flywithu/ntel-gpu-tools
1
0
Fork 0
mirror of https://github.com/tiagovignatti/intel-gpu-tools.git synced 2025-06-09 17:06:14 +00:00
Code Issues Releases Wiki Activity

igt/gem_exec_lut_handle: Hide exec latency

Browse Source 
The first test tries to rewrite relocations in an active batch, which is
a useful test and measurement. However, the overhead of the exec may
dominate and so we want a measurement without that overhead as well.
Using a pool of batches should allow for the oldest to idle whilst we
setup the next (and so the wait should be non-existent).

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
This commit is contained in:
Chris Wilson 2015-04-14 19:02:19 +01:00
parent ecb03266b2
commit beddb3be3f
1 changed files with 31 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

38
tests/gem_exec_lut_handle.c
View File

@ -52,6 +52,7 @@ IGT_TEST_DESCRIPTION("Exercises the basic execbuffer using the handle LUT"
#define SKIP_RELOC 0x1
#define NO_RELOC 0x2
#define CYCLE_BATCH 0x4
int target[MAX_NUM_RELOC];
struct drm_i915_gem_exec_object2 gem_exec[MAX_NUM_EXEC+1];
@ -83,12 +84,14 @@ static int has_exec_lut(int fd)
igt_simple_main
{
uint32_t batch[2] = {MI_BATCH_BUFFER_END};
int fd, n, m, count;
uint32_t cycle[16];
int fd, n, m, count, c;
const struct {
const char *name;
unsigned int flags;
} pass[] = {
{ .name = "relocation", .flags = 0 },
{ .name = "cycle-relocation", .flags = CYCLE_BATCH },
{ .name = "skip-relocs", .flags = SKIP_RELOC },
{ .name = "no-relocs", .flags = SKIP_RELOC | NO_RELOC },
{ .name = NULL },
@ -102,8 +105,11 @@ igt_simple_main
for (n = 0; n < MAX_NUM_EXEC; n++)
gem_exec[n].handle = gem_create(fd, 4096);
gem_exec[n].handle = gem_create(fd, 4096);
gem_write(fd, gem_exec[n].handle, 0, batch, sizeof(batch));
for (n = 0; n < 16; n++) {
cycle[n] = gem_create(fd, 4096);
gem_write(fd, cycle[n], 0, batch, sizeof(batch));
}
gem_exec[MAX_NUM_EXEC].handle = cycle[0];
memset(gem_reloc, 0, sizeof(gem_reloc));
for (n = 0; n < MAX_NUM_RELOC; n++) {
@ -145,13 +151,22 @@ igt_simple_main
gem_execbuf(fd,&execbuf);
gettimeofday(&start, NULL);
for (count = 0; count < 1000; count++) {
if ((p->flags & SKIP_RELOC) == 0)
if ((p->flags & SKIP_RELOC) == 0) {
for (j = 0; j < m; j++)
gem_reloc[j].presumed_offset = 0;
if (p->flags & CYCLE_BATCH) {
c = (c + 1) % 16;
gem_exec[MAX_NUM_EXEC].handle = cycle[c];
}
}
gem_execbuf(fd, &execbuf);
}
gettimeofday(&end, NULL);
gem_sync(fd, gem_exec[MAX_NUM_EXEC].handle);
c = 16;
do
gem_sync(fd, cycle[--c]);
while (c != 0);
gem_exec[MAX_NUM_EXEC].handle = cycle[c];
elapsed[i][1] = ELAPSED(&start, &end);
execbuf.flags &= ~LOCAL_I915_EXEC_HANDLE_LUT;
@ -161,13 +176,22 @@ igt_simple_main
gem_execbuf(fd,&execbuf);
gettimeofday(&start, NULL);
for (count = 0; count < 1000; count++) {
if ((p->flags & SKIP_RELOC) == 0)
if ((p->flags & SKIP_RELOC) == 0) {
for (j = 0; j < m; j++)
gem_reloc[j].presumed_offset = 0;
if (p->flags & CYCLE_BATCH) {
c = (c + 1) % 16;
gem_exec[MAX_NUM_EXEC].handle = cycle[c];
}
}
gem_execbuf(fd, &execbuf);
}
gettimeofday(&end, NULL);
gem_sync(fd, gem_exec[MAX_NUM_EXEC].handle);
c = 16;
do
gem_sync(fd, cycle[--c]);
while (c != 0);
gem_exec[MAX_NUM_EXEC].handle = cycle[c];
elapsed[i][0] = ELAPSED(&start, &end);
}