radeonsi: add R600_DEBUG=nofmask to disable MSAA compression

For testing.

Tested-by: Dieter Nützel <[email protected]>

5 files changed, 17 insertions, 14 deletions

diff --git a/src/gallium/drivers/radeon/r600_pipe_common.h b/src/gallium/drivers/radeon/r600_pipe_common.h
index 79419036665..4df039d33a4 100644
--- a/src/gallium/drivers/radeon/r600_pipe_common.h
+++ b/src/gallium/drivers/radeon/r600_pipe_common.h
@@ -110,6 +110,7 @@ enum {
 	DBG_NO_DCC_FB,
 	DBG_NO_DCC_MSAA,
 	DBG_DCC_MSAA,
+	DBG_NO_FMASK,
 
 	/* Tests: */
 	DBG_TEST_DMA,
diff --git a/src/gallium/drivers/radeon/r600_texture.c b/src/gallium/drivers/radeon/r600_texture.c
index 50c5dbe3e1a..ae9623a829d 100644
--- a/src/gallium/drivers/radeon/r600_texture.c
+++ b/src/gallium/drivers/radeon/r600_texture.c
@@ -1268,12 +1268,13 @@ r600_texture_create_object(struct pipe_screen *screen,
 				r600_texture_allocate_htile(sscreen, rtex);
 		}
 	} else {
-		if (base->nr_samples > 1) {
-			if (!buf) {
-				r600_texture_allocate_fmask(sscreen, rtex);
-				r600_texture_allocate_cmask(sscreen, rtex);
-				rtex->cmask_buffer = &rtex->resource;
-			}
+		if (base->nr_samples > 1 &&
+		    !buf &&
+		    !(sscreen->debug_flags & DBG(NO_FMASK))) {
+			r600_texture_allocate_fmask(sscreen, rtex);
+			r600_texture_allocate_cmask(sscreen, rtex);
+			rtex->cmask_buffer = &rtex->resource;
+
 			if (!rtex->fmask.size || !rtex->cmask.size) {
 				FREE(rtex);
 				return NULL;
diff --git a/src/gallium/drivers/radeonsi/si_pipe.c b/src/gallium/drivers/radeonsi/si_pipe.c
index 1cc08c5feed..cbc8689044b 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.c
+++ b/src/gallium/drivers/radeonsi/si_pipe.c
@@ -88,6 +88,7 @@ static const struct debug_named_value debug_options[] = {
 	{ "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: */
 	{ "testdma", DBG(TEST_DMA), "Invoke SDMA tests and exit." },
diff --git a/src/gallium/drivers/radeonsi/si_pipe.h b/src/gallium/drivers/radeonsi/si_pipe.h
index 2053dcb9fcd..dbb04ed7e45 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.h
+++ b/src/gallium/drivers/radeonsi/si_pipe.h
@@ -285,6 +285,7 @@ struct si_framebuffer {
 	ubyte				nr_samples:5; /* at most 16xAA */
 	ubyte				log_samples:3; /* at most 4 = 16xAA */
 	ubyte				compressed_cb_mask;
+	ubyte				uncompressed_cb_mask;
 	ubyte				color_is_int8;
 	ubyte				color_is_int10;
 	ubyte				dirty_cbufs;
diff --git a/src/gallium/drivers/radeonsi/si_state.c b/src/gallium/drivers/radeonsi/si_state.c
index b4165a4669b..c9ed7256f04 100644
--- a/src/gallium/drivers/radeonsi/si_state.c
+++ b/src/gallium/drivers/radeonsi/si_state.c
@@ -2784,8 +2784,7 @@ static void si_set_framebuffer_state(struct pipe_context *ctx,
 	 *
 	 * Only flush and wait for CB if there is actually a bound color buffer.
 	 */
-	if (sctx->framebuffer.nr_samples <= 1 &&
-	    sctx->framebuffer.state.nr_cbufs)
+	if (sctx->framebuffer.uncompressed_cb_mask)
 		si_make_CB_shader_coherent(sctx, sctx->framebuffer.nr_samples,
 					   sctx->framebuffer.CB_has_shader_readable_metadata);
 
@@ -2829,6 +2828,7 @@ static void si_set_framebuffer_state(struct pipe_context *ctx,
 	sctx->framebuffer.color_is_int10 = 0;
 
 	sctx->framebuffer.compressed_cb_mask = 0;
+	sctx->framebuffer.uncompressed_cb_mask = 0;
 	sctx->framebuffer.nr_samples = util_framebuffer_get_num_samples(state);
 	sctx->framebuffer.log_samples = util_logbase2(sctx->framebuffer.nr_samples);
 	sctx->framebuffer.any_dst_linear = false;
@@ -2861,9 +2861,10 @@ static void si_set_framebuffer_state(struct pipe_context *ctx,
 		if (surf->color_is_int10)
 			sctx->framebuffer.color_is_int10 |= 1 << i;
 
-		if (rtex->fmask.size) {
+		if (rtex->fmask.size)
 			sctx->framebuffer.compressed_cb_mask |= 1 << i;
-		}
+		else
+			sctx->framebuffer.uncompressed_cb_mask |= 1 << i;
 
 		if (rtex->surface.is_linear)
 			sctx->framebuffer.any_dst_linear = true;
@@ -4449,8 +4450,7 @@ static void si_texture_barrier(struct pipe_context *ctx, unsigned flags)
 	si_update_fb_dirtiness_after_rendering(sctx);
 
 	/* Multisample surfaces are flushed in si_decompress_textures. */
-	if (sctx->framebuffer.nr_samples <= 1 &&
-	    sctx->framebuffer.state.nr_cbufs)
+	if (sctx->framebuffer.uncompressed_cb_mask)
 		si_make_CB_shader_coherent(sctx, sctx->framebuffer.nr_samples,
 					   sctx->framebuffer.CB_has_shader_readable_metadata);
 }
@@ -4493,8 +4493,7 @@ static void si_memory_barrier(struct pipe_context *ctx, unsigned flags)
 	 * si_decompress_textures when needed.
 	 */
 	if (flags & PIPE_BARRIER_FRAMEBUFFER &&
-	    sctx->framebuffer.nr_samples <= 1 &&
-	    sctx->framebuffer.state.nr_cbufs) {
+	    sctx->framebuffer.uncompressed_cb_mask) {
 		sctx->b.flags |= SI_CONTEXT_FLUSH_AND_INV_CB;
 
 		if (sctx->b.chip_class <= VI)