svn commit: r520806 - in head/math/mkl-dnn: . files
Piotr Kubaj
pkubaj at FreeBSD.org
Tue Dec 24 13:40:03 UTC 2019
Author: pkubaj
Date: Tue Dec 24 13:40:02 2019
New Revision: 520806
URL: https://svnweb.freebsd.org/changeset/ports/520806
Log:
math/mkl-dnn: fix build on non-x86
Remove hardcoded -msse4.1 and instead add it to our Makefile, but only on amd64 and i386.
PR: 242847
Approved by: yuri (maintainer)
Added:
head/math/mkl-dnn/files/patch-cmake_platform.cmake (contents, props changed)
Modified:
head/math/mkl-dnn/Makefile
Modified: head/math/mkl-dnn/Makefile
==============================================================================
--- head/math/mkl-dnn/Makefile Tue Dec 24 13:36:50 2019 (r520805)
+++ head/math/mkl-dnn/Makefile Tue Dec 24 13:40:02 2019 (r520806)
@@ -11,8 +11,6 @@ COMMENT= Intel(R) Math Kernel Library for Deep Neural
LICENSE= APACHE20
LICENSE_FILE= ${WRKSRC}/LICENSE
-ONLY_FOR_ARCHS= amd64 i386
-ONLY_FOR_ARCHS_REASON= Intel MKL-DNN supports Intel 64 architecture and compatible architectures
BROKEN_i386= error: shift count >= width of type (and other errors), see https://github.com/intel/mkl-dnn/issues/315
TEST_DEPENDS= bash:shells/bash
@@ -29,6 +27,9 @@ OPTIONS_DEFAULT= OPENMP
OPENMP_CMAKE_BOOL= USE_OPENMP
OPENMP_LIB_DEPENDS= libomp.so:devel/openmp
OPENMP_BROKEN_OFF= still requires omp.h, see https://github.com/intel/mkl-dnn/issues/581
+
+CXXFLAGS_amd64= -msse4.1
+CFXXLAGS_i386= -msse4.1
post-install:
@${RM} -r ${STAGEDIR}${PREFIX}/share/doc
Added: head/math/mkl-dnn/files/patch-cmake_platform.cmake
==============================================================================
--- /dev/null 00:00:00 1970 (empty, because file is newly added)
+++ head/math/mkl-dnn/files/patch-cmake_platform.cmake Tue Dec 24 13:40:02 2019 (r520806)
@@ -0,0 +1,18 @@
+--- cmake/platform.cmake.orig 2019-12-24 01:47:09 UTC
++++ cmake/platform.cmake
+@@ -89,7 +89,6 @@ elseif(UNIX OR MINGW)
+ append(CMAKE_CCXX_NOEXCEPT_FLAGS "-fno-exceptions")
+ # compiler specific settings
+ if(CMAKE_CXX_COMPILER_ID MATCHES "Clang")
+- set(DEF_ARCH_OPT_FLAGS "-msse4.1")
+ # Clang cannot vectorize some loops with #pragma omp simd and gets
+ # very upset. Tell it that it's okay and that we love it
+ # unconditionally.
+@@ -133,7 +132,6 @@ elseif(UNIX OR MINGW)
+ append(CMAKE_CCXX_SANITIZER_FLAGS "-g -fno-omit-frame-pointer")
+ endif()
+ elseif("${CMAKE_CXX_COMPILER_ID}" STREQUAL "GNU")
+- set(DEF_ARCH_OPT_FLAGS "-msse4.1")
+ # suppress warning on assumptions made regarding overflow (#146)
+ append(CMAKE_CCXX_NOWARN_FLAGS "-Wno-strict-overflow")
+ elseif(CMAKE_CXX_COMPILER_ID STREQUAL "Intel")
More information about the svn-ports-all
mailing list