flywithu/ntel-gpu-tools
1
0
Fork 0
mirror of https://github.com/tiagovignatti/intel-gpu-tools.git synced 2025-06-23 07:46:15 +00:00
Code Issues Releases Wiki Activity

benchmarks/gem_mmap: Convert to run over a fixed period

Browse Source 
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
This commit is contained in:
Chris Wilson 2016-01-02 16:07:34 +00:00
parent 9b90234414
commit e21368c53a
1 changed files with 41 additions and 21 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

62
benchmarks/gem_mmap.c
View File

@ -46,10 +46,10 @@
#define OBJECT_SIZE (1<<23) #define OBJECT_SIZE (1<<23)
static uint64_t elapsed(const struct timespec *start, static double elapsed(const struct timespec *start,
const struct timespec *end) const struct timespec *end)
{ {
return 1000000000ULL*(end->tv_sec - start->tv_sec) + (end->tv_nsec - start->tv_nsec); return (end->tv_sec - start->tv_sec) + 1e-9*(end->tv_nsec - start->tv_nsec);
} }
int main(int argc, char **argv) int main(int argc, char **argv)
@ -58,11 +58,13 @@ int main(int argc, char **argv)
enum map {CPU, GTT, WC} map = CPU; enum map {CPU, GTT, WC} map = CPU;
enum dir {READ, WRITE, CLEAR, FAULT} dir = READ; enum dir {READ, WRITE, CLEAR, FAULT} dir = READ;
int tiling = I915_TILING_NONE; int tiling = I915_TILING_NONE;
struct timespec start, end;
void *buf = malloc(OBJECT_SIZE); void *buf = malloc(OBJECT_SIZE);
uint32_t handle; uint32_t handle;
void *ptr, *src, *dst; void *ptr, *src, *dst;
int reps = 13; int reps = 1;
int c, size; int loops;
int c;
while ((c = getopt (argc, argv, "m:d:r:t:")) != -1) { while ((c = getopt (argc, argv, "m:d:r:t:")) != -1) {
switch (c) { switch (c) {
@ -140,38 +142,56 @@ int main(int argc, char **argv)
dst = ptr; dst = ptr;
} }
for (size = 1; size <= OBJECT_SIZE; size <<= 1) { clock_gettime(CLOCK_MONOTONIC, &start);
igt_stats_t stats; switch (dir) {
int n; case CLEAR:
case FAULT:
memset(dst, 0, OBJECT_SIZE);
break;
default:
memcpy(dst, src, OBJECT_SIZE);
break;
}
clock_gettime(CLOCK_MONOTONIC, &end);
igt_stats_init_with_size(&stats, reps); loops = 2 / elapsed(&start, &end);
while (reps--) {
for (n = 0; n < reps; n++) { clock_gettime(CLOCK_MONOTONIC, &start);
struct timespec start, end; for (c = 0; c < loops; c++) {
int page; int page;
clock_gettime(CLOCK_MONOTONIC, &start);
switch (dir) { switch (dir) {
case CLEAR: case CLEAR:
memset(dst, 0, size); memset(dst, 0, OBJECT_SIZE);
break; break;
case FAULT: case FAULT:
munmap(ptr, OBJECT_SIZE);
switch (map) {
case CPU:
ptr = gem_mmap__cpu(fd, handle, 0, OBJECT_SIZE, PROT_WRITE);
break;
case GTT:
ptr = gem_mmap__gtt(fd, handle, OBJECT_SIZE, PROT_WRITE);
break;
case WC:
ptr = gem_mmap__wc(fd, handle, 0, OBJECT_SIZE, PROT_WRITE);
break;
default:
abort();
}
for (page = 0; page < OBJECT_SIZE; page += 4096) { for (page = 0; page < OBJECT_SIZE; page += 4096) {
uint32_t *x = (uint32_t *)ptr + page/4; uint32_t *x = (uint32_t *)ptr + page/4;
__asm__ __volatile__("": : :"memory");
page += *x; /* should be zero! */ page += *x; /* should be zero! */
} }
break; break;
default: default:
memcpy(dst, src, size); memcpy(dst, src, OBJECT_SIZE);
break; break;
} }
clock_gettime(CLOCK_MONOTONIC, &end);
igt_stats_push(&stats, elapsed(&start, &end));
} }
clock_gettime(CLOCK_MONOTONIC, &end);
printf("%7.3f\n", igt_stats_get_trimean(&stats)/1000); printf("%7.3f\n", OBJECT_SIZE / elapsed(&start, &end) * loops / (1024*1024));
igt_stats_fini(&stats);
} }
return 0; return 0;