[Mesa-dev] [PATCH 10/11] radeonsi: don't enable DCC statistics gathering for small surfaces

[Marek Olšák]

From: Marek Olšák <marek.olsak at amd.com>

---
 src/gallium/drivers/radeonsi/si_clear.c | 30 +++++++++++++------------
 1 file changed, 16 insertions(+), 14 deletions(-)

diff --git a/src/gallium/drivers/radeonsi/si_clear.c b/src/gallium/drivers/radeonsi/si_clear.c
index ef843966f2a..f366de32d0f 100644
--- a/src/gallium/drivers/radeonsi/si_clear.c
+++ b/src/gallium/drivers/radeonsi/si_clear.c
@@ -435,46 +435,48 @@ static void si_do_fast_color_clear(struct si_context *sctx,
 		 */
 		if (tex->buffer.b.is_shared &&
 		    !(tex->buffer.external_usage & PIPE_HANDLE_USAGE_EXPLICIT_FLUSH))
 			continue;
 
 		if (sctx->chip_class <= VI &&
 		    tex->surface.u.legacy.level[0].mode == RADEON_SURF_MODE_1D &&
 		    !sctx->screen->info.htile_cmask_support_1d_tiling)
 			continue;
 
-		/* Fast clear is the most appropriate place to enable DCC for
-		 * displayable surfaces.
-		 */
-		vi_separate_dcc_try_enable(sctx, tex);
-
-		/* RB+ isn't supported with a CMASK clear only on Stoney,
-		 * so all clears are considered to be hypothetically slow
-		 * clears, which is weighed when determining whether to
-		 * enable separate DCC.
-		 */
-		if (tex->dcc_gather_statistics &&
-		    sctx->family == CHIP_STONEY)
-			tex->num_slow_clears++;
-
 		bool need_decompress_pass = false;
 
 		/* Use a slow clear for small surfaces where the cost of
 		 * the eliminate pass can be higher than the benefit of fast
 		 * clear. The closed driver does this, but the numbers may differ.
 		 *
 		 * This helps on both dGPUs and APUs, even small APUs like Mullins.
 		 */
 		bool too_small = tex->buffer.b.b.nr_samples <= 1 &&
 				 tex->buffer.b.b.width0 *
 				 tex->buffer.b.b.height0 <= 512 * 512;
 
+		/* Fast clear is the most appropriate place to enable DCC for
+		 * displayable surfaces.
+		 */
+		if (!too_small) {
+			vi_separate_dcc_try_enable(sctx, tex);
+
+			/* RB+ isn't supported with a CMASK clear only on Stoney,
+			 * so all clears are considered to be hypothetically slow
+			 * clears, which is weighed when determining whether to
+			 * enable separate DCC.
+			 */
+			if (tex->dcc_gather_statistics &&
+			    sctx->family == CHIP_STONEY)
+				tex->num_slow_clears++;
+		}
+
 		/* Try to clear DCC first, otherwise try CMASK. */
 		if (vi_dcc_enabled(tex, 0)) {
 			uint32_t reset_value;
 			bool eliminate_needed;
 
 			if (sctx->screen->debug_flags & DBG(NO_DCC_CLEAR))
 				continue;
 
 			/* This can only occur with MSAA. */
 			if (sctx->chip_class == VI &&
-- 
2.17.1