4 files changed, 31 insertions, 20 deletions

diff --git a/src/gallium/drivers/radeonsi/si_pipe.h b/src/gallium/drivers/radeonsi/si_pipe.h
index 0240a3c191d..e7617bc4975 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.h
+++ b/src/gallium/drivers/radeonsi/si_pipe.h
@@ -450,7 +450,7 @@ static inline struct tgsi_shader_info *si_get_vs_info(struct si_context *sctx)
 static inline struct si_shader* si_get_vs_state(struct si_context *sctx)
 {
 	if (sctx->gs_shader.current)
-		return sctx->gs_shader.current->gs_copy_shader;
+		return sctx->gs_shader.cso->gs_copy_shader;
 	else if (sctx->tes_shader.current)
 		return sctx->tes_shader.current;
 	else
diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
index e7553991fe5..38f65f3b770 100644
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -5951,7 +5951,7 @@ static const char *si_get_shader_name(struct si_shader *shader,
 		else
 			return "Tessellation Evaluation Shader as VS";
 	case PIPE_SHADER_GEOMETRY:
-		if (shader->gs_copy_shader == NULL)
+		if (shader->is_gs_copy_shader)
 			return "GS Copy Shader as VS";
 		else
 			return "Geometry Shader";
@@ -6080,7 +6080,7 @@ static void si_llvm_build_ret(struct si_shader_context *ctx, LLVMValueRef ret)
 }
 
 /* Generate code for the hardware VS shader stage to go with a geometry shader */
-static struct si_shader *
+struct si_shader *
 si_generate_gs_copy_shader(struct si_screen *sscreen,
 			   LLVMTargetMachineRef tm,
 			   struct si_shader_selector *gs_selector,
@@ -6103,6 +6103,8 @@ si_generate_gs_copy_shader(struct si_screen *sscreen,
 		return NULL;
 
 	shader->selector = gs_selector;
+	shader->is_gs_copy_shader = true;
+
 	si_init_shader_ctx(&ctx, sscreen, shader, tm);
 	ctx.type = PIPE_SHADER_VERTEX;
 	ctx.is_gs_copy_shader = true;
@@ -7150,13 +7152,6 @@ int si_compile_tgsi_shader(struct si_screen *sscreen,
 			shader->info.num_input_vgprs += 1;
 	}
 
-	if (ctx.type == PIPE_SHADER_GEOMETRY) {
-		shader->gs_copy_shader =
-			si_generate_gs_copy_shader(sscreen, tm, shader->selector, debug);
-		if (!shader->gs_copy_shader)
-			return -1;
-	}
-
 	return 0;
 }
 
@@ -8098,11 +8093,6 @@ int si_shader_create(struct si_screen *sscreen, LLVMTargetMachineRef tm,
 
 void si_shader_destroy(struct si_shader *shader)
 {
-	if (shader->gs_copy_shader) {
-		si_shader_destroy(shader->gs_copy_shader);
-		FREE(shader->gs_copy_shader);
-	}
-
 	if (shader->scratch_bo)
 		r600_resource_reference(&shader->scratch_bo, NULL);
 
diff --git a/src/gallium/drivers/radeonsi/si_shader.h b/src/gallium/drivers/radeonsi/si_shader.h
index 6c7a05f01bf..91f9cbffd8e 100644
--- a/src/gallium/drivers/radeonsi/si_shader.h
+++ b/src/gallium/drivers/radeonsi/si_shader.h
@@ -257,6 +257,8 @@ struct si_shader_selector {
 	 */
 	struct si_shader	*main_shader_part;
 
+	struct si_shader	*gs_copy_shader;
+
 	struct tgsi_token       *tokens;
 	struct pipe_stream_output_info  so;
 	struct tgsi_shader_info		info;
@@ -444,12 +446,12 @@ struct si_shader {
 	struct si_shader_part		*prolog;
 	struct si_shader_part		*epilog;
 
-	struct si_shader		*gs_copy_shader;
 	struct si_pm4_state		*pm4;
 	struct r600_resource		*bo;
 	struct r600_resource		*scratch_bo;
 	union si_shader_key		key;
 	bool				is_binary_shared;
+	bool				is_gs_copy_shader;
 
 	/* The following data is all that's needed for binary shaders. */
 	struct radeon_shader_binary	binary;
@@ -471,6 +473,11 @@ struct si_shader_part {
 };
 
 /* si_shader.c */
+struct si_shader *
+si_generate_gs_copy_shader(struct si_screen *sscreen,
+			   LLVMTargetMachineRef tm,
+			   struct si_shader_selector *gs_selector,
+			   struct pipe_debug_callback *debug);
 int si_compile_tgsi_shader(struct si_screen *sscreen,
 			   LLVMTargetMachineRef tm,
 			   struct si_shader *shader,
diff --git a/src/gallium/drivers/radeonsi/si_state_shaders.c b/src/gallium/drivers/radeonsi/si_state_shaders.c
index ebe7a75a2f8..4c647cbbf02 100644
--- a/src/gallium/drivers/radeonsi/si_state_shaders.c
+++ b/src/gallium/drivers/radeonsi/si_state_shaders.c
@@ -836,7 +836,6 @@ static void si_shader_init_pm4_state(struct si_screen *sscreen,
 		break;
 	case PIPE_SHADER_GEOMETRY:
 		si_shader_gs(shader);
-		si_shader_vs(sscreen, shader->gs_copy_shader, shader->selector);
 		break;
 	case PIPE_SHADER_FRAGMENT:
 		si_shader_ps(shader);
@@ -1236,6 +1235,17 @@ void si_init_shader_selector_async(void *job, int thread_index)
 					      false, sel->is_debug_context))
 			fprintf(stderr, "radeonsi: can't create a monolithic shader\n");
 	}
+
+	/* The GS copy shader is always pre-compiled. */
+	if (sel->type == PIPE_SHADER_GEOMETRY) {
+		sel->gs_copy_shader = si_generate_gs_copy_shader(sscreen, tm, sel, debug);
+		if (!sel->gs_copy_shader) {
+			fprintf(stderr, "radeonsi: can't create GS copy shader\n");
+			return;
+		}
+
+		si_shader_vs(sscreen, sel->gs_copy_shader, sel);
+	}
 }
 
 static void *si_create_shader_selector(struct pipe_context *ctx,
@@ -1518,8 +1528,10 @@ static void si_delete_shader(struct si_context *sctx, struct si_shader *shader)
 				si_pm4_delete_state(sctx, vs, shader->pm4);
 			break;
 		case PIPE_SHADER_GEOMETRY:
-			si_pm4_delete_state(sctx, gs, shader->pm4);
-			si_pm4_delete_state(sctx, vs, shader->gs_copy_shader->pm4);
+			if (shader->is_gs_copy_shader)
+				si_pm4_delete_state(sctx, vs, shader->pm4);
+			else
+				si_pm4_delete_state(sctx, gs, shader->pm4);
 			break;
 		case PIPE_SHADER_FRAGMENT:
 			si_pm4_delete_state(sctx, ps, shader->pm4);
@@ -1559,6 +1571,8 @@ static void si_delete_shader_selector(struct pipe_context *ctx, void *state)
 
 	if (sel->main_shader_part)
 		si_delete_shader(sctx, sel->main_shader_part);
+	if (sel->gs_copy_shader)
+		si_delete_shader(sctx, sel->gs_copy_shader);
 
 	util_queue_fence_destroy(&sel->ready);
 	pipe_mutex_destroy(sel->mutex);
@@ -2215,7 +2229,7 @@ bool si_update_shaders(struct si_context *sctx)
 		if (r)
 			return false;
 		si_pm4_bind_state(sctx, gs, sctx->gs_shader.current->pm4);
-		si_pm4_bind_state(sctx, vs, sctx->gs_shader.current->gs_copy_shader->pm4);
+		si_pm4_bind_state(sctx, vs, sctx->gs_shader.cso->gs_copy_shader->pm4);
 		si_update_so(sctx, sctx->gs_shader.cso);
 
 		if (!si_update_gs_ring_buffers(sctx))