Merge pull request #11 from xianyi/develop

sync with upstream

.travis.yml

@@ -162,16 +162,16 @@ matrix:
       before_script:
         - COMMON_FLAGS="DYNAMIC_ARCH=1 TARGET=NEHALEM NUM_THREADS=32"
         - brew update
-        - brew install gcc # for gfortran
+        - brew install gcc@8 # for gfortran
       script:
         - travis_wait 45 make QUIET_MAKE=1 $COMMON_FLAGS $BTYPE
       env:
-        - BTYPE="BINARY=64 INTERFACE64=1"
+        - BTYPE="BINARY=64 INTERFACE64=1 FC=gfortran-8"
 
     - <<: *test-macos
       osx_image: xcode8.3
       env:
-        - BTYPE="BINARY=32"
+        - BTYPE="BINARY=32 FC=gfortran-8"
 
 # whitelist
 branches:

common_arm64.h

@@ -103,12 +103,14 @@ static inline int blas_quickdivide(blasint x, blasint y){
 
 #if defined(ASSEMBLER) && !defined(NEEDPARAM)
 
-#define PROLOGUE \
-	.text ;\
-	.align	4 ;\
-	.global	REALNAME ;\
-	.type	REALNAME, %function ;\
+.macro PROLOGUE 
+	.text ;
+	.p2align 2 ;
+	.global	REALNAME ;
+	.type	REALNAME, %function ;
 REALNAME:
+.endm
+
 
 #define EPILOGUE
 

kernel/arm64/nrm2.S

@@ -54,37 +54,37 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #if !defined(DOUBLE)
 	ldr	s4, [X], #4
 	fcmp	s4, REGZERO
-	beq	KERNEL_F1_NEXT_\@
+	beq	2f     /* KERNEL_F1_NEXT_\@ */
 	fabs	s4, s4
 	fcmp	SCALE, s4
-	bge	KERNEL_F1_SCALE_GE_X_\@
+	bge	1f     /* KERNEL_F1_SCALE_GE_X_\@ */
 	fdiv	s2, SCALE, s4
 	fmul	s2, s2, s2
 	fmul	s3, SSQ, s2
 	fadd	SSQ, REGONE, s3
 	fmov	SCALE, s4
-	b	KERNEL_F1_NEXT_\@
-KERNEL_F1_SCALE_GE_X_\@:
+	b	2f     /* KERNEL_F1_NEXT_\@ */
+1:               /* KERNEL_F1_SCALE_GE_X_\@: */
 	fdiv	s2, s4, SCALE
 	fmla	SSQ, s2, v2.s[0]
 #else
 	ldr	d4, [X], #8
 	fcmp	d4, REGZERO
-	beq	KERNEL_F1_NEXT_\@
+	beq	2f     /* KERNEL_F1_NEXT_\@ */
 	fabs	d4, d4
 	fcmp	SCALE, d4
-	bge	KERNEL_F1_SCALE_GE_X_\@
+	bge	1f     /* KERNEL_F1_SCALE_GE_X_\@ */
 	fdiv	d2, SCALE, d4
 	fmul	d2, d2, d2
 	fmul	d3, SSQ, d2
 	fadd	SSQ, REGONE, d3
 	fmov	SCALE, d4
-	b	KERNEL_F1_NEXT_\@
-KERNEL_F1_SCALE_GE_X_\@:
+	b	2f     /* KERNEL_F1_NEXT_\@ */
+1:                     /* KERNEL_F1_SCALE_GE_X_\@: */
 	fdiv	d2, d4, SCALE
 	fmla	SSQ, d2, v2.d[0]
 #endif
-KERNEL_F1_NEXT_\@:
+2:                     /* KERNEL_F1_NEXT_\@: */
 .endm
 
 .macro KERNEL_S1

kernel/arm64/znrm2.S

@@ -54,69 +54,69 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #if !defined(DOUBLE)
 	ldr	s4, [X], #4
 	fcmp	s4, REGZERO
-	beq	KERNEL_F1_NEXT_\@
+	beq	2f    /* KERNEL_F1_NEXT_\@ */
 	fabs	s4, s4
 	fcmp	SCALE, s4
-	bge	KERNEL_F1_SCALE_GE_XR_\@
+	bge	1f    /* KERNEL_F1_SCALE_GE_XR_\@ */
 	fdiv	s2, SCALE, s4
 	fmul	s2, s2, s2
 	fmul	s3, SSQ, s2
 	fadd	SSQ, REGONE, s3
 	fmov	SCALE, s4
-	b	KERNEL_F1_NEXT_\@
-KERNEL_F1_SCALE_GE_XR_\@:
+	b	2f    /* KERNEL_F1_NEXT_\@ */
+1:                    /* KERNEL_F1_SCALE_GE_XR_\@: */
 	fdiv	s2, s4, SCALE
 	fmla	SSQ, s2, v2.s[0]
-KERNEL_F1_NEXT_\@:
+2:                    /* KERNEL_F1_NEXT_\@: */
 	ldr	s5, [X], #4
 	fcmp	s5, REGZERO
-	beq	KERNEL_F1_END_\@
+	beq	4f    /* KERNEL_F1_END_\@ */
 	fabs	s5, s5
 	fcmp	SCALE, s5
-	bge	KERNEL_F1_SCALE_GE_XI_\@
+	bge	3f    /* KERNEL_F1_SCALE_GE_XI_\@ */
 	fdiv	s2, SCALE, s5
 	fmul	s2, s2, s2
 	fmul	s3, SSQ, s2
 	fadd	SSQ, REGONE, s3
 	fmov	SCALE, s5
-	b	KERNEL_F1_END_\@
-KERNEL_F1_SCALE_GE_XI_\@:
+	b	4f    /* KERNEL_F1_END_\@ */
+3:                    /* KERNEL_F1_SCALE_GE_XI_\@: */
 	fdiv	s2, s5, SCALE
 	fmla	SSQ, s2, v2.s[0]
 #else
 	ldr	d4, [X], #8
 	fcmp	d4, REGZERO
-	beq	KERNEL_F1_NEXT_\@
+	beq	2f    /* KERNEL_F1_NEXT_\@ */
 	fabs	d4, d4
 	fcmp	SCALE, d4
-	bge	KERNEL_F1_SCALE_GE_XR_\@
+	bge	1f    /* KERNEL_F1_SCALE_GE_XR_\@ */
 	fdiv	d2, SCALE, d4
 	fmul	d2, d2, d2
 	fmul	d3, SSQ, d2
 	fadd	SSQ, REGONE, d3
 	fmov	SCALE, d4
-	b	KERNEL_F1_NEXT_\@
-KERNEL_F1_SCALE_GE_XR_\@:
+	b	2f    /* KERNEL_F1_NEXT_\@ */
+1:                    /* KERNEL_F1_SCALE_GE_XR_\@: */
 	fdiv	d2, d4, SCALE
 	fmla	SSQ, d2, v2.d[0]
-KERNEL_F1_NEXT_\@:
+2:                   /* KERNEL_F1_NEXT_\@: */
 	ldr	d5, [X], #8
 	fcmp	d5, REGZERO
-	beq	KERNEL_F1_END_\@
+	beq	4f  /* KERNEL_F1_END_\@ */
 	fabs	d5, d5
 	fcmp	SCALE, d5
-	bge	KERNEL_F1_SCALE_GE_XI_\@
+	bge	3f    /* KERNEL_F1_SCALE_GE_XI_\@ */
 	fdiv	d2, SCALE, d5
 	fmul	d2, d2, d2
 	fmul	d3, SSQ, d2
 	fadd	SSQ, REGONE, d3
 	fmov	SCALE, d5
-	b	KERNEL_F1_END_\@
-KERNEL_F1_SCALE_GE_XI_\@:
+	b	4f    /* KERNEL_F1_END_\@ */
+3:                    /* KERNEL_F1_SCALE_GE_XI_\@: */
 	fdiv	d2, d5, SCALE
 	fmla	SSQ, d2, v2.d[0]
 #endif
-KERNEL_F1_END_\@:
+4:                    /* KERNEL_F1_END_\@: */
 .endm
 
 .macro KERNEL_S1

kernel/power/caxpy_power8.S

@@ -34,9 +34,9 @@ caxpy_k:
 	lfs 0,4(10)
 	fmuls 10,2,10
 #ifdef CONJ	
-	fmsubs 11,11,1,10
-#else	
 	fmadds 11,11,1,10
+#else	
+	fmsubs 11,11,1,10
 #endif
 	fadds 12,12,11
 	stfs 12,0(10)
@@ -241,8 +241,13 @@ caxpy_k:
 	lfsx 12,8,5
 	lfsx 0,10,5
 	fmuls 11,2,11
+#ifdef CONJ
 	fmsubs 12,1,12,11
 	fsubs 0,0,12
+#else
+	fmadds 12,1,12,11
+	fadds  0,0,12
+#endif
 	stfsx 0,10,5
 	ble 7,.L39
 	sldi 6,6,2

kernel/power/cdot_power9.S

@@ -1,10 +1,16 @@
-	.file	"cdot.c"
+#define ASSEMBLER
+#include "common.h"
+/*
+.file	"cdot.c"
 	.abiversion 2
 	.section	".text"
 	.align 2
 	.p2align 4,,15
 	.globl cdot_k
 	.type	cdot_k, @function
+*/
+	PROLOGUE
+	
 cdot_k:
 .LCF0:
 0:	addis 2,12,.TOC.-.LCF0@ha

kernel/power/sgemm_logic_power9.S

@@ -136,8 +136,8 @@ LSGEMM_L8x16_BEGIN:
 #endif 
  
     ZERO8x16 
-	mtctr		L 
 	ble		LSGEMM_L8x16_SUB0
+	mtctr		L 
     bl      LSGEMM_L8x16_LMAIN_SUB
 	andi.		L,	T12,	127
 	ble		LSGEMM_L8x16_SAVE
@@ -146,7 +146,7 @@ LSGEMM_L8x16_BEGIN:
 LSGEMM_L8x16_SUB0:
 #if defined(TRMMKERNEL)
 	andi.		L,	T11,	255
-    cmpwi   T11,128
+    cmpwi   T11,129
 #else
 	andi.		L,	K,	255
     cmpwi   K,129