5 files changed, 13 insertions, 17 deletions

diff --git a/src/gallium/drivers/radeonsi/si_hw_context.c b/src/gallium/drivers/radeonsi/si_hw_context.c
index fbac95d5b29..28aed79a0bf 100644
--- a/src/gallium/drivers/radeonsi/si_hw_context.c
+++ b/src/gallium/drivers/radeonsi/si_hw_context.c
@@ -193,6 +193,7 @@ void si_begin_new_cs(struct si_context *ctx)
 	si_mark_atom_dirty(ctx, &ctx->msaa_sample_locs);
 	si_mark_atom_dirty(ctx, &ctx->msaa_config);
 	si_mark_atom_dirty(ctx, &ctx->sample_mask.atom);
+	si_mark_atom_dirty(ctx, &ctx->cb_target_mask);
 	si_mark_atom_dirty(ctx, &ctx->blend_color.atom);
 	si_mark_atom_dirty(ctx, &ctx->db_render_state);
 	si_mark_atom_dirty(ctx, &ctx->stencil_ref.atom);
diff --git a/src/gallium/drivers/radeonsi/si_pipe.h b/src/gallium/drivers/radeonsi/si_pipe.h
index 6eb0cb3bc92..e5d5d10851f 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.h
+++ b/src/gallium/drivers/radeonsi/si_pipe.h
@@ -203,6 +203,7 @@ struct si_context {
 	struct si_clip_state		clip_state;
 	struct r600_atom		clip_regs;
 	struct si_sample_mask		sample_mask;
+	struct r600_atom		cb_target_mask;
 	struct r600_atom		msaa_sample_locs;
 	struct r600_atom		msaa_config;
 	int				ps_iter_samples;
diff --git a/src/gallium/drivers/radeonsi/si_state.c b/src/gallium/drivers/radeonsi/si_state.c
index d52b9379bad..0c1448eb2d5 100644
--- a/src/gallium/drivers/radeonsi/si_state.c
+++ b/src/gallium/drivers/radeonsi/si_state.c
@@ -249,23 +249,18 @@ static unsigned si_pack_float_12p4(float x)
  *
  * Another reason is to avoid a hang with dual source blending.
  */
-void si_update_fb_blend_state(struct si_context *sctx)
+static void si_emit_cb_target_mask(struct si_context *sctx, struct r600_atom *atom)
 {
-	struct si_pm4_state *pm4;
+	struct radeon_winsys_cs *cs = sctx->b.rings.gfx.cs;
 	struct si_state_blend *blend = sctx->queued.named.blend;
 	uint32_t mask = 0, i;
 
-	if (blend == NULL)
-		return;
-
-	pm4 = CALLOC_STRUCT(si_pm4_state);
-	if (pm4 == NULL)
-		return;
-
 	for (i = 0; i < sctx->framebuffer.state.nr_cbufs; i++)
 		if (sctx->framebuffer.state.cbufs[i])
 			mask |= 0xf << (4*i);
-	mask &= blend->cb_target_mask;
+
+	if (blend)
+		mask &= blend->cb_target_mask;
 
 	/* Avoid a hang that happens when dual source blending is enabled
 	 * but there is not enough color outputs. This is undefined behavior,
@@ -277,8 +272,7 @@ void si_update_fb_blend_state(struct si_context *sctx)
 	    (sctx->ps_shader->ps_colors_written & 0x3) != 0x3)
 		mask = 0;
 
-	si_pm4_set_reg(pm4, R_028238_CB_TARGET_MASK, mask);
-	si_pm4_set_state(sctx, fb_blend, pm4);
+	r600_write_context_reg(cs, R_028238_CB_TARGET_MASK, mask);
 }
 
 /*
@@ -439,7 +433,7 @@ static void si_bind_blend_state(struct pipe_context *ctx, void *state)
 {
 	struct si_context *sctx = (struct si_context *)ctx;
 	si_pm4_bind_state(sctx, blend, (struct si_state_blend *)state);
-	si_update_fb_blend_state(sctx);
+	si_mark_atom_dirty(sctx, &sctx->cb_target_mask);
 }
 
 static void si_delete_blend_state(struct pipe_context *ctx, void *state)
@@ -2164,7 +2158,7 @@ static void si_set_framebuffer_state(struct pipe_context *ctx,
 	}
 
 	si_update_fb_rs_state(sctx);
-	si_update_fb_blend_state(sctx);
+	si_mark_atom_dirty(sctx, &sctx->cb_target_mask);
 
 	sctx->framebuffer.atom.num_dw = state->nr_cbufs*16 + (8 - state->nr_cbufs)*3;
 	sctx->framebuffer.atom.num_dw += state->zsbuf ? 26 : 4;
@@ -3063,6 +3057,7 @@ void si_init_state_functions(struct si_context *sctx)
 	si_init_atom(sctx, &sctx->db_render_state, &sctx->atoms.s.db_render_state, si_emit_db_render_state, 10);
 	si_init_atom(sctx, &sctx->msaa_config, &sctx->atoms.s.msaa_config, si_emit_msaa_config, 10);
 	si_init_atom(sctx, &sctx->sample_mask.atom, &sctx->atoms.s.sample_mask, si_emit_sample_mask, 4);
+	si_init_atom(sctx, &sctx->cb_target_mask, &sctx->atoms.s.cb_target_mask, si_emit_cb_target_mask, 3);
 	si_init_atom(sctx, &sctx->blend_color.atom, &sctx->atoms.s.blend_color, si_emit_blend_color, 6);
 	si_init_atom(sctx, &sctx->clip_regs, &sctx->atoms.s.clip_regs, si_emit_clip_regs, 6);
 	si_init_atom(sctx, &sctx->clip_state.atom, &sctx->atoms.s.clip_state, si_emit_clip_state, 2+6*4);
diff --git a/src/gallium/drivers/radeonsi/si_state.h b/src/gallium/drivers/radeonsi/si_state.h
index 60483db19b5..40be40a6fa4 100644
--- a/src/gallium/drivers/radeonsi/si_state.h
+++ b/src/gallium/drivers/radeonsi/si_state.h
@@ -91,7 +91,6 @@ union si_state {
 		struct si_state_rasterizer	*rasterizer;
 		struct si_state_dsa		*dsa;
 		struct si_pm4_state		*fb_rs;
-		struct si_pm4_state		*fb_blend;
 		struct si_pm4_state		*ta_bordercolor_base;
 		struct si_pm4_state		*ls;
 		struct si_pm4_state		*hs;
@@ -116,6 +115,7 @@ union si_state_atoms {
 		struct r600_atom *db_render_state;
 		struct r600_atom *msaa_config;
 		struct r600_atom *sample_mask;
+		struct r600_atom *cb_target_mask;
 		struct r600_atom *blend_color;
 		struct r600_atom *clip_regs;
 		struct r600_atom *clip_state;
@@ -269,7 +269,6 @@ void si_init_atom(struct si_context *sctx, struct r600_atom *atom,
 		  struct r600_atom **list_elem,
 		  void (*emit_func)(struct si_context *ctx, struct r600_atom *state),
 		  unsigned num_dw);
-void si_update_fb_blend_state(struct si_context *sctx);
 boolean si_is_format_supported(struct pipe_screen *screen,
                                enum pipe_format format,
                                enum pipe_texture_target target,
diff --git a/src/gallium/drivers/radeonsi/si_state_shaders.c b/src/gallium/drivers/radeonsi/si_state_shaders.c
index af91af97ce8..702af8c803e 100644
--- a/src/gallium/drivers/radeonsi/si_state_shaders.c
+++ b/src/gallium/drivers/radeonsi/si_state_shaders.c
@@ -873,7 +873,7 @@ static void si_bind_ps_shader(struct pipe_context *ctx, void *state)
 	}
 
 	sctx->ps_shader = sel;
-	si_update_fb_blend_state(sctx);
+	si_mark_atom_dirty(sctx, &sctx->cb_target_mask);
 }
 
 static void si_delete_shader_selector(struct pipe_context *ctx,