[Mesa-dev] [PATCH] radeon/llvm: Run standard optimization passes on compute shader modules

[Tom Stellard]

From: Tom Stellard <thomas.stellard at amd.com>

The SROA and function inliner passes are espically important, because
they optimize away unsupported features: functions and indirect
private memory access.
---
 src/gallium/drivers/r600/evergreen_compute.c  |  2 +-
 src/gallium/drivers/radeon/radeon_llvm_util.c | 16 ++++++++++++++++
 2 files changed, 17 insertions(+), 1 deletion(-)

diff --git a/src/gallium/drivers/r600/evergreen_compute.c b/src/gallium/drivers/r600/evergreen_compute.c
index 5f67759..2d73ee5 100644
--- a/src/gallium/drivers/r600/evergreen_compute.c
+++ b/src/gallium/drivers/r600/evergreen_compute.c
@@ -214,7 +214,7 @@ void *evergreen_create_compute_state(
 	shader->private_size = cso->req_private_mem;
 	shader->input_size = cso->req_input_mem;
 
-#ifdef HAVE_OPENCL 
+#ifdef HAVE_OPENCL
 	shader->num_kernels = radeon_llvm_get_num_kernels(code, header->num_bytes);
 	shader->kernels = CALLOC(sizeof(struct r600_kernel), shader->num_kernels);
 
diff --git a/src/gallium/drivers/radeon/radeon_llvm_util.c b/src/gallium/drivers/radeon/radeon_llvm_util.c
index e3b0d63..5a7bcd9 100644
--- a/src/gallium/drivers/radeon/radeon_llvm_util.c
+++ b/src/gallium/drivers/radeon/radeon_llvm_util.c
@@ -29,6 +29,7 @@
 
 #include <llvm-c/BitReader.h>
 #include <llvm-c/Core.h>
+#include <llvm-c/Transforms/PassManagerBuilder.h>
 
 LLVMModuleRef radeon_llvm_parse_bitcode(const unsigned char * bitcode,
 							unsigned bitcode_len)
@@ -50,6 +51,19 @@ unsigned radeon_llvm_get_num_kernels(const unsigned char *bitcode,
 	return LLVMGetNamedMetadataNumOperands(mod, "opencl.kernels");
 }
 
+static void radeon_llvm_optimize(LLVMModuleRef mod)
+{
+	LLVMPassManagerBuilderRef builder = LLVMPassManagerBuilderCreate();
+	LLVMPassManagerRef pass_manager = LLVMCreatePassManager();
+
+	LLVMPassManagerBuilderUseInlinerWithThreshold(builder, 1000000000);
+	LLVMPassManagerBuilderPopulateModulePassManager(builder, pass_manager);
+
+	LLVMRunPassManager(pass_manager, mod);
+	LLVMPassManagerBuilderDispose(builder);
+	LLVMDisposePassManager(pass_manager);
+}
+
 LLVMModuleRef radeon_llvm_get_kernel_module(unsigned index,
 		const unsigned char *bitcode, unsigned bitcode_len)
 {
@@ -72,5 +86,7 @@ LLVMModuleRef radeon_llvm_get_kernel_module(unsigned index,
 		LLVMDeleteFunction(kernel_function);
 	}
 	FREE(kernel_metadata);
+	radeon_llvm_optimize(mod);
 	return mod;
 }
+
-- 
1.8.1.5