mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2024-10-19 19:12:56 +02:00
dd9ab77318
Currently the default C calling convention functions are treated the same as compute kernels. Make this explicit so the default calling convention can be changed to a non-kernel. Converted with perl -pi -e 's/define void/define amdgpu_kernel void/' on the relevant test directories (and undoing in one place that actually wanted a non-kernel). llvm-svn: 298444
29 lines
1.0 KiB
LLVM
29 lines
1.0 KiB
LLVM
; RUN: llc -march=amdgcn -verify-machineinstrs -verify-coalescing < %s
|
|
|
|
; The original and requires materializing a 64-bit immediate for
|
|
; s_and_b64. This is split into 2 x v_and_i32, part of the immediate
|
|
; is folded through the reg_sequence into the v_and_i32 operand, and
|
|
; only half of the result is ever used.
|
|
;
|
|
; During live interval construction, the first sub register def is
|
|
; incorrectly marked as dead.
|
|
|
|
declare i32 @llvm.amdgcn.workitem.id.x() #1
|
|
|
|
define amdgpu_kernel void @dead_def_subregister(i32 addrspace(1)* noalias %out, i64 addrspace(1)* noalias %in) #0 {
|
|
%tid = call i32 @llvm.amdgcn.workitem.id.x() #1
|
|
%in.gep = getelementptr i64, i64 addrspace(1)* %in, i32 %tid
|
|
%val = load i64, i64 addrspace(1)* %in.gep
|
|
|
|
%lshr = shl i64 %val, 24
|
|
%and1 = and i64 %lshr, 2190433320969 ; (255 << 33) | 9
|
|
%vec = bitcast i64 %and1 to <2 x i32>
|
|
%elt1 = extractelement <2 x i32> %vec, i32 1
|
|
|
|
store i32 %elt1, i32 addrspace(1)* %out
|
|
ret void
|
|
}
|
|
|
|
attributes #0 = { nounwind }
|
|
attributes #1 = { nounwind readnone }
|