[igt-dev] [PATCH i-g-t] igt/gem_ppgtt: Unroll 32b memset

* [igt-dev] [PATCH i-g-t] igt/gem_ppgtt: Unroll 32b memset
@ 2018-10-29 19:37 Chris Wilson
  2018-10-30  2:46 ` [igt-dev] ✓ Fi.CI.BAT: success for " Patchwork
                   ` (2 more replies)
  0 siblings, 3 replies; 4+ messages in thread
From: Chris Wilson @ 2018-10-29 19:37 UTC (permalink / raw)
  To: igt-dev

Manually unroll the 32b memset in create_bo() to set a cacheline at a
time, for a 2x speed improve of the whole test.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 tests/i915/gem_ppgtt.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

diff --git a/tests/i915/gem_ppgtt.c b/tests/i915/gem_ppgtt.c
index af5e3e07..86fe59f2 100644
--- a/tests/i915/gem_ppgtt.c
+++ b/tests/i915/gem_ppgtt.c
@@ -47,16 +47,20 @@
 static drm_intel_bo *create_bo(drm_intel_bufmgr *bufmgr,
 			       uint32_t pixel)
 {
+	uint64_t value = (uint64_t)pixel << 32 | pixel, *v;
 	drm_intel_bo *bo;
-	uint32_t *v;
 
 	bo = drm_intel_bo_alloc(bufmgr, "surface", SIZE, 4096);
 	igt_assert(bo);
 
 	do_or_die(drm_intel_bo_map(bo, 1));
 	v = bo->virtual;
-	for (int i = 0; i < SIZE/4; i++)
-		v[i] = pixel;
+	for (int i = 0; i < SIZE / sizeof(value); i += 8) {
+		v[i + 0] = value; v[i + 1] = value;
+		v[i + 2] = value; v[i + 3] = value;
+		v[i + 4] = value; v[i + 5] = value;
+		v[i + 6] = value; v[i + 7] = value;
+	}
 	drm_intel_bo_unmap(bo);
 
 	return bo;
-- 
2.19.1

_______________________________________________
igt-dev mailing list
igt-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/igt-dev

^ permalink raw reply related	[flat|nested] 4+ messages in thread