gallium/radeon: add can_sample_z/s flags for textures

v2: adjust r600_init_color_surface as well

Reviewed-by: Marek Olšák <[email protected]>

5 files changed, 34 insertions, 24 deletions

diff --git a/src/gallium/drivers/r600/r600_blit.c b/src/gallium/drivers/r600/r600_blit.c
index edf972650f9..6e5675b246c 100644
--- a/src/gallium/drivers/r600/r600_blit.c
+++ b/src/gallium/drivers/r600/r600_blit.c
@@ -277,8 +277,7 @@ void r600_decompress_depth_textures(struct r600_context *rctx,
 		tex = (struct r600_texture *)view->texture;
 		assert(tex->is_depth && !tex->is_flushing_texture);
 
-		if (rctx->b.chip_class >= EVERGREEN ||
-		    r600_can_read_depth(tex)) {
+		if (r600_can_sample_zs(tex, rview->is_stencil_sampler)) {
 			r600_blit_decompress_depth_in_place(rctx, tex,
 						   rview->is_stencil_sampler,
 						   view->u.tex.first_level, view->u.tex.last_level,
@@ -374,8 +373,7 @@ static bool r600_decompress_subresource(struct pipe_context *ctx,
 	struct r600_texture *rtex = (struct r600_texture*)tex;
 
 	if (rtex->is_depth && !rtex->is_flushing_texture) {
-		if (rctx->b.chip_class >= EVERGREEN ||
-		    r600_can_read_depth(rtex)) {
+		if (r600_can_sample_zs(rtex, false)) {
 			r600_blit_decompress_depth_in_place(rctx, rtex, false,
 						   level, level,
 						   first_layer, last_layer);
diff --git a/src/gallium/drivers/r600/r600_pipe.h b/src/gallium/drivers/r600/r600_pipe.h
index 0dd538bcebe..e1b2aeddf01 100644
--- a/src/gallium/drivers/r600/r600_pipe.h
+++ b/src/gallium/drivers/r600/r600_pipe.h
@@ -926,14 +926,6 @@ static inline unsigned r600_pack_float_12p4(float x)
 	       x >= 4096 ? 0xffff : x * 16;
 }
 
-/* Return if the depth format can be read without the DB->CB copy on r6xx-r7xx. */
-static inline bool r600_can_read_depth(struct r600_texture *rtex)
-{
-	return rtex->resource.b.b.nr_samples <= 1 &&
-	       (rtex->resource.b.b.format == PIPE_FORMAT_Z16_UNORM ||
-		rtex->resource.b.b.format == PIPE_FORMAT_Z32_FLOAT);
-}
-
 static inline unsigned r600_get_flush_flags(enum r600_coherency coher)
 {
 	switch (coher) {
diff --git a/src/gallium/drivers/r600/r600_state.c b/src/gallium/drivers/r600/r600_state.c
index e805d332e3e..ea551be1489 100644
--- a/src/gallium/drivers/r600/r600_state.c
+++ b/src/gallium/drivers/r600/r600_state.c
@@ -706,7 +706,13 @@ r600_create_sampler_view_custom(struct pipe_context *ctx,
 		return NULL;
 	}
 
-	if (tmp->is_depth && !tmp->is_flushing_texture && !r600_can_read_depth(tmp)) {
+	if (state->format == PIPE_FORMAT_X24S8_UINT ||
+	    state->format == PIPE_FORMAT_S8X24_UINT ||
+	    state->format == PIPE_FORMAT_X32_S8X24_UINT ||
+	    state->format == PIPE_FORMAT_S8_UINT)
+		view->is_stencil_sampler = true;
+
+	if (tmp->is_depth && !r600_can_sample_zs(tmp, view->is_stencil_sampler)) {
 		if (!r600_init_flushed_depth_texture(ctx, texture, NULL)) {
 			FREE(view);
 			return NULL;
@@ -744,12 +750,6 @@ r600_create_sampler_view_custom(struct pipe_context *ctx,
 		break;
 	}
 
-	if (state->format == PIPE_FORMAT_X24S8_UINT ||
-	    state->format == PIPE_FORMAT_S8X24_UINT ||
-	    state->format == PIPE_FORMAT_X32_S8X24_UINT ||
-	    state->format == PIPE_FORMAT_S8_UINT)
-		view->is_stencil_sampler = true;
-
 	view->tex_resource = &tmp->resource;
 	view->tex_resource_words[0] = (S_038000_DIM(r600_tex_dim(texture->target, texture->nr_samples)) |
 				       S_038000_TILE_MODE(array_mode) |
@@ -842,7 +842,7 @@ static void r600_init_color_surface(struct r600_context *rctx,
 	int i;
 	bool blend_bypass = 0, blend_clamp = 1, do_endian_swap = FALSE;
 
-	if (rtex->is_depth && !rtex->is_flushing_texture && !r600_can_read_depth(rtex)) {
+	if (rtex->is_depth && !rtex->is_flushing_texture && !r600_can_sample_zs(rtex, false)) {
 		r600_init_flushed_depth_texture(&rctx->b.b, surf->base.texture, NULL);
 		rtex = rtex->flushed_depth_texture;
 		assert(rtex);
diff --git a/src/gallium/drivers/radeon/r600_pipe_common.h b/src/gallium/drivers/radeon/r600_pipe_common.h
index 1ad69f85bb4..1840640e421 100644
--- a/src/gallium/drivers/radeon/r600_pipe_common.h
+++ b/src/gallium/drivers/radeon/r600_pipe_common.h
@@ -242,6 +242,8 @@ struct r600_texture {
 	uint64_t			size;
 	unsigned			num_level0_transfers;
 	bool				is_depth;
+	bool				can_sample_z;
+	bool				can_sample_s;
 	unsigned			dirty_level_mask; /* each bit says if that mipmap is compressed */
 	unsigned			stencil_dirty_level_mask; /* each bit says if that mipmap is compressed */
 	struct r600_texture		*flushed_depth_texture;
@@ -872,6 +874,13 @@ r600_get_sampler_view_priority(struct r600_resource *res)
 	return RADEON_PRIO_SAMPLER_TEXTURE;
 }
 
+static inline bool
+r600_can_sample_zs(struct r600_texture *tex, bool stencil_sampler)
+{
+	return (stencil_sampler && tex->can_sample_s) ||
+	       (!stencil_sampler && tex->can_sample_z);
+}
+
 #define COMPUTE_DBG(rscreen, fmt, args...) \
 	do { \
 		if ((rscreen->b.debug_flags & DBG_COMPUTE)) fprintf(stderr, fmt, ##args); \
diff --git a/src/gallium/drivers/radeon/r600_texture.c b/src/gallium/drivers/radeon/r600_texture.c
index 614dd0e87f0..6b88fc14396 100644
--- a/src/gallium/drivers/radeon/r600_texture.c
+++ b/src/gallium/drivers/radeon/r600_texture.c
@@ -1027,11 +1027,22 @@ r600_texture_create_object(struct pipe_screen *screen,
 	rtex->ps_draw_ratio = 0;
 
 	if (rtex->is_depth) {
-		if (!(base->flags & (R600_RESOURCE_FLAG_TRANSFER |
-				     R600_RESOURCE_FLAG_FLUSHED_DEPTH)) &&
-		    !(rscreen->debug_flags & DBG_NO_HYPERZ)) {
+		if (base->flags & (R600_RESOURCE_FLAG_TRANSFER |
+				   R600_RESOURCE_FLAG_FLUSHED_DEPTH) ||
+		    rscreen->chip_class >= EVERGREEN) {
+			rtex->can_sample_z = true;
+			rtex->can_sample_s = true;
+		} else {
+			if (rtex->resource.b.b.nr_samples <= 1 &&
+			    (rtex->resource.b.b.format == PIPE_FORMAT_Z16_UNORM ||
+			     rtex->resource.b.b.format == PIPE_FORMAT_Z32_FLOAT))
+				rtex->can_sample_z = true;
+		}
 
-			r600_texture_allocate_htile(rscreen, rtex);
+		if (!(base->flags & (R600_RESOURCE_FLAG_TRANSFER |
+				     R600_RESOURCE_FLAG_FLUSHED_DEPTH))) {
+			if (!(rscreen->debug_flags & DBG_NO_HYPERZ))
+				r600_texture_allocate_htile(rscreen, rtex);
 		}
 	} else {
 		if (base->nr_samples > 1) {