radeonsi/gfx10: flush gfx cs on ngg -> legacy transition

with a sequence like this:

  glClear(STENCIL)
  glBeginTransformFeedback()
  ...
  glEndTransformFeedback()
  glClear(STENCIL)

The second clear sometimes may produce an unexpected result.

Calling si_flush_gfx_cs() when doing ngg -> legacy transition seems to be a
valid workaround (both for the synthetic reproducer and the real Blender bug).

Using flush flags or events (BOTTOM_OF_PIPE_TS, RESET_TO_LOWEST_VGT) didn't help.

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/2941
Cc: mesa-stable
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7750>
(cherry picked from commit 0b3bd7c516)
This commit is contained in:
Pierre-Eric Pelloux-Prayer 2020-11-23 13:48:42 +01:00 committed by Dylan Baker
parent 99aec40f08
commit 2a74f0ceb6
2 changed files with 7 additions and 2 deletions

View file

@ -2020,7 +2020,7 @@
"description": "radeonsi/gfx10: flush gfx cs on ngg -> legacy transition",
"nominated": true,
"nomination_type": 0,
"resolution": 0,
"resolution": 1,
"master_sha": null,
"because_sha": null
},

View file

@ -2954,8 +2954,13 @@ bool si_update_ngg(struct si_context *sctx)
* VGT_FLUSH is also emitted at the beginning of IBs when legacy GS ring
* pointers are set.
*/
if ((sctx->chip_class == GFX10 || sctx->family == CHIP_SIENNA_CICHLID) && !new_ngg)
if ((sctx->chip_class == GFX10 || sctx->family == CHIP_SIENNA_CICHLID) && !new_ngg) {
sctx->flags |= SI_CONTEXT_VGT_FLUSH;
if (sctx->chip_class == GFX10) {
/* Workaround for https://gitlab.freedesktop.org/mesa/mesa/-/issues/2941 */
si_flush_gfx_cs(sctx, RADEON_FLUSH_ASYNC_START_NEXT_GFX_IB_NOW, NULL);
}
}
sctx->ngg = new_ngg;
sctx->last_gs_out_prim = -1; /* reset this so that it gets updated */