ClusterCockpit/cc-metric-collector
1
0
Fork 0
mirror of https://github.com/ClusterCockpit/cc-metric-collector.git synced 2025-07-31 08:56:06 +02:00
Code Issues Packages Projects Releases Wiki Activity

Add likwid collector

Browse Source
This commit is contained in:
Thomas Roehl
2021-03-25 14:47:10 +01:00
parent 4fddcb9741
commit a6ac0c5373
670 changed files with 24926 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

30
collectors/likwid/groups/power8/BRANCH.txt Normal file
View File

@@ -0,0 +1,30 @@
SHORT Branch prediction miss rate/ratio
EVENTSET
PMC0 PM_BR_PRED_BR_CMPL
PMC1 PM_BR_PRED_CCACHE_CMPL
PMC2 PM_BR_PRED_CR_CMPL
PMC3 PM_BR_MPRED_CMPL
PMC4 PM_RUN_INST_CMPL
PMC5 PM_RUN_CYC
METRICS
Runtime (RDTSC) [s] time
CPI PMC5/PMC4
Branch rate (PMC0+PMC1+PMC2)/PMC4
Branch misprediction rate PMC3/PMC4
Branch misprediction ratio PMC4/(PMC0+PMC1+PMC2)
Instructions per branch PMC4/(PMC0+PMC1+PMC2)
LONG
Formulas:
Branch rate = BR_INST_RETIRED_ALL_BRANCHES/INSTR_RETIRED_ANY
Branch misprediction rate = BR_MISP_RETIRED_ALL_BRANCHES/INSTR_RETIRED_ANY
Branch misprediction ratio = BR_MISP_RETIRED_ALL_BRANCHES/BR_INST_RETIRED_ALL_BRANCHES
Instructions per branch = INSTR_RETIRED_ANY/BR_INST_RETIRED_ALL_BRANCHES
-
The rates state how often in average a branch or a mispredicted branch occured
per instruction retired in total. The Branch misprediction ratio sets directly
into relation what ratio of all branch instruction where mispredicted.
Instructions per branch is 1/Branch rate.

35
collectors/likwid/groups/power8/CPISTACK1.txt Normal file
View File

@@ -0,0 +1,35 @@
SHORT First level of IBM CPI stack
EVENTSET
PMC0 PM_CMPLU_STALL_THRD
PMC1 PM_GCT_EMPTY_CYC
PMC3 PM_CMPLU_STALL
PMC4 PM_RUN_INST_CMPL
PMC5 PM_RUN_CYC
METRICS
CPI PMC5/PMC4
Stall cycles PMC3
Stall cycle ratio PMC3/PMC5
Thread blocked cycles PMC0
Thread blocked cycle ratio PMC0/PMC5
GCT empty cycles PMC1
GCT empty cycle ratio PMC1/PM5
LONG
Formulas:
Stall cycles = PM_CMPLU_STALL
Stall cycle ratio = PM_CMPLU_STALL/PM_RUN_CYC
Thread blocked cycles = PM_CMPLU_STALL_THRD
Thread blocked cycle ratio = PM_CMPLU_STALL_THRD/PM_RUN_CYC
GCT empty cycles = PM_GCT_EMPTY_CYC
GCT empty cycle ratio = PM_GCT_EMPTY_CYC/PM_RUN_CYC
--
First level of IBM CPI stack. IBM names Stalled Cycles, Waiting to Complete,
Thread Blocked, Completion Table Empty, Other and Completion Cycles. For some
there are no clearly identifiable events, so this group concentrates on
Stalled Cycles (PM_CMPLU_STALL), Thread Blocked (PM_CMPLU_STALL_THRD),
Completion Table Empty (PM_GCT_EMPTY_CYC) and Other (PM_CMPLU_STALL_OTHER_CMPL).

23
collectors/likwid/groups/power8/DATA.txt Normal file
View File

@@ -0,0 +1,23 @@
SHORT Load to store ratio
EVENTSET
PMC0 PM_LD_CMPL
PMC1 PM_ST_CMPL
PMC4 PM_RUN_INST_CMPL
PMC5 PM_RUN_CYC
METRICS
Runtime (RDTSC) [s] time
CPI PMC5/PMC4
Load to store ratio PMC0/PMC1
Load ratio PMC0/PMC4
Store ratio PMC1/PMC4
LONG
Formulas:
Load to store ratio = PM_LD_CMPL/PM_ST_CMPL
Load ratio = PM_LD_CMPL/PM_RUN_INST_CMPL
Store ratio = PM_ST_CMPL/PM_RUN_INST_CMPL
-
This is a metric to determine your load to store ratio.

24
collectors/likwid/groups/power8/FLOPS_1_2.txt Normal file
View File

@@ -0,0 +1,24 @@
SHORT Group 121 as used in IBM Parallel Environment Developer Edition
EVENTSET
PMC0 PM_VSU0_1FLOP
PMC1 PM_VSU1_1FLOP
PMC2 PM_VSU0_2FLOP
PMC3 PM_VSU1_2FLOP
PMC4 PM_RUN_INST_CMPL
PMC5 PM_RUN_CYC
METRICS
CPI PMC5/PMC4
One FLOP ops PMC0+PMC1
Two FLOPs ops PMC2+PMC3
[MFLOP/s] 1E-6*(PMC0+PMC1+((PMC2+PMC3)*2))/time
LONG
Formulas:
CPI = PM_RUN_CYC/PM_RUN_INST_CMPL
One FLOP ops = PM_VSU0_1FLOP+PM_VSU1_1FLOP
Two FLOPs ops = PM_VSU0_2FLOP+PM_VSU1_2FLOP
[MFLOP/s] = 1E-6*(PM_VSU0_1FLOP+PM_VSU1_1FLOP+((PM_VSU0_2FLOP+PM_VSU1_2FLOP)*2))/time
--
Group 121 from web page http://www.ibm.com/support/knowledgecenter/en/SSFK5S_2.2.0/com.ibm.cluster.pedev.v2r2.pedev100.doc/bl7ug_power8metrics.htm

24
collectors/likwid/groups/power8/FLOPS_4_8.txt Normal file
View File

@@ -0,0 +1,24 @@
SHORT Group 122 as used in IBM Parallel Environment Developer Edition
EVENTSET
PMC0 PM_VSU0_4FLOP
PMC1 PM_VSU1_4FLOP
PMC2 PM_VSU0_8FLOP
PMC3 PM_VSU1_8FLOP
PMC4 PM_RUN_INST_CMPL
PMC5 PM_RUN_CYC
METRICS
CPI PMC5/PMC4
Four FLOPs ops PMC0+PMC1
Eight FLOPs ops PMC2+PMC3
MFLOP/s 1E-6*(((PMC0+PMC1)*4.0)+((PMC2+PMC3)*8.0))/time
LONG
Formulas:
CPI = PM_RUN_CYC/PM_RUN_INST_CMPL
Four FLOPs ops = PM_VSU0_4FLOP+PM_VSU1_4FLOP
Eight FLOPs ops = PM_VSU0_8FLOP+PM_VSU1_8FLOP
MFLOP/s = 1E-6*(((PM_VSU0_4FLOP+PM_VSU1_4FLOP)*4.0)+((PM_VSU0_8FLOP+PM_VSU1_8FLOP)*8.0))/time
--
Group 122 from web page http://www.ibm.com/support/knowledgecenter/en/SSFK5S_2.2.0/com.ibm.cluster.pedev.v2r2.pedev100.doc/bl7ug_power8metrics.htm

27
collectors/likwid/groups/power8/FLOPS_DP.txt Normal file
View File

@@ -0,0 +1,27 @@
SHORT Double Precision MFlops/s
EVENTSET
PMC0 PM_VSU0_DP_2FLOP
PMC1 PM_VSU0_DP_FMA
PMC2 PM_VSU0_DP_FSQRT_FDIV
PMC3 PM_VSU0_SCALAR_DP_ISSUED
PMC4 PM_RUN_INST_CMPL
PMC5 PM_RUN_CYC
METRICS
Runtime (RDTSC) [s] time
CPI PMC5/PMC4
DP [MFLOP/s] 1.0E-06*((PMC0*2.0)+PMC2+(PMC1*4.0))/time
DP VSX [MFLOP/s] 1.0E-06*((PMC1*4.0)+(PMC0*2.0))/time
Packed [MUOPS/s] 1.0E-06*(PMC1)/time
Scalar [MUOPS/s] 1.0E-06*(PMC0+PMC2)/time
LONG
Formulas:
CPI = PM_RUN_CYC/PM_RUN_INST_CMPL
DP [MFLOP/s] = 1.0E-06*(PM_VSU0_SCALAR_DP_ISSUED+PM_VSU1_SCALAR_DP_ISSUED+(PM_VSU0_VECTOR_DP_ISSUED+PM_VSU1_VECTOR_DP_ISSUED)*4)/runtime
DP VSX [MFLOP/s] = 1.0E-06*((PM_VSU0_VECTOR_DP_ISSUED+PM_VSU1_VECTOR_DP_ISSUED)*4)/runtime
Packed [MUOPS/s] = 1.0E-06*(PM_VSU0_VECTOR_DP_ISSUED+PM_VSU1_VECTOR_DP_ISSUED)/runtime
Scalar [MUOPS/s] = 1.0E-06*(PM_VSU0_SCALAR_DP_ISSUED+PM_VSU1_SCALAR_DP_ISSUED)/runtime
--

27
collectors/likwid/groups/power8/FLOPS_DP2.txt Normal file
View File

@@ -0,0 +1,27 @@
SHORT Double Precision MFlops/s
EVENTSET
PMC0 PM_VSU1_DP_2FLOP
PMC1 PM_VSU1_DP_FMA
PMC2 PM_VSU1_DP_FSQRT_FDIV
PMC3 PM_VSU1_SCALAR_DP_ISSUED
PMC4 PM_RUN_INST_CMPL
PMC5 PM_RUN_CYC
METRICS
Runtime (RDTSC) [s] time
CPI PMC5/PMC4
DP [MFLOP/s] 1.0E-06*(PMC0+PMC2+(PMC1)*4.0)/time
DP VSX [MFLOP/s] 1.0E-06*((PMC1)*4.0)/time
Packed [MUOPS/s] 1.0E-06*(PMC1)/time
Scalar [MUOPS/s] 1.0E-06*(PMC0+PMC2)/time
LONG
Formulas:
CPI = PM_RUN_CYC/PM_RUN_INST_CMPL
DP [MFLOP/s] = 1.0E-06*(PM_VSU0_SCALAR_DP_ISSUED+PM_VSU1_SCALAR_DP_ISSUED+(PM_VSU0_VECTOR_DP_ISSUED+PM_VSU1_VECTOR_DP_ISSUED)*4)/runtime
DP VSX [MFLOP/s] = 1.0E-06*((PM_VSU0_VECTOR_DP_ISSUED+PM_VSU1_VECTOR_DP_ISSUED)*4)/runtime
Packed [MUOPS/s] = 1.0E-06*(PM_VSU0_VECTOR_DP_ISSUED+PM_VSU1_VECTOR_DP_ISSUED)/runtime
Scalar [MUOPS/s] = 1.0E-06*(PM_VSU0_SCALAR_DP_ISSUED+PM_VSU1_SCALAR_DP_ISSUED)/runtime
--

28
collectors/likwid/groups/power8/FLOPS_FMA.txt Normal file
View File

@@ -0,0 +1,28 @@
SHORT Group 124 as used in IBM Parallel Environment Developer Edition
EVENTSET
PMC0 PM_VSU0_DP_FMA
PMC1 PM_VSU1_DP_FMA
PMC2 PM_VSU0_FMA
PMC3 PM_VSU1_FMA
PMC4 PM_RUN_INST_CMPL
PMC5 PM_RUN_CYC
METRICS
CPI PMC5/PMC4
DP FMAs PMC0+PMC1
Scalar FMAs PMC2+PMC3
DP FMA [MFLOP/s] 1E-6*(PMC0+PMC1)*4.0/time
Scalar FMA [MFLOP/s] 1E-6*(PMC2+PMC3)*2.0/time
[MFLOP/s] 1E-6*(((PMC0+PMC1)*4.0)+((PMC2+PMC3)*2.0))/time
LONG
Formulas:
CPI = PM_RUN_CYC/PM_RUN_INST_CMPL
DP FMAs = PM_VSU0_DP_FMA+PM_VSU1_DP_FMA
Scalar FMAs = PM_VSU0_FMA+PM_VSU1_FMA
DP FMA [MFLOP/s] = 1E-6*(PM_VSU0_DP_FMA+PM_VSU1_DP_FMA)*4.0/runtime
Scalar FMA [MFLOP/s] = 1E-6*(PM_VSU0_FMA+PM_VSU1_FMA)*2.0/runtime
[MFLOP/s] = 1E-6*(((PM_VSU0_DP_FMA+PM_VSU1_DP_FMA)*4.0)+((PM_VSU0_FMA+PM_VSU1_FMA)*2.0))/runtime
--
Group 124 from web page http://www.ibm.com/support/knowledgecenter/en/SSFK5S_2.2.0/com.ibm.cluster.pedev.v2r2.pedev100.doc/bl7ug_power8metrics.htm

27
collectors/likwid/groups/power8/FLOPS_SP.txt Normal file
View File

@@ -0,0 +1,27 @@
SHORT Double Precision MFlops/s
EVENTSET
PMC0 PM_VSU0_SINGLE
PMC1 PM_VSU0_VECTOR_SP_ISSUED
PMC2 PM_VSU1_SINGLE
PMC3 PM_VSU1_VECTOR_SP_ISSUED
PMC4 PM_RUN_INST_CMPL
PMC5 PM_RUN_CYC
METRICS
Runtime (RDTSC) [s] time
CPI PMC5/PMC4
SP [MFLOP/s] 1.0E-06*(((PMC0-PMC1)+(PMC2-PMC3))*4.0+(PMC1+PMC3)*8.0)/time
SP VSX [MFLOP/s] 1.0E-06*((PMC1+PMC3)*8.0)/time
Packed [MUOPS/s] 1.0E-06*(PMC1+PMC3)/time
Scalar [MUOPS/s] 1.0E-06*(PMC0+PMC2)/time
LONG
Formulas:
CPI = PM_RUN_CYC/PM_RUN_INST_CMPL
SP [MFLOP/s] = 1.0E-06*(PM_VSU0_SINGLE+PM_VSU1_SINGLE+(PM_VSU0_VECTOR_SP_ISSUED+PM_VSU1_VECTOR_SP_ISSUED)*8)/runtime
SP VSX [MFLOP/s] = 1.0E-06*((PM_VSU0_VECTOR_SP_ISSUED+PM_VSU1_VECTOR_SP_ISSUED)*8)/runtime
Packed [MUOPS/s] = 1.0E-06*(PM_VSU0_VECTOR_SP_ISSUED+PM_VSU1_VECTOR_SP_ISSUED)/runtime
Scalar [MUOPS/s] = 1.0E-06*(PM_VSU0_SINGLE+PM_VSU1_SINGLE)/runtime
--

23
collectors/likwid/groups/power8/FLOPS_VSU0.txt Normal file
View File

@@ -0,0 +1,23 @@
SHORT Double Precision MFlops/s performed by VSU pipe 0
EVENTSET
PMC0 PM_VSU0_DP_2FLOP
PMC1 PM_VSU0_DP_FMA
PMC2 PM_VSU0_DP_FSQRT_FDIV
PMC3 PM_VSU0_1FLOP
PMC4 PM_RUN_INST_CMPL
PMC5 PM_RUN_CYC
METRICS
Runtime (RDTSC) [s] time
CPI PMC5/PMC4
[MFLOP/s] 1.0E-06*((PMC0*2.0)+(PMC2*8.0)+(PMC1*4.0)+PMC3)/time
VSX [MFLOP/s] 1.0E-06*(PMC1*4.0)/time
LONG
Formulas:
CPI = PM_RUN_CYC/PM_RUN_INST_CMPL
[MFLOP/s] = 1.0E-06*(PM_VSU0_SCALAR_DP_ISSUED+PM_VSU0_SCALAR_DP_ISSUED+(PM_VSU0_VECTOR_DP_ISSUED+PM_VSU0_VECTOR_DP_ISSUED)*4)/runtime
VSX [MFLOP/s] = 1.0E-06*((PM_VSU0_VECTOR_DP_ISSUED+PM_VSU1_VECTOR_DP_ISSUED)*4)/runtime
--

22
collectors/likwid/groups/power8/FLOPS_VSU1.txt Normal file
View File

@@ -0,0 +1,22 @@
SHORT Double Precision MFlops/s performed by VSU pipe 1
EVENTSET
PMC0 PM_VSU1_DP_2FLOP
PMC1 PM_VSU1_DP_FMA
PMC2 PM_VSU1_DP_FSQRT_FDIV
PMC3 PM_VSU1_1FLOP
PMC4 PM_RUN_INST_CMPL
PMC5 PM_RUN_CYC
METRICS
Runtime (RDTSC) [s] time
CPI PMC5/PMC4
[MFLOP/s] 1.0E-06*((PMC0*2.0)+(PMC2*8.0)+(PMC1*4.0)+PMC3)/time
VSX [MFLOP/s] 1.0E-06*(PMC1*4.0)/time
LONG
Formulas:
CPI = PM_RUN_CYC/PM_RUN_INST_CMPL
[MFLOP/s] = 1.0E-06*(PM_VSU1_SCALAR_DP_ISSUED+PM_VSU1_SCALAR_DP_ISSUED+(PM_VSU1_VECTOR_DP_ISSUED+PM_VSU1_VECTOR_DP_ISSUED)*4)/runtime
VSX [MFLOP/s] = 1.0E-06*((PM_VSU0_VECTOR_DP_ISSUED+PM_VSU1_VECTOR_DP_ISSUED)*4)/runtime
--

29
collectors/likwid/groups/power8/FLOPS_VSX.txt Normal file
View File

@@ -0,0 +1,29 @@
SHORT Vectorized MFlops/s
EVENTSET
PMC0 PM_VSU0_VECTOR_DP_ISSUED
PMC1 PM_VSU1_VECTOR_DP_ISSUED
PMC2 PM_VSU0_VECTOR_SP_ISSUED
PMC3 PM_VSU1_VECTOR_SP_ISSUED
PMC4 PM_RUN_INST_CMPL
PMC5 PM_RUN_CYC
METRICS
Runtime (RDTSC) [s] time
CPI PMC5/PMC4
[MFLOP/s] 1.0E-06*((PMC0+PMC1)*4.0+(PMC2+PMC3)*8.0)/time
DP [MFLOP/s] 1.0E-06*((PMC0+PMC1)*4.0)/time
SP [MFLOP/s] 1.0E-06*((PMC2+PMC3)*8.0)/time
DP [MUOPS/s] 1.0E-06*(PMC0+PMC1)/time
SP [MUOPS/s] 1.0E-06*(PMC2+PMC3)/time
LONG
Formulas:
CPI = PM_RUN_CYC/PM_RUN_INST_CMPL
[MFLOP/s] = 1.0E-06*((PM_VSU0_VECTOR_DP_ISSUED+PM_VSU1_VECTOR_DP_ISSUED)*4.0+(PM_VSU0_VECTOR_SP_ISSUED+PM_VSU1_VECTOR_SP_ISSUED)*8.0)/runtime
DP [MFLOP/s] = 1.0E-06*((PM_VSU0_VECTOR_DP_ISSUED+PM_VSU1_VECTOR_DP_ISSUED)*4.0)/runtime
SP [MFLOP/s] = 1.0E-06*((PM_VSU0_VECTOR_SP_ISSUED+PM_VSU1_VECTOR_SP_ISSUED)*8.0)/runtime
DP [MUOPS/s] = 1.0E-06*(PM_VSU0_VECTOR_DP_ISSUED+PM_VSU1_VECTOR_DP_ISSUED)/runtime
SP [MUOPS/s] = 1.0E-06*(PM_VSU0_VECTOR_SP_ISSUED+PM_VSU1_VECTOR_SP_ISSUED)/runtime
--

22
collectors/likwid/groups/power8/ICACHE.txt Normal file
View File

@@ -0,0 +1,22 @@
SHORT Instruction cache miss rate/ratio
EVENTSET
PMC0 PM_INST_FROM_L1
PMC1 PM_L1_ICACHE_MISS
PMC4 PM_RUN_INST_CMPL
PMC5 PM_RUN_CYC
METRICS
Runtime (RDTSC) [s] time
CPI PMC5/PMC4
L1I request rate PMC0/PMC4
L1I miss rate PMC1/PMC4
L1I miss ratio PMC1/PMC0
LONG
Formulas:
L1I request rate = ICACHE_ACCESSES / INSTR_RETIRED_ANY
L1I miss rate = ICACHE_MISSES / INSTR_RETIRED_ANY
L1I miss ratio = ICACHE_MISSES / ICACHE_ACCESSES
-
This group measures some L1 instruction cache metrics.

33
collectors/likwid/groups/power8/L1.txt Normal file
View File

@@ -0,0 +1,33 @@
SHORT L2 cache bandwidth in MBytes/s
EVENTSET
PMC0 PM_LD_REF_L1
PMC1 PM_ST_CMPL
PMC2 PM_LSU_L1_PREF
PMC4 PM_RUN_INST_CMPL
PMC5 PM_RUN_CYC
METRICS
Runtime (RDTSC) [s] time
CPI PMC5/PMC4
L2D load bandwidth [MBytes/s] 1.0E-06*((PMC0+PMC2)/2)*64.0/time
L2D load data volume [GBytes] 1.0E-09*((PMC0+PMC2)/2)*64.0
L2D store bandwidth [MBytes/s] 1.0E-06*((PMC1/2))*64.0/time
L2D store data volume [GBytes] 1.0E-09*((PMC1/2))*64.0
L2 bandwidth [MBytes/s] 1.0E-06*((PMC1+PMC0+PMC2)/2)*64.0/time
L2 data volume [GBytes] 1.0E-09*((PMC1+PMC0+PMC2)/2)*64.0
LONG
Formulas:
CPI = PM_RUN_CYC/PM_RUN_INST_CMPL
L2D load bandwidth [MBytes/s] = 1.0E-06*(PM_DATA_FROM_L2/2)*128.0/time
L2D load data volume [GBytes] = 1.0E-09*(PM_DATA_FROM_L2/2)*128.0
L2D store bandwidth [MBytes/s] = 1.0E-06*(PM_ST_MISS_L1)*128.0/time
L2D store data volume [GBytes] = 1.0E-09*(PM_ST_MISS_L1)*128.0
L2 bandwidth [MBytes/s] = 1.0E-06*(PM_DATA_FROM_L2/2 + PM_ST_MISS_L1)*128.0/time
L2 data volume [GBytes] = 1.0E-09*(PM_DATA_FROM_L2/2 + PM_ST_MISS_L1)*128.0
-
Profiling group to measure L2 cache bandwidth. The bandwidth is computed by the
number of cacheline loaded from the L2 to the L1 data cache. There is currently no
event to get the evicted data volume.

32
collectors/likwid/groups/power8/L2.txt Normal file
View File

@@ -0,0 +1,32 @@
SHORT L2 cache bandwidth in MBytes/s
EVENTSET
PMC0 PM_L2_ST
PMC2 PM_LD_MISS_L1
PMC4 PM_RUN_INST_CMPL
PMC5 PM_RUN_CYC
METRICS
Runtime (RDTSC) [s] time
CPI PMC5/PMC4
L2D load bandwidth [MBytes/s] 1.0E-06*(PMC2/2)*128.0/time
L2D load data volume [GBytes] 1.0E-09*(PMC2/2)*128.0
L2D store bandwidth [MBytes/s] 1.0E-06*(PMC0/2)*128.0/time
L2D store data volume [GBytes] 1.0E-09*(PMC0/2)*128.0
L2 bandwidth [MBytes/s] 1.0E-06*((PMC0+PMC2)/2)*128.0/time
L2 data volume [GBytes] 1.0E-09*((PMC0+PMC2)/2)*128.0
LONG
Formulas:
CPI = PM_RUN_CYC/PM_RUN_INST_CMPL
L2D load bandwidth [MBytes/s] = 1.0E-06*(PM_DATA_FROM_L2/2)*128.0/time
L2D load data volume [GBytes] = 1.0E-09*(PM_DATA_FROM_L2/2)*128.0
L2D store bandwidth [MBytes/s] = 1.0E-06*(PM_ST_CMPL/2)*128.0/time
L2D store data volume [GBytes] = 1.0E-09*(PM_ST_CMPL/2)*128.0
L2 bandwidth [MBytes/s] = 1.0E-06*((PM_DATA_FROM_L2 + PM_ST_CMPL))*128.0/time
L2 data volume [GBytes] = 1.0E-09*((PM_DATA_FROM_L2 + PM_ST_CMPL))*128.0
-
Profiling group to measure L2 cache bandwidth. The bandwidth is computed by the
number of cacheline loaded from the L2 to the L1 data cache. There is currently no
event to get the evicted data volume.

40
collectors/likwid/groups/power8/L2CACHE.txt Normal file
View File

@@ -0,0 +1,40 @@
SHORT L2 cache miss rate/ratio
EVENTSET
PMC0 PM_L2_ST_MISS
PMC1 PM_L2_LD_MISS
PMC2 PM_L2_LD_DISP
PMC3 PM_L2_ST_DISP
PMC4 PM_RUN_INST_CMPL
PMC5 PM_RUN_CYC
METRICS
Runtime (RDTSC) [s] time
CPI PMC5/PMC4
L2 request rate = (PMC2+PMC3)/PMC4
L2 miss rate = (PMC0+PMC1)/PMC4
L2 miss ratio = (PMC0+PMC1)/(PMC2+PMC3)
LONG
Formulas:
L2 request rate = (PM_L2_LD_DISP+PM_L2_ST_DISP)/PM_RUN_INST_CMPL
L2 miss rate = (PM_L2_LD_MISS+PM_L2_ST_MISS)/PM_RUN_INST_CMPL
L2 miss ratio = (PM_L2_LD_MISS+PM_L2_ST_MISS)/(PM_L2_LD_DISP+PM_L2_ST_DISP)
L2 load request rate = PM_L2_LD_DISP/PM_RUN_INST_CMPL
L2 store request rate = PM_L2_ST_DISP/PM_RUN_INST_CMPL
L2 load miss rate = PM_L2_LD_MISS/PM_RUN_INST_CMPL
L2 store miss rate = PM_L2_ST_DISP/PM_RUN_INST_CMPL
L2 load miss ratio = PM_L2_LD_MISS/(PM_L2_LD_DISP+PM_L2_ST_DISP)
L2 store miss ratio = PM_L2_ST_MISS/(PM_L2_LD_DISP+PM_L2_ST_DISP)
-
This group measures the locality of your data accesses with regard to the
L2 Cache. L2 request rate tells you how data intensive your code is
or how many Data accesses you have in average per instruction.
The L2 miss rate gives a measure how often it was necessary to get
cachelines from memory. And finally L2 miss ratio tells you how many of your
memory references required a cacheline to be loaded from a higher level.
While the Data cache miss rate might be given by your algorithm you should
try to get Data cache miss ratio as low as possible by increasing your cache reuse.

31
collectors/likwid/groups/power8/L3.txt Normal file
View File

@@ -0,0 +1,31 @@
SHORT L3 cache bandwidth in MBytes/s
EVENTSET
PMC0 PM_L3_LD_PREF
PMC3 PM_DATA_FROM_L3
PMC4 PM_RUN_INST_CMPL
PMC5 PM_RUN_CYC
METRICS
Runtime (RDTSC) [s] time
CPI PMC5/PMC4
L3D load bandwidth [MBytes/s] 1.0E-06*(PMC3+(PMC0-PMC3))*128.0/time
L3D load data volume [GBytes] 1.0E-09*(PMC3+(PMC0-PMC3))*128.0
L3 bandwidth [MBytes/s] 1.0E-06*(PMC3+(PMC0-PMC3))*128.0/time
L3 data volume [GBytes] 1.0E-09*(PMC3+(PMC0-PMC3))*128.0
Loads from local L3 per cycle 100.0*(PMC3+(PMC0-PMC3))/PMC5
LONG
Formulas:
CPI = PM_RUN_CYC/PM_RUN_INST_CMPL
L3D load bandwidth [MBytes/s] = 1.0E-06*(PM_DATA_FROM_L3)*128.0/time
L3D load data volume [GBytes] = 1.0E-09*(PM_DATA_FROM_L3)*128.0
L3D evict bandwidth [MBytes/s] = 1.0E-06*(PM_L2_CASTOUT_MOD)*128.0/time
L3D evict data volume [GBytes] = 1.0E-09*(PM_L2_CASTOUT_MOD)*128.0
L3 bandwidth [MBytes/s] = 1.0E-06*(PM_DATA_FROM_L3+PM_L2_CASTOUT_MOD)*128.0/time
L3 data volume [GBytes] = 1.0E-09*(PM_DATA_FROM_L3+PM_L2_CASTOUT_MOD)*128.0
-
Profiling group to measure L3 cache bandwidth. The bandwidth is computed by the
number of cacheline loaded from the L3 to the L2 data cache. There is currently no
event to get the evicted data volume.

30
collectors/likwid/groups/power8/MEM.txt Normal file
View File

@@ -0,0 +1,30 @@
SHORT Main memory bandwidth in MBytes/s
EVENTSET
PMC0 PM_L3_CO_MEPF
PMC1 PM_DATA_ALL_FROM_MEMORY
PMC3 PM_L3_PF_ON_CHIP_MEM
PMC4 PM_RUN_INST_CMPL
PMC5 PM_RUN_CYC
METRICS
Runtime (RDTSC) [s] time
CPI PMC5/PMC4
Memory load bandwidth [MBytes/s] 1.0E-06*(PMC1+PMC3)*128.0/time
Memory load data volume [GBytes] 1.0E-09*(PMC1+PMC3)*128.0
Memory evict bandwidth [MBytes/s] 1.0E-06*(PMC0)*128.0/time
Memory evict data volume [GBytes] 1.0E-09*(PMC0)*128.0
Memory bandwidth [MBytes/s] 1.0E-06*(PMC1+PMC3+PMC0)*128.0/time
Memory data volume [GBytes] 1.0E-09*(PMC1+PMC3+PMC0)*128.0
LONG
Formulas:
CPI = PM_RUN_CYC / PM_RUN_INST_CMPL
Memory load bandwidth [MBytes/s] = 1.0E-06* (PM_DATA_ALL_FROM_MEMORY)*128/time
Memory load data volume [GBytes] = 1.0E-09* (PM_DATA_ALL_FROM_MEMORY)*128
Memory evict bandwidth [MBytes/s] = 1.0E-06* (PM_MEM_CO)*128/time
Memory evict data volume [GBytes] = 1.0E-09* (PM_MEM_CO)*128
Memory bandwidth [MBytes/s] = 1.0E-06* (PM_DATA_ALL_FROM_MEMORY+PM_MEM_CO)*128/time
Memory data volume [GBytes] = 1.0E-09* (PM_DATA_ALL_FROM_MEMORY+PM_MEM_CO)*128
--
This group uses the core-local events to measure data traffic from memory.

29
collectors/likwid/groups/power8/NUMA.txt Normal file
View File

@@ -0,0 +1,29 @@
SHORT Memory bandwidth in MBytes/s for local and remote memory
EVENTSET
PMC1 PM_DATA_ALL_FROM_LMEM
PMC3 PM_DATA_ALL_FROM_DMEM
PMC4 PM_RUN_INST_CMPL
PMC5 PM_RUN_CYC
METRICS
Runtime (RDTSC) [s] time
CPI PMC5/PMC4
Local bandwidth [MBytes/s] 1.0E-06*(PMC1)*128.0/time
Local data volume [GBytes] 1.0E-09*(PMC1)*128.0
Remote bandwidth [MBytes/s] 1.0E-06*(PMC3)*128.0/time
Remote data volume [GBytes] 1.0E-09*(PMC3)*128.0
Memory load bandwidth [MBytes/s] 1.0E-06*(PMC1+PMC3)*128.0/time
Memory load data volume [GBytes] 1.0E-09*(PMC1+PMC3)*128.0
LONG
Formulas:
CPI = PM_RUN_CYC / PM_RUN_INST_CMPL
Memory load bandwidth [MBytes/s] = 1.0E-06* (PM_DATA_ALL_FROM_MEMORY)*128/time
Memory load data volume [GBytes] = 1.0E-09* (PM_DATA_ALL_FROM_MEMORY)*128
Memory evict bandwidth [MBytes/s] = 1.0E-06* (PM_MEM_CO)*128/time
Memory evict data volume [GBytes] = 1.0E-09* (PM_MEM_CO)*128
Memory bandwidth [MBytes/s] = 1.0E-06* (PM_DATA_ALL_FROM_MEMORY+PM_MEM_CO)*128/time
Memory data volume [GBytes] = 1.0E-09* (PM_DATA_ALL_FROM_MEMORY+PM_MEM_CO)*128
--
This group measures the NUMA traffic by separating local from remote memory data transfers.

33
collectors/likwid/groups/power8/STALLS1.txt Normal file
View File

@@ -0,0 +1,33 @@
SHORT Completion stalls (group 1)
EVENTSET
PMC0 PM_CMPLU_STALL_THRD
PMC1 PM_CMPLU_STALL_DCACHE_MISS
PMC2 PM_CMPLU_STALL_COQ_FULL
PMC3 PM_CMPLU_STALL
PMC4 PM_RUN_INST_CMPL
PMC5 PM_RUN_CYC
METRICS
Runtime time
CPI PMC5/PMC4
Completion stall cycles PMC3
Stall cycles by thread conflict PMC0
Stall ratio by thread conflict [%] PMC0/PMC3*100.0
Stall cycles by d-cache miss PMC1
Stall ratio by d-cache miss [%] PMC1/PMC3*100.0
Stall cycles by full castout queue PMC2
Stall ratio by full castout queue [%] PMC2/PMC3*100.0
LONG
Formulas:
CPI = PM_RUN_CYC / PM_RUN_INST_CMPL
Completion stall cycles = PM_CMPLU_STALL
Stall cycles by thread conflict = PM_CMPLU_STALL_THRD
Stall ratio by thread conflict [%] = PM_CMPLU_STALL_THRD/PM_CMPLU_STALL*100
Stall cycles by d-cache miss = PM_CMPLU_STALL_DCACHE_MISS
Stall ratio by d-cache miss [%] = PM_CMPLU_STALL_DCACHE_MISS/PM_CMPLU_STALL*100
Stall cycles by full castout queue = PM_CMPLU_STALL_COQ_FULL
Stall ratio by full castout queue [%] = PM_CMPLU_STALL_COQ_FULL/PM_CMPLU_STALL*100
--

32
collectors/likwid/groups/power8/STALLS2.txt Normal file
View File

@@ -0,0 +1,32 @@
SHORT Completion stalls (group 2)
EVENTSET
PMC0 PM_CMPLU_STALL
PMC1 PM_CMPLU_STALL_LSU
PMC2 PM_CMPLU_STALL_FLUSH
PMC3 PM_CMPLU_STALL_BRU
PMC4 PM_RUN_INST_CMPL
PMC5 PM_RUN_CYC
METRICS
CPI PMC5/PMC4
Stall cycles PMC0
Stall cycles by load/store unit PMC1
Stall ratio by load/store unit [%] PMC1/PMC0*100.0
Stall cycles by pipeline flush PMC2
Stall ratio by pipeline flush [%] PMC2/PMC0*100.0
Stall cycles by branch unit PMC3
Stall ratio by branch unit [%] PMC3/PMC0*100.0
LONG
Formulas:
CPI = PM_RUN_CYC / PM_RUN_INST_CMPL
Stall cycles = PM_CMPLU_STALL
Stall cycles by load/store unit = PM_CMPLU_STALL_LSU
Stall ratio by load/store unit [%] = PM_CMPLU_STALL_LSU/PM_CMPLU_STALL*100.0
Stall cycles by pipeline flush = PM_CMPLU_STALL_FLUSH
Stall ratio by pipeline flush [%] = PM_CMPLU_STALL_FLUSH/PM_CMPLU_STALL*100.0
Stall cycles by branch unit = PM_CMPLU_STALL_BRU
Stall ratio by branch unit [%] = PM_CMPLU_STALL_BRU/PM_CMPLU_STALL*100.0
--

37
collectors/likwid/groups/power8/TLB_DATA.txt Normal file
View File

@@ -0,0 +1,37 @@
SHORT L1 Data TLB miss rate/ratio
EVENTSET
PMC0 PM_DTLB_MISS_16G
PMC1 PM_DTLB_MISS_4K
PMC2 PM_DTLB_MISS_64K
PMC3 PM_DTLB_MISS_16M
PMC4 PM_RUN_INST_CMPL
PMC5 PM_RUN_CYC
METRICS
Runtime (RDTSC) [s] time
CPI PMC5/PMC4
L1 DTLB 4K misses PMC1
L1 DTLB 4K miss rate PMC1/PMC4
L1 DTLB 64K misses PMC2
L1 DTLB 64K miss rate PMC2/PMC4
L1 DTLB 16M misses PMC3
L1 DTLB 16M miss rate PMC3/PMC4
L1 DTLB 16G misses PMC0
L1 DTLB 16G miss rate PMC0/PMC4
LONG
Formulas:
CPI = PM_RUN_CYC / PM_RUN_INST_CMPL
L1 DTLB 4K misses = PM_DTLB_MISS_4K
L1 DTLB 4K miss rate = PM_DTLB_MISS_4K/PM_RUN_INST_CMPL
L1 DTLB 64K misses = PM_DTLB_MISS_64K
L1 DTLB 64K miss rate = PM_DTLB_MISS_64K/PM_RUN_INST_CMPL
L1 DTLB 16M misses = PM_DTLB_MISS_16M
L1 DTLB 16M miss rate = PM_DTLB_MISS_16M/PM_RUN_INST_CMPL
L1 DTLB 16G misses = PM_DTLB_MISS_16G
L1 DTLB 16G miss rate = PM_DTLB_MISS_16G/PM_RUN_INST_CMPL
--
The DTLB load and store miss rates gives a measure how often a TLB miss occured
per instruction.

21
collectors/likwid/groups/power8/TLB_INSTR.txt Normal file
View File

@@ -0,0 +1,21 @@
SHORT L1 Instruction TLB miss rate/ratio
EVENTSET
PMC2 PM_ITLB_MISS
PMC4 PM_RUN_INST_CMPL
PMC5 PM_RUN_CYC
METRICS
Runtime (RDTSC) [s] time
CPI PMC5/PMC4
L1 ITLB misses PMC2
L1 ITLB miss rate PMC2/PMC4
LONG
Formulas:
CPI = PM_RUN_CYC / PM_RUN_INST_CMPL
L1 ITLB misses = PM_ITLB_MISS
L1 ITLB miss rate = PM_ITLB_MISS/PM_RUN_INST_CMPL
--
The ITLB miss rate gives a measure how often a TLB miss occured per instruction.

24
collectors/likwid/groups/power8/USEFUL.txt Normal file
View File

@@ -0,0 +1,24 @@
SHORT Rate of useful instructions
EVENTSET
PMC0 PM_IOPS_CMPL
PMC1 PM_INST_DISP
PMC2 PM_IOPS_DISP
PMC4 PM_RUN_INST_CMPL
PMC5 PM_RUN_CYC
METRICS
CPI PMC5/PMC4
Useful instr. rate PMC4/PMC1*100.0
Useful uops rate PMC0/PMC2*100.0
LONG
Formulas:
CPI = PM_RUN_CYC / PM_RUN_INST_CMPL
Useful instr. rate = PM_RUN_INST_CMPL/PM_INST_DISP*100.0
Useful uops rate = PM_IOPS_CMPL/PM_IOPS_DISP*100.0
--
This groups measures how many of the dispatched instructions and internal operations (uops) are
acutally completed. These metrics show the speculatively dispatches instructions compared to the
completed instructions.