Compare commits
No commits in common. "main" and "gromacs_masking" have entirely different histories.
main
...
gromacs_ma
176
.clang-format
176
.clang-format
@ -1,176 +0,0 @@
|
||||
---
|
||||
Language: Cpp
|
||||
# BasedOnStyle: WebKit
|
||||
AccessModifierOffset: -4
|
||||
AlignAfterOpenBracket: DontAlign
|
||||
AlignArrayOfStructures: None
|
||||
AlignConsecutiveAssignments: Consecutive
|
||||
AlignConsecutiveBitFields: None
|
||||
AlignConsecutiveDeclarations: None
|
||||
AlignConsecutiveMacros: Consecutive
|
||||
AlignEscapedNewlines: Right
|
||||
AlignOperands: Align
|
||||
AlignTrailingComments: true
|
||||
AllowAllArgumentsOnNextLine: false
|
||||
AllowAllParametersOfDeclarationOnNextLine: true
|
||||
AllowShortEnumsOnASingleLine: true
|
||||
AllowShortBlocksOnASingleLine: Never
|
||||
AllowShortCaseLabelsOnASingleLine: false
|
||||
AllowShortFunctionsOnASingleLine: All
|
||||
AllowShortLambdasOnASingleLine: All
|
||||
AllowShortIfStatementsOnASingleLine: OnlyFirstIf
|
||||
AllowShortLoopsOnASingleLine: false
|
||||
AlwaysBreakAfterDefinitionReturnType: None
|
||||
AlwaysBreakAfterReturnType: None
|
||||
AlwaysBreakBeforeMultilineStrings: false
|
||||
AlwaysBreakTemplateDeclarations: MultiLine
|
||||
AttributeMacros:
|
||||
- __capability
|
||||
BinPackArguments: false
|
||||
BinPackParameters: false
|
||||
BraceWrapping:
|
||||
AfterCaseLabel: false
|
||||
AfterClass: false
|
||||
AfterControlStatement: Never
|
||||
AfterEnum: false
|
||||
AfterFunction: true
|
||||
AfterNamespace: false
|
||||
AfterObjCDeclaration: false
|
||||
AfterStruct: false
|
||||
AfterUnion: false
|
||||
AfterExternBlock: false
|
||||
BeforeCatch: false
|
||||
BeforeElse: false
|
||||
BeforeLambdaBody: false
|
||||
BeforeWhile: false
|
||||
IndentBraces: false
|
||||
SplitEmptyFunction: true
|
||||
SplitEmptyRecord: true
|
||||
SplitEmptyNamespace: true
|
||||
BreakBeforeBinaryOperators: None
|
||||
BreakBeforeBraces: WebKit
|
||||
BreakBeforeInheritanceComma: false
|
||||
BreakInheritanceList: BeforeColon
|
||||
BreakBeforeTernaryOperators: true
|
||||
BreakConstructorInitializersBeforeComma: false
|
||||
BreakConstructorInitializers: BeforeComma
|
||||
BreakAfterJavaFieldAnnotations: false
|
||||
BreakStringLiterals: true
|
||||
ColumnLimit: 90
|
||||
CommentPragmas: '^ IWYU pragma:'
|
||||
CompactNamespaces: false
|
||||
ConstructorInitializerIndentWidth: 4
|
||||
ContinuationIndentWidth: 4
|
||||
Cpp11BracedListStyle: false
|
||||
DeriveLineEnding: true
|
||||
DerivePointerAlignment: false
|
||||
DisableFormat: false
|
||||
EmptyLineAfterAccessModifier: Never
|
||||
EmptyLineBeforeAccessModifier: LogicalBlock
|
||||
ExperimentalAutoDetectBinPacking: false
|
||||
BasedOnStyle: ''
|
||||
ConstructorInitializerAllOnOneLineOrOnePerLine: false
|
||||
AllowAllConstructorInitializersOnNextLine: true
|
||||
FixNamespaceComments: false
|
||||
ForEachMacros:
|
||||
- foreach
|
||||
- Q_FOREACH
|
||||
- BOOST_FOREACH
|
||||
IfMacros:
|
||||
- KJ_IF_MAYBE
|
||||
IncludeBlocks: Preserve
|
||||
IncludeCategories:
|
||||
- Regex: '^"(llvm|llvm-c|clang|clang-c)/'
|
||||
Priority: 2
|
||||
SortPriority: 0
|
||||
CaseSensitive: false
|
||||
- Regex: '^(<|"(gtest|gmock|isl|json)/)'
|
||||
Priority: 3
|
||||
SortPriority: 0
|
||||
CaseSensitive: false
|
||||
- Regex: '.*'
|
||||
Priority: 1
|
||||
SortPriority: 0
|
||||
CaseSensitive: false
|
||||
IncludeIsMainRegex: '(Test)?$'
|
||||
IncludeIsMainSourceRegex: ''
|
||||
IndentAccessModifiers: false
|
||||
IndentCaseLabels: false
|
||||
IndentCaseBlocks: false
|
||||
IndentGotoLabels: true
|
||||
IndentPPDirectives: None
|
||||
IndentExternBlock: AfterExternBlock
|
||||
IndentWidth: 4
|
||||
IndentWrappedFunctionNames: false
|
||||
InsertTrailingCommas: None
|
||||
JavaScriptQuotes: Leave
|
||||
JavaScriptWrapImports: true
|
||||
KeepEmptyLinesAtTheStartOfBlocks: true
|
||||
LambdaBodyIndentation: Signature
|
||||
MacroBlockBegin: ''
|
||||
MacroBlockEnd: ''
|
||||
MaxEmptyLinesToKeep: 1
|
||||
NamespaceIndentation: Inner
|
||||
ObjCBinPackProtocolList: Auto
|
||||
ObjCBlockIndentWidth: 4
|
||||
ObjCBreakBeforeNestedBlockParam: true
|
||||
ObjCSpaceAfterProperty: true
|
||||
ObjCSpaceBeforeProtocolList: true
|
||||
PenaltyBreakAssignment: 200
|
||||
PenaltyBreakBeforeFirstCallParameter: 19
|
||||
PenaltyBreakComment: 300
|
||||
PenaltyBreakFirstLessLess: 120
|
||||
PenaltyBreakString: 1000
|
||||
PenaltyBreakTemplateDeclaration: 10
|
||||
PenaltyExcessCharacter: 1000000
|
||||
PenaltyReturnTypeOnItsOwnLine: 60
|
||||
PenaltyIndentedWhitespace: 0
|
||||
PointerAlignment: Left
|
||||
PPIndentWidth: -1
|
||||
ReferenceAlignment: Pointer
|
||||
ReflowComments: true
|
||||
ShortNamespaceLines: 1
|
||||
SortIncludes: CaseSensitive
|
||||
SortJavaStaticImport: Before
|
||||
SortUsingDeclarations: true
|
||||
SpaceAfterCStyleCast: false
|
||||
SpaceAfterLogicalNot: false
|
||||
SpaceAfterTemplateKeyword: true
|
||||
SpaceBeforeAssignmentOperators: true
|
||||
SpaceBeforeCaseColon: false
|
||||
SpaceBeforeCpp11BracedList: true
|
||||
SpaceBeforeCtorInitializerColon: true
|
||||
SpaceBeforeInheritanceColon: true
|
||||
SpaceBeforeParens: ControlStatements
|
||||
SpaceAroundPointerQualifiers: Default
|
||||
SpaceBeforeRangeBasedForLoopColon: true
|
||||
SpaceInEmptyBlock: false
|
||||
SpaceInEmptyParentheses: false
|
||||
SpacesBeforeTrailingComments: 1
|
||||
SpacesInAngles: Never
|
||||
SpacesInConditionalStatement: false
|
||||
SpacesInContainerLiterals: true
|
||||
SpacesInCStyleCastParentheses: false
|
||||
SpacesInLineCommentPrefix:
|
||||
Minimum: 1
|
||||
Maximum: -1
|
||||
SpacesInParentheses: false
|
||||
SpacesInSquareBrackets: false
|
||||
SpaceBeforeSquareBrackets: false
|
||||
BitFieldColonSpacing: Both
|
||||
Standard: Latest
|
||||
StatementAttributeLikeMacros:
|
||||
- Q_EMIT
|
||||
StatementMacros:
|
||||
- Q_UNUSED
|
||||
- QT_REQUIRE_VERSION
|
||||
TabWidth: 8
|
||||
UseCRLF: false
|
||||
UseTab: Never
|
||||
WhitespaceSensitiveMacros:
|
||||
- STRINGIZE
|
||||
- PP_STRINGIZE
|
||||
- BOOST_PP_STRINGIZE
|
||||
- NS_SWIFT_NAME
|
||||
- CF_SWIFT_NAME
|
||||
...
|
||||
14
.clang-tidy
14
.clang-tidy
@ -1,14 +0,0 @@
|
||||
---
|
||||
Checks: 'clang-diagnostic-*,clang-analyzer-*,clang-bugprone-*,readability-identifier-naming'
|
||||
WarningsAsErrors: true
|
||||
HeaderFilterRegex: '.*'
|
||||
AnalyzeTemporaryDtors: false
|
||||
CheckOptions:
|
||||
- key: readability-identifier-naming.StructCase
|
||||
value: 'CamelCase'
|
||||
- key: readability-identifier-naming.FunctionCase
|
||||
value: 'camelBack'
|
||||
- key: readability-identifier-naming.VariableCase
|
||||
value: 'camelBack'
|
||||
- key: readability-identifier-naming.GlobalConstantCase
|
||||
value: 'UPPER_CASE'
|
||||
3
.clangd
3
.clangd
@ -1,3 +0,0 @@
|
||||
CompileFlags:
|
||||
Add: [-I/Users/jan/prg/MD-Bench/src/verletlist/, -I/Users/jan/prg/MD-Bench/src/common/, -DALIGNMENT=64]
|
||||
Compiler: clang
|
||||
129
Makefile
129
Makefile
@ -1,32 +1,116 @@
|
||||
#CONFIGURE BUILD SYSTEM
|
||||
TAG = $(OPT_TAG)-$(TOOLCHAIN)-$(DATA_TYPE)
|
||||
TARGET = MDBench-$(TAG)
|
||||
BUILD_DIR = ./build/build-$(TAG)
|
||||
SRC_ROOT = ./src
|
||||
SRC_DIR = $(SRC_ROOT)/$(OPT_SCHEME)
|
||||
COMMON_DIR = $(SRC_ROOT)/common
|
||||
CUDA_DIR = $(SRC_DIR)/cuda
|
||||
MAKE_DIR = ./make
|
||||
IDENTIFIER = $(OPT_SCHEME)-$(TAG)-$(ISA)-$(DATA_TYPE)
|
||||
TARGET = MDBench-$(IDENTIFIER)
|
||||
BUILD_DIR = ./build-$(IDENTIFIER)
|
||||
SRC_DIR = ./$(OPT_SCHEME)
|
||||
ASM_DIR = ./asm
|
||||
COMMON_DIR = ./common
|
||||
CUDA_DIR = ./$(SRC_DIR)/cuda
|
||||
MAKE_DIR = ./
|
||||
Q ?= @
|
||||
|
||||
#DO NOT EDIT BELOW
|
||||
include config.mk
|
||||
include $(MAKE_DIR)/include_$(TOOLCHAIN).mk
|
||||
include $(MAKE_DIR)/config.mk
|
||||
include $(MAKE_DIR)/include_$(TAG).mk
|
||||
include $(MAKE_DIR)/include_LIKWID.mk
|
||||
ifneq ($(strip $(ISA)),NONE)
|
||||
include $(MAKE_DIR)/include_ISA.mk
|
||||
endif
|
||||
INCLUDES += -I./$(SRC_DIR) -I./$(COMMON_DIR)
|
||||
include $(MAKE_DIR)/include_GROMACS.mk
|
||||
INCLUDES += -I./$(SRC_DIR)/includes -I./$(COMMON_DIR)/includes
|
||||
|
||||
VPATH = $(SRC_DIR) $(COMMON_DIR) $(CUDA_DIR)
|
||||
ifeq ($(strip $(DATA_LAYOUT)),AOS)
|
||||
DEFINES += -DAOS
|
||||
endif
|
||||
ifeq ($(strip $(DATA_TYPE)),SP)
|
||||
DEFINES += -DPRECISION=1
|
||||
else
|
||||
DEFINES += -DPRECISION=2
|
||||
endif
|
||||
|
||||
ifneq ($(ASM_SYNTAX), ATT)
|
||||
ASFLAGS += -masm=intel
|
||||
endif
|
||||
|
||||
ifeq ($(strip $(EXPLICIT_TYPES)),true)
|
||||
DEFINES += -DEXPLICIT_TYPES
|
||||
endif
|
||||
|
||||
ifeq ($(strip $(MEM_TRACER)),true)
|
||||
DEFINES += -DMEM_TRACER
|
||||
endif
|
||||
|
||||
ifeq ($(strip $(INDEX_TRACER)),true)
|
||||
DEFINES += -DINDEX_TRACER
|
||||
endif
|
||||
|
||||
ifeq ($(strip $(COMPUTE_STATS)),true)
|
||||
DEFINES += -DCOMPUTE_STATS
|
||||
endif
|
||||
|
||||
ifeq ($(strip $(XTC_OUTPUT)),true)
|
||||
DEFINES += -DXTC_OUTPUT
|
||||
endif
|
||||
|
||||
ifeq ($(strip $(USE_REFERENCE_VERSION)),true)
|
||||
DEFINES += -DUSE_REFERENCE_VERSION
|
||||
endif
|
||||
|
||||
ifeq ($(strip $(HALF_NEIGHBOR_LISTS_CHECK_CJ)),true)
|
||||
DEFINES += -DHALF_NEIGHBOR_LISTS_CHECK_CJ
|
||||
endif
|
||||
|
||||
ifeq ($(strip $(DEBUG)),true)
|
||||
DEFINES += -DDEBUG
|
||||
endif
|
||||
|
||||
ifneq ($(VECTOR_WIDTH),)
|
||||
DEFINES += -DVECTOR_WIDTH=$(VECTOR_WIDTH)
|
||||
endif
|
||||
|
||||
ifeq ($(strip $(__SIMD_KERNEL__)),true)
|
||||
DEFINES += -D__SIMD_KERNEL__
|
||||
endif
|
||||
|
||||
ifeq ($(strip $(__SSE__)),true)
|
||||
DEFINES += -D__ISA_SSE__
|
||||
endif
|
||||
|
||||
ifeq ($(strip $(__ISA_AVX__)),true)
|
||||
DEFINES += -D__ISA_AVX__
|
||||
endif
|
||||
|
||||
ifeq ($(strip $(__ISA_AVX_FMA__)),true)
|
||||
DEFINES += -D__ISA_AVX_FMA__
|
||||
endif
|
||||
|
||||
ifeq ($(strip $(__ISA_AVX2__)),true)
|
||||
DEFINES += -D__ISA_AVX2__
|
||||
endif
|
||||
|
||||
ifeq ($(strip $(__ISA_AVX512__)),true)
|
||||
DEFINES += -D__ISA_AVX512__
|
||||
endif
|
||||
|
||||
ifeq ($(strip $(ENABLE_OMP_SIMD)),true)
|
||||
DEFINES += -DENABLE_OMP_SIMD
|
||||
endif
|
||||
|
||||
ifeq ($(strip $(USE_SIMD_KERNEL)),true)
|
||||
DEFINES += -DUSE_SIMD_KERNEL
|
||||
endif
|
||||
|
||||
VPATH = $(SRC_DIR) $(ASM_DIR) $(CUDA_DIR)
|
||||
ASM = $(patsubst $(SRC_DIR)/%.c, $(BUILD_DIR)/%.s,$(wildcard $(SRC_DIR)/*.c))
|
||||
OBJ = $(filter-out $(BUILD_DIR)/main%, $(patsubst $(SRC_DIR)/%.c, $(BUILD_DIR)/%.o,$(wildcard $(SRC_DIR)/*.c)))
|
||||
OBJ += $(patsubst $(COMMON_DIR)/%.c, $(BUILD_DIR)/%.o,$(wildcard $(COMMON_DIR)/*.c))
|
||||
OVERWRITE:= $(patsubst $(ASM_DIR)/%-new.s, $(BUILD_DIR)/%.o,$(wildcard $(ASM_DIR)/*-new.s))
|
||||
OBJ = $(filter-out $(BUILD_DIR)/main% $(OVERWRITE),$(patsubst $(SRC_DIR)/%.c, $(BUILD_DIR)/%.o,$(wildcard $(SRC_DIR)/*.c)))
|
||||
OBJ += $(patsubst $(ASM_DIR)/%.s, $(BUILD_DIR)/%.o,$(wildcard $(ASM_DIR)/*.s))
|
||||
OBJ += $(patsubst $(COMMON_DIR)/%.c, $(BUILD_DIR)/%-common.o,$(wildcard $(COMMON_DIR)/*.c))
|
||||
ifeq ($(strip $(TAG)),NVCC)
|
||||
OBJ += $(patsubst $(CUDA_DIR)/%.cu, $(BUILD_DIR)/%-cuda.o,$(wildcard $(CUDA_DIR)/*.cu))
|
||||
endif
|
||||
CPPFLAGS := $(CPPFLAGS) $(DEFINES) $(OPTIONS) $(INCLUDES)
|
||||
|
||||
# $(warning $(OBJ))
|
||||
|
||||
ifneq ($(VARIANT),)
|
||||
.DEFAULT_GOAL := ${TARGET}-$(VARIANT)
|
||||
DEFINES += -DVARIANT=$(VARIANT)
|
||||
@ -45,6 +129,11 @@ $(BUILD_DIR)/%.o: %.c
|
||||
$(Q)$(CC) -c $(CPPFLAGS) $(CFLAGS) $< -o $@
|
||||
$(Q)$(CC) $(CPPFLAGS) -MT $@ -MM $< > $(BUILD_DIR)/$*.d
|
||||
|
||||
$(BUILD_DIR)/%-common.o: $(COMMON_DIR)/%.c
|
||||
$(info ===> COMPILE $@)
|
||||
$(Q)$(CC) -c $(CPPFLAGS) $(CFLAGS) $< -o $@
|
||||
$(Q)$(CC) $(CPPFLAGS) -MT $@ -MM $< > $(BUILD_DIR)/$*.d
|
||||
|
||||
$(BUILD_DIR)/%-cuda.o: %.cu
|
||||
$(info ===> COMPILE $@)
|
||||
$(Q)$(CC) -c $(CPPFLAGS) $(CFLAGS) $< -o $@
|
||||
@ -63,16 +152,18 @@ $(BUILD_DIR)/%.o: %.s
|
||||
clean:
|
||||
$(info ===> CLEAN)
|
||||
@rm -rf $(BUILD_DIR)
|
||||
@rm -rf MDBench-$(IDENTIFIER)
|
||||
@rm -f tags
|
||||
|
||||
cleanall:
|
||||
$(info ===> CLEAN)
|
||||
@rm -rf build
|
||||
@rm -rf build-*
|
||||
@rm -rf MDBench-*
|
||||
@rm -f tags
|
||||
|
||||
distclean: clean
|
||||
$(info ===> DIST CLEAN)
|
||||
@rm -f $(TARGET)
|
||||
@rm -f $(TARGET)*
|
||||
@rm -f tags
|
||||
|
||||
info:
|
||||
@ -86,6 +177,6 @@ tags:
|
||||
$(Q)ctags -R
|
||||
|
||||
$(BUILD_DIR):
|
||||
@mkdir -p $(BUILD_DIR)
|
||||
@mkdir $(BUILD_DIR)
|
||||
|
||||
-include $(OBJ:.o=.d)
|
||||
|
||||
54
README.md
54
README.md
@ -1,14 +1,34 @@
|
||||
# MD-Bench
|
||||
|
||||
MD-Bench is a toolbox for the performance engineering of short-range force
|
||||
calculation kernels on molecular-dynamics applications. It aims at covering all
|
||||
available state-of-the-art algorithms from different community codes such as
|
||||
LAMMPS and GROMACS.
|
||||
|
||||
|
||||
MD-Bench is a toolbox for the performance engineering of short-range force calculation kernels on molecular-dynamics applications.
|
||||
It aims at covering all available state-of-the-art algorithms from different community codes such as LAMMPS and GROMACS.
|
||||
|
||||
Apart from that, many tools to study and evaluate the in-depth performance of such kernels on distinct hardware are offered, like gather-bench, a standalone benchmark that mimics the data movement from MD kernels and the stubbed force calculation cases that focus on isolating the impacts caused by memory latencies and control flow divergence contributions in the overall performance.
|
||||
|
||||
<table>
|
||||
<thead>
|
||||
<tr>
|
||||
<th>Verlet Lists</th>
|
||||
<th>GROMACS MxN</th>
|
||||
<th>Stubbed cases</th>
|
||||
</tr>
|
||||
</thead>
|
||||
<tbody>
|
||||
<tr>
|
||||
<td><a target="_blank" rel="noopener noreferrer" href="figures/verlet_v2.png"><img src="figures/verlet_v2.png" alt="Image" title="Verlet Lists" style="width: 100%;"></a></td>
|
||||
<td><a target="_blank" rel="noopener noreferrer" href="figures/gromacs_mxn_v2.png"><img src="figures/gromacs_mxn_v2.png" alt="Image" title="GROMACS MxN" style="width: 90%;"></a></td>
|
||||
<td><a target="_blank" rel="noopener noreferrer" href="figures/stub_new_v3.png"><img src="figures/stub_new_v3.png" alt="Image" title="Stubbed cases" style="width: 100%;"></a></td>
|
||||
</tr>
|
||||
</tbody>
|
||||
</table>
|
||||
|
||||
<!--  -->
|
||||
|
||||
## Build instructions
|
||||
|
||||
Properly configure your building by changing `config.mk` file. The following
|
||||
options are available:
|
||||
Properly configure your building by changing `config.mk` file. The following options are available:
|
||||
|
||||
- **TAG:** Compiler tag (available options: GCC, CLANG, ICC, ONEAPI, NVCC).
|
||||
- **ISA:** Instruction set (available options: SSE, AVX, AVX\_FMA, AVX2, AVX512).
|
||||
@ -25,18 +45,15 @@ options are available:
|
||||
- **COMPUTE\_STATS:** Compute statistics.
|
||||
|
||||
Configurations for LAMMPS Verlet Lists optimization scheme:
|
||||
|
||||
- **ENABLE\_OMP\_SIMD:** Use omp simd pragma on half neighbor-lists kernels.
|
||||
- **USE\_SIMD\_KERNEL:** Compile kernel with explicit SIMD intrinsics.
|
||||
|
||||
Configurations for GROMACS MxN optimization scheme:
|
||||
|
||||
- **USE\_REFERENCE\_VERSION:** Use reference version (only for correction purposes).
|
||||
- **XTC\_OUTPUT:** Enable XTC output.
|
||||
- **HALF\_NEIGHBOR\_LISTS\_CHECK\_CJ:** Check if j-clusters are local when decreasing the reaction force.
|
||||
|
||||
Configurations for CUDA:
|
||||
|
||||
- **USE\_CUDA\_HOST\_MEMORY:** Use CUDA host memory to optimize host-device transfers.
|
||||
|
||||
When done, just use `make` to compile the code.
|
||||
@ -51,14 +68,11 @@ Use the following command to run a simulation:
|
||||
./MD-Bench-<TAG>-<OPT_SCHEME> [OPTION]...
|
||||
```
|
||||
|
||||
Where `TAG` and `OPT_SCHEME` correspond to the building options with the same
|
||||
name. Without any options, a Copper FCC lattice system with size 32x32x32
|
||||
(131072 atoms) over 200 time-steps using the Lennard-Jones potential (sigma=1.0,
|
||||
epsilon=1.0) is simulated.
|
||||
Where `TAG` and `OPT_SCHEME` correspond to the building options with the same name.
|
||||
Without any options, a Copper FCC lattice system with size 32x32x32 (131072 atoms) over 200 time-steps using the Lennard-Jones potential (sigma=1.0, epsilon=1.0) is simulated.
|
||||
|
||||
The default behavior and other options can be changed using the following parameters:
|
||||
|
||||
```sh
|
||||
```
|
||||
-p <string>: file to read parameters from (can be specified more than once)
|
||||
-f <string>: force field (lj or eam), default lj
|
||||
-i <string>: input file with atom positions (dump)
|
||||
@ -78,17 +92,11 @@ TBD
|
||||
|
||||
## Citations
|
||||
|
||||
Rafael Ravedutti Lucio Machado, Jan Eitzinger, Harald Köstler, and Gerhard
|
||||
Wellein: MD-Bench: A generic proxy-app toolbox for state-of-the-art molecular
|
||||
dynamics algorithms. Accepted for [PPAM](https://ppam.edu.pl/) 2022, the 14th
|
||||
International Conference on Parallel Processing and Applied Mathematics, Gdansk,
|
||||
Poland, September 11-14, 2022. PPAM 2022 Best Paper Award. Preprint:
|
||||
[arXiv:2207.13094](https://arxiv.org/abs/2207.13094)
|
||||
Rafael Ravedutti Lucio Machado, Jan Eitzinger, Harald Köstler, and Gerhard Wellein: MD-Bench: A generic proxy-app toolbox for state-of-the-art molecular dynamics algorithms. Accepted for [PPAM](https://ppam.edu.pl/) 2022, the 14th International Conference on Parallel Processing and Applied Mathematics, Gdansk, Poland, September 11-14, 2022. PPAM 2022 Best Paper Award. Preprint: [arXiv:2207.13094](https://arxiv.org/abs/2207.13094)
|
||||
|
||||
## Credits
|
||||
|
||||
MD-Bench is developed by the Erlangen National High Performance Computing Center
|
||||
([NHR@FAU](https://hpc.fau.de/)) at the University of Erlangen-Nürnberg.
|
||||
MD-Bench is developed by the Erlangen National High Performance Computing Center ([NHR@FAU](https://hpc.fau.de/)) at the University of Erlangen-Nürnberg.
|
||||
|
||||
## License
|
||||
|
||||
|
||||
0
asm/.gitkeep
Normal file
0
asm/.gitkeep
Normal file
626
asm/unused/force-mem-only-with-likwid.s
Normal file
626
asm/unused/force-mem-only-with-likwid.s
Normal file
@ -0,0 +1,626 @@
|
||||
# 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/mnt/opt/likwid-5.2-dev/include -I./src/includes -S -D_GNU_SOURCE -DLIKWID_PERFMON -DAOS -DPRECISION=2 -DN";
|
||||
# mark_description "EIGHBORS_LOOP_RUNS=1 -DVECTOR_WIDTH=8 -DALIGNMENT=64 -restrict -Ofast -xCORE-AVX512 -qopt-zmm-usage=high -o ";
|
||||
# mark_description "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, int, int)
|
||||
computeForce:
|
||||
# parameter 1: %rdi
|
||||
# parameter 2: %rsi
|
||||
# parameter 3: %rdx
|
||||
# parameter 4: %ecx
|
||||
# parameter 5: %r8d
|
||||
# parameter 6: %r9d
|
||||
..B1.1: # Preds ..B1.0
|
||||
# Execution count [1.00e+00]
|
||||
.cfi_startproc
|
||||
..___tag_value_computeForce.1:
|
||||
..L2:
|
||||
#121.112
|
||||
pushq %rbp #121.112
|
||||
.cfi_def_cfa_offset 16
|
||||
movq %rsp, %rbp #121.112
|
||||
.cfi_def_cfa 6, 16
|
||||
.cfi_offset 6, -16
|
||||
andq $-64, %rsp #121.112
|
||||
pushq %r12 #121.112
|
||||
pushq %r13 #121.112
|
||||
pushq %r14 #121.112
|
||||
pushq %r15 #121.112
|
||||
pushq %rbx #121.112
|
||||
subq $88, %rsp #121.112
|
||||
xorl %eax, %eax #124.16
|
||||
.cfi_escape 0x10, 0x03, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xd8, 0xff, 0xff, 0xff, 0x22
|
||||
.cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xf8, 0xff, 0xff, 0xff, 0x22
|
||||
.cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xf0, 0xff, 0xff, 0xff, 0x22
|
||||
.cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xe8, 0xff, 0xff, 0xff, 0x22
|
||||
.cfi_escape 0x10, 0x0f, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x22
|
||||
movq %rdx, %r15 #121.112
|
||||
movq %rsi, %r12 #121.112
|
||||
movq %rdi, %rbx #121.112
|
||||
..___tag_value_computeForce.11:
|
||||
# getTimeStamp()
|
||||
call getTimeStamp #124.16
|
||||
..___tag_value_computeForce.12:
|
||||
# LOE rbx r12 r15 xmm0
|
||||
..B1.51: # Preds ..B1.1
|
||||
# Execution count [1.00e+00]
|
||||
vmovsd %xmm0, 24(%rsp) #124.16[spill]
|
||||
# LOE rbx r12 r15
|
||||
..B1.2: # Preds ..B1.51
|
||||
# Execution count [1.00e+00]
|
||||
movl 4(%r12), %r13d #125.18
|
||||
movq 64(%r12), %r9 #127.20
|
||||
movq 72(%r12), %r14 #127.45
|
||||
movq 80(%r12), %r8 #127.70
|
||||
vmovsd 72(%rbx), %xmm2 #129.27
|
||||
vmovsd 8(%rbx), %xmm1 #130.23
|
||||
vmovsd (%rbx), %xmm0 #131.24
|
||||
testl %r13d, %r13d #134.24
|
||||
jle ..B1.43 # Prob 50% #134.24
|
||||
# LOE r8 r9 r12 r14 r15 r13d xmm0 xmm1 xmm2
|
||||
..B1.3: # Preds ..B1.2
|
||||
# Execution count [1.00e+00]
|
||||
xorl %ebx, %ebx #134.5
|
||||
movl %r13d, %edx #134.5
|
||||
xorl %ecx, %ecx #134.5
|
||||
movl $1, %esi #134.5
|
||||
xorl %eax, %eax #135.17
|
||||
shrl $1, %edx #134.5
|
||||
je ..B1.7 # Prob 9% #134.5
|
||||
# LOE rax rdx rcx rbx r8 r9 r12 r14 r15 esi r13d xmm0 xmm1 xmm2
|
||||
..B1.5: # Preds ..B1.3 ..B1.5
|
||||
# Execution count [2.50e+00]
|
||||
movq %rax, (%rcx,%r9) #135.9
|
||||
incq %rbx #134.5
|
||||
movq %rax, (%rcx,%r14) #136.9
|
||||
movq %rax, (%rcx,%r8) #137.9
|
||||
movq %rax, 8(%rcx,%r9) #135.9
|
||||
movq %rax, 8(%rcx,%r14) #136.9
|
||||
movq %rax, 8(%rcx,%r8) #137.9
|
||||
addq $16, %rcx #134.5
|
||||
cmpq %rdx, %rbx #134.5
|
||||
jb ..B1.5 # Prob 63% #134.5
|
||||
# LOE rax rdx rcx rbx r8 r9 r12 r14 r15 r13d xmm0 xmm1 xmm2
|
||||
..B1.6: # Preds ..B1.5
|
||||
# Execution count [9.00e-01]
|
||||
lea 1(%rbx,%rbx), %esi #135.9
|
||||
# LOE rax r8 r9 r12 r14 r15 esi r13d xmm0 xmm1 xmm2
|
||||
..B1.7: # Preds ..B1.3 ..B1.6
|
||||
# Execution count [1.00e+00]
|
||||
lea -1(%rsi), %edx #134.5
|
||||
cmpl %r13d, %edx #134.5
|
||||
jae ..B1.9 # Prob 9% #134.5
|
||||
# LOE rax r8 r9 r12 r14 r15 esi r13d xmm0 xmm1 xmm2
|
||||
..B1.8: # Preds ..B1.7
|
||||
# Execution count [9.00e-01]
|
||||
movslq %esi, %rsi #134.5
|
||||
movq %rax, -8(%r9,%rsi,8) #135.9
|
||||
movq %rax, -8(%r14,%rsi,8) #136.9
|
||||
movq %rax, -8(%r8,%rsi,8) #137.9
|
||||
# LOE r8 r9 r12 r14 r15 r13d xmm0 xmm1 xmm2
|
||||
..B1.9: # Preds ..B1.7 ..B1.8
|
||||
# Execution count [5.00e-01]
|
||||
movl $.L_2__STRING.0, %edi #141.5
|
||||
movq %r8, 32(%rsp) #141.5[spill]
|
||||
movq %r9, 80(%rsp) #141.5[spill]
|
||||
vmovsd %xmm2, (%rsp) #141.5[spill]
|
||||
vmovsd %xmm1, 8(%rsp) #141.5[spill]
|
||||
vmovsd %xmm0, 16(%rsp) #141.5[spill]
|
||||
..___tag_value_computeForce.18:
|
||||
# likwid_markerStartRegion(const char *)
|
||||
call likwid_markerStartRegion #141.5
|
||||
..___tag_value_computeForce.19:
|
||||
# LOE r12 r14 r15 r13d
|
||||
..B1.10: # Preds ..B1.9
|
||||
# Execution count [9.00e-01]
|
||||
vmovsd 16(%rsp), %xmm0 #[spill]
|
||||
xorl %esi, %esi #143.15
|
||||
vmovsd (%rsp), %xmm2 #[spill]
|
||||
xorl %eax, %eax #143.5
|
||||
vmulsd %xmm2, %xmm2, %xmm13 #129.45
|
||||
xorl %edi, %edi #143.5
|
||||
vmovdqu32 .L_2il0floatpacket.0(%rip), %ymm16 #173.13
|
||||
vmulsd .L_2il0floatpacket.3(%rip), %xmm0, %xmm0 #197.45
|
||||
vmovdqu .L_2il0floatpacket.1(%rip), %ymm15 #173.13
|
||||
vmovups .L_2il0floatpacket.4(%rip), %zmm5 #197.58
|
||||
vmovsd 8(%rsp), %xmm1 #[spill]
|
||||
vbroadcastsd %xmm13, %zmm14 #129.25
|
||||
vbroadcastsd %xmm1, %zmm13 #130.21
|
||||
vbroadcastsd %xmm0, %zmm9 #197.45
|
||||
movslq %r13d, %r13 #143.5
|
||||
movq 24(%r15), %r10 #145.25
|
||||
movslq 16(%r15), %rdx #144.43
|
||||
movq 8(%r15), %rcx #144.19
|
||||
movq 32(%rsp), %r8 #[spill]
|
||||
movq 16(%r12), %rbx #146.25
|
||||
shlq $2, %rdx #126.5
|
||||
movq %r13, 64(%rsp) #143.5[spill]
|
||||
movq %r10, 72(%rsp) #143.5[spill]
|
||||
# LOE rax rdx rcx rbx rsi rdi r8 r14 ymm15 ymm16 zmm5 zmm9 zmm13 zmm14
|
||||
..B1.11: # Preds ..B1.41 ..B1.10
|
||||
# Execution count [5.00e+00]
|
||||
movq 72(%rsp), %r9 #145.25[spill]
|
||||
vxorpd %xmm24, %xmm24, %xmm24 #149.22
|
||||
vmovapd %xmm24, %xmm18 #150.22
|
||||
movl (%r9,%rax,4), %r10d #145.25
|
||||
vmovapd %xmm18, %xmm4 #151.22
|
||||
vmovsd (%rdi,%rbx), %xmm10 #146.25
|
||||
vmovsd 8(%rdi,%rbx), %xmm6 #147.25
|
||||
vmovsd 16(%rdi,%rbx), %xmm12 #148.25
|
||||
testl %r10d, %r10d #173.32
|
||||
jle ..B1.41 # Prob 50% #173.32
|
||||
# LOE rax rdx rcx rbx rsi rdi r8 r14 r10d xmm4 xmm6 xmm10 xmm12 xmm18 xmm24 ymm15 ymm16 zmm5 zmm9 zmm13 zmm14
|
||||
..B1.12: # Preds ..B1.11
|
||||
# Execution count [4.50e+00]
|
||||
vpxord %zmm8, %zmm8, %zmm8 #149.22
|
||||
vmovaps %zmm8, %zmm7 #150.22
|
||||
vmovaps %zmm7, %zmm11 #151.22
|
||||
cmpl $8, %r10d #173.13
|
||||
jl ..B1.48 # Prob 10% #173.13
|
||||
# LOE rax rdx rcx rbx rsi rdi r8 r14 r10d xmm6 xmm10 xmm12 ymm15 ymm16 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.13: # Preds ..B1.12
|
||||
# Execution count [4.50e+00]
|
||||
cmpl $1200, %r10d #173.13
|
||||
jl ..B1.47 # Prob 10% #173.13
|
||||
# LOE rax rdx rcx rbx rsi rdi r8 r14 r10d xmm6 xmm10 xmm12 ymm15 ymm16 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.14: # Preds ..B1.13
|
||||
# Execution count [4.50e+00]
|
||||
movq %rdx, %r15 #144.43
|
||||
imulq %rsi, %r15 #144.43
|
||||
addq %rcx, %r15 #126.5
|
||||
movq %r15, %r11 #173.13
|
||||
andq $63, %r11 #173.13
|
||||
testl $3, %r11d #173.13
|
||||
je ..B1.16 # Prob 50% #173.13
|
||||
# LOE rax rdx rcx rbx rsi rdi r8 r14 r15 r10d r11d xmm6 xmm10 xmm12 ymm15 ymm16 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.15: # Preds ..B1.14
|
||||
# Execution count [2.25e+00]
|
||||
xorl %r11d, %r11d #173.13
|
||||
jmp ..B1.18 # Prob 100% #173.13
|
||||
# LOE rax rdx rcx rbx rsi rdi r8 r14 r15 r10d r11d xmm6 xmm10 xmm12 ymm15 ymm16 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.16: # Preds ..B1.14
|
||||
# Execution count [2.25e+00]
|
||||
testl %r11d, %r11d #173.13
|
||||
je ..B1.18 # Prob 50% #173.13
|
||||
# LOE rax rdx rcx rbx rsi rdi r8 r14 r15 r10d r11d xmm6 xmm10 xmm12 ymm15 ymm16 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.17: # Preds ..B1.16
|
||||
# Execution count [2.50e+01]
|
||||
negl %r11d #173.13
|
||||
addl $64, %r11d #173.13
|
||||
shrl $2, %r11d #173.13
|
||||
cmpl %r11d, %r10d #173.13
|
||||
cmovl %r10d, %r11d #173.13
|
||||
# LOE rax rdx rcx rbx rsi rdi r8 r14 r15 r10d r11d xmm6 xmm10 xmm12 ymm15 ymm16 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.18: # Preds ..B1.15 ..B1.17 ..B1.16
|
||||
# Execution count [5.00e+00]
|
||||
movl %r10d, %r13d #173.13
|
||||
subl %r11d, %r13d #173.13
|
||||
andl $7, %r13d #173.13
|
||||
negl %r13d #173.13
|
||||
addl %r10d, %r13d #173.13
|
||||
cmpl $1, %r11d #173.13
|
||||
jb ..B1.26 # Prob 50% #173.13
|
||||
# LOE rax rdx rcx rbx rsi rdi r8 r14 r15 r10d r11d r13d xmm6 xmm10 xmm12 ymm15 ymm16 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.19: # Preds ..B1.18
|
||||
# Execution count [4.50e+00]
|
||||
vmovdqa %ymm15, %ymm4 #173.13
|
||||
xorl %r12d, %r12d #173.13
|
||||
vpbroadcastd %r11d, %ymm3 #173.13
|
||||
vbroadcastsd %xmm10, %zmm2 #146.23
|
||||
vbroadcastsd %xmm6, %zmm1 #147.23
|
||||
vbroadcastsd %xmm12, %zmm0 #148.23
|
||||
movslq %r11d, %r9 #173.13
|
||||
movq %r8, 32(%rsp) #173.13[spill]
|
||||
movq %r14, (%rsp) #173.13[spill]
|
||||
# LOE rax rdx rcx rbx rsi rdi r9 r12 r15 r10d r11d r13d xmm6 xmm10 xmm12 ymm3 ymm4 ymm15 ymm16 zmm0 zmm1 zmm2 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.20: # Preds ..B1.24 ..B1.19
|
||||
# Execution count [2.50e+01]
|
||||
vpcmpgtd %ymm4, %ymm3, %k3 #173.13
|
||||
vmovdqu32 (%r15,%r12,4), %ymm17{%k3}{z} #174.25
|
||||
kmovw %k3, %r14d #173.13
|
||||
vpaddd %ymm17, %ymm17, %ymm18 #175.40
|
||||
vpaddd %ymm18, %ymm17, %ymm17 #175.40
|
||||
# LOE rax rdx rcx rbx rsi rdi r9 r12 r15 r10d r11d r13d r14d xmm6 xmm10 xmm12 ymm3 ymm4 ymm15 ymm16 ymm17 zmm0 zmm1 zmm2 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14 k3
|
||||
..B1.23: # Preds ..B1.20
|
||||
# Execution count [1.25e+01]
|
||||
kmovw %k3, %k1 #175.40
|
||||
kmovw %k3, %k2 #175.40
|
||||
vpxord %zmm18, %zmm18, %zmm18 #175.40
|
||||
vpxord %zmm19, %zmm19, %zmm19 #175.40
|
||||
vpxord %zmm20, %zmm20, %zmm20 #175.40
|
||||
vgatherdpd 16(%rbx,%ymm17,8), %zmm18{%k1} #175.40
|
||||
vgatherdpd 8(%rbx,%ymm17,8), %zmm19{%k2} #175.40
|
||||
vgatherdpd (%rbx,%ymm17,8), %zmm20{%k3} #175.40
|
||||
# LOE rax rdx rcx rbx rsi rdi r9 r12 r15 r10d r11d r13d r14d xmm6 xmm10 xmm12 ymm3 ymm4 ymm15 ymm16 zmm0 zmm1 zmm2 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14 zmm18 zmm19 zmm20
|
||||
..B1.24: # Preds ..B1.23
|
||||
# Execution count [2.50e+01]
|
||||
addq $8, %r12 #173.13
|
||||
#vpaddd %ymm16, %ymm4, %ymm4 #173.13
|
||||
#vsubpd %zmm18, %zmm0, %zmm29 #177.40
|
||||
#vsubpd %zmm19, %zmm1, %zmm27 #176.40
|
||||
#vsubpd %zmm20, %zmm2, %zmm26 #175.40
|
||||
#vmulpd %zmm27, %zmm27, %zmm25 #178.53
|
||||
#vfmadd231pd %zmm26, %zmm26, %zmm25 #178.53
|
||||
#vfmadd231pd %zmm29, %zmm29, %zmm25 #178.67
|
||||
#vrcp14pd %zmm25, %zmm24 #195.42
|
||||
#vcmppd $1, %zmm14, %zmm25, %k2 #194.26
|
||||
#vfpclasspd $30, %zmm24, %k0 #195.42
|
||||
#kmovw %k2, %r8d #194.26
|
||||
#knotw %k0, %k1 #195.42
|
||||
#vmovaps %zmm25, %zmm17 #195.42
|
||||
#andl %r8d, %r14d #194.26
|
||||
#vfnmadd213pd .L_2il0floatpacket.9(%rip){1to8}, %zmm24, %zmm17 #195.42
|
||||
#kmovw %r14d, %k3 #198.21
|
||||
#vmulpd %zmm17, %zmm17, %zmm18 #195.42
|
||||
#vfmadd213pd %zmm24, %zmm17, %zmm24{%k1} #195.42
|
||||
#vfmadd213pd %zmm24, %zmm18, %zmm24{%k1} #195.42
|
||||
#vmulpd %zmm13, %zmm24, %zmm19 #196.42
|
||||
#vmulpd %zmm9, %zmm24, %zmm21 #197.58
|
||||
#vmulpd %zmm19, %zmm24, %zmm22 #196.48
|
||||
#vmulpd %zmm22, %zmm24, %zmm20 #196.54
|
||||
#vfmsub213pd %zmm5, %zmm22, %zmm24 #197.58
|
||||
#vmulpd %zmm21, %zmm20, %zmm23 #197.65
|
||||
#vmulpd %zmm24, %zmm23, %zmm28 #197.71
|
||||
#vfmadd231pd %zmm26, %zmm28, %zmm8{%k3} #198.21
|
||||
#vfmadd231pd %zmm27, %zmm28, %zmm7{%k3} #199.21
|
||||
#vfmadd231pd %zmm29, %zmm28, %zmm11{%k3} #200.21
|
||||
cmpq %r9, %r12 #173.13
|
||||
jb ..B1.20 # Prob 82% #173.13
|
||||
# LOE rax rdx rcx rbx rsi rdi r9 r12 r15 r10d r11d r13d xmm6 xmm10 xmm12 ymm3 ymm4 ymm15 ymm16 zmm0 zmm1 zmm2 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.25: # Preds ..B1.24
|
||||
# Execution count [4.50e+00]
|
||||
movq 32(%rsp), %r8 #[spill]
|
||||
movq (%rsp), %r14 #[spill]
|
||||
cmpl %r11d, %r10d #173.13
|
||||
je ..B1.40 # Prob 10% #173.13
|
||||
# LOE rax rdx rcx rbx rsi rdi r8 r14 r10d r11d r13d xmm6 xmm10 xmm12 ymm15 ymm16 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.26: # Preds ..B1.25 ..B1.18 ..B1.47
|
||||
# Execution count [2.50e+01]
|
||||
lea 8(%r11), %r9d #173.13
|
||||
cmpl %r9d, %r13d #173.13
|
||||
jl ..B1.34 # Prob 50% #173.13
|
||||
# LOE rax rdx rcx rbx rsi rdi r8 r14 r10d r11d r13d xmm6 xmm10 xmm12 ymm15 ymm16 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.27: # Preds ..B1.26
|
||||
# Execution count [4.50e+00]
|
||||
movq %rdx, %r12 #144.43
|
||||
imulq %rsi, %r12 #144.43
|
||||
vbroadcastsd %xmm10, %zmm1 #146.23
|
||||
vbroadcastsd %xmm6, %zmm0 #147.23
|
||||
vbroadcastsd %xmm12, %zmm2 #148.23
|
||||
movslq %r11d, %r9 #173.13
|
||||
addq %rcx, %r12 #126.5
|
||||
movq %rdi, 8(%rsp) #126.5[spill]
|
||||
movq %rdx, 16(%rsp) #126.5[spill]
|
||||
movq %rcx, 40(%rsp) #126.5[spill]
|
||||
movq %rax, 48(%rsp) #126.5[spill]
|
||||
movq %rsi, 56(%rsp) #126.5[spill]
|
||||
movq %r8, 32(%rsp) #126.5[spill]
|
||||
movq %r14, (%rsp) #126.5[spill]
|
||||
# LOE rbx r9 r12 r10d r11d r13d xmm6 xmm10 xmm12 ymm15 ymm16 zmm0 zmm1 zmm2 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.28: # Preds ..B1.32 ..B1.27
|
||||
# Execution count [2.50e+01]
|
||||
vmovdqu (%r12,%r9,4), %ymm3 #174.25
|
||||
vpaddd %ymm3, %ymm3, %ymm4 #175.40
|
||||
vpaddd %ymm4, %ymm3, %ymm3 #175.40
|
||||
movl (%r12,%r9,4), %r14d #174.25
|
||||
movl 4(%r12,%r9,4), %r8d #174.25
|
||||
movl 8(%r12,%r9,4), %edi #174.25
|
||||
movl 12(%r12,%r9,4), %esi #174.25
|
||||
lea (%r14,%r14,2), %r14d #175.40
|
||||
movl 16(%r12,%r9,4), %ecx #174.25
|
||||
lea (%r8,%r8,2), %r8d #175.40
|
||||
movl 20(%r12,%r9,4), %edx #174.25
|
||||
lea (%rdi,%rdi,2), %edi #175.40
|
||||
movl 24(%r12,%r9,4), %eax #174.25
|
||||
lea (%rsi,%rsi,2), %esi #175.40
|
||||
movl 28(%r12,%r9,4), %r15d #174.25
|
||||
lea (%rcx,%rcx,2), %ecx #175.40
|
||||
lea (%rdx,%rdx,2), %edx #175.40
|
||||
lea (%rax,%rax,2), %eax #175.40
|
||||
lea (%r15,%r15,2), %r15d #175.40
|
||||
# LOE rbx r9 r12 eax edx ecx esi edi r8d r10d r11d r13d r14d r15d xmm6 xmm10 xmm12 ymm3 ymm15 ymm16 zmm0 zmm1 zmm2 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.31: # Preds ..B1.28
|
||||
# Execution count [1.25e+01]
|
||||
vpcmpeqb %xmm0, %xmm0, %k1 #175.40
|
||||
vpcmpeqb %xmm0, %xmm0, %k2 #175.40
|
||||
vpcmpeqb %xmm0, %xmm0, %k3 #175.40
|
||||
vpxord %zmm4, %zmm4, %zmm4 #175.40
|
||||
vpxord %zmm17, %zmm17, %zmm17 #175.40
|
||||
vpxord %zmm18, %zmm18, %zmm18 #175.40
|
||||
vgatherdpd 16(%rbx,%ymm3,8), %zmm4{%k1} #175.40
|
||||
vgatherdpd 8(%rbx,%ymm3,8), %zmm17{%k2} #175.40
|
||||
vgatherdpd (%rbx,%ymm3,8), %zmm18{%k3} #175.40
|
||||
# LOE rbx r9 r12 r10d r11d r13d xmm6 xmm10 xmm12 ymm15 ymm16 zmm0 zmm1 zmm2 zmm4 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14 zmm17 zmm18
|
||||
..B1.32: # Preds ..B1.31
|
||||
# Execution count [2.50e+01]
|
||||
addl $8, %r11d #173.13
|
||||
addq $8, %r9 #173.13
|
||||
#vsubpd %zmm4, %zmm2, %zmm26 #177.40
|
||||
#vsubpd %zmm17, %zmm0, %zmm24 #176.40
|
||||
#vsubpd %zmm18, %zmm1, %zmm23 #175.40
|
||||
#vmulpd %zmm24, %zmm24, %zmm3 #178.53
|
||||
#vfmadd231pd %zmm23, %zmm23, %zmm3 #178.53
|
||||
#vfmadd231pd %zmm26, %zmm26, %zmm3 #178.67
|
||||
#vrcp14pd %zmm3, %zmm22 #195.42
|
||||
#vcmppd $1, %zmm14, %zmm3, %k2 #194.26
|
||||
#vfpclasspd $30, %zmm22, %k0 #195.42
|
||||
#vfnmadd213pd .L_2il0floatpacket.9(%rip){1to8}, %zmm22, %zmm3 #195.42
|
||||
#knotw %k0, %k1 #195.42
|
||||
#vmulpd %zmm3, %zmm3, %zmm4 #195.42
|
||||
#vfmadd213pd %zmm22, %zmm3, %zmm22{%k1} #195.42
|
||||
#vfmadd213pd %zmm22, %zmm4, %zmm22{%k1} #195.42
|
||||
#vmulpd %zmm13, %zmm22, %zmm17 #196.42
|
||||
#vmulpd %zmm9, %zmm22, %zmm19 #197.58
|
||||
#vmulpd %zmm17, %zmm22, %zmm20 #196.48
|
||||
#vmulpd %zmm20, %zmm22, %zmm18 #196.54
|
||||
#vfmsub213pd %zmm5, %zmm20, %zmm22 #197.58
|
||||
#vmulpd %zmm19, %zmm18, %zmm21 #197.65
|
||||
#vmulpd %zmm22, %zmm21, %zmm25 #197.71
|
||||
#vfmadd231pd %zmm23, %zmm25, %zmm8{%k2} #198.21
|
||||
#vfmadd231pd %zmm24, %zmm25, %zmm7{%k2} #199.21
|
||||
#vfmadd231pd %zmm26, %zmm25, %zmm11{%k2} #200.21
|
||||
cmpl %r13d, %r11d #173.13
|
||||
jb ..B1.28 # Prob 82% #173.13
|
||||
# LOE rbx r9 r12 r10d r11d r13d xmm6 xmm10 xmm12 ymm15 ymm16 zmm0 zmm1 zmm2 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.33: # Preds ..B1.32
|
||||
# Execution count [4.50e+00]
|
||||
movq 8(%rsp), %rdi #[spill]
|
||||
movq 16(%rsp), %rdx #[spill]
|
||||
movq 40(%rsp), %rcx #[spill]
|
||||
movq 48(%rsp), %rax #[spill]
|
||||
movq 56(%rsp), %rsi #[spill]
|
||||
movq 32(%rsp), %r8 #[spill]
|
||||
movq (%rsp), %r14 #[spill]
|
||||
# LOE rax rdx rcx rbx rsi rdi r8 r14 r10d r13d xmm6 xmm10 xmm12 ymm15 ymm16 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.34: # Preds ..B1.33 ..B1.26 ..B1.48
|
||||
# Execution count [5.00e+00]
|
||||
lea 1(%r13), %r9d #173.13
|
||||
cmpl %r10d, %r9d #173.13
|
||||
ja ..B1.40 # Prob 50% #173.13
|
||||
# LOE rax rdx rcx rbx rsi rdi r8 r14 r10d r13d xmm6 xmm10 xmm12 ymm15 ymm16 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.35: # Preds ..B1.34
|
||||
# Execution count [2.50e+01]
|
||||
imulq %rdx, %rsi #144.43
|
||||
vbroadcastsd %xmm10, %zmm4 #146.23
|
||||
subl %r13d, %r10d #173.13
|
||||
addq %rcx, %rsi #126.5
|
||||
vpbroadcastd %r10d, %ymm0 #173.13
|
||||
vpcmpgtd %ymm15, %ymm0, %k3 #173.13
|
||||
movslq %r13d, %r13 #173.13
|
||||
kmovw %k3, %r9d #173.13
|
||||
vmovdqu32 (%rsi,%r13,4), %ymm1{%k3}{z} #174.25
|
||||
vpaddd %ymm1, %ymm1, %ymm2 #175.40
|
||||
vpaddd %ymm2, %ymm1, %ymm0 #175.40
|
||||
# LOE rax rdx rcx rbx rdi r8 r14 r9d xmm6 xmm12 ymm0 ymm15 ymm16 zmm4 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14 k3
|
||||
..B1.38: # Preds ..B1.35
|
||||
# Execution count [1.25e+01]
|
||||
kmovw %k3, %k1 #175.40
|
||||
kmovw %k3, %k2 #175.40
|
||||
vpxord %zmm1, %zmm1, %zmm1 #175.40
|
||||
vpxord %zmm2, %zmm2, %zmm2 #175.40
|
||||
vpxord %zmm3, %zmm3, %zmm3 #175.40
|
||||
vgatherdpd 16(%rbx,%ymm0,8), %zmm1{%k1} #175.40
|
||||
vgatherdpd 8(%rbx,%ymm0,8), %zmm2{%k2} #175.40
|
||||
vgatherdpd (%rbx,%ymm0,8), %zmm3{%k3} #175.40
|
||||
# LOE rax rdx rcx rbx rdi r8 r14 r9d xmm6 xmm12 ymm15 ymm16 zmm1 zmm2 zmm3 zmm4 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.39: # Preds ..B1.38
|
||||
# Execution count [2.50e+01]
|
||||
#vbroadcastsd %xmm6, %zmm6 #147.23
|
||||
#vbroadcastsd %xmm12, %zmm12 #148.23
|
||||
#vsubpd %zmm1, %zmm12, %zmm23 #177.40
|
||||
#vsubpd %zmm2, %zmm6, %zmm21 #176.40
|
||||
#vsubpd %zmm3, %zmm4, %zmm20 #175.40
|
||||
#vmulpd %zmm21, %zmm21, %zmm19 #178.53
|
||||
#vfmadd231pd %zmm20, %zmm20, %zmm19 #178.53
|
||||
#vfmadd231pd %zmm23, %zmm23, %zmm19 #178.67
|
||||
#vrcp14pd %zmm19, %zmm18 #195.42
|
||||
#vcmppd $1, %zmm14, %zmm19, %k2 #194.26
|
||||
#vfpclasspd $30, %zmm18, %k0 #195.42
|
||||
#kmovw %k2, %esi #194.26
|
||||
#knotw %k0, %k1 #195.42
|
||||
#vmovaps %zmm19, %zmm0 #195.42
|
||||
#andl %esi, %r9d #194.26
|
||||
#vfnmadd213pd .L_2il0floatpacket.9(%rip){1to8}, %zmm18, %zmm0 #195.42
|
||||
#kmovw %r9d, %k3 #198.21
|
||||
#vmulpd %zmm0, %zmm0, %zmm1 #195.42
|
||||
#vfmadd213pd %zmm18, %zmm0, %zmm18{%k1} #195.42
|
||||
#vfmadd213pd %zmm18, %zmm1, %zmm18{%k1} #195.42
|
||||
#vmulpd %zmm13, %zmm18, %zmm2 #196.42
|
||||
#vmulpd %zmm9, %zmm18, %zmm4 #197.58
|
||||
#vmulpd %zmm2, %zmm18, %zmm10 #196.48
|
||||
#vmulpd %zmm10, %zmm18, %zmm3 #196.54
|
||||
#vfmsub213pd %zmm5, %zmm10, %zmm18 #197.58
|
||||
#vmulpd %zmm4, %zmm3, %zmm17 #197.65
|
||||
#vmulpd %zmm18, %zmm17, %zmm22 #197.71
|
||||
#vfmadd231pd %zmm20, %zmm22, %zmm8{%k3} #198.21
|
||||
#vfmadd231pd %zmm21, %zmm22, %zmm7{%k3} #199.21
|
||||
#vfmadd231pd %zmm23, %zmm22, %zmm11{%k3} #200.21
|
||||
# LOE rax rdx rcx rbx rdi r8 r14 ymm15 ymm16 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.40: # Preds ..B1.25 ..B1.39 ..B1.34
|
||||
# Execution count [4.50e+00]
|
||||
vmovups .L_2il0floatpacket.10(%rip), %zmm19 #151.22
|
||||
vpermd %zmm11, %zmm19, %zmm0 #151.22
|
||||
vpermd %zmm7, %zmm19, %zmm6 #150.22
|
||||
vpermd %zmm8, %zmm19, %zmm20 #149.22
|
||||
vaddpd %zmm11, %zmm0, %zmm11 #151.22
|
||||
vaddpd %zmm7, %zmm6, %zmm7 #150.22
|
||||
vaddpd %zmm8, %zmm20, %zmm8 #149.22
|
||||
vpermpd $78, %zmm11, %zmm1 #151.22
|
||||
vpermpd $78, %zmm7, %zmm10 #150.22
|
||||
vpermpd $78, %zmm8, %zmm21 #149.22
|
||||
vaddpd %zmm1, %zmm11, %zmm2 #151.22
|
||||
vaddpd %zmm10, %zmm7, %zmm12 #150.22
|
||||
vaddpd %zmm21, %zmm8, %zmm22 #149.22
|
||||
vpermpd $177, %zmm2, %zmm3 #151.22
|
||||
vpermpd $177, %zmm12, %zmm17 #150.22
|
||||
vpermpd $177, %zmm22, %zmm23 #149.22
|
||||
vaddpd %zmm3, %zmm2, %zmm4 #151.22
|
||||
vaddpd %zmm17, %zmm12, %zmm18 #150.22
|
||||
vaddpd %zmm23, %zmm22, %zmm24 #149.22
|
||||
# LOE rax rdx rcx rbx rdi r8 r14 xmm4 xmm18 xmm24 ymm15 ymm16 zmm5 zmm9 zmm13 zmm14
|
||||
..B1.41: # Preds ..B1.40 ..B1.11
|
||||
# Execution count [5.00e+00]
|
||||
movq 80(%rsp), %rsi #208.9[spill]
|
||||
addq $24, %rdi #143.5
|
||||
vaddsd (%rsi,%rax,8), %xmm24, %xmm0 #208.9
|
||||
vmovsd %xmm0, (%rsi,%rax,8) #208.9
|
||||
movslq %eax, %rsi #143.32
|
||||
vaddsd (%r14,%rax,8), %xmm18, %xmm1 #209.9
|
||||
vmovsd %xmm1, (%r14,%rax,8) #209.9
|
||||
incq %rsi #143.32
|
||||
vaddsd (%r8,%rax,8), %xmm4, %xmm2 #210.9
|
||||
vmovsd %xmm2, (%r8,%rax,8) #210.9
|
||||
incq %rax #143.5
|
||||
cmpq 64(%rsp), %rax #143.5[spill]
|
||||
jb ..B1.11 # Prob 82% #143.5
|
||||
jmp ..B1.44 # Prob 100% #143.5
|
||||
# LOE rax rdx rcx rbx rsi rdi r8 r14 ymm15 ymm16 zmm5 zmm9 zmm13 zmm14
|
||||
..B1.43: # Preds ..B1.2
|
||||
# Execution count [5.00e-01]
|
||||
movl $.L_2__STRING.0, %edi #141.5
|
||||
..___tag_value_computeForce.48:
|
||||
# likwid_markerStartRegion(const char *)
|
||||
call likwid_markerStartRegion #141.5
|
||||
..___tag_value_computeForce.49:
|
||||
# LOE
|
||||
..B1.44: # Preds ..B1.41 ..B1.43
|
||||
# Execution count [1.00e+00]
|
||||
movl $.L_2__STRING.0, %edi #219.5
|
||||
vzeroupper #219.5
|
||||
..___tag_value_computeForce.50:
|
||||
# likwid_markerStopRegion(const char *)
|
||||
call likwid_markerStopRegion #219.5
|
||||
..___tag_value_computeForce.51:
|
||||
# LOE
|
||||
..B1.45: # Preds ..B1.44
|
||||
# Execution count [1.00e+00]
|
||||
xorl %eax, %eax #221.16
|
||||
..___tag_value_computeForce.52:
|
||||
# getTimeStamp()
|
||||
call getTimeStamp #221.16
|
||||
..___tag_value_computeForce.53:
|
||||
# LOE xmm0
|
||||
..B1.46: # Preds ..B1.45
|
||||
# Execution count [1.00e+00]
|
||||
vsubsd 24(%rsp), %xmm0, %xmm0 #224.14[spill]
|
||||
addq $88, %rsp #224.14
|
||||
.cfi_restore 3
|
||||
popq %rbx #224.14
|
||||
.cfi_restore 15
|
||||
popq %r15 #224.14
|
||||
.cfi_restore 14
|
||||
popq %r14 #224.14
|
||||
.cfi_restore 13
|
||||
popq %r13 #224.14
|
||||
.cfi_restore 12
|
||||
popq %r12 #224.14
|
||||
movq %rbp, %rsp #224.14
|
||||
popq %rbp #224.14
|
||||
.cfi_def_cfa 7, 8
|
||||
.cfi_restore 6
|
||||
ret #224.14
|
||||
.cfi_def_cfa 6, 16
|
||||
.cfi_escape 0x10, 0x03, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xd8, 0xff, 0xff, 0xff, 0x22
|
||||
.cfi_offset 6, -16
|
||||
.cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xf8, 0xff, 0xff, 0xff, 0x22
|
||||
.cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xf0, 0xff, 0xff, 0xff, 0x22
|
||||
.cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xe8, 0xff, 0xff, 0xff, 0x22
|
||||
.cfi_escape 0x10, 0x0f, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x22
|
||||
# LOE
|
||||
..B1.47: # Preds ..B1.13
|
||||
# Execution count [4.50e-01]: Infreq
|
||||
movl %r10d, %r13d #173.13
|
||||
xorl %r11d, %r11d #173.13
|
||||
andl $-8, %r13d #173.13
|
||||
jmp ..B1.26 # Prob 100% #173.13
|
||||
# LOE rax rdx rcx rbx rsi rdi r8 r14 r10d r11d r13d xmm6 xmm10 xmm12 ymm15 ymm16 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.48: # Preds ..B1.12
|
||||
# Execution count [4.50e-01]: Infreq
|
||||
xorl %r13d, %r13d #173.13
|
||||
jmp ..B1.34 # Prob 100% #173.13
|
||||
.align 16,0x90
|
||||
# LOE rax rdx rcx rbx rsi rdi r8 r14 r10d r13d xmm6 xmm10 xmm12 ymm15 ymm16 zmm5 zmm7 zmm8 zmm9 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
|
||||
.section .rodata.str1.4, "aMS",@progbits,1
|
||||
.align 4
|
||||
.align 4
|
||||
.L_2__STRING.0:
|
||||
.long 1668444006
|
||||
.word 101
|
||||
.type .L_2__STRING.0,@object
|
||||
.size .L_2__STRING.0,6
|
||||
.data
|
||||
.section .note.GNU-stack, ""
|
||||
# End
|
||||
585
asm/unused/force-mem-only.s
Normal file
585
asm/unused/force-mem-only.s
Normal file
@ -0,0 +1,585 @@
|
||||
# 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 -DNEIGHBORS_LOOP_RUNS=1 -DVECTOR_WIDTH=8 -DALIGNMENT=6";
|
||||
# mark_description "4 -restrict -Ofast -xCORE-AVX512 -qopt-zmm-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:
|
||||
#103.87
|
||||
pushq %rbp #103.87
|
||||
.cfi_def_cfa_offset 16
|
||||
movq %rsp, %rbp #103.87
|
||||
.cfi_def_cfa 6, 16
|
||||
.cfi_offset 6, -16
|
||||
andq $-64, %rsp #103.87
|
||||
pushq %r12 #103.87
|
||||
pushq %r13 #103.87
|
||||
pushq %r14 #103.87
|
||||
subq $104, %rsp #103.87
|
||||
xorl %eax, %eax #106.16
|
||||
.cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xf8, 0xff, 0xff, 0xff, 0x22
|
||||
.cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xf0, 0xff, 0xff, 0xff, 0x22
|
||||
.cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xe8, 0xff, 0xff, 0xff, 0x22
|
||||
movq %rdx, %r14 #103.87
|
||||
movq %rsi, %r13 #103.87
|
||||
movq %rdi, %r12 #103.87
|
||||
..___tag_value_computeForce.9:
|
||||
# getTimeStamp()
|
||||
call getTimeStamp #106.16
|
||||
..___tag_value_computeForce.10:
|
||||
# LOE rbx r12 r13 r14 r15 xmm0
|
||||
..B1.48: # Preds ..B1.1
|
||||
# Execution count [1.00e+00]
|
||||
vmovsd %xmm0, 16(%rsp) #106.16[spill]
|
||||
# LOE rbx r12 r13 r14 r15
|
||||
..B1.2: # Preds ..B1.48
|
||||
# Execution count [1.00e+00]
|
||||
movl 4(%r13), %ecx #107.18
|
||||
movq 64(%r13), %r11 #109.20
|
||||
movq 72(%r13), %r10 #109.45
|
||||
movq 80(%r13), %r9 #109.70
|
||||
vmovsd 72(%r12), %xmm2 #111.27
|
||||
vmovsd 8(%r12), %xmm1 #112.23
|
||||
vmovsd (%r12), %xmm0 #113.24
|
||||
testl %ecx, %ecx #116.24
|
||||
jle ..B1.42 # Prob 50% #116.24
|
||||
# LOE rbx r9 r10 r11 r13 r14 r15 ecx xmm0 xmm1 xmm2
|
||||
..B1.3: # Preds ..B1.2
|
||||
# Execution count [1.00e+00]
|
||||
xorl %edi, %edi #116.5
|
||||
movl %ecx, %edx #116.5
|
||||
xorl %esi, %esi #116.5
|
||||
movl $1, %r8d #116.5
|
||||
xorl %eax, %eax #117.17
|
||||
shrl $1, %edx #116.5
|
||||
je ..B1.7 # Prob 9% #116.5
|
||||
# LOE rax rdx rbx rsi rdi r9 r10 r11 r13 r14 r15 ecx r8d xmm0 xmm1 xmm2
|
||||
..B1.5: # Preds ..B1.3 ..B1.5
|
||||
# Execution count [2.50e+00]
|
||||
movq %rax, (%rsi,%r11) #117.9
|
||||
incq %rdi #116.5
|
||||
movq %rax, (%rsi,%r10) #118.9
|
||||
movq %rax, (%rsi,%r9) #119.9
|
||||
movq %rax, 8(%rsi,%r11) #117.9
|
||||
movq %rax, 8(%rsi,%r10) #118.9
|
||||
movq %rax, 8(%rsi,%r9) #119.9
|
||||
addq $16, %rsi #116.5
|
||||
cmpq %rdx, %rdi #116.5
|
||||
jb ..B1.5 # Prob 63% #116.5
|
||||
# LOE rax rdx rbx rsi rdi r9 r10 r11 r13 r14 r15 ecx xmm0 xmm1 xmm2
|
||||
..B1.6: # Preds ..B1.5
|
||||
# Execution count [9.00e-01]
|
||||
lea 1(%rdi,%rdi), %r8d #117.9
|
||||
# LOE rax rbx r9 r10 r11 r13 r14 r15 ecx r8d xmm0 xmm1 xmm2
|
||||
..B1.7: # Preds ..B1.3 ..B1.6
|
||||
# Execution count [1.00e+00]
|
||||
lea -1(%r8), %edx #116.5
|
||||
cmpl %ecx, %edx #116.5
|
||||
jae ..B1.9 # Prob 9% #116.5
|
||||
# LOE rax rbx r9 r10 r11 r13 r14 r15 ecx r8d xmm0 xmm1 xmm2
|
||||
..B1.8: # Preds ..B1.7
|
||||
# Execution count [9.00e-01]
|
||||
movslq %r8d, %r8 #116.5
|
||||
movq %rax, -8(%r11,%r8,8) #117.9
|
||||
movq %rax, -8(%r10,%r8,8) #118.9
|
||||
movq %rax, -8(%r9,%r8,8) #119.9
|
||||
# LOE rbx r9 r10 r11 r13 r14 r15 ecx xmm0 xmm1 xmm2
|
||||
..B1.9: # Preds ..B1.7 ..B1.8
|
||||
# Execution count [9.00e-01]
|
||||
vmulsd %xmm2, %xmm2, %xmm13 #111.45
|
||||
xorl %edi, %edi #124.15
|
||||
vmovdqu32 .L_2il0floatpacket.0(%rip), %ymm16 #153.13
|
||||
vmulsd .L_2il0floatpacket.3(%rip), %xmm0, %xmm0 #177.45
|
||||
vmovdqu .L_2il0floatpacket.1(%rip), %ymm15 #153.13
|
||||
vmovups .L_2il0floatpacket.4(%rip), %zmm5 #177.58
|
||||
vbroadcastsd %xmm13, %zmm14 #111.25
|
||||
vbroadcastsd %xmm1, %zmm13 #112.21
|
||||
vbroadcastsd %xmm0, %zmm9 #177.45
|
||||
movq 16(%r13), %rdx #127.25
|
||||
xorl %r8d, %r8d #124.5
|
||||
movslq %ecx, %r12 #124.5
|
||||
xorl %eax, %eax #124.5
|
||||
movq 24(%r14), %r13 #126.25
|
||||
movslq 16(%r14), %rcx #125.43
|
||||
movq 8(%r14), %rsi #125.19
|
||||
shlq $2, %rcx #108.5
|
||||
movq %r12, 80(%rsp) #124.5[spill]
|
||||
movq %r13, 88(%rsp) #124.5[spill]
|
||||
movq %r11, 96(%rsp) #124.5[spill]
|
||||
movq %r15, 8(%rsp) #124.5[spill]
|
||||
movq %rbx, (%rsp) #124.5[spill]
|
||||
.cfi_escape 0x10, 0x03, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x80, 0xff, 0xff, 0xff, 0x22
|
||||
.cfi_escape 0x10, 0x0f, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x88, 0xff, 0xff, 0xff, 0x22
|
||||
# LOE rax rdx rcx rsi rdi r8 r9 r10 ymm15 ymm16 zmm5 zmm9 zmm13 zmm14
|
||||
..B1.10: # Preds ..B1.40 ..B1.9
|
||||
# Execution count [5.00e+00]
|
||||
movq 88(%rsp), %rbx #126.25[spill]
|
||||
vxorpd %xmm24, %xmm24, %xmm24 #130.22
|
||||
vmovapd %xmm24, %xmm18 #131.22
|
||||
movl (%rbx,%r8,4), %r11d #126.25
|
||||
vmovapd %xmm18, %xmm4 #132.22
|
||||
vmovsd (%rax,%rdx), %xmm10 #127.25
|
||||
vmovsd 8(%rax,%rdx), %xmm6 #128.25
|
||||
vmovsd 16(%rax,%rdx), %xmm12 #129.25
|
||||
testl %r11d, %r11d #153.32
|
||||
jle ..B1.40 # Prob 50% #153.32
|
||||
# LOE rax rdx rcx rsi rdi r8 r9 r10 r11d xmm4 xmm6 xmm10 xmm12 xmm18 xmm24 ymm15 ymm16 zmm5 zmm9 zmm13 zmm14
|
||||
..B1.11: # Preds ..B1.10
|
||||
# Execution count [4.50e+00]
|
||||
vpxord %zmm8, %zmm8, %zmm8 #130.22
|
||||
vmovaps %zmm8, %zmm7 #131.22
|
||||
vmovaps %zmm7, %zmm11 #132.22
|
||||
cmpl $8, %r11d #153.13
|
||||
jl ..B1.45 # Prob 10% #153.13
|
||||
# LOE rax rdx rcx rsi rdi r8 r9 r10 r11d xmm6 xmm10 xmm12 ymm15 ymm16 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.12: # Preds ..B1.11
|
||||
# Execution count [4.50e+00]
|
||||
cmpl $1200, %r11d #153.13
|
||||
jl ..B1.44 # Prob 10% #153.13
|
||||
# LOE rax rdx rcx rsi rdi r8 r9 r10 r11d xmm6 xmm10 xmm12 ymm15 ymm16 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.13: # Preds ..B1.12
|
||||
# Execution count [4.50e+00]
|
||||
movq %rcx, %r15 #125.43
|
||||
imulq %rdi, %r15 #125.43
|
||||
addq %rsi, %r15 #108.5
|
||||
movq %r15, %r12 #153.13
|
||||
andq $63, %r12 #153.13
|
||||
testl $3, %r12d #153.13
|
||||
je ..B1.15 # Prob 50% #153.13
|
||||
# LOE rax rdx rcx rsi rdi r8 r9 r10 r15 r11d r12d xmm6 xmm10 xmm12 ymm15 ymm16 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.14: # Preds ..B1.13
|
||||
# Execution count [2.25e+00]
|
||||
xorl %r12d, %r12d #153.13
|
||||
jmp ..B1.17 # Prob 100% #153.13
|
||||
# LOE rax rdx rcx rsi rdi r8 r9 r10 r15 r11d r12d xmm6 xmm10 xmm12 ymm15 ymm16 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.15: # Preds ..B1.13
|
||||
# Execution count [2.25e+00]
|
||||
testl %r12d, %r12d #153.13
|
||||
je ..B1.17 # Prob 50% #153.13
|
||||
# LOE rax rdx rcx rsi rdi r8 r9 r10 r15 r11d r12d xmm6 xmm10 xmm12 ymm15 ymm16 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.16: # Preds ..B1.15
|
||||
# Execution count [2.50e+01]
|
||||
negl %r12d #153.13
|
||||
addl $64, %r12d #153.13
|
||||
shrl $2, %r12d #153.13
|
||||
cmpl %r12d, %r11d #153.13
|
||||
cmovl %r11d, %r12d #153.13
|
||||
# LOE rax rdx rcx rsi rdi r8 r9 r10 r15 r11d r12d xmm6 xmm10 xmm12 ymm15 ymm16 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.17: # Preds ..B1.14 ..B1.16 ..B1.15
|
||||
# Execution count [5.00e+00]
|
||||
movl %r11d, %r14d #153.13
|
||||
subl %r12d, %r14d #153.13
|
||||
andl $7, %r14d #153.13
|
||||
negl %r14d #153.13
|
||||
addl %r11d, %r14d #153.13
|
||||
cmpl $1, %r12d #153.13
|
||||
jb ..B1.25 # Prob 50% #153.13
|
||||
# LOE rax rdx rcx rsi rdi r8 r9 r10 r15 r11d r12d r14d xmm6 xmm10 xmm12 ymm15 ymm16 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.18: # Preds ..B1.17
|
||||
# Execution count [4.50e+00]
|
||||
vmovdqa %ymm15, %ymm4 #153.13
|
||||
xorl %r13d, %r13d #153.13
|
||||
vpbroadcastd %r12d, %ymm3 #153.13
|
||||
vbroadcastsd %xmm10, %zmm2 #127.23
|
||||
vbroadcastsd %xmm6, %zmm1 #128.23
|
||||
vbroadcastsd %xmm12, %zmm0 #129.23
|
||||
movslq %r12d, %rbx #153.13
|
||||
movq %r9, 24(%rsp) #153.13[spill]
|
||||
movq %r10, 32(%rsp) #153.13[spill]
|
||||
# LOE rax rdx rcx rbx rsi rdi r8 r13 r15 r11d r12d r14d xmm6 xmm10 xmm12 ymm3 ymm4 ymm15 ymm16 zmm0 zmm1 zmm2 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.19: # Preds ..B1.23 ..B1.18
|
||||
# Execution count [2.50e+01]
|
||||
vpcmpgtd %ymm4, %ymm3, %k3 #153.13
|
||||
vmovdqu32 (%r15,%r13,4), %ymm17{%k3}{z} #154.25
|
||||
kmovw %k3, %r10d #153.13
|
||||
vpaddd %ymm17, %ymm17, %ymm18 #155.40
|
||||
vpaddd %ymm18, %ymm17, %ymm17 #155.40
|
||||
# LOE rax rdx rcx rbx rsi rdi r8 r13 r15 r10d r11d r12d r14d xmm6 xmm10 xmm12 ymm3 ymm4 ymm15 ymm16 ymm17 zmm0 zmm1 zmm2 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14 k3
|
||||
..B1.22: # Preds ..B1.19
|
||||
# Execution count [1.25e+01]
|
||||
kmovw %k3, %k1 #155.40
|
||||
kmovw %k3, %k2 #155.40
|
||||
vpxord %zmm18, %zmm18, %zmm18 #155.40
|
||||
vpxord %zmm19, %zmm19, %zmm19 #155.40
|
||||
vpxord %zmm20, %zmm20, %zmm20 #155.40
|
||||
vgatherdpd 16(%rdx,%ymm17,8), %zmm18{%k1} #155.40
|
||||
vgatherdpd 8(%rdx,%ymm17,8), %zmm19{%k2} #155.40
|
||||
vgatherdpd (%rdx,%ymm17,8), %zmm20{%k3} #155.40
|
||||
# LOE rax rdx rcx rbx rsi rdi r8 r13 r15 r10d r11d r12d r14d xmm6 xmm10 xmm12 ymm3 ymm4 ymm15 ymm16 zmm0 zmm1 zmm2 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14 zmm18 zmm19 zmm20
|
||||
..B1.23: # Preds ..B1.22
|
||||
# Execution count [2.50e+01]
|
||||
addq $8, %r13 #153.13
|
||||
#vpaddd %ymm16, %ymm4, %ymm4 #153.13
|
||||
#vsubpd %zmm18, %zmm0, %zmm29 #157.40
|
||||
#vsubpd %zmm19, %zmm1, %zmm27 #156.40
|
||||
#vsubpd %zmm20, %zmm2, %zmm26 #155.40
|
||||
#vmulpd %zmm27, %zmm27, %zmm25 #158.53
|
||||
#vfmadd231pd %zmm26, %zmm26, %zmm25 #158.53
|
||||
#vfmadd231pd %zmm29, %zmm29, %zmm25 #158.67
|
||||
#vrcp14pd %zmm25, %zmm24 #175.42
|
||||
#vcmppd $1, %zmm14, %zmm25, %k2 #174.26
|
||||
#vfpclasspd $30, %zmm24, %k0 #175.42
|
||||
#kmovw %k2, %r9d #174.26
|
||||
#knotw %k0, %k1 #175.42
|
||||
#vmovaps %zmm25, %zmm17 #175.42
|
||||
#andl %r9d, %r10d #174.26
|
||||
#vfnmadd213pd .L_2il0floatpacket.9(%rip){1to8}, %zmm24, %zmm17 #175.42
|
||||
#kmovw %r10d, %k3 #178.21
|
||||
#vmulpd %zmm17, %zmm17, %zmm18 #175.42
|
||||
#vfmadd213pd %zmm24, %zmm17, %zmm24{%k1} #175.42
|
||||
#vfmadd213pd %zmm24, %zmm18, %zmm24{%k1} #175.42
|
||||
#vmulpd %zmm13, %zmm24, %zmm19 #176.42
|
||||
#vmulpd %zmm9, %zmm24, %zmm21 #177.58
|
||||
#vmulpd %zmm19, %zmm24, %zmm22 #176.48
|
||||
#vmulpd %zmm22, %zmm24, %zmm20 #176.54
|
||||
#vfmsub213pd %zmm5, %zmm22, %zmm24 #177.58
|
||||
#vmulpd %zmm21, %zmm20, %zmm23 #177.65
|
||||
#vmulpd %zmm24, %zmm23, %zmm28 #177.71
|
||||
#vfmadd231pd %zmm26, %zmm28, %zmm8{%k3} #178.21
|
||||
#vfmadd231pd %zmm27, %zmm28, %zmm7{%k3} #179.21
|
||||
#vfmadd231pd %zmm29, %zmm28, %zmm11{%k3} #180.21
|
||||
cmpq %rbx, %r13 #153.13
|
||||
jb ..B1.19 # Prob 82% #153.13
|
||||
# LOE rax rdx rcx rbx rsi rdi r8 r13 r15 r11d r12d r14d xmm6 xmm10 xmm12 ymm3 ymm4 ymm15 ymm16 zmm0 zmm1 zmm2 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.24: # Preds ..B1.23
|
||||
# Execution count [4.50e+00]
|
||||
movq 24(%rsp), %r9 #[spill]
|
||||
movq 32(%rsp), %r10 #[spill]
|
||||
cmpl %r12d, %r11d #153.13
|
||||
je ..B1.39 # Prob 10% #153.13
|
||||
# LOE rax rdx rcx rsi rdi r8 r9 r10 r11d r12d r14d xmm6 xmm10 xmm12 ymm15 ymm16 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.25: # Preds ..B1.24 ..B1.17 ..B1.44
|
||||
# Execution count [2.50e+01]
|
||||
lea 8(%r12), %ebx #153.13
|
||||
cmpl %ebx, %r14d #153.13
|
||||
jl ..B1.33 # Prob 50% #153.13
|
||||
# LOE rax rdx rcx rsi rdi r8 r9 r10 r11d r12d r14d xmm6 xmm10 xmm12 ymm15 ymm16 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.26: # Preds ..B1.25
|
||||
# Execution count [4.50e+00]
|
||||
movq %rcx, %r13 #125.43
|
||||
imulq %rdi, %r13 #125.43
|
||||
vbroadcastsd %xmm10, %zmm1 #127.23
|
||||
vbroadcastsd %xmm6, %zmm0 #128.23
|
||||
vbroadcastsd %xmm12, %zmm2 #129.23
|
||||
movslq %r12d, %rbx #153.13
|
||||
addq %rsi, %r13 #108.5
|
||||
movq %rax, 40(%rsp) #108.5[spill]
|
||||
movq %rcx, 48(%rsp) #108.5[spill]
|
||||
movq %rsi, 56(%rsp) #108.5[spill]
|
||||
movq %r8, 64(%rsp) #108.5[spill]
|
||||
movq %rdi, 72(%rsp) #108.5[spill]
|
||||
movq %r9, 24(%rsp) #108.5[spill]
|
||||
movq %r10, 32(%rsp) #108.5[spill]
|
||||
# LOE rdx rbx r13 r11d r12d r14d xmm6 xmm10 xmm12 ymm15 ymm16 zmm0 zmm1 zmm2 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.27: # Preds ..B1.31 ..B1.26
|
||||
# Execution count [2.50e+01]
|
||||
vmovdqu (%r13,%rbx,4), %ymm3 #154.25
|
||||
vpaddd %ymm3, %ymm3, %ymm4 #155.40
|
||||
vpaddd %ymm4, %ymm3, %ymm3 #155.40
|
||||
movl (%r13,%rbx,4), %r10d #154.25
|
||||
movl 4(%r13,%rbx,4), %r9d #154.25
|
||||
movl 8(%r13,%rbx,4), %r8d #154.25
|
||||
movl 12(%r13,%rbx,4), %edi #154.25
|
||||
lea (%r10,%r10,2), %r10d #155.40
|
||||
movl 16(%r13,%rbx,4), %esi #154.25
|
||||
lea (%r9,%r9,2), %r9d #155.40
|
||||
movl 20(%r13,%rbx,4), %ecx #154.25
|
||||
lea (%r8,%r8,2), %r8d #155.40
|
||||
movl 24(%r13,%rbx,4), %eax #154.25
|
||||
lea (%rdi,%rdi,2), %edi #155.40
|
||||
movl 28(%r13,%rbx,4), %r15d #154.25
|
||||
lea (%rsi,%rsi,2), %esi #155.40
|
||||
lea (%rcx,%rcx,2), %ecx #155.40
|
||||
lea (%rax,%rax,2), %eax #155.40
|
||||
lea (%r15,%r15,2), %r15d #155.40
|
||||
# LOE rdx rbx r13 eax ecx esi edi r8d r9d r10d r11d r12d r14d r15d xmm6 xmm10 xmm12 ymm3 ymm15 ymm16 zmm0 zmm1 zmm2 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.30: # Preds ..B1.27
|
||||
# Execution count [1.25e+01]
|
||||
vpcmpeqb %xmm0, %xmm0, %k1 #155.40
|
||||
vpcmpeqb %xmm0, %xmm0, %k2 #155.40
|
||||
vpcmpeqb %xmm0, %xmm0, %k3 #155.40
|
||||
vpxord %zmm4, %zmm4, %zmm4 #155.40
|
||||
vpxord %zmm17, %zmm17, %zmm17 #155.40
|
||||
vpxord %zmm18, %zmm18, %zmm18 #155.40
|
||||
vgatherdpd 16(%rdx,%ymm3,8), %zmm4{%k1} #155.40
|
||||
vgatherdpd 8(%rdx,%ymm3,8), %zmm17{%k2} #155.40
|
||||
vgatherdpd (%rdx,%ymm3,8), %zmm18{%k3} #155.40
|
||||
# LOE rdx rbx r13 r11d r12d r14d xmm6 xmm10 xmm12 ymm15 ymm16 zmm0 zmm1 zmm2 zmm4 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14 zmm17 zmm18
|
||||
..B1.31: # Preds ..B1.30
|
||||
# Execution count [2.50e+01]
|
||||
addl $8, %r12d #153.13
|
||||
addq $8, %rbx #153.13
|
||||
#vsubpd %zmm4, %zmm2, %zmm26 #157.40
|
||||
#vsubpd %zmm17, %zmm0, %zmm24 #156.40
|
||||
#vsubpd %zmm18, %zmm1, %zmm23 #155.40
|
||||
#vmulpd %zmm24, %zmm24, %zmm3 #158.53
|
||||
#vfmadd231pd %zmm23, %zmm23, %zmm3 #158.53
|
||||
#vfmadd231pd %zmm26, %zmm26, %zmm3 #158.67
|
||||
#vrcp14pd %zmm3, %zmm22 #175.42
|
||||
#vcmppd $1, %zmm14, %zmm3, %k2 #174.26
|
||||
#vfpclasspd $30, %zmm22, %k0 #175.42
|
||||
#vfnmadd213pd .L_2il0floatpacket.9(%rip){1to8}, %zmm22, %zmm3 #175.42
|
||||
#knotw %k0, %k1 #175.42
|
||||
#vmulpd %zmm3, %zmm3, %zmm4 #175.42
|
||||
#vfmadd213pd %zmm22, %zmm3, %zmm22{%k1} #175.42
|
||||
#vfmadd213pd %zmm22, %zmm4, %zmm22{%k1} #175.42
|
||||
#vmulpd %zmm13, %zmm22, %zmm17 #176.42
|
||||
#vmulpd %zmm9, %zmm22, %zmm19 #177.58
|
||||
#vmulpd %zmm17, %zmm22, %zmm20 #176.48
|
||||
#vmulpd %zmm20, %zmm22, %zmm18 #176.54
|
||||
#vfmsub213pd %zmm5, %zmm20, %zmm22 #177.58
|
||||
#vmulpd %zmm19, %zmm18, %zmm21 #177.65
|
||||
#vmulpd %zmm22, %zmm21, %zmm25 #177.71
|
||||
#vfmadd231pd %zmm23, %zmm25, %zmm8{%k2} #178.21
|
||||
#vfmadd231pd %zmm24, %zmm25, %zmm7{%k2} #179.21
|
||||
#vfmadd231pd %zmm26, %zmm25, %zmm11{%k2} #180.21
|
||||
cmpl %r14d, %r12d #153.13
|
||||
jb ..B1.27 # Prob 82% #153.13
|
||||
# LOE rdx rbx r13 r11d r12d r14d xmm6 xmm10 xmm12 ymm15 ymm16 zmm0 zmm1 zmm2 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.32: # Preds ..B1.31
|
||||
# Execution count [4.50e+00]
|
||||
movq 40(%rsp), %rax #[spill]
|
||||
movq 48(%rsp), %rcx #[spill]
|
||||
movq 56(%rsp), %rsi #[spill]
|
||||
movq 64(%rsp), %r8 #[spill]
|
||||
movq 72(%rsp), %rdi #[spill]
|
||||
movq 24(%rsp), %r9 #[spill]
|
||||
movq 32(%rsp), %r10 #[spill]
|
||||
# LOE rax rdx rcx rsi rdi r8 r9 r10 r11d r14d xmm6 xmm10 xmm12 ymm15 ymm16 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.33: # Preds ..B1.32 ..B1.25 ..B1.45
|
||||
# Execution count [5.00e+00]
|
||||
lea 1(%r14), %ebx #153.13
|
||||
cmpl %r11d, %ebx #153.13
|
||||
ja ..B1.39 # Prob 50% #153.13
|
||||
# LOE rax rdx rcx rsi rdi r8 r9 r10 r11d r14d xmm6 xmm10 xmm12 ymm15 ymm16 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.34: # Preds ..B1.33
|
||||
# Execution count [2.50e+01]
|
||||
imulq %rcx, %rdi #125.43
|
||||
vbroadcastsd %xmm10, %zmm4 #127.23
|
||||
subl %r14d, %r11d #153.13
|
||||
addq %rsi, %rdi #108.5
|
||||
vpbroadcastd %r11d, %ymm0 #153.13
|
||||
vpcmpgtd %ymm15, %ymm0, %k3 #153.13
|
||||
movslq %r14d, %r14 #153.13
|
||||
vmovdqu32 (%rdi,%r14,4), %ymm1{%k3}{z} #154.25
|
||||
kmovw %k3, %edi #153.13
|
||||
vpaddd %ymm1, %ymm1, %ymm2 #155.40
|
||||
vpaddd %ymm2, %ymm1, %ymm0 #155.40
|
||||
# LOE rax rdx rcx rsi r8 r9 r10 edi xmm6 xmm12 ymm0 ymm15 ymm16 zmm4 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14 k3
|
||||
..B1.37: # Preds ..B1.34
|
||||
# Execution count [1.25e+01]
|
||||
kmovw %k3, %k1 #155.40
|
||||
kmovw %k3, %k2 #155.40
|
||||
vpxord %zmm1, %zmm1, %zmm1 #155.40
|
||||
vpxord %zmm2, %zmm2, %zmm2 #155.40
|
||||
vpxord %zmm3, %zmm3, %zmm3 #155.40
|
||||
vgatherdpd 16(%rdx,%ymm0,8), %zmm1{%k1} #155.40
|
||||
vgatherdpd 8(%rdx,%ymm0,8), %zmm2{%k2} #155.40
|
||||
vgatherdpd (%rdx,%ymm0,8), %zmm3{%k3} #155.40
|
||||
# LOE rax rdx rcx rsi r8 r9 r10 edi xmm6 xmm12 ymm15 ymm16 zmm1 zmm2 zmm3 zmm4 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.38: # Preds ..B1.37
|
||||
# Execution count [2.50e+01]
|
||||
#vbroadcastsd %xmm6, %zmm6 #128.23
|
||||
#vbroadcastsd %xmm12, %zmm12 #129.23
|
||||
#vsubpd %zmm1, %zmm12, %zmm23 #157.40
|
||||
#vsubpd %zmm2, %zmm6, %zmm21 #156.40
|
||||
#vsubpd %zmm3, %zmm4, %zmm20 #155.40
|
||||
#vmulpd %zmm21, %zmm21, %zmm19 #158.53
|
||||
#vfmadd231pd %zmm20, %zmm20, %zmm19 #158.53
|
||||
#vfmadd231pd %zmm23, %zmm23, %zmm19 #158.67
|
||||
#vrcp14pd %zmm19, %zmm18 #175.42
|
||||
#vcmppd $1, %zmm14, %zmm19, %k2 #174.26
|
||||
#vfpclasspd $30, %zmm18, %k0 #175.42
|
||||
#kmovw %k2, %ebx #174.26
|
||||
#knotw %k0, %k1 #175.42
|
||||
#vmovaps %zmm19, %zmm0 #175.42
|
||||
#andl %ebx, %edi #174.26
|
||||
#vfnmadd213pd .L_2il0floatpacket.9(%rip){1to8}, %zmm18, %zmm0 #175.42
|
||||
#kmovw %edi, %k3 #178.21
|
||||
#vmulpd %zmm0, %zmm0, %zmm1 #175.42
|
||||
#vfmadd213pd %zmm18, %zmm0, %zmm18{%k1} #175.42
|
||||
#vfmadd213pd %zmm18, %zmm1, %zmm18{%k1} #175.42
|
||||
#vmulpd %zmm13, %zmm18, %zmm2 #176.42
|
||||
#vmulpd %zmm9, %zmm18, %zmm4 #177.58
|
||||
#vmulpd %zmm2, %zmm18, %zmm10 #176.48
|
||||
#vmulpd %zmm10, %zmm18, %zmm3 #176.54
|
||||
#vfmsub213pd %zmm5, %zmm10, %zmm18 #177.58
|
||||
#vmulpd %zmm4, %zmm3, %zmm17 #177.65
|
||||
#vmulpd %zmm18, %zmm17, %zmm22 #177.71
|
||||
#vfmadd231pd %zmm20, %zmm22, %zmm8{%k3} #178.21
|
||||
#vfmadd231pd %zmm21, %zmm22, %zmm7{%k3} #179.21
|
||||
#vfmadd231pd %zmm23, %zmm22, %zmm11{%k3} #180.21
|
||||
# LOE rax rdx rcx rsi r8 r9 r10 ymm15 ymm16 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.39: # Preds ..B1.24 ..B1.38 ..B1.33
|
||||
# Execution count [4.50e+00]
|
||||
vmovups .L_2il0floatpacket.10(%rip), %zmm19 #132.22
|
||||
vpermd %zmm11, %zmm19, %zmm0 #132.22
|
||||
vpermd %zmm7, %zmm19, %zmm6 #131.22
|
||||
vpermd %zmm8, %zmm19, %zmm20 #130.22
|
||||
vaddpd %zmm11, %zmm0, %zmm11 #132.22
|
||||
vaddpd %zmm7, %zmm6, %zmm7 #131.22
|
||||
vaddpd %zmm8, %zmm20, %zmm8 #130.22
|
||||
vpermpd $78, %zmm11, %zmm1 #132.22
|
||||
vpermpd $78, %zmm7, %zmm10 #131.22
|
||||
vpermpd $78, %zmm8, %zmm21 #130.22
|
||||
vaddpd %zmm1, %zmm11, %zmm2 #132.22
|
||||
vaddpd %zmm10, %zmm7, %zmm12 #131.22
|
||||
vaddpd %zmm21, %zmm8, %zmm22 #130.22
|
||||
vpermpd $177, %zmm2, %zmm3 #132.22
|
||||
vpermpd $177, %zmm12, %zmm17 #131.22
|
||||
vpermpd $177, %zmm22, %zmm23 #130.22
|
||||
vaddpd %zmm3, %zmm2, %zmm4 #132.22
|
||||
vaddpd %zmm17, %zmm12, %zmm18 #131.22
|
||||
vaddpd %zmm23, %zmm22, %zmm24 #130.22
|
||||
# LOE rax rdx rcx rsi r8 r9 r10 xmm4 xmm18 xmm24 ymm15 ymm16 zmm5 zmm9 zmm13 zmm14
|
||||
..B1.40: # Preds ..B1.39 ..B1.10
|
||||
# Execution count [5.00e+00]
|
||||
movq 96(%rsp), %rbx #188.9[spill]
|
||||
addq $24, %rax #124.5
|
||||
movslq %r8d, %rdi #124.32
|
||||
incq %rdi #124.32
|
||||
#vaddsd (%rbx,%r8,8), %xmm24, %xmm0 #188.9
|
||||
#vmovsd %xmm0, (%rbx,%r8,8) #188.9
|
||||
#vaddsd (%r10,%r8,8), %xmm18, %xmm1 #189.9
|
||||
#vmovsd %xmm1, (%r10,%r8,8) #189.9
|
||||
#vaddsd (%r9,%r8,8), %xmm4, %xmm2 #190.9
|
||||
#vmovsd %xmm2, (%r9,%r8,8) #190.9
|
||||
incq %r8 #124.5
|
||||
cmpq 80(%rsp), %r8 #124.5[spill]
|
||||
jb ..B1.10 # Prob 82% #124.5
|
||||
# LOE rax rdx rcx rsi rdi r8 r9 r10 ymm15 ymm16 zmm5 zmm9 zmm13 zmm14
|
||||
..B1.41: # Preds ..B1.40
|
||||
# Execution count [9.00e-01]
|
||||
movq 8(%rsp), %r15 #[spill]
|
||||
.cfi_restore 15
|
||||
movq (%rsp), %rbx #[spill]
|
||||
.cfi_restore 3
|
||||
# LOE rbx r15
|
||||
..B1.42: # Preds ..B1.2 ..B1.41
|
||||
# Execution count [1.00e+00]
|
||||
xorl %eax, %eax #201.16
|
||||
vzeroupper #201.16
|
||||
..___tag_value_computeForce.43:
|
||||
# getTimeStamp()
|
||||
call getTimeStamp #201.16
|
||||
..___tag_value_computeForce.44:
|
||||
# LOE rbx r15 xmm0
|
||||
..B1.43: # Preds ..B1.42
|
||||
# Execution count [1.00e+00]
|
||||
vsubsd 16(%rsp), %xmm0, %xmm0 #204.14[spill]
|
||||
addq $104, %rsp #204.14
|
||||
.cfi_restore 14
|
||||
popq %r14 #204.14
|
||||
.cfi_restore 13
|
||||
popq %r13 #204.14
|
||||
.cfi_restore 12
|
||||
popq %r12 #204.14
|
||||
movq %rbp, %rsp #204.14
|
||||
popq %rbp #204.14
|
||||
.cfi_def_cfa 7, 8
|
||||
.cfi_restore 6
|
||||
ret #204.14
|
||||
.cfi_def_cfa 6, 16
|
||||
.cfi_escape 0x10, 0x03, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x80, 0xff, 0xff, 0xff, 0x22
|
||||
.cfi_offset 6, -16
|
||||
.cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xf8, 0xff, 0xff, 0xff, 0x22
|
||||
.cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xf0, 0xff, 0xff, 0xff, 0x22
|
||||
.cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xe8, 0xff, 0xff, 0xff, 0x22
|
||||
.cfi_escape 0x10, 0x0f, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x88, 0xff, 0xff, 0xff, 0x22
|
||||
# LOE
|
||||
..B1.44: # Preds ..B1.12
|
||||
# Execution count [4.50e-01]: Infreq
|
||||
movl %r11d, %r14d #153.13
|
||||
xorl %r12d, %r12d #153.13
|
||||
andl $-8, %r14d #153.13
|
||||
jmp ..B1.25 # Prob 100% #153.13
|
||||
# LOE rax rdx rcx rsi rdi r8 r9 r10 r11d r12d r14d xmm6 xmm10 xmm12 ymm15 ymm16 zmm5 zmm7 zmm8 zmm9 zmm11 zmm13 zmm14
|
||||
..B1.45: # Preds ..B1.11
|
||||
# Execution count [4.50e-01]: Infreq
|
||||
xorl %r14d, %r14d #153.13
|
||||
jmp ..B1.33 # Prob 100% #153.13
|
||||
.align 16,0x90
|
||||
# LOE rax rdx rcx rsi rdi r8 r9 r10 r11d r14d xmm6 xmm10 xmm12 ymm15 ymm16 zmm5 zmm7 zmm8 zmm9 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
|
||||
324
asm/unused/force.s
Normal file
324
asm/unused/force.s
Normal file
@ -0,0 +1,324 @@
|
||||
.intel_syntax noprefix
|
||||
|
||||
.text
|
||||
.align 16,0x90
|
||||
.globl computeForce
|
||||
computeForce:
|
||||
# parameter 1: rdi Parameter*
|
||||
# parameter 2: rsi Atom*
|
||||
# parameter 3: rdx Neighbor*
|
||||
push rbp
|
||||
push r12
|
||||
push r13
|
||||
push r14
|
||||
push r15
|
||||
push rbx
|
||||
#call getTimeStamp # xmm0 <- getTimeStamp()
|
||||
#vmovsd QWORD PTR [-56+rsp], xmm0 # [-56+rsp] <- xmm0 [spill]
|
||||
mov r9d, DWORD PTR [4+rsi] # r9d <- atom->Nlocal
|
||||
vmovsd xmm2, QWORD PTR [96+rdi] # xmm2 <- param->cutforce
|
||||
vmovsd xmm1, QWORD PTR [32+rdi] # xmm1 <- param->sigma6
|
||||
vmovsd xmm0, QWORD PTR [24+rdi] # xmm0 <- param->epsilon
|
||||
mov r13, QWORD PTR [64+rsi] # r13 <- atom->fx
|
||||
mov r14, QWORD PTR [72+rsi] # r14 <- atom->fy
|
||||
mov rdi, QWORD PTR [80+rsi] # rdi <- atom->fz
|
||||
test r9d, r9d # atom->Nlocal <= 0
|
||||
jle ..atom_loop_exit
|
||||
xor r10d, r10d # r10d <- 0
|
||||
mov ecx, r9d # ecx <- atom->Nlocal
|
||||
xor r8d, r8d # r8d <- 0
|
||||
mov r11d, 1 # r11d <- 1
|
||||
xor eax, eax # eax <- 0
|
||||
shr ecx, 1 # ecx <- atom->Nlocal >> 1
|
||||
je ..zero_last_element # ecx == 0
|
||||
|
||||
# Init forces to zero loop (unroll factor = 2)
|
||||
..init_force_loop:
|
||||
mov QWORD PTR [r8+r13], rax # fx[i] <- 0
|
||||
mov QWORD PTR [r8+r14], rax # fy[i] <- 0
|
||||
mov QWORD PTR [r8+rdi], rax # fz[i] <- 0
|
||||
mov QWORD PTR [8+r8+r13], rax # fx[i] <- 0
|
||||
mov QWORD PTR [8+r8+r14], rax # fy[i] <- 0
|
||||
mov QWORD PTR [8+r8+rdi], rax # fz[i] <- 0
|
||||
add r8, 16 # i++
|
||||
inc r10 # i++
|
||||
cmp r10, rcx # i < Nlocal
|
||||
jb ..init_force_loop
|
||||
|
||||
# Trick to make r11d contain value of last element to be zeroed plus 1
|
||||
# Maybe we can directly put r10+10 here and zero r11d above, then remove the -1 below
|
||||
lea r11d, DWORD PTR [1+r10+r10] # r11d <- i * 2 + 1
|
||||
..zero_last_element:
|
||||
lea ecx, DWORD PTR [-1+r11] # ecx <- i * 2
|
||||
cmp ecx, r9d # i >= Nlocal
|
||||
jae ..before_atom_loop
|
||||
|
||||
# Set last element to zero
|
||||
movsxd r11, r11d # r11 <- i * 2
|
||||
mov QWORD PTR [-8+r13+r11*8], rax # fx[i] <- 0
|
||||
mov QWORD PTR [-8+r14+r11*8], rax # fy[i] <- 0
|
||||
mov QWORD PTR [-8+rdi+r11*8], rax # fz[i] <- 0
|
||||
|
||||
# Initialize registers to be used within atom loop
|
||||
..before_atom_loop:
|
||||
vmulsd xmm15, xmm2, xmm2 # xmm15 <- cutforcesq
|
||||
vmovdqu32 ymm18, YMMWORD PTR .L_2il0floatpacket.0[rip] # ymm18 <- [8, ...]
|
||||
vmulsd xmm0, xmm0, QWORD PTR .L_2il0floatpacket.3[rip] # xmm0 <- 48 * epsilon
|
||||
vmovdqu32 ymm17, YMMWORD PTR .L_2il0floatpacket.1[rip] # ymm17 <- [0..7]
|
||||
vmovups zmm7, ZMMWORD PTR .L_2il0floatpacket.4[rip] # zmm7 <- [0.5, ...]
|
||||
vbroadcastsd zmm16, xmm15 # zmm16 <- [cutforcesq, ...]
|
||||
vbroadcastsd zmm15, xmm1 # zmm15 <- [param->sigma6, ...]
|
||||
vbroadcastsd zmm14, xmm0 # zmm14 <- [48 * epsilon, ...]
|
||||
movsxd r9, r9d # r9 <- atom->Nlocal
|
||||
xor r10d, r10d # r10d <- 0 (i)
|
||||
mov rcx, QWORD PTR [24+rdx] # rcx <- neighbor->numneigh
|
||||
mov r11, QWORD PTR [8+rdx] # r11 <- neighbor->neighbors
|
||||
movsxd r12, DWORD PTR [16+rdx] # r12 <- neighbor->maxneighs
|
||||
mov rdx, QWORD PTR [16+rsi] # rdx <- atom->x
|
||||
### AOS
|
||||
xor eax, eax
|
||||
### SOA
|
||||
#mov rax, QWORD PTR [24+rsi] # rax <- atom->y
|
||||
#mov rsi, QWORD PTR [32+rsi] # rsi <- atom->z
|
||||
###
|
||||
shl r12, 2 # r12 <- neighbor->maxneighs * 4
|
||||
|
||||
# Register spilling
|
||||
mov QWORD PTR [-32+rsp], r9 # [-32+rsp] <- atom->Nlocal
|
||||
mov QWORD PTR [-24+rsp], rcx # [-24+rsp] <- neighbor->numneigh
|
||||
mov QWORD PTR [-16+rsp], r14 # [-16+rsp] <- atom->fy
|
||||
mov QWORD PTR [-8+rsp], r13 # [-8+rsp] <- atom->fx
|
||||
mov QWORD PTR [-40+rsp], r15 # [-40+rsp] <- r15
|
||||
mov QWORD PTR [-48+rsp], rbx # [-48+rsp] <- rbx
|
||||
|
||||
..atom_loop_begin:
|
||||
mov rcx, QWORD PTR [-24+rsp] # rcx <- neighbor->numneigh
|
||||
vxorpd xmm25, xmm25, xmm25 # xmm25 <- 0 (fix)
|
||||
vmovapd xmm20, xmm25 # xmm20 <- 0 (fiy)
|
||||
mov r13d, DWORD PTR [rcx+r10*4] # r13d <- neighbor->numneigh[i] (numneighs)
|
||||
vmovapd xmm4, xmm20 # xmm4 <- 0 (fiz)
|
||||
|
||||
### AOS
|
||||
vmovsd xmm8, QWORD PTR[rdx+rax] # xmm8 <- atom->x[i * 3]
|
||||
vmovsd xmm9, QWORD PTR[8+rdx+rax] # xmm9 <- atom->x[i * 3 + 1]
|
||||
vmovsd xmm10, QWORD PTR[16+rdx+rax] # xmm10 <- atom->x[i * 3 + 2]
|
||||
### SOA
|
||||
#vmovsd xmm8, QWORD PTR [rdx+r10*8] # xmm8 <- atom->x[i]
|
||||
#vmovsd xmm9, QWORD PTR [rax+r10*8] # xmm9 <- atom->y[i]
|
||||
#vmovsd xmm10, QWORD PTR [rsi+r10*8] # xmm10 <- atom->z[i]
|
||||
###
|
||||
vbroadcastsd zmm0, xmm8 # zmm0 <- atom_x(i)
|
||||
vbroadcastsd zmm1, xmm9 # zmm1 <- atom_y(i)
|
||||
vbroadcastsd zmm2, xmm10 # zmm2 <- atom_z(i)
|
||||
test r13d, r13d # numneighs <= 0
|
||||
jle ..atom_loop_exit
|
||||
|
||||
vpxord zmm13, zmm13, zmm13 # zmm13 <- 0 (fix)
|
||||
vmovaps zmm12, zmm13 # zmm12 <- 0 (fiy)
|
||||
vmovaps zmm11, zmm12 # zmm11 <- 0 (fiz)
|
||||
mov rcx, r12 # rcx <- neighbor->maxneighs * 4
|
||||
imul rcx, r10 # rcx <- neighbor->maxneighs * 4 * i
|
||||
add rcx, r11 # rcx <- &neighbor->neighbors[neighbor->maxneighs * i]
|
||||
xor r9d, r9d # r9d <- 0 (k)
|
||||
mov r14d, r13d # r14d <- numneighs
|
||||
cmp r14d, 8
|
||||
jl ..compute_forces_remainder
|
||||
|
||||
..compute_forces:
|
||||
vpcmpeqb k1, xmm0, xmm0
|
||||
vpcmpeqb k2, xmm0, xmm0
|
||||
vpcmpeqb k3, xmm0, xmm0
|
||||
vmovdqu ymm3, YMMWORD PTR [rcx+r9*4]
|
||||
vpxord zmm5, zmm5, zmm5
|
||||
vpxord zmm6, zmm6, zmm6
|
||||
|
||||
### AOS
|
||||
vpaddd ymm4, ymm3, ymm3
|
||||
vpaddd ymm3, ymm3, ymm4
|
||||
vpxord zmm4, zmm4, zmm4
|
||||
vgatherdpd zmm4{k1}, [rdx+ymm3*8]
|
||||
vgatherdpd zmm5{k2}, [8+rdx+ymm3*8]
|
||||
vgatherdpd zmm6{k3}, [16+rdx+ymm3*8]
|
||||
### SOA
|
||||
#vpxord zmm4, zmm4, zmm4
|
||||
#vgatherdpd zmm5{k2}, [rax+ymm3*8]
|
||||
#vgatherdpd zmm4{k1}, [rdx+ymm3*8]
|
||||
#vgatherdpd zmm6{k3}, [rsi+ymm3*8]
|
||||
###
|
||||
|
||||
vsubpd zmm29, zmm1, zmm5 # zmm29 <- atom_y(i) - atom_y(j) -- dely
|
||||
vsubpd zmm28, zmm0, zmm4 # zmm28 <- atom_x(i) - atom_x(j) -- delx
|
||||
vsubpd zmm31, zmm2, zmm6 # zmm31 <- atom_z(i) - atom_z(j) -- delz
|
||||
vmulpd zmm20, zmm29, zmm29 # zmm20 <- dely * dely
|
||||
vfmadd231pd zmm20, zmm28, zmm28 # zmm20 <- dely * dely + delx * delx
|
||||
vfmadd231pd zmm20, zmm31, zmm31 # zmm20 <- zmm20 + delz * delz -- rsq
|
||||
|
||||
# Cutoff radius condition
|
||||
vrcp14pd zmm27, zmm20 # zmm27 <- 1.0 / rsq (sr2)
|
||||
vcmppd k5, zmm20, zmm16, 1 # k5 <- rsq < cutforcesq
|
||||
vmulpd zmm22, zmm27, zmm15 # zmm22 <- sr2 * sigma6
|
||||
vmulpd zmm24, zmm27, zmm14 # zmm24 <- 48.0 * epsilon * sr2
|
||||
vmulpd zmm25, zmm27, zmm22 # zmm25 <- sr2 * sigma6 * sr2
|
||||
vmulpd zmm23, zmm27, zmm25 # zmm23 <- sr2 * sigma6 * sr2 * sr2
|
||||
vfmsub213pd zmm27, zmm25, zmm7 # zmm27 <- sr2 * sigma * sr2 * sr2 - 0.5
|
||||
vmulpd zmm26, zmm23, zmm24 # zmm26 <- 48.0 * epsilon * sr2 * sr2 * sigma6 * sr2
|
||||
vmulpd zmm30, zmm26, zmm27 # zmm30 <- force
|
||||
vfmadd231pd zmm13{k5}, zmm30, zmm28 # fix += force * delx
|
||||
vfmadd231pd zmm12{k5}, zmm30, zmm29 # fiy += force * dely
|
||||
vfmadd231pd zmm11{k5}, zmm30, zmm31 # fiz += force * delz
|
||||
sub r14d, 8
|
||||
add r9, 8
|
||||
cmp r14d, 8
|
||||
jge ..compute_forces
|
||||
|
||||
# Check if there are remaining neighbors to be computed
|
||||
..compute_forces_remainder:
|
||||
test r14d, r14d
|
||||
jle ..sum_up_forces
|
||||
|
||||
vpbroadcastd ymm4, r14d
|
||||
vpcmpgtd k1, ymm4, ymm17
|
||||
kmovw r15d, k1
|
||||
vmovdqu32 ymm3{k1}{z}, YMMWORD PTR [rcx+r9*4]
|
||||
kmovw k2, k1
|
||||
kmovw k3, k1
|
||||
vpxord zmm5, zmm5, zmm5
|
||||
vpxord zmm6, zmm6, zmm6
|
||||
|
||||
### AOS
|
||||
vpaddd ymm4, ymm3, ymm3
|
||||
vpaddd ymm3, ymm3, ymm4
|
||||
vpxord zmm4, zmm4, zmm4
|
||||
vgatherdpd zmm4{k1}, [rdx+ymm3*8]
|
||||
vgatherdpd zmm5{k2}, [8+rdx+ymm3*8]
|
||||
vgatherdpd zmm6{k3}, [16+rdx+ymm3*8]
|
||||
#### SOA
|
||||
#vpxord zmm4, zmm4, zmm4
|
||||
#vgatherdpd zmm5{k2}, [rax+ymm3*8]
|
||||
#vgatherdpd zmm4{k1}, [rdx+ymm3*8]
|
||||
#vgatherdpd zmm6{k3}, [rsi+ymm3*8]
|
||||
###
|
||||
|
||||
vsubpd zmm29, zmm1, zmm5 # zmm29 <- atom_y(i) - atom_y(j) -- dely
|
||||
vsubpd zmm28, zmm0, zmm4 # zmm28 <- atom_x(i) - atom_x(j) -- delx
|
||||
vsubpd zmm31, zmm2, zmm6 # zmm31 <- atom_z(i) - atom_z(j) -- delz
|
||||
vmulpd zmm20, zmm29, zmm29 # zmm20 <- dely * dely
|
||||
vfmadd231pd zmm20, zmm28, zmm28 # zmm20 <- dely * dely + delx * delx
|
||||
vfmadd231pd zmm20, zmm31, zmm31 # zmm20 <- zmm20 + delz * delz -- rsq
|
||||
|
||||
# Cutoff radius condition
|
||||
vrcp14pd zmm27, zmm20 # zmm27 <- 1.0 / rsq (sr2)
|
||||
vcmppd k5, zmm20, zmm16, 1 # k5 <- rsq < cutforcesq
|
||||
kmovw r9d, k5 # r9d <- rsq < cutforcesq
|
||||
and r15d, r9d # r15d <- rsq < cutforcesq && k < numneighs
|
||||
kmovw k3, r15d # k3 <- rsq < cutforcesq && k < numneighs
|
||||
vmulpd zmm22, zmm27, zmm15 # zmm22 <- sr2 * sigma6
|
||||
vmulpd zmm24, zmm27, zmm14 # zmm24 <- 48.0 * epsilon * sr2
|
||||
vmulpd zmm25, zmm27, zmm22 # zmm25 <- sr2 * sigma6 * sr2
|
||||
vmulpd zmm23, zmm27, zmm25 # zmm23 <- sr2 * sigma6 * sr2 * sr2
|
||||
vfmsub213pd zmm27, zmm25, zmm7 # zmm27 <- sr2 * sigma * sr2 * sr2 - 0.5
|
||||
vmulpd zmm26, zmm23, zmm24 # zmm26 <- 48.0 * epsilon * sr2 * sr2 * sigma6 * sr2
|
||||
vmulpd zmm30, zmm26, zmm27 # zmm30 <- force
|
||||
vfmadd231pd zmm13{k3}, zmm30, zmm28 # fix += force * delx
|
||||
vfmadd231pd zmm12{k3}, zmm30, zmm29 # fiy += force * dely
|
||||
vfmadd231pd zmm11{k3}, zmm30, zmm31 # fiz += force * delz
|
||||
|
||||
# Forces are currently separated in different lanes of zmm registers, hence it is necessary to permutate
|
||||
# and add them (reduction) to obtain the final contribution for the current atom
|
||||
..sum_up_forces:
|
||||
vmovups zmm10, ZMMWORD PTR .L_2il0floatpacket.6[rip]
|
||||
vpermd zmm0, zmm10, zmm11
|
||||
vpermd zmm5, zmm10, zmm12
|
||||
vpermd zmm21, zmm10, zmm13
|
||||
vaddpd zmm11, zmm0, zmm11
|
||||
vaddpd zmm12, zmm5, zmm12
|
||||
vaddpd zmm13, zmm21, zmm13
|
||||
vpermpd zmm1, zmm11, 78
|
||||
vpermpd zmm6, zmm12, 78
|
||||
vpermpd zmm22, zmm13, 78
|
||||
vaddpd zmm2, zmm11, zmm1
|
||||
vaddpd zmm8, zmm12, zmm6
|
||||
vaddpd zmm23, zmm13, zmm22
|
||||
vpermpd zmm3, zmm2, 177
|
||||
vpermpd zmm9, zmm8, 177
|
||||
vpermpd zmm24, zmm23, 177
|
||||
vaddpd zmm4, zmm2, zmm3
|
||||
vaddpd zmm20, zmm8, zmm9
|
||||
vaddpd zmm25, zmm23, zmm24
|
||||
|
||||
..atom_loop_exit:
|
||||
mov rcx, QWORD PTR [-8+rsp] #84.9[spill]
|
||||
mov rbx, QWORD PTR [-16+rsp] #85.9[spill]
|
||||
|
||||
### AOS
|
||||
add rax, 24
|
||||
###
|
||||
|
||||
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 ..atom_loop_begin
|
||||
vzeroupper #93.12
|
||||
vxorpd xmm0, xmm0, xmm0 #93.12
|
||||
#call getTimeStamp # xmm0 <- getTimeStamp()
|
||||
#vsubsd xmm0, xmm0, QWORD PTR [-56+rsp] # xmm0 <- E-S
|
||||
pop rbx
|
||||
pop r15
|
||||
pop r14 #93.12
|
||||
pop r13 #93.12
|
||||
pop r12 #93.12
|
||||
pop rbp #93.12
|
||||
ret #93.12
|
||||
|
||||
.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
|
||||
326
asm/unused/force_lj.s
Normal file
326
asm/unused/force_lj.s
Normal file
@ -0,0 +1,326 @@
|
||||
.intel_syntax noprefix
|
||||
|
||||
.text
|
||||
.align 16,0x90
|
||||
.globl computeForceLJ
|
||||
computeForceLJ:
|
||||
# parameter 1: rdi Parameter*
|
||||
# parameter 2: rsi Atom*
|
||||
# parameter 3: rdx Neighbor*
|
||||
push rbp
|
||||
push r12
|
||||
push r13
|
||||
push r14
|
||||
push r15
|
||||
push rbx
|
||||
mov r9d, DWORD PTR [4+rsi] # r9d <- atom->Nlocal
|
||||
vmovsd xmm2, QWORD PTR [96+rdi] # xmm2 <- param->cutforce
|
||||
vmovsd xmm1, QWORD PTR [32+rdi] # xmm1 <- param->sigma6
|
||||
vmovsd xmm0, QWORD PTR [24+rdi] # xmm0 <- param->epsilon
|
||||
mov r13, QWORD PTR [64+rsi] # r13 <- atom->fx
|
||||
mov r14, QWORD PTR [72+rsi] # r14 <- atom->fy
|
||||
mov rdi, QWORD PTR [80+rsi] # rdi <- atom->fz
|
||||
test r9d, r9d # atom->Nlocal <= 0
|
||||
jle ..atom_loop_exit
|
||||
xor r10d, r10d # r10d <- 0
|
||||
mov ecx, r9d # ecx <- atom->Nlocal
|
||||
xor r8d, r8d # r8d <- 0
|
||||
mov r11d, 1 # r11d <- 1
|
||||
xor eax, eax # eax <- 0
|
||||
shr ecx, 1 # ecx <- atom->Nlocal >> 1
|
||||
je ..zero_last_element # ecx == 0
|
||||
|
||||
# Init forces to zero loop (unroll factor = 2)
|
||||
..init_force_loop:
|
||||
mov QWORD PTR [r8+r13], rax # fx[i] <- 0
|
||||
mov QWORD PTR [r8+r14], rax # fy[i] <- 0
|
||||
mov QWORD PTR [r8+rdi], rax # fz[i] <- 0
|
||||
mov QWORD PTR [8+r8+r13], rax # fx[i] <- 0
|
||||
mov QWORD PTR [8+r8+r14], rax # fy[i] <- 0
|
||||
mov QWORD PTR [8+r8+rdi], rax # fz[i] <- 0
|
||||
add r8, 16 # i++
|
||||
inc r10 # i++
|
||||
cmp r10, rcx # i < Nlocal
|
||||
jb ..init_force_loop
|
||||
|
||||
# Trick to make r11d contain value of last element to be zeroed plus 1
|
||||
# Maybe we can directly put r10+10 here and zero r11d above, then remove the -1 below
|
||||
lea r11d, DWORD PTR [1+r10+r10] # r11d <- i * 2 + 1
|
||||
..zero_last_element:
|
||||
lea ecx, DWORD PTR [-1+r11] # ecx <- i * 2
|
||||
cmp ecx, r9d # i >= Nlocal
|
||||
jae ..before_atom_loop
|
||||
|
||||
# Set last element to zero
|
||||
movsxd r11, r11d # r11 <- i * 2
|
||||
mov QWORD PTR [-8+r13+r11*8], rax # fx[i] <- 0
|
||||
mov QWORD PTR [-8+r14+r11*8], rax # fy[i] <- 0
|
||||
mov QWORD PTR [-8+rdi+r11*8], rax # fz[i] <- 0
|
||||
|
||||
# Initialize registers to be used within atom loop
|
||||
..before_atom_loop:
|
||||
vmulsd xmm15, xmm2, xmm2 # xmm15 <- cutforcesq
|
||||
vmovdqu32 ymm18, YMMWORD PTR .L_2il0floatpacket.0[rip] # ymm18 <- [8, ...]
|
||||
vmulsd xmm0, xmm0, QWORD PTR .L_2il0floatpacket.3[rip] # xmm0 <- 48 * epsilon
|
||||
vmovdqu32 ymm17, YMMWORD PTR .L_2il0floatpacket.1[rip] # ymm17 <- [0..7]
|
||||
vmovups zmm7, ZMMWORD PTR .L_2il0floatpacket.4[rip] # zmm7 <- [0.5, ...]
|
||||
vbroadcastsd zmm16, xmm15 # zmm16 <- [cutforcesq, ...]
|
||||
vbroadcastsd zmm15, xmm1 # zmm15 <- [param->sigma6, ...]
|
||||
vbroadcastsd zmm14, xmm0 # zmm14 <- [48 * epsilon, ...]
|
||||
movsxd r9, r9d # r9 <- atom->Nlocal
|
||||
xor r10d, r10d # r10d <- 0 (i)
|
||||
mov rcx, QWORD PTR [24+rdx] # rcx <- neighbor->numneigh
|
||||
mov r11, QWORD PTR [8+rdx] # r11 <- neighbor->neighbors
|
||||
movsxd r12, DWORD PTR [16+rdx] # r12 <- neighbor->maxneighs
|
||||
mov rdx, QWORD PTR [16+rsi] # rdx <- atom->x
|
||||
### AOS
|
||||
xor eax, eax
|
||||
### SOA
|
||||
#mov rax, QWORD PTR [24+rsi] # rax <- atom->y
|
||||
#mov rsi, QWORD PTR [32+rsi] # rsi <- atom->z
|
||||
###
|
||||
shl r12, 2 # r12 <- neighbor->maxneighs * 4
|
||||
|
||||
# Register spilling
|
||||
mov QWORD PTR [-32+rsp], r9 # [-32+rsp] <- atom->Nlocal
|
||||
mov QWORD PTR [-24+rsp], rcx # [-24+rsp] <- neighbor->numneigh
|
||||
mov QWORD PTR [-16+rsp], r14 # [-16+rsp] <- atom->fy
|
||||
mov QWORD PTR [-8+rsp], r13 # [-8+rsp] <- atom->fx
|
||||
mov QWORD PTR [-40+rsp], r15 # [-40+rsp] <- r15
|
||||
mov QWORD PTR [-48+rsp], rbx # [-48+rsp] <- rbx
|
||||
#sub rsp, 64
|
||||
#call getTimeStamp # xmm0 <- getTimeStamp()
|
||||
#vmovsd QWORD PTR [-56+rsp], xmm0 # [-56+rsp] <- xmm0 [spill]
|
||||
#add rsp, 64
|
||||
|
||||
..atom_loop_begin:
|
||||
mov rcx, QWORD PTR [-24+rsp] # rcx <- neighbor->numneigh
|
||||
vxorpd xmm25, xmm25, xmm25 # xmm25 <- 0 (fix)
|
||||
vmovapd xmm20, xmm25 # xmm20 <- 0 (fiy)
|
||||
mov r13d, DWORD PTR [rcx+r10*4] # r13d <- neighbor->numneigh[i] (numneighs)
|
||||
vmovapd xmm4, xmm20 # xmm4 <- 0 (fiz)
|
||||
|
||||
### AOS
|
||||
vmovsd xmm8, QWORD PTR[rdx+rax] # xmm8 <- atom->x[i * 3]
|
||||
vmovsd xmm9, QWORD PTR[8+rdx+rax] # xmm9 <- atom->x[i * 3 + 1]
|
||||
vmovsd xmm10, QWORD PTR[16+rdx+rax] # xmm10 <- atom->x[i * 3 + 2]
|
||||
### SOA
|
||||
#vmovsd xmm8, QWORD PTR [rdx+r10*8] # xmm8 <- atom->x[i]
|
||||
#vmovsd xmm9, QWORD PTR [rax+r10*8] # xmm9 <- atom->y[i]
|
||||
#vmovsd xmm10, QWORD PTR [rsi+r10*8] # xmm10 <- atom->z[i]
|
||||
###
|
||||
vbroadcastsd zmm0, xmm8 # zmm0 <- atom_x(i)
|
||||
vbroadcastsd zmm1, xmm9 # zmm1 <- atom_y(i)
|
||||
vbroadcastsd zmm2, xmm10 # zmm2 <- atom_z(i)
|
||||
test r13d, r13d # numneighs <= 0
|
||||
jle ..atom_loop_exit
|
||||
|
||||
vpxord zmm13, zmm13, zmm13 # zmm13 <- 0 (fix)
|
||||
vmovaps zmm12, zmm13 # zmm12 <- 0 (fiy)
|
||||
vmovaps zmm11, zmm12 # zmm11 <- 0 (fiz)
|
||||
mov rcx, r12 # rcx <- neighbor->maxneighs * 4
|
||||
imul rcx, r10 # rcx <- neighbor->maxneighs * 4 * i
|
||||
add rcx, r11 # rcx <- &neighbor->neighbors[neighbor->maxneighs * i]
|
||||
xor r9d, r9d # r9d <- 0 (k)
|
||||
mov r14d, r13d # r14d <- numneighs
|
||||
cmp r14d, 8
|
||||
jl ..compute_forces_remainder
|
||||
|
||||
..compute_forces:
|
||||
vpcmpeqb k1, xmm0, xmm0
|
||||
vpcmpeqb k2, xmm0, xmm0
|
||||
vpcmpeqb k3, xmm0, xmm0
|
||||
vmovdqu ymm3, YMMWORD PTR [rcx+r9*4]
|
||||
vpxord zmm5, zmm5, zmm5
|
||||
vpxord zmm6, zmm6, zmm6
|
||||
|
||||
### AOS
|
||||
vpaddd ymm4, ymm3, ymm3
|
||||
vpaddd ymm3, ymm3, ymm4
|
||||
vpxord zmm4, zmm4, zmm4
|
||||
vgatherdpd zmm4{k1}, [rdx+ymm3*8]
|
||||
vgatherdpd zmm5{k2}, [8+rdx+ymm3*8]
|
||||
vgatherdpd zmm6{k3}, [16+rdx+ymm3*8]
|
||||
### SOA
|
||||
#vpxord zmm4, zmm4, zmm4
|
||||
#vgatherdpd zmm5{k2}, [rax+ymm3*8]
|
||||
#vgatherdpd zmm4{k1}, [rdx+ymm3*8]
|
||||
#vgatherdpd zmm6{k3}, [rsi+ymm3*8]
|
||||
###
|
||||
|
||||
vsubpd zmm29, zmm1, zmm5 # zmm29 <- atom_y(i) - atom_y(j) -- dely
|
||||
vsubpd zmm28, zmm0, zmm4 # zmm28 <- atom_x(i) - atom_x(j) -- delx
|
||||
vsubpd zmm31, zmm2, zmm6 # zmm31 <- atom_z(i) - atom_z(j) -- delz
|
||||
vmulpd zmm20, zmm29, zmm29 # zmm20 <- dely * dely
|
||||
vfmadd231pd zmm20, zmm28, zmm28 # zmm20 <- dely * dely + delx * delx
|
||||
vfmadd231pd zmm20, zmm31, zmm31 # zmm20 <- zmm20 + delz * delz -- rsq
|
||||
|
||||
# Cutoff radius condition
|
||||
vrcp14pd zmm27, zmm20 # zmm27 <- 1.0 / rsq (sr2)
|
||||
vcmppd k5, zmm20, zmm16, 1 # k5 <- rsq < cutforcesq
|
||||
vmulpd zmm22, zmm27, zmm15 # zmm22 <- sr2 * sigma6
|
||||
vmulpd zmm24, zmm27, zmm14 # zmm24 <- 48.0 * epsilon * sr2
|
||||
vmulpd zmm25, zmm27, zmm22 # zmm25 <- sr2 * sigma6 * sr2
|
||||
vmulpd zmm23, zmm27, zmm25 # zmm23 <- sr2 * sigma6 * sr2 * sr2
|
||||
vfmsub213pd zmm27, zmm25, zmm7 # zmm27 <- sr2 * sigma * sr2 * sr2 - 0.5
|
||||
vmulpd zmm26, zmm23, zmm24 # zmm26 <- 48.0 * epsilon * sr2 * sr2 * sigma6 * sr2
|
||||
vmulpd zmm30, zmm26, zmm27 # zmm30 <- force
|
||||
vfmadd231pd zmm13{k5}, zmm30, zmm28 # fix += force * delx
|
||||
vfmadd231pd zmm12{k5}, zmm30, zmm29 # fiy += force * dely
|
||||
vfmadd231pd zmm11{k5}, zmm30, zmm31 # fiz += force * delz
|
||||
sub r14d, 8
|
||||
add r9, 8
|
||||
cmp r14d, 8
|
||||
jge ..compute_forces
|
||||
|
||||
# Check if there are remaining neighbors to be computed
|
||||
..compute_forces_remainder:
|
||||
test r14d, r14d
|
||||
jle ..sum_up_forces
|
||||
|
||||
vpbroadcastd ymm4, r14d
|
||||
vpcmpgtd k1, ymm4, ymm17
|
||||
kmovw r15d, k1
|
||||
vmovdqu32 ymm3{k1}{z}, YMMWORD PTR [rcx+r9*4]
|
||||
kmovw k2, k1
|
||||
kmovw k3, k1
|
||||
vpxord zmm5, zmm5, zmm5
|
||||
vpxord zmm6, zmm6, zmm6
|
||||
|
||||
### AOS
|
||||
vpaddd ymm4, ymm3, ymm3
|
||||
vpaddd ymm3, ymm3, ymm4
|
||||
vpxord zmm4, zmm4, zmm4
|
||||
vgatherdpd zmm4{k1}, [rdx+ymm3*8]
|
||||
vgatherdpd zmm5{k2}, [8+rdx+ymm3*8]
|
||||
vgatherdpd zmm6{k3}, [16+rdx+ymm3*8]
|
||||
#### SOA
|
||||
#vpxord zmm4, zmm4, zmm4
|
||||
#vgatherdpd zmm5{k2}, [rax+ymm3*8]
|
||||
#vgatherdpd zmm4{k1}, [rdx+ymm3*8]
|
||||
#vgatherdpd zmm6{k3}, [rsi+ymm3*8]
|
||||
###
|
||||
|
||||
vsubpd zmm29, zmm1, zmm5 # zmm29 <- atom_y(i) - atom_y(j) -- dely
|
||||
vsubpd zmm28, zmm0, zmm4 # zmm28 <- atom_x(i) - atom_x(j) -- delx
|
||||
vsubpd zmm31, zmm2, zmm6 # zmm31 <- atom_z(i) - atom_z(j) -- delz
|
||||
vmulpd zmm20, zmm29, zmm29 # zmm20 <- dely * dely
|
||||
vfmadd231pd zmm20, zmm28, zmm28 # zmm20 <- dely * dely + delx * delx
|
||||
vfmadd231pd zmm20, zmm31, zmm31 # zmm20 <- zmm20 + delz * delz -- rsq
|
||||
|
||||
# Cutoff radius condition
|
||||
vrcp14pd zmm27, zmm20 # zmm27 <- 1.0 / rsq (sr2)
|
||||
vcmppd k5, zmm20, zmm16, 1 # k5 <- rsq < cutforcesq
|
||||
kmovw r9d, k5 # r9d <- rsq < cutforcesq
|
||||
and r15d, r9d # r15d <- rsq < cutforcesq && k < numneighs
|
||||
kmovw k3, r15d # k3 <- rsq < cutforcesq && k < numneighs
|
||||
vmulpd zmm22, zmm27, zmm15 # zmm22 <- sr2 * sigma6
|
||||
vmulpd zmm24, zmm27, zmm14 # zmm24 <- 48.0 * epsilon * sr2
|
||||
vmulpd zmm25, zmm27, zmm22 # zmm25 <- sr2 * sigma6 * sr2
|
||||
vmulpd zmm23, zmm27, zmm25 # zmm23 <- sr2 * sigma6 * sr2 * sr2
|
||||
vfmsub213pd zmm27, zmm25, zmm7 # zmm27 <- sr2 * sigma * sr2 * sr2 - 0.5
|
||||
vmulpd zmm26, zmm23, zmm24 # zmm26 <- 48.0 * epsilon * sr2 * sr2 * sigma6 * sr2
|
||||
vmulpd zmm30, zmm26, zmm27 # zmm30 <- force
|
||||
vfmadd231pd zmm13{k3}, zmm30, zmm28 # fix += force * delx
|
||||
vfmadd231pd zmm12{k3}, zmm30, zmm29 # fiy += force * dely
|
||||
vfmadd231pd zmm11{k3}, zmm30, zmm31 # fiz += force * delz
|
||||
|
||||
# Forces are currently separated in different lanes of zmm registers, hence it is necessary to permutate
|
||||
# and add them (reduction) to obtain the final contribution for the current atom
|
||||
..sum_up_forces:
|
||||
vmovups zmm10, ZMMWORD PTR .L_2il0floatpacket.6[rip]
|
||||
vpermd zmm0, zmm10, zmm11
|
||||
vpermd zmm5, zmm10, zmm12
|
||||
vpermd zmm21, zmm10, zmm13
|
||||
vaddpd zmm11, zmm0, zmm11
|
||||
vaddpd zmm12, zmm5, zmm12
|
||||
vaddpd zmm13, zmm21, zmm13
|
||||
vpermpd zmm1, zmm11, 78
|
||||
vpermpd zmm6, zmm12, 78
|
||||
vpermpd zmm22, zmm13, 78
|
||||
vaddpd zmm2, zmm11, zmm1
|
||||
vaddpd zmm8, zmm12, zmm6
|
||||
vaddpd zmm23, zmm13, zmm22
|
||||
vpermpd zmm3, zmm2, 177
|
||||
vpermpd zmm9, zmm8, 177
|
||||
vpermpd zmm24, zmm23, 177
|
||||
vaddpd zmm4, zmm2, zmm3
|
||||
vaddpd zmm20, zmm8, zmm9
|
||||
vaddpd zmm25, zmm23, zmm24
|
||||
|
||||
..atom_loop_exit:
|
||||
mov rcx, QWORD PTR [-8+rsp] #84.9[spill]
|
||||
mov rbx, QWORD PTR [-16+rsp] #85.9[spill]
|
||||
|
||||
### AOS
|
||||
add rax, 24
|
||||
###
|
||||
|
||||
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 ..atom_loop_begin
|
||||
vzeroupper #93.12
|
||||
vxorpd xmm0, xmm0, xmm0 #93.12
|
||||
#call getTimeStamp # xmm0 <- getTimeStamp()
|
||||
#vsubsd xmm0, xmm0, QWORD PTR [-56+rsp] # xmm0 <- E-S
|
||||
pop rbx
|
||||
pop r15
|
||||
pop r14 #93.12
|
||||
pop r13 #93.12
|
||||
pop r12 #93.12
|
||||
pop rbp #93.12
|
||||
ret #93.12
|
||||
|
||||
.type computeForceLJ,@function
|
||||
.size computeForceLJ,.-computeForceLJ
|
||||
|
||||
|
||||
..LNcomputeForce.0:
|
||||
.data
|
||||
# -- End computeForceLJ
|
||||
.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
|
||||
1541
asm/unused/force_lj_lammps_avx512_dp_no_newton_raphson.s
Normal file
1541
asm/unused/force_lj_lammps_avx512_dp_no_newton_raphson.s
Normal file
File diff suppressed because it is too large
Load Diff
1421
asm/unused/force_lj_lammps_avx512_sp_no_newton_raphson.s
Normal file
1421
asm/unused/force_lj_lammps_avx512_sp_no_newton_raphson.s
Normal file
File diff suppressed because it is too large
Load Diff
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -21,7 +21,6 @@ typedef struct {
|
||||
char* input_file;
|
||||
char* vtk_file;
|
||||
char* xtc_file;
|
||||
char* write_atom_file;
|
||||
MD_FLOAT epsilon;
|
||||
MD_FLOAT sigma;
|
||||
MD_FLOAT sigma6;
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -48,13 +48,11 @@ static inline MD_FLOAT simd_incr_reduced_sum(MD_FLOAT *m, MD_SIMD_FLOAT v0, MD_S
|
||||
t2 = _mm256_permute2f128_pd(t0, t1, 0x21);
|
||||
t0 = _mm256_add_pd(t0, t2);
|
||||
t1 = _mm256_add_pd(t1, t2);
|
||||
t0 = _mm256_blend_pd(t0, t1, 0xC);
|
||||
//t0 = _mm256_blend_pd(t0, t1, 0b1100);
|
||||
t0 = _mm256_blend_pd(t0, t1, 0b1100);
|
||||
t1 = _mm256_add_pd(t0, _mm256_load_pd(m));
|
||||
_mm256_store_pd(m, t1);
|
||||
|
||||
t0 = _mm256_add_pd(t0, _mm256_permute_pd(t0, 0x5));
|
||||
//t0 = _mm256_add_pd(t0, _mm256_permute_pd(t0, 0b0101));
|
||||
t0 = _mm256_add_pd(t0, _mm256_permute_pd(t0, 0b0101));
|
||||
a0 = _mm256_castpd256_pd128(t0);
|
||||
a1 = _mm256_extractf128_pd(t0, 0x1);
|
||||
a0 = _mm_add_sd(a0, a1);
|
||||
@ -93,7 +91,7 @@ static inline void simd_h_decr3(MD_FLOAT *m, MD_SIMD_FLOAT a0, MD_SIMD_FLOAT a1,
|
||||
}
|
||||
|
||||
// Functions used in LAMMPS kernel
|
||||
#define simd_gather(vidx, m, s) _mm256_i32gather_pd(m, vidx, s);
|
||||
static inline MD_SIMD_FLOAT simd_gather(MD_SIMD_INT vidx, const MD_FLOAT *m, int s) { return _mm256_i32gather_pd(m, vidx, s); }
|
||||
static inline MD_SIMD_INT simd_int_broadcast(int scalar) { return _mm_set1_epi32(scalar); }
|
||||
static inline MD_SIMD_INT simd_int_zero() { return _mm_setzero_si128(); }
|
||||
static inline MD_SIMD_INT simd_int_seq() { return _mm_set_epi32(3, 2, 1, 0); }
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -7,7 +7,8 @@
|
||||
#ifndef __TIMING_H_
|
||||
#define __TIMING_H_
|
||||
|
||||
extern double getTimeStamp(void);
|
||||
extern double getTimeResolution(void);
|
||||
extern double getTimeStamp();
|
||||
extern double getTimeResolution();
|
||||
extern double getTimeStamp_();
|
||||
|
||||
#endif
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -7,41 +7,40 @@
|
||||
#ifndef __UTIL_H_
|
||||
#define __UTIL_H_
|
||||
|
||||
#include <stdio.h>
|
||||
#ifndef MIN
|
||||
#define MIN(x, y) ((x) < (y) ? (x) : (y))
|
||||
# define MIN(x,y) ((x)<(y)?(x):(y))
|
||||
#endif
|
||||
|
||||
#ifndef MAX
|
||||
#define MAX(x, y) ((x) > (y) ? (x) : (y))
|
||||
# define MAX(x,y) ((x)>(y)?(x):(y))
|
||||
#endif
|
||||
|
||||
#ifndef ABS
|
||||
#define ABS(a) ((a) >= 0 ? (a) : -(a))
|
||||
# define ABS(a) ((a) >= 0 ? (a) : -(a))
|
||||
#endif
|
||||
|
||||
#define DEBUG_MESSAGE debug_printf
|
||||
|
||||
#ifndef MAXLINE
|
||||
#define MAXLINE 4096
|
||||
# define MAXLINE 4096
|
||||
#endif
|
||||
|
||||
#define FF_LJ 0
|
||||
#define FF_EAM 1
|
||||
#define FF_DEM 2
|
||||
#define FF_LJ 0
|
||||
#define FF_EAM 1
|
||||
#define FF_DEM 2
|
||||
|
||||
#if PRECISION == 1
|
||||
#define PRECISION_STRING "single"
|
||||
# define PRECISION_STRING "single"
|
||||
#else
|
||||
#define PRECISION_STRING "double"
|
||||
# define PRECISION_STRING "double"
|
||||
#endif
|
||||
|
||||
extern double myrandom(int *);
|
||||
extern double myrandom(int*);
|
||||
extern void random_reset(int *seed, int ibase, double *coord);
|
||||
extern int str2ff(const char *string);
|
||||
extern const char *ff2str(int ff);
|
||||
extern const char* ff2str(int ff);
|
||||
extern int get_num_threads();
|
||||
extern void readline(char *line, FILE *fp);
|
||||
extern void debug_printf(const char *format, ...);
|
||||
extern int get_cuda_num_threads(void);
|
||||
|
||||
#endif
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -17,7 +17,6 @@ void initParameter(Parameter *param) {
|
||||
param->vtk_file = NULL;
|
||||
param->xtc_file = NULL;
|
||||
param->eam_file = NULL;
|
||||
param->write_atom_file = NULL;
|
||||
param->force_field = FF_LJ;
|
||||
param->epsilon = 1.0;
|
||||
param->sigma = 1.0;
|
||||
@ -170,11 +169,6 @@ void printParameter(Parameter *param) {
|
||||
printf("\tNumber of timesteps: %d\n", param->ntimes);
|
||||
printf("\tReport stats every (timesteps): %d\n", param->nstat);
|
||||
printf("\tReneighbor every (timesteps): %d\n", param->reneigh_every);
|
||||
#ifdef SORT_ATOMS
|
||||
printf("\tSort atoms when reneighboring: yes\n");
|
||||
#else
|
||||
printf("\tSort atoms when reneighboring: no\n");
|
||||
#endif
|
||||
printf("\tPrune every (timesteps): %d\n", param->prune_every);
|
||||
printf("\tOutput positions every (timesteps): %d\n", param->x_out_every);
|
||||
printf("\tOutput velocities every (timesteps): %d\n", param->v_out_every);
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,21 +1,27 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
*/
|
||||
#include <stdlib.h>
|
||||
#include <time.h>
|
||||
|
||||
double getTimeStamp(void)
|
||||
double getTimeStamp()
|
||||
{
|
||||
struct timespec ts;
|
||||
clock_gettime(CLOCK_MONOTONIC, &ts);
|
||||
return (double)ts.tv_sec + (double)ts.tv_nsec * 1.e-9;
|
||||
}
|
||||
|
||||
double getTimeResolution(void)
|
||||
double getTimeResolution()
|
||||
{
|
||||
struct timespec ts;
|
||||
clock_getres(CLOCK_MONOTONIC, &ts);
|
||||
return (double)ts.tv_sec + (double)ts.tv_nsec * 1.e-9;
|
||||
}
|
||||
|
||||
double getTimeStamp_()
|
||||
{
|
||||
return getTimeStamp();
|
||||
}
|
||||
@ -1,39 +1,38 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
*/
|
||||
#include <errno.h>
|
||||
#include <stdarg.h>
|
||||
#include <stdio.h>
|
||||
#include <stdlib.h>
|
||||
#include <string.h>
|
||||
#include <util.h>
|
||||
|
||||
/* Park/Miller RNG w/out MASKING, so as to be like f90s version */
|
||||
#define IA 16807
|
||||
#define IM 2147483647
|
||||
#define AM (1.0 / IM)
|
||||
#define IQ 127773
|
||||
#define IR 2836
|
||||
#define IA 16807
|
||||
#define IM 2147483647
|
||||
#define AM (1.0/IM)
|
||||
#define IQ 127773
|
||||
#define IR 2836
|
||||
#define MASK 123459876
|
||||
|
||||
double myrandom(int* seed)
|
||||
{
|
||||
int k = (*seed) / IQ;
|
||||
double myrandom(int* seed) {
|
||||
int k= (*seed) / IQ;
|
||||
double ans;
|
||||
|
||||
*seed = IA * (*seed - k * IQ) - IR * k;
|
||||
if (*seed < 0) *seed += IM;
|
||||
if(*seed < 0) *seed += IM;
|
||||
ans = AM * (*seed);
|
||||
return ans;
|
||||
}
|
||||
|
||||
void random_reset(int* seed, int ibase, double* coord)
|
||||
{
|
||||
void random_reset(int *seed, int ibase, double *coord) {
|
||||
int i;
|
||||
char* str = (char*)&ibase;
|
||||
int n = sizeof(int);
|
||||
char *str = (char *) &ibase;
|
||||
int n = sizeof(int);
|
||||
unsigned int hash = 0;
|
||||
|
||||
for (i = 0; i < n; i++) {
|
||||
@ -42,8 +41,8 @@ void random_reset(int* seed, int ibase, double* coord)
|
||||
hash ^= (hash >> 6);
|
||||
}
|
||||
|
||||
str = (char*)coord;
|
||||
n = 3 * sizeof(double);
|
||||
str = (char *) coord;
|
||||
n = 3 * sizeof(double);
|
||||
for (i = 0; i < n; i++) {
|
||||
hash += str[i];
|
||||
hash += (hash << 10);
|
||||
@ -62,59 +61,45 @@ void random_reset(int* seed, int ibase, double* coord)
|
||||
|
||||
// warm up the RNG
|
||||
|
||||
for (i = 0; i < 5; i++)
|
||||
myrandom(seed);
|
||||
// save = 0;
|
||||
for (i = 0; i < 5; i++) myrandom(seed);
|
||||
//save = 0;
|
||||
}
|
||||
|
||||
int str2ff(const char* string)
|
||||
{
|
||||
if (strncmp(string, "lj", 2) == 0) return FF_LJ;
|
||||
if (strncmp(string, "eam", 3) == 0) return FF_EAM;
|
||||
if (strncmp(string, "dem", 3) == 0) return FF_DEM;
|
||||
int str2ff(const char *string) {
|
||||
if(strncmp(string, "lj", 2) == 0) return FF_LJ;
|
||||
if(strncmp(string, "eam", 3) == 0) return FF_EAM;
|
||||
if(strncmp(string, "dem", 3) == 0) return FF_DEM;
|
||||
return -1;
|
||||
}
|
||||
|
||||
const char* ff2str(int ff)
|
||||
{
|
||||
if (ff == FF_LJ) {
|
||||
return "lj";
|
||||
}
|
||||
if (ff == FF_EAM) {
|
||||
return "eam";
|
||||
}
|
||||
if (ff == FF_DEM) {
|
||||
return "dem";
|
||||
}
|
||||
const char* ff2str(int ff) {
|
||||
if(ff == FF_LJ) { return "lj"; }
|
||||
if(ff == FF_EAM) { return "eam"; }
|
||||
if(ff == FF_DEM) { return "dem"; }
|
||||
return "invalid";
|
||||
}
|
||||
|
||||
int get_cuda_num_threads(void)
|
||||
{
|
||||
const char* num_threads_env = getenv("NUM_THREADS");
|
||||
int get_num_threads() {
|
||||
const char *num_threads_env = getenv("NUM_THREADS");
|
||||
return (num_threads_env == NULL) ? 32 : atoi(num_threads_env);
|
||||
}
|
||||
|
||||
void readline(char* line, FILE* fp)
|
||||
{
|
||||
if (fgets(line, MAXLINE, fp) == NULL) {
|
||||
if (errno != 0) {
|
||||
void readline(char *line, FILE *fp) {
|
||||
if(fgets(line, MAXLINE, fp) == NULL) {
|
||||
if(errno != 0) {
|
||||
perror("readline()");
|
||||
exit(-1);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
void debug_printf(const char* format, ...)
|
||||
{
|
||||
#ifdef DEBUG
|
||||
void debug_printf(const char *format, ...) {
|
||||
#ifdef DEBUG
|
||||
va_list arg;
|
||||
int ret;
|
||||
|
||||
va_start(arg, format);
|
||||
if ((vfprintf(stdout, format, arg)) < 0) {
|
||||
perror("debug_printf()");
|
||||
}
|
||||
if((vfprintf(stdout, format, arg)) < 0) { perror("debug_printf()"); }
|
||||
va_end(arg);
|
||||
#endif
|
||||
#endif
|
||||
}
|
||||
113
config.mk
113
config.mk
@ -1,23 +1,20 @@
|
||||
# Compiler tool chain (GCC/CLANG/ICC/ICX/ONEAPI/NVCC)
|
||||
TOOLCHAIN ?= CLANG
|
||||
# Instruction set for instrinsic kernels (NONE/SSE/AVX/AVX_FMA/AVX2/AVX512)
|
||||
ISA ?= ARM
|
||||
SIMD ?= NONE
|
||||
# Optimization scheme (verletlist/clusterpair/clusters_per_bin)
|
||||
OPT_SCHEME ?= verletlist
|
||||
# Compiler tag (GCC/CLANG/ICC/ICX/ONEAPI/NVCC)
|
||||
TAG ?= ICC
|
||||
# Instruction set (SSE/AVX/AVX_FMA/AVX2/AVX512)
|
||||
ISA ?= AVX512
|
||||
# Optimization scheme (lammps/gromacs/clusters_per_bin)
|
||||
OPT_SCHEME ?= gromacs
|
||||
# Enable likwid (true or false)
|
||||
ENABLE_LIKWID ?= false
|
||||
ENABLE_LIKWID ?= true
|
||||
# SP or DP
|
||||
DATA_TYPE ?= DP
|
||||
# AOS or SOA
|
||||
DATA_LAYOUT ?= AOS
|
||||
# Assembly syntax to generate (ATT/INTEL)
|
||||
ASM_SYNTAX ?= INTEL
|
||||
ASM_SYNTAX ?= ATT
|
||||
# Debug
|
||||
DEBUG ?= false
|
||||
|
||||
# Sort atoms when reneighboring (true or false)
|
||||
SORT_ATOMS ?= true
|
||||
# Explicitly store and load atom types (true or false)
|
||||
EXPLICIT_TYPES ?= false
|
||||
# Trace memory addresses for cache simulator (true or false)
|
||||
@ -29,7 +26,7 @@ COMPUTE_STATS ?= true
|
||||
|
||||
# Configurations for lammps optimization scheme
|
||||
# Use omp simd pragma when running with half neighbor-lists
|
||||
ENABLE_OMP_SIMD ?= false
|
||||
ENABLE_OMP_SIMD ?= true
|
||||
# Use kernel with explicit SIMD intrinsics
|
||||
USE_SIMD_KERNEL ?= false
|
||||
|
||||
@ -39,7 +36,7 @@ USE_REFERENCE_VERSION ?= false
|
||||
# Enable XTC output
|
||||
XTC_OUTPUT ?= false
|
||||
# Check if cj is local when decreasing reaction force
|
||||
HALF_NEIGHBOR_LISTS_CHECK_CJ ?= true
|
||||
HALF_NEIGHBOR_LISTS_CHECK_CJ ?= false
|
||||
|
||||
# Configurations for CUDA
|
||||
# Use CUDA host memory to optimize transfers
|
||||
@ -48,93 +45,3 @@ USE_CUDA_HOST_MEMORY ?= false
|
||||
#Feature options
|
||||
OPTIONS = -DALIGNMENT=64
|
||||
#OPTIONS += More options
|
||||
|
||||
#DO NOT EDIT BELOW
|
||||
ifeq ($(strip $(DATA_LAYOUT)),AOS)
|
||||
DEFINES += -DAOS
|
||||
endif
|
||||
ifeq ($(strip $(DATA_TYPE)),SP)
|
||||
DEFINES += -DPRECISION=1
|
||||
else
|
||||
DEFINES += -DPRECISION=2
|
||||
endif
|
||||
|
||||
ifneq ($(ASM_SYNTAX), ATT)
|
||||
ASFLAGS += -masm=intel
|
||||
endif
|
||||
|
||||
ifeq ($(strip $(SORT_ATOMS)),true)
|
||||
DEFINES += -DSORT_ATOMS
|
||||
endif
|
||||
|
||||
ifeq ($(strip $(EXPLICIT_TYPES)),true)
|
||||
DEFINES += -DEXPLICIT_TYPES
|
||||
endif
|
||||
|
||||
ifeq ($(strip $(MEM_TRACER)),true)
|
||||
DEFINES += -DMEM_TRACER
|
||||
endif
|
||||
|
||||
ifeq ($(strip $(INDEX_TRACER)),true)
|
||||
DEFINES += -DINDEX_TRACER
|
||||
endif
|
||||
|
||||
ifeq ($(strip $(COMPUTE_STATS)),true)
|
||||
DEFINES += -DCOMPUTE_STATS
|
||||
endif
|
||||
|
||||
ifeq ($(strip $(XTC_OUTPUT)),true)
|
||||
DEFINES += -DXTC_OUTPUT
|
||||
endif
|
||||
|
||||
ifeq ($(strip $(USE_REFERENCE_VERSION)),true)
|
||||
DEFINES += -DUSE_REFERENCE_VERSION
|
||||
endif
|
||||
|
||||
ifeq ($(strip $(HALF_NEIGHBOR_LISTS_CHECK_CJ)),true)
|
||||
DEFINES += -DHALF_NEIGHBOR_LISTS_CHECK_CJ
|
||||
endif
|
||||
|
||||
ifeq ($(strip $(DEBUG)),true)
|
||||
DEFINES += -DDEBUG
|
||||
endif
|
||||
|
||||
ifneq ($(VECTOR_WIDTH),)
|
||||
DEFINES += -DVECTOR_WIDTH=$(VECTOR_WIDTH)
|
||||
endif
|
||||
|
||||
ifeq ($(strip $(__SIMD_KERNEL__)),true)
|
||||
DEFINES += -D__SIMD_KERNEL__
|
||||
endif
|
||||
|
||||
ifeq ($(strip $(__SSE__)),true)
|
||||
DEFINES += -D__ISA_SSE__
|
||||
endif
|
||||
|
||||
ifeq ($(strip $(__ISA_AVX__)),true)
|
||||
DEFINES += -D__ISA_AVX__
|
||||
endif
|
||||
|
||||
ifeq ($(strip $(__ISA_AVX_FMA__)),true)
|
||||
DEFINES += -D__ISA_AVX_FMA__
|
||||
endif
|
||||
|
||||
ifeq ($(strip $(__ISA_AVX2__)),true)
|
||||
DEFINES += -D__ISA_AVX2__
|
||||
endif
|
||||
|
||||
ifeq ($(strip $(__ISA_AVX512__)),true)
|
||||
DEFINES += -D__ISA_AVX512__
|
||||
endif
|
||||
|
||||
ifeq ($(strip $(ENABLE_OMP_SIMD)),true)
|
||||
DEFINES += -DENABLE_OMP_SIMD
|
||||
endif
|
||||
|
||||
ifeq ($(strip $(OPT_SCHEME)),verletlist)
|
||||
OPT_TAG = VL
|
||||
endif
|
||||
|
||||
ifneq ($(strip $(SIMD)),NONE)
|
||||
TOOLCHAIN = $(TOOLCHAIN)-$(ISA)-$(SIMD)
|
||||
endif
|
||||
|
||||
@ -6,7 +6,7 @@ dt 0.001
|
||||
temp 80
|
||||
x_out_freq 500
|
||||
v_out_freq 5
|
||||
cutforce 1.8
|
||||
skin 0.1
|
||||
cutforce 0.9
|
||||
skin 0.05
|
||||
reneigh_every 100
|
||||
nstat 125000
|
||||
|
||||
BIN
figures/features-v3.png
Normal file
BIN
figures/features-v3.png
Normal file
Binary file not shown.
|
After Width: | Height: | Size: 273 KiB |
BIN
figures/gather_bench.png
Normal file
BIN
figures/gather_bench.png
Normal file
Binary file not shown.
|
After Width: | Height: | Size: 98 KiB |
523
figures/gather_bench.svg
Normal file
523
figures/gather_bench.svg
Normal file
@ -0,0 +1,523 @@
|
||||
<?xml version="1.0" encoding="UTF-8" standalone="no"?>
|
||||
<!-- Created with Inkscape (http://www.inkscape.org/) -->
|
||||
|
||||
<svg
|
||||
width="297mm"
|
||||
height="210mm"
|
||||
viewBox="0 0 297 210"
|
||||
version="1.1"
|
||||
id="svg5"
|
||||
inkscape:version="1.1.2 (0a00cf5339, 2022-02-04)"
|
||||
sodipodi:docname="gather_bench.svg"
|
||||
xmlns:inkscape="http://www.inkscape.org/namespaces/inkscape"
|
||||
xmlns:sodipodi="http://sodipodi.sourceforge.net/DTD/sodipodi-0.dtd"
|
||||
xmlns:xlink="http://www.w3.org/1999/xlink"
|
||||
xmlns="http://www.w3.org/2000/svg"
|
||||
xmlns:svg="http://www.w3.org/2000/svg">
|
||||
<sodipodi:namedview
|
||||
id="namedview7"
|
||||
pagecolor="#ffffff"
|
||||
bordercolor="#666666"
|
||||
borderopacity="1.0"
|
||||
inkscape:pageshadow="2"
|
||||
inkscape:pageopacity="0.0"
|
||||
inkscape:pagecheckerboard="0"
|
||||
inkscape:document-units="mm"
|
||||
showgrid="false"
|
||||
inkscape:zoom="0.73508842"
|
||||
inkscape:cx="551.63432"
|
||||
inkscape:cy="348.25743"
|
||||
inkscape:window-width="1920"
|
||||
inkscape:window-height="1011"
|
||||
inkscape:window-x="0"
|
||||
inkscape:window-y="165"
|
||||
inkscape:window-maximized="1"
|
||||
inkscape:current-layer="layer1" />
|
||||
<defs
|
||||
id="defs2">
|
||||
<rect
|
||||
x="144.01516"
|
||||
y="304.36604"
|
||||
width="248.99777"
|
||||
height="100.91557"
|
||||
id="rect79475" />
|
||||
<rect
|
||||
x="309.01869"
|
||||
y="43.698615"
|
||||
width="552.19421"
|
||||
height="71.390348"
|
||||
id="rect65238" />
|
||||
<rect
|
||||
x="762.55856"
|
||||
y="341.3838"
|
||||
width="277.62756"
|
||||
height="105.0235"
|
||||
id="rect47632" />
|
||||
<linearGradient
|
||||
inkscape:collect="always"
|
||||
id="linearGradient40704">
|
||||
<stop
|
||||
style="stop-color:#ccffaa;stop-opacity:1;"
|
||||
offset="0"
|
||||
id="stop40700" />
|
||||
<stop
|
||||
style="stop-color:#ccffaa;stop-opacity:0;"
|
||||
offset="1"
|
||||
id="stop40702" />
|
||||
</linearGradient>
|
||||
<marker
|
||||
style="overflow:visible;"
|
||||
id="Arrow2Mend"
|
||||
refX="0.0"
|
||||
refY="0.0"
|
||||
orient="auto"
|
||||
inkscape:stockid="Arrow2Mend"
|
||||
inkscape:isstock="true">
|
||||
<path
|
||||
transform="scale(0.6) rotate(180) translate(0,0)"
|
||||
d="M 8.7185878,4.0337352 L -2.2072895,0.016013256 L 8.7185884,-4.0017078 C 6.9730900,-1.6296469 6.9831476,1.6157441 8.7185878,4.0337352 z "
|
||||
style="stroke:context-stroke;fill-rule:evenodd;fill:context-stroke;stroke-width:0.62500000;stroke-linejoin:round;"
|
||||
id="path39486" />
|
||||
</marker>
|
||||
<marker
|
||||
style="overflow:visible;"
|
||||
id="Arrow1Mend"
|
||||
refX="0.0"
|
||||
refY="0.0"
|
||||
orient="auto"
|
||||
inkscape:stockid="Arrow1Mend"
|
||||
inkscape:isstock="true">
|
||||
<path
|
||||
transform="scale(0.4) rotate(180) translate(10,0)"
|
||||
style="fill-rule:evenodd;fill:context-stroke;stroke:context-stroke;stroke-width:1.0pt;"
|
||||
d="M 0.0,0.0 L 5.0,-5.0 L -12.5,0.0 L 5.0,5.0 L 0.0,0.0 z "
|
||||
id="path39468" />
|
||||
</marker>
|
||||
<marker
|
||||
style="overflow:visible;"
|
||||
id="Arrow1Lend"
|
||||
refX="0.0"
|
||||
refY="0.0"
|
||||
orient="auto"
|
||||
inkscape:stockid="Arrow1Lend"
|
||||
inkscape:isstock="true">
|
||||
<path
|
||||
transform="scale(0.8) rotate(180) translate(12.5,0)"
|
||||
style="fill-rule:evenodd;fill:context-stroke;stroke:context-stroke;stroke-width:1.0pt;"
|
||||
d="M 0.0,0.0 L 5.0,-5.0 L -12.5,0.0 L 5.0,5.0 L 0.0,0.0 z "
|
||||
id="path39462" />
|
||||
</marker>
|
||||
<rect
|
||||
x="707.09731"
|
||||
y="616.36746"
|
||||
width="407.71288"
|
||||
height="417.08306"
|
||||
id="rect24254" />
|
||||
<rect
|
||||
x="47.404365"
|
||||
y="100.3268"
|
||||
width="398.49855"
|
||||
height="110.16514"
|
||||
id="rect5050" />
|
||||
<rect
|
||||
x="47.404366"
|
||||
y="100.3268"
|
||||
width="398.49854"
|
||||
height="110.16514"
|
||||
id="rect5050-3" />
|
||||
<rect
|
||||
x="47.404366"
|
||||
y="100.3268"
|
||||
width="398.49854"
|
||||
height="110.16514"
|
||||
id="rect5050-3-5" />
|
||||
<rect
|
||||
x="47.404366"
|
||||
y="100.3268"
|
||||
width="398.49854"
|
||||
height="110.16514"
|
||||
id="rect5050-3-5-6" />
|
||||
<rect
|
||||
x="47.404366"
|
||||
y="100.3268"
|
||||
width="398.49854"
|
||||
height="110.16514"
|
||||
id="rect5050-3-5-6-1" />
|
||||
<rect
|
||||
x="47.404366"
|
||||
y="100.3268"
|
||||
width="398.49854"
|
||||
height="110.16514"
|
||||
id="rect5050-0" />
|
||||
<rect
|
||||
x="47.404366"
|
||||
y="100.3268"
|
||||
width="398.49854"
|
||||
height="110.16514"
|
||||
id="rect5050-0-6" />
|
||||
<rect
|
||||
x="47.404366"
|
||||
y="100.3268"
|
||||
width="398.49854"
|
||||
height="110.16514"
|
||||
id="rect5050-0-6-2" />
|
||||
<rect
|
||||
x="47.404366"
|
||||
y="100.3268"
|
||||
width="398.49854"
|
||||
height="110.16514"
|
||||
id="rect5050-0-6-2-8" />
|
||||
<marker
|
||||
style="overflow:visible"
|
||||
id="Arrow2Mend-2"
|
||||
refX="0"
|
||||
refY="0"
|
||||
orient="auto"
|
||||
inkscape:stockid="Arrow2Mend"
|
||||
inkscape:isstock="true">
|
||||
<path
|
||||
transform="scale(-0.6)"
|
||||
d="M 8.7185878,4.0337352 -2.2072895,0.01601326 8.7185884,-4.0017078 c -1.7454984,2.3720609 -1.7354408,5.6174519 -6e-7,8.035443 z"
|
||||
style="fill:context-stroke;fill-rule:evenodd;stroke:context-stroke;stroke-width:0.625;stroke-linejoin:round"
|
||||
id="path39486-3" />
|
||||
</marker>
|
||||
<marker
|
||||
style="overflow:visible"
|
||||
id="Arrow2Mend-2-5"
|
||||
refX="0"
|
||||
refY="0"
|
||||
orient="auto"
|
||||
inkscape:stockid="Arrow2Mend"
|
||||
inkscape:isstock="true">
|
||||
<path
|
||||
transform="scale(-0.6)"
|
||||
d="M 8.7185878,4.0337352 -2.2072895,0.01601326 8.7185884,-4.0017078 c -1.7454984,2.3720609 -1.7354408,5.6174519 -6e-7,8.035443 z"
|
||||
style="fill:context-stroke;fill-rule:evenodd;stroke:context-stroke;stroke-width:0.625;stroke-linejoin:round"
|
||||
id="path39486-3-9" />
|
||||
</marker>
|
||||
<marker
|
||||
style="overflow:visible"
|
||||
id="Arrow2Mend-2-5-2"
|
||||
refX="0"
|
||||
refY="0"
|
||||
orient="auto"
|
||||
inkscape:stockid="Arrow2Mend"
|
||||
inkscape:isstock="true">
|
||||
<path
|
||||
transform="scale(-0.6)"
|
||||
d="M 8.7185878,4.0337352 -2.2072895,0.01601326 8.7185884,-4.0017078 c -1.7454984,2.3720609 -1.7354408,5.6174519 -6e-7,8.035443 z"
|
||||
style="fill:context-stroke;fill-rule:evenodd;stroke:context-stroke;stroke-width:0.625;stroke-linejoin:round"
|
||||
id="path39486-3-9-8" />
|
||||
</marker>
|
||||
<linearGradient
|
||||
inkscape:collect="always"
|
||||
xlink:href="#linearGradient40704"
|
||||
id="linearGradient40706"
|
||||
x1="324.58157"
|
||||
y1="127.35331"
|
||||
x2="363.61096"
|
||||
y2="98.957848"
|
||||
gradientUnits="userSpaceOnUse" />
|
||||
<rect
|
||||
x="47.404366"
|
||||
y="100.3268"
|
||||
width="398.49854"
|
||||
height="110.16514"
|
||||
id="rect5050-3-5-6-1-7" />
|
||||
<rect
|
||||
x="309.01868"
|
||||
y="43.698616"
|
||||
width="552.19421"
|
||||
height="71.39035"
|
||||
id="rect65238-1" />
|
||||
</defs>
|
||||
<g
|
||||
inkscape:label="Layer 1"
|
||||
inkscape:groupmode="layer"
|
||||
id="layer1">
|
||||
<rect
|
||||
style="fill:#d5d5ff;stroke:#091600;stroke-width:1.5;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0"
|
||||
id="rect55834"
|
||||
width="250.31726"
|
||||
height="74.676537"
|
||||
x="25.257824"
|
||||
y="97.277718" />
|
||||
<rect
|
||||
style="fill:#d5f6ff;stroke:#091600;stroke-width:1.5;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0"
|
||||
id="rect55832"
|
||||
width="250.35208"
|
||||
height="64.461151"
|
||||
x="25.256891"
|
||||
y="32.817505" />
|
||||
<rect
|
||||
style="fill:#ccffaa;stroke:#091600;stroke-width:1.31891"
|
||||
id="rect6462"
|
||||
width="82.385742"
|
||||
height="20.525751"
|
||||
x="28.355024"
|
||||
y="48.740646" />
|
||||
<text
|
||||
xml:space="preserve"
|
||||
transform="matrix(0.26458333,0,0,0.26458333,17.244577,26.206534)"
|
||||
id="text5048"
|
||||
style="font-style:normal;font-weight:normal;font-size:40px;line-height:1.25;font-family:sans-serif;letter-spacing:0px;word-spacing:0px;white-space:pre;shape-inside:url(#rect5050);fill:#000000;fill-opacity:1;stroke:none"><tspan
|
||||
x="47.404297"
|
||||
y="135.7168"
|
||||
id="tspan82948"><tspan
|
||||
style="font-weight:bold;-inkscape-font-specification:'sans-serif Bold'"
|
||||
id="tspan82946">gather-bench</tspan></tspan></text>
|
||||
<rect
|
||||
style="fill:#de87aa;stroke:#091600;stroke-width:1.5;stroke-miterlimit:4;stroke-dasharray:none"
|
||||
id="rect6462-9"
|
||||
width="18.764017"
|
||||
height="20.965076"
|
||||
x="39.518955"
|
||||
y="140.726" />
|
||||
<text
|
||||
xml:space="preserve"
|
||||
transform="matrix(0.33667319,0,0,0.33667319,25.589293,109.42998)"
|
||||
id="text5048-3"
|
||||
style="font-style:normal;font-weight:normal;font-size:40px;line-height:1.25;font-family:sans-serif;letter-spacing:0px;word-spacing:0px;white-space:pre;shape-inside:url(#rect5050-0);fill:#000000;fill-opacity:1;stroke:none"><tspan
|
||||
x="47.404297"
|
||||
y="135.7168"
|
||||
id="tspan82950">L1</tspan></text>
|
||||
<rect
|
||||
style="fill:#de87aa;stroke:#091600;stroke-width:1.5;stroke-miterlimit:4;stroke-dasharray:none"
|
||||
id="rect6462-9-0"
|
||||
width="21.653919"
|
||||
height="24.193966"
|
||||
x="97.687294"
|
||||
y="138.51564" />
|
||||
<text
|
||||
xml:space="preserve"
|
||||
transform="matrix(0.3885252,0,0,0.3885252,81.212654,102.39964)"
|
||||
id="text5048-3-6"
|
||||
style="font-style:normal;font-weight:normal;font-size:40px;line-height:1.25;font-family:sans-serif;letter-spacing:0px;word-spacing:0px;white-space:pre;shape-inside:url(#rect5050-0-6);fill:#000000;fill-opacity:1;stroke:none"><tspan
|
||||
x="47.404297"
|
||||
y="135.7168"
|
||||
id="tspan82952">L2</tspan></text>
|
||||
<rect
|
||||
style="fill:#de87aa;stroke:#091600;stroke-width:1.5;stroke-miterlimit:4;stroke-dasharray:none"
|
||||
id="rect6462-9-0-6"
|
||||
width="27.217058"
|
||||
height="30.409672"
|
||||
x="149.19933"
|
||||
y="134.83977" />
|
||||
<text
|
||||
xml:space="preserve"
|
||||
transform="matrix(0.48834178,0,0,0.48834178,128.49215,89.445174)"
|
||||
id="text5048-3-6-1"
|
||||
style="font-style:normal;font-weight:normal;font-size:40px;line-height:1.25;font-family:sans-serif;letter-spacing:0px;word-spacing:0px;white-space:pre;shape-inside:url(#rect5050-0-6-2);fill:#000000;fill-opacity:1;stroke:none"><tspan
|
||||
x="47.404297"
|
||||
y="135.7168"
|
||||
id="tspan82954">L3</tspan></text>
|
||||
<rect
|
||||
style="fill:#eeaaff;stroke:#091600;stroke-width:1.5;stroke-miterlimit:4;stroke-dasharray:none"
|
||||
id="rect6462-9-0-6-7"
|
||||
width="61.032539"
|
||||
height="29.96501"
|
||||
x="204.01265"
|
||||
y="135.61238" />
|
||||
<text
|
||||
xml:space="preserve"
|
||||
transform="matrix(0.48834178,0,0,0.48834178,182.37007,89.995434)"
|
||||
id="text5048-3-6-1-9"
|
||||
style="font-style:normal;font-weight:normal;font-size:40px;line-height:1.25;font-family:sans-serif;letter-spacing:0px;word-spacing:0px;white-space:pre;shape-inside:url(#rect5050-0-6-2-8);fill:#000000;fill-opacity:1;stroke:none"><tspan
|
||||
x="47.404297"
|
||||
y="135.7168"
|
||||
id="tspan82956">DRAM</tspan></text>
|
||||
<rect
|
||||
style="fill:#ffccaa;stroke:#091600;stroke-width:1.10636"
|
||||
id="rect6462-6"
|
||||
width="74.980759"
|
||||
height="15.869514"
|
||||
x="126.09525"
|
||||
y="38.773243" />
|
||||
<text
|
||||
xml:space="preserve"
|
||||
transform="matrix(0.26458333,0,0,0.26458333,115.65481,14.295323)"
|
||||
id="text5048-7"
|
||||
style="font-style:normal;font-weight:normal;font-size:40px;line-height:1.25;font-family:sans-serif;letter-spacing:0px;word-spacing:0px;white-space:pre;shape-inside:url(#rect5050-3);fill:#000000;fill-opacity:1;stroke:none"><tspan
|
||||
x="47.404297"
|
||||
y="135.7168"
|
||||
id="tspan82958">Single gather</tspan></text>
|
||||
<rect
|
||||
style="fill:#ffccaa;stroke:#091600;stroke-width:1.03971"
|
||||
id="rect6462-6-3"
|
||||
width="66.071701"
|
||||
height="15.904838"
|
||||
x="126.90776"
|
||||
y="63.642746" />
|
||||
<text
|
||||
xml:space="preserve"
|
||||
transform="matrix(0.26458333,0,0,0.26458333,116.63325,39.114393)"
|
||||
id="text5048-7-5"
|
||||
style="font-style:normal;font-weight:normal;font-size:40px;line-height:1.25;font-family:sans-serif;letter-spacing:0px;word-spacing:0px;white-space:pre;shape-inside:url(#rect5050-3-5);fill:#000000;fill-opacity:1;stroke:none"><tspan
|
||||
x="47.404297"
|
||||
y="135.7168"
|
||||
id="tspan82960">MD gathers</tspan></text>
|
||||
<rect
|
||||
style="fill:#afe9dd;stroke:#091600;stroke-width:1.02848"
|
||||
id="rect6462-6-3-2"
|
||||
width="64.479698"
|
||||
height="15.947394"
|
||||
x="206.65364"
|
||||
y="52.98967" />
|
||||
<text
|
||||
xml:space="preserve"
|
||||
transform="matrix(0.26458333,0,0,0.26458333,196.01512,28.482594)"
|
||||
id="text5048-7-5-9"
|
||||
style="font-style:normal;font-weight:normal;font-size:40px;line-height:1.25;font-family:sans-serif;letter-spacing:0px;word-spacing:0px;white-space:pre;shape-inside:url(#rect5050-3-5-6);fill:#000000;fill-opacity:1;stroke:none"><tspan
|
||||
x="47.404297"
|
||||
y="135.7168"
|
||||
id="tspan82962">Contiguous</tspan></text>
|
||||
<rect
|
||||
style="fill:#afe9dd;stroke:#091600;stroke-width:0.987323"
|
||||
id="rect6462-6-3-2-2"
|
||||
width="59.269382"
|
||||
height="15.988551"
|
||||
x="208.16559"
|
||||
y="76.856781" />
|
||||
<text
|
||||
xml:space="preserve"
|
||||
transform="matrix(0.26458333,0,0,0.26458333,197.58604,52.220445)"
|
||||
id="text5048-7-5-9-7"
|
||||
style="font-style:normal;font-weight:normal;font-size:40px;line-height:1.25;font-family:sans-serif;letter-spacing:0px;word-spacing:0px;white-space:pre;shape-inside:url(#rect5050-3-5-6-1);fill:#000000;fill-opacity:1;stroke:none"><tspan
|
||||
x="47.404297"
|
||||
y="135.7168"
|
||||
id="tspan82964">"Random"</tspan></text>
|
||||
<text
|
||||
xml:space="preserve"
|
||||
transform="scale(0.26458333)"
|
||||
id="text24252"
|
||||
style="fill:black;fill-opacity:1;stroke:none;font-family:sans-serif;font-style:normal;font-weight:normal;font-size:40px;line-height:1.25;letter-spacing:0px;word-spacing:0px;white-space:pre;shape-inside:url(#rect24254)" />
|
||||
<path
|
||||
style="fill:none;fill-rule:evenodd;stroke:#000000;stroke-width:1;stroke-linecap:butt;stroke-linejoin:miter;stroke-miterlimit:4;stroke-dasharray:none;stroke-opacity:1"
|
||||
d="M 193.10512,71.273276 206.30683,61.033513"
|
||||
id="path39049"
|
||||
inkscape:connector-type="polyline"
|
||||
inkscape:connector-curvature="0" />
|
||||
<path
|
||||
style="fill:none;fill-rule:evenodd;stroke:#000000;stroke-width:1;stroke-linecap:butt;stroke-linejoin:miter;stroke-miterlimit:4;stroke-dasharray:none;stroke-opacity:1"
|
||||
d="M 193.08841,71.196939 207.86207,84.43804"
|
||||
id="path39053"
|
||||
inkscape:connector-type="polyline"
|
||||
inkscape:connector-curvature="0" />
|
||||
<path
|
||||
style="fill:none;fill-rule:evenodd;stroke:#000000;stroke-width:1.39816;stroke-linecap:butt;stroke-linejoin:miter;stroke-miterlimit:4;stroke-dasharray:none;stroke-opacity:1"
|
||||
d="m 58.548229,151.24436 38.298093,0.25023"
|
||||
id="path39219"
|
||||
inkscape:connector-type="polyline"
|
||||
inkscape:connector-curvature="0" />
|
||||
<path
|
||||
style="fill:none;fill-rule:evenodd;stroke:#000000;stroke-width:1.24847;stroke-linecap:butt;stroke-linejoin:miter;stroke-miterlimit:4;stroke-dasharray:none;stroke-opacity:1"
|
||||
d="m 119.19252,150.09399 29.28333,0.26095"
|
||||
id="path39219-2"
|
||||
inkscape:connector-type="polyline"
|
||||
inkscape:connector-curvature="0" />
|
||||
<path
|
||||
style="fill:none;fill-rule:evenodd;stroke:#000000;stroke-width:1;stroke-linecap:butt;stroke-linejoin:miter;stroke-miterlimit:4;stroke-dasharray:none;stroke-opacity:1"
|
||||
d="m 177.02022,150.44367 26.36623,0.26095"
|
||||
id="path39219-2-0"
|
||||
inkscape:connector-type="polyline"
|
||||
inkscape:connector-curvature="0" />
|
||||
<path
|
||||
style="fill:none;fill-rule:evenodd;stroke:#000000;stroke-width:1;stroke-linecap:butt;stroke-linejoin:miter;stroke-miterlimit:4;stroke-dasharray:3, 1;stroke-dashoffset:0;stroke-opacity:1;marker-end:url(#Arrow2Mend)"
|
||||
d="m 48.145458,92.71788 -0.644819,47.57709"
|
||||
id="path39377"
|
||||
inkscape:connector-type="polyline"
|
||||
inkscape:connector-curvature="0" />
|
||||
<path
|
||||
style="display:inline;fill:none;fill-rule:evenodd;stroke:#000000;stroke-width:1;stroke-linecap:butt;stroke-linejoin:miter;stroke-miterlimit:4;stroke-dasharray:3, 1;stroke-dashoffset:0;stroke-opacity:1;marker-end:url(#Arrow2Mend-2)"
|
||||
d="M 48.121208,92.873762 106.60807,137.41946"
|
||||
id="path39377-7"
|
||||
inkscape:connector-type="polyline"
|
||||
inkscape:connector-curvature="0" />
|
||||
<path
|
||||
style="display:inline;fill:none;fill-rule:evenodd;stroke:#000000;stroke-width:1;stroke-linecap:butt;stroke-linejoin:miter;stroke-miterlimit:4;stroke-dasharray:3, 1;stroke-dashoffset:0;stroke-opacity:1;marker-end:url(#Arrow2Mend-2-5)"
|
||||
d="M 48.073928,92.825143 158.88023,133.04546"
|
||||
id="path39377-7-2"
|
||||
inkscape:connector-type="polyline"
|
||||
inkscape:connector-curvature="0" />
|
||||
<path
|
||||
style="display:inline;fill:none;fill-rule:evenodd;stroke:#000000;stroke-width:1;stroke-linecap:butt;stroke-linejoin:miter;stroke-miterlimit:4;stroke-dasharray:3, 1;stroke-dashoffset:0;stroke-opacity:1;marker-end:url(#Arrow2Mend-2-5-2)"
|
||||
d="M 48.051946,92.813593 233.0959,134.16596"
|
||||
id="path39377-7-2-9"
|
||||
inkscape:connector-type="polyline"
|
||||
inkscape:connector-curvature="0" />
|
||||
<rect
|
||||
style="fill:#e9afaf;stroke:#091600;stroke-width:1.34518"
|
||||
id="rect6462-6-3-2-2-3"
|
||||
width="65.880661"
|
||||
height="26.700579"
|
||||
x="38.104012"
|
||||
y="80.530182" />
|
||||
<path
|
||||
style="display:inline;fill:none;fill-rule:evenodd;stroke:#000000;stroke-width:0.264583px;stroke-linecap:butt;stroke-linejoin:miter;stroke-opacity:1"
|
||||
d="m 77.365612,69.678744 h 2e-6"
|
||||
id="path39808"
|
||||
inkscape:connector-type="polyline"
|
||||
inkscape:connector-curvature="0" />
|
||||
<path
|
||||
style="display:inline;fill:none;fill-rule:evenodd;stroke:#000000;stroke-width:1;stroke-linecap:butt;stroke-linejoin:miter;stroke-miterlimit:4;stroke-dasharray:none;stroke-opacity:1"
|
||||
d="m 111.64767,59.183009 6.84466,0.03069"
|
||||
id="path41004"
|
||||
inkscape:connector-type="polyline"
|
||||
inkscape:connector-curvature="0" />
|
||||
<path
|
||||
style="display:inline;fill:none;fill-rule:evenodd;stroke:#000000;stroke-width:1;stroke-linecap:butt;stroke-linejoin:miter;stroke-miterlimit:4;stroke-dasharray:none;stroke-opacity:1"
|
||||
d="m 119.03378,47.056357 -0.58704,25.198541"
|
||||
id="path41006"
|
||||
inkscape:connector-type="polyline"
|
||||
inkscape:connector-curvature="0" />
|
||||
<path
|
||||
style="display:inline;fill:none;fill-rule:evenodd;stroke:#000000;stroke-width:1.02423;stroke-linecap:butt;stroke-linejoin:miter;stroke-miterlimit:4;stroke-dasharray:none;stroke-opacity:1"
|
||||
d="m 118.07503,72.254897 7.94998,-0.05784"
|
||||
id="path41008"
|
||||
inkscape:connector-type="polyline"
|
||||
inkscape:connector-curvature="0" />
|
||||
<path
|
||||
style="fill:none;fill-rule:evenodd;stroke:#000000;stroke-width:0.882836;stroke-linecap:butt;stroke-linejoin:miter;stroke-miterlimit:4;stroke-dasharray:none;stroke-opacity:1"
|
||||
d="m 118.26666,47.054814 7.69322,0.173925"
|
||||
id="path41112"
|
||||
inkscape:connector-type="polyline"
|
||||
inkscape:connector-curvature="0" />
|
||||
<path
|
||||
style="fill:none;fill-rule:evenodd;stroke:#000000;stroke-width:1;stroke-linecap:butt;stroke-linejoin:miter;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1"
|
||||
d="M 68.213642,69.068864 67.910274,80.302728"
|
||||
id="path55728"
|
||||
inkscape:connector-type="polyline"
|
||||
inkscape:connector-curvature="0" />
|
||||
<text
|
||||
xml:space="preserve"
|
||||
transform="matrix(0.26458333,0,0,0.26458333,-1.3782637,4.0412367)"
|
||||
id="text65236"
|
||||
style="font-style:normal;font-weight:normal;font-size:53.3333px;line-height:1.25;font-family:sans-serif;letter-spacing:0px;word-spacing:0px;white-space:pre;shape-inside:url(#rect65238);fill:#000000;fill-opacity:1;stroke:none"><tspan
|
||||
x="309.01953"
|
||||
y="90.886691"
|
||||
id="tspan82968"><tspan
|
||||
style="font-weight:bold;-inkscape-font-specification:'sans-serif Bold'"
|
||||
id="tspan82966">Application Level</tspan></tspan></text>
|
||||
<text
|
||||
xml:space="preserve"
|
||||
transform="matrix(0.26458333,0,0,0.26458333,2.7015103,160.71919)"
|
||||
id="text65236-2"
|
||||
style="font-style:normal;font-weight:normal;font-size:53.3333px;line-height:1.25;font-family:sans-serif;letter-spacing:0px;word-spacing:0px;white-space:pre;shape-inside:url(#rect65238-1);fill:#000000;fill-opacity:1;stroke:none"><tspan
|
||||
x="309.01953"
|
||||
y="90.886691"
|
||||
id="tspan82972"><tspan
|
||||
style="font-weight:bold;-inkscape-font-specification:'sans-serif Bold'"
|
||||
id="tspan82970">Hardware Level</tspan></tspan></text>
|
||||
<text
|
||||
xml:space="preserve"
|
||||
transform="matrix(0.26458333,0,0,0.26458333,2.3490396,0.57331532)"
|
||||
id="text79473"
|
||||
style="font-style:normal;font-weight:normal;font-size:40px;line-height:1.25;font-family:sans-serif;letter-spacing:0px;word-spacing:0px;white-space:pre;shape-inside:url(#rect79475);fill:#000000;fill-opacity:1;stroke:none"><tspan
|
||||
x="144.01562"
|
||||
y="339.75586"
|
||||
id="tspan82974">vgather </tspan><tspan
|
||||
x="144.01562"
|
||||
y="389.75586"
|
||||
id="tspan82976">instructions</tspan></text>
|
||||
</g>
|
||||
</svg>
|
||||
|
After Width: | Height: | Size: 21 KiB |
BIN
figures/gromacs_mxn_v2.pdf
Normal file
BIN
figures/gromacs_mxn_v2.pdf
Normal file
Binary file not shown.
BIN
figures/gromacs_mxn_v2.png
Normal file
BIN
figures/gromacs_mxn_v2.png
Normal file
Binary file not shown.
|
After Width: | Height: | Size: 128 KiB |
BIN
figures/stub_new_v3.pdf
Normal file
BIN
figures/stub_new_v3.pdf
Normal file
Binary file not shown.
BIN
figures/stub_new_v3.png
Normal file
BIN
figures/stub_new_v3.png
Normal file
Binary file not shown.
|
After Width: | Height: | Size: 52 KiB |
BIN
figures/verlet_v2.pdf
Normal file
BIN
figures/verlet_v2.pdf
Normal file
Binary file not shown.
BIN
figures/verlet_v2.png
Normal file
BIN
figures/verlet_v2.png
Normal file
Binary file not shown.
|
After Width: | Height: | Size: 62 KiB |
1
gather-bench
Submodule
1
gather-bench
Submodule
@ -0,0 +1 @@
|
||||
Subproject commit 2f654cb043359197be07e0fa362324dab8899a33
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -45,7 +45,7 @@ static inline void gmx_load_simd_4xn_interactions(
|
||||
double computeForceLJ_ref(Parameter *param, Atom *atom, Neighbor *neighbor, Stats *stats) {
|
||||
DEBUG_MESSAGE("computeForceLJ begin\n");
|
||||
int Nlocal = atom->Nlocal;
|
||||
int *neighs;
|
||||
NeighborCluster* neighs;
|
||||
MD_FLOAT cutforcesq = param->cutforce * param->cutforce;
|
||||
MD_FLOAT sigma6 = param->sigma6;
|
||||
MD_FLOAT epsilon = param->epsilon;
|
||||
@ -66,7 +66,7 @@ double computeForceLJ_ref(Parameter *param, Atom *atom, Neighbor *neighbor, Stat
|
||||
{
|
||||
LIKWID_MARKER_START("force");
|
||||
|
||||
#pragma omp for schedule(runtime)
|
||||
#pragma omp for
|
||||
for(int ci = 0; ci < atom->Nclusters_local; ci++) {
|
||||
int ci_cj0 = CJ0_FROM_CI(ci);
|
||||
int ci_cj1 = CJ1_FROM_CI(ci);
|
||||
@ -77,7 +77,7 @@ double computeForceLJ_ref(Parameter *param, Atom *atom, Neighbor *neighbor, Stat
|
||||
int numneighs = neighbor->numneigh[ci];
|
||||
|
||||
for(int k = 0; k < numneighs; k++) {
|
||||
int cj = neighs[k];
|
||||
int cj = neighs[k].cj;
|
||||
int cj_vec_base = CJ_VECTOR_BASE_INDEX(cj);
|
||||
int any = 0;
|
||||
MD_FLOAT *cj_x = &atom->cl_x[cj_vec_base];
|
||||
@ -158,7 +158,7 @@ double computeForceLJ_ref(Parameter *param, Atom *atom, Neighbor *neighbor, Stat
|
||||
double computeForceLJ_2xnn_half(Parameter *param, Atom *atom, Neighbor *neighbor, Stats *stats) {
|
||||
DEBUG_MESSAGE("computeForceLJ_2xnn begin\n");
|
||||
int Nlocal = atom->Nlocal;
|
||||
int *neighs;
|
||||
NeighborCluster* neighs;
|
||||
MD_FLOAT cutforcesq = param->cutforce * param->cutforce;
|
||||
MD_FLOAT sigma6 = param->sigma6;
|
||||
MD_FLOAT epsilon = param->epsilon;
|
||||
@ -213,7 +213,7 @@ double computeForceLJ_2xnn_half(Parameter *param, Atom *atom, Neighbor *neighbor
|
||||
#endif
|
||||
*/
|
||||
|
||||
#pragma omp for schedule(runtime)
|
||||
#pragma omp for
|
||||
for(int ci = 0; ci < atom->Nclusters_local; ci++) {
|
||||
int ci_cj0 = CJ0_FROM_CI(ci);
|
||||
#if CLUSTER_M > CLUSTER_N
|
||||
@ -240,9 +240,9 @@ double computeForceLJ_2xnn_half(Parameter *param, Atom *atom, Neighbor *neighbor
|
||||
MD_SIMD_FLOAT fiz2 = simd_zero();
|
||||
|
||||
for(int k = 0; k < numneighs_masked; k++) {
|
||||
int cj = neighs[k];
|
||||
int cj = neighs[k].cj;
|
||||
int cj_vec_base = CJ_VECTOR_BASE_INDEX(cj);
|
||||
//int imask = neighs_imask[k];
|
||||
//int imask = neighs[k].imask;
|
||||
MD_FLOAT *cj_x = &atom->cl_x[cj_vec_base];
|
||||
MD_FLOAT *cj_f = &atom->cl_f[cj_vec_base];
|
||||
//MD_SIMD_MASK interact0;
|
||||
@ -331,7 +331,7 @@ double computeForceLJ_2xnn_half(Parameter *param, Atom *atom, Neighbor *neighbor
|
||||
}
|
||||
|
||||
for(int k = numneighs_masked; k < numneighs; k++) {
|
||||
int cj = neighs[k];
|
||||
int cj = neighs[k].cj;
|
||||
int cj_vec_base = CJ_VECTOR_BASE_INDEX(cj);
|
||||
MD_FLOAT *cj_x = &atom->cl_x[cj_vec_base];
|
||||
MD_FLOAT *cj_f = &atom->cl_f[cj_vec_base];
|
||||
@ -401,7 +401,7 @@ double computeForceLJ_2xnn_half(Parameter *param, Atom *atom, Neighbor *neighbor
|
||||
double computeForceLJ_2xnn_full(Parameter *param, Atom *atom, Neighbor *neighbor, Stats *stats) {
|
||||
DEBUG_MESSAGE("computeForceLJ_2xnn begin\n");
|
||||
int Nlocal = atom->Nlocal;
|
||||
int *neighs;
|
||||
NeighborCluster* neighs;
|
||||
MD_FLOAT cutforcesq = param->cutforce * param->cutforce;
|
||||
MD_FLOAT sigma6 = param->sigma6;
|
||||
MD_FLOAT epsilon = param->epsilon;
|
||||
@ -427,7 +427,7 @@ double computeForceLJ_2xnn_full(Parameter *param, Atom *atom, Neighbor *neighbor
|
||||
{
|
||||
LIKWID_MARKER_START("force");
|
||||
|
||||
#pragma omp for schedule(runtime)
|
||||
#pragma omp for
|
||||
for(int ci = 0; ci < atom->Nclusters_local; ci++) {
|
||||
int ci_cj0 = CJ0_FROM_CI(ci);
|
||||
#if CLUSTER_M > CLUSTER_N
|
||||
@ -454,8 +454,9 @@ double computeForceLJ_2xnn_full(Parameter *param, Atom *atom, Neighbor *neighbor
|
||||
MD_SIMD_FLOAT fiz2 = simd_zero();
|
||||
|
||||
for(int k = 0; k < numneighs_masked; k++) {
|
||||
int cj = neighs[k];
|
||||
int cj = neighs[k].cj;
|
||||
int cj_vec_base = CJ_VECTOR_BASE_INDEX(cj);
|
||||
int imask = neighs[k].imask;
|
||||
MD_FLOAT *cj_x = &atom->cl_x[cj_vec_base];
|
||||
unsigned int mask0, mask1, mask2, mask3;
|
||||
|
||||
@ -506,7 +507,7 @@ double computeForceLJ_2xnn_full(Parameter *param, Atom *atom, Neighbor *neighbor
|
||||
}
|
||||
|
||||
for(int k = numneighs_masked; k < numneighs; k++) {
|
||||
int cj = neighs[k];
|
||||
int cj = neighs[k].cj;
|
||||
int cj_vec_base = CJ_VECTOR_BASE_INDEX(cj);
|
||||
MD_FLOAT *cj_x = &atom->cl_x[cj_vec_base];
|
||||
|
||||
@ -569,7 +570,7 @@ double computeForceLJ_2xnn(Parameter *param, Atom *atom, Neighbor *neighbor, Sta
|
||||
double computeForceLJ_4xn_half(Parameter *param, Atom *atom, Neighbor *neighbor, Stats *stats) {
|
||||
DEBUG_MESSAGE("computeForceLJ_4xn begin\n");
|
||||
int Nlocal = atom->Nlocal;
|
||||
int *neighs;
|
||||
NeighborCluster* neighs;
|
||||
MD_FLOAT cutforcesq = param->cutforce * param->cutforce;
|
||||
MD_FLOAT sigma6 = param->sigma6;
|
||||
MD_FLOAT epsilon = param->epsilon;
|
||||
@ -595,7 +596,7 @@ double computeForceLJ_4xn_half(Parameter *param, Atom *atom, Neighbor *neighbor,
|
||||
{
|
||||
LIKWID_MARKER_START("force");
|
||||
|
||||
#pragma omp for schedule(runtime)
|
||||
#pragma omp for
|
||||
for(int ci = 0; ci < atom->Nclusters_local; ci++) {
|
||||
int ci_cj0 = CJ0_FROM_CI(ci);
|
||||
#if CLUSTER_M > CLUSTER_N
|
||||
@ -634,8 +635,9 @@ double computeForceLJ_4xn_half(Parameter *param, Atom *atom, Neighbor *neighbor,
|
||||
MD_SIMD_FLOAT fiz3 = simd_zero();
|
||||
|
||||
for(int k = 0; k < numneighs_masked; k++) {
|
||||
int cj = neighs[k];
|
||||
int cj = neighs[k].cj;
|
||||
int cj_vec_base = CJ_VECTOR_BASE_INDEX(cj);
|
||||
int imask = neighs[k].imask;
|
||||
MD_FLOAT *cj_x = &atom->cl_x[cj_vec_base];
|
||||
MD_FLOAT *cj_f = &atom->cl_f[cj_vec_base];
|
||||
MD_SIMD_FLOAT xj_tmp = simd_load(&cj_x[CL_X_OFFSET]);
|
||||
@ -739,8 +741,9 @@ double computeForceLJ_4xn_half(Parameter *param, Atom *atom, Neighbor *neighbor,
|
||||
}
|
||||
|
||||
for(int k = numneighs_masked; k < numneighs; k++) {
|
||||
int cj = neighs[k];
|
||||
int cj = neighs[k].cj;
|
||||
int cj_vec_base = CJ_VECTOR_BASE_INDEX(cj);
|
||||
int imask = neighs[k].imask;
|
||||
MD_FLOAT *cj_x = &atom->cl_x[cj_vec_base];
|
||||
MD_FLOAT *cj_f = &atom->cl_f[cj_vec_base];
|
||||
MD_SIMD_FLOAT xj_tmp = simd_load(&cj_x[CL_X_OFFSET]);
|
||||
@ -843,7 +846,7 @@ double computeForceLJ_4xn_half(Parameter *param, Atom *atom, Neighbor *neighbor,
|
||||
double computeForceLJ_4xn_full(Parameter *param, Atom *atom, Neighbor *neighbor, Stats *stats) {
|
||||
DEBUG_MESSAGE("computeForceLJ_4xn begin\n");
|
||||
int Nlocal = atom->Nlocal;
|
||||
int *neighs;
|
||||
NeighborCluster* neighs;
|
||||
MD_FLOAT cutforcesq = param->cutforce * param->cutforce;
|
||||
MD_FLOAT sigma6 = param->sigma6;
|
||||
MD_FLOAT epsilon = param->epsilon;
|
||||
@ -869,7 +872,7 @@ double computeForceLJ_4xn_full(Parameter *param, Atom *atom, Neighbor *neighbor,
|
||||
{
|
||||
LIKWID_MARKER_START("force");
|
||||
|
||||
#pragma omp for schedule(runtime)
|
||||
#pragma omp for
|
||||
for(int ci = 0; ci < atom->Nclusters_local; ci++) {
|
||||
int ci_cj0 = CJ0_FROM_CI(ci);
|
||||
#if CLUSTER_M > CLUSTER_N
|
||||
@ -908,8 +911,9 @@ double computeForceLJ_4xn_full(Parameter *param, Atom *atom, Neighbor *neighbor,
|
||||
MD_SIMD_FLOAT fiz3 = simd_zero();
|
||||
|
||||
for(int k = 0; k < numneighs_masked; k++) {
|
||||
int cj = neighs[k];
|
||||
int cj = neighs[k].cj;
|
||||
int cj_vec_base = CJ_VECTOR_BASE_INDEX(cj);
|
||||
int imask = neighs[k].imask;
|
||||
MD_FLOAT *cj_x = &atom->cl_x[cj_vec_base];
|
||||
MD_SIMD_FLOAT xj_tmp = simd_load(&cj_x[CL_X_OFFSET]);
|
||||
MD_SIMD_FLOAT yj_tmp = simd_load(&cj_x[CL_Y_OFFSET]);
|
||||
@ -987,8 +991,9 @@ double computeForceLJ_4xn_full(Parameter *param, Atom *atom, Neighbor *neighbor,
|
||||
}
|
||||
|
||||
for(int k = numneighs_masked; k < numneighs; k++) {
|
||||
int cj = neighs[k];
|
||||
int cj = neighs[k].cj;
|
||||
int cj_vec_base = CJ_VECTOR_BASE_INDEX(cj);
|
||||
int imask = neighs[k].imask;
|
||||
MD_FLOAT *cj_x = &atom->cl_x[cj_vec_base];
|
||||
MD_SIMD_FLOAT xj_tmp = simd_load(&cj_x[CL_X_OFFSET]);
|
||||
MD_SIMD_FLOAT yj_tmp = simd_load(&cj_x[CL_Y_OFFSET]);
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -25,6 +25,11 @@
|
||||
#define NBNXN_INTERACTION_MASK_DIAG_J8_0 0xf0f8fcfeU
|
||||
#define NBNXN_INTERACTION_MASK_DIAG_J8_1 0x0080c0e0U
|
||||
|
||||
typedef struct {
|
||||
int cj;
|
||||
unsigned int imask;
|
||||
} NeighborCluster;
|
||||
|
||||
typedef struct {
|
||||
int every;
|
||||
int ncalls;
|
||||
@ -32,8 +37,7 @@ typedef struct {
|
||||
int* numneigh;
|
||||
int* numneigh_masked;
|
||||
int half_neigh;
|
||||
int* neighbors;
|
||||
unsigned int* neighbors_imask;
|
||||
NeighborCluster* neighbors;
|
||||
} Neighbor;
|
||||
|
||||
extern void initNeighbor(Neighbor*, Parameter*);
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -60,15 +60,18 @@ void init(Parameter *param) {
|
||||
param->eam_file = NULL;
|
||||
}
|
||||
|
||||
void createNeighbors(Atom *atom, Neighbor *neighbor, int pattern, int nneighs, int nreps, int masked) {
|
||||
// Show debug messages
|
||||
#define DEBUG(msg) printf(msg)
|
||||
// Do not show debug messages
|
||||
//#define DEBUG(msg)
|
||||
|
||||
|
||||
void createNeighbors(Atom *atom, Neighbor *neighbor, int pattern, int nneighs, int nreps) {
|
||||
const int maxneighs = nneighs * nreps;
|
||||
const int jfac = MAX(1, CLUSTER_N / CLUSTER_M);
|
||||
const int ncj = atom->Nclusters_local / jfac;
|
||||
const unsigned int imask = NBNXN_INTERACTION_MASK_ALL;
|
||||
neighbor->numneigh = (int*) malloc(atom->Nclusters_max * sizeof(int));
|
||||
neighbor->numneigh_masked = (int*) malloc(atom->Nclusters_max * sizeof(int));
|
||||
neighbor->neighbors = (int*) malloc(atom->Nclusters_max * maxneighs * sizeof(int));
|
||||
neighbor->neighbors_imask = (unsigned int*) malloc(atom->Nclusters_max * maxneighs * sizeof(unsigned int));
|
||||
|
||||
if(pattern == P_RAND && ncj <= nneighs) {
|
||||
fprintf(stderr, "Error: P_RAND: Number of j-clusters should be higher than number of j-cluster neighbors per i-cluster!\n");
|
||||
@ -77,7 +80,6 @@ void createNeighbors(Atom *atom, Neighbor *neighbor, int pattern, int nneighs, i
|
||||
|
||||
for(int ci = 0; ci < atom->Nclusters_local; ci++) {
|
||||
int *neighptr = &(neighbor->neighbors[ci * neighbor->maxneighs]);
|
||||
unsigned int *neighptr_imask = &(neighbor->neighbors_imask[ci * neighbor->maxneighs]);
|
||||
int j = (pattern == P_SEQ) ? CJ0_FROM_CI(ci) : 0;
|
||||
int m = (pattern == P_SEQ) ? ncj : nneighs;
|
||||
int k = 0;
|
||||
@ -88,7 +90,6 @@ void createNeighbors(Atom *atom, Neighbor *neighbor, int pattern, int nneighs, i
|
||||
do {
|
||||
int cj = rand() % ncj;
|
||||
neighptr[k] = cj;
|
||||
neighptr_imask[k] = imask;
|
||||
found = 0;
|
||||
for(int l = 0; l < k; l++) {
|
||||
if(neighptr[l] == cj) {
|
||||
@ -98,7 +99,6 @@ void createNeighbors(Atom *atom, Neighbor *neighbor, int pattern, int nneighs, i
|
||||
} while(found == 1);
|
||||
} else {
|
||||
neighptr[k] = j;
|
||||
neighptr_imask[k] = imask;
|
||||
j = (j + 1) % m;
|
||||
}
|
||||
}
|
||||
@ -106,12 +106,10 @@ void createNeighbors(Atom *atom, Neighbor *neighbor, int pattern, int nneighs, i
|
||||
for(int r = 1; r < nreps; r++) {
|
||||
for(int k = 0; k < nneighs; k++) {
|
||||
neighptr[r * nneighs + k] = neighptr[k];
|
||||
neighptr_imask[r * nneighs + k] = neighptr_imask[k];
|
||||
}
|
||||
}
|
||||
|
||||
neighbor->numneigh[ci] = nneighs * nreps;
|
||||
neighbor->numneigh_masked[ci] = (masked == 1) ? (nneighs * nreps) : 0;
|
||||
}
|
||||
}
|
||||
|
||||
@ -127,13 +125,12 @@ int main(int argc, const char *argv[]) {
|
||||
int niclusters = 256; // Number of local i-clusters
|
||||
int iclusters_natoms = CLUSTER_M; // Number of valid atoms within i-clusters
|
||||
int nneighs = 9; // Number of j-cluster neighbors per i-cluster
|
||||
int masked = 0; // Use masked loop
|
||||
int nreps = 1;
|
||||
int csv = 0;
|
||||
|
||||
LIKWID_MARKER_INIT;
|
||||
LIKWID_MARKER_REGISTER("force");
|
||||
DEBUG_MESSAGE("Initializing parameters...\n");
|
||||
DEBUG("Initializing parameters...\n");
|
||||
init(¶m);
|
||||
|
||||
for(int i = 0; i < argc; i++) {
|
||||
@ -159,10 +156,6 @@ int main(int argc, const char *argv[]) {
|
||||
param.eam_file = strdup(argv[++i]);
|
||||
continue;
|
||||
}
|
||||
if((strcmp(argv[i], "-m") == 0)) {
|
||||
masked = 1;
|
||||
continue;
|
||||
}
|
||||
if((strcmp(argv[i], "-n") == 0) || (strcmp(argv[i], "--nsteps") == 0)) {
|
||||
param.ntimes = atoi(argv[++i]);
|
||||
continue;
|
||||
@ -213,11 +206,11 @@ int main(int argc, const char *argv[]) {
|
||||
}
|
||||
|
||||
if(param.force_field == FF_EAM) {
|
||||
DEBUG_MESSAGE("Initializing EAM parameters...\n");
|
||||
DEBUG("Initializing EAM parameters...\n");
|
||||
initEam(&eam, ¶m);
|
||||
}
|
||||
|
||||
DEBUG_MESSAGE("Initializing atoms...\n");
|
||||
DEBUG("Initializing atoms...\n");
|
||||
initAtom(atom);
|
||||
initStats(&stats);
|
||||
|
||||
@ -233,7 +226,7 @@ int main(int argc, const char *argv[]) {
|
||||
atom->cutforcesq[i] = param.cutforce * param.cutforce;
|
||||
}
|
||||
|
||||
DEBUG_MESSAGE("Creating atoms...\n");
|
||||
DEBUG("Creating atoms...\n");
|
||||
while(atom->Nmax < niclusters * iclusters_natoms) {
|
||||
growAtom(atom);
|
||||
}
|
||||
@ -288,13 +281,13 @@ int main(int argc, const char *argv[]) {
|
||||
printf("Estimated neighborlist data volume (kB): %.4f\n", estim_neighbors_volume / 1000.0);
|
||||
}
|
||||
|
||||
DEBUG_MESSAGE("Defining j-clusters...\n");
|
||||
DEBUG("Defining j-clusters...\n");
|
||||
defineJClusters(atom);
|
||||
DEBUG_MESSAGE("Initializing neighbor lists...\n");
|
||||
DEBUG("Initializing neighbor lists...\n");
|
||||
initNeighbor(&neighbor, ¶m);
|
||||
DEBUG_MESSAGE("Creating neighbor lists...\n");
|
||||
createNeighbors(atom, &neighbor, pattern, nneighs, nreps, masked);
|
||||
DEBUG_MESSAGE("Computing forces...\n");
|
||||
DEBUG("Creating neighbor lists...\n");
|
||||
createNeighbors(atom, &neighbor, pattern, nneighs, nreps);
|
||||
DEBUG("Computing forces...\n");
|
||||
|
||||
double T_accum = 0.0;
|
||||
for(int i = 0; i < param.ntimes; i++) {
|
||||
@ -1,13 +1,11 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
*/
|
||||
#include <stdio.h>
|
||||
#include <string.h>
|
||||
#include <math.h>
|
||||
#include <omp.h>
|
||||
//--
|
||||
#include <likwid-marker.h>
|
||||
//--
|
||||
@ -119,7 +117,7 @@ int main(int argc, char** argv) {
|
||||
|
||||
initParameter(¶m);
|
||||
for(int i = 0; i < argc; i++) {
|
||||
if((strcmp(argv[i], "-p") == 0) || (strcmp(argv[i], "--param") == 0)) {
|
||||
if((strcmp(argv[i], "-p") == 0)) {
|
||||
readParameter(¶m, argv[++i]);
|
||||
continue;
|
||||
}
|
||||
@ -310,30 +308,6 @@ int main(int argc, char** argv) {
|
||||
printf("TOTAL %.2fs FORCE %.2fs NEIGH %.2fs REST %.2fs\n",
|
||||
timer[TOTAL], timer[FORCE], timer[NEIGH], timer[TOTAL]-timer[FORCE]-timer[NEIGH]);
|
||||
printf(HLINE);
|
||||
|
||||
int nthreads = 0;
|
||||
int chunkSize = 0;
|
||||
omp_sched_t schedKind;
|
||||
char schedType[10];
|
||||
#pragma omp parallel
|
||||
#pragma omp master
|
||||
{
|
||||
omp_get_schedule(&schedKind, &chunkSize);
|
||||
|
||||
switch (schedKind)
|
||||
{
|
||||
case omp_sched_static: strcpy(schedType, "static"); break;
|
||||
case omp_sched_dynamic: strcpy(schedType, "dynamic"); break;
|
||||
case omp_sched_guided: strcpy(schedType, "guided"); break;
|
||||
case omp_sched_auto: strcpy(schedType, "auto"); break;
|
||||
}
|
||||
|
||||
nthreads = omp_get_max_threads();
|
||||
}
|
||||
|
||||
printf("Num threads: %d\n", nthreads);
|
||||
printf("Schedule: (%s,%d)\n", schedType, chunkSize);
|
||||
|
||||
printf("Performance: %.2f million atom updates per second\n",
|
||||
1e-6 * (double) atom.Natoms * param.ntimes / timer[TOTAL]);
|
||||
#ifdef COMPUTE_STATS
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -58,7 +58,6 @@ void initNeighbor(Neighbor *neighbor, Parameter *param) {
|
||||
neighbor->numneigh = NULL;
|
||||
neighbor->numneigh_masked = NULL;
|
||||
neighbor->neighbors = NULL;
|
||||
neighbor->neighbors_imask = NULL;
|
||||
}
|
||||
|
||||
void setupNeighbor(Parameter *param, Atom *atom) {
|
||||
@ -230,13 +229,10 @@ void buildNeighbor(Atom *atom, Neighbor *neighbor) {
|
||||
if(atom->Nclusters_local > nmax) {
|
||||
nmax = atom->Nclusters_local;
|
||||
if(neighbor->numneigh) free(neighbor->numneigh);
|
||||
if(neighbor->numneigh_masked) free(neighbor->numneigh_masked);
|
||||
if(neighbor->neighbors) free(neighbor->neighbors);
|
||||
if(neighbor->neighbors_imask) free(neighbor->neighbors_imask);
|
||||
neighbor->numneigh = (int*) malloc(nmax * sizeof(int));
|
||||
neighbor->numneigh_masked = (int*) malloc(nmax * sizeof(int));
|
||||
neighbor->neighbors = (int*) malloc(nmax * neighbor->maxneighs * sizeof(int));
|
||||
neighbor->neighbors_imask = (unsigned int*) malloc(nmax * neighbor->maxneighs * sizeof(unsigned int));
|
||||
neighbor->neighbors = (NeighborCluster*) malloc(nmax * neighbor->maxneighs * sizeof(NeighborCluster));
|
||||
}
|
||||
|
||||
MD_FLOAT bbx = 0.5 * (binsizex + binsizex);
|
||||
@ -252,8 +248,7 @@ void buildNeighbor(Atom *atom, Neighbor *neighbor) {
|
||||
|
||||
for(int ci = 0; ci < atom->Nclusters_local; ci++) {
|
||||
int ci_cj1 = CJ1_FROM_CI(ci);
|
||||
int *neighptr = &(neighbor->neighbors[ci * neighbor->maxneighs]);
|
||||
unsigned int *neighptr_imask = &(neighbor->neighbors_imask[ci * neighbor->maxneighs]);
|
||||
NeighborCluster *neighptr = &(neighbor->neighbors[ci * neighbor->maxneighs]);
|
||||
int n = 0, nmasked = 0;
|
||||
int ibin = atom->icluster_bin[ci];
|
||||
MD_FLOAT ibb_xmin = atom->iclusters[ci].bbminx;
|
||||
@ -329,17 +324,15 @@ void buildNeighbor(Atom *atom, Neighbor *neighbor) {
|
||||
imask = get_imask_simd_4xn(1, ci, cj);
|
||||
#endif
|
||||
|
||||
if(n < neighbor->maxneighs) {
|
||||
if(imask == NBNXN_INTERACTION_MASK_ALL) {
|
||||
neighptr[n] = cj;
|
||||
neighptr_imask[n] = imask;
|
||||
} else {
|
||||
neighptr[n] = neighptr[nmasked];
|
||||
neighptr_imask[n] = neighptr_imask[nmasked];
|
||||
neighptr[nmasked] = cj;
|
||||
neighptr_imask[nmasked] = imask;
|
||||
nmasked++;
|
||||
}
|
||||
if(imask == NBNXN_INTERACTION_MASK_ALL) {
|
||||
neighptr[n].cj = cj;
|
||||
neighptr[n].imask = imask;
|
||||
} else {
|
||||
neighptr[n].cj = neighptr[nmasked].cj;
|
||||
neighptr[n].imask = neighptr[nmasked].imask;
|
||||
neighptr[nmasked].cj = cj;
|
||||
neighptr[nmasked].imask = imask;
|
||||
nmasked++;
|
||||
}
|
||||
|
||||
n++;
|
||||
@ -364,8 +357,8 @@ void buildNeighbor(Atom *atom, Neighbor *neighbor) {
|
||||
// Fill neighbor list with dummy values to fit vector width
|
||||
if(CLUSTER_N < VECTOR_WIDTH) {
|
||||
while(n % (VECTOR_WIDTH / CLUSTER_N)) {
|
||||
neighptr[n] = atom->dummy_cj; // Last cluster is always a dummy cluster
|
||||
neighptr_imask[n] = 0;
|
||||
neighptr[n].cj = atom->dummy_cj; // Last cluster is always a dummy cluster
|
||||
neighptr[n].imask = 0;
|
||||
n++;
|
||||
}
|
||||
}
|
||||
@ -382,12 +375,10 @@ void buildNeighbor(Atom *atom, Neighbor *neighbor) {
|
||||
}
|
||||
|
||||
if(resize) {
|
||||
neighbor->maxneighs = new_maxneighs * 1.2;
|
||||
fprintf(stdout, "RESIZE %d\n", neighbor->maxneighs);
|
||||
neighbor->maxneighs = new_maxneighs * 1.2;
|
||||
free(neighbor->neighbors);
|
||||
free(neighbor->neighbors_imask);
|
||||
neighbor->neighbors = (int *) malloc(nmax * neighbor->maxneighs * sizeof(int));
|
||||
neighbor->neighbors_imask = (unsigned int *) malloc(nmax * neighbor->maxneighs * sizeof(unsigned int));
|
||||
neighbor->neighbors = (NeighborCluster*) malloc(atom->Nmax * neighbor->maxneighs * sizeof(int));
|
||||
}
|
||||
}
|
||||
|
||||
@ -442,21 +433,20 @@ void pruneNeighbor(Parameter *param, Atom *atom, Neighbor *neighbor) {
|
||||
MD_FLOAT cutsq = cutneighsq;
|
||||
|
||||
for(int ci = 0; ci < atom->Nclusters_local; ci++) {
|
||||
int *neighs = &neighbor->neighbors[ci * neighbor->maxneighs];
|
||||
unsigned int *neighs_imask = &neighbor->neighbors_imask[ci * neighbor->maxneighs];
|
||||
NeighborCluster *neighs = &neighbor->neighbors[ci * neighbor->maxneighs];
|
||||
int numneighs = neighbor->numneigh[ci];
|
||||
int numneighs_masked = neighbor->numneigh_masked[ci];
|
||||
int k = 0;
|
||||
|
||||
// Remove dummy clusters if necessary
|
||||
if(CLUSTER_N < VECTOR_WIDTH) {
|
||||
while(neighs[numneighs - 1] == atom->dummy_cj) {
|
||||
while(neighs[numneighs - 1].cj == atom->dummy_cj) {
|
||||
numneighs--;
|
||||
}
|
||||
}
|
||||
|
||||
while(k < numneighs) {
|
||||
int cj = neighs[k];
|
||||
int cj = neighs[k].cj;
|
||||
if(atomDistanceInRange(atom, ci, cj, cutsq)) {
|
||||
k++;
|
||||
} else {
|
||||
@ -471,8 +461,8 @@ void pruneNeighbor(Parameter *param, Atom *atom, Neighbor *neighbor) {
|
||||
// Readd dummy clusters if necessary
|
||||
if(CLUSTER_N < VECTOR_WIDTH) {
|
||||
while(numneighs % (VECTOR_WIDTH / CLUSTER_N)) {
|
||||
neighs[numneighs] = atom->dummy_cj; // Last cluster is always a dummy cluster
|
||||
neighs_imask[numneighs] = 0;
|
||||
neighs[numneighs].cj = atom->dummy_cj; // Last cluster is always a dummy cluster
|
||||
neighs[numneighs].imask = 0;
|
||||
numneighs++;
|
||||
}
|
||||
}
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -13,8 +13,7 @@ void traceAddresses(Parameter *param, Atom *atom, Neighbor *neighbor, int timest
|
||||
MEM_TRACER_INIT;
|
||||
INDEX_TRACER_INIT;
|
||||
int Nlocal = atom->Nlocal;
|
||||
int *neighs;
|
||||
unsigned int *neighs_imask;
|
||||
NeighborCluster* neighs;
|
||||
//MD_FLOAT* fx = atom->fx; MD_FLOAT* fy = atom->fy; MD_FLOAT* fz = atom->fz;
|
||||
|
||||
INDEX_TRACE_NATOMS(Nlocal, atom->Nghost, neighbor->maxneighs);
|
||||
@ -35,7 +34,7 @@ void traceAddresses(Parameter *param, Atom *atom, Neighbor *neighbor, int timest
|
||||
DIST_TRACE(neighs, numneighs);
|
||||
|
||||
for(int k = 0; k < numneighs; k++) {
|
||||
int j = neighs[k];
|
||||
int j = neighs[k].cj;
|
||||
MEM_TRACE(j, 'R');
|
||||
MEM_TRACE(atom_x(j), 'R');
|
||||
MEM_TRACE(atom_y(j), 'R');
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,18 +1,17 @@
|
||||
CC = /opt/homebrew/Cellar/llvm/18.1.5/bin/clang
|
||||
CC = clang
|
||||
LINKER = $(CC)
|
||||
|
||||
ANSI_CFLAGS = -ansi
|
||||
ANSI_CFLAGS += -std=c99
|
||||
ANSI_CFLAGS += -pedantic
|
||||
# ANSI_CFLAGS += -Wextra
|
||||
ANSI_CFLAGS += -Wextra
|
||||
|
||||
CFLAGS = -Ofast -march=native $(ANSI_CFLAGS) -Xpreprocessor -fopenmp #-g
|
||||
CFLAGS = -Ofast -march=native $(ANSI_CFLAGS) #-Xpreprocessor -fopenmp -g
|
||||
#CFLAGS = -Ofast -march=core-avx2 $(ANSI_CFLAGS) #-Xpreprocessor -fopenmp -g
|
||||
#CFLAGS = -O3 -march=cascadelake $(ANSI_CFLAGS) #-Xpreprocessor -fopenmp -g
|
||||
#CFLAGS = -Ofast $(ANSI_CFLAGS) -g #-Xpreprocessor -fopenmp -g
|
||||
ASFLAGS = #-masm=intel
|
||||
ASFLAGS = -masm=intel
|
||||
LFLAGS =
|
||||
DEFINES = -D_GNU_SOURCE
|
||||
# MacOSX with Apple Silicon and homebrew
|
||||
INCLUDES = -I/opt/homebrew/Cellar/libomp/18.1.5/include/
|
||||
LIBS = -lm -L/opt/homebrew/Cellar/libomp/18.1.5/lib/ -lomp
|
||||
INCLUDES =
|
||||
LIBS = -lm #-lomp
|
||||
11
include_GROMACS.mk
Normal file
11
include_GROMACS.mk
Normal file
@ -0,0 +1,11 @@
|
||||
GROMACS_PATH=/apps/Gromacs/2018.1-mkl
|
||||
GROMACS_INC ?= -I${GROMACS_PATH}/include
|
||||
GROMACS_DEFINES ?=
|
||||
GROMACS_LIB ?= -L${GROMACS_PATH}/lib64
|
||||
|
||||
ifeq ($(strip $(XTC_OUTPUT)),true)
|
||||
INCLUDES += ${GROMACS_INC}
|
||||
DEFINES += ${GROMACS_DEFINES}
|
||||
LIBS += -lgromacs
|
||||
LFLAGS += ${GROMACS_LIB}
|
||||
endif
|
||||
@ -1,7 +1,7 @@
|
||||
CC = icc
|
||||
LINKER = $(CC)
|
||||
|
||||
OPENMP = -qopenmp
|
||||
OPENMP = #-qopenmp
|
||||
PROFILE = #-profile-functions -g -pg
|
||||
|
||||
ifeq ($(ISA),AVX512)
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -502,21 +502,6 @@ int readAtom_in(Atom* atom, Parameter* param) {
|
||||
return natoms;
|
||||
}
|
||||
|
||||
void writeAtom(Atom *atom, Parameter *param) {
|
||||
FILE *fp = fopen(param->write_atom_file, "w");
|
||||
|
||||
for(int i = 0; i < atom->Nlocal; i++) {
|
||||
fprintf(fp, "%d,%f,%f,%f,%f,%f,%f,%f,0\n",
|
||||
atom->type[i], 1.0,
|
||||
atom_x(i), atom_y(i), atom_z(i),
|
||||
atom_vx(i), atom_vy(i), atom_vz(i));
|
||||
}
|
||||
|
||||
fclose(fp);
|
||||
fprintf(stdout, "Wrote input data to %s, grid size: %f, %f, %f\n",
|
||||
param->write_atom_file, param->xprd, param->yprd, param->zprd);
|
||||
}
|
||||
|
||||
void growAtom(Atom *atom) {
|
||||
DeviceAtom *d_atom = &(atom->d_atom);
|
||||
int nold = atom->Nmax;
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -29,7 +29,7 @@ extern "C" {
|
||||
}
|
||||
|
||||
// cuda kernel
|
||||
__global__ void calc_force(DeviceAtom a, MD_FLOAT cutforcesq, MD_FLOAT sigma6, MD_FLOAT epsilon, int Nlocal, int neigh_maxneighs, int *neigh_neighbors, int *neigh_numneigh, int ntypes) {
|
||||
__global__ void calc_force(DeviceAtom a, MD_FLOAT cutforcesq, MD_FLOAT sigma6, MD_FLOAT epsilon, int Nlocal, int neigh_maxneighs, int *neigh_neighbors, int *neigh_numneigh) {
|
||||
const int i = blockIdx.x * blockDim.x + threadIdx.x;
|
||||
if(i >= Nlocal) {
|
||||
return;
|
||||
@ -46,10 +46,6 @@ __global__ void calc_force(DeviceAtom a, MD_FLOAT cutforcesq, MD_FLOAT sigma6, M
|
||||
MD_FLOAT fiy = 0;
|
||||
MD_FLOAT fiz = 0;
|
||||
|
||||
#ifdef EXPLICIT_TYPES
|
||||
const int type_i = atom->type[i];
|
||||
#endif
|
||||
|
||||
for(int k = 0; k < numneighs; k++) {
|
||||
int j = neigh_neighbors[Nlocal * k + i];
|
||||
MD_FLOAT delx = xtmp - atom_x(j);
|
||||
@ -59,7 +55,7 @@ __global__ void calc_force(DeviceAtom a, MD_FLOAT cutforcesq, MD_FLOAT sigma6, M
|
||||
|
||||
#ifdef EXPLICIT_TYPES
|
||||
const int type_j = atom->type[j];
|
||||
const int type_ij = type_i * ntypes + type_j;
|
||||
const int type_ij = type_i * atom->ntypes + type_j;
|
||||
const MD_FLOAT cutforcesq = atom->cutforcesq[type_ij];
|
||||
const MD_FLOAT sigma6 = atom->sigma6[type_ij];
|
||||
const MD_FLOAT epsilon = atom->epsilon[type_ij];
|
||||
@ -113,7 +109,7 @@ extern "C" {
|
||||
|
||||
void finalIntegrate_cuda(bool reneigh, Parameter *param, Atom *atom) {
|
||||
const int Nlocal = atom->Nlocal;
|
||||
const int num_threads_per_block = get_cuda_num_threads();
|
||||
const int num_threads_per_block = get_num_threads();
|
||||
const int num_blocks = ceil((float)Nlocal / (float)num_threads_per_block);
|
||||
|
||||
kernel_final_integrate <<< num_blocks, num_threads_per_block >>> (param->dtforce, Nlocal, atom->d_atom);
|
||||
@ -127,7 +123,7 @@ void finalIntegrate_cuda(bool reneigh, Parameter *param, Atom *atom) {
|
||||
|
||||
void initialIntegrate_cuda(bool reneigh, Parameter *param, Atom *atom) {
|
||||
const int Nlocal = atom->Nlocal;
|
||||
const int num_threads_per_block = get_cuda_num_threads();
|
||||
const int num_threads_per_block = get_num_threads();
|
||||
const int num_blocks = ceil((float)Nlocal / (float)num_threads_per_block);
|
||||
|
||||
kernel_initial_integrate <<< num_blocks, num_threads_per_block >>> (param->dtforce, param->dt, Nlocal, atom->d_atom);
|
||||
@ -140,11 +136,13 @@ void initialIntegrate_cuda(bool reneigh, Parameter *param, Atom *atom) {
|
||||
}
|
||||
|
||||
double computeForceLJFullNeigh_cuda(Parameter *param, Atom *atom, Neighbor *neighbor) {
|
||||
const int num_threads_per_block = get_cuda_num_threads();
|
||||
const int num_threads_per_block = get_num_threads();
|
||||
int Nlocal = atom->Nlocal;
|
||||
#ifndef EXPLICIT_TYPES
|
||||
MD_FLOAT cutforcesq = param->cutforce * param->cutforce;
|
||||
MD_FLOAT sigma6 = param->sigma6;
|
||||
MD_FLOAT epsilon = param->epsilon;
|
||||
#endif
|
||||
|
||||
/*
|
||||
int nDevices;
|
||||
@ -167,7 +165,7 @@ double computeForceLJFullNeigh_cuda(Parameter *param, Atom *atom, Neighbor *neig
|
||||
double S = getTimeStamp();
|
||||
LIKWID_MARKER_START("force");
|
||||
|
||||
calc_force <<< num_blocks, num_threads_per_block >>> (atom->d_atom, cutforcesq, sigma6, epsilon, Nlocal, neighbor->maxneighs, neighbor->d_neighbor.neighbors, neighbor->d_neighbor.numneigh, atom->ntypes);
|
||||
calc_force <<< num_blocks, num_threads_per_block >>> (atom->d_atom, cutforcesq, sigma6, epsilon, Nlocal, neighbor->maxneighs, neighbor->d_neighbor.neighbors, neighbor->d_neighbor.numneigh);
|
||||
cuda_assert("calc_force", cudaPeekAtLastError());
|
||||
cuda_assert("calc_force", cudaDeviceSynchronize());
|
||||
cudaProfilerStop();
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -120,7 +120,7 @@ __global__ void binatoms_kernel(DeviceAtom a, int nall, int* bincount, int* bins
|
||||
|
||||
__global__ void compute_neighborhood(
|
||||
DeviceAtom a, DeviceNeighbor neigh, Neighbor_params np, int nlocal, int maxneighs, int nstencil, int* stencil,
|
||||
int* bins, int atoms_per_bin, int *bincount, int *new_maxneighs, MD_FLOAT cutneighsq, int ntypes) {
|
||||
int* bins, int atoms_per_bin, int *bincount, int *new_maxneighs, MD_FLOAT cutneighsq) {
|
||||
|
||||
const int i = blockIdx.x * blockDim.x + threadIdx.x;
|
||||
if(i >= nlocal) {
|
||||
@ -157,7 +157,7 @@ __global__ void compute_neighborhood(
|
||||
|
||||
#ifdef EXPLICIT_TYPES
|
||||
int type_j = atom->type[j];
|
||||
const MD_FLOAT cutoff = atom->cutneighsq[type_i * ntypes + type_j];
|
||||
const MD_FLOAT cutoff = atom->cutneighsq[type_i * atom->ntypes + type_j];
|
||||
#else
|
||||
const MD_FLOAT cutoff = cutneighsq;
|
||||
#endif
|
||||
@ -206,7 +206,7 @@ void binatoms_cuda(Atom *atom, Binning *c_binning, int *c_resize_needed, Neighbo
|
||||
|
||||
void buildNeighbor_cuda(Atom *atom, Neighbor *neighbor) {
|
||||
DeviceNeighbor *d_neighbor = &(neighbor->d_neighbor);
|
||||
const int num_threads_per_block = get_cuda_num_threads();
|
||||
const int num_threads_per_block = get_num_threads();
|
||||
int nall = atom->Nlocal + atom->Nghost;
|
||||
|
||||
cudaProfilerStart();
|
||||
@ -269,7 +269,7 @@ void buildNeighbor_cuda(Atom *atom, Neighbor *neighbor) {
|
||||
np, atom->Nlocal, neighbor->maxneighs, nstencil, c_stencil,
|
||||
c_binning.bins, c_binning.atoms_per_bin, c_binning.bincount,
|
||||
c_new_maxneighs,
|
||||
cutneighsq, atom->ntypes);
|
||||
cutneighsq);
|
||||
|
||||
cuda_assert("compute_neighborhood", cudaPeekAtLastError());
|
||||
cuda_assert("compute_neighborhood", cudaDeviceSynchronize());
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -65,7 +65,7 @@ __global__ void computePbcUpdate(DeviceAtom a, int nlocal, int nghost, int* PBCx
|
||||
/* update coordinates of ghost atoms */
|
||||
/* uses mapping created in setupPbc */
|
||||
void updatePbc_cuda(Atom *atom, Parameter *param, bool reneigh) {
|
||||
const int num_threads_per_block = get_cuda_num_threads();
|
||||
const int num_threads_per_block = get_num_threads();
|
||||
|
||||
if(reneigh) {
|
||||
memcpyToGPU(atom->d_atom.x, atom->x, sizeof(MD_FLOAT) * atom->Nmax * 3);
|
||||
@ -98,7 +98,7 @@ void updatePbc_cuda(Atom *atom, Parameter *param, bool reneigh) {
|
||||
}
|
||||
|
||||
void updateAtomsPbc_cuda(Atom* atom, Parameter *param) {
|
||||
const int num_threads_per_block = get_cuda_num_threads();
|
||||
const int num_threads_per_block = get_num_threads();
|
||||
MD_FLOAT xprd = param->xprd;
|
||||
MD_FLOAT yprd = param->yprd;
|
||||
MD_FLOAT zprd = param->zprd;
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -14,7 +14,6 @@ void initDevice(Atom *atom, Neighbor *neighbor) {
|
||||
|
||||
d_atom->epsilon = (MD_FLOAT *) allocateGPU(sizeof(MD_FLOAT) * atom->ntypes * atom->ntypes);
|
||||
d_atom->sigma6 = (MD_FLOAT *) allocateGPU(sizeof(MD_FLOAT) * atom->ntypes * atom->ntypes);
|
||||
d_atom->cutneighsq = (MD_FLOAT *) allocateGPU(sizeof(MD_FLOAT) * atom->ntypes * atom->ntypes);
|
||||
d_atom->cutforcesq = (MD_FLOAT *) allocateGPU(sizeof(MD_FLOAT) * atom->ntypes * atom->ntypes);
|
||||
d_neighbor->neighbors = (int *) allocateGPU(sizeof(int) * atom->Nmax * neighbor->maxneighs);
|
||||
d_neighbor->numneigh = (int *) allocateGPU(sizeof(int) * atom->Nmax);
|
||||
@ -23,7 +22,6 @@ void initDevice(Atom *atom, Neighbor *neighbor) {
|
||||
memcpyToGPU(d_atom->vx, atom->vx, sizeof(MD_FLOAT) * atom->Nmax * 3);
|
||||
memcpyToGPU(d_atom->sigma6, atom->sigma6, sizeof(MD_FLOAT) * atom->ntypes * atom->ntypes);
|
||||
memcpyToGPU(d_atom->epsilon, atom->epsilon, sizeof(MD_FLOAT) * atom->ntypes * atom->ntypes);
|
||||
memcpyToGPU(d_atom->cutneighsq, atom->cutneighsq, sizeof(MD_FLOAT) * atom->ntypes * atom->ntypes);
|
||||
memcpyToGPU(d_atom->cutforcesq, atom->cutforcesq, sizeof(MD_FLOAT) * atom->ntypes * atom->ntypes);
|
||||
memcpyToGPU(d_atom->type, atom->type, sizeof(int) * atom->Nmax);
|
||||
}
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
272
lammps/force_lj.c
Normal file
272
lammps/force_lj.c
Normal file
@ -0,0 +1,272 @@
|
||||
/*
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
*/
|
||||
#include <stdio.h>
|
||||
#include <stdlib.h>
|
||||
//---
|
||||
#include <atom.h>
|
||||
#include <likwid-marker.h>
|
||||
#include <neighbor.h>
|
||||
#include <parameter.h>
|
||||
#include <stats.h>
|
||||
#include <timing.h>
|
||||
|
||||
#ifdef __SIMD_KERNEL__
|
||||
#include <simd.h>
|
||||
#endif
|
||||
|
||||
double computeForceLJFullNeigh_plain_c(Parameter *param, Atom *atom, Neighbor *neighbor, Stats *stats) {
|
||||
int Nlocal = atom->Nlocal;
|
||||
int* neighs;
|
||||
#ifndef EXPLICIT_TYPES
|
||||
MD_FLOAT cutforcesq = param->cutforce * param->cutforce;
|
||||
MD_FLOAT sigma6 = param->sigma6;
|
||||
MD_FLOAT epsilon = param->epsilon;
|
||||
#endif
|
||||
const MD_FLOAT num1 = 1.0;
|
||||
const MD_FLOAT num48 = 48.0;
|
||||
const MD_FLOAT num05 = 0.5;
|
||||
|
||||
for(int i = 0; i < Nlocal; i++) {
|
||||
atom_fx(i) = 0.0;
|
||||
atom_fy(i) = 0.0;
|
||||
atom_fz(i) = 0.0;
|
||||
}
|
||||
double S = getTimeStamp();
|
||||
|
||||
#pragma omp parallel
|
||||
{
|
||||
LIKWID_MARKER_START("force");
|
||||
|
||||
#pragma omp for
|
||||
for(int i = 0; i < Nlocal; i++) {
|
||||
neighs = &neighbor->neighbors[i * neighbor->maxneighs];
|
||||
int numneighs = neighbor->numneigh[i];
|
||||
MD_FLOAT xtmp = atom_x(i);
|
||||
MD_FLOAT ytmp = atom_y(i);
|
||||
MD_FLOAT ztmp = atom_z(i);
|
||||
MD_FLOAT fix = 0;
|
||||
MD_FLOAT fiy = 0;
|
||||
MD_FLOAT fiz = 0;
|
||||
|
||||
#ifdef EXPLICIT_TYPES
|
||||
const int type_i = atom->type[i];
|
||||
#endif
|
||||
|
||||
for(int k = 0; k < numneighs; k++) {
|
||||
int j = neighs[k];
|
||||
MD_FLOAT delx = xtmp - atom_x(j);
|
||||
MD_FLOAT dely = ytmp - atom_y(j);
|
||||
MD_FLOAT delz = ztmp - atom_z(j);
|
||||
MD_FLOAT rsq = delx * delx + dely * dely + delz * delz;
|
||||
|
||||
#ifdef EXPLICIT_TYPES
|
||||
const int type_j = atom->type[j];
|
||||
const int type_ij = type_i * atom->ntypes + type_j;
|
||||
const MD_FLOAT cutforcesq = atom->cutforcesq[type_ij];
|
||||
const MD_FLOAT sigma6 = atom->sigma6[type_ij];
|
||||
const MD_FLOAT epsilon = atom->epsilon[type_ij];
|
||||
#endif
|
||||
|
||||
if(rsq < cutforcesq) {
|
||||
MD_FLOAT sr2 = num1 / rsq;
|
||||
MD_FLOAT sr6 = sr2 * sr2 * sr2 * sigma6;
|
||||
MD_FLOAT force = num48 * sr6 * (sr6 - num05) * sr2 * epsilon;
|
||||
fix += delx * force;
|
||||
fiy += dely * force;
|
||||
fiz += delz * force;
|
||||
#ifdef USE_REFERENCE_VERSION
|
||||
addStat(stats->atoms_within_cutoff, 1);
|
||||
} else {
|
||||
addStat(stats->atoms_outside_cutoff, 1);
|
||||
#endif
|
||||
}
|
||||
}
|
||||
|
||||
atom_fx(i) += fix;
|
||||
atom_fy(i) += fiy;
|
||||
atom_fz(i) += fiz;
|
||||
|
||||
addStat(stats->total_force_neighs, numneighs);
|
||||
addStat(stats->total_force_iters, (numneighs + VECTOR_WIDTH - 1) / VECTOR_WIDTH);
|
||||
}
|
||||
|
||||
LIKWID_MARKER_STOP("force");
|
||||
}
|
||||
|
||||
double E = getTimeStamp();
|
||||
return E-S;
|
||||
}
|
||||
|
||||
double computeForceLJHalfNeigh(Parameter *param, Atom *atom, Neighbor *neighbor, Stats *stats) {
|
||||
int Nlocal = atom->Nlocal;
|
||||
int* neighs;
|
||||
#ifndef EXPLICIT_TYPES
|
||||
MD_FLOAT cutforcesq = param->cutforce * param->cutforce;
|
||||
MD_FLOAT sigma6 = param->sigma6;
|
||||
MD_FLOAT epsilon = param->epsilon;
|
||||
#endif
|
||||
const MD_FLOAT num1 = 1.0;
|
||||
const MD_FLOAT num48 = 48.0;
|
||||
const MD_FLOAT num05 = 0.5;
|
||||
|
||||
for(int i = 0; i < Nlocal; i++) {
|
||||
atom_fx(i) = 0.0;
|
||||
atom_fy(i) = 0.0;
|
||||
atom_fz(i) = 0.0;
|
||||
}
|
||||
|
||||
double S = getTimeStamp();
|
||||
|
||||
#pragma omp parallel
|
||||
{
|
||||
LIKWID_MARKER_START("forceLJ-halfneigh");
|
||||
|
||||
#pragma omp for
|
||||
for(int i = 0; i < Nlocal; i++) {
|
||||
neighs = &neighbor->neighbors[i * neighbor->maxneighs];
|
||||
int numneighs = neighbor->numneigh[i];
|
||||
MD_FLOAT xtmp = atom_x(i);
|
||||
MD_FLOAT ytmp = atom_y(i);
|
||||
MD_FLOAT ztmp = atom_z(i);
|
||||
MD_FLOAT fix = 0;
|
||||
MD_FLOAT fiy = 0;
|
||||
MD_FLOAT fiz = 0;
|
||||
|
||||
#ifdef EXPLICIT_TYPES
|
||||
const int type_i = atom->type[i];
|
||||
#endif
|
||||
|
||||
// Pragma required to vectorize the inner loop
|
||||
#ifdef ENABLE_OMP_SIMD
|
||||
#pragma omp simd reduction(+: fix,fiy,fiz)
|
||||
#endif
|
||||
for(int k = 0; k < numneighs; k++) {
|
||||
int j = neighs[k];
|
||||
MD_FLOAT delx = xtmp - atom_x(j);
|
||||
MD_FLOAT dely = ytmp - atom_y(j);
|
||||
MD_FLOAT delz = ztmp - atom_z(j);
|
||||
MD_FLOAT rsq = delx * delx + dely * dely + delz * delz;
|
||||
|
||||
#ifdef EXPLICIT_TYPES
|
||||
const int type_j = atom->type[j];
|
||||
const int type_ij = type_i * atom->ntypes + type_j;
|
||||
const MD_FLOAT cutforcesq = atom->cutforcesq[type_ij];
|
||||
const MD_FLOAT sigma6 = atom->sigma6[type_ij];
|
||||
const MD_FLOAT epsilon = atom->epsilon[type_ij];
|
||||
#endif
|
||||
|
||||
if(rsq < cutforcesq) {
|
||||
MD_FLOAT sr2 = num1 / rsq;
|
||||
MD_FLOAT sr6 = sr2 * sr2 * sr2 * sigma6;
|
||||
MD_FLOAT force = num48 * sr6 * (sr6 - num05) * sr2 * epsilon;
|
||||
fix += delx * force;
|
||||
fiy += dely * force;
|
||||
fiz += delz * force;
|
||||
|
||||
// We do not need to update forces for ghost atoms
|
||||
if(j < Nlocal) {
|
||||
atom_fx(j) -= delx * force;
|
||||
atom_fy(j) -= dely * force;
|
||||
atom_fz(j) -= delz * force;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
atom_fx(i) += fix;
|
||||
atom_fy(i) += fiy;
|
||||
atom_fz(i) += fiz;
|
||||
|
||||
addStat(stats->total_force_neighs, numneighs);
|
||||
addStat(stats->total_force_iters, (numneighs + VECTOR_WIDTH - 1) / VECTOR_WIDTH);
|
||||
}
|
||||
|
||||
LIKWID_MARKER_STOP("forceLJ-halfneigh");
|
||||
}
|
||||
|
||||
double E = getTimeStamp();
|
||||
return E-S;
|
||||
}
|
||||
|
||||
double computeForceLJFullNeigh_simd(Parameter *param, Atom *atom, Neighbor *neighbor, Stats *stats) {
|
||||
int Nlocal = atom->Nlocal;
|
||||
int* neighs;
|
||||
MD_FLOAT cutforcesq = param->cutforce * param->cutforce;
|
||||
MD_FLOAT sigma6 = param->sigma6;
|
||||
MD_FLOAT epsilon = param->epsilon;
|
||||
|
||||
for(int i = 0; i < Nlocal; i++) {
|
||||
atom_fx(i) = 0.0;
|
||||
atom_fy(i) = 0.0;
|
||||
atom_fz(i) = 0.0;
|
||||
}
|
||||
|
||||
double S = getTimeStamp();
|
||||
|
||||
#ifndef __SIMD_KERNEL__
|
||||
fprintf(stderr, "Error: SIMD kernel not implemented for specified instruction set!");
|
||||
exit(-1);
|
||||
#else
|
||||
MD_SIMD_FLOAT cutforcesq_vec = simd_broadcast(cutforcesq);
|
||||
MD_SIMD_FLOAT sigma6_vec = simd_broadcast(sigma6);
|
||||
MD_SIMD_FLOAT eps_vec = simd_broadcast(epsilon);
|
||||
MD_SIMD_FLOAT c48_vec = simd_broadcast(48.0);
|
||||
MD_SIMD_FLOAT c05_vec = simd_broadcast(0.5);
|
||||
|
||||
|
||||
#pragma omp parallel
|
||||
{
|
||||
LIKWID_MARKER_START("force");
|
||||
|
||||
#pragma omp for
|
||||
for(int i = 0; i < Nlocal; i++) {
|
||||
neighs = &neighbor->neighbors[i * neighbor->maxneighs];
|
||||
int numneighs = neighbor->numneigh[i];
|
||||
MD_SIMD_INT numneighs_vec = simd_int_broadcast(numneighs);
|
||||
MD_SIMD_FLOAT xtmp = simd_broadcast(atom_x(i));
|
||||
MD_SIMD_FLOAT ytmp = simd_broadcast(atom_y(i));
|
||||
MD_SIMD_FLOAT ztmp = simd_broadcast(atom_z(i));
|
||||
MD_SIMD_FLOAT fix = simd_zero();
|
||||
MD_SIMD_FLOAT fiy = simd_zero();
|
||||
MD_SIMD_FLOAT fiz = simd_zero();
|
||||
|
||||
for(int k = 0; k < numneighs; k += VECTOR_WIDTH) {
|
||||
// If the last iteration of this loop is separated from the rest, this mask can be set only there
|
||||
MD_SIMD_MASK mask_numneighs = simd_mask_int_cond_lt(simd_int_add(simd_int_broadcast(k), simd_int_seq()), numneighs_vec);
|
||||
MD_SIMD_INT j = simd_int_mask_load(&neighs[k], mask_numneighs);
|
||||
#ifdef AOS
|
||||
MD_SIMD_INT j3 = simd_int_add(simd_int_add(j, j), j); // j * 3
|
||||
MD_SIMD_FLOAT delx = xtmp - simd_gather(j3, &(atom->x[0]), sizeof(MD_FLOAT));
|
||||
MD_SIMD_FLOAT dely = ytmp - simd_gather(j3, &(atom->x[1]), sizeof(MD_FLOAT));
|
||||
MD_SIMD_FLOAT delz = ztmp - simd_gather(j3, &(atom->x[2]), sizeof(MD_FLOAT));
|
||||
#else
|
||||
MD_SIMD_FLOAT delx = xtmp - simd_gather(j, atom->x, sizeof(MD_FLOAT));
|
||||
MD_SIMD_FLOAT dely = ytmp - simd_gather(j, atom->y, sizeof(MD_FLOAT));
|
||||
MD_SIMD_FLOAT delz = ztmp - simd_gather(j, atom->z, sizeof(MD_FLOAT));
|
||||
#endif
|
||||
MD_SIMD_FLOAT rsq = simd_fma(delx, delx, simd_fma(dely, dely, simd_mul(delz, delz)));
|
||||
MD_SIMD_MASK cutoff_mask = simd_mask_and(mask_numneighs, simd_mask_cond_lt(rsq, cutforcesq_vec));
|
||||
MD_SIMD_FLOAT sr2 = simd_reciprocal(rsq);
|
||||
MD_SIMD_FLOAT sr6 = simd_mul(sr2, simd_mul(sr2, simd_mul(sr2, sigma6_vec)));
|
||||
MD_SIMD_FLOAT force = simd_mul(c48_vec, simd_mul(sr6, simd_mul(simd_sub(sr6, c05_vec), simd_mul(sr2, eps_vec))));
|
||||
|
||||
fix = simd_masked_add(fix, simd_mul(delx, force), cutoff_mask);
|
||||
fiy = simd_masked_add(fiy, simd_mul(dely, force), cutoff_mask);
|
||||
fiz = simd_masked_add(fiz, simd_mul(delz, force), cutoff_mask);
|
||||
}
|
||||
|
||||
atom_fx(i) += simd_h_reduce_sum(fix);
|
||||
atom_fy(i) += simd_h_reduce_sum(fiy);
|
||||
atom_fz(i) += simd_h_reduce_sum(fiz);
|
||||
}
|
||||
|
||||
LIKWID_MARKER_STOP("force");
|
||||
}
|
||||
#endif
|
||||
|
||||
double E = getTimeStamp();
|
||||
return E-S;
|
||||
}
|
||||
102
lammps/includes/atom.h
Normal file
102
lammps/includes/atom.h
Normal file
@ -0,0 +1,102 @@
|
||||
/*
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
*/
|
||||
#include <parameter.h>
|
||||
|
||||
#ifndef __ATOM_H_
|
||||
#define __ATOM_H_
|
||||
|
||||
#ifdef CUDA_TARGET
|
||||
# define KERNEL_NAME "CUDA"
|
||||
# define computeForceLJFullNeigh computeForceLJFullNeigh_cuda
|
||||
# define initialIntegrate initialIntegrate_cuda
|
||||
# define finalIntegrate finalIntegrate_cuda
|
||||
# define buildNeighbor buildNeighbor_cuda
|
||||
# define updatePbc updatePbc_cuda
|
||||
# define updateAtomsPbc updateAtomsPbc_cuda
|
||||
#else
|
||||
# ifdef USE_SIMD_KERNEL
|
||||
# define KERNEL_NAME "SIMD"
|
||||
# define computeForceLJFullNeigh computeForceLJFullNeigh_simd
|
||||
# else
|
||||
# define KERNEL_NAME "plain-C"
|
||||
# define computeForceLJFullNeigh computeForceLJFullNeigh_plain_c
|
||||
# endif
|
||||
# define initialIntegrate initialIntegrate_cpu
|
||||
# define finalIntegrate finalIntegrate_cpu
|
||||
# define buildNeighbor buildNeighbor_cpu
|
||||
# define updatePbc updatePbc_cpu
|
||||
# define updateAtomsPbc updateAtomsPbc_cpu
|
||||
#endif
|
||||
|
||||
typedef struct {
|
||||
MD_FLOAT *x, *y, *z;
|
||||
MD_FLOAT *vx, *vy, *vz;
|
||||
MD_FLOAT *fx, *fy, *fz;
|
||||
int *border_map;
|
||||
int *type;
|
||||
MD_FLOAT *epsilon;
|
||||
MD_FLOAT *sigma6;
|
||||
MD_FLOAT *cutforcesq;
|
||||
MD_FLOAT *cutneighsq;
|
||||
} DeviceAtom;
|
||||
|
||||
typedef struct {
|
||||
int Natoms, Nlocal, Nghost, Nmax;
|
||||
MD_FLOAT *x, *y, *z;
|
||||
MD_FLOAT *vx, *vy, *vz;
|
||||
MD_FLOAT *fx, *fy, *fz;
|
||||
int *border_map;
|
||||
int *type;
|
||||
int ntypes;
|
||||
MD_FLOAT *epsilon;
|
||||
MD_FLOAT *sigma6;
|
||||
MD_FLOAT *cutforcesq;
|
||||
MD_FLOAT *cutneighsq;
|
||||
|
||||
// DEM
|
||||
MD_FLOAT *radius;
|
||||
MD_FLOAT *av;
|
||||
MD_FLOAT *r;
|
||||
|
||||
// Device data
|
||||
DeviceAtom d_atom;
|
||||
} Atom;
|
||||
|
||||
extern void initAtom(Atom*);
|
||||
extern void createAtom(Atom*, Parameter*);
|
||||
extern int readAtom(Atom*, Parameter*);
|
||||
extern int readAtom_pdb(Atom*, Parameter*);
|
||||
extern int readAtom_gro(Atom*, Parameter*);
|
||||
extern int readAtom_dmp(Atom*, Parameter*);
|
||||
extern int readAtom_in(Atom*, Parameter*);
|
||||
extern void growAtom(Atom*);
|
||||
|
||||
#ifdef AOS
|
||||
# define POS_DATA_LAYOUT "AoS"
|
||||
# define atom_x(i) atom->x[(i) * 3 + 0]
|
||||
# define atom_y(i) atom->x[(i) * 3 + 1]
|
||||
# define atom_z(i) atom->x[(i) * 3 + 2]
|
||||
# define atom_vx(i) atom->vx[(i) * 3 + 0]
|
||||
# define atom_vy(i) atom->vx[(i) * 3 + 1]
|
||||
# define atom_vz(i) atom->vx[(i) * 3 + 2]
|
||||
# define atom_fx(i) atom->fx[(i) * 3 + 0]
|
||||
# define atom_fy(i) atom->fx[(i) * 3 + 1]
|
||||
# define atom_fz(i) atom->fx[(i) * 3 + 2]
|
||||
#else
|
||||
# define POS_DATA_LAYOUT "SoA"
|
||||
# define atom_x(i) atom->x[i]
|
||||
# define atom_y(i) atom->y[i]
|
||||
# define atom_z(i) atom->z[i]
|
||||
# define atom_vx(i) atom->vx[i]
|
||||
# define atom_vy(i) atom->vy[i]
|
||||
# define atom_vz(i) atom->vz[i]
|
||||
# define atom_fx(i) atom->fx[i]
|
||||
# define atom_fy(i) atom->fy[i]
|
||||
# define atom_fz(i) atom->fz[i]
|
||||
#endif
|
||||
|
||||
#endif
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -11,7 +11,7 @@
|
||||
|
||||
#ifndef __PBC_H_
|
||||
#define __PBC_H_
|
||||
extern void initPbc(Atom*);
|
||||
extern void initPbc();
|
||||
extern void updatePbc_cpu(Atom*, Parameter*, bool);
|
||||
extern void updateAtomsPbc_cpu(Atom*, Parameter*);
|
||||
extern void setupPbc(Atom*, Parameter*);
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -59,6 +59,12 @@ void init(Parameter *param) {
|
||||
param->eam_file = NULL;
|
||||
}
|
||||
|
||||
// Show debug messages
|
||||
#define DEBUG(msg) printf(msg)
|
||||
// Do not show debug messages
|
||||
//#define DEBUG(msg)
|
||||
|
||||
|
||||
void createNeighbors(Atom *atom, Neighbor *neighbor, int pattern, int nneighs, int nreps) {
|
||||
const int maxneighs = nneighs * nreps;
|
||||
neighbor->numneigh = (int*) malloc(atom->Nmax * sizeof(int));
|
||||
@ -119,7 +125,7 @@ int main(int argc, const char *argv[]) {
|
||||
|
||||
LIKWID_MARKER_INIT;
|
||||
LIKWID_MARKER_REGISTER("force");
|
||||
DEBUG_MESSAGE("Initializing parameters...\n");
|
||||
DEBUG("Initializing parameters...\n");
|
||||
init(¶m);
|
||||
|
||||
for(int i = 0; i < argc; i++) {
|
||||
@ -190,11 +196,11 @@ int main(int argc, const char *argv[]) {
|
||||
}
|
||||
|
||||
if(param.force_field == FF_EAM) {
|
||||
DEBUG_MESSAGE("Initializing EAM parameters...\n");
|
||||
DEBUG("Initializing EAM parameters...\n");
|
||||
initEam(&eam, ¶m);
|
||||
}
|
||||
|
||||
DEBUG_MESSAGE("Initializing atoms...\n");
|
||||
DEBUG("Initializing atoms...\n");
|
||||
initAtom(atom);
|
||||
initStats(&stats);
|
||||
|
||||
@ -210,7 +216,7 @@ int main(int argc, const char *argv[]) {
|
||||
atom->cutforcesq[i] = param.cutforce * param.cutforce;
|
||||
}
|
||||
|
||||
DEBUG_MESSAGE("Creating atoms...\n");
|
||||
DEBUG("Creating atoms...\n");
|
||||
for(int i = 0; i < natoms; ++i) {
|
||||
while(atom->Nlocal > atom->Nmax - natoms) {
|
||||
growAtom(atom);
|
||||
@ -241,11 +247,11 @@ int main(int argc, const char *argv[]) {
|
||||
printf("Estimated neighborlist data volume (kB): %.4f\n", estim_neighbors_volume / 1000.0);
|
||||
}
|
||||
|
||||
DEBUG_MESSAGE("Initializing neighbor lists...\n");
|
||||
DEBUG("Initializing neighbor lists...\n");
|
||||
initNeighbor(&neighbor, ¶m);
|
||||
DEBUG_MESSAGE("Creating neighbor lists...\n");
|
||||
DEBUG("Creating neighbor lists...\n");
|
||||
createNeighbors(atom, &neighbor, pattern, nneighs, nreps);
|
||||
DEBUG_MESSAGE("Computing forces...\n");
|
||||
DEBUG("Computing forces...\n");
|
||||
|
||||
double T_accum = 0.0;
|
||||
for(int i = 0; i < param.ntimes; i++) {
|
||||
285
lammps/main.c
Normal file
285
lammps/main.c
Normal file
@ -0,0 +1,285 @@
|
||||
/*
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
*/
|
||||
#include <stdlib.h>
|
||||
#include <stdio.h>
|
||||
#include <string.h>
|
||||
#include <unistd.h>
|
||||
#include <limits.h>
|
||||
#include <math.h>
|
||||
#include <float.h>
|
||||
|
||||
#include <likwid-marker.h>
|
||||
|
||||
#include <allocate.h>
|
||||
#include <atom.h>
|
||||
#include <device.h>
|
||||
#include <eam.h>
|
||||
#include <integrate.h>
|
||||
#include <thermo.h>
|
||||
#include <timing.h>
|
||||
#include <neighbor.h>
|
||||
#include <parameter.h>
|
||||
#include <pbc.h>
|
||||
#include <stats.h>
|
||||
#include <timers.h>
|
||||
#include <util.h>
|
||||
#include <vtk.h>
|
||||
|
||||
#define HLINE "----------------------------------------------------------------------------\n"
|
||||
|
||||
extern double computeForceLJFullNeigh_plain_c(Parameter*, Atom*, Neighbor*, Stats*);
|
||||
extern double computeForceLJFullNeigh_simd(Parameter*, Atom*, Neighbor*, Stats*);
|
||||
extern double computeForceLJHalfNeigh(Parameter*, Atom*, Neighbor*, Stats*);
|
||||
extern double computeForceEam(Eam*, Parameter*, Atom*, Neighbor*, Stats*);
|
||||
extern double computeForceDemFullNeigh(Parameter*, Atom*, Neighbor*, Stats*);
|
||||
|
||||
#ifdef CUDA_TARGET
|
||||
extern double computeForceLJFullNeigh_cuda(Parameter*, Atom*, Neighbor*);
|
||||
#endif
|
||||
|
||||
double setup(Parameter *param, Eam *eam, Atom *atom, Neighbor *neighbor, Stats *stats) {
|
||||
if(param->force_field == FF_EAM) { initEam(eam, param); }
|
||||
double S, E;
|
||||
param->lattice = pow((4.0 / param->rho), (1.0 / 3.0));
|
||||
param->xprd = param->nx * param->lattice;
|
||||
param->yprd = param->ny * param->lattice;
|
||||
param->zprd = param->nz * param->lattice;
|
||||
|
||||
S = getTimeStamp();
|
||||
initAtom(atom);
|
||||
initPbc(atom);
|
||||
initStats(stats);
|
||||
initNeighbor(neighbor, param);
|
||||
if(param->input_file == NULL) {
|
||||
createAtom(atom, param);
|
||||
} else {
|
||||
readAtom(atom, param);
|
||||
}
|
||||
|
||||
setupNeighbor(param);
|
||||
setupThermo(param, atom->Natoms);
|
||||
if(param->input_file == NULL) { adjustThermo(param, atom); }
|
||||
setupPbc(atom, param);
|
||||
initDevice(atom, neighbor);
|
||||
updatePbc(atom, param, true);
|
||||
buildNeighbor(atom, neighbor);
|
||||
E = getTimeStamp();
|
||||
return E-S;
|
||||
}
|
||||
|
||||
double reneighbour(Parameter *param, Atom *atom, Neighbor *neighbor) {
|
||||
double S, E;
|
||||
S = getTimeStamp();
|
||||
LIKWID_MARKER_START("reneighbour");
|
||||
updateAtomsPbc(atom, param);
|
||||
setupPbc(atom, param);
|
||||
updatePbc(atom, param, true);
|
||||
//sortAtom(atom);
|
||||
buildNeighbor(atom, neighbor);
|
||||
LIKWID_MARKER_STOP("reneighbour");
|
||||
E = getTimeStamp();
|
||||
return E-S;
|
||||
}
|
||||
|
||||
void printAtomState(Atom *atom) {
|
||||
printf("Atom counts: Natoms=%d Nlocal=%d Nghost=%d Nmax=%d\n", atom->Natoms, atom->Nlocal, atom->Nghost, atom->Nmax);
|
||||
// int nall = atom->Nlocal + atom->Nghost;
|
||||
// for (int i=0; i<nall; i++) {
|
||||
// printf("%d %f %f %f\n", i, atom->x[i], atom->y[i], atom->z[i]);
|
||||
// }
|
||||
}
|
||||
|
||||
double computeForce(Eam *eam, Parameter *param, Atom *atom, Neighbor *neighbor, Stats *stats) {
|
||||
if(param->force_field == FF_EAM) {
|
||||
return computeForceEam(eam, param, atom, neighbor, stats);
|
||||
} else if(param->force_field == FF_DEM) {
|
||||
if(param->half_neigh) {
|
||||
fprintf(stderr, "Error: DEM cannot use half neighbor-lists!\n");
|
||||
return 0.0;
|
||||
} else {
|
||||
return computeForceDemFullNeigh(param, atom, neighbor, stats);
|
||||
}
|
||||
}
|
||||
|
||||
if(param->half_neigh) {
|
||||
return computeForceLJHalfNeigh(param, atom, neighbor, stats);
|
||||
}
|
||||
|
||||
#ifdef CUDA_TARGET
|
||||
return computeForceLJFullNeigh(param, atom, neighbor);
|
||||
#else
|
||||
return computeForceLJFullNeigh(param, atom, neighbor, stats);
|
||||
#endif
|
||||
}
|
||||
|
||||
void writeInput(Parameter *param, Atom *atom) {
|
||||
FILE *fpin = fopen("input.in", "w");
|
||||
fprintf(fpin, "0,%f,0,%f,0,%f\n", param->xprd, param->yprd, param->zprd);
|
||||
|
||||
for(int i = 0; i < atom->Nlocal; i++) {
|
||||
fprintf(fpin, "1,%f,%f,%f,%f,%f,%f\n", atom_x(i), atom_y(i), atom_z(i), atom_vx(i), atom_vy(i), atom_vz(i));
|
||||
}
|
||||
|
||||
fclose(fpin);
|
||||
}
|
||||
|
||||
int main(int argc, char** argv) {
|
||||
double timer[NUMTIMER];
|
||||
Eam eam;
|
||||
Atom atom;
|
||||
Neighbor neighbor;
|
||||
Stats stats;
|
||||
Parameter param;
|
||||
|
||||
LIKWID_MARKER_INIT;
|
||||
#pragma omp parallel
|
||||
{
|
||||
LIKWID_MARKER_REGISTER("force");
|
||||
//LIKWID_MARKER_REGISTER("reneighbour");
|
||||
//LIKWID_MARKER_REGISTER("pbc");
|
||||
}
|
||||
|
||||
initParameter(¶m);
|
||||
for(int i = 0; i < argc; i++) {
|
||||
if((strcmp(argv[i], "-p") == 0)) {
|
||||
readParameter(¶m, argv[++i]);
|
||||
continue;
|
||||
}
|
||||
if((strcmp(argv[i], "-f") == 0)) {
|
||||
if((param.force_field = str2ff(argv[++i])) < 0) {
|
||||
fprintf(stderr, "Invalid force field!\n");
|
||||
exit(-1);
|
||||
}
|
||||
continue;
|
||||
}
|
||||
if((strcmp(argv[i], "-i") == 0)) {
|
||||
param.input_file = strdup(argv[++i]);
|
||||
continue;
|
||||
}
|
||||
if((strcmp(argv[i], "-e") == 0)) {
|
||||
param.eam_file = strdup(argv[++i]);
|
||||
continue;
|
||||
}
|
||||
if((strcmp(argv[i], "-n") == 0) || (strcmp(argv[i], "--nsteps") == 0)) {
|
||||
param.ntimes = atoi(argv[++i]);
|
||||
continue;
|
||||
}
|
||||
if((strcmp(argv[i], "-nx") == 0)) {
|
||||
param.nx = atoi(argv[++i]);
|
||||
continue;
|
||||
}
|
||||
if((strcmp(argv[i], "-ny") == 0)) {
|
||||
param.ny = atoi(argv[++i]);
|
||||
continue;
|
||||
}
|
||||
if((strcmp(argv[i], "-nz") == 0)) {
|
||||
param.nz = atoi(argv[++i]);
|
||||
continue;
|
||||
}
|
||||
if((strcmp(argv[i], "-half") == 0)) {
|
||||
param.half_neigh = atoi(argv[++i]);
|
||||
continue;
|
||||
}
|
||||
if((strcmp(argv[i], "-r") == 0) || (strcmp(argv[i], "--radius") == 0)) {
|
||||
param.cutforce = atof(argv[++i]);
|
||||
continue;
|
||||
}
|
||||
if((strcmp(argv[i], "-s") == 0) || (strcmp(argv[i], "--skin") == 0)) {
|
||||
param.skin = atof(argv[++i]);
|
||||
continue;
|
||||
}
|
||||
if((strcmp(argv[i], "--freq") == 0)) {
|
||||
param.proc_freq = atof(argv[++i]);
|
||||
continue;
|
||||
}
|
||||
if((strcmp(argv[i], "--vtk") == 0)) {
|
||||
param.vtk_file = strdup(argv[++i]);
|
||||
continue;
|
||||
}
|
||||
if((strcmp(argv[i], "-h") == 0) || (strcmp(argv[i], "--help") == 0)) {
|
||||
printf("MD Bench: A minimalistic re-implementation of miniMD\n");
|
||||
printf(HLINE);
|
||||
printf("-p <string>: file to read parameters from (can be specified more than once)\n");
|
||||
printf("-f <string>: force field (lj, eam or dem), default lj\n");
|
||||
printf("-i <string>: input file with atom positions (dump)\n");
|
||||
printf("-e <string>: input file for EAM\n");
|
||||
printf("-n / --nsteps <int>: set number of timesteps for simulation\n");
|
||||
printf("-nx/-ny/-nz <int>: set linear dimension of systembox in x/y/z direction\n");
|
||||
printf("-r / --radius <real>: set cutoff radius\n");
|
||||
printf("-s / --skin <real>: set skin (verlet buffer)\n");
|
||||
printf("--freq <real>: processor frequency (GHz)\n");
|
||||
printf("--vtk <string>: VTK file for visualization\n");
|
||||
printf(HLINE);
|
||||
exit(EXIT_SUCCESS);
|
||||
}
|
||||
}
|
||||
|
||||
param.cutneigh = param.cutforce + param.skin;
|
||||
setup(¶m, &eam, &atom, &neighbor, &stats);
|
||||
printParameter(¶m);
|
||||
printf(HLINE);
|
||||
|
||||
printf("step\ttemp\t\tpressure\n");
|
||||
computeThermo(0, ¶m, &atom);
|
||||
#if defined(MEM_TRACER) || defined(INDEX_TRACER)
|
||||
traceAddresses(¶m, &atom, &neighbor, n + 1);
|
||||
#endif
|
||||
|
||||
//writeInput(¶m, &atom);
|
||||
|
||||
timer[FORCE] = computeForce(&eam, ¶m, &atom, &neighbor, &stats);
|
||||
timer[NEIGH] = 0.0;
|
||||
timer[TOTAL] = getTimeStamp();
|
||||
|
||||
if(param.vtk_file != NULL) {
|
||||
write_atoms_to_vtk_file(param.vtk_file, &atom, 0);
|
||||
}
|
||||
|
||||
for(int n = 0; n < param.ntimes; n++) {
|
||||
bool reneigh = (n + 1) % param.reneigh_every == 0;
|
||||
initialIntegrate(reneigh, ¶m, &atom);
|
||||
if((n + 1) % param.reneigh_every) {
|
||||
updatePbc(&atom, ¶m, false);
|
||||
} else {
|
||||
timer[NEIGH] += reneighbour(¶m, &atom, &neighbor);
|
||||
}
|
||||
|
||||
#if defined(MEM_TRACER) || defined(INDEX_TRACER)
|
||||
traceAddresses(¶m, &atom, &neighbor, n + 1);
|
||||
#endif
|
||||
|
||||
timer[FORCE] += computeForce(&eam, ¶m, &atom, &neighbor, &stats);
|
||||
finalIntegrate(reneigh, ¶m, &atom);
|
||||
|
||||
if(!((n + 1) % param.nstat) && (n+1) < param.ntimes) {
|
||||
#ifdef CUDA_TARGET
|
||||
memcpyFromGPU(atom.x, atom.d_atom.x, atom.Nmax * sizeof(MD_FLOAT) * 3);
|
||||
#endif
|
||||
computeThermo(n + 1, ¶m, &atom);
|
||||
}
|
||||
|
||||
if(param.vtk_file != NULL) {
|
||||
write_atoms_to_vtk_file(param.vtk_file, &atom, n + 1);
|
||||
}
|
||||
}
|
||||
|
||||
timer[TOTAL] = getTimeStamp() - timer[TOTAL];
|
||||
computeThermo(-1, ¶m, &atom);
|
||||
|
||||
printf(HLINE);
|
||||
printf("System: %d atoms %d ghost atoms, Steps: %d\n", atom.Natoms, atom.Nghost, param.ntimes);
|
||||
printf("TOTAL %.2fs FORCE %.2fs NEIGH %.2fs REST %.2fs\n",
|
||||
timer[TOTAL], timer[FORCE], timer[NEIGH], timer[TOTAL]-timer[FORCE]-timer[NEIGH]);
|
||||
printf(HLINE);
|
||||
printf("Performance: %.2f million atom updates per second\n",
|
||||
1e-6 * (double) atom.Natoms * param.ntimes / timer[TOTAL]);
|
||||
#ifdef COMPUTE_STATS
|
||||
displayStatistics(&atom, ¶m, &stats, timer);
|
||||
#endif
|
||||
LIKWID_MARKER_CLOSE;
|
||||
return EXIT_SUCCESS;
|
||||
}
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -326,45 +326,45 @@ void sortAtom(Atom* atom) {
|
||||
int Nmax = atom->Nmax;
|
||||
int* binpos = bincount;
|
||||
|
||||
for(int i = 1; i < mbins; i++) {
|
||||
binpos[i] += binpos[i - 1];
|
||||
for(int i=1; i<mbins; i++) {
|
||||
binpos[i] += binpos[i-1];
|
||||
}
|
||||
|
||||
#ifdef AOS
|
||||
#ifdef AOS
|
||||
MD_FLOAT* new_x = (MD_FLOAT*) malloc(Nmax * sizeof(MD_FLOAT) * 3);
|
||||
MD_FLOAT* new_vx = (MD_FLOAT*) malloc(Nmax * sizeof(MD_FLOAT) * 3);
|
||||
#else
|
||||
#else
|
||||
MD_FLOAT* new_x = (MD_FLOAT*) malloc(Nmax * sizeof(MD_FLOAT));
|
||||
MD_FLOAT* new_y = (MD_FLOAT*) malloc(Nmax * sizeof(MD_FLOAT));
|
||||
MD_FLOAT* new_z = (MD_FLOAT*) malloc(Nmax * sizeof(MD_FLOAT));
|
||||
MD_FLOAT* new_vx = (MD_FLOAT*) malloc(Nmax * sizeof(MD_FLOAT));
|
||||
MD_FLOAT* new_vy = (MD_FLOAT*) malloc(Nmax * sizeof(MD_FLOAT));
|
||||
MD_FLOAT* new_vz = (MD_FLOAT*) malloc(Nmax * sizeof(MD_FLOAT));
|
||||
#endif
|
||||
#endif
|
||||
MD_FLOAT* old_x = atom->x; MD_FLOAT* old_y = atom->y; MD_FLOAT* old_z = atom->z;
|
||||
MD_FLOAT* old_vx = atom->vx; MD_FLOAT* old_vy = atom->vy; MD_FLOAT* old_vz = atom->vz;
|
||||
|
||||
for(int mybin = 0; mybin < mbins; mybin++) {
|
||||
int start = mybin > 0 ? binpos[mybin - 1] : 0;
|
||||
for(int mybin = 0; mybin<mbins; mybin++) {
|
||||
int start = mybin>0?binpos[mybin-1]:0;
|
||||
int count = binpos[mybin] - start;
|
||||
for(int k = 0; k < count; k++) {
|
||||
for(int k=0; k<count; k++) {
|
||||
int new_i = start + k;
|
||||
int old_i = bins[mybin * atoms_per_bin + k];
|
||||
#ifdef AOS
|
||||
#ifdef AOS
|
||||
new_x[new_i * 3 + 0] = old_x[old_i * 3 + 0];
|
||||
new_x[new_i * 3 + 1] = old_x[old_i * 3 + 1];
|
||||
new_x[new_i * 3 + 2] = old_x[old_i * 3 + 2];
|
||||
new_vx[new_i * 3 + 0] = old_vx[old_i * 3 + 0];
|
||||
new_vx[new_i * 3 + 1] = old_vx[old_i * 3 + 1];
|
||||
new_vx[new_i * 3 + 2] = old_vx[old_i * 3 + 2];
|
||||
#else
|
||||
#else
|
||||
new_x[new_i] = old_x[old_i];
|
||||
new_y[new_i] = old_y[old_i];
|
||||
new_z[new_i] = old_z[old_i];
|
||||
new_vx[new_i] = old_vx[old_i];
|
||||
new_vy[new_i] = old_vy[old_i];
|
||||
new_vz[new_i] = old_vz[old_i];
|
||||
#endif
|
||||
#endif
|
||||
}
|
||||
}
|
||||
|
||||
@ -372,7 +372,7 @@ void sortAtom(Atom* atom) {
|
||||
free(atom->vx);
|
||||
atom->x = new_x;
|
||||
atom->vx = new_vx;
|
||||
#ifndef AOS
|
||||
#ifndef AOS
|
||||
free(atom->y);
|
||||
free(atom->z);
|
||||
free(atom->vy);
|
||||
@ -381,5 +381,5 @@ void sortAtom(Atom* atom) {
|
||||
atom->z = new_z;
|
||||
atom->vy = new_vy;
|
||||
atom->vz = new_vz;
|
||||
#endif
|
||||
#endif
|
||||
}
|
||||
171
lammps/pbc.c
Normal file
171
lammps/pbc.c
Normal file
@ -0,0 +1,171 @@
|
||||
/*
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
*/
|
||||
#include <stdbool.h>
|
||||
#include <stdlib.h>
|
||||
#include <stdio.h>
|
||||
//---
|
||||
#include <pbc.h>
|
||||
#include <atom.h>
|
||||
#include <allocate.h>
|
||||
|
||||
#define DELTA 20000
|
||||
|
||||
int NmaxGhost;
|
||||
int *PBCx, *PBCy, *PBCz;
|
||||
|
||||
static void growPbc(Atom*);
|
||||
|
||||
/* exported subroutines */
|
||||
void initPbc(Atom* atom) {
|
||||
NmaxGhost = 0;
|
||||
atom->border_map = NULL;
|
||||
PBCx = NULL; PBCy = NULL; PBCz = NULL;
|
||||
}
|
||||
|
||||
/* update coordinates of ghost atoms */
|
||||
/* uses mapping created in setupPbc */
|
||||
void updatePbc_cpu(Atom *atom, Parameter *param, bool doReneighbor) {
|
||||
int *border_map = atom->border_map;
|
||||
int nlocal = atom->Nlocal;
|
||||
MD_FLOAT xprd = param->xprd;
|
||||
MD_FLOAT yprd = param->yprd;
|
||||
MD_FLOAT zprd = param->zprd;
|
||||
|
||||
for(int i = 0; i < atom->Nghost; i++) {
|
||||
atom_x(nlocal + i) = atom_x(border_map[i]) + PBCx[i] * xprd;
|
||||
atom_y(nlocal + i) = atom_y(border_map[i]) + PBCy[i] * yprd;
|
||||
atom_z(nlocal + i) = atom_z(border_map[i]) + PBCz[i] * zprd;
|
||||
}
|
||||
}
|
||||
|
||||
/* relocate atoms that have left domain according
|
||||
* to periodic boundary conditions */
|
||||
void updateAtomsPbc_cpu(Atom *atom, Parameter *param) {
|
||||
MD_FLOAT xprd = param->xprd;
|
||||
MD_FLOAT yprd = param->yprd;
|
||||
MD_FLOAT zprd = param->zprd;
|
||||
|
||||
for(int i = 0; i < atom->Nlocal; i++) {
|
||||
if(atom_x(i) < 0.0) {
|
||||
atom_x(i) += xprd;
|
||||
} else if(atom_x(i) >= xprd) {
|
||||
atom_x(i) -= xprd;
|
||||
}
|
||||
|
||||
if(atom_y(i) < 0.0) {
|
||||
atom_y(i) += yprd;
|
||||
} else if(atom_y(i) >= yprd) {
|
||||
atom_y(i) -= yprd;
|
||||
}
|
||||
|
||||
if(atom_z(i) < 0.0) {
|
||||
atom_z(i) += zprd;
|
||||
} else if(atom_z(i) >= zprd) {
|
||||
atom_z(i) -= zprd;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/* setup periodic boundary conditions by
|
||||
* defining ghost atoms around domain
|
||||
* only creates mapping and coordinate corrections
|
||||
* that are then enforced in updatePbc */
|
||||
#define ADDGHOST(dx,dy,dz) \
|
||||
Nghost++; \
|
||||
border_map[Nghost] = i; \
|
||||
PBCx[Nghost] = dx; \
|
||||
PBCy[Nghost] = dy; \
|
||||
PBCz[Nghost] = dz; \
|
||||
atom->type[atom->Nlocal + Nghost] = atom->type[i]
|
||||
|
||||
void setupPbc(Atom *atom, Parameter *param) {
|
||||
int *border_map = atom->border_map;
|
||||
MD_FLOAT xprd = param->xprd;
|
||||
MD_FLOAT yprd = param->yprd;
|
||||
MD_FLOAT zprd = param->zprd;
|
||||
MD_FLOAT Cutneigh = param->cutneigh;
|
||||
int Nghost = -1;
|
||||
|
||||
for(int i = 0; i < atom->Nlocal; i++) {
|
||||
if (atom->Nlocal + Nghost + 7 >= atom->Nmax) {
|
||||
growAtom(atom);
|
||||
}
|
||||
|
||||
if (Nghost + 7 >= NmaxGhost) {
|
||||
growPbc(atom);
|
||||
border_map = atom->border_map;
|
||||
}
|
||||
|
||||
MD_FLOAT x = atom_x(i);
|
||||
MD_FLOAT y = atom_y(i);
|
||||
MD_FLOAT z = atom_z(i);
|
||||
|
||||
/* Setup ghost atoms */
|
||||
/* 6 planes */
|
||||
if(param->pbc_x != 0) {
|
||||
if (x < Cutneigh) { ADDGHOST(+1,0,0); }
|
||||
if (x >= (xprd-Cutneigh)) { ADDGHOST(-1,0,0); }
|
||||
}
|
||||
|
||||
if(param->pbc_y != 0) {
|
||||
if (y < Cutneigh) { ADDGHOST(0,+1,0); }
|
||||
if (y >= (yprd-Cutneigh)) { ADDGHOST(0,-1,0); }
|
||||
}
|
||||
|
||||
if(param->pbc_z != 0) {
|
||||
if (z < Cutneigh) { ADDGHOST(0,0,+1); }
|
||||
if (z >= (zprd-Cutneigh)) { ADDGHOST(0,0,-1); }
|
||||
}
|
||||
|
||||
/* 8 corners */
|
||||
if(param->pbc_x != 0 && param->pbc_y != 0 && param->pbc_z != 0) {
|
||||
if (x < Cutneigh && y < Cutneigh && z < Cutneigh) { ADDGHOST(+1,+1,+1); }
|
||||
if (x < Cutneigh && y >= (yprd-Cutneigh) && z < Cutneigh) { ADDGHOST(+1,-1,+1); }
|
||||
if (x < Cutneigh && y >= Cutneigh && z >= (zprd-Cutneigh)) { ADDGHOST(+1,+1,-1); }
|
||||
if (x < Cutneigh && y >= (yprd-Cutneigh) && z >= (zprd-Cutneigh)) { ADDGHOST(+1,-1,-1); }
|
||||
if (x >= (xprd-Cutneigh) && y < Cutneigh && z < Cutneigh) { ADDGHOST(-1,+1,+1); }
|
||||
if (x >= (xprd-Cutneigh) && y >= (yprd-Cutneigh) && z < Cutneigh) { ADDGHOST(-1,-1,+1); }
|
||||
if (x >= (xprd-Cutneigh) && y < Cutneigh && z >= (zprd-Cutneigh)) { ADDGHOST(-1,+1,-1); }
|
||||
if (x >= (xprd-Cutneigh) && y >= (yprd-Cutneigh) && z >= (zprd-Cutneigh)) { ADDGHOST(-1,-1,-1); }
|
||||
}
|
||||
|
||||
/* 12 edges */
|
||||
if(param->pbc_x != 0 && param->pbc_z != 0) {
|
||||
if (x < Cutneigh && z < Cutneigh) { ADDGHOST(+1,0,+1); }
|
||||
if (x < Cutneigh && z >= (zprd-Cutneigh)) { ADDGHOST(+1,0,-1); }
|
||||
if (x >= (xprd-Cutneigh) && z < Cutneigh) { ADDGHOST(-1,0,+1); }
|
||||
if (x >= (xprd-Cutneigh) && z >= (zprd-Cutneigh)) { ADDGHOST(-1,0,-1); }
|
||||
}
|
||||
|
||||
if(param->pbc_y != 0 && param->pbc_z != 0) {
|
||||
if (y < Cutneigh && z < Cutneigh) { ADDGHOST(0,+1,+1); }
|
||||
if (y < Cutneigh && z >= (zprd-Cutneigh)) { ADDGHOST(0,+1,-1); }
|
||||
if (y >= (yprd-Cutneigh) && z < Cutneigh) { ADDGHOST(0,-1,+1); }
|
||||
if (y >= (yprd-Cutneigh) && z >= (zprd-Cutneigh)) { ADDGHOST(0,-1,-1); }
|
||||
}
|
||||
|
||||
if(param->pbc_x != 0 && param->pbc_y != 0) {
|
||||
if (y < Cutneigh && x < Cutneigh) { ADDGHOST(+1,+1,0); }
|
||||
if (y < Cutneigh && x >= (xprd-Cutneigh)) { ADDGHOST(-1,+1,0); }
|
||||
if (y >= (yprd-Cutneigh) && x < Cutneigh) { ADDGHOST(+1,-1,0); }
|
||||
if (y >= (yprd-Cutneigh) && x >= (xprd-Cutneigh)) { ADDGHOST(-1,-1,0); }
|
||||
}
|
||||
}
|
||||
// increase by one to make it the ghost atom count
|
||||
atom->Nghost = Nghost + 1;
|
||||
}
|
||||
|
||||
/* internal subroutines */
|
||||
void growPbc(Atom* atom) {
|
||||
int nold = NmaxGhost;
|
||||
NmaxGhost += DELTA;
|
||||
|
||||
atom->border_map = (int*) reallocate(atom->border_map, ALIGNMENT, NmaxGhost * sizeof(int), nold * sizeof(int));
|
||||
PBCx = (int*) reallocate(PBCx, ALIGNMENT, NmaxGhost * sizeof(int), nold * sizeof(int));
|
||||
PBCy = (int*) reallocate(PBCy, ALIGNMENT, NmaxGhost * sizeof(int), nold * sizeof(int));
|
||||
PBCz = (int*) reallocate(PBCz, ALIGNMENT, NmaxGhost * sizeof(int), nold * sizeof(int));
|
||||
}
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
@ -1,5 +1,5 @@
|
||||
/*
|
||||
* Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
|
||||
* Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
|
||||
* All rights reserved. This file is part of MD-Bench.
|
||||
* Use of this source code is governed by a LGPL-3.0
|
||||
* license that can be found in the LICENSE file.
|
||||
88
likwid-outputs/csx-lammps-dp-mem_dp-stub.out
Normal file
88
likwid-outputs/csx-lammps-dp-mem_dp-stub.out
Normal file
@ -0,0 +1,88 @@
|
||||
--------------------------------------------------------------------------------
|
||||
CPU name: Intel(R) Xeon(R) Gold 6248 CPU @ 2.50GHz
|
||||
CPU type: Intel Cascadelake SP processor
|
||||
CPU clock: 2.49 GHz
|
||||
--------------------------------------------------------------------------------
|
||||
Initializing parameters...
|
||||
Initializing atoms...
|
||||
Creating atoms...
|
||||
Pattern: seq
|
||||
Number of timesteps: 200
|
||||
Number of atoms: 256
|
||||
Number of neighbors per atom: 1024
|
||||
Number of times to replicate neighbor lists: 1
|
||||
Estimated total data volume (kB): 1062.9120
|
||||
Estimated atom data volume (kB): 6.1440
|
||||
Estimated neighborlist data volume (kB): 1050.6240
|
||||
Initializing neighbor lists...
|
||||
Creating neighbor lists...
|
||||
Computing forces...
|
||||
Total time: 0.2735, Mega atom updates/s: 0.1872
|
||||
Cycles per atom: 10682.8568, Cycles per neighbor: 10.4325
|
||||
Statistics:
|
||||
Vector width: 8, Processor frequency: 2.0000 GHz
|
||||
Average neighbors per atom: 1018.9055
|
||||
Average SIMD iterations per atom: 127.3632
|
||||
Total number of computed pair interactions: 52428800
|
||||
Total number of SIMD iterations: 6553600
|
||||
Useful read data volume for force computation: 1.47GB
|
||||
Cycles/SIMD iteration: 83.4598
|
||||
--------------------------------------------------------------------------------
|
||||
Region force, Group 1: MEM_DP
|
||||
+-------------------+------------+
|
||||
| Region Info | HWThread 0 |
|
||||
+-------------------+------------+
|
||||
| RDTSC Runtime [s] | 0.110776 |
|
||||
| call count | 200 |
|
||||
+-------------------+------------+
|
||||
|
||||
+------------------------------------------+---------+------------+
|
||||
| Event | Counter | HWThread 0 |
|
||||
+------------------------------------------+---------+------------+
|
||||
| INSTR_RETIRED_ANY | FIXC0 | 267036300 |
|
||||
| CPU_CLK_UNHALTED_CORE | FIXC1 | 219034500 |
|
||||
| CPU_CLK_UNHALTED_REF | FIXC2 | 273793400 |
|
||||
| PWR_PKG_ENERGY | PWR0 | 10.9296 |
|
||||
| PWR_DRAM_ENERGY | PWR3 | 0 |
|
||||
| FP_ARITH_INST_RETIRED_128B_PACKED_DOUBLE | PMC0 | 0 |
|
||||
| FP_ARITH_INST_RETIRED_SCALAR_DOUBLE | PMC1 | 159400 |
|
||||
| FP_ARITH_INST_RETIRED_256B_PACKED_DOUBLE | PMC2 | 0 |
|
||||
| FP_ARITH_INST_RETIRED_512B_PACKED_DOUBLE | PMC3 | 197068800 |
|
||||
| CAS_COUNT_RD | MBOX0C0 | 8643 |
|
||||
| CAS_COUNT_WR | MBOX0C1 | 1367 |
|
||||
| CAS_COUNT_RD | MBOX1C0 | 9124 |
|
||||
| CAS_COUNT_WR | MBOX1C1 | 1354 |
|
||||
| CAS_COUNT_RD | MBOX2C0 | 9138 |
|
||||
| CAS_COUNT_WR | MBOX2C1 | 1356 |
|
||||
| CAS_COUNT_RD | MBOX3C0 | 5586 |
|
||||
| CAS_COUNT_WR | MBOX3C1 | 1297 |
|
||||
| CAS_COUNT_RD | MBOX4C0 | 5328 |
|
||||
| CAS_COUNT_WR | MBOX4C1 | 1269 |
|
||||
| CAS_COUNT_RD | MBOX5C0 | 5280 |
|
||||
| CAS_COUNT_WR | MBOX5C1 | 1295 |
|
||||
+------------------------------------------+---------+------------+
|
||||
|
||||
+-----------------------------------+------------+
|
||||
| Metric | HWThread 0 |
|
||||
+-----------------------------------+------------+
|
||||
| Runtime (RDTSC) [s] | 0.1108 |
|
||||
| Runtime unhalted [s] | 0.0878 |
|
||||
| Clock [MHz] | 1995.2564 |
|
||||
| CPI | 0.8202 |
|
||||
| Energy [J] | 10.9296 |
|
||||
| Power [W] | 98.6643 |
|
||||
| Energy DRAM [J] | 0 |
|
||||
| Power DRAM [W] | 0 |
|
||||
| DP [MFLOP/s] | 14233.3287 |
|
||||
| AVX DP [MFLOP/s] | 14231.8898 |
|
||||
| Packed [MUOPS/s] | 1778.9862 |
|
||||
| Scalar [MUOPS/s] | 1.4389 |
|
||||
| Memory read bandwidth [MBytes/s] | 24.9001 |
|
||||
| Memory read data volume [GBytes] | 0.0028 |
|
||||
| Memory write bandwidth [MBytes/s] | 4.5861 |
|
||||
| Memory write data volume [GBytes] | 0.0005 |
|
||||
| Memory bandwidth [MBytes/s] | 29.4863 |
|
||||
| Memory data volume [GBytes] | 0.0033 |
|
||||
| Operational intensity | 482.7104 |
|
||||
+-----------------------------------+------------+
|
||||
|
||||
168
likwid-outputs/csx-lammps-dp-mem_dp.out
Normal file
168
likwid-outputs/csx-lammps-dp-mem_dp.out
Normal file
@ -0,0 +1,168 @@
|
||||
--------------------------------------------------------------------------------
|
||||
CPU name: Intel(R) Xeon(R) Gold 6248 CPU @ 2.50GHz
|
||||
CPU type: Intel Cascadelake SP processor
|
||||
CPU clock: 2.49 GHz
|
||||
--------------------------------------------------------------------------------
|
||||
Parameters:
|
||||
Force field: lj
|
||||
Kernel: plain-C
|
||||
Data layout: AoS
|
||||
Floating-point precision: double
|
||||
Unit cells (nx, ny, nz): 32, 32, 32
|
||||
Domain box sizes (x, y, z): 5.374708e+01, 5.374708e+01, 5.374708e+01
|
||||
Periodic (x, y, z): 1, 1, 1
|
||||
Lattice size: 1.679596e+00
|
||||
Epsilon: 1.000000e+00
|
||||
Sigma: 1.000000e+00
|
||||
Spring constant: 1.000000e+00
|
||||
Damping constant: 1.000000e+00
|
||||
Temperature: 1.440000e+00
|
||||
RHO: 8.442000e-01
|
||||
Mass: 1.000000e+00
|
||||
Number of types: 4
|
||||
Number of timesteps: 200
|
||||
Report stats every (timesteps): 100
|
||||
Reneighbor every (timesteps): 20
|
||||
Prune every (timesteps): 1000
|
||||
Output positions every (timesteps): 20
|
||||
Output velocities every (timesteps): 5
|
||||
Delta time (dt): 5.000000e-03
|
||||
Cutoff radius: 2.500000e+00
|
||||
Skin: 3.000000e-01
|
||||
Half neighbor lists: 0
|
||||
Processor frequency (GHz): 2.0000
|
||||
----------------------------------------------------------------------------
|
||||
step temp pressure
|
||||
0 1.440000e+00 1.215639e+00
|
||||
100 8.200895e-01 6.923143e-01
|
||||
200 7.961495e-01 6.721043e-01
|
||||
----------------------------------------------------------------------------
|
||||
System: 131072 atoms 47265 ghost atoms, Steps: 200
|
||||
TOTAL 11.50s FORCE 5.28s NEIGH 5.91s REST 0.31s
|
||||
----------------------------------------------------------------------------
|
||||
Performance: 2.28 million atom updates per second
|
||||
Statistics:
|
||||
Vector width: 8, Processor frequency: 2.0000 GHz
|
||||
Average neighbors per atom: 76.0352
|
||||
Average SIMD iterations per atom: 9.9181
|
||||
Total number of computed pair interactions: 2003182862
|
||||
Total number of SIMD iterations: 261297661
|
||||
Useful read data volume for force computation: 57.46GB
|
||||
Cycles/SIMD iteration: 40.4432
|
||||
--------------------------------------------------------------------------------
|
||||
Region force, Group 1: MEM_DP
|
||||
+-------------------+------------+
|
||||
| Region Info | HWThread 0 |
|
||||
+-------------------+------------+
|
||||
| RDTSC Runtime [s] | 5.115807 |
|
||||
| call count | 201 |
|
||||
+-------------------+------------+
|
||||
|
||||
+------------------------------------------+---------+-------------+
|
||||
| Event | Counter | HWThread 0 |
|
||||
+------------------------------------------+---------+-------------+
|
||||
| INSTR_RETIRED_ANY | FIXC0 | 12592470000 |
|
||||
| CPU_CLK_UNHALTED_CORE | FIXC1 | 10196910000 |
|
||||
| CPU_CLK_UNHALTED_REF | FIXC2 | 12746120000 |
|
||||
| PWR_PKG_ENERGY | PWR0 | 307.9429 |
|
||||
| PWR_DRAM_ENERGY | PWR3 | 0 |
|
||||
| FP_ARITH_INST_RETIRED_128B_PACKED_DOUBLE | PMC0 | 0 |
|
||||
| FP_ARITH_INST_RETIRED_SCALAR_DOUBLE | PMC1 | 79042240 |
|
||||
| FP_ARITH_INST_RETIRED_256B_PACKED_DOUBLE | PMC2 | 0 |
|
||||
| FP_ARITH_INST_RETIRED_512B_PACKED_DOUBLE | PMC3 | 8076039000 |
|
||||
| CAS_COUNT_RD | MBOX0C0 | 22734550 |
|
||||
| CAS_COUNT_WR | MBOX0C1 | 1147714 |
|
||||
| CAS_COUNT_RD | MBOX1C0 | 22755180 |
|
||||
| CAS_COUNT_WR | MBOX1C1 | 1144415 |
|
||||
| CAS_COUNT_RD | MBOX2C0 | 22762780 |
|
||||
| CAS_COUNT_WR | MBOX2C1 | 1129051 |
|
||||
| CAS_COUNT_RD | MBOX3C0 | 22905660 |
|
||||
| CAS_COUNT_WR | MBOX3C1 | 1143324 |
|
||||
| CAS_COUNT_RD | MBOX4C0 | 22914860 |
|
||||
| CAS_COUNT_WR | MBOX4C1 | 1169116 |
|
||||
| CAS_COUNT_RD | MBOX5C0 | 22890220 |
|
||||
| CAS_COUNT_WR | MBOX5C1 | 1180739 |
|
||||
+------------------------------------------+---------+-------------+
|
||||
|
||||
+-----------------------------------+------------+
|
||||
| Metric | HWThread 0 |
|
||||
+-----------------------------------+------------+
|
||||
| Runtime (RDTSC) [s] | 5.1158 |
|
||||
| Runtime unhalted [s] | 4.0885 |
|
||||
| Clock [MHz] | 1995.2508 |
|
||||
| CPI | 0.8098 |
|
||||
| Energy [J] | 307.9429 |
|
||||
| Power [W] | 60.1944 |
|
||||
| Energy DRAM [J] | 0 |
|
||||
| Power DRAM [W] | 0 |
|
||||
| DP [MFLOP/s] | 12644.6041 |
|
||||
| AVX DP [MFLOP/s] | 12629.1535 |
|
||||
| Packed [MUOPS/s] | 1578.6442 |
|
||||
| Scalar [MUOPS/s] | 15.4506 |
|
||||
| Memory read bandwidth [MBytes/s] | 1713.4438 |
|
||||
| Memory read data volume [GBytes] | 8.7656 |
|
||||
| Memory write bandwidth [MBytes/s] | 86.5003 |
|
||||
| Memory write data volume [GBytes] | 0.4425 |
|
||||
| Memory bandwidth [MBytes/s] | 1799.9442 |
|
||||
| Memory data volume [GBytes] | 9.2082 |
|
||||
| Operational intensity | 7.0250 |
|
||||
+-----------------------------------+------------+
|
||||
|
||||
Region reneighbour, Group 1: MEM_DP
|
||||
+-------------------+------------+
|
||||
| Region Info | HWThread 0 |
|
||||
+-------------------+------------+
|
||||
| RDTSC Runtime [s] | 5.897385 |
|
||||
| call count | 10 |
|
||||
+-------------------+------------+
|
||||
|
||||
+------------------------------------------+---------+-------------+
|
||||
| Event | Counter | HWThread 0 |
|
||||
+------------------------------------------+---------+-------------+
|
||||
| INSTR_RETIRED_ANY | FIXC0 | 18212540000 |
|
||||
| CPU_CLK_UNHALTED_CORE | FIXC1 | 11728500000 |
|
||||
| CPU_CLK_UNHALTED_REF | FIXC2 | 14660630000 |
|
||||
| PWR_PKG_ENERGY | PWR0 | 338.9000 |
|
||||
| PWR_DRAM_ENERGY | PWR3 | 0 |
|
||||
| FP_ARITH_INST_RETIRED_128B_PACKED_DOUBLE | PMC0 | 0 |
|
||||
| FP_ARITH_INST_RETIRED_SCALAR_DOUBLE | PMC1 | 6240402000 |
|
||||
| FP_ARITH_INST_RETIRED_256B_PACKED_DOUBLE | PMC2 | 0 |
|
||||
| FP_ARITH_INST_RETIRED_512B_PACKED_DOUBLE | PMC3 | 983040 |
|
||||
| CAS_COUNT_RD | MBOX0C0 | 2086787 |
|
||||
| CAS_COUNT_WR | MBOX0C1 | 1115626 |
|
||||
| CAS_COUNT_RD | MBOX1C0 | 2089964 |
|
||||
| CAS_COUNT_WR | MBOX1C1 | 1117021 |
|
||||
| CAS_COUNT_RD | MBOX2C0 | 2103832 |
|
||||
| CAS_COUNT_WR | MBOX2C1 | 1117965 |
|
||||
| CAS_COUNT_RD | MBOX3C0 | 2086930 |
|
||||
| CAS_COUNT_WR | MBOX3C1 | 1102471 |
|
||||
| CAS_COUNT_RD | MBOX4C0 | 2094688 |
|
||||
| CAS_COUNT_WR | MBOX4C1 | 1103018 |
|
||||
| CAS_COUNT_RD | MBOX5C0 | 2097438 |
|
||||
| CAS_COUNT_WR | MBOX5C1 | 1102525 |
|
||||
+------------------------------------------+---------+-------------+
|
||||
|
||||
+-----------------------------------+------------+
|
||||
| Metric | HWThread 0 |
|
||||
+-----------------------------------+------------+
|
||||
| Runtime (RDTSC) [s] | 5.8974 |
|
||||
| Runtime unhalted [s] | 4.7026 |
|
||||
| Clock [MHz] | 1995.2473 |
|
||||
| CPI | 0.6440 |
|
||||
| Energy [J] | 338.9000 |
|
||||
| Power [W] | 57.4661 |
|
||||
| Energy DRAM [J] | 0 |
|
||||
| Power DRAM [W] | 0 |
|
||||
| DP [MFLOP/s] | 1059.4978 |
|
||||
| AVX DP [MFLOP/s] | 1.3335 |
|
||||
| Packed [MUOPS/s] | 0.1667 |
|
||||
| Scalar [MUOPS/s] | 1058.1643 |
|
||||
| Memory read bandwidth [MBytes/s] | 136.3006 |
|
||||
| Memory read data volume [GBytes] | 0.8038 |
|
||||
| Memory write bandwidth [MBytes/s] | 72.2612 |
|
||||
| Memory write data volume [GBytes] | 0.4262 |
|
||||
| Memory bandwidth [MBytes/s] | 208.5618 |
|
||||
| Memory data volume [GBytes] | 1.2300 |
|
||||
| Operational intensity | 5.0800 |
|
||||
+-----------------------------------+------------+
|
||||
|
||||
Some files were not shown because too many files have changed in this diff Show More
Loading…
x
Reference in New Issue
Block a user