[Mesa-dev] [PATCH] r600/llvm: use killgt info from llvm

[Vincent Lejeune]

---
 src/gallium/drivers/r600/r600_llvm.c   | 2 ++
 src/gallium/drivers/r600/r600_llvm.h   | 1 +
 src/gallium/drivers/r600/r600_shader.c | 8 ++++++--
 3 files changed, 9 insertions(+), 2 deletions(-)

diff --git a/src/gallium/drivers/r600/r600_llvm.c b/src/gallium/drivers/r600/r600_llvm.c
index 2050be2..83d7340 100644
--- a/src/gallium/drivers/r600/r600_llvm.c
+++ b/src/gallium/drivers/r600/r600_llvm.c
@@ -556,6 +556,7 @@ unsigned r600_llvm_compile(
 	unsigned * inst_byte_count,
 	enum radeon_family family,
 	struct r600_bytecode *bc,
+	boolean *use_kill,
 	unsigned dump)
 {
 	unsigned r;
@@ -566,6 +567,7 @@ unsigned r600_llvm_compile(
 	*inst_byte_count = binary.code_size;
 	bc->ngpr = util_le32_to_cpu(*(uint32_t*)binary.config);
 	bc->nstack = util_le32_to_cpu(*(uint32_t*)(binary.config + 4));
+	*use_kill = util_le32_to_cpu(*(uint32_t*)(binary.config + 8));
 	return r;
 }
 
diff --git a/src/gallium/drivers/r600/r600_llvm.h b/src/gallium/drivers/r600/r600_llvm.h
index 919dd24..50bbca6 100644
--- a/src/gallium/drivers/r600/r600_llvm.h
+++ b/src/gallium/drivers/r600/r600_llvm.h
@@ -22,6 +22,7 @@ unsigned r600_llvm_compile(
 	unsigned * inst_byte_count,
 	enum radeon_family family,
 	struct r600_bytecode *bc,
+	boolean *use_kill,
 	unsigned dump);
 
 #endif /* defined R600_USE_LLVM || defined HAVE_OPENCL */
diff --git a/src/gallium/drivers/r600/r600_shader.c b/src/gallium/drivers/r600/r600_shader.c
index 0204f80..25f900f 100644
--- a/src/gallium/drivers/r600/r600_shader.c
+++ b/src/gallium/drivers/r600/r600_shader.c
@@ -272,6 +272,7 @@ int r600_compute_shader_create(struct pipe_context * ctx,
 	unsigned byte_count;
 	struct r600_shader_ctx shader_ctx;
 	bool dump = (r600_ctx->screen->debug_flags & DBG_CS) != 0;
+	boolean use_kill;
 
 	shader_ctx.bc = bytecode;
 	r600_bytecode_init(shader_ctx.bc, r600_ctx->chip_class, r600_ctx->family,
@@ -279,7 +280,7 @@ int r600_compute_shader_create(struct pipe_context * ctx,
 	shader_ctx.bc->type = TGSI_PROCESSOR_COMPUTE;
 	shader_ctx.bc->isa = r600_ctx->isa;
 	r600_llvm_compile(mod, &bytes, &byte_count, r600_ctx->family,
-				shader_ctx.bc, dump);
+				shader_ctx.bc, &use_kill, dump);
 	r600_bytecode_from_byte_stream(&shader_ctx, bytes, byte_count);
 	if (shader_ctx.bc->chip_class == CAYMAN) {
 		cm_bytecode_add_cf_end(shader_ctx.bc);
@@ -1444,6 +1445,7 @@ static int r600_shader_from_tgsi(struct r600_screen *rscreen,
 	if (use_llvm) {
 		struct radeon_llvm_context radeon_llvm_ctx;
 		LLVMModuleRef mod;
+		boolean use_kill;
 		bool dump = r600_can_dump_shader(rscreen, ctx.type);
 
 		memset(&radeon_llvm_ctx, 0, sizeof(radeon_llvm_ctx));
@@ -1461,13 +1463,15 @@ static int r600_shader_from_tgsi(struct r600_screen *rscreen,
 		mod = r600_tgsi_llvm(&radeon_llvm_ctx, tokens);
 
 		if (r600_llvm_compile(mod, &inst_bytes, &inst_byte_count,
-				      rscreen->family, ctx.bc, dump)) {
+				      rscreen->family, ctx.bc, &use_kill, dump)) {
 			FREE(inst_bytes);
 			radeon_llvm_dispose(&radeon_llvm_ctx);
 			use_llvm = 0;
 			fprintf(stderr, "R600 LLVM backend failed to compile "
 				"shader.  Falling back to TGSI\n");
 		} else {
+			if (use_kill)
+				ctx.shader->uses_kill = use_kill;
 			ctx.file_offset[TGSI_FILE_OUTPUT] =
 					ctx.file_offset[TGSI_FILE_INPUT];
 		}
-- 
1.8.1.4