r600g: avoid reemiting literal, avoid scheduling empty cs

Signed-off-by: Jerome Glisse <jglisse@redhat.com>

6 files changed, 40 insertions, 32 deletions

diff --git a/src/gallium/drivers/r600/r600_asm.c b/src/gallium/drivers/r600/r600_asm.c
index 16c98504ad..ae818bf19b 100644
--- a/src/gallium/drivers/r600/r600_asm.c
+++ b/src/gallium/drivers/r600/r600_asm.c
@@ -179,12 +179,13 @@ int r600_bc_add_literal(struct r600_bc *bc, const u32 *value)
 		return -EINVAL;
 	}
 	alu = LIST_ENTRY(struct r600_bc_alu, bc->cf_last->alu.prev, list);
-	if (!alu->last || !alu->nliteral) {
+	if (!alu->last || !alu->nliteral || alu->literal_added) {
 		return 0;
 	}
 	memcpy(alu->value, value, 4 * 4);
 	bc->cf_last->ndw += alu->nliteral;
 	bc->ndw += alu->nliteral;
+	alu->literal_added = 1;
 	return 0;
 }
 
diff --git a/src/gallium/drivers/r600/r600_asm.h b/src/gallium/drivers/r600/r600_asm.h
index 3fd94dbda0..10d98afaf0 100644
--- a/src/gallium/drivers/r600/r600_asm.h
+++ b/src/gallium/drivers/r600/r600_asm.h
@@ -48,6 +48,7 @@ struct r600_bc_alu {
 	unsigned			last;
 	unsigned			is_op3;
 	unsigned			nliteral;
+	unsigned			literal_added;
 	u32				value[4];
 };
 
diff --git a/src/gallium/drivers/r600/r600_context.c b/src/gallium/drivers/r600/r600_context.c
index 052eb1cd6d..edde80c660 100644
--- a/src/gallium/drivers/r600/r600_context.c
+++ b/src/gallium/drivers/r600/r600_context.c
@@ -54,15 +54,18 @@ void r600_flush(struct pipe_context *ctx, unsigned flags,
 	/* FIXME dumping should be removed once shader support instructions
 	 * without throwing bad code
 	 */
+	if (!rctx->ctx->cpm4)
+		goto out;
 	sprintf(dname, "gallium-%08d.bof", dc);
 	if (dc < 1)
 		radeon_ctx_dump_bof(rctx->ctx, dname);
 #if 1
 	radeon_ctx_submit(rctx->ctx);
 #endif
+	dc++;
+out:
 	rctx->ctx = radeon_ctx_decref(rctx->ctx);
 	rctx->ctx = radeon_ctx(rscreen->rw);
-	dc++;
 }
 
 static void r600_init_config(struct r600_context *rctx)
diff --git a/src/gallium/drivers/r600/radeon.h b/src/gallium/drivers/r600/radeon.h
index 00cff41b4f..8f00a4895a 100644
--- a/src/gallium/drivers/r600/radeon.h
+++ b/src/gallium/drivers/r600/radeon.h
@@ -157,6 +157,37 @@ int radeon_ctx_submit(struct radeon_ctx *ctx);
 void radeon_ctx_dump_bof(struct radeon_ctx *ctx, const char *file);
 
 /*
+ * radeon context functions
+ */
+#pragma pack(1)
+struct radeon_cs_reloc {
+	uint32_t	handle;
+	uint32_t	read_domain;
+	uint32_t	write_domain;
+	uint32_t	flags;
+};
+#pragma pack()
+
+struct radeon_ctx {
+	int				refcount;
+	struct radeon			*radeon;
+	u32				*pm4;
+	u32				cpm4;
+	u32				draw_cpm4;
+	unsigned			id;
+	unsigned			next_id;
+	unsigned			nreloc;
+	struct radeon_cs_reloc		*reloc;
+	unsigned			nbo;
+	struct radeon_bo		**bo;
+	unsigned			ndraw;
+	struct radeon_draw		*cdraw;
+	struct radeon_draw		**draw;
+	unsigned			nstate;
+	struct radeon_state		**state;
+};
+
+/*
  * R600/R700
  */
 
diff --git a/src/gallium/winsys/r600/drm/radeon_ctx.c b/src/gallium/winsys/r600/drm/radeon_ctx.c
index 6b0eba0b28..ff70ce6de7 100644
--- a/src/gallium/winsys/r600/drm/radeon_ctx.c
+++ b/src/gallium/winsys/r600/drm/radeon_ctx.c
@@ -151,6 +151,8 @@ int radeon_ctx_submit(struct radeon_ctx *ctx)
 	uint64_t chunk_array[2];
 	int r = 0;
 
+	if (!ctx->cpm4)
+		return 0;
 #if 0
 	for (r = 0; r < ctx->cpm4; r++) {
 		fprintf(stderr, "0x%08X\n", ctx->pm4[r]);
diff --git a/src/gallium/winsys/r600/drm/radeon_priv.h b/src/gallium/winsys/r600/drm/radeon_priv.h
index b91421f438..96c0d060f7 100644
--- a/src/gallium/winsys/r600/drm/radeon_priv.h
+++ b/src/gallium/winsys/r600/drm/radeon_priv.h
@@ -68,36 +68,6 @@ extern int radeon_is_family_compatible(unsigned family1, unsigned family2);
 extern int radeon_reg_id(struct radeon *radeon, unsigned offset, unsigned *typeid, unsigned *stateid, unsigned *id);
 extern unsigned radeon_type_from_id(struct radeon *radeon, unsigned id);
 
-/*
- * radeon context functions
- */
-#pragma pack(1)
-struct radeon_cs_reloc {
-	uint32_t	handle;
-	uint32_t	read_domain;
-	uint32_t	write_domain;
-	uint32_t	flags;
-};
-#pragma pack()
-
-struct radeon_ctx {
-	int				refcount;
-	struct radeon			*radeon;
-	u32				*pm4;
-	u32				cpm4;
-	u32				draw_cpm4;
-	unsigned			id;
-	unsigned			next_id;
-	unsigned			nreloc;
-	struct radeon_cs_reloc		*reloc;
-	unsigned			nbo;
-	struct radeon_bo		**bo;
-	unsigned			ndraw;
-	struct radeon_draw		*cdraw;
-	struct radeon_draw		**draw;
-	unsigned			nstate;
-	struct radeon_state		**state;
-};
 
 int radeon_ctx_set_bo_new(struct radeon_ctx *ctx, struct radeon_bo *bo);
 struct radeon_bo *radeon_ctx_get_bo(struct radeon_ctx *ctx, unsigned reloc);