[llvm] [AMDGPU] Fold copies of constant physical registers into their uses (PR #154183)

[Stanislav Mekhanoshin via llvm-commits]

rampitec wrote:

> Thanks for reverting it.

Sure. I cannot fix it quickly.

> For local repro I used the cmake config:
> 
> > -DCMAKE_BUILD_TYPE=Release -DCMAKE_C_COMPILER_LAUNCHER=ccache -DCMAKE_CXX_COMPILER_LAUNCHER=ccache -DLLVM_ENABLE_ASSERTIONS=ON '-DLLVM_LIT_ARGS=-vv --show-unsupported --timeout 120 --show-xfail -j 64'  '-DLLVM_TARGETS_TO_BUILD=X86;AMDGPU' '-DLLVM_ENABLE_RUNTIMES=compiler-rt;libunwind;libcxx;libcxxabi;openmp;offload' '-DLLVM_ENABLE_PROJECTS=clang;lld;llvm' -DCLANG_DEFAULT_LINKER=lld -DLIBCXX_ENABLE_SHARED=OFF -DLIBCXX_ENABLE_STATIC=ON -DLIBCXX_INSTALL_LIBRARY=OFF -DLIBCXX_INSTALL_HEADERS=OFF -DLIBCXXABI_ENABLE_SHARED=OFF -DLIBCXXABI_ENABLE_STATIC=ON -DLIBCXXABI_INSTALL_STATIC_LIBRARY=OFF -DLLVM_ENABLE_Z3_SOLVER=OFF -DLLVM_ENABLE_ZLIB=ON -DLLVM_ENABLE_PER_TARGET_RUNTIME_DIR=ON -DCMAKE_CXX_STANDARD=17 -DBUILD_SHARED_LIBS=ON -DLLVM_ENABLE_LIBCXX=ON -DCLANG_DEFAULT_RTLIB=compiler-rt -DCLANG_DEFAULT_UNWINDLIB=libgcc -DLIBOMPTARGET_PLUGINS_TO_BUILD='amdgpu;host' -DPACKAGE_VENDOR='LLVM' -DLLVM_RUNTIME_TARGETS='default;amdgcn-amd-amdhsa' -DRUNTIMES_amdgcn-amd-amdhsa_LLVM_ENABLE_RUNTIMES=libc

Yep, already did that.
 
> and run the test with:
> 
> > ninja  check-libc-amdgcn-amd-amdhsa

That is where I really cannot see what's happening. I really unable to run the test as this is an offline builder:
```
-- No GPU architecture detected or provided, tests will not be built
```
Do you happen to know:
1) On which gpu arch did it fail? Unclear from the log.
2) Is there a way to force the gpu arch just to build the test?

In essence all I need is an IR of the failing test, all I need is to compare what it compiles to before and after the change. If by the chance you have the IR I would appreciate it.


https://github.com/llvm/llvm-project/pull/154183