[llvm] [AMDGPU][test] Fix the wrong os names in triple. NFC (PR #93501)

Lu Weining via llvm-commits llvm-commits at lists.llvm.org
Tue May 28 05:49:53 PDT 2024


SixWeining wrote:

> LGTM but the test will need update, this would have defaulted to unknown

lower-work-group-id-intrinsics-hsa.ll is the case but lower-work-group-id-intrinsics-pal.ll is not.

If use `-mtriple=amdgcn-amd-amdhsa`, we got error:
```
error: <unknown>:0:0: in function _amdgpu_cs_main void (): unsupported non-compute shaders with HSA
                                               
error: <unknown>:0:0: in function caller void (): unsupported non-compute shaders with HSA

llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.h:910: void llvm::SIMachineFunctionInfo::setScratchRSrcReg(llvm::Register): Assertion `Reg != 0 && "S
hould never be unset"' failed.
```

I suppose  that lower-work-group-id-intrinsics-pal.ll only needs RUN lines with `-mtriple=amdgcn-amd-amdpal` while lower-work-group-id-intrinsics-hsa.ll only needs RUN lines with `-mtriple=amdgcn-amd-amdhsa`.

https://github.com/llvm/llvm-project/pull/93501


More information about the llvm-commits mailing list