radeonsi: fix a Unigine Heaven hang when drirc is missing

Cc: 10.6 11.0 <[email protected]>
Reviewed-by: Alex Deucher <[email protected]>
Acked-by: Christian König <[email protected]>

4 files changed, 28 insertions, 1 deletions

diff --git a/src/gallium/drivers/radeonsi/si_shader.h b/src/gallium/drivers/radeonsi/si_shader.h
index 423b849b7c3..ad32473b91e 100644
--- a/src/gallium/drivers/radeonsi/si_shader.h
+++ b/src/gallium/drivers/radeonsi/si_shader.h
@@ -190,6 +190,7 @@ struct si_shader_selector {
 	uint64_t	inputs_read;
 	uint64_t	outputs_written;
 	uint32_t	patch_outputs_written;
+	uint32_t	ps_colors_written;
 };
 
 /* Valid shader configurations:
diff --git a/src/gallium/drivers/radeonsi/si_state.c b/src/gallium/drivers/radeonsi/si_state.c
index 806ab5f0e22..1ca5e4667b2 100644
--- a/src/gallium/drivers/radeonsi/si_state.c
+++ b/src/gallium/drivers/radeonsi/si_state.c
@@ -29,6 +29,7 @@
 #include "sid.h"
 #include "radeon/r600_cs.h"
 
+#include "util/u_dual_blend.h"
 #include "util/u_format.h"
 #include "util/u_format_s3tc.h"
 #include "util/u_memory.h"
@@ -233,8 +234,10 @@ static unsigned si_pack_float_12p4(float x)
  * - The COLOR1 format isn't INVALID because of possible dual-source blending,
  *   so COLOR1 is enabled pretty much all the time.
  * So CB_TARGET_MASK is the only register that can disable COLOR1.
+ *
+ * Another reason is to avoid a hang with dual source blending.
  */
-static void si_update_fb_blend_state(struct si_context *sctx)
+void si_update_fb_blend_state(struct si_context *sctx)
 {
 	struct si_pm4_state *pm4;
 	struct si_state_blend *blend = sctx->queued.named.blend;
@@ -252,6 +255,16 @@ static void si_update_fb_blend_state(struct si_context *sctx)
 			mask |= 0xf << (4*i);
 	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,
+	 * so disable color writes completely.
+	 *
+	 * Reproducible with Unigine Heaven 4.0 and drirc missing.
+	 */
+	if (blend->dual_src_blend &&
+	    (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);
 }
@@ -343,6 +356,7 @@ static void *si_create_blend_state_mode(struct pipe_context *ctx,
 		return NULL;
 
 	blend->alpha_to_one = state->alpha_to_one;
+	blend->dual_src_blend = util_blend_state_is_dual(state, 0);
 
 	if (state->logicop_enable) {
 		color_control |= S_028808_ROP3(state->logicop_func | (state->logicop_func << 4));
diff --git a/src/gallium/drivers/radeonsi/si_state.h b/src/gallium/drivers/radeonsi/si_state.h
index 118c5622c62..242db8afd45 100644
--- a/src/gallium/drivers/radeonsi/si_state.h
+++ b/src/gallium/drivers/radeonsi/si_state.h
@@ -39,6 +39,7 @@ struct si_state_blend {
 	struct si_pm4_state	pm4;
 	uint32_t		cb_target_mask;
 	bool			alpha_to_one;
+	bool			dual_src_blend;
 };
 
 struct si_state_sample_mask {
@@ -251,6 +252,7 @@ void si_shader_change_notify(struct si_context *sctx);
 /* si_state.c */
 struct si_shader_selector;
 
+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 a09f588b356..b223e060be3 100644
--- a/src/gallium/drivers/radeonsi/si_state_shaders.c
+++ b/src/gallium/drivers/radeonsi/si_state_shaders.c
@@ -713,6 +713,15 @@ static void *si_create_shader_state(struct pipe_context *ctx,
 			}
 		}
 		break;
+	case PIPE_SHADER_FRAGMENT:
+		for (i = 0; i < sel->info.num_outputs; i++) {
+			unsigned name = sel->info.output_semantic_name[i];
+			unsigned index = sel->info.output_semantic_index[i];
+
+			if (name == TGSI_SEMANTIC_COLOR)
+				sel->ps_colors_written |= 1 << index;
+		}
+		break;
 	}
 
 	if (sscreen->b.debug_flags & DBG_PRECOMPILE)
@@ -840,6 +849,7 @@ static void si_bind_ps_shader(struct pipe_context *ctx, void *state)
 	}
 
 	sctx->ps_shader = sel;
+	si_update_fb_blend_state(sctx);
 }
 
 static void si_delete_shader_selector(struct pipe_context *ctx,