| Message ID | 20190322092155.1656-7-chris@chris-wilson.co.uk (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [i-g-t,01/24] i915/gem_exec_latency: Measure the latency of context switching | expand |
On 22/03/2019 09:21, Chris Wilson wrote: > To make the demonstration of the cheeky preemption more impactful, make > the second context a nop to contrast the first being 1024 > MI_STORE_DWORD_IMM. Then if we execute and wait on the second context > before executing the first, the client latency is even more drastically > reduced. > > To more clearly show any effect on wait reordering, measure the > alternative path and present both. > > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> > --- > tests/i915/gem_sync.c | 40 +++++++++++++++++++++++++++++----------- > 1 file changed, 29 insertions(+), 11 deletions(-) > > diff --git a/tests/i915/gem_sync.c b/tests/i915/gem_sync.c > index fb209977d..3e4feff32 100644 > --- a/tests/i915/gem_sync.c > +++ b/tests/i915/gem_sync.c > @@ -651,7 +651,7 @@ switch_ring(int fd, unsigned ring, int num_children, int timeout) > struct drm_i915_gem_relocation_entry reloc[1024]; > struct drm_i915_gem_execbuffer2 execbuf; > } contexts[2]; > - double start, elapsed; > + double elapsed, baseline; > unsigned long cycles; > > for (int i = 0; i < ARRAY_SIZE(contexts); i++) { > @@ -679,7 +679,7 @@ switch_ring(int fd, unsigned ring, int num_children, int timeout) > c->object[1].handle = gem_create(fd, sz); > > c->object[1].relocs_ptr = to_user_pointer(c->reloc); > - c->object[1].relocation_count = 1024; > + c->object[1].relocation_count = 1024 * i; > > batch = gem_mmap__cpu(fd, c->object[1].handle, 0, sz, > PROT_WRITE | PROT_READ); > @@ -688,7 +688,7 @@ switch_ring(int fd, unsigned ring, int num_children, int timeout) > > memset(c->reloc, 0, sizeof(c->reloc)); > b = batch; > - for (int r = 0; r < 1024; r++) { > + for (int r = 0; r < c->object[1].relocation_count; r++) { > uint64_t offset; > > c->reloc[r].presumed_offset = c->object[0].offset; > @@ -722,26 +722,44 @@ switch_ring(int fd, unsigned ring, int num_children, int timeout) > } > > cycles = 0; > - elapsed = 0; > - start = gettime(); > - do { > + baseline = 0; > + igt_until_timeout(timeout) { > do { > double this; > > - gem_execbuf(fd, &contexts[0].execbuf); > gem_execbuf(fd, &contexts[1].execbuf); > + gem_execbuf(fd, &contexts[0].execbuf); > > this = gettime(); > gem_sync(fd, contexts[1].object[1].handle); > - elapsed += gettime() - this; > + gem_sync(fd, contexts[0].object[1].handle); > + baseline += gettime() - this; > + } while (++cycles & 1023); > + } > + baseline /= cycles; > + > + cycles = 0; > + elapsed = 0; > + igt_until_timeout(timeout) { > + do { > + double this; > > + gem_execbuf(fd, &contexts[1].execbuf); > + gem_execbuf(fd, &contexts[0].execbuf); > + > + this = gettime(); > gem_sync(fd, contexts[0].object[1].handle); > + elapsed += gettime() - this; > + > + gem_sync(fd, contexts[1].object[1].handle); > } while (++cycles & 1023); > - } while ((gettime() - start) < timeout); > - igt_info("%s%sompleted %ld cycles: %.3f us\n", > + } > + elapsed /= cycles; > + > + igt_info("%s%sompleted %ld cycles: %.3f us, baseline %.3f us\n", > names[child % num_engines] ?: "", > names[child % num_engines] ? " c" : "C", > - cycles, elapsed*1e6/cycles); > + cycles, elapsed*1e6, baseline*1e6); > > for (int i = 0; i < ARRAY_SIZE(contexts); i++) { > gem_close(fd, contexts[i].object[1].handle); > Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Regards, Tvrtko
diff --git a/tests/i915/gem_sync.c b/tests/i915/gem_sync.c index fb209977d..3e4feff32 100644 --- a/tests/i915/gem_sync.c +++ b/tests/i915/gem_sync.c @@ -651,7 +651,7 @@ switch_ring(int fd, unsigned ring, int num_children, int timeout) struct drm_i915_gem_relocation_entry reloc[1024]; struct drm_i915_gem_execbuffer2 execbuf; } contexts[2]; - double start, elapsed; + double elapsed, baseline; unsigned long cycles; for (int i = 0; i < ARRAY_SIZE(contexts); i++) { @@ -679,7 +679,7 @@ switch_ring(int fd, unsigned ring, int num_children, int timeout) c->object[1].handle = gem_create(fd, sz); c->object[1].relocs_ptr = to_user_pointer(c->reloc); - c->object[1].relocation_count = 1024; + c->object[1].relocation_count = 1024 * i; batch = gem_mmap__cpu(fd, c->object[1].handle, 0, sz, PROT_WRITE | PROT_READ); @@ -688,7 +688,7 @@ switch_ring(int fd, unsigned ring, int num_children, int timeout) memset(c->reloc, 0, sizeof(c->reloc)); b = batch; - for (int r = 0; r < 1024; r++) { + for (int r = 0; r < c->object[1].relocation_count; r++) { uint64_t offset; c->reloc[r].presumed_offset = c->object[0].offset; @@ -722,26 +722,44 @@ switch_ring(int fd, unsigned ring, int num_children, int timeout) } cycles = 0; - elapsed = 0; - start = gettime(); - do { + baseline = 0; + igt_until_timeout(timeout) { do { double this; - gem_execbuf(fd, &contexts[0].execbuf); gem_execbuf(fd, &contexts[1].execbuf); + gem_execbuf(fd, &contexts[0].execbuf); this = gettime(); gem_sync(fd, contexts[1].object[1].handle); - elapsed += gettime() - this; + gem_sync(fd, contexts[0].object[1].handle); + baseline += gettime() - this; + } while (++cycles & 1023); + } + baseline /= cycles; + + cycles = 0; + elapsed = 0; + igt_until_timeout(timeout) { + do { + double this; + gem_execbuf(fd, &contexts[1].execbuf); + gem_execbuf(fd, &contexts[0].execbuf); + + this = gettime(); gem_sync(fd, contexts[0].object[1].handle); + elapsed += gettime() - this; + + gem_sync(fd, contexts[1].object[1].handle); } while (++cycles & 1023); - } while ((gettime() - start) < timeout); - igt_info("%s%sompleted %ld cycles: %.3f us\n", + } + elapsed /= cycles; + + igt_info("%s%sompleted %ld cycles: %.3f us, baseline %.3f us\n", names[child % num_engines] ?: "", names[child % num_engines] ? " c" : "C", - cycles, elapsed*1e6/cycles); + cycles, elapsed*1e6, baseline*1e6); for (int i = 0; i < ARRAY_SIZE(contexts); i++) { gem_close(fd, contexts[i].object[1].handle);
To make the demonstration of the cheeky preemption more impactful, make the second context a nop to contrast the first being 1024 MI_STORE_DWORD_IMM. Then if we execute and wait on the second context before executing the first, the client latency is even more drastically reduced. To more clearly show any effect on wait reordering, measure the alternative path and present both. Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> --- tests/i915/gem_sync.c | 40 +++++++++++++++++++++++++++++----------- 1 file changed, 29 insertions(+), 11 deletions(-)