From 1bea71a151b552ddfed1633a9ab7b4cf6d003bd6 Mon Sep 17 00:00:00 2001
From: Samuel Audet <samuel.audet@gmail.com>
Date: Thu, 19 Nov 2020 11:35:57 +0900
Subject: [PATCH] Fix build for OpenBLAS on Linux with AVX-512

---
 ci/install-travis.sh | 2 ++
 openblas/cppbuild.sh | 9 ++++-----
 2 files changed, 6 insertions(+), 5 deletions(-)

diff --git a/ci/install-travis.sh b/ci/install-travis.sh
index bcb62c6d150..383bda9b50f 100755
--- a/ci/install-travis.sh
+++ b/ci/install-travis.sh
@@ -70,6 +70,8 @@ if [[ "$OS" == "linux-x86" ]] || [[ "$OS" == "linux-x86_64" ]] || [[ "$OS" =~ an
       docker exec -ti $DOCKER_CONTAINER_ID /bin/bash -xec "rpm -i --force --nodeps $HOME/*.rpm"
     fi
   fi
+  # https://gcc.gnu.org/legacy-ml/gcc-patches/2018-01/msg01962.html
+  docker exec -ti $DOCKER_CONTAINER_ID /bin/bash -xec "sed -i 's/_mm512_abs_pd (__m512 __A)/_mm512_abs_pd (__m512d __A)/g' /opt/rh/devtoolset-7/root/usr/lib/gcc/x86_64-redhat-linux/7/include/avx512fintrin.h"
   docker exec -ti $DOCKER_CONTAINER_ID /bin/bash -xec "rm -f /usr/lib/libgfortran.so.3* /usr/lib64/libgfortran.so.3*" # not required for GCC 7+
   docker exec -ti $DOCKER_CONTAINER_ID /bin/bash -xec "source scl_source enable $SCL_ENABLE || true; gcc --version"
   docker exec -ti $DOCKER_CONTAINER_ID /bin/bash -xec "gpg --version"
diff --git a/openblas/cppbuild.sh b/openblas/cppbuild.sh
index 5fb22b32f77..c55cd2d8510 100755
--- a/openblas/cppbuild.sh
+++ b/openblas/cppbuild.sh
@@ -26,9 +26,6 @@ cp lapack-netlib/LAPACKE/include/*.h ../include
 # remove broken cross-compiler workaround on Mac
 sedinplace '/if (($os eq "Darwin")/,/}/d' c_check ../OpenBLAS-$OPENBLAS_VERSION-nolapack/c_check
 
-# fix AVX512 on Linux
-sedinplace 's/_mm512_abs_pd(_mm512_load_pd/_mm512_abs_pd((__m512)_mm512_load_pd/g' kernel/x86_64/dasum_microk_skylakex-2.c ../OpenBLAS-$OPENBLAS_VERSION-nolapack/kernel/x86_64/dasum_microk_skylakex-2.c
-
 # blas (requires fortran, e.g. sudo yum install gcc-gfortran)
 export FEXTRALIB="-lgfortran"
 export CROSS_SUFFIX=
@@ -36,6 +33,7 @@ export HOSTCC=gcc
 export NO_LAPACK=0
 export NUM_THREADS=64
 export NO_AFFINITY=1
+export NO_AVX512=1
 case $PLATFORM in
     android-arm)
         patch -Np1 < ../../../OpenBLAS-android.patch
@@ -157,6 +155,7 @@ case $PLATFORM in
         export BINARY=64
         export DYNAMIC_ARCH=1
         export TARGET=NEHALEM
+        export NO_AVX512=0
         ;;
     linux-ppc64le)
         # patch to use less buggy generic kernels
@@ -232,12 +231,12 @@ case $PLATFORM in
         ;;
 esac
 
-make -s -j $MAKEJ libs netlib shared "CROSS_SUFFIX=$CROSS_SUFFIX" "CC=$CC" "FC=$FC" "HOSTCC=$HOSTCC" BINARY=$BINARY COMMON_PROF= F_COMPILER=GFORTRAN "FEXTRALIB=$FEXTRALIB" USE_OPENMP=0 NUM_THREADS=$NUM_THREADS
+make -s -j $MAKEJ libs netlib shared "CROSS_SUFFIX=$CROSS_SUFFIX" "CC=$CC" "FC=$FC" "HOSTCC=$HOSTCC" BINARY=$BINARY COMMON_PROF= F_COMPILER=GFORTRAN "FEXTRALIB=$FEXTRALIB" USE_OPENMP=0 NUM_THREADS=$NUM_THREADS NO_AVX512=$NO_AVX512
 make install "PREFIX=$INSTALL_PATH"
 
 unset DYNAMIC_ARCH
 cd ../OpenBLAS-$OPENBLAS_VERSION-nolapack/
-make -s -j $MAKEJ libs netlib shared "CROSS_SUFFIX=$CROSS_SUFFIX" "CC=$CC" "FC=$FC" "HOSTCC=$HOSTCC" BINARY=$BINARY COMMON_PROF= F_COMPILER=GFORTRAN "FEXTRALIB=$FEXTRALIB" USE_OPENMP=0 NUM_THREADS=$NUM_THREADS NO_LAPACK=1 LIBNAMESUFFIX=nolapack
+make -s -j $MAKEJ libs netlib shared "CROSS_SUFFIX=$CROSS_SUFFIX" "CC=$CC" "FC=$FC" "HOSTCC=$HOSTCC" BINARY=$BINARY COMMON_PROF= F_COMPILER=GFORTRAN "FEXTRALIB=$FEXTRALIB" USE_OPENMP=0 NUM_THREADS=$NUM_THREADS NO_AVX512=$NO_AVX512 NO_LAPACK=1 LIBNAMESUFFIX=nolapack
 make install "PREFIX=$INSTALL_PATH" NO_LAPACK=1 LIBNAMESUFFIX=nolapack
 
 unset CC