radeonsi: move DB_RENDER_CONTROL into draw_vbo

So that I can add fast depth clear.

Reviewed-by: Michel Dänzer <[email protected]>

5 files changed, 46 insertions, 58 deletions

diff --git a/src/gallium/drivers/radeonsi/si_blit.c b/src/gallium/drivers/radeonsi/si_blit.c
index 2806b468ee5..87c8fa9c634 100644
--- a/src/gallium/drivers/radeonsi/si_blit.c
+++ b/src/gallium/drivers/radeonsi/si_blit.c
@@ -114,7 +114,6 @@ static void si_blit_decompress_depth(struct pipe_context *ctx,
 	unsigned layer, level, sample, checked_last_layer, max_layer, max_sample;
 	float depth = 1.0f;
 	const struct util_format_description *desc;
-	void **custom_dsa;
 	struct r600_texture *flushed_depth_texture = staging ?
 			staging : texture->flushed_depth_texture;
 
@@ -124,20 +123,13 @@ static void si_blit_decompress_depth(struct pipe_context *ctx,
 	max_sample = u_max_sample(&texture->resource.b.b);
 
 	desc = util_format_description(flushed_depth_texture->resource.b.b.format);
-	switch (util_format_has_depth(desc) | util_format_has_stencil(desc) << 1) {
-	default:
-		assert(!"No depth or stencil to uncompress");
-		return;
-	case 3:
-		custom_dsa = sctx->custom_dsa_flush_depth_stencil;
-		break;
-	case 2:
-		custom_dsa = sctx->custom_dsa_flush_stencil;
-		break;
-	case 1:
-		custom_dsa = sctx->custom_dsa_flush_depth;
-		break;
-	}
+
+	if (util_format_has_depth(desc))
+		sctx->dbcb_depth_copy_enabled = true;
+	if (util_format_has_stencil(desc))
+		sctx->dbcb_stencil_copy_enabled = true;
+
+	assert(sctx->dbcb_depth_copy_enabled || sctx->dbcb_stencil_copy_enabled);
 
 	for (level = first_level; level <= last_level; level++) {
 		if (!staging && !(texture->dirty_level_mask & (1 << level)))
@@ -152,6 +144,8 @@ static void si_blit_decompress_depth(struct pipe_context *ctx,
 			for (sample = first_sample; sample <= last_sample; sample++) {
 				struct pipe_surface *zsurf, *cbsurf, surf_tmpl;
 
+				sctx->dbcb_copy_sample = sample;
+
 				surf_tmpl.format = texture->resource.b.b.format;
 				surf_tmpl.u.tex.level = level;
 				surf_tmpl.u.tex.first_layer = layer;
@@ -165,7 +159,7 @@ static void si_blit_decompress_depth(struct pipe_context *ctx,
 
 				si_blitter_begin(ctx, SI_DECOMPRESS);
 				util_blitter_custom_depth_stencil(sctx->blitter, zsurf, cbsurf, 1 << sample,
-								  custom_dsa[sample], depth);
+								  sctx->custom_dsa_flush, depth);
 				si_blitter_end(ctx);
 
 				pipe_surface_reference(&zsurf, NULL);
@@ -181,6 +175,9 @@ static void si_blit_decompress_depth(struct pipe_context *ctx,
 			texture->dirty_level_mask &= ~(1 << level);
 		}
 	}
+
+	sctx->dbcb_depth_copy_enabled = false;
+	sctx->dbcb_stencil_copy_enabled = false;
 }
 
 static void si_blit_decompress_depth_in_place(struct si_context *sctx,
@@ -191,6 +188,8 @@ static void si_blit_decompress_depth_in_place(struct si_context *sctx,
 	struct pipe_surface *zsurf, surf_tmpl = {{0}};
 	unsigned layer, max_layer, checked_last_layer, level;
 
+	sctx->db_inplace_flush_enabled = true;
+
 	surf_tmpl.format = texture->resource.b.b.format;
 
 	for (level = first_level; level <= last_level; level++) {
@@ -212,7 +211,7 @@ static void si_blit_decompress_depth_in_place(struct si_context *sctx,
 
 			si_blitter_begin(&sctx->b.b, SI_DECOMPRESS);
 			util_blitter_custom_depth_stencil(sctx->blitter, zsurf, NULL, ~0,
-							  sctx->custom_dsa_flush_inplace,
+							  sctx->custom_dsa_flush,
 							  1.0f);
 			si_blitter_end(&sctx->b.b);
 
@@ -225,6 +224,8 @@ static void si_blit_decompress_depth_in_place(struct si_context *sctx,
 			texture->dirty_level_mask &= ~(1 << level);
 		}
 	}
+
+	sctx->db_inplace_flush_enabled = false;
 }
 
 void si_flush_depth_textures(struct si_context *sctx,
diff --git a/src/gallium/drivers/radeonsi/si_pipe.c b/src/gallium/drivers/radeonsi/si_pipe.c
index 6c0011238ed..abd7d4f6a94 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.c
+++ b/src/gallium/drivers/radeonsi/si_pipe.c
@@ -49,12 +49,7 @@ static void si_destroy_context(struct pipe_context *context)
 	if (sctx->dummy_pixel_shader) {
 		sctx->b.b.delete_fs_state(&sctx->b.b, sctx->dummy_pixel_shader);
 	}
-	for (int i = 0; i < 8; i++) {
-		sctx->b.b.delete_depth_stencil_alpha_state(&sctx->b.b, sctx->custom_dsa_flush_depth_stencil[i]);
-		sctx->b.b.delete_depth_stencil_alpha_state(&sctx->b.b, sctx->custom_dsa_flush_depth[i]);
-		sctx->b.b.delete_depth_stencil_alpha_state(&sctx->b.b, sctx->custom_dsa_flush_stencil[i]);
-	}
-	sctx->b.b.delete_depth_stencil_alpha_state(&sctx->b.b, sctx->custom_dsa_flush_inplace);
+	sctx->b.b.delete_depth_stencil_alpha_state(&sctx->b.b, sctx->custom_dsa_flush);
 	sctx->b.b.delete_blend_state(&sctx->b.b, sctx->custom_blend_resolve);
 	sctx->b.b.delete_blend_state(&sctx->b.b, sctx->custom_blend_decompress);
 	sctx->b.b.delete_blend_state(&sctx->b.b, sctx->custom_blend_fastclear);
diff --git a/src/gallium/drivers/radeonsi/si_pipe.h b/src/gallium/drivers/radeonsi/si_pipe.h
index 10f7e23c360..c4f38001c75 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.h
+++ b/src/gallium/drivers/radeonsi/si_pipe.h
@@ -86,10 +86,7 @@ struct si_framebuffer {
 struct si_context {
 	struct r600_common_context	b;
 	struct blitter_context		*blitter;
-	void				*custom_dsa_flush_depth_stencil[8];
-	void				*custom_dsa_flush_depth[8];
-	void				*custom_dsa_flush_stencil[8];
-	void				*custom_dsa_flush_inplace;
+	void				*custom_dsa_flush;
 	void				*custom_blend_resolve;
 	void				*custom_blend_decompress;
 	void				*custom_blend_fastclear;
@@ -161,6 +158,12 @@ struct si_context {
 	/* SI state handling */
 	union si_state	queued;
 	union si_state	emitted;
+
+	/* Additional DB state. */
+	bool dbcb_depth_copy_enabled;
+	bool dbcb_stencil_copy_enabled;
+	unsigned dbcb_copy_sample;
+	bool db_inplace_flush_enabled;
 };
 
 /* si_blit.c */
diff --git a/src/gallium/drivers/radeonsi/si_state.c b/src/gallium/drivers/radeonsi/si_state.c
index c4915f1c1e6..cbb29a91532 100644
--- a/src/gallium/drivers/radeonsi/si_state.c
+++ b/src/gallium/drivers/radeonsi/si_state.c
@@ -757,7 +757,6 @@ static void *si_create_dsa_state(struct pipe_context *ctx,
 	struct si_state_dsa *dsa = CALLOC_STRUCT(si_state_dsa);
 	struct si_pm4_state *pm4 = &dsa->pm4;
 	unsigned db_depth_control;
-	unsigned db_render_control;
 	uint32_t db_stencil_control = 0;
 
 	if (dsa == NULL) {
@@ -802,9 +801,7 @@ static void *si_create_dsa_state(struct pipe_context *ctx,
 	}
 
 	/* misc */
-	db_render_control = 0;
 	si_pm4_set_reg(pm4, R_028800_DB_DEPTH_CONTROL, db_depth_control);
-	si_pm4_set_reg(pm4, R_028000_DB_RENDER_CONTROL, db_render_control);
 	si_pm4_set_reg(pm4, R_02842C_DB_STENCIL_CONTROL, db_stencil_control);
 
 	return dsa;
@@ -828,28 +825,11 @@ static void si_delete_dsa_state(struct pipe_context *ctx, void *state)
 	si_pm4_delete_state(sctx, dsa, (struct si_state_dsa *)state);
 }
 
-static void *si_create_db_flush_dsa(struct si_context *sctx, bool copy_depth,
-				    bool copy_stencil, int sample)
+static void *si_create_db_flush_dsa(struct si_context *sctx)
 {
-	struct pipe_depth_stencil_alpha_state dsa;
-        struct si_state_dsa *state;
+	struct pipe_depth_stencil_alpha_state dsa = {};
 
-	memset(&dsa, 0, sizeof(dsa));
-
-	state = sctx->b.b.create_depth_stencil_alpha_state(&sctx->b.b, &dsa);
-	if (copy_depth || copy_stencil) {
-		si_pm4_set_reg(&state->pm4, R_028000_DB_RENDER_CONTROL,
-			       S_028000_DEPTH_COPY(copy_depth) |
-			       S_028000_STENCIL_COPY(copy_stencil) |
-			       S_028000_COPY_CENTROID(1) |
-			       S_028000_COPY_SAMPLE(sample));
-	} else {
-		si_pm4_set_reg(&state->pm4, R_028000_DB_RENDER_CONTROL,
-			       S_028000_DEPTH_COMPRESS_DISABLE(1) |
-			       S_028000_STENCIL_COMPRESS_DISABLE(1));
-	}
-
-        return state;
+	return sctx->b.b.create_depth_stencil_alpha_state(&sctx->b.b, &dsa);
 }
 
 /*
@@ -2933,8 +2913,6 @@ static void si_need_gfx_cs_space(struct pipe_context *ctx, unsigned num_dw,
 
 void si_init_state_functions(struct si_context *sctx)
 {
-	int i;
-
 	si_init_atom(&sctx->framebuffer.atom, &sctx->atoms.s.framebuffer, si_emit_framebuffer_state, 0);
 
 	sctx->b.b.create_blend_state = si_create_blend_state;
@@ -2950,12 +2928,7 @@ void si_init_state_functions(struct si_context *sctx)
 	sctx->b.b.bind_depth_stencil_alpha_state = si_bind_dsa_state;
 	sctx->b.b.delete_depth_stencil_alpha_state = si_delete_dsa_state;
 
-	for (i = 0; i < 8; i++) {
-		sctx->custom_dsa_flush_depth_stencil[i] = si_create_db_flush_dsa(sctx, true, true, i);
-		sctx->custom_dsa_flush_depth[i] = si_create_db_flush_dsa(sctx, true, false, i);
-		sctx->custom_dsa_flush_stencil[i] = si_create_db_flush_dsa(sctx, false, true, i);
-	}
-	sctx->custom_dsa_flush_inplace = si_create_db_flush_dsa(sctx, false, false, 0);
+	sctx->custom_dsa_flush = si_create_db_flush_dsa(sctx);
 	sctx->custom_blend_resolve = si_create_blend_custom(sctx, V_028808_CB_RESOLVE);
 	sctx->custom_blend_decompress = si_create_blend_custom(sctx, V_028808_CB_FMASK_DECOMPRESS);
 	sctx->custom_blend_fastclear = si_create_blend_custom(sctx, V_028808_CB_ELIMINATE_FAST_CLEAR);
diff --git a/src/gallium/drivers/radeonsi/si_state_draw.c b/src/gallium/drivers/radeonsi/si_state_draw.c
index 3926e5ff15b..eb5869ccc68 100644
--- a/src/gallium/drivers/radeonsi/si_state_draw.c
+++ b/src/gallium/drivers/radeonsi/si_state_draw.c
@@ -739,6 +739,22 @@ static void si_state_draw(struct si_context *sctx,
 		}
 	}
 
+	/* DB_RENDER_CONTROL */
+	if (sctx->dbcb_depth_copy_enabled ||
+	    sctx->dbcb_stencil_copy_enabled) {
+		si_pm4_set_reg(pm4, R_028000_DB_RENDER_CONTROL,
+			       S_028000_DEPTH_COPY(sctx->dbcb_depth_copy_enabled) |
+			       S_028000_STENCIL_COPY(sctx->dbcb_stencil_copy_enabled) |
+			       S_028000_COPY_CENTROID(1) |
+			       S_028000_COPY_SAMPLE(sctx->dbcb_copy_sample));
+	} else if (sctx->db_inplace_flush_enabled) {
+		si_pm4_set_reg(pm4, R_028000_DB_RENDER_CONTROL,
+			       S_028000_DEPTH_COMPRESS_DISABLE(1) |
+			       S_028000_STENCIL_COMPRESS_DISABLE(1));
+	} else {
+		si_pm4_set_reg(pm4, R_028000_DB_RENDER_CONTROL, 0);
+	}
+
 	if (info->count_from_stream_output) {
 		struct r600_so_target *t =
 			(struct r600_so_target*)info->count_from_stream_output;