Merge pull request #64 from xianyi/develop

rebase

Author: martin-frbg

Makefile.system

@@ -797,11 +797,11 @@ endif
 ifeq ($(USE_OPENMP), 1)
 FCOMMON_OPT += -fopenmp
 endif
-ifeq ($(OSNAME), Linux)
-ifeq ($(ARCH), x86_64)
-FLANG_VENDOR := $(shell expr `$(FC) --version|cut -f 1 -d "."|head -1`)
-endif
-endif
+#ifeq ($(OSNAME), Linux)
+#ifeq ($(ARCH), x86_64)
+#FLANG_VENDOR := $(shell expr `$(FC) --version|cut -f 1 -d "."|head -1`)
+#endif
+#endif
 endif
 
 ifeq ($(F_COMPILER), G77)
@@ -1276,7 +1276,8 @@ endif
 
 override CFLAGS     += $(COMMON_OPT) $(CCOMMON_OPT) -I$(TOPDIR)
 override PFLAGS     += $(COMMON_OPT) $(CCOMMON_OPT) -I$(TOPDIR) -DPROFILE $(COMMON_PROF)
-ifeq ($(FLANG_VENDOR),AOCC)
+#ifeq ($(FLANG_VENDOR),AOCC)
+ifeq ($(F_COMPILER),FLANG)
 override FFLAGS     += $(filter-out -O2 -O3,$(COMMON_OPT)) -O1 $(FCOMMON_OPT)
 else
 override FFLAGS     += $(COMMON_OPT) $(FCOMMON_OPT)

cmake/fc.cmake

@@ -16,6 +16,7 @@ if (${F_COMPILER} STREQUAL "FLANG")
   if (USE_OPENMP)
     set(FCOMMON_OPT "${FCOMMON_OPT} -fopenmp")
   endif ()
+  set(FCOMMON_OPT "${FCOMMON_OPT} -frecursive")
 endif ()
 
 if (${F_COMPILER} STREQUAL "G77")

cmake/system.cmake

@@ -417,6 +417,15 @@ if (${CMAKE_C_COMPILER} STREQUAL "LSB" OR ${CMAKE_SYSTEM_NAME} STREQUAL "Windows
   set(LAPACK_CFLAGS "${LAPACK_CFLAGS} -DLAPACK_COMPLEX_STRUCTURE")
 endif ()
 
+if ("${CMAKE_BUILD_TYPE}" STREQUAL "Release")
+if ("${F_COMPILER}" STREQUAL "FLANG")
+  set(FILTER_FLAGS "-O2;-O3")
+  foreach (FILTER_FLAG ${FILTER_FLAGS})
+    string(REPLACE ${FILTER_FLAG} "-O1" CMAKE_Fortran_FLAGS_RELEASE ${CMAKE_Fortran_FLAGS_RELEASE})
+  endforeach ()
+endif ()
+endif ()
+
 if (NOT DEFINED SUFFIX)
   set(SUFFIX o)
 endif ()

exports/Makefile

@@ -155,8 +155,12 @@ ifeq ($(F_COMPILER), INTEL)
 	-Wl,--whole-archive $< -Wl,--no-whole-archive \
 	-Wl,-soname,$(INTERNALNAME) $(EXTRALIB)
 	$(CC) $(CFLAGS) $(LDFLAGS) -w -o linktest linktest.c ../$(LIBSONAME) $(FEXTRALIB) && echo OK.
+else ifeq ($(F_COMPILER), FLANG)
+	$(FC) $(FFLAGS) $(LDFLAGS) -shared -o ../$(LIBSONAME) \
+	-Wl,--whole-archive $< -Wl,--no-whole-archive \
+	-Wl,-soname,$(INTERNALNAME) $(EXTRALIB)
+	$(CC) $(CFLAGS) $(LDFLAGS) -w -o linktest linktest.c ../$(LIBSONAME) $(FEXTRALIB) && echo OK.
 else
-
 ifneq ($(C_COMPILER), LSB)
 	$(CC) $(CFLAGS) $(LDFLAGS) -shared -o ../$(LIBSONAME) \
 	-Wl,--whole-archive $< -Wl,--no-whole-archive \

kernel/arm64/sgemm_kernel_8x8_cortexa53.S

@@ -681,12 +681,12 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 .macro INIT8x4
 	fmov		s16, wzr
 	fmov		s17, wzr
+	fmov		s18, wzr
+	fmov		s19, s16
 	fmov		s20, wzr
 	fmov		s21, s16
-	fmov		s24, wzr
-	fmov		s25, s16
-	fmov		s28, wzr
-	fmov		s29, s16
+	fmov		s22, wzr
+	fmov		s23, s16
 .endm
 
 .macro KERNEL8x4_I
@@ -765,14 +765,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	fmla	v21.4s, v3.4s, v6.s[2]
 	fmla	v22.4s, v2.4s, v6.s[3]
 	fmla	v23.4s, v3.4s, v6.s[3]
-	fmla	v24.4s, v2.4s, v7.s[0]
-	fmla	v25.4s, v3.4s, v7.s[0]
-	fmla	v26.4s, v2.4s, v7.s[1]
-	fmla	v27.4s, v3.4s, v7.s[1]
-	fmla	v28.4s, v2.4s, v7.s[2]
-	fmla	v29.4s, v3.4s, v7.s[2]
-	fmla	v30.4s, v2.4s, v7.s[3]
-	fmla	v31.4s, v3.4s, v7.s[3]
 .endm
 
 .macro KERNEL8x4_SUB