We were pointing the color calc and depth stencil states blindly
to an offset of 1k from bb start. This was foolhardy as it collides
with other state in the batch and results in a wrecked state upload.
Chris noticed that with snb gt1, it takes 10 seconds for renderstate batch
to complete. However pointing the states to a known valid 64 aligned
zero blocks didn't work to reduce the upload time either.
Cave in and omit uploading color calc and depth stencil states.
Reported-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
Cc: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
Signed-off-by: Mika Kuoppala <mika.kuoppala@xxxxxxxxx>
---
tools/null_state_gen/intel_renderstate_gen6.c | 6 ++++--
1 file changed, 4 insertions(+), 2 deletions(-)
diff --git a/tools/null_state_gen/intel_renderstate_gen6.c b/tools/null_state_gen/intel_renderstate_gen6.c
index 5c1b7f97..c64e17fd 100644
--- a/tools/null_state_gen/intel_renderstate_gen6.c
+++ b/tools/null_state_gen/intel_renderstate_gen6.c
@@ -221,8 +221,10 @@ gen6_emit_cc(struct intel_batchbuffer *batch, uint32_t blend)
{
OUT_BATCH(GEN6_3DSTATE_CC_STATE_POINTERS | (4 - 2));
OUT_BATCH_STATE_OFFSET(blend | 1);
- OUT_BATCH(1024 | 1);
- OUT_BATCH(1024 | 1);
+ /* no color calc */
+ OUT_BATCH(0);
+ /* no depth stencil */
+ OUT_BATCH(0);
}
static void
--
2.11.0
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
