [r8+rbx*4] #68.21
kmovw k1, k5 #69.36
vmovaps zmm5, zmm19 #71.36
kmovw k3, k5 #71.36
vgatherdpd zmm5{k3}, QWORD PTR [rsi+ymm1*8] #71.36
vgatherdpd zmm4{k2}, QWORD PTR [rax+ymm1*8] #70.36
vgatherdpd zmm3{k1}, QWORD PTR [rdx+ymm1*8] #69.36
vsubpd zmm30, zmm10, zmm5 #71.36
vsubpd zmm28, zmm9, zmm4 #70.36
vsubpd zmm27, zmm2, zmm3 #69.36
vmulpd zmm26, zmm28, zmm28 #72.49
vfmadd231pd zmm26, zmm27, zmm27 #72.49
vfmadd231pd zmm26, zmm30, zmm30 #72.63
vrcp14pd zmm25, zmm26 #75.38
vcmppd k6{k5}, zmm26, zmm16, 1 #74.22
vfpclasspd k0, zmm25, 30 #75.38
vmovaps zmm6, zmm26 #75.38
vfnmadd213pd zmm6, zmm25, QWORD BCST .L_2il0floatpacket.5[rip] #75.38
knotw k4, k0 #75.38
vmulpd zmm8, zmm6, zmm6 #75.38
vfmadd213pd zmm25{k4}, zmm6, zmm25 #75.38
vfmadd213pd zmm25{k4}, zmm8, zmm25 #75.38
vmulpd zmm20, zmm25, zmm15 #76.38
vmulpd zmm22, zmm25, zmm14 #77.54
vmulpd zmm23, zmm25, zmm20 #76.44
vmulpd zmm21, zmm25, zmm23 #76.50
vfmsub213pd zmm25, zmm23, zmm7 #77.54
vmulpd zmm24, zmm21, zmm22 #77.61
vmulpd zmm29, zmm24, zmm25 #77.67
vfmadd231pd zmm13{k6}, zmm29, zmm27 #78.17
vfmadd231pd zmm12{k6}, zmm29, zmm28 #79.17
vfmadd231pd zmm11{k6}, zmm29, zmm30 #80.17
# LOE rax rdx rbp rsi rdi r10 r11 r12 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.26: # Preds ..B1.19 ..B1.25 ..B1.24
# Execution count [4.50e+00]
vmovups zmm10, ZMMWORD PTR .L_2il0floatpacket.6[rip] #63.22
vpermd zmm0, zmm10, zmm11 #63.22
vpermd zmm5, zmm10, zmm12 #62.22
vpermd zmm21, zmm10, zmm13 #61.22
vaddpd zmm11, zmm0, zmm11 #63.22
vaddpd zmm12, zmm5, zmm12 #62.22
vaddpd zmm13, zmm21, zmm13 #61.22
vpermpd zmm1, zmm11, 78 #63.22
vpermpd zmm6, zmm12, 78 #62.22
vpermpd zmm22, zmm13, 78 #61.22
vaddpd zmm2, zmm11, zmm1 #63.22
vaddpd zmm8, zmm12, zmm6 #62.22
vaddpd zmm23, zmm13, zmm22 #61.22
vpermpd zmm3, zmm2, 177 #63.22
vpermpd zmm9, zmm8, 177 #62.22
vpermpd zmm24, zmm23, 177 #61.22
vaddpd zmm4, zmm2, zmm3 #63.22
vaddpd zmm20, zmm8, zmm9 #62.22
vaddpd zmm25, zmm23, zmm24 #61.22
# LOE rax rdx rbp rsi rdi r10 r11 r12 xmm4 xmm20 xmm25 ymm17 ymm18 zmm7 zmm14 zmm15 zmm16 zmm19
..B1.27: # Preds ..B1.26 ..B1.9
# Execution count [5.00e+00]
mov rcx, QWORD PTR [-8+rsp] #84.9[spill]
mov rbx, QWORD PTR [-16+rsp] #85.9[spill]
movsxd r8, r10d #55.32
inc r8 #55.32
vaddsd xmm0, xmm25, QWORD PTR [rcx+r10*8] #84.9
vmovsd QWORD PTR [rcx+r10*8], xmm0 #84.9
vaddsd xmm1, xmm20, QWORD PTR [rbx+r10*8] #85.9
vmovsd QWORD PTR [rbx+r10*8], xmm1 #85.9
vaddsd xmm2, xmm4, QWORD PTR [rdi+r10*8] #86.9
vmovsd QWORD PTR [rdi+r10*8], xmm2 #86.9
inc r10 #55.5
cmp r10, QWORD PTR [-32+rsp] #55.5[spill]
jb ..B1.9 # Prob 82% #55.5
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 ymm17 ymm18 zmm7 zmm14 zmm15 zmm16 zmm19
..B1.28: # Preds ..B1.27
# Execution count [9.00e-01]
mov r15, QWORD PTR [-40+rsp] #[spill]
.cfi_restore 15
mov rbx, QWORD PTR [-48+rsp] #[spill]
.cfi_restore 3
# LOE rbx rbp r15
..B1.30: # Preds ..B1.1 ..B1.28
# Execution count [1.00e+00]
vzeroupper #93.12
vxorpd xmm0, xmm0, xmm0 #93.12
.cfi_restore 14
pop r14 #93.12
.cfi_def_cfa_offset 24
.cfi_restore 13
pop r13 #93.12
.cfi_def_cfa_offset 16
.cfi_restore 12
pop r12 #93.12
.cfi_def_cfa_offset 8
ret #93.12
.cfi_def_cfa_offset 32
.cfi_offset 3, -80
.cfi_offset 12, -16
.cfi_offset 13, -24
.cfi_offset 14, -32
.cfi_offset 15, -72
# LOE
..B1.31: # Preds ..B1.11
# Execution count [4.50e-01]: Infreq
mov ebx, r13d #67.9
xor r9d, r9d #67.9
and ebx, -8 #67.9
jmp ..B1.20 # Prob 100% #67.9
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 ebx r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.32: # Preds ..B1.10
# Execution count [4.50e-01]: Infreq
xor ebx, ebx #67.9
jmp ..B1.24 # Prob 100% #67.9
.align 16,0x90
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 ebx r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
.cfi_endproc
# mark_end;
.type computeForce,@function
.size computeForce,.-computeForce
..LNcomputeForce.0:
.data
# -- End computeForce
.section .rodata, "a"
.align 64
.align 64
.L_2il0floatpacket.2:
.long 0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000
.type .L_2il0floatpacket.2,@object
.size .L_2il0floatpacket.2,64
.align 64
.L_2il0floatpacket.4:
.long 0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000
.type .L_2il0floatpacket.4,@object
.size .L_2il0floatpacket.4,64
.align 64
.L_2il0floatpacket.6:
.long 0x00000008,0x00000009,0x0000000a,0x0000000b,0x0000000c,0x0000000d,0x0000000e,0x0000000f,0x00000008,0x00000009,0x0000000a,0x0000000b,0x0000000c,0x0000000d,0x0000000e,0x0000000f
.type .L_2il0floatpacket.6,@object
.size .L_2il0floatpacket.6,64
.align 32
.L_2il0floatpacket.0:
.long 0x00000008,0x00000008,0x00000008,0x00000008,0x00000008,0x00000008,0x00000008,0x00000008
.type .L_2il0floatpacket.0,@object
.size .L_2il0floatpacket.0,32
.align 32
.L_2il0floatpacket.1:
.long 0x00000000,0x00000001,0x00000002,0x00000003,0x00000004,0x00000005,0x00000006,0x00000007
.type .L_2il0floatpacket.1,@object
.size .L_2il0floatpacket.1,32
.align 8
.L_2il0floatpacket.3:
.long 0x00000000,0x40480000
.type .L_2il0floatpacket.3,@object
.size .L_2il0floatpacket.3,8
.align 8
.L_2il0floatpacket.5:
.long 0x00000000,0x3ff00000
.type .L_2il0floatpacket.5,@object
.size .L_2il0floatpacket.5,8
.data
.section .note.GNU-stack, ""
# End

475
asm/avx512/force_soa_lt1200_markers.s
View File

@ -1,475 +0,0 @@
# mark_description "Intel(R) C Intel(R) 64 Compiler for applications running on Intel(R) 64, Version 19.0.5.281 Build 20190815";
# mark_description "-I./src/includes -S -D_GNU_SOURCE -DPRECISION=2 -DALIGNMENT=64 -restrict -Ofast -xCORE-AVX512 -qopt-zmm-usag";
# mark_description "e=high -o ICC/force.s";
.file "force.c"
.text
..TXTST0:
.L_2__routine_start_computeForce_0:
# -- Begin computeForce
.text
# mark_begin;
.align 16,0x90
.globl computeForce
# --- computeForce(Parameter *, Atom *, Neighbor *, int)
computeForce:
# parameter 1: %rdi
# parameter 2: %rsi
# parameter 3: %rdx
# parameter 4: %ecx
..B1.1: # Preds ..B1.0
# Execution count [1.00e+00]
.cfi_startproc
..___tag_value_computeForce.1:
..L2:
#35.1
pushq %r12 #35.1
.cfi_def_cfa_offset 16
.cfi_offset 12, -16
pushq %r13 #35.1
.cfi_def_cfa_offset 24
.cfi_offset 13, -24
pushq %r14 #35.1
.cfi_def_cfa_offset 32
.cfi_offset 14, -32
movl 4(%rsi), %r9d #36.18
vmovsd 72(%rdi), %xmm2 #38.27
vmovsd 8(%rdi), %xmm1 #39.23
vmovsd (%rdi), %xmm0 #40.24
movq 64(%rsi), %r13 #41.20
movq 72(%rsi), %r14 #41.45
movq 80(%rsi), %rdi #41.70
testl %r9d, %r9d #44.24
jle ..B1.30 # Prob 50% #44.24
# LOE rdx rbx rbp rsi rdi r13 r14 r15 r9d xmm0 xmm1 xmm2
..B1.2: # Preds ..B1.1
# Execution count [1.00e+00]
xorl %r10d, %r10d #44.5
movl %r9d, %ecx #44.5
xorl %r8d, %r8d #44.5
movl $1, %r11d #44.5
xorl %eax, %eax #45.17
shrl $1, %ecx #44.5
je ..B1.6 # Prob 9% #44.5
# LOE rax rdx rcx rbx rbp rsi rdi r8 r10 r13 r14 r15 r9d r11d xmm0 xmm1 xmm2
..B1.4: # Preds ..B1.2 ..B1.4
# Execution count [2.50e+00]
movq %rax, (%r8,%r13) #45.9
incq %r10 #44.5
movq %rax, (%r8,%r14) #46.9
movq %rax, (%r8,%rdi) #47.9
movq %rax, 8(%r8,%r13) #45.9
movq %rax, 8(%r8,%r14) #46.9
movq %rax, 8(%r8,%rdi) #47.9
addq $16, %r8 #44.5
cmpq %rcx, %r10 #44.5
jb ..B1.4 # Prob 63% #44.5
# LOE rax rdx rcx rbx rbp rsi rdi r8 r10 r13 r14 r15 r9d xmm0 xmm1 xmm2
..B1.5: # Preds ..B1.4
# Execution count [9.00e-01]
lea 1(%r10,%r10), %r11d #45.9
# LOE rax rdx rbx rbp rsi rdi r13 r14 r15 r9d r11d xmm0 xmm1 xmm2
..B1.6: # Preds ..B1.2 ..B1.5
# Execution count [1.00e+00]
lea -1(%r11), %ecx #44.5
cmpl %r9d, %ecx #44.5
jae ..B1.8 # Prob 9% #44.5
# LOE rax rdx rbx rbp rsi rdi r13 r14 r15 r9d r11d xmm0 xmm1 xmm2
..B1.7: # Preds ..B1.6
# Execution count [9.00e-01]
movslq %r11d, %r11 #44.5
movq %rax, -8(%r13,%r11,8) #45.9
movq %rax, -8(%r14,%r11,8) #46.9
movq %rax, -8(%rdi,%r11,8) #47.9
# LOE rdx rbx rbp rsi rdi r13 r14 r15 r9d xmm0 xmm1 xmm2
..B1.8: # Preds ..B1.6 ..B1.7
# Execution count [9.00e-01]
vmulsd %xmm2, %xmm2, %xmm15 #38.45
xorl %r8d, %r8d #55.15
vmovdqu32 .L_2il0floatpacket.0(%rip), %ymm18 #67.9
vmulsd .L_2il0floatpacket.3(%rip), %xmm0, %xmm0 #77.41
vmovdqu32 .L_2il0floatpacket.1(%rip), %ymm17 #67.9
vmovups .L_2il0floatpacket.4(%rip), %zmm7 #77.54
vbroadcastsd %xmm15, %zmm16 #38.25
vbroadcastsd %xmm1, %zmm15 #39.21
vbroadcastsd %xmm0, %zmm14 #77.41
movslq %r9d, %r9 #55.5
xorl %r10d, %r10d #55.5
movq 24(%rdx), %rcx #57.25
movq 8(%rdx), %r11 #56.19
movslq 16(%rdx), %r12 #56.43
movq 16(%rsi), %rdx #58.25
movq 24(%rsi), %rax #59.25
movq 32(%rsi), %rsi #60.25
shlq $2, %r12 #37.5
movq %r9, -32(%rsp) #60.25[spill]
movq %rcx, -24(%rsp) #60.25[spill]
movq %r14, -16(%rsp) #60.25[spill]
movq %r13, -8(%rsp) #60.25[spill]
movq %r15, -40(%rsp) #60.25[spill]
movq %rbx, -48(%rsp) #60.25[spill]
vpxord %zmm19, %zmm19, %zmm19 #61.22
.cfi_offset 3, -80
.cfi_offset 15, -72
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 ymm17 ymm18 zmm7 zmm14 zmm15 zmm16 zmm19
..B1.9: # Preds ..B1.27 ..B1.8
# Execution count [5.00e+00]
movq -24(%rsp), %rcx #57.25[spill]
vxorpd %xmm25, %xmm25, %xmm25 #61.22
vmovapd %xmm25, %xmm20 #62.22
movl (%rcx,%r10,4), %r13d #57.25
vmovapd %xmm20, %xmm4 #63.22
vmovsd (%rdx,%r10,8), %xmm8 #58.25
vmovsd (%rax,%r10,8), %xmm9 #59.25
vmovsd (%rsi,%r10,8), %xmm10 #60.25
testl %r13d, %r13d #67.28
jle ..B1.27 # Prob 50% #67.28
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 r13d xmm4 xmm8 xmm9 xmm10 xmm20 xmm25 ymm17 ymm18 zmm7 zmm14 zmm15 zmm16 zmm19
..B1.10: # Preds ..B1.9
# Execution count [4.50e+00]
vmovaps %zmm19, %zmm13 #61.22
vmovaps %zmm13, %zmm12 #62.22
vmovaps %zmm12, %zmm11 #63.22
cmpl $8, %r13d #67.9
jl ..B1.32 # Prob 10% #67.9
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.11: # Preds ..B1.10
# Execution count [4.50e+00]
cmpl $1200, %r13d #67.9
jl ..B1.31 # Prob 10% #67.9
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.12: # Preds ..B1.11
# Execution count [4.50e+00]
movq %r12, %rcx #56.43
imulq %r8, %rcx #56.43
addq %r11, %rcx #37.5
movq %rcx, %r9 #67.9
andq $63, %r9 #67.9
testl $3, %r9d #67.9
je ..B1.14 # Prob 50% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.13: # Preds ..B1.12
# Execution count [2.25e+00]
xorl %r9d, %r9d #67.9
jmp ..B1.16 # Prob 100% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.14: # Preds ..B1.12
# Execution count [2.25e+00]
testl %r9d, %r9d #67.9
je ..B1.16 # Prob 50% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.15: # Preds ..B1.14
# Execution count [2.50e+01]
negl %r9d #67.9
addl $64, %r9d #67.9
shrl $2, %r9d #67.9
cmpl %r9d, %r13d #67.9
cmovl %r13d, %r9d #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.16: # Preds ..B1.13 ..B1.15 ..B1.14
# Execution count [5.00e+00]
movl %r13d, %ebx #67.9
subl %r9d, %ebx #67.9
andl $7, %ebx #67.9
negl %ebx #67.9
addl %r13d, %ebx #67.9
cmpl $1, %r9d #67.9
jb ..B1.20 # Prob 50% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 ebx r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.17: # Preds ..B1.16
# Execution count [4.50e+00]
vmovdqa32 %ymm17, %ymm4 #67.9
xorl %r15d, %r15d #67.9
vpbroadcastd %r9d, %ymm3 #67.9
vbroadcastsd %xmm8, %zmm2 #58.23
vbroadcastsd %xmm9, %zmm5 #59.23
vbroadcastsd %xmm10, %zmm6 #60.23
movslq %r9d, %r14 #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 r14 r15 ebx r9d r13d xmm8 xmm9 xmm10 ymm3 ymm4 ymm17 ymm18 zmm2 zmm5 zmm6 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.18: # Preds ..B1.18 ..B1.17
# Execution count [2.50e+01]
vpcmpgtd %ymm4, %ymm3, %k5 #67.9
vpaddd %ymm18, %ymm4, %ymm4 #67.9
vmovdqu32 (%rcx,%r15,4), %ymm20{%k5}{z} #68.21
vmovaps %zmm19, %zmm22 #70.36
addq $8, %r15 #67.9
kmovw %k5, %k2 #70.36
vmovaps %zmm19, %zmm21 #69.36
kmovw %k5, %k1 #69.36
vmovaps %zmm19, %zmm23 #71.36
kmovw %k5, %k3 #71.36
vgatherdpd (%rsi,%ymm20,8), %zmm23{%k3} #71.36
vgatherdpd (%rax,%ymm20,8), %zmm22{%k2} #70.36
vgatherdpd (%rdx,%ymm20,8), %zmm21{%k1} #69.36
vsubpd %zmm22, %zmm5, %zmm0 #70.36
vsubpd %zmm21, %zmm2, %zmm1 #69.36
vsubpd %zmm23, %zmm6, %zmm21 #71.36
vmulpd %zmm0, %zmm0, %zmm20 #72.49
vfmadd231pd %zmm1, %zmm1, %zmm20 #72.49
vfmadd231pd %zmm21, %zmm21, %zmm20 #72.63
vrcp14pd %zmm20, %zmm31 #75.38
vcmppd $1, %zmm16, %zmm20, %k6{%k5} #74.22
vfpclasspd $30, %zmm31, %k0 #75.38
vmovaps %zmm20, %zmm24 #75.38
vfnmadd213pd .L_2il0floatpacket.5(%rip){1to8}, %zmm31, %zmm24 #75.38
knotw %k0, %k4 #75.38
vmulpd %zmm24, %zmm24, %zmm25 #75.38
vfmadd213pd %zmm31, %zmm24, %zmm31{%k4} #75.38
vfmadd213pd %zmm31, %zmm25, %zmm31{%k4} #75.38
vmulpd %zmm15, %zmm31, %zmm26 #76.38
vmulpd %zmm14, %zmm31, %zmm28 #77.54
vmulpd %zmm26, %zmm31, %zmm29 #76.44
vmulpd %zmm29, %zmm31, %zmm27 #76.50
vfmsub213pd %zmm7, %zmm29, %zmm31 #77.54
vmulpd %zmm28, %zmm27, %zmm30 #77.61
vmulpd %zmm31, %zmm30, %zmm24 #77.67
vfmadd231pd %zmm1, %zmm24, %zmm13{%k6} #78.17
vfmadd231pd %zmm0, %zmm24, %zmm12{%k6} #79.17
vfmadd231pd %zmm21, %zmm24, %zmm11{%k6} #80.17
cmpq %r14, %r15 #67.9
jb ..B1.18 # Prob 82% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 r14 r15 ebx r9d r13d xmm8 xmm9 xmm10 ymm3 ymm4 ymm17 ymm18 zmm2 zmm5 zmm6 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.19: # Preds ..B1.18
# Execution count [4.50e+00]
cmpl %r9d, %r13d #67.9
je ..B1.26 # Prob 10% #67.9
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 ebx r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.20: # Preds ..B1.19 ..B1.16 ..B1.31
# Execution count [2.50e+01]
lea 8(%r9), %ecx #67.9
cmpl %ecx, %ebx #67.9
jl ..B1.24 # Prob 50% #67.9
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 ebx r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.21: # Preds ..B1.20
# Execution count [4.50e+00]
movq %r12, %rcx #56.43
imulq %r8, %rcx #56.43
vbroadcastsd %xmm8, %zmm0 #58.23
vbroadcastsd %xmm9, %zmm1 #59.23
vbroadcastsd %xmm10, %zmm2 #60.23
movslq %r9d, %r14 #67.9
addq %r11, %rcx #37.5
movl $111,%ebx #IACA/OSACA START MARKER
.byte 100,103,144 #IACA/OSACA START MARKER
# LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 r14 ebx r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm0 zmm1 zmm2 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.22: # Preds ..B1.22 ..B1.21
# Execution count [2.50e+01]
vpcmpeqb %xmm0, %xmm0, %k2 #70.36
addl $8, %r9d #67.9
vpcmpeqb %xmm0, %xmm0, %k1 #69.36
vpcmpeqb %xmm0, %xmm0, %k3 #71.36
vmovdqu (%rcx,%r14,4), %ymm3 #68.21
addq $8, %r14 #67.9
vpxord %zmm5, %zmm5, %zmm5 #70.36
vpxord %zmm4, %zmm4, %zmm4 #69.36
vpxord %zmm6, %zmm6, %zmm6 #71.36
vgatherdpd (%rax,%ymm3,8), %zmm5{%k2} #70.36
vgatherdpd (%rdx,%ymm3,8), %zmm4{%k1} #69.36
vgatherdpd (%rsi,%ymm3,8), %zmm6{%k3} #71.36
vsubpd %zmm5, %zmm1, %zmm29 #70.36
vsubpd %zmm4, %zmm0, %zmm28 #69.36
vsubpd %zmm6, %zmm2, %zmm31 #71.36
vmulpd %zmm29, %zmm29, %zmm20 #72.49
vfmadd231pd %zmm28, %zmm28, %zmm20 #72.49
vfmadd231pd %zmm31, %zmm31, %zmm20 #72.63
vrcp14pd %zmm20, %zmm27 #75.38
vcmppd $1, %zmm16, %zmm20, %k5 #74.22
vfpclasspd $30, %zmm27, %k0 #75.38
vfnmadd213pd .L_2il0floatpacket.5(%rip){1to8}, %zmm27, %zmm20 #75.38
knotw %k0, %k4 #75.38
vmulpd %zmm20, %zmm20, %zmm21 #75.38
vfmadd213pd %zmm27, %zmm20, %zmm27{%k4} #75.38
vfmadd213pd %zmm27, %zmm21, %zmm27{%k4} #75.38
vmulpd %zmm15, %zmm27, %zmm22 #76.38
vmulpd %zmm14, %zmm27, %zmm24 #77.54
vmulpd %zmm22, %zmm27, %zmm25 #76.44
vmulpd %zmm25, %zmm27, %zmm23 #76.50
vfmsub213pd %zmm7, %zmm25, %zmm27 #77.54
vmulpd %zmm24, %zmm23, %zmm26 #77.61
vmulpd %zmm27, %zmm26, %zmm30 #77.67
vfmadd231pd %zmm28, %zmm30, %zmm13{%k5} #78.17
vfmadd231pd %zmm29, %zmm30, %zmm12{%k5} #79.17
vfmadd231pd %zmm31, %zmm30, %zmm11{%k5} #80.17
cmpl %ebx, %r9d #67.9
jb ..B1.22 # Prob 82% #67.9
movl $222,%ebx #IACA/OSACA END MARKER
.byte 100,103,144 #IACA/OSACA END MARKER
# LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 r14 ebx r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm0 zmm1 zmm2 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.24: # Preds ..B1.22 ..B1.20 ..B1.32
# Execution count [5.00e+00]
lea 1(%rbx), %ecx #67.9
cmpl %r13d, %ecx #67.9
ja ..B1.26 # Prob 50% #67.9
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 ebx r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.25: # Preds ..B1.24
# Execution count [2.50e+01]
imulq %r12, %r8 #56.43
vbroadcastsd %xmm9, %zmm9 #59.23
vbroadcastsd %xmm8, %zmm2 #58.23
vbroadcastsd %xmm10, %zmm10 #60.23
subl %ebx, %r13d #67.9
addq %r11, %r8 #37.5
vpbroadcastd %r13d, %ymm0 #67.9
vpcmpgtd %ymm17, %ymm0, %k5 #67.9
movslq %ebx, %rbx #67.9
vmovaps %zmm19, %zmm4 #70.36
kmovw %k5, %k2 #70.36
vmovaps %zmm19, %zmm3 #69.36
vmovdqu32 (%r8,%rbx,4), %ymm1{%k5}{z} #68.21
kmovw %k5, %k1 #69.36
vmovaps %zmm19, %zmm5 #71.36
kmovw %k5, %k3 #71.36
vgatherdpd (%rsi,%ymm1,8), %zmm5{%k3} #71.36
vgatherdpd (%rax,%ymm1,8), %zmm4{%k2} #70.36
vgatherdpd (%rdx,%ymm1,8), %zmm3{%k1} #69.36
vsubpd %zmm5, %zmm10, %zmm30 #71.36
vsubpd %zmm4, %zmm9, %zmm28 #70.36
vsubpd %zmm3, %zmm2, %zmm27 #69.36
vmulpd %zmm28, %zmm28, %zmm26 #72.49
vfmadd231pd %zmm27, %zmm27, %zmm26 #72.49
vfmadd231pd %zmm30, %zmm30, %zmm26 #72.63
vrcp14pd %zmm26, %zmm25 #75.38
vcmppd $1, %zmm16, %zmm26, %k6{%k5} #74.22
vfpclasspd $30, %zmm25, %k0 #75.38
vmovaps %zmm26, %zmm6 #75.38
vfnmadd213pd .L_2il0floatpacket.5(%rip){1to8}, %zmm25, %zmm6 #75.38
knotw %k0, %k4 #75.38
vmulpd %zmm6, %zmm6, %zmm8 #75.38
vfmadd213pd %zmm25, %zmm6, %zmm25{%k4} #75.38
vfmadd213pd %zmm25, %zmm8, %zmm25{%k4} #75.38
vmulpd %zmm15, %zmm25, %zmm20 #76.38
vmulpd %zmm14, %zmm25, %zmm22 #77.54
vmulpd %zmm20, %zmm25, %zmm23 #76.44
vmulpd %zmm23, %zmm25, %zmm21 #76.50
vfmsub213pd %zmm7, %zmm23, %zmm25 #77.54
vmulpd %zmm22, %zmm21, %zmm24 #77.61
vmulpd %zmm25, %zmm24, %zmm29 #77.67
vfmadd231pd %zmm27, %zmm29, %zmm13{%k6} #78.17
vfmadd231pd %zmm28, %zmm29, %zmm12{%k6} #79.17
vfmadd231pd %zmm30, %zmm29, %zmm11{%k6} #80.17
# LOE rax rdx rbp rsi rdi r10 r11 r12 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.26: # Preds ..B1.19 ..B1.25 ..B1.24
# Execution count [4.50e+00]
vmovups .L_2il0floatpacket.6(%rip), %zmm10 #63.22
vpermd %zmm11, %zmm10, %zmm0 #63.22
vpermd %zmm12, %zmm10, %zmm5 #62.22
vpermd %zmm13, %zmm10, %zmm21 #61.22
vaddpd %zmm11, %zmm0, %zmm11 #63.22
vaddpd %zmm12, %zmm5, %zmm12 #62.22
vaddpd %zmm13, %zmm21, %zmm13 #61.22
vpermpd $78, %zmm11, %zmm1 #63.22
vpermpd $78, %zmm12, %zmm6 #62.22
vpermpd $78, %zmm13, %zmm22 #61.22
vaddpd %zmm1, %zmm11, %zmm2 #63.22
vaddpd %zmm6, %zmm12, %zmm8 #62.22
vaddpd %zmm22, %zmm13, %zmm23 #61.22
vpermpd $177, %zmm2, %zmm3 #63.22
vpermpd $177, %zmm8, %zmm9 #62.22
vpermpd $177, %zmm23, %zmm24 #61.22
vaddpd %zmm3, %zmm2, %zmm4 #63.22
vaddpd %zmm9, %zmm8, %zmm20 #62.22
vaddpd %zmm24, %zmm23, %zmm25 #61.22
# LOE rax rdx rbp rsi rdi r10 r11 r12 xmm4 xmm20 xmm25 ymm17 ymm18 zmm7 zmm14 zmm15 zmm16 zmm19
..B1.27: # Preds ..B1.26 ..B1.9
# Execution count [5.00e+00]
movq -8(%rsp), %rcx #84.9[spill]
movq -16(%rsp), %rbx #85.9[spill]
movslq %r10d, %r8 #55.32
incq %r8 #55.32
vaddsd (%rcx,%r10,8), %xmm25, %xmm0 #84.9
vmovsd %xmm0, (%rcx,%r10,8) #84.9
vaddsd (%rbx,%r10,8), %xmm20, %xmm1 #85.9
vmovsd %xmm1, (%rbx,%r10,8) #85.9
vaddsd (%rdi,%r10,8), %xmm4, %xmm2 #86.9
vmovsd %xmm2, (%rdi,%r10,8) #86.9
incq %r10 #55.5
cmpq -32(%rsp), %r10 #55.5[spill]
jb ..B1.9 # Prob 82% #55.5
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 ymm17 ymm18 zmm7 zmm14 zmm15 zmm16 zmm19
..B1.28: # Preds ..B1.27
# Execution count [9.00e-01]
movq -40(%rsp), %r15 #[spill]
.cfi_restore 15
movq -48(%rsp), %rbx #[spill]
.cfi_restore 3
# LOE rbx rbp r15
..B1.30: # Preds ..B1.1 ..B1.28
# Execution count [1.00e+00]
vzeroupper #93.12
vxorpd %xmm0, %xmm0, %xmm0 #93.12
.cfi_restore 14
popq %r14 #93.12
.cfi_def_cfa_offset 24
.cfi_restore 13
popq %r13 #93.12
.cfi_def_cfa_offset 16
.cfi_restore 12
popq %r12 #93.12
.cfi_def_cfa_offset 8
ret #93.12
.cfi_def_cfa_offset 32
.cfi_offset 3, -80
.cfi_offset 12, -16
.cfi_offset 13, -24
.cfi_offset 14, -32
.cfi_offset 15, -72
# LOE
..B1.31: # Preds ..B1.11
# Execution count [4.50e-01]: Infreq
movl %r13d, %ebx #67.9
xorl %r9d, %r9d #67.9
andl $-8, %ebx #67.9
jmp ..B1.20 # Prob 100% #67.9
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 ebx r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.32: # Preds ..B1.10
# Execution count [4.50e-01]: Infreq
xorl %ebx, %ebx #67.9
jmp ..B1.24 # Prob 100% #67.9
.align 16,0x90
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 ebx r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
.cfi_endproc
# mark_end;
.type computeForce,@function
.size computeForce,.-computeForce
..LNcomputeForce.0:
.data
# -- End computeForce
.section .rodata, "a"
.align 64
.align 64
.L_2il0floatpacket.2:
.long 0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000
.type .L_2il0floatpacket.2,@object
.size .L_2il0floatpacket.2,64
.align 64
.L_2il0floatpacket.4:
.long 0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000
.type .L_2il0floatpacket.4,@object
.size .L_2il0floatpacket.4,64
.align 64
.L_2il0floatpacket.6:
.long 0x00000008,0x00000009,0x0000000a,0x0000000b,0x0000000c,0x0000000d,0x0000000e,0x0000000f,0x00000008,0x00000009,0x0000000a,0x0000000b,0x0000000c,0x0000000d,0x0000000e,0x0000000f
.type .L_2il0floatpacket.6,@object
.size .L_2il0floatpacket.6,64
.align 32
.L_2il0floatpacket.0:
.long 0x00000008,0x00000008,0x00000008,0x00000008,0x00000008,0x00000008,0x00000008,0x00000008
.type .L_2il0floatpacket.0,@object
.size .L_2il0floatpacket.0,32
.align 32
.L_2il0floatpacket.1:
.long 0x00000000,0x00000001,0x00000002,0x00000003,0x00000004,0x00000005,0x00000006,0x00000007
.type .L_2il0floatpacket.1,@object
.size .L_2il0floatpacket.1,32
.align 8
.L_2il0floatpacket.3:
.long 0x00000000,0x40480000
.type .L_2il0floatpacket.3,@object
.size .L_2il0floatpacket.3,8
.align 8
.L_2il0floatpacket.5:
.long 0x00000000,0x3ff00000
.type .L_2il0floatpacket.5,@object
.size .L_2il0floatpacket.5,8
.data
.section .note.GNU-stack, ""
# End

475
asm/avx512/force_soa_lt8_markers.s
View File

@ -1,475 +0,0 @@
# mark_description "Intel(R) C Intel(R) 64 Compiler for applications running on Intel(R) 64, Version 19.0.5.281 Build 20190815";
# mark_description "-I./src/includes -S -D_GNU_SOURCE -DPRECISION=2 -DALIGNMENT=64 -restrict -Ofast -xCORE-AVX512 -qopt-zmm-usag";
# mark_description "e=high -o ICC/force.s";
.file "force.c"
.text
..TXTST0:
.L_2__routine_start_computeForce_0:
# -- Begin computeForce
.text
# mark_begin;
.align 16,0x90
.globl computeForce
# --- computeForce(Parameter *, Atom *, Neighbor *, int)
computeForce:
# parameter 1: %rdi
# parameter 2: %rsi
# parameter 3: %rdx
# parameter 4: %ecx
..B1.1: # Preds ..B1.0
# Execution count [1.00e+00]
.cfi_startproc
..___tag_value_computeForce.1:
..L2:
#35.1
pushq %r12 #35.1
.cfi_def_cfa_offset 16
.cfi_offset 12, -16
pushq %r13 #35.1
.cfi_def_cfa_offset 24
.cfi_offset 13, -24
pushq %r14 #35.1
.cfi_def_cfa_offset 32
.cfi_offset 14, -32
movl 4(%rsi), %r9d #36.18
vmovsd 72(%rdi), %xmm2 #38.27
vmovsd 8(%rdi), %xmm1 #39.23
vmovsd (%rdi), %xmm0 #40.24
movq 64(%rsi), %r13 #41.20
movq 72(%rsi), %r14 #41.45
movq 80(%rsi), %rdi #41.70
testl %r9d, %r9d #44.24
jle ..B1.30 # Prob 50% #44.24
# LOE rdx rbx rbp rsi rdi r13 r14 r15 r9d xmm0 xmm1 xmm2
..B1.2: # Preds ..B1.1
# Execution count [1.00e+00]
xorl %r10d, %r10d #44.5
movl %r9d, %ecx #44.5
xorl %r8d, %r8d #44.5
movl $1, %r11d #44.5
xorl %eax, %eax #45.17
shrl $1, %ecx #44.5
je ..B1.6 # Prob 9% #44.5
# LOE rax rdx rcx rbx rbp rsi rdi r8 r10 r13 r14 r15 r9d r11d xmm0 xmm1 xmm2
..B1.4: # Preds ..B1.2 ..B1.4
# Execution count [2.50e+00]
movq %rax, (%r8,%r13) #45.9
incq %r10 #44.5
movq %rax, (%r8,%r14) #46.9
movq %rax, (%r8,%rdi) #47.9
movq %rax, 8(%r8,%r13) #45.9
movq %rax, 8(%r8,%r14) #46.9
movq %rax, 8(%r8,%rdi) #47.9
addq $16, %r8 #44.5
cmpq %rcx, %r10 #44.5
jb ..B1.4 # Prob 63% #44.5
# LOE rax rdx rcx rbx rbp rsi rdi r8 r10 r13 r14 r15 r9d xmm0 xmm1 xmm2
..B1.5: # Preds ..B1.4
# Execution count [9.00e-01]
lea 1(%r10,%r10), %r11d #45.9
# LOE rax rdx rbx rbp rsi rdi r13 r14 r15 r9d r11d xmm0 xmm1 xmm2
..B1.6: # Preds ..B1.2 ..B1.5
# Execution count [1.00e+00]
lea -1(%r11), %ecx #44.5
cmpl %r9d, %ecx #44.5
jae ..B1.8 # Prob 9% #44.5
# LOE rax rdx rbx rbp rsi rdi r13 r14 r15 r9d r11d xmm0 xmm1 xmm2
..B1.7: # Preds ..B1.6
# Execution count [9.00e-01]
movslq %r11d, %r11 #44.5
movq %rax, -8(%r13,%r11,8) #45.9
movq %rax, -8(%r14,%r11,8) #46.9
movq %rax, -8(%rdi,%r11,8) #47.9
# LOE rdx rbx rbp rsi rdi r13 r14 r15 r9d xmm0 xmm1 xmm2
..B1.8: # Preds ..B1.6 ..B1.7
# Execution count [9.00e-01]
vmulsd %xmm2, %xmm2, %xmm15 #38.45
xorl %r8d, %r8d #55.15
vmovdqu32 .L_2il0floatpacket.0(%rip), %ymm18 #67.9
vmulsd .L_2il0floatpacket.3(%rip), %xmm0, %xmm0 #77.41
vmovdqu32 .L_2il0floatpacket.1(%rip), %ymm17 #67.9
vmovups .L_2il0floatpacket.4(%rip), %zmm7 #77.54
vbroadcastsd %xmm15, %zmm16 #38.25
vbroadcastsd %xmm1, %zmm15 #39.21
vbroadcastsd %xmm0, %zmm14 #77.41
movslq %r9d, %r9 #55.5
xorl %r10d, %r10d #55.5
movq 24(%rdx), %rcx #57.25
movq 8(%rdx), %r11 #56.19
movslq 16(%rdx), %r12 #56.43
movq 16(%rsi), %rdx #58.25
movq 24(%rsi), %rax #59.25
movq 32(%rsi), %rsi #60.25
shlq $2, %r12 #37.5
movq %r9, -32(%rsp) #60.25[spill]
movq %rcx, -24(%rsp) #60.25[spill]
movq %r14, -16(%rsp) #60.25[spill]
movq %r13, -8(%rsp) #60.25[spill]
movq %r15, -40(%rsp) #60.25[spill]
movq %rbx, -48(%rsp) #60.25[spill]
vpxord %zmm19, %zmm19, %zmm19 #61.22
.cfi_offset 3, -80
.cfi_offset 15, -72
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 ymm17 ymm18 zmm7 zmm14 zmm15 zmm16 zmm19
..B1.9: # Preds ..B1.27 ..B1.8
# Execution count [5.00e+00]
movq -24(%rsp), %rcx #57.25[spill]
vxorpd %xmm25, %xmm25, %xmm25 #61.22
vmovapd %xmm25, %xmm20 #62.22
movl (%rcx,%r10,4), %r13d #57.25
vmovapd %xmm20, %xmm4 #63.22
vmovsd (%rdx,%r10,8), %xmm8 #58.25
vmovsd (%rax,%r10,8), %xmm9 #59.25
vmovsd (%rsi,%r10,8), %xmm10 #60.25
testl %r13d, %r13d #67.28
jle ..B1.27 # Prob 50% #67.28
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 r13d xmm4 xmm8 xmm9 xmm10 xmm20 xmm25 ymm17 ymm18 zmm7 zmm14 zmm15 zmm16 zmm19
..B1.10: # Preds ..B1.9
# Execution count [4.50e+00]
vmovaps %zmm19, %zmm13 #61.22
vmovaps %zmm13, %zmm12 #62.22
vmovaps %zmm12, %zmm11 #63.22
cmpl $8, %r13d #67.9
jl ..B1.32 # Prob 10% #67.9
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.11: # Preds ..B1.10
# Execution count [4.50e+00]
cmpl $1200, %r13d #67.9
jl ..B1.31 # Prob 10% #67.9
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.12: # Preds ..B1.11
# Execution count [4.50e+00]
movq %r12, %rcx #56.43
imulq %r8, %rcx #56.43
addq %r11, %rcx #37.5
movq %rcx, %r9 #67.9
andq $63, %r9 #67.9
testl $3, %r9d #67.9
je ..B1.14 # Prob 50% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.13: # Preds ..B1.12
# Execution count [2.25e+00]
xorl %r9d, %r9d #67.9
jmp ..B1.16 # Prob 100% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.14: # Preds ..B1.12
# Execution count [2.25e+00]
testl %r9d, %r9d #67.9
je ..B1.16 # Prob 50% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.15: # Preds ..B1.14
# Execution count [2.50e+01]
negl %r9d #67.9
addl $64, %r9d #67.9
shrl $2, %r9d #67.9
cmpl %r9d, %r13d #67.9
cmovl %r13d, %r9d #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.16: # Preds ..B1.13 ..B1.15 ..B1.14
# Execution count [5.00e+00]
movl %r13d, %ebx #67.9
subl %r9d, %ebx #67.9
andl $7, %ebx #67.9
negl %ebx #67.9
addl %r13d, %ebx #67.9
cmpl $1, %r9d #67.9
jb ..B1.20 # Prob 50% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 ebx r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.17: # Preds ..B1.16
# Execution count [4.50e+00]
vmovdqa32 %ymm17, %ymm4 #67.9
xorl %r15d, %r15d #67.9
vpbroadcastd %r9d, %ymm3 #67.9
vbroadcastsd %xmm8, %zmm2 #58.23
vbroadcastsd %xmm9, %zmm5 #59.23
vbroadcastsd %xmm10, %zmm6 #60.23
movslq %r9d, %r14 #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 r14 r15 ebx r9d r13d xmm8 xmm9 xmm10 ymm3 ymm4 ymm17 ymm18 zmm2 zmm5 zmm6 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.18: # Preds ..B1.18 ..B1.17
# Execution count [2.50e+01]
vpcmpgtd %ymm4, %ymm3, %k5 #67.9
vpaddd %ymm18, %ymm4, %ymm4 #67.9
vmovdqu32 (%rcx,%r15,4), %ymm20{%k5}{z} #68.21
vmovaps %zmm19, %zmm22 #70.36
addq $8, %r15 #67.9
kmovw %k5, %k2 #70.36
vmovaps %zmm19, %zmm21 #69.36
kmovw %k5, %k1 #69.36
vmovaps %zmm19, %zmm23 #71.36
kmovw %k5, %k3 #71.36
vgatherdpd (%rsi,%ymm20,8), %zmm23{%k3} #71.36
vgatherdpd (%rax,%ymm20,8), %zmm22{%k2} #70.36
vgatherdpd (%rdx,%ymm20,8), %zmm21{%k1} #69.36
vsubpd %zmm22, %zmm5, %zmm0 #70.36
vsubpd %zmm21, %zmm2, %zmm1 #69.36
vsubpd %zmm23, %zmm6, %zmm21 #71.36
vmulpd %zmm0, %zmm0, %zmm20 #72.49
vfmadd231pd %zmm1, %zmm1, %zmm20 #72.49
vfmadd231pd %zmm21, %zmm21, %zmm20 #72.63
vrcp14pd %zmm20, %zmm31 #75.38
vcmppd $1, %zmm16, %zmm20, %k6{%k5} #74.22
vfpclasspd $30, %zmm31, %k0 #75.38
vmovaps %zmm20, %zmm24 #75.38
vfnmadd213pd .L_2il0floatpacket.5(%rip){1to8}, %zmm31, %zmm24 #75.38
knotw %k0, %k4 #75.38
vmulpd %zmm24, %zmm24, %zmm25 #75.38
vfmadd213pd %zmm31, %zmm24, %zmm31{%k4} #75.38
vfmadd213pd %zmm31, %zmm25, %zmm31{%k4} #75.38
vmulpd %zmm15, %zmm31, %zmm26 #76.38
vmulpd %zmm14, %zmm31, %zmm28 #77.54
vmulpd %zmm26, %zmm31, %zmm29 #76.44
vmulpd %zmm29, %zmm31, %zmm27 #76.50
vfmsub213pd %zmm7, %zmm29, %zmm31 #77.54
vmulpd %zmm28, %zmm27, %zmm30 #77.61
vmulpd %zmm31, %zmm30, %zmm24 #77.67
vfmadd231pd %zmm1, %zmm24, %zmm13{%k6} #78.17
vfmadd231pd %zmm0, %zmm24, %zmm12{%k6} #79.17
vfmadd231pd %zmm21, %zmm24, %zmm11{%k6} #80.17
cmpq %r14, %r15 #67.9
jb ..B1.18 # Prob 82% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 r14 r15 ebx r9d r13d xmm8 xmm9 xmm10 ymm3 ymm4 ymm17 ymm18 zmm2 zmm5 zmm6 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.19: # Preds ..B1.18
# Execution count [4.50e+00]
cmpl %r9d, %r13d #67.9
je ..B1.26 # Prob 10% #67.9
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 ebx r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.20: # Preds ..B1.19 ..B1.16 ..B1.31
# Execution count [2.50e+01]
lea 8(%r9), %ecx #67.9
cmpl %ecx, %ebx #67.9
jl ..B1.24 # Prob 50% #67.9
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 ebx r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.21: # Preds ..B1.20
# Execution count [4.50e+00]
movq %r12, %rcx #56.43
imulq %r8, %rcx #56.43
vbroadcastsd %xmm8, %zmm0 #58.23
vbroadcastsd %xmm9, %zmm1 #59.23
vbroadcastsd %xmm10, %zmm2 #60.23
movslq %r9d, %r14 #67.9
addq %r11, %rcx #37.5
# LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 r14 ebx r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm0 zmm1 zmm2 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.22: # Preds ..B1.22 ..B1.21
# Execution count [2.50e+01]
vpcmpeqb %xmm0, %xmm0, %k2 #70.36
addl $8, %r9d #67.9
vpcmpeqb %xmm0, %xmm0, %k1 #69.36
vpcmpeqb %xmm0, %xmm0, %k3 #71.36
vmovdqu (%rcx,%r14,4), %ymm3 #68.21
addq $8, %r14 #67.9
vpxord %zmm5, %zmm5, %zmm5 #70.36
vpxord %zmm4, %zmm4, %zmm4 #69.36
vpxord %zmm6, %zmm6, %zmm6 #71.36
vgatherdpd (%rax,%ymm3,8), %zmm5{%k2} #70.36
vgatherdpd (%rdx,%ymm3,8), %zmm4{%k1} #69.36
vgatherdpd (%rsi,%ymm3,8), %zmm6{%k3} #71.36
vsubpd %zmm5, %zmm1, %zmm29 #70.36
vsubpd %zmm4, %zmm0, %zmm28 #69.36
vsubpd %zmm6, %zmm2, %zmm31 #71.36
vmulpd %zmm29, %zmm29, %zmm20 #72.49
vfmadd231pd %zmm28, %zmm28, %zmm20 #72.49
vfmadd231pd %zmm31, %zmm31, %zmm20 #72.63
vrcp14pd %zmm20, %zmm27 #75.38
vcmppd $1, %zmm16, %zmm20, %k5 #74.22
vfpclasspd $30, %zmm27, %k0 #75.38
vfnmadd213pd .L_2il0floatpacket.5(%rip){1to8}, %zmm27, %zmm20 #75.38
knotw %k0, %k4 #75.38
vmulpd %zmm20, %zmm20, %zmm21 #75.38
vfmadd213pd %zmm27, %zmm20, %zmm27{%k4} #75.38
vfmadd213pd %zmm27, %zmm21, %zmm27{%k4} #75.38
vmulpd %zmm15, %zmm27, %zmm22 #76.38
vmulpd %zmm14, %zmm27, %zmm24 #77.54
vmulpd %zmm22, %zmm27, %zmm25 #76.44
vmulpd %zmm25, %zmm27, %zmm23 #76.50
vfmsub213pd %zmm7, %zmm25, %zmm27 #77.54
vmulpd %zmm24, %zmm23, %zmm26 #77.61
vmulpd %zmm27, %zmm26, %zmm30 #77.67
vfmadd231pd %zmm28, %zmm30, %zmm13{%k5} #78.17
vfmadd231pd %zmm29, %zmm30, %zmm12{%k5} #79.17
vfmadd231pd %zmm31, %zmm30, %zmm11{%k5} #80.17
cmpl %ebx, %r9d #67.9
jb ..B1.22 # Prob 82% #67.9
# LOE rax rdx rcx rbp rsi rdi r8 r10 r11 r12 r14 ebx r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm0 zmm1 zmm2 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.24: # Preds ..B1.22 ..B1.20 ..B1.32
# Execution count [5.00e+00]
lea 1(%rbx), %ecx #67.9
cmpl %r13d, %ecx #67.9
ja ..B1.26 # Prob 50% #67.9
movl $111,%ebx #IACA/OSACA START MARKER
.byte 100,103,144 #IACA/OSACA START MARKER
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 ebx r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.25: # Preds ..B1.24
# Execution count [2.50e+01]
imulq %r12, %r8 #56.43
vbroadcastsd %xmm9, %zmm9 #59.23
vbroadcastsd %xmm8, %zmm2 #58.23
vbroadcastsd %xmm10, %zmm10 #60.23
subl %ebx, %r13d #67.9
addq %r11, %r8 #37.5
vpbroadcastd %r13d, %ymm0 #67.9
vpcmpgtd %ymm17, %ymm0, %k5 #67.9
movslq %ebx, %rbx #67.9
vmovaps %zmm19, %zmm4 #70.36
kmovw %k5, %k2 #70.36
vmovaps %zmm19, %zmm3 #69.36
vmovdqu32 (%r8,%rbx,4), %ymm1{%k5}{z} #68.21
kmovw %k5, %k1 #69.36
vmovaps %zmm19, %zmm5 #71.36
kmovw %k5, %k3 #71.36
vgatherdpd (%rsi,%ymm1,8), %zmm5{%k3} #71.36
vgatherdpd (%rax,%ymm1,8), %zmm4{%k2} #70.36
vgatherdpd (%rdx,%ymm1,8), %zmm3{%k1} #69.36
vsubpd %zmm5, %zmm10, %zmm30 #71.36
vsubpd %zmm4, %zmm9, %zmm28 #70.36
vsubpd %zmm3, %zmm2, %zmm27 #69.36
vmulpd %zmm28, %zmm28, %zmm26 #72.49
vfmadd231pd %zmm27, %zmm27, %zmm26 #72.49
vfmadd231pd %zmm30, %zmm30, %zmm26 #72.63
vrcp14pd %zmm26, %zmm25 #75.38
vcmppd $1, %zmm16, %zmm26, %k6{%k5} #74.22
vfpclasspd $30, %zmm25, %k0 #75.38
vmovaps %zmm26, %zmm6 #75.38
vfnmadd213pd .L_2il0floatpacket.5(%rip){1to8}, %zmm25, %zmm6 #75.38
knotw %k0, %k4 #75.38
vmulpd %zmm6, %zmm6, %zmm8 #75.38
vfmadd213pd %zmm25, %zmm6, %zmm25{%k4} #75.38
vfmadd213pd %zmm25, %zmm8, %zmm25{%k4} #75.38
vmulpd %zmm15, %zmm25, %zmm20 #76.38
vmulpd %zmm14, %zmm25, %zmm22 #77.54
vmulpd %zmm20, %zmm25, %zmm23 #76.44
vmulpd %zmm23, %zmm25, %zmm21 #76.50
vfmsub213pd %zmm7, %zmm23, %zmm25 #77.54
vmulpd %zmm22, %zmm21, %zmm24 #77.61
vmulpd %zmm25, %zmm24, %zmm29 #77.67
vfmadd231pd %zmm27, %zmm29, %zmm13{%k6} #78.17
vfmadd231pd %zmm28, %zmm29, %zmm12{%k6} #79.17
vfmadd231pd %zmm30, %zmm29, %zmm11{%k6} #80.17
movl $222,%ebx #IACA/OSACA END MARKER
.byte 100,103,144 #IACA/OSACA END MARKER
# LOE rax rdx rbp rsi rdi r10 r11 r12 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.26: # Preds ..B1.19 ..B1.25 ..B1.24
# Execution count [4.50e+00]
vmovups .L_2il0floatpacket.6(%rip), %zmm10 #63.22
vpermd %zmm11, %zmm10, %zmm0 #63.22
vpermd %zmm12, %zmm10, %zmm5 #62.22
vpermd %zmm13, %zmm10, %zmm21 #61.22
vaddpd %zmm11, %zmm0, %zmm11 #63.22
vaddpd %zmm12, %zmm5, %zmm12 #62.22
vaddpd %zmm13, %zmm21, %zmm13 #61.22
vpermpd $78, %zmm11, %zmm1 #63.22
vpermpd $78, %zmm12, %zmm6 #62.22
vpermpd $78, %zmm13, %zmm22 #61.22
vaddpd %zmm1, %zmm11, %zmm2 #63.22
vaddpd %zmm6, %zmm12, %zmm8 #62.22
vaddpd %zmm22, %zmm13, %zmm23 #61.22
vpermpd $177, %zmm2, %zmm3 #63.22
vpermpd $177, %zmm8, %zmm9 #62.22
vpermpd $177, %zmm23, %zmm24 #61.22
vaddpd %zmm3, %zmm2, %zmm4 #63.22
vaddpd %zmm9, %zmm8, %zmm20 #62.22
vaddpd %zmm24, %zmm23, %zmm25 #61.22
# LOE rax rdx rbp rsi rdi r10 r11 r12 xmm4 xmm20 xmm25 ymm17 ymm18 zmm7 zmm14 zmm15 zmm16 zmm19
..B1.27: # Preds ..B1.26 ..B1.9
# Execution count [5.00e+00]
movq -8(%rsp), %rcx #84.9[spill]
movq -16(%rsp), %rbx #85.9[spill]
movslq %r10d, %r8 #55.32
incq %r8 #55.32
vaddsd (%rcx,%r10,8), %xmm25, %xmm0 #84.9
vmovsd %xmm0, (%rcx,%r10,8) #84.9
vaddsd (%rbx,%r10,8), %xmm20, %xmm1 #85.9
vmovsd %xmm1, (%rbx,%r10,8) #85.9
vaddsd (%rdi,%r10,8), %xmm4, %xmm2 #86.9
vmovsd %xmm2, (%rdi,%r10,8) #86.9
incq %r10 #55.5
cmpq -32(%rsp), %r10 #55.5[spill]
jb ..B1.9 # Prob 82% #55.5
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 ymm17 ymm18 zmm7 zmm14 zmm15 zmm16 zmm19
..B1.28: # Preds ..B1.27
# Execution count [9.00e-01]
movq -40(%rsp), %r15 #[spill]
.cfi_restore 15
movq -48(%rsp), %rbx #[spill]
.cfi_restore 3
# LOE rbx rbp r15
..B1.30: # Preds ..B1.1 ..B1.28
# Execution count [1.00e+00]
vzeroupper #93.12
vxorpd %xmm0, %xmm0, %xmm0 #93.12
.cfi_restore 14
popq %r14 #93.12
.cfi_def_cfa_offset 24
.cfi_restore 13
popq %r13 #93.12
.cfi_def_cfa_offset 16
.cfi_restore 12
popq %r12 #93.12
.cfi_def_cfa_offset 8
ret #93.12
.cfi_def_cfa_offset 32
.cfi_offset 3, -80
.cfi_offset 12, -16
.cfi_offset 13, -24
.cfi_offset 14, -32
.cfi_offset 15, -72
# LOE
..B1.31: # Preds ..B1.11
# Execution count [4.50e-01]: Infreq
movl %r13d, %ebx #67.9
xorl %r9d, %r9d #67.9
andl $-8, %ebx #67.9
jmp ..B1.20 # Prob 100% #67.9
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 ebx r9d r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
..B1.32: # Preds ..B1.10
# Execution count [4.50e-01]: Infreq
xorl %ebx, %ebx #67.9
jmp ..B1.24 # Prob 100% #67.9
.align 16,0x90
# LOE rax rdx rbp rsi rdi r8 r10 r11 r12 ebx r13d xmm8 xmm9 xmm10 ymm17 ymm18 zmm7 zmm11 zmm12 zmm13 zmm14 zmm15 zmm16 zmm19
.cfi_endproc
# mark_end;
.type computeForce,@function
.size computeForce,.-computeForce
..LNcomputeForce.0:
.data
# -- End computeForce
.section .rodata, "a"
.align 64
.align 64
.L_2il0floatpacket.2:
.long 0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000,0x00000000,0x3ff00000
.type .L_2il0floatpacket.2,@object
.size .L_2il0floatpacket.2,64
.align 64
.L_2il0floatpacket.4:
.long 0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000,0x00000000,0x3fe00000
.type .L_2il0floatpacket.4,@object
.size .L_2il0floatpacket.4,64
.align 64
.L_2il0floatpacket.6:
.long 0x00000008,0x00000009,0x0000000a,0x0000000b,0x0000000c,0x0000000d,0x0000000e,0x0000000f,0x00000008,0x00000009,0x0000000a,0x0000000b,0x0000000c,0x0000000d,0x0000000e,0x0000000f
.type .L_2il0floatpacket.6,@object
.size .L_2il0floatpacket.6,64
.align 32
.L_2il0floatpacket.0:
.long 0x00000008,0x00000008,0x00000008,0x00000008,0x00000008,0x00000008,0x00000008,0x00000008
.type .L_2il0floatpacket.0,@object
.size .L_2il0floatpacket.0,32
.align 32
.L_2il0floatpacket.1:
.long 0x00000000,0x00000001,0x00000002,0x00000003,0x00000004,0x00000005,0x00000006,0x00000007
.type .L_2il0floatpacket.1,@object
.size .L_2il0floatpacket.1,32
.align 8
.L_2il0floatpacket.3:
.long 0x00000000,0x40480000
.type .L_2il0floatpacket.3,@object
.size .L_2il0floatpacket.3,8
.align 8
.L_2il0floatpacket.5:
.long 0x00000000,0x3ff00000
.type .L_2il0floatpacket.5,@object
.size .L_2il0floatpacket.5,8
.data
.section .note.GNU-stack, ""
# End

450
scripts/results_aos_casclakesp2.txt
View File

@ -1,450 +0,0 @@
200,4x4x2,8,256,21.5040,6.1440,9.2160,0.0016,31.3286,76.6074,10.9439
200,4x4x4,8,512,43.0080,12.2880,18.4320,0.0034,30.5402,78.5851,11.2264
200,4x4x8,8,1024,86.0160,24.5760,36.8640,0.0068,30.2173,79.4248,11.3464
200,4x4x16,8,2048,172.0320,49.1520,73.7280,0.0137,30.0069,79.9815,11.4259
200,4x4x32,8,4096,344.0640,98.3040,147.4560,0.0272,30.1585,79.5795,11.3685
200,4x4x64,8,8192,688.1280,196.6080,294.9120,0.0570,28.7247,83.5518,11.9360
200,4x8x2,8,512,43.0080,12.2880,18.4320,0.0034,30.4999,78.6887,11.2412
200,4x8x4,8,1024,86.0160,24.5760,36.8640,0.0069,29.8712,80.3449,11.4778
200,4x8x8,8,2048,172.0320,49.1520,73.7280,0.0136,30.2043,79.4588,11.3513
200,4x8x16,8,4096,344.0640,98.3040,147.4560,0.0270,30.2881,79.2391,11.3199
200,4x8x32,8,8192,688.1280,196.6080,294.9120,0.0573,28.5762,83.9861,11.9980
200,4x8x64,8,16384,1376.2560,393.2160,589.8240,0.1153,28.4258,84.4303,12.0615
200,4x16x2,8,1024,86.0160,24.5760,36.8640,0.0066,31.0076,77.4004,11.0572
200,4x16x4,8,2048,172.0320,49.1520,73.7280,0.0132,30.9218,77.6152,11.0879
200,4x16x8,8,4096,344.0640,98.3040,147.4560,0.0272,30.1032,79.7258,11.3894
200,4x16x16,8,8192,688.1280,196.6080,294.9120,0.0567,28.9101,83.0160,11.8594
200,4x16x32,8,16384,1376.2560,393.2160,589.8240,0.1156,28.3389,84.6893,12.0985
200,4x16x64,8,32768,2752.5120,786.4320,1179.6480,0.2277,28.7869,83.3713,11.9102
200,4x32x2,8,2048,172.0320,49.1520,73.7280,0.0136,30.2157,79.4289,11.3470
200,4x32x4,8,4096,344.0640,98.3040,147.4560,0.0275,29.7785,80.5951,11.5136
200,4x32x8,8,8192,688.1280,196.6080,294.9120,0.0566,28.9500,82.9015,11.8431
200,4x32x16,8,16384,1376.2560,393.2160,589.8240,0.1167,28.0786,85.4744,12.2106
200,4x32x32,8,32768,2752.5120,786.4320,1179.6480,0.2276,28.7961,83.3446,11.9064
200,4x32x64,8,65536,5505.0240,1572.8640,2359.2960,0.4624,28.3432,84.6765,12.0966
200,4x64x2,8,4096,344.0640,98.3040,147.4560,0.0272,30.1718,79.5444,11.3635
200,4x64x4,8,8192,688.1280,196.6080,294.9120,0.0574,28.5211,84.1482,12.0212
200,4x64x8,8,16384,1376.2560,393.2160,589.8240,0.1145,28.6206,83.8558,11.9794
200,4x64x16,8,32768,2752.5120,786.4320,1179.6480,0.2235,29.3180,81.8611,11.6944
200,4x64x32,8,65536,5505.0240,1572.8640,2359.2960,0.4677,28.0274,85.6305,12.2329
200,4x64x64,8,131072,11010.0480,3145.7280,4718.5920,1.2438,21.0760,113.8738,16.2677
200,8x4x2,8,512,43.0080,12.2880,18.4320,0.0034,30.3789,79.0023,11.2860
200,8x4x4,8,1024,86.0160,24.5760,36.8640,0.0068,30.0176,79.9532,11.4219
200,8x4x8,8,2048,172.0320,49.1520,73.7280,0.0136,30.2149,79.4309,11.3473
200,8x4x16,8,4096,344.0640,98.3040,147.4560,0.0269,30.5030,78.6807,11.2401
200,8x4x32,8,8192,688.1280,196.6080,294.9120,0.0581,28.1906,85.1349,12.1621
200,8x4x64,8,16384,1376.2560,393.2160,589.8240,0.1151,28.4795,84.2712,12.0387
200,8x8x2,8,1024,86.0160,24.5760,36.8640,0.0068,30.1697,79.5501,11.3643
200,8x8x4,8,2048,172.0320,49.1520,73.7280,0.0137,29.9307,80.1853,11.4550
200,8x8x8,8,4096,344.0640,98.3040,147.4560,0.0267,30.6941,78.1910,11.1701
200,8x8x16,8,8192,688.1280,196.6080,294.9120,0.0579,28.3008,84.8033,12.1148
200,8x8x32,8,16384,1376.2560,393.2160,589.8240,0.1165,28.1230,85.3395,12.1914
200,8x8x64,8,32768,2752.5120,786.4320,1179.6480,0.2284,28.6881,83.6583,11.9512
200,8x16x2,8,2048,172.0320,49.1520,73.7280,0.0137,29.9813,80.0498,11.4357
200,8x16x4,8,4096,344.0640,98.3040,147.4560,0.0274,29.9487,80.1369,11.4481
200,8x16x8,8,8192,688.1280,196.6080,294.9120,0.0569,28.7964,83.3438,11.9063
200,8x16x16,8,16384,1376.2560,393.2160,589.8240,0.1150,28.5022,84.2039,12.0291
200,8x16x32,8,32768,2752.5120,786.4320,1179.6480,0.2281,28.7316,83.5316,11.9331
200,8x16x64,8,65536,5505.0240,1572.8640,2359.2960,0.4686,27.9702,85.8055,12.2579
200,8x32x2,8,4096,344.0640,98.3040,147.4560,0.0268,30.6043,78.4204,11.2029
200,8x32x4,8,8192,688.1280,196.6080,294.9120,0.0562,29.1740,82.2649,11.7521
200,8x32x8,8,16384,1376.2560,393.2160,589.8240,0.1158,28.2999,84.8059,12.1151
200,8x32x16,8,32768,2752.5120,786.4320,1179.6480,0.2275,28.8054,83.3178,11.9025
200,8x32x32,8,65536,5505.0240,1572.8640,2359.2960,0.4667,28.0853,85.4540,12.2077
200,8x32x64,8,131072,11010.0480,3145.7280,4718.5920,1.2352,21.2224,113.0883,16.1555
200,8x64x2,8,8192,688.1280,196.6080,294.9120,0.0581,28.2059,85.0885,12.1555
200,8x64x4,8,16384,1376.2560,393.2160,589.8240,0.1165,28.1387,85.2918,12.1845
200,8x64x8,8,32768,2752.5120,786.4320,1179.6480,0.2266,28.9194,82.9892,11.8556
200,8x64x16,8,65536,5505.0240,1572.8640,2359.2960,0.4618,28.3802,84.5661,12.0809
200,8x64x32,8,131072,11010.0480,3145.7280,4718.5920,1.2085,21.6913,110.6434,15.8062
200,8x64x64,8,262144,22020.0960,6291.4560,9437.1840,3.2471,16.1463,148.6409,21.2344
200,16x4x2,8,1024,86.0160,24.5760,36.8640,0.0068,30.3205,79.1543,11.3078
200,16x4x4,8,2048,172.0320,49.1520,73.7280,0.0136,30.2035,79.4611,11.3516
200,16x4x8,8,4096,344.0640,98.3040,147.4560,0.0273,30.0167,79.9555,11.4222
200,16x4x16,8,8192,688.1280,196.6080,294.9120,0.0571,28.7024,83.6168,11.9453
200,16x4x32,8,16384,1376.2560,393.2160,589.8240,0.1154,28.4058,84.4899,12.0700
200,16x4x64,8,32768,2752.5120,786.4320,1179.6480,0.2190,29.9276,80.1934,11.4562
200,16x8x2,8,2048,172.0320,49.1520,73.7280,0.0138,29.7709,80.6158,11.5165
200,16x8x4,8,4096,344.0640,98.3040,147.4560,0.0273,30.0582,79.8451,11.4064
200,16x8x8,8,8192,688.1280,196.6080,294.9120,0.0577,28.3944,84.5238,12.0748
200,16x8x16,8,16384,1376.2560,393.2160,589.8240,0.1113,29.4382,81.5267,11.6467
200,16x8x32,8,32768,2752.5120,786.4320,1179.6480,0.2247,29.1619,82.2992,11.7570
200,16x8x64,8,65536,5505.0240,1572.8640,2359.2960,0.4572,28.6660,83.7229,11.9604
200,16x16x2,8,4096,344.0640,98.3040,147.4560,0.0269,30.4826,78.7334,11.2476
200,16x16x4,8,8192,688.1280,196.6080,294.9120,0.0575,28.5172,84.1597,12.0228
200,16x16x8,8,16384,1376.2560,393.2160,589.8240,0.1145,28.6233,83.8476,11.9782
200,16x16x16,8,32768,2752.5120,786.4320,1179.6480,0.2246,29.1736,82.2660,11.7523
200,16x16x32,8,65536,5505.0240,1572.8640,2359.2960,0.4568,28.6941,83.6410,11.9487
200,16x16x64,8,131072,11010.0480,3145.7280,4718.5920,1.2084,21.6932,110.6338,15.8048
200,16x32x2,8,8192,688.1280,196.6080,294.9120,0.0575,28.4700,84.2994,12.0428
200,16x32x4,8,16384,1376.2560,393.2160,589.8240,0.1143,28.6568,83.7497,11.9642
200,16x32x8,8,32768,2752.5120,786.4320,1179.6480,0.2298,28.5246,84.1378,12.0197
200,16x32x16,8,65536,5505.0240,1572.8640,2359.2960,0.4589,28.5652,84.0182,12.0026
200,16x32x32,8,131072,11010.0480,3145.7280,4718.5920,1.2003,21.8406,109.8869,15.6981
200,16x32x64,8,262144,22020.0960,6291.4560,9437.1840,3.1864,16.4538,145.8629,20.8376
200,16x64x2,8,16384,1376.2560,393.2160,589.8240,0.1155,28.3812,84.5629,12.0804
200,16x64x4,8,32768,2752.5120,786.4320,1179.6480,0.2266,28.9186,82.9915,11.8559
200,16x64x8,8,65536,5505.0240,1572.8640,2359.2960,0.4500,29.1273,82.3969,11.7710
200,16x64x16,8,131072,11010.0480,3145.7280,4718.5920,1.2058,21.7407,110.3920,15.7703
200,16x64x32,8,262144,22020.0960,6291.4560,9437.1840,3.2580,16.0921,149.1411,21.3059
200,16x64x64,8,524288,44040.1920,12582.9120,18874.3680,7.0624,14.8472,161.6462,23.0923
200,32x4x2,8,2048,172.0320,49.1520,73.7280,0.0136,30.2024,79.4640,11.3520
200,32x4x4,8,4096,344.0640,98.3040,147.4560,0.0272,30.1067,79.7165,11.3881
200,32x4x8,8,8192,688.1280,196.6080,294.9120,0.0558,29.3475,81.7786,11.6827
200,32x4x16,8,16384,1376.2560,393.2160,589.8240,0.1131,28.9697,82.8452,11.8350
200,32x4x32,8,32768,2752.5120,786.4320,1179.6480,0.2266,28.9188,82.9910,11.8559
200,32x4x64,8,65536,5505.0240,1572.8640,2359.2960,0.4571,28.6719,83.7058,11.9580
200,32x8x2,8,4096,344.0640,98.3040,147.4560,0.0267,30.6548,78.2913,11.1845
200,32x8x4,8,8192,688.1280,196.6080,294.9120,0.0565,28.9951,82.7725,11.8246
200,32x8x8,8,16384,1376.2560,393.2160,589.8240,0.1161,28.2313,85.0119,12.1446
200,32x8x16,8,32768,2752.5120,786.4320,1179.6480,0.2236,29.3134,81.8739,11.6963
200,32x8x32,8,65536,5505.0240,1572.8640,2359.2960,0.4577,28.6354,83.8122,11.9732
200,32x8x64,8,131072,11010.0480,3145.7280,4718.5920,1.2065,21.7269,110.4620,15.7803
200,32x16x2,8,8192,688.1280,196.6080,294.9120,0.0564,29.0619,82.5825,11.7975
200,32x16x4,8,16384,1376.2560,393.2160,589.8240,0.1149,28.5307,84.1199,12.0171
200,32x16x8,8,32768,2752.5120,786.4320,1179.6480,0.2242,29.2341,82.0959,11.7280
200,32x16x16,8,65536,5505.0240,1572.8640,2359.2960,0.4530,28.9334,82.9492,11.8499
200,32x16x32,8,131072,11010.0480,3145.7280,4718.5920,1.2210,21.4702,111.7830,15.9690
200,32x16x64,8,262144,22020.0960,6291.4560,9437.1840,3.2488,16.1378,148.7188,21.2455
200,32x32x2,8,16384,1376.2560,393.2160,589.8240,0.1138,28.8006,83.3317,11.9045
200,32x32x4,8,32768,2752.5120,786.4320,1179.6480,0.2283,28.7093,83.5966,11.9424
200,32x32x8,8,65536,5505.0240,1572.8640,2359.2960,0.4490,29.1907,82.2179,11.7454
200,32x32x16,8,131072,11010.0480,3145.7280,4718.5920,1.2358,21.2121,113.1431,16.1633
200,32x32x32,8,262144,22020.0960,6291.4560,9437.1840,3.2416,16.1738,148.3884,21.1983
200,32x32x64,8,524288,44040.1920,12582.9120,18874.3680,7.0780,14.8145,162.0034,23.1433
200,32x64x2,8,32768,2752.5120,786.4320,1179.6480,0.2284,28.6984,83.6283,11.9469
200,32x64x4,8,65536,5505.0240,1572.8640,2359.2960,0.4556,28.7686,83.4243,11.9178
200,32x64x8,8,131072,11010.0480,3145.7280,4718.5920,1.2309,21.2978,112.6880,16.0983
200,32x64x16,8,262144,22020.0960,6291.4560,9437.1840,3.2217,16.2735,147.4793,21.0685
200,32x64x32,8,524288,44040.1920,12582.9120,18874.3680,7.1806,14.6029,164.3510,23.4787
200,32x64x64,8,1048576,88080.3840,25165.8240,37748.7360,14.3679,14.5961,164.4279,23.4897
200,64x4x2,8,4096,344.0640,98.3040,147.4560,0.0273,30.0281,79.9251,11.4179
200,64x4x4,8,8192,688.1280,196.6080,294.9120,0.0570,28.7561,83.4607,11.9230
200,64x4x8,8,16384,1376.2560,393.2160,589.8240,0.1145,28.6263,83.8391,11.9770
200,64x4x16,8,32768,2752.5120,786.4320,1179.6480,0.2273,28.8273,83.2544,11.8935
200,64x4x32,8,65536,5505.0240,1572.8640,2359.2960,0.4798,27.3181,87.8537,12.5505
200,64x4x64,8,131072,11010.0480,3145.7280,4718.5920,1.2127,21.6162,111.0279,15.8611
200,64x8x2,8,8192,688.1280,196.6080,294.9120,0.0565,28.9784,82.8203,11.8315
200,64x8x4,8,16384,1376.2560,393.2160,589.8240,0.1131,28.9691,82.8468,11.8353
200,64x8x8,8,32768,2752.5120,786.4320,1179.6480,0.2251,29.1085,82.4501,11.7786
200,64x8x16,8,65536,5505.0240,1572.8640,2359.2960,0.4546,28.8315,83.2422,11.8917
200,64x8x32,8,131072,11010.0480,3145.7280,4718.5920,1.2090,21.6824,110.6888,15.8127
200,64x8x64,8,262144,22020.0960,6291.4560,9437.1840,3.2163,16.3010,147.2298,21.0328
200,64x16x2,8,16384,1376.2560,393.2160,589.8240,0.1152,28.4384,84.3931,12.0562
200,64x16x4,8,32768,2752.5120,786.4320,1179.6480,0.2273,28.8373,83.2257,11.8894
200,64x16x8,8,65536,5505.0240,1572.8640,2359.2960,0.4812,27.2382,88.1115,12.5874
200,64x16x16,8,131072,11010.0480,3145.7280,4718.5920,1.2232,21.4307,111.9891,15.9984
200,64x16x32,8,262144,22020.0960,6291.4560,9437.1840,3.1955,16.4069,146.2800,20.8971
200,64x16x64,8,524288,44040.1920,12582.9120,18874.3680,7.1952,14.5732,164.6858,23.5265
200,64x32x2,8,32768,2752.5120,786.4320,1179.6480,0.2265,28.9330,82.9502,11.8500
200,64x32x4,8,65536,5505.0240,1572.8640,2359.2960,0.5024,26.0894,91.9913,13.1416
200,64x32x8,8,131072,11010.0480,3145.7280,4718.5920,1.2188,21.5088,111.5821,15.9403
200,64x32x16,8,262144,22020.0960,6291.4560,9437.1840,3.2184,16.2901,147.3285,21.0469
200,64x32x32,8,524288,44040.1920,12582.9120,18874.3680,7.0597,14.8529,161.5841,23.0834
200,64x32x64,8,1048576,88080.3840,25165.8240,37748.7360,14.5806,14.3832,166.8613,23.8373
200,64x64x2,8,65536,5505.0240,1572.8640,2359.2960,0.4546,28.8336,83.2361,11.8909
200,64x64x4,8,131072,11010.0480,3145.7280,4718.5920,1.2400,21.1414,113.5215,16.2174
200,64x64x8,8,262144,22020.0960,6291.4560,9437.1840,3.2102,16.3320,146.9510,20.9930
200,64x64x16,8,524288,44040.1920,12582.9120,18874.3680,7.1222,14.7227,163.0139,23.2877
200,64x64x32,8,1048576,88080.3840,25165.8240,37748.7360,14.4043,14.5593,164.8437,23.5491
200,64x64x64,8,2097152,176160.7680,50331.6480,75497.4720,28.8310,14.5479,164.9724,23.5675
200,4x4x2,16,512,59.3920,12.2880,34.8160,0.0055,18.7721,127.8494,8.5233
200,4x4x4,16,1024,118.7840,24.5760,69.6320,0.0108,18.8919,127.0386,8.4692
200,4x4x8,16,2048,237.5680,49.1520,139.2640,0.0217,18.8390,127.3953,8.4930
200,4x4x16,16,4096,475.1360,98.3040,278.5280,0.0438,18.7125,128.2565,8.5504
200,4x4x32,16,8192,950.2720,196.6080,557.0560,0.0894,18.3265,130.9576,8.7305
200,4x4x64,16,16384,1900.5440,393.2160,1114.1120,0.1766,18.5508,129.3744,8.6250
200,4x8x2,16,1024,118.7840,24.5760,69.6320,0.0108,18.9210,126.8434,8.4562
200,4x8x4,16,2048,237.5680,49.1520,139.2640,0.0217,18.9109,126.9106,8.4607
200,4x8x8,16,4096,475.1360,98.3040,278.5280,0.0441,18.5708,129.2352,8.6157
200,4x8x16,16,8192,950.2720,196.6080,557.0560,0.0883,18.5564,129.3351,8.6223
200,4x8x32,16,16384,1900.5440,393.2160,1114.1120,0.1758,18.6440,128.7276,8.5818
200,4x8x64,16,32768,3801.0880,786.4320,2228.2240,0.3520,18.6169,128.9148,8.5943
200,4x16x2,16,2048,237.5680,49.1520,139.2640,0.0218,18.8228,127.5050,8.5003
200,4x16x4,16,4096,475.1360,98.3040,278.5280,0.0437,18.7451,128.0334,8.5356
200,4x16x8,16,8192,950.2720,196.6080,557.0560,0.0892,18.3753,130.6102,8.7073
200,4x16x16,16,16384,1900.5440,393.2160,1114.1120,0.1773,18.4770,129.8910,8.6594
200,4x16x32,16,32768,3801.0880,786.4320,2228.2240,0.3472,18.8739,127.1597,8.4773
200,4x16x64,16,65536,7602.1760,1572.8640,4456.4480,0.7188,18.2342,131.6211,8.7747
200,4x32x2,16,4096,475.1360,98.3040,278.5280,0.0432,18.9414,126.7065,8.4471
200,4x32x4,16,8192,950.2720,196.6080,557.0560,0.0887,18.4681,129.9539,8.6636
200,4x32x8,16,16384,1900.5440,393.2160,1114.1120,0.1779,18.4177,130.3097,8.6873
200,4x32x16,16,32768,3801.0880,786.4320,2228.2240,0.3515,18.6457,128.7157,8.5810
200,4x32x32,16,65536,7602.1760,1572.8640,4456.4480,0.7172,18.2746,131.3300,8.7553
200,4x32x64,16,131072,15204.3520,3145.7280,8912.8960,1.7695,14.8150,161.9980,10.7999
200,4x64x2,16,8192,950.2720,196.6080,557.0560,0.0878,18.6524,128.6696,8.5780
200,4x64x4,16,16384,1900.5440,393.2160,1114.1120,0.1791,18.2972,131.1673,8.7445
200,4x64x8,16,32768,3801.0880,786.4320,2228.2240,0.3522,18.6056,128.9933,8.5996
200,4x64x16,16,65536,7602.1760,1572.8640,4456.4480,0.7149,18.3332,130.9103,8.7274
200,4x64x32,16,131072,15204.3520,3145.7280,8912.8960,1.7628,14.8711,161.3873,10.7592
200,4x64x64,16,262144,30408.7040,6291.4560,17825.7920,4.0027,13.0982,183.2306,12.2154
200,8x4x2,16,1024,118.7840,24.5760,69.6320,0.0107,19.1269,125.4779,8.3652
200,8x4x4,16,2048,237.5680,49.1520,139.2640,0.0216,18.9319,126.7701,8.4513
200,8x4x8,16,4096,475.1360,98.3040,278.5280,0.0433,18.9287,126.7916,8.4528
200,8x4x16,16,8192,950.2720,196.6080,557.0560,0.0888,18.4489,130.0893,8.6726
200,8x4x32,16,16384,1900.5440,393.2160,1114.1120,0.1747,18.7517,127.9884,8.5326
200,8x4x64,16,32768,3801.0880,786.4320,2228.2240,0.3523,18.6023,129.0163,8.6011
200,8x8x2,16,2048,237.5680,49.1520,139.2640,0.0216,18.9696,126.5184,8.4346
200,8x8x4,16,4096,475.1360,98.3040,278.5280,0.0436,18.7761,127.8222,8.5215
200,8x8x8,16,8192,950.2720,196.6080,557.0560,0.0889,18.4242,130.2632,8.6842
200,8x8x16,16,16384,1900.5440,393.2160,1114.1120,0.1783,18.3747,130.6141,8.7076
200,8x8x32,16,32768,3801.0880,786.4320,2228.2240,0.3517,18.6356,128.7861,8.5857
200,8x8x64,16,65536,7602.1760,1572.8640,4456.4480,0.7266,18.0382,133.0508,8.8701
200,8x16x2,16,4096,475.1360,98.3040,278.5280,0.0439,18.6448,128.7220,8.5815
200,8x16x4,16,8192,950.2720,196.6080,557.0560,0.0902,18.1590,132.1657,8.8110
200,8x16x8,16,16384,1900.5440,393.2160,1114.1120,0.1773,18.4786,129.8799,8.6587
200,8x16x16,16,32768,3801.0880,786.4320,2228.2240,0.3516,18.6372,128.7747,8.5850
200,8x16x32,16,65536,7602.1760,1572.8640,4456.4480,0.7154,18.3217,130.9920,8.7328
200,8x16x64,16,131072,15204.3520,3145.7280,8912.8960,1.7763,14.7579,162.6247,10.8416
200,8x32x2,16,8192,950.2720,196.6080,557.0560,0.0881,18.6023,129.0164,8.6011
200,8x32x4,16,16384,1900.5440,393.2160,1114.1120,0.1782,18.3872,130.5254,8.7017
200,8x32x8,16,32768,3801.0880,786.4320,2228.2240,0.3524,18.5996,129.0351,8.6023
200,8x32x16,16,65536,7602.1760,1572.8640,4456.4480,0.7284,17.9940,133.3779,8.8919
200,8x32x32,16,131072,15204.3520,3145.7280,8912.8960,1.7605,14.8904,161.1771,10.7451
200,8x32x64,16,262144,30408.7040,6291.4560,17825.7920,4.0180,13.0484,183.9300,12.2620
200,8x64x2,16,16384,1900.5440,393.2160,1114.1120,0.1789,18.3191,131.0111,8.7341
200,8x64x4,16,32768,3801.0880,786.4320,2228.2240,0.3529,18.5730,129.2201,8.6147
200,8x64x8,16,65536,7602.1760,1572.8640,4456.4480,0.7318,17.9113,133.9936,8.9329
200,8x64x16,16,131072,15204.3520,3145.7280,8912.8960,1.7637,14.8629,161.4761,10.7651
200,8x64x32,16,262144,30408.7040,6291.4560,17825.7920,3.9553,13.2553,181.0596,12.0706
200,8x64x64,16,524288,60817.4080,12582.9120,35651.5840,8.4619,12.3918,193.6766,12.9118
200,16x4x2,16,2048,237.5680,49.1520,139.2640,0.0217,18.8926,127.0335,8.4689
200,16x4x4,16,4096,475.1360,98.3040,278.5280,0.0435,18.8480,127.3345,8.4890
200,16x4x8,16,8192,950.2720,196.6080,557.0560,0.0873,18.7736,127.8388,8.5226
200,16x4x16,16,16384,1900.5440,393.2160,1114.1120,0.1762,18.6018,129.0198,8.6013
200,16x4x32,16,32768,3801.0880,786.4320,2228.2240,0.3528,18.5762,129.1979,8.6132
200,16x4x64,16,65536,7602.1760,1572.8640,4456.4480,0.7303,17.9482,133.7180,8.9145
200,16x8x2,16,4096,475.1360,98.3040,278.5280,0.0438,18.7211,128.1975,8.5465
200,16x8x4,16,8192,950.2720,196.6080,557.0560,0.0895,18.3125,131.0580,8.7372
200,16x8x8,16,16384,1900.5440,393.2160,1114.1120,0.1794,18.2652,131.3976,8.7598
200,16x8x16,16,32768,3801.0880,786.4320,2228.2240,0.3518,18.6292,128.8299,8.5887
200,16x8x32,16,65536,7602.1760,1572.8640,4456.4480,0.7323,17.8998,134.0798,8.9387
200,16x8x64,16,131072,15204.3520,3145.7280,8912.8960,1.7832,14.7006,163.2583,10.8839
200,16x16x2,16,8192,950.2720,196.6080,557.0560,0.0876,18.7029,128.3222,8.5548
200,16x16x4,16,16384,1900.5440,393.2160,1114.1120,0.1778,18.4260,130.2506,8.6834
200,16x16x8,16,32768,3801.0880,786.4320,2228.2240,0.3532,18.5531,129.3583,8.6239
200,16x16x16,16,65536,7602.1760,1572.8640,4456.4480,0.7349,17.8365,134.5554,8.9704
200,16x16x32,16,131072,15204.3520,3145.7280,8912.8960,1.7469,15.0064,159.9322,10.6621
200,16x16x64,16,262144,30408.7040,6291.4560,17825.7920,3.9916,13.1349,182.7194,12.1813
200,16x32x2,16,16384,1900.5440,393.2160,1114.1120,0.1752,18.7055,128.3043,8.5536
200,16x32x4,16,32768,3801.0880,786.4320,2228.2240,0.3481,18.8292,127.4617,8.4974
200,16x32x8,16,65536,7602.1760,1572.8640,4456.4480,0.7822,16.7564,143.2289,9.5486
200,16x32x16,16,131072,15204.3520,3145.7280,8912.8960,1.7702,14.8089,162.0652,10.8043
200,16x32x32,16,262144,30408.7040,6291.4560,17825.7920,4.0154,13.0569,183.8103,12.2540
200,16x32x64,16,524288,60817.4080,12582.9120,35651.5840,8.4181,12.4562,192.6749,12.8450
200,16x64x2,16,32768,3801.0880,786.4320,2228.2240,0.3513,18.6578,128.6324,8.5755
200,16x64x4,16,65536,7602.1760,1572.8640,4456.4480,0.7349,17.8363,134.5568,8.9705
200,16x64x8,16,131072,15204.3520,3145.7280,8912.8960,1.7682,14.8256,161.8822,10.7921
200,16x64x16,16,262144,30408.7040,6291.4560,17825.7920,3.9491,13.2760,180.7771,12.0518
200,16x64x32,16,524288,60817.4080,12582.9120,35651.5840,8.4483,12.4117,193.3659,12.8911
200,16x64x64,16,1048576,121634.8160,25165.8240,71303.1680,16.9824,12.3490,194.3483,12.9566
200,32x4x2,16,4096,475.1360,98.3040,278.5280,0.0432,18.9428,126.6975,8.4465
200,32x4x4,16,8192,950.2720,196.6080,557.0560,0.0892,18.3690,130.6548,8.7103
200,32x4x8,16,16384,1900.5440,393.2160,1114.1120,0.1791,18.2945,131.1872,8.7458
200,32x4x16,16,32768,3801.0880,786.4320,2228.2240,0.3513,18.6530,128.6659,8.5777
200,32x4x32,16,65536,7602.1760,1572.8640,4456.4480,0.7125,18.3966,130.4586,8.6972
200,32x4x64,16,131072,15204.3520,3145.7280,8912.8960,1.7588,14.9051,161.0192,10.7346
200,32x8x2,16,8192,950.2720,196.6080,557.0560,0.0873,18.7703,127.8618,8.5241
200,32x8x4,16,16384,1900.5440,393.2160,1114.1120,0.1779,18.4195,130.2968,8.6865
200,32x8x8,16,32768,3801.0880,786.4320,2228.2240,0.3521,18.6148,128.9296,8.5953
200,32x8x16,16,65536,7602.1760,1572.8640,4456.4480,0.7187,18.2386,131.5892,8.7726
200,32x8x32,16,131072,15204.3520,3145.7280,8912.8960,1.7793,14.7330,162.8995,10.8600
200,32x8x64,16,262144,30408.7040,6291.4560,17825.7920,4.0073,13.0833,183.4398,12.2293
200,32x16x2,16,16384,1900.5440,393.2160,1114.1120,0.1772,18.4919,129.7865,8.6524
200,32x16x4,16,32768,3801.0880,786.4320,2228.2240,0.3529,18.5716,129.2299,8.6153
200,32x16x8,16,65536,7602.1760,1572.8640,4456.4480,0.7414,17.6795,135.7505,9.0500
200,32x16x16,16,131072,15204.3520,3145.7280,8912.8960,1.7616,14.8806,161.2834,10.7522
200,32x16x32,16,262144,30408.7040,6291.4560,17825.7920,3.9904,13.1388,182.6649,12.1777
200,32x16x64,16,524288,60817.4080,12582.9120,35651.5840,8.4067,12.4731,192.4144,12.8276
200,32x32x2,16,32768,3801.0880,786.4320,2228.2240,0.3511,18.6636,128.5925,8.5728
200,32x32x4,16,65536,7602.1760,1572.8640,4456.4480,0.7480,17.5225,136.9670,9.1311
200,32x32x8,16,131072,15204.3520,3145.7280,8912.8960,1.7637,14.8634,161.4702,10.7647
200,32x32x16,16,262144,30408.7040,6291.4560,17825.7920,3.9943,13.1260,182.8430,12.1895
200,32x32x32,16,524288,60817.4080,12582.9120,35651.5840,8.4761,12.3709,194.0033,12.9336
200,32x32x64,16,1048576,121634.8160,25165.8240,71303.1680,17.1450,12.2318,196.2091,13.0806
200,32x64x2,16,65536,7602.1760,1572.8640,4456.4480,0.7103,18.4529,130.0611,8.6707
200,32x64x4,16,131072,15204.3520,3145.7280,8912.8960,1.7754,14.7651,162.5455,10.8364
200,32x64x8,16,262144,30408.7040,6291.4560,17825.7920,3.9735,13.1946,181.8919,12.1261
200,32x64x16,16,524288,60817.4080,12582.9120,35651.5840,8.3487,12.5597,191.0876,12.7392
200,32x64x32,16,1048576,121634.8160,25165.8240,71303.1680,17.1582,12.2224,196.3605,13.0907
200,32x64x64,16,2097152,243269.6320,50331.6480,142606.3360,34.0888,12.3040,195.0579,13.0039
200,64x4x2,16,8192,950.2720,196.6080,557.0560,0.0891,18.3835,130.5519,8.7035
200,64x4x4,16,16384,1900.5440,393.2160,1114.1120,0.1777,18.4426,130.1336,8.6756
200,64x4x8,16,32768,3801.0880,786.4320,2228.2240,0.3487,18.7918,127.7156,8.5144
200,64x4x16,16,65536,7602.1760,1572.8640,4456.4480,0.7352,17.8288,134.6140,8.9743
200,64x4x32,16,131072,15204.3520,3145.7280,8912.8960,1.7606,14.8899,161.1832,10.7455
200,64x4x64,16,262144,30408.7040,6291.4560,17825.7920,3.9950,13.1237,182.8747,12.1916
200,64x8x2,16,16384,1900.5440,393.2160,1114.1120,0.1780,18.4140,130.3357,8.6890
200,64x8x4,16,32768,3801.0880,786.4320,2228.2240,0.3542,18.5019,129.7162,8.6477
200,64x8x8,16,65536,7602.1760,1572.8640,4456.4480,0.7467,17.5530,136.7290,9.1153
200,64x8x16,16,131072,15204.3520,3145.7280,8912.8960,1.7498,14.9809,160.2035,10.6802
200,64x8x32,16,262144,30408.7040,6291.4560,17825.7920,3.9909,13.1372,182.6874,12.1792
200,64x8x64,16,524288,60817.4080,12582.9120,35651.5840,8.3531,12.5531,191.1874,12.7458
200,64x16x2,16,32768,3801.0880,786.4320,2228.2240,0.3518,18.6265,128.8490,8.5899
200,64x16x4,16,65536,7602.1760,1572.8640,4456.4480,0.7085,18.5008,129.7240,8.6483
200,64x16x8,16,131072,15204.3520,3145.7280,8912.8960,1.7465,15.0097,159.8963,10.6598
200,64x16x16,16,262144,30408.7040,6291.4560,17825.7920,3.9689,13.2098,181.6827,12.1122
200,64x16x32,16,524288,60817.4080,12582.9120,35651.5840,8.3589,12.5444,191.3201,12.7547
200,64x16x64,16,1048576,121634.8160,25165.8240,71303.1680,17.0777,12.2800,195.4392,13.0293
200,64x32x2,16,65536,7602.1760,1572.8640,4456.4480,0.7205,18.1912,131.9318,8.7955
200,64x32x4,16,131072,15204.3520,3145.7280,8912.8960,1.7700,14.8100,162.0526,10.8035
200,64x32x8,16,262144,30408.7040,6291.4560,17825.7920,4.0141,13.0612,183.7507,12.2500
200,64x32x16,16,524288,60817.4080,12582.9120,35651.5840,8.3834,12.5077,191.8811,12.7921
200,64x32x32,16,1048576,121634.8160,25165.8240,71303.1680,17.0771,12.2805,195.4318,13.0288
200,64x32x64,16,2097152,243269.6320,50331.6480,142606.3360,34.2614,12.2421,196.0452,13.0697
200,64x64x2,16,131072,15204.3520,3145.7280,8912.8960,1.7763,14.7575,162.6287,10.8419
200,64x64x4,16,262144,30408.7040,6291.4560,17825.7920,3.9779,13.1802,182.0917,12.1394
200,64x64x8,16,524288,60817.4080,12582.9120,35651.5840,8.4382,12.4266,193.1347,12.8756
200,64x64x16,16,1048576,121634.8160,25165.8240,71303.1680,16.9325,12.3853,193.7776,12.9185
200,64x64x32,16,2097152,243269.6320,50331.6480,142606.3360,33.9828,12.3424,194.4513,12.9634
200,64x64x64,16,4194304,486539.2640,100663.2960,285212.6720,68.5011,12.2459,195.9834,13.0656
200,4x4x2,32,1024,184.3200,24.5760,135.1680,0.0174,11.7952,203.4727,6.5636
200,4x4x4,32,2048,368.6400,49.1520,270.3360,0.0353,11.6039,206.8273,6.6718
200,4x4x8,32,4096,737.2800,98.3040,540.6720,0.0708,11.5768,207.3121,6.6875
200,4x4x16,32,8192,1474.5600,196.6080,1081.3440,0.1432,11.4416,209.7618,6.7665
200,4x4x32,32,16384,2949.1200,393.2160,2162.6880,0.2832,11.5706,207.4224,6.6910
200,4x4x64,32,32768,5898.2400,786.4320,4325.3760,0.5676,11.5465,207.8548,6.7050
200,4x8x2,32,2048,368.6400,49.1520,270.3360,0.0348,11.7767,203.7917,6.5739
200,4x8x4,32,4096,737.2800,98.3040,540.6720,0.0719,11.4004,210.5186,6.7909
200,4x8x8,32,8192,1474.5600,196.6080,1081.3440,0.1435,11.4208,210.1434,6.7788
200,4x8x16,32,16384,2949.1200,393.2160,2162.6880,0.2865,11.4374,209.8376,6.7690
200,4x8x32,32,32768,5898.2400,786.4320,4325.3760,0.5630,11.6406,206.1756,6.6508
200,4x8x64,32,65536,11796.4800,1572.8640,8650.7520,1.1707,11.1959,214.3643,6.9150
200,4x16x2,32,4096,737.2800,98.3040,540.6720,0.0711,11.5250,208.2427,6.7175
200,4x16x4,32,8192,1474.5600,196.6080,1081.3440,0.1421,11.5337,208.0852,6.7124
200,4x16x8,32,16384,2949.1200,393.2160,2162.6880,0.2871,11.4149,210.2514,6.7823
200,4x16x16,32,32768,5898.2400,786.4320,4325.3760,0.5671,11.5565,207.6753,6.6992
200,4x16x32,32,65536,11796.4800,1572.8640,8650.7520,1.1876,11.0367,217.4561,7.0147
200,4x16x64,32,131072,23592.9600,3145.7280,17301.5040,2.6114,10.0384,239.0823,7.7123
200,4x32x2,32,8192,1474.5600,196.6080,1081.3440,0.1424,11.5080,208.5502,6.7274
200,4x32x4,32,16384,2949.1200,393.2160,2162.6880,0.2859,11.4602,209.4209,6.7555
200,4x32x8,32,32768,5898.2400,786.4320,4325.3760,0.5648,11.6029,206.8443,6.6724
200,4x32x16,32,65536,11796.4800,1572.8640,8650.7520,1.1923,10.9931,218.3185,7.0425
200,4x32x32,32,131072,23592.9600,3145.7280,17301.5040,2.6085,10.0497,238.8139,7.7037
200,4x32x64,32,262144,47185.9200,6291.4560,34603.0080,5.5256,9.4883,252.9427,8.1594
200,4x64x2,32,16384,2949.1200,393.2160,2162.6880,0.2875,11.3992,210.5419,6.7917
200,4x64x4,32,32768,5898.2400,786.4320,4325.3760,0.5719,11.4588,209.4459,6.7563
200,4x64x8,32,65536,11796.4800,1572.8640,8650.7520,1.1988,10.9339,219.5006,7.0807
200,4x64x16,32,131072,23592.9600,3145.7280,17301.5040,2.6122,10.0354,239.1525,7.7146
200,4x64x32,32,262144,47185.9200,6291.4560,34603.0080,5.4710,9.5830,250.4447,8.0789
200,4x64x64,32,524288,94371.8400,12582.9120,69206.0160,11.2159,9.3491,256.7105,8.2810
200,8x4x2,32,2048,368.6400,49.1520,270.3360,0.0353,11.5927,207.0267,6.6783
200,8x4x4,32,4096,737.2800,98.3040,540.6720,0.0720,11.3755,210.9803,6.8058
200,8x4x8,32,8192,1474.5600,196.6080,1081.3440,0.1429,11.4632,209.3662,6.7537
200,8x4x16,32,16384,2949.1200,393.2160,2162.6880,0.2858,11.4667,209.3019,6.7517
200,8x4x32,32,32768,5898.2400,786.4320,4325.3760,0.5713,11.4723,209.1999,6.7484
200,8x4x64,32,65536,11796.4800,1572.8640,8650.7520,1.1912,11.0032,218.1182,7.0361
200,8x8x2,32,4096,737.2800,98.3040,540.6720,0.0712,11.5041,208.6211,6.7297
200,8x8x4,32,8192,1474.5600,196.6080,1081.3440,0.1432,11.4410,209.7719,6.7668
200,8x8x8,32,16384,2949.1200,393.2160,2162.6880,0.2860,11.4576,209.4685,6.7570
200,8x8x16,32,32768,5898.2400,786.4320,4325.3760,0.5698,11.5010,208.6770,6.7315
200,8x8x32,32,65536,11796.4800,1572.8640,8650.7520,1.1786,11.1205,215.8170,6.9618
200,8x8x64,32,131072,23592.9600,3145.7280,17301.5040,2.6113,10.0388,239.0718,7.7120
200,8x16x2,32,8192,1474.5600,196.6080,1081.3440,0.1433,11.4328,209.9230,6.7717
200,8x16x4,32,16384,2949.1200,393.2160,2162.6880,0.2864,11.4420,209.7532,6.7662
200,8x16x8,32,32768,5898.2400,786.4320,4325.3760,0.5645,11.6087,206.7407,6.6691
200,8x16x16,32,65536,11796.4800,1572.8640,8650.7520,1.1612,11.2876,212.6224,6.8588
200,8x16x32,32,131072,23592.9600,3145.7280,17301.5040,2.6158,10.0214,239.4873,7.7254
200,8x16x64,32,262144,47185.9200,6291.4560,34603.0080,5.5078,9.5191,252.1257,8.1331
200,8x32x2,32,16384,2949.1200,393.2160,2162.6880,0.2867,11.4279,210.0125,6.7746
200,8x32x4,32,32768,5898.2400,786.4320,4325.3760,0.5686,11.5263,208.2201,6.7168
200,8x32x8,32,65536,11796.4800,1572.8640,8650.7520,1.1619,11.2809,212.7493,6.8629
200,8x32x16,32,131072,23592.9600,3145.7280,17301.5040,2.6084,10.0499,238.8074,7.7035
200,8x32x32,32,262144,47185.9200,6291.4560,34603.0080,5.5168,9.5035,252.5382,8.1464
200,8x32x64,32,524288,94371.8400,12582.9120,69206.0160,11.2449,9.3249,257.3764,8.3025
200,8x64x2,32,32768,5898.2400,786.4320,4325.3760,0.5702,11.4941,208.8035,6.7356
200,8x64x4,32,65536,11796.4800,1572.8640,8650.7520,1.1710,11.1933,214.4132,6.9166
200,8x64x8,32,131072,23592.9600,3145.7280,17301.5040,2.6195,10.0076,239.8179,7.7361
200,8x64x16,32,262144,47185.9200,6291.4560,34603.0080,5.4891,9.5514,251.2729,8.1056
200,8x64x32,32,524288,94371.8400,12582.9120,69206.0160,11.2059,9.3574,256.4817,8.2736
200,8x64x64,32,1048576,188743.6800,25165.8240,138412.0320,22.7045,9.2367,259.8329,8.3817
200,16x4x2,32,4096,737.2800,98.3040,540.6720,0.0714,11.4723,209.2005,6.7484
200,16x4x4,32,8192,1474.5600,196.6080,1081.3440,0.1413,11.5911,207.0553,6.6792
200,16x4x8,32,16384,2949.1200,393.2160,2162.6880,0.2868,11.4257,210.0525,6.7759
200,16x4x16,32,32768,5898.2400,786.4320,4325.3760,0.5702,11.4944,208.7970,6.7354
200,16x4x32,32,65536,11796.4800,1572.8640,8650.7520,1.1638,11.2621,213.1049,6.8744
200,16x4x64,32,131072,23592.9600,3145.7280,17301.5040,2.6182,10.0125,239.6999,7.7323
200,16x8x2,32,8192,1474.5600,196.6080,1081.3440,0.1429,11.4679,209.2805,6.7510
200,16x8x4,32,16384,2949.1200,393.2160,2162.6880,0.2874,11.4027,210.4763,6.7896
200,16x8x8,32,32768,5898.2400,786.4320,4325.3760,0.5699,11.4998,208.6998,6.7323
200,16x8x16,32,65536,11796.4800,1572.8640,8650.7520,1.1751,11.1544,215.1621,6.9407
200,16x8x32,32,131072,23592.9600,3145.7280,17301.5040,2.6103,10.0427,238.9795,7.7090
200,16x8x64,32,262144,47185.9200,6291.4560,34603.0080,5.5152,9.5062,252.4679,8.1441
200,16x16x2,32,16384,2949.1200,393.2160,2162.6880,0.2872,11.4108,210.3278,6.7848
200,16x16x4,32,32768,5898.2400,786.4320,4325.3760,0.5607,11.6888,205.3246,6.6234
200,16x16x8,32,65536,11796.4800,1572.8640,8650.7520,1.1865,11.0473,217.2482,7.0080
200,16x16x16,32,131072,23592.9600,3145.7280,17301.5040,2.6115,10.0381,239.0898,7.7126
200,16x16x32,32,262144,47185.9200,6291.4560,34603.0080,5.4684,9.5876,250.3234,8.0749
200,16x16x64,32,524288,94371.8400,12582.9120,69206.0160,11.2851,9.2917,258.2944,8.3321
200,16x32x2,32,32768,5898.2400,786.4320,4325.3760,0.5698,11.5016,208.6667,6.7312
200,16x32x4,32,65536,11796.4800,1572.8640,8650.7520,1.1800,11.1079,216.0618,6.9697
200,16x32x8,32,131072,23592.9600,3145.7280,17301.5040,2.6096,10.0455,238.9130,7.7069
200,16x32x16,32,262144,47185.9200,6291.4560,34603.0080,5.4894,9.5510,251.2829,8.1059
200,16x32x32,32,524288,94371.8400,12582.9120,69206.0160,11.2821,9.2941,258.2271,8.3299
200,16x32x64,32,1048576,188743.6800,25165.8240,138412.0320,22.7474,9.2193,260.3233,8.3975
200,16x64x2,32,65536,11796.4800,1572.8640,8650.7520,1.1741,11.1639,214.9782,6.9348
200,16x64x4,32,131072,23592.9600,3145.7280,17301.5040,2.6114,10.0384,239.0820,7.7123
200,16x64x8,32,262144,47185.9200,6291.4560,34603.0080,5.5263,9.4872,252.9729,8.1604
200,16x64x16,32,524288,94371.8400,12582.9120,69206.0160,11.3133,9.2686,258.9398,8.3529
200,16x64x32,32,1048576,188743.6800,25165.8240,138412.0320,22.8068,9.1953,261.0030,8.4195
200,16x64x64,32,2097152,377487.3600,50331.6480,276824.0640,45.4358,9.2313,259.9857,8.3866
200,32x4x2,32,8192,1474.5600,196.6080,1081.3440,0.1441,11.3732,211.0217,6.8072
200,32x4x4,32,16384,2949.1200,393.2160,2162.6880,0.2867,11.4281,210.0080,6.7745
200,32x4x8,32,32768,5898.2400,786.4320,4325.3760,0.5704,11.4899,208.8797,6.7381
200,32x4x16,32,65536,11796.4800,1572.8640,8650.7520,1.1706,11.1972,214.3391,6.9142
200,32x4x32,32,131072,23592.9600,3145.7280,17301.5040,2.6585,9.8606,243.3935,7.8514
200,32x4x64,32,262144,47185.9200,6291.4560,34603.0080,5.5104,9.5145,252.2460,8.1370
200,32x8x2,32,16384,2949.1200,393.2160,2162.6880,0.2863,11.4448,209.7019,6.7646
200,32x8x4,32,32768,5898.2400,786.4320,4325.3760,0.5707,11.4842,208.9832,6.7414
200,32x8x8,32,65536,11796.4800,1572.8640,8650.7520,1.1623,11.2773,212.8178,6.8651
200,32x8x16,32,131072,23592.9600,3145.7280,17301.5040,2.6152,10.0240,239.4244,7.7234
200,32x8x32,32,262144,47185.9200,6291.4560,34603.0080,5.4955,9.5403,251.5632,8.1149
200,32x8x64,32,524288,94371.8400,12582.9120,69206.0160,11.1827,9.3768,255.9510,8.2565
200,32x16x2,32,32768,5898.2400,786.4320,4325.3760,0.5703,11.4915,208.8507,6.7371
200,32x16x4,32,65536,11796.4800,1572.8640,8650.7520,1.1691,11.2114,214.0682,6.9054
200,32x16x8,32,131072,23592.9600,3145.7280,17301.5040,2.6149,10.0251,239.3981,7.7225
200,32x16x16,32,262144,47185.9200,6291.4560,34603.0080,5.5122,9.5114,252.3295,8.1397
200,32x16x32,32,524288,94371.8400,12582.9120,69206.0160,11.2002,9.3621,256.3522,8.2694
200,32x16x64,32,1048576,188743.6800,25165.8240,138412.0320,22.8756,9.1676,261.7905,8.4449
200,32x32x2,32,65536,11796.4800,1572.8640,8650.7520,1.1770,11.1358,215.5214,6.9523
200,32x32x4,32,131072,23592.9600,3145.7280,17301.5040,2.6142,10.0275,239.3407,7.7207
200,32x32x8,32,262144,47185.9200,6291.4560,34603.0080,5.5252,9.4890,252.9249,8.1589
200,32x32x16,32,524288,94371.8400,12582.9120,69206.0160,11.2002,9.3621,256.3532,8.2695
200,32x32x32,32,1048576,188743.6800,25165.8240,138412.0320,22.6890,9.2430,259.6552,8.3760
200,32x32x64,32,2097152,377487.3600,50331.6480,276824.0640,45.7402,9.1698,261.7278,8.4428
200,32x64x2,32,131072,23592.9600,3145.7280,17301.5040,2.6267,9.9801,240.4777,7.7573
200,32x64x4,32,262144,47185.9200,6291.4560,34603.0080,5.5090,9.5170,252.1814,8.1349
200,32x64x8,32,524288,94371.8400,12582.9120,69206.0160,11.3300,9.2549,259.3221,8.3652
200,32x64x16,32,1048576,188743.6800,25165.8240,138412.0320,22.7104,9.2343,259.9002,8.3839
200,32x64x32,32,2097152,377487.3600,50331.6480,276824.0640,45.6291,9.1922,261.0918,8.4223
200,32x64x64,32,4194304,754974.7200,100663.2960,553648.1280,90.8981,9.2286,260.0616,8.3891
200,64x4x2,32,16384,2949.1200,393.2160,2162.6880,0.2861,11.4553,209.5105,6.7584
200,64x4x4,32,32768,5898.2400,786.4320,4325.3760,0.5686,11.5252,208.2389,6.7174
200,64x4x8,32,65536,11796.4800,1572.8640,8650.7520,1.1681,11.2211,213.8826,6.8994
200,64x4x16,32,131072,23592.9600,3145.7280,17301.5040,2.5998,10.0834,238.0152,7.6779
200,64x4x32,32,262144,47185.9200,6291.4560,34603.0080,5.5037,9.5261,251.9403,8.1271
200,64x4x64,32,524288,94371.8400,12582.9120,69206.0160,11.2344,9.3336,257.1346,8.2947
200,64x8x2,32,32768,5898.2400,786.4320,4325.3760,0.5687,11.5235,208.2693,6.7184
200,64x8x4,32,65536,11796.4800,1572.8640,8650.7520,1.1901,11.0134,217.9169,7.0296
200,64x8x8,32,131072,23592.9600,3145.7280,17301.5040,2.6161,10.0205,239.5079,7.7261
200,64x8x16,32,262144,47185.9200,6291.4560,34603.0080,5.4939,9.5431,251.4914,8.1126
200,64x8x32,32,524288,94371.8400,12582.9120,69206.0160,11.2474,9.3229,257.4317,8.3042
200,64x8x64,32,1048576,188743.6800,25165.8240,138412.0320,22.7123,9.2336,259.9214,8.3846
200,64x16x2,32,65536,11796.4800,1572.8640,8650.7520,1.1692,11.2105,214.0853,6.9060
200,64x16x4,32,131072,23592.9600,3145.7280,17301.5040,2.6176,10.0147,239.6479,7.7306
200,64x16x8,32,262144,47185.9200,6291.4560,34603.0080,5.5163,9.5043,252.5162,8.1457
200,64x16x16,32,524288,94371.8400,12582.9120,69206.0160,11.2024,9.3603,256.4033,8.2711
200,64x16x32,32,1048576,188743.6800,25165.8240,138412.0320,22.7399,9.2224,260.2373,8.3948
200,64x16x64,32,2097152,377487.3600,50331.6480,276824.0640,45.6910,9.1797,261.4461,8.4337
200,64x32x2,32,131072,23592.9600,3145.7280,17301.5040,2.6132,10.0314,239.2490,7.7177
200,64x32x4,32,262144,47185.9200,6291.4560,34603.0080,5.4943,9.5423,251.5104,8.1132
200,64x32x8,32,524288,94371.8400,12582.9120,69206.0160,11.2644,9.3088,257.8213,8.3168
200,64x32x16,32,1048576,188743.6800,25165.8240,138412.0320,22.6715,9.2502,259.4542,8.3695
200,64x32x32,32,2097152,377487.3600,50331.6480,276824.0640,45.5770,9.2027,260.7935,8.4127
200,64x32x64,32,4194304,754974.7200,100663.2960,553648.1280,91.1329,9.2048,260.7334,8.4108
200,64x64x2,32,262144,47185.9200,6291.4560,34603.0080,5.4815,9.5647,250.9236,8.0943
200,64x64x4,32,524288,94371.8400,12582.9120,69206.0160,11.2564,9.3154,257.6377,8.3109
200,64x64x8,32,1048576,188743.6800,25165.8240,138412.0320,22.5064,9.3180,257.5657,8.3086
200,64x64x16,32,2097152,377487.3600,50331.6480,276824.0640,45.3069,9.2575,259.2479,8.3628
200,64x64x32,32,4194304,754974.7200,100663.2960,553648.1280,90.8428,9.2342,259.9032,8.3840
200,64x64x64,32,8388608,1509949.4400,201326.5920,1107296.2560,185.4831,9.0451,265.3357,8.5592

450
scripts/results_aos_medusa.txt
View File

@ -1,450 +0,0 @@
200,4x4x2,8,256,21.5040,6.1440,9.2160,0.0012,41.1057,80.3051,11.4722
200,4x4x4,8,512,43.0080,12.2880,18.4320,0.0025,40.3647,81.7793,11.6828
200,4x4x8,8,1024,86.0160,24.5760,36.8640,0.0051,39.8752,82.7833,11.8262
200,4x4x16,8,2048,172.0320,49.1520,73.7280,0.0101,40.6017,81.3020,11.6146
200,4x4x32,8,4096,344.0640,98.3040,147.4560,0.0204,40.2231,82.0673,11.7239
200,4x4x64,8,8192,688.1280,196.6080,294.9120,0.0472,34.6834,95.1752,13.5965
200,4x8x2,8,512,43.0080,12.2880,18.4320,0.0025,40.4194,81.6686,11.6669
200,4x8x4,8,1024,86.0160,24.5760,36.8640,0.0052,39.3566,83.8740,11.9820
200,4x8x8,8,2048,172.0320,49.1520,73.7280,0.0102,39.9745,82.5777,11.7968
200,4x8x16,8,4096,344.0640,98.3040,147.4560,0.0204,40.1524,82.2118,11.7445
200,4x8x32,8,8192,688.1280,196.6080,294.9120,0.0466,35.1455,93.9238,13.4177
200,4x8x64,8,16384,1376.2560,393.2160,589.8240,0.0929,35.2833,93.5569,13.3653
200,4x16x2,8,1024,86.0160,24.5760,36.8640,0.0052,39.5220,83.5232,11.9319
200,4x16x4,8,2048,172.0320,49.1520,73.7280,0.0102,40.0184,82.4871,11.7839
200,4x16x8,8,4096,344.0640,98.3040,147.4560,0.0207,39.5628,83.4369,11.9196
200,4x16x16,8,8192,688.1280,196.6080,294.9120,0.0468,35.0227,94.2531,13.4647
200,4x16x32,8,16384,1376.2560,393.2160,589.8240,0.0952,34.4175,95.9105,13.7015
200,4x16x64,8,32768,2752.5120,786.4320,1179.6480,0.1878,34.8933,94.6026,13.5147
200,4x32x2,8,2048,172.0320,49.1520,73.7280,0.0102,40.0417,82.4391,11.7770
200,4x32x4,8,4096,344.0640,98.3040,147.4560,0.0205,39.8714,82.7911,11.8273
200,4x32x8,8,8192,688.1280,196.6080,294.9120,0.0458,35.8069,92.1889,13.1698
200,4x32x16,8,16384,1376.2560,393.2160,589.8240,0.0955,34.3126,96.2037,13.7434
200,4x32x32,8,32768,2752.5120,786.4320,1179.6480,0.1895,34.5920,95.4266,13.6324
200,4x32x64,8,65536,5505.0240,1572.8640,2359.2960,0.3960,33.1002,99.7276,14.2468
200,4x64x2,8,4096,344.0640,98.3040,147.4560,0.0203,40.4125,81.6827,11.6690
200,4x64x4,8,8192,688.1280,196.6080,294.9120,0.0456,35.9633,91.7881,13.1126
200,4x64x8,8,16384,1376.2560,393.2160,589.8240,0.0934,35.0939,94.0620,13.4374
200,4x64x16,8,32768,2752.5120,786.4320,1179.6480,0.1896,34.5699,95.4876,13.6411
200,4x64x32,8,65536,5505.0240,1572.8640,2359.2960,0.3912,33.5037,98.5264,14.0752
200,4x64x64,8,131072,11010.0480,3145.7280,4718.5920,1.1277,23.2467,141.9985,20.2855
200,8x4x2,8,512,43.0080,12.2880,18.4320,0.0026,40.0831,82.3538,11.7648
200,8x4x4,8,1024,86.0160,24.5760,36.8640,0.0052,39.7363,83.0727,11.8675
200,8x4x8,8,2048,172.0320,49.1520,73.7280,0.0103,39.8957,82.7408,11.8201
200,8x4x16,8,4096,344.0640,98.3040,147.4560,0.0209,39.2829,84.0314,12.0045
200,8x4x32,8,8192,688.1280,196.6080,294.9120,0.0458,35.7476,92.3419,13.1917
200,8x4x64,8,16384,1376.2560,393.2160,589.8240,0.0946,34.6359,95.3057,13.6151
200,8x8x2,8,1024,86.0160,24.5760,36.8640,0.0051,40.0148,82.4944,11.7849
200,8x8x4,8,2048,172.0320,49.1520,73.7280,0.0103,39.6144,83.3284,11.9041
200,8x8x8,8,4096,344.0640,98.3040,147.4560,0.0205,39.8786,82.7763,11.8252
200,8x8x16,8,8192,688.1280,196.6080,294.9120,0.0464,35.3339,93.4229,13.3461
200,8x8x32,8,16384,1376.2560,393.2160,589.8240,0.0954,34.3429,96.1188,13.7313
200,8x8x64,8,32768,2752.5120,786.4320,1179.6480,0.1857,35.2944,93.5275,13.3611
200,8x16x2,8,2048,172.0320,49.1520,73.7280,0.0103,39.6254,83.3052,11.9007
200,8x16x4,8,4096,344.0640,98.3040,147.4560,0.0210,39.0435,84.5466,12.0781
200,8x16x8,8,8192,688.1280,196.6080,294.9120,0.0464,35.3417,93.4023,13.3432
200,8x16x16,8,16384,1376.2560,393.2160,589.8240,0.0952,34.4276,95.8824,13.6975
200,8x16x32,8,32768,2752.5120,786.4320,1179.6480,0.1901,34.4669,95.7730,13.6819
200,8x16x64,8,65536,5505.0240,1572.8640,2359.2960,0.3949,33.1919,99.4520,14.2074
200,8x32x2,8,4096,344.0640,98.3040,147.4560,0.0207,39.5242,83.5184,11.9312
200,8x32x4,8,8192,688.1280,196.6080,294.9120,0.0453,36.1420,91.3341,13.0477
200,8x32x8,8,16384,1376.2560,393.2160,589.8240,0.0943,34.7606,94.9638,13.5663
200,8x32x16,8,32768,2752.5120,786.4320,1179.6480,0.1897,34.5562,95.5256,13.6465
200,8x32x32,8,65536,5505.0240,1572.8640,2359.2960,0.3924,33.4054,98.8165,14.1166
200,8x32x64,8,131072,11010.0480,3145.7280,4718.5920,1.1113,23.5885,139.9413,19.9916
200,8x64x2,8,8192,688.1280,196.6080,294.9120,0.0464,35.3179,93.4654,13.3522
200,8x64x4,8,16384,1376.2560,393.2160,589.8240,0.0954,34.3379,96.1328,13.7333
200,8x64x8,8,32768,2752.5120,786.4320,1179.6480,0.1836,35.6978,92.4708,13.2101
200,8x64x16,8,65536,5505.0240,1572.8640,2359.2960,0.3942,33.2522,99.2717,14.1817
200,8x64x32,8,131072,11010.0480,3145.7280,4718.5920,1.1300,23.1979,142.2975,20.3282
200,8x64x64,8,262144,22020.0960,6291.4560,9437.1840,2.7875,18.8084,175.5068,25.0724
200,16x4x2,8,1024,86.0160,24.5760,36.8640,0.0051,39.8587,82.8175,11.8311
200,16x4x4,8,2048,172.0320,49.1520,73.7280,0.0102,40.2620,81.9879,11.7126
200,16x4x8,8,4096,344.0640,98.3040,147.4560,0.0205,39.9633,82.6008,11.8001
200,16x4x16,8,8192,688.1280,196.6080,294.9120,0.0467,35.0947,94.0599,13.4371
200,16x4x32,8,16384,1376.2560,393.2160,589.8240,0.0951,34.4516,95.8156,13.6879
200,16x4x64,8,32768,2752.5120,786.4320,1179.6480,0.1862,35.1990,93.7812,13.3973
200,16x8x2,8,2048,172.0320,49.1520,73.7280,0.0103,39.7484,83.0474,11.8639
200,16x8x4,8,4096,344.0640,98.3040,147.4560,0.0208,39.2926,84.0107,12.0015
200,16x8x8,8,8192,688.1280,196.6080,294.9120,0.0457,35.8223,92.1492,13.1642
200,16x8x16,8,16384,1376.2560,393.2160,589.8240,0.0955,34.3243,96.1708,13.7387
200,16x8x32,8,32768,2752.5120,786.4320,1179.6480,0.1852,35.3809,93.2990,13.3284
200,16x8x64,8,65536,5505.0240,1572.8640,2359.2960,0.3926,33.3876,98.8690,14.1241
200,16x16x2,8,4096,344.0640,98.3040,147.4560,0.0206,39.8296,82.8780,11.8397
200,16x16x4,8,8192,688.1280,196.6080,294.9120,0.0465,35.2205,93.7237,13.3891
200,16x16x8,8,16384,1376.2560,393.2160,589.8240,0.0950,34.4971,95.6893,13.6699
200,16x16x16,8,32768,2752.5120,786.4320,1179.6480,0.1854,35.3446,93.3949,13.3421
200,16x16x32,8,65536,5505.0240,1572.8640,2359.2960,0.3933,33.3269,99.0492,14.1499
200,16x16x64,8,131072,11010.0480,3145.7280,4718.5920,1.1221,23.3617,141.2995,20.1856
200,16x32x2,8,8192,688.1280,196.6080,294.9120,0.0465,35.2689,93.5953,13.3708
200,16x32x4,8,16384,1376.2560,393.2160,589.8240,0.0933,35.1150,94.0053,13.4293
200,16x32x8,8,32768,2752.5120,786.4320,1179.6480,0.1890,34.6767,95.1937,13.5991
200,16x32x16,8,65536,5505.0240,1572.8640,2359.2960,0.3917,33.4612,98.6515,14.0931
200,16x32x32,8,131072,11010.0480,3145.7280,4718.5920,1.1154,23.5026,140.4528,20.0647
200,16x32x64,8,262144,22020.0960,6291.4560,9437.1840,2.8702,18.2666,180.7123,25.8160
200,16x64x2,8,16384,1376.2560,393.2160,589.8240,0.0935,35.0393,94.2084,13.4583
200,16x64x4,8,32768,2752.5120,786.4320,1179.6480,0.1859,35.2627,93.6118,13.3731
200,16x64x8,8,65536,5505.0240,1572.8640,2359.2960,0.3860,33.9565,97.2126,13.8875
200,16x64x16,8,131072,11010.0480,3145.7280,4718.5920,1.1278,23.2434,142.0187,20.2884
200,16x64x32,8,262144,22020.0960,6291.4560,9437.1840,2.7969,18.7451,176.0992,25.1570
200,16x64x64,8,524288,44040.1920,12582.9120,18874.3680,6.0921,17.2122,191.7829,27.3976
200,32x4x2,8,2048,172.0320,49.1520,73.7280,0.0104,39.5116,83.5452,11.9350
200,32x4x4,8,4096,344.0640,98.3040,147.4560,0.0210,39.0648,84.5007,12.0715
200,32x4x8,8,8192,688.1280,196.6080,294.9120,0.0466,35.1938,93.7948,13.3993
200,32x4x16,8,16384,1376.2560,393.2160,589.8240,0.0943,34.7486,94.9965,13.5709
200,32x4x32,8,32768,2752.5120,786.4320,1179.6480,0.1895,34.5862,95.4426,13.6347
200,32x4x64,8,65536,5505.0240,1572.8640,2359.2960,0.3987,32.8783,100.4005,14.3429
200,32x8x2,8,4096,344.0640,98.3040,147.4560,0.0205,39.8861,82.7606,11.8229
200,32x8x4,8,8192,688.1280,196.6080,294.9120,0.0461,35.5303,92.9066,13.2724
200,32x8x8,8,16384,1376.2560,393.2160,589.8240,0.0945,34.6607,95.2374,13.6053
200,32x8x16,8,32768,2752.5120,786.4320,1179.6480,0.1869,35.0584,94.1573,13.4510
200,32x8x32,8,65536,5505.0240,1572.8640,2359.2960,0.3925,33.3962,98.8436,14.1205
200,32x8x64,8,131072,11010.0480,3145.7280,4718.5920,1.1238,23.3268,141.5113,20.2159
200,32x16x2,8,8192,688.1280,196.6080,294.9120,0.0457,35.8343,92.1184,13.1598
200,32x16x4,8,16384,1376.2560,393.2160,589.8240,0.0964,33.9864,97.1271,13.8753
200,32x16x8,8,32768,2752.5120,786.4320,1179.6480,0.1880,34.8516,94.7160,13.5309
200,32x16x16,8,65536,5505.0240,1572.8640,2359.2960,0.3958,33.1143,99.6850,14.2407
200,32x16x32,8,131072,11010.0480,3145.7280,4718.5920,1.1231,23.3408,141.4262,20.2037
200,32x16x64,8,262144,22020.0960,6291.4560,9437.1840,2.8733,18.2471,180.9059,25.8437
200,32x32x2,8,16384,1376.2560,393.2160,589.8240,0.0956,34.2772,96.3030,13.7576
200,32x32x4,8,32768,2752.5120,786.4320,1179.6480,0.1897,34.5447,95.5574,13.6511
200,32x32x8,8,65536,5505.0240,1572.8640,2359.2960,0.3997,32.7961,100.6523,14.3789
200,32x32x16,8,131072,11010.0480,3145.7280,4718.5920,1.1139,23.5333,140.2696,20.0385
200,32x32x32,8,262144,22020.0960,6291.4560,9437.1840,2.8750,18.2358,181.0173,25.8596
200,32x32x64,8,524288,44040.1920,12582.9120,18874.3680,6.0705,17.2733,191.1044,27.3006
200,32x64x2,8,32768,2752.5120,786.4320,1179.6480,0.1856,35.3144,93.4747,13.3535
200,32x64x4,8,65536,5505.0240,1572.8640,2359.2960,0.3870,33.8709,97.4582,13.9226
200,32x64x8,8,131072,11010.0480,3145.7280,4718.5920,1.1255,23.2919,141.7230,20.2461
200,32x64x16,8,262144,22020.0960,6291.4560,9437.1840,2.8226,18.5745,177.7170,25.3881
200,32x64x32,8,524288,44040.1920,12582.9120,18874.3680,6.1089,17.1648,192.3122,27.4732
200,32x64x64,8,1048576,88080.3840,25165.8240,37748.7360,12.4440,16.8527,195.8737,27.9820
200,64x4x2,8,4096,344.0640,98.3040,147.4560,0.0206,39.8061,82.9270,11.8467
200,64x4x4,8,8192,688.1280,196.6080,294.9120,0.0468,35.0209,94.2580,13.4654
200,64x4x8,8,16384,1376.2560,393.2160,589.8240,0.0953,34.3759,96.0265,13.7181
200,64x4x16,8,32768,2752.5120,786.4320,1179.6480,0.1905,34.3953,95.9725,13.7104
200,64x4x32,8,65536,5505.0240,1572.8640,2359.2960,0.4040,32.4474,101.7340,14.5334
200,64x4x64,8,131072,11010.0480,3145.7280,4718.5920,1.1196,23.4137,140.9856,20.1408
200,64x8x2,8,8192,688.1280,196.6080,294.9120,0.0469,34.9211,94.5273,13.5039
200,64x8x4,8,16384,1376.2560,393.2160,589.8240,0.0931,35.1895,93.8065,13.4009
200,64x8x8,8,32768,2752.5120,786.4320,1179.6480,0.1876,34.9341,94.4921,13.4989
200,64x8x16,8,65536,5505.0240,1572.8640,2359.2960,0.4015,32.6446,101.1194,14.4456
200,64x8x32,8,131072,11010.0480,3145.7280,4718.5920,1.1355,23.0863,142.9852,20.4265
200,64x8x64,8,262144,22020.0960,6291.4560,9437.1840,2.7960,18.7514,176.0401,25.1486
200,64x16x2,8,16384,1376.2560,393.2160,589.8240,0.0947,34.6156,95.3617,13.6231
200,64x16x4,8,32768,2752.5120,786.4320,1179.6480,0.1877,34.9064,94.5672,13.5096
200,64x16x8,8,65536,5505.0240,1572.8640,2359.2960,0.3966,33.0458,99.8916,14.2702
200,64x16x16,8,131072,11010.0480,3145.7280,4718.5920,1.1021,23.7865,138.7759,19.8251
200,64x16x32,8,262144,22020.0960,6291.4560,9437.1840,2.7874,18.8091,175.5005,25.0715
200,64x16x64,8,524288,44040.1920,12582.9120,18874.3680,6.1138,17.1511,192.4663,27.4952
200,64x32x2,8,32768,2752.5120,786.4320,1179.6480,0.1897,34.5455,95.5553,13.6508
200,64x32x4,8,65536,5505.0240,1572.8640,2359.2960,0.3956,33.1363,99.6189,14.2313
200,64x32x8,8,131072,11010.0480,3145.7280,4718.5920,1.1204,23.3981,141.0800,20.1543
200,64x32x16,8,262144,22020.0960,6291.4560,9437.1840,2.8010,18.7181,176.3538,25.1934
200,64x32x32,8,524288,44040.1920,12582.9120,18874.3680,6.1574,17.0296,193.8395,27.6914
200,64x32x64,8,1048576,88080.3840,25165.8240,37748.7360,12.3240,17.0169,193.9841,27.7120
200,64x64x2,8,65536,5505.0240,1572.8640,2359.2960,0.3888,33.7099,97.9237,13.9891
200,64x64x4,8,131072,11010.0480,3145.7280,4718.5920,1.1338,23.1208,142.7720,20.3960
200,64x64x8,8,262144,22020.0960,6291.4560,9437.1840,2.8212,18.5837,177.6291,25.3756
200,64x64x16,8,524288,44040.1920,12582.9120,18874.3680,6.1008,17.1875,192.0577,27.4368
200,64x64x32,8,1048576,88080.3840,25165.8240,37748.7360,12.4366,16.8628,195.7566,27.9652
200,64x64x64,8,2097152,176160.7680,50331.6480,75497.4720,25.4410,16.4864,200.2254,-61.7980
200,4x4x2,16,512,59.3920,12.2880,34.8160,0.0041,24.7324,133.4689,8.8979
200,4x4x4,16,1024,118.7840,24.5760,69.6320,0.0083,24.6887,133.7047,8.9136
200,4x4x8,16,2048,237.5680,49.1520,139.2640,0.0163,25.0903,131.5646,8.7710
200,4x4x16,16,4096,475.1360,98.3040,278.5280,0.0329,24.8829,132.6616,8.8441
200,4x4x32,16,8192,950.2720,196.6080,557.0560,0.0680,24.0966,136.9905,9.1327
200,4x4x64,16,16384,1900.5440,393.2160,1114.1120,0.1371,23.8973,138.1328,9.2089
200,4x8x2,16,1024,118.7840,24.5760,69.6320,0.0082,25.0360,131.8501,8.7900
200,4x8x4,16,2048,237.5680,49.1520,139.2640,0.0164,24.9590,132.2571,8.8171
200,4x8x8,16,4096,475.1360,98.3040,278.5280,0.0335,24.4561,134.9764,8.9984
200,4x8x16,16,8192,950.2720,196.6080,557.0560,0.0685,23.9268,137.9623,9.1975
200,4x8x32,16,16384,1900.5440,393.2160,1114.1120,0.1375,23.8339,138.5001,9.2333
200,4x8x64,16,32768,3801.0880,786.4320,2228.2240,0.2740,23.9222,137.9890,9.1993
200,4x16x2,16,2048,237.5680,49.1520,139.2640,0.0165,24.8840,132.6557,8.8437
200,4x16x4,16,4096,475.1360,98.3040,278.5280,0.0331,24.7245,133.5111,8.9007
200,4x16x8,16,8192,950.2720,196.6080,557.0560,0.0681,24.0635,137.1784,9.1452
200,4x16x16,16,16384,1900.5440,393.2160,1114.1120,0.1379,23.7594,138.9344,9.2623
200,4x16x32,16,32768,3801.0880,786.4320,2228.2240,0.2743,23.8917,138.1649,9.2110
200,4x16x64,16,65536,7602.1760,1572.8640,4456.4480,0.5828,22.4908,146.7711,9.7847
200,4x32x2,16,4096,475.1360,98.3040,278.5280,0.0332,24.6499,133.9152,8.9277
200,4x32x4,16,8192,950.2720,196.6080,557.0560,0.0691,23.7219,139.1539,9.2769
200,4x32x8,16,16384,1900.5440,393.2160,1114.1120,0.1368,23.9567,137.7903,9.1860
200,4x32x16,16,32768,3801.0880,786.4320,2228.2240,0.2735,23.9648,137.7438,9.1829
200,4x32x32,16,65536,7602.1760,1572.8640,4456.4480,0.5886,22.2698,148.2277,9.8818
200,4x32x64,16,131072,15204.3520,3145.7280,8912.8960,1.4720,17.8088,185.3577,12.3572
200,4x64x2,16,8192,950.2720,196.6080,557.0560,0.0682,24.0084,137.4936,9.1662
200,4x64x4,16,16384,1900.5440,393.2160,1114.1120,0.1372,23.8862,138.1971,9.2131
200,4x64x8,16,32768,3801.0880,786.4320,2228.2240,0.2742,23.9051,138.0878,9.2059
200,4x64x16,16,65536,7602.1760,1572.8640,4456.4480,0.5849,22.4077,147.3151,9.8210
200,4x64x32,16,131072,15204.3520,3145.7280,8912.8960,1.4458,18.1316,182.0577,12.1372
200,4x64x64,16,262144,30408.7040,6291.4560,17825.7920,3.1137,16.8378,196.0465,13.0698
200,8x4x2,16,1024,118.7840,24.5760,69.6320,0.0083,24.7704,133.2639,8.8843
200,8x4x4,16,2048,237.5680,49.1520,139.2640,0.0164,24.9790,132.1512,8.8101
200,8x4x8,16,4096,475.1360,98.3040,278.5280,0.0327,25.0873,131.5805,8.7720
200,8x4x16,16,8192,950.2720,196.6080,557.0560,0.0683,23.9916,137.5897,9.1726
200,8x4x32,16,16384,1900.5440,393.2160,1114.1120,0.1374,23.8546,138.3798,9.2253
200,8x4x64,16,32768,3801.0880,786.4320,2228.2240,0.2742,23.8990,138.1229,9.2082
200,8x8x2,16,2048,237.5680,49.1520,139.2640,0.0163,25.1862,131.0636,8.7376
200,8x8x4,16,4096,475.1360,98.3040,278.5280,0.0333,24.6350,133.9965,8.9331
200,8x8x8,16,8192,950.2720,196.6080,557.0560,0.0681,24.0493,137.2596,9.1506
200,8x8x16,16,16384,1900.5440,393.2160,1114.1120,0.1373,23.8580,138.3605,9.2240
200,8x8x32,16,32768,3801.0880,786.4320,2228.2240,0.2736,23.9502,137.8279,9.1885
200,8x8x64,16,65536,7602.1760,1572.8640,4456.4480,0.5870,22.3294,147.8318,9.8555
200,8x16x2,16,4096,475.1360,98.3040,278.5280,0.0331,24.7333,133.4638,8.8976
200,8x16x4,16,8192,950.2720,196.6080,557.0560,0.0686,23.8895,138.1779,9.2119
200,8x16x8,16,16384,1900.5440,393.2160,1114.1120,0.1371,23.8939,138.1526,9.2102
200,8x16x16,16,32768,3801.0880,786.4320,2228.2240,0.2742,23.9045,138.0909,9.2061
200,8x16x32,16,65536,7602.1760,1572.8640,4456.4480,0.5851,22.4035,147.3428,9.8229
200,8x16x64,16,131072,15204.3520,3145.7280,8912.8960,1.4449,18.1431,181.9423,12.1295
200,8x32x2,16,8192,950.2720,196.6080,557.0560,0.0685,23.9332,137.9258,9.1951
200,8x32x4,16,16384,1900.5440,393.2160,1114.1120,0.1354,24.1941,136.4383,9.0959
200,8x32x8,16,32768,3801.0880,786.4320,2228.2240,0.2742,23.9018,138.1068,9.2071
200,8x32x16,16,65536,7602.1760,1572.8640,4456.4480,0.5858,22.3747,147.5327,9.8355
200,8x32x32,16,131072,15204.3520,3145.7280,8912.8960,1.4728,17.7993,185.4566,12.3638
200,8x32x64,16,262144,30408.7040,6291.4560,17825.7920,3.1009,16.9075,195.2388,13.0159
200,8x64x2,16,16384,1900.5440,393.2160,1114.1120,0.1362,24.0517,137.2460,9.1497
200,8x64x4,16,32768,3801.0880,786.4320,2228.2240,0.2729,24.0178,137.4398,9.1627
200,8x64x8,16,65536,7602.1760,1572.8640,4456.4480,0.5883,22.2787,148.1687,9.8779
200,8x64x16,16,131072,15204.3520,3145.7280,8912.8960,1.4518,18.0563,182.8169,12.1878
200,8x64x32,16,262144,30408.7040,6291.4560,17825.7920,3.1235,16.7852,196.6615,13.1108
200,8x64x64,16,524288,60817.4080,12582.9120,35651.5840,6.4154,16.3447,201.9617,13.4641
200,16x4x2,16,2048,237.5680,49.1520,139.2640,0.0164,25.0188,131.9407,8.7960
200,16x4x4,16,4096,475.1360,98.3040,278.5280,0.0327,25.0635,131.7055,8.7804
200,16x4x8,16,8192,950.2720,196.6080,557.0560,0.0681,24.0625,137.1844,9.1456
200,16x4x16,16,16384,1900.5440,393.2160,1114.1120,0.1353,24.2172,136.3079,9.0872
200,16x4x32,16,32768,3801.0880,786.4320,2228.2240,0.2736,23.9531,137.8109,9.1874
200,16x4x64,16,65536,7602.1760,1572.8640,4456.4480,0.5894,22.2374,148.4434,9.8962
200,16x8x2,16,4096,475.1360,98.3040,278.5280,0.0331,24.7414,133.4200,8.8947
200,16x8x4,16,8192,950.2720,196.6080,557.0560,0.0681,24.0635,137.1789,9.1453
200,16x8x8,16,16384,1900.5440,393.2160,1114.1120,0.1377,23.7946,138.7287,9.2486
200,16x8x16,16,32768,3801.0880,786.4320,2228.2240,0.2731,23.9998,137.5431,9.1695
200,16x8x32,16,65536,7602.1760,1572.8640,4456.4480,0.5872,22.3223,147.8793,9.8586
200,16x8x64,16,131072,15204.3520,3145.7280,8912.8960,1.4703,17.8294,185.1437,12.3429
200,16x16x2,16,8192,950.2720,196.6080,557.0560,0.0687,23.8416,138.4552,9.2303
200,16x16x4,16,16384,1900.5440,393.2160,1114.1120,0.1375,23.8375,138.4791,9.2319
200,16x16x8,16,32768,3801.0880,786.4320,2228.2240,0.2727,24.0302,137.3686,9.1579
200,16x16x16,16,65536,7602.1760,1572.8640,4456.4480,0.5883,22.2792,148.1652,9.8777
200,16x16x32,16,131072,15204.3520,3145.7280,8912.8960,1.4533,18.0375,183.0080,12.2005
200,16x16x64,16,262144,30408.7040,6291.4560,17825.7920,3.1294,16.7536,197.0325,13.1355
200,16x32x2,16,16384,1900.5440,393.2160,1114.1120,0.1378,23.7792,138.8186,9.2546
200,16x32x4,16,32768,3801.0880,786.4320,2228.2240,0.2724,24.0554,137.2249,9.1483
200,16x32x8,16,65536,7602.1760,1572.8640,4456.4480,0.5840,22.4451,147.0702,9.8047
200,16x32x16,16,131072,15204.3520,3145.7280,8912.8960,1.4553,18.0128,183.2583,12.2172
200,16x32x32,16,262144,30408.7040,6291.4560,17825.7920,3.1316,16.7421,197.1674,13.1445
200,16x32x64,16,524288,60817.4080,12582.9120,35651.5840,6.5036,16.1231,204.7372,13.6491
200,16x64x2,16,32768,3801.0880,786.4320,2228.2240,0.2731,23.9947,137.5723,9.1715
200,16x64x4,16,65536,7602.1760,1572.8640,4456.4480,0.5852,22.3992,147.3712,9.8247
200,16x64x8,16,131072,15204.3520,3145.7280,8912.8960,1.4516,18.0592,182.7882,12.1859
200,16x64x16,16,262144,30408.7040,6291.4560,17825.7920,3.1097,16.8599,195.7903,13.0527
200,16x64x32,16,524288,60817.4080,12582.9120,35651.5840,6.5056,16.1181,204.8003,13.6534
200,16x64x64,16,1048576,121634.8160,25165.8240,71303.1680,13.1053,16.0023,206.2823,-37.6428
200,32x4x2,16,4096,475.1360,98.3040,278.5280,0.0328,24.9512,132.2984,8.8199
200,32x4x4,16,8192,950.2720,196.6080,557.0560,0.0682,24.0216,137.4177,9.1612
200,32x4x8,16,16384,1900.5440,393.2160,1114.1120,0.1377,23.7965,138.7181,9.2479
200,32x4x16,16,32768,3801.0880,786.4320,2228.2240,0.2730,24.0046,137.5155,9.1677
200,32x4x32,16,65536,7602.1760,1572.8640,4456.4480,0.5829,22.4881,146.7890,9.7859
200,32x4x64,16,131072,15204.3520,3145.7280,8912.8960,1.4484,18.0992,182.3833,12.1589
200,32x8x2,16,8192,950.2720,196.6080,557.0560,0.0677,24.1845,136.4923,9.0995
200,32x8x4,16,16384,1900.5440,393.2160,1114.1120,0.1374,23.8416,138.4555,9.2304
200,32x8x8,16,32768,3801.0880,786.4320,2228.2240,0.2738,23.9371,137.9030,9.1935
200,32x8x16,16,65536,7602.1760,1572.8640,4456.4480,0.5882,22.2849,148.1270,9.8751
200,32x8x32,16,131072,15204.3520,3145.7280,8912.8960,1.4562,18.0018,183.3708,12.2247
200,32x8x64,16,262144,30408.7040,6291.4560,17825.7920,3.0955,16.9373,194.8949,12.9930
200,32x16x2,16,16384,1900.5440,393.2160,1114.1120,0.1377,23.7916,138.7464,9.2498
200,32x16x4,16,32768,3801.0880,786.4320,2228.2240,0.2742,23.8984,138.1263,9.2084
200,32x16x8,16,65536,7602.1760,1572.8640,4456.4480,0.5860,22.3662,147.5890,9.8393
200,32x16x16,16,131072,15204.3520,3145.7280,8912.8960,1.4521,18.0527,182.8531,12.1902
200,32x16x32,16,262144,30408.7040,6291.4560,17825.7920,3.1149,16.8315,196.1201,13.0747
200,32x16x64,16,524288,60817.4080,12582.9120,35651.5840,6.4845,16.1705,204.1378,13.6092
200,32x32x2,16,32768,3801.0880,786.4320,2228.2240,0.2740,23.9190,138.0077,9.2005
200,32x32x4,16,65536,7602.1760,1572.8640,4456.4480,0.5801,22.5958,146.0891,9.7393
200,32x32x8,16,131072,15204.3520,3145.7280,8912.8960,1.4427,18.1702,181.6708,12.1114
200,32x32x16,16,262144,30408.7040,6291.4560,17825.7920,3.1059,16.8805,195.5508,13.0367
200,32x32x32,16,524288,60817.4080,12582.9120,35651.5840,6.5648,15.9727,206.6647,13.7776
200,32x32x64,16,1048576,121634.8160,25165.8240,71303.1680,13.4465,15.5962,211.6535,-38.6229
200,32x64x2,16,65536,7602.1760,1572.8640,4456.4480,0.5794,22.6212,145.9248,9.7283
200,32x64x4,16,131072,15204.3520,3145.7280,8912.8960,1.4597,17.9582,183.8159,12.2544
200,32x64x8,16,262144,30408.7040,6291.4560,17825.7920,3.1126,16.8442,195.9724,13.0648
200,32x64x16,16,524288,60817.4080,12582.9120,35651.5840,6.5076,16.1131,204.8649,13.6577
200,32x64x32,16,1048576,121634.8160,25165.8240,71303.1680,13.2052,15.8813,207.8550,-37.9298
200,32x64x64,16,2097152,243269.6320,50331.6480,142606.3360,26.5533,15.7958,208.9801,43.9034
200,64x4x2,16,8192,950.2720,196.6080,557.0560,0.0676,24.2458,136.1475,9.0765
200,64x4x4,16,16384,1900.5440,393.2160,1114.1120,0.1387,23.6314,139.6870,9.3125
200,64x4x8,16,32768,3801.0880,786.4320,2228.2240,0.2728,24.0216,137.4180,9.1612
200,64x4x16,16,65536,7602.1760,1572.8640,4456.4480,0.5876,22.3071,147.9798,9.8653
200,64x4x32,16,131072,15204.3520,3145.7280,8912.8960,1.4662,17.8796,184.6236,12.3082
200,64x4x64,16,262144,30408.7040,6291.4560,17825.7920,3.1136,16.8386,196.0381,13.0692
200,64x8x2,16,16384,1900.5440,393.2160,1114.1120,0.1372,23.8911,138.1686,9.2112
200,64x8x4,16,32768,3801.0880,786.4320,2228.2240,0.2738,23.9316,137.9348,9.1957
200,64x8x8,16,65536,7602.1760,1572.8640,4456.4480,0.5872,22.3233,147.8722,9.8581
200,64x8x16,16,131072,15204.3520,3145.7280,8912.8960,1.4458,18.1310,182.0638,12.1376
200,64x8x32,16,262144,30408.7040,6291.4560,17825.7920,3.1368,16.7141,197.4976,13.1665
200,64x8x64,16,524288,60817.4080,12582.9120,35651.5840,6.4162,16.3427,201.9857,13.4657
200,64x16x2,16,32768,3801.0880,786.4320,2228.2240,0.2740,23.9176,138.0158,9.2011
200,64x16x4,16,65536,7602.1760,1572.8640,4456.4480,0.5800,22.5988,146.0695,9.7380
200,64x16x8,16,131072,15204.3520,3145.7280,8912.8960,1.4595,17.9611,183.7857,12.2524
200,64x16x16,16,262144,30408.7040,6291.4560,17825.7920,3.0984,16.9211,195.0822,13.0055
200,64x16x32,16,524288,60817.4080,12582.9120,35651.5840,6.4903,16.1561,204.3191,13.6213
200,64x16x64,16,1048576,121634.8160,25165.8240,71303.1680,13.1687,15.9252,207.2811,-37.8250
200,64x32x2,16,65536,7602.1760,1572.8640,4456.4480,0.5855,22.3859,147.4589,9.8306
200,64x32x4,16,131072,15204.3520,3145.7280,8912.8960,1.4746,17.7769,185.6908,12.3794
200,64x32x8,16,262144,30408.7040,6291.4560,17825.7920,3.1073,16.8730,195.6382,13.0425
200,64x32x16,16,524288,60817.4080,12582.9120,35651.5840,6.4985,16.1356,204.5791,13.6386
200,64x32x32,16,1048576,121634.8160,25165.8240,71303.1680,13.3353,15.7263,209.9025,-38.3034
200,64x32x64,16,2097152,243269.6320,50331.6480,142606.3360,26.9030,15.5905,211.7316,44.4814
200,64x64x2,16,131072,15204.3520,3145.7280,8912.8960,1.4540,18.0290,183.0936,12.2062
200,64x64x4,16,262144,30408.7040,6291.4560,17825.7920,3.1314,16.7431,197.1556,13.1437
200,64x64x8,16,524288,60817.4080,12582.9120,35651.5840,6.4141,16.3480,201.9211,13.4614
200,64x64x16,16,1048576,121634.8160,25165.8240,71303.1680,13.2231,15.8597,208.1375,-37.9813
200,64x64x32,16,2097152,243269.6320,50331.6480,142606.3360,26.2168,15.9985,206.3317,43.3470
200,64x64x64,16,4194304,486539.2640,100663.2960,285212.6720,53.6958,15.6225,211.2984,-586.9401
200,4x4x2,32,1024,184.3200,24.5760,135.1680,0.0134,15.2346,216.6777,6.9896
200,4x4x4,32,2048,368.6400,49.1520,270.3360,0.0266,15.4085,214.2323,6.9107
200,4x4x8,32,4096,737.2800,98.3040,540.6720,0.0545,15.0233,219.7251,7.0879
200,4x4x16,32,8192,1474.5600,196.6080,1081.3440,0.1093,14.9944,220.1483,7.1016
200,4x4x32,32,16384,2949.1200,393.2160,2162.6880,0.2193,14.9437,220.8951,7.1256
200,4x4x64,32,32768,5898.2400,786.4320,4325.3760,0.4379,14.9647,220.5860,7.1157
200,4x8x2,32,2048,368.6400,49.1520,270.3360,0.0267,15.3207,215.4599,6.9503
200,4x8x4,32,4096,737.2800,98.3040,540.6720,0.0546,15.0008,220.0543,7.0985
200,4x8x8,32,8192,1474.5600,196.6080,1081.3440,0.1094,14.9822,220.3278,7.1073
200,4x8x16,32,16384,2949.1200,393.2160,2162.6880,0.2191,14.9583,220.6805,7.1187
200,4x8x32,32,32768,5898.2400,786.4320,4325.3760,0.4374,14.9848,220.2903,7.1061
200,4x8x64,32,65536,11796.4800,1572.8640,8650.7520,0.9374,13.9828,236.0765,7.6154
200,4x16x2,32,4096,737.2800,98.3040,540.6720,0.0541,15.1494,217.8961,7.0289
200,4x16x4,32,8192,1474.5600,196.6080,1081.3440,0.1089,15.0502,219.3323,7.0752
200,4x16x8,32,16384,2949.1200,393.2160,2162.6880,0.2189,14.9698,220.5113,7.1133
200,4x16x16,32,32768,5898.2400,786.4320,4325.3760,0.4385,14.9443,220.8864,7.1254
200,4x16x32,32,65536,11796.4800,1572.8640,8650.7520,0.9337,14.0372,235.1611,7.5858
200,4x16x64,32,131072,23592.9600,3145.7280,17301.5040,2.0220,12.9647,254.6154,8.2134
200,4x32x2,32,8192,1474.5600,196.6080,1081.3440,0.1087,15.0688,219.0623,7.0665
200,4x32x4,32,16384,2949.1200,393.2160,2162.6880,0.2187,14.9847,220.2915,7.1062
200,4x32x8,32,32768,5898.2400,786.4320,4325.3760,0.4346,15.0800,218.8995,7.0613
200,4x32x16,32,65536,11796.4800,1572.8640,8650.7520,0.9384,13.9682,236.3232,7.6233
200,4x32x32,32,131072,23592.9600,3145.7280,17301.5040,2.0175,12.9935,254.0510,8.1952
200,4x32x64,32,262144,47185.9200,6291.4560,34603.0080,4.1609,12.6004,261.9751,8.4508
200,4x64x2,32,16384,2949.1200,393.2160,2162.6880,0.2187,14.9811,220.3445,7.1079
200,4x64x4,32,32768,5898.2400,786.4320,4325.3760,0.4385,14.9470,220.8474,7.1241
200,4x64x8,32,65536,11796.4800,1572.8640,8650.7520,0.9425,13.9075,237.3533,7.6566
200,4x64x16,32,131072,23592.9600,3145.7280,17301.5040,2.0193,12.9817,254.2807,8.2026
200,4x64x32,32,262144,47185.9200,6291.4560,34603.0080,4.1060,12.7687,258.5221,8.3394
200,4x64x64,32,524288,94371.8400,12582.9120,69206.0160,8.5023,12.3329,267.6588,-26.8734
200,8x4x2,32,2048,368.6400,49.1520,270.3360,0.0269,15.2501,216.4575,6.9825
200,8x4x4,32,4096,737.2800,98.3040,540.6720,0.0546,15.0017,220.0414,7.0981
200,8x4x8,32,8192,1474.5600,196.6080,1081.3440,0.1090,15.0299,219.6287,7.0848
200,8x4x16,32,16384,2949.1200,393.2160,2162.6880,0.2186,14.9877,220.2466,7.1047
200,8x4x32,32,32768,5898.2400,786.4320,4325.3760,0.4382,14.9568,220.7025,7.1194
200,8x4x64,32,65536,11796.4800,1572.8640,8650.7520,0.9368,13.9922,235.9171,7.6102
200,8x8x2,32,4096,737.2800,98.3040,540.6720,0.0542,15.1196,218.3266,7.0428
200,8x8x4,32,8192,1474.5600,196.6080,1081.3440,0.1092,15.0007,220.0570,7.0986
200,8x8x8,32,16384,2949.1200,393.2160,2162.6880,0.2159,15.1799,217.4583,7.0148
200,8x8x16,32,32768,5898.2400,786.4320,4325.3760,0.4374,14.9815,220.3387,7.1077
200,8x8x32,32,65536,11796.4800,1572.8640,8650.7520,0.9386,13.9650,236.3773,7.6251
200,8x8x64,32,131072,23592.9600,3145.7280,17301.5040,2.0236,12.9544,254.8161,8.2199
200,8x16x2,32,8192,1474.5600,196.6080,1081.3440,0.1093,14.9932,220.1666,7.1021
200,8x16x4,32,16384,2949.1200,393.2160,2162.6880,0.2187,14.9802,220.3572,7.1083
200,8x16x8,32,32768,5898.2400,786.4320,4325.3760,0.4388,14.9344,221.0333,7.1301
200,8x16x16,32,65536,11796.4800,1572.8640,8650.7520,0.9327,14.0529,234.8988,7.5774
200,8x16x32,32,131072,23592.9600,3145.7280,17301.5040,2.0367,12.8711,256.4670,8.2731
200,8x16x64,32,262144,47185.9200,6291.4560,34603.0080,4.1018,12.7819,258.2559,8.3308
200,8x32x2,32,16384,2949.1200,393.2160,2162.6880,0.2188,14.9790,220.3750,7.1089
200,8x32x4,32,32768,5898.2400,786.4320,4325.3760,0.4386,14.9426,220.9122,7.1262
200,8x32x8,32,65536,11796.4800,1572.8640,8650.7520,0.9360,14.0039,235.7196,7.6039
200,8x32x16,32,131072,23592.9600,3145.7280,17301.5040,2.0310,12.9071,255.7514,8.2500
200,8x32x32,32,262144,47185.9200,6291.4560,34603.0080,4.1461,12.6452,261.0469,8.4209
200,8x32x64,32,524288,94371.8400,12582.9120,69206.0160,8.3325,12.5842,262.3128,-26.3366
200,8x64x2,32,32768,5898.2400,786.4320,4325.3760,0.4385,14.9469,220.8478,7.1241
200,8x64x4,32,65536,11796.4800,1572.8640,8650.7520,0.9403,13.9390,236.8176,7.6393
200,8x64x8,32,131072,23592.9600,3145.7280,17301.5040,2.0259,12.9397,255.1062,8.2292
200,8x64x16,32,262144,47185.9200,6291.4560,34603.0080,4.1068,12.7662,258.5733,8.3411
200,8x64x32,32,524288,94371.8400,12582.9120,69206.0160,8.4559,12.4005,266.1989,-26.7268
200,8x64x64,32,1048576,188743.6800,25165.8240,138412.0320,17.0485,12.3011,268.3504,-26.9428
200,16x4x2,32,4096,737.2800,98.3040,540.6720,0.0545,15.0353,219.5506,7.0823
200,16x4x4,32,8192,1474.5600,196.6080,1081.3440,0.1083,15.1228,218.2803,7.0413
200,16x4x8,32,16384,2949.1200,393.2160,2162.6880,0.2193,14.9435,220.8982,7.1257
200,16x4x16,32,32768,5898.2400,786.4320,4325.3760,0.4371,14.9928,220.1717,7.1023
200,16x4x32,32,65536,11796.4800,1572.8640,8650.7520,0.9377,13.9786,236.1473,7.6177
200,16x4x64,32,131072,23592.9600,3145.7280,17301.5040,2.0308,12.9084,255.7246,8.2492
200,16x8x2,32,8192,1474.5600,196.6080,1081.3440,0.1092,14.9993,220.0768,7.0993
200,16x8x4,32,16384,2949.1200,393.2160,2162.6880,0.2156,15.2008,217.1597,7.0052
200,16x8x8,32,32768,5898.2400,786.4320,4325.3760,0.4348,15.0713,219.0254,7.0653
200,16x8x16,32,65536,11796.4800,1572.8640,8650.7520,0.9381,13.9724,236.2512,7.6210
200,16x8x32,32,131072,23592.9600,3145.7280,17301.5040,2.0138,13.0174,253.5827,8.1801
200,16x8x64,32,262144,47185.9200,6291.4560,34603.0080,4.1049,12.7722,258.4526,8.3372
200,16x16x2,32,16384,2949.1200,393.2160,2162.6880,0.2196,14.9241,221.1865,7.1350
200,16x16x4,32,32768,5898.2400,786.4320,4325.3760,0.4374,14.9833,220.3117,7.1068
200,16x16x8,32,65536,11796.4800,1572.8640,8650.7520,0.9391,13.9572,236.5088,7.6293
200,16x16x16,32,131072,23592.9600,3145.7280,17301.5040,2.0364,12.8726,256.4356,8.2721
200,16x16x32,32,262144,47185.9200,6291.4560,34603.0080,4.1083,12.7618,258.6626,8.3440
200,16x16x64,32,524288,94371.8400,12582.9120,69206.0160,8.5071,12.3259,267.8100,-26.8886
200,16x32x2,32,32768,5898.2400,786.4320,4325.3760,0.4374,14.9829,220.3184,7.1070
200,16x32x4,32,65536,11796.4800,1572.8640,8650.7520,0.9337,14.0376,235.1549,7.5856
200,16x32x8,32,131072,23592.9600,3145.7280,17301.5040,2.0285,12.9228,255.4397,8.2400
200,16x32x16,32,262144,47185.9200,6291.4560,34603.0080,4.1517,12.6282,261.3994,8.4322
200,16x32x32,32,524288,94371.8400,12582.9120,69206.0160,8.3354,12.5797,262.4065,-26.3460
200,16x32x64,32,1048576,188743.6800,25165.8240,138412.0320,17.1852,12.2033,270.5016,-27.1588
200,16x64x2,32,65536,11796.4800,1572.8640,8650.7520,0.9372,13.9857,236.0275,7.6138
200,16x64x4,32,131072,23592.9600,3145.7280,17301.5040,2.0181,12.9897,254.1243,8.1976
200,16x64x8,32,262144,47185.9200,6291.4560,34603.0080,4.1102,12.7557,258.7871,8.3480
200,16x64x16,32,524288,94371.8400,12582.9120,69206.0160,8.5143,12.3155,268.0368,-26.9113
200,16x64x32,32,1048576,188743.6800,25165.8240,138412.0320,16.7462,12.5231,263.5925,-26.4651
200,16x64x64,32,2097152,377487.3600,50331.6480,276824.0640,34.3474,12.2114,270.3206,965.4307
200,32x4x2,32,8192,1474.5600,196.6080,1081.3440,0.1091,15.0122,219.8883,7.0932
200,32x4x4,32,16384,2949.1200,393.2160,2162.6880,0.2191,14.9554,220.7223,7.1201
200,32x4x8,32,32768,5898.2400,786.4320,4325.3760,0.4372,14.9912,220.1951,7.1031
200,32x4x16,32,65536,11796.4800,1572.8640,8650.7520,0.9347,14.0224,235.4083,7.5938
200,32x4x32,32,131072,23592.9600,3145.7280,17301.5040,2.0181,12.9898,254.1230,8.1975
200,32x4x64,32,262144,47185.9200,6291.4560,34603.0080,4.0964,12.7986,257.9188,8.3200
200,32x8x2,32,16384,2949.1200,393.2160,2162.6880,0.2191,14.9549,220.7300,7.1203
200,32x8x4,32,32768,5898.2400,786.4320,4325.3760,0.4380,14.9628,220.6131,7.1166
200,32x8x8,32,65536,11796.4800,1572.8640,8650.7520,0.9370,13.9889,235.9727,7.6120
200,32x8x16,32,131072,23592.9600,3145.7280,17301.5040,2.0005,13.1036,251.9156,8.1263
200,32x8x32,32,262144,47185.9200,6291.4560,34603.0080,4.1569,12.6125,261.7241,8.4427
200,32x8x64,32,524288,94371.8400,12582.9120,69206.0160,8.4536,12.4039,266.1250,-26.7194
200,32x16x2,32,32768,5898.2400,786.4320,4325.3760,0.4376,14.9762,220.4158,7.1102
200,32x16x4,32,65536,11796.4800,1572.8640,8650.7520,0.9325,14.0565,234.8381,7.5754
200,32x16x8,32,131072,23592.9600,3145.7280,17301.5040,2.0189,12.9848,254.2212,8.2007
200,32x16x16,32,262144,47185.9200,6291.4560,34603.0080,4.1342,12.6818,260.2952,8.3966
200,32x16x32,32,524288,94371.8400,12582.9120,69206.0160,8.4425,12.4201,265.7779,-26.6845
200,32x16x64,32,1048576,188743.6800,25165.8240,138412.0320,16.8367,12.4558,265.0161,-26.6080
200,32x32x2,32,65536,11796.4800,1572.8640,8650.7520,0.9372,13.9857,236.0276,7.6138
200,32x32x4,32,131072,23592.9600,3145.7280,17301.5040,2.0228,12.9595,254.7174,8.2167
200,32x32x8,32,262144,47185.9200,6291.4560,34603.0080,4.1124,12.7491,258.9207,8.3523
200,32x32x16,32,524288,94371.8400,12582.9120,69206.0160,8.5060,12.3275,267.7762,-26.8852
200,32x32x32,32,1048576,188743.6800,25165.8240,138412.0320,17.1890,12.2005,270.5622,-27.1649
200,32x32x64,32,2097152,377487.3600,50331.6480,276824.0640,33.5418,12.5047,263.9803,942.7868
200,32x64x2,32,131072,23592.9600,3145.7280,17301.5040,2.1203,12.3637,266.9921,8.6126
200,32x64x4,32,262144,47185.9200,6291.4560,34603.0080,4.1809,12.5402,263.2341,8.4914
200,32x64x8,32,524288,94371.8400,12582.9120,69206.0160,8.3133,12.6132,261.7105,-26.2762
200,32x64x16,32,1048576,188743.6800,25165.8240,138412.0320,16.6965,12.5605,262.8088,-26.3864
200,32x64x32,32,2097152,377487.3600,50331.6480,276824.0640,33.7448,12.4295,265.5785,948.4946
200,32x64x64,32,4194304,754974.7200,100663.2960,553648.1280,68.5976,12.2287,269.9385,964.0660
200,64x4x2,32,16384,2949.1200,393.2160,2162.6880,0.2194,14.9371,220.9935,7.1288
200,64x4x4,32,32768,5898.2400,786.4320,4325.3760,0.4381,14.9604,220.6498,7.1177
200,64x4x8,32,65536,11796.4800,1572.8640,8650.7520,0.9245,14.1782,232.8227,7.5104
200,64x4x16,32,131072,23592.9600,3145.7280,17301.5040,2.0184,12.9880,254.1572,8.1986
200,64x4x32,32,262144,47185.9200,6291.4560,34603.0080,4.1707,12.5708,262.5927,8.4707
200,64x4x64,32,524288,94371.8400,12582.9120,69206.0160,8.2933,12.6437,261.0782,-26.2127
200,64x8x2,32,32768,5898.2400,786.4320,4325.3760,0.4371,14.9927,220.1744,7.1024
200,64x8x4,32,65536,11796.4800,1572.8640,8650.7520,0.9394,13.9534,236.5739,7.6314
200,64x8x8,32,131072,23592.9600,3145.7280,17301.5040,2.0204,12.9749,254.4147,8.2069
200,64x8x16,32,262144,47185.9200,6291.4560,34603.0080,4.1678,12.5794,262.4123,8.4649
200,64x8x32,32,524288,94371.8400,12582.9120,69206.0160,8.3336,12.5825,262.3480,-26.3402
200,64x8x64,32,1048576,188743.6800,25165.8240,138412.0320,17.0020,12.3347,267.6182,-26.8693
200,64x16x2,32,65536,11796.4800,1572.8640,8650.7520,0.9336,14.0390,235.1313,7.5849
200,64x16x4,32,131072,23592.9600,3145.7280,17301.5040,2.0218,12.9657,254.5951,8.2127
200,64x16x8,32,262144,47185.9200,6291.4560,34603.0080,4.1207,12.7233,259.4454,8.3692
200,64x16x16,32,524288,94371.8400,12582.9120,69206.0160,8.3560,12.5488,263.0527,-26.4109
200,64x16x32,32,1048576,188743.6800,25165.8240,138412.0320,16.7676,12.5072,263.9284,-26.4988
200,64x16x64,32,2097152,377487.3600,50331.6480,276824.0640,33.4795,12.5280,263.4900,941.0357
200,64x32x2,32,131072,23592.9600,3145.7280,17301.5040,2.0220,12.9643,254.6225,8.2136
200,64x32x4,32,262144,47185.9200,6291.4560,34603.0080,4.0872,12.8275,257.3383,8.3012
200,64x32x8,32,524288,94371.8400,12582.9120,69206.0160,8.4447,12.4169,265.8471,-26.6915
200,64x32x16,32,1048576,188743.6800,25165.8240,138412.0320,17.0434,12.3048,268.2701,-26.9347
200,64x32x32,32,2097152,377487.3600,50331.6480,276824.0640,33.4924,12.5231,263.5920,941.3999
200,64x32x64,32,4194304,754974.7200,100663.2960,553648.1280,68.4806,12.2496,269.4779,962.4209
200,64x64x2,32,262144,47185.9200,6291.4560,34603.0080,4.1498,12.6341,261.2772,8.4283
200,64x64x4,32,524288,94371.8400,12582.9120,69206.0160,8.4875,12.3543,267.1943,-26.8267
200,64x64x8,32,1048576,188743.6800,25165.8240,138412.0320,16.7591,12.5135,263.7955,-26.4855
200,64x64x16,32,2097152,377487.3600,50331.6480,276824.0640,33.6675,12.4580,264.9700,946.3215
200,64x64x32,32,4194304,754974.7200,100663.2960,553648.1280,68.4488,12.2553,269.3528,961.9742
200,64x64x64,32,8388608,1509949.4400,201326.5920,1107296.2560,136.3839,12.3015,268.3420,958.3643

900
scripts/results_aos_skylakesp2.txt
View File

@ -1,900 +0,0 @@
200,4x4x2,8,256,21.5040,6.1440,9.2160,0.0017,30.3465,108.7769,15.5396
200,4x4x4,8,512,43.0080,12.2880,18.4320,0.0035,29.4117,112.2341,16.0334
200,4x4x8,8,1024,86.0160,24.5760,36.8640,0.0068,29.9117,110.3582,15.7655
200,4x4x16,8,2048,172.0320,49.1520,73.7280,0.0140,29.2747,112.7597,16.1085
200,4x4x32,8,4096,344.0640,98.3040,147.4560,0.0277,29.6271,111.4183,15.9169
200,4x4x64,8,8192,688.1280,196.6080,294.9120,0.0588,27.8858,118.3758,16.9108
200,4x8x2,8,512,43.0080,12.2880,18.4320,0.0035,29.2900,112.7006,16.1001
200,4x8x4,8,1024,86.0160,24.5760,36.8640,0.0071,28.8960,114.2372,16.3196
200,4x8x8,8,2048,172.0320,49.1520,73.7280,0.0138,29.7567,110.9330,15.8476
200,4x8x16,8,4096,344.0640,98.3040,147.4560,0.0283,28.9645,113.9670,16.2810
200,4x8x32,8,8192,688.1280,196.6080,294.9120,0.0589,27.8278,118.6226,16.9461
200,4x8x64,8,16384,1376.2560,393.2160,589.8240,0.1161,28.2126,117.0045,16.7149
200,4x16x2,8,1024,86.0160,24.5760,36.8640,0.0070,29.1692,113.1674,16.1668
200,4x16x4,8,2048,172.0320,49.1520,73.7280,0.0138,29.7129,111.0967,15.8710
200,4x16x8,8,4096,344.0640,98.3040,147.4560,0.0281,29.1939,113.0715,16.1531
200,4x16x16,8,8192,688.1280,196.6080,294.9120,0.0575,28.5109,115.7801,16.5400
200,4x16x32,8,16384,1376.2560,393.2160,589.8240,0.1183,27.7097,119.1278,17.0183
200,4x16x64,8,32768,2752.5120,786.4320,1179.6480,0.2349,27.9053,118.2930,16.8990
200,4x32x2,8,2048,172.0320,49.1520,73.7280,0.0143,28.6695,115.1397,16.4485
200,4x32x4,8,4096,344.0640,98.3040,147.4560,0.0284,28.8051,114.5978,16.3711
200,4x32x8,8,8192,688.1280,196.6080,294.9120,0.0586,27.9695,118.0216,16.8602
200,4x32x16,8,16384,1376.2560,393.2160,589.8240,0.1180,27.7722,118.8600,16.9800
200,4x32x32,8,32768,2752.5120,786.4320,1179.6480,0.2282,28.7202,114.9367,16.4195
200,4x32x64,8,65536,5505.0240,1572.8640,2359.2960,0.5880,22.2904,148.0906,21.1558
200,4x64x2,8,4096,344.0640,98.3040,147.4560,0.0281,29.1687,113.1693,16.1670
200,4x64x4,8,8192,688.1280,196.6080,294.9120,0.0587,27.9318,118.1807,16.8830
200,4x64x8,8,16384,1376.2560,393.2160,589.8240,0.1181,27.7405,118.9959,16.9994
200,4x64x16,8,32768,2752.5120,786.4320,1179.6480,0.2362,27.7451,118.9758,16.9965
200,4x64x32,8,65536,5505.0240,1572.8640,2359.2960,0.5785,22.6560,145.7012,20.8145
200,4x64x64,8,131072,11010.0480,3145.7280,4718.5920,1.4375,18.2361,181.0141,25.8592
200,8x4x2,8,512,43.0080,12.2880,18.4320,0.0035,29.3622,112.4235,16.0605
200,8x4x4,8,1024,86.0160,24.5760,36.8640,0.0068,29.9101,110.3639,15.7663
200,8x4x8,8,2048,172.0320,49.1520,73.7280,0.0141,29.0417,113.6642,16.2377
200,8x4x16,8,4096,344.0640,98.3040,147.4560,0.0281,29.1617,113.1963,16.1709
200,8x4x32,8,8192,688.1280,196.6080,294.9120,0.0585,28.0111,117.8460,16.8351
200,8x4x64,8,16384,1376.2560,393.2160,589.8240,0.1183,27.7088,119.1318,17.0188
200,8x8x2,8,1024,86.0160,24.5760,36.8640,0.0069,29.8459,110.6015,15.8002
200,8x8x4,8,2048,172.0320,49.1520,73.7280,0.0139,29.4463,112.1024,16.0146
200,8x8x8,8,4096,344.0640,98.3040,147.4560,0.0277,29.5525,111.6996,15.9571
200,8x8x16,8,8192,688.1280,196.6080,294.9120,0.0579,28.3017,116.6362,16.6623
200,8x8x32,8,16384,1376.2560,393.2160,589.8240,0.1145,28.6203,115.3378,16.4768
200,8x8x64,8,32768,2752.5120,786.4320,1179.6480,0.2280,28.7474,114.8278,16.4040
200,8x16x2,8,2048,172.0320,49.1520,73.7280,0.0140,29.3212,112.5807,16.0830
200,8x16x4,8,4096,344.0640,98.3040,147.4560,0.0277,29.5887,111.5627,15.9375
200,8x16x8,8,8192,688.1280,196.6080,294.9120,0.0572,28.6587,115.1831,16.4547
200,8x16x16,8,16384,1376.2560,393.2160,589.8240,0.1178,27.8274,118.6242,16.9463
200,8x16x32,8,32768,2752.5120,786.4320,1179.6480,0.2344,27.9549,118.0829,16.8690
200,8x16x64,8,65536,5505.0240,1572.8640,2359.2960,0.5755,22.7750,144.9394,20.7056
200,8x32x2,8,4096,344.0640,98.3040,147.4560,0.0283,28.9347,114.0844,16.2978
200,8x32x4,8,8192,688.1280,196.6080,294.9120,0.0586,27.9741,118.0018,16.8574
200,8x32x8,8,16384,1376.2560,393.2160,589.8240,0.1156,28.3512,116.4326,16.6332
200,8x32x16,8,32768,2752.5120,786.4320,1179.6480,0.2286,28.6691,115.1416,16.4488
200,8x32x32,8,65536,5505.0240,1572.8640,2359.2960,0.5765,22.7365,145.1848,20.7407
200,8x32x64,8,131072,11010.0480,3145.7280,4718.5920,1.4350,18.2676,180.7024,25.8146
200,8x64x2,8,8192,688.1280,196.6080,294.9120,0.0584,28.0706,117.5963,16.7995
200,8x64x4,8,16384,1376.2560,393.2160,589.8240,0.1148,28.5532,115.6087,16.5155
200,8x64x8,8,32768,2752.5120,786.4320,1179.6480,0.2361,27.7596,118.9138,16.9877
200,8x64x16,8,65536,5505.0240,1572.8640,2359.2960,0.5871,22.3246,147.8639,21.1234
200,8x64x32,8,131072,11010.0480,3145.7280,4718.5920,1.4440,18.1544,181.8290,25.9756
200,8x64x64,8,262144,22020.0960,6291.4560,9437.1840,3.0284,17.3122,190.6752,27.2393
200,16x4x2,8,1024,86.0160,24.5760,36.8640,0.0070,29.1585,113.2087,16.1727
200,16x4x4,8,2048,172.0320,49.1520,73.7280,0.0138,29.6911,111.1780,15.8826
200,16x4x8,8,4096,344.0640,98.3040,147.4560,0.0283,28.9797,113.9074,16.2725
200,16x4x16,8,8192,688.1280,196.6080,294.9120,0.0587,27.9252,118.2088,16.8870
200,16x4x32,8,16384,1376.2560,393.2160,589.8240,0.1183,27.7026,119.1586,17.0227
200,16x4x64,8,32768,2752.5120,786.4320,1179.6480,0.2341,27.9895,117.9373,16.8482
200,16x8x2,8,2048,172.0320,49.1520,73.7280,0.0142,28.7942,114.6411,16.3773
200,16x8x4,8,4096,344.0640,98.3040,147.4560,0.0283,28.9327,114.0924,16.2989
200,16x8x8,8,8192,688.1280,196.6080,294.9120,0.0570,28.7438,114.8422,16.4060
200,16x8x16,8,16384,1376.2560,393.2160,589.8240,0.1146,28.5861,115.4759,16.4966
200,16x8x32,8,32768,2752.5120,786.4320,1179.6480,0.2326,28.1699,117.1817,16.7402
200,16x8x64,8,65536,5505.0240,1572.8640,2359.2960,0.5820,22.5202,146.5794,20.9399
200,16x16x2,8,4096,344.0640,98.3040,147.4560,0.0286,28.6319,115.2912,16.4702
200,16x16x4,8,8192,688.1280,196.6080,294.9120,0.0577,28.4037,116.2174,16.6025
200,16x16x8,8,16384,1376.2560,393.2160,589.8240,0.1157,28.3232,116.5474,16.6496
200,16x16x16,8,32768,2752.5120,786.4320,1179.6480,0.2272,28.8476,114.4289,16.3470
200,16x16x32,8,65536,5505.0240,1572.8640,2359.2960,0.5853,22.3948,147.4001,21.0572
200,16x16x64,8,131072,11010.0480,3145.7280,4718.5920,1.4359,18.2569,180.8082,25.8297
200,16x32x2,8,8192,688.1280,196.6080,294.9120,0.0592,27.6563,119.3580,17.0511
200,16x32x4,8,16384,1376.2560,393.2160,589.8240,0.1175,27.8953,118.3353,16.9050
200,16x32x8,8,32768,2752.5120,786.4320,1179.6480,0.2339,28.0208,117.8051,16.8293
200,16x32x16,8,65536,5505.0240,1572.8640,2359.2960,0.5877,22.3036,148.0030,21.1433
200,16x32x32,8,131072,11010.0480,3145.7280,4718.5920,1.4163,18.5085,178.3503,25.4786
200,16x32x64,8,262144,22020.0960,6291.4560,9437.1840,3.0250,17.3321,190.4557,27.2080
200,16x64x2,8,16384,1376.2560,393.2160,589.8240,0.1184,27.6850,119.2343,17.0335
200,16x64x4,8,32768,2752.5120,786.4320,1179.6480,0.2318,28.2692,116.7704,16.6815
200,16x64x8,8,65536,5505.0240,1572.8640,2359.2960,0.5828,22.4888,146.7843,20.9692
200,16x64x16,8,131072,11010.0480,3145.7280,4718.5920,1.4080,18.6177,177.3041,25.3292
200,16x64x32,8,262144,22020.0960,6291.4560,9437.1840,3.0668,17.0956,193.0906,27.5844
200,16x64x64,8,524288,44040.1920,12582.9120,18874.3680,6.2702,16.7233,197.3894,28.1985
200,32x4x2,8,2048,172.0320,49.1520,73.7280,0.0138,29.7469,110.9697,15.8528
200,32x4x4,8,4096,344.0640,98.3040,147.4560,0.0280,29.2260,112.9475,16.1354
200,32x4x8,8,8192,688.1280,196.6080,294.9120,0.0592,27.6594,119.3447,17.0492
200,32x4x16,8,16384,1376.2560,393.2160,589.8240,0.1176,27.8590,118.4893,16.9270
200,32x4x32,8,32768,2752.5120,786.4320,1179.6480,0.2335,28.0716,117.5922,16.7989
200,32x4x64,8,65536,5505.0240,1572.8640,2359.2960,0.5712,22.9486,143.8433,20.5490
200,32x8x2,8,4096,344.0640,98.3040,147.4560,0.0278,29.4205,112.2006,16.0287
200,32x8x4,8,8192,688.1280,196.6080,294.9120,0.0582,28.1302,117.3473,16.7639
200,32x8x8,8,16384,1376.2560,393.2160,589.8240,0.1179,27.8007,118.7378,16.9625
200,32x8x16,8,32768,2752.5120,786.4320,1179.6480,0.2297,28.5347,115.6839,16.5263
200,32x8x32,8,65536,5505.0240,1572.8640,2359.2960,0.5933,22.0925,149.4173,21.3453
200,32x8x64,8,131072,11010.0480,3145.7280,4718.5920,1.4077,18.6223,177.2608,25.3230
200,32x16x2,8,8192,688.1280,196.6080,294.9120,0.0584,28.0354,117.7439,16.8206
200,32x16x4,8,16384,1376.2560,393.2160,589.8240,0.1151,28.4770,115.9181,16.5597
200,32x16x8,8,32768,2752.5120,786.4320,1179.6480,0.2362,27.7498,118.9559,16.9937
200,32x16x16,8,65536,5505.0240,1572.8640,2359.2960,0.5820,22.5213,146.5721,20.9389
200,32x16x32,8,131072,11010.0480,3145.7280,4718.5920,1.4174,18.4944,178.4868,25.4981
200,32x16x64,8,262144,22020.0960,6291.4560,9437.1840,3.0733,17.0595,193.4992,27.6427
200,32x32x2,8,16384,1376.2560,393.2160,589.8240,0.1152,28.4535,116.0140,16.5734
200,32x32x4,8,32768,2752.5120,786.4320,1179.6480,0.2266,28.9244,114.1250,16.3036
200,32x32x8,8,65536,5505.0240,1572.8640,2359.2960,0.5804,22.5825,146.1755,20.8822
200,32x32x16,8,131072,11010.0480,3145.7280,4718.5920,1.4367,18.2460,180.9162,25.8452
200,32x32x32,8,262144,22020.0960,6291.4560,9437.1840,3.0375,17.2607,191.2436,27.3205
200,32x32x64,8,524288,44040.1920,12582.9120,18874.3680,6.2335,16.8215,196.2366,28.0338
200,32x64x2,8,32768,2752.5120,786.4320,1179.6480,0.2302,28.4738,115.9312,16.5616
200,32x64x4,8,65536,5505.0240,1572.8640,2359.2960,0.5861,22.3652,147.5953,21.0850
200,32x64x8,8,131072,11010.0480,3145.7280,4718.5920,1.4180,18.4875,178.5533,25.5076
200,32x64x16,8,262144,22020.0960,6291.4560,9437.1840,3.0346,17.2770,191.0631,27.2947
200,32x64x32,8,524288,44040.1920,12582.9120,18874.3680,6.3025,16.6376,198.4062,28.3437
200,32x64x64,8,1048576,88080.3840,25165.8240,37748.7360,12.7814,16.4079,201.1838,28.7405
200,64x4x2,8,4096,344.0640,98.3040,147.4560,0.0283,28.9709,113.9419,16.2774
200,64x4x4,8,8192,688.1280,196.6080,294.9120,0.0582,28.1578,117.2320,16.7474
200,64x4x8,8,16384,1376.2560,393.2160,589.8240,0.1183,27.7027,119.1582,17.0226
200,64x4x16,8,32768,2752.5120,786.4320,1179.6480,0.2278,28.7749,114.7179,16.3883
200,64x4x32,8,65536,5505.0240,1572.8640,2359.2960,0.5870,22.3280,147.8415,21.1202
200,64x4x64,8,131072,11010.0480,3145.7280,4718.5920,1.4253,18.3927,179.4738,25.6391
200,64x8x2,8,8192,688.1280,196.6080,294.9120,0.0583,28.1031,117.4602,16.7800
200,64x8x4,8,16384,1376.2560,393.2160,589.8240,0.1160,28.2362,116.9066,16.7009
200,64x8x8,8,32768,2752.5120,786.4320,1179.6480,0.2346,27.9345,118.1694,16.8813
200,64x8x16,8,65536,5505.0240,1572.8640,2359.2960,0.5759,22.7579,145.0484,20.7212
200,64x8x32,8,131072,11010.0480,3145.7280,4718.5920,1.4412,18.1889,181.4842,25.9263
200,64x8x64,8,262144,22020.0960,6291.4560,9437.1840,3.0304,17.3007,190.8014,27.2573
200,64x16x2,8,16384,1376.2560,393.2160,589.8240,0.1152,28.4352,116.0884,16.5841
200,64x16x4,8,32768,2752.5120,786.4320,1179.6480,0.2329,28.1400,117.3061,16.7580
200,64x16x8,8,65536,5505.0240,1572.8640,2359.2960,0.5882,22.2847,148.1284,21.1612
200,64x16x16,8,131072,11010.0480,3145.7280,4718.5920,1.4200,18.4609,178.8099,25.5443
200,64x16x32,8,262144,22020.0960,6291.4560,9437.1840,3.0645,17.1084,192.9467,27.5638
200,64x16x64,8,524288,44040.1920,12582.9120,18874.3680,6.2823,16.6910,197.7714,28.2531
200,64x32x2,8,32768,2752.5120,786.4320,1179.6480,0.2325,28.1827,117.1285,16.7326
200,64x32x4,8,65536,5505.0240,1572.8640,2359.2960,0.5814,22.5459,146.4124,20.9161
200,64x32x8,8,131072,11010.0480,3145.7280,4718.5920,1.4327,18.2977,180.4052,25.7722
200,64x32x16,8,262144,22020.0960,6291.4560,9437.1840,3.0435,17.2262,191.6265,27.3752
200,64x32x32,8,524288,44040.1920,12582.9120,18874.3680,6.2562,16.7606,196.9494,28.1356
200,64x32x64,8,1048576,88080.3840,25165.8240,37748.7360,12.8218,16.3562,201.8198,28.8314
200,64x64x2,8,65536,5505.0240,1572.8640,2359.2960,0.5803,22.5887,146.1352,20.8765
200,64x64x4,8,131072,11010.0480,3145.7280,4718.5920,1.4161,18.5111,178.3255,25.4751
200,64x64x8,8,262144,22020.0960,6291.4560,9437.1840,3.0288,17.3098,190.7007,27.2430
200,64x64x16,8,524288,44040.1920,12582.9120,18874.3680,6.2470,16.7853,196.6601,28.0943
200,64x64x32,8,1048576,88080.3840,25165.8240,37748.7360,12.7711,16.4211,201.0219,28.7174
200,64x64x64,8,2097152,176160.7680,50331.6480,75497.4720,25.5759,16.3994,201.2874,-62.1257
200,4x4x2,16,512,59.3920,12.2880,34.8160,0.0056,18.4446,178.9684,11.9312
200,4x4x4,16,1024,118.7840,24.5760,69.6320,0.0112,18.2150,181.2240,12.0816
200,4x4x8,16,2048,237.5680,49.1520,139.2640,0.0224,18.3113,180.2714,12.0181
200,4x4x16,16,4096,475.1360,98.3040,278.5280,0.0447,18.3455,179.9354,11.9957
200,4x4x32,16,8192,950.2720,196.6080,557.0560,0.0905,18.1055,182.3198,12.1547
200,4x4x64,16,16384,1900.5440,393.2160,1114.1120,0.1819,18.0138,183.2489,12.2166
200,4x8x2,16,1024,118.7840,24.5760,69.6320,0.0111,18.4483,178.9328,11.9289
200,4x8x4,16,2048,237.5680,49.1520,139.2640,0.0222,18.4121,179.2842,11.9523
200,4x8x8,16,4096,475.1360,98.3040,278.5280,0.0452,18.1169,182.2060,12.1471
200,4x8x16,16,8192,950.2720,196.6080,557.0560,0.0921,17.7870,185.5850,12.3723
200,4x8x32,16,16384,1900.5440,393.2160,1114.1120,0.1818,18.0274,183.1101,12.2073
200,4x8x64,16,32768,3801.0880,786.4320,2228.2240,0.3687,17.7742,185.7186,12.3812
200,4x16x2,16,2048,237.5680,49.1520,139.2640,0.0224,18.3173,180.2120,12.0141
200,4x16x4,16,4096,475.1360,98.3040,278.5280,0.0459,17.8317,185.1201,12.3413
200,4x16x8,16,8192,950.2720,196.6080,557.0560,0.0907,18.0668,182.7105,12.1807
200,4x16x16,16,16384,1900.5440,393.2160,1114.1120,0.1836,17.8502,184.9274,12.3285
200,4x16x32,16,32768,3801.0880,786.4320,2228.2240,0.3632,18.0417,182.9653,12.1977
200,4x16x64,16,65536,7602.1760,1572.8640,4456.4480,0.8405,15.5939,211.6859,14.1124
200,4x32x2,16,4096,475.1360,98.3040,278.5280,0.0453,18.1000,182.3761,12.1584
200,4x32x4,16,8192,950.2720,196.6080,557.0560,0.0906,18.0813,182.5645,12.1710
200,4x32x8,16,16384,1900.5440,393.2160,1114.1120,0.1839,17.8212,185.2292,12.3486
200,4x32x16,16,32768,3801.0880,786.4320,2228.2240,0.3639,18.0100,183.2872,12.2191
200,4x32x32,16,65536,7602.1760,1572.8640,4456.4480,0.8393,15.6170,211.3718,14.0915
200,4x32x64,16,131072,15204.3520,3145.7280,8912.8960,1.7298,15.1546,217.8222,14.5215
200,4x64x2,16,8192,950.2720,196.6080,557.0560,0.0914,17.9284,184.1209,12.2747
200,4x64x4,16,16384,1900.5440,393.2160,1114.1120,0.1847,17.7401,186.0756,12.4050
200,4x64x8,16,32768,3801.0880,786.4320,2228.2240,0.3655,17.9308,184.0962,12.2731
200,4x64x16,16,65536,7602.1760,1572.8640,4456.4480,0.8365,15.6689,210.6717,14.0448
200,4x64x32,16,131072,15204.3520,3145.7280,8912.8960,1.7282,15.1685,217.6226,14.5082
200,4x64x64,16,262144,30408.7040,6291.4560,17825.7920,3.4849,15.0444,219.4168,14.6278
200,8x4x2,16,1024,118.7840,24.5760,69.6320,0.0111,18.4278,179.1319,11.9421
200,8x4x4,16,2048,237.5680,49.1520,139.2640,0.0226,18.1638,181.7354,12.1157
200,8x4x8,16,4096,475.1360,98.3040,278.5280,0.0453,18.0726,182.6524,12.1768
200,8x4x16,16,8192,950.2720,196.6080,557.0560,0.0920,17.8152,185.2914,12.3528
200,8x4x32,16,16384,1900.5440,393.2160,1114.1120,0.1823,17.9748,183.6461,12.2431
200,8x4x64,16,32768,3801.0880,786.4320,2228.2240,0.3658,17.9157,184.2522,12.2835
200,8x8x2,16,2048,237.5680,49.1520,139.2640,0.0222,18.4192,179.2152,11.9477
200,8x8x4,16,4096,475.1360,98.3040,278.5280,0.0444,18.4360,179.0522,11.9368
200,8x8x8,16,8192,950.2720,196.6080,557.0560,0.0916,17.8863,184.5548,12.3037
200,8x8x16,16,16384,1900.5440,393.2160,1114.1120,0.1839,17.8220,185.2203,12.3480
200,8x8x32,16,32768,3801.0880,786.4320,2228.2240,0.3664,17.8863,184.5542,12.3036
200,8x8x64,16,65536,7602.1760,1572.8640,4456.4480,0.8380,15.6408,211.0501,14.0700
200,8x16x2,16,4096,475.1360,98.3040,278.5280,0.0453,18.0660,182.7192,12.1813
200,8x16x4,16,8192,950.2720,196.6080,557.0560,0.0919,17.8192,185.2493,12.3500
200,8x16x8,16,16384,1900.5440,393.2160,1114.1120,0.1840,17.8059,185.3877,12.3592
200,8x16x16,16,32768,3801.0880,786.4320,2228.2240,0.3670,17.8573,184.8539,12.3236
200,8x16x32,16,65536,7602.1760,1572.8640,4456.4480,0.8359,15.6812,210.5074,14.0338
200,8x16x64,16,131072,15204.3520,3145.7280,8912.8960,1.7319,15.1360,218.0891,14.5393
200,8x32x2,16,8192,950.2720,196.6080,557.0560,0.0904,18.1311,182.0625,12.1375
200,8x32x4,16,16384,1900.5440,393.2160,1114.1120,0.1824,17.9638,183.7583,12.2506
200,8x32x8,16,32768,3801.0880,786.4320,2228.2240,0.3648,17.9625,183.7718,12.2515
200,8x32x16,16,65536,7602.1760,1572.8640,4456.4480,0.8361,15.6759,210.5784,14.0386
200,8x32x32,16,131072,15204.3520,3145.7280,8912.8960,1.7322,15.1336,218.1245,14.5416
200,8x32x64,16,262144,30408.7040,6291.4560,17825.7920,3.4793,15.0688,219.0625,14.6042
200,8x64x2,16,16384,1900.5440,393.2160,1114.1120,0.1843,17.7795,185.6637,12.3776
200,8x64x4,16,32768,3801.0880,786.4320,2228.2240,0.3625,18.0764,182.6137,12.1742
200,8x64x8,16,65536,7602.1760,1572.8640,4456.4480,0.8262,15.8645,208.0749,13.8717
200,8x64x16,16,131072,15204.3520,3145.7280,8912.8960,1.7275,15.1748,217.5316,14.5021
200,8x64x32,16,262144,30408.7040,6291.4560,17825.7920,3.4846,15.0457,219.3977,14.6265
200,8x64x64,16,524288,60817.4080,12582.9120,35651.5840,7.0206,14.9356,221.0153,14.7344
200,16x4x2,16,2048,237.5680,49.1520,139.2640,0.0226,18.1117,182.2579,12.1505
200,16x4x4,16,4096,475.1360,98.3040,278.5280,0.0449,18.2408,180.9682,12.0645
200,16x4x8,16,8192,950.2720,196.6080,557.0560,0.0919,17.8350,185.0856,12.3390
200,16x4x16,16,16384,1900.5440,393.2160,1114.1120,0.1816,18.0462,182.9196,12.1946
200,16x4x32,16,32768,3801.0880,786.4320,2228.2240,0.3678,17.8165,185.2778,12.3519
200,16x4x64,16,65536,7602.1760,1572.8640,4456.4480,0.8404,15.5960,211.6567,14.1104
200,16x8x2,16,4096,475.1360,98.3040,278.5280,0.0450,18.2228,181.1466,12.0764
200,16x8x4,16,8192,950.2720,196.6080,557.0560,0.0908,18.0431,182.9508,12.1967
200,16x8x8,16,16384,1900.5440,393.2160,1114.1120,0.1820,18.0031,183.3572,12.2238
200,16x8x16,16,32768,3801.0880,786.4320,2228.2240,0.3693,17.7458,186.0162,12.4011
200,16x8x32,16,65536,7602.1760,1572.8640,4456.4480,0.8347,15.7034,210.2096,14.0140
200,16x8x64,16,131072,15204.3520,3145.7280,8912.8960,1.7291,15.1608,217.7328,14.5155
200,16x16x2,16,8192,950.2720,196.6080,557.0560,0.0909,18.0184,183.2018,12.2135
200,16x16x4,16,16384,1900.5440,393.2160,1114.1120,0.1824,17.9630,183.7665,12.2511
200,16x16x8,16,32768,3801.0880,786.4320,2228.2240,0.3699,17.7162,186.3271,12.4218
200,16x16x16,16,65536,7602.1760,1572.8640,4456.4480,0.8338,15.7207,209.9783,13.9986
200,16x16x32,16,131072,15204.3520,3145.7280,8912.8960,1.7287,15.1644,217.6807,14.5120
200,16x16x64,16,262144,30408.7040,6291.4560,17825.7920,3.4878,15.0321,219.5966,14.6398
200,16x32x2,16,16384,1900.5440,393.2160,1114.1120,0.1840,17.8096,185.3495,12.3566
200,16x32x4,16,32768,3801.0880,786.4320,2228.2240,0.3640,18.0034,183.3538,12.2236
200,16x32x8,16,65536,7602.1760,1572.8640,4456.4480,0.8352,15.6942,210.3329,14.0222
200,16x32x16,16,131072,15204.3520,3145.7280,8912.8960,1.7275,15.1744,217.5369,14.5025
200,16x32x32,16,262144,30408.7040,6291.4560,17825.7920,3.4834,15.0510,219.3216,14.6214
200,16x32x64,16,524288,60817.4080,12582.9120,35651.5840,7.0276,14.9209,221.2334,14.7489
200,16x64x2,16,32768,3801.0880,786.4320,2228.2240,0.3636,18.0229,183.1558,12.2104
200,16x64x4,16,65536,7602.1760,1572.8640,4456.4480,0.8278,15.8337,208.4795,13.8986
200,16x64x8,16,131072,15204.3520,3145.7280,8912.8960,1.7258,15.1897,217.3186,14.4879
200,16x64x16,16,262144,30408.7040,6291.4560,17825.7920,3.4981,14.9878,220.2464,14.6831
200,16x64x32,16,524288,60817.4080,12582.9120,35651.5840,7.0395,14.8955,221.6102,14.7740
200,16x64x64,16,1048576,121634.8160,25165.8240,71303.1680,14.1277,14.8442,222.3762,-40.5796
200,32x4x2,16,4096,475.1360,98.3040,278.5280,0.0454,18.0286,183.0981,12.2065
200,32x4x4,16,8192,950.2720,196.6080,557.0560,0.0921,17.7984,185.4660,12.3644
200,32x4x8,16,16384,1900.5440,393.2160,1114.1120,0.1820,18.0071,183.3167,12.2211
200,32x4x16,16,32768,3801.0880,786.4320,2228.2240,0.3688,17.7722,185.7397,12.3826
200,32x4x32,16,65536,7602.1760,1572.8640,4456.4480,0.8394,15.6154,211.3937,14.0929
200,32x4x64,16,131072,15204.3520,3145.7280,8912.8960,1.7272,15.1777,217.4896,14.4993
200,32x8x2,16,8192,950.2720,196.6080,557.0560,0.0922,17.7751,185.7089,12.3806
200,32x8x4,16,16384,1900.5440,393.2160,1114.1120,0.1847,17.7401,186.0755,12.4050
200,32x8x8,16,32768,3801.0880,786.4320,2228.2240,0.3650,17.9548,183.8507,12.2567
200,32x8x16,16,65536,7602.1760,1572.8640,4456.4480,0.8342,15.7128,210.0837,14.0056
200,32x8x32,16,131072,15204.3520,3145.7280,8912.8960,1.7343,15.1151,218.3913,14.5594
200,32x8x64,16,262144,30408.7040,6291.4560,17825.7920,3.4881,15.0309,219.6142,14.6409
200,32x16x2,16,16384,1900.5440,393.2160,1114.1120,0.1842,17.7899,185.5548,12.3703
200,32x16x4,16,32768,3801.0880,786.4320,2228.2240,0.3635,18.0279,183.1051,12.2070
200,32x16x8,16,65536,7602.1760,1572.8640,4456.4480,0.8402,15.5994,211.6101,14.1073
200,32x16x16,16,131072,15204.3520,3145.7280,8912.8960,1.7321,15.1345,218.1105,14.5407
200,32x16x32,16,262144,30408.7040,6291.4560,17825.7920,3.4956,14.9983,220.0914,14.6728
200,32x16x64,16,524288,60817.4080,12582.9120,35651.5840,7.0035,14.9722,220.4748,14.6983
200,32x32x2,16,32768,3801.0880,786.4320,2228.2240,0.3676,17.8286,185.1519,12.3435
200,32x32x4,16,65536,7602.1760,1572.8640,4456.4480,0.8358,15.6819,210.4980,14.0332
200,32x32x8,16,131072,15204.3520,3145.7280,8912.8960,1.7292,15.1599,217.7456,14.5164
200,32x32x16,16,262144,30408.7040,6291.4560,17825.7920,3.4836,15.0501,219.3342,14.6223
200,32x32x32,16,524288,60817.4080,12582.9120,35651.5840,7.0147,14.9483,220.8277,14.7218
200,32x32x64,16,1048576,121634.8160,25165.8240,71303.1680,14.1095,14.8634,222.0885,-40.5271
200,32x64x2,16,65536,7602.1760,1572.8640,4456.4480,0.8323,15.7487,209.6045,13.9736
200,32x64x4,16,131072,15204.3520,3145.7280,8912.8960,1.7337,15.1204,218.3147,14.5543
200,32x64x8,16,262144,30408.7040,6291.4560,17825.7920,3.4854,15.0424,219.4469,14.6298
200,32x64x16,16,524288,60817.4080,12582.9120,35651.5840,6.9997,14.9803,220.3564,14.6904
200,32x64x32,16,1048576,121634.8160,25165.8240,71303.1680,14.1000,14.8734,221.9396,-40.4999
200,32x64x64,16,2097152,243269.6320,50331.6480,142606.3360,28.5006,14.7165,224.3054,47.1230
200,64x4x2,16,8192,950.2720,196.6080,557.0560,0.0928,17.6461,187.0664,12.4711
200,64x4x4,16,16384,1900.5440,393.2160,1114.1120,0.1819,18.0106,183.2808,12.2187
200,64x4x8,16,32768,3801.0880,786.4320,2228.2240,0.3692,17.7502,185.9693,12.3980
200,64x4x16,16,65536,7602.1760,1572.8640,4456.4480,0.8337,15.7213,209.9699,13.9980
200,64x4x32,16,131072,15204.3520,3145.7280,8912.8960,1.7324,15.1317,218.1512,14.5434
200,64x4x64,16,262144,30408.7040,6291.4560,17825.7920,3.4837,15.0497,219.3394,14.6226
200,64x8x2,16,16384,1900.5440,393.2160,1114.1120,0.1839,17.8183,185.2592,12.3506
200,64x8x4,16,32768,3801.0880,786.4320,2228.2240,0.3694,17.7413,186.0627,12.4042
200,64x8x8,16,65536,7602.1760,1572.8640,4456.4480,0.8317,15.7603,209.4500,13.9633
200,64x8x16,16,131072,15204.3520,3145.7280,8912.8960,1.7289,15.1629,217.7025,14.5135
200,64x8x32,16,262144,30408.7040,6291.4560,17825.7920,3.4803,15.0645,219.1241,14.6083
200,64x8x64,16,524288,60817.4080,12582.9120,35651.5840,7.0112,14.9557,220.7180,14.7145
200,64x16x2,16,32768,3801.0880,786.4320,2228.2240,0.3702,17.7021,186.4753,12.4317
200,64x16x4,16,65536,7602.1760,1572.8640,4456.4480,0.8310,15.7727,209.2857,13.9524
200,64x16x8,16,131072,15204.3520,3145.7280,8912.8960,1.7311,15.1428,217.9918,14.5328
200,64x16x16,16,262144,30408.7040,6291.4560,17825.7920,3.4783,15.0730,219.0016,14.6001
200,64x16x32,16,524288,60817.4080,12582.9120,35651.5840,7.1692,14.6261,225.6924,15.0462
200,64x16x64,16,1048576,121634.8160,25165.8240,71303.1680,14.0752,14.8996,221.5499,-40.4288
200,64x32x2,16,65536,7602.1760,1572.8640,4456.4480,0.8301,15.7907,209.0467,13.9364
200,64x32x4,16,131072,15204.3520,3145.7280,8912.8960,1.7349,15.1099,218.4658,14.5644
200,64x32x8,16,262144,30408.7040,6291.4560,17825.7920,3.4954,14.9994,220.0760,14.6717
200,64x32x16,16,524288,60817.4080,12582.9120,35651.5840,7.0085,14.9614,220.6340,14.7089
200,64x32x32,16,1048576,121634.8160,25165.8240,71303.1680,14.0931,14.8807,221.8305,-40.4800
200,64x32x64,16,2097152,243269.6320,50331.6480,142606.3360,28.3617,14.7886,223.2118,46.8932
200,64x64x2,16,131072,15204.3520,3145.7280,8912.8960,1.7273,15.1768,217.5037,14.5002
200,64x64x4,16,262144,30408.7040,6291.4560,17825.7920,3.4938,15.0061,219.9772,14.6651
200,64x64x8,16,524288,60817.4080,12582.9120,35651.5840,7.0357,14.9037,221.4885,14.7659
200,64x64x16,16,1048576,121634.8160,25165.8240,71303.1680,14.1410,14.8303,222.5848,-40.6177
200,64x64x32,16,2097152,243269.6320,50331.6480,142606.3360,28.4706,14.7320,224.0696,47.0734
200,64x64x64,16,4194304,486539.2640,100663.2960,285212.6720,56.5696,14.8288,222.6070,-618.3528
200,4x4x2,32,1024,184.3200,24.5760,135.1680,0.0184,11.1141,297.0107,9.5810
200,4x4x4,32,2048,368.6400,49.1520,270.3360,0.0363,11.2779,292.6971,9.4418
200,4x4x8,32,4096,737.2800,98.3040,540.6720,0.0742,11.0449,298.8711,9.6410
200,4x4x16,32,8192,1474.5600,196.6080,1081.3440,0.1481,11.0605,298.4503,9.6274
200,4x4x32,32,16384,2949.1200,393.2160,2162.6880,0.2949,11.1106,297.1028,9.5840
200,4x4x64,32,32768,5898.2400,786.4320,4325.3760,0.5983,10.9532,301.3734,9.7217
200,4x8x2,32,2048,368.6400,49.1520,270.3360,0.0368,11.1204,296.8413,9.5755
200,4x8x4,32,4096,737.2800,98.3040,540.6720,0.0742,11.0352,299.1326,9.6494
200,4x8x8,32,8192,1474.5600,196.6080,1081.3440,0.1483,11.0465,298.8270,9.6396
200,4x8x16,32,16384,2949.1200,393.2160,2162.6880,0.2961,11.0655,298.3154,9.6231
200,4x8x32,32,32768,5898.2400,786.4320,4325.3760,0.5986,10.9475,301.5306,9.7268
200,4x8x64,32,65536,11796.4800,1572.8640,8650.7520,1.2720,10.3045,320.3467,10.3338
200,4x16x2,32,4096,737.2800,98.3040,540.6720,0.0743,11.0186,299.5856,9.6641
200,4x16x4,32,8192,1474.5600,196.6080,1081.3440,0.1484,11.0422,298.9442,9.6434
200,4x16x8,32,16384,2949.1200,393.2160,2162.6880,0.2966,11.0488,298.7648,9.6376
200,4x16x16,32,32768,5898.2400,786.4320,4325.3760,0.5989,10.9436,301.6379,9.7303
200,4x16x32,32,65536,11796.4800,1572.8640,8650.7520,1.2770,10.2640,321.6110,10.3745
200,4x16x64,32,131072,23592.9600,3145.7280,17301.5040,2.6173,10.0157,329.5828,10.6317
200,4x32x2,32,8192,1474.5600,196.6080,1081.3440,0.1488,11.0087,299.8546,9.6727
200,4x32x4,32,16384,2949.1200,393.2160,2162.6880,0.2971,11.0310,299.2466,9.6531
200,4x32x8,32,32768,5898.2400,786.4320,4325.3760,0.5925,11.0611,298.4334,9.6269
200,4x32x16,32,65536,11796.4800,1572.8640,8650.7520,1.2776,10.2593,321.7576,10.3793
200,4x32x32,32,131072,23592.9600,3145.7280,17301.5040,2.6138,10.0294,329.1324,10.6172
200,4x32x64,32,262144,47185.9200,6291.4560,34603.0080,5.2766,9.9362,332.2209,10.7168
200,4x64x2,32,16384,2949.1200,393.2160,2162.6880,0.2965,11.0517,298.6869,9.6351
200,4x64x4,32,32768,5898.2400,786.4320,4325.3760,0.5938,11.0371,299.0820,9.6478
200,4x64x8,32,65536,11796.4800,1572.8640,8650.7520,1.2778,10.2578,321.8033,10.3808
200,4x64x16,32,131072,23592.9600,3145.7280,17301.5040,2.6155,10.0229,329.3472,10.6241
200,4x64x32,32,262144,47185.9200,6291.4560,34603.0080,5.2779,9.9336,332.3065,10.7196
200,4x64x64,32,524288,94371.8400,12582.9120,69206.0160,10.6334,9.8612,334.7478,-33.6092
200,8x4x2,32,2048,368.6400,49.1520,270.3360,0.0367,11.1554,295.9115,9.5455
200,8x4x4,32,4096,737.2800,98.3040,540.6720,0.0740,11.0655,298.3153,9.6231
200,8x4x8,32,8192,1474.5600,196.6080,1081.3440,0.1486,11.0227,299.4742,9.6605
200,8x4x16,32,16384,2949.1200,393.2160,2162.6880,0.2971,11.0293,299.2935,9.6546
200,8x4x32,32,32768,5898.2400,786.4320,4325.3760,0.6002,10.9192,302.3105,9.7520
200,8x4x64,32,65536,11796.4800,1572.8640,8650.7520,1.2766,10.2675,321.4994,10.3709
200,8x8x2,32,4096,737.2800,98.3040,540.6720,0.0741,11.0621,298.4065,9.6260
200,8x8x4,32,8192,1474.5600,196.6080,1081.3440,0.1473,11.1195,296.8666,9.5763
200,8x8x8,32,16384,2949.1200,393.2160,2162.6880,0.2977,11.0065,299.9133,9.6746
200,8x8x16,32,32768,5898.2400,786.4320,4325.3760,0.5911,11.0872,297.7294,9.6042
200,8x8x32,32,65536,11796.4800,1572.8640,8650.7520,1.2751,10.2790,321.1408,10.3594
200,8x8x64,32,131072,23592.9600,3145.7280,17301.5040,2.6235,9.9923,330.3550,10.6566
200,8x16x2,32,8192,1474.5600,196.6080,1081.3440,0.1479,11.0789,297.9548,9.6114
200,8x16x4,32,16384,2949.1200,393.2160,2162.6880,0.2968,11.0408,298.9830,9.6446
200,8x16x8,32,32768,5898.2400,786.4320,4325.3760,0.5965,10.9874,300.4339,9.6914
200,8x16x16,32,65536,11796.4800,1572.8640,8650.7520,1.2750,10.2803,321.0997,10.3581
200,8x16x32,32,131072,23592.9600,3145.7280,17301.5040,2.6219,9.9984,330.1535,10.6501
200,8x16x64,32,262144,47185.9200,6291.4560,34603.0080,5.2861,9.9183,332.8200,10.7361
200,8x32x2,32,16384,2949.1200,393.2160,2162.6880,0.2988,10.9680,300.9674,9.7086
200,8x32x4,32,32768,5898.2400,786.4320,4325.3760,0.6009,10.9058,302.6840,9.7640
200,8x32x8,32,65536,11796.4800,1572.8640,8650.7520,1.2771,10.2631,321.6368,10.3754
200,8x32x16,32,131072,23592.9600,3145.7280,17301.5040,2.6137,10.0297,329.1236,10.6169
200,8x32x32,32,262144,47185.9200,6291.4560,34603.0080,5.2858,9.9188,332.8016,10.7355
200,8x32x64,32,524288,94371.8400,12582.9120,69206.0160,10.6487,9.8470,335.2298,-33.6576
200,8x64x2,32,32768,5898.2400,786.4320,4325.3760,0.5968,10.9820,300.5841,9.6963
200,8x64x4,32,65536,11796.4800,1572.8640,8650.7520,1.2759,10.2733,321.3197,10.3652
200,8x64x8,32,131072,23592.9600,3145.7280,17301.5040,2.6176,10.0147,329.6169,10.6328
200,8x64x16,32,262144,47185.9200,6291.4560,34603.0080,5.2745,9.9401,332.0889,10.7125
200,8x64x32,32,524288,94371.8400,12582.9120,69206.0160,10.6213,9.8724,334.3654,-33.5708
200,8x64x64,32,1048576,188743.6800,25165.8240,138412.0320,21.3695,9.8138,336.3646,-33.7715
200,16x4x2,32,4096,737.2800,98.3040,540.6720,0.0735,11.1455,296.1731,9.5540
200,16x4x4,32,8192,1474.5600,196.6080,1081.3440,0.1485,11.0346,299.1487,9.6500
200,16x4x8,32,16384,2949.1200,393.2160,2162.6880,0.2974,11.0192,299.5679,9.6635
200,16x4x16,32,32768,5898.2400,786.4320,4325.3760,0.5902,11.1047,297.2608,9.5891
200,16x4x32,32,65536,11796.4800,1572.8640,8650.7520,1.2747,10.2823,321.0372,10.3560
200,16x4x64,32,131072,23592.9600,3145.7280,17301.5040,2.6211,10.0014,330.0542,10.6469
200,16x8x2,32,8192,1474.5600,196.6080,1081.3440,0.1464,11.1916,294.9529,9.5146
200,16x8x4,32,16384,2949.1200,393.2160,2162.6880,0.2968,11.0406,298.9861,9.6447
200,16x8x8,32,32768,5898.2400,786.4320,4325.3760,0.5997,10.9280,302.0694,9.7442
200,16x8x16,32,65536,11796.4800,1572.8640,8650.7520,1.2737,10.2908,320.7705,10.3474
200,16x8x32,32,131072,23592.9600,3145.7280,17301.5040,2.6176,10.0146,329.6179,10.6328
200,16x8x64,32,262144,47185.9200,6291.4560,34603.0080,5.2740,9.9410,332.0605,10.7116
200,16x16x2,32,16384,2949.1200,393.2160,2162.6880,0.2980,10.9960,300.1990,9.6838
200,16x16x4,32,32768,5898.2400,786.4320,4325.3760,0.5998,10.9255,302.1364,9.7463
200,16x16x8,32,65536,11796.4800,1572.8640,8650.7520,1.2795,10.2436,322.2487,10.3951
200,16x16x16,32,131072,23592.9600,3145.7280,17301.5040,2.6165,10.0187,329.4831,10.6285
200,16x16x32,32,262144,47185.9200,6291.4560,34603.0080,5.2867,9.9172,332.8567,10.7373
200,16x16x64,32,524288,94371.8400,12582.9120,69206.0160,10.6383,9.8566,334.9024,-33.6247
200,16x32x2,32,32768,5898.2400,786.4320,4325.3760,0.5986,10.9484,301.5060,9.7260
200,16x32x4,32,65536,11796.4800,1572.8640,8650.7520,1.2760,10.2722,321.3535,10.3662
200,16x32x8,32,131072,23592.9600,3145.7280,17301.5040,2.6148,10.0254,329.2634,10.6214
200,16x32x16,32,262144,47185.9200,6291.4560,34603.0080,5.2901,9.9108,333.0721,10.7443
200,16x32x32,32,524288,94371.8400,12582.9120,69206.0160,10.6227,9.8711,334.4096,-33.5753
200,16x32x64,32,1048576,188743.6800,25165.8240,138412.0320,21.3623,9.8171,336.2509,-33.7601
200,16x64x2,32,65536,11796.4800,1572.8640,8650.7520,1.2790,10.2483,322.1007,10.3903
200,16x64x4,32,131072,23592.9600,3145.7280,17301.5040,2.6180,10.0131,329.6669,10.6344
200,16x64x8,32,262144,47185.9200,6291.4560,34603.0080,5.2855,9.9195,332.7805,10.7349
200,16x64x16,32,524288,94371.8400,12582.9120,69206.0160,10.6399,9.8552,334.9509,-33.6296
200,16x64x32,32,1048576,188743.6800,25165.8240,138412.0320,21.3698,9.8136,336.3697,-33.7721
200,16x64x64,32,2097152,377487.3600,50331.6480,276824.0640,43.1537,9.7195,339.6280,1212.9573
200,32x4x2,32,8192,1474.5600,196.6080,1081.3440,0.1487,11.0146,299.6924,9.6675
200,32x4x4,32,16384,2949.1200,393.2160,2162.6880,0.2971,11.0283,299.3212,9.6555
200,32x4x8,32,32768,5898.2400,786.4320,4325.3760,0.6031,10.8658,303.7984,9.7999
200,32x4x16,32,65536,11796.4800,1572.8640,8650.7520,1.2738,10.2897,320.8047,10.3485
200,32x4x32,32,131072,23592.9600,3145.7280,17301.5040,2.6164,10.0192,329.4670,10.6280
200,32x4x64,32,262144,47185.9200,6291.4560,34603.0080,5.2957,9.9002,333.4274,10.7557
200,32x8x2,32,16384,2949.1200,393.2160,2162.6880,0.2961,11.0654,298.3166,9.6231
200,32x8x4,32,32768,5898.2400,786.4320,4325.3760,0.5972,10.9736,300.8139,9.7037
200,32x8x8,32,65536,11796.4800,1572.8640,8650.7520,1.2758,10.2741,321.2934,10.3643
200,32x8x16,32,131072,23592.9600,3145.7280,17301.5040,2.6237,9.9914,330.3840,10.6575
200,32x8x32,32,262144,47185.9200,6291.4560,34603.0080,5.2833,9.9235,332.6437,10.7304
200,32x8x64,32,524288,94371.8400,12582.9120,69206.0160,10.6768,9.8211,336.1128,-33.7463
200,32x16x2,32,32768,5898.2400,786.4320,4325.3760,0.6002,10.9182,302.3380,9.7528
200,32x16x4,32,65536,11796.4800,1572.8640,8650.7520,1.2746,10.2830,321.0152,10.3553
200,32x16x8,32,131072,23592.9600,3145.7280,17301.5040,2.6157,10.0219,329.3779,10.6251
200,32x16x16,32,262144,47185.9200,6291.4560,34603.0080,5.2852,9.9200,332.7618,10.7343
200,32x16x32,32,524288,94371.8400,12582.9120,69206.0160,10.6337,9.8609,334.7568,-33.6101
200,32x16x64,32,1048576,188743.6800,25165.8240,138412.0320,21.3454,9.8248,335.9852,-33.7335
200,32x32x2,32,65536,11796.4800,1572.8640,8650.7520,1.2752,10.2788,321.1461,10.3596
200,32x32x4,32,131072,23592.9600,3145.7280,17301.5040,2.6087,10.0488,328.4974,10.5967
200,32x32x8,32,262144,47185.9200,6291.4560,34603.0080,5.2916,9.9080,333.1646,10.7472
200,32x32x16,32,524288,94371.8400,12582.9120,69206.0160,10.6600,9.8365,335.5854,-33.6933
200,32x32x32,32,1048576,188743.6800,25165.8240,138412.0320,21.2981,9.8467,335.2408,-33.6587
200,32x32x64,32,2097152,377487.3600,50331.6480,276824.0640,44.2297,9.4830,348.0965,1243.2018
200,32x64x2,32,131072,23592.9600,3145.7280,17301.5040,2.6099,10.0444,328.6422,10.6014
200,32x64x4,32,262144,47185.9200,6291.4560,34603.0080,5.2859,9.9185,332.8113,10.7358
200,32x64x8,32,524288,94371.8400,12582.9120,69206.0160,10.6227,9.8711,334.4097,-33.5753
200,32x64x16,32,1048576,188743.6800,25165.8240,138412.0320,21.3381,9.8282,335.8697,-33.7219
200,32x64x32,32,2097152,377487.3600,50331.6480,276824.0640,43.4748,9.6477,342.1553,1221.9831
200,32x64x64,32,4194304,754974.7200,100663.2960,553648.1280,85.5724,9.8029,336.7359,1202.6281
200,64x4x2,32,16384,2949.1200,393.2160,2162.6880,0.2976,11.0103,299.8114,9.6713
200,64x4x4,32,32768,5898.2400,786.4320,4325.3760,0.5956,11.0036,299.9923,9.6772
200,64x4x8,32,65536,11796.4800,1572.8640,8650.7520,1.2751,10.2796,321.1224,10.3588
200,64x4x16,32,131072,23592.9600,3145.7280,17301.5040,2.6173,10.0158,329.5782,10.6316
200,64x4x32,32,262144,47185.9200,6291.4560,34603.0080,5.2761,9.9371,332.1908,10.7158
200,64x4x64,32,524288,94371.8400,12582.9120,69206.0160,10.6702,9.8271,335.9071,-33.7256
200,64x8x2,32,32768,5898.2400,786.4320,4325.3760,0.5979,10.9603,301.1766,9.7154
200,64x8x4,32,65536,11796.4800,1572.8640,8650.7520,1.2730,10.2965,320.5931,10.3417
200,64x8x8,32,131072,23592.9600,3145.7280,17301.5040,2.6162,10.0199,329.4430,10.6272
200,64x8x16,32,262144,47185.9200,6291.4560,34603.0080,5.2847,9.9208,332.7343,10.7334
200,64x8x32,32,524288,94371.8400,12582.9120,69206.0160,10.6461,9.8494,335.1478,-33.6494
200,64x8x64,32,1048576,188743.6800,25165.8240,138412.0320,21.2812,9.8545,334.9748,-33.6320
200,64x16x2,32,65536,11796.4800,1572.8640,8650.7520,1.2672,10.3434,319.1396,10.2948
200,64x16x4,32,131072,23592.9600,3145.7280,17301.5040,2.6178,10.0139,329.6408,10.6336
200,64x16x8,32,262144,47185.9200,6291.4560,34603.0080,5.2912,9.9087,333.1431,10.7466
200,64x16x16,32,524288,94371.8400,12582.9120,69206.0160,10.6585,9.8379,335.5386,-33.6886
200,64x16x32,32,1048576,188743.6800,25165.8240,138412.0320,21.3308,9.8316,335.7558,-33.7104
200,64x16x64,32,2097152,377487.3600,50331.6480,276824.0640,42.6895,9.8252,335.9745,1199.9088
200,64x32x2,32,131072,23592.9600,3145.7280,17301.5040,2.6194,10.0078,329.8425,10.6401
200,64x32x4,32,262144,47185.9200,6291.4560,34603.0080,5.2909,9.9093,333.1209,10.7458
200,64x32x8,32,524288,94371.8400,12582.9120,69206.0160,10.6534,9.8426,335.3774,-33.6724
200,64x32x16,32,1048576,188743.6800,25165.8240,138412.0320,21.3611,9.8176,336.2327,-33.7583
200,64x32x32,32,2097152,377487.3600,50331.6480,276824.0640,43.1672,9.7164,339.7341,1213.3361
200,64x32x64,32,4194304,754974.7200,100663.2960,553648.1280,86.3298,9.7169,339.7161,1213.2719
200,64x64x2,32,262144,47185.9200,6291.4560,34603.0080,5.2673,9.9537,331.6370,10.6980
200,64x64x4,32,524288,94371.8400,12582.9120,69206.0160,10.6459,9.8496,335.1422,-33.6488
200,64x64x8,32,1048576,188743.6800,25165.8240,138412.0320,21.2973,9.8471,335.2271,-33.6573
200,64x64x16,32,2097152,377487.3600,50331.6480,276824.0640,42.8531,9.7876,337.2623,1204.5083
200,64x64x32,32,4194304,754974.7200,100663.2960,553648.1280,85.7954,9.7775,337.6135,1205.7624
200,64x64x64,32,8388608,1509949.4400,201326.5920,1107296.2560,175.9705,9.5341,346.2307,1236.5383
1000,4x4x2,8,256,21.5040,6.1440,9.2160,0.0083,30.9670,77.5342,11.0763
1000,4x4x4,8,512,43.0080,12.2880,18.4320,0.0174,29.3700,81.7501,11.6786
1000,4x4x8,8,1024,86.0160,24.5760,36.8640,0.0343,29.8299,80.4898,11.4985
1000,4x4x16,8,2048,172.0320,49.1520,73.7280,0.0691,29.6372,81.0130,11.5733
1000,4x4x32,8,4096,344.0640,98.3040,147.4560,0.1401,29.2325,82.1345,11.7335
1000,4x4x64,8,8192,688.1280,196.6080,294.9120,0.2881,28.4384,84.4280,12.0611
1000,4x8x2,8,512,43.0080,12.2880,18.4320,0.0170,30.1311,79.6851,11.3836
1000,4x8x4,8,1024,86.0160,24.5760,36.8640,0.0348,29.4583,81.5050,11.6436
1000,4x8x8,8,2048,172.0320,49.1520,73.7280,0.0689,29.7161,80.7979,11.5426
1000,4x8x16,8,4096,344.0640,98.3040,147.4560,0.1404,29.1689,82.3138,11.7591
1000,4x8x32,8,8192,688.1280,196.6080,294.9120,0.2892,28.3229,84.7724,12.1103
1000,4x8x64,8,16384,1376.2560,393.2160,589.8240,0.5720,28.6452,83.8185,11.9741
1000,4x16x2,8,1024,86.0160,24.5760,36.8640,0.0356,28.7810,83.4231,11.9176
1000,4x16x4,8,2048,172.0320,49.1520,73.7280,0.0688,29.7735,80.6421,11.5203
1000,4x16x8,8,4096,344.0640,98.3040,147.4560,0.1422,28.8142,83.3269,11.9038
1000,4x16x16,8,8192,688.1280,196.6080,294.9120,0.2883,28.4145,84.4992,12.0713
1000,4x16x32,8,16384,1376.2560,393.2160,589.8240,0.5915,27.6978,86.6857,12.3837
1000,4x16x64,8,32768,2752.5120,786.4320,1179.6480,1.1573,28.3139,84.7992,12.1142
1000,4x32x2,8,2048,172.0320,49.1520,73.7280,0.0694,29.5152,81.3478,11.6211
1000,4x32x4,8,4096,344.0640,98.3040,147.4560,0.1408,29.0865,82.5470,11.7924
1000,4x32x8,8,8192,688.1280,196.6080,294.9120,0.2926,27.9986,85.7544,12.2506
1000,4x32x16,8,16384,1376.2560,393.2160,589.8240,0.5922,27.6675,86.7806,12.3972
1000,4x32x32,8,32768,2752.5120,786.4320,1179.6480,1.1478,28.5484,84.1027,12.0147
1000,4x32x64,8,65536,5505.0240,1572.8640,2359.2960,2.9154,22.4796,106.8080,15.2583
1000,4x64x2,8,4096,344.0640,98.3040,147.4560,0.1386,29.5472,81.2597,11.6085
1000,4x64x4,8,8192,688.1280,196.6080,294.9120,0.2868,28.5604,84.0673,12.0096
1000,4x64x8,8,16384,1376.2560,393.2160,589.8240,0.5812,28.1914,85.1677,12.1668
1000,4x64x16,8,32768,2752.5120,786.4320,1179.6480,1.1536,28.4058,84.5250,12.0750
1000,4x64x32,8,65536,5505.0240,1572.8640,2359.2960,2.9057,22.5544,106.4539,15.2077
1000,4x64x64,8,131072,11010.0480,3145.7280,4718.5920,7.0310,18.6421,128.7947,18.3992
1000,8x4x2,8,512,43.0080,12.2880,18.4320,0.0170,30.0859,79.8048,11.4007
1000,8x4x4,8,1024,86.0160,24.5760,36.8640,0.0343,29.8537,80.4257,11.4894
1000,8x4x8,8,2048,172.0320,49.1520,73.7280,0.0697,29.3996,81.6679,11.6668
1000,8x4x16,8,4096,344.0640,98.3040,147.4560,0.1376,29.7726,80.6447,11.5207
1000,8x4x32,8,8192,688.1280,196.6080,294.9120,0.2908,28.1687,85.2363,12.1766
1000,8x4x64,8,16384,1376.2560,393.2160,589.8240,0.5743,28.5301,84.1566,12.0224
1000,8x8x2,8,1024,86.0160,24.5760,36.8640,0.0349,29.3367,81.8430,11.6919
1000,8x8x4,8,2048,172.0320,49.1520,73.7280,0.0694,29.5090,81.3651,11.6236
1000,8x8x8,8,4096,344.0640,98.3040,147.4560,0.1426,28.7195,83.6016,11.9431
1000,8x8x16,8,8192,688.1280,196.6080,294.9120,0.2867,28.5694,84.0408,12.0058
1000,8x8x32,8,16384,1376.2560,393.2160,589.8240,0.5736,28.5656,84.0520,12.0074
1000,8x8x64,8,32768,2752.5120,786.4320,1179.6480,1.1689,28.0333,85.6482,12.2355
1000,8x16x2,8,2048,172.0320,49.1520,73.7280,0.0711,28.7897,83.3979,11.9140
1000,8x16x4,8,4096,344.0640,98.3040,147.4560,0.1372,29.8442,80.4512,11.4930
1000,8x16x8,8,8192,688.1280,196.6080,294.9120,0.2941,27.8531,86.2021,12.3146
1000,8x16x16,8,16384,1376.2560,393.2160,589.8240,0.5863,27.9445,85.9205,12.2744
1000,8x16x32,8,32768,2752.5120,786.4320,1179.6480,1.1456,28.6026,83.9434,11.9919
1000,8x16x64,8,65536,5505.0240,1572.8640,2359.2960,2.8886,22.6878,105.8276,15.1182
1000,8x32x2,8,4096,344.0640,98.3040,147.4560,0.1396,29.3343,81.8496,11.6928
1000,8x32x4,8,8192,688.1280,196.6080,294.9120,0.2935,27.9113,86.0226,12.2889
1000,8x32x8,8,16384,1376.2560,393.2160,589.8240,0.5721,28.6397,83.8346,11.9764
1000,8x32x16,8,32768,2752.5120,786.4320,1179.6480,1.1710,27.9820,85.8051,12.2579
1000,8x32x32,8,65536,5505.0240,1572.8640,2359.2960,2.8921,22.6607,105.9545,15.1364
1000,8x32x64,8,131072,11010.0480,3145.7280,4718.5920,7.1243,18.3979,130.5043,18.6435
1000,8x64x2,8,8192,688.1280,196.6080,294.9120,0.2906,28.1914,85.1678,12.1668
1000,8x64x4,8,16384,1376.2560,393.2160,589.8240,0.5911,27.7178,86.6230,12.3747
1000,8x64x8,8,32768,2752.5120,786.4320,1179.6480,1.1648,28.1307,85.3515,12.1931
1000,8x64x16,8,65536,5505.0240,1572.8640,2359.2960,2.9356,22.3249,107.5479,15.3640
1000,8x64x32,8,131072,11010.0480,3145.7280,4718.5920,7.1261,18.3933,130.5367,18.6481
1000,8x64x64,8,262144,22020.0960,6291.4560,9437.1840,15.3614,17.0651,140.6965,20.0995
1000,16x4x2,8,1024,86.0160,24.5760,36.8640,0.0358,28.6355,83.8469,11.9781
1000,16x4x4,8,2048,172.0320,49.1520,73.7280,0.0695,29.4591,81.5028,11.6433
1000,16x4x8,8,4096,344.0640,98.3040,147.4560,0.1411,29.0285,82.7118,11.8160
1000,16x4x16,8,8192,688.1280,196.6080,294.9120,0.2947,27.8009,86.3642,12.3377
1000,16x4x32,8,16384,1376.2560,393.2160,589.8240,0.5916,27.6949,86.6947,12.3850
1000,16x4x64,8,32768,2752.5120,786.4320,1179.6480,1.1713,27.9752,85.8260,12.2609
1000,16x8x2,8,2048,172.0320,49.1520,73.7280,0.0690,29.7017,80.8372,11.5482
1000,16x8x4,8,4096,344.0640,98.3040,147.4560,0.1416,28.9211,83.0191,11.8599
1000,16x8x8,8,8192,688.1280,196.6080,294.9120,0.2923,28.0278,85.6651,12.2379
1000,16x8x16,8,16384,1376.2560,393.2160,589.8240,0.5905,27.7469,86.5321,12.3617
1000,16x8x32,8,32768,2752.5120,786.4320,1179.6480,1.1391,28.7668,83.4641,11.9234
1000,16x8x64,8,65536,5505.0240,1572.8640,2359.2960,2.9559,22.1712,108.2939,15.4706
1000,16x16x2,8,4096,344.0640,98.3040,147.4560,0.1393,29.4145,81.6263,11.6609
1000,16x16x4,8,8192,688.1280,196.6080,294.9120,0.2840,28.8456,83.2363,11.8909
1000,16x16x8,8,16384,1376.2560,393.2160,589.8240,0.5932,27.6181,86.9358,12.4194
1000,16x16x16,8,32768,2752.5120,786.4320,1179.6480,1.1747,27.8951,86.0725,12.2961
1000,16x16x32,8,65536,5505.0240,1572.8640,2359.2960,2.8668,22.8603,105.0292,15.0042
1000,16x16x64,8,131072,11010.0480,3145.7280,4718.5920,7.1309,18.3808,130.6257,18.6608
1000,16x32x2,8,8192,688.1280,196.6080,294.9120,0.2898,28.2676,84.9383,12.1340
1000,16x32x4,8,16384,1376.2560,393.2160,589.8240,0.5868,27.9203,85.9949,12.2850
1000,16x32x8,8,32768,2752.5120,786.4320,1179.6480,1.1752,27.8833,86.1089,12.3013
1000,16x32x16,8,65536,5505.0240,1572.8640,2359.2960,2.9443,22.2585,107.8690,15.4099
1000,16x32x32,8,131072,11010.0480,3145.7280,4718.5920,7.0845,18.5011,129.7760,18.5394
1000,16x32x64,8,262144,22020.0960,6291.4560,9437.1840,15.2361,17.2054,139.5488,19.9355
1000,16x64x2,8,16384,1376.2560,393.2160,589.8240,0.5867,27.9261,85.9769,12.2824
1000,16x64x4,8,32768,2752.5120,786.4320,1179.6480,1.1651,28.1240,85.3720,12.1960
1000,16x64x8,8,65536,5505.0240,1572.8640,2359.2960,2.8820,22.7399,105.5853,15.0836
1000,16x64x16,8,131072,11010.0480,3145.7280,4718.5920,7.1805,18.2538,131.5341,18.7906
1000,16x64x32,8,262144,22020.0960,6291.4560,9437.1840,15.0906,17.3714,138.2160,19.7451
1000,16x64x64,8,524288,44040.1920,12582.9120,18874.3680,31.9739,16.3974,146.4258,20.9180
1000,32x4x2,8,2048,172.0320,49.1520,73.7280,0.0689,29.7095,80.8160,11.5451
1000,32x4x4,8,4096,344.0640,98.3040,147.4560,0.1378,29.7248,80.7743,11.5392
1000,32x4x8,8,8192,688.1280,196.6080,294.9120,0.2894,28.3078,84.8177,12.1168
1000,32x4x16,8,16384,1376.2560,393.2160,589.8240,0.5759,28.4518,84.3883,12.0555
1000,32x4x32,8,32768,2752.5120,786.4320,1179.6480,1.1651,28.1237,85.3730,12.1961
1000,32x4x64,8,65536,5505.0240,1572.8640,2359.2960,2.8862,22.7065,105.7406,15.1058
1000,32x8x2,8,4096,344.0640,98.3040,147.4560,0.1368,29.9362,80.2040,11.4577
1000,32x8x4,8,8192,688.1280,196.6080,294.9120,0.2835,28.8986,83.0837,11.8691
1000,32x8x8,8,16384,1376.2560,393.2160,589.8240,0.5787,28.3102,84.8104,12.1158
1000,32x8x16,8,32768,2752.5120,786.4320,1179.6480,1.1465,28.5810,84.0069,12.0010
1000,32x8x32,8,65536,5505.0240,1572.8640,2359.2960,2.8498,22.9967,104.4064,14.9152
1000,32x8x64,8,131072,11010.0480,3145.7280,4718.5920,7.0561,18.5757,129.2547,18.4650
1000,32x16x2,8,8192,688.1280,196.6080,294.9120,0.2852,28.7235,83.5902,11.9415
1000,32x16x4,8,16384,1376.2560,393.2160,589.8240,0.5837,28.0686,85.5404,12.2201
1000,32x16x8,8,32768,2752.5120,786.4320,1179.6480,1.1504,28.4845,84.2913,12.0416
1000,32x16x16,8,65536,5505.0240,1572.8640,2359.2960,2.8993,22.6042,106.2193,15.1742
1000,32x16x32,8,131072,11010.0480,3145.7280,4718.5920,7.0918,18.4823,129.9079,18.5583
1000,32x16x64,8,262144,22020.0960,6291.4560,9437.1840,15.4199,17.0004,141.2322,20.1760
1000,32x32x2,8,16384,1376.2560,393.2160,589.8240,0.5807,28.2129,85.1028,12.1575
1000,32x32x4,8,32768,2752.5120,786.4320,1179.6480,1.1824,27.7128,86.6386,12.3769
1000,32x32x8,8,65536,5505.0240,1572.8640,2359.2960,2.8990,22.6067,106.2074,15.1725
1000,32x32x16,8,131072,11010.0480,3145.7280,4718.5920,7.0475,18.5983,129.0979,18.4426
1000,32x32x32,8,262144,22020.0960,6291.4560,9437.1840,15.3367,17.0926,140.4703,20.0672
1000,32x32x64,8,524288,44040.1920,12582.9120,18874.3680,31.7109,16.5334,145.2215,20.7459
1000,32x64x2,8,32768,2752.5120,786.4320,1179.6480,1.1432,28.6634,83.7653,11.9665
1000,32x64x4,8,65536,5505.0240,1572.8640,2359.2960,2.8859,22.7092,105.7282,15.1040
1000,32x64x8,8,131072,11010.0480,3145.7280,4718.5920,7.0480,18.5970,129.1068,18.4438
1000,32x64x16,8,262144,22020.0960,6291.4560,9437.1840,15.1667,17.2842,138.9128,19.8447
1000,32x64x32,8,524288,44040.1920,12582.9120,18874.3680,31.3721,16.7119,143.6699,20.5243
1000,32x64x64,8,1048576,88080.3840,25165.8240,37748.7360,64.7684,16.1896,148.3048,21.1864
1000,64x4x2,8,4096,344.0640,98.3040,147.4560,0.1401,29.2284,82.1462,11.7352
1000,64x4x4,8,8192,688.1280,196.6080,294.9120,0.2859,28.6484,83.8092,11.9727
1000,64x4x8,8,16384,1376.2560,393.2160,589.8240,0.5852,27.9971,85.7588,12.2513
1000,64x4x16,8,32768,2752.5120,786.4320,1179.6480,1.1721,27.9556,85.8862,12.2695
1000,64x4x32,8,65536,5505.0240,1572.8640,2359.2960,2.9070,22.5442,106.5020,15.2146
1000,64x4x64,8,131072,11010.0480,3145.7280,4718.5920,7.0141,18.6868,128.4864,18.3552
1000,64x8x2,8,8192,688.1280,196.6080,294.9120,0.2957,27.7067,86.6578,12.3797
1000,64x8x4,8,16384,1376.2560,393.2160,589.8240,0.5773,28.3785,84.6063,12.0866
1000,64x8x8,8,32768,2752.5120,786.4320,1179.6480,1.1553,28.3623,84.6545,12.0935
1000,64x8x16,8,65536,5505.0240,1572.8640,2359.2960,2.9517,22.2026,108.1404,15.4486
1000,64x8x32,8,131072,11010.0480,3145.7280,4718.5920,7.3935,17.7279,135.4359,19.3480
1000,64x8x64,8,262144,22020.0960,6291.4560,9437.1840,15.1413,17.3132,138.6802,19.8115
1000,64x16x2,8,16384,1376.2560,393.2160,589.8240,0.5724,28.6252,83.8772,11.9825
1000,64x16x4,8,32768,2752.5120,786.4320,1179.6480,1.1533,28.4121,84.5064,12.0723
1000,64x16x8,8,65536,5505.0240,1572.8640,2359.2960,2.9298,22.3690,107.3362,15.3337
1000,64x16x16,8,131072,11010.0480,3145.7280,4718.5920,7.0712,18.5361,129.5313,18.5045
1000,64x16x32,8,262144,22020.0960,6291.4560,9437.1840,15.2023,17.2437,139.2391,19.8913
1000,64x16x64,8,524288,44040.1920,12582.9120,18874.3680,31.2550,16.7745,143.1337,20.4477
1000,64x32x2,8,32768,2752.5120,786.4320,1179.6480,1.1872,27.6006,86.9909,12.4273
1000,64x32x4,8,65536,5505.0240,1572.8640,2359.2960,2.9028,22.5766,106.3489,15.1927
1000,64x32x8,8,131072,11010.0480,3145.7280,4718.5920,7.0732,18.5308,129.5680,18.5097
1000,64x32x16,8,262144,22020.0960,6291.4560,9437.1840,15.3410,17.0878,140.5094,20.0728
1000,64x32x32,8,524288,44040.1920,12582.9120,18874.3680,31.6923,16.5431,145.1361,20.7337
1000,64x32x64,8,1048576,88080.3840,25165.8240,37748.7360,64.2230,16.3271,147.0560,21.0080
1000,64x64x2,8,65536,5505.0240,1572.8640,2359.2960,2.8779,22.7720,105.4365,15.0624
1000,64x64x4,8,131072,11010.0480,3145.7280,4718.5920,7.0810,18.5103,129.7112,18.5302
1000,64x64x8,8,262144,22020.0960,6291.4560,9437.1840,15.1939,17.2532,139.1623,19.8803
1000,64x64x16,8,524288,44040.1920,12582.9120,18874.3680,31.4185,16.6872,143.8826,20.5547
1000,64x64x32,8,1048576,88080.3840,25165.8240,37748.7360,63.5853,16.4909,145.5959,20.7994
1000,64x64x64,8,2097152,176160.7680,50331.6480,75497.4720,125.9547,16.6500,144.2038,20.6005
1000,4x4x2,16,512,59.3920,12.2880,34.8160,0.0280,18.2876,131.2913,8.7528
1000,4x4x4,16,1024,118.7840,24.5760,69.6320,0.0601,17.0372,140.9266,9.3951
1000,4x4x8,16,2048,237.5680,49.1520,139.2640,0.1130,18.1276,132.4500,8.8300
1000,4x4x16,16,4096,475.1360,98.3040,278.5280,0.2272,18.0306,133.1625,8.8775
1000,4x4x32,16,8192,950.2720,196.6080,557.0560,0.4596,17.8244,134.7032,8.9802
1000,4x4x64,16,16384,1900.5440,393.2160,1114.1120,0.9065,18.0747,132.8377,8.8558
1000,4x8x2,16,1024,118.7840,24.5760,69.6320,0.0562,18.2071,131.8715,8.7914
1000,4x8x4,16,2048,237.5680,49.1520,139.2640,0.1116,18.3563,130.7997,8.7200
1000,4x8x8,16,4096,475.1360,98.3040,278.5280,0.2282,17.9456,133.7934,8.9196
1000,4x8x16,16,8192,950.2720,196.6080,557.0560,0.4603,17.7979,134.9035,8.9936
1000,4x8x32,16,16384,1900.5440,393.2160,1114.1120,0.9207,17.7954,134.9228,8.9949
1000,4x8x64,16,32768,3801.0880,786.4320,2228.2240,1.8206,17.9982,133.4024,8.8935
1000,4x16x2,16,2048,237.5680,49.1520,139.2640,0.1129,18.1431,132.3366,8.8224
1000,4x16x4,16,4096,475.1360,98.3040,278.5280,0.2266,18.0738,132.8439,8.8563
1000,4x16x8,16,8192,950.2720,196.6080,557.0560,0.4606,17.7846,135.0044,9.0003
1000,4x16x16,16,16384,1900.5440,393.2160,1114.1120,0.9065,18.0746,132.8387,8.8559
1000,4x16x32,16,32768,3801.0880,786.4320,2228.2240,1.8156,18.0480,133.0340,8.8689
1000,4x16x64,16,65536,7602.1760,1572.8640,4456.4480,4.1961,15.6185,153.7282,10.2485
1000,4x32x2,16,4096,475.1360,98.3040,278.5280,0.2227,18.3938,130.5333,8.7022
1000,4x32x4,16,8192,950.2720,196.6080,557.0560,0.4560,17.9654,133.6456,8.9097
1000,4x32x8,16,16384,1900.5440,393.2160,1114.1120,0.9065,18.0730,132.8501,8.8567
1000,4x32x16,16,32768,3801.0880,786.4320,2228.2240,1.8438,17.7718,135.1017,9.0068
1000,4x32x32,16,65536,7602.1760,1572.8640,4456.4480,4.1937,15.6272,153.6423,10.2428
1000,4x32x64,16,131072,15204.3520,3145.7280,8912.8960,8.6363,15.1770,158.2004,10.5467
1000,4x64x2,16,8192,950.2720,196.6080,557.0560,0.4594,17.8324,134.6428,8.9762
1000,4x64x4,16,16384,1900.5440,393.2160,1114.1120,0.9249,17.7138,135.5442,9.0363
1000,4x64x8,16,32768,3801.0880,786.4320,2228.2240,1.8450,17.7609,135.1846,9.0123
1000,4x64x16,16,65536,7602.1760,1572.8640,4456.4480,4.1901,15.6405,153.5118,10.2341
1000,4x64x32,16,131072,15204.3520,3145.7280,8912.8960,8.6339,15.1811,158.1575,10.5438
1000,4x64x64,16,262144,30408.7040,6291.4560,17825.7920,17.4726,15.0031,160.0333,10.6689
1000,8x4x2,16,1024,118.7840,24.5760,69.6320,0.0556,18.4242,130.3179,8.6879
1000,8x4x4,16,2048,237.5680,49.1520,139.2640,0.1115,18.3608,130.7676,8.7178
1000,8x4x8,16,4096,475.1360,98.3040,278.5280,0.2238,18.3034,131.1781,8.7452
1000,8x4x16,16,8192,950.2720,196.6080,557.0560,0.4585,17.8671,134.3814,8.9588
1000,8x4x32,16,16384,1900.5440,393.2160,1114.1120,0.9209,17.7921,134.9472,8.9965
1000,8x4x64,16,32768,3801.0880,786.4320,2228.2240,1.8148,18.0560,132.9752,8.8650
1000,8x8x2,16,2048,237.5680,49.1520,139.2640,0.1129,18.1326,132.4138,8.8276
1000,8x8x4,16,4096,475.1360,98.3040,278.5280,0.2261,18.1147,132.5445,8.8363
1000,8x8x8,16,8192,950.2720,196.6080,557.0560,0.4553,17.9909,133.4562,8.8971
1000,8x8x16,16,16384,1900.5440,393.2160,1114.1120,0.9142,17.9220,133.9695,8.9313
1000,8x8x32,16,32768,3801.0880,786.4320,2228.2240,1.8324,17.8830,134.2618,8.9508
1000,8x8x64,16,65536,7602.1760,1572.8640,4456.4480,4.1905,15.6392,153.5243,10.2350
1000,8x16x2,16,4096,475.1360,98.3040,278.5280,0.2247,18.2328,131.6857,8.7790
1000,8x16x4,16,8192,950.2720,196.6080,557.0560,0.4584,17.8694,134.3637,8.9576
1000,8x16x8,16,16384,1900.5440,393.2160,1114.1120,0.9115,17.9756,133.5702,8.9047
1000,8x16x16,16,32768,3801.0880,786.4320,2228.2240,1.8386,17.8221,134.7201,8.9813
1000,8x16x32,16,65536,7602.1760,1572.8640,4456.4480,4.1788,15.6830,153.0953,10.2064
1000,8x16x64,16,131072,15204.3520,3145.7280,8912.8960,8.6652,15.1262,158.7312,10.5821
1000,8x32x2,16,8192,950.2720,196.6080,557.0560,0.4603,17.7970,134.9102,8.9940
1000,8x32x4,16,16384,1900.5440,393.2160,1114.1120,0.9139,17.9273,133.9299,8.9287
1000,8x32x8,16,32768,3801.0880,786.4320,2228.2240,1.8440,17.7700,135.1152,9.0077
1000,8x32x16,16,65536,7602.1760,1572.8640,4456.4480,4.8293,13.5705,176.9282,11.7952
1000,8x32x32,16,131072,15204.3520,3145.7280,8912.8960,8.6609,15.1338,158.6510,10.5767
1000,8x32x64,16,262144,30408.7040,6291.4560,17825.7920,17.4388,15.0322,159.7238,10.6483
1000,8x64x2,16,16384,1900.5440,393.2160,1114.1120,0.9137,17.9323,133.8928,8.9262
1000,8x64x4,16,32768,3801.0880,786.4320,2228.2240,1.8088,18.1163,132.5328,8.8355
1000,8x64x8,16,65536,7602.1760,1572.8640,4456.4480,4.1850,15.6597,153.3232,10.2215
1000,8x64x16,16,131072,15204.3520,3145.7280,8912.8960,8.6250,15.1968,157.9938,10.5329
1000,8x64x32,16,262144,30408.7040,6291.4560,17825.7920,17.4081,15.0587,159.4423,10.6295
1000,8x64x64,16,524288,60817.4080,12582.9120,35651.5840,35.3398,14.8356,161.8402,10.7893
1000,16x4x2,16,2048,237.5680,49.1520,139.2640,0.1124,18.2263,131.7324,8.7822
1000,16x4x4,16,4096,475.1360,98.3040,278.5280,0.2265,18.0842,132.7675,8.8512
1000,16x4x8,16,8192,950.2720,196.6080,557.0560,0.4601,17.8055,134.8461,8.9897
1000,16x4x16,16,16384,1900.5440,393.2160,1114.1120,0.9200,17.8089,134.8199,8.9880
1000,16x4x32,16,32768,3801.0880,786.4320,2228.2240,1.8303,17.9035,134.1081,8.9405
1000,16x4x64,16,65536,7602.1760,1572.8640,4456.4480,4.1983,15.6100,153.8118,10.2541
1000,16x8x2,16,4096,475.1360,98.3040,278.5280,0.2231,18.3580,130.7878,8.7192
1000,16x8x4,16,8192,950.2720,196.6080,557.0560,0.4627,17.7039,135.6196,9.0413
1000,16x8x8,16,16384,1900.5440,393.2160,1114.1120,0.9071,18.0624,132.9284,8.8619
1000,16x8x16,16,32768,3801.0880,786.4320,2228.2240,1.8383,17.8252,134.6967,8.9798
1000,16x8x32,16,65536,7602.1760,1572.8640,4456.4480,4.1920,15.6337,153.5790,10.2386
1000,16x8x64,16,131072,15204.3520,3145.7280,8912.8960,8.6293,15.1892,158.0725,10.5382
1000,16x16x2,16,8192,950.2720,196.6080,557.0560,0.4582,17.8769,134.3077,8.9538
1000,16x16x4,16,16384,1900.5440,393.2160,1114.1120,0.9178,17.8520,134.4944,8.9663
1000,16x16x8,16,32768,3801.0880,786.4320,2228.2240,1.8323,17.8835,134.2578,8.9505
1000,16x16x16,16,65536,7602.1760,1572.8640,4456.4480,4.1813,15.6735,153.1887,10.2126
1000,16x16x32,16,131072,15204.3520,3145.7280,8912.8960,8.6485,15.1555,158.4247,10.5616
1000,16x16x64,16,262144,30408.7040,6291.4560,17825.7920,17.3921,15.0726,159.2954,10.6197
1000,16x32x2,16,16384,1900.5440,393.2160,1114.1120,0.9191,17.8264,134.6879,8.9792
1000,16x32x4,16,32768,3801.0880,786.4320,2228.2240,1.8266,17.9393,133.8401,8.9227
1000,16x32x8,16,65536,7602.1760,1572.8640,4456.4480,4.1715,15.7105,152.8278,10.1885
1000,16x32x16,16,131072,15204.3520,3145.7280,8912.8960,8.6513,15.1506,158.4756,10.5650
1000,16x32x32,16,262144,30408.7040,6291.4560,17825.7920,17.5022,14.9778,160.3039,10.6869
1000,16x32x64,16,524288,60817.4080,12582.9120,35651.5840,35.1069,14.9340,160.7738,10.7183
1000,16x64x2,16,32768,3801.0880,786.4320,2228.2240,1.8369,17.8391,134.5922,8.9728
1000,16x64x4,16,65536,7602.1760,1572.8640,4456.4480,4.1644,15.7371,152.5693,10.1713
1000,16x64x8,16,131072,15204.3520,3145.7280,8912.8960,8.6170,15.2109,157.8476,10.5232
1000,16x64x16,16,262144,30408.7040,6291.4560,17825.7920,17.4472,15.0250,159.8003,10.6534
1000,16x64x32,16,524288,60817.4080,12582.9120,35651.5840,35.0089,14.9758,160.3249,10.6883
1000,16x64x64,16,1048576,121634.8160,25165.8240,71303.1680,72.1388,14.5355,165.1814,11.0121
1000,32x4x2,16,4096,475.1360,98.3040,278.5280,0.2250,18.2023,131.9062,8.7937
1000,32x4x4,16,8192,950.2720,196.6080,557.0560,0.4598,17.8161,134.7657,8.9844
1000,32x4x8,16,16384,1900.5440,393.2160,1114.1120,0.9061,18.0816,132.7866,8.8524
1000,32x4x16,16,32768,3801.0880,786.4320,2228.2240,1.8442,17.7680,135.1302,9.0087
1000,32x4x32,16,65536,7602.1760,1572.8640,4456.4480,4.2043,15.5878,154.0308,10.2687
1000,32x4x64,16,131072,15204.3520,3145.7280,8912.8960,8.6253,15.1962,158.0005,10.5334
1000,32x8x2,16,8192,950.2720,196.6080,557.0560,0.4542,18.0361,133.1217,8.8748
1000,32x8x4,16,16384,1900.5440,393.2160,1114.1120,0.9193,17.8219,134.7221,8.9815
1000,32x8x8,16,32768,3801.0880,786.4320,2228.2240,1.8215,17.9900,133.4632,8.8975
1000,32x8x16,16,65536,7602.1760,1572.8640,4456.4480,4.1587,15.7590,152.3578,10.1572
1000,32x8x32,16,131072,15204.3520,3145.7280,8912.8960,8.6452,15.1613,158.3642,10.5576
1000,32x8x64,16,262144,30408.7040,6291.4560,17825.7920,17.4339,15.0365,159.6785,10.6452
1000,32x16x2,16,16384,1900.5440,393.2160,1114.1120,0.9196,17.8166,134.7621,8.9841
1000,32x16x4,16,32768,3801.0880,786.4320,2228.2240,1.8172,18.0323,133.1500,8.8767
1000,32x16x8,16,65536,7602.1760,1572.8640,4456.4480,4.1820,15.6710,153.2125,10.2142
1000,32x16x16,16,131072,15204.3520,3145.7280,8912.8960,8.6406,15.1693,158.2806,10.5520
1000,32x16x32,16,262144,30408.7040,6291.4560,17825.7920,17.4204,15.0481,159.5547,10.6370
1000,32x16x64,16,524288,60817.4080,12582.9120,35651.5840,35.0026,14.9786,160.2959,10.6864
1000,32x32x2,16,32768,3801.0880,786.4320,2228.2240,1.8324,17.8826,134.2647,8.9510
1000,32x32x4,16,65536,7602.1760,1572.8640,4456.4480,4.1733,15.7036,152.8946,10.1930
1000,32x32x8,16,131072,15204.3520,3145.7280,8912.8960,8.6484,15.1556,158.4234,10.5616
1000,32x32x16,16,262144,30408.7040,6291.4560,17825.7920,17.4350,15.0355,159.6888,10.6459
1000,32x32x32,16,524288,60817.4080,12582.9120,35651.5840,35.0854,14.9432,160.6750,10.7117
1000,32x32x64,16,1048576,121634.8160,25165.8240,71303.1680,70.6692,14.8378,161.8165,10.7878
1000,32x64x2,16,65536,7602.1760,1572.8640,4456.4480,4.1885,15.6465,153.4529,10.2302
1000,32x64x4,16,131072,15204.3520,3145.7280,8912.8960,8.6618,15.1321,158.6692,10.5779
1000,32x64x8,16,262144,30408.7040,6291.4560,17825.7920,17.4122,15.0552,159.4796,10.6320
1000,32x64x16,16,524288,60817.4080,12582.9120,35651.5840,34.9968,14.9810,160.2693,10.6846
1000,32x64x32,16,1048576,121634.8160,25165.8240,71303.1680,71.3457,14.6971,163.3653,10.8910
1000,32x64x64,16,2097152,243269.6320,50331.6480,142606.3360,142.7880,14.6872,163.4760,10.8984
1000,64x4x2,16,8192,950.2720,196.6080,557.0560,0.4560,17.9634,133.6604,8.9107
1000,64x4x4,16,16384,1900.5440,393.2160,1114.1120,0.9158,17.8900,134.2094,8.9473
1000,64x4x8,16,32768,3801.0880,786.4320,2228.2240,1.8438,17.7723,135.0980,9.0065
1000,64x4x16,16,65536,7602.1760,1572.8640,4456.4480,4.1660,15.7313,152.6255,10.1750
1000,64x4x32,16,131072,15204.3520,3145.7280,8912.8960,8.6348,15.1795,158.1737,10.5449
1000,64x4x64,16,262144,30408.7040,6291.4560,17825.7920,17.5210,14.9617,160.4760,10.6984
1000,64x8x2,16,16384,1900.5440,393.2160,1114.1120,0.9088,18.0283,133.1792,8.8786
1000,64x8x4,16,32768,3801.0880,786.4320,2228.2240,1.8309,17.8973,134.1545,8.9436
1000,64x8x8,16,65536,7602.1760,1572.8640,4456.4480,4.2025,15.5945,153.9650,10.2643
1000,64x8x16,16,131072,15204.3520,3145.7280,8912.8960,8.6268,15.1937,158.0264,10.5351
1000,64x8x32,16,262144,30408.7040,6291.4560,17825.7920,17.4245,15.0446,159.5921,10.6395
1000,64x8x64,16,524288,60817.4080,12582.9120,35651.5840,35.3317,14.8390,161.8029,10.7869
1000,64x16x2,16,32768,3801.0880,786.4320,2228.2240,1.8250,17.9552,133.7218,8.9148
1000,64x16x4,16,65536,7602.1760,1572.8640,4456.4480,4.1726,15.7062,152.8698,10.1913
1000,64x16x8,16,131072,15204.3520,3145.7280,8912.8960,8.6631,15.1299,158.6928,10.5795
1000,64x16x16,16,262144,30408.7040,6291.4560,17825.7920,17.3823,15.0811,159.2060,10.6137
1000,64x16x32,16,524288,60817.4080,12582.9120,35651.5840,34.9747,14.9905,160.1681,10.6779
1000,64x16x64,16,1048576,121634.8160,25165.8240,71303.1680,70.8759,14.7945,162.2896,10.8193
1000,64x32x2,16,65536,7602.1760,1572.8640,4456.4480,4.2008,15.6010,153.9008,10.2601
1000,64x32x4,16,131072,15204.3520,3145.7280,8912.8960,8.6586,15.1378,158.6100,10.5740
1000,64x32x8,16,262144,30408.7040,6291.4560,17825.7920,17.4436,15.0281,159.7671,10.6511
1000,64x32x16,16,524288,60817.4080,12582.9120,35651.5840,35.1817,14.9023,161.1163,10.7411
1000,64x32x32,16,1048576,121634.8160,25165.8240,71303.1680,71.6125,14.6424,163.9763,10.9318
1000,64x32x64,16,2097152,243269.6320,50331.6480,142606.3360,143.0502,14.6603,163.7762,10.9184
1000,64x64x2,16,131072,15204.3520,3145.7280,8912.8960,8.6345,15.1801,158.1676,10.5445
1000,64x64x4,16,262144,30408.7040,6291.4560,17825.7920,17.5245,14.9587,160.5082,10.7005
1000,64x64x8,16,524288,60817.4080,12582.9120,35651.5840,35.1115,14.9321,160.7948,10.7197
1000,64x64x16,16,1048576,121634.8160,25165.8240,71303.1680,71.1214,14.7435,162.8518,10.8568
1000,64x64x32,16,2097152,243269.6320,50331.6480,142606.3360,143.4827,14.6161,164.2713,10.9514
1000,64x64x64,16,4194304,486539.2640,100663.2960,285212.6720,283.9210,-0.3545,-6772.0256,-451.4684
1000,4x4x2,32,1024,184.3200,24.5760,135.1680,0.0902,11.3486,211.5671,6.8247
1000,4x4x4,32,2048,368.6400,49.1520,270.3360,0.1812,11.3004,212.4704,6.8539
1000,4x4x8,32,4096,737.2800,98.3040,540.6720,0.3681,11.1280,215.7622,6.9601
1000,4x4x16,32,8192,1474.5600,196.6080,1081.3440,0.7372,11.1121,216.0702,6.9700
1000,4x4x32,32,16384,2949.1200,393.2160,2162.6880,1.4782,11.0841,216.6173,6.9877
1000,4x4x64,32,32768,5898.2400,786.4320,4325.3760,2.9977,10.9311,219.6483,7.0854
1000,4x8x2,32,2048,368.6400,49.1520,270.3360,0.1831,11.1830,214.7001,6.9258
1000,4x8x4,32,4096,737.2800,98.3040,540.6720,0.3708,11.0452,217.3798,7.0123
1000,4x8x8,32,8192,1474.5600,196.6080,1081.3440,0.7423,11.0353,217.5744,7.0185
1000,4x8x16,32,16384,2949.1200,393.2160,2162.6880,1.4850,11.0329,217.6212,7.0200
1000,4x8x32,32,32768,5898.2400,786.4320,4325.3760,2.9915,10.9538,219.1933,7.0708
1000,4x8x64,32,65536,11796.4800,1572.8640,8650.7520,6.3738,10.2822,233.5111,7.5326
1000,4x16x2,32,4096,737.2800,98.3040,540.6720,0.3720,11.0106,218.0628,7.0343
1000,4x16x4,32,8192,1474.5600,196.6080,1081.3440,0.7425,11.0324,217.6320,7.0204
1000,4x16x8,32,16384,2949.1200,393.2160,2162.6880,1.4858,11.0268,217.7423,7.0239
1000,4x16x16,32,32768,5898.2400,786.4320,4325.3760,2.9797,10.9969,218.3335,7.0430
1000,4x16x32,32,65536,11796.4800,1572.8640,8650.7520,6.3757,10.2791,233.5812,7.5349
1000,4x16x64,32,131072,23592.9600,3145.7280,17301.5040,13.1968,9.9321,241.7404,7.7981
1000,4x32x2,32,8192,1474.5600,196.6080,1081.3440,0.7442,11.0081,218.1128,7.0359
1000,4x32x4,32,16384,2949.1200,393.2160,2162.6880,1.4890,11.0032,218.2089,7.0390
1000,4x32x8,32,32768,5898.2400,786.4320,4325.3760,2.9839,10.9815,218.6413,7.0529
1000,4x32x16,32,65536,11796.4800,1572.8640,8650.7520,6.3759,10.2787,233.5892,7.5351
1000,4x32x32,32,131072,23592.9600,3145.7280,17301.5040,13.0549,10.0401,239.1414,7.7142
1000,4x32x64,32,262144,47185.9200,6291.4560,34603.0080,26.5356,9.8790,243.0420,7.8401
1000,4x64x2,32,16384,2949.1200,393.2160,2162.6880,1.4778,11.0870,216.5590,6.9858
1000,4x64x4,32,32768,5898.2400,786.4320,4325.3760,2.9831,10.9847,218.5774,7.0509
1000,4x64x8,32,65536,11796.4800,1572.8640,8650.7520,6.3862,10.2621,233.9666,7.5473
1000,4x64x16,32,131072,23592.9600,3145.7280,17301.5040,13.1334,9.9800,240.5807,7.7607
1000,4x64x32,32,262144,47185.9200,6291.4560,34603.0080,26.4259,9.9200,242.0371,7.8076
1000,4x64x64,32,524288,94371.8400,12582.9120,69206.0160,53.8714,9.7322,246.7066,7.9583
1000,8x4x2,32,2048,368.6400,49.1520,270.3360,0.1834,11.1683,214.9841,6.9350
1000,8x4x4,32,4096,737.2800,98.3040,540.6720,0.3706,11.0535,217.2172,7.0070
1000,8x4x8,32,8192,1474.5600,196.6080,1081.3440,0.7387,11.0895,216.5115,6.9842
1000,8x4x16,32,16384,2949.1200,393.2160,2162.6880,1.4829,11.0488,217.3094,7.0100
1000,8x4x32,32,32768,5898.2400,786.4320,4325.3760,2.9946,10.9422,219.4260,7.0783
1000,8x4x64,32,65536,11796.4800,1572.8640,8650.7520,6.3676,10.2921,233.2858,7.5253
1000,8x8x2,32,4096,737.2800,98.3040,540.6720,0.3697,11.0802,216.6930,6.9901
1000,8x8x4,32,8192,1474.5600,196.6080,1081.3440,0.7463,10.9762,218.7468,7.0563
1000,8x8x8,32,16384,2949.1200,393.2160,2162.6880,1.4900,10.9957,218.3571,7.0438
1000,8x8x16,32,32768,5898.2400,786.4320,4325.3760,2.9865,10.9722,218.8253,7.0589
1000,8x8x32,32,65536,11796.4800,1572.8640,8650.7520,6.3751,10.2800,233.5598,7.5342
1000,8x8x64,32,131072,23592.9600,3145.7280,17301.5040,13.0598,10.0363,239.2326,7.7172
1000,8x16x2,32,8192,1474.5600,196.6080,1081.3440,0.7391,11.0836,216.6270,6.9880
1000,8x16x4,32,16384,2949.1200,393.2160,2162.6880,1.4744,11.1125,216.0630,6.9698
1000,8x16x8,32,32768,5898.2400,786.4320,4325.3760,2.9826,10.9866,218.5399,7.0497
1000,8x16x16,32,65536,11796.4800,1572.8640,8650.7520,6.3835,10.2665,233.8681,7.5441
1000,8x16x32,32,131072,23592.9600,3145.7280,17301.5040,13.0897,10.0134,239.7787,7.7348
1000,8x16x64,32,262144,47185.9200,6291.4560,34603.0080,26.4301,9.9184,242.0754,7.8089
1000,8x32x2,32,16384,2949.1200,393.2160,2162.6880,1.4847,11.0349,217.5818,7.0188
1000,8x32x4,32,32768,5898.2400,786.4320,4325.3760,2.9873,10.9692,218.8857,7.0608
1000,8x32x8,32,65536,11796.4800,1572.8640,8650.7520,6.3839,10.2658,233.8828,7.5446
1000,8x32x16,32,131072,23592.9600,3145.7280,17301.5040,13.0760,10.0239,239.5278,7.7267
1000,8x32x32,32,262144,47185.9200,6291.4560,34603.0080,26.4446,9.9130,242.2084,7.8132
1000,8x32x64,32,524288,94371.8400,12582.9120,69206.0160,53.2448,9.8468,243.8367,7.8657
1000,8x64x2,32,32768,5898.2400,786.4320,4325.3760,2.9830,10.9850,218.5698,7.0506
1000,8x64x4,32,65536,11796.4800,1572.8640,8650.7520,6.3740,10.2818,233.5201,7.5329
1000,8x64x8,32,131072,23592.9600,3145.7280,17301.5040,13.0821,10.0192,239.6396,7.7303
1000,8x64x16,32,262144,47185.9200,6291.4560,34603.0080,26.4722,9.9026,242.4609,7.8213
1000,8x64x32,32,524288,94371.8400,12582.9120,69206.0160,53.9286,9.7219,246.9686,7.9667
1000,8x64x64,32,1048576,188743.6800,25165.8240,138412.0320,108.0731,9.7025,247.4627,7.9827
1000,16x4x2,32,4096,737.2800,98.3040,540.6720,0.3691,11.0977,216.3506,6.9791
1000,16x4x4,32,8192,1474.5600,196.6080,1081.3440,0.7398,11.0733,216.8286,6.9945
1000,16x4x8,32,16384,2949.1200,393.2160,2162.6880,1.4866,11.0209,217.8594,7.0277
1000,16x4x16,32,32768,5898.2400,786.4320,4325.3760,2.9906,10.9569,219.1318,7.0688
1000,16x4x32,32,65536,11796.4800,1572.8640,8650.7520,6.3724,10.2844,233.4603,7.5310
1000,16x4x64,32,131072,23592.9600,3145.7280,17301.5040,13.0838,10.0178,239.6722,7.7314
1000,16x8x2,32,8192,1474.5600,196.6080,1081.3440,0.7357,11.1354,215.6189,6.9554
1000,16x8x4,32,16384,2949.1200,393.2160,2162.6880,1.4879,11.0118,218.0381,7.0335
1000,16x8x8,32,32768,5898.2400,786.4320,4325.3760,2.9957,10.9384,219.5020,7.0807
1000,16x8x16,32,65536,11796.4800,1572.8640,8650.7520,6.3740,10.2818,233.5198,7.5329
1000,16x8x32,32,131072,23592.9600,3145.7280,17301.5040,13.0834,10.0182,239.6639,7.7311
1000,16x8x64,32,262144,47185.9200,6291.4560,34603.0080,26.4693,9.9037,242.4344,7.8205
1000,16x16x2,32,16384,2949.1200,393.2160,2162.6880,1.4863,11.0236,217.8045,7.0260
1000,16x16x4,32,32768,5898.2400,786.4320,4325.3760,2.9977,10.9310,219.6508,7.0855
1000,16x16x8,32,65536,11796.4800,1572.8640,8650.7520,6.3658,10.2951,233.2189,7.5232
1000,16x16x16,32,131072,23592.9600,3145.7280,17301.5040,13.0569,10.0385,239.1787,7.7154
1000,16x16x32,32,262144,47185.9200,6291.4560,34603.0080,26.4757,9.9013,242.4937,7.8224
1000,16x16x64,32,524288,94371.8400,12582.9120,69206.0160,53.2599,9.8439,243.9062,7.8679
1000,16x32x2,32,32768,5898.2400,786.4320,4325.3760,2.9927,10.9495,219.2798,7.0735
1000,16x32x4,32,65536,11796.4800,1572.8640,8650.7520,6.3878,10.2596,234.0254,7.5492
1000,16x32x8,32,131072,23592.9600,3145.7280,17301.5040,13.0650,10.0323,239.3277,7.7202
1000,16x32x16,32,262144,47185.9200,6291.4560,34603.0080,26.4047,9.9279,241.8432,7.8014
1000,16x32x32,32,524288,94371.8400,12582.9120,69206.0160,53.7802,9.7487,246.2890,7.9448
1000,16x32x64,32,1048576,188743.6800,25165.8240,138412.0320,107.4784,9.7562,246.1009,7.9387
1000,16x64x2,32,65536,11796.4800,1572.8640,8650.7520,6.3794,10.2731,233.7173,7.5393
1000,16x64x4,32,131072,23592.9600,3145.7280,17301.5040,13.0662,10.0314,239.3486,7.7209
1000,16x64x8,32,262144,47185.9200,6291.4560,34603.0080,26.7841,9.7873,245.3178,7.9135
1000,16x64x16,32,524288,94371.8400,12582.9120,69206.0160,53.3277,9.8314,244.2168,7.8780
1000,16x64x32,32,1048576,188743.6800,25165.8240,138412.0320,107.9683,9.7119,247.2227,7.9749
1000,16x64x64,32,2097152,377487.3600,50331.6480,276824.0640,214.3031,9.7859,245.3527,7.9146
1000,32x4x2,32,8192,1474.5600,196.6080,1081.3440,0.7426,11.0309,217.6612,7.0213
1000,32x4x4,32,16384,2949.1200,393.2160,2162.6880,1.4894,11.0005,218.2626,7.0407
1000,32x4x8,32,32768,5898.2400,786.4320,4325.3760,2.9852,10.9767,218.7364,7.0560
1000,32x4x16,32,65536,11796.4800,1572.8640,8650.7520,6.3911,10.2542,234.1482,7.5532
1000,32x4x32,32,131072,23592.9600,3145.7280,17301.5040,13.0791,10.0215,239.5854,7.7286
1000,32x4x64,32,262144,47185.9200,6291.4560,34603.0080,26.4275,9.9194,242.0518,7.8081
1000,32x8x2,32,16384,2949.1200,393.2160,2162.6880,1.4820,11.0557,217.1738,7.0056
1000,32x8x4,32,32768,5898.2400,786.4320,4325.3760,2.9772,11.0063,218.1474,7.0370
1000,32x8x8,32,65536,11796.4800,1572.8640,8650.7520,6.3678,10.2917,233.2947,7.5256
1000,32x8x16,32,131072,23592.9600,3145.7280,17301.5040,13.0689,10.0293,239.3989,7.7225
1000,32x8x32,32,262144,47185.9200,6291.4560,34603.0080,26.4323,9.9176,242.0958,7.8095
1000,32x8x64,32,524288,94371.8400,12582.9120,69206.0160,53.2360,9.8484,243.7967,7.8644
1000,32x16x2,32,32768,5898.2400,786.4320,4325.3760,2.9766,11.0085,218.1046,7.0356
1000,32x16x4,32,65536,11796.4800,1572.8640,8650.7520,6.3747,10.2807,233.5454,7.5337
1000,32x16x8,32,131072,23592.9600,3145.7280,17301.5040,13.4372,9.7544,246.1454,7.9402
1000,32x16x16,32,262144,47185.9200,6291.4560,34603.0080,26.4996,9.8924,242.7121,7.8294
1000,32x16x32,32,524288,94371.8400,12582.9120,69206.0160,54.7582,9.5746,250.7676,8.0893
1000,32x16x64,32,1048576,188743.6800,25165.8240,138412.0320,108.2222,9.6891,247.8041,7.9937
1000,32x32x2,32,65536,11796.4800,1572.8640,8650.7520,6.3767,10.2773,233.6206,7.5361
1000,32x32x4,32,131072,23592.9600,3145.7280,17301.5040,13.7663,9.5212,252.1739,8.1346
1000,32x32x8,32,262144,47185.9200,6291.4560,34603.0080,26.4448,9.9129,242.2106,7.8132
1000,32x32x16,32,524288,94371.8400,12582.9120,69206.0160,53.3462,9.8280,244.3011,7.8807
1000,32x32x32,32,1048576,188743.6800,25165.8240,138412.0320,108.0679,9.7029,247.4509,7.9823
1000,32x32x64,32,2097152,377487.3600,50331.6480,276824.0640,215.9838,9.7098,247.2768,7.9767
1000,32x64x2,32,131072,23592.9600,3145.7280,17301.5040,13.0597,10.0364,239.2292,7.7171
1000,32x64x4,32,262144,47185.9200,6291.4560,34603.0080,26.3953,9.9315,241.7572,7.7986
1000,32x64x8,32,524288,94371.8400,12582.9120,69206.0160,53.1987,9.8553,243.6259,7.8589
1000,32x64x16,32,1048576,188743.6800,25165.8240,138412.0320,106.6929,9.8280,244.3024,7.8807
1000,32x64x32,32,2097152,377487.3600,50331.6480,276824.0640,214.7515,9.7655,245.8660,7.9312
1000,32x64x64,32,4194304,754974.7200,100663.2960,553648.1280,431.4337,-0.2333,-10290.4677,-331.9506
1000,64x4x2,32,16384,2949.1200,393.2160,2162.6880,1.4879,11.0118,218.0398,7.0335
1000,64x4x4,32,32768,5898.2400,786.4320,4325.3760,2.9847,10.9788,218.6950,7.0547
1000,64x4x8,32,65536,11796.4800,1572.8640,8650.7520,6.3789,10.2739,233.6995,7.5387
1000,64x4x16,32,131072,23592.9600,3145.7280,17301.5040,13.0762,10.0237,239.5328,7.7269
1000,64x4x32,32,262144,47185.9200,6291.4560,34603.0080,26.4055,9.9276,241.8499,7.8016
1000,64x4x64,32,524288,94371.8400,12582.9120,69206.0160,53.9779,9.7130,247.1940,7.9740
1000,64x8x2,32,32768,5898.2400,786.4320,4325.3760,2.9891,10.9623,219.0227,7.0652
1000,64x8x4,32,65536,11796.4800,1572.8640,8650.7520,6.3605,10.3036,233.0252,7.5169
1000,64x8x8,32,131072,23592.9600,3145.7280,17301.5040,13.0487,10.0449,239.0275,7.7106
1000,64x8x16,32,262144,47185.9200,6291.4560,34603.0080,26.4589,9.9076,242.3398,7.8174
1000,64x8x32,32,524288,94371.8400,12582.9120,69206.0160,53.3738,9.8229,244.4278,7.8848
1000,64x8x64,32,1048576,188743.6800,25165.8240,138412.0320,106.7776,9.8202,244.4963,7.8870
1000,64x16x2,32,65536,11796.4800,1572.8640,8650.7520,6.3749,10.2803,233.5543,7.5340
1000,64x16x4,32,131072,23592.9600,3145.7280,17301.5040,13.1027,10.0035,240.0170,7.7425
1000,64x16x8,32,262144,47185.9200,6291.4560,34603.0080,26.4519,9.9102,242.2757,7.8153
1000,64x16x16,32,524288,94371.8400,12582.9120,69206.0160,53.8700,9.7325,246.7002,7.9581
1000,64x16x32,32,1048576,188743.6800,25165.8240,138412.0320,107.7415,9.7323,246.7034,7.9582
1000,64x16x64,32,2097152,377487.3600,50331.6480,276824.0640,213.3675,9.8288,244.2815,7.8800
1000,64x32x2,32,131072,23592.9600,3145.7280,17301.5040,13.0621,10.0345,239.2736,7.7185
1000,64x32x4,32,262144,47185.9200,6291.4560,34603.0080,26.3724,9.9401,241.5467,7.7918
1000,64x32x8,32,524288,94371.8400,12582.9120,69206.0160,53.1027,9.8731,243.1863,7.8447
1000,64x32x16,32,1048576,188743.6800,25165.8240,138412.0320,106.7040,9.8270,244.3278,7.8815
1000,64x32x32,32,2097152,377487.3600,50331.6480,276824.0640,213.9692,9.8012,244.9703,7.9023
1000,64x32x64,32,4194304,754974.7200,100663.2960,553648.1280,428.8920,-0.2347,-10229.8417,-329.9949
1000,64x64x2,32,262144,47185.9200,6291.4560,34603.0080,27.0668,9.6851,247.9069,7.9970
1000,64x64x4,32,524288,94371.8400,12582.9120,69206.0160,54.7363,9.5784,250.6672,8.0860
1000,64x64x8,32,1048576,188743.6800,25165.8240,138412.0320,108.3671,9.6762,248.1359,8.0044
1000,64x64x16,32,2097152,377487.3600,50331.6480,276824.0640,216.9378,9.6671,248.3690,8.0119
1000,64x64x32,32,4194304,754974.7200,100663.2960,553648.1280,433.8169,-0.2320,-10347.3116,-333.7842
1000,64x64x64,32,8388608,1509949.4400,201326.5920,1107296.2560,878.1040,-0.2293,-10472.1773,-337.8122

107
scripts/results_soa_broadep2_iln1000.txt
View File

@ -1,107 +0,0 @@
200,4x4x2,8,256,21.5040,6.1440,9.2160,2.1353,23.9782,95.9620,13.7089
200,4x4x4,8,512,43.0080,12.2880,18.4320,4.3181,23.7144,97.0298,13.8614
200,4x4x8,8,1024,86.0160,24.5760,36.8640,8.5109,24.0632,95.6231,13.6604
200,4x4x16,8,2048,172.0320,49.1520,73.7280,17.0205,24.0651,95.6158,13.6594
200,4x8x2,8,512,43.0080,12.2880,18.4320,4.2564,24.0579,95.6442,13.6635
200,4x8x4,8,1024,86.0160,24.5760,36.8640,8.6350,23.7174,97.0173,13.8596
200,4x8x8,8,2048,172.0320,49.1520,73.7280,17.0207,24.0649,95.6166,13.6595
200,4x8x16,8,4096,344.0640,98.3040,147.4560,34.6027,23.6744,97.1935,13.8848
200,4x16x2,8,1024,86.0160,24.5760,36.8640,8.5135,24.0559,95.6524,13.6646
200,4x16x4,8,2048,172.0320,49.1520,73.7280,17.0300,24.0517,95.6689,13.6670
200,4x16x8,8,4096,344.0640,98.3040,147.4560,34.4582,23.7737,96.7876,13.8268
200,4x16x16,8,8192,688.1280,196.6080,294.9120,68.3538,23.9694,95.9974,13.7139
200,8x4x2,8,512,43.0080,12.2880,18.4320,4.2576,24.0512,95.6708,13.6673
200,8x4x4,8,1024,86.0160,24.5760,36.8640,8.5643,23.9133,96.2224,13.7461
200,8x4x8,8,2048,172.0320,49.1520,73.7280,17.0216,24.0636,95.6216,13.6602
200,8x4x16,8,4096,344.0640,98.3040,147.4560,34.3277,23.8641,96.4210,13.7744
200,8x8x2,8,1024,86.0160,24.5760,36.8640,8.5150,24.0516,95.6693,13.6670
200,8x8x4,8,2048,172.0320,49.1520,73.7280,17.0235,24.0608,95.6327,13.6618
200,8x8x8,8,4096,344.0640,98.3040,147.4560,34.0518,24.0575,95.6459,13.6637
200,8x8x16,8,8192,688.1280,196.6080,294.9120,68.0984,24.0593,95.6386,13.6627
200,8x16x2,8,2048,172.0320,49.1520,73.7280,17.0230,24.0616,95.6297,13.6614
200,8x16x4,8,4096,344.0640,98.3040,147.4560,34.0451,24.0622,95.6273,13.6610
200,8x16x8,8,8192,688.1280,196.6080,294.9120,68.1014,24.0583,95.6428,13.6633
200,8x16x16,8,16384,1376.2560,393.2160,589.8240,136.4647,24.0121,95.8268,13.6895
200,16x4x2,8,1024,86.0160,24.5760,36.8640,8.6901,23.5671,97.6360,13.9480
200,16x4x4,8,2048,172.0320,49.1520,73.7280,17.4044,23.5343,97.7723,13.9675
200,16x4x8,8,4096,344.0640,98.3040,147.4560,34.0569,24.0539,95.6603,13.6658
200,16x4x16,8,8192,688.1280,196.6080,294.9120,68.7236,23.8404,96.5168,13.7881
200,16x8x2,8,2048,172.0320,49.1520,73.7280,17.1338,23.9059,96.2523,13.7503
200,16x8x4,8,4096,344.0640,98.3040,147.4560,34.2669,23.9064,96.2502,13.7500
200,16x8x8,8,8192,688.1280,196.6080,294.9120,70.0776,23.3798,98.4182,14.0597
200,16x8x16,8,16384,1376.2560,393.2160,589.8240,136.2386,24.0519,95.6680,13.6669
200,16x16x2,8,4096,344.0640,98.3040,147.4560,34.0482,24.0600,95.6358,13.6623
200,16x16x4,8,8192,688.1280,196.6080,294.9120,68.6491,23.8663,96.4121,13.7732
200,16x16x8,8,16384,1376.2560,393.2160,589.8240,137.8710,23.7671,96.8144,13.8306
200,16x16x16,8,32768,2752.5120,786.4320,1179.6480,277.3699,23.6277,97.3859,13.9123
200,4x4x2,16,512,59.3920,12.2880,34.8160,7.3151,13.9984,164.3763,10.9584
200,4x4x4,16,1024,118.7840,24.5760,69.6320,14.4104,14.2120,161.9059,10.7937
200,4x4x8,16,2048,237.5680,49.1520,139.2640,29.7442,13.7708,167.0931,11.1395
200,4x4x16,16,4096,475.1360,98.3040,278.5280,57.9628,14.1332,162.8081,10.8539
200,4x8x2,16,1024,118.7840,24.5760,69.6320,14.4086,14.2137,161.8861,10.7924
200,4x8x4,16,2048,237.5680,49.1520,139.2640,29.2763,13.9908,164.4650,10.9643
200,4x8x8,16,4096,475.1360,98.3040,278.5280,57.6454,14.2110,161.9167,10.7944
200,4x8x16,16,8192,950.2720,196.6080,557.0560,117.4588,13.9487,164.9614,10.9974
200,4x16x2,16,2048,237.5680,49.1520,139.2640,28.8261,14.2094,161.9356,10.7957
200,4x16x4,16,4096,475.1360,98.3040,278.5280,58.6978,13.9562,164.8725,10.9915
200,4x16x8,16,8192,950.2720,196.6080,557.0560,115.3036,14.2094,161.9346,10.7956
200,4x16x16,16,16384,1900.5440,393.2160,1114.1120,234.2918,13.9860,164.5219,10.9681
200,8x4x2,16,1024,118.7840,24.5760,69.6320,14.4086,14.2137,161.8855,10.7924
200,8x4x4,16,2048,237.5680,49.1520,139.2640,28.8230,14.2109,161.9184,10.7946
200,8x4x8,16,4096,475.1360,98.3040,278.5280,59.1780,13.8430,166.2214,11.0814
200,8x4x16,16,8192,950.2720,196.6080,557.0560,117.4585,13.9488,164.9609,10.9974
200,8x8x2,16,2048,237.5680,49.1520,139.2640,28.8173,14.2137,161.8863,10.7924
200,8x8x4,16,4096,475.1360,98.3040,278.5280,58.4082,14.0254,164.0591,10.9373
200,8x8x8,16,8192,950.2720,196.6080,557.0560,115.2800,14.2123,161.9015,10.7934
200,8x8x16,16,16384,1900.5440,393.2160,1114.1120,232.9699,14.0653,163.5937,10.9062
200,8x16x2,16,4096,475.1360,98.3040,278.5280,59.3376,13.8058,166.6696,11.1113
200,8x16x4,16,8192,950.2720,196.6080,557.0560,115.2970,14.2103,161.9253,10.7950
200,8x16x8,16,16384,1900.5440,393.2160,1114.1120,240.4558,13.6275,168.8503,11.2567
200,8x16x16,16,32768,3801.0880,786.4320,2228.2240,462.5401,14.1687,162.4000,10.8267
200,16x4x2,16,2048,237.5680,49.1520,139.2640,28.8184,14.2131,161.8927,10.7928
200,16x4x4,16,4096,475.1360,98.3040,278.5280,57.9551,14.1351,162.7864,10.8524
200,16x4x8,16,8192,950.2720,196.6080,557.0560,115.7986,14.1487,162.6298,10.8420
200,16x4x16,16,16384,1900.5440,393.2160,1114.1120,236.1090,13.8783,165.7980,11.0532
200,16x8x2,16,4096,475.1360,98.3040,278.5280,59.4157,13.7876,166.8891,11.1259
200,16x8x4,16,8192,950.2720,196.6080,557.0560,117.0833,13.9935,164.4339,10.9623
200,16x8x8,16,16384,1900.5440,393.2160,1114.1120,231.0443,14.1826,162.2415,10.8161
200,16x8x16,16,32768,3801.0880,786.4320,2228.2240,461.2008,14.2099,161.9298,10.7953
200,16x16x2,16,8192,950.2720,196.6080,557.0560,118.1060,13.8723,165.8703,11.0580
200,16x16x4,16,16384,1900.5440,393.2160,1114.1120,231.5209,14.1534,162.5762,10.8384
200,16x16x8,16,32768,3801.0880,786.4320,2228.2240,466.6747,14.0432,163.8517,10.9234
200,16x16x16,16,65536,7602.1760,1572.8640,4456.4480,938.5839,13.9649,164.7706,10.9847
200,4x4x2,32,1024,184.3200,24.5760,135.1680,27.1988,7.5298,305.5877,9.8577
200,4x4x4,32,2048,368.6400,49.1520,270.3360,51.8929,7.8932,291.5174,9.4038
200,4x4x8,32,4096,737.2800,98.3040,540.6720,104.9857,7.8030,294.8878,9.5125
200,4x4x16,32,8192,1474.5600,196.6080,1081.3440,210.0029,7.8018,294.9320,9.5139
200,4x8x2,32,2048,368.6400,49.1520,270.3360,51.8941,7.8930,291.5240,9.4040
200,4x8x4,32,4096,737.2800,98.3040,540.6720,104.9772,7.8036,294.8640,9.5117
200,4x8x8,32,8192,1474.5600,196.6080,1081.3440,212.3359,7.7161,298.2085,9.6196
200,4x8x16,32,16384,2949.1200,393.2160,2162.6880,415.1714,7.8926,291.5373,9.4044
200,4x16x2,32,4096,737.2800,98.3040,540.6720,107.1892,7.6426,301.0771,9.7122
200,4x16x4,32,8192,1474.5600,196.6080,1081.3440,213.7067,7.6666,300.1338,9.6817
200,4x16x8,32,16384,2949.1200,393.2160,2162.6880,429.5336,7.6287,301.6226,9.7298
200,4x16x16,32,32768,5898.2400,786.4320,4325.3760,835.8309,7.8408,293.4642,9.4666
200,8x4x2,32,2048,368.6400,49.1520,270.3360,53.2960,7.6854,299.3996,9.6581
200,8x4x4,32,4096,737.2800,98.3040,540.6720,106.0529,7.7244,297.8853,9.6092
200,8x4x8,32,8192,1474.5600,196.6080,1081.3440,213.3924,7.6779,299.6924,9.6675
200,8x4x16,32,16384,2949.1200,393.2160,2162.6880,416.1584,7.8739,292.2304,9.4268
200,8x8x2,32,4096,737.2800,98.3040,540.6720,104.1101,7.8686,292.4283,9.4332
200,8x8x4,32,8192,1474.5600,196.6080,1081.3440,211.0920,7.7615,296.4616,9.5633
200,8x8x8,32,16384,2949.1200,393.2160,2162.6880,421.6304,7.7717,296.0729,9.5507
200,8x8x16,32,32768,5898.2400,786.4320,4325.3760,837.8503,7.8219,294.1732,9.4895
200,8x16x2,32,8192,1474.5600,196.6080,1081.3440,208.5033,7.8579,292.8260,9.4460
200,8x16x4,32,16384,2949.1200,393.2160,2162.6880,420.4673,7.7932,295.2561,9.5244
200,8x16x8,32,32768,5898.2400,786.4320,4325.3760,840.7054,7.7954,295.1757,9.5218
200,8x16x16,32,65536,11796.4800,1572.8640,8650.7520,1661.9291,7.8867,291.7556,9.4115
200,16x4x2,32,4096,737.2800,98.3040,540.6720,105.6303,7.7554,296.6983,9.5709
200,16x4x4,32,8192,1474.5600,196.6080,1081.3440,207.5775,7.8930,291.5258,9.4041
200,16x4x8,32,16384,2949.1200,393.2160,2162.6880,420.5695,7.7913,295.3279,9.5267
200,16x4x16,32,32768,5898.2400,786.4320,4325.3760,833.0888,7.8666,292.5014,9.4355
200,16x8x2,32,8192,1474.5600,196.6080,1081.3440,210.1080,7.7979,295.0796,9.5187
200,16x8x4,32,16384,2949.1200,393.2160,2162.6880,428.5183,7.6468,300.9096,9.7068
200,16x8x8,32,32768,5898.2400,786.4320,4325.3760,845.6674,7.7496,296.9178,9.5780
200,16x8x16,32,65536,11796.4800,1572.8640,8650.7520,1706.5013,7.6807,299.5803,9.6639
200,16x16x2,32,16384,2949.1200,393.2160,2162.6880,416.8327,7.8612,292.7039,9.4421
200,16x16x4,32,32768,5898.2400,786.4320,4325.3760,844.7536,7.7580,296.5970,9.5676
200,16x16x8,32,65536,11796.4800,1572.8640,8650.7520,1706.9248,7.6788,299.6547,9.6663

450
scripts/results_soa_casclakesp2.txt
View File

@ -1,450 +0,0 @@
200,4x4x2,8,256,21.5040,6.1440,9.2160,0.0017,30.1968,79.4787,11.3541
200,4x4x4,8,512,43.0080,12.2880,18.4320,0.0034,30.0857,79.7722,11.3960
200,4x4x8,8,1024,86.0160,24.5760,36.8640,0.0071,29.0411,82.6414,11.8059
200,4x4x16,8,2048,172.0320,49.1520,73.7280,0.0142,28.9227,82.9799,11.8543
200,4x4x32,8,4096,344.0640,98.3040,147.4560,0.0281,29.1749,82.2624,11.7518
200,4x4x64,8,8192,688.1280,196.6080,294.9120,0.0587,27.9148,85.9760,12.2823
200,4x8x2,8,512,43.0080,12.2880,18.4320,0.0035,29.3791,81.6907,11.6701
200,4x8x4,8,1024,86.0160,24.5760,36.8640,0.0073,28.1253,85.3324,12.1903
200,4x8x8,8,2048,172.0320,49.1520,73.7280,0.0142,28.9410,82.9273,11.8468
200,4x8x16,8,4096,344.0640,98.3040,147.4560,0.0287,28.5778,83.9811,11.9973
200,4x8x32,8,8192,688.1280,196.6080,294.9120,0.0587,27.9348,85.9143,12.2735
200,4x8x64,8,16384,1376.2560,393.2160,589.8240,0.1193,27.4757,87.3498,12.4785
200,4x16x2,8,1024,86.0160,24.5760,36.8640,0.0069,29.6841,80.8515,11.5502
200,4x16x4,8,2048,172.0320,49.1520,73.7280,0.0140,29.1786,82.2521,11.7503
200,4x16x8,8,4096,344.0640,98.3040,147.4560,0.0281,29.1882,82.2250,11.7464
200,4x16x16,8,8192,688.1280,196.6080,294.9120,0.0601,27.2622,88.0339,12.5763
200,4x16x32,8,16384,1376.2560,393.2160,589.8240,0.1194,27.4458,87.4450,12.4921
200,4x16x64,8,32768,2752.5120,786.4320,1179.6480,0.2367,27.6917,86.6685,12.3812
200,4x32x2,8,2048,172.0320,49.1520,73.7280,0.0144,28.4002,84.5065,12.0724
200,4x32x4,8,4096,344.0640,98.3040,147.4560,0.0281,29.2045,82.1792,11.7399
200,4x32x8,8,8192,688.1280,196.6080,294.9120,0.0578,28.3343,84.7028,12.1004
200,4x32x16,8,16384,1376.2560,393.2160,589.8240,0.1187,27.6149,86.9095,12.4156
200,4x32x32,8,32768,2752.5120,786.4320,1179.6480,0.2353,27.8557,86.1585,12.3084
200,4x32x64,8,65536,5505.0240,1572.8640,2359.2960,0.5017,26.1265,91.8606,13.1229
200,4x64x2,8,4096,344.0640,98.3040,147.4560,0.0286,28.6918,83.6476,11.9497
200,4x64x4,8,8192,688.1280,196.6080,294.9120,0.0583,28.1218,85.3430,12.1919
200,4x64x8,8,16384,1376.2560,393.2160,589.8240,0.1196,27.3912,87.6195,12.5171
200,4x64x16,8,32768,2752.5120,786.4320,1179.6480,0.2342,27.9881,85.7507,12.2501
200,4x64x32,8,65536,5505.0240,1572.8640,2359.2960,0.4861,26.9639,89.0080,12.7154
200,4x64x64,8,131072,11010.0480,3145.7280,4718.5920,1.2546,20.8949,114.8605,16.4086
200,8x4x2,8,512,43.0080,12.2880,18.4320,0.0035,29.1498,82.3334,11.7619
200,8x4x4,8,1024,86.0160,24.5760,36.8640,0.0070,29.0834,82.5213,11.7888
200,8x4x8,8,2048,172.0320,49.1520,73.7280,0.0139,29.4496,81.4952,11.6422
200,8x4x16,8,4096,344.0640,98.3040,147.4560,0.0281,29.1828,82.2403,11.7486
200,8x4x32,8,8192,688.1280,196.6080,294.9120,0.0593,27.6198,86.8941,12.4134
200,8x4x64,8,16384,1376.2560,393.2160,589.8240,0.1177,27.8367,86.2171,12.3167
200,8x8x2,8,1024,86.0160,24.5760,36.8640,0.0072,28.5962,83.9274,11.9896
200,8x8x4,8,2048,172.0320,49.1520,73.7280,0.0141,29.0048,82.7449,11.8207
200,8x8x8,8,4096,344.0640,98.3040,147.4560,0.0282,29.1001,82.4738,11.7820
200,8x8x16,8,8192,688.1280,196.6080,294.9120,0.0588,27.8416,86.2020,12.3146
200,8x8x32,8,16384,1376.2560,393.2160,589.8240,0.1175,27.8776,86.0907,12.2987
200,8x8x64,8,32768,2752.5120,786.4320,1179.6480,0.2373,27.6184,86.8985,12.4141
200,8x16x2,8,2048,172.0320,49.1520,73.7280,0.0138,29.6202,81.0259,11.5751
200,8x16x4,8,4096,344.0640,98.3040,147.4560,0.0282,29.0022,82.7524,11.8218
200,8x16x8,8,8192,688.1280,196.6080,294.9120,0.0590,27.7595,86.4568,12.3510
200,8x16x16,8,16384,1376.2560,393.2160,589.8240,0.1195,27.4101,87.5590,12.5084
200,8x16x32,8,32768,2752.5120,786.4320,1179.6480,0.2345,27.9510,85.8646,12.2664
200,8x16x64,8,65536,5505.0240,1572.8640,2359.2960,0.4899,26.7532,89.7089,12.8156
200,8x32x2,8,4096,344.0640,98.3040,147.4560,0.0281,29.1831,82.2394,11.7485
200,8x32x4,8,8192,688.1280,196.6080,294.9120,0.0591,27.7320,86.5425,12.3632
200,8x32x8,8,16384,1376.2560,393.2160,589.8240,0.1190,27.5256,87.1916,12.4559
200,8x32x16,8,32768,2752.5120,786.4320,1179.6480,0.2357,27.8106,86.2980,12.3283
200,8x32x32,8,65536,5505.0240,1572.8640,2359.2960,0.4782,27.4087,87.5635,12.5091
200,8x32x64,8,131072,11010.0480,3145.7280,4718.5920,1.2611,20.7865,115.4596,16.4942
200,8x64x2,8,8192,688.1280,196.6080,294.9120,0.0598,27.4201,87.5270,12.5039
200,8x64x4,8,16384,1376.2560,393.2160,589.8240,0.1188,27.5857,87.0017,12.4288
200,8x64x8,8,32768,2752.5120,786.4320,1179.6480,0.2378,27.5646,87.0682,12.4383
200,8x64x16,8,65536,5505.0240,1572.8640,2359.2960,0.4982,26.3088,91.2243,13.0320
200,8x64x32,8,131072,11010.0480,3145.7280,4718.5920,1.2420,21.1071,113.7059,16.2437
200,8x64x64,8,262144,22020.0960,6291.4560,9437.1840,3.2871,15.9501,150.4694,21.4956
200,16x4x2,8,1024,86.0160,24.5760,36.8640,0.0071,28.9960,82.7700,11.8243
200,16x4x4,8,2048,172.0320,49.1520,73.7280,0.0141,29.0035,82.7486,11.8212
200,16x4x8,8,4096,344.0640,98.3040,147.4560,0.0281,29.1943,82.2077,11.7440
200,16x4x16,8,8192,688.1280,196.6080,294.9120,0.0580,28.2245,85.0324,12.1475
200,16x4x32,8,16384,1376.2560,393.2160,589.8240,0.1190,27.5459,87.1274,12.4468
200,16x4x64,8,32768,2752.5120,786.4320,1179.6480,0.2315,28.3093,84.7778,12.1111
200,16x8x2,8,2048,172.0320,49.1520,73.7280,0.0138,29.6944,80.8234,11.5462
200,16x8x4,8,4096,344.0640,98.3040,147.4560,0.0283,28.9408,82.9278,11.8468
200,16x8x8,8,8192,688.1280,196.6080,294.9120,0.0588,27.8805,86.0817,12.2974
200,16x8x16,8,16384,1376.2560,393.2160,589.8240,0.1164,28.1587,85.2312,12.1759
200,16x8x32,8,32768,2752.5120,786.4320,1179.6480,0.2351,27.8775,86.0909,12.2987
200,16x8x64,8,65536,5505.0240,1572.8640,2359.2960,0.4827,27.1555,88.3799,12.6257
200,16x16x2,8,4096,344.0640,98.3040,147.4560,0.0278,29.5090,81.3312,11.6187
200,16x16x4,8,8192,688.1280,196.6080,294.9120,0.0582,28.1319,85.3123,12.1875
200,16x16x8,8,16384,1376.2560,393.2160,589.8240,0.1171,27.9783,85.7808,12.2544
200,16x16x16,8,32768,2752.5120,786.4320,1179.6480,0.2342,27.9800,85.7755,12.2536
200,16x16x32,8,65536,5505.0240,1572.8640,2359.2960,0.5129,25.5535,93.9207,13.4172
200,16x16x64,8,131072,11010.0480,3145.7280,4718.5920,1.2518,20.9419,114.6028,16.3718
200,16x32x2,8,8192,688.1280,196.6080,294.9120,0.0598,27.3917,87.6179,12.5168
200,16x32x4,8,16384,1376.2560,393.2160,589.8240,0.1198,27.3519,87.7454,12.5351
200,16x32x8,8,32768,2752.5120,786.4320,1179.6480,0.2359,27.7824,86.3856,12.3408
200,16x32x16,8,65536,5505.0240,1572.8640,2359.2960,0.4728,27.7198,86.5806,12.3687
200,16x32x32,8,131072,11010.0480,3145.7280,4718.5920,1.2741,20.5744,116.6497,16.6642
200,16x32x64,8,262144,22020.0960,6291.4560,9437.1840,3.2988,15.8931,151.0087,21.5727
200,16x64x2,8,16384,1376.2560,393.2160,589.8240,0.1198,27.3624,87.7117,12.5302
200,16x64x4,8,32768,2752.5120,786.4320,1179.6480,0.2345,27.9526,85.8595,12.2656
200,16x64x8,8,65536,5505.0240,1572.8640,2359.2960,0.4795,27.3355,87.7979,12.5426
200,16x64x16,8,131072,11010.0480,3145.7280,4718.5920,1.2697,20.6461,116.2446,16.6064
200,16x64x32,8,262144,22020.0960,6291.4560,9437.1840,3.2978,15.8981,150.9610,21.5659
200,16x64x64,8,524288,44040.1920,12582.9120,18874.3680,7.2019,14.5598,164.8376,23.5482
200,32x4x2,8,2048,172.0320,49.1520,73.7280,0.0143,28.7329,83.5278,11.9325
200,32x4x4,8,4096,344.0640,98.3040,147.4560,0.0284,28.8290,83.2496,11.8928
200,32x4x8,8,8192,688.1280,196.6080,294.9120,0.0581,28.2211,85.0427,12.1490
200,32x4x16,8,16384,1376.2560,393.2160,589.8240,0.1172,27.9499,85.8678,12.2668
200,32x4x32,8,32768,2752.5120,786.4320,1179.6480,0.2343,27.9743,85.7930,12.2561
200,32x4x64,8,65536,5505.0240,1572.8640,2359.2960,0.4899,26.7521,89.7124,12.8161
200,32x8x2,8,4096,344.0640,98.3040,147.4560,0.0286,28.6625,83.7332,11.9619
200,32x8x4,8,8192,688.1280,196.6080,294.9120,0.0578,28.3611,84.6231,12.0890
200,32x8x8,8,16384,1376.2560,393.2160,589.8240,0.1186,27.6276,86.8697,12.4100
200,32x8x16,8,32768,2752.5120,786.4320,1179.6480,0.2345,27.9497,85.8684,12.2669
200,32x8x32,8,65536,5505.0240,1572.8640,2359.2960,0.4789,27.3714,87.6827,12.5261
200,32x8x64,8,131072,11010.0480,3145.7280,4718.5920,1.2396,21.1468,113.4925,16.2132
200,32x16x2,8,8192,688.1280,196.6080,294.9120,0.0587,27.9346,85.9149,12.2736
200,32x16x4,8,16384,1376.2560,393.2160,589.8240,0.1190,27.5368,87.1562,12.4509
200,32x16x8,8,32768,2752.5120,786.4320,1179.6480,0.2349,27.9022,86.0146,12.2878
200,32x16x16,8,65536,5505.0240,1572.8640,2359.2960,0.4823,27.1772,88.3092,12.6156
200,32x16x32,8,131072,11010.0480,3145.7280,4718.5920,1.2440,21.0724,113.8929,16.2704
200,32x16x64,8,262144,22020.0960,6291.4560,9437.1840,3.2937,15.9178,150.7744,21.5392
200,32x32x2,8,16384,1376.2560,393.2160,589.8240,0.1180,27.7590,86.4583,12.3512
200,32x32x4,8,32768,2752.5120,786.4320,1179.6480,0.2350,27.8876,86.0597,12.2942
200,32x32x8,8,65536,5505.0240,1572.8640,2359.2960,0.4704,27.8658,86.1270,12.3039
200,32x32x16,8,131072,11010.0480,3145.7280,4718.5920,1.2534,20.9146,114.7524,16.3932
200,32x32x32,8,262144,22020.0960,6291.4560,9437.1840,3.2411,16.1762,148.3658,21.1951
200,32x32x64,8,524288,44040.1920,12582.9120,18874.3680,7.2101,14.5431,165.0270,23.5753
200,32x64x2,8,32768,2752.5120,786.4320,1179.6480,0.2339,28.0142,85.6710,12.2387
200,32x64x4,8,65536,5505.0240,1572.8640,2359.2960,0.4701,27.8800,86.0834,12.2976
200,32x64x8,8,131072,11010.0480,3145.7280,4718.5920,1.2565,20.8636,115.0326,16.4332
200,32x64x16,8,262144,22020.0960,6291.4560,9437.1840,3.3016,15.8796,151.1373,21.5910
200,32x64x32,8,524288,44040.1920,12582.9120,18874.3680,7.1664,14.6318,164.0262,23.4323
200,32x64x64,8,1048576,88080.3840,25165.8240,37748.7360,14.7563,14.2119,168.8722,24.1246
200,64x4x2,8,4096,344.0640,98.3040,147.4560,0.0284,28.8520,83.1830,11.8833
200,64x4x4,8,8192,688.1280,196.6080,294.9120,0.0599,27.3595,87.7210,12.5316
200,64x4x8,8,16384,1376.2560,393.2160,589.8240,0.1155,28.3599,84.6266,12.0895
200,64x4x16,8,32768,2752.5120,786.4320,1179.6480,0.2271,28.8639,83.1489,11.8784
200,64x4x32,8,65536,5505.0240,1572.8640,2359.2960,0.4784,27.3953,87.6063,12.5152
200,64x4x64,8,131072,11010.0480,3145.7280,4718.5920,1.2466,21.0279,114.1338,16.3048
200,64x8x2,8,8192,688.1280,196.6080,294.9120,0.0595,27.5465,87.1256,12.4465
200,64x8x4,8,16384,1376.2560,393.2160,589.8240,0.1201,27.2917,87.9388,12.5627
200,64x8x8,8,32768,2752.5120,786.4320,1179.6480,0.2349,27.8954,86.0358,12.2908
200,64x8x16,8,65536,5505.0240,1572.8640,2359.2960,0.4647,28.2078,85.0829,12.1547
200,64x8x32,8,131072,11010.0480,3145.7280,4718.5920,1.2417,21.1119,113.6802,16.2400
200,64x8x64,8,262144,22020.0960,6291.4560,9437.1840,3.2842,15.9640,150.3379,21.4768
200,64x16x2,8,16384,1376.2560,393.2160,589.8240,0.1161,28.2349,85.0010,12.1430
200,64x16x4,8,32768,2752.5120,786.4320,1179.6480,0.2347,27.9189,85.9634,12.2805
200,64x16x8,8,65536,5505.0240,1572.8640,2359.2960,0.5039,26.0132,92.2608,13.1801
200,64x16x16,8,131072,11010.0480,3145.7280,4718.5920,1.2615,20.7799,115.4963,16.4995
200,64x16x32,8,262144,22020.0960,6291.4560,9437.1840,3.2610,16.0774,149.2775,21.3254
200,64x16x64,8,524288,44040.1920,12582.9120,18874.3680,7.1767,14.6108,164.2617,23.4660
200,64x32x2,8,32768,2752.5120,786.4320,1179.6480,0.2356,27.8138,86.2881,12.3269
200,64x32x4,8,65536,5505.0240,1572.8640,2359.2960,0.4749,27.5997,86.9576,12.4225
200,64x32x8,8,131072,11010.0480,3145.7280,4718.5920,1.2495,20.9806,114.3916,16.3417
200,64x32x16,8,262144,22020.0960,6291.4560,9437.1840,3.2536,16.1139,148.9395,21.2771
200,64x32x32,8,524288,44040.1920,12582.9120,18874.3680,7.2381,14.4870,165.6663,23.6666
200,64x32x64,8,1048576,88080.3840,25165.8240,37748.7360,14.6710,14.2945,167.8963,23.9852
200,64x64x2,8,65536,5505.0240,1572.8640,2359.2960,0.4799,27.3151,87.8633,12.5519
200,64x64x4,8,131072,11010.0480,3145.7280,4718.5920,1.2605,20.7972,115.3999,16.4857
200,64x64x8,8,262144,22020.0960,6291.4560,9437.1840,3.2808,15.9804,150.1841,21.4549
200,64x64x16,8,524288,44040.1920,12582.9120,18874.3680,7.1934,14.5769,164.6442,23.5206
200,64x64x32,8,1048576,88080.3840,25165.8240,37748.7360,14.6673,14.2981,167.8542,23.9792
200,64x64x64,8,2097152,176160.7680,50331.6480,75497.4720,29.2180,14.3552,167.1865,23.8838
200,4x4x2,16,512,59.3920,12.2880,34.8160,0.0056,18.2455,131.5393,8.7693
200,4x4x4,16,1024,118.7840,24.5760,69.6320,0.0113,18.1440,132.2754,8.8184
200,4x4x8,16,2048,237.5680,49.1520,139.2640,0.0227,18.0297,133.1133,8.8742
200,4x4x16,16,4096,475.1360,98.3040,278.5280,0.0448,18.2657,131.3938,8.7596
200,4x4x32,16,8192,950.2720,196.6080,557.0560,0.0911,17.9803,133.4791,8.8986
200,4x4x64,16,16384,1900.5440,393.2160,1114.1120,0.1844,17.7724,135.0411,9.0027
200,4x8x2,16,1024,118.7840,24.5760,69.6320,0.0113,18.1733,132.0615,8.8041
200,4x8x4,16,2048,237.5680,49.1520,139.2640,0.0225,18.2026,131.8491,8.7899
200,4x8x8,16,4096,475.1360,98.3040,278.5280,0.0456,17.9674,133.5754,8.9050
200,4x8x16,16,8192,950.2720,196.6080,557.0560,0.0923,17.7468,135.2353,9.0157
200,4x8x32,16,16384,1900.5440,393.2160,1114.1120,0.1850,17.7143,135.4836,9.0322
200,4x8x64,16,32768,3801.0880,786.4320,2228.2240,0.3665,17.8797,134.2308,8.9487
200,4x16x2,16,2048,237.5680,49.1520,139.2640,0.0233,17.5763,136.5478,9.1032
200,4x16x4,16,4096,475.1360,98.3040,278.5280,0.0450,18.2126,131.7769,8.7851
200,4x16x8,16,8192,950.2720,196.6080,557.0560,0.0911,17.9818,133.4684,8.8979
200,4x16x16,16,16384,1900.5440,393.2160,1114.1120,0.1868,17.5420,136.8145,9.1210
200,4x16x32,16,32768,3801.0880,786.4320,2228.2240,0.3681,17.8024,134.8134,8.9876
200,4x16x64,16,65536,7602.1760,1572.8640,4456.4480,0.7630,17.1780,139.7133,9.3142
200,4x32x2,16,4096,475.1360,98.3040,278.5280,0.0456,17.9525,133.6859,8.9124
200,4x32x4,16,8192,950.2720,196.6080,557.0560,0.0938,17.4605,137.4528,9.1635
200,4x32x8,16,16384,1900.5440,393.2160,1114.1120,0.1827,17.9368,133.8034,8.9202
200,4x32x16,16,32768,3801.0880,786.4320,2228.2240,0.3653,17.9395,133.7829,8.9189
200,4x32x32,16,65536,7602.1760,1572.8640,4456.4480,0.7845,16.7085,143.6395,9.5760
200,4x32x64,16,131072,15204.3520,3145.7280,8912.8960,1.8176,14.4226,166.4053,11.0937
200,4x64x2,16,8192,950.2720,196.6080,557.0560,0.0918,17.8519,134.4391,8.9626
200,4x64x4,16,16384,1900.5440,393.2160,1114.1120,0.1855,17.6617,135.8875,9.0592
200,4x64x8,16,32768,3801.0880,786.4320,2228.2240,0.3674,17.8396,134.5323,8.9688
200,4x64x16,16,65536,7602.1760,1572.8640,4456.4480,0.7768,16.8730,142.2392,9.4826
200,4x64x32,16,131072,15204.3520,3145.7280,8912.8960,1.8113,14.4725,165.8318,11.0555
200,4x64x64,16,262144,30408.7040,6291.4560,17825.7920,4.0660,12.8944,186.1278,12.4085
200,8x4x2,16,1024,118.7840,24.5760,69.6320,0.0113,18.1771,132.0345,8.8023
200,8x4x4,16,2048,237.5680,49.1520,139.2640,0.0223,18.3984,130.4461,8.6964
200,8x4x8,16,4096,475.1360,98.3040,278.5280,0.0452,18.1085,132.5347,8.8356
200,8x4x16,16,8192,950.2720,196.6080,557.0560,0.0923,17.7436,135.2601,9.0173
200,8x4x32,16,16384,1900.5440,393.2160,1114.1120,0.1851,17.7047,135.5574,9.0372
200,8x4x64,16,32768,3801.0880,786.4320,2228.2240,0.3667,17.8705,134.2992,8.9533
200,8x8x2,16,2048,237.5680,49.1520,139.2640,0.0227,18.0258,133.1425,8.8762
200,8x8x4,16,4096,475.1360,98.3040,278.5280,0.0454,18.0636,132.8640,8.8576
200,8x8x8,16,8192,950.2720,196.6080,557.0560,0.0916,17.8941,134.1221,8.9415
200,8x8x16,16,16384,1900.5440,393.2160,1114.1120,0.1851,17.7060,135.5472,9.0365
200,8x8x32,16,32768,3801.0880,786.4320,2228.2240,0.3608,18.1644,132.1268,8.8085
200,8x8x64,16,65536,7602.1760,1572.8640,4456.4480,0.7464,17.5607,136.6689,9.1113
200,8x16x2,16,4096,475.1360,98.3040,278.5280,0.0454,18.0396,133.0408,8.8694
200,8x16x4,16,8192,950.2720,196.6080,557.0560,0.0912,17.9607,133.6254,8.9084
200,8x16x8,16,16384,1900.5440,393.2160,1114.1120,0.1824,17.9620,133.6156,8.9077
200,8x16x16,16,32768,3801.0880,786.4320,2228.2240,0.3660,17.9062,134.0320,8.9355
200,8x16x32,16,65536,7602.1760,1572.8640,4456.4480,0.7780,16.8469,142.4592,9.4973
200,8x16x64,16,131072,15204.3520,3145.7280,8912.8960,1.8134,14.4562,166.0193,11.0680
200,8x32x2,16,8192,950.2720,196.6080,557.0560,0.0914,17.9222,133.9118,8.9275
200,8x32x4,16,16384,1900.5440,393.2160,1114.1120,0.1850,17.7102,135.5155,9.0344
200,8x32x8,16,32768,3801.0880,786.4320,2228.2240,0.3659,17.9109,133.9967,8.9331
200,8x32x16,16,65536,7602.1760,1572.8640,4456.4480,0.8239,15.9087,150.8613,10.0574
200,8x32x32,16,131072,15204.3520,3145.7280,8912.8960,1.8117,14.4692,165.8694,11.0580
200,8x32x64,16,262144,30408.7040,6291.4560,17825.7920,4.0171,13.0514,183.8886,12.2592
200,8x64x2,16,16384,1900.5440,393.2160,1114.1120,0.1860,17.6194,136.2134,9.0809
200,8x64x4,16,32768,3801.0880,786.4320,2228.2240,0.3671,17.8511,134.4456,8.9630
200,8x64x8,16,65536,7602.1760,1572.8640,4456.4480,0.7499,17.4789,137.3082,9.1539
200,8x64x16,16,131072,15204.3520,3145.7280,8912.8960,1.8146,14.4464,166.1315,11.0754
200,8x64x32,16,262144,30408.7040,6291.4560,17825.7920,4.0309,13.0066,184.5212,12.3014
200,8x64x64,16,524288,60817.4080,12582.9120,35651.5840,8.5774,12.2249,196.3206,13.0880
200,16x4x2,16,2048,237.5680,49.1520,139.2640,0.0225,18.1741,132.0560,8.8037
200,16x4x4,16,4096,475.1360,98.3040,278.5280,0.0457,17.9366,133.8046,8.9203
200,16x4x8,16,8192,950.2720,196.6080,557.0560,0.0930,17.6142,136.2538,9.0836
200,16x4x16,16,16384,1900.5440,393.2160,1114.1120,0.1846,17.7515,135.1995,9.0133
200,16x4x32,16,32768,3801.0880,786.4320,2228.2240,0.3653,17.9405,133.7757,8.9184
200,16x4x64,16,65536,7602.1760,1572.8640,4456.4480,0.7407,17.6949,135.6324,9.0422
200,16x8x2,16,4096,475.1360,98.3040,278.5280,0.0455,18.0129,133.2379,8.8825
200,16x8x4,16,8192,950.2720,196.6080,557.0560,0.0923,17.7417,135.2748,9.0183
200,16x8x8,16,16384,1900.5440,393.2160,1114.1120,0.1829,17.9163,133.9561,8.9304
200,16x8x16,16,32768,3801.0880,786.4320,2228.2240,0.3665,17.8840,134.1982,8.9465
200,16x8x32,16,65536,7602.1760,1572.8640,4456.4480,0.7461,17.5671,136.6192,9.1079
200,16x8x64,16,131072,15204.3520,3145.7280,8912.8960,1.8047,14.5254,165.2281,11.0152
200,16x16x2,16,8192,950.2720,196.6080,557.0560,0.0923,17.7442,135.2553,9.0170
200,16x16x4,16,16384,1900.5440,393.2160,1114.1120,0.1851,17.7021,135.5774,9.0385
200,16x16x8,16,32768,3801.0880,786.4320,2228.2240,0.3663,17.8925,134.1344,8.9423
200,16x16x16,16,65536,7602.1760,1572.8640,4456.4480,0.7613,17.2167,139.3996,9.2933
200,16x16x32,16,131072,15204.3520,3145.7280,8912.8960,1.8171,14.4263,166.3633,11.0909
200,16x16x64,16,262144,30408.7040,6291.4560,17825.7920,4.0583,12.9189,185.7738,12.3849
200,16x32x2,16,16384,1900.5440,393.2160,1114.1120,0.1846,17.7495,135.2148,9.0143
200,16x32x4,16,32768,3801.0880,786.4320,2228.2240,0.3663,17.8904,134.1501,8.9433
200,16x32x8,16,65536,7602.1760,1572.8640,4456.4480,0.7477,17.5298,136.9097,9.1273
200,16x32x16,16,131072,15204.3520,3145.7280,8912.8960,1.8009,14.5562,164.8782,10.9919
200,16x32x32,16,262144,30408.7040,6291.4560,17825.7920,4.0799,12.8505,186.7638,12.4509
200,16x32x64,16,524288,60817.4080,12582.9120,35651.5840,8.5664,12.2405,196.0700,13.0713
200,16x64x2,16,32768,3801.0880,786.4320,2228.2240,0.3630,18.0539,132.9349,8.8623
200,16x64x4,16,65536,7602.1760,1572.8640,4456.4480,0.7584,17.2830,138.8645,9.2576
200,16x64x8,16,131072,15204.3520,3145.7280,8912.8960,1.8135,14.4554,166.0274,11.0685
200,16x64x16,16,262144,30408.7040,6291.4560,17825.7920,4.0762,12.8621,186.5942,12.4396
200,16x64x32,16,524288,60817.4080,12582.9120,35651.5840,8.5873,12.2108,196.5473,13.1032
200,16x64x64,16,1048576,121634.8160,25165.8240,71303.1680,17.2827,12.1344,197.7849,13.1857
200,32x4x2,16,4096,475.1360,98.3040,278.5280,0.0456,17.9516,133.6928,8.9129
200,32x4x4,16,8192,950.2720,196.6080,557.0560,0.0939,17.4453,137.5727,9.1715
200,32x4x8,16,16384,1900.5440,393.2160,1114.1120,0.1851,17.7068,135.5413,9.0361
200,32x4x16,16,32768,3801.0880,786.4320,2228.2240,0.3663,17.8912,134.1438,8.9429
200,32x4x32,16,65536,7602.1760,1572.8640,4456.4480,0.7391,17.7335,135.3367,9.0224
200,32x4x64,16,131072,15204.3520,3145.7280,8912.8960,1.7899,14.6456,163.8720,10.9248
200,32x8x2,16,8192,950.2720,196.6080,557.0560,0.0920,17.8125,134.7366,8.9824
200,32x8x4,16,16384,1900.5440,393.2160,1114.1120,0.1853,17.6791,135.7533,9.0502
200,32x8x8,16,32768,3801.0880,786.4320,2228.2240,0.3653,17.9421,133.7634,8.9176
200,32x8x16,16,65536,7602.1760,1572.8640,4456.4480,0.7749,16.9148,141.8872,9.4591
200,32x8x32,16,131072,15204.3520,3145.7280,8912.8960,1.8102,14.4812,165.7323,11.0488
200,32x8x64,16,262144,30408.7040,6291.4560,17825.7920,4.0739,12.8695,186.4867,12.4324
200,32x16x2,16,16384,1900.5440,393.2160,1114.1120,0.1844,17.7741,135.0281,9.0019
200,32x16x4,16,32768,3801.0880,786.4320,2228.2240,0.3635,18.0300,133.1117,8.8741
200,32x16x8,16,65536,7602.1760,1572.8640,4456.4480,0.7638,17.1596,139.8630,9.3242
200,32x16x16,16,131072,15204.3520,3145.7280,8912.8960,1.8171,14.4267,166.3587,11.0906
200,32x16x32,16,262144,30408.7040,6291.4560,17825.7920,4.0640,12.9008,186.0355,12.4024
200,32x16x64,16,524288,60817.4080,12582.9120,35651.5840,8.5080,12.3246,194.7333,12.9822
200,32x32x2,16,32768,3801.0880,786.4320,2228.2240,0.3675,17.8332,134.5804,8.9720
200,32x32x4,16,65536,7602.1760,1572.8640,4456.4480,0.7585,17.2815,138.8771,9.2585
200,32x32x8,16,131072,15204.3520,3145.7280,8912.8960,1.8147,14.4452,166.1447,11.0763
200,32x32x16,16,262144,30408.7040,6291.4560,17825.7920,4.0587,12.9176,185.7936,12.3862
200,32x32x32,16,524288,60817.4080,12582.9120,35651.5840,8.6098,12.1788,197.0638,13.1376
200,32x32x64,16,1048576,121634.8160,25165.8240,71303.1680,17.3969,12.0547,199.0917,13.2728
200,32x64x2,16,65536,7602.1760,1572.8640,4456.4480,0.7403,17.7044,135.5592,9.0373
200,32x64x4,16,131072,15204.3520,3145.7280,8912.8960,1.8081,14.4986,165.5333,11.0356
200,32x64x8,16,262144,30408.7040,6291.4560,17825.7920,4.0864,12.8300,187.0616,12.4708
200,32x64x16,16,524288,60817.4080,12582.9120,35651.5840,8.5361,12.2840,195.3760,13.0251
200,32x64x32,16,1048576,121634.8160,25165.8240,71303.1680,17.3914,12.0585,199.0291,13.2686
200,32x64x64,16,2097152,243269.6320,50331.6480,142606.3360,35.0665,11.9610,200.6520,13.3768
200,64x4x2,16,8192,950.2720,196.6080,557.0560,0.0914,17.9340,133.8241,8.9216
200,64x4x4,16,16384,1900.5440,393.2160,1114.1120,0.1833,17.8749,134.2662,8.9511
200,64x4x8,16,32768,3801.0880,786.4320,2228.2240,0.3669,17.8618,134.3650,8.9577
200,64x4x16,16,65536,7602.1760,1572.8640,4456.4480,0.7622,17.1955,139.5710,9.3047
200,64x4x32,16,131072,15204.3520,3145.7280,8912.8960,1.8131,14.4585,165.9924,11.0662
200,64x4x64,16,262144,30408.7040,6291.4560,17825.7920,4.0475,12.9534,185.2790,12.3519
200,64x8x2,16,16384,1900.5440,393.2160,1114.1120,0.1831,17.8992,134.0843,8.9390
200,64x8x4,16,32768,3801.0880,786.4320,2228.2240,0.3669,17.8643,134.3464,8.9564
200,64x8x8,16,65536,7602.1760,1572.8640,4456.4480,0.7450,17.5925,136.4221,9.0948
200,64x8x16,16,131072,15204.3520,3145.7280,8912.8960,1.8160,14.4350,166.2626,11.0842
200,64x8x32,16,262144,30408.7040,6291.4560,17825.7920,4.0604,12.9123,185.8690,12.3913
200,64x8x64,16,524288,60817.4080,12582.9120,35651.5840,8.5079,12.3248,194.7298,12.9820
200,64x16x2,16,32768,3801.0880,786.4320,2228.2240,0.3655,17.9308,133.8476,8.9232
200,64x16x4,16,65536,7602.1760,1572.8640,4456.4480,0.7456,17.5799,136.5194,9.1013
200,64x16x8,16,131072,15204.3520,3145.7280,8912.8960,1.8031,14.5389,165.0747,11.0050
200,64x16x16,16,262144,30408.7040,6291.4560,17825.7920,4.0631,12.9035,185.9954,12.3997
200,64x16x32,16,524288,60817.4080,12582.9120,35651.5840,8.5217,12.3048,195.0458,13.0031
200,64x16x64,16,1048576,121634.8160,25165.8240,71303.1680,17.4178,12.0403,199.3308,13.2887
200,64x32x2,16,65536,7602.1760,1572.8640,4456.4480,0.8037,16.3093,147.1549,9.8103
200,64x32x4,16,131072,15204.3520,3145.7280,8912.8960,1.8168,14.4292,166.3290,11.0886
200,64x32x8,16,262144,30408.7040,6291.4560,17825.7920,4.0649,12.8978,186.0779,12.4052
200,64x32x16,16,524288,60817.4080,12582.9120,35651.5840,8.5959,12.1986,196.7439,13.1163
200,64x32x32,16,1048576,121634.8160,25165.8240,71303.1680,17.3318,12.1000,198.3464,13.2231
200,64x32x64,16,2097152,243269.6320,50331.6480,142606.3360,35.1403,11.9359,201.0746,13.4050
200,64x64x2,16,131072,15204.3520,3145.7280,8912.8960,1.8047,14.5253,165.2288,11.0153
200,64x64x4,16,262144,30408.7040,6291.4560,17825.7920,4.0375,12.9854,184.8230,12.3215
200,64x64x8,16,524288,60817.4080,12582.9120,35651.5840,8.5280,12.2957,195.1895,13.0126
200,64x64x16,16,1048576,121634.8160,25165.8240,71303.1680,17.2131,12.1835,196.9883,13.1326
200,64x64x32,16,2097152,243269.6320,50331.6480,142606.3360,34.7267,12.0780,198.7080,13.2472
200,64x64x64,16,4194304,486539.2640,100663.2960,285212.6720,69.9977,11.9841,200.2649,13.3510
200,4x4x2,32,1024,184.3200,24.5760,135.1680,0.0184,11.1445,215.3526,6.9469
200,4x4x4,32,2048,368.6400,49.1520,270.3360,0.0367,11.1484,215.2784,6.9445
200,4x4x8,32,4096,737.2800,98.3040,540.6720,0.0749,10.9335,219.5084,7.0809
200,4x4x16,32,8192,1474.5600,196.6080,1081.3440,0.1489,11.0031,218.1212,7.0362
200,4x4x32,32,16384,2949.1200,393.2160,2162.6880,0.2979,10.9983,218.2151,7.0392
200,4x4x64,32,32768,5898.2400,786.4320,4325.3760,0.5936,11.0407,217.3782,7.0122
200,4x8x2,32,2048,368.6400,49.1520,270.3360,0.0368,11.1377,215.4848,6.9511
200,4x8x4,32,4096,737.2800,98.3040,540.6720,0.0747,10.9650,218.8788,7.0606
200,4x8x8,32,8192,1474.5600,196.6080,1081.3440,0.1491,10.9856,218.4685,7.0474
200,4x8x16,32,16384,2949.1200,393.2160,2162.6880,0.2987,10.9688,218.8018,7.0581
200,4x8x32,32,32768,5898.2400,786.4320,4325.3760,0.5901,11.1053,216.1135,6.9714
200,4x8x64,32,65536,11796.4800,1572.8640,8650.7520,1.2135,10.8010,222.2016,7.1678
200,4x16x2,32,4096,737.2800,98.3040,540.6720,0.0746,10.9759,218.6605,7.0536
200,4x16x4,32,8192,1474.5600,196.6080,1081.3440,0.1495,10.9556,219.0657,7.0666
200,4x16x8,32,16384,2949.1200,393.2160,2162.6880,0.2986,10.9744,218.6898,7.0545
200,4x16x16,32,32768,5898.2400,786.4320,4325.3760,0.5924,11.0635,216.9289,6.9977
200,4x16x32,32,65536,11796.4800,1572.8640,8650.7520,1.2140,10.7970,222.2843,7.1705
200,4x16x64,32,131072,23592.9600,3145.7280,17301.5040,2.6870,9.7559,246.0042,7.9356
200,4x32x2,32,8192,1474.5600,196.6080,1081.3440,0.1479,11.0780,216.6453,6.9886
200,4x32x4,32,16384,2949.1200,393.2160,2162.6880,0.2975,11.0133,217.9181,7.0296
200,4x32x8,32,32768,5898.2400,786.4320,4325.3760,0.5923,11.0647,216.9061,6.9970
200,4x32x16,32,65536,11796.4800,1572.8640,8650.7520,1.2194,10.7489,223.2783,7.2025
200,4x32x32,32,131072,23592.9600,3145.7280,17301.5040,2.6961,9.7231,246.8337,7.9624
200,4x32x64,32,262144,47185.9200,6291.4560,34603.0080,5.6436,9.2899,258.3444,8.3337
200,4x64x2,32,16384,2949.1200,393.2160,2162.6880,0.2979,10.9978,218.2247,7.0395
200,4x64x4,32,32768,5898.2400,786.4320,4325.3760,0.5905,11.0991,216.2334,6.9753
200,4x64x8,32,65536,11796.4800,1572.8640,8650.7520,1.2377,10.5903,226.6227,7.3104
200,4x64x16,32,131072,23592.9600,3145.7280,17301.5040,2.7026,9.6995,247.4349,7.9818
200,4x64x32,32,262144,47185.9200,6291.4560,34603.0080,5.6421,9.2925,258.2736,8.3314
200,4x64x64,32,524288,94371.8400,12582.9120,69206.0160,11.5057,9.1135,263.3446,8.4950
200,8x4x2,32,2048,368.6400,49.1520,270.3360,0.0366,11.1998,214.2904,6.9126
200,8x4x4,32,4096,737.2800,98.3040,540.6720,0.0742,11.0403,217.3856,7.0124
200,8x4x8,32,8192,1474.5600,196.6080,1081.3440,0.1492,10.9849,218.4817,7.0478
200,8x4x16,32,16384,2949.1200,393.2160,2162.6880,0.2991,10.9560,219.0584,7.0664
200,8x4x32,32,32768,5898.2400,786.4320,4325.3760,0.5907,11.0940,216.3330,6.9785
200,8x4x64,32,65536,11796.4800,1572.8640,8650.7520,1.2175,10.7656,222.9332,7.1914
200,8x8x2,32,4096,737.2800,98.3040,540.6720,0.0743,11.0187,217.8120,7.0262
200,8x8x4,32,8192,1474.5600,196.6080,1081.3440,0.1495,10.9595,218.9885,7.0641
200,8x8x8,32,16384,2949.1200,393.2160,2162.6880,0.2999,10.9274,219.6313,7.0849
200,8x8x16,32,32768,5898.2400,786.4320,4325.3760,0.5869,11.1662,214.9339,6.9334
200,8x8x32,32,65536,11796.4800,1572.8640,8650.7520,1.2165,10.7745,222.7480,7.1854
200,8x8x64,32,131072,23592.9600,3145.7280,17301.5040,2.7471,9.5424,251.5081,8.1132
200,8x16x2,32,8192,1474.5600,196.6080,1081.3440,0.1490,10.9992,218.1970,7.0386
200,8x16x4,32,16384,2949.1200,393.2160,2162.6880,0.2983,10.9837,218.5056,7.0486
200,8x16x8,32,32768,5898.2400,786.4320,4325.3760,0.5901,11.1053,216.1125,6.9714
200,8x16x16,32,65536,11796.4800,1572.8640,8650.7520,1.2063,10.8653,220.8859,7.1254
200,8x16x32,32,131072,23592.9600,3145.7280,17301.5040,2.6882,9.7516,246.1127,7.9391
200,8x16x64,32,262144,47185.9200,6291.4560,34603.0080,5.6302,9.3121,257.7289,8.3138
200,8x32x2,32,16384,2949.1200,393.2160,2162.6880,0.2976,11.0089,218.0051,7.0324
200,8x32x4,32,32768,5898.2400,786.4320,4325.3760,0.5926,11.0594,217.0096,7.0003
200,8x32x8,32,65536,11796.4800,1572.8640,8650.7520,1.2088,10.8433,221.3346,7.1398
200,8x32x16,32,131072,23592.9600,3145.7280,17301.5040,2.7010,9.7055,247.2829,7.9769
200,8x32x32,32,262144,47185.9200,6291.4560,34603.0080,5.6443,9.2887,258.3778,8.3348
200,8x32x64,32,524288,94371.8400,12582.9120,69206.0160,11.4754,9.1376,262.6509,8.4726
200,8x64x2,32,32768,5898.2400,786.4320,4325.3760,0.5909,11.0900,216.4113,6.9810
200,8x64x4,32,65536,11796.4800,1572.8640,8650.7520,1.2180,10.7614,223.0192,7.1942
200,8x64x8,32,131072,23592.9600,3145.7280,17301.5040,2.6995,9.7110,247.1430,7.9724
200,8x64x16,32,262144,47185.9200,6291.4560,34603.0080,5.6213,9.3268,257.3236,8.3008
200,8x64x32,32,524288,94371.8400,12582.9120,69206.0160,11.5317,9.0930,263.9393,8.5142
200,8x64x64,32,1048576,188743.6800,25165.8240,138412.0320,23.0906,9.0823,264.2507,8.5242
200,16x4x2,32,4096,737.2800,98.3040,540.6720,0.0734,11.1635,214.9868,6.9351
200,16x4x4,32,8192,1474.5600,196.6080,1081.3440,0.1493,10.9774,218.6311,7.0526
200,16x4x8,32,16384,2949.1200,393.2160,2162.6880,0.2981,10.9922,218.3376,7.0431
200,16x4x16,32,32768,5898.2400,786.4320,4325.3760,0.5933,11.0465,217.2627,7.0085
200,16x4x32,32,65536,11796.4800,1572.8640,8650.7520,1.2348,10.6149,226.0980,7.2935
200,16x4x64,32,131072,23592.9600,3145.7280,17301.5040,2.6964,9.7218,246.8669,7.9634
200,16x8x2,32,8192,1474.5600,196.6080,1081.3440,0.1470,11.1473,215.2982,6.9451
200,16x8x4,32,16384,2949.1200,393.2160,2162.6880,0.2982,10.9872,218.4350,7.0463
200,16x8x8,32,32768,5898.2400,786.4320,4325.3760,0.5910,11.0884,216.4431,6.9820
200,16x8x16,32,65536,11796.4800,1572.8640,8650.7520,1.2354,10.6093,226.2169,7.2973
200,16x8x32,32,131072,23592.9600,3145.7280,17301.5040,2.6843,9.7659,245.7535,7.9275
200,16x8x64,32,262144,47185.9200,6291.4560,34603.0080,5.6402,9.2956,258.1872,8.3286
200,16x16x2,32,16384,2949.1200,393.2160,2162.6880,0.2986,10.9722,218.7351,7.0560
200,16x16x4,32,32768,5898.2400,786.4320,4325.3760,0.5930,11.0511,217.1736,7.0056
200,16x16x8,32,65536,11796.4800,1572.8640,8650.7520,1.2187,10.7551,223.1490,7.1984
200,16x16x16,32,131072,23592.9600,3145.7280,17301.5040,2.6955,9.7253,246.7789,7.9606
200,16x16x32,32,262144,47185.9200,6291.4560,34603.0080,5.6300,9.3124,257.7221,8.3136
200,16x16x64,32,524288,94371.8400,12582.9120,69206.0160,11.4678,9.1436,262.4776,8.4670
200,16x32x2,32,32768,5898.2400,786.4320,4325.3760,0.5932,11.0481,217.2322,7.0075
200,16x32x4,32,65536,11796.4800,1572.8640,8650.7520,1.2189,10.7536,223.1802,7.1994
200,16x32x8,32,131072,23592.9600,3145.7280,17301.5040,2.6839,9.7675,245.7140,7.9263
200,16x32x16,32,262144,47185.9200,6291.4560,34603.0080,5.6155,9.3364,257.0593,8.2922
200,16x32x32,32,524288,94371.8400,12582.9120,69206.0160,11.5181,9.1037,263.6285,8.5041
200,16x32x64,32,1048576,188743.6800,25165.8240,138412.0320,23.2101,9.0355,265.6190,8.5684
200,16x64x2,32,65536,11796.4800,1572.8640,8650.7520,1.2148,10.7894,222.4400,7.1755
200,16x64x4,32,131072,23592.9600,3145.7280,17301.5040,2.7171,9.6480,248.7573,8.0244
200,16x64x8,32,262144,47185.9200,6291.4560,34603.0080,5.6323,9.3086,257.8258,8.3170
200,16x64x16,32,524288,94371.8400,12582.9120,69206.0160,11.4916,9.1247,263.0214,8.4846
200,16x64x32,32,1048576,188743.6800,25165.8240,138412.0320,23.1721,9.0503,265.1837,8.5543
200,16x64x64,32,2097152,377487.3600,50331.6480,276824.0640,46.4225,9.0351,265.6319,8.5688
200,32x4x2,32,8192,1474.5600,196.6080,1081.3440,0.1491,10.9908,218.3644,7.0440
200,32x4x4,32,16384,2949.1200,393.2160,2162.6880,0.2980,10.9973,218.2357,7.0399
200,32x4x8,32,32768,5898.2400,786.4320,4325.3760,0.5923,11.0640,216.9201,6.9974
200,32x4x16,32,65536,11796.4800,1572.8640,8650.7520,1.2190,10.7526,223.2011,7.2000
200,32x4x32,32,131072,23592.9600,3145.7280,17301.5040,2.6910,9.7414,246.3719,7.9475
200,32x4x64,32,262144,47185.9200,6291.4560,34603.0080,5.6597,9.2635,259.0816,8.3575
200,32x8x2,32,16384,2949.1200,393.2160,2162.6880,0.2994,10.9438,219.3016,7.0742
200,32x8x4,32,32768,5898.2400,786.4320,4325.3760,0.5909,11.0911,216.3901,6.9803
200,32x8x8,32,65536,11796.4800,1572.8640,8650.7520,1.2158,10.7807,222.6192,7.1813
200,32x8x16,32,131072,23592.9600,3145.7280,17301.5040,2.6899,9.7456,246.2645,7.9440
200,32x8x32,32,262144,47185.9200,6291.4560,34603.0080,5.6445,9.2884,258.3855,8.3350
200,32x8x64,32,524288,94371.8400,12582.9120,69206.0160,11.4730,9.1395,262.5972,8.4709
200,32x16x2,32,32768,5898.2400,786.4320,4325.3760,0.5929,11.0536,217.1243,7.0040
200,32x16x4,32,65536,11796.4800,1572.8640,8650.7520,1.2112,10.8220,221.7706,7.1539
200,32x16x8,32,131072,23592.9600,3145.7280,17301.5040,2.6860,9.7596,245.9114,7.9326
200,32x16x16,32,262144,47185.9200,6291.4560,34603.0080,5.6322,9.3087,257.8220,8.3168
200,32x16x32,32,524288,94371.8400,12582.9120,69206.0160,11.5069,9.1126,263.3713,8.4958
200,32x16x64,32,1048576,188743.6800,25165.8240,138412.0320,23.1910,9.0429,265.4001,8.5613
200,32x32x2,32,65536,11796.4800,1572.8640,8650.7520,1.2062,10.8663,220.8666,7.1247
200,32x32x4,32,131072,23592.9600,3145.7280,17301.5040,2.6866,9.7575,245.9652,7.9344
200,32x32x8,32,262144,47185.9200,6291.4560,34603.0080,5.6328,9.3077,257.8506,8.3178
200,32x32x16,32,524288,94371.8400,12582.9120,69206.0160,11.4603,9.1496,262.3058,8.4615
200,32x32x32,32,1048576,188743.6800,25165.8240,138412.0320,23.2151,9.0336,265.6754,8.5702
200,32x32x64,32,2097152,377487.3600,50331.6480,276824.0640,46.3930,9.0408,265.4628,8.5633
200,32x64x2,32,131072,23592.9600,3145.7280,17301.5040,2.6934,9.7330,246.5840,7.9543
200,32x64x4,32,262144,47185.9200,6291.4560,34603.0080,5.6297,9.3129,257.7073,8.3131
200,32x64x8,32,524288,94371.8400,12582.9120,69206.0160,11.4733,9.1393,262.6034,8.4711
200,32x64x16,32,1048576,188743.6800,25165.8240,138412.0320,23.1572,9.0562,265.0131,8.5488
200,32x64x32,32,2097152,377487.3600,50331.6480,276824.0640,46.4403,9.0316,265.7335,8.5720
200,32x64x64,32,4194304,754974.7200,100663.2960,553648.1280,93.1158,9.0088,266.4065,8.5938
200,64x4x2,32,16384,2949.1200,393.2160,2162.6880,0.2983,10.9862,218.4566,7.0470
200,64x4x4,32,32768,5898.2400,786.4320,4325.3760,0.5919,11.0721,216.7605,6.9923
200,64x4x8,32,65536,11796.4800,1572.8640,8650.7520,1.2059,10.8690,220.8113,7.1229
200,64x4x16,32,131072,23592.9600,3145.7280,17301.5040,2.7223,9.6296,249.2308,8.0397
200,64x4x32,32,262144,47185.9200,6291.4560,34603.0080,5.6259,9.3192,257.5338,8.3075
200,64x4x64,32,524288,94371.8400,12582.9120,69206.0160,11.5348,9.0905,264.0113,8.5165
200,64x8x2,32,32768,5898.2400,786.4320,4325.3760,0.5919,11.0719,216.7642,6.9924
200,64x8x4,32,65536,11796.4800,1572.8640,8650.7520,1.2060,10.8686,220.8192,7.1232
200,64x8x8,32,131072,23592.9600,3145.7280,17301.5040,2.6973,9.7187,246.9454,7.9660
200,64x8x16,32,262144,47185.9200,6291.4560,34603.0080,5.6342,9.3055,257.9115,8.3197
200,64x8x32,32,524288,94371.8400,12582.9120,69206.0160,11.4968,9.1206,263.1412,8.4884
200,64x8x64,32,1048576,188743.6800,25165.8240,138412.0320,23.1210,9.0703,264.5987,8.5354
200,64x16x2,32,65536,11796.4800,1572.8640,8650.7520,1.2136,10.8004,222.2134,7.1682
200,64x16x4,32,131072,23592.9600,3145.7280,17301.5040,2.6854,9.7618,245.8551,7.9308
200,64x16x8,32,262144,47185.9200,6291.4560,34603.0080,5.6319,9.3092,257.8084,8.3164
200,64x16x16,32,524288,94371.8400,12582.9120,69206.0160,11.4978,9.1198,263.1639,8.4892
200,64x16x32,32,1048576,188743.6800,25165.8240,138412.0320,23.1832,9.0460,265.3105,8.5584
200,64x16x64,32,2097152,377487.3600,50331.6480,276824.0640,46.4555,9.0287,265.8204,8.5749
200,64x32x2,32,131072,23592.9600,3145.7280,17301.5040,2.6859,9.7600,245.9017,7.9323
200,64x32x4,32,262144,47185.9200,6291.4560,34603.0080,5.6275,9.3166,257.6054,8.3099
200,64x32x8,32,524288,94371.8400,12582.9120,69206.0160,11.4766,9.1367,262.6781,8.4735
200,64x32x16,32,1048576,188743.6800,25165.8240,138412.0320,23.1329,9.0657,264.7348,8.5398
200,64x32x32,32,2097152,377487.3600,50331.6480,276824.0640,46.5382,9.0126,266.2938,8.5901
200,64x32x64,32,4194304,754974.7200,100663.2960,553648.1280,92.7335,9.0459,265.3127,8.5585
200,64x64x2,32,262144,47185.9200,6291.4560,34603.0080,5.6284,9.3150,257.6485,8.3112
200,64x64x4,32,524288,94371.8400,12582.9120,69206.0160,11.4938,9.1230,263.0719,8.4862
200,64x64x8,32,1048576,188743.6800,25165.8240,138412.0320,23.1701,9.0511,265.1611,8.5536
200,64x64x16,32,2097152,377487.3600,50331.6480,276824.0640,46.5385,9.0125,266.2957,8.5902
200,64x64x32,32,4194304,754974.7200,100663.2960,553648.1280,93.0355,9.0166,266.1767,8.5863
200,64x64x64,32,8388608,1509949.4400,201326.5920,1107296.2560,187.8248,8.9324,268.6855,8.6673

363
scripts/results_soa_casclakesp2_iln100.txt
View File

@ -1,363 +0,0 @@
200,4x4x2,8,256,21.5040,6.1440,9.2160,0.0724,70.7407,35.3403,5.0486
200,4x4x4,8,512,43.0080,12.2880,18.4320,0.1449,70.6729,35.3742,5.0535
200,4x4x8,8,1024,86.0160,24.5760,36.8640,0.2903,70.5584,35.4316,5.0617
200,4x4x16,8,2048,172.0320,49.1520,73.7280,0.5804,70.5670,35.4273,5.0610
200,4x4x32,8,4096,344.0640,98.3040,147.4560,1.1651,70.3143,35.5546,5.0792
200,4x4x64,8,8192,688.1280,196.6080,294.9120,2.3287,70.3554,35.5339,5.0763
200,4x8x2,8,512,43.0080,12.2880,18.4320,0.1451,70.5941,35.4137,5.0591
200,4x8x4,8,1024,86.0160,24.5760,36.8640,0.2900,70.6120,35.4047,5.0578
200,4x8x8,8,2048,172.0320,49.1520,73.7280,0.5800,70.6225,35.3995,5.0571
200,4x8x16,8,4096,344.0640,98.3040,147.4560,1.1606,70.5872,35.4172,5.0596
200,4x8x32,8,8192,688.1280,196.6080,294.9120,2.3285,70.3622,35.5305,5.0758
200,4x8x64,8,16384,1376.2560,393.2160,589.8240,4.6577,70.3519,35.5356,5.0765
200,4x16x2,8,1024,86.0160,24.5760,36.8640,0.2900,70.6244,35.3985,5.0569
200,4x16x4,8,2048,172.0320,49.1520,73.7280,0.5802,70.5984,35.4116,5.0588
200,4x16x8,8,4096,344.0640,98.3040,147.4560,1.1645,70.3467,35.5383,5.0769
200,4x16x16,8,8192,688.1280,196.6080,294.9120,2.3288,70.3539,35.5346,5.0764
200,4x16x32,8,16384,1376.2560,393.2160,589.8240,4.6600,70.3172,35.5532,5.0790
200,4x16x64,8,32768,2752.5120,786.4320,1179.6480,9.3183,70.3306,35.5464,5.0781
200,4x32x2,8,2048,172.0320,49.1520,73.7280,0.5801,70.6086,35.4065,5.0581
200,4x32x4,8,4096,344.0640,98.3040,147.4560,1.1606,70.5830,35.4193,5.0599
200,4x32x8,8,8192,688.1280,196.6080,294.9120,2.3294,70.3369,35.5432,5.0776
200,4x32x16,8,16384,1376.2560,393.2160,589.8240,4.6553,70.3893,35.5168,5.0738
200,4x32x32,8,32768,2752.5120,786.4320,1179.6480,9.3281,70.2565,35.5839,5.0834
200,4x32x64,8,65536,5505.0240,1572.8640,2359.2960,18.6943,70.1133,35.6566,5.0938
200,4x64x2,8,4096,344.0640,98.3040,147.4560,1.1606,70.5854,35.4181,5.0597
200,4x64x4,8,8192,688.1280,196.6080,294.9120,2.3293,70.3399,35.5417,5.0774
200,4x64x8,8,16384,1376.2560,393.2160,589.8240,4.6605,70.3106,35.5565,5.0795
200,4x64x16,8,32768,2752.5120,786.4320,1179.6480,9.3203,70.3157,35.5539,5.0791
200,4x64x32,8,65536,5505.0240,1572.8640,2359.2960,18.7065,70.0674,35.6799,5.0971
200,4x64x64,8,131072,11010.0480,3145.7280,4718.5920,37.8230,69.3080,36.0709,5.1530
200,8x4x2,8,512,43.0080,12.2880,18.4320,0.1450,70.6204,35.4005,5.0572
200,8x4x4,8,1024,86.0160,24.5760,36.8640,0.2904,70.5318,35.4450,5.0636
200,8x4x8,8,2048,172.0320,49.1520,73.7280,0.5842,70.1134,35.6565,5.0938
200,8x4x16,8,4096,344.0640,98.3040,147.4560,1.1659,70.2662,35.5790,5.0827
200,8x4x32,8,8192,688.1280,196.6080,294.9120,2.3293,70.3383,35.5425,5.0775
200,8x4x64,8,16384,1376.2560,393.2160,589.8240,4.6570,70.3624,35.5303,5.0758
200,8x8x2,8,1024,86.0160,24.5760,36.8640,0.2902,70.5777,35.4219,5.0603
200,8x8x4,8,2048,172.0320,49.1520,73.7280,0.5806,70.5534,35.4342,5.0620
200,8x8x8,8,4096,344.0640,98.3040,147.4560,1.1619,70.5074,35.4573,5.0653
200,8x8x16,8,8192,688.1280,196.6080,294.9120,2.3311,70.2836,35.5702,5.0815
200,8x8x32,8,16384,1376.2560,393.2160,589.8240,4.6565,70.3707,35.5261,5.0752
200,8x8x64,8,32768,2752.5120,786.4320,1179.6480,9.3263,70.2702,35.5770,5.0824
200,8x16x2,8,2048,172.0320,49.1520,73.7280,0.5807,70.5301,35.4459,5.0637
200,8x16x4,8,4096,344.0640,98.3040,147.4560,1.1617,70.5187,35.4516,5.0645
200,8x16x8,8,8192,688.1280,196.6080,294.9120,2.3301,70.3158,35.5539,5.0791
200,8x16x16,8,16384,1376.2560,393.2160,589.8240,4.6612,70.2990,35.5624,5.0803
200,8x16x32,8,32768,2752.5120,786.4320,1179.6480,9.3193,70.3226,35.5504,5.0786
200,8x16x64,8,65536,5505.0240,1572.8640,2359.2960,18.7155,70.0339,35.6970,5.0996
200,8x32x2,8,4096,344.0640,98.3040,147.4560,1.1650,70.3184,35.5525,5.0789
200,8x32x4,8,8192,688.1280,196.6080,294.9120,2.3298,70.3239,35.5498,5.0785
200,8x32x8,8,16384,1376.2560,393.2160,589.8240,4.6583,70.3430,35.5401,5.0772
200,8x32x16,8,32768,2752.5120,786.4320,1179.6480,9.3185,70.3292,35.5471,5.0782
200,8x32x32,8,65536,5505.0240,1572.8640,2359.2960,18.6322,70.3469,35.5382,5.0769
200,8x32x64,8,131072,11010.0480,3145.7280,4718.5920,37.8024,69.3458,36.0512,5.1502
200,8x64x2,8,8192,688.1280,196.6080,294.9120,2.3291,70.3437,35.5398,5.0771
200,8x64x4,8,16384,1376.2560,393.2160,589.8240,4.6581,70.3470,35.5381,5.0769
200,8x64x8,8,32768,2752.5120,786.4320,1179.6480,9.3259,70.2733,35.5754,5.0822
200,8x64x16,8,65536,5505.0240,1572.8640,2359.2960,18.6892,70.1325,35.6468,5.0924
200,8x64x32,8,131072,11010.0480,3145.7280,4718.5920,37.7789,69.3891,36.0287,5.1470
200,8x64x64,8,262144,22020.0960,6291.4560,9437.1840,76.8891,68.1876,36.6636,5.2377
200,16x4x2,8,1024,86.0160,24.5760,36.8640,0.2901,70.6085,35.4065,5.0581
200,16x4x4,8,2048,172.0320,49.1520,73.7280,0.5798,70.6425,35.3895,5.0556
200,16x4x8,8,4096,344.0640,98.3040,147.4560,1.1648,70.3298,35.5468,5.0781
200,16x4x16,8,8192,688.1280,196.6080,294.9120,2.3279,70.3809,35.5210,5.0744
200,16x4x32,8,16384,1376.2560,393.2160,589.8240,4.6856,69.9334,35.7483,5.1069
200,16x4x64,8,32768,2752.5120,786.4320,1179.6480,9.3227,70.2975,35.5631,5.0804
200,16x8x2,8,2048,172.0320,49.1520,73.7280,0.5802,70.5913,35.4151,5.0593
200,16x8x4,8,4096,344.0640,98.3040,147.4560,1.1611,70.5552,35.4332,5.0619
200,16x8x8,8,8192,688.1280,196.6080,294.9120,2.3278,70.3847,35.5191,5.0742
200,16x8x16,8,16384,1376.2560,393.2160,589.8240,4.6602,70.3152,35.5542,5.0792
200,16x8x32,8,32768,2752.5120,786.4320,1179.6480,9.3197,70.3199,35.5518,5.0788
200,16x8x64,8,65536,5505.0240,1572.8640,2359.2960,18.6500,70.2798,35.5721,5.0817
200,16x16x2,8,4096,344.0640,98.3040,147.4560,1.1600,70.6186,35.4014,5.0573
200,16x16x4,8,8192,688.1280,196.6080,294.9120,2.3273,70.3983,35.5122,5.0732
200,16x16x8,8,16384,1376.2560,393.2160,589.8240,4.6616,70.2940,35.5649,5.0807
200,16x16x16,8,32768,2752.5120,786.4320,1179.6480,9.3175,70.3367,35.5433,5.0776
200,16x16x32,8,65536,5505.0240,1572.8640,2359.2960,18.6637,70.2283,35.5982,5.0855
200,16x16x64,8,131072,11010.0480,3145.7280,4718.5920,37.7619,69.4202,36.0126,5.1447
200,16x32x2,8,8192,688.1280,196.6080,294.9120,2.3295,70.3338,35.5448,5.0778
200,16x32x4,8,16384,1376.2560,393.2160,589.8240,4.6611,70.3016,35.5611,5.0802
200,16x32x8,8,32768,2752.5120,786.4320,1179.6480,9.3429,70.1450,35.6405,5.0915
200,16x32x16,8,65536,5505.0240,1572.8640,2359.2960,18.6794,70.1691,35.6282,5.0897
200,16x32x32,8,131072,11010.0480,3145.7280,4718.5920,37.8122,69.3279,36.0605,5.1515
200,16x32x64,8,262144,22020.0960,6291.4560,9437.1840,76.9227,68.1578,36.6796,5.2399
200,16x64x2,8,16384,1376.2560,393.2160,589.8240,4.6548,70.3959,35.5134,5.0733
200,16x64x4,8,32768,2752.5120,786.4320,1179.6480,9.3245,70.2833,35.5703,5.0815
200,16x64x8,8,65536,5505.0240,1572.8640,2359.2960,18.6449,70.2990,35.5624,5.0803
200,16x64x16,8,131072,11010.0480,3145.7280,4718.5920,37.8426,69.2722,36.0895,5.1556
200,16x64x32,8,262144,22020.0960,6291.4560,9437.1840,77.0980,68.0028,36.7632,5.2519
200,16x64x64,8,524288,44040.1920,12582.9120,18874.3680,155.0182,67.6421,36.9592,5.2799
200,32x4x2,8,2048,172.0320,49.1520,73.7280,0.5831,70.2406,35.5919,5.0846
200,32x4x4,8,4096,344.0640,98.3040,147.4560,1.1612,70.5480,35.4368,5.0624
200,32x4x8,8,8192,688.1280,196.6080,294.9120,2.3289,70.3507,35.5362,5.0766
200,32x4x16,8,16384,1376.2560,393.2160,589.8240,4.6557,70.3825,35.5202,5.0743
200,32x4x32,8,32768,2752.5120,786.4320,1179.6480,9.3166,70.3433,35.5400,5.0771
200,32x4x64,8,65536,5505.0240,1572.8640,2359.2960,18.6629,70.2311,35.5967,5.0852
200,32x8x2,8,4096,344.0640,98.3040,147.4560,1.1615,70.5290,35.4464,5.0638
200,32x8x4,8,8192,688.1280,196.6080,294.9120,2.3308,70.2941,35.5649,5.0807
200,32x8x8,8,16384,1376.2560,393.2160,589.8240,4.6586,70.3385,35.5424,5.0775
200,32x8x16,8,32768,2752.5120,786.4320,1179.6480,9.3158,70.3491,35.5371,5.0767
200,32x8x32,8,65536,5505.0240,1572.8640,2359.2960,18.6605,70.2404,35.5921,5.0846
200,32x8x64,8,131072,11010.0480,3145.7280,4718.5920,37.8379,69.2809,36.0850,5.1550
200,32x16x2,8,8192,688.1280,196.6080,294.9120,2.3253,70.4600,35.4811,5.0687
200,32x16x4,8,16384,1376.2560,393.2160,589.8240,4.6573,70.3576,35.5328,5.0761
200,32x16x8,8,32768,2752.5120,786.4320,1179.6480,9.3170,70.3403,35.5415,5.0774
200,32x16x16,8,65536,5505.0240,1572.8640,2359.2960,18.6424,70.3087,35.5575,5.0796
200,32x16x32,8,131072,11010.0480,3145.7280,4718.5920,37.8021,69.3463,36.0509,5.1501
200,32x16x64,8,262144,22020.0960,6291.4560,9437.1840,77.1611,67.9472,36.7933,5.2562
200,32x32x2,8,16384,1376.2560,393.2160,589.8240,4.6552,70.3899,35.5165,5.0738
200,32x32x4,8,32768,2752.5120,786.4320,1179.6480,9.3462,70.1208,35.6528,5.0933
200,32x32x8,8,65536,5505.0240,1572.8640,2359.2960,18.6614,70.2368,35.5939,5.0848
200,32x32x16,8,131072,11010.0480,3145.7280,4718.5920,37.7842,69.3793,36.0338,5.1477
200,32x32x32,8,262144,22020.0960,6291.4560,9437.1840,76.8843,68.1918,36.6613,5.2373
200,32x32x64,8,524288,44040.1920,12582.9120,18874.3680,155.0573,67.6251,36.9685,5.2812
200,32x64x2,8,32768,2752.5120,786.4320,1179.6480,9.3212,70.3083,35.5577,5.0797
200,32x64x4,8,65536,5505.0240,1572.8640,2359.2960,18.6546,70.2625,35.5808,5.0830
200,32x64x8,8,131072,11010.0480,3145.7280,4718.5920,37.8010,69.3485,36.0498,5.1500
200,32x64x16,8,262144,22020.0960,6291.4560,9437.1840,76.9422,68.1405,36.6889,5.2413
200,32x64x32,8,524288,44040.1920,12582.9120,18874.3680,155.7954,67.3047,37.1445,5.3064
200,32x64x64,8,1048576,88080.3840,25165.8240,37748.7360,311.0378,67.4244,37.0786,5.2969
200,64x4x2,8,4096,344.0640,98.3040,147.4560,1.1608,70.5721,35.4248,5.0607
200,64x4x4,8,8192,688.1280,196.6080,294.9120,2.3302,70.3130,35.5553,5.0793
200,64x4x8,8,16384,1376.2560,393.2160,589.8240,4.6540,70.4077,35.5075,5.0725
200,64x4x16,8,32768,2752.5120,786.4320,1179.6480,9.3333,70.2171,35.6039,5.0863
200,64x4x32,8,65536,5505.0240,1572.8640,2359.2960,18.6453,70.2976,35.5631,5.0804
200,64x4x64,8,131072,11010.0480,3145.7280,4718.5920,37.8258,69.3029,36.0735,5.1534
200,64x8x2,8,8192,688.1280,196.6080,294.9120,2.3290,70.3481,35.5375,5.0768
200,64x8x4,8,16384,1376.2560,393.2160,589.8240,4.6544,70.4028,35.5099,5.0728
200,64x8x8,8,32768,2752.5120,786.4320,1179.6480,9.3230,70.2951,35.5643,5.0806
200,64x8x16,8,65536,5505.0240,1572.8640,2359.2960,18.6754,70.1844,35.6204,5.0886
200,64x8x32,8,131072,11010.0480,3145.7280,4718.5920,37.7690,69.4072,36.0193,5.1456
200,64x8x64,8,262144,22020.0960,6291.4560,9437.1840,76.9719,68.1142,36.7030,5.2433
200,64x16x2,8,16384,1376.2560,393.2160,589.8240,4.6547,70.3970,35.5129,5.0733
200,64x16x4,8,32768,2752.5120,786.4320,1179.6480,9.3174,70.3375,35.5429,5.0776
200,64x16x8,8,65536,5505.0240,1572.8640,2359.2960,18.6668,70.2166,35.6041,5.0863
200,64x16x16,8,131072,11010.0480,3145.7280,4718.5920,37.7946,69.3602,36.0437,5.1491
200,64x16x32,8,262144,22020.0960,6291.4560,9437.1840,76.8550,68.2178,36.6473,5.2353
200,64x16x64,8,524288,44040.1920,12582.9120,18874.3680,155.0076,67.6467,36.9567,5.2795
200,64x32x2,8,32768,2752.5120,786.4320,1179.6480,9.3208,70.3114,35.5561,5.0794
200,64x32x4,8,65536,5505.0240,1572.8640,2359.2960,18.6700,70.2048,35.6101,5.0872
200,64x32x8,8,131072,11010.0480,3145.7280,4718.5920,37.7702,69.4050,36.0205,5.1458
200,64x32x16,8,262144,22020.0960,6291.4560,9437.1840,76.9311,68.1503,36.6836,5.2405
200,64x32x32,8,524288,44040.1920,12582.9120,18874.3680,155.0291,67.6374,36.9618,5.2803
200,64x32x64,8,1048576,88080.3840,25165.8240,37748.7360,311.1067,67.4094,37.0868,5.2981
200,64x64x2,8,65536,5505.0240,1572.8640,2359.2960,18.6574,70.2521,35.5861,5.0837
200,64x64x4,8,131072,11010.0480,3145.7280,4718.5920,37.7664,69.4120,36.0168,5.1453
200,64x64x8,8,262144,22020.0960,6291.4560,9437.1840,76.8818,68.1940,36.6601,5.2372
200,64x64x16,8,524288,44040.1920,12582.9120,18874.3680,155.6472,67.3688,37.1092,5.3013
200,64x64x32,8,1048576,88080.3840,25165.8240,37748.7360,311.2255,67.3837,37.1010,5.3001
200,64x64x64,8,2097152,176160.7680,50331.6480,75497.4720,622.2886,67.4013,37.0913,5.2988
200,4x4x2,16,512,59.3920,12.2880,34.8160,0.2845,35.9949,69.4543,4.6303
200,4x4x4,16,1024,118.7840,24.5760,69.6320,0.5659,36.1878,69.0841,4.6056
200,4x4x8,16,2048,237.5680,49.1520,139.2640,1.1282,36.3067,68.8579,4.5905
200,4x4x16,16,4096,475.1360,98.3040,278.5280,2.2700,36.0884,69.2744,4.6183
200,4x4x32,16,8192,950.2720,196.6080,557.0560,4.5076,36.3478,68.7800,4.5853
200,4x4x64,16,16384,1900.5440,393.2160,1114.1120,9.0420,36.2396,68.9854,4.5990
200,4x8x2,16,1024,118.7840,24.5760,69.6320,0.5647,36.2673,68.9326,4.5955
200,4x8x4,16,2048,237.5680,49.1520,139.2640,1.1283,36.3018,68.8672,4.5911
200,4x8x8,16,4096,475.1360,98.3040,278.5280,2.2682,36.1169,69.2197,4.6146
200,4x8x16,16,8192,950.2720,196.6080,557.0560,4.5107,36.3223,68.8283,4.5886
200,4x8x32,16,16384,1900.5440,393.2160,1114.1120,9.0316,36.2814,68.9058,4.5937
200,4x8x64,16,32768,3801.0880,786.4320,2228.2240,18.1464,36.1151,69.2232,4.6149
200,4x16x2,16,2048,237.5680,49.1520,139.2640,1.1343,36.1088,69.2352,4.6157
200,4x16x4,16,4096,475.1360,98.3040,278.5280,2.2682,36.1174,69.2187,4.6146
200,4x16x8,16,8192,950.2720,196.6080,557.0560,4.5095,36.3319,68.8100,4.5873
200,4x16x16,16,16384,1900.5440,393.2160,1114.1120,9.0313,36.2828,68.9031,4.5935
200,4x16x32,16,32768,3801.0880,786.4320,2228.2240,18.0596,36.2887,68.8921,4.5928
200,4x16x64,16,65536,7602.1760,1572.8640,4456.4480,36.2157,36.1920,69.0760,4.6051
200,4x32x2,16,4096,475.1360,98.3040,278.5280,2.2771,35.9759,69.4911,4.6327
200,4x32x4,16,8192,950.2720,196.6080,557.0560,4.5199,36.2489,68.9676,4.5978
200,4x32x8,16,16384,1900.5440,393.2160,1114.1120,9.1423,35.8423,69.7501,4.6500
200,4x32x16,16,32768,3801.0880,786.4320,2228.2240,18.0634,36.2812,68.9063,4.5938
200,4x32x32,16,65536,7602.1760,1572.8640,4456.4480,36.5631,35.8481,69.7386,4.6492
200,4x32x64,16,131072,15204.3520,3145.7280,8912.8960,73.5555,35.6389,70.1480,4.6765
200,4x64x2,16,8192,950.2720,196.6080,557.0560,4.5115,36.3160,68.8402,4.5893
200,4x64x4,16,16384,1900.5440,393.2160,1114.1120,9.0323,36.2786,68.9111,4.5941
200,4x64x8,16,32768,3801.0880,786.4320,2228.2240,18.0756,36.2566,68.9529,4.5969
200,4x64x16,16,65536,7602.1760,1572.8640,4456.4480,36.2052,36.2025,69.0559,4.6037
200,4x64x32,16,131072,15204.3520,3145.7280,8912.8960,73.4933,35.6691,70.0887,4.6726
200,4x64x64,16,262144,30408.7040,6291.4560,17825.7920,148.5317,35.2981,70.8254,4.7217
200,8x4x2,16,1024,118.7840,24.5760,69.6320,0.5665,36.1522,69.1521,4.6101
200,8x4x4,16,2048,237.5680,49.1520,139.2640,1.1390,35.9629,69.5161,4.6344
200,8x4x8,16,4096,475.1360,98.3040,278.5280,2.2692,36.1002,69.2516,4.6168
200,8x4x16,16,8192,950.2720,196.6080,557.0560,4.5118,36.3140,68.8439,4.5896
200,8x4x32,16,16384,1900.5440,393.2160,1114.1120,9.0309,36.2842,68.9004,4.5934
200,8x4x64,16,32768,3801.0880,786.4320,2228.2240,18.0565,36.2949,68.8802,4.5920
200,8x8x2,16,2048,237.5680,49.1520,139.2640,1.1283,36.3024,68.8660,4.5911
200,8x8x4,16,4096,475.1360,98.3040,278.5280,2.2735,36.0331,69.3807,4.6254
200,8x8x8,16,8192,950.2720,196.6080,557.0560,4.5120,36.3122,68.8474,4.5898
200,8x8x16,16,16384,1900.5440,393.2160,1114.1120,9.0231,36.3155,68.8411,4.5894
200,8x8x32,16,32768,3801.0880,786.4320,2228.2240,18.0499,36.3082,68.8550,4.5903
200,8x8x64,16,65536,7602.1760,1572.8640,4456.4480,36.1667,36.2410,68.9826,4.5988
200,8x16x2,16,4096,475.1360,98.3040,278.5280,2.2698,36.0910,69.2694,4.6180
200,8x16x4,16,8192,950.2720,196.6080,557.0560,4.5133,36.3020,68.8668,4.5911
200,8x16x8,16,16384,1900.5440,393.2160,1114.1120,9.0286,36.2934,68.8830,4.5922
200,8x16x16,16,32768,3801.0880,786.4320,2228.2240,18.0824,36.2429,68.9789,4.5986
200,8x16x32,16,65536,7602.1760,1572.8640,4456.4480,36.1922,36.2156,69.0311,4.6021
200,8x16x64,16,131072,15204.3520,3145.7280,8912.8960,73.6442,35.5960,70.2325,4.6822
200,8x32x2,16,8192,950.2720,196.6080,557.0560,4.5166,36.2750,68.9179,4.5945
200,8x32x4,16,16384,1900.5440,393.2160,1114.1120,9.0241,36.3116,68.8485,4.5899
200,8x32x8,16,32768,3801.0880,786.4320,2228.2240,18.0582,36.2915,68.8867,4.5924
200,8x32x16,16,65536,7602.1760,1572.8640,4456.4480,36.6601,35.7534,69.9235,4.6616
200,8x32x32,16,131072,15204.3520,3145.7280,8912.8960,73.6473,35.5945,70.2356,4.6824
200,8x32x64,16,262144,30408.7040,6291.4560,17825.7920,148.6287,35.2750,70.8717,4.7248
200,8x64x2,16,16384,1900.5440,393.2160,1114.1120,9.0336,36.2733,68.9212,4.5947
200,8x64x4,16,32768,3801.0880,786.4320,2228.2240,18.0749,36.2580,68.9503,4.5967
200,8x64x8,16,65536,7602.1760,1572.8640,4456.4480,36.3902,36.0185,69.4088,4.6273
200,8x64x16,16,131072,15204.3520,3145.7280,8912.8960,73.5268,35.6528,70.1206,4.6747
200,8x64x32,16,262144,30408.7040,6291.4560,17825.7920,148.3429,35.3430,70.7354,4.7157
200,8x64x64,16,524288,60817.4080,12582.9120,35651.5840,297.9615,35.1917,71.0396,4.7360
200,16x4x2,16,2048,237.5680,49.1520,139.2640,1.1309,36.2181,69.0264,4.6018
200,16x4x4,16,4096,475.1360,98.3040,278.5280,2.2770,35.9779,69.4871,4.6325
200,16x4x8,16,8192,950.2720,196.6080,557.0560,4.5217,36.2345,68.9949,4.5997
200,16x4x16,16,16384,1900.5440,393.2160,1114.1120,9.0326,36.2776,68.9131,4.5942
200,16x4x32,16,32768,3801.0880,786.4320,2228.2240,18.0961,36.2155,69.0313,4.6021
200,16x4x64,16,65536,7602.1760,1572.8640,4456.4480,36.2045,36.2032,69.0547,4.6036
200,16x8x2,16,4096,475.1360,98.3040,278.5280,2.2714,36.0657,69.3179,4.6212
200,16x8x4,16,8192,950.2720,196.6080,557.0560,4.5179,36.2648,68.9374,4.5958
200,16x8x8,16,16384,1900.5440,393.2160,1114.1120,9.0243,36.3109,68.8498,4.5900
200,16x8x16,16,32768,3801.0880,786.4320,2228.2240,18.0528,36.3024,68.8660,4.5911
200,16x8x32,16,65536,7602.1760,1572.8640,4456.4480,36.1828,36.2250,69.0131,4.6009
200,16x8x64,16,131072,15204.3520,3145.7280,8912.8960,73.5654,35.6341,70.1574,4.6772
200,16x16x2,16,8192,950.2720,196.6080,557.0560,4.5168,36.2739,68.9201,4.5947
200,16x16x4,16,16384,1900.5440,393.2160,1114.1120,9.0385,36.2536,68.9586,4.5972
200,16x16x8,16,32768,3801.0880,786.4320,2228.2240,18.0888,36.2302,69.0032,4.6002
200,16x16x16,16,65536,7602.1760,1572.8640,4456.4480,36.3237,36.0844,69.2819,4.6188
200,16x16x32,16,131072,15204.3520,3145.7280,8912.8960,73.5941,35.6202,70.1848,4.6790
200,16x16x64,16,262144,30408.7040,6291.4560,17825.7920,148.4666,35.3135,70.7944,4.7196
200,16x32x2,16,16384,1900.5440,393.2160,1114.1120,9.0241,36.3117,68.8482,4.5899
200,16x32x4,16,32768,3801.0880,786.4320,2228.2240,18.0509,36.3063,68.8585,4.5906
200,16x32x8,16,65536,7602.1760,1572.8640,4456.4480,36.3694,36.0391,69.3692,4.6246
200,16x32x16,16,131072,15204.3520,3145.7280,8912.8960,73.6294,35.6032,70.2185,4.6812
200,16x32x32,16,262144,30408.7040,6291.4560,17825.7920,148.3355,35.3448,70.7319,4.7155
200,16x32x64,16,524288,60817.4080,12582.9120,35651.5840,297.8748,35.2019,71.0189,4.7346
200,16x64x2,16,32768,3801.0880,786.4320,2228.2240,18.0701,36.2676,68.9320,4.5955
200,16x64x4,16,65536,7602.1760,1572.8640,4456.4480,36.3114,36.0966,69.2585,4.6172
200,16x64x8,16,131072,15204.3520,3145.7280,8912.8960,73.5786,35.6278,70.1700,4.6780
200,16x64x16,16,262144,30408.7040,6291.4560,17825.7920,148.4242,35.3236,70.7742,4.7183
200,16x64x32,16,524288,60817.4080,12582.9120,35651.5840,297.8513,35.2047,71.0133,4.7342
200,16x64x64,16,1048576,121634.8160,25165.8240,71303.1680,597.7505,35.0841,71.2574,4.7505
200,32x4x2,16,4096,475.1360,98.3040,278.5280,2.2755,36.0003,69.4438,4.6296
200,32x4x4,16,8192,950.2720,196.6080,557.0560,4.5120,36.3122,68.8474,4.5898
200,32x4x8,16,16384,1900.5440,393.2160,1114.1120,9.0302,36.2872,68.8948,4.5930
200,32x4x16,16,32768,3801.0880,786.4320,2228.2240,18.0486,36.3108,68.8499,4.5900
200,32x4x32,16,65536,7602.1760,1572.8640,4456.4480,36.5587,35.8524,69.7302,4.6487
200,32x4x64,16,131072,15204.3520,3145.7280,8912.8960,73.5152,35.6585,70.1096,4.6740
200,32x8x2,16,8192,950.2720,196.6080,557.0560,4.5142,36.2943,68.8814,4.5921
200,32x8x4,16,16384,1900.5440,393.2160,1114.1120,9.0295,36.2901,68.8893,4.5926
200,32x8x8,16,32768,3801.0880,786.4320,2228.2240,18.0602,36.2875,68.8943,4.5930
200,32x8x16,16,65536,7602.1760,1572.8640,4456.4480,36.1865,36.2213,69.0202,4.6013
200,32x8x32,16,131072,15204.3520,3145.7280,8912.8960,73.5540,35.6397,70.1466,4.6764
200,32x8x64,16,262144,30408.7040,6291.4560,17825.7920,148.6246,35.2760,70.8697,4.7246
200,32x16x2,16,16384,1900.5440,393.2160,1114.1120,9.0264,36.3024,68.8660,4.5911
200,32x16x4,16,32768,3801.0880,786.4320,2228.2240,18.1538,36.1004,69.2513,4.6168
200,32x16x8,16,65536,7602.1760,1572.8640,4456.4480,36.1831,36.2247,69.0138,4.6009
200,32x16x16,16,131072,15204.3520,3145.7280,8912.8960,73.5389,35.6470,70.1321,4.6755
200,32x16x32,16,262144,30408.7040,6291.4560,17825.7920,148.3446,35.3426,70.7362,4.7157
200,32x16x64,16,524288,60817.4080,12582.9120,35651.5840,297.8388,35.2062,71.0103,4.7340
200,32x32x2,16,32768,3801.0880,786.4320,2228.2240,18.0716,36.2646,68.9377,4.5958
200,32x32x4,16,65536,7602.1760,1572.8640,4456.4480,36.1953,36.2124,69.0371,4.6025
200,32x32x8,16,131072,15204.3520,3145.7280,8912.8960,73.6102,35.6124,70.2002,4.6800
200,32x32x16,16,262144,30408.7040,6291.4560,17825.7920,148.5063,35.3041,70.8133,4.7209
200,32x32x32,16,524288,60817.4080,12582.9120,35651.5840,297.8050,35.2102,71.0022,4.7335
200,32x32x64,16,1048576,121634.8160,25165.8240,71303.1680,597.1303,35.1205,71.1835,4.7456
200,32x64x2,16,65536,7602.1760,1572.8640,4456.4480,36.2228,36.1850,69.0895,4.6060
200,32x64x4,16,131072,15204.3520,3145.7280,8912.8960,73.5889,35.6228,70.1798,4.6787
200,32x64x8,16,262144,30408.7040,6291.4560,17825.7920,148.2803,35.3579,70.7056,4.7137
200,32x64x16,16,524288,60817.4080,12582.9120,35651.5840,298.5905,35.1175,71.1895,4.7460
200,32x64x32,16,1048576,121634.8160,25165.8240,71303.1680,596.9741,35.1297,71.1649,4.7443
200,32x64x64,16,2097152,243269.6320,50331.6480,142606.3360,1198.2872,35.0025,71.4235,4.7616
200,64x4x2,16,8192,950.2720,196.6080,557.0560,4.5187,36.2584,68.9496,4.5966
200,64x4x4,16,16384,1900.5440,393.2160,1114.1120,9.0256,36.3056,68.8599,4.5907
200,64x4x8,16,32768,3801.0880,786.4320,2228.2240,18.0590,36.2898,68.8898,4.5927
200,64x4x16,16,65536,7602.1760,1572.8640,4456.4480,36.1871,36.2207,69.0214,4.6014
200,64x4x32,16,131072,15204.3520,3145.7280,8912.8960,73.5339,35.6494,70.1274,4.6752
200,64x4x64,16,262144,30408.7040,6291.4560,17825.7920,148.3662,35.3374,70.7465,4.7164
200,64x8x2,16,16384,1900.5440,393.2160,1114.1120,9.1411,35.8467,69.7414,4.6494
200,64x8x4,16,32768,3801.0880,786.4320,2228.2240,18.2628,35.8849,69.6672,4.6445
200,64x8x8,16,65536,7602.1760,1572.8640,4456.4480,36.2778,36.1301,69.1944,4.6130
200,64x8x16,16,131072,15204.3520,3145.7280,8912.8960,73.8568,35.4935,70.4354,4.6957
200,64x8x32,16,262144,30408.7040,6291.4560,17825.7920,148.6318,35.2743,70.8732,4.7249
200,64x8x64,16,524288,60817.4080,12582.9120,35651.5840,297.9060,35.1982,71.0263,4.7351
200,64x16x2,16,32768,3801.0880,786.4320,2228.2240,18.0935,36.2208,69.0211,4.6014
200,64x16x4,16,65536,7602.1760,1572.8640,4456.4480,36.2339,36.1739,69.1106,4.6074
200,64x16x8,16,131072,15204.3520,3145.7280,8912.8960,73.5469,35.6431,70.1398,4.6760
200,64x16x16,16,262144,30408.7040,6291.4560,17825.7920,148.6804,35.2628,70.8963,4.7264
200,64x16x32,16,524288,60817.4080,12582.9120,35651.5840,298.4753,35.1311,71.1621,4.7441
200,64x16x64,16,1048576,121634.8160,25165.8240,71303.1680,596.9407,35.1317,71.1609,4.7441
200,64x32x2,16,65536,7602.1760,1572.8640,4456.4480,36.1765,36.2312,69.0012,4.6001
200,64x32x4,16,131072,15204.3520,3145.7280,8912.8960,73.5663,35.6337,70.1583,4.6772
200,64x32x8,16,262144,30408.7040,6291.4560,17825.7920,148.5674,35.2896,70.8424,4.7228
200,64x32x16,16,524288,60817.4080,12582.9120,35651.5840,297.8978,35.1992,71.0244,4.7350
200,64x32x32,16,1048576,121634.8160,25165.8240,71303.1680,597.0268,35.1266,71.1711,4.7447
200,64x32x64,16,2097152,243269.6320,50331.6480,142606.3360,1193.9364,35.1300,71.1642,4.7443
200,64x64x2,16,131072,15204.3520,3145.7280,8912.8960,73.6033,35.6158,70.1935,4.6796
200,64x64x4,16,262144,30408.7040,6291.4560,17825.7920,148.5687,35.2893,70.8431,4.7229
200,1x1x1,8,8,0.6720,0.1920,0.2880,0.0023,70.5477,35.4370,5.0624
200,1x1x2,8,16,1.3440,0.3840,0.5760,0.0045,70.3495,35.5368,5.0767
200,1x2x1,8,16,1.3440,0.3840,0.5760,0.0046,70.0866,35.6701,5.0957
200,1x2x2,8,32,2.6880,0.7680,1.1520,0.0092,69.8415,35.7953,5.1136
200,2x1x1,8,16,1.3440,0.3840,0.5760,0.0046,70.0568,35.6853,5.0979
200,2x1x2,8,32,2.6880,0.7680,1.1520,0.0091,70.4164,35.5031,5.0719
200,2x2x1,8,32,2.6880,0.7680,1.1520,0.0092,69.8992,35.7658,5.1094
200,2x2x2,8,64,5.3760,1.5360,2.3040,0.0181,70.8343,35.2936,5.0419
200,1x1x1,16,16,1.8560,0.3840,1.0880,0.0089,36.0799,69.2906,4.6194
200,1x1x2,16,32,3.7120,0.7680,2.1760,0.0178,36.0316,69.3836,4.6256
200,1x2x1,16,32,3.7120,0.7680,2.1760,0.0178,36.0079,69.4292,4.6286
200,1x2x2,16,64,7.4240,1.5360,4.3520,0.0355,36.0231,69.4000,4.6267
200,2x1x1,16,32,3.7120,0.7680,2.1760,0.0178,36.0312,69.3843,4.6256
200,2x1x2,16,64,7.4240,1.5360,4.3520,0.0357,35.9018,69.6344,4.6423
200,2x2x1,16,64,7.4240,1.5360,4.3520,0.0355,36.0121,69.4211,4.6281
200,2x2x2,16,128,14.8480,3.0720,8.7040,0.0712,35.9431,69.5544,4.6370
200,1x1x1,32,32,5.7600,0.7680,4.2240,0.0347,18.4608,135.4223,4.3685
200,1x1x2,32,64,11.5200,1.5360,8.4480,0.0693,18.4827,135.2617,4.3633
200,1x2x1,32,64,11.5200,1.5360,8.4480,0.0692,18.4850,135.2449,4.3627
200,1x2x2,32,128,23.0400,3.0720,16.8960,0.1385,18.4893,135.2135,4.3617
200,2x1x1,32,64,11.5200,1.5360,8.4480,0.0693,18.4790,135.2890,4.3642
200,2x1x2,32,128,23.0400,3.0720,16.8960,0.1390,18.4216,135.7101,4.3777
200,2x2x1,32,128,23.0400,3.0720,16.8960,0.1385,18.4819,135.2674,4.3635
200,2x2x2,32,256,46.0800,6.1440,33.7920,0.2774,18.4575,135.4459,4.3692
200,1x1x1,64,64,19.7120,1.5360,16.6400,0.1343,9.5315,262.2869,4.1633
200,1x1x2,64,128,39.4240,3.0720,33.2800,0.2685,9.5328,262.2517,4.1627
200,1x2x1,64,128,39.4240,3.0720,33.2800,0.2680,9.5529,261.7008,4.1540
200,1x2x2,64,256,78.8480,6.1440,66.5600,0.5369,9.5358,262.1698,4.1614
200,2x1x1,64,128,39.4240,3.0720,33.2800,0.2680,9.5531,261.6939,4.1539
200,2x1x2,64,256,78.8480,6.1440,66.5600,0.5369,9.5354,262.1801,4.1616
200,2x2x1,64,256,78.8480,6.1440,66.5600,0.5358,9.5553,261.6343,4.1529
200,2x2x2,64,512,157.6960,12.2880,133.1200,1.0721,9.5509,261.7547,4.1548
200,1x1x1,128,128,72.1920,3.0720,66.0480,0.5453,4.6950,532.4769,4.1927
200,1x1x2,128,256,144.3840,6.1440,132.0960,1.0901,4.6966,532.2970,4.1913
200,1x2x1,128,256,144.3840,6.1440,132.0960,1.0882,4.7051,531.3436,4.1838
200,1x2x2,128,512,288.7680,12.2880,264.1920,2.1802,4.6968,532.2760,4.1911
200,2x1x1,128,256,144.3840,6.1440,132.0960,1.0883,4.7046,531.3984,4.1842
200,2x1x2,128,512,288.7680,12.2880,264.1920,2.1759,4.7061,531.2260,4.1829
200,2x2x1,128,512,288.7680,12.2880,264.1920,2.1761,4.7057,531.2684,4.1832
200,2x2x2,128,1024,577.5360,24.5760,528.3840,4.3527,4.7051,531.3391,4.1838
200,1x1x1,256,256,275.4560,6.1440,263.1680,2.1305,2.4032,1040.2863,4.0796
200,1x1x2,256,512,550.9120,12.2880,526.3360,4.2607,2.4034,1040.2135,4.0793
200,1x2x1,256,512,550.9120,12.2880,526.3360,4.2612,2.4031,1040.3325,4.0797
200,1x2x2,256,1024,1101.8240,24.5760,1052.6720,8.5423,2.3975,1042.7643,4.0893
200,2x1x1,256,512,550.9120,12.2880,526.3360,4.2616,2.4029,1040.4180,4.0801
200,2x1x2,256,1024,1101.8240,24.5760,1052.6720,8.5271,2.4018,1040.9040,4.0820
200,2x2x1,256,1024,1101.8240,24.5760,1052.6720,8.5308,2.4007,1041.3556,4.0837
200,2x2x2,256,2048,2203.6480,49.1520,2105.3440,17.0679,2.3998,1041.7399,4.0853
200,1x1x1,512,512,1075.2000,12.2880,1050.6240,8.4481,1.2121,2062.5241,4.0363
200,1x1x2,512,1024,2150.4000,24.5760,2101.2480,16.8998,1.2118,2062.9650,4.0371
200,1x2x1,512,1024,2150.4000,24.5760,2101.2480,16.9053,1.2115,2063.6406,4.0384
200,1x2x2,512,2048,4300.8000,49.1520,4202.4960,33.8109,1.2114,2063.6513,4.0385
200,2x1x1,512,1024,2150.4000,24.5760,2101.2480,16.8985,1.2119,2062.8109,4.0368
200,2x1x2,512,2048,4300.8000,49.1520,4202.4960,33.7934,1.2121,2062.5859,4.0364
200,2x2x1,512,2048,4300.8000,49.1520,4202.4960,33.7967,1.2120,2062.7883,4.0368
200,2x2x2,512,4096,8601.6000,98.3040,8404.9920,67.6227,1.2114,2063.6823,4.0385
200,1x1x1,1024,1024,4247.5520,24.5760,4198.4000,33.6446,0.6087,4107.0054,4.0147
200,1x1x2,1024,2048,8495.1040,49.1520,8396.8000,67.3630,0.6080,4111.5120,4.0191
200,1x2x1,1024,2048,8495.1040,49.1520,8396.8000,67.3655,0.6080,4111.6644,4.0192
200,1x2x2,1024,4096,16990.2080,98.3040,16793.6000,134.7595,0.6079,4112.5325,4.0201
200,2x1x1,1024,2048,8495.1040,49.1520,8396.8000,67.3755,0.6079,4112.2737,4.0198
200,2x1x2,1024,4096,16990.2080,98.3040,16793.6000,134.6290,0.6085,4108.5514,4.0162
200,2x2x1,1024,4096,16990.2080,98.3040,16793.6000,134.7230,0.6081,4111.4207,4.0190
200,2x2x2,1024,8192,33980.4160,196.6080,33587.2000,269.4353,0.6081,4111.2566,4.0188
200,1x1x1,2048,2048,16883.7120,49.1520,16785.4080,137.8137,0.2972,8411.4836,4.1092
200,1x1x1,4096,4096,67321.8560,98.3040,67125.2480,546.8174,0.1498,16687.5438,4.0751
200,1x1x1,8192,8192,268861.4400,196.6080,268468.2240,2211.8457,0.0741,33750.0866,4.1204

142
scripts/results_soa_casclakesp2_iln1000.txt
View File

@ -1,142 +0,0 @@
200,4x4x2,8,256,21.5040,6.1440,9.2160,0.7405,69.1424,34.7110,4.9587
200,4x4x4,8,512,43.0080,12.2880,18.4320,1.4818,69.1066,34.7290,4.9613
200,4x4x8,8,1024,86.0160,24.5760,36.8640,2.9627,69.1261,34.7191,4.9599
200,4x4x16,8,2048,172.0320,49.1520,73.7280,5.9244,69.1378,34.7133,4.9590
200,4x8x2,8,512,43.0080,12.2880,18.4320,1.4813,69.1298,34.7173,4.9596
200,4x8x4,8,1024,86.0160,24.5760,36.8640,2.9633,69.1126,34.7259,4.9608
200,4x8x8,8,2048,172.0320,49.1520,73.7280,5.9254,69.1265,34.7189,4.9598
200,4x8x16,8,4096,344.0640,98.3040,147.4560,11.8621,69.0600,34.7524,4.9646
200,4x16x2,8,1024,86.0160,24.5760,36.8640,2.9705,68.9438,34.8110,4.9730
200,4x16x4,8,2048,172.0320,49.1520,73.7280,5.9338,69.0288,34.7681,4.9669
200,4x16x8,8,4096,344.0640,98.3040,147.4560,11.8492,69.1354,34.7145,4.9592
200,4x16x16,8,8192,688.1280,196.6080,294.9120,23.7239,69.0612,34.7518,4.9645
200,8x4x2,8,512,43.0080,12.2880,18.4320,1.4814,69.1239,34.7203,4.9600
200,8x4x4,8,1024,86.0160,24.5760,36.8640,2.9622,69.1383,34.7130,4.9590
200,8x4x8,8,2048,172.0320,49.1520,73.7280,5.9256,69.1234,34.7205,4.9601
200,8x4x16,8,4096,344.0640,98.3040,147.4560,11.8548,69.1026,34.7310,4.9616
200,8x8x2,8,1024,86.0160,24.5760,36.8640,2.9625,69.1319,34.7162,4.9595
200,8x8x4,8,2048,172.0320,49.1520,73.7280,5.9247,69.1339,34.7152,4.9593
200,8x8x8,8,4096,344.0640,98.3040,147.4560,11.8484,69.1400,34.7122,4.9589
200,8x8x16,8,8192,688.1280,196.6080,294.9120,23.7201,69.0721,34.7463,4.9638
200,8x16x2,8,2048,172.0320,49.1520,73.7280,5.9244,69.1376,34.7134,4.9591
200,8x16x4,8,4096,344.0640,98.3040,147.4560,11.8496,69.1331,34.7156,4.9594
200,8x16x8,8,8192,688.1280,196.6080,294.9120,23.7200,69.0724,34.7461,4.9637
200,8x16x16,8,16384,1376.2560,393.2160,589.8240,47.4404,-21.4620,-111.8254,-15.9751
200,16x4x2,8,1024,86.0160,24.5760,36.8640,2.9624,69.1338,34.7153,4.9593
200,16x4x4,8,2048,172.0320,49.1520,73.7280,5.9251,69.1300,34.7172,4.9596
200,16x4x8,8,4096,344.0640,98.3040,147.4560,11.8495,69.1335,34.7155,4.9594
200,16x4x16,8,8192,688.1280,196.6080,294.9120,23.7169,69.0814,34.7416,4.9631
200,16x8x2,8,2048,172.0320,49.1520,73.7280,5.9246,69.1360,34.7142,4.9592
200,16x8x4,8,4096,344.0640,98.3040,147.4560,11.8498,69.1319,34.7163,4.9595
200,16x8x8,8,8192,688.1280,196.6080,294.9120,23.7234,69.0625,34.7511,4.9644
200,16x8x16,8,16384,1376.2560,393.2160,589.8240,47.4499,-21.4578,-111.8477,-15.9782
200,16x16x2,8,4096,344.0640,98.3040,147.4560,11.8552,69.1005,34.7320,4.9617
200,16x16x4,8,8192,688.1280,196.6080,294.9120,23.7215,69.0682,34.7483,4.9640
200,4x4x2,8,256,21.5040,6.1440,9.2160,0.7406,69.1332,34.7156,4.9594
200,4x4x4,8,512,43.0080,12.2880,18.4320,1.4813,69.1306,34.7169,4.9596
200,4x4x8,8,1024,86.0160,24.5760,36.8640,2.9622,69.1383,34.7130,4.9590
200,4x4x16,8,2048,172.0320,49.1520,73.7280,5.9253,69.1267,34.7188,4.9598
200,4x8x2,8,512,43.0080,12.2880,18.4320,1.4813,69.1298,34.7173,4.9596
200,4x8x4,8,1024,86.0160,24.5760,36.8640,2.9625,69.1308,34.7168,4.9595
200,4x8x8,8,2048,172.0320,49.1520,73.7280,5.9247,69.1340,34.7152,4.9593
200,4x8x16,8,4096,344.0640,98.3040,147.4560,11.8482,69.1412,34.7116,4.9588
200,4x16x2,8,1024,86.0160,24.5760,36.8640,2.9625,69.1310,34.7167,4.9595
200,4x16x4,8,2048,172.0320,49.1520,73.7280,5.9254,69.1263,34.7191,4.9599
200,4x16x8,8,4096,344.0640,98.3040,147.4560,11.8488,69.1375,34.7134,4.9591
200,4x16x16,8,8192,688.1280,196.6080,294.9120,23.7265,69.0536,34.7556,4.9651
200,8x4x2,8,512,43.0080,12.2880,18.4320,1.4814,69.1244,34.7200,4.9600
200,8x4x4,8,1024,86.0160,24.5760,36.8640,2.9622,69.1375,34.7134,4.9591
200,8x4x8,8,2048,172.0320,49.1520,73.7280,5.9251,69.1301,34.7172,4.9596
200,8x4x16,8,4096,344.0640,98.3040,147.4560,11.8497,69.1326,34.7159,4.9594
200,8x8x2,8,1024,86.0160,24.5760,36.8640,2.9623,69.1364,34.7140,4.9591
200,8x8x4,8,2048,172.0320,49.1520,73.7280,5.9250,69.1311,34.7166,4.9595
200,8x8x8,8,4096,344.0640,98.3040,147.4560,11.8488,69.1378,34.7133,4.9590
200,8x8x16,8,8192,688.1280,196.6080,294.9120,23.7217,69.0677,34.7485,4.9641
200,8x16x2,8,2048,172.0320,49.1520,73.7280,5.9246,69.1355,34.7145,4.9592
200,8x16x4,8,4096,344.0640,98.3040,147.4560,11.8491,69.1358,34.7143,4.9592
200,8x16x8,8,8192,688.1280,196.6080,294.9120,23.7200,69.0725,34.7461,4.9637
200,8x16x16,8,16384,1376.2560,393.2160,589.8240,47.4484,-21.4584,-111.8442,-15.9777
200,16x4x2,8,1024,86.0160,24.5760,36.8640,2.9624,69.1343,34.7151,4.9593
200,16x4x4,8,2048,172.0320,49.1520,73.7280,5.9239,69.1432,34.7106,4.9587
200,16x4x8,8,4096,344.0640,98.3040,147.4560,11.8490,69.1367,34.7138,4.9591
200,16x4x16,8,8192,688.1280,196.6080,294.9120,23.7192,69.0749,34.7449,4.9636
200,16x8x2,8,2048,172.0320,49.1520,73.7280,5.9251,69.1301,34.7171,4.9596
200,16x8x4,8,4096,344.0640,98.3040,147.4560,11.8496,69.1329,34.7157,4.9594
200,16x8x8,8,8192,688.1280,196.6080,294.9120,23.7224,69.0655,34.7496,4.9642
200,16x8x16,8,16384,1376.2560,393.2160,589.8240,47.4415,-21.4615,-111.8280,-15.9754
200,16x16x2,8,4096,344.0640,98.3040,147.4560,11.8502,69.1297,34.7173,4.9596
200,16x16x4,8,8192,688.1280,196.6080,294.9120,23.7177,69.0792,34.7428,4.9633
200,16x16x8,8,16384,1376.2560,393.2160,589.8240,47.4485,-21.4584,-111.8445,-15.9778
200,16x16x16,8,32768,2752.5120,786.4320,1179.6480,94.8632,-21.4660,-111.8047,-15.9721
200,4x4x2,16,512,59.3920,12.2880,34.8160,2.8749,35.6181,67.3814,4.4921
200,4x4x4,16,1024,118.7840,24.5760,69.6320,5.7553,35.5849,67.4444,4.4963
200,4x4x8,16,2048,237.5680,49.1520,139.2640,11.5277,35.5317,67.5452,4.5030
200,4x4x16,16,4096,475.1360,98.3040,278.5280,23.0679,35.5125,67.5818,4.5055
200,4x8x2,16,1024,118.7840,24.5760,69.6320,5.7516,35.6076,67.4014,4.4934
200,4x8x4,16,2048,237.5680,49.1520,139.2640,11.5048,35.6026,67.4108,4.4941
200,4x8x8,16,4096,475.1360,98.3040,278.5280,23.0343,35.5643,67.4834,4.4989
200,4x8x16,16,8192,950.2720,196.6080,557.0560,46.1992,35.4638,67.6746,4.5116
200,4x16x2,16,2048,237.5680,49.1520,139.2640,11.5339,35.5126,67.5817,4.5054
200,4x16x4,16,4096,475.1360,98.3040,278.5280,23.0366,35.5607,67.4902,4.4993
200,4x16x8,16,8192,950.2720,196.6080,557.0560,46.1666,35.4888,67.6269,4.5085
200,4x16x16,16,16384,1900.5440,393.2160,1114.1120,92.3461,-11.0256,-217.6759,-14.5117
200,8x4x2,16,1024,118.7840,24.5760,69.6320,5.7504,35.6148,67.3877,4.4925
200,8x4x4,16,2048,237.5680,49.1520,139.2640,11.5363,35.5054,67.5954,4.5064
200,8x4x8,16,4096,475.1360,98.3040,278.5280,23.0449,35.5480,67.5143,4.5010
200,8x4x16,16,8192,950.2720,196.6080,557.0560,46.2153,35.4515,67.6981,4.5132
200,8x8x2,16,2048,237.5680,49.1520,139.2640,11.5019,35.6114,67.3942,4.4929
200,8x8x4,16,4096,475.1360,98.3040,278.5280,23.0126,35.5980,67.4196,4.4946
200,8x8x8,16,8192,950.2720,196.6080,557.0560,46.1823,35.4768,67.6499,4.5100
200,8x8x16,16,16384,1900.5440,393.2160,1114.1120,92.3762,-11.0220,-217.7469,-14.5165
200,8x16x2,16,4096,475.1360,98.3040,278.5280,23.0367,35.5606,67.4904,4.4994
200,8x16x4,16,8192,950.2720,196.6080,557.0560,46.1673,35.4883,67.6279,4.5085
200,8x16x8,16,16384,1900.5440,393.2160,1114.1120,92.4053,-11.0185,-217.8156,-14.5210
200,8x16x16,16,32768,3801.0880,786.4320,2228.2240,184.6536,-11.0279,-217.6306,-14.5087
200,16x4x2,16,2048,237.5680,49.1520,139.2640,11.4984,35.6223,67.3736,4.4916
200,16x4x4,16,4096,475.1360,98.3040,278.5280,23.0553,35.5319,67.5449,4.5030
200,16x4x8,16,8192,950.2720,196.6080,557.0560,46.1639,35.4909,67.6229,4.5082
200,16x4x16,16,16384,1900.5440,393.2160,1114.1120,92.3485,-11.0253,-217.6818,-14.5121
200,16x8x2,16,4096,475.1360,98.3040,278.5280,23.0221,35.5832,67.4475,4.4965
200,16x8x4,16,8192,950.2720,196.6080,557.0560,46.1950,35.4671,67.6684,4.5112
200,16x8x8,16,16384,1900.5440,393.2160,1114.1120,92.3825,-11.0212,-217.7618,-14.5175
200,16x8x16,16,32768,3801.0880,786.4320,2228.2240,184.4897,-11.0377,-217.4373,-14.4958
200,16x16x2,16,8192,950.2720,196.6080,557.0560,46.1887,35.4719,67.6592,4.5106
200,16x16x4,16,16384,1900.5440,393.2160,1114.1120,92.4000,-11.0191,-217.8031,-14.5202
200,16x16x8,16,32768,3801.0880,786.4320,2228.2240,184.6561,-11.0277,-217.6335,-14.5089
200,16x16x16,16,65536,7602.1760,1572.8640,4456.4480,369.2723,0.6020,3986.7795,265.7853
200,4x4x2,32,1024,184.3200,24.5760,135.1680,11.4223,17.9299,133.8547,4.3179
200,4x4x4,32,2048,368.6400,49.1520,270.3360,22.8232,17.9466,133.7300,4.3139
200,4x4x8,32,4096,737.2800,98.3040,540.6720,45.6452,17.9471,133.7263,4.3138
200,4x4x16,32,8192,1474.5600,196.6080,1081.3440,91.3233,17.9407,133.7744,4.3153
200,4x8x2,32,2048,368.6400,49.1520,270.3360,22.8306,17.9408,133.7732,4.3153
200,4x8x4,32,4096,737.2800,98.3040,540.6720,45.6485,17.9458,133.7357,4.3141
200,4x8x8,32,8192,1474.5600,196.6080,1081.3440,91.3158,17.9421,133.7634,4.3149
200,4x8x16,32,16384,2949.1200,393.2160,2162.6880,182.6220,-5.5753,-430.4722,-13.8862
200,4x16x2,32,4096,737.2800,98.3040,540.6720,45.6489,17.9457,133.7371,4.3141
200,4x16x4,32,8192,1474.5600,196.6080,1081.3440,91.2890,17.9474,133.7241,4.3137
200,4x16x8,32,16384,2949.1200,393.2160,2162.6880,182.6145,-5.5755,-430.4547,-13.8856
200,4x16x16,32,32768,5898.2400,786.4320,4325.3760,365.2677,-5.5749,-430.5002,-13.8871
200,8x4x2,32,2048,368.6400,49.1520,270.3360,22.8301,17.9412,133.7701,4.3152
200,8x4x4,32,4096,737.2800,98.3040,540.6720,45.6689,17.9378,133.7956,4.3160
200,8x4x8,32,8192,1474.5600,196.6080,1081.3440,91.3143,17.9424,133.7611,4.3149
200,8x4x16,32,16384,2949.1200,393.2160,2162.6880,182.6332,-5.5749,-430.4987,-13.8871
200,8x8x2,32,4096,737.2800,98.3040,540.6720,45.6519,17.9445,133.7457,4.3144
200,8x8x4,32,8192,1474.5600,196.6080,1081.3440,91.3215,17.9410,133.7717,4.3152
200,8x8x8,32,16384,2949.1200,393.2160,2162.6880,182.6451,-5.5746,-430.5267,-13.8880
200,8x8x16,32,32768,5898.2400,786.4320,4325.3760,365.1989,-5.5760,-430.4192,-13.8845
200,8x16x2,32,8192,1474.5600,196.6080,1081.3440,91.3015,17.9449,133.7425,4.3143
200,8x16x4,32,16384,2949.1200,393.2160,2162.6880,182.6251,-5.5752,-430.4795,-13.8864
200,8x16x8,32,32768,5898.2400,786.4320,4325.3760,365.2087,-5.5758,-430.4307,-13.8849
200,8x16x16,32,65536,11796.4800,1572.8640,8650.7520,730.5781,0.3043,7887.5501,254.4371
200,16x4x2,32,4096,737.2800,98.3040,540.6720,45.6592,17.9416,133.7671,4.3151
200,16x4x4,32,8192,1474.5600,196.6080,1081.3440,91.3092,17.9434,133.7537,4.3146
200,16x4x8,32,16384,2949.1200,393.2160,2162.6880,182.6471,-5.5745,-430.5315,-13.8881
200,16x4x16,32,32768,5898.2400,786.4320,4325.3760,365.2706,-5.5749,-430.5036,-13.8872
200,16x8x2,32,8192,1474.5600,196.6080,1081.3440,91.3464,17.9361,133.8082,4.3164
200,16x8x4,32,16384,2949.1200,393.2160,2162.6880,182.6559,-5.5742,-430.5521,-13.8888
200,16x8x8,32,32768,5898.2400,786.4320,4325.3760,365.2269,-5.5755,-430.4521,-13.8856
200,16x8x16,32,65536,11796.4800,1572.8640,8650.7520,730.5707,0.3043,7887.4697,254.4345
200,16x16x2,32,16384,2949.1200,393.2160,2162.6880,182.6406,-5.5747,-430.5160,-13.8876
200,16x16x4,32,32768,5898.2400,786.4320,4325.3760,365.2364,-5.5754,-430.4633,-13.8859
200,16x16x8,32,65536,11796.4800,1572.8640,8650.7520,730.5352,0.3043,7887.0871,254.4222
200,16x16x16,32,131072,23592.9600,3145.7280,17301.5040,1461.1927,0.3043,7887.7469,254.4434

450
scripts/results_soa_skylakesp2.txt
View File

@ -1,450 +0,0 @@
200,4x4x2,8,256,21.5040,6.1440,9.2160,0.0016,31.1175,106.0818,15.1545
200,4x4x4,8,512,43.0080,12.2880,18.4320,0.0035,29.3957,112.2953,16.0422
200,4x4x8,8,1024,86.0160,24.5760,36.8640,0.0070,29.1148,113.3787,16.1970
200,4x4x16,8,2048,172.0320,49.1520,73.7280,0.0143,28.6699,115.1380,16.4483
200,4x4x32,8,4096,344.0640,98.3040,147.4560,0.0279,29.3329,112.5357,16.0765
200,4x4x64,8,8192,688.1280,196.6080,294.9120,0.0589,27.8187,118.6611,16.9516
200,4x8x2,8,512,43.0080,12.2880,18.4320,0.0034,30.1387,109.5268,15.6467
200,4x8x4,8,1024,86.0160,24.5760,36.8640,0.0070,29.1126,113.3874,16.1982
200,4x8x8,8,2048,172.0320,49.1520,73.7280,0.0138,29.7694,110.8857,15.8408
200,4x8x16,8,4096,344.0640,98.3040,147.4560,0.0283,28.9748,113.9266,16.2752
200,4x8x32,8,8192,688.1280,196.6080,294.9120,0.0567,28.9172,114.1536,16.3077
200,4x8x64,8,16384,1376.2560,393.2160,589.8240,0.1181,27.7523,118.9452,16.9922
200,4x16x2,8,1024,86.0160,24.5760,36.8640,0.0070,29.3553,112.4497,16.0642
200,4x16x4,8,2048,172.0320,49.1520,73.7280,0.0141,29.0617,113.5858,16.2265
200,4x16x8,8,4096,344.0640,98.3040,147.4560,0.0283,28.9315,114.0972,16.2996
200,4x16x16,8,8192,688.1280,196.6080,294.9120,0.0581,28.1803,117.1387,16.7341
200,4x16x32,8,16384,1376.2560,393.2160,589.8240,0.1182,27.7240,119.0667,17.0095
200,4x16x64,8,32768,2752.5120,786.4320,1179.6480,0.2292,28.5891,115.4635,16.4948
200,4x32x2,8,2048,172.0320,49.1520,73.7280,0.0138,29.5889,111.5622,15.9375
200,4x32x4,8,4096,344.0640,98.3040,147.4560,0.0277,29.6082,111.4894,15.9271
200,4x32x8,8,8192,688.1280,196.6080,294.9120,0.0573,28.6036,115.4052,16.4865
200,4x32x16,8,16384,1376.2560,393.2160,589.8240,0.1180,27.7805,118.8243,16.9749
200,4x32x32,8,32768,2752.5120,786.4320,1179.6480,0.2315,28.3076,116.6117,16.6588
200,4x32x64,8,65536,5505.0240,1572.8640,2359.2960,0.5896,22.2305,148.4894,21.2128
200,4x64x2,8,4096,344.0640,98.3040,147.4560,0.0275,29.8382,110.6301,15.8043
200,4x64x4,8,8192,688.1280,196.6080,294.9120,0.0590,27.7490,118.9591,16.9942
200,4x64x8,8,16384,1376.2560,393.2160,589.8240,0.1172,27.9699,118.0198,16.8600
200,4x64x16,8,32768,2752.5120,786.4320,1179.6480,0.2337,28.0471,117.6948,16.8135
200,4x64x32,8,65536,5505.0240,1572.8640,2359.2960,0.5894,22.2368,148.4475,21.2068
200,4x64x64,8,131072,11010.0480,3145.7280,4718.5920,1.4231,18.4202,179.2050,25.6007
200,8x4x2,8,512,43.0080,12.2880,18.4320,0.0034,30.1707,109.4109,15.6301
200,8x4x4,8,1024,86.0160,24.5760,36.8640,0.0070,29.1336,113.3055,16.1865
200,8x4x8,8,2048,172.0320,49.1520,73.7280,0.0141,29.0357,113.6876,16.2411
200,8x4x16,8,4096,344.0640,98.3040,147.4560,0.0283,28.9620,113.9771,16.2824
200,8x4x32,8,8192,688.1280,196.6080,294.9120,0.0589,27.7938,118.7673,16.9668
200,8x4x64,8,16384,1376.2560,393.2160,589.8240,0.1177,27.8454,118.5474,16.9353
200,8x8x2,8,1024,86.0160,24.5760,36.8640,0.0071,28.7228,114.9259,16.4180
200,8x8x4,8,2048,172.0320,49.1520,73.7280,0.0143,28.6016,115.4133,16.4876
200,8x8x8,8,4096,344.0640,98.3040,147.4560,0.0277,29.5796,111.5972,15.9425
200,8x8x16,8,8192,688.1280,196.6080,294.9120,0.0588,27.8794,118.4030,16.9147
200,8x8x32,8,16384,1376.2560,393.2160,589.8240,0.1171,27.9865,117.9497,16.8500
200,8x8x64,8,32768,2752.5120,786.4320,1179.6480,0.2282,28.7237,114.9226,16.4175
200,8x16x2,8,2048,172.0320,49.1520,73.7280,0.0142,28.7967,114.6314,16.3759
200,8x16x4,8,4096,344.0640,98.3040,147.4560,0.0283,28.9346,114.0847,16.2978
200,8x16x8,8,8192,688.1280,196.6080,294.9120,0.0570,28.7606,114.7752,16.3965
200,8x16x16,8,16384,1376.2560,393.2160,589.8240,0.1179,27.7814,118.8205,16.9744
200,8x16x32,8,32768,2752.5120,786.4320,1179.6480,0.2265,28.9373,114.0741,16.2963
200,8x16x64,8,65536,5505.0240,1572.8640,2359.2960,0.5811,22.5540,146.3601,20.9086
200,8x32x2,8,4096,344.0640,98.3040,147.4560,0.0286,28.6660,115.1539,16.4506
200,8x32x4,8,8192,688.1280,196.6080,294.9120,0.0586,27.9693,118.0222,16.8603
200,8x32x8,8,16384,1376.2560,393.2160,589.8240,0.1156,28.3397,116.4796,16.6399
200,8x32x16,8,32768,2752.5120,786.4320,1179.6480,0.2275,28.8080,114.5864,16.3695
200,8x32x32,8,65536,5505.0240,1572.8640,2359.2960,0.5878,22.2982,148.0390,21.1484
200,8x32x64,8,131072,11010.0480,3145.7280,4718.5920,1.4210,18.4483,178.9322,25.5617
200,8x64x2,8,8192,688.1280,196.6080,294.9120,0.0587,27.9060,118.2900,16.8986
200,8x64x4,8,16384,1376.2560,393.2160,589.8240,0.1172,27.9557,118.0798,16.8685
200,8x64x8,8,32768,2752.5120,786.4320,1179.6480,0.2330,28.1226,117.3790,16.7684
200,8x64x16,8,65536,5505.0240,1572.8640,2359.2960,0.5877,22.3042,147.9991,21.1427
200,8x64x32,8,131072,11010.0480,3145.7280,4718.5920,1.4194,18.4681,178.7404,25.5343
200,8x64x64,8,262144,22020.0960,6291.4560,9437.1840,3.0295,17.3061,190.7417,27.2488
200,16x4x2,8,1024,86.0160,24.5760,36.8640,0.0072,28.5625,115.5712,16.5102
200,16x4x4,8,2048,172.0320,49.1520,73.7280,0.0139,29.4813,111.9692,15.9956
200,16x4x8,8,4096,344.0640,98.3040,147.4560,0.0281,29.1991,113.0515,16.1502
200,16x4x16,8,8192,688.1280,196.6080,294.9120,0.0571,28.7037,115.0028,16.4290
200,16x4x32,8,16384,1376.2560,393.2160,589.8240,0.1143,28.6746,115.1193,16.4456
200,16x4x64,8,32768,2752.5120,786.4320,1179.6480,0.2279,28.7624,114.7679,16.3954
200,16x8x2,8,2048,172.0320,49.1520,73.7280,0.0142,28.8441,114.4430,16.3490
200,16x8x4,8,4096,344.0640,98.3040,147.4560,0.0284,28.8597,114.3809,16.3401
200,16x8x8,8,8192,688.1280,196.6080,294.9120,0.0571,28.6922,115.0488,16.4355
200,16x8x16,8,16384,1376.2560,393.2160,589.8240,0.1139,28.7644,114.7598,16.3943
200,16x8x32,8,32768,2752.5120,786.4320,1179.6480,0.2279,28.7548,114.7984,16.3998
200,16x8x64,8,65536,5505.0240,1572.8640,2359.2960,0.5880,22.2899,148.0941,21.1563
200,16x16x2,8,4096,344.0640,98.3040,147.4560,0.0285,28.7502,114.8167,16.4024
200,16x16x4,8,8192,688.1280,196.6080,294.9120,0.0571,28.7103,114.9760,16.4251
200,16x16x8,8,16384,1376.2560,393.2160,589.8240,0.1174,27.9047,118.2953,16.8993
200,16x16x16,8,32768,2752.5120,786.4320,1179.6480,0.2328,28.1544,117.2462,16.7495
200,16x16x32,8,65536,5505.0240,1572.8640,2359.2960,0.5841,22.4409,147.0975,21.0139
200,16x16x64,8,131072,11010.0480,3145.7280,4718.5920,1.4255,18.3902,179.4976,25.6425
200,16x32x2,8,8192,688.1280,196.6080,294.9120,0.0588,27.8632,118.4715,16.9245
200,16x32x4,8,16384,1376.2560,393.2160,589.8240,0.1147,28.5722,115.5317,16.5045
200,16x32x8,8,32768,2752.5120,786.4320,1179.6480,0.2358,27.7892,118.7872,16.9696
200,16x32x16,8,65536,5505.0240,1572.8640,2359.2960,0.5700,22.9961,143.5462,20.5066
200,16x32x32,8,131072,11010.0480,3145.7280,4718.5920,1.4499,18.0804,182.5737,26.0820
200,16x32x64,8,262144,22020.0960,6291.4560,9437.1840,3.0517,17.1800,192.1420,27.4489
200,16x64x2,8,16384,1376.2560,393.2160,589.8240,0.1171,27.9769,117.9904,16.8558
200,16x64x4,8,32768,2752.5120,786.4320,1179.6480,0.2327,28.1641,117.2060,16.7437
200,16x64x8,8,65536,5505.0240,1572.8640,2359.2960,0.5765,22.7359,145.1890,20.7413
200,16x64x16,8,131072,11010.0480,3145.7280,4718.5920,1.4113,18.5747,177.7145,25.3878
200,16x64x32,8,262144,22020.0960,6291.4560,9437.1840,3.1046,16.8873,195.4718,27.9245
200,16x64x64,8,524288,44040.1920,12582.9120,18874.3680,6.3321,16.5596,199.3400,28.4771
200,32x4x2,8,2048,172.0320,49.1520,73.7280,0.0140,29.3028,112.6513,16.0930
200,32x4x4,8,4096,344.0640,98.3040,147.4560,0.0279,29.4024,112.2697,16.0385
200,32x4x8,8,8192,688.1280,196.6080,294.9120,0.0571,28.7071,114.9890,16.4270
200,32x4x16,8,16384,1376.2560,393.2160,589.8240,0.1176,27.8618,118.4778,16.9254
200,32x4x32,8,32768,2752.5120,786.4320,1179.6480,0.2354,27.8446,118.5509,16.9358
200,32x4x64,8,65536,5505.0240,1572.8640,2359.2960,0.5903,22.2057,148.6558,21.2365
200,32x8x2,8,4096,344.0640,98.3040,147.4560,0.0275,29.7629,110.9100,15.8443
200,32x8x4,8,8192,688.1280,196.6080,294.9120,0.0588,27.8758,118.4181,16.9169
200,32x8x8,8,16384,1376.2560,393.2160,589.8240,0.1173,27.9366,118.1605,16.8801
200,32x8x16,8,32768,2752.5120,786.4320,1179.6480,0.2267,28.9068,114.1947,16.3135
200,32x8x32,8,65536,5505.0240,1572.8640,2359.2960,0.5834,22.4680,146.9204,20.9886
200,32x8x64,8,131072,11010.0480,3145.7280,4718.5920,1.4399,18.2055,181.3185,25.9026
200,32x16x2,8,8192,688.1280,196.6080,294.9120,0.0594,27.5914,119.6387,17.0912
200,32x16x4,8,16384,1376.2560,393.2160,589.8240,0.1171,27.9773,117.9887,16.8555
200,32x16x8,8,32768,2752.5120,786.4320,1179.6480,0.2352,27.8634,118.4709,16.9244
200,32x16x16,8,65536,5505.0240,1572.8640,2359.2960,0.5805,22.5780,146.2041,20.8863
200,32x16x32,8,131072,11010.0480,3145.7280,4718.5920,1.4311,18.3173,180.2122,25.7446
200,32x16x64,8,262144,22020.0960,6291.4560,9437.1840,3.0673,17.0929,193.1212,27.5887
200,32x32x2,8,16384,1376.2560,393.2160,589.8240,0.1147,28.5593,115.5841,16.5120
200,32x32x4,8,32768,2752.5120,786.4320,1179.6480,0.2338,28.0300,117.7668,16.8238
200,32x32x8,8,65536,5505.0240,1572.8640,2359.2960,0.5872,22.3231,147.8738,21.1248
200,32x32x16,8,131072,11010.0480,3145.7280,4718.5920,1.4118,18.5683,177.7761,25.3966
200,32x32x32,8,262144,22020.0960,6291.4560,9437.1840,3.0614,17.1258,192.7496,27.5357
200,32x32x64,8,524288,44040.1920,12582.9120,18874.3680,6.3631,16.4789,200.3168,28.6167
200,32x64x2,8,32768,2752.5120,786.4320,1179.6480,0.2330,28.1262,117.3640,16.7663
200,32x64x4,8,65536,5505.0240,1572.8640,2359.2960,0.5818,22.5288,146.5236,20.9319
200,32x64x8,8,131072,11010.0480,3145.7280,4718.5920,1.4092,18.6030,177.4448,25.3493
200,32x64x16,8,262144,22020.0960,6291.4560,9437.1840,3.0520,17.1785,192.1593,27.4513
200,32x64x32,8,524288,44040.1920,12582.9120,18874.3680,6.3449,16.5264,199.7414,28.5345
200,32x64x64,8,1048576,88080.3840,25165.8240,37748.7360,12.8872,16.2732,202.8493,28.9785
200,64x4x2,8,4096,344.0640,98.3040,147.4560,0.0278,29.4982,111.9052,15.9865
200,64x4x4,8,8192,688.1280,196.6080,294.9120,0.0573,28.5718,115.5334,16.5048
200,64x4x8,8,16384,1376.2560,393.2160,589.8240,0.1152,28.4428,116.0573,16.5796
200,64x4x16,8,32768,2752.5120,786.4320,1179.6480,0.2336,28.0505,117.6807,16.8115
200,64x4x32,8,65536,5505.0240,1572.8640,2359.2960,0.5798,22.6063,146.0209,20.8601
200,64x4x64,8,131072,11010.0480,3145.7280,4718.5920,1.4209,18.4488,178.9273,25.5610
200,64x8x2,8,8192,688.1280,196.6080,294.9120,0.0586,27.9668,118.0326,16.8618
200,64x8x4,8,16384,1376.2560,393.2160,589.8240,0.1160,28.2470,116.8621,16.6946
200,64x8x8,8,32768,2752.5120,786.4320,1179.6480,0.2294,28.5742,115.5238,16.5034
200,64x8x16,8,65536,5505.0240,1572.8640,2359.2960,0.5870,22.3285,147.8382,21.1197
200,64x8x32,8,131072,11010.0480,3145.7280,4718.5920,1.4191,18.4732,178.6914,25.5273
200,64x8x64,8,262144,22020.0960,6291.4560,9437.1840,3.0748,17.0510,193.5953,27.6565
200,64x16x2,8,16384,1376.2560,393.2160,589.8240,0.1159,28.2836,116.7109,16.6730
200,64x16x4,8,32768,2752.5120,786.4320,1179.6480,0.2332,28.0997,117.4745,16.7821
200,64x16x8,8,65536,5505.0240,1572.8640,2359.2960,0.5861,22.3650,147.5965,21.0852
200,64x16x16,8,131072,11010.0480,3145.7280,4718.5920,1.4183,18.4828,178.5985,25.5141
200,64x16x32,8,262144,22020.0960,6291.4560,9437.1840,3.0343,17.2785,191.0463,27.2923
200,64x16x64,8,524288,44040.1920,12582.9120,18874.3680,6.3269,16.5733,199.1752,28.4536
200,64x32x2,8,32768,2752.5120,786.4320,1179.6480,0.2275,28.8109,114.5747,16.3678
200,64x32x4,8,65536,5505.0240,1572.8640,2359.2960,0.5870,22.3278,147.8425,21.1204
200,64x32x8,8,131072,11010.0480,3145.7280,4718.5920,1.4124,18.5597,177.8585,25.4084
200,64x32x16,8,262144,22020.0960,6291.4560,9437.1840,3.0282,17.3134,190.6611,27.2373
200,64x32x32,8,524288,44040.1920,12582.9120,18874.3680,6.3631,16.4791,200.3148,28.6164
200,64x32x64,8,1048576,88080.3840,25165.8240,37748.7360,12.7459,16.4536,200.6248,28.6607
200,64x64x2,8,65536,5505.0240,1572.8640,2359.2960,0.5816,22.5346,146.4856,20.9265
200,64x64x4,8,131072,11010.0480,3145.7280,4718.5920,1.4185,18.4802,178.6234,25.5176
200,64x64x8,8,262144,22020.0960,6291.4560,9437.1840,3.0653,17.1039,192.9967,27.5710
200,64x64x16,8,524288,44040.1920,12582.9120,18874.3680,6.2591,16.7527,197.0425,28.1489
200,64x64x32,8,1048576,88080.3840,25165.8240,37748.7360,12.7873,16.4002,201.2777,28.7540
200,64x64x64,8,2097152,176160.7680,50331.6480,75497.4720,25.3414,16.5512,199.4419,-61.5561
200,4x4x2,16,512,59.3920,12.2880,34.8160,0.0056,18.2960,180.4222,12.0281
200,4x4x4,16,1024,118.7840,24.5760,69.6320,0.0113,18.0639,182.7398,12.1827
200,4x4x8,16,2048,237.5680,49.1520,139.2640,0.0232,17.6695,186.8195,12.4546
200,4x4x16,16,4096,475.1360,98.3040,278.5280,0.0451,18.1703,181.6698,12.1113
200,4x4x32,16,8192,950.2720,196.6080,557.0560,0.0915,17.9014,184.3990,12.2933
200,4x4x64,16,16384,1900.5440,393.2160,1114.1120,0.1840,17.8046,185.4010,12.3601
200,4x8x2,16,1024,118.7840,24.5760,69.6320,0.0112,18.2238,181.1364,12.0758
200,4x8x4,16,2048,237.5680,49.1520,139.2640,0.0225,18.2084,181.2901,12.0860
200,4x8x8,16,4096,475.1360,98.3040,278.5280,0.0452,18.1204,182.1706,12.1447
200,4x8x16,16,8192,950.2720,196.6080,557.0560,0.0908,18.0397,182.9853,12.1990
200,4x8x32,16,16384,1900.5440,393.2160,1114.1120,0.1838,17.8319,185.1175,12.3412
200,4x8x64,16,32768,3801.0880,786.4320,2228.2240,0.3640,18.0025,183.3630,12.2242
200,4x16x2,16,2048,237.5680,49.1520,139.2640,0.0227,18.0239,183.1456,12.2097
200,4x16x4,16,4096,475.1360,98.3040,278.5280,0.0451,18.1620,181.7526,12.1168
200,4x16x8,16,8192,950.2720,196.6080,557.0560,0.0923,17.7419,186.0569,12.4038
200,4x16x16,16,16384,1900.5440,393.2160,1114.1120,0.1816,18.0477,182.9043,12.1936
200,4x16x32,16,32768,3801.0880,786.4320,2228.2240,0.3646,17.9729,183.6656,12.2444
200,4x16x64,16,65536,7602.1760,1572.8640,4456.4480,0.8363,15.6732,210.6148,14.0410
200,4x32x2,16,4096,475.1360,98.3040,278.5280,0.0449,18.2632,180.7461,12.0497
200,4x32x4,16,8192,950.2720,196.6080,557.0560,0.0921,17.7950,185.5015,12.3668
200,4x32x8,16,16384,1900.5440,393.2160,1114.1120,0.1838,17.8299,185.1388,12.3426
200,4x32x16,16,32768,3801.0880,786.4320,2228.2240,0.3655,17.9329,184.0750,12.2717
200,4x32x32,16,65536,7602.1760,1572.8640,4456.4480,0.8389,15.6246,211.2694,14.0846
200,4x32x64,16,131072,15204.3520,3145.7280,8912.8960,1.7340,15.1178,218.3526,14.5568
200,4x64x2,16,8192,950.2720,196.6080,557.0560,0.0907,18.0613,182.7666,12.1844
200,4x64x4,16,16384,1900.5440,393.2160,1114.1120,0.1839,17.8213,185.2282,12.3485
200,4x64x8,16,32768,3801.0880,786.4320,2228.2240,0.3693,17.7464,186.0092,12.4006
200,4x64x16,16,65536,7602.1760,1572.8640,4456.4480,0.8366,15.6679,210.6855,14.0457
200,4x64x32,16,131072,15204.3520,3145.7280,8912.8960,1.7225,15.2190,216.9000,14.4600
200,4x64x64,16,262144,30408.7040,6291.4560,17825.7920,3.4820,15.0573,219.2297,14.6153
200,8x4x2,16,1024,118.7840,24.5760,69.6320,0.0114,18.0138,183.2482,12.2165
200,8x4x4,16,2048,237.5680,49.1520,139.2640,0.0225,18.2077,181.2973,12.0865
200,8x4x8,16,4096,475.1360,98.3040,278.5280,0.0454,18.0407,182.9754,12.1984
200,8x4x16,16,8192,950.2720,196.6080,557.0560,0.0918,17.8433,184.9994,12.3333
200,8x4x32,16,16384,1900.5440,393.2160,1114.1120,0.1826,17.9495,183.9044,12.2603
200,8x4x64,16,32768,3801.0880,786.4320,2228.2240,0.3641,18.0005,183.3834,12.2256
200,8x8x2,16,2048,237.5680,49.1520,139.2640,0.0224,18.3071,180.3121,12.0208
200,8x8x4,16,4096,475.1360,98.3040,278.5280,0.0453,18.0714,182.6647,12.1776
200,8x8x8,16,8192,950.2720,196.6080,557.0560,0.0906,18.0807,182.5707,12.1714
200,8x8x16,16,16384,1900.5440,393.2160,1114.1120,0.1838,17.8322,185.1142,12.3409
200,8x8x32,16,32768,3801.0880,786.4320,2228.2240,0.3640,18.0066,183.3212,12.2214
200,8x8x64,16,65536,7602.1760,1572.8640,4456.4480,0.8361,15.6765,210.5705,14.0380
200,8x16x2,16,4096,475.1360,98.3040,278.5280,0.0445,18.4011,179.3913,11.9594
200,8x16x4,16,8192,950.2720,196.6080,557.0560,0.0908,18.0523,182.8576,12.1905
200,8x16x8,16,16384,1900.5440,393.2160,1114.1120,0.1840,17.8071,185.3758,12.3584
200,8x16x16,16,32768,3801.0880,786.4320,2228.2240,0.3677,17.8222,185.2185,12.3479
200,8x16x32,16,65536,7602.1760,1572.8640,4456.4480,0.8354,15.6894,210.3971,14.0265
200,8x16x64,16,131072,15204.3520,3145.7280,8912.8960,1.7307,15.1470,217.9303,14.5287
200,8x32x2,16,8192,950.2720,196.6080,557.0560,0.0915,17.8982,184.4316,12.2954
200,8x32x4,16,16384,1900.5440,393.2160,1114.1120,0.1839,17.8177,185.2656,12.3510
200,8x32x8,16,32768,3801.0880,786.4320,2228.2240,0.3630,18.0547,182.8336,12.1889
200,8x32x16,16,65536,7602.1760,1572.8640,4456.4480,0.8365,15.6691,210.6693,14.0446
200,8x32x32,16,131072,15204.3520,3145.7280,8912.8960,1.7251,15.1956,217.2335,14.4822
200,8x32x64,16,262144,30408.7040,6291.4560,17825.7920,3.4910,15.0184,219.7975,14.6532
200,8x64x2,16,16384,1900.5440,393.2160,1114.1120,0.1841,17.8028,185.4207,12.3614
200,8x64x4,16,32768,3801.0880,786.4320,2228.2240,0.3677,17.8242,185.1975,12.3465
200,8x64x8,16,65536,7602.1760,1572.8640,4456.4480,0.8365,15.6697,210.6616,14.0441
200,8x64x16,16,131072,15204.3520,3145.7280,8912.8960,1.7215,15.2276,216.7771,14.4518
200,8x64x32,16,262144,30408.7040,6291.4560,17825.7920,3.4890,15.0267,219.6760,14.6451
200,8x64x64,16,524288,60817.4080,12582.9120,35651.5840,7.0182,14.9408,220.9380,14.7292
200,16x4x2,16,2048,237.5680,49.1520,139.2640,0.0225,18.2315,181.0606,12.0707
200,16x4x4,16,4096,475.1360,98.3040,278.5280,0.0454,18.0499,182.8815,12.1921
200,16x4x8,16,8192,950.2720,196.6080,557.0560,0.0919,17.8201,185.2398,12.3493
200,16x4x16,16,16384,1900.5440,393.2160,1114.1120,0.1826,17.9490,183.9098,12.2607
200,16x4x32,16,32768,3801.0880,786.4320,2228.2240,0.3682,17.8013,185.4364,12.3624
200,16x4x64,16,65536,7602.1760,1572.8640,4456.4480,0.8284,15.8219,208.6348,13.9090
200,16x8x2,16,4096,475.1360,98.3040,278.5280,0.0453,18.0897,182.4799,12.1653
200,16x8x4,16,8192,950.2720,196.6080,557.0560,0.0921,17.7910,185.5429,12.3695
200,16x8x8,16,16384,1900.5440,393.2160,1114.1120,0.1841,17.7950,185.5011,12.3667
200,16x8x16,16,32768,3801.0880,786.4320,2228.2240,0.3644,17.9861,183.5306,12.2354
200,16x8x32,16,65536,7602.1760,1572.8640,4456.4480,0.8360,15.6776,210.5549,14.0370
200,16x8x64,16,131072,15204.3520,3145.7280,8912.8960,1.7218,15.2250,216.8139,14.4543
200,16x16x2,16,8192,950.2720,196.6080,557.0560,0.0919,17.8267,185.1722,12.3448
200,16x16x4,16,16384,1900.5440,393.2160,1114.1120,0.1816,18.0469,182.9119,12.1941
200,16x16x8,16,32768,3801.0880,786.4320,2228.2240,0.3628,18.0620,182.7598,12.1840
200,16x16x16,16,65536,7602.1760,1572.8640,4456.4480,0.8371,15.6579,210.8203,14.0547
200,16x16x32,16,131072,15204.3520,3145.7280,8912.8960,1.7275,15.1749,217.5302,14.5020
200,16x16x64,16,262144,30408.7040,6291.4560,17825.7920,3.4822,15.0561,219.2462,14.6164
200,16x32x2,16,16384,1900.5440,393.2160,1114.1120,0.1814,18.0647,182.7321,12.1821
200,16x32x4,16,32768,3801.0880,786.4320,2228.2240,0.3693,17.7452,186.0226,12.4015
200,16x32x8,16,65536,7602.1760,1572.8640,4456.4480,0.8312,15.7692,209.3325,13.9555
200,16x32x16,16,131072,15204.3520,3145.7280,8912.8960,1.7303,15.1499,217.8895,14.5260
200,16x32x32,16,262144,30408.7040,6291.4560,17825.7920,3.4813,15.0603,219.1857,14.6124
200,16x32x64,16,524288,60817.4080,12582.9120,35651.5840,7.0206,14.9358,221.0133,14.7342
200,16x64x2,16,32768,3801.0880,786.4320,2228.2240,0.3677,17.8227,185.2129,12.3475
200,16x64x4,16,65536,7602.1760,1572.8640,4456.4480,0.8286,15.8193,208.6690,13.9113
200,16x64x8,16,131072,15204.3520,3145.7280,8912.8960,1.7333,15.1243,218.2585,14.5506
200,16x64x16,16,262144,30408.7040,6291.4560,17825.7920,3.4808,15.0624,219.1549,14.6103
200,16x64x32,16,524288,60817.4080,12582.9120,35651.5840,7.0109,14.9565,220.7072,14.7138
200,16x64x64,16,1048576,121634.8160,25165.8240,71303.1680,14.0846,14.8897,221.6970,-40.4557
200,32x4x2,16,4096,475.1360,98.3040,278.5280,0.0455,18.0158,183.2281,12.2152
200,32x4x4,16,8192,950.2720,196.6080,557.0560,0.0910,18.0058,183.3293,12.2220
200,32x4x8,16,16384,1900.5440,393.2160,1114.1120,0.1824,17.9626,183.7708,12.2514
200,32x4x16,16,32768,3801.0880,786.4320,2228.2240,0.3649,17.9604,183.7931,12.2529
200,32x4x32,16,65536,7602.1760,1572.8640,4456.4480,0.8305,15.7818,209.1647,13.9443
200,32x4x64,16,131072,15204.3520,3145.7280,8912.8960,1.7326,15.1299,218.1779,14.5452
200,32x8x2,16,8192,950.2720,196.6080,557.0560,0.0914,17.9206,184.2011,12.2801
200,32x8x4,16,16384,1900.5440,393.2160,1114.1120,0.1817,18.0318,183.0659,12.2044
200,32x8x8,16,32768,3801.0880,786.4320,2228.2240,0.3631,18.0476,182.9056,12.1937
200,32x8x16,16,65536,7602.1760,1572.8640,4456.4480,0.8336,15.7244,209.9284,13.9952
200,32x8x32,16,131072,15204.3520,3145.7280,8912.8960,1.7333,15.1241,218.2616,14.5508
200,32x8x64,16,262144,30408.7040,6291.4560,17825.7920,3.4862,15.0390,219.4955,14.6330
200,32x16x2,16,16384,1900.5440,393.2160,1114.1120,0.1817,18.0327,183.0564,12.2038
200,32x16x4,16,32768,3801.0880,786.4320,2228.2240,0.3708,17.6763,186.7474,12.4498
200,32x16x8,16,65536,7602.1760,1572.8640,4456.4480,0.8334,15.7281,209.8790,13.9919
200,32x16x16,16,131072,15204.3520,3145.7280,8912.8960,1.7301,15.1518,217.8613,14.5241
200,32x16x32,16,262144,30408.7040,6291.4560,17825.7920,3.4846,15.0461,219.3931,14.6262
200,32x16x64,16,524288,60817.4080,12582.9120,35651.5840,7.0117,14.9548,220.7325,14.7155
200,32x32x2,16,32768,3801.0880,786.4320,2228.2240,0.3650,17.9532,183.8669,12.2578
200,32x32x4,16,65536,7602.1760,1572.8640,4456.4480,0.8263,15.8627,208.0976,13.8732
200,32x32x8,16,131072,15204.3520,3145.7280,8912.8960,1.7333,15.1243,218.2578,14.5505
200,32x32x16,16,262144,30408.7040,6291.4560,17825.7920,3.4864,15.0382,219.5082,14.6339
200,32x32x32,16,524288,60817.4080,12582.9120,35651.5840,7.0269,14.9223,221.2125,14.7475
200,32x32x64,16,1048576,121634.8160,25165.8240,71303.1680,14.0684,14.9068,221.4424,-40.4092
200,32x64x2,16,65536,7602.1760,1572.8640,4456.4480,0.8343,15.7113,210.1042,14.0069
200,32x64x4,16,131072,15204.3520,3145.7280,8912.8960,1.7294,15.1578,217.7754,14.5184
200,32x64x8,16,262144,30408.7040,6291.4560,17825.7920,3.4826,15.0546,219.2679,14.6179
200,32x64x16,16,524288,60817.4080,12582.9120,35651.5840,7.0182,14.9408,220.9389,14.7293
200,32x64x32,16,1048576,121634.8160,25165.8240,71303.1680,14.0401,14.9369,220.9959,-40.3277
200,32x64x64,16,2097152,243269.6320,50331.6480,142606.3360,28.4436,14.7460,223.8568,47.0287
200,64x4x2,16,8192,950.2720,196.6080,557.0560,0.0907,18.0594,182.7859,12.1857
200,64x4x4,16,16384,1900.5440,393.2160,1114.1120,0.1821,17.9935,183.4551,12.2303
200,64x4x8,16,32768,3801.0880,786.4320,2228.2240,0.3673,17.8419,185.0140,12.3343
200,64x4x16,16,65536,7602.1760,1572.8640,4456.4480,0.8270,15.8488,208.2813,13.8854
200,64x4x32,16,131072,15204.3520,3145.7280,8912.8960,1.7268,15.1806,217.4484,14.4966
200,64x4x64,16,262144,30408.7040,6291.4560,17825.7920,3.5042,14.9618,220.6282,14.7085
200,64x8x2,16,16384,1900.5440,393.2160,1114.1120,0.1841,17.7998,185.4515,12.3634
200,64x8x4,16,32768,3801.0880,786.4320,2228.2240,0.3677,17.8227,185.2128,12.3475
200,64x8x8,16,65536,7602.1760,1572.8640,4456.4480,0.8403,15.5980,211.6294,14.1086
200,64x8x16,16,131072,15204.3520,3145.7280,8912.8960,1.7286,15.1654,217.6671,14.5111
200,64x8x32,16,262144,30408.7040,6291.4560,17825.7920,3.4831,15.0521,219.3045,14.6203
200,64x8x64,16,524288,60817.4080,12582.9120,35651.5840,7.0128,14.9524,220.7673,14.7178
200,64x16x2,16,32768,3801.0880,786.4320,2228.2240,0.3696,17.7295,186.1872,12.4125
200,64x16x4,16,65536,7602.1760,1572.8640,4456.4480,0.8355,15.6870,210.4292,14.0286
200,64x16x8,16,131072,15204.3520,3145.7280,8912.8960,1.7271,15.1781,217.4841,14.4989
200,64x16x16,16,262144,30408.7040,6291.4560,17825.7920,3.4838,15.0493,219.3463,14.6231
200,64x16x32,16,524288,60817.4080,12582.9120,35651.5840,6.9985,14.9828,220.3193,14.6880
200,64x16x64,16,1048576,121634.8160,25165.8240,71303.1680,14.1217,14.8505,222.2818,-40.5624
200,64x32x2,16,65536,7602.1760,1572.8640,4456.4480,0.8328,15.7387,209.7378,13.9825
200,64x32x4,16,131072,15204.3520,3145.7280,8912.8960,1.7318,15.1367,218.0798,14.5387
200,64x32x8,16,262144,30408.7040,6291.4560,17825.7920,3.4835,15.0505,219.3286,14.6219
200,64x32x16,16,524288,60817.4080,12582.9120,35651.5840,7.0249,14.9267,221.1479,14.7432
200,64x32x32,16,1048576,121634.8160,25165.8240,71303.1680,14.0693,14.9059,221.4558,-40.4116
200,64x32x64,16,2097152,243269.6320,50331.6480,142606.3360,28.5260,14.7034,224.5056,47.1650
200,64x64x2,16,131072,15204.3520,3145.7280,8912.8960,1.7298,15.1548,217.8186,14.5212
200,64x64x4,16,262144,30408.7040,6291.4560,17825.7920,3.4835,15.0508,219.3240,14.6216
200,64x64x8,16,524288,60817.4080,12582.9120,35651.5840,7.0038,14.9715,220.4853,14.6990
200,64x64x16,16,1048576,121634.8160,25165.8240,71303.1680,14.0787,14.8960,221.6037,-40.4386
200,64x64x32,16,2097152,243269.6320,50331.6480,142606.3360,28.2532,14.8454,222.3579,46.7138
200,64x64x64,16,4194304,486539.2640,100663.2960,285212.6720,56.7032,14.7939,223.1326,-619.8127
200,4x4x2,32,1024,184.3200,24.5760,135.1680,0.0184,11.1343,296.4700,9.5635
200,4x4x4,32,2048,368.6400,49.1520,270.3360,0.0367,11.1623,295.7287,9.5396
200,4x4x8,32,4096,737.2800,98.3040,540.6720,0.0742,11.0459,298.8427,9.6401
200,4x4x16,32,8192,1474.5600,196.6080,1081.3440,0.1479,11.0768,298.0102,9.6132
200,4x4x32,32,16384,2949.1200,393.2160,2162.6880,0.2940,11.1459,296.1616,9.5536
200,4x4x64,32,32768,5898.2400,786.4320,4325.3760,0.5993,10.9363,301.8385,9.7367
200,4x8x2,32,2048,368.6400,49.1520,270.3360,0.0363,11.2914,292.3454,9.4305
200,4x8x4,32,4096,737.2800,98.3040,540.6720,0.0743,11.0324,299.2096,9.6519
200,4x8x8,32,8192,1474.5600,196.6080,1081.3440,0.1485,11.0358,299.1171,9.6489
200,4x8x16,32,16384,2949.1200,393.2160,2162.6880,0.2969,11.0373,299.0780,9.6477
200,4x8x32,32,32768,5898.2400,786.4320,4325.3760,0.5939,11.0341,299.1637,9.6504
200,4x8x64,32,65536,11796.4800,1572.8640,8650.7520,1.2794,10.2451,322.2030,10.3936
200,4x16x2,32,4096,737.2800,98.3040,540.6720,0.0740,11.0639,298.3586,9.6245
200,4x16x4,32,8192,1474.5600,196.6080,1081.3440,0.1462,11.2070,294.5475,9.5015
200,4x16x8,32,16384,2949.1200,393.2160,2162.6880,0.2939,11.1506,296.0374,9.5496
200,4x16x16,32,32768,5898.2400,786.4320,4325.3760,0.5989,10.9423,301.6720,9.7314
200,4x16x32,32,65536,11796.4800,1572.8640,8650.7520,1.2748,10.2822,321.0414,10.3562
200,4x16x64,32,131072,23592.9600,3145.7280,17301.5040,2.6346,9.9501,331.7548,10.7018
200,4x32x2,32,8192,1474.5600,196.6080,1081.3440,0.1485,11.0330,299.1930,9.6514
200,4x32x4,32,16384,2949.1200,393.2160,2162.6880,0.2971,11.0277,299.3378,9.6561
200,4x32x8,32,32768,5898.2400,786.4320,4325.3760,0.5972,10.9748,300.7808,9.7026
200,4x32x16,32,65536,11796.4800,1572.8640,8650.7520,1.2769,10.2653,321.5702,10.3732
200,4x32x32,32,131072,23592.9600,3145.7280,17301.5040,2.6224,9.9963,330.2237,10.6524
200,4x32x64,32,262144,47185.9200,6291.4560,34603.0080,5.3057,9.8816,334.0550,10.7760
200,4x64x2,32,16384,2949.1200,393.2160,2162.6880,0.2971,11.0311,299.2447,9.6531
200,4x64x4,32,32768,5898.2400,786.4320,4325.3760,0.6015,10.8958,302.9599,9.7729
200,4x64x8,32,65536,11796.4800,1572.8640,8650.7520,1.2819,10.2248,322.8439,10.4143
200,4x64x16,32,131072,23592.9600,3145.7280,17301.5040,2.6160,10.0209,329.4108,10.6262
200,4x64x32,32,262144,47185.9200,6291.4560,34603.0080,5.2811,9.9276,332.5078,10.7261
200,4x64x64,32,524288,94371.8400,12582.9120,69206.0160,10.6519,9.8441,335.3289,-33.6676
200,8x4x2,32,2048,368.6400,49.1520,270.3360,0.0368,11.1330,296.5056,9.5647
200,8x4x4,32,4096,737.2800,98.3040,540.6720,0.0740,11.0736,298.0970,9.6160
200,8x4x8,32,8192,1474.5600,196.6080,1081.3440,0.1483,11.0501,298.7301,9.6365
200,8x4x16,32,16384,2949.1200,393.2160,2162.6880,0.2933,11.1714,295.4874,9.5319
200,8x4x32,32,32768,5898.2400,786.4320,4325.3760,0.5981,10.9567,301.2773,9.7186
200,8x4x64,32,65536,11796.4800,1572.8640,8650.7520,1.2789,10.2488,322.0863,10.3899
200,8x8x2,32,4096,737.2800,98.3040,540.6720,0.0739,11.0912,297.6237,9.6008
200,8x8x4,32,8192,1474.5600,196.6080,1081.3440,0.1482,11.0545,298.6125,9.6327
200,8x8x8,32,16384,2949.1200,393.2160,2162.6880,0.2974,11.0180,299.5995,9.6645
200,8x8x16,32,32768,5898.2400,786.4320,4325.3760,0.5904,11.0999,297.3889,9.5932
200,8x8x32,32,65536,11796.4800,1572.8640,8650.7520,1.2774,10.2611,321.6992,10.3774
200,8x8x64,32,131072,23592.9600,3145.7280,17301.5040,2.6137,10.0297,329.1226,10.6169
200,8x16x2,32,8192,1474.5600,196.6080,1081.3440,0.1483,11.0470,298.8150,9.6392
200,8x16x4,32,16384,2949.1200,393.2160,2162.6880,0.2975,11.0131,299.7339,9.6688
200,8x16x8,32,32768,5898.2400,786.4320,4325.3760,0.5982,10.9562,301.2905,9.7190
200,8x16x16,32,65536,11796.4800,1572.8640,8650.7520,1.2760,10.2721,321.3560,10.3663
200,8x16x32,32,131072,23592.9600,3145.7280,17301.5040,2.6146,10.0263,329.2351,10.6205
200,8x16x64,32,262144,47185.9200,6291.4560,34603.0080,5.2848,9.9206,332.7404,10.7336
200,8x32x2,32,16384,2949.1200,393.2160,2162.6880,0.2967,11.0446,298.8779,9.6412
200,8x32x4,32,32768,5898.2400,786.4320,4325.3760,0.5983,10.9536,301.3623,9.7214
200,8x32x8,32,65536,11796.4800,1572.8640,8650.7520,1.2785,10.2522,321.9806,10.3865
200,8x32x16,32,131072,23592.9600,3145.7280,17301.5040,2.6162,10.0200,329.4416,10.6271
200,8x32x32,32,262144,47185.9200,6291.4560,34603.0080,5.2818,9.9263,332.5517,10.7275
200,8x32x64,32,524288,94371.8400,12582.9120,69206.0160,10.6656,9.8314,335.7605,-33.7109
200,8x64x2,32,32768,5898.2400,786.4320,4325.3760,0.5981,10.9568,301.2731,9.7185
200,8x64x4,32,65536,11796.4800,1572.8640,8650.7520,1.2759,10.2726,321.3416,10.3659
200,8x64x8,32,131072,23592.9600,3145.7280,17301.5040,2.6177,10.0143,329.6278,10.6332
200,8x64x16,32,262144,47185.9200,6291.4560,34603.0080,5.2981,9.8957,333.5791,10.7606
200,8x64x32,32,524288,94371.8400,12582.9120,69206.0160,10.6569,9.8394,335.4876,-33.6835
200,8x64x64,32,1048576,188743.6800,25165.8240,138412.0320,22.0848,9.4959,347.6238,-34.9020
200,16x4x2,32,4096,737.2800,98.3040,540.6720,0.0734,11.1599,295.7909,9.5416
200,16x4x4,32,8192,1474.5600,196.6080,1081.3440,0.1483,11.0457,298.8502,9.6403
200,16x4x8,32,16384,2949.1200,393.2160,2162.6880,0.2969,11.0351,299.1364,9.6496
200,16x4x16,32,32768,5898.2400,786.4320,4325.3760,0.6002,10.9182,302.3381,9.7528
200,16x4x32,32,65536,11796.4800,1572.8640,8650.7520,1.2791,10.2475,322.1284,10.3912
200,16x4x64,32,131072,23592.9600,3145.7280,17301.5040,2.6199,10.0058,329.9076,10.6422
200,16x8x2,32,8192,1474.5600,196.6080,1081.3440,0.1471,11.1387,296.3547,9.5598
200,16x8x4,32,16384,2949.1200,393.2160,2162.6880,0.2965,11.0523,298.6703,9.6345
200,16x8x8,32,32768,5898.2400,786.4320,4325.3760,0.6002,10.9183,302.3378,9.7528
200,16x8x16,32,65536,11796.4800,1572.8640,8650.7520,1.2728,10.2978,320.5541,10.3405
200,16x8x32,32,131072,23592.9600,3145.7280,17301.5040,2.6176,10.0146,329.6201,10.6329
200,16x8x64,32,262144,47185.9200,6291.4560,34603.0080,5.3039,9.8849,333.9450,10.7724
200,16x16x2,32,16384,2949.1200,393.2160,2162.6880,0.2931,11.1816,295.2182,9.5232
200,16x16x4,32,32768,5898.2400,786.4320,4325.3760,0.5996,10.9300,302.0115,9.7423
200,16x16x8,32,65536,11796.4800,1572.8640,8650.7520,1.2755,10.2760,321.2350,10.3624
200,16x16x16,32,131072,23592.9600,3145.7280,17301.5040,2.6117,10.0374,328.8698,10.6087
200,16x16x32,32,262144,47185.9200,6291.4560,34603.0080,5.2872,9.9162,332.8884,10.7383
200,16x16x64,32,524288,94371.8400,12582.9120,69206.0160,10.6518,9.8441,335.3272,-33.6674
200,16x32x2,32,32768,5898.2400,786.4320,4325.3760,0.5951,11.0130,299.7369,9.6689
200,16x32x4,32,65536,11796.4800,1572.8640,8650.7520,1.2710,10.3123,320.1036,10.3259
200,16x32x8,32,131072,23592.9600,3145.7280,17301.5040,2.6168,10.0177,329.5162,10.6296
200,16x32x16,32,262144,47185.9200,6291.4560,34603.0080,5.2696,9.9493,331.7830,10.7027
200,16x32x32,32,524288,94371.8400,12582.9120,69206.0160,10.6700,9.8274,335.8990,-33.7248
200,16x32x64,32,1048576,188743.6800,25165.8240,138412.0320,21.3463,9.8244,335.9986,-33.7348
200,16x64x2,32,65536,11796.4800,1572.8640,8650.7520,1.2744,10.2853,320.9430,10.3530
200,16x64x4,32,131072,23592.9600,3145.7280,17301.5040,2.6233,9.9930,330.3298,10.6558
200,16x64x8,32,262144,47185.9200,6291.4560,34603.0080,5.2958,9.9000,333.4339,10.7559
200,16x64x16,32,524288,94371.8400,12582.9120,69206.0160,10.6676,9.8295,335.8249,-33.7174
200,16x64x32,32,1048576,188743.6800,25165.8240,138412.0320,21.3328,9.8306,335.7866,-33.7135
200,16x64x64,32,2097152,377487.3600,50331.6480,276824.0640,43.2515,9.6975,340.3977,1215.7060
200,32x4x2,32,8192,1474.5600,196.6080,1081.3440,0.1467,11.1717,295.4785,9.5316
200,32x4x4,32,16384,2949.1200,393.2160,2162.6880,0.2932,11.1753,295.3833,9.5285
200,32x4x8,32,32768,5898.2400,786.4320,4325.3760,0.5962,10.9931,300.2802,9.6865
200,32x4x16,32,65536,11796.4800,1572.8640,8650.7520,1.2727,10.2991,320.5145,10.3392
200,32x4x32,32,131072,23592.9600,3145.7280,17301.5040,2.6218,9.9985,330.1501,10.6500
200,32x4x64,32,262144,47185.9200,6291.4560,34603.0080,5.2997,9.8928,333.6772,10.7638
200,32x8x2,32,16384,2949.1200,393.2160,2162.6880,0.2982,10.9877,300.4260,9.6912
200,32x8x4,32,32768,5898.2400,786.4320,4325.3760,0.5999,10.9247,302.1584,9.7470
200,32x8x8,32,65536,11796.4800,1572.8640,8650.7520,1.2737,10.2905,320.7800,10.3477
200,32x8x16,32,131072,23592.9600,3145.7280,17301.5040,2.6220,9.9977,330.1746,10.6508
200,32x8x32,32,262144,47185.9200,6291.4560,34603.0080,5.2694,9.9497,331.7683,10.7022
200,32x8x64,32,524288,94371.8400,12582.9120,69206.0160,10.6605,9.8361,335.6015,-33.6949
200,32x16x2,32,32768,5898.2400,786.4320,4325.3760,0.5981,10.9581,301.2397,9.7174
200,32x16x4,32,65536,11796.4800,1572.8640,8650.7520,1.2765,10.2682,321.4795,10.3703
200,32x16x8,32,131072,23592.9600,3145.7280,17301.5040,2.6141,10.0280,329.1777,10.6186
200,32x16x16,32,262144,47185.9200,6291.4560,34603.0080,5.2722,9.9444,331.9453,10.7079
200,32x16x32,32,524288,94371.8400,12582.9120,69206.0160,10.6403,9.8548,334.9650,-33.6310
200,32x16x64,32,1048576,188743.6800,25165.8240,138412.0320,21.3337,9.8302,335.8002,-33.7149
200,32x32x2,32,65536,11796.4800,1572.8640,8650.7520,1.2759,10.2729,321.3320,10.3655
200,32x32x4,32,131072,23592.9600,3145.7280,17301.5040,2.6127,10.0333,329.0044,10.6130
200,32x32x8,32,262144,47185.9200,6291.4560,34603.0080,5.2880,9.9147,332.9401,10.7400
200,32x32x16,32,524288,94371.8400,12582.9120,69206.0160,10.6384,9.8565,334.9068,-33.6252
200,32x32x32,32,1048576,188743.6800,25165.8240,138412.0320,21.3562,9.8199,336.1558,-33.7506
200,32x32x64,32,2097152,377487.3600,50331.6480,276824.0640,42.9627,9.7627,338.1252,1207.5901
200,32x64x2,32,131072,23592.9600,3145.7280,17301.5040,2.6187,10.0106,329.7513,10.6371
200,32x64x4,32,262144,47185.9200,6291.4560,34603.0080,5.2917,9.9078,333.1709,10.7474
200,32x64x8,32,524288,94371.8400,12582.9120,69206.0160,10.6475,9.8481,335.1905,-33.6537
200,32x64x16,32,1048576,188743.6800,25165.8240,138412.0320,21.3369,9.8288,335.8512,-33.7200
200,32x64x32,32,2097152,377487.3600,50331.6480,276824.0640,42.7122,9.8199,336.1533,1200.5474
200,32x64x64,32,4194304,754974.7200,100663.2960,553648.1280,86.3545,9.7141,339.8136,1213.6201
200,64x4x2,32,16384,2949.1200,393.2160,2162.6880,0.2938,11.1517,296.0095,9.5487
200,64x4x4,32,32768,5898.2400,786.4320,4325.3760,0.5981,10.9565,301.2815,9.7188
200,64x4x8,32,65536,11796.4800,1572.8640,8650.7520,1.2762,10.2702,321.4139,10.3682
200,64x4x16,32,131072,23592.9600,3145.7280,17301.5040,2.6196,10.0069,329.8709,10.6410
200,64x4x32,32,262144,47185.9200,6291.4560,34603.0080,5.3022,9.8882,333.8331,10.7688
200,64x4x64,32,524288,94371.8400,12582.9120,69206.0160,10.6430,9.8523,335.0502,-33.6396
200,64x8x2,32,32768,5898.2400,786.4320,4325.3760,0.5980,10.9588,301.2187,9.7167
200,64x8x4,32,65536,11796.4800,1572.8640,8650.7520,1.2720,10.3042,320.3540,10.3340
200,64x8x8,32,131072,23592.9600,3145.7280,17301.5040,2.6136,10.0298,329.1186,10.6167
200,64x8x16,32,262144,47185.9200,6291.4560,34603.0080,5.2814,9.9272,332.5222,10.7265
200,64x8x32,32,524288,94371.8400,12582.9120,69206.0160,10.6593,9.8372,335.5619,-33.6910
200,64x8x64,32,1048576,188743.6800,25165.8240,138412.0320,21.3728,9.8123,336.4161,-33.7767
200,64x16x2,32,65536,11796.4800,1572.8640,8650.7520,1.2737,10.2904,320.7839,10.3479
200,64x16x4,32,131072,23592.9600,3145.7280,17301.5040,2.6248,9.9870,330.5291,10.6622
200,64x16x8,32,262144,47185.9200,6291.4560,34603.0080,5.2807,9.9284,332.4815,10.7252
200,64x16x16,32,524288,94371.8400,12582.9120,69206.0160,10.6290,9.8653,334.6079,-33.5952
200,64x16x32,32,1048576,188743.6800,25165.8240,138412.0320,21.3837,9.8072,336.5883,-33.7940
200,64x16x64,32,2097152,377487.3600,50331.6480,276824.0640,42.8648,9.7850,337.3547,1204.8384
200,64x32x2,32,131072,23592.9600,3145.7280,17301.5040,2.6133,10.0311,329.0782,10.6154
200,64x32x4,32,262144,47185.9200,6291.4560,34603.0080,5.2990,9.8941,333.6320,10.7623
200,64x32x8,32,524288,94371.8400,12582.9120,69206.0160,10.6473,9.8483,335.1850,-33.6531
200,64x32x16,32,1048576,188743.6800,25165.8240,138412.0320,21.3139,9.8394,335.4895,-33.6837
200,64x32x32,32,2097152,377487.3600,50331.6480,276824.0640,42.7194,9.8183,336.2104,1200.7513
200,64x32x64,32,4194304,754974.7200,100663.2960,553648.1280,86.0053,9.7536,338.4393,1208.7117
200,64x64x2,32,262144,47185.9200,6291.4560,34603.0080,5.2792,9.9312,332.3876,10.7222
200,64x64x4,32,524288,94371.8400,12582.9120,69206.0160,10.6422,9.8530,335.0261,-33.6372
200,64x64x8,32,1048576,188743.6800,25165.8240,138412.0320,21.3188,9.8371,335.5660,-33.6914
200,64x64x16,32,2097152,377487.3600,50331.6480,276824.0640,42.7500,9.8112,336.4510,1201.6108
200,64x64x32,32,4194304,754974.7200,100663.2960,553648.1280,86.0358,9.7501,338.5595,1209.1409
200,64x64x64,32,8388608,1509949.4400,201326.5920,1107296.2560,175.8321,9.5416,345.9584,1235.5657

110
scripts/results_soa_skylakesp2_iln1000.txt
View File

@ -1,110 +0,0 @@
200,4x4x2,8,256,21.5040,6.1440,9.2160,0.8212,62.3493,38.5089,5.5013
200,4x4x4,8,512,43.0080,12.2880,18.4320,1.6918,60.5255,39.6693,5.6670
200,4x4x8,8,1024,86.0160,24.5760,36.8640,3.3338,61.4308,39.0847,5.5835
200,4x4x16,8,2048,172.0320,49.1520,73.7280,6.7673,60.5267,39.6685,5.6669
200,4x4x2,8,256,21.5040,6.1440,9.2160,0.7389,69.2906,34.6511,4.9502
200,4x4x4,8,512,43.0080,12.2880,18.4320,1.4781,69.2763,34.6583,4.9512
200,4x4x8,8,1024,86.0160,24.5760,36.8640,2.9563,69.2761,34.6584,4.9512
200,4x4x16,8,2048,172.0320,49.1520,73.7280,5.9123,69.2798,34.6566,4.9509
200,4x4x32,8,4096,344.0640,98.3040,147.4560,11.8247,69.2789,34.6570,4.9510
200,4x4x64,8,8192,688.1280,196.6080,294.9120,23.6678,69.2247,34.6841,4.9549
200,4x8x2,8,512,43.0080,12.2880,18.4320,1.4780,69.2831,34.6549,4.9507
200,4x8x4,8,1024,86.0160,24.5760,36.8640,2.9563,69.2756,34.6587,4.9512
200,4x8x8,8,2048,172.0320,49.1520,73.7280,5.9131,69.2697,34.6616,4.9517
200,4x8x16,8,4096,344.0640,98.3040,147.4560,11.8363,69.2107,34.6912,4.9559
200,4x8x32,8,8192,688.1280,196.6080,294.9120,23.6674,69.2260,34.6835,4.9548
200,4x8x64,8,16384,1376.2560,393.2160,589.8240,47.3354,69.2252,34.6839,4.9548
200,4x16x2,8,1024,86.0160,24.5760,36.8640,2.9563,69.2749,34.6590,4.9513
200,4x16x4,8,2048,172.0320,49.1520,73.7280,5.9120,69.2832,34.6549,4.9507
200,4x16x8,8,4096,344.0640,98.3040,147.4560,11.8249,69.2774,34.6578,4.9511
200,4x16x16,8,8192,688.1280,196.6080,294.9120,23.6676,69.2255,34.6837,4.9548
200,4x16x32,8,16384,1376.2560,393.2160,589.8240,47.3407,69.2174,34.6878,4.9554
200,4x16x64,8,32768,2752.5120,786.4320,1179.6480,94.6451,69.2439,34.6745,4.9535
200,4x32x2,8,2048,172.0320,49.1520,73.7280,5.9164,69.2315,34.6807,4.9544
200,4x32x4,8,4096,344.0640,98.3040,147.4560,11.8253,69.2752,34.6589,4.9513
200,4x32x8,8,8192,688.1280,196.6080,294.9120,23.6684,69.2231,34.6850,4.9550
200,4x32x16,8,16384,1376.2560,393.2160,589.8240,47.3354,69.2252,34.6839,4.9548
200,4x32x32,8,32768,2752.5120,786.4320,1179.6480,95.1617,68.8680,34.8638,4.9805
200,4x32x64,8,65536,5505.0240,1572.8640,2359.2960,192.9801,67.9200,35.3504,5.0501
200,4x64x2,8,4096,344.0640,98.3040,147.4560,11.8252,69.2760,34.6585,4.9512
200,4x64x4,8,8192,688.1280,196.6080,294.9120,23.6676,69.2254,34.6838,4.9548
200,4x64x8,8,16384,1376.2560,393.2160,589.8240,47.6913,68.7085,34.9447,4.9921
200,4x64x16,8,32768,2752.5120,786.4320,1179.6480,94.6478,69.2420,34.6755,4.9536
200,4x64x32,8,65536,5505.0240,1572.8640,2359.2960,193.7782,67.6402,35.4966,5.0709
200,4x64x64,8,131072,11010.0480,3145.7280,4718.5920,384.4290,68.1905,35.2102,5.0300
200,8x4x2,8,512,43.0080,12.2880,18.4320,1.4779,69.2853,34.6538,4.9505
200,8x4x4,8,1024,86.0160,24.5760,36.8640,2.9567,69.2662,34.6634,4.9519
200,8x4x8,8,2048,172.0320,49.1520,73.7280,5.9137,69.2624,34.6653,4.9522
200,8x4x16,8,4096,344.0640,98.3040,147.4560,11.8243,69.2808,34.6561,4.9509
200,8x4x32,8,8192,688.1280,196.6080,294.9120,23.6673,69.2263,34.6833,4.9548
200,8x4x64,8,16384,1376.2560,393.2160,589.8240,48.1857,68.0036,35.3069,5.0438
200,8x8x2,8,1024,86.0160,24.5760,36.8640,3.0384,67.4040,35.6211,5.0887
200,8x8x4,8,2048,172.0320,49.1520,73.7280,5.9124,69.2776,34.6577,4.9511
200,8x8x8,8,4096,344.0640,98.3040,147.4560,11.8272,69.2643,34.6643,4.9520
200,8x8x16,8,8192,688.1280,196.6080,294.9120,24.0210,68.2070,35.2016,5.0288
200,8x8x32,8,16384,1376.2560,393.2160,589.8240,48.3581,67.7612,35.4333,5.0619
200,8x8x64,8,32768,2752.5120,786.4320,1179.6480,95.4919,68.6299,34.9847,4.9978
200,8x16x2,8,2048,172.0320,49.1520,73.7280,5.9123,69.2792,34.6569,4.9510
200,8x16x4,8,4096,344.0640,98.3040,147.4560,11.8242,69.2818,34.6556,4.9508
200,8x16x8,8,8192,688.1280,196.6080,294.9120,23.6666,69.2284,34.6823,4.9546
200,8x16x16,8,16384,1376.2560,393.2160,589.8240,47.3484,69.2062,34.6934,4.9562
200,8x16x32,8,32768,2752.5120,786.4320,1179.6480,94.6531,69.2381,34.6775,4.9539
200,8x16x64,8,65536,5505.0240,1572.8640,2359.2960,189.5620,69.1447,34.7243,4.9606
200,8x32x2,8,4096,344.0640,98.3040,147.4560,11.8244,69.2807,34.6561,4.9509
200,8x32x4,8,8192,688.1280,196.6080,294.9120,23.6701,69.2182,34.6874,4.9553
200,8x32x8,8,16384,1376.2560,393.2160,589.8240,47.3417,69.2159,34.6886,4.9555
200,8x32x16,8,32768,2752.5120,786.4320,1179.6480,94.6549,69.2368,34.6781,4.9540
200,8x32x32,8,65536,5505.0240,1572.8640,2359.2960,193.0646,67.8902,35.3659,5.0523
200,8x32x64,8,131072,11010.0480,3145.7280,4718.5920,385.1419,68.0643,35.2755,5.0394
200,8x64x2,8,8192,688.1280,196.6080,294.9120,23.6709,69.2158,34.6886,4.9555
200,8x64x4,8,16384,1376.2560,393.2160,589.8240,48.3610,67.7570,35.4354,5.0622
200,8x64x8,8,32768,2752.5120,786.4320,1179.6480,94.6743,69.2226,34.6852,4.9550
200,8x64x16,8,65536,5505.0240,1572.8640,2359.2960,190.5796,68.7754,34.9107,4.9872
200,8x64x32,8,131072,11010.0480,3145.7280,4718.5920,385.7335,67.9599,35.3297,5.0471
200,8x64x64,8,262144,22020.0960,6291.4560,9437.1840,771.4760,67.9591,35.3301,5.0472
200,16x4x2,8,1024,86.0160,24.5760,36.8640,2.9560,69.2828,34.6550,4.9507
200,16x4x4,8,2048,172.0320,49.1520,73.7280,6.6205,61.8689,38.8079,5.5440
200,16x4x8,8,4096,344.0640,98.3040,147.4560,12.3473,66.3466,36.1888,5.1698
200,16x4x16,8,8192,688.1280,196.6080,294.9120,23.6669,69.2274,34.6828,4.9547
200,16x4x32,8,16384,1376.2560,393.2160,589.8240,47.3374,69.2223,34.6854,4.9551
200,16x4x64,8,32768,2752.5120,786.4320,1179.6480,95.5037,68.6214,34.9891,4.9984
200,16x8x2,8,2048,172.0320,49.1520,73.7280,5.9119,69.2840,34.6545,4.9506
200,16x8x4,8,4096,344.0640,98.3040,147.4560,11.8263,69.2694,34.6618,4.9517
200,16x8x8,8,8192,688.1280,196.6080,294.9120,24.5150,66.8326,35.9256,5.1322
200,16x8x16,8,16384,1376.2560,393.2160,589.8240,47.3509,69.2025,34.6953,4.9565
200,16x8x32,8,32768,2752.5120,786.4320,1179.6480,95.0141,68.9750,34.8097,4.9728
200,16x8x64,8,65536,5505.0240,1572.8640,2359.2960,193.1030,67.8767,35.3730,5.0533
200,16x16x2,8,4096,344.0640,98.3040,147.4560,11.8264,69.2689,34.6620,4.9517
200,16x16x4,8,8192,688.1280,196.6080,294.9120,23.6685,69.2228,34.6851,4.9550
200,16x16x8,8,16384,1376.2560,393.2160,589.8240,48.5268,67.5256,35.5569,5.0796
200,16x16x16,8,32768,2752.5120,786.4320,1179.6480,95.7067,68.4759,35.0634,5.0091
200,16x16x32,8,65536,5505.0240,1572.8640,2359.2960,192.0975,68.2320,35.1888,5.0270
200,16x16x64,8,131072,11010.0480,3145.7280,4718.5920,383.9017,68.2841,35.1619,5.0231
200,16x32x2,8,8192,688.1280,196.6080,294.9120,23.6651,69.2326,34.6802,4.9543
200,16x32x4,8,16384,1376.2560,393.2160,589.8240,47.3449,69.2113,34.6909,4.9558
200,16x32x8,8,32768,2752.5120,786.4320,1179.6480,98.2238,66.7211,35.9856,5.1408
200,16x32x16,8,65536,5505.0240,1572.8640,2359.2960,193.6725,67.6771,35.4773,5.0682
200,16x32x32,8,131072,11010.0480,3145.7280,4718.5920,381.8818,68.6453,34.9769,4.9967
200,16x32x64,8,262144,22020.0960,6291.4560,9437.1840,767.4085,68.3193,35.1438,5.0205
200,16x64x2,8,16384,1376.2560,393.2160,589.8240,47.6878,68.7136,34.9421,4.9917
200,16x64x4,8,32768,2752.5120,786.4320,1179.6480,102.4829,63.9482,37.5460,5.3637
200,16x64x8,8,65536,5505.0240,1572.8640,2359.2960,192.9239,67.9398,35.3401,5.0486
200,16x64x16,8,131072,11010.0480,3145.7280,4718.5920,381.0739,68.7909,34.9029,4.9861
200,16x64x32,8,262144,22020.0960,6291.4560,9437.1840,766.5159,68.3988,35.1029,5.0147
200,16x64x64,8,524288,44040.1920,12582.9120,18874.3680,1541.4866,68.0237,35.2965,5.0424
200,32x4x2,8,2048,172.0320,49.1520,73.7280,5.9127,69.2747,34.6591,4.9513
200,32x4x4,8,4096,344.0640,98.3040,147.4560,11.8241,69.2825,34.6552,4.9507
200,32x4x8,8,8192,688.1280,196.6080,294.9120,23.6679,69.2245,34.6842,4.9549
200,32x4x16,8,16384,1376.2560,393.2160,589.8240,47.3350,69.2258,34.6836,4.9548
200,32x4x32,8,32768,2752.5120,786.4320,1179.6480,95.6797,68.4952,35.0535,5.0076
200,32x4x64,8,65536,5505.0240,1572.8640,2359.2960,189.5597,69.1455,34.7239,4.9606
200,32x8x2,8,4096,344.0640,98.3040,147.4560,11.8242,69.2816,34.6557,4.9508
200,32x8x4,8,8192,688.1280,196.6080,294.9120,23.6709,69.2159,34.6886,4.9555
200,32x8x8,8,16384,1376.2560,393.2160,589.8240,47.3351,69.2256,34.6837,4.9548
200,32x8x16,8,32768,2752.5120,786.4320,1179.6480,94.6586,69.2341,34.6794,4.9542
200,32x8x32,8,65536,5505.0240,1572.8640,2359.2960,191.5835,68.4151,35.0946,5.0135
200,32x8x64,8,131072,11010.0480,3145.7280,4718.5920,383.8995,68.2845,35.1617,5.0231
200,32x16x2,8,8192,688.1280,196.6080,294.9120,24.3550,67.2715,35.6912,5.0987
200,32x16x4,8,16384,1376.2560,393.2160,589.8240,48.0213,68.2364,35.1865,5.0266
200,32x16x8,8,32768,2752.5120,786.4320,1179.6480,97.0362,67.5377,35.5505,5.0786
200,32x16x16,8,65536,5505.0240,1572.8640,2359.2960,193.7625,67.6457,35.4938,5.0705