blas_bm.cc

#include "benchmark/benchmark.h"
#include "cblas.h"

#include <memory>

static void BM_GEMM_unique_ptr(benchmark::State& state) {
  std::unique_ptr<float> A(new float[state.range(0) * state.range(0)]);
  std::unique_ptr<float> B(new float[state.range(0) * state.range(0)]);
  std::unique_ptr<float> C(new float[state.range(0) * state.range(0)]);

  //openblas_set_num_threads(1);
  for (auto _ : state) {
    cblas_sgemm(CblasColMajor, CblasNoTrans, CblasTrans,
        state.range(0),
        state.range(0),
        state.range(0),
        1, // alpha
        A.get(),
        state.range(0),
        B.get(),
        state.range(0),
        1, // beta,
        C.get(),
        state.range(0));
    benchmark::DoNotOptimize(C.get());
  }
}

static void BM_GEMM_raw_ptr(benchmark::State& state) {
  float* A = new float[state.range(0) * state.range(0)];
  float* B = new float[state.range(0) * state.range(0)];
  float* C = new float[state.range(0) * state.range(0)];

  //openblas_set_num_threads(1);
  for (auto _ : state) {
    cblas_sgemm(CblasColMajor, CblasNoTrans, CblasTrans,
        state.range(0),
        state.range(0),
        state.range(0),
        1, // alpha
        A,
        state.range(0),
        B,
        state.range(0),
        1, // beta,
        C,
        state.range(0));
    benchmark::DoNotOptimize(C);
  }

  delete[] C;
  delete[] B;
  delete[] A;
}

BENCHMARK(BM_GEMM_unique_ptr)->Range(4, 1024);
BENCHMARK(BM_GEMM_raw_ptr)->Range(4, 1024);

command

$ clang++-5.0 -std=c++11 ./blas_bm.cc -I ~/git/benchmark/include/ -L ~/git/benchmark/build/src/ -lbenchmark_main -lbenchmark -lopenblas -pthread

output

./a.out 
2020-06-05 16:09:38
Running ./a.out
Run on (4 X 3900 MHz CPU s)
CPU Caches:
  L1 Data 32 KiB (x2)
  L1 Instruction 32 KiB (x2)
  L2 Unified 256 KiB (x2)
  L3 Unified 4096 KiB (x1)
Load Average: 3.57, 3.55, 3.11
***WARNING*** CPU scaling is enabled, the benchmark real time measurements may be noisy and will incur extra overhead.
------------------------------------------------------------------
Benchmark                        Time             CPU   Iterations
------------------------------------------------------------------
BM_GEMM_unique_ptr/4           465 ns          448 ns      1974826
BM_GEMM_unique_ptr/8           321 ns          321 ns      1889871
BM_GEMM_unique_ptr/64        10502 ns        10499 ns        65323
BM_GEMM_unique_ptr/512     3810300 ns      2597048 ns          387
BM_GEMM_unique_ptr/1024   31456758 ns     18933383 ns           49
BM_GEMM_raw_ptr/4              255 ns          254 ns      2978025
BM_GEMM_raw_ptr/8              337 ns          322 ns      2221654
BM_GEMM_raw_ptr/64           10354 ns        10121 ns        72125
BM_GEMM_raw_ptr/512        6574122 ns      2887697 ns          339
BM_GEMM_raw_ptr/1024      22505692 ns     17119469 ns           47