Merge pull request #20 from xianyi/develop

Rebase

kernel/Makefile

@@ -5,6 +5,11 @@ endif
 TOPDIR	= ..
 include $(TOPDIR)/Makefile.system
 
+
+ifeq ($(C_COMPILER), GCC)
+GCCVERSIONGTEQ9 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 9)
+endif
+
 AVX2OPT = 
 ifeq ($(C_COMPILER), GCC)
 # AVX2 support was added in 4.7.0

kernel/power/KERNEL.POWER9

@@ -89,14 +89,30 @@ ZTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
 #SMINKERNEL   = ../arm/min.c
 #DMINKERNEL   = ../arm/min.c
 #
+ifneq ($(GCCVERSIONGTEQ9),1)
 ISAMAXKERNEL = isamax_power9.S
+else
+ISAMAXKERNEL = isamax.c
+endif
 IDAMAXKERNEL = idamax.c
+ifneq ($(GCCVERSIONGTEQ9),1)
 ICAMAXKERNEL = icamax_power9.S
+else
+ICAMAXKERNEL = icamax.c
+endif
 IZAMAXKERNEL = izamax.c
 #
+ifneq ($(GCCVERSIONGTEQ9),1)
 ISAMINKERNEL = isamin_power9.S
+else
+ISAMINKERNEL = isamin.c
+endif
 IDAMINKERNEL = idamin.c
+ifneq ($(GCCVERSIONGTEQ9),1)
 ICAMINKERNEL = icamin_power9.S
+else
+ICAMINKERNEL = icamin.c
+endif
 IZAMINKERNEL = izamin.c
 #
 #ISMAXKERNEL  = ../arm/imax.c
@@ -112,7 +128,11 @@ ZASUMKERNEL  = zasum.c
 #
 SAXPYKERNEL  = saxpy.c
 DAXPYKERNEL  = daxpy.c
+ifneq ($(GCCVERSIONGTEQ9),1)
 CAXPYKERNEL  = caxpy_power9.S
+else
+CAXPYKERNEL  = caxpy.c
+endif
 ZAXPYKERNEL  = zaxpy.c
 #
 SCOPYKERNEL  = scopy.c
@@ -123,7 +143,11 @@ ZCOPYKERNEL  = zcopy.c
 SDOTKERNEL   =  sdot.c
 DDOTKERNEL   =  ddot.c
 DSDOTKERNEL  =  sdot.c
+ifneq ($(GCCVERSIONGTEQ9),1)
 CDOTKERNEL   =  cdot_power9.S
+else
+CDOTKERNEL   =  cdot.c
+endif
 ZDOTKERNEL   =  zdot.c
 #
 SNRM2KERNEL  = ../arm/nrm2.c

kernel/power/caxpy_power8.S

@@ -16,7 +16,11 @@
 0:	addis 2,12,.TOC.-.LCF0@ha
 	addi 2,2,.TOC.-.LCF0@l
 #if _CALL_ELF ==2
+#ifdef CONJ
+	.localentry	caxpyc_k,.-caxpyc_k
+#else
 	.localentry	caxpy_k,.-caxpy_k
+#endif
 #endif
 	mr. 7,3
 	ble 0,.L33
@@ -517,7 +521,11 @@
 	.long 0
 	.byte 0,0,0,0,0,4,0,0
 #if _CALL_ELF ==2
+#ifdef CONJ
+	.size	caxpyc_k,.-caxpyc_k
+#else
 	.size	caxpy_k,.-caxpy_k
+#endif
 #endif
 	.section	.rodata
 	.align 4

kernel/power/caxpy_power9.S

@@ -17,7 +17,11 @@ caxpy_k:
 .LCF0:
 0:	addis 2,12,.TOC.-.LCF0@ha
 	addi 2,2,.TOC.-.LCF0@l
+#ifdef CONJ
+	.localentry	caxpyc_k,.-caxpyc_k
+#else
 	.localentry	caxpy_k,.-caxpy_k
+#endif
 	mr. 7,3
 	ble 0,.L33
 	cmpdi 7,9,1
@@ -474,7 +478,11 @@ caxpy_k:
 	b .L13
 	.long 0
 	.byte 0,0,0,0,0,1,0,0
+#ifdef CONJ
+	.size	caxpyc_k,.-caxpyc_k
+#else
 	.size	caxpy_k,.-caxpy_k
+#endif
 	.section	.rodata
 	.align 4
 	.set	.LANCHOR0,. + 0

kernel/power/cdot_power9.S

@@ -13,10 +13,7 @@
 	
 cdot_k:
 .LCF0:
-0:	addis 2,12,.TOC.-.LCF0@ha
-	addi 2,2,.TOC.-.LCF0@l
-	.localentry	cdot_k,.-cdot_k
-	mr. 9,3
+0:	mr. 9,3
 	ble 0,.L10
 	cmpdi 7,5,1
 	beq 7,.L18

kernel/x86_64/dsymv_L_microk_skylakex-2.c

@@ -33,6 +33,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #define HAVE_KERNEL_4x4 1
 
+#if defined(__clang_patchlevel__) && __clang_major__ == 9 && __clang_minor__ == 0 && __clang_patchlevel__ == 0
+#pragma clang optimize off
+#endif
+
 static void dsymv_kernel_4x4(BLASLONG from, BLASLONG to, FLOAT **a, FLOAT *x, FLOAT *y, FLOAT *temp1, FLOAT *temp2)
 {
 
@@ -156,6 +160,11 @@ static void dsymv_kernel_4x4(BLASLONG from, BLASLONG to, FLOAT **a, FLOAT *x, FL
 	temp2[2] += half_accum2[0];
 	temp2[3] += half_accum3[0];
 }
+
+#if defined(__clang_patchlevel__) && __clang_major__ == 9 && __clang_minor__ == 0 && __clang_patchlevel__ == 0
+#pragma clang optimize on
+#endif
+
 #else
 #include "dsymv_L_microk_haswell-2.c"
 #endif