Merge pull request #5029 from martin-frbg/issue5020

Add support for compiling with Intel oneAPI 2025.0 on MS Windows

CMakeLists.txt

@@ -4,6 +4,7 @@
 
 cmake_minimum_required(VERSION 3.16.0)
 
+set (CMAKE_ASM_SOURCE_FILE_EXTENSIONS "S")
 project(OpenBLAS C ASM)
 
 set(OpenBLAS_MAJOR_VERSION 0)

cmake/f_check.cmake

@@ -45,13 +45,15 @@ if (NOT ONLY_CBLAS)
 
   # TODO: detect whether underscore needed, set #defines and BU appropriately - use try_compile
   # TODO: set FEXTRALIB flags a la f_check?
-
+  if (NOT (${CMAKE_SYSTEM_NAME} MATCHES "Windows" AND x${CMAKE_Fortran_COMPILER_ID} MATCHES "IntelLLVM"))
   set(BU "_")
   file(APPEND ${TARGET_CONF_TEMP}
     "#define BUNDERSCORE _\n"
     "#define NEEDBUNDERSCORE 1\n"
     "#define NEED2UNDERSCORES 0\n")
-
+  else ()
+	  set (FCOMMON_OPT "${FCOMMON_OPT} /fp:precise /recursive /names:lowercase /assume:nounderscore")
+  endif()
 else ()
 
   #When we only build CBLAS, we set NOFORTRAN=2

cmake/prebuild.cmake

@@ -58,7 +58,7 @@ set(TARGET_CONF_TEMP "${PROJECT_BINARY_DIR}/${TARGET_CONF}.tmp")
 
 # c_check
 set(FU "")
-if (APPLE OR (MSVC AND NOT ${CMAKE_C_COMPILER_ID} MATCHES "Clang"))
+if (APPLE OR (MSVC AND NOT (${CMAKE_C_COMPILER_ID} MATCHES "Clang" OR ${CMAKE_C_COMPILER_ID} MATCHES "IntelLLVM")))
   set(FU "_")
 endif()
 if(MINGW AND NOT MINGW64)
@@ -1433,7 +1433,9 @@ else(NOT CMAKE_CROSSCOMPILING)
     message(STATUS "MSVC")
     set(GETARCH_FLAGS ${GETARCH_FLAGS} -DFORCE_GENERIC)
   else()
-    list(APPEND GETARCH_SRC ${PROJECT_SOURCE_DIR}/cpuid.S)
+    if ("${CMAKE_SYSTEM_NAME}" STREQUAL "Darwin")
+      list(APPEND GETARCH_SRC ${PROJECT_SOURCE_DIR}/cpuid.S)
+    endif()
     if (DEFINED TARGET_CORE)
     set(GETARCH_FLAGS ${GETARCH_FLAGS} -DFORCE_${TARGET_CORE})
   endif ()

cmake/system.cmake

@@ -672,6 +672,9 @@ endif ()
 if (${CMAKE_C_COMPILER} STREQUAL "LSB" OR ${CMAKE_SYSTEM_NAME} STREQUAL "Windows")
   set(LAPACK_CFLAGS "${LAPACK_CFLAGS} -DLAPACK_COMPLEX_STRUCTURE")
 endif ()
+if (${CMAKE_C_COMPILER_ID} MATCHES "IntelLLVM" AND ${CMAKE_SYSTEM_NAME} STREQUAL "Windows")
+	set(LAPACK_CFLAGS "${LAPACK_CFLAGS} -DNOCHANGE")
+endif ()
 
 if ("${CMAKE_BUILD_TYPE}" STREQUAL "Release")
 if ("${F_COMPILER}" STREQUAL "FLANG")

ctest/cblas_test.h

@@ -10,6 +10,10 @@
 #define int long
 #endif
 
+#if defined(_MSC_VER) && defined(__INTEL_CLANG_COMPILER)
+//#define LAPACK_COMPLEX_STRUCTURE
+#define NOCHANGE
+#endif
 /* e.g. mingw64/x86_64-w64-mingw32/include/winerror.h */
 #ifdef FAILED
 #undef FAILED

lapack-netlib/LAPACKE/include/lapacke_config.h

@@ -67,8 +67,14 @@ extern "C" {
 #define lapack_logical    lapack_int
 #endif
 
+#if defined(_MSC_VER) && defined(__INTEL_CLANG_COMPILER)
+#define LAPACK_COMPLEX_STRUCTURE
+#define LAPACK_GLOBAL(lcname,UCNAME)  lcname
+#define NOCHANGE
+#endif
+
 #ifndef LAPACK_COMPLEX_CUSTOM
-#if defined(_MSC_VER)
+#if defined(_MSC_VER) && !defined(__INTEL_CLANG_COMPILER)
     #define _CRT_USE_C_COMPLEX_H
     #include <complex.h>
     #define LAPACK_COMPLEX_CUSTOM

utest/openblas_utest.h

@@ -36,7 +36,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include <stdlib.h>
 #include "ctest.h"
-
+#if defined(_MSC_VER) && defined(__INTEL_CLANG_COMPILER)
+//#define LAPACK_COMPLEX_STRUCTURE
+#define NOCHANGE
+#endif
 #include <common.h>
 #include <math.h>
 

utest/test_extensions/common.c

@@ -206,7 +206,7 @@ void ztranspose(blasint rows, blasint cols, double *alpha, double *a_src, int ld
  * param lda_dst - leading dimension of output matrix A
  * param conj specifies conjugation
  */
-void scopy(blasint rows, blasint cols, float alpha, float *a_src, int lda_src, 
+void my_scopy(blasint rows, blasint cols, float alpha, float *a_src, int lda_src, 
            float *a_dst, blasint lda_dst)
 {
     blasint i, j;
@@ -217,7 +217,7 @@ void scopy(blasint rows, blasint cols, float alpha, float *a_src, int lda_src,
     }
 }
 
-void dcopy(blasint rows, blasint cols, double alpha, double *a_src, int lda_src, 
+void my_dcopy(blasint rows, blasint cols, double alpha, double *a_src, int lda_src, 
            double *a_dst, blasint lda_dst)
 {
     blasint i, j;
@@ -228,7 +228,7 @@ void dcopy(blasint rows, blasint cols, double alpha, double *a_src, int lda_src,
     }
 }
 
-void ccopy(blasint rows, blasint cols, float *alpha, float *a_src, int lda_src, 
+void my_ccopy(blasint rows, blasint cols, float *alpha, float *a_src, int lda_src, 
            float *a_dst, blasint lda_dst, int conj)
 {
     blasint i, j;
@@ -243,7 +243,7 @@ void ccopy(blasint rows, blasint cols, float *alpha, float *a_src, int lda_src,
     }
 }
 
-void zcopy(blasint rows, blasint cols, double *alpha, double *a_src, int lda_src, 
+void my_zcopy(blasint rows, blasint cols, double *alpha, double *a_src, int lda_src, 
            double *a_dst, blasint lda_dst, int conj)
 {
     blasint i, j;

utest/test_extensions/common.h

@@ -65,12 +65,12 @@ extern void ctranspose(blasint rows, blasint cols, float *alpha, float *a_src, i
 extern void ztranspose(blasint rows, blasint cols, double *alpha, double *a_src, int lda_src, 
                 double *a_dst, blasint lda_dst, int conj);
 
-extern void scopy(blasint rows, blasint cols, float alpha, float *a_src, int lda_src, 
+extern void my_scopy(blasint rows, blasint cols, float alpha, float *a_src, int lda_src, 
            float *a_dst, blasint lda_dst);
-extern void dcopy(blasint rows, blasint cols, double alpha, double *a_src, int lda_src, 
+extern void my_dcopy(blasint rows, blasint cols, double alpha, double *a_src, int lda_src, 
            double *a_dst, blasint lda_dst);
-extern void ccopy(blasint rows, blasint cols, float *alpha, float *a_src, int lda_src, 
+extern void my_ccopy(blasint rows, blasint cols, float *alpha, float *a_src, int lda_src, 
            float *a_dst, blasint lda_dst, int conj);
-extern void zcopy(blasint rows, blasint cols, double *alpha, double *a_src, int lda_src, 
+extern void my_zcopy(blasint rows, blasint cols, double *alpha, double *a_src, int lda_src, 
            double *a_dst, blasint lda_dst, int conj);                
-#endif
+#endif

utest/test_extensions/test_cimatcopy.c

@@ -91,7 +91,7 @@ static float check_cimatcopy(char api, char order, char trans, blasint rows, bla
         ctranspose(m, n, alpha, data_cimatcopy.a_test, lda_src, data_cimatcopy.a_verify, lda_dst, conj);
     } 
     else {
-        ccopy(m, n, alpha, data_cimatcopy.a_test, lda_src, data_cimatcopy.a_verify, lda_dst, conj);
+        my_ccopy(m, n, alpha, data_cimatcopy.a_test, lda_src, data_cimatcopy.a_verify, lda_dst, conj);
     }
 
     if (api == 'F') {

utest/test_extensions/test_comatcopy.c

@@ -92,7 +92,7 @@ static float check_comatcopy(char api, char order, char trans, blasint rows, bla
         ctranspose(m, n, alpha, data_comatcopy.a_test, lda, data_comatcopy.b_verify, ldb, conj);
     } 
     else {
-        ccopy(m, n, alpha, data_comatcopy.a_test, lda, data_comatcopy.b_verify, ldb, conj);
+        my_ccopy(m, n, alpha, data_comatcopy.a_test, lda, data_comatcopy.b_verify, ldb, conj);
     }
 
     if (api == 'F') {

utest/test_extensions/test_dimatcopy.c

@@ -86,7 +86,7 @@ static double check_dimatcopy(char api, char order, char trans, blasint rows, bl
         dtranspose(m, n, alpha, data_dimatcopy.a_test, lda_src, data_dimatcopy.a_verify, lda_dst);
     } 
     else {
-        dcopy(m, n, alpha, data_dimatcopy.a_test, lda_src, data_dimatcopy.a_verify, lda_dst);
+        my_dcopy(m, n, alpha, data_dimatcopy.a_test, lda_src, data_dimatcopy.a_verify, lda_dst);
     }
 
     if (api == 'F') {

utest/test_extensions/test_domatcopy.c

@@ -87,7 +87,7 @@ static double check_domatcopy(char api, char order, char trans, blasint rows, bl
         dtranspose(m, n, alpha, data_domatcopy.a_test, lda, data_domatcopy.b_verify, ldb);
     } 
     else {
-        dcopy(m, n, alpha, data_domatcopy.a_test, lda, data_domatcopy.b_verify, ldb);
+        my_dcopy(m, n, alpha, data_domatcopy.a_test, lda, data_domatcopy.b_verify, ldb);
     }
 
     if (api == 'F') {

utest/test_extensions/test_simatcopy.c

@@ -86,7 +86,7 @@ static float check_simatcopy(char api, char order, char trans, blasint rows, bla
         stranspose(m, n, alpha, data_simatcopy.a_test, lda_src, data_simatcopy.a_verify, lda_dst);
     } 
     else {
-        scopy(m, n, alpha, data_simatcopy.a_test, lda_src, data_simatcopy.a_verify, lda_dst);
+        my_scopy(m, n, alpha, data_simatcopy.a_test, lda_src, data_simatcopy.a_verify, lda_dst);
     }
 
     if (api == 'F') {

utest/test_extensions/test_somatcopy.c

@@ -87,7 +87,7 @@ static float check_somatcopy(char api, char order, char trans, blasint rows, bla
         stranspose(m, n, alpha, data_somatcopy.a_test, lda, data_somatcopy.b_verify, ldb);
     } 
     else {
-        scopy(m, n, alpha, data_somatcopy.a_test, lda, data_somatcopy.b_verify, ldb);
+        my_scopy(m, n, alpha, data_somatcopy.a_test, lda, data_somatcopy.b_verify, ldb);
     }
 
     if (api == 'F') {

utest/test_extensions/test_zimatcopy.c

@@ -91,7 +91,7 @@ static double check_zimatcopy(char api, char order, char trans, blasint rows, bl
         ztranspose(m, n, alpha, data_zimatcopy.a_test, lda_src, data_zimatcopy.a_verify, lda_dst, conj);
     } 
     else {
-        zcopy(m, n, alpha, data_zimatcopy.a_test, lda_src, data_zimatcopy.a_verify, lda_dst, conj);
+        my_zcopy(m, n, alpha, data_zimatcopy.a_test, lda_src, data_zimatcopy.a_verify, lda_dst, conj);
     }
 
     if (api == 'F') {

utest/test_extensions/test_zomatcopy.c

@@ -92,7 +92,7 @@ static double check_zomatcopy(char api, char order, char trans, blasint rows, bl
         ztranspose(m, n, alpha, data_zomatcopy.a_test, lda, data_zomatcopy.b_verify, ldb, conj);
     } 
     else {
-        zcopy(m, n, alpha, data_zomatcopy.a_test, lda, data_zomatcopy.b_verify, ldb, conj);
+        my_zcopy(m, n, alpha, data_zomatcopy.a_test, lda, data_zomatcopy.b_verify, ldb, conj);
     }
 
     if (api == 'F') {