Merge pull request #72 from xianyi/develop

rebase

Makefile.system

@@ -617,6 +617,7 @@ DYNAMIC_CORE += POWER8
 ifneq ($(C_COMPILER), GCC)
 DYNAMIC_CORE += POWER9
 DYNAMIC_CORE += POWER10
+override LDFLAGS += -Wl,-no-power10-stubs
 endif
 ifeq ($(C_COMPILER), GCC)
 ifeq ($(GCCVERSIONGT5), 1)
@@ -626,9 +627,11 @@ $(info, OpenBLAS: Your gcc version is too old to build the POWER9 kernels.)
 endif
 ifeq ($(GCCVERSIONGTEQ11), 1)
 DYNAMIC_CORE += POWER10
-else ifeq ($(GCCVERSIONEQ10), 1)
+override LDFLAGS += -Wl,-no-power10-stubs
+else ifeq ($(GCCVERSIONGTEQ10), 1)
 ifeq ($(GCCMINORVERSIONGTEQ2), 1)
 DYNAMIC_CORE += POWER10
+override LDFLAGS += -Wl,-no-power10-stubs
 endif
 else
 $(info, OpenBLAS: Your gcc version is too old to build the POWER10 kernels.)

README.md

@@ -58,6 +58,10 @@ Examples:
   ```sh
   make BINARY=64 CC=mips64el-unknown-linux-gnu-gcc FC=mips64el-unknown-linux-gnu-gfortran HOSTCC=gcc TARGET=LOONGSON3A
   ```
+  or same with the newer mips-crosscompiler put out by Loongson that defaults to the 32bit ABI:
+  ```sh
+  make HOSTCC=gcc CC='/opt/mips-loongson-gcc7.3-linux-gnu/2019.06-29/bin/mips-linux-gnu-gcc -mabi=64' FC='/opt/mips-loongson-gcc7.3-linux-gnu/2019.06-29/bin/mips-linux-gnu-gfortran -mabi=64' TARGET=LOONGSON3A
+  ```
 
 * On an x86 box, compile this library for a loongson3a CPU with loongcc (based on Open64) compiler:
   ```sh

c_check

@@ -249,6 +249,28 @@ if (($architecture eq "x86") || ($architecture eq "x86_64")) {
     }
 }
 
+$c11_atomics = 0;
+if ($data =~ /HAVE_C11/) {
+    eval "use File::Temp qw(tempfile)";
+    if ($@){ 
+       warn "could not load PERL module File::Temp, so could not check compiler compatibility with C11";
+       $c11_atomics = 0;
+    } else {
+       ($fh,$tmpf) = tempfile( SUFFIX => '.c' , UNLINK => 1 );
+       print $tmpf "#include <stdatomic.h>\nint main(void){}\n";
+       $args = " -c -o $tmpf.o $tmpf";
+       my @cmd = ("$compiler_name $flags $args >/dev/null 2>/dev/null");
+       system(@cmd) == 0;
+       if ($? != 0) {
+           $c11_atomics = 0;
+       } else {
+           $c11_atomics = 1;
+       }
+       unlink("$tmpf.o");
+    }
+}
+
+
 $data = `$compiler_name $flags -S ctest1.c && grep globl ctest1.s | head -n 1 && rm -f ctest1.s`;
 
 $data =~ /globl\s([_\.]*)(.*)/;
@@ -352,6 +374,8 @@ print CONFFILE "#define __32BIT__\t1\n"  if $binformat eq bin32;
 print CONFFILE "#define __64BIT__\t1\n"  if $binformat eq bin64;
 print CONFFILE "#define FUNDERSCORE\t$need_fu\n" if $need_fu ne "";
 print CONFFILE "#define HAVE_MSA\t1\n"  if $have_msa eq 1;
+print CONFFILE "#define HAVE_C11\t1\n" if $c11_atomics eq 1;
+
 
 if ($os eq "LINUX") {
 

cmake/f_check.cmake

@@ -21,7 +21,15 @@
 # NEED2UNDERSCORES
 
 if (NOT NO_LAPACK)
-  enable_language(Fortran)
+  include(CheckLanguage)
+  check_language(Fortran)
+  if(CMAKE_Fortran_COMPILER)
+    enable_language(Fortran)
+  else()
+  message(STATUS "No Fortran compiler found, can build only BLAS but not LAPACK")
+  set (NOFORTRAN 1)
+  set (NO_LAPACK 1)
+  endif()
 else()
   include(CMakeForceCompiler)
   CMAKE_FORCE_Fortran_COMPILER(gfortran GNU)

common.h

@@ -681,7 +681,7 @@ __declspec(dllimport) int __cdecl omp_in_parallel(void);
 __declspec(dllimport) int __cdecl omp_get_num_procs(void);
 #endif
 
-#if (__STDC_VERSION__ >= 201112L)
+#ifdef HAVE_C11
 #if defined(C_GCC) && ( __GNUC__ < 7) 
 // workaround for GCC bug 65467
 #ifndef _Atomic

ctest.c

@@ -153,3 +153,6 @@ ARCH_ARM
 ARCH_ARM64
 #endif
 
+#if (defined(__STDC_VERSION__) && __STDC_VERSION__ >= 201112L)
+HAVE_C11
+#endif

driver/level3/level3_gemm3m_thread.c

@@ -91,7 +91,7 @@
 #endif
 
 typedef struct {
-#if __STDC_VERSION__ >= 201112L
+#ifdef HAVE_C11
   _Atomic
 #else
   volatile

driver/level3/level3_syrk_threaded.c

@@ -67,7 +67,7 @@
 #endif
 
 typedef struct {
-#if __STDC_VERSION__ >= 201112L
+#ifdef HAVE_C11
 _Atomic
 #else 
   volatile

driver/others/blas_server.c

@@ -141,7 +141,7 @@ typedef struct {
 
 } thread_status_t;
 
-#if (__STDC_VERSION__ >= 201112L)
+#ifdef HAVE_C11
 #define	atomic_load_queue(p)		__atomic_load_n(p, __ATOMIC_RELAXED)
 #define	atomic_store_queue(p, v)	__atomic_store_n(p, v, __ATOMIC_RELAXED)
 #else

driver/others/blas_server_omp.c

@@ -55,7 +55,7 @@
 int blas_server_avail = 0;
 
 static void * blas_thread_buffer[MAX_PARALLEL_NUMBER][MAX_CPU_NUMBER];
-#if __STDC_VERSION__ >= 201112L
+#ifdef HAVE_C11
 static atomic_bool blas_buffer_inuse[MAX_PARALLEL_NUMBER];
 #else
 static _Bool blas_buffer_inuse[MAX_PARALLEL_NUMBER];
@@ -320,7 +320,7 @@ int exec_blas(BLASLONG num, blas_queue_t *queue){
 
   while(true) {
     for(i=0; i < MAX_PARALLEL_NUMBER; i++) {
-#if __STDC_VERSION__ >= 201112L
+#ifdef HAVE_C11
       _Bool inuse = false;
       if(atomic_compare_exchange_weak(&blas_buffer_inuse[i], &inuse, true)) {
 #else
@@ -345,7 +345,7 @@ int exec_blas(BLASLONG num, blas_queue_t *queue){
     exec_threads(&queue[i], buf_index);
   }
 
-#if __STDC_VERSION__ >= 201112L
+#ifdef HAVE_C11
   atomic_store(&blas_buffer_inuse[buf_index], false);
 #else
   blas_buffer_inuse[buf_index] = false;

driver/others/memory.c

@@ -1095,7 +1095,7 @@ static BLASULONG base_address      = 0UL;
 static BLASULONG base_address      = BASE_ADDRESS;
 #endif
 
-#if __STDC_VERSION__ >= 201112L
+#ifdef HAVE_C11
 static _Atomic int memory_initialized = 0;
 #else
 static volatile int memory_initialized = 0;

lapack/getrf/getrf_parallel.c

@@ -68,7 +68,7 @@ double sqrt(double);
 #define GETRF_FACTOR 1.00
 
 
-#if (__STDC_VERSION__ >= 201112L)
+#ifdef HAVE_C11
 #define	atomic_load_long(p)		__atomic_load_n(p, __ATOMIC_RELAXED)
 #define	atomic_store_long(p, v)		__atomic_store_n(p, v, __ATOMIC_RELAXED)
 #else

lapack/getrf/potrf_parallel.c

@@ -101,7 +101,7 @@ static FLOAT dm1 = -1.;
 #endif
 
 typedef struct {
-#if __STDC_VERSION__ >= 201112L
+#ifdef HAVE_C11
   _Atomic 
 #else
   volatile