[Reason] To support maximum CUDA features in offline tests
+ Add defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 600 restriction for atomicAdd on doubles in atomics.cu.
So if LLVM < 7 and --cuda-gpu-arch doesn't work, __CUDA_ARCH__ is unset too (350 by default in clang);
if LLVM >= 7 --cuda-gpu-arch is used and __CUDA_ARCH__ is set based on it.
[ROCm/hip commit: 39e7d213cf]
[Reason] To support maximum CUDA features in offline tests
+ Add CUDA_VERSION >= 800 restriction for atomics.cu
[TODO] Find a way to use or exclude atomicAdd for doubles if LLVM < 7, because
LLVM 6.0.1 and older do not use --cuda-gpu-arch in clang's Driver code at all (option is only declared)
[ROCm/hip commit: 14b4df126c]
[REASON]
1. hip-clang is fine with the templated kernel launch, brackets are unneeded: HIP_KERNEL_NAME(...) __VA_ARGS__
2. HCC is not, thus: HIP_KERNEL_NAME(...) (__VA_ARGS__)
[TODO] Clean-up entirely kernel name wrapping when HCC is finally obsolete.
+ Update perl generation, hipify-perl, and affected tests accordingly.
[ROCm/hip commit: 6bb9913e8a]
cudaMemcpyToSymbol, cudaMemcpyToSymbolAsync, cudaGetSymbolSize, cudaGetSymbolAddress, cudaMemcpyFromSymbol, cudaMemcpyFromSymbolAsync
+ Add a corresponding cudaSymbolFuncCall matcher.
+ Add device_symbols.cu test for the above 6 functions, update existed.
+ Fix dim3() type cast issue, update affected tests.
TODO: Do the same in hipify-perl
[ROCm/hip commit: 3722d5b4b9]
+ Do not treat somenamespace::device_function_name as a device function
+ Fix generation of warnUnsupportedDeviceFunctions function in hipify-clang
+ Update hipify-perl based on hipify-clang -perl generation
+ Update device test math_functions.cu for hipify-perl
[Restrictions]
- hipify-perl is yet unable to handle function declarations in user namespaces
- hipify-perl is yet unable to handle using directive
[ROCm/hip commit: 4f59ec25fe]
+ Add a corresponding matcher cudaDeviceFuncCall to match only (__device__ or __global__) and not __host__ functions.
+ Add a corresponding device functions mapping:
only unsupported are listed, cause supported are exactly the same as of CUDA and do not need transformation;
make FindAndReplace for device functions separated from host API calls.
+ Add a test to distinguish device functions and user-defined.
[ROCm/hip commit: 6602fadc16]
Evgeny Mankov