aaron/ispc
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

added ispc timers w/o task

Browse Source
This commit is contained in:
Evghenii
2013-11-13 19:13:04 +01:00
parent e71259006c
commit 01df6ed4a9
3 changed files with 13 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

BIN
examples_cuda/sort/sort
View File

Binary file not shown.

24
examples_cuda/sort/sort.cpp
View File

@@ -92,7 +92,7 @@ int main (int argc, char *argv[])
srand (0); srand (0);
#if 0 #if 1
for (i = 0; i < m; i ++) for (i = 0; i < m; i ++)
{ {
for (j = 0; j < n; j ++) code [j] = random() % l; for (j = 0; j < n; j ++) code [j] = random() % l;
@@ -100,15 +100,20 @@ int main (int argc, char *argv[])
reset_and_start_timer(); reset_and_start_timer();
const double t0 = rtc(); const double t0 = rtc();
sort_ispc (n, code, order, 1); sort_ispc (n, code, order, 13*4*2+1);
tISPC1 += (rtc() - t0); //get_elapsed_mcycles(); tISPC2 += (rtc() - t0); //get_elapsed_mcycles();
if (argc != 3) if (argc != 3)
progressbar (i, m); progressbar (i, m);
} }
printf("[sort ispc]:\t[%.3f] million cycles\n", tISPC1); printf("[sort ispc + task]:\t[%.3f] million cycles\n", tISPC2);
for (int i = 0; i < n-1; i++)
{
assert(code[i+1] >= code[i]);
}
#endif #endif
srand (0); srand (0);
@@ -120,20 +125,15 @@ int main (int argc, char *argv[])
reset_and_start_timer(); reset_and_start_timer();
const double t0 = rtc(); const double t0 = rtc();
sort_ispc (n, code, order, 0); sort_ispc (n, code, order, 1);
tISPC2 += (rtc() - t0); // get_elapsed_mcycles(); tISPC1 += (rtc() - t0); // get_elapsed_mcycles();
if (argc != 3) if (argc != 3)
progressbar (i, m); progressbar (i, m);
} }
printf("[sort ispc + tasks]:\t[%.3f] million cycles\n", tISPC2); printf("[sort ispc]:\t[%.3f] million cycles\n", tISPC1);
for (int i = 0; i < n-1; i++)
{
assert(code[i+1] >= code[i]);
}
srand (0); srand (0);

2
examples_cuda/sort/sort_cu.cpp
View File

@@ -351,7 +351,7 @@ int main (int argc, char *argv[])
createContext(); createContext();
/*******************/ /*******************/
int ntask = 13*4*2; int ntask = 13*4;
devicePtr d_code = deviceMalloc(n*sizeof(int)); devicePtr d_code = deviceMalloc(n*sizeof(int));
devicePtr d_order = deviceMalloc(n*sizeof(int)); devicePtr d_order = deviceMalloc(n*sizeof(int));
devicePtr d_pair = deviceMalloc(n*2*sizeof(int)); devicePtr d_pair = deviceMalloc(n*2*sizeof(int));