[PATCH] D60907: [OpenMP][WIP] Add math functions support in OpenMP offloading

gtbercea updated this revision to Diff 196619. gtbercea added a comment. - Use macros. Repository: rC Clang CHANGES SINCE LAST ACTION https://reviews.llvm.org/D60907/new/ https://reviews.llvm.org/D60907 Files: include/clang/Driver/ToolChain.h lib/Driver/ToolChains/Clang.cpp lib/Driv

[PATCH] D60907: [OpenMP][WIP] Add math functions support in OpenMP offloading

gtbercea added a subscriber: gregrodgers. gtbercea added a comment. @gregrodgers Repository: rC Clang CHANGES SINCE LAST ACTION https://reviews.llvm.org/D60907/new/ https://reviews.llvm.org/D60907 ___ cfe-commits mailing list cfe-commits@lists

[PATCH] D60907: [OpenMP][WIP] Add math functions support in OpenMP offloading

gtbercea added a comment. In D60907#1473406 , @Hahnfeld wrote: > So the scheme is: `pow` is defined in `__clang_openmp_math.h` to call > `__kmpc_pow`. This lives in `libomptarget-nvptx` (both bc and static lib) and > just calls `pow` which works because

[PATCH] D60907: [OpenMP][WIP] Add math functions support in OpenMP offloading

Hahnfeld added a comment. So the scheme is: `pow` is defined in `__clang_openmp_math.h` to call `__kmpc_pow`. This lives in `libomptarget-nvptx` (both bc and static lib) and just calls `pow` which works because `nvcc` and Clang in CUDA mode make sure that the call gets routed into `libdevice`?

[PATCH] D60907: [OpenMP][WIP] Add math functions support in OpenMP offloading

gtbercea updated this revision to Diff 195915. gtbercea edited the summary of this revision. gtbercea added a comment. - Address comments. Repository: rC Clang CHANGES SINCE LAST ACTION https://reviews.llvm.org/D60907/new/ https://reviews.llvm.org/D60907 Files: include/clang/Driver/Tool

[PATCH] D60907: [OpenMP][WIP] Add math functions support in OpenMP offloading

jdoerfert added a comment. To follow up on my comment why this is NVPTX specific: Is there a reason why this has to happen in the Cuda ToolChain part? I would have assumed us to add the declarations similar to the ones provided in `__clang_openmp_math.h` whenever we may compile for a target. So,

[PATCH] D60907: [OpenMP][WIP] Add math functions support in OpenMP offloading

jdoerfert added inline comments. Comment at: include/clang/Driver/ToolChain.h:575 + /// Add arguments to use system-specific CUDA includes. + virtual void AddMathDeviceFunctions(const llvm::opt::ArgList &DriverArgs, Copy & Past comment Comm

[PATCH] D60907: [OpenMP][WIP] Add math functions support in OpenMP offloading

ABataev added inline comments. Comment at: lib/Headers/__clang_openmp_math.h:2 + +#ifndef __CLANG_OMP_CMATH_H__ +#define __CLANG_OMP_CMATH_H__ Why `__CLANG_OMP_CMATH_H__`? Your file is `..._math.h`, not `..._cmath.h`. Plus, seems to me, you're missing standard h

[PATCH] D60907: [OpenMP][WIP] Add math functions support in OpenMP offloading

ABataev added inline comments. Comment at: lib/Headers/__clang_openmp_math.h:14 +double __kmpc_pow(double, double); +double __kmpc_sin(double); + Also, versions for float and long double Comment at: lib/Headers/__clang_openmp_math.h:21 +// Defi

[PATCH] D60907: [OpenMP][WIP] Add math functions support in OpenMP offloading

gtbercea created this revision. gtbercea added reviewers: ABataev, hfinkel, caomhin. Herald added subscribers: cfe-commits, jdoerfert, guansong, mgorny. Herald added a project: clang. gtbercea added a reviewer: tra. gtbercea added parent revisions: D60906: [OpenMP][libomptarget][WIP] Add math func