Statistics
| Branch: | Tag: | Revision:

root / volk / lib / qa_16s_branch_4_state_8_aligned16.cc @ 4c6316e1

History | View | Annotate | Download (4 kB)

1
#include <volk/volk.h>
2
#include <qa_16s_branch_4_state_8_aligned16.h>
3
#include <cstdlib>
4
#include <ctime>
5


6
//test for ssse3



7


8
#ifndef LV_HAVE_SSSE3



9


10
void qa_16s_branch_4_state_8_aligned16::t1() {



11
  printf("ssse3 not available... no test performed\n");



12
}



13


14
#else



15


16
void qa_16s_branch_4_state_8_aligned16::t1() {



17
  const int num_iters = 1000000;



18
  const int vlen = 32;



19


20
  static char permute0[16]__attribute__((aligned(16))) = {0x0e, 0x0f, 0x0a, 0x0b, 0x04, 0x05, 0x00, 0x01, 0x0c, 0x0d, 0x08, 0x09, 0x06, 0x07, 0x02, 0x03};



21
  static char permute1[16]__attribute__((aligned(16))) = {0x0c, 0x0d, 0x08, 0x09, 0x06, 0x07, 0x02, 0x03, 0x0e, 0x0f, 0x0a, 0x0b, 0x04, 0x05, 0x00, 0x01};



22
  static char permute2[16]__attribute__((aligned(16))) = {0x02, 0x03, 0x06, 0x07, 0x08, 0x09, 0x0c, 0x0d, 0x00, 0x01, 0x04, 0x05, 0x0a, 0x0b, 0x0e, 0x0f};



23
  static char permute3[16]__attribute__((aligned(16))) = {0x00, 0x01, 0x04, 0x05, 0x0a, 0x0b, 0x0e, 0x0f, 0x02, 0x03, 0x06, 0x07, 0x08, 0x09, 0x0c, 0x0d};



24
  static char* permuters[4] = {permute0, permute1, permute2, permute3};



25
  



26
  unsigned int num_bytes = vlen << 1;



27


28
  volk_environment_init();



29
  clock_t start, end;



30
  double total;



31
  



32
  __VOLK_ATTR_ALIGNED(16) short target[vlen];



33
  __VOLK_ATTR_ALIGNED(16) short target2[vlen];



34
  __VOLK_ATTR_ALIGNED(16) short target3[vlen];



35
  



36
  __VOLK_ATTR_ALIGNED(16) short src0[vlen];



37
  __VOLK_ATTR_ALIGNED(16) short permute_indexes[vlen] =  {



38
7, 5, 2, 0, 6, 4, 3, 1, 6, 4, 3, 1, 7, 5, 2, 0, 1, 3, 4, 6, 0, 2, 5, 7, 0, 2, 5, 7, 1, 3, 4, 6 };



39
  __VOLK_ATTR_ALIGNED(16) short cntl0[vlen] = {



40
    0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000 };



41
  __VOLK_ATTR_ALIGNED(16) short cntl1[vlen] = {



42
    0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0xffff, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000 };



43
  __VOLK_ATTR_ALIGNED(16) short cntl2[vlen] = {



44
    0x0000, 0xffff, 0xffff, 0x0000, 0x0000, 0xffff, 0xffff, 0x0000, 0xffff, 0x0000, 0x0000, 0xffff, 0xffff, 0x0000, 0x0000, 0xffff, 0xffff, 0x0000, 0x0000, 0xffff, 0xffff, 0x0000, 0x0000, 0xffff, 0x0000, 0xffff, 0xffff, 0x0000, 0x0000, 0xffff, 0xffff, 0x0000 };



45
  __VOLK_ATTR_ALIGNED(16) short cntl3[vlen] =  {



46
    0xffff, 0xffff, 0x0000, 0x0000, 0xffff, 0xffff, 0x0000, 0x0000, 0x0000, 0x0000, 0xffff, 0xffff, 0x0000, 0x0000, 0xffff, 0xffff, 0xffff, 0xffff, 0x0000, 0x0000, 0xffff, 0xffff, 0x0000, 0x0000, 0x0000, 0x0000, 0xffff, 0xffff, 0x0000, 0x0000, 0xffff, 0xffff };



47
  __VOLK_ATTR_ALIGNED(16) short scalars[4] = {1, 2, 3, 4};



48
  



49
  



50


51
  for(int i = 0; i < vlen; ++i) {



52
    src0[i] = i;



53
    



54
  }



55
  



56


57
  printf("16s_branch_4_state_8_aligned\n");



58
  



59
  



60
  start = clock();



61
  for(int i = 0; i < num_iters; ++i) {



62
    volk_16s_permute_and_scalar_add_aligned16_manual(target, src0, permute_indexes, cntl0, cntl1, cntl2, cntl3, scalars, num_bytes, "sse2");



63
  }



64
  end = clock();



65
  



66
  total = (double)(end-start)/(double)CLOCKS_PER_SEC;



67


68
  printf("permute_and_scalar_add_time: %f\n", total);



69
  



70
  



71


72
  start = clock();



73
  for(int i = 0; i < num_iters; ++i) {



74
    volk_16s_branch_4_state_8_aligned16_manual(target2, src0, permuters, cntl2, cntl3, scalars, "ssse3");



75
  }



76
  end = clock();



77


78
  total = (double)(end-start)/(double)CLOCKS_PER_SEC;



79


80
  printf("branch_4_state_8_time, ssse3: %f\n", total);



81
  



82
  start = clock();



83
  for(int i = 0; i < num_iters; ++i) {



84
    volk_16s_branch_4_state_8_aligned16_manual(target3, src0, permuters, cntl2, cntl3, scalars, "generic");



85
  }



86
  end = clock();



87
  



88
  total = (double)(end-start)/(double)CLOCKS_PER_SEC;



89


90
  printf("permute_and_scalar_add_time, generic: %f\n", total);



91
  



92
  



93
  



94
  for(int i = 0; i < vlen; ++i) {



95
    printf("psa... %d, b4s8... %d\n", target[i], target3[i]);



96
  }



97
  



98
  for(int i = 0; i < vlen; ++i) {



99
    



100
    CPPUNIT_ASSERT(target[i] == target2[i]);



101
    CPPUNIT_ASSERT(target[i] == target3[i]);



102
  }



103
}



104


105


106
#endif