devtools/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2026-04-10 11:34:33 +08:00
Code Issues Packages Projects Releases Wiki Activity

updates based on PR feedback

Browse Source 
There are two major changes (and a few minor ones which are not listed here...see PR discussion for details)

1. Eigen::half implementations for HIP and CUDA have been merged.
This means that
- `CUDA/Half.h` and `HIP/hcc/Half.h` got merged to a new file `GPU/Half.h`
- `CUDA/PacketMathHalf.h` and `HIP/hcc/PacketMathHalf.h` got merged to a new file `GPU/PacketMathHalf.h`
- `CUDA/TypeCasting.h` and `HIP/hcc/TypeCasting.h` got merged to a new file `GPU/TypeCasting.h`

After this change the `HIP/hcc` directory only contains one file `math_constants.h`. That will go away too once that file becomes a part of the HIP install.

2. new macros EIGEN_GPUCC, EIGEN_GPU_COMPILE_PHASE and EIGEN_HAS_GPU_FP16 have been added and the code has been updated to use them where appropriate.
- `EIGEN_GPUCC` is the same as `(EIGEN_CUDACC || EIGEN_HIPCC)`
- `EIGEN_GPU_DEVICE_COMPILE` is the same as `(EIGEN_CUDA_ARCH || EIGEN_HIP_DEVICE_COMPILE)`
- `EIGEN_HAS_GPU_FP16` is the same as `(EIGEN_HAS_CUDA_FP16 or EIGEN_HAS_HIP_FP16)`
This commit is contained in:
Deven Desai
2018-06-14 10:21:54 -04:00
parent ba972fb6b4
commit b6cc0961b1
29 changed files with 425 additions and 185 deletions
Download Patch File Download Diff File Expand all files Collapse all files

71
Eigen/Core
View File

@@ -99,6 +99,61 @@
#define EIGEN_DONT_VECTORIZE
#endif
#if defined(EIGEN_CUDACC) || defined(EIGEN_HIPCC)
//
// If either EIGEN_CUDACC or EIGEN_HIPCC is defined, then define EIGEN_GPUCC
//
#define EIGEN_GPUCC
//
// EIGEN_HIPCC implies the HIP compiler and is used to tweak Eigen code for use in HIP kernels
// EIGEN_CUDACC implies the CUDA compiler and is used to tweak Eigen code for use in CUDA kernels
//
// In most cases the same tweaks are required to the Eigen code to enable in both the HIP and CUDA kernels.
// For those cases, the corresponding code should be guarded with
// #if defined(EIGEN_GPUCC)
// instead of
// #if defined(EIGEN_CUDACC) || defined(EIGEN_HIPCC)
//
// For cases where the tweak is specific to HIP, the code should be guarded with
// #if defined(EIGEN_HIPCC)
//
// For cases where the tweak is specific to CUDA, the code should be guarded with
// #if defined(EIGEN_CUDACC)
//
#endif
#if defined(EIGEN_CUDA_ARCH) || defined(EIGEN_HIP_DEVICE_COMPILE)
//
// If either EIGEN_CUDA_ARCH or EIGEN_HIP_DEVICE_COMPILE is defined, then define EIGEN_GPU_COMPILE_PHASE
//
#define EIGEN_GPU_COMPILE_PHASE
//
// GPU compilers (HIPCC, NVCC) typically do two passes over the source code,
// + one to compile the source for the "host" (ie CPU)
// + another to compile the source for the "device" (ie. GPU)
//
// Code that needs to enabled only during the either the "host" or "device" compilation phase
// needs to be guarded with a macro that indicates the current compilation phase
//
// EIGEN_HIP_DEVICE_COMPILE implies the device compilation phase in HIP
// EIGEN_CUDA_ARCH implies the device compilation phase in CUDA
//
// In most cases, the "host" / "device" specific code is the same for both HIP and CUDA
// For those cases, the code should be guarded with
// #if defined(EIGEN_GPU_COMPILE_PHASE)
// instead of
// #if defined(EIGEN_CUDA_ARCH) || defined(EIGEN_HIP_DEVICE_COMPILE)
//
// For cases where the tweak is specific to HIP, the code should be guarded with
// #if defined(EIGEN_HIP_DEVICE_COMPILE)
//
// For cases where the tweak is specific to CUDA, the code should be guarded with
// #if defined(EIGEN_CUDA_ARCH)
//
#endif
// When compiling CUDA device code with NVCC, or HIP device code with HIPCC
// pull in math functions from the global namespace. In host mode, and when
// device doee with clang, use the std versions.
@@ -312,6 +367,10 @@
#endif
#endif
#if defined(EIGEN_HAS_CUDA_FP16) || defined(EIGEN_HAS_HIP_FP16)
#define EIGEN_HAS_GPU_FP16
#endif
#if (defined _OPENMP) && (!defined EIGEN_DONT_PARALLELIZE)
#define EIGEN_HAS_OPENMP
#endif
@@ -475,15 +534,9 @@ using std::ptrdiff_t;
#endif
// Half float support
#if defined EIGEN_USE_HIP
#include "src/Core/arch/HIP/hcc/Half.h"
#include "src/Core/arch/HIP/hcc/PacketMathHalf.h"
#include "src/Core/arch/HIP/hcc/TypeCasting.h"
#else
#include "src/Core/arch/CUDA/Half.h"
#include "src/Core/arch/CUDA/PacketMathHalf.h"
#include "src/Core/arch/CUDA/TypeCasting.h"
#endif
#include "src/Core/arch/GPU/Half.h"
#include "src/Core/arch/GPU/PacketMathHalf.h"
#include "src/Core/arch/GPU/TypeCasting.h"
#if defined EIGEN_VECTORIZE_CUDA
#include "src/Core/arch/CUDA/PacketMath.h"