radv: setup llvm target data layout

Ported from radeonsi, pointed out by Tom.

"This prevents LLVM from using sext instructions for local memory
offsets and allows the backend to fold immediate offsets into the
instruction. This also prevents some incorrect code generation for
ptrtoint and inttoptr instructions."

Cc: "13.0 17.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Tom Stellard <tstellar@redhat.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>
(cherry picked from commit b8ee70384a)
[Emil Velikov: resolve trivial conflicts]
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>

Conflicts:
	src/amd/common/ac_nir_to_llvm.c
This commit is contained in:
Dave Airlie 2017-03-14 06:50:59 +10:00 committed by Emil Velikov
parent dac86c5d3c
commit 1d143f0018

View file

@ -4499,6 +4499,13 @@ LLVMModuleRef ac_translate_nir_to_llvm(LLVMTargetMachineRef tm,
memset(shader_info, 0, sizeof(*shader_info));
LLVMSetTarget(ctx.module, "amdgcn--");
LLVMTargetDataRef data_layout = LLVMCreateTargetDataLayout(tm);
char *data_layout_str = LLVMCopyStringRepOfTargetData(data_layout);
LLVMSetDataLayout(ctx.module, data_layout_str);
LLVMDisposeTargetData(data_layout);
LLVMDisposeMessage(data_layout_str);
setup_types(&ctx);
ctx.builder = LLVMCreateBuilderInContext(ctx.context);