Apple Accelerate sgemv benchmark

accelerate_gemv.cpp

#include <algorithm>
#include <chrono>
#include <iostream>
#include <iterator>
#include <random>
#include <vector>

#include <Accelerate/Accelerate.h>
#include <cblas.h>

using namespace std;

// Compile with:
//
// c++ -std=c++17 accelerate_gemv.cpp \
//  -framework Accelerate \
//  -I/Library/Developer/CommandLineTools/SDKs/MacOSX.sdk/System/Library/Frameworks/Accelerate.framework/Versions/Current/Frameworks/vecLib.framework/Headers/
int main(int, char **) {
  int iters = 1000;
  int m = 1;
  /* int n = 2048; */
  int n = 51865;
  int k = 512;

  int lda = n;
  int incx = 1;
  float alpha = 1.;
  float beta = 0.;
  int incy = 1;

  random_device rnd_device;
  mt19937 mersenne_engine{rnd_device()};
  uniform_int_distribution<int> dist{1, 52};

  auto gen = [&dist, &mersenne_engine]() { return dist(mersenne_engine); };

  vector<float> a(n * k);
  generate(begin(a), end(a), gen);

  vector<float> x(k);
  generate(begin(x), end(x), gen);

  vector<float> y(n);

  auto start = std::chrono::steady_clock::now();
  for (int i = 0; i < iters; i++) {
    cblas_sgemv(CblasRowMajor, CblasNoTrans, k, n, alpha, a.data(), lda,
                x.data(), incx, beta, y.data(), incy);
  }
  auto end = std::chrono::steady_clock::now();
  auto duration =
      std::chrono::duration_cast<std::chrono::microseconds>(end - start)
          .count();
  auto flops =
      static_cast<double>(2 * m * n * k * static_cast<int64_t>(iters)) /
      (static_cast<double>(duration) / 1000000.0);
  auto gflops = flops / 1e9;

  std::cout << "duration (s) " << duration << std::endl;
  std::cout << "gflops " << gflops << std::endl;
}

Created as part of comparisons with rten in robertknight/rten#42 (comment).