Merge pull request #3320 from martin-frbg/issue3318

Empirical workaround for numpy SVD NaN problem from issue 3318

Author: martin-frbg

cmake/cc.cmake

@@ -124,9 +124,9 @@ if (NOT DYNAMIC_ARCH)
 	if (HAVE_AVX)
         set (CCOMMON_OPT  "${CCOMMON_OPT} -mavx")
 	endif ()
-	if (HAVE_FMA3)
-	set (CCOMMON_OPT  "${CCOMMON_OPT} -mfma")
-	endif ()
+	#	if (HAVE_FMA3)
+	#set (CCOMMON_OPT  "${CCOMMON_OPT} -mfma")
+	#endif ()
 	if (HAVE_SSE)
 	set (CCOMMON_OPT  "${CCOMMON_OPT} -msse")
 	endif ()

cmake/system.cmake

@@ -186,11 +186,11 @@ if (DEFINED TARGET)
       	  set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -mavx2")
 	endif()
   endif()
-  if (DEFINED HAVE_FMA3)
-	if (NOT NO_AVX2)
-    set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -mfma")
-	endif()
-  endif()
+  #  if (DEFINED HAVE_FMA3)
+  #	if (NOT NO_AVX2)
+  #  set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -mfma")
+  #	endif()
+  #  endif()
     if (DEFINED HAVE_SSE)
     set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -msse")
   endif()

cmake/utils.cmake

@@ -311,7 +311,15 @@ function(GenerateNamedObjects sources_in)
       configure_file(${new_source_file}.tmp ${new_source_file} COPYONLY)
       file(REMOVE ${new_source_file}.tmp)
       list(APPEND SRC_LIST_OUT ${new_source_file})
-
+      message (STATUS ${new_source_file})
+      if (DEFINED HAVE_FMA3)
+        if ( ${new_source_file} MATCHES "(s|d?)rot_k.*c")
+		set_source_files_properties(${new_source_file} PROPERTIES COMPILE_OPTIONS "-mfma")
+        endif ()
+        if ( ${new_source_file} MATCHES "dgemv_t_k.*c")
+		set_source_files_properties(${new_source_file} PROPERTIES COMPILE_OPTIONS "-mfma")
+        endif ()
+      endif ()
     endforeach ()
   endforeach ()
 

kernel/Makefile.L2

@@ -1,3 +1,10 @@
+FMAFLAG=
+ifndef OLDGCC
+ifdef HAVE_FMA3
+FMAFLAG = -mfma
+endif
+endif
+
 ### GEMV ###
 
 ifndef SGEMVNKERNEL
@@ -263,7 +270,7 @@ $(KDIR)dgemv_n$(TSUFFIX).$(SUFFIX)  $(KDIR)dgemv_n$(TSUFFIX).$(PSUFFIX)  : $(KER
 	$(CC) -c $(CFLAGS) -DDOUBLE -UCOMPLEX  -UTRANS $< -o $@
 
 $(KDIR)dgemv_t$(TSUFFIX).$(SUFFIX)  $(KDIR)dgemv_t$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(DGEMVTKERNEL) $(TOPDIR)/common.h $(GEMVDEP)
-	$(CC) -c $(CFLAGS) -DDOUBLE -UCOMPLEX  -DTRANS $< -o $@
+	$(CC) -c $(CFLAGS) $(FMAFLAG) -DDOUBLE -UCOMPLEX  -DTRANS $< -o $@
 endif
 
 $(KDIR)qgemv_n$(TSUFFIX).$(SUFFIX)  $(KDIR)qgemv_n$(TSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(QGEMVNKERNEL)