GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-06-04 18:54:00 +08:00
Code Issues Packages Projects Releases Wiki Activity

Do not disable alignment with EIGEN_GPUCC

Browse Source
This commit is contained in:
Eugene Zhulenev 2018-12-03 15:54:10 -08:00
parent 0ea7ae7213
commit fd0fbfa9b5
2 changed files with 7 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
Eigen/src/Core/util/ConfigureVectorization.h
View File

@ -10,13 +10,6 @@
#ifndef EIGEN_CONFIGURE_VECTORIZATION_H #ifndef EIGEN_CONFIGURE_VECTORIZATION_H
#define EIGEN_CONFIGURE_VECTORIZATION_H #define EIGEN_CONFIGURE_VECTORIZATION_H
// FIXME: not sure why this is needed, perhaps it is not needed anymore.
#ifdef __NVCC__
#ifndef EIGEN_DONT_VECTORIZE
#define EIGEN_DONT_VECTORIZE
#endif
#endif
//------------------------------------------------------------------------------------------ //------------------------------------------------------------------------------------------
// Static and dynamic alignment control // Static and dynamic alignment control
// //
@ -183,7 +176,13 @@
//---------------------------------------------------------------------- //----------------------------------------------------------------------
// If we are compiling for GPU we should also disable vectorization because
// all the packet functions are not marked as __device__ functions.
#ifdef EIGEN_GPUCC
#ifndef EIGEN_DONT_VECTORIZE
#define EIGEN_DONT_VECTORIZE
#endif
#endif
// if alignment is disabled, then disable vectorization. Note: EIGEN_MAX_ALIGN_BYTES is the proper check, it takes into // if alignment is disabled, then disable vectorization. Note: EIGEN_MAX_ALIGN_BYTES is the proper check, it takes into
// account both the user's will (EIGEN_MAX_ALIGN_BYTES,EIGEN_DONT_ALIGN) and our own platform checks // account both the user's will (EIGEN_MAX_ALIGN_BYTES,EIGEN_DONT_ALIGN) and our own platform checks

4
Eigen/src/Core/util/Macros.h
View File

@ -742,10 +742,6 @@
// All functions callable from CUDA/HIP code must be qualified with __device__ // All functions callable from CUDA/HIP code must be qualified with __device__
#ifdef EIGEN_GPUCC #ifdef EIGEN_GPUCC
#ifndef EIGEN_DONT_VECTORIZE
#define EIGEN_DONT_VECTORIZE
#endif
#define EIGEN_DEVICE_FUNC __host__ __device__ #define EIGEN_DEVICE_FUNC __host__ __device__
#else #else
#define EIGEN_DEVICE_FUNC #define EIGEN_DEVICE_FUNC