Book a Demo!
CoCalc Logo Icon
StoreFeaturesDocsShareSupportNewsAboutPoliciesSign UpSign In
emscripten-core
GitHub Repository: emscripten-core/emscripten
 Path: blob/main/test/benchmark/benchmark_ssse3.cpp
6174 views
1
/*

2
 * Copyright 2020 The Emscripten Authors.  All rights reserved.

3
 * Emscripten is available under two separate licenses, the MIT license and the

4
 * University of Illinois/NCSA Open Source License.  Both these licenses can be

5
 * found in the LICENSE file.

6
 */

7
#include <tmmintrin.h>

8
#include "benchmark_sse.h"

9


10
int main() {

11
  printf ("{ \"workload\": %u, \"results\": [\n", N);

12


13
  float *src_flt = alloc_float_buffer();

14
  float *src2_flt = alloc_float_buffer();

15
  float *dst_flt = alloc_float_buffer();

16
  for(int i = 0; i < N; ++i) src_flt[i] = (float)(1.0 + (double)rand() / RAND_MAX);

17
  for(int i = 0; i < N; ++i) src2_flt[i] = (float)(1.0 + (double)rand() / RAND_MAX);

18


19
  double *src_dbl = alloc_double_buffer();

20
  double *src2_dbl = alloc_double_buffer();

21
  double *dst_dbl = alloc_double_buffer();

22
  for(int i = 0; i < N; ++i) src_dbl[i] = 1.0 + (double)rand() / RAND_MAX;

23
  for(int i = 0; i < N; ++i) src2_dbl[i] = 1.0 + (double)rand() / RAND_MAX;

24


25
  int *src_int = alloc_int_buffer();

26
  int *src2_int = alloc_int_buffer();

27
  int *dst_int = alloc_int_buffer();

28
  for(int i = 0; i < N; ++i) src_int[i] = rand();

29
  for(int i = 0; i < N; ++i) src2_int[i] = rand();

30


31
  float scalarTime = 0.f;

32


33
  // Benchmarks start:

34
  SETCHART("abs");

35
  UNARYOP_I_I("_mm_abs_epi8", _mm_abs_epi8, _mm_load_si128((__m128i*)src_int));

36
  UNARYOP_I_I("_mm_abs_epi16", _mm_abs_epi16, _mm_load_si128((__m128i*)src_int));

37
  UNARYOP_I_I("_mm_abs_epi32", _mm_abs_epi32, _mm_load_si128((__m128i*)src_int));

38


39
  // TODO _mm_alignr_epi8

40


41
  SETCHART("horizontal add");

42
  BINARYOP_I_II("_mm_hadd_epi16", _mm_hadd_epi16, _mm_load_si128((__m128i*)src_int), _mm_load_si128((__m128i*)src2_int));

43
  BINARYOP_I_II("_mm_hadd_epi32", _mm_hadd_epi32, _mm_load_si128((__m128i*)src_int), _mm_load_si128((__m128i*)src2_int));

44
  BINARYOP_I_II("_mm_hadds_epi16", _mm_hadds_epi16, _mm_load_si128((__m128i*)src_int), _mm_load_si128((__m128i*)src2_int));

45


46
  SETCHART("horizontal sub");

47
  BINARYOP_I_II("_mm_hsub_epi16", _mm_hsub_epi16, _mm_load_si128((__m128i*)src_int), _mm_load_si128((__m128i*)src2_int));

48
  BINARYOP_I_II("_mm_hsub_epi32", _mm_hsub_epi32, _mm_load_si128((__m128i*)src_int), _mm_load_si128((__m128i*)src2_int));

49
  BINARYOP_I_II("_mm_hsubs_epi16", _mm_hsubs_epi16, _mm_load_si128((__m128i*)src_int), _mm_load_si128((__m128i*)src2_int));

50


51
  SETCHART("mul & shuffle");

52
  BINARYOP_I_II("_mm_maddubs_epi16", _mm_maddubs_epi16, _mm_load_si128((__m128i*)src_int), _mm_load_si128((__m128i*)src2_int));

53
  BINARYOP_I_II("_mm_mulhrs_epi16", _mm_mulhrs_epi16, _mm_load_si128((__m128i*)src_int), _mm_load_si128((__m128i*)src2_int));

54
  BINARYOP_I_II("_mm_shuffle_epi8", _mm_shuffle_epi8, _mm_load_si128((__m128i*)src_int), _mm_load_si128((__m128i*)src2_int));

55


56
  SETCHART("sign");

57
  BINARYOP_I_II("_mm_sign_epi8", _mm_sign_epi8, _mm_load_si128((__m128i*)src_int), _mm_load_si128((__m128i*)src2_int));

58
  BINARYOP_I_II("_mm_sign_epi16", _mm_sign_epi16, _mm_load_si128((__m128i*)src_int), _mm_load_si128((__m128i*)src2_int));

59
  BINARYOP_I_II("_mm_sign_epi32", _mm_sign_epi32, _mm_load_si128((__m128i*)src_int), _mm_load_si128((__m128i*)src2_int));

60


61
  // Benchmarks end:

62
  printf("]}\n");

63
}

64


65