+ Add a corresponding test kernel_launch_01.cu
+ Add isBefore() check to avoid crash on Replacement with negative length
TODO:
+ Compatibility with former LLVM versions
+ More complicated kernel launch tests
By implicit unconditional passing -fno-delayed-template-parsing option (which appeared in LLVM 3.8.0, thus doesn't need compatibility wrapping) to hipify-clang.
[Reason] To parse uncalled template functions otherwise they are not parsed without calling, thus not hipified.
Affects cub_03.cu test, which has uncalled global template function.
[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.
[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)
+ affects cuDNN and CUB tests, paths to libraries of which are defined by CUDA_DNN_ROOT_DIR and CUDA_CUB_ROOT_DIR
+ Warn about excluding and why, for instance:
"WARN: cuDNN tests are excluded due to unset CUDA_DNN_ROOT_DIR"
+ Add one matcher (will be more)
+ Update Maps and Statistics
+ Add cub_01.cu unit test
+ Update lit harness to support standalone CUB
+ Update README.md
+ Update hipify-perl (only CUB header is supported for now)
[IMPORTANT]
clang (and hipify-clang) works correctly only with official NVLabs version on GitHub.
Compilation of CUB from official CUDA release has conflicts with THRUST.
Thus, to compile CUB sources, option "-I" should be specified to the cloned CUB from NVLAB on GitHub.
[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.
+ Perl part of [#1458]
+ Affected functions: hipFuncSetCacheConfig, hipFuncGetAttributes
+ Implement function generateHostFunctions() in hipify-clang for that purposes
+ Update hipify-perl accordingly
+ Affected functions: hipFuncSetCacheConfig, hipFuncGetAttributes
+ Add a corresponding Matcher cudaReinterpretCastArgFuncCall
+ Add reinterpret_cast.cu test
TODO: Do the same for hipify-perl
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
+ 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
+ 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.
+ Start to translate preprocessor's false conditional blocks too:
based on clang's https://reviews.llvm.org/D66597;
available only starting from LLVM 10.0 or trunk.
+ Option -skip-excluded-preprocessor-conditional-blocks for skipping excluded conditional blocks:
the default behavior for hipify-clang built with LLVM < 10.0;
false by default for hipify-clang built with LLVM 10 or trunk.
+ Add 4 preprocessor unit tests, 2 of which are LLVM 10.0 only
+ Update couple of existing tests by setting -skip-excluded-preprocessor-conditional-blocks option:
update lit testing accordingly
...while including HIP main header file, which is inserted now after #indef controlling macro, or after #pragma once, if it's occurred earlier.
+ Add a couple of unit tests.
ToDo: Check backward compatibility on older clang versions.
[Reason] To be compatible with CUDA [#1133]
Update HIP code, hipify-clang, tests and docs
[TODO] Add support of the corresponding functions on nvcc fallback path
+ Add missing cuda_profiler_api.h to hip/hip_profile.h transformation.
NOTE: HIP Profiler API is under development. This is NOT WORKING example.
TODO: Find out a way to generate HIP_SCOPED_MARKER, HIP_BEGIN_MARKER, HIP_END_MARKER, declared in hip/hip_profile.h in particular place (signatures are to obtain).
+ Set -D__LP64__ in case of 64-bit hipify-clang binary
[partial workaround for clang's bug https://bugs.llvm.org/show_bug.cgi?id=38811]
C:/GIT/LLVM/trunk/llvm-64-release-vs2017/dist/lib/clang/9.0.0\include\__clang_cuda_device_functions.h(1609,45): error GEF7559A7: no matching function for call to 'roundf'
__DEVICE__ long lroundf(float __a) { return roundf(__a); }
#if defined(__LP64__)
__DEVICE__ long lround(double __a) { return llround(__a); }
__DEVICE__ long lroundf(float __a) { return llroundf(__a); } // ok: llroundf should be used when 64-bit
#else
__DEVICE__ long lround(double __a) { return round(__a); }
__DEVICE__ long lroundf(float __a) { return roundf(__a); } // error
#endif
+ Print more system info while testing in the following form:
========================================
CUDA 9.0 - will be used for testing
LLVM 9.0.0svn - will be used for testing
AMD64 - Platform architecture
Windows 10 - Platform OS
64 - hipify-clang binary bitness
32 - python 3.7.2 binary bitness
========================================
root
Maneesh Gupta
Evgeny Mankov