From 54c8cf842dca2557dd6ae67b4903084cf54c1ae9 Mon Sep 17 00:00:00 2001
From: Yuri Victorovich <yuri@FreeBSD.org>
Date: Fri, 6 Nov 2020 18:51:36 +0000
Subject: [PATCH] math/onednn: Add options for SIMD optimization level; Use
 standard BUILD_TESTING cmake variable instead of BUILD_TESTS

---
 math/onednn/Makefile                         | 31 +++++++++++++-------
 math/onednn/files/patch-CMakeLists.txt       |  2 +-
 math/onednn/files/patch-cmake_platform.cmake | 24 +++++++++++++--
 3 files changed, 43 insertions(+), 14 deletions(-)

diff --git a/math/onednn/Makefile b/math/onednn/Makefile
index 67a730f1cb7b..13d581277ef4 100644
--- a/math/onednn/Makefile
+++ b/math/onednn/Makefile
@@ -3,7 +3,7 @@
 PORTNAME=	onednn
 DISTVERSIONPREFIX=	v
 DISTVERSION=	1.7
-PORTREVISION=	1
+PORTREVISION=	2
 CATEGORIES=	math # machine-learning
 
 MAINTAINER=	yuri@FreeBSD.org
@@ -23,30 +23,39 @@ GH_ACCOUNT=	oneapi-src
 GH_PROJECT=	oneDNN
 USE_LDCONFIG=	yes
 
+CMAKE_OFF=	BUILD_TESTING
+
 OPTIONS_DEFINE=		OPENMP
-OPTIONS_DEFAULT=	OPENMP
+OPTIONS_DEFAULT=	OPENMP SIMD_DEFAULT
+OPTIONS_SINGLE=		SIMD
+OPTIONS_SINGLE_SIMD=	SIMD_DEFAULT SIMD_NATIVE
 
 OPENMP_CMAKE_BOOL=	USE_OPENMP
 OPENMP_BROKEN_OFF=	still requires omp.h, see https://github.com/intel/mkl-dnn/issues/581
 
-CXXFLAGS_amd64=	-msse4.1
-CXXFLAGS_i386=	-msse4.1
+SIMD_DEFAULT_DESC=	Default, no non-default SIMD instructions are used
+
+SIMD_NATIVE_DESC=	Optimize for this CPU
+SIMD_NATIVE_CXXFLAGS=	-march=native
+
+SIMD_SSE41_DESC=	Use SSE4.1 instructions
+SIMD_SSE41_CXXFLAGS=	-msse4.1
+
+ARCH_LOCAL!=	/usr/bin/uname -p # because OPTIONS_SINGLE_SIMD doesn't support per-ARCH values OPTIONS_SINGLE_SIMD_{arch}, like OPTIONS_DEFINE_{arch}
+
+.if ${ARCH_LOCAL} == i386 || ${ARCH_LOCAL} == amd64
+OPTIONS_SINGLE_SIMD+=	SIMD_SSE41
+.endif
 
 CXXFLAGS:=	${CXXFLAGS:S/-O2/-O3/} # clang writes wrong binary code when -O2 optimization is used and one testcase is failing, see https://github.com/oneapi-src/oneDNN/issues/873#issuecomment-722778910
 
-.include <bsd.port.options.mk>
-
-.if ${ARCH} != aarch64 && ${ARCH} != amd64 && ${ARCH} != powerpc64
-CMAKE_ARGS+=	-DDNNL_TARGET_ARCH:STRING="GENERIC"
-.endif
-
 post-install:
 	@${RM} -r ${STAGEDIR}${PREFIX}/share/doc
 
 do-test: # some tests are known to fail: https://github.com/intel/mkl-dnn/issues/350 (the upstream failed to reproduce it and closed the report)
 	@${REINPLACE_CMD} 's| /bin/bash | ${LOCALBASE}/bin/bash |' ${WRKSRC}/tests/CMakeLists.txt
 	@cd ${BUILD_WRKSRC} && \
-		${SETENV} ${CONFIGURE_ENV} ${CMAKE_BIN} ${CMAKE_ARGS} -DBUILD_TESTS=ON ${CMAKE_SOURCE_PATH} && \
+		${SETENV} ${CONFIGURE_ENV} ${CMAKE_BIN} ${CMAKE_ARGS} -DBUILD_TESTING=ON ${CMAKE_SOURCE_PATH} && \
 		${SETENV} ${MAKE_ENV} ${MAKE_CMD} ${MAKE_ARGS} ${ALL_TARGET} && \
 		${SETENV} ${MAKE_ENV} ${MAKE_CMD} ${MAKE_ARGS} test
 
diff --git a/math/onednn/files/patch-CMakeLists.txt b/math/onednn/files/patch-CMakeLists.txt
index 79cc4d19a98c..b5465477d487 100644
--- a/math/onednn/files/patch-CMakeLists.txt
+++ b/math/onednn/files/patch-CMakeLists.txt
@@ -20,7 +20,7 @@
 +if(BUILD_EXAMPLES)
 +  add_subdirectory(examples)
 +endif()
-+if(BUILD_TESTS)
++if(BUILD_TESTING)
 +  add_subdirectory(tests)
 +endif()
  
diff --git a/math/onednn/files/patch-cmake_platform.cmake b/math/onednn/files/patch-cmake_platform.cmake
index eb4bf3e72db7..ce18aaa0fa84 100644
--- a/math/onednn/files/patch-cmake_platform.cmake
+++ b/math/onednn/files/patch-cmake_platform.cmake
@@ -27,8 +27,28 @@
               set(DEF_ARCH_OPT_FLAGS "-msse4.1")
          endif()
          # Clang cannot vectorize some loops with #pragma omp simd and gets
-@@ -203,7 +203,7 @@ elseif(UNIX OR MINGW)
-              if (CMAKE_SYSTEM_PROCESSOR STREQUAL CMAKE_HOST_SYSTEM_PROCESSOR)
+@@ -186,24 +186,24 @@ elseif(UNIX OR MINGW)
+         if(DNNL_TARGET_ARCH STREQUAL "AARCH64")
+              set(DEF_ARCH_OPT_FLAGS "-O3")
+              # For native compilation tune for the host processor
+-             if (CMAKE_SYSTEM_PROCESSOR STREQUAL CMAKE_HOST_SYSTEM_PROCESSOR)
++             if (FALSE AND CMAKE_SYSTEM_PROCESSOR STREQUAL CMAKE_HOST_SYSTEM_PROCESSOR)
+                  append(DEF_ARCH_OPT_FLAGS "-mcpu=native")
+              endif()
+         elseif(DNNL_TARGET_ARCH STREQUAL "PPC64")
+              set(DEF_ARCH_OPT_FLAGS "-O3")
+              # In GCC, -ftree-vectorize is turned on under -O3 since 2007.
+              # For native compilation tune for the host processor
+-             if (CMAKE_SYSTEM_PROCESSOR STREQUAL CMAKE_HOST_SYSTEM_PROCESSOR)
++             if (FALSE AND CMAKE_SYSTEM_PROCESSOR STREQUAL CMAKE_HOST_SYSTEM_PROCESSOR)
+                  append(DEF_ARCH_OPT_FLAGS "-mcpu=native")
+              endif()
+         elseif(DNNL_TARGET_ARCH STREQUAL "S390X")
+              set(DEF_ARCH_OPT_FLAGS "-O3")
+              # In GCC, -ftree-vectorize is turned on under -O3 since 2007.
+              # For native compilation tune for the host processor
+-             if (CMAKE_SYSTEM_PROCESSOR STREQUAL CMAKE_HOST_SYSTEM_PROCESSOR)
++             if (FALSE AND CMAKE_SYSTEM_PROCESSOR STREQUAL CMAKE_HOST_SYSTEM_PROCESSOR)
                   append(DEF_ARCH_OPT_FLAGS "-march=native")
               endif()
 -        elseif(DNNL_TARGET_ARCH STREQUAL "X64")