On Mon, May 02, 2022 at 02:14:02PM -0700, Matt Roper wrote:
On Mon, May 02, 2022 at 02:03:28PM -0700, Lucas De Marchi wrote:
On Mon, May 02, 2022 at 09:34:09AM -0700, Matt Roper wrote:
> From: Ayaz A Siddiqui <ayaz.siddiqui@xxxxxxxxx>
...
> @@ -2002,11 +2002,18 @@ engine_fake_wa_init(struct intel_engine_cs *engine, struct i915_wa_list *wal)
> * Streamers on Gen12 onward platforms.
> */
> if (GRAPHICS_VER(engine->i915) >= 12) {
> - mocs = engine->gt->mocs.uc_index;
> + if (HAS_L3_CCS_READ(engine->i915) &&
> + engine->class == COMPUTE_CLASS)
> + mocs_r = engine->gt->mocs.wb_index;
> + else
> + mocs_r = engine->gt->mocs.uc_index;
shouldn't we add a warning in get_mocs_settings() if HAS_L3_CCS_READ(engine->i915)
and mocs.wb_index is 0 (since index 0 shouldn't really be used in latest
platforms)?
We should be careful about that assumption...index 0 is valid on DG2
today, although HAS_L3_CCS_READ() doesn't apply there. And a couple
platforms in the future we're also going to have index 0 being valid on
a platform where HAS_L3_CCS_READ() is true (bspec 71582). Index 0 would
still be the wrong entry to pick for WB behavior there, but it is a
legitimate entry in general.
ok, but comment is more about "forgetting to initialize it in
get_mocs_settings() and then using it here". Using 0 as "it was not
initialized" may be an easy way to do that.
Lucas De Marchi
