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
33 lines
1.3 KiB
LLVM
33 lines
1.3 KiB
LLVM
; RUN: llc -verify-machineinstrs -march=amdgcn -mcpu=SI < %s
|
|
; RUN: llc -verify-machineinstrs -march=amdgcn -mcpu=tonga < %s
|
|
|
|
; This works because promote allocas pass replaces these with LDS atomics.
|
|
|
|
; Private atomics have no real use, but at least shouldn't crash on it.
|
|
define amdgpu_kernel void @atomicrmw_private(i32 addrspace(1)* %out, i32 %in) nounwind {
|
|
entry:
|
|
%tmp = alloca [2 x i32]
|
|
%tmp1 = getelementptr inbounds [2 x i32], [2 x i32]* %tmp, i32 0, i32 0
|
|
%tmp2 = getelementptr inbounds [2 x i32], [2 x i32]* %tmp, i32 0, i32 1
|
|
store i32 0, i32* %tmp1
|
|
store i32 1, i32* %tmp2
|
|
%tmp3 = getelementptr inbounds [2 x i32], [2 x i32]* %tmp, i32 0, i32 %in
|
|
%tmp4 = atomicrmw add i32* %tmp3, i32 7 acq_rel
|
|
store i32 %tmp4, i32 addrspace(1)* %out
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @cmpxchg_private(i32 addrspace(1)* %out, i32 %in) nounwind {
|
|
entry:
|
|
%tmp = alloca [2 x i32]
|
|
%tmp1 = getelementptr inbounds [2 x i32], [2 x i32]* %tmp, i32 0, i32 0
|
|
%tmp2 = getelementptr inbounds [2 x i32], [2 x i32]* %tmp, i32 0, i32 1
|
|
store i32 0, i32* %tmp1
|
|
store i32 1, i32* %tmp2
|
|
%tmp3 = getelementptr inbounds [2 x i32], [2 x i32]* %tmp, i32 0, i32 %in
|
|
%tmp4 = cmpxchg i32* %tmp3, i32 0, i32 1 acq_rel monotonic
|
|
%val = extractvalue { i32, i1 } %tmp4, 0
|
|
store i32 %val, i32 addrspace(1)* %out
|
|
ret void
|
|
}
|