fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2025-12-23 17:40:11 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

panfrost: Skip rasterizer discard draws without side effects

Browse source 
Minor optimization, but more importantly fixes an interaction of IDVS
with rasterizer discard.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14154>
This commit is contained in:
Alyssa Rosenzweig 2021-12-10 14:21:02 -05:00 • committed by Marge Bot
parent f5412409db
commit 79356b2e5f
1 changed files with 16 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

16
src/gallium/drivers/panfrost/pan_cmdstream.c
View file

@ -3106,6 +3106,22 @@ panfrost_draw_vbo(struct pipe_context *pipe,
/* Do some common setup */ /* Do some common setup */
struct panfrost_batch *batch = panfrost_get_batch_for_fbo(ctx); struct panfrost_batch *batch = panfrost_get_batch_for_fbo(ctx);
/* If rasterization discard is enabled but the vertex shader does not
* have side effects (including transform feedback), skip the draw
* altogether. This is always an optimization. Additionally, this is
* required for Index-Driven Vertex Shading, since IDVS always
* rasterizes. The compiler will not use IDVS if the vertex shader has
* side effects. So the only problem case is rasterizer discard with a
* shader without side effects -- but these draws are useless.
*/
if (panfrost_batch_skip_rasterization(batch)) {
struct panfrost_shader_state *vs =
panfrost_get_shader_state(ctx, PIPE_SHADER_VERTEX);
if (!vs->info.writes_global)
return;
}
/* Don't add too many jobs to a single batch. Hardware has a hard limit /* Don't add too many jobs to a single batch. Hardware has a hard limit
* of 65536 jobs, but we choose a smaller soft limit (arbitrary) to * of 65536 jobs, but we choose a smaller soft limit (arbitrary) to
* avoid the risk of timeouts. This might not be a good idea. */ * avoid the risk of timeouts. This might not be a good idea. */