From 214fbcee1549a152ec1a2ef21d3f00479f6cc299 Mon Sep 17 00:00:00 2001
From: Rafael Cardoso Fernandes Sousa <rafaelcfsousa@ibm.com>
Date: Thu, 9 Dec 2021 08:28:17 -0600
Subject: [PATCH] Fix cmake for power

---
 cmake/cc.cmake               | 28 ++++++++++++++++++++++++++++
 cmake/system.cmake           | 34 ++++++++++++++++++++++++++++++++--
 cmake/system_check.cmake     |  6 +++---
 driver/others/CMakeLists.txt |  2 ++
 4 files changed, 65 insertions(+), 5 deletions(-)

diff --git a/cmake/cc.cmake b/cmake/cc.cmake
index b28209c0c..06bc14986 100644
--- a/cmake/cc.cmake
+++ b/cmake/cc.cmake
@@ -161,6 +161,34 @@ if (${CORE} STREQUAL ARMV8SVE)
   endif ()
 endif ()
 
+if (${CORE} STREQUAL POWER10)
+  if (NOT DYNAMIC_ARCH)
+    execute_process(COMMAND ${CMAKE_C_COMPILER} -dumpversion OUTPUT_VARIABLE GCC_VERSION)
+    if (${GCC_VERSION} VERSION_GREATER 10.2 OR ${GCC_VERSION} VERSION_EQUAL 10.2)
+      set (CCOMMON_OPT  "${CCOMMON_OPT} -mcpu=power10 -mtune=power10 -mvsx -fno-fast-math")
+    else ()
+      message(FATAL_ERROR "Compiler GCC.${GCC_VERSION} does not support Power10." )
+    endif()
+  endif ()
+endif ()
+
+if (${CORE} STREQUAL POWER9)
+  if (NOT DYNAMIC_ARCH)
+    execute_process(COMMAND ${CMAKE_C_COMPILER} -dumpversion OUTPUT_VARIABLE GCC_VERSION)
+    if (${GCC_VERSION} VERSION_GREATER 5.0 OR ${GCC_VERSION} VERSION_EQUAL 5.0)
+      set (CCOMMON_OPT  "${CCOMMON_OPT} -mcpu=power9 -mtune=power9 -mvsx -fno-fast-math")
+    else ()
+      set (CCOMMON_OPT  "${CCOMMON_OPT} -mcpu=power8 -mtune=power8 -mvsx -fno-fast-math")
+      message(WARNING "Compiler GCC.${GCC_VERSION} does not fully support Power9.")
+    endif ()
+  endif ()
+endif ()
+
+if (${CORE} STREQUAL POWER8)
+  if (NOT DYNAMIC_ARCH)
+    set (CCOMMON_OPT  "${CCOMMON_OPT} -mcpu=power8 -mtune=power8 -mvsx -fno-fast-math")
+  endif ()
+endif ()
 
 if (NOT DYNAMIC_ARCH)
 	if (HAVE_AVX2)
diff --git a/cmake/system.cmake b/cmake/system.cmake
index 410cf01e5..b4cce20f5 100644
--- a/cmake/system.cmake
+++ b/cmake/system.cmake
@@ -42,6 +42,9 @@ if (DEFINED BINARY AND DEFINED TARGET AND BINARY EQUAL 32)
   if (${TARGET} STREQUAL "ARMV8" OR ${TARGET} STREQUAL "CORTEXA57" OR ${TARGET} STREQUAL "CORTEXA53" OR ${TARGET} STREQUAL "CORTEXA55")
     set(TARGET "ARMV7")
   endif ()
+  if (${TARGET} STREQUAL "POWER8" OR ${TARGET} STREQUAL "POWER9" OR ${TARGET} STREQUAL "POWER10")
+    set(TARGET "POWER6")
+  endif ()
 endif ()
 
 
@@ -102,6 +105,12 @@ if (CMAKE_C_COMPILER STREQUAL loongcc)
   set(GETARCH_FLAGS	"${GETARCH_FLAGS} -static")
 endif ()
 
+if (POWER)
+  set (NO_WARMUP 1)
+  execute_process(COMMAND bash -c "as -v < /dev/null 2>&1 | grep GNU 2>&1 >/dev/null" RESULT_VARIABLE HAVE_GAS)
+  set(GETARCH_FLAGS	"${GETARCH_FLAGS} -DHAVE_GAS=${HAVE_GAS}")
+endif()
+
 #if don't use Fortran, it will only compile CBLAS.
 if (ONLY_CBLAS)
   set(NO_LAPACK 1)
@@ -222,6 +231,27 @@ if (DEFINED TARGET)
     if (DEFINED HAVE_SSE4_1)
     set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -msse4.1")
   endif()
+
+  if (${TARGET} STREQUAL POWER10)
+    execute_process(COMMAND ${CMAKE_C_COMPILER} -dumpversion OUTPUT_VARIABLE GCC_VERSION)
+    if (${GCC_VERSION} VERSION_GREATER 10.2 OR ${GCC_VERSION} VERSION_EQUAL 10.2)
+      set (KERNEL_DEFINITIONS  "${KERNEL_DEFINITIONS} -mcpu=power10 -mtune=power10 -mvsx -fno-fast-math")
+    else ()
+      message(FATAL_ERROR "Compiler GCC.${GCC_VERSION} does not support Power10.")
+    endif()
+  endif()
+  if (${TARGET} STREQUAL POWER9)
+    execute_process(COMMAND ${CMAKE_C_COMPILER} -dumpversion OUTPUT_VARIABLE GCC_VERSION)
+    if (${GCC_VERSION} VERSION_GREATER 5.0 OR ${GCC_VERSION} VERSION_EQUAL 5.0)
+      set (KERNEL_DEFINITIONS  "${KERNEL_DEFINITIONS} -mcpu=power9 -mtune=power9 -mvsx -fno-fast-math")
+    else ()
+      set (KERNEL_DEFINITIONS  "${KERNEL_DEFINITIONS} -mcpu=power8 -mtune=power8 -mvsx -fno-fast-math")
+      message(WARNING "Compiler GCC.${GCC_VERSION} does not support fully Power9.")
+    endif()
+  endif()
+  if (${TARGET} STREQUAL POWER8)
+    set (KERNEL_DEFINITIONS  "${KERNEL_DEFINITIONS} -mcpu=power8 -mtune=power8 -mvsx -fno-fast-math")
+  endif()
 endif()
 if (DEFINED BINARY)
   message(STATUS "Compiling a ${BINARY}-bit binary.")
@@ -279,7 +309,7 @@ if (NEED_PIC)
   endif()
 endif ()
 
-if (X86_64)
+if (X86_64 OR ${CORE} STREQUAL POWER10)
   set(SMALL_MATRIX_OPT TRUE)
 endif ()
 if (SMALL_MATRIX_OPT)
@@ -287,7 +317,7 @@ if (SMALL_MATRIX_OPT)
 endif ()
 
 if (DYNAMIC_ARCH)
-  if (X86 OR X86_64 OR ARM64 OR PPC)
+  if (X86 OR X86_64 OR ARM64 OR POWER)
     set(CCOMMON_OPT "${CCOMMON_OPT} -DDYNAMIC_ARCH")
     if (DYNAMIC_OLDER)
       set(CCOMMON_OPT "${CCOMMON_OPT} -DDYNAMIC_OLDER")
diff --git a/cmake/system_check.cmake b/cmake/system_check.cmake
index f71ec4555..86ce3dfb0 100644
--- a/cmake/system_check.cmake
+++ b/cmake/system_check.cmake
@@ -35,7 +35,7 @@ if(CMAKE_CL_64 OR MINGW64)
 elseif(MINGW OR (MSVC AND NOT CMAKE_CROSSCOMPILING))
   set(X86 1)
 elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "ppc.*|power.*|Power.*")
-  set(PPC 1)
+  set(POWER 1)
 elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "mips64.*")
   set(MIPS64 1)
 elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "loongarch64.*")
@@ -88,7 +88,7 @@ if (X86_64)
   set(ARCH "x86_64")
 elseif(X86)
   set(ARCH "x86")
-elseif(PPC)
+elseif(POWER)
   set(ARCH "power")
 elseif(MIPS32)
   set(ARCH "mips")
@@ -103,7 +103,7 @@ else()
 endif ()
 
 if (NOT BINARY)
-  if (X86_64 OR ARM64 OR PPC OR MIPS64 OR LOONGARCH64)
+  if (X86_64 OR ARM64 OR POWER OR MIPS64 OR LOONGARCH64)
     set(BINARY 64)
   else ()
     set(BINARY 32)
diff --git a/driver/others/CMakeLists.txt b/driver/others/CMakeLists.txt
index a07e00b3b..1a38740a3 100644
--- a/driver/others/CMakeLists.txt
+++ b/driver/others/CMakeLists.txt
@@ -49,6 +49,8 @@ GenerateNamedObjects("openblas_get_config.c;openblas_get_parallel.c" "" "" 0 ""
 if (DYNAMIC_ARCH)
   if (ARM64)
     list(APPEND COMMON_SOURCES dynamic_arm64.c)
+  elseif (POWER)
+    list(APPEND COMMON_SOURCES dynamic_power.c)
   else ()  
     list(APPEND COMMON_SOURCES dynamic.c)
   endif ()