[Mesa-dev] [PATCH 52/61] radeonsi/gfx9: 2nd shader of merged shaders should hold a reference of the 1st

[Marek Olšák]

From: Marek Olšák <marek.olsak at amd.com>

---
 src/gallium/drivers/radeonsi/si_shader.h        |  1 +
 src/gallium/drivers/radeonsi/si_state_shaders.c | 35 ++++++++++++++++++-------
 2 files changed, 26 insertions(+), 10 deletions(-)

diff --git a/src/gallium/drivers/radeonsi/si_shader.h b/src/gallium/drivers/radeonsi/si_shader.h
index 514bb3c..5b665b5 100644
--- a/src/gallium/drivers/radeonsi/si_shader.h
+++ b/src/gallium/drivers/radeonsi/si_shader.h
@@ -498,20 +498,21 @@ struct si_shader_info {
 	char			face_vgpr_index;
 	bool			uses_instanceid;
 	ubyte			nr_pos_exports;
 	ubyte			nr_param_exports;
 };
 
 struct si_shader {
 	struct si_compiler_ctx_state	compiler_ctx_state;
 
 	struct si_shader_selector	*selector;
+	struct si_shader_selector	*previous_stage_sel; /* for refcounting */
 	struct si_shader		*next_variant;
 
 	struct si_shader_part		*prolog;
 	struct si_shader		*previous_stage; /* for GFX9 */
 	struct si_shader_part		*prolog2;
 	struct si_shader_part		*epilog;
 
 	struct si_pm4_state		*pm4;
 	struct r600_resource		*bo;
 	struct r600_resource		*scratch_bo;
diff --git a/src/gallium/drivers/radeonsi/si_state_shaders.c b/src/gallium/drivers/radeonsi/si_state_shaders.c
index c5fa01d..94fc0e4 100644
--- a/src/gallium/drivers/radeonsi/si_state_shaders.c
+++ b/src/gallium/drivers/radeonsi/si_state_shaders.c
@@ -1517,20 +1517,21 @@ static void si_shader_selector_reference(struct si_context *sctx,
 }
 
 /* Select the hw shader variant depending on the current state. */
 static int si_shader_select_with_key(struct si_screen *sscreen,
 				     struct si_shader_ctx_state *state,
 				     struct si_compiler_ctx_state *compiler_state,
 				     struct si_shader_key *key,
 				     int thread_index)
 {
 	struct si_shader_selector *sel = state->cso;
+	struct si_shader_selector *previous_stage_sel = NULL;
 	struct si_shader *current = state->current;
 	struct si_shader *iter, *shader = NULL;
 
 	if (unlikely(sscreen->b.debug_flags & DBG_NO_OPT_VARIANT)) {
 		memset(&key->opt, 0, sizeof(key->opt));
 	}
 
 again:
 	/* Check if we don't need to change anything.
 	 * This path is also used for most shaders that don't need multiple
@@ -1584,64 +1585,77 @@ again:
 	/* Build a new shader. */
 	shader = CALLOC_STRUCT(si_shader);
 	if (!shader) {
 		mtx_unlock(&sel->mutex);
 		return -ENOMEM;
 	}
 	shader->selector = sel;
 	shader->key = *key;
 	shader->compiler_ctx_state = *compiler_state;
 
+	/* If this is a merged shader, get the first shader's selector. */
+	if (sscreen->b.chip_class >= GFX9) {
+		if (sel->type == PIPE_SHADER_TESS_CTRL)
+			previous_stage_sel  = key->part.tcs.ls;
+		else if (sel->type == PIPE_SHADER_GEOMETRY)
+			previous_stage_sel  = key->part.gs.es;
+	}
+
 	/* Compile the main shader part if it doesn't exist. This can happen
 	 * if the initial guess was wrong. */
 	bool is_pure_monolithic =
 		sscreen->use_monolithic_shaders ||
 		memcmp(&key->mono, &zeroed.mono, sizeof(key->mono)) != 0;
 
 	if (!is_pure_monolithic) {
 		bool ok;
 
 		/* Make sure the main shader part is present. This is needed
 		 * for shaders that can be compiled as VS, LS, or ES, and only
 		 * one of them is compiled at creation.
 		 *
 		 * For merged shaders, check that the starting shader's main
 		 * part is present.
 		 */
-		if (sscreen->b.chip_class >= GFX9 &&
-		    (sel->type == PIPE_SHADER_TESS_CTRL ||
-		     sel->type == PIPE_SHADER_GEOMETRY)) {
-			struct si_shader_selector *shader1 = NULL;
+		if (previous_stage_sel) {
 			struct si_shader_key shader1_key = zeroed;
 
-			if (sel->type == PIPE_SHADER_TESS_CTRL) {
-				shader1 = key->part.tcs.ls;
+			if (sel->type == PIPE_SHADER_TESS_CTRL)
 				shader1_key.as_ls = 1;
-			} else if (sel->type == PIPE_SHADER_GEOMETRY) {
-				shader1 = key->part.gs.es;
+			else if (sel->type == PIPE_SHADER_GEOMETRY)
 				shader1_key.as_es = 1;
-			} else
+			else
 				assert(0);
 
-			ok = si_check_missing_main_part(sscreen, shader1,
+			ok = si_check_missing_main_part(sscreen,
+							previous_stage_sel,
 							compiler_state, &shader1_key);
 		} else {
 			ok = si_check_missing_main_part(sscreen, sel,
 							compiler_state, key);
 		}
 		if (!ok) {
 			FREE(shader);
 			mtx_unlock(&sel->mutex);
 			return -ENOMEM; /* skip the draw call */
 		}
 	}
 
+	/* Keep the reference to the 1st shader of merged shaders, so that
+	 * Gallium can't destroy it before we destroy the 2nd shader.
+	 *
+	 * Set sctx = NULL, because it's unused if we're not releasing
+	 * the shader, and we don't have any sctx here.
+	 */
+	si_shader_selector_reference(NULL, &shader->previous_stage_sel,
+				     previous_stage_sel);
+
 	/* Monolithic-only shaders don't make a distinction between optimized
 	 * and unoptimized. */
 	shader->is_monolithic =
 		is_pure_monolithic ||
 		memcmp(&key->opt, &zeroed.opt, sizeof(key->opt)) != 0;
 
 	shader->is_optimized =
 		!is_pure_monolithic &&
 		memcmp(&key->opt, &zeroed.opt, sizeof(key->opt)) != 0;
 	if (shader->is_optimized)
@@ -2237,20 +2251,21 @@ static void si_delete_shader(struct si_context *sctx, struct si_shader *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);
 			break;
 		}
 	}
 
+	si_shader_selector_reference(sctx, &shader->previous_stage_sel, NULL);
 	si_shader_destroy(shader);
 	free(shader);
 }
 
 static void si_destroy_shader_selector(struct si_context *sctx,
 				       struct si_shader_selector *sel)
 {
 	struct si_shader *p = sel->first_variant, *c;
 	struct si_shader_ctx_state *current_shader[SI_NUM_SHADERS] = {
 		[PIPE_SHADER_VERTEX] = &sctx->vs_shader,
-- 
2.7.4