flywithu/ntel-gpu-tools
1
0
Fork 0
mirror of https://github.com/tiagovignatti/intel-gpu-tools.git synced 2025-06-09 08:56:11 +00:00
Code Issues Releases Wiki Activity

gem_exec_lut_handle: Use linear regression analysis

Browse Source 
Simultaneously make the results more robust and compact by performing a
linear regression to compute the amount of time required to perform the
exec array walk and the relocations.
This commit is contained in:
Chris Wilson 2013-08-13 23:51:09 +01:00
parent 629f48fa0e
commit 6a746c82ab
1 changed files with 36 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

43
tests/gem_exec_lut_handle.c
View File

@ -148,27 +148,56 @@ int main(int argc, char **argv)
for (p = pass; p->name != NULL; p++) { for (p = pass; p->name != NULL; p++) {
for (n = 1; n <= MAX_NUM_EXEC; n *= 2) { for (n = 1; n <= MAX_NUM_EXEC; n *= 2) {
for (m = 1; m <= MAX_NUM_RELOC; m *= 2) { double elapsed[16][2];
double s_x, s_y, s_xx, s_xy;
double A, B;
int i, j;
for (i = 0, m = 1; m <= MAX_NUM_RELOC; m *= 2, i++) {
struct timeval start, end; struct timeval start, end;
double elapsed[2];
gettimeofday(&start, NULL); gettimeofday(&start, NULL);
for (count = 0; count < 1000; count++) for (count = 0; count < 1000; count++)
do_or_die(exec(fd, n, m, 0 | p->flags)); do_or_die(exec(fd, n, m, 0 | p->flags));
gettimeofday(&end, NULL); gettimeofday(&end, NULL);
gem_sync(fd, gem_exec[MAX_NUM_EXEC].handle); gem_sync(fd, gem_exec[MAX_NUM_EXEC].handle);
elapsed[0] = ELAPSED(&start, &end) / 1000.; elapsed[i][0] = ELAPSED(&start, &end);
gettimeofday(&start, NULL); gettimeofday(&start, NULL);
for (count = 0; count < 1000; count++) for (count = 0; count < 1000; count++)
do_or_die(exec(fd, n, m, USE_LUT | p->flags)); do_or_die(exec(fd, n, m, USE_LUT | p->flags));
gettimeofday(&end, NULL); gettimeofday(&end, NULL);
gem_sync(fd, gem_exec[MAX_NUM_EXEC].handle); gem_sync(fd, gem_exec[MAX_NUM_EXEC].handle);
elapsed[1] = ELAPSED(&start, &end) / 1000.; elapsed[i][1] = ELAPSED(&start, &end);
printf("%s: buffer_count=%d, reloc_count=%d: old=%f us, lut=%f us\n",
p->name, n, m, elapsed[0], elapsed[1]);
} }
printf("%s: buffers=%4d:", p->name, n);
s_x = s_y = s_xx = s_xy = 0;
for (j = 0; j < i; j++) {
int k = 1 << j;
s_x += k;
s_y += elapsed[j][0];
s_xx += k * k;
s_xy += k * elapsed[j][0];
}
B = (s_xy - s_x * s_y / j) / (s_xx - s_x * s_x / j);
A = s_y / j - B * s_x / j;
printf(" old=%7.0f + %.1f*reloc,", A, B);
s_x = s_y = s_xx = s_xy = 0;
for (j = 0; j < i; j++) {
int k = 1 << j;
s_x += k;
s_y += elapsed[j][1];
s_xx += k * k;
s_xy += k * elapsed[j][1];
}
B = (s_xy - s_x * s_y / j) / (s_xx - s_x * s_x / j);
A = s_y / j - B * s_x / j;
printf(" lut=%7.0f + %.1f*reloc (ns)", A, B);
printf("\n");
} }
} }