radeonsi: fully enable 2x DCC MSAA for array and non-array textures

The clear code is exactly the same as for 1 sample buffers -
just clear the whole thing.

Reviewed-by: Nicolai Hähnle <[email protected]>

4 files changed, 20 insertions, 14 deletions

diff --git a/src/gallium/drivers/radeonsi/si_clear.c b/src/gallium/drivers/radeonsi/si_clear.c
index b08a9558b4d..f2df9483c12 100644
--- a/src/gallium/drivers/radeonsi/si_clear.c
+++ b/src/gallium/drivers/radeonsi/si_clear.c
@@ -211,19 +211,20 @@ void vi_dcc_clear_level(struct si_context *sctx,
 	if (sctx->chip_class >= GFX9) {
 		/* Mipmap level clears aren't implemented. */
 		assert(rtex->resource.b.b.last_level == 0);
-		/* MSAA needs a different clear size. */
-		assert(rtex->resource.b.b.nr_samples <= 1);
+		/* 4x and 8x MSAA needs a sophisticated compute shader for
+		 * the clear. See AMDVLK. */
+		assert(rtex->resource.b.b.nr_samples <= 2);
 		clear_size = rtex->surface.dcc_size;
 	} else {
 		unsigned num_layers = util_num_layers(&rtex->resource.b.b, level);
 
 		/* If this is 0, fast clear isn't possible. (can occur with MSAA) */
 		assert(rtex->surface.u.legacy.level[level].dcc_fast_clear_size);
-		/* Layered MSAA DCC fast clears need to clear dcc_fast_clear_size
-		 * bytes for each layer. This is not currently implemented, and
-		 * therefore MSAA DCC isn't even enabled with multiple layers.
+		/* Layered 4x and 8x MSAA DCC fast clears need to clear
+		 * dcc_fast_clear_size bytes for each layer. A compute shader
+		 * would be more efficient than separate per-layer clear operations.
 		 */
-		assert(rtex->resource.b.b.nr_samples <= 1 || num_layers == 1);
+		assert(rtex->resource.b.b.nr_samples <= 2 || num_layers == 1);
 
 		dcc_offset += rtex->surface.u.legacy.level[level].dcc_offset;
 		clear_size = rtex->surface.u.legacy.level[level].dcc_fast_clear_size *
diff --git a/src/gallium/drivers/radeonsi/si_pipe.c b/src/gallium/drivers/radeonsi/si_pipe.c
index 327dd7c4242..3eb89ff0287 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.c
+++ b/src/gallium/drivers/radeonsi/si_pipe.c
@@ -90,7 +90,6 @@ static const struct debug_named_value debug_options[] = {
 	{ "nodccclear", DBG(NO_DCC_CLEAR), "Disable DCC fast clear." },
 	{ "nodccfb", DBG(NO_DCC_FB), "Disable separate DCC on the main framebuffer" },
 	{ "nodccmsaa", DBG(NO_DCC_MSAA), "Disable DCC for MSAA" },
-	{ "dccmsaa", DBG(DCC_MSAA), "Enable DCC for MSAA" },
 	{ "nofmask", DBG(NO_FMASK), "Disable MSAA compression" },
 
 	/* Tests: */
@@ -984,9 +983,7 @@ struct pipe_screen *radeonsi_screen_create(struct radeon_winsys *ws,
 	}
 
 	sscreen->dcc_msaa_allowed =
-		!(sscreen->debug_flags & DBG(NO_DCC_MSAA)) &&
-		(sscreen->debug_flags & DBG(DCC_MSAA) ||
-		 sscreen->info.chip_class == VI);
+		!(sscreen->debug_flags & DBG(NO_DCC_MSAA));
 
 	sscreen->cpdma_prefetch_writes_memory = sscreen->info.chip_class <= VI;
 
diff --git a/src/gallium/drivers/radeonsi/si_pipe.h b/src/gallium/drivers/radeonsi/si_pipe.h
index 351c9f4cd38..f73e0d1aac3 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.h
+++ b/src/gallium/drivers/radeonsi/si_pipe.h
@@ -154,7 +154,6 @@ enum {
 	DBG_NO_DCC_CLEAR,
 	DBG_NO_DCC_FB,
 	DBG_NO_DCC_MSAA,
-	DBG_DCC_MSAA,
 	DBG_NO_FMASK,
 
 	/* Tests: */
diff --git a/src/gallium/drivers/radeonsi/si_texture.c b/src/gallium/drivers/radeonsi/si_texture.c
index 0a2939bdd16..9ff0ba3e9a8 100644
--- a/src/gallium/drivers/radeonsi/si_texture.c
+++ b/src/gallium/drivers/radeonsi/si_texture.c
@@ -268,9 +268,18 @@ static int si_init_surface(struct si_screen *sscreen,
 	if (sscreen->info.chip_class >= VI &&
 	    (ptex->flags & SI_RESOURCE_FLAG_DISABLE_DCC ||
 	     ptex->format == PIPE_FORMAT_R9G9B9E5_FLOAT ||
-	     /* DCC MSAA array textures are disallowed due to incomplete clear impl. */
-	     (ptex->nr_samples >= 2 &&
-	      (!sscreen->dcc_msaa_allowed || ptex->array_size > 1))))
+	     (ptex->nr_samples >= 2 && !sscreen->dcc_msaa_allowed)))
+		flags |= RADEON_SURF_DISABLE_DCC;
+
+	/* VI: DCC clear for 4x and 8x MSAA array textures unimplemented. */
+	if (sscreen->info.chip_class == VI &&
+	    ptex->nr_samples >= 4 &&
+	    ptex->array_size > 1)
+		flags |= RADEON_SURF_DISABLE_DCC;
+
+	/* GFX9: DCC clear for 4x and 8x MSAA textures unimplemented. */
+	if (sscreen->info.chip_class >= GFX9 &&
+	    ptex->nr_samples >= 4)
 		flags |= RADEON_SURF_DISABLE_DCC;
 
 	if (ptex->bind & PIPE_BIND_SCANOUT || is_scanout) {