portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-04 19:25:56 +01:00
Code Releases Activity

Add GPU specific simd targets

Browse Source
This commit is contained in:
paboyle 2018-06-13 19:55:30 +01:00
parent eb7d34a4cc
commit 7a4c142955
1 changed files with 14 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
configure.ac
View File

@ -251,6 +251,8 @@ esac
##################### Compiler dependent choices ##################### Compiler dependent choices
case ${CXX} in case ${CXX} in
nvcc) nvcc)
# CXX="nvcc -keep -v -x cu "
# CXXLD="nvcc -v -link"
CXX="nvcc -x cu " CXX="nvcc -x cu "
CXXLD="nvcc -link" CXXLD="nvcc -link"
CXXFLAGS="$CXXFLAGS -Xcompiler -fno-strict-aliasing" CXXFLAGS="$CXXFLAGS -Xcompiler -fno-strict-aliasing"
@ -264,6 +266,12 @@ esac
case ${ax_cv_cxx_compiler_vendor} in case ${ax_cv_cxx_compiler_vendor} in
clang|gnu) clang|gnu)
case ${ac_SIMD} in case ${ac_SIMD} in
GPU)
AC_DEFINE([GPU],[1],[GPU float4 vectors])
SIMD_FLAGS='';;
VGPU)
AC_DEFINE([GPU_VEC],[1],[GPU vectorised 512bit])
SIMD_FLAGS='';;
SSE4) SSE4)
AC_DEFINE([SSE4],[1],[SSE4 intrinsics]) AC_DEFINE([SSE4],[1],[SSE4 intrinsics])
case ${ac_SFW_FP16} in case ${ac_SFW_FP16} in
@ -313,6 +321,12 @@ case ${ax_cv_cxx_compiler_vendor} in
esac;; esac;;
intel) intel)
case ${ac_SIMD} in case ${ac_SIMD} in
GPU)
AC_DEFINE([GPU],[1],[GPU float4 vectors])
SIMD_FLAGS='';;
VGPU)
AC_DEFINE([GPU_VEC],[1],[GPU vectorised ])
SIMD_FLAGS='';;
SSE4) SSE4)
AC_DEFINE([SSE4],[1],[SSE4 intrinsics]) AC_DEFINE([SSE4],[1],[SSE4 intrinsics])
SIMD_FLAGS='-msse4.2 -xsse4.2';; SIMD_FLAGS='-msse4.2 -xsse4.2';;
@ -351,13 +365,6 @@ esac
AM_CXXFLAGS="$SIMD_FLAGS $AM_CXXFLAGS" AM_CXXFLAGS="$SIMD_FLAGS $AM_CXXFLAGS"
AM_CFLAGS="$SIMD_FLAGS $AM_CFLAGS" AM_CFLAGS="$SIMD_FLAGS $AM_CFLAGS"
case ${ac_SIMD} in
AVX512|KNL)
AC_DEFINE([TEST_ZMM],[1],[compile ZMM test]);;
*)
;;
esac
############### Precision selection ############### Precision selection
AC_ARG_ENABLE([precision], AC_ARG_ENABLE([precision],
[AC_HELP_STRING([--enable-precision=single|double], [AC_HELP_STRING([--enable-precision=single|double],