updates based on PR feedback

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

View File

@@ -9,7 +9,7 @@
#include "main.h"
#include <Eigen/src/Core/arch/CUDA/Half.h>
#include <Eigen/src/Core/arch/GPU/Half.h>
// Make sure it's possible to forward declare Eigen::half
namespace Eigen {

View File

@@ -68,9 +68,19 @@
// are defined here and any not-parenthesized min/max call will cause a
// compiler error.
#if !defined(__HIPCC__)
// HIP headers include the <thread> header which contains not-parenthesized
// calls to "max", triggering the following check and causing the compile to fail
// so disabling the following checks for HIP
//
// HIP header files include the following files
// <thread>
// <regex>
// <unordered_map>
// which seem to contain not-parenthesized calls to "max"/"min", triggering the following check and causing the compile to fail
//
// Including those header files before the following macro definition for "min" / "max", only partially resolves the issue
// This is because other HIP header files also define "isnan" / "isinf" / "isfinite" functions, which are needed in other
// headers.
//
// So instead choosing to simply disable this check for HIP
//
#define min(A,B) please_protect_your_min_with_parentheses
#define max(A,B) please_protect_your_max_with_parentheses
#define isnan(X) please_protect_your_isnan_with_parentheses