flywithu/ntel-gpu-tools
1
0
Fork 0
mirror of https://github.com/tiagovignatti/intel-gpu-tools.git synced 2025-06-21 23:06:29 +00:00
Code Issues Releases Wiki Activity

benchmarks/gem_blt: Fixup a couple of non-llc foibles

Browse Source 
When extending the batch for multiple copies, we need to remember to
flag it as being in the CPU write domain so that the new values get
flushed out to main memory before execution. We also have to be careful
not to specify NO_RELOC for the extended batch as the execobjects will
have been updated but we write the wrong presumed offsets. Subsequent
iterations will be correct and we can tell the kernel then to skip the
relocations entirely.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
This commit is contained in:
Chris Wilson 2015-11-12 10:50:17 +00:00
parent 143a216058
commit 4c14aa18c1
1 changed files with 11 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
benchmarks/gem_blt.c
View File

@ -166,7 +166,7 @@ static int run(int object, int batch, int count, int set, int reps)
struct drm_i915_gem_exec_object2 exec[3]; struct drm_i915_gem_exec_object2 exec[3];
struct drm_i915_gem_relocation_entry *reloc; struct drm_i915_gem_relocation_entry *reloc;
uint32_t *buf, handle, src, dst; uint32_t *buf, handle, src, dst;
int fd, len, gen, size; int fd, len, gen, size, nreloc;
int ring; int ring;
size = ALIGN(batch * 64, 4096); size = ALIGN(batch * 64, 4096);
@ -183,17 +183,18 @@ static int run(int object, int batch, int count, int set, int reps)
dst = gem_create(fd, object); dst = gem_create(fd, object);
len = gem_linear_blt(fd, buf, 0, 0, 1, object, reloc); len = gem_linear_blt(fd, buf, 0, 0, 1, object, reloc);
if (has_64bit_reloc)
nreloc = len > 56 ? 4 : 2;
else
nreloc = len > 40 ? 4 : 2;
memset(exec, 0, sizeof(exec)); memset(exec, 0, sizeof(exec));
exec[0].handle = src; exec[0].handle = src;
exec[1].handle = dst; exec[1].handle = dst;
exec[2].handle = handle; exec[2].handle = handle;
if (has_64bit_reloc)
exec[2].relocation_count = len > 56 ? 4 : 2;
else
exec[2].relocation_count = len > 40 ? 4 : 2;
exec[2].relocs_ptr = (uintptr_t)reloc; exec[2].relocs_ptr = (uintptr_t)reloc;
exec[2].relocation_count = nreloc;
ring = 0; ring = 0;
if (gen >= 6) if (gen >= 6)
@ -205,7 +206,6 @@ static int run(int object, int batch, int count, int set, int reps)
execbuf.batch_len = len; execbuf.batch_len = len;
execbuf.flags = ring; execbuf.flags = ring;
execbuf.flags |= LOCAL_I915_EXEC_HANDLE_LUT; execbuf.flags |= LOCAL_I915_EXEC_HANDLE_LUT;
execbuf.flags |= LOCAL_I915_EXEC_NO_RELOC;
if (__gem_execbuf(fd, &execbuf)) { if (__gem_execbuf(fd, &execbuf)) {
gem_set_domain(fd, handle, I915_GEM_DOMAIN_CPU, I915_GEM_DOMAIN_CPU); gem_set_domain(fd, handle, I915_GEM_DOMAIN_CPU, I915_GEM_DOMAIN_CPU);
@ -217,24 +217,25 @@ static int run(int object, int batch, int count, int set, int reps)
gem_sync(fd, handle); gem_sync(fd, handle);
if (batch > 1) { if (batch > 1) {
int nreloc = exec[2].relocation_count;
if (execbuf.flags & LOCAL_I915_EXEC_HANDLE_LUT) { if (execbuf.flags & LOCAL_I915_EXEC_HANDLE_LUT) {
src = 0; src = 0;
dst = 1; dst = 1;
} }
gem_set_domain(fd, handle, I915_GEM_DOMAIN_CPU, I915_GEM_DOMAIN_CPU);
for (int i = 1; i < batch; i++) { for (int i = 1; i < batch; i++) {
len = gem_linear_blt(fd, buf, len - 8, len = gem_linear_blt(fd, buf, len - 8,
src, dst, object, src, dst, object,
reloc + exec[2].relocation_count); reloc + nreloc * i);
exec[2].relocation_count += nreloc;
} }
exec[2].relocation_count = nreloc * batch;
execbuf.batch_len = len; execbuf.batch_len = len;
gem_execbuf(fd, &execbuf); gem_execbuf(fd, &execbuf);
gem_sync(fd, handle); gem_sync(fd, handle);
} }
if (execbuf.flags & LOCAL_I915_EXEC_HANDLE_LUT)
execbuf.flags |= LOCAL_I915_EXEC_NO_RELOC;
while (reps--) { while (reps--) {
double min = HUGE_VAL; double min = HUGE_VAL;