devtools/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2026-04-10 11:34:33 +08:00
Code Issues Packages Projects Releases Wiki Activity

Extended the tensor benchmark suite to support types other than floats

Browse Source
This commit is contained in:
Benoit Steiner
2016-02-23 05:28:02 +00:00
parent f442a5a5b3
commit 8cb9bfab87
3 changed files with 74 additions and 74 deletions
Download Patch File Download Diff File Expand all files Collapse all files

42
bench/tensors/tensor_benchmarks_cpu.cc
View File

@@ -9,13 +9,13 @@ Eigen::ThreadPool pool(threads); \
Eigen::ThreadPoolDevice device(&pool, threads);
// Simple functions
#define BM_FuncCPU(FUNC, THREADS) \
static void BM_##FUNC##_##THREADS##T(int iters, int N) { \
StopBenchmarkTiming(); \
CREATE_THREAD_POOL(THREADS); \
BenchmarkSuite<Eigen::ThreadPoolDevice> suite(device, N); \
suite.FUNC(iters); \
} \
#define BM_FuncCPU(FUNC, THREADS) \
static void BM_##FUNC##_##THREADS##T(int iters, int N) { \
StopBenchmarkTiming(); \
CREATE_THREAD_POOL(THREADS); \
BenchmarkSuite<Eigen::ThreadPoolDevice, float> suite(device, N); \
suite.FUNC(iters); \
} \
BENCHMARK_RANGE(BM_##FUNC##_##THREADS##T, 10, 5000);
BM_FuncCPU(memcpy, 4);
@@ -80,19 +80,19 @@ BM_FuncCPU(colReduction, 12);
// Contractions
#define BM_FuncWithInputDimsCPU(FUNC, D1, D2, D3, THREADS) \
static void BM_##FUNC##_##D1##x##D2##x##D3##_##THREADS##T(int iters, int N) {\
StopBenchmarkTiming(); \
if (THREADS == 1) { \
Eigen::DefaultDevice device; \
BenchmarkSuite<Eigen::DefaultDevice> suite(device, D1, D2, D3); \
suite.FUNC(iters); \
} else { \
CREATE_THREAD_POOL(THREADS); \
BenchmarkSuite<Eigen::ThreadPoolDevice> suite(device, D1, D2, D3); \
suite.FUNC(iters); \
} \
} \
#define BM_FuncWithInputDimsCPU(FUNC, D1, D2, D3, THREADS) \
static void BM_##FUNC##_##D1##x##D2##x##D3##_##THREADS##T(int iters, int N) { \
StopBenchmarkTiming(); \
if (THREADS == 1) { \
Eigen::DefaultDevice device; \
BenchmarkSuite<Eigen::DefaultDevice, float> suite(device, D1, D2, D3); \
suite.FUNC(iters); \
} else { \
CREATE_THREAD_POOL(THREADS); \
BenchmarkSuite<Eigen::ThreadPoolDevice, float> suite(device, D1, D2, D3); \
suite.FUNC(iters); \
} \
} \
BENCHMARK_RANGE(BM_##FUNC##_##D1##x##D2##x##D3##_##THREADS##T, 10, 5000);
@@ -138,7 +138,7 @@ BM_FuncWithInputDimsCPU(contraction, N, N, 1, 16);
static void BM_##FUNC##_##DIM1##x##DIM2##_##THREADS##T(int iters, int N) { \
StopBenchmarkTiming(); \
CREATE_THREAD_POOL(THREADS); \
BenchmarkSuite<Eigen::ThreadPoolDevice> suite(device, N); \
BenchmarkSuite<Eigen::ThreadPoolDevice, float> suite(device, N); \
suite.FUNC(iters, DIM1, DIM2); \
} \
BENCHMARK_RANGE(BM_##FUNC##_##DIM1##x##DIM2##_##THREADS##T, 128, 5000);