Merge pull request #2072 from martin-frbg/sum

Add (C)BLAS extension ?sum

cblas.h

@@ -73,6 +73,11 @@ double cblas_dasum (OPENBLAS_CONST blasint n, OPENBLAS_CONST double *x, OPENBLAS
 float  cblas_scasum(OPENBLAS_CONST blasint n, OPENBLAS_CONST void  *x, OPENBLAS_CONST blasint incx);
 double cblas_dzasum(OPENBLAS_CONST blasint n, OPENBLAS_CONST void *x, OPENBLAS_CONST blasint incx);
 
+float  cblas_ssum (OPENBLAS_CONST blasint n, OPENBLAS_CONST float  *x, OPENBLAS_CONST blasint incx);
+double cblas_dsum (OPENBLAS_CONST blasint n, OPENBLAS_CONST double *x, OPENBLAS_CONST blasint incx);
+float  cblas_scsum(OPENBLAS_CONST blasint n, OPENBLAS_CONST void  *x, OPENBLAS_CONST blasint incx);
+double cblas_dzsum(OPENBLAS_CONST blasint n, OPENBLAS_CONST void *x, OPENBLAS_CONST blasint incx);
+
 float  cblas_snrm2 (OPENBLAS_CONST blasint N, OPENBLAS_CONST float  *X, OPENBLAS_CONST blasint incX);
 double cblas_dnrm2 (OPENBLAS_CONST blasint N, OPENBLAS_CONST double *X, OPENBLAS_CONST blasint incX);
 float  cblas_scnrm2(OPENBLAS_CONST blasint N, OPENBLAS_CONST void  *X, OPENBLAS_CONST blasint incX);

cmake/kernel.cmake

@@ -107,6 +107,12 @@ macro(SetDefaultL1)
   set(DAXPBYKERNEL ../arm/axpby.c)
   set(CAXPBYKERNEL ../arm/zaxpby.c)
   set(ZAXPBYKERNEL ../arm/zaxpby.c)
+  set(SSUMKERNEL sum.S)
+  set(DSUMKERNEL sum.S)
+  set(CSUMKERNEL zsum.S)
+  set(ZSUMKERNEL zsum.S)
+  set(QSUMKERNEL sum.S)
+  set(XSUMKERNEL zsum.S)
 endmacro ()
 
 macro(SetDefaultL2)
@@ -162,4 +168,4 @@ macro(SetDefaultL3)
   set(DGEADD_KERNEL ../generic/geadd.c)
   set(CGEADD_KERNEL ../generic/zgeadd.c)
   set(ZGEADD_KERNEL ../generic/zgeadd.c)
-endmacro ()
+endmacro ()

common_c.h

@@ -19,6 +19,7 @@
 #define	CDOTC_K			cdotc_k
 #define	CNRM2_K			cnrm2_k
 #define	CSCAL_K			cscal_k
+#define	CSUM_K			csum_k
 #define	CSWAP_K			cswap_k
 #define	CROT_K			csrot_k
 
@@ -249,6 +250,7 @@
 #define	CDOTC_K			gotoblas -> cdotc_k
 #define	CNRM2_K			gotoblas -> cnrm2_k
 #define	CSCAL_K			gotoblas -> cscal_k
+#define	CSUM_K			gotoblas -> csum_k
 #define	CSWAP_K			gotoblas -> cswap_k
 #define	CROT_K			gotoblas -> csrot_k
 

common_d.h

@@ -19,6 +19,7 @@
 #define	DDOTC_K			ddot_k
 #define	DNRM2_K			dnrm2_k
 #define	DSCAL_K			dscal_k
+#define	DSUM_K			dsum_k
 #define	DSWAP_K			dswap_k
 #define	DROT_K			drot_k
 
@@ -174,6 +175,7 @@
 #define	DDOTC_K			gotoblas -> ddot_k
 #define	DNRM2_K			gotoblas -> dnrm2_k
 #define	DSCAL_K			gotoblas -> dscal_k
+#define	DSUM_K			gotoblas -> dsum_k
 #define	DSWAP_K			gotoblas -> dswap_k
 #define	DROT_K			gotoblas -> drot_k
 

common_interface.h

@@ -122,6 +122,13 @@ xdouble BLASFUNC(qasum) (blasint *, xdouble *, blasint *);
 double BLASFUNC(dzasum)(blasint *, double *, blasint *);
 xdouble BLASFUNC(qxasum)(blasint *, xdouble *, blasint *);
 
+FLOATRET  BLASFUNC(ssum) (blasint *, float  *, blasint *);
+FLOATRET  BLASFUNC(scsum)(blasint *, float  *, blasint *);
+double BLASFUNC(dsum) (blasint *, double *, blasint *);
+xdouble BLASFUNC(qsum) (blasint *, xdouble *, blasint *);
+double BLASFUNC(dzsum)(blasint *, double *, blasint *);
+xdouble BLASFUNC(qxsum)(blasint *, xdouble *, blasint *);
+
 blasint    BLASFUNC(isamax)(blasint *, float  *, blasint *);
 blasint    BLASFUNC(idamax)(blasint *, double *, blasint *);
 blasint    BLASFUNC(iqamax)(blasint *, xdouble *, blasint *);

common_level1.h

@@ -100,6 +100,13 @@ float   casum_k (BLASLONG, float  *, BLASLONG);
 double  zasum_k (BLASLONG, double *, BLASLONG);
 xdouble xasum_k (BLASLONG, xdouble *, BLASLONG);
 
+float   ssum_k (BLASLONG, float  *, BLASLONG);
+double  dsum_k (BLASLONG, double *, BLASLONG);
+xdouble qsum_k (BLASLONG, xdouble *, BLASLONG);
+float   csum_k (BLASLONG, float  *, BLASLONG);
+double  zsum_k (BLASLONG, double *, BLASLONG);
+xdouble xsum_k (BLASLONG, xdouble *, BLASLONG);
+
 float   samax_k (BLASLONG, float  *, BLASLONG);
 double  damax_k (BLASLONG, double *, BLASLONG);
 xdouble qamax_k (BLASLONG, xdouble *, BLASLONG);

common_macro.h

@@ -66,6 +66,7 @@
 #define	DOTC_K			QDOTC_K
 #define	NRM2_K			QNRM2_K
 #define	SCAL_K			QSCAL_K
+#define	SUM_K			QSUM_K
 #define	SWAP_K			QSWAP_K
 #define	ROT_K			QROT_K
 
@@ -356,6 +357,7 @@
 #define	DOTC_K			DDOTC_K
 #define	NRM2_K			DNRM2_K
 #define	SCAL_K			DSCAL_K
+#define	SUM_K			DSUM_K
 #define	SWAP_K			DSWAP_K
 #define	ROT_K			DROT_K
 
@@ -658,6 +660,7 @@
 #define	DOTC_K			SDOTC_K
 #define	NRM2_K			SNRM2_K
 #define	SCAL_K			SSCAL_K
+#define	SUM_K			SSUM_K
 #define	SWAP_K			SSWAP_K
 #define	ROT_K			SROT_K
 
@@ -962,6 +965,7 @@
 #define	DOTC_K			XDOTC_K
 #define	NRM2_K			XNRM2_K
 #define	SCAL_K			XSCAL_K
+#define	SUM_K			XSUM_K
 #define	SWAP_K			XSWAP_K
 #define	ROT_K			XROT_K
 
@@ -1363,6 +1367,7 @@
 #define	DOTC_K			ZDOTC_K
 #define	NRM2_K			ZNRM2_K
 #define	SCAL_K			ZSCAL_K
+#define	SUM_K			ZSUM_K
 #define	SWAP_K			ZSWAP_K
 #define	ROT_K			ZROT_K
 
@@ -1785,6 +1790,7 @@
 #define	DOTC_K			CDOTC_K
 #define	NRM2_K			CNRM2_K
 #define	SCAL_K			CSCAL_K
+#define	SUM_K			CSUM_K
 #define	SWAP_K			CSWAP_K
 #define	ROT_K			CROT_K
 

common_param.h

@@ -63,6 +63,7 @@ BLASLONG (*ismin_k) (BLASLONG, float *, BLASLONG);
 
   float  (*snrm2_k) (BLASLONG, float *, BLASLONG);
   float  (*sasum_k) (BLASLONG, float *, BLASLONG);
+  float  (*ssum_k)  (BLASLONG, float *, BLASLONG);
   int    (*scopy_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG);
   float  (*sdot_k)  (BLASLONG, float *, BLASLONG, float *, BLASLONG);
   double (*dsdot_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG);
@@ -154,6 +155,7 @@ BLASLONG (*idmin_k) (BLASLONG, double *, BLASLONG);
 
   double (*dnrm2_k) (BLASLONG, double *, BLASLONG);
   double (*dasum_k) (BLASLONG, double *, BLASLONG);
+  double (*dsum_k)  (BLASLONG, double *, BLASLONG);
   int    (*dcopy_k) (BLASLONG, double *, BLASLONG, double *, BLASLONG);
   double (*ddot_k)  (BLASLONG, double *, BLASLONG, double *, BLASLONG);
   int    (*drot_k)  (BLASLONG, double *, BLASLONG, double *, BLASLONG, double, double);
@@ -245,6 +247,7 @@ BLASLONG (*iqmin_k) (BLASLONG, xdouble *, BLASLONG);
 
  xdouble (*qnrm2_k) (BLASLONG, xdouble *, BLASLONG);
  xdouble (*qasum_k) (BLASLONG, xdouble *, BLASLONG);
+ xdouble (*qsum_k)  (BLASLONG, xdouble *, BLASLONG);
   int    (*qcopy_k) (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
  xdouble (*qdot_k)  (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
   int    (*qrot_k)  (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG, xdouble, xdouble);
@@ -332,6 +335,7 @@ BLASLONG (*icamin_k)(BLASLONG, float *, BLASLONG);
 
   float (*cnrm2_k) (BLASLONG, float *, BLASLONG);
   float (*casum_k) (BLASLONG, float *, BLASLONG);
+  float (*csum_k)  (BLASLONG, float *, BLASLONG);
   int    (*ccopy_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG);
   openblas_complex_float (*cdotu_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG);
   openblas_complex_float (*cdotc_k) (BLASLONG, float *, BLASLONG, float *, BLASLONG);
@@ -495,6 +499,7 @@ BLASLONG (*izamin_k)(BLASLONG, double *, BLASLONG);
 
   double (*znrm2_k) (BLASLONG, double *, BLASLONG);
   double (*zasum_k) (BLASLONG, double *, BLASLONG);
+  double (*zsum_k)  (BLASLONG, double *, BLASLONG);
   int    (*zcopy_k) (BLASLONG, double *, BLASLONG, double *, BLASLONG);
   openblas_complex_double (*zdotu_k) (BLASLONG, double *, BLASLONG, double *, BLASLONG);
   openblas_complex_double (*zdotc_k) (BLASLONG, double *, BLASLONG, double *, BLASLONG);
@@ -660,6 +665,7 @@ BLASLONG (*ixamin_k)(BLASLONG, xdouble *, BLASLONG);
 
   xdouble (*xnrm2_k) (BLASLONG, xdouble *, BLASLONG);
   xdouble (*xasum_k) (BLASLONG, xdouble *, BLASLONG);
+  xdouble (*xsum_k) (BLASLONG, xdouble *, BLASLONG);
   int    (*xcopy_k) (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
   openblas_complex_xdouble (*xdotu_k) (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);
   openblas_complex_xdouble (*xdotc_k) (BLASLONG, xdouble *, BLASLONG, xdouble *, BLASLONG);

common_q.h

@@ -19,6 +19,7 @@
 #define	QDOTC_K			qdot_k
 #define	QNRM2_K			qnrm2_k
 #define	QSCAL_K			qscal_k
+#define	QSUM_K			qsum_k
 #define	QSWAP_K			qswap_k
 #define	QROT_K			qrot_k
 
@@ -161,6 +162,7 @@
 #define	QDOTC_K			gotoblas -> qdot_k
 #define	QNRM2_K			gotoblas -> qnrm2_k
 #define	QSCAL_K			gotoblas -> qscal_k
+#define	QSUM_K			gotoblas -> qsum_k
 #define	QSWAP_K			gotoblas -> qswap_k
 #define	QROT_K			gotoblas -> qrot_k
 

common_s.h

@@ -12,6 +12,7 @@
 #define	ISMAX_K			ismax_k
 #define	ISMIN_K			ismin_k
 #define	SASUM_K			sasum_k
+#define	SSUM_K			ssum_k
 #define	SAXPYU_K		saxpy_k
 #define	SAXPYC_K		saxpy_k
 #define	SCOPY_K			scopy_k
@@ -170,6 +171,7 @@
 #define	ISMAX_K			gotoblas -> ismax_k
 #define	ISMIN_K			gotoblas -> ismin_k
 #define	SASUM_K			gotoblas -> sasum_k
+#define	SSUM_K			gotoblas -> ssum_k
 #define	SAXPYU_K		gotoblas -> saxpy_k
 #define	SAXPYC_K		gotoblas -> saxpy_k
 #define	SCOPY_K			gotoblas -> scopy_k

common_x.h

@@ -19,6 +19,7 @@
 #define	XDOTC_K			xdotc_k
 #define	XNRM2_K			xnrm2_k
 #define	XSCAL_K			xscal_k
+#define	XSUM_K			xsum_k
 #define	XSWAP_K			xswap_k
 #define	XROT_K			xqrot_k
 
@@ -227,6 +228,7 @@
 #define	XDOTC_K			gotoblas -> xdotc_k
 #define	XNRM2_K			gotoblas -> xnrm2_k
 #define	XSCAL_K			gotoblas -> xscal_k
+#define	XSUM_K			gotoblas -> xsum_k
 #define	XSWAP_K			gotoblas -> xswap_k
 #define	XROT_K			gotoblas -> xqrot_k
 

common_z.h

@@ -19,6 +19,7 @@
 #define	ZDOTC_K			zdotc_k
 #define	ZNRM2_K			znrm2_k
 #define	ZSCAL_K			zscal_k
+#define	ZSUM_K			zsum_k
 #define	ZSWAP_K			zswap_k
 #define	ZROT_K			zdrot_k
 
@@ -249,6 +250,7 @@
 #define	ZDOTC_K			gotoblas -> zdotc_k
 #define	ZNRM2_K			gotoblas -> znrm2_k
 #define	ZSCAL_K			gotoblas -> zscal_k
+#define	ZSUM_K			gotoblas -> zsum_k
 #define	ZSWAP_K			gotoblas -> zswap_k
 #define	ZROT_K			gotoblas -> zdrot_k
 

interface/CMakeLists.txt

@@ -12,6 +12,7 @@ set(BLAS1_REAL_ONLY_SOURCES
   rotm.c rotmg.c # N.B. these do not have complex counterparts
   rot.c
   asum.c
+  sum.c
 )
 
 # these will have 'z' prepended for the complex version
@@ -124,6 +125,7 @@ foreach (float_type ${FLOAT_TYPES})
     GenerateNamedObjects("max.c" "USE_ABS;USE_MIN" "scamin" ${CBLAS_FLAG} "" "" true "COMPLEX")
     GenerateNamedObjects("max.c" "USE_ABS" "scamax" ${CBLAS_FLAG} "" "" true "COMPLEX")
     GenerateNamedObjects("asum.c" "" "scasum" ${CBLAS_FLAG} "" "" true "COMPLEX")
+    GenerateNamedObjects("sum.c" "" "scsum" ${CBLAS_FLAG} "" "" true "COMPLEX")
   endif ()
   if (${float_type} STREQUAL "ZCOMPLEX")
     GenerateNamedObjects("zscal.c" "SSCAL" "dscal" ${CBLAS_FLAG} "" "" false "ZCOMPLEX")
@@ -132,6 +134,7 @@ foreach (float_type ${FLOAT_TYPES})
     GenerateNamedObjects("max.c" "USE_ABS;USE_MIN" "dzamin" ${CBLAS_FLAG} "" "" true "ZCOMPLEX")
     GenerateNamedObjects("max.c" "USE_ABS" "dzamax" ${CBLAS_FLAG} "" "" true "ZCOMPLEX")
     GenerateNamedObjects("asum.c" "" "dzasum" ${CBLAS_FLAG} "" "" true "ZCOMPLEX")
+    GenerateNamedObjects("sum.c" "" "dzsum" ${CBLAS_FLAG} "" "" true "ZCOMPLEX")
   endif ()
 endforeach ()
 

interface/Makefile

@@ -25,7 +25,7 @@ SBLAS1OBJS    = \
 		saxpy.$(SUFFIX) sswap.$(SUFFIX) \
 		scopy.$(SUFFIX) sscal.$(SUFFIX) \
 		sdot.$(SUFFIX) sdsdot.$(SUFFIX) dsdot.$(SUFFIX) \
-		sasum.$(SUFFIX) snrm2.$(SUFFIX) \
+		sasum.$(SUFFIX) ssum.$(SUFFIX) snrm2.$(SUFFIX) \
 		smax.$(SUFFIX) samax.$(SUFFIX) ismax.$(SUFFIX) isamax.$(SUFFIX) \
 		smin.$(SUFFIX) samin.$(SUFFIX) ismin.$(SUFFIX) isamin.$(SUFFIX) \
 		srot.$(SUFFIX) srotg.$(SUFFIX) srotm.$(SUFFIX) srotmg.$(SUFFIX) \
@@ -51,7 +51,7 @@ DBLAS1OBJS    = \
 		daxpy.$(SUFFIX) dswap.$(SUFFIX) \
 		dcopy.$(SUFFIX) dscal.$(SUFFIX) \
 		ddot.$(SUFFIX) \
-		dasum.$(SUFFIX) dnrm2.$(SUFFIX) \
+		dasum.$(SUFFIX) dsum.$(SUFFIX) dnrm2.$(SUFFIX) \
 		dmax.$(SUFFIX) damax.$(SUFFIX) idmax.$(SUFFIX) idamax.$(SUFFIX) \
 		dmin.$(SUFFIX) damin.$(SUFFIX) idmin.$(SUFFIX) idamin.$(SUFFIX) \
 		drot.$(SUFFIX) drotg.$(SUFFIX) drotm.$(SUFFIX) drotmg.$(SUFFIX) \
@@ -76,7 +76,7 @@ CBLAS1OBJS    = \
 		caxpy.$(SUFFIX) caxpyc.$(SUFFIX) cswap.$(SUFFIX) \
 		ccopy.$(SUFFIX) cscal.$(SUFFIX) csscal.$(SUFFIX) \
 		cdotc.$(SUFFIX)  cdotu.$(SUFFIX) \
-		scasum.$(SUFFIX) scnrm2.$(SUFFIX) \
+		scasum.$(SUFFIX) scsum.$(SUFFIX) scnrm2.$(SUFFIX) \
 		scamax.$(SUFFIX) icamax.$(SUFFIX) \
 		scamin.$(SUFFIX) icamin.$(SUFFIX) \
 		csrot.$(SUFFIX) crotg.$(SUFFIX) \
@@ -105,7 +105,7 @@ ZBLAS1OBJS    = \
 		zaxpy.$(SUFFIX) zaxpyc.$(SUFFIX) zswap.$(SUFFIX) \
 		zcopy.$(SUFFIX) zscal.$(SUFFIX) zdscal.$(SUFFIX) \
 		zdotc.$(SUFFIX)  zdotu.$(SUFFIX) \
-		dzasum.$(SUFFIX)  dznrm2.$(SUFFIX) \
+		dzasum.$(SUFFIX) dzsum.$(SUFFIX) dznrm2.$(SUFFIX) \
 		dzamax.$(SUFFIX) izamax.$(SUFFIX) \
 		dzamin.$(SUFFIX) izamin.$(SUFFIX) \
 		zdrot.$(SUFFIX) zrotg.$(SUFFIX) \
@@ -146,7 +146,7 @@ QBLAS1OBJS    = \
 		qaxpy.$(SUFFIX) qswap.$(SUFFIX) \
 		qcopy.$(SUFFIX) qscal.$(SUFFIX) \
 		qdot.$(SUFFIX) \
-		qasum.$(SUFFIX) qnrm2.$(SUFFIX) \
+		qasum.$(SUFFIX) qsum.$(SUFFIX) qnrm2.$(SUFFIX) \
 		qmax.$(SUFFIX) qamax.$(SUFFIX) iqmax.$(SUFFIX) iqamax.$(SUFFIX) \
 		qmin.$(SUFFIX) qamin.$(SUFFIX) iqmin.$(SUFFIX) iqamin.$(SUFFIX) \
 		qrot.$(SUFFIX) qrotg.$(SUFFIX) qrotm.$(SUFFIX) qrotmg.$(SUFFIX) \
@@ -168,7 +168,7 @@ XBLAS1OBJS    = \
 		xaxpy.$(SUFFIX) xaxpyc.$(SUFFIX) xswap.$(SUFFIX) \
 		xcopy.$(SUFFIX) xscal.$(SUFFIX) xqscal.$(SUFFIX) \
 		xdotc.$(SUFFIX)  xdotu.$(SUFFIX) \
-		qxasum.$(SUFFIX) qxnrm2.$(SUFFIX) \
+		qxasum.$(SUFFIX) qxsum.$(SUFFIX) qxnrm2.$(SUFFIX) \
 		qxamax.$(SUFFIX) ixamax.$(SUFFIX) \
 		qxamin.$(SUFFIX) ixamin.$(SUFFIX) \
 		xqrot.$(SUFFIX) xrotg.$(SUFFIX) \
@@ -203,7 +203,7 @@ ifdef QUAD_PRECISION
 QBLAS1OBJS    = \
 		qaxpy.$(SUFFIX) qswap.$(SUFFIX) \
 		qcopy.$(SUFFIX) qscal.$(SUFFIX) \
-		qasum.$(SUFFIX) qnrm2.$(SUFFIX) \
+		qasum.$(SUFFIX) qsum.$(SUFFIX) qnrm2.$(SUFFIX) \
 		qmax.$(SUFFIX) qamax.$(SUFFIX) iqmax.$(SUFFIX) iqamax.$(SUFFIX) \
 		qmin.$(SUFFIX) qamin.$(SUFFIX) iqmin.$(SUFFIX) iqamin.$(SUFFIX) \
 		qrot.$(SUFFIX) qrotg.$(SUFFIX) qrotm.$(SUFFIX) qrotmg.$(SUFFIX) \
@@ -224,7 +224,7 @@ QBLAS3OBJS    = \
 XBLAS1OBJS    = \
 		xaxpy.$(SUFFIX) xaxpyc.$(SUFFIX) xswap.$(SUFFIX) \
 		xcopy.$(SUFFIX) xscal.$(SUFFIX) xqscal.$(SUFFIX) \
-		qxasum.$(SUFFIX) qxnrm2.$(SUFFIX) \
+		qxasum.$(SUFFIX) qxsum.$(SUFFIX) qxnrm2.$(SUFFIX) \
 		qxamax.$(SUFFIX) ixamax.$(SUFFIX) \
 		qxamin.$(SUFFIX) ixamin.$(SUFFIX) \
 		xqrot.$(SUFFIX) xrotg.$(SUFFIX) \
@@ -264,7 +264,7 @@ CSBLAS1OBJS   = \
 	cblas_scopy.$(SUFFIX) cblas_sdot.$(SUFFIX) cblas_sdsdot.$(SUFFIX) cblas_dsdot.$(SUFFIX) \
 	cblas_srot.$(SUFFIX) cblas_srotg.$(SUFFIX) cblas_srotm.$(SUFFIX) cblas_srotmg.$(SUFFIX) \
 	cblas_sscal.$(SUFFIX) cblas_sswap.$(SUFFIX) cblas_snrm2.$(SUFFIX) cblas_saxpby.$(SUFFIX) \
-	cblas_ismin.$(SUFFIX) cblas_ismax.$(SUFFIX)
+	cblas_ismin.$(SUFFIX) cblas_ismax.$(SUFFIX) cblas_ssum.$(SUFFIX)
 
 CSBLAS2OBJS   = \
 	cblas_sgemv.$(SUFFIX) cblas_sger.$(SUFFIX) cblas_ssymv.$(SUFFIX) cblas_strmv.$(SUFFIX) \
@@ -282,7 +282,7 @@ CDBLAS1OBJS   = \
 	cblas_dcopy.$(SUFFIX) cblas_ddot.$(SUFFIX) \
 	cblas_drot.$(SUFFIX) cblas_drotg.$(SUFFIX) cblas_drotm.$(SUFFIX) cblas_drotmg.$(SUFFIX) \
 	cblas_dscal.$(SUFFIX) cblas_dswap.$(SUFFIX) cblas_dnrm2.$(SUFFIX) cblas_daxpby.$(SUFFIX) \
-	cblas_idmin.$(SUFFIX) cblas_idmax.$(SUFFIX)
+	cblas_idmin.$(SUFFIX) cblas_idmax.$(SUFFIX) cblas_dsum.$(SUFFIX)
 
 CDBLAS2OBJS   = \
 	cblas_dgemv.$(SUFFIX) cblas_dger.$(SUFFIX) cblas_dsymv.$(SUFFIX) cblas_dtrmv.$(SUFFIX) \
@@ -303,7 +303,7 @@ CCBLAS1OBJS   = \
 	cblas_cscal.$(SUFFIX) cblas_csscal.$(SUFFIX) \
 	cblas_cswap.$(SUFFIX) cblas_scnrm2.$(SUFFIX) \
 	cblas_caxpby.$(SUFFIX) \
-	cblas_icmin.$(SUFFIX) cblas_icmax.$(SUFFIX)
+	cblas_icmin.$(SUFFIX) cblas_icmax.$(SUFFIX) cblas_scsum.$(SUFFIX)
 
 CCBLAS2OBJS   = \
 	cblas_cgemv.$(SUFFIX) cblas_cgerc.$(SUFFIX) cblas_cgeru.$(SUFFIX) \
@@ -330,7 +330,7 @@ CZBLAS1OBJS   = \
 	cblas_zscal.$(SUFFIX) cblas_zdscal.$(SUFFIX) \
 	cblas_zswap.$(SUFFIX) cblas_dznrm2.$(SUFFIX) \
 	cblas_zaxpby.$(SUFFIX) \
-	cblas_izmin.$(SUFFIX) cblas_izmax.$(SUFFIX)
+	cblas_izmin.$(SUFFIX) cblas_izmax.$(SUFFIX) cblas_dzsum.$(SUFFIX)
 
 
 CZBLAS2OBJS   = \
@@ -565,6 +565,24 @@ dzasum.$(SUFFIX) dzasum.$(PSUFFIX) : asum.c
 qxasum.$(SUFFIX) qxasum.$(PSUFFIX) : asum.c
 	$(CC) $(CFLAGS) -c $< -o $(@F)
 
+ssum.$(SUFFIX) ssum.$(PSUFFIX) : sum.c
+	$(CC) $(CFLAGS) -c $< -o $(@F)
+
+dsum.$(SUFFIX) dsum.$(PSUFFIX) : sum.c
+	$(CC) $(CFLAGS) -c $< -o $(@F)
+
+qsum.$(SUFFIX) qsum.$(PSUFFIX) : sum.c
+	$(CC) $(CFLAGS) -c $< -o $(@F)
+
+scsum.$(SUFFIX) scsum.$(PSUFFIX) : sum.c
+	$(CC) $(CFLAGS) -c $< -o $(@F)
+
+dzsum.$(SUFFIX) dzsum.$(PSUFFIX) : sum.c
+	$(CC) $(CFLAGS) -c $< -o $(@F)
+
+qxsum.$(SUFFIX) qxsum.$(PSUFFIX) : sum.c
+	$(CC) $(CFLAGS) -c $< -o $(@F)
+
 snrm2.$(SUFFIX) snrm2.$(PSUFFIX) : nrm2.c
 	$(CC) $(CFLAGS) -c $< -o $(@F)
 
@@ -1412,6 +1430,18 @@ cblas_scasum.$(SUFFIX) cblas_scasum.$(PSUFFIX) : asum.c
 cblas_dzasum.$(SUFFIX) cblas_dzasum.$(PSUFFIX) : asum.c
 	$(CC) $(CFLAGS) -DCBLAS -c $< -o $(@F)
 
+cblas_ssum.$(SUFFIX) cblas_ssum.$(PSUFFIX) : sum.c
+	$(CC) $(CFLAGS) -DCBLAS -c $< -o $(@F)
+
+cblas_dsum.$(SUFFIX) cblas_dsum.$(PSUFFIX) : sum.c
+	$(CC) $(CFLAGS) -DCBLAS -c $< -o $(@F)
+
+cblas_scsum.$(SUFFIX) cblas_scsum.$(PSUFFIX) : sum.c
+	$(CC) $(CFLAGS) -DCBLAS -c $< -o $(@F)
+
+cblas_dzsum.$(SUFFIX) cblas_dzsum.$(PSUFFIX) : sum.c
+	$(CC) $(CFLAGS) -DCBLAS -c $< -o $(@F)
+
 cblas_sdsdot.$(SUFFIX) cblas_sdsdot.$(PSUFFIX) : sdsdot.c
 	$(CC) $(CFLAGS) -DCBLAS -c $< -o $(@F)
 
@@ -1419,7 +1449,7 @@ cblas_dsdot.$(SUFFIX) cblas_dsdot.$(PSUFFIX) : dsdot.c
 	$(CC) $(CFLAGS) -DCBLAS -c $< -o $(@F)
 
 cblas_sdot.$(SUFFIX) cblas_sdot.$(PSUFFIX) : dot.c
-	$(CC) $(CFLAGS) -DCBLAS -c $< -o $(@F)
+		$(CC) $(CFLAGS) -DCBLAS -c $< -o $(@F)
 
 cblas_ddot.$(SUFFIX) cblas_ddot.$(PSUFFIX) : dot.c
 	$(CC) $(CFLAGS) -DCBLAS -c $< -o $(@F)

interface/sum.c

@@ -0,0 +1,97 @@
+/*********************************************************************/
+/* Copyright 2009, 2010 The University of Texas at Austin.           */
+/* All rights reserved.                                              */
+/*                                                                   */
+/* Redistribution and use in source and binary forms, with or        */
+/* without modification, are permitted provided that the following   */
+/* conditions are met:                                               */
+/*                                                                   */
+/*   1. Redistributions of source code must retain the above         */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer.                                                  */
+/*                                                                   */
+/*   2. Redistributions in binary form must reproduce the above      */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer in the documentation and/or other materials       */
+/*      provided with the distribution.                              */
+/*                                                                   */
+/*    THIS  SOFTWARE IS PROVIDED  BY THE  UNIVERSITY OF  TEXAS AT    */
+/*    AUSTIN  ``AS IS''  AND ANY  EXPRESS OR  IMPLIED WARRANTIES,    */
+/*    INCLUDING, BUT  NOT LIMITED  TO, THE IMPLIED  WARRANTIES OF    */
+/*    MERCHANTABILITY  AND FITNESS FOR  A PARTICULAR  PURPOSE ARE    */
+/*    DISCLAIMED.  IN  NO EVENT SHALL THE UNIVERSITY  OF TEXAS AT    */
+/*    AUSTIN OR CONTRIBUTORS BE  LIABLE FOR ANY DIRECT, INDIRECT,    */
+/*    INCIDENTAL,  SPECIAL, EXEMPLARY,  OR  CONSEQUENTIAL DAMAGES    */
+/*    (INCLUDING, BUT  NOT LIMITED TO,  PROCUREMENT OF SUBSTITUTE    */
+/*    GOODS  OR  SERVICES; LOSS  OF  USE,  DATA,  OR PROFITS;  OR    */
+/*    BUSINESS INTERRUPTION) HOWEVER CAUSED  AND ON ANY THEORY OF    */
+/*    LIABILITY, WHETHER  IN CONTRACT, STRICT  LIABILITY, OR TORT    */
+/*    (INCLUDING NEGLIGENCE OR OTHERWISE)  ARISING IN ANY WAY OUT    */
+/*    OF  THE  USE OF  THIS  SOFTWARE,  EVEN  IF ADVISED  OF  THE    */
+/*    POSSIBILITY OF SUCH DAMAGE.                                    */
+/*                                                                   */
+/* The views and conclusions contained in the software and           */
+/* documentation are those of the authors and should not be          */
+/* interpreted as representing official policies, either expressed   */
+/* or implied, of The University of Texas at Austin.                 */
+/*********************************************************************/
+
+#include <stdio.h>
+#include "common.h"
+#ifdef FUNCTION_PROFILE
+#include "functable.h"
+#endif
+
+#ifndef CBLAS
+
+FLOATRET NAME(blasint *N, FLOAT *x, blasint *INCX){
+
+  BLASLONG n    = *N;
+  BLASLONG incx = *INCX;
+  FLOATRET ret;
+
+  PRINT_DEBUG_NAME;
+
+  if (n <= 0) return 0;
+
+  IDEBUG_START;
+
+  FUNCTION_PROFILE_START();
+
+  ret = (FLOATRET)SUM_K(n, x, incx);
+
+  FUNCTION_PROFILE_END(COMPSIZE, n, n);
+
+  IDEBUG_END;
+
+  return ret;
+}
+
+#else
+#ifdef COMPLEX
+FLOAT CNAME(blasint n, void *vx, blasint incx){
+  FLOAT *x = (FLOAT*) vx;
+#else
+FLOAT CNAME(blasint n, FLOAT *x, blasint incx){
+#endif
+
+  FLOAT ret;
+
+  PRINT_DEBUG_CNAME;
+
+  if (n <= 0) return 0;
+
+  IDEBUG_START;
+
+  FUNCTION_PROFILE_START();
+
+  ret = SUM_K(n, x, incx);
+
+  FUNCTION_PROFILE_END(COMPSIZE, n, n);
+
+  IDEBUG_END;
+
+  return ret;
+}
+
+#endif

kernel/CMakeLists.txt

@@ -65,6 +65,7 @@ function (build_core TARGET_CORE KDIR TSUFFIX KERNEL_DEFINITIONS)
       GenerateNamedObjects("${KERNELDIR}/${${float_char}SCALKERNEL}" "" "scal_k" false "" "" false ${float_type})
       GenerateNamedObjects("${KERNELDIR}/${${float_char}SWAPKERNEL}" "" "swap_k" false "" "" false ${float_type})
       GenerateNamedObjects("${KERNELDIR}/${${float_char}AXPBYKERNEL}" "" "axpby_k" false "" "" false ${float_type})
+      GenerateNamedObjects("${KERNELDIR}/${${float_char}SUMKERNEL}" "" "sum_k" false "" "" false ${float_type})
 
       if (${float_type} STREQUAL "COMPLEX" OR ${float_type} STREQUAL "ZCOMPLEX")
         GenerateNamedObjects("${KERNELDIR}/${${float_char}AXPYKERNEL}" "CONJ" "axpyc_k" false "" "" false ${float_type})

kernel/Makefile.L1

@@ -340,6 +340,32 @@ ifndef XSCALKERNEL
 XSCALKERNEL = zscal.S
 endif
 
+### SUM ###
+
+ifndef SSUMKERNEL
+SSUMKERNEL =  sum.S
+endif
+
+ifndef DSUMKERNEL
+DSUMKERNEL =  sum.S
+endif
+
+ifndef CSUMKERNEL
+CSUMKERNEL = zsum.S
+endif
+
+ifndef ZSUMKERNEL
+ZSUMKERNEL = zsum.S
+endif
+
+ifndef QSUMKERNEL
+QSUMKERNEL =  sum.S
+endif
+
+ifndef XSUMKERNEL
+XSUMKERNEL = zsum.S
+endif
+
 ### SWAP ###
 
 ifndef SSWAPKERNEL
@@ -453,7 +479,7 @@ endif
 SBLASOBJS	+= \
 	 samax_k$(TSUFFIX).$(SUFFIX)  samin_k$(TSUFFIX).$(SUFFIX)  smax_k$(TSUFFIX).$(SUFFIX)  smin_k$(TSUFFIX).$(SUFFIX) \
 	isamax_k$(TSUFFIX).$(SUFFIX) isamin_k$(TSUFFIX).$(SUFFIX) ismax_k$(TSUFFIX).$(SUFFIX) ismin_k$(TSUFFIX).$(SUFFIX) \
-	sasum_k$(TSUFFIX).$(SUFFIX) saxpy_k$(TSUFFIX).$(SUFFIX) scopy_k$(TSUFFIX).$(SUFFIX) \
+	sasum_k$(TSUFFIX).$(SUFFIX) ssum_k$(TSUFFIX).$(SUFFIX) saxpy_k$(TSUFFIX).$(SUFFIX) scopy_k$(TSUFFIX).$(SUFFIX) \
 	sdot_k$(TSUFFIX).$(SUFFIX) sdsdot_k$(TSUFFIX).$(SUFFIX) dsdot_k$(TSUFFIX).$(SUFFIX) \
 	snrm2_k$(TSUFFIX).$(SUFFIX) srot_k$(TSUFFIX).$(SUFFIX) sscal_k$(TSUFFIX).$(SUFFIX) sswap_k$(TSUFFIX).$(SUFFIX) \
 	saxpby_k$(TSUFFIX).$(SUFFIX)
@@ -463,31 +489,32 @@ DBLASOBJS	+= \
 	idamax_k$(TSUFFIX).$(SUFFIX) idamin_k$(TSUFFIX).$(SUFFIX) idmax_k$(TSUFFIX).$(SUFFIX) idmin_k$(TSUFFIX).$(SUFFIX) \
 	dasum_k$(TSUFFIX).$(SUFFIX) daxpy_k$(TSUFFIX).$(SUFFIX) dcopy_k$(TSUFFIX).$(SUFFIX) ddot_k$(TSUFFIX).$(SUFFIX) \
 	dnrm2_k$(TSUFFIX).$(SUFFIX) drot_k$(TSUFFIX).$(SUFFIX) dscal_k$(TSUFFIX).$(SUFFIX) dswap_k$(TSUFFIX).$(SUFFIX) \
-	daxpby_k$(TSUFFIX).$(SUFFIX)
+	daxpby_k$(TSUFFIX).$(SUFFIX) dsum_k$(TSUFFIX).$(SUFFIX)
 
 QBLASOBJS	+= \
 	 qamax_k$(TSUFFIX).$(SUFFIX)  qamin_k$(TSUFFIX).$(SUFFIX)  qmax_k$(TSUFFIX).$(SUFFIX)  qmin_k$(TSUFFIX).$(SUFFIX) \
 	iqamax_k$(TSUFFIX).$(SUFFIX) iqamin_k$(TSUFFIX).$(SUFFIX) iqmax_k$(TSUFFIX).$(SUFFIX) iqmin_k$(TSUFFIX).$(SUFFIX) \
 	qasum_k$(TSUFFIX).$(SUFFIX) qaxpy_k$(TSUFFIX).$(SUFFIX) qcopy_k$(TSUFFIX).$(SUFFIX) qdot_k$(TSUFFIX).$(SUFFIX) \
-	qnrm2_k$(TSUFFIX).$(SUFFIX) qrot_k$(TSUFFIX).$(SUFFIX) qscal_k$(TSUFFIX).$(SUFFIX) qswap_k$(TSUFFIX).$(SUFFIX)
+	qnrm2_k$(TSUFFIX).$(SUFFIX) qrot_k$(TSUFFIX).$(SUFFIX) qscal_k$(TSUFFIX).$(SUFFIX) qswap_k$(TSUFFIX).$(SUFFIX) \
+	qsum_k$(TSUFFIX).$(SUFFIX)
 
 CBLASOBJS	+= \
 	camax_k$(TSUFFIX).$(SUFFIX)  camin_k$(TSUFFIX).$(SUFFIX) icamax_k$(TSUFFIX).$(SUFFIX) icamin_k$(TSUFFIX).$(SUFFIX) \
 	casum_k$(TSUFFIX).$(SUFFIX) caxpy_k$(TSUFFIX).$(SUFFIX) caxpyc_k$(TSUFFIX).$(SUFFIX) ccopy_k$(TSUFFIX).$(SUFFIX) \
 	cdotc_k$(TSUFFIX).$(SUFFIX) cdotu_k$(TSUFFIX).$(SUFFIX) cnrm2_k$(TSUFFIX).$(SUFFIX) csrot_k$(TSUFFIX).$(SUFFIX) \
-	cscal_k$(TSUFFIX).$(SUFFIX) cswap_k$(TSUFFIX).$(SUFFIX) caxpby_k$(TSUFFIX).$(SUFFIX)
+	cscal_k$(TSUFFIX).$(SUFFIX) cswap_k$(TSUFFIX).$(SUFFIX) caxpby_k$(TSUFFIX).$(SUFFIX) csum_k$(TSUFFIX).$(SUFFIX)
 
 ZBLASOBJS	+= \
 	 zamax_k$(TSUFFIX).$(SUFFIX)  zamin_k$(TSUFFIX).$(SUFFIX) izamax_k$(TSUFFIX).$(SUFFIX) izamin_k$(TSUFFIX).$(SUFFIX) \
 	zasum_k$(TSUFFIX).$(SUFFIX) zaxpy_k$(TSUFFIX).$(SUFFIX) zaxpyc_k$(TSUFFIX).$(SUFFIX) zcopy_k$(TSUFFIX).$(SUFFIX) \
 	zdotc_k$(TSUFFIX).$(SUFFIX) zdotu_k$(TSUFFIX).$(SUFFIX) znrm2_k$(TSUFFIX).$(SUFFIX) zdrot_k$(TSUFFIX).$(SUFFIX) \
-	zscal_k$(TSUFFIX).$(SUFFIX) zswap_k$(TSUFFIX).$(SUFFIX) zaxpby_k$(TSUFFIX).$(SUFFIX)
+	zscal_k$(TSUFFIX).$(SUFFIX) zswap_k$(TSUFFIX).$(SUFFIX) zaxpby_k$(TSUFFIX).$(SUFFIX) zsum_k$(TSUFFIX).$(SUFFIX)
 
 XBLASOBJS	+= \
 	 xamax_k$(TSUFFIX).$(SUFFIX)  xamin_k$(TSUFFIX).$(SUFFIX) ixamax_k$(TSUFFIX).$(SUFFIX) ixamin_k$(TSUFFIX).$(SUFFIX) \
 	xasum_k$(TSUFFIX).$(SUFFIX) xaxpy_k$(TSUFFIX).$(SUFFIX) xaxpyc_k$(TSUFFIX).$(SUFFIX) xcopy_k$(TSUFFIX).$(SUFFIX) \
 	xdotc_k$(TSUFFIX).$(SUFFIX) xdotu_k$(TSUFFIX).$(SUFFIX) xnrm2_k$(TSUFFIX).$(SUFFIX) xqrot_k$(TSUFFIX).$(SUFFIX) \
-	xscal_k$(TSUFFIX).$(SUFFIX) xswap_k$(TSUFFIX).$(SUFFIX)
+	xscal_k$(TSUFFIX).$(SUFFIX) xswap_k$(TSUFFIX).$(SUFFIX) xsum_k$(TSUFFIX).$(SUFFIX)
 
 ### AMAX ###
 
@@ -617,7 +644,7 @@ $(KDIR)idmin_k$(TSUFFIX).$(SUFFIX)  $(KDIR)idmin_k$(TPSUFFIX).$(PSUFFIX)  : $(KE
 $(KDIR)iqmin_k$(TSUFFIX).$(SUFFIX)  $(KDIR)iqmin_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(IQMINKERNEL)
 	$(CC) -c $(CFLAGS) -UCOMPLEX -DXDOUBLE -UUSE_ABS -DUSE_MIN $< -o $@
 
-
+### ASUM ###
 $(KDIR)sasum_k$(TSUFFIX).$(SUFFIX)   $(KDIR)sasum_k$(TPSUFFIX).$(PSUFFIX)   : $(KERNELDIR)/$(SASUMKERNEL)
 	$(CC) -c $(CFLAGS) -UCOMPLEX -UDOUBLE $< -o $@
 
@@ -636,6 +663,26 @@ $(KDIR)zasum_k$(TSUFFIX).$(SUFFIX)  $(KDIR)zasum_k$(TPSUFFIX).$(PSUFFIX)  : $(KE
 $(KDIR)xasum_k$(TSUFFIX).$(SUFFIX)  $(KDIR)xasum_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(XASUMKERNEL)
 	$(CC) -c $(CFLAGS) -DCOMPLEX -DXDOUBLE $< -o $@
 
+### SUM ###
+$(KDIR)ssum_k$(TSUFFIX).$(SUFFIX)   $(KDIR)ssum_k$(TPSUFFIX).$(PSUFFIX)   : $(KERNELDIR)/$(SSUMKERNEL)
+	$(CC) -c $(CFLAGS) -UCOMPLEX -UDOUBLE $< -o $@
+
+$(KDIR)dsum_k$(TSUFFIX).$(SUFFIX)   $(KDIR)dsum_k$(TPSUFFIX).$(PSUFFIX)   : $(KERNELDIR)/$(DSUMKERNEL)
+	$(CC) -c $(CFLAGS) -UCOMPLEX -DDOUBLE $< -o $@
+
+$(KDIR)qsum_k$(TSUFFIX).$(SUFFIX)   $(KDIR)qsum_k$(TPSUFFIX).$(PSUFFIX)   : $(KERNELDIR)/$(QSUMKERNEL)
+	$(CC) -c $(CFLAGS) -UCOMPLEX -DXDOUBLE $< -o $@
+
+$(KDIR)csum_k$(TSUFFIX).$(SUFFIX)  $(KDIR)csum_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(CSUMKERNEL)
+	$(CC) -c $(CFLAGS) -DCOMPLEX -UDOUBLE $< -o $@
+
+$(KDIR)zsum_k$(TSUFFIX).$(SUFFIX)  $(KDIR)zsum_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(ZSUMKERNEL)
+	$(CC) -c $(CFLAGS) -DCOMPLEX -DDOUBLE $< -o $@
+
+$(KDIR)xsum_k$(TSUFFIX).$(SUFFIX)  $(KDIR)xsum_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(XSUMKERNEL)
+	$(CC) -c $(CFLAGS) -DCOMPLEX -DXDOUBLE $< -o $@
+
+### AXPY ###
 $(KDIR)saxpy_k$(TSUFFIX).$(SUFFIX)  $(KDIR)saxpy_k$(TPSUFFIX).$(PSUFFIX)  : $(KERNELDIR)/$(SAXPYKERNEL)
 	$(CC) -c $(CFLAGS) -UCOMPLEX -UDOUBLE $< -o $@
 

kernel/alpha/sum.S

@@ -0,0 +1,206 @@
+/*********************************************************************/
+/* Copyright 2009, 2010 The University of Texas at Austin.           */
+/* All rights reserved.                                              */
+/*                                                                   */
+/* Redistribution and use in source and binary forms, with or        */
+/* without modification, are permitted provided that the following   */
+/* conditions are met:                                               */
+/*                                                                   */
+/*   1. Redistributions of source code must retain the above         */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer.                                                  */
+/*                                                                   */
+/*   2. Redistributions in binary form must reproduce the above      */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer in the documentation and/or other materials       */
+/*      provided with the distribution.                              */
+/*                                                                   */
+/*    THIS  SOFTWARE IS PROVIDED  BY THE  UNIVERSITY OF  TEXAS AT    */
+/*    AUSTIN  ``AS IS''  AND ANY  EXPRESS OR  IMPLIED WARRANTIES,    */
+/*    INCLUDING, BUT  NOT LIMITED  TO, THE IMPLIED  WARRANTIES OF    */
+/*    MERCHANTABILITY  AND FITNESS FOR  A PARTICULAR  PURPOSE ARE    */
+/*    DISCLAIMED.  IN  NO EVENT SHALL THE UNIVERSITY  OF TEXAS AT    */
+/*    AUSTIN OR CONTRIBUTORS BE  LIABLE FOR ANY DIRECT, INDIRECT,    */
+/*    INCIDENTAL,  SPECIAL, EXEMPLARY,  OR  CONSEQUENTIAL DAMAGES    */
+/*    (INCLUDING, BUT  NOT LIMITED TO,  PROCUREMENT OF SUBSTITUTE    */
+/*    GOODS  OR  SERVICES; LOSS  OF  USE,  DATA,  OR PROFITS;  OR    */
+/*    BUSINESS INTERRUPTION) HOWEVER CAUSED  AND ON ANY THEORY OF    */
+/*    LIABILITY, WHETHER  IN CONTRACT, STRICT  LIABILITY, OR TORT    */
+/*    (INCLUDING NEGLIGENCE OR OTHERWISE)  ARISING IN ANY WAY OUT    */
+/*    OF  THE  USE OF  THIS  SOFTWARE,  EVEN  IF ADVISED  OF  THE    */
+/*    POSSIBILITY OF SUCH DAMAGE.                                    */
+/*                                                                   */
+/* The views and conclusions contained in the software and           */
+/* documentation are those of the authors and should not be          */
+/* interpreted as representing official policies, either expressed   */
+/* or implied, of The University of Texas at Austin.                 */
+/*********************************************************************/
+
+#define ASSEMBLER
+#include "common.h"
+#include "version.h"
+
+#define PREFETCHSIZE	88
+
+#define N	$16
+#define X	$17
+#define INCX	$18
+#define I	$19
+
+#define s0	$f0
+#define s1	$f1
+#define s2	$f10
+#define s3	$f11
+
+#define a0	$f12
+#define a1	$f13
+#define a2	$f14
+#define a3	$f15
+#define a4	$f16
+#define a5	$f17
+#define a6	$f18
+#define a7	$f19
+
+#define t0	$f20
+#define t1	$f21
+#define t2	$f22
+#define t3	$f23
+
+	PROLOGUE
+	PROFCODE
+
+	fclr	s0
+	unop
+	fclr	t0
+	ble	N,  $L999
+
+	sra	N, 3, I
+	fclr	s1
+	fclr	s2
+	ble	I, $L15
+
+	LD	a0,  0 * SIZE(X)
+	fclr	t1
+	SXADDQ	INCX, X, X
+	fclr	t2
+
+	LD	a1,  0 * SIZE(X)
+	fclr	t3
+	SXADDQ	INCX, X, X
+	fclr	s3
+
+	LD	a2,  0 * SIZE(X)
+	SXADDQ	INCX, X, X
+	LD	a3,  0 * SIZE(X)
+	SXADDQ	INCX, X, X
+
+	LD	a4,  0 * SIZE(X)
+	SXADDQ	INCX, X, X
+	LD	a5,  0 * SIZE(X)
+	SXADDQ	INCX, X, X
+
+	lda	I,  -1(I)
+	ble	I, $L13
+	.align 4
+
+$L12:
+	ADD	s0, t0, s0
+	ldl	$31, PREFETCHSIZE * 2 * SIZE(X)
+	fmov	a0, t0
+	lda	I,  -1(I)
+
+	ADD	s1, t1, s1
+	LD	a6,  0 * SIZE(X)
+	fmov	a1, t1
+	SXADDQ	INCX, X, X
+
+	ADD	s2, t2, s2
+	LD	a7,  0 * SIZE(X)
+	fmov	a2, t2
+	SXADDQ	INCX, X, X
+
+	ADD	s3, t3, s3
+	LD	a0,  0 * SIZE(X)
+	fmov	a3, t3
+	SXADDQ	INCX, X, X
+
+	ADD	s0, t0, s0
+	LD	a1,  0 * SIZE(X)
+	fmov	a4, t0
+	SXADDQ	INCX, X, X
+
+	ADD	s1, t1, s1
+	LD	a2,  0 * SIZE(X)
+	fmov	a5, t1
+	SXADDQ	INCX, X, X
+
+	ADD	s2, t2, s2
+	LD	a3,  0 * SIZE(X)
+	fmov	a6, t2
+	SXADDQ	INCX, X, X
+
+	ADD	s3, t3, s3
+	LD	a4,  0 * SIZE(X)
+	fmov	a7, t3
+	SXADDQ	INCX, X, X
+
+	LD	a5,  0 * SIZE(X)
+	unop
+	SXADDQ	INCX, X, X
+	bne	I, $L12
+	.align 4
+
+$L13:
+	ADD	s0, t0, s0
+	LD	a6,  0 * SIZE(X)
+	fmov	a0, t0
+	SXADDQ	INCX, X, X
+
+	ADD	s1, t1, s1
+	LD	a7,  0 * SIZE(X)
+	fmov	a1, t1
+	SXADDQ	INCX, X, X
+
+	ADD	s2, t2, s2
+	fmov	a2, t2
+	ADD	s3, t3, s3
+	fmov	a3, t3
+
+	ADD	s0, t0, s0
+	fmov	a4, t0
+	ADD	s1, t1, s1
+	fmov	a5, t1
+	ADD	s2, t2, s2
+	fmov	a6, t2
+	ADD	s3, t3, s3
+	fmov	a7, t3
+
+	ADD	s1, t1, s1
+	ADD	s2, t2, s2
+	ADD	s3, t3, s3
+
+	ADD	s0, s1, s0
+	ADD	s2, s3, s2
+	.align 4
+
+$L15:
+	and	N, 7, I
+	ADD	s0, s2, s0
+	unop
+	ble	I, $L999
+	.align 4
+
+$L17:
+	ADD	s0, t0, s0
+	LD	a0,  0 * SIZE(X)
+	SXADDQ	INCX, X, X
+	fmov	a0, t0
+
+	lda	I,  -1(I)
+	bne	I, $L17
+	.align 4
+
+$L999:
+	ADD	s0, t0, s0
+	ret
+	EPILOGUE

kernel/alpha/zsum.S

@@ -0,0 +1,208 @@
+/*********************************************************************/
+/* Copyright 2009, 2010 The University of Texas at Austin.           */
+/* All rights reserved.                                              */
+/*                                                                   */
+/* Redistribution and use in source and binary forms, with or        */
+/* without modification, are permitted provided that the following   */
+/* conditions are met:                                               */
+/*                                                                   */
+/*   1. Redistributions of source code must retain the above         */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer.                                                  */
+/*                                                                   */
+/*   2. Redistributions in binary form must reproduce the above      */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer in the documentation and/or other materials       */
+/*      provided with the distribution.                              */
+/*                                                                   */
+/*    THIS  SOFTWARE IS PROVIDED  BY THE  UNIVERSITY OF  TEXAS AT    */
+/*    AUSTIN  ``AS IS''  AND ANY  EXPRESS OR  IMPLIED WARRANTIES,    */
+/*    INCLUDING, BUT  NOT LIMITED  TO, THE IMPLIED  WARRANTIES OF    */
+/*    MERCHANTABILITY  AND FITNESS FOR  A PARTICULAR  PURPOSE ARE    */
+/*    DISCLAIMED.  IN  NO EVENT SHALL THE UNIVERSITY  OF TEXAS AT    */
+/*    AUSTIN OR CONTRIBUTORS BE  LIABLE FOR ANY DIRECT, INDIRECT,    */
+/*    INCIDENTAL,  SPECIAL, EXEMPLARY,  OR  CONSEQUENTIAL DAMAGES    */
+/*    (INCLUDING, BUT  NOT LIMITED TO,  PROCUREMENT OF SUBSTITUTE    */
+/*    GOODS  OR  SERVICES; LOSS  OF  USE,  DATA,  OR PROFITS;  OR    */
+/*    BUSINESS INTERRUPTION) HOWEVER CAUSED  AND ON ANY THEORY OF    */
+/*    LIABILITY, WHETHER  IN CONTRACT, STRICT  LIABILITY, OR TORT    */
+/*    (INCLUDING NEGLIGENCE OR OTHERWISE)  ARISING IN ANY WAY OUT    */
+/*    OF  THE  USE OF  THIS  SOFTWARE,  EVEN  IF ADVISED  OF  THE    */
+/*    POSSIBILITY OF SUCH DAMAGE.                                    */
+/*                                                                   */
+/* The views and conclusions contained in the software and           */
+/* documentation are those of the authors and should not be          */
+/* interpreted as representing official policies, either expressed   */
+/* or implied, of The University of Texas at Austin.                 */
+/*********************************************************************/
+
+#define ASSEMBLER
+#include "common.h"
+#include "version.h"
+
+#define PREFETCHSIZE	88
+
+#define N	$16
+#define X	$17
+#define INCX	$18
+#define I	$19
+
+#define s0	$f0
+#define s1	$f1
+#define s2	$f10
+#define s3	$f11
+
+#define a0	$f12
+#define a1	$f13
+#define a2	$f14
+#define a3	$f15
+#define a4	$f16
+#define a5	$f17
+#define a6	$f18
+#define a7	$f19
+
+#define t0	$f20
+#define t1	$f21
+#define t2	$f22
+#define t3	$f23
+
+	PROLOGUE
+	PROFCODE
+
+	fclr	s0
+	unop
+	fclr	t0
+	addq	INCX, INCX, INCX
+
+	fclr	s1
+	unop
+	fclr	t1
+	ble	N,  $L999
+
+	fclr	s2
+	sra	N, 2, I
+	fclr	s3
+	ble	I, $L15
+
+	LD	a0,  0 * SIZE(X)
+	fclr	t2
+	LD	a1,  1 * SIZE(X)
+	SXADDQ	INCX, X, X
+
+	LD	a2,  0 * SIZE(X)
+	fclr	t3
+	LD	a3,  1 * SIZE(X)
+	SXADDQ	INCX, X, X
+
+	LD	a4,  0 * SIZE(X)
+	LD	a5,  1 * SIZE(X)
+	SXADDQ	INCX, X, X
+	lda	I,  -1(I)
+
+	ble	I, $L13
+	.align 4
+
+$L12:
+	ADD	s0, t0, s0
+	ldl	$31, PREFETCHSIZE * SIZE(X)
+	fmov	a0, t0
+	lda	I,  -1(I)
+
+	ADD	s1, t1, s1
+	LD	a6,  0 * SIZE(X)
+	fmov	a1, t1
+	unop
+
+	ADD	s2, t2, s2
+	LD	a7,  1 * SIZE(X)
+	fmov	a2, t2
+	SXADDQ	INCX, X, X
+
+	ADD	s3, t3, s3
+	LD	a0,  0 * SIZE(X)
+	fmov	a3, t3
+	unop
+
+	ADD	s0, t0, s0
+	LD	a1,  1 * SIZE(X)
+	fmov	a4, t0
+	SXADDQ	INCX, X, X
+
+	ADD	s1, t1, s1
+	LD	a2,  0 * SIZE(X)
+	fmov	a5, t1
+	unop
+
+	ADD	s2, t2, s2
+	LD	a3,  1 * SIZE(X)
+	fmov	a6, t2
+	SXADDQ	INCX, X, X
+
+	ADD	s3, t3, s3
+	LD	a4,  0 * SIZE(X)
+	fmov	a7, t3
+	unop
+
+	LD	a5,  1 * SIZE(X)
+	unop
+	SXADDQ	INCX, X, X
+	bne	I, $L12
+	.align 4
+
+$L13:
+	ADD	s0, t0, s0
+	LD	a6,  0 * SIZE(X)
+	fmov	a0, t0
+
+	ADD	s1, t1, s1
+	LD	a7,  1 * SIZE(X)
+	fmov	a1, t1
+	SXADDQ	INCX, X, X
+
+	ADD	s2, t2, s2
+	fmov	a2, t2
+	ADD	s3, t3, s3
+	fmov	a3, t3
+
+	ADD	s0, t0, s0
+	fmov	a4, t0
+	ADD	s1, t1, s1
+	fmov	a5, t1
+	ADD	s2, t2, s2
+	fmov	a6, t2
+	ADD	s3, t3, s3
+	fmov	a7, t3
+
+	ADD	s2, t2, s2
+	ADD	s3, t3, s3
+
+	.align 4
+
+$L15:
+	ADD	s0, s2, s0
+	and	N, 3, I
+	ADD	s1, s3, s1
+	ble	I, $L999
+	.align 4
+
+$L17:
+	ADD	s0, t0, s0
+	LD	a0,  0 * SIZE(X)
+	fmov	a0, t0
+	lda	I,  -1(I)
+
+	ADD	s1, t1, s1
+	LD	a1,  1 * SIZE(X)
+	fmov	a1, t1
+	SXADDQ	INCX, X, X
+
+	bne	I, $L17
+	.align 4
+
+$L999:
+	ADD	s0, t0, s0
+	ADD	s1, t1, s1
+
+	ADD	s0, s1, s0
+	ret
+	EPILOGUE

kernel/arm/KERNEL.ARMV5

@@ -35,6 +35,11 @@ DASUMKERNEL  = ../arm/asum.c
 CASUMKERNEL  = ../arm/zasum.c
 ZASUMKERNEL  = ../arm/zasum.c
 
+SSUMKERNEL  = ../arm/sum.c
+DSUMKERNEL  = ../arm/sum.c
+CSUMKERNEL  = ../arm/zsum.c
+ZSUMKERNEL  = ../arm/zsum.c
+
 SAXPYKERNEL  = ../arm/axpy.c
 DAXPYKERNEL  = ../arm/axpy.c
 CAXPYKERNEL  = ../arm/zaxpy.c

kernel/arm/KERNEL.ARMV6

@@ -37,6 +37,9 @@ DASUMKERNEL  = asum_vfp.S
 CASUMKERNEL  = asum_vfp.S
 ZASUMKERNEL  = asum_vfp.S
 
+SSUMKERNEL  = sum_vfp.S
+DSUMKERNEL  = sum_vfp.S
+
 SAXPYKERNEL  = axpy_vfp.S
 DAXPYKERNEL  = axpy_vfp.S
 CAXPYKERNEL  = axpy_vfp.S

kernel/arm/sum.c

@@ -0,0 +1,51 @@
+/***************************************************************************
+Copyright (c) 2013, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+/**************************************************************************************
+* trivial copy of asum.c with the ABS() removed                                       *
+**************************************************************************************/
+
+
+#include "common.h"
+#include <math.h>
+
+FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
+{
+	BLASLONG i=0;
+	FLOAT sumf = 0.0;
+	if (n <= 0 || inc_x <= 0) return(sumf);
+
+	n *= inc_x;
+	while(i < n)
+	{
+		sumf += x[i];
+		i += inc_x;
+	}
+	return(sumf);
+}
+
+

kernel/arm/sum_vfp.S

@@ -0,0 +1,425 @@
+/***************************************************************************
+Copyright (c) 2013, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+/**************************************************************************************
+* trivial copy of asum_vfp.S with the in-place vabs.f64 calls removed                                    *
+**************************************************************************************/
+
+#define ASSEMBLER
+#include "common.h"
+
+#define STACKSIZE 256
+
+#define	N	r0
+#define	X	r1
+#define	INC_X	r2
+
+
+#define I	r12
+
+#define X_PRE	512
+
+/**************************************************************************************
+* Macro definitions
+**************************************************************************************/
+
+#if	!defined(COMPLEX)
+
+#if	defined(DOUBLE)
+
+.macro KERNEL_F4
+
+	pld	[ X, #X_PRE  ]
+	vldmia.f64	X!, { d4 - d5 }
+	vadd.f64   d0  , d0,  d4
+	vldmia.f64	X!, { d6 - d7 }
+	vadd.f64   d1  , d1,  d5
+	vadd.f64   d0  , d0,  d6
+	vadd.f64   d1  , d1,  d7
+
+.endm
+
+.macro KERNEL_F1
+
+	vldmia.f64	X!, { d4 }
+	vadd.f64   d0  , d0,  d4
+
+.endm
+
+
+.macro KERNEL_S4
+
+	vldmia.f64	X, { d4 }
+	vadd.f64   d0  , d0,  d4
+	add	X, X, INC_X
+
+	vldmia.f64	X, { d4 }
+	vadd.f64   d0  , d0,  d4
+	add	X, X, INC_X
+
+	vldmia.f64	X, { d4 }
+	vadd.f64   d0  , d0,  d4
+	add	X, X, INC_X
+
+	vldmia.f64	X, { d4 }
+	vadd.f64   d0  , d0,  d4
+	add	X, X, INC_X
+
+.endm
+
+
+.macro KERNEL_S1
+
+	vldmia.f64	X, { d4 }
+	vadd.f64   d0  , d0,  d4
+	add	X, X, INC_X
+
+.endm
+
+#else
+
+.macro KERNEL_F4
+
+	vldmia.f32	X!, { s4 - s5 }
+	vadd.f32   s0  , s0,  s4
+	vldmia.f32	X!, { s6 - s7 }
+	vadd.f32   s1  , s1,  s5
+	vadd.f32   s0  , s0,  s6
+	vadd.f32   s1  , s1,  s7
+
+.endm
+
+.macro KERNEL_F1
+
+	vldmia.f32	X!, { s4 }
+	vadd.f32   s0  , s0,  s4
+
+.endm
+
+
+.macro KERNEL_S4
+
+	vldmia.f32	X, { s4 }
+	vadd.f32   s0  , s0,  s4
+	add	X, X, INC_X
+
+	vldmia.f32	X, { s4 }
+	vadd.f32   s0  , s0,  s4
+	add	X, X, INC_X
+
+	vldmia.f32	X, { s4 }
+	vadd.f32   s0  , s0,  s4
+	add	X, X, INC_X
+
+	vldmia.f32	X, { s4 }
+	vadd.f32   s0  , s0,  s4
+	add	X, X, INC_X
+
+.endm
+
+
+.macro KERNEL_S1
+
+	vldmia.f32	X, { s4 }
+	vadd.f32   s0  , s0,  s4
+	add	X, X, INC_X
+
+.endm
+
+
+#endif
+
+#else
+
+#if	defined(DOUBLE)
+
+.macro KERNEL_F4
+
+	pld	[ X, #X_PRE  ]
+	vldmia.f64	X!, { d4 - d5 }
+	vadd.f64   d0  , d0,  d4
+	vldmia.f64	X!, { d6 - d7 }
+	vadd.f64   d1  , d1,  d5
+	vadd.f64   d0  , d0,  d6
+	vadd.f64   d1  , d1,  d7
+
+	pld	[ X, #X_PRE  ]
+	vldmia.f64	X!, { d4 - d5 }
+	vadd.f64   d0  , d0,  d4
+	vldmia.f64	X!, { d6 - d7 }
+	vadd.f64   d1  , d1,  d5
+	vadd.f64   d0  , d0,  d6
+	vadd.f64   d1  , d1,  d7
+
+
+.endm
+
+.macro KERNEL_F1
+
+	vldmia.f64	X!, { d4 }
+	vadd.f64   d0  , d0,  d4
+
+	vldmia.f64	X!, { d4 }
+	vadd.f64   d0  , d0,  d4
+
+
+.endm
+
+
+.macro KERNEL_S4
+
+	vldmia.f64	X, { d4 -d5 }
+	vadd.f64   d0  , d0,  d4
+	vadd.f64   d0  , d0,  d5
+	add	X, X, INC_X
+
+	vldmia.f64	X, { d4 -d5 }
+	vadd.f64   d0  , d0,  d4
+	vadd.f64   d0  , d0,  d5
+	add	X, X, INC_X
+
+	vldmia.f64	X, { d4 -d5 }
+	vadd.f64   d0  , d0,  d4
+	vadd.f64   d0  , d0,  d5
+	add	X, X, INC_X
+
+	vldmia.f64	X, { d4 -d5 }
+	vadd.f64   d0  , d0,  d4
+	vadd.f64   d0  , d0,  d5
+	add	X, X, INC_X
+
+.endm
+
+
+.macro KERNEL_S1
+
+	vldmia.f64	X, { d4 -d5 }
+	vadd.f64   d0  , d0,  d4
+	vadd.f64   d0  , d0,  d5
+	add	X, X, INC_X
+
+.endm
+
+#else
+
+.macro KERNEL_F4
+
+	pld	[ X, #X_PRE  ]
+	vldmia.f32	X!, { s4 - s5 }
+	vadd.f32   s0  , s0,  s4
+	vldmia.f32	X!, { s6 - s7 }
+	vadd.f32   s1  , s1,  s5
+	vadd.f32   s0  , s0,  s6
+	vadd.f32   s1  , s1,  s7
+
+	vldmia.f32	X!, { s4 - s5 }
+	vadd.f32   s0  , s0,  s4
+	vldmia.f32	X!, { s6 - s7 }
+	vadd.f32   s1  , s1,  s5
+	vadd.f32   s0  , s0,  s6
+	vadd.f32   s1  , s1,  s7
+
+
+.endm
+
+.macro KERNEL_F1
+
+	vldmia.f32	X!, { s4 }
+	vadd.f32   s0  , s0,  s4
+
+	vldmia.f32	X!, { s4 }
+	vadd.f32   s0  , s0,  s4
+
+.endm
+
+
+.macro KERNEL_S4
+
+	vldmia.f32	X, { s4 -s5 }
+	vadd.f32   s0  , s0,  s4
+	vadd.f32   s0  , s0,  s5
+	add	X, X, INC_X
+
+	vldmia.f32	X, { s4 -s5 }
+	vadd.f32   s0  , s0,  s4
+	vadd.f32   s0  , s0,  s5
+	add	X, X, INC_X
+
+	vldmia.f32	X, { s4 -s5 }
+	vadd.f32   s0  , s0,  s4
+	vadd.f32   s0  , s0,  s5
+	add	X, X, INC_X
+
+	vldmia.f32	X, { s4 -s5 }
+	vadd.f32   s0  , s0,  s4
+	vadd.f32   s0  , s0,  s5
+	add	X, X, INC_X
+
+.endm
+
+
+.macro KERNEL_S1
+
+	vldmia.f32	X, { s4 -s5 }
+	vadd.f32   s0  , s0,  s4
+	vadd.f32   s0  , s0,  s5
+	add	X, X, INC_X
+
+.endm
+
+#endif
+
+#endif
+
+/**************************************************************************************
+* End of macro definitions
+**************************************************************************************/
+
+	PROLOGUE
+
+	.align 5
+
+        movs    r12, #0                                          // clear floating point register
+        vmov    s0, r12
+        vmov    s1, r12
+#if     defined(DOUBLE)
+        vcvt.f64.f32    d0, s0
+        vcvt.f64.f32    d1, s1
+#endif
+
+	cmp	N, #0
+	ble	asum_kernel_L999
+
+	cmp	INC_X, #0
+	beq	asum_kernel_L999
+
+	cmp	INC_X, #1
+	bne	asum_kernel_S_BEGIN
+
+
+asum_kernel_F_BEGIN:
+
+	asrs	I, N, #2					// I = N / 4
+	ble	asum_kernel_F1
+
+	.align 5
+
+asum_kernel_F4:
+
+#if !defined(DOUBLE) && !defined(COMPLEX)
+	pld	[ X, #X_PRE  ]
+#endif
+	KERNEL_F4
+
+	subs	I, I, #1
+	ble	asum_kernel_F1
+
+	KERNEL_F4
+
+	subs	I, I, #1
+	bne	asum_kernel_F4
+
+asum_kernel_F1:
+
+	ands	I, N, #3
+	ble	asum_kernel_L999
+
+asum_kernel_F10:
+
+	KERNEL_F1
+
+	subs    I, I, #1
+        bne     asum_kernel_F10
+
+	b	asum_kernel_L999
+
+asum_kernel_S_BEGIN:
+
+#if defined(COMPLEX)
+
+#if defined(DOUBLE)
+	lsl	INC_X, INC_X, #4				// INC_X * SIZE * 2
+#else
+	lsl	INC_X, INC_X, #3				// INC_X * SIZE * 2
+#endif
+
+#else
+
+#if defined(DOUBLE)
+	lsl	INC_X, INC_X, #3				// INC_X * SIZE
+#else
+	lsl	INC_X, INC_X, #2				// INC_X * SIZE
+#endif
+
+#endif
+
+	asrs	I, N, #2					// I = N / 4
+	ble	asum_kernel_S1
+
+	.align 5
+
+asum_kernel_S4:
+
+	KERNEL_S4
+
+	subs	I, I, #1
+	bne	asum_kernel_S4
+
+asum_kernel_S1:
+
+	ands	I, N, #3
+	ble	asum_kernel_L999
+
+asum_kernel_S10:
+
+	KERNEL_S1
+
+	subs    I, I, #1
+        bne     asum_kernel_S10
+
+
+asum_kernel_L999:
+
+
+#if defined(DOUBLE)
+	vadd.f64	d0 , d0, d1				// set return value
+#else
+	vadd.f32	s0 , s0, s1				// set return value
+#endif
+
+#if !defined(__ARM_PCS_VFP)
+#if !defined(DOUBLE)
+	vmov	r0, s0
+#else
+	vmov	r0, r1, d0
+#endif
+#endif
+
+	bx	lr
+
+	EPILOGUE
+

kernel/arm/zsum.c

@@ -0,0 +1,57 @@
+/***************************************************************************
+Copyright (c) 2013, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+/**************************************************************************************
+* trivial copy of zasum.c with the ABS() removed                                      *
+**************************************************************************************/
+
+
+#include "common.h"
+#include <math.h>
+
+#define CSUM1(x,i)	x[i]+x[i+1]
+
+FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
+{
+	BLASLONG i=0;
+	FLOAT sumf = 0.0;
+	BLASLONG inc_x2;
+
+	if (n <= 0 || inc_x <= 0) return(sumf);
+
+	inc_x2 = 2 * inc_x;
+
+	n *= inc_x2;
+	while(i < n)
+	{
+		sumf += CSUM1(x,i);
+		i += inc_x2;
+	}
+	return(sumf);
+}
+
+

kernel/arm64/csum.S

@@ -0,0 +1,164 @@
+/*******************************************************************************
+Copyright (c) 2019, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*******************************************************************************/
+
+#define ASSEMBLER
+#include "common.h"
+
+#define	N	x0	/* vector length */
+#define	X	x1	/* X vector address */
+#define	INC_X	x2	/* X stride */
+#define I	x5	/* loop variable */
+
+/*******************************************************************************
+* Macro definitions
+*******************************************************************************/
+
+#define REG0	wzr
+#define SUMF	s0
+#define TMPF	s1
+#define TMPVF	{v1.s}[0]
+#define SZ	4
+
+/******************************************************************************/
+
+.macro KERNEL_F1
+	ld1	{v1.2s}, [X], #8
+	ext	v2.8b, v1.8b, v1.8b, #4
+	fadd	TMPF, TMPF, s2
+	fadd	SUMF, SUMF, TMPF
+.endm
+
+.macro KERNEL_F8
+	ld1	{v1.4s, v2.4s, v3.4s, v4.4s}, [X]
+	add	X, X, #64
+
+	PRFM	PLDL1KEEP, [X, #1024]
+
+	fadd	v1.4s, v1.4s, v2.4s
+	fadd	v3.4s, v3.4s, v4.4s
+	fadd	v0.4s, v0.4s, v1.4s
+	fadd	v0.4s, v0.4s, v3.4s
+.endm
+
+.macro KERNEL_F8_FINALIZE
+	ext	v1.16b, v0.16b, v0.16b, #8
+	fadd	v0.2s, v0.2s, v1.2s
+	faddp	SUMF, v0.2s
+.endm
+
+.macro INIT_S
+	lsl	INC_X, INC_X, #3
+.endm
+
+.macro KERNEL_S1
+	ld1	{v1.2s}, [X], INC_X
+	ext	v2.8b, v1.8b, v1.8b, #4
+	fadd	TMPF, TMPF, s2
+	fadd	SUMF, SUMF, TMPF
+
+.endm
+
+/*******************************************************************************
+* End of macro definitions
+*******************************************************************************/
+
+	PROLOGUE
+
+	fmov	SUMF, REG0
+	fmov	s1, SUMF
+
+	cmp	N, xzr
+	ble	.Lcsum_kernel_L999
+	cmp	INC_X, xzr
+	ble	.Lcsum_kernel_L999
+
+	cmp	INC_X, #1
+	bne	.Lcsum_kernel_S_BEGIN
+
+.Lcsum_kernel_F_BEGIN:
+
+	asr	I, N, #3
+	cmp	I, xzr
+	beq	.Lcsum_kernel_F1
+
+.Lcsum_kernel_F8:
+
+	KERNEL_F8
+
+	subs	I, I, #1
+	bne	.Lcsum_kernel_F8
+
+	KERNEL_F8_FINALIZE
+
+.Lcsum_kernel_F1:
+
+	ands	I, N, #7
+	ble	.Lcsum_kernel_L999
+
+.Lcsum_kernel_F10:
+
+	KERNEL_F1
+
+	subs    I, I, #1
+        bne     .Lcsum_kernel_F10
+
+.Lcsum_kernel_L999:
+	ret
+
+.Lcsum_kernel_S_BEGIN:
+
+	INIT_S
+
+	asr	I, N, #2
+	cmp	I, xzr
+	ble	.Lcsum_kernel_S1
+
+.Lcsum_kernel_S4:
+
+	KERNEL_S1
+	KERNEL_S1
+	KERNEL_S1
+	KERNEL_S1
+
+	subs	I, I, #1
+	bne	.Lcsum_kernel_S4
+
+.Lcsum_kernel_S1:
+
+	ands	I, N, #3
+	ble	.Lcsum_kernel_L999
+
+.Lcsum_kernel_S10:
+
+	KERNEL_S1
+
+	subs    I, I, #1
+        bne     .Lcsum_kernel_S10
+
+	ret
+
+	EPILOGUE

kernel/arm64/sum.S

@@ -0,0 +1,186 @@
+/*******************************************************************************
+Copyright (c) 2019, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*******************************************************************************/
+
+#define ASSEMBLER
+#include "common.h"
+
+#define	N	x0	/* vector length */
+#define	X	x1	/* X vector address */
+#define	INC_X	x2	/* X stride */
+#define I	x5	/* loop variable */
+
+/*******************************************************************************
+* Macro definitions
+*******************************************************************************/
+
+#if !defined(DOUBLE)
+#define REG0	wzr
+#define SUMF	s0
+#define TMPF	s1
+#define TMPVF	{v1.s}[0]
+#define SZ	4
+#else
+#define REG0	xzr
+#define SUMF	d0
+#define TMPF	d1
+#define TMPVF	{v1.d}[0]
+#define SZ	8
+#endif
+
+/******************************************************************************/
+
+.macro KERNEL_F1
+	ldr	TMPF, [X], #SZ
+	fadd	SUMF, SUMF, TMPF
+.endm
+
+.macro KERNEL_F8
+#if !defined(DOUBLE)
+	ld1	{v1.4s, v2.4s}, [X], #32	// Load [X3, X2, X1, X0]
+	fadd	v1.4s, v1.4s, v2.4s		// [X3+X1, X2+X0]
+	fadd	v0.4s, v0.4s, v1.4s		// [X3+X1, X2+X0]
+	PRFM	PLDL1KEEP, [X, #1024]
+#else // DOUBLE
+	ld1	{v2.2d, v3.2d, v4.2d, v5.2d}, [X]
+	add	X, X, #64
+
+	PRFM	PLDL1KEEP, [X, #1024]
+
+	fadd	v2.2d, v2.2d, v3.2d
+	fadd	v4.2d, v4.2d, v5.2d
+	fadd	v0.2d, v0.2d, v2.2d
+	fadd	v0.2d, v0.2d, v4.2d
+#endif
+.endm
+
+.macro KERNEL_F8_FINALIZE
+#if !defined(DOUBLE)
+	ext	v1.16b, v0.16b, v0.16b, #8
+	fadd	v0.2s, v0.2s, v1.2s
+	faddp	SUMF, v0.2s
+#else
+	faddp	SUMF, v0.2d
+#endif
+.endm
+
+.macro INIT_S
+#if !defined(DOUBLE)
+	lsl	INC_X, INC_X, #2
+#else
+	lsl	INC_X, INC_X, #3
+#endif
+.endm
+
+.macro KERNEL_S1
+	ld1	TMPVF, [X], INC_X
+	fadd	SUMF, SUMF, TMPF
+.endm
+
+/*******************************************************************************
+* End of macro definitions
+*******************************************************************************/
+
+	PROLOGUE
+
+	fmov	SUMF, REG0
+#if !defined(DOUBLE)
+	fmov	s1, SUMF
+#else
+	fmov	d1, SUMF
+#endif
+
+	cmp	N, xzr
+	ble	.Lsum_kernel_L999
+	cmp	INC_X, xzr
+	ble	.Lsum_kernel_L999
+
+	cmp	INC_X, #1
+	bne	.Lsum_kernel_S_BEGIN
+
+.Lsum_kernel_F_BEGIN:
+
+	asr	I, N, #3
+	cmp	I, xzr
+	beq	.Lsum_kernel_F1
+
+.Lsum_kernel_F8:
+
+	KERNEL_F8
+
+	subs	I, I, #1
+	bne	.Lsum_kernel_F8
+
+	KERNEL_F8_FINALIZE
+
+.Lsum_kernel_F1:
+
+	ands	I, N, #7
+	ble	.Lsum_kernel_L999
+
+.Lsum_kernel_F10:
+
+	KERNEL_F1
+
+	subs    I, I, #1
+        bne     .Lsum_kernel_F10
+
+.Lsum_kernel_L999:
+	ret
+
+.Lsum_kernel_S_BEGIN:
+
+	INIT_S
+
+	asr	I, N, #2
+	cmp	I, xzr
+	ble	.Lsum_kernel_S1
+
+.Lsum_kernel_S4:
+
+	KERNEL_S1
+	KERNEL_S1
+	KERNEL_S1
+	KERNEL_S1
+
+	subs	I, I, #1
+	bne	.Lsum_kernel_S4
+
+.Lsum_kernel_S1:
+
+	ands	I, N, #3
+	ble	.Lsum_kernel_L999
+
+.Lsum_kernel_S10:
+
+	KERNEL_S1
+
+	subs    I, I, #1
+        bne     .Lsum_kernel_S10
+
+	ret
+
+	EPILOGUE

kernel/arm64/zsum.S

@@ -0,0 +1,158 @@
+/*******************************************************************************
+Copyright (c) 2015, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*******************************************************************************/
+
+#define ASSEMBLER
+#include "common.h"
+
+#define	N	x0	/* vector length */
+#define	X	x1	/* X vector address */
+#define	INC_X	x2	/* X stride */
+#define I	x5	/* loop variable */
+
+/*******************************************************************************
+* Macro definitions
+*******************************************************************************/
+
+#define REG0	xzr
+#define SUMF	d0
+#define TMPF	d1
+#define TMPVF	{v1.d}[0]
+#define SZ	8
+
+/******************************************************************************/
+
+.macro KERNEL_F1
+	ld1	{v1.2d}, [X], #16
+	faddp	TMPF, v1.2d
+	fadd	SUMF, SUMF, TMPF
+.endm
+
+.macro KERNEL_F4
+	ld1	{v1.2d, v2.2d, v3.2d, v4.2d}, [X], #64
+
+	fadd	v1.2d, v1.2d, v2.2d
+	fadd	v3.2d, v3.2d, v4.2d
+
+	fadd	v0.2d, v0.2d, v1.2d
+	fadd	v0.2d, v0.2d, v3.2d
+
+	PRFM	PLDL1KEEP, [X, #1024]
+.endm
+
+.macro KERNEL_F4_FINALIZE
+	faddp	SUMF, v0.2d
+.endm
+
+.macro INIT_S
+	lsl	INC_X, INC_X, #4
+.endm
+
+.macro KERNEL_S1
+	ld1	{v1.2d}, [X], INC_X
+	faddp	TMPF, v1.2d
+	fadd	SUMF, SUMF, TMPF
+.endm
+
+/*******************************************************************************
+* End of macro definitions
+*******************************************************************************/
+
+	PROLOGUE
+
+	fmov	SUMF, REG0
+
+	cmp	N, xzr
+	ble	.Lzsum_kernel_L999
+	cmp	INC_X, xzr
+	ble	.Lzsum_kernel_L999
+
+	cmp	INC_X, #1
+	bne	.Lzsum_kernel_S_BEGIN
+
+.Lzsum_kernel_F_BEGIN:
+
+	asr	I, N, #2
+	cmp	I, xzr
+	beq	.Lzsum_kernel_F1
+
+.Lzsum_kernel_F4:
+
+	KERNEL_F4
+
+	subs	I, I, #1
+	bne	.Lzsum_kernel_F4
+
+	KERNEL_F4_FINALIZE
+
+.Lzsum_kernel_F1:
+
+	ands	I, N, #3
+	ble	.Lzsum_kernel_L999
+
+.Lzsum_kernel_F10:
+
+	KERNEL_F1
+
+	subs    I, I, #1
+        bne     .Lzsum_kernel_F10
+
+.Lzsum_kernel_L999:
+	ret
+
+.Lzsum_kernel_S_BEGIN:
+
+	INIT_S
+
+	asr	I, N, #2
+	cmp	I, xzr
+	ble	.Lzsum_kernel_S1
+
+.Lzsum_kernel_S4:
+
+	KERNEL_S1
+	KERNEL_S1
+	KERNEL_S1
+	KERNEL_S1
+
+	subs	I, I, #1
+	bne	.Lzsum_kernel_S4
+
+.Lzsum_kernel_S1:
+
+	ands	I, N, #3
+	ble	.Lzsum_kernel_L999
+
+.Lzsum_kernel_S10:
+
+	KERNEL_S1
+
+	subs    I, I, #1
+        bne     .Lzsum_kernel_S10
+
+	ret
+
+	EPILOGUE

kernel/ia64/KERNEL

@@ -60,6 +60,10 @@ CASUMKERNEL  = asum.S
 ZASUMKERNEL  = asum.S
 XASUMKERNEL  = asum.S
 
+CSUMKERNEL  = sum.S
+ZSUMKERNEL  = sum.S
+XSUMKERNEL  = sum.S
+
 CNRM2KERNEL  = nrm2.S
 ZNRM2KERNEL  = nrm2.S
 XNRM2KERNEL  = nrm2.S

kernel/ia64/sum.S

@@ -0,0 +1,358 @@
+/*********************************************************************/
+/* Copyright 2009, 2010 The University of Texas at Austin.           */
+/* Copyright 2019, The OpenBLAS project                              */
+/* All rights reserved.                                              */
+/*                                                                   */
+/* Redistribution and use in source and binary forms, with or        */
+/* without modification, are permitted provided that the following   */
+/* conditions are met:                                               */
+/*                                                                   */
+/*   1. Redistributions of source code must retain the above         */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer.                                                  */
+/*                                                                   */
+/*   2. Redistributions in binary form must reproduce the above      */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer in the documentation and/or other materials       */
+/*      provided with the distribution.                              */
+/*                                                                   */
+/*    THIS  SOFTWARE IS PROVIDED  BY THE  UNIVERSITY OF  TEXAS AT    */
+/*    AUSTIN  ``AS IS''  AND ANY  EXPRESS OR  IMPLIED WARRANTIES,    */
+/*    INCLUDING, BUT  NOT LIMITED  TO, THE IMPLIED  WARRANTIES OF    */
+/*    MERCHANTABILITY  AND FITNESS FOR  A PARTICULAR  PURPOSE ARE    */
+/*    DISCLAIMED.  IN  NO EVENT SHALL THE UNIVERSITY  OF TEXAS AT    */
+/*    AUSTIN OR CONTRIBUTORS BE  LIABLE FOR ANY DIRECT, INDIRECT,    */
+/*    INCIDENTAL,  SPECIAL, EXEMPLARY,  OR  CONSEQUENTIAL DAMAGES    */
+/*    (INCLUDING, BUT  NOT LIMITED TO,  PROCUREMENT OF SUBSTITUTE    */
+/*    GOODS  OR  SERVICES; LOSS  OF  USE,  DATA,  OR PROFITS;  OR    */
+/*    BUSINESS INTERRUPTION) HOWEVER CAUSED  AND ON ANY THEORY OF    */
+/*    LIABILITY, WHETHER  IN CONTRACT, STRICT  LIABILITY, OR TORT    */
+/*    (INCLUDING NEGLIGENCE OR OTHERWISE)  ARISING IN ANY WAY OUT    */
+/*    OF  THE  USE OF  THIS  SOFTWARE,  EVEN  IF ADVISED  OF  THE    */
+/*    POSSIBILITY OF SUCH DAMAGE.                                    */
+/*                                                                   */
+/* The views and conclusions contained in the software and           */
+/* documentation are those of the authors and should not be          */
+/* interpreted as representing official policies, either expressed   */
+/* or implied, of The University of Texas at Austin.                 */
+/*********************************************************************/
+
+#define ASSEMBLER
+#include "common.h"
+
+#ifdef XDOUBLE
+#define PREFETCH_SIZE ( 8 * 16 +  4)
+#elif defined(DOUBLE)
+#define PREFETCH_SIZE (16 * 16 +  8)
+#else
+#define PREFETCH_SIZE (32 * 16 + 16)
+#endif
+
+#ifndef COMPLEX
+#define COMPADD	0
+#define STRIDE INCX
+#else
+#define COMPADD	1
+#define STRIDE SIZE
+#endif
+
+#define PRE1	r2
+
+#define I	r17
+#define J	r18
+#define INCX16	r21
+
+#define PR	r30
+#define ARLC	r31
+
+#define N	r32
+#define X	r33
+#define INCX	r34
+
+
+	PROLOGUE
+	.prologue
+	PROFCODE
+	{ .mfi
+	adds	PRE1 = PREFETCH_SIZE * SIZE, X
+	mov	f8   = f0
+	.save ar.lc, ARLC
+	mov	ARLC = ar.lc
+	}
+	;;
+	.body
+#ifdef F_INTERFACE
+	{ .mmi
+	LDINT	N    = [N]
+	LDINT	INCX = [INCX]
+	nop.i	0
+	}
+	;;
+#ifndef USE64BITINT
+	{ .mii
+	nop.m	0
+	sxt4	N = N
+	sxt4	INCX = INCX
+	}
+	;;
+#endif
+#endif
+	{ .mmi
+	cmp.lt	p0, p6 = r0, INCX
+	cmp.lt	p0, p7 = r0, N
+	shr	I =  N, (4 - COMPADD)
+	}
+	{ .mbb
+	and	J = ((1 << (4 - COMPADD)) - 1), N
+	(p6) 	br.ret.sptk.many b0
+	(p7) 	br.ret.sptk.many b0
+	}
+	;;
+	{ .mfi
+	adds	I = -1, I
+	mov	f10 = f0
+	mov	PR = pr
+	}
+	{ .mfi
+	cmp.eq	p9, p0  =   r0, J
+	mov	f9  = f0
+	tbit.z	p0, p12 = N, 3 - COMPADD
+	}
+	;;
+	{ .mmi
+	cmp.eq	p16, p0 = r0, r0
+	cmp.ne	p17, p0 = r0, r0
+	mov	ar.ec= 3
+	}
+	{ .mfi
+	cmp.ne	p18, p0 = r0, r0
+	mov	f11 = f0
+	shl	INCX = INCX, BASE_SHIFT + COMPADD
+	}
+	;;
+	{ .mmi
+#ifdef XDOUBLE
+	shladd	INCX16  = INCX, (3 - COMPADD), r0
+#else
+	shladd	INCX16  = INCX, (4 - COMPADD), r0
+#endif
+	cmp.ne	p19, p0 = r0, r0
+	mov	ar.lc = I
+	}
+	{ .mmb
+	cmp.gt	p8 ,p0  =   r0, I
+#ifdef COMPLEX
+       adds	INCX = - SIZE, INCX
+#else
+	nop.m	0
+#endif
+	(p8) br.cond.dpnt  .L55
+	}
+	;;
+	.align 32
+
+.L52:
+	{ .mmf
+	(p16) lfetch.nt1 [PRE1], INCX16
+	(p16) LDFD	f32  = [X], STRIDE
+	}
+	{ .mfb
+	(p19) FADD	f8  = f8,  f71
+	}
+	;;
+	{ .mmf
+	(p16) LDFD	f35  = [X], INCX
+	}
+	{ .mfb
+	(p19) FADD	f9  = f9,  f74
+	}
+	;;
+	{ .mmf
+	(p16) LDFD	f38  = [X], STRIDE
+	}
+	{ .mfb
+	(p19) FADD	f10 = f10, f77
+	}
+	;;
+	{ .mmf
+	(p16) LDFD	f41  = [X], INCX
+	}
+	{ .mfb
+	(p19) FADD	f11 = f11, f80
+	}
+	;;
+	{ .mmf
+	(p16) LDFD	f44  = [X], STRIDE
+	}
+	{ .mfb
+	(p18) FADD	f8  = f8,  f34
+	}
+	;;
+	{ .mmf
+	(p16) LDFD	f47  = [X], INCX
+	}
+	{ .mfb
+	(p18) FADD	f9  = f9,  f37
+	}
+	;;
+	{ .mmf
+	(p16) LDFD	f50  = [X], STRIDE
+	}
+	{ .mfb
+	(p18) FADD	f10 = f10, f40
+	}
+	;;
+	{ .mmf
+	(p16) LDFD	f53  = [X], INCX
+	}
+	{ .mfb
+	(p18) FADD	f11 = f11, f43
+	}
+	;;
+	{ .mmf
+#ifdef XDOUBLE
+	(p16) lfetch.nt1 [PRE1], INCX16
+#endif
+	(p16) LDFD	f56  = [X], STRIDE
+	}
+	{ .mfb
+	(p18) FADD	f8  = f8,  f46
+	}
+	;;
+	{ .mmf
+	(p16) LDFD	f59  = [X], INCX
+	}
+	{ .mfb
+	(p18) FADD	f9  = f9,  f49
+	}
+	;;
+	{ .mmf
+	(p16) LDFD	f62  = [X], STRIDE
+	}
+	{ .mfb
+	(p18) FADD	f10 = f10, f52
+	}
+	;;
+	{ .mmf
+	(p16) LDFD	f65  = [X], INCX
+	}
+	{ .mfb
+	(p18) FADD	f11 = f11, f55
+	}
+	;;
+	{ .mmf
+	(p16) LDFD	f68  = [X], STRIDE
+	}
+	{ .mfb
+	(p18) FADD	f8  = f8,  f58
+	}
+	;;
+	{ .mmf
+	(p16) LDFD	f71  = [X], INCX
+	}
+	{ .mfb
+	(p18) FADD	f9  = f9,  f61
+	}
+	;;
+	{ .mmf
+	(p16) LDFD	f74  = [X], STRIDE
+	}
+	{ .mfb
+	(p18) FADD	f10 = f10, f64
+	}
+	;;
+	{ .mmf
+	(p16) LDFD	f77  = [X], INCX
+	}
+	{ .mfb
+	(p18) FADD	f11 = f11, f67
+	br.ctop.sptk.few .L52
+	}
+	;;
+	FADD	f8  = f8,  f71
+	FADD	f9  = f9,  f74
+	FADD	f10 = f10, f77
+	FADD	f11 = f11, f80
+	.align 32
+	;;
+.L55:
+	(p12) LDFD	f32  = [X], STRIDE
+	(p9) br.cond.dptk .L998
+	;;
+	(p12) LDFD	f33  = [X], INCX
+	;;
+	(p12) LDFD	f34  = [X], STRIDE
+	;;
+	(p12) LDFD	f35  = [X], INCX
+	tbit.z	p0, p13 = N, (2 - COMPADD)
+	;;
+	(p12) LDFD	f36  = [X], STRIDE
+	tbit.z	p0, p14 = N, (1 - COMPADD)
+	;;
+	(p12) LDFD	f37  = [X], INCX
+#ifndef COMPLEX
+	tbit.z	p0, p15 = N, 0
+#endif
+	;;
+	(p12) LDFD	f38  = [X], STRIDE
+	;;
+	(p12) LDFD	f39  = [X], INCX
+	;;
+	(p13) LDFD	f40  = [X], STRIDE
+	;;
+	(p13) LDFD	f41  = [X], INCX
+	;;
+	(p13) LDFD	f42  = [X], STRIDE
+	(p12) FADD	f8  = f8,  f32
+	;;
+	(p13) LDFD	f43  = [X], INCX
+	(p12) FADD	f9  = f9,  f33
+	;;
+	(p14) LDFD	f44  = [X], STRIDE
+	(p12) FADD	f10 = f10, f34
+	;;
+	(p14) LDFD	f45  = [X], INCX
+	(p12) FADD	f11 = f11, f35
+	;;
+#ifndef COMPLEX
+	(p15) LDFD	f46  = [X]
+#endif
+	(p12) FADD	f8  = f8,  f36
+	;;
+	(p12) FADD	f9  = f9,  f37
+	(p12) FADD	f10 = f10, f38
+	(p12) FADD	f11 = f11, f39
+	;;
+	(p13) FADD	f8  = f8,  f40
+	(p13) FADD	f9  = f9,  f41
+#ifndef COMPLEX
+#endif
+	(p13) FADD	f10 = f10, f42
+	;;
+	(p13) FADD	f11 = f11, f43
+	(p14) FADD	f8  = f8,  f44
+	(p14) FADD	f9  = f9,  f45
+#ifndef COMPLEX
+	(p15) FADD	f10 = f10, f46
+#endif
+	;;
+	.align 32
+
+.L998:
+	{ .mfi
+	FADD	f8  = f8,  f9
+	mov	ar.lc  = ARLC
+	}
+	{ .mmf
+	FADD	f10 = f10, f11
+	}
+	;;
+	{ .mii
+	mov	pr = PR, -65474
+	}
+	;;
+	{ .mfb
+	FADD	f8  = f8,  f10
+	br.ret.sptk.many b0
+	}
+	EPILOGUE

kernel/mips/KERNEL.P5600

@@ -30,6 +30,11 @@ IDMAXKERNEL  = ../mips/imax.c
 ISMINKERNEL  = ../mips/imin.c
 IDMINKERNEL  = ../mips/imin.c
 
+SSUMKERNEL  = ../mips/sum.c
+DSUMKERNEL  = ../mips/sum.c
+CSUMKERNEL  = ../mips/zsum.c
+ZSUMKERNEL  = ../mips/zsum.c
+
 ifdef HAVE_MSA
 SASUMKERNEL  = ../mips/sasum_msa.c
 DASUMKERNEL  = ../mips/dasum_msa.c

kernel/mips/sum.c

@@ -0,0 +1,47 @@
+/***************************************************************************
+Copyright (c) 2016, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include "common.h"
+#include <math.h>
+
+
+FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
+{
+	BLASLONG i=0;
+	FLOAT sumf = 0.0;
+	if (n <= 0 || inc_x <= 0) return(sumf);
+
+	n *= inc_x;
+	while(i < n)
+	{
+		sumf += x[i];
+		i += inc_x;
+	}
+	return(sumf);
+}
+
+

kernel/mips/zsum.c

@@ -0,0 +1,52 @@
+/***************************************************************************
+Copyright (c) 2016, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include "common.h"
+#include <math.h>
+
+#define CSUM1(x,i)	x[i]+x[i+1]
+
+FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
+{
+	BLASLONG i=0;
+	FLOAT sumf = 0.0;
+	BLASLONG inc_x2;
+
+	if (n <= 0 || inc_x <= 0) return(sumf);
+
+	inc_x2 = 2 * inc_x;
+
+	n *= inc_x2;
+	while(i < n)
+	{
+		sumf += CSUM1(x,i);
+		i += inc_x2;
+	}
+	return(sumf);
+}
+
+

kernel/mips64/sum.S

@@ -0,0 +1,332 @@
+/*********************************************************************/
+/* Copyright 2009, 2010 The University of Texas at Austin.           */
+/* All rights reserved.                                              */
+/*                                                                   */
+/* Redistribution and use in source and binary forms, with or        */
+/* without modification, are permitted provided that the following   */
+/* conditions are met:                                               */
+/*                                                                   */
+/*   1. Redistributions of source code must retain the above         */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer.                                                  */
+/*                                                                   */
+/*   2. Redistributions in binary form must reproduce the above      */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer in the documentation and/or other materials       */
+/*      provided with the distribution.                              */
+/*                                                                   */
+/*    THIS  SOFTWARE IS PROVIDED  BY THE  UNIVERSITY OF  TEXAS AT    */
+/*    AUSTIN  ``AS IS''  AND ANY  EXPRESS OR  IMPLIED WARRANTIES,    */
+/*    INCLUDING, BUT  NOT LIMITED  TO, THE IMPLIED  WARRANTIES OF    */
+/*    MERCHANTABILITY  AND FITNESS FOR  A PARTICULAR  PURPOSE ARE    */
+/*    DISCLAIMED.  IN  NO EVENT SHALL THE UNIVERSITY  OF TEXAS AT    */
+/*    AUSTIN OR CONTRIBUTORS BE  LIABLE FOR ANY DIRECT, INDIRECT,    */
+/*    INCIDENTAL,  SPECIAL, EXEMPLARY,  OR  CONSEQUENTIAL DAMAGES    */
+/*    (INCLUDING, BUT  NOT LIMITED TO,  PROCUREMENT OF SUBSTITUTE    */
+/*    GOODS  OR  SERVICES; LOSS  OF  USE,  DATA,  OR PROFITS;  OR    */
+/*    BUSINESS INTERRUPTION) HOWEVER CAUSED  AND ON ANY THEORY OF    */
+/*    LIABILITY, WHETHER  IN CONTRACT, STRICT  LIABILITY, OR TORT    */
+/*    (INCLUDING NEGLIGENCE OR OTHERWISE)  ARISING IN ANY WAY OUT    */
+/*    OF  THE  USE OF  THIS  SOFTWARE,  EVEN  IF ADVISED  OF  THE    */
+/*    POSSIBILITY OF SUCH DAMAGE.                                    */
+/*                                                                   */
+/* The views and conclusions contained in the software and           */
+/* documentation are those of the authors and should not be          */
+/* interpreted as representing official policies, either expressed   */
+/* or implied, of The University of Texas at Austin.                 */
+/*********************************************************************/
+
+#define ASSEMBLER
+#include "common.h"
+
+#define N	$4
+#define	X	$5
+#define INCX	$6
+
+#define I	$2
+#define TEMP	$3
+
+#define a1	$f2
+#define a2	$f3
+#define a3	$f4
+#define a4	$f5
+#define a5	$f6
+#define a6	$f7
+#define a7	$f8
+#define a8	$f9
+
+#define t1	$f10
+#define t2	$f11
+#define t3	$f12
+#define t4	$f13
+
+#define s1	$f0
+#define s2	$f1
+
+	PROLOGUE
+
+#ifdef F_INTERFACE
+	LDINT	N,     0(N)
+	LDINT	INCX,  0(INCX)
+#endif
+
+	MTC	$0,  s1
+
+	MTC	$0,  s2
+	dsll	INCX, INCX, BASE_SHIFT
+
+	blez	N, .L999
+	li	TEMP, SIZE
+
+	bne	INCX, TEMP, .L20
+	dsra	I, N, 3
+
+	blez	I, .L15
+	NOP
+
+	LD	a1,  0 * SIZE(X)
+	LD	a2,  1 * SIZE(X)
+	LD	a3,  2 * SIZE(X)
+	LD	a4,  3 * SIZE(X)
+
+	LD	a5,  4 * SIZE(X)
+	MOV	t1, a1
+	LD	a6,  5 * SIZE(X)
+	MOV	t2, a2
+	LD	a7,  6 * SIZE(X)
+	MOV	t3, a3
+
+	MOV	t4, a4
+	daddiu	I, I, -1
+
+	blez	I, .L13
+	LD	a8,  7 * SIZE(X)
+	.align 3
+
+.L12:
+	ADD	s1, s1, t1
+	LD	a1,  8 * SIZE(X)
+
+	MOV	t1, a5
+	daddiu	I, I, -1
+
+	ADD	s2, s2, t2
+	LD	a2,  9 * SIZE(X)
+
+	MOV	t2, a6
+	NOP
+
+	ADD	s1, s1, t3
+	LD	a3, 10 * SIZE(X)
+
+	MOV	t3, a7
+	NOP
+
+	ADD	s2, s2, t4
+	LD	a4, 11 * SIZE(X)
+
+	MOV	t4, a8
+	daddiu	X, X, 8 * SIZE
+
+	ADD	s1, s1, t1
+	LD	a5,  4 * SIZE(X)
+
+	MOV	t1, a1
+	NOP
+
+	ADD	s2, s2, t2
+	LD	a6,  5 * SIZE(X)
+
+	MOV	t2, a2
+	NOP
+
+	ADD	s1, s1, t3
+	LD	a7,  6 * SIZE(X)
+
+	MOV	t3, a3
+	NOP
+
+	ADD	s2, s2, t4
+	LD	a8,  7 * SIZE(X)
+
+	bgtz	I, .L12
+	MOV	t4, a4
+	.align 3
+
+.L13:
+	ADD	s1, s1, t1
+	daddiu	X, X, 8 * SIZE
+
+	MOV	t1, a5
+	NOP
+
+	ADD	s2, s2, t2
+	MOV	t2, a6
+
+	ADD	s1, s1, t3
+	MOV	t3, a7
+
+	ADD	s2, s2, t4
+	MOV	t4, a8
+
+	ADD	s1, s1, t1
+	ADD	s2, s2, t2
+	ADD	s1, s1, t3
+	ADD	s2, s2, t4
+	.align 3
+
+.L15:
+	andi	I,  N, 7
+
+	blez	I, .L999
+	NOP
+	.align	3
+
+.L16:
+	LD	a1,  0 * SIZE(X)
+	daddiu	I, I, -1
+
+	MOV	t1, a1
+
+	ADD	s1, s1, t1
+
+	bgtz	I, .L16
+	daddiu	X, X, SIZE
+
+	j	.L999
+	NOP
+	.align 3
+
+.L20:
+	blez	I, .L25
+	NOP
+
+	LD	a1,  0 * SIZE(X)
+	daddu	X, X, INCX
+
+	LD	a2,  0 * SIZE(X)
+	daddu	X, X, INCX
+
+	LD	a3,  0 * SIZE(X)
+	daddu	X, X, INCX
+
+	LD	a4,  0 * SIZE(X)
+	daddu	X, X, INCX
+
+	LD	a5,  0 * SIZE(X)
+	daddu	X, X, INCX
+
+	LD	a6,  0 * SIZE(X)
+	daddu	X, X, INCX
+
+	MOV	t1, a1
+	LD	a7,  0 * SIZE(X)
+
+	MOV	t2, a2
+	daddu	X, X, INCX
+
+	MOV	t3, a3
+	LD	a8,  0 * SIZE(X)
+
+	MOV	t4, a4
+	daddiu	I, I, -1
+
+	blez	I, .L24
+	daddu	X, X, INCX
+	.align 3
+
+.L23:
+	ADD	s1, s1, t1
+	LD	a1,  0 * SIZE(X)
+
+	MOV	t1, a5
+	daddu	X, X, INCX
+
+	ADD	s2, s2, t2
+	LD	a2,  0 * SIZE(X)
+
+	MOV	t2, a6
+	daddu	X, X, INCX
+
+	ADD	s1, s1, t3
+	LD	a3,  0 * SIZE(X)
+
+	MOV	t3, a7
+	daddu	X, X, INCX
+
+	ADD	s2, s2, t4
+	LD	a4,  0 * SIZE(X)
+
+	MOV	t4, a8
+	daddu	X, X, INCX
+
+	ADD	s1, s1, t1
+	LD	a5,  0 * SIZE(X)
+
+	MOV	t1, a1
+	daddu	X, X, INCX
+
+	ADD	s2, s2, t2
+	LD	a6,  0 * SIZE(X)
+
+	MOV	t2, a2
+	daddu	X, X, INCX
+
+	ADD	s1, s1, t3
+	LD	a7,  0 * SIZE(X)
+
+	MOV	t3, a3
+	daddu	X, X, INCX
+
+	ADD	s2, s2, t4
+	LD	a8,  0 * SIZE(X)
+
+	MOV	t4, a4
+	daddiu	I, I, -1
+
+	bgtz	I, .L23
+	daddu	X, X, INCX
+	.align 3
+
+.L24:
+	ADD	s1, s1, t1
+	MOV	t1, a5
+
+	ADD	s2, s2, t2
+	MOV	t2, a6
+
+	ADD	s1, s1, t3
+	MOV	t3, a7
+
+	ADD	s2, s2, t4
+	MOV	t4, a8
+
+	ADD	s1, s1, t1
+	ADD	s2, s2, t2
+	ADD	s1, s1, t3
+	ADD	s2, s2, t4
+	.align 3
+
+.L25:
+	andi	I,  N, 7
+
+	blez	I, .L999
+	NOP
+	.align	3
+
+.L26:
+	LD	a1,  0 * SIZE(X)
+	daddiu	I, I, -1
+
+	MOV	t1, a1
+	daddu	X, X, INCX
+
+	bgtz	I, .L26
+	ADD	s1, s1, t1
+	.align 3
+
+.L999:
+	j	$31
+	ADD	s1, s1, s2
+
+	EPILOGUE

kernel/mips64/zsum.S

@@ -0,0 +1,204 @@
+/*********************************************************************/
+/* Copyright 2009, 2010 The University of Texas at Austin.           */
+/* All rights reserved.                                              */
+/*                                                                   */
+/* Redistribution and use in source and binary forms, with or        */
+/* without modification, are permitted provided that the following   */
+/* conditions are met:                                               */
+/*                                                                   */
+/*   1. Redistributions of source code must retain the above         */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer.                                                  */
+/*                                                                   */
+/*   2. Redistributions in binary form must reproduce the above      */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer in the documentation and/or other materials       */
+/*      provided with the distribution.                              */
+/*                                                                   */
+/*    THIS  SOFTWARE IS PROVIDED  BY THE  UNIVERSITY OF  TEXAS AT    */
+/*    AUSTIN  ``AS IS''  AND ANY  EXPRESS OR  IMPLIED WARRANTIES,    */
+/*    INCLUDING, BUT  NOT LIMITED  TO, THE IMPLIED  WARRANTIES OF    */
+/*    MERCHANTABILITY  AND FITNESS FOR  A PARTICULAR  PURPOSE ARE    */
+/*    DISCLAIMED.  IN  NO EVENT SHALL THE UNIVERSITY  OF TEXAS AT    */
+/*    AUSTIN OR CONTRIBUTORS BE  LIABLE FOR ANY DIRECT, INDIRECT,    */
+/*    INCIDENTAL,  SPECIAL, EXEMPLARY,  OR  CONSEQUENTIAL DAMAGES    */
+/*    (INCLUDING, BUT  NOT LIMITED TO,  PROCUREMENT OF SUBSTITUTE    */
+/*    GOODS  OR  SERVICES; LOSS  OF  USE,  DATA,  OR PROFITS;  OR    */
+/*    BUSINESS INTERRUPTION) HOWEVER CAUSED  AND ON ANY THEORY OF    */
+/*    LIABILITY, WHETHER  IN CONTRACT, STRICT  LIABILITY, OR TORT    */
+/*    (INCLUDING NEGLIGENCE OR OTHERWISE)  ARISING IN ANY WAY OUT    */
+/*    OF  THE  USE OF  THIS  SOFTWARE,  EVEN  IF ADVISED  OF  THE    */
+/*    POSSIBILITY OF SUCH DAMAGE.                                    */
+/*                                                                   */
+/* The views and conclusions contained in the software and           */
+/* documentation are those of the authors and should not be          */
+/* interpreted as representing official policies, either expressed   */
+/* or implied, of The University of Texas at Austin.                 */
+/*********************************************************************/
+
+#define ASSEMBLER
+#include "common.h"
+
+#define N	$4
+#define	X	$5
+#define INCX	$6
+
+#define I	$2
+#define TEMP	$3
+
+#define a1	$f2
+#define a2	$f3
+#define a3	$f4
+#define a4	$f5
+#define a5	$f6
+#define a6	$f7
+#define a7	$f8
+#define a8	$f9
+
+#define t1	$f10
+#define t2	$f11
+#define t3	$f12
+#define t4	$f13
+
+#define s1	$f0
+#define s2	$f1
+
+	PROLOGUE
+
+#ifdef F_INTERFACE
+	LDINT	N,     0(N)
+	LDINT	INCX,  0(INCX)
+#endif
+
+	MTC	$0,  s1
+
+	MTC	$0,  s2
+	dsll	INCX, INCX, ZBASE_SHIFT
+
+	blez	N, .L999
+	dsra	I, N, 2
+
+	blez	I, .L25
+	NOP
+
+	LD	a1,  0 * SIZE(X)
+	LD	a2,  1 * SIZE(X)
+	daddu	X, X, INCX
+
+	LD	a3,  0 * SIZE(X)
+	LD	a4,  1 * SIZE(X)
+	daddu	X, X, INCX
+
+	LD	a5,  0 * SIZE(X)
+	LD	a6,  1 * SIZE(X)
+	daddu	X, X, INCX
+
+	MOV	t1, a1
+	MOV	t2, a2
+
+	LD	a7,  0 * SIZE(X)
+	LD	a8,  1 * SIZE(X)
+
+	MOV	t3, a3
+	MOV	t4, a4
+	daddiu	I, I, -1
+
+	blez	I, .L24
+	daddu	X, X, INCX
+	.align 3
+
+.L23:
+	ADD	s1, s1, t1
+	LD	a1,  0 * SIZE(X)
+
+	MOV	t1, a5
+	daddiu	I, I, -1
+
+	ADD	s2, s2, t2
+	LD	a2,  1 * SIZE(X)
+
+	MOV	t2, a6
+	daddu	X, X, INCX
+
+	ADD	s1, s1, t3
+	LD	a3,  0 * SIZE(X)
+
+	MOV	t3, a7
+	NOP
+
+	ADD	s2, s2, t4
+	LD	a4,  1 * SIZE(X)
+
+	MOV	t4, a8
+	daddu	X, X, INCX
+
+	ADD	s1, s1, t1
+	LD	a5,  0 * SIZE(X)
+
+	MOV	t1, a1
+	NOP
+
+	ADD	s2, s2, t2
+	LD	a6,  1 * SIZE(X)
+
+	MOV	t2, a2
+	daddu	X, X, INCX
+
+	ADD	s1, s1, t3
+	LD	a7,  0 * SIZE(X)
+
+	MOV	t3, a3
+	LD	a8,  1 * SIZE(X)
+
+	ADD	s2, s2, t4
+	daddu	X, X, INCX
+
+	bgtz	I, .L23
+	MOV	t4, a4
+	.align 3
+
+.L24:
+	ADD	s1, s1, t1
+	MOV	t1, a5
+
+	ADD	s2, s2, t2
+	MOV	t2, a6
+
+	ADD	s1, s1, t3
+	MOV	t3, a7
+
+	ADD	s2, s2, t4
+	MOV	t4, a8
+
+	ADD	s1, s1, t1
+	ADD	s2, s2, t2
+	ADD	s1, s1, t3
+	ADD	s2, s2, t4
+	.align 3
+
+.L25:
+	andi	I,  N, 3
+
+	blez	I, .L999
+	NOP
+	.align	3
+
+.L26:
+	LD	a1,  0 * SIZE(X)
+	LD	a2,  1 * SIZE(X)
+
+	MOV	t1, a1
+	daddiu	I, I, -1
+	MOV	t2, a2
+	daddu	X, X, INCX
+
+	ADD	s1, s1, t1
+	bgtz	I, .L26
+	ADD	s2, s2, t2
+	.align 3
+
+.L999:
+	j	$31
+	ADD	s1, s1, s2
+
+	EPILOGUE

kernel/power/sum.S

@@ -0,0 +1,446 @@
+/*********************************************************************/
+/* Copyright 2009, 2010 The University of Texas at Austin.           */
+/* All rights reserved.                                              */
+/*                                                                   */
+/* Redistribution and use in source and binary forms, with or        */
+/* without modification, are permitted provided that the following   */
+/* conditions are met:                                               */
+/*                                                                   */
+/*   1. Redistributions of source code must retain the above         */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer.                                                  */
+/*                                                                   */
+/*   2. Redistributions in binary form must reproduce the above      */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer in the documentation and/or other materials       */
+/*      provided with the distribution.                              */
+/*                                                                   */
+/*    THIS  SOFTWARE IS PROVIDED  BY THE  UNIVERSITY OF  TEXAS AT    */
+/*    AUSTIN  ``AS IS''  AND ANY  EXPRESS OR  IMPLIED WARRANTIES,    */
+/*    INCLUDING, BUT  NOT LIMITED  TO, THE IMPLIED  WARRANTIES OF    */
+/*    MERCHANTABILITY  AND FITNESS FOR  A PARTICULAR  PURPOSE ARE    */
+/*    DISCLAIMED.  IN  NO EVENT SHALL THE UNIVERSITY  OF TEXAS AT    */
+/*    AUSTIN OR CONTRIBUTORS BE  LIABLE FOR ANY DIRECT, INDIRECT,    */
+/*    INCIDENTAL,  SPECIAL, EXEMPLARY,  OR  CONSEQUENTIAL DAMAGES    */
+/*    (INCLUDING, BUT  NOT LIMITED TO,  PROCUREMENT OF SUBSTITUTE    */
+/*    GOODS  OR  SERVICES; LOSS  OF  USE,  DATA,  OR PROFITS;  OR    */
+/*    BUSINESS INTERRUPTION) HOWEVER CAUSED  AND ON ANY THEORY OF    */
+/*    LIABILITY, WHETHER  IN CONTRACT, STRICT  LIABILITY, OR TORT    */
+/*    (INCLUDING NEGLIGENCE OR OTHERWISE)  ARISING IN ANY WAY OUT    */
+/*    OF  THE  USE OF  THIS  SOFTWARE,  EVEN  IF ADVISED  OF  THE    */
+/*    POSSIBILITY OF SUCH DAMAGE.                                    */
+/*                                                                   */
+/* The views and conclusions contained in the software and           */
+/* documentation are those of the authors and should not be          */
+/* interpreted as representing official policies, either expressed   */
+/* or implied, of The University of Texas at Austin.                 */
+/*********************************************************************/
+
+#define ASSEMBLER
+#include "common.h"
+
+#define N	r3
+#define X	r4
+#define INCX	r5
+
+#define PREA	r8
+
+#define FZERO	f0
+
+#define STACKSIZE 160
+
+	PROLOGUE
+	PROFCODE
+
+	addi	SP, SP, -STACKSIZE
+	li	r0,   0
+
+	stfd	f14,    0(SP)
+	stfd	f15,    8(SP)
+	stfd	f16,   16(SP)
+	stfd	f17,   24(SP)
+
+	stfd	f18,   32(SP)
+	stfd	f19,   40(SP)
+	stfd	f20,   48(SP)
+	stfd	f21,   56(SP)
+
+	stfd	f22,   64(SP)
+	stfd	f23,   72(SP)
+	stfd	f24,   80(SP)
+	stfd	f25,   88(SP)
+
+	stfd	f26,   96(SP)
+	stfd	f27,  104(SP)
+	stfd	f28,  112(SP)
+	stfd	f29,  120(SP)
+
+	stfd	f30,  128(SP)
+	stfd	f31,  136(SP)
+
+	stw	r0,   144(SP)
+	lfs	FZERO,144(SP)
+
+#ifdef F_INTERFACE
+	LDINT	N,    0(N)
+	LDINT	INCX, 0(INCX)
+#endif
+
+	slwi	INCX, INCX, BASE_SHIFT
+
+	fmr	f1,  FZERO
+	fmr	f2,  FZERO
+	fmr	f3,  FZERO
+	fmr	f4,  FZERO
+	fmr	f5,  FZERO
+	fmr	f6,  FZERO
+	fmr	f7,  FZERO
+
+	li	PREA, L1_PREFETCHSIZE
+
+	cmpwi	cr0, N, 0
+	ble-	LL(999)
+
+	cmpwi	cr0, INCX, 0
+	ble-	LL(999)
+
+	cmpwi	cr0, INCX, SIZE
+	bne-	cr0, LL(100)
+
+	srawi.	r0, N, 4
+	mtspr	CTR, r0
+	beq-	cr0, LL(50)
+	.align 4
+
+	LFD	f8,    0 * SIZE(X)
+	LFD	f9,    1 * SIZE(X)
+	LFD	f10,   2 * SIZE(X)
+	LFD	f11,   3 * SIZE(X)
+	LFD	f12,   4 * SIZE(X)
+	LFD	f13,   5 * SIZE(X)
+	LFD	f14,   6 * SIZE(X)
+	LFD	f15,   7 * SIZE(X)
+
+	LFD	f24,   8 * SIZE(X)
+	LFD	f25,   9 * SIZE(X)
+	LFD	f26,  10 * SIZE(X)
+	LFD	f27,  11 * SIZE(X)
+	LFD	f28,  12 * SIZE(X)
+	LFD	f29,  13 * SIZE(X)
+	LFD	f30,  14 * SIZE(X)
+	LFD	f31,  15 * SIZE(X)
+
+	fmr	f16, f8
+	fmr	f17, f9
+	fmr	f18, f10
+	fmr	f19, f11
+
+	fmr	f20, f12
+	fmr	f21, f13
+	fmr	f22, f14
+	fmr	f23, f15
+	bdz	LL(20)
+	.align 4
+
+LL(10):
+	FADD	f0, f0, f16
+	fmr	f16, f24
+	FADD	f1, f1, f17
+	fmr	f17, f25
+
+	FADD	f2, f2, f18
+	fmr	f18, f26
+	FADD	f3, f3, f19
+	fmr	f19, f27
+
+	LFD	f8,   16 * SIZE(X)
+	LFD	f9,   17 * SIZE(X)
+	LFD	f10,  18 * SIZE(X)
+	LFD	f11,  19 * SIZE(X)
+
+	FADD	f4, f4, f20
+	fmr	f20, f28
+	FADD	f5, f5, f21
+	fmr	f21, f29
+
+	FADD	f6, f6, f22
+	fmr	f22, f30
+	FADD	f7, f7, f23
+	fmr	f23, f31
+
+	LFD	f12,  20 * SIZE(X)
+	LFD	f13,  21 * SIZE(X)
+	LFD	f14,  22 * SIZE(X)
+	LFD	f15,  23 * SIZE(X)
+
+	FADD	f0, f0, f16
+	fmr	f16, f8
+	FADD	f1, f1, f17
+	fmr	f17, f9
+
+	FADD	f2, f2, f18
+	fmr	f18, f10
+	FADD	f3, f3, f19
+	fmr	f19, f11
+
+	LFD	f24,  24 * SIZE(X)
+	LFD	f25,  25 * SIZE(X)
+	LFD	f26,  26 * SIZE(X)
+	LFD	f27,  27 * SIZE(X)
+
+	FADD	f4, f4, f20
+	fmr	f20, f12
+	FADD	f5, f5, f21
+	fmr	f21, f13
+
+	FADD	f6, f6, f22
+	fmr	f22, f14
+	FADD	f7, f7, f23
+	fmr	f23, f15
+
+	LFD	f28,  28 * SIZE(X)
+	LFD	f29,  29 * SIZE(X)
+	LFD	f30,  30 * SIZE(X)
+	LFD	f31,  31 * SIZE(X)
+
+#ifndef POWER6
+	L1_PREFETCH	X, PREA
+#endif
+	addi	X, X, 16 * SIZE
+#ifdef POWER6
+	L1_PREFETCH	X, PREA
+#endif
+
+	bdnz	LL(10)
+	.align 4
+
+LL(20):
+	FADD	f0, f0, f16
+	fmr	f16, f24
+	FADD	f1, f1, f17
+	fmr	f17, f25
+
+	FADD	f2, f2, f18
+	fmr	f18, f26
+	FADD	f3, f3, f19
+	fmr	f19, f27
+
+	FADD	f4, f4, f20
+	fmr	f20, f28
+	FADD	f5, f5, f21
+	fmr	f21, f29
+
+	FADD	f6, f6, f22
+	fmr	f22, f30
+	FADD	f7, f7, f23
+	fmr	f23, f31
+
+	FADD	f0, f0, f16
+	FADD	f1, f1, f17
+	FADD	f2, f2, f18
+	FADD	f3, f3, f19
+
+	FADD	f4, f4, f20
+	FADD	f5, f5, f21
+	FADD	f6, f6, f22
+	FADD	f7, f7, f23
+	addi	X, X, 16 * SIZE
+	.align 4
+
+LL(50):
+	andi.	r0,  N, 15
+	mtspr	CTR, r0
+	beq	LL(999)
+	.align 4
+
+LL(60):
+	LFD	f8,  0 * SIZE(X)
+	addi	X, X,  1 * SIZE
+
+	FADD	f0, f0,  f8
+
+	bdnz	LL(60)
+	b	LL(999)
+	.align 4
+
+LL(100):
+	sub	X, X, INCX
+
+	srawi.	r0, N, 4
+	mtspr	CTR,  r0
+	beq-	LL(150)
+
+	LFDUX	f8,    X, INCX
+	LFDUX	f9,    X, INCX
+	LFDUX	f10,   X, INCX
+	LFDUX	f11,   X, INCX
+	LFDUX	f12,   X, INCX
+	LFDUX	f13,   X, INCX
+	LFDUX	f14,   X, INCX
+	LFDUX	f15,   X, INCX
+
+	LFDUX	f24,   X, INCX
+	LFDUX	f25,   X, INCX
+	LFDUX	f26,   X, INCX
+	LFDUX	f27,   X, INCX
+	LFDUX	f28,   X, INCX
+	LFDUX	f29,   X, INCX
+	LFDUX	f30,   X, INCX
+	LFDUX	f31,   X, INCX
+
+	fmr	f16, f8
+	fmr	f17, f9
+	fmr	f18, f10
+	fmr	f19, f11
+
+	fmr	f20, f12
+	fmr	f21, f13
+	fmr	f22, f14
+	fmr	f23, f15
+	bdz	LL(120)
+	.align 4
+
+LL(110):
+	FADD	f0, f0, f16
+	fmr	f16, f24
+	FADD	f1, f1, f17
+	fmr	f17, f25
+
+	FADD	f2, f2, f18
+	fmr	f18, f26
+	FADD	f3, f3, f19
+	fmr	f19, f27
+
+	LFDUX	f8,    X, INCX
+	LFDUX	f9,    X, INCX
+	LFDUX	f10,   X, INCX
+	LFDUX	f11,   X, INCX
+
+	FADD	f4, f4, f20
+	fmr	f20, f28
+	FADD	f5, f5, f21
+	fmr	f21, f29
+
+	FADD	f6, f6, f22
+	fmr	f22, f30
+	FADD	f7, f7, f23
+	fmr	f23, f31
+
+	LFDUX	f12,   X, INCX
+	LFDUX	f13,   X, INCX
+	LFDUX	f14,   X, INCX
+	LFDUX	f15,   X, INCX
+
+	FADD	f0, f0, f16
+	fmr	f16, f8
+	FADD	f1, f1, f17
+	fmr	f17, f9
+
+	FADD	f2, f2, f18
+	fmr	f18, f10
+	FADD	f3, f3, f19
+	fmr	f19, f11
+
+	LFDUX	f24,   X, INCX
+	LFDUX	f25,   X, INCX
+	LFDUX	f26,   X, INCX
+	LFDUX	f27,   X, INCX
+
+	FADD	f4, f4, f20
+	fmr	f20, f12
+	FADD	f5, f5, f21
+	fmr	f21, f13
+
+	FADD	f6, f6, f22
+	fmr	f22, f14
+	FADD	f7, f7, f23
+	fmr	f23, f15
+
+	LFDUX	f28,   X, INCX
+	LFDUX	f29,   X, INCX
+	LFDUX	f30,   X, INCX
+	LFDUX	f31,   X, INCX
+	bdnz	LL(110)
+	.align 4
+
+LL(120):
+	FADD	f0, f0, f16
+	fmr	f16, f24
+	FADD	f1, f1, f17
+	fmr	f17, f25
+
+	FADD	f2, f2, f18
+	fmr	f18, f26
+	FADD	f3, f3, f19
+	fmr	f19, f27
+
+	FADD	f4, f4, f20
+	fmr	f20, f28
+	FADD	f5, f5, f21
+	fmr	f21, f29
+
+	FADD	f6, f6, f22
+	fmr	f22, f30
+	FADD	f7, f7, f23
+	fmr	f23, f31
+
+	FADD	f0, f0, f16
+	FADD	f1, f1, f17
+	FADD	f2, f2, f18
+	FADD	f3, f3, f19
+
+	FADD	f4, f4, f20
+	FADD	f5, f5, f21
+	FADD	f6, f6, f22
+	FADD	f7, f7, f23
+	.align 4
+
+LL(150):
+	andi.	r0,  N, 15
+	mtspr	CTR, r0
+	beq	LL(999)
+	.align 4
+
+LL(160):
+	LFDUX	f8,    X, INCX
+	FADD	f0,  f0, f8
+	bdnz	LL(160)
+	.align 4
+
+LL(999):
+	FADD	f0,  f0,  f1
+	FADD	f2,  f2,  f3
+	FADD	f4,  f4,  f5
+	FADD	f6,  f6,  f7
+
+	FADD	f0,  f0,  f2
+	FADD	f4,  f4,  f6
+	FADD	f1,  f0,  f4
+
+	lfd	f14,    0(SP)
+	lfd	f15,    8(SP)
+	lfd	f16,   16(SP)
+	lfd	f17,   24(SP)
+
+	lfd	f18,   32(SP)
+	lfd	f19,   40(SP)
+	lfd	f20,   48(SP)
+	lfd	f21,   56(SP)
+
+	lfd	f22,   64(SP)
+	lfd	f23,   72(SP)
+	lfd	f24,   80(SP)
+	lfd	f25,   88(SP)
+
+	lfd	f26,   96(SP)
+	lfd	f27,  104(SP)
+	lfd	f28,  112(SP)
+	lfd	f29,  120(SP)
+
+	lfd	f30,  128(SP)
+	lfd	f31,  136(SP)
+
+	addi	SP, SP, STACKSIZE
+	blr
+
+	EPILOGUE

kernel/power/zsum.S

@@ -0,0 +1,452 @@
+/*********************************************************************/
+/* Copyright 2009, 2010 The University of Texas at Austin.           */
+/* All rights reserved.                                              */
+/*                                                                   */
+/* Redistribution and use in source and binary forms, with or        */
+/* without modification, are permitted provided that the following   */
+/* conditions are met:                                               */
+/*                                                                   */
+/*   1. Redistributions of source code must retain the above         */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer.                                                  */
+/*                                                                   */
+/*   2. Redistributions in binary form must reproduce the above      */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer in the documentation and/or other materials       */
+/*      provided with the distribution.                              */
+/*                                                                   */
+/*    THIS  SOFTWARE IS PROVIDED  BY THE  UNIVERSITY OF  TEXAS AT    */
+/*    AUSTIN  ``AS IS''  AND ANY  EXPRESS OR  IMPLIED WARRANTIES,    */
+/*    INCLUDING, BUT  NOT LIMITED  TO, THE IMPLIED  WARRANTIES OF    */
+/*    MERCHANTABILITY  AND FITNESS FOR  A PARTICULAR  PURPOSE ARE    */
+/*    DISCLAIMED.  IN  NO EVENT SHALL THE UNIVERSITY  OF TEXAS AT    */
+/*    AUSTIN OR CONTRIBUTORS BE  LIABLE FOR ANY DIRECT, INDIRECT,    */
+/*    INCIDENTAL,  SPECIAL, EXEMPLARY,  OR  CONSEQUENTIAL DAMAGES    */
+/*    (INCLUDING, BUT  NOT LIMITED TO,  PROCUREMENT OF SUBSTITUTE    */
+/*    GOODS  OR  SERVICES; LOSS  OF  USE,  DATA,  OR PROFITS;  OR    */
+/*    BUSINESS INTERRUPTION) HOWEVER CAUSED  AND ON ANY THEORY OF    */
+/*    LIABILITY, WHETHER  IN CONTRACT, STRICT  LIABILITY, OR TORT    */
+/*    (INCLUDING NEGLIGENCE OR OTHERWISE)  ARISING IN ANY WAY OUT    */
+/*    OF  THE  USE OF  THIS  SOFTWARE,  EVEN  IF ADVISED  OF  THE    */
+/*    POSSIBILITY OF SUCH DAMAGE.                                    */
+/*                                                                   */
+/* The views and conclusions contained in the software and           */
+/* documentation are those of the authors and should not be          */
+/* interpreted as representing official policies, either expressed   */
+/* or implied, of The University of Texas at Austin.                 */
+/*********************************************************************/
+
+#define ASSEMBLER
+#include "common.h"
+
+#define N	r3
+#define X	r4
+#define INCX	r5
+
+#define INCXM1	r9
+#define PREA	r8
+
+#define FZERO	f0
+
+#define STACKSIZE 160
+
+	PROLOGUE
+	PROFCODE
+
+	addi	SP, SP, -STACKSIZE
+	li	r0,   0
+
+	stfd	f14,    0(SP)
+	stfd	f15,    8(SP)
+	stfd	f16,   16(SP)
+	stfd	f17,   24(SP)
+
+	stfd	f18,   32(SP)
+	stfd	f19,   40(SP)
+	stfd	f20,   48(SP)
+	stfd	f21,   56(SP)
+
+	stfd	f22,   64(SP)
+	stfd	f23,   72(SP)
+	stfd	f24,   80(SP)
+	stfd	f25,   88(SP)
+
+	stfd	f26,   96(SP)
+	stfd	f27,  104(SP)
+	stfd	f28,  112(SP)
+	stfd	f29,  120(SP)
+
+	stfd	f30,  128(SP)
+	stfd	f31,  136(SP)
+
+	stw	r0,   144(SP)
+	lfs	FZERO,144(SP)
+
+#ifdef F_INTERFACE
+	LDINT	N,    0(N)
+	LDINT	INCX, 0(INCX)
+#endif
+
+	slwi	INCX, INCX, ZBASE_SHIFT
+	subi	INCXM1, INCX, SIZE
+
+	fmr	f1,  FZERO
+	fmr	f2,  FZERO
+	fmr	f3,  FZERO
+	fmr	f4,  FZERO
+	fmr	f5,  FZERO
+	fmr	f6,  FZERO
+	fmr	f7,  FZERO
+
+	li	PREA, L1_PREFETCHSIZE
+
+	cmpwi	cr0, N, 0
+	ble-	LL(999)
+
+	cmpwi	cr0, INCX, 0
+	ble-	LL(999)
+
+	cmpwi	cr0, INCX, 2 * SIZE
+	bne-	cr0, LL(100)
+
+	srawi.	r0, N, 3
+	mtspr	CTR, r0
+	beq-	cr0, LL(50)
+	.align 4
+
+	LFD	f8,    0 * SIZE(X)
+	LFD	f9,    1 * SIZE(X)
+	LFD	f10,   2 * SIZE(X)
+	LFD	f11,   3 * SIZE(X)
+	LFD	f12,   4 * SIZE(X)
+	LFD	f13,   5 * SIZE(X)
+	LFD	f14,   6 * SIZE(X)
+	LFD	f15,   7 * SIZE(X)
+
+	LFD	f24,   8 * SIZE(X)
+	LFD	f25,   9 * SIZE(X)
+	LFD	f26,  10 * SIZE(X)
+	LFD	f27,  11 * SIZE(X)
+	LFD	f28,  12 * SIZE(X)
+	LFD	f29,  13 * SIZE(X)
+	LFD	f30,  14 * SIZE(X)
+	LFD	f31,  15 * SIZE(X)
+
+	fmr	f16, f8
+	fmr	f17, f9
+	fmr	f18, f10
+	fmr	f19, f11
+
+	fmr	f20, f12
+	fmr	f21, f13
+	fmr	f22, f14
+	fmr	f23, f15
+	bdz	LL(20)
+	.align 4
+
+LL(10):
+	FADD	f0, f0, f16
+	fmr	f16, f24
+	FADD	f1, f1, f17
+	fmr	f17, f25
+
+	FADD	f2, f2, f18
+	fmr	f18, f26
+	FADD	f3, f3, f19
+	fmr	f19, f27
+
+	LFD	f8,   16 * SIZE(X)
+	LFD	f9,   17 * SIZE(X)
+	LFD	f10,  18 * SIZE(X)
+	LFD	f11,  19 * SIZE(X)
+
+	FADD	f4, f4, f20
+	fmr	f20, f28
+	FADD	f5, f5, f21
+	fmr	f21, f29
+
+	FADD	f6, f6, f22
+	fmr	f22, f30
+	FADD	f7, f7, f23
+	fmr	f23, f31
+
+	LFD	f12,  20 * SIZE(X)
+	LFD	f13,  21 * SIZE(X)
+	LFD	f14,  22 * SIZE(X)
+	LFD	f15,  23 * SIZE(X)
+
+	FADD	f0, f0, f16
+	fmr	f16, f8
+	FADD	f1, f1, f17
+	fmr	f17, f9
+
+	FADD	f2, f2, f18
+	fmr	f18, f10
+	FADD	f3, f3, f19
+	fmr	f19, f11
+
+	LFD	f24,  24 * SIZE(X)
+	LFD	f25,  25 * SIZE(X)
+	LFD	f26,  26 * SIZE(X)
+	LFD	f27,  27 * SIZE(X)
+
+	FADD	f4, f4, f20
+	fmr	f20, f12
+	FADD	f5, f5, f21
+	fmr	f21, f13
+
+	FADD	f6, f6, f22
+	fmr	f22, f14
+	FADD	f7, f7, f23
+	fmr	f23, f15
+
+	LFD	f28,  28 * SIZE(X)
+	LFD	f29,  29 * SIZE(X)
+	LFD	f30,  30 * SIZE(X)
+	LFD	f31,  31 * SIZE(X)
+
+#ifndef POWER6
+	L1_PREFETCH	X, PREA
+#endif
+	addi	X, X, 16 * SIZE
+#ifdef POWER6
+	L1_PREFETCH	X, PREA
+#endif
+
+	bdnz	LL(10)
+	.align 4
+
+LL(20):
+	FADD	f0, f0, f16
+	fmr	f16, f24
+	FADD	f1, f1, f17
+	fmr	f17, f25
+
+	FADD	f2, f2, f18
+	fmr	f18, f26
+	FADD	f3, f3, f19
+	fmr	f19, f27
+
+	FADD	f4, f4, f20
+	fmr	f20, f28
+	FADD	f5, f5, f21
+	fmr	f21, f29
+
+	FADD	f6, f6, f22
+	fmr	f22, f30
+	FADD	f7, f7, f23
+	fmr	f23, f31
+
+	FADD	f0, f0, f16
+	FADD	f1, f1, f17
+	FADD	f2, f2, f18
+	FADD	f3, f3, f19
+
+	FADD	f4, f4, f20
+	FADD	f5, f5, f21
+	FADD	f6, f6, f22
+	FADD	f7, f7, f23
+	addi	X, X, 16 * SIZE
+	.align 4
+
+LL(50):
+	andi.	r0,  N, 7
+	mtspr	CTR, r0
+	beq	LL(999)
+	.align 4
+
+LL(60):
+	LFD	f8,  0 * SIZE(X)
+	LFD	f9,  1 * SIZE(X)
+	addi	X, X,  2 * SIZE
+
+	FADD	f0, f0,  f8
+	FADD	f1, f1,  f9
+
+	bdnz	LL(60)
+	b	LL(999)
+	.align 4
+
+LL(100):
+	sub	X, X, INCXM1
+
+	srawi.	r0, N, 3
+	mtspr	CTR,  r0
+	beq-	LL(150)
+
+	LFDX	f8,    X, INCXM1
+	LFDUX	f9,    X, INCX
+	LFDX	f10,   X, INCXM1
+	LFDUX	f11,   X, INCX
+	LFDX	f12,   X, INCXM1
+	LFDUX	f13,   X, INCX
+	LFDX	f14,   X, INCXM1
+	LFDUX	f15,   X, INCX
+
+	LFDX	f24,   X, INCXM1
+	LFDUX	f25,   X, INCX
+	LFDX	f26,   X, INCXM1
+	LFDUX	f27,   X, INCX
+	LFDX	f28,   X, INCXM1
+	LFDUX	f29,   X, INCX
+	LFDX	f30,   X, INCXM1
+	LFDUX	f31,   X, INCX
+
+	fmr	f16, f8
+	fmr	f17, f9
+	fmr	f18, f10
+	fmr	f19, f11
+
+	fmr	f20, f12
+	fmr	f21, f13
+	fmr	f22, f14
+	fmr	f23, f15
+	bdz	LL(120)
+	.align 4
+
+LL(110):
+	FADD	f0, f0, f16
+	fmr	f16, f24
+	FADD	f1, f1, f17
+	fmr	f17, f25
+
+	FADD	f2, f2, f18
+	fmr	f18, f26
+	FADD	f3, f3, f19
+	fmr	f19, f27
+
+	LFDX	f8,    X, INCXM1
+	LFDUX	f9,    X, INCX
+	LFDX	f10,   X, INCXM1
+	LFDUX	f11,   X, INCX
+
+	FADD	f4, f4, f20
+	fmr	f20, f28
+	FADD	f5, f5, f21
+	fmr	f21, f29
+
+	FADD	f6, f6, f22
+	fmr	f22, f30
+	FADD	f7, f7, f23
+	fmr	f23, f31
+
+	LFDX	f12,   X, INCXM1
+	LFDUX	f13,   X, INCX
+	LFDX	f14,   X, INCXM1
+	LFDUX	f15,   X, INCX
+
+	FADD	f0, f0, f16
+	fmr	f16, f8
+	FADD	f1, f1, f17
+	fmr	f17, f9
+
+	FADD	f2, f2, f18
+	fmr	f18, f10
+	FADD	f3, f3, f19
+	fmr	f19, f11
+
+	LFDX	f24,   X, INCXM1
+	LFDUX	f25,   X, INCX
+	LFDX	f26,   X, INCXM1
+	LFDUX	f27,   X, INCX
+
+	FADD	f4, f4, f20
+	fmr	f20, f12
+	FADD	f5, f5, f21
+	fmr	f21, f13
+
+	FADD	f6, f6, f22
+	fmr	f22, f14
+	FADD	f7, f7, f23
+	fmr	f23, f15
+
+	LFDX	f28,   X, INCXM1
+	LFDUX	f29,   X, INCX
+	LFDX	f30,   X, INCXM1
+	LFDUX	f31,   X, INCX
+	bdnz	LL(110)
+	.align 4
+
+LL(120):
+	FADD	f0, f0, f16
+	fmr	f16, f24
+	FADD	f1, f1, f17
+	fmr	f17, f25
+
+	FADD	f2, f2, f18
+	fmr	f18, f26
+	FADD	f3, f3, f19
+	fmr	f19, f27
+
+	FADD	f4, f4, f20
+	fmr	f20, f28
+	FADD	f5, f5, f21
+	fmr	f21, f29
+
+	FADD	f6, f6, f22
+	fmr	f22, f30
+	FADD	f7, f7, f23
+	fmr	f23, f31
+
+	FADD	f0, f0, f16
+	FADD	f1, f1, f17
+	FADD	f2, f2, f18
+	FADD	f3, f3, f19
+
+	FADD	f4, f4, f20
+	FADD	f5, f5, f21
+	FADD	f6, f6, f22
+	FADD	f7, f7, f23
+	.align 4
+
+LL(150):
+	andi.	r0,  N, 7
+	mtspr	CTR, r0
+	beq	LL(999)
+	.align 4
+
+LL(160):
+	LFDX	f8,    X, INCXM1
+	LFDUX	f9,    X, INCX
+	FADD	f0,  f0, f8
+	FADD	f1,  f1, f9
+	bdnz	LL(160)
+	.align 4
+
+LL(999):
+	FADD	f0,  f0,  f1
+	FADD	f2,  f2,  f3
+	FADD	f4,  f4,  f5
+	FADD	f6,  f6,  f7
+
+	FADD	f0,  f0,  f2
+	FADD	f4,  f4,  f6
+	FADD	f1,  f0,  f4
+
+	lfd	f14,    0(SP)
+	lfd	f15,    8(SP)
+	lfd	f16,   16(SP)
+	lfd	f17,   24(SP)
+
+	lfd	f18,   32(SP)
+	lfd	f19,   40(SP)
+	lfd	f20,   48(SP)
+	lfd	f21,   56(SP)
+
+	lfd	f22,   64(SP)
+	lfd	f23,   72(SP)
+	lfd	f24,   80(SP)
+	lfd	f25,   88(SP)
+
+	lfd	f26,   96(SP)
+	lfd	f27,  104(SP)
+	lfd	f28,  112(SP)
+	lfd	f29,  120(SP)
+
+	lfd	f30,  128(SP)
+	lfd	f31,  136(SP)
+
+	addi	SP, SP, STACKSIZE
+	blr
+
+	EPILOGUE

kernel/setparam-ref.c

@@ -70,7 +70,7 @@ gotoblas_t TABLE_NAME = {
 
   samax_kTS,  samin_kTS,  smax_kTS,  smin_kTS,
   isamax_kTS, isamin_kTS, ismax_kTS, ismin_kTS,
-  snrm2_kTS,  sasum_kTS,  scopy_kTS, sdot_kTS,
+  snrm2_kTS,  sasum_kTS, ssum_kTS, scopy_kTS, sdot_kTS,
   dsdot_kTS,
   srot_kTS,   saxpy_kTS,  sscal_kTS, sswap_kTS,
   sgemv_nTS,  sgemv_tTS, sger_kTS,
@@ -126,7 +126,7 @@ gotoblas_t TABLE_NAME = {
 
   damax_kTS,  damin_kTS,  dmax_kTS,  dmin_kTS,
   idamax_kTS, idamin_kTS, idmax_kTS, idmin_kTS,
-  dnrm2_kTS,  dasum_kTS,  dcopy_kTS, ddot_kTS,
+  dnrm2_kTS,  dasum_kTS,  dsum_kTS, dcopy_kTS, ddot_kTS,
   drot_kTS,   daxpy_kTS,  dscal_kTS, dswap_kTS,
   dgemv_nTS,  dgemv_tTS,  dger_kTS,
   dsymv_LTS,  dsymv_UTS,
@@ -178,7 +178,7 @@ gotoblas_t TABLE_NAME = {
 
   qamax_kTS,  qamin_kTS,  qmax_kTS,  qmin_kTS,
   iqamax_kTS, iqamin_kTS, iqmax_kTS, iqmin_kTS,
-  qnrm2_kTS,  qasum_kTS,  qcopy_kTS, qdot_kTS,
+  qnrm2_kTS,  qasum_kTS,  qsum_kTS, qcopy_kTS, qdot_kTS,
   qrot_kTS,   qaxpy_kTS,  qscal_kTS, qswap_kTS,
   qgemv_nTS,  qgemv_tTS,  qger_kTS,
   qsymv_LTS,  qsymv_UTS,
@@ -234,7 +234,7 @@ gotoblas_t TABLE_NAME = {
 #endif
 
   camax_kTS, camin_kTS, icamax_kTS, icamin_kTS,
-  cnrm2_kTS, casum_kTS, ccopy_kTS,
+  cnrm2_kTS, casum_kTS, csum_kTS, ccopy_kTS,
   cdotu_kTS, cdotc_kTS, csrot_kTS,
   caxpy_kTS, caxpyc_kTS, cscal_kTS, cswap_kTS,
 
@@ -369,7 +369,7 @@ gotoblas_t TABLE_NAME = {
 #endif
 
   zamax_kTS, zamin_kTS, izamax_kTS, izamin_kTS,
-  znrm2_kTS, zasum_kTS, zcopy_kTS,
+  znrm2_kTS, zasum_kTS, zsum_kTS, zcopy_kTS,
   zdotu_kTS, zdotc_kTS, zdrot_kTS,
   zaxpy_kTS, zaxpyc_kTS, zscal_kTS, zswap_kTS,
 
@@ -500,7 +500,7 @@ gotoblas_t TABLE_NAME = {
   XGEMM_DEFAULT_UNROLL_M, XGEMM_DEFAULT_UNROLL_N, MAX(XGEMM_DEFAULT_UNROLL_M, XGEMM_DEFAULT_UNROLL_N),
 
   xamax_kTS, xamin_kTS, ixamax_kTS, ixamin_kTS,
-  xnrm2_kTS, xasum_kTS, xcopy_kTS,
+  xnrm2_kTS, xasum_kTS, xsum_kTS, xcopy_kTS,
   xdotu_kTS, xdotc_kTS, xqrot_kTS,
   xaxpy_kTS, xaxpyc_kTS, xscal_kTS, xswap_kTS,
 

kernel/sparc/sum.S

@@ -0,0 +1,325 @@
+/*********************************************************************/
+/* Copyright 2009, 2010 The University of Texas at Austin.           */
+/* All rights reserved.                                              */
+/*                                                                   */
+/* Redistribution and use in source and binary forms, with or        */
+/* without modification, are permitted provided that the following   */
+/* conditions are met:                                               */
+/*                                                                   */
+/*   1. Redistributions of source code must retain the above         */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer.                                                  */
+/*                                                                   */
+/*   2. Redistributions in binary form must reproduce the above      */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer in the documentation and/or other materials       */
+/*      provided with the distribution.                              */
+/*                                                                   */
+/*    THIS  SOFTWARE IS PROVIDED  BY THE  UNIVERSITY OF  TEXAS AT    */
+/*    AUSTIN  ``AS IS''  AND ANY  EXPRESS OR  IMPLIED WARRANTIES,    */
+/*    INCLUDING, BUT  NOT LIMITED  TO, THE IMPLIED  WARRANTIES OF    */
+/*    MERCHANTABILITY  AND FITNESS FOR  A PARTICULAR  PURPOSE ARE    */
+/*    DISCLAIMED.  IN  NO EVENT SHALL THE UNIVERSITY  OF TEXAS AT    */
+/*    AUSTIN OR CONTRIBUTORS BE  LIABLE FOR ANY DIRECT, INDIRECT,    */
+/*    INCIDENTAL,  SPECIAL, EXEMPLARY,  OR  CONSEQUENTIAL DAMAGES    */
+/*    (INCLUDING, BUT  NOT LIMITED TO,  PROCUREMENT OF SUBSTITUTE    */
+/*    GOODS  OR  SERVICES; LOSS  OF  USE,  DATA,  OR PROFITS;  OR    */
+/*    BUSINESS INTERRUPTION) HOWEVER CAUSED  AND ON ANY THEORY OF    */
+/*    LIABILITY, WHETHER  IN CONTRACT, STRICT  LIABILITY, OR TORT    */
+/*    (INCLUDING NEGLIGENCE OR OTHERWISE)  ARISING IN ANY WAY OUT    */
+/*    OF  THE  USE OF  THIS  SOFTWARE,  EVEN  IF ADVISED  OF  THE    */
+/*    POSSIBILITY OF SUCH DAMAGE.                                    */
+/*                                                                   */
+/* The views and conclusions contained in the software and           */
+/* documentation are those of the authors and should not be          */
+/* interpreted as representing official policies, either expressed   */
+/* or implied, of The University of Texas at Austin.                 */
+/*********************************************************************/
+
+#define ASSEMBLER
+#include "common.h"
+
+#define N	%i0
+#define X	%i1
+#define INCX	%i2
+#define I	%i3
+
+#ifdef DOUBLE
+#define c1	%f0
+#define c2	%f2
+#define t1	%f8
+#define t2	%f10
+#define t3	%f12
+#define t4	%f14
+
+#define a1	%f16
+#define a2	%f18
+#define a3	%f20
+#define a4	%f22
+#define a5	%f24
+#define a6	%f26
+#define a7	%f28
+#define a8	%f30
+#else
+#define c1	%f0
+#define c2	%f1
+#define t1	%f4
+#define t2	%f5
+#define t3	%f6
+#define t4	%f7
+
+#define a1	%f8
+#define a2	%f9
+#define a3	%f10
+#define a4	%f11
+#define a5	%f12
+#define a6	%f13
+#define a7	%f14
+#define a8	%f15
+#endif
+
+	PROLOGUE
+	SAVESP
+
+	FCLR(0)
+
+	sll	INCX, BASE_SHIFT, INCX
+
+	FMOV	c1, c2
+	FMOV	c1, t1
+	FMOV	c1, t2
+	FMOV	c1, t3
+	FMOV	c1, t4
+
+	cmp	INCX, 0
+	ble	.LL19
+	cmp	INCX, SIZE
+	bne	.LL50
+
+	sra	N, 3, I
+	cmp	I, 0
+	ble,pn	%icc, .LL15
+	nop
+
+	LDF	[X +  0 * SIZE], a1
+	add	I, -1, I
+	LDF	[X +  1 * SIZE], a2
+	cmp	I, 0
+	LDF	[X +  2 * SIZE], a3
+	LDF	[X +  3 * SIZE], a4
+	LDF	[X +  4 * SIZE], a5
+	LDF	[X +  5 * SIZE], a6
+	LDF	[X +  6 * SIZE], a7
+	LDF	[X +  7 * SIZE], a8
+
+	ble,pt	%icc, .LL12
+	add	X, 8 * SIZE, X
+
+#define PREFETCHSIZE 128
+
+.LL11:
+	FADD	c1, t1, c1
+	prefetch [X  + PREFETCHSIZE * SIZE], 0
+	FMOV	a1, t1
+	LDF	[X +  0 * SIZE], a1
+
+	FADD	c2, t2, c2
+	add	I, -1, I
+	FMOV	a2, t2
+	LDF	[X +  1 * SIZE], a2
+
+	FADD	c1, t3, c1
+	cmp	I, 0
+	FMOV	a3, t3
+	LDF	[X +  2 * SIZE], a3
+
+	FADD	c2, t4, c2
+	nop
+	FMOV	a4, t4
+	LDF	[X +  3 * SIZE], a4
+
+	FADD	c1, t1, c1
+	nop
+	FMOV	a5, t1
+	LDF	[X +  4 * SIZE], a5
+
+	FADD	c2, t2, c2
+	nop
+	FMOV	a6, t2
+	LDF	[X +  5 * SIZE], a6
+
+	FADD	c1, t3, c1
+	FMOV	a7, t3
+	LDF	[X +  6 * SIZE], a7
+	add	X, 8 * SIZE, X
+
+	FADD	c2, t4, c2
+	FMOV	a8, t4
+	bg,pt	%icc, .LL11
+	LDF	[X -  1 * SIZE], a8
+
+.LL12:
+	FADD	c1, t1, c1
+	FMOV	a1, t1
+	FADD	c2, t2, c2
+	FMOV	a2, t2
+
+	FADD	c1, t3, c1
+	FMOV	a3, t3
+	FADD	c2, t4, c2
+	FMOV	a4, t4
+
+	FADD	c1, t1, c1
+	FMOV	a5, t1
+	FADD	c2, t2, c2
+	FMOV	a6, t2
+
+	FADD	c1, t3, c1
+	FMOV	a7, t3
+	FADD	c2, t4, c2
+	FMOV	a8, t4
+
+.LL15:
+	and	N, 7, I
+	cmp	I,  0
+	ble,a,pn %icc, .LL19
+	nop
+
+.LL16:
+	LDF	[X +  0 * SIZE], a1
+	add	I, -1, I
+	cmp	I, 0
+	FADD	c1, t1, c1
+	FMOV	a1, t1
+	bg,pt	%icc, .LL16
+	add	X, 1 * SIZE, X
+
+.LL19:
+	FADD	c1, t1, c1
+	FADD	c2, t2, c2
+	FADD	c1, t3, c1
+	FADD	c2, t4, c2
+
+	FADD	c1, c2, c1
+	return	%i7 + 8
+	clr	%g0
+
+.LL50:
+	sra	N, 3, I
+	cmp	I, 0
+	ble,pn	%icc, .LL55
+	nop
+
+	LDF	[X +  0 * SIZE], a1
+	add	X, INCX, X
+	LDF	[X +  0 * SIZE], a2
+	add	X, INCX, X
+	LDF	[X +  0 * SIZE], a3
+	add	X, INCX, X
+	LDF	[X +  0 * SIZE], a4
+	add	X, INCX, X
+	LDF	[X +  0 * SIZE], a5
+	add	X, INCX, X
+	LDF	[X +  0 * SIZE], a6
+	add	X, INCX, X
+	add	I, -1, I
+	LDF	[X +  0 * SIZE], a7
+	cmp	I, 0
+	add	X, INCX, X
+	LDF	[X +  0 * SIZE], a8
+
+	ble,pt	%icc, .LL52
+	add	X, INCX, X
+
+.LL51:
+	FADD	c1, t1, c1
+	add	I, -1, I
+	FMOV	a1, t1
+	LDF	[X +  0 * SIZE], a1
+	add	X, INCX, X
+
+	FADD	c2, t2, c2
+	cmp	I, 0
+	FMOV	a2, t2
+	LDF	[X +  0 * SIZE], a2
+	add	X, INCX, X
+
+	FADD	c1, t3, c1
+	FMOV	a3, t3
+	LDF	[X +  0 * SIZE], a3
+	add	X, INCX, X
+
+	FADD	c2, t4, c2
+	FMOV	a4, t4
+	LDF	[X +  0 * SIZE], a4
+	add	X, INCX, X
+
+	FADD	c1, t1, c1
+	FMOV	a5, t1
+	LDF	[X +  0 * SIZE], a5
+	add	X, INCX, X
+
+	FADD	c2, t2, c2
+	FMOV	a6, t2
+	LDF	[X +  0 * SIZE], a6
+	add	X, INCX, X
+
+	FADD	c1, t3, c1
+	FMOV	a7, t3
+	LDF	[X +  0 * SIZE], a7
+	add	X, INCX, X
+
+	FADD	c2, t4, c2
+	FMOV	a8, t4
+	LDF	[X +  0 * SIZE], a8
+
+	bg,pt	%icc, .LL51
+	add	X, INCX, X
+
+.LL52:
+	FADD	c1, t1, c1
+	FMOV	a1, t1
+	FADD	c2, t2, c2
+	FMOV	a2, t2
+
+	FADD	c1, t3, c1
+	FMOV	a3, t3
+	FADD	c2, t4, c2
+	FMOV	a4, t4
+
+	FADD	c1, t1, c1
+	FMOV	a5, t1
+	FADD	c2, t2, c2
+	FMOV	a6, t2
+
+	FADD	c1, t3, c1
+	FMOV	a7, t3
+	FADD	c2, t4, c2
+	FMOV	a8, t4
+
+.LL55:
+	and	N, 7, I
+	cmp	I,  0
+	ble,a,pn %icc, .LL59
+	nop
+
+.LL56:
+	LDF	[X +  0 * SIZE], a1
+	FADD	c1, t1, c1
+	add	I, -1, I
+	FMOV	a1, t1
+	cmp	I, 0
+	bg,pt	%icc, .LL56
+	add	X, INCX, X
+
+.LL59:
+	FADD	c1, t1, c1
+	FADD	c2, t2, c2
+	FADD	c1, t3, c1
+	FADD	c2, t4, c2
+
+	FADD	c1, c2, c1
+	return	%i7 + 8
+	clr	%o0
+
+	EPILOGUE

kernel/sparc/zsum.S

@@ -0,0 +1,327 @@
+/*********************************************************************/
+/* Copyright 2009, 2010 The University of Texas at Austin.           */
+/* All rights reserved.                                              */
+/*                                                                   */
+/* Redistribution and use in source and binary forms, with or        */
+/* without modification, are permitted provided that the following   */
+/* conditions are met:                                               */
+/*                                                                   */
+/*   1. Redistributions of source code must retain the above         */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer.                                                  */
+/*                                                                   */
+/*   2. Redistributions in binary form must reproduce the above      */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer in the documentation and/or other materials       */
+/*      provided with the distribution.                              */
+/*                                                                   */
+/*    THIS  SOFTWARE IS PROVIDED  BY THE  UNIVERSITY OF  TEXAS AT    */
+/*    AUSTIN  ``AS IS''  AND ANY  EXPRESS OR  IMPLIED WARRANTIES,    */
+/*    INCLUDING, BUT  NOT LIMITED  TO, THE IMPLIED  WARRANTIES OF    */
+/*    MERCHANTABILITY  AND FITNESS FOR  A PARTICULAR  PURPOSE ARE    */
+/*    DISCLAIMED.  IN  NO EVENT SHALL THE UNIVERSITY  OF TEXAS AT    */
+/*    AUSTIN OR CONTRIBUTORS BE  LIABLE FOR ANY DIRECT, INDIRECT,    */
+/*    INCIDENTAL,  SPECIAL, EXEMPLARY,  OR  CONSEQUENTIAL DAMAGES    */
+/*    (INCLUDING, BUT  NOT LIMITED TO,  PROCUREMENT OF SUBSTITUTE    */
+/*    GOODS  OR  SERVICES; LOSS  OF  USE,  DATA,  OR PROFITS;  OR    */
+/*    BUSINESS INTERRUPTION) HOWEVER CAUSED  AND ON ANY THEORY OF    */
+/*    LIABILITY, WHETHER  IN CONTRACT, STRICT  LIABILITY, OR TORT    */
+/*    (INCLUDING NEGLIGENCE OR OTHERWISE)  ARISING IN ANY WAY OUT    */
+/*    OF  THE  USE OF  THIS  SOFTWARE,  EVEN  IF ADVISED  OF  THE    */
+/*    POSSIBILITY OF SUCH DAMAGE.                                    */
+/*                                                                   */
+/* The views and conclusions contained in the software and           */
+/* documentation are those of the authors and should not be          */
+/* interpreted as representing official policies, either expressed   */
+/* or implied, of The University of Texas at Austin.                 */
+/*********************************************************************/
+
+#define ASSEMBLER
+#include "common.h"
+
+#define N	%i0
+#define X	%i1
+#define INCX	%i2
+#define I	%i3
+
+#ifdef DOUBLE
+#define c1	%f0
+#define c2	%f2
+#define t1	%f8
+#define t2	%f10
+#define t3	%f12
+#define t4	%f14
+
+#define a1	%f16
+#define a2	%f18
+#define a3	%f20
+#define a4	%f22
+#define a5	%f24
+#define a6	%f26
+#define a7	%f28
+#define a8	%f30
+#else
+#define c1	%f0
+#define c2	%f1
+#define t1	%f4
+#define t2	%f5
+#define t3	%f6
+#define t4	%f7
+
+#define a1	%f8
+#define a2	%f9
+#define a3	%f10
+#define a4	%f11
+#define a5	%f12
+#define a6	%f13
+#define a7	%f14
+#define a8	%f15
+#endif
+
+	PROLOGUE
+	SAVESP
+
+	FCLR(0)
+
+	sll	INCX, ZBASE_SHIFT, INCX
+
+	FMOV	c1, c2
+	FMOV	c1, t1
+	FMOV	c1, t2
+	FMOV	c1, t3
+	FMOV	c1, t4
+
+	cmp	INCX, 0
+	ble	.LL19
+	nop
+
+	cmp	INCX, 2 * SIZE
+	bne	.LL50
+	nop
+
+	sra	N, 2, I
+	cmp	I, 0
+	ble,pn	%icc, .LL15
+	nop
+
+	LDF	[X +  0 * SIZE], a1
+	add	I, -1, I
+	LDF	[X +  1 * SIZE], a2
+	cmp	I, 0
+	LDF	[X +  2 * SIZE], a3
+	LDF	[X +  3 * SIZE], a4
+	LDF	[X +  4 * SIZE], a5
+	LDF	[X +  5 * SIZE], a6
+	LDF	[X +  6 * SIZE], a7
+	LDF	[X +  7 * SIZE], a8
+
+	ble,pt	%icc, .LL12
+	add	X, 8 * SIZE, X
+
+#define PREFETCHSIZE 32
+
+.LL11:
+	FADD	c1, t1, c1
+	prefetch [X  + PREFETCHSIZE * SIZE], 0
+	FMOV	a1, t1
+	LDF	[X +  0 * SIZE], a1
+
+	FADD	c2, t2, c2
+	add	I, -1, I
+	FMOV	a2, t2
+	LDF	[X +  1 * SIZE], a2
+
+	FADD	c1, t3, c1
+	cmp	I, 0
+	FMOV	a3, t3
+	LDF	[X +  2 * SIZE], a3
+
+	FADD	c2, t4, c2
+	nop
+	FMOV	a4, t4
+	LDF	[X +  3 * SIZE], a4
+
+	FADD	c1, t1, c1
+	nop
+	FMOV	a5, t1
+	LDF	[X +  4 * SIZE], a5
+
+	FADD	c2, t2, c2
+	nop
+	FMOV	a6, t2
+	LDF	[X +  5 * SIZE], a6
+
+	FADD	c1, t3, c1
+	FMOV	a7, t3
+	LDF	[X +  6 * SIZE], a7
+	add	X, 8 * SIZE, X
+
+	FADD	c2, t4, c2
+	FMOV	a8, t4
+	bg,pt	%icc, .LL11
+	LDF	[X -  1 * SIZE], a8
+
+.LL12:
+	FADD	c1, t1, c1
+	FMOV	a1, t1
+	FADD	c2, t2, c2
+	FMOV	a2, t2
+
+	FADD	c1, t3, c1
+	FMOV	a3, t3
+	FADD	c2, t4, c2
+	FMOV	a4, t4
+
+	FADD	c1, t1, c1
+	FMOV	a5, t1
+	FADD	c2, t2, c2
+	FMOV	a6, t2
+
+	FADD	c1, t3, c1
+	FMOV	a7, t3
+	FADD	c2, t4, c2
+	FMOV	a8, t4
+
+.LL15:
+	and	N, 3, I
+	cmp	I,  0
+	ble,a,pn %icc, .LL19
+	nop
+
+.LL16:
+	LDF	[X +  0 * SIZE], a1
+	LDF	[X +  1 * SIZE], a2
+	add	I, -1, I
+	cmp	I, 0
+	FADD	c1, t1, c1
+	FADD	c2, t2, c2
+	FMOV	a1, t1
+	FMOV	a2, t2
+	bg,pt	%icc, .LL16
+	add	X, 2 * SIZE, X
+
+.LL19:
+	FADD	c1, t1, c1
+	FADD	c2, t2, c2
+	FADD	c1, t3, c1
+	FADD	c2, t4, c2
+
+	FADD	c1, c2, c1
+	return	%i7 + 8
+	clr	%g0
+
+.LL50:
+	sra	N, 2, I
+	cmp	I, 0
+	ble,pn	%icc, .LL55
+	nop
+
+	LDF	[X +  0 * SIZE], a1
+	LDF	[X +  1 * SIZE], a2
+	add	X, INCX, X
+	LDF	[X +  0 * SIZE], a3
+	LDF	[X +  1 * SIZE], a4
+	add	X, INCX, X
+	LDF	[X +  0 * SIZE], a5
+	LDF	[X +  1 * SIZE], a6
+	add	X, INCX, X
+	add	I, -1, I
+	LDF	[X +  0 * SIZE], a7
+	cmp	I, 0
+	LDF	[X +  1 * SIZE], a8
+
+	ble,pt	%icc, .LL52
+	add	X, INCX, X
+
+.LL51:
+	FADD	c1, t1, c1
+	add	I, -1, I
+	FMOV	a1, t1
+	LDF	[X +  0 * SIZE], a1
+
+	FADD	c2, t2, c2
+	cmp	I, 0
+	FMOV	a2, t2
+	LDF	[X +  1 * SIZE], a2
+	add	X, INCX, X
+
+	FADD	c1, t3, c1
+	FMOV	a3, t3
+	LDF	[X +  0 * SIZE], a3
+
+	FADD	c2, t4, c2
+	FMOV	a4, t4
+	LDF	[X +  1 * SIZE], a4
+	add	X, INCX, X
+
+	FADD	c1, t1, c1
+	FMOV	a5, t1
+	LDF	[X +  0 * SIZE], a5
+
+	FADD	c2, t2, c2
+	FMOV	a6, t2
+	LDF	[X +  1 * SIZE], a6
+	add	X, INCX, X
+
+	FADD	c1, t3, c1
+	FMOV	a7, t3
+	LDF	[X +  0 * SIZE], a7
+
+	FADD	c2, t4, c2
+	FMOV	a8, t4
+	LDF	[X +  1 * SIZE], a8
+
+	bg,pt	%icc, .LL51
+	add	X, INCX, X
+
+.LL52:
+	FADD	c1, t1, c1
+	FMOV	a1, t1
+	FADD	c2, t2, c2
+	FMOV	a2, t2
+
+	FADD	c1, t3, c1
+	FMOV	a3, t3
+	FADD	c2, t4, c2
+	FMOV	a4, t4
+
+	FADD	c1, t1, c1
+	FMOV	a5, t1
+	FADD	c2, t2, c2
+	FMOV	a6, t2
+
+	FADD	c1, t3, c1
+	FMOV	a7, t3
+	FADD	c2, t4, c2
+	FMOV	a8, t4
+
+.LL55:
+	and	N, 3, I
+	cmp	I,  0
+	ble,a,pn %icc, .LL59
+	nop
+
+.LL56:
+	LDF	[X +  0 * SIZE], a1
+	LDF	[X +  1 * SIZE], a2
+	FADD	c1, t1, c1
+	FADD	c2, t2, c2
+	add	I, -1, I
+	FMOV	a1, t1
+	FMOV	a2, t2
+	cmp	I, 0
+	bg,pt	%icc, .LL56
+	add	X, INCX, X
+
+.LL59:
+	FADD	c1, t1, c1
+	FADD	c2, t2, c2
+	FADD	c1, t3, c1
+	FADD	c2, t4, c2
+
+	FADD	c1, c2, c1
+
+	return	%i7 + 8
+	clr	%o0
+
+	EPILOGUE

kernel/x86/KERNEL.generic

@@ -94,6 +94,11 @@ DASUMKERNEL  = ../arm/asum.c
 CASUMKERNEL  = ../arm/zasum.c
 ZASUMKERNEL  = ../arm/zasum.c
 
+SSUMKERNEL  = ../arm/sum.c
+DSUMKERNEL  = ../arm/sum.c
+CSUMKERNEL  = ../arm/zsum.c
+ZSUMKERNEL  = ../arm/zsum.c
+
 SAXPYKERNEL  = ../arm/axpy.c
 DAXPYKERNEL  = ../arm/axpy.c
 CAXPYKERNEL  = ../arm/zaxpy.c

kernel/x86/sum.S

@@ -0,0 +1,207 @@
+/*********************************************************************/
+/* Copyright 2009, 2010 The University of Texas at Austin.           */
+/* All rights reserved.                                              */
+/*                                                                   */
+/* Redistribution and use in source and binary forms, with or        */
+/* without modification, are permitted provided that the following   */
+/* conditions are met:                                               */
+/*                                                                   */
+/*   1. Redistributions of source code must retain the above         */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer.                                                  */
+/*                                                                   */
+/*   2. Redistributions in binary form must reproduce the above      */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer in the documentation and/or other materials       */
+/*      provided with the distribution.                              */
+/*                                                                   */
+/*    THIS  SOFTWARE IS PROVIDED  BY THE  UNIVERSITY OF  TEXAS AT    */
+/*    AUSTIN  ``AS IS''  AND ANY  EXPRESS OR  IMPLIED WARRANTIES,    */
+/*    INCLUDING, BUT  NOT LIMITED  TO, THE IMPLIED  WARRANTIES OF    */
+/*    MERCHANTABILITY  AND FITNESS FOR  A PARTICULAR  PURPOSE ARE    */
+/*    DISCLAIMED.  IN  NO EVENT SHALL THE UNIVERSITY  OF TEXAS AT    */
+/*    AUSTIN OR CONTRIBUTORS BE  LIABLE FOR ANY DIRECT, INDIRECT,    */
+/*    INCIDENTAL,  SPECIAL, EXEMPLARY,  OR  CONSEQUENTIAL DAMAGES    */
+/*    (INCLUDING, BUT  NOT LIMITED TO,  PROCUREMENT OF SUBSTITUTE    */
+/*    GOODS  OR  SERVICES; LOSS  OF  USE,  DATA,  OR PROFITS;  OR    */
+/*    BUSINESS INTERRUPTION) HOWEVER CAUSED  AND ON ANY THEORY OF    */
+/*    LIABILITY, WHETHER  IN CONTRACT, STRICT  LIABILITY, OR TORT    */
+/*    (INCLUDING NEGLIGENCE OR OTHERWISE)  ARISING IN ANY WAY OUT    */
+/*    OF  THE  USE OF  THIS  SOFTWARE,  EVEN  IF ADVISED  OF  THE    */
+/*    POSSIBILITY OF SUCH DAMAGE.                                    */
+/*                                                                   */
+/* The views and conclusions contained in the software and           */
+/* documentation are those of the authors and should not be          */
+/* interpreted as representing official policies, either expressed   */
+/* or implied, of The University of Texas at Austin.                 */
+/*********************************************************************/
+
+#define ASSEMBLER
+#include "common.h"
+
+#define STACK	 8
+#define ARGS	 0
+
+#define STACK_M		 4 + STACK + ARGS(%esp)
+#define STACK_X		 8 + STACK + ARGS(%esp)
+#define STACK_INCX	12 + STACK + ARGS(%esp)
+
+#define M	%edx
+#define X	%ecx
+#define INCX	%esi
+
+#define I	%eax
+
+#include "l1param.h"
+
+	PROLOGUE
+
+	pushl	%esi
+	pushl	%ebx
+
+	PROFCODE
+
+#if defined(F_INTERFACE_GFORT) || defined(F_INTERFACE_G95)
+	EMMS
+#endif
+
+	movl	STACK_M,    M
+	movl	STACK_X,    X
+	movl	STACK_INCX, INCX
+
+#ifdef F_INTERFACE
+	movl	(M),    M
+	movl	(INCX), INCX
+#endif
+
+	fldz
+	testl	M, M
+	jle	.L999
+	testl	INCX, INCX
+	jle	.L999
+
+	sall	$BASE_SHIFT, INCX
+	fldz
+	fldz
+	fldz
+	cmpl	$SIZE, INCX
+	jne	.L40
+
+	movl	M,  I
+	sarl	$3, I
+	jle	.L20
+	ALIGN_4
+
+.L10:
+#ifdef PREFETCH
+	PREFETCH (PREFETCHSIZE +  0) - PREOFFSET(X)
+#endif
+
+	FLD	0 * SIZE(X)
+	FLD	1 * SIZE(X)
+	FLD	2 * SIZE(X)
+	FLD	3 * SIZE(X)
+
+	faddp	%st, %st(7)
+	faddp	%st, %st(5)
+	faddp	%st, %st(3)
+	faddp	%st, %st(1)
+
+	FLD	4 * SIZE(X)
+	FLD	5 * SIZE(X)
+	FLD	6 * SIZE(X)
+	FLD	7 * SIZE(X)
+
+	addl	$8 * SIZE, X
+
+	faddp	%st, %st(7)
+	faddp	%st, %st(5)
+	faddp	%st, %st(3)
+	faddp	%st, %st(1)
+
+	decl	I
+	jg	.L10
+	ALIGN_4
+
+.L20:
+	movl	M,  I
+	andl	$7, I
+	jle	.L998
+	ALIGN_4
+
+
+.L21:
+	FLD	(X)
+	faddp	%st,%st(1)
+	addl	$1 * SIZE, X
+	decl	I
+	jg	.L21
+	jmp	.L998
+	ALIGN_4
+
+.L40:
+	movl	M,  I
+	sarl	$3, I
+	jle	.L60
+	ALIGN_4
+
+.L50:
+	FLD	(X)
+	addl	INCX, X
+	FLD	(X)
+	addl	INCX, X
+	FLD	(X)
+	addl	INCX, X
+	FLD	(X)
+	addl	INCX, X
+
+	faddp	%st, %st(7)
+	faddp	%st, %st(5)
+	faddp	%st, %st(3)
+	faddp	%st, %st(1)
+
+	FLD	(X)
+	addl	INCX, X
+	FLD	(X)
+	addl	INCX, X
+	FLD	(X)
+	addl	INCX, X
+	FLD	(X)
+	addl	INCX, X
+
+	faddp	%st, %st(7)
+	faddp	%st, %st(5)
+	faddp	%st, %st(3)
+	faddp	%st, %st(1)
+
+	decl	I
+	jg	.L50
+	ALIGN_4
+
+.L60:
+	movl	M,  I
+	andl	$7, I
+	jle	.L998
+	ALIGN_4
+
+
+.L61:
+	FLD	(X)
+	addl	INCX, X
+	faddp	%st,%st(1)
+	decl	I
+	jg	.L61
+	ALIGN_4
+
+.L998:
+	faddp	%st,%st(2)
+	faddp	%st,%st(1)
+	faddp	%st,%st(1)
+	ALIGN_4
+
+.L999:
+	popl	%ebx
+	popl	%esi
+	ret
+
+	EPILOGUE

kernel/x86/zsum.S

@@ -0,0 +1,208 @@
+/*********************************************************************/
+/* Copyright 2009, 2010 The University of Texas at Austin.           */
+/* All rights reserved.                                              */
+/*                                                                   */
+/* Redistribution and use in source and binary forms, with or        */
+/* without modification, are permitted provided that the following   */
+/* conditions are met:                                               */
+/*                                                                   */
+/*   1. Redistributions of source code must retain the above         */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer.                                                  */
+/*                                                                   */
+/*   2. Redistributions in binary form must reproduce the above      */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer in the documentation and/or other materials       */
+/*      provided with the distribution.                              */
+/*                                                                   */
+/*    THIS  SOFTWARE IS PROVIDED  BY THE  UNIVERSITY OF  TEXAS AT    */
+/*    AUSTIN  ``AS IS''  AND ANY  EXPRESS OR  IMPLIED WARRANTIES,    */
+/*    INCLUDING, BUT  NOT LIMITED  TO, THE IMPLIED  WARRANTIES OF    */
+/*    MERCHANTABILITY  AND FITNESS FOR  A PARTICULAR  PURPOSE ARE    */
+/*    DISCLAIMED.  IN  NO EVENT SHALL THE UNIVERSITY  OF TEXAS AT    */
+/*    AUSTIN OR CONTRIBUTORS BE  LIABLE FOR ANY DIRECT, INDIRECT,    */
+/*    INCIDENTAL,  SPECIAL, EXEMPLARY,  OR  CONSEQUENTIAL DAMAGES    */
+/*    (INCLUDING, BUT  NOT LIMITED TO,  PROCUREMENT OF SUBSTITUTE    */
+/*    GOODS  OR  SERVICES; LOSS  OF  USE,  DATA,  OR PROFITS;  OR    */
+/*    BUSINESS INTERRUPTION) HOWEVER CAUSED  AND ON ANY THEORY OF    */
+/*    LIABILITY, WHETHER  IN CONTRACT, STRICT  LIABILITY, OR TORT    */
+/*    (INCLUDING NEGLIGENCE OR OTHERWISE)  ARISING IN ANY WAY OUT    */
+/*    OF  THE  USE OF  THIS  SOFTWARE,  EVEN  IF ADVISED  OF  THE    */
+/*    POSSIBILITY OF SUCH DAMAGE.                                    */
+/*                                                                   */
+/* The views and conclusions contained in the software and           */
+/* documentation are those of the authors and should not be          */
+/* interpreted as representing official policies, either expressed   */
+/* or implied, of The University of Texas at Austin.                 */
+/*********************************************************************/
+
+#define ASSEMBLER
+#include "common.h"
+
+#define STACK	 8
+#define ARGS	 0
+
+#define STACK_M		 4 + STACK + ARGS(%esp)
+#define STACK_X		 8 + STACK + ARGS(%esp)
+#define STACK_INCX	12 + STACK + ARGS(%esp)
+
+#define M	%edx
+#define X	%ecx
+#define INCX	%esi
+
+#define I	%eax
+
+#include "l1param.h"
+
+	PROLOGUE
+
+	pushl	%esi
+	pushl	%ebx
+
+	PROFCODE
+
+#if defined(F_INTERFACE_GFORT) || defined(F_INTERFACE_G95)
+	EMMS
+#endif
+
+	movl	STACK_M,    M
+	movl	STACK_X,    X
+	movl	STACK_INCX, INCX
+
+#ifdef F_INTERFACE
+	movl	(M),    M
+	movl	(INCX), INCX
+#endif
+
+	fldz
+	testl	M,  M
+	jle	.L999
+	testl	INCX, INCX
+	jle	.L999
+
+	sall	$ZBASE_SHIFT, INCX
+
+	fldz
+	fldz
+	fldz
+	cmpl	$SIZE * 2, INCX
+	jne	.L40
+
+	movl	M,  I
+	sarl	$2, I
+	jle	.L20
+	ALIGN_4
+
+.L10:
+#ifdef PREFETCH
+	PREFETCH (PREFETCHSIZE +  0) - PREOFFSET(X)
+#endif
+
+	FLD	0 * SIZE(X)
+	FLD	1 * SIZE(X)
+	FLD	2 * SIZE(X)
+	FLD	3 * SIZE(X)
+
+	faddp	%st, %st(7)
+	faddp	%st, %st(5)
+	faddp	%st, %st(3)
+	faddp	%st, %st(1)
+
+	FLD	4 * SIZE(X)
+	FLD	5 * SIZE(X)
+	FLD	6 * SIZE(X)
+	FLD	7 * SIZE(X)
+
+	addl	$8 * SIZE, X
+
+	faddp	%st, %st(7)
+	faddp	%st, %st(5)
+	faddp	%st, %st(3)
+	faddp	%st, %st(1)
+
+	decl	I
+	jg	.L10
+	ALIGN_4
+
+.L20:
+	movl	M,  I
+	andl	$3, I
+	jle	.L998
+	ALIGN_4
+
+
+.L21:
+	FLD	0 * SIZE(X)
+	FLD	1 * SIZE(X)
+	faddp	%st,%st(3)
+	faddp	%st,%st(1)
+	addl	$2 * SIZE, X
+	decl	I
+	jg	.L21
+	jmp	.L998
+	ALIGN_4
+
+.L40:
+	movl	M,  I
+	sarl	$2, I
+	jle	.L60
+	ALIGN_4
+
+.L50:
+	FLD	0 * SIZE(X)
+	FLD	1 * SIZE(X)
+	addl	INCX, X
+	FLD	0 * SIZE(X)
+	FLD	1 * SIZE(X)
+	addl	INCX, X
+
+	faddp	%st, %st(7)
+	faddp	%st, %st(5)
+	faddp	%st, %st(3)
+	faddp	%st, %st(1)
+
+	FLD	0 * SIZE(X)
+	FLD	1 * SIZE(X)
+	addl	INCX, X
+	FLD	0 * SIZE(X)
+	FLD	1 * SIZE(X)
+	addl	INCX, X
+
+	faddp	%st, %st(7)
+	faddp	%st, %st(5)
+	faddp	%st, %st(3)
+	faddp	%st, %st(1)
+
+	decl	I
+	jg	.L50
+	ALIGN_4
+
+.L60:
+	movl	M,  I
+	andl	$3, I
+	jle	.L998
+	ALIGN_4
+
+
+.L61:
+	FLD	0 * SIZE(X)
+	FLD	1 * SIZE(X)
+	addl	INCX, X
+	faddp	%st,%st(3)
+	faddp	%st,%st(1)
+	decl	I
+	jg	.L61
+	ALIGN_4
+
+.L998:
+	faddp	%st,%st(2)
+	faddp	%st,%st(1)
+	faddp	%st,%st(1)
+	ALIGN_4
+
+.L999:
+	popl	%ebx
+	popl	%esi
+	ret
+
+	EPILOGUE

kernel/x86_64/KERNEL.generic

@@ -94,6 +94,11 @@ DASUMKERNEL  = ../arm/asum.c
 CASUMKERNEL  = ../arm/zasum.c
 ZASUMKERNEL  = ../arm/zasum.c
 
+SSUMKERNEL  = ../arm/sum.c
+DSUMKERNEL  = ../arm/sum.c
+CSUMKERNEL  = ../arm/zsum.c
+ZSUMKERNEL  = ../arm/zsum.c
+
 SAXPYKERNEL  = ../arm/axpy.c
 DAXPYKERNEL  = ../arm/axpy.c
 CAXPYKERNEL  = ../arm/zaxpy.c

kernel/x86_64/sum.S

@@ -0,0 +1,179 @@
+/*********************************************************************/
+/* Copyright 2009, 2010 The University of Texas at Austin.           */
+/* All rights reserved.                                              */
+/*                                                                   */
+/* Redistribution and use in source and binary forms, with or        */
+/* without modification, are permitted provided that the following   */
+/* conditions are met:                                               */
+/*                                                                   */
+/*   1. Redistributions of source code must retain the above         */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer.                                                  */
+/*                                                                   */
+/*   2. Redistributions in binary form must reproduce the above      */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer in the documentation and/or other materials       */
+/*      provided with the distribution.                              */
+/*                                                                   */
+/*    THIS  SOFTWARE IS PROVIDED  BY THE  UNIVERSITY OF  TEXAS AT    */
+/*    AUSTIN  ``AS IS''  AND ANY  EXPRESS OR  IMPLIED WARRANTIES,    */
+/*    INCLUDING, BUT  NOT LIMITED  TO, THE IMPLIED  WARRANTIES OF    */
+/*    MERCHANTABILITY  AND FITNESS FOR  A PARTICULAR  PURPOSE ARE    */
+/*    DISCLAIMED.  IN  NO EVENT SHALL THE UNIVERSITY  OF TEXAS AT    */
+/*    AUSTIN OR CONTRIBUTORS BE  LIABLE FOR ANY DIRECT, INDIRECT,    */
+/*    INCIDENTAL,  SPECIAL, EXEMPLARY,  OR  CONSEQUENTIAL DAMAGES    */
+/*    (INCLUDING, BUT  NOT LIMITED TO,  PROCUREMENT OF SUBSTITUTE    */
+/*    GOODS  OR  SERVICES; LOSS  OF  USE,  DATA,  OR PROFITS;  OR    */
+/*    BUSINESS INTERRUPTION) HOWEVER CAUSED  AND ON ANY THEORY OF    */
+/*    LIABILITY, WHETHER  IN CONTRACT, STRICT  LIABILITY, OR TORT    */
+/*    (INCLUDING NEGLIGENCE OR OTHERWISE)  ARISING IN ANY WAY OUT    */
+/*    OF  THE  USE OF  THIS  SOFTWARE,  EVEN  IF ADVISED  OF  THE    */
+/*    POSSIBILITY OF SUCH DAMAGE.                                    */
+/*                                                                   */
+/* The views and conclusions contained in the software and           */
+/* documentation are those of the authors and should not be          */
+/* interpreted as representing official policies, either expressed   */
+/* or implied, of The University of Texas at Austin.                 */
+/*********************************************************************/
+
+#define ASSEMBLER
+#include "common.h"
+
+#define M	ARG1
+#define X	ARG2
+#define INCX	ARG3
+
+#define I	%rax
+
+#include "l1param.h"
+
+	PROLOGUE
+	PROFCODE
+
+	fldz
+	testq	M, M
+	jle	.L999
+	testq	INCX, INCX
+	jle	.L999
+
+	salq	$BASE_SHIFT, INCX
+
+	fldz
+	fldz
+	fldz
+	cmpq	$SIZE, INCX
+	jne	.L40
+
+	movq	M, I
+	sarq	$3,   I
+	jle	.L20
+	ALIGN_4
+
+.L10:
+#ifdef PREFETCH
+	PREFETCH (PREFETCHSIZE +  0) - PREOFFSET(X)
+#endif
+
+	FLD	0 * SIZE(X)
+	FLD	1 * SIZE(X)
+	FLD	2 * SIZE(X)
+	FLD	3 * SIZE(X)
+
+	faddp	%st, %st(7)
+	faddp	%st, %st(5)
+	faddp	%st, %st(3)
+	faddp	%st, %st(1)
+
+	FLD	4 * SIZE(X)
+	FLD	5 * SIZE(X)
+	FLD	6 * SIZE(X)
+	FLD	7 * SIZE(X)
+
+	addq	$8 * SIZE, X
+
+	faddp	%st, %st(7)
+	faddp	%st, %st(5)
+	faddp	%st, %st(3)
+	faddp	%st, %st(1)
+
+	decq	I
+	jg	.L10
+	ALIGN_4
+
+.L20:
+	andq	$7,  M
+	jle	.L998
+	ALIGN_4
+
+.L21:
+	FLD	(X)
+	faddp	%st,%st(1)
+	addq	$1 * SIZE, X
+	decq	M
+	jg	.L21
+	jmp	.L998
+	ALIGN_4
+
+.L40:
+	movq	M, I
+	sarq	$3,   I
+	jle	.L60
+	ALIGN_4
+
+.L50:
+	FLD	(X)
+	addq	INCX, X
+	FLD	(X)
+	addq	INCX, X
+	FLD	(X)
+	addq	INCX, X
+	FLD	(X)
+	addq	INCX, X
+
+	faddp	%st, %st(7)
+	faddp	%st, %st(5)
+	faddp	%st, %st(3)
+	faddp	%st, %st(1)
+
+	FLD	(X)
+	addq	INCX, X
+	FLD	(X)
+	addq	INCX, X
+	FLD	(X)
+	addq	INCX, X
+	FLD	(X)
+	addq	INCX, X
+
+	faddp	%st, %st(7)
+	faddp	%st, %st(5)
+	faddp	%st, %st(3)
+	faddp	%st, %st(1)
+
+	decq	I
+	jg	.L50
+	ALIGN_4
+
+.L60:
+	andq	$7,  M
+	jle	.L998
+	ALIGN_4
+
+
+.L61:
+	FLD	(X)
+	addq	INCX, X
+	faddp	%st,%st(1)
+	decq	M
+	jg	.L61
+	ALIGN_4
+
+.L998:
+	faddp	%st,%st(2)
+	faddp	%st,%st(1)
+	faddp	%st,%st(1)
+	ALIGN_4
+
+.L999:
+	ret
+
+	EPILOGUE

kernel/x86_64/zsum.S

@@ -0,0 +1,180 @@
+/*********************************************************************/
+/* Copyright 2009, 2010 The University of Texas at Austin.           */
+/* All rights reserved.                                              */
+/*                                                                   */
+/* Redistribution and use in source and binary forms, with or        */
+/* without modification, are permitted provided that the following   */
+/* conditions are met:                                               */
+/*                                                                   */
+/*   1. Redistributions of source code must retain the above         */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer.                                                  */
+/*                                                                   */
+/*   2. Redistributions in binary form must reproduce the above      */
+/*      copyright notice, this list of conditions and the following  */
+/*      disclaimer in the documentation and/or other materials       */
+/*      provided with the distribution.                              */
+/*                                                                   */
+/*    THIS  SOFTWARE IS PROVIDED  BY THE  UNIVERSITY OF  TEXAS AT    */
+/*    AUSTIN  ``AS IS''  AND ANY  EXPRESS OR  IMPLIED WARRANTIES,    */
+/*    INCLUDING, BUT  NOT LIMITED  TO, THE IMPLIED  WARRANTIES OF    */
+/*    MERCHANTABILITY  AND FITNESS FOR  A PARTICULAR  PURPOSE ARE    */
+/*    DISCLAIMED.  IN  NO EVENT SHALL THE UNIVERSITY  OF TEXAS AT    */
+/*    AUSTIN OR CONTRIBUTORS BE  LIABLE FOR ANY DIRECT, INDIRECT,    */
+/*    INCIDENTAL,  SPECIAL, EXEMPLARY,  OR  CONSEQUENTIAL DAMAGES    */
+/*    (INCLUDING, BUT  NOT LIMITED TO,  PROCUREMENT OF SUBSTITUTE    */
+/*    GOODS  OR  SERVICES; LOSS  OF  USE,  DATA,  OR PROFITS;  OR    */
+/*    BUSINESS INTERRUPTION) HOWEVER CAUSED  AND ON ANY THEORY OF    */
+/*    LIABILITY, WHETHER  IN CONTRACT, STRICT  LIABILITY, OR TORT    */
+/*    (INCLUDING NEGLIGENCE OR OTHERWISE)  ARISING IN ANY WAY OUT    */
+/*    OF  THE  USE OF  THIS  SOFTWARE,  EVEN  IF ADVISED  OF  THE    */
+/*    POSSIBILITY OF SUCH DAMAGE.                                    */
+/*                                                                   */
+/* The views and conclusions contained in the software and           */
+/* documentation are those of the authors and should not be          */
+/* interpreted as representing official policies, either expressed   */
+/* or implied, of The University of Texas at Austin.                 */
+/*********************************************************************/
+
+#define ASSEMBLER
+#include "common.h"
+
+#define M	ARG1
+#define X	ARG2
+#define INCX	ARG3
+
+#define I	%rax
+
+#include "l1param.h"
+
+	PROLOGUE
+	PROFCODE
+
+	fldz
+	testq	M, M
+	jle	.L999
+	testq	INCX, INCX
+	jle	.L999
+
+	salq	$ZBASE_SHIFT, INCX
+
+	fldz
+	fldz
+	fldz
+	cmpq	$SIZE * 2, INCX
+	jne	.L40
+
+	movq	M, I
+	sarq	$2,   I
+	jle	.L20
+	ALIGN_4
+
+.L10:
+#ifdef PREFETCH
+	PREFETCH (PREFETCHSIZE +  0) - PREOFFSET(X)
+#endif
+
+	FLD	0 * SIZE(X)
+	FLD	1 * SIZE(X)
+	FLD	2 * SIZE(X)
+	FLD	3 * SIZE(X)
+
+	faddp	%st, %st(7)
+	faddp	%st, %st(5)
+	faddp	%st, %st(3)
+	faddp	%st, %st(1)
+
+	FLD	4 * SIZE(X)
+	FLD	5 * SIZE(X)
+	FLD	6 * SIZE(X)
+	FLD	7 * SIZE(X)
+
+	addq	$8 * SIZE, X
+
+	faddp	%st, %st(7)
+	faddp	%st, %st(5)
+	faddp	%st, %st(3)
+	faddp	%st, %st(1)
+
+	decq	I
+	jg	.L10
+	ALIGN_4
+
+.L20:
+	andq	$3,  M
+	jle	.L998
+	ALIGN_4
+
+
+.L21:
+	FLD	0 * SIZE(X)
+	FLD	1 * SIZE(X)
+	faddp	%st,%st(3)
+	faddp	%st,%st(1)
+	addq	$2 * SIZE, X
+	decq	M
+	jg	.L21
+	jmp	.L998
+	ALIGN_4
+
+.L40:
+	movq	M, I
+	sarq	$2,   I
+	jle	.L60
+	ALIGN_4
+
+.L50:
+	FLD	0 * SIZE(X)
+	FLD	1 * SIZE(X)
+	addq	INCX, X
+	FLD	0 * SIZE(X)
+	FLD	1 * SIZE(X)
+	addq	INCX, X
+
+	faddp	%st, %st(7)
+	faddp	%st, %st(5)
+	faddp	%st, %st(3)
+	faddp	%st, %st(1)
+
+	FLD	0 * SIZE(X)
+	FLD	1 * SIZE(X)
+	addq	INCX, X
+	FLD	0 * SIZE(X)
+	FLD	1 * SIZE(X)
+	addq	INCX, X
+
+	faddp	%st, %st(7)
+	faddp	%st, %st(5)
+	faddp	%st, %st(3)
+	faddp	%st, %st(1)
+
+	decq	I
+	jg	.L50
+	ALIGN_4
+
+.L60:
+	andq	$3,  M
+	jle	.L998
+	ALIGN_4
+
+
+.L61:
+	FLD	0 * SIZE(X)
+	FLD	1 * SIZE(X)
+	addq	INCX, X
+	faddp	%st,%st(3)
+	faddp	%st,%st(1)
+	decq	M
+	jg	.L61
+	ALIGN_4
+
+.L998:
+	faddp	%st,%st(2)
+	faddp	%st,%st(1)
+	faddp	%st,%st(1)
+	ALIGN_4
+
+.L999:
+	ret
+
+	EPILOGUE

kernel/zarch/KERNEL.Z13

@@ -35,6 +35,11 @@ DASUMKERNEL  = dasum.c
 CASUMKERNEL  = ../arm/zasum.c
 ZASUMKERNEL  = zasum.c
 
+SSUMKERNEL  = ../arm/asum.c
+DSUMKERNEL  = dasum.c
+CSUMKERNEL  = ../arm/zasum.c
+ZSUMKERNEL  = zasum.c
+
 SAXPYKERNEL  = ../arm/axpy.c
 DAXPYKERNEL  = daxpy.c
 CAXPYKERNEL  = ../arm/zaxpy.c

kernel/zarch/KERNEL.Z14

@@ -35,6 +35,11 @@ DASUMKERNEL  = dasum.c
 CASUMKERNEL  = casum.c
 ZASUMKERNEL  = zasum.c
 
+SSUMKERNEL  = ssum.c
+DSUMKERNEL  = dsum.c
+CSUMKERNEL  = csum.c
+ZSUMKERNEL  = zsum.c
+
 SAXPYKERNEL  = saxpy.c
 DAXPYKERNEL  = daxpy.c
 CAXPYKERNEL  = caxpy.c

kernel/zarch/KERNEL.ZARCH_GENERIC

@@ -35,6 +35,11 @@ DASUMKERNEL  = ../arm/asum.c
 CASUMKERNEL  = ../arm/zasum.c
 ZASUMKERNEL  = ../arm/zasum.c
 
+SSUMKERNEL  = ../arm/sum.c
+DSUMKERNEL  = ../arm/sum.c
+CSUMKERNEL  = ../arm/zsum.c
+ZSUMKERNEL  = ../arm/zsum.c
+
 SAXPYKERNEL  = ../arm/axpy.c
 DAXPYKERNEL  = ../arm/axpy.c
 CAXPYKERNEL  = ../arm/zaxpy.c

kernel/zarch/csum.c

@@ -0,0 +1,137 @@
+/***************************************************************************
+Copyright (c) 2013-2019, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include "common.h"
+#include <math.h>
+
+static FLOAT csum_kernel_32(BLASLONG n, FLOAT *x) {
+  FLOAT sum;
+
+  __asm__("vzero   %%v24\n\t"
+    "vzero   %%v25\n\t"
+    "vzero   %%v26\n\t"
+    "vzero   %%v27\n\t"
+    "vzero   %%v28\n\t"
+    "vzero   %%v29\n\t"
+    "vzero   %%v30\n\t"
+    "vzero   %%v31\n\t"
+    "srlg  %[n],%[n],5\n\t"
+    "xgr %%r1,%%r1\n\t"
+    "0:\n\t"
+    "pfd  1, 1024(%%r1,%[x])\n\t"
+    "vl  %%v16, 0(%%r1,%[x])\n\t"
+    "vl  %%v17, 16(%%r1,%[x])\n\t"
+    "vl  %%v18, 32(%%r1,%[x])\n\t"
+    "vl  %%v19, 48(%%r1,%[x])\n\t"
+    "vl  %%v20, 64(%%r1,%[x])\n\t"
+    "vl  %%v21, 80(%%r1,%[x])\n\t"
+    "vl  %%v22, 96(%%r1,%[x])\n\t"
+    "vl  %%v23, 112(%%r1,%[x])\n\t"
+    "vfasb   %%v24,%%v24,%%v16\n\t"
+    "vfasb   %%v25,%%v25,%%v17\n\t"
+    "vfasb   %%v26,%%v26,%%v18\n\t"
+    "vfasb   %%v27,%%v27,%%v19\n\t"
+    "vfasb   %%v28,%%v28,%%v20\n\t"
+    "vfasb   %%v29,%%v29,%%v21\n\t"
+    "vfasb   %%v30,%%v30,%%v22\n\t"
+    "vfasb   %%v31,%%v31,%%v23\n\t"
+    "vl  %%v16, 128(%%r1,%[x])\n\t"
+    "vl  %%v17, 144(%%r1,%[x])\n\t"
+    "vl  %%v18, 160(%%r1,%[x])\n\t"
+    "vl  %%v19, 176(%%r1,%[x])\n\t"
+    "vl  %%v20, 192(%%r1,%[x])\n\t"
+    "vl  %%v21, 208(%%r1,%[x])\n\t"
+    "vl  %%v22, 224(%%r1,%[x])\n\t"
+    "vl  %%v23, 240(%%r1,%[x])\n\t"
+    "vfasb   %%v24,%%v24,%%v16\n\t"
+    "vfasb   %%v25,%%v25,%%v17\n\t"
+    "vfasb   %%v26,%%v26,%%v18\n\t"
+    "vfasb   %%v27,%%v27,%%v19\n\t"
+    "vfasb   %%v28,%%v28,%%v20\n\t"
+    "vfasb   %%v29,%%v29,%%v21\n\t"
+    "vfasb   %%v30,%%v30,%%v22\n\t"
+    "vfasb   %%v31,%%v31,%%v23\n\t"
+    "agfi  %%r1,256\n\t"
+    "brctg %[n],0b\n\t"
+    "vfasb   %%v24,%%v24,%%v25\n\t"
+    "vfasb   %%v24,%%v24,%%v26\n\t"
+    "vfasb   %%v24,%%v24,%%v27\n\t"
+    "vfasb   %%v24,%%v24,%%v28\n\t"
+    "vfasb   %%v24,%%v24,%%v29\n\t"
+    "vfasb   %%v24,%%v24,%%v30\n\t"
+    "vfasb   %%v24,%%v24,%%v31\n\t"
+    "veslg   %%v25,%%v24,32\n\t"
+    "vfasb   %%v24,%%v24,%%v25\n\t"
+    "vrepf   %%v25,%%v24,2\n\t"
+    "vfasb   %%v24,%%v24,%%v25\n\t"
+    "vstef   %%v24,%[asum],0"
+    : [sum] "=Q"(sum),[n] "+&r"(n)
+    : "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x)
+    : "cc", "r1", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23",
+       "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31");
+
+  return sum;
+}
+
+FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x) {
+  BLASLONG i = 0;
+  BLASLONG ip = 0;
+  FLOAT sumf = 0.0;
+  BLASLONG n1;
+  BLASLONG inc_x2;
+
+  if (n <= 0 || inc_x <= 0)
+    return (sumf);
+
+  if (inc_x == 1) {
+
+    n1 = n & -32;
+    if (n1 > 0) {
+
+      sumf = csum_kernel_32(n1, x);
+      i = n1;
+      ip = 2 * n1;
+    }
+
+    while (i < n) {
+      sumf += x[ip] + x[ip + 1];
+      i++;
+      ip += 2;
+    }
+
+  } else {
+    inc_x2 = 2 * inc_x;
+
+    while (i < n) {
+      sumf += x[ip] + x[ip + 1];
+      ip += inc_x2;
+      i++;
+    }
+
+  }
+  return (sumf);
+}

kernel/zarch/dsum.c

@@ -0,0 +1,148 @@
+/***************************************************************************
+Copyright (c) 2013-2019, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include "common.h"
+#include <math.h>
+
+static FLOAT dsum_kernel_32(BLASLONG n, FLOAT *x) {
+  FLOAT sum;
+
+  __asm__("vzero   %%v24\n\t"
+    "vzero   %%v25\n\t"
+    "vzero   %%v26\n\t"
+    "vzero   %%v27\n\t"
+    "vzero   %%v28\n\t"
+    "vzero   %%v29\n\t"
+    "vzero   %%v30\n\t"
+    "vzero   %%v31\n\t"
+    "srlg  %[n],%[n],5\n\t"
+    "xgr %%r1,%%r1\n\t"
+    "0:\n\t"
+    "pfd  1, 1024(%%r1,%[x])\n\t"
+    "vl  %%v16, 0(%%r1,%[x])\n\t"
+    "vl  %%v17, 16(%%r1,%[x])\n\t"
+    "vl  %%v18, 32(%%r1,%[x])\n\t"
+    "vl  %%v19, 48(%%r1,%[x])\n\t"
+    "vl  %%v20, 64(%%r1,%[x])\n\t"
+    "vl  %%v21, 80(%%r1,%[x])\n\t"
+    "vl  %%v22, 96(%%r1,%[x])\n\t"
+    "vl  %%v23, 112(%%r1,%[x])\n\t"
+    "vfadb   %%v24,%%v24,%%v16\n\t"
+    "vfadb   %%v25,%%v25,%%v17\n\t"
+    "vfadb   %%v26,%%v26,%%v18\n\t"
+    "vfadb   %%v27,%%v27,%%v19\n\t"
+    "vfadb   %%v28,%%v28,%%v20\n\t"
+    "vfadb   %%v29,%%v29,%%v21\n\t"
+    "vfadb   %%v30,%%v30,%%v22\n\t"
+    "vfadb   %%v31,%%v31,%%v23\n\t"
+    "vl  %%v16, 128(%%r1,%[x])\n\t"
+    "vl  %%v17, 144(%%r1,%[x])\n\t"
+    "vl  %%v18, 160(%%r1,%[x])\n\t"
+    "vl  %%v19, 176(%%r1,%[x])\n\t"
+    "vl  %%v20, 192(%%r1,%[x])\n\t"
+    "vl  %%v21, 208(%%r1,%[x])\n\t"
+    "vl  %%v22, 224(%%r1,%[x])\n\t"
+    "vl  %%v23, 240(%%r1,%[x])\n\t"
+    "vfadb   %%v24,%%v24,%%v16\n\t"
+    "vfadb   %%v25,%%v25,%%v17\n\t"
+    "vfadb   %%v26,%%v26,%%v18\n\t"
+    "vfadb   %%v27,%%v27,%%v19\n\t"
+    "vfadb   %%v28,%%v28,%%v20\n\t"
+    "vfadb   %%v29,%%v29,%%v21\n\t"
+    "vfadb   %%v30,%%v30,%%v22\n\t"
+    "vfadb   %%v31,%%v31,%%v23\n\t"
+    "agfi  %%r1,256\n\t"
+    "brctg %[n],0b\n\t"
+    "vfadb   %%v24,%%v24,%%v25\n\t"
+    "vfadb   %%v24,%%v24,%%v26\n\t"
+    "vfadb   %%v24,%%v24,%%v27\n\t"
+    "vfadb   %%v24,%%v24,%%v28\n\t"
+    "vfadb   %%v24,%%v24,%%v29\n\t"
+    "vfadb   %%v24,%%v24,%%v30\n\t"
+    "vfadb   %%v24,%%v24,%%v31\n\t"
+    "vrepg   %%v25,%%v24,1\n\t"
+    "vfadb   %%v24,%%v24,%%v25\n\t"
+    "vsteg   %%v24,%[asum],0"
+    : [sum] "=Q"(sum),[n] "+&r"(n)
+    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x)
+    : "cc", "r1", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23",
+       "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31");
+
+  return sum;
+}
+
+FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x) {
+  BLASLONG i = 0;
+  BLASLONG j = 0;
+  FLOAT sumf = 0.0;
+  BLASLONG n1;
+
+  if (n <= 0 || inc_x <= 0)
+    return sumf;
+
+  if (inc_x == 1) {
+
+    n1 = n & -32;
+
+    if (n1 > 0) {
+
+      sumf = dsum_kernel_32(n1, x);
+      i = n1;
+    }
+
+    while (i < n) {
+      sumf += x[i];
+      i++;
+    }
+
+  } else {
+    BLASLONG n1 = n & -4;
+    register FLOAT sum1, sum2;
+    sum1 = 0.0;
+    sum2 = 0.0;
+    while (j < n1) {
+
+      sum1 += x[i];
+      sum2 += x[i + inc_x];
+      sum1 += x[i + 2 * inc_x];
+      sum2 += x[i + 3 * inc_x];
+
+      i += inc_x * 4;
+      j += 4;
+
+    }
+    sumf = sum1 + sum2;
+    while (j < n) {
+
+      sumf += x[i];
+      i += inc_x;
+      j++;
+    }
+
+  }
+  return sumf;
+}

kernel/zarch/ssum.c

@@ -0,0 +1,151 @@
+/***************************************************************************
+Copyright (c) 2013-2019, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include "common.h"
+#include <math.h>
+
+
+static FLOAT ssum_kernel_64(BLASLONG n, FLOAT *x) {
+  FLOAT sum;
+
+  __asm__("vzero   %%v24\n\t"
+    "vzero   %%v25\n\t"
+    "vzero   %%v26\n\t"
+    "vzero   %%v27\n\t"
+    "vzero   %%v28\n\t"
+    "vzero   %%v29\n\t"
+    "vzero   %%v30\n\t"
+    "vzero   %%v31\n\t"
+    "srlg  %[n],%[n],6\n\t"
+    "xgr %%r1,%%r1\n\t"
+    "0:\n\t"
+    "pfd  1, 1024(%%r1,%[x])\n\t"
+    "vl  %%v16, 0(%%r1,%[x])\n\t"
+    "vl  %%v17, 16(%%r1,%[x])\n\t"
+    "vl  %%v18, 32(%%r1,%[x])\n\t"
+    "vl  %%v19, 48(%%r1,%[x])\n\t"
+    "vl  %%v20, 64(%%r1,%[x])\n\t"
+    "vl  %%v21, 80(%%r1,%[x])\n\t"
+    "vl  %%v22, 96(%%r1,%[x])\n\t"
+    "vl  %%v23, 112(%%r1,%[x])\n\t"
+    "vfasb   %%v24,%%v24,%%v16\n\t"
+    "vfasb   %%v25,%%v25,%%v17\n\t"
+    "vfasb   %%v26,%%v26,%%v18\n\t"
+    "vfasb   %%v27,%%v27,%%v19\n\t"
+    "vfasb   %%v28,%%v28,%%v20\n\t"
+    "vfasb   %%v29,%%v29,%%v21\n\t"
+    "vfasb   %%v30,%%v30,%%v22\n\t"
+    "vfasb   %%v31,%%v31,%%v23\n\t"
+    "vl  %%v16, 128(%%r1,%[x])\n\t"
+    "vl  %%v17, 144(%%r1,%[x])\n\t"
+    "vl  %%v18, 160(%%r1,%[x])\n\t"
+    "vl  %%v19, 176(%%r1,%[x])\n\t"
+    "vl  %%v20, 192(%%r1,%[x])\n\t"
+    "vl  %%v21, 208(%%r1,%[x])\n\t"
+    "vl  %%v22, 224(%%r1,%[x])\n\t"
+    "vl  %%v23, 240(%%r1,%[x])\n\t"
+    "vfasb   %%v24,%%v24,%%v16\n\t"
+    "vfasb   %%v25,%%v25,%%v17\n\t"
+    "vfasb   %%v26,%%v26,%%v18\n\t"
+    "vfasb   %%v27,%%v27,%%v19\n\t"
+    "vfasb   %%v28,%%v28,%%v20\n\t"
+    "vfasb   %%v29,%%v29,%%v21\n\t"
+    "vfasb   %%v30,%%v30,%%v22\n\t"
+    "vfasb   %%v31,%%v31,%%v23\n\t"
+    "agfi  %%r1,256\n\t"
+    "brctg %[n],0b\n\t"
+    "vfasb   %%v24,%%v24,%%v25\n\t"
+    "vfasb   %%v24,%%v24,%%v26\n\t"
+    "vfasb   %%v24,%%v24,%%v27\n\t"
+    "vfasb   %%v24,%%v24,%%v28\n\t"
+    "vfasb   %%v24,%%v24,%%v29\n\t"
+    "vfasb   %%v24,%%v24,%%v30\n\t"
+    "vfasb   %%v24,%%v24,%%v31\n\t"
+    "veslg   %%v25,%%v24,32\n\t"
+    "vfasb   %%v24,%%v24,%%v25\n\t"
+    "vrepf   %%v25,%%v24,2\n\t"
+    "vfasb   %%v24,%%v24,%%v25\n\t"
+    "vstef   %%v24,%[asum],0"
+    : [sum] "=Q"(sum),[n] "+&r"(n)
+    : "m"(*(const struct { FLOAT x[n]; } *) x),[x] "a"(x)
+    : "cc", "r1", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23",
+       "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31");
+
+  return sum;
+}
+
+FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x) {
+  BLASLONG i = 0;
+  BLASLONG j = 0;
+  FLOAT sumf = 0.0;
+  BLASLONG n1;
+
+  if (n <= 0 || inc_x <= 0)
+    return sumf;
+
+  if (inc_x == 1) {
+
+    n1 = n & -64;
+
+    if (n1 > 0) {
+
+      sumf = ssum_kernel_64(n1, x);
+      i = n1;
+    }
+
+    while (i < n) {
+      sumf += x[i];
+      i++;
+    }
+
+  } else {
+    BLASLONG n1 = n & -4;
+    register FLOAT sum1, sum2;
+    sum1 = 0.0;
+    sum2 = 0.0;
+    while (j < n1) {
+
+      sum1 += x[i];
+      sum2 += x[i + inc_x];
+      sum1 += x[i + 2 * inc_x];
+      sum2 += x[i + 3 * inc_x];
+
+      i += inc_x * 4;
+      j += 4;
+
+    }
+    sumf = sum1 + sum2;
+    while (j < n) {
+
+      sumf += x[i];
+      i += inc_x;
+      j++;
+    }
+
+  }
+  return sumf;
+}

kernel/zarch/zsum.c

@@ -0,0 +1,136 @@
+/***************************************************************************
+Copyright (c) 2013-2019, The OpenBLAS Project
+All rights reserved.
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+1. Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+2. Redistributions in binary form must reproduce the above copyright
+notice, this list of conditions and the following disclaimer in
+the documentation and/or other materials provided with the
+distribution.
+3. Neither the name of the OpenBLAS project nor the names of
+its contributors may be used to endorse or promote products
+derived from this software without specific prior written permission.
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
+USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*****************************************************************************/
+
+#include "common.h"
+#include <math.h>
+
+
+static FLOAT zsum_kernel_16(BLASLONG n, FLOAT *x) {
+  FLOAT sum;
+
+  __asm__("vzero   %%v24\n\t"
+    "vzero   %%v25\n\t"
+    "vzero   %%v26\n\t"
+    "vzero   %%v27\n\t"
+    "vzero   %%v28\n\t"
+    "vzero   %%v29\n\t"
+    "vzero   %%v30\n\t"
+    "vzero   %%v31\n\t"
+    "srlg  %[n],%[n],4\n\t"
+    "xgr %%r1,%%r1\n\t"
+    "0:\n\t"
+    "pfd  1, 1024(%%r1,%[x])\n\t"
+    "vl  %%v16, 0(%%r1,%[x])\n\t"
+    "vl  %%v17, 16(%%r1,%[x])\n\t"
+    "vl  %%v18, 32(%%r1,%[x])\n\t"
+    "vl  %%v19, 48(%%r1,%[x])\n\t"
+    "vl  %%v20, 64(%%r1,%[x])\n\t"
+    "vl  %%v21, 80(%%r1,%[x])\n\t"
+    "vl  %%v22, 96(%%r1,%[x])\n\t"
+    "vl  %%v23, 112(%%r1,%[x])\n\t"
+    "vfadb   %%v24,%%v24,%%v16\n\t"
+    "vfadb   %%v25,%%v25,%%v17\n\t"
+    "vfadb   %%v26,%%v26,%%v18\n\t"
+    "vfadb   %%v27,%%v27,%%v19\n\t"
+    "vfadb   %%v28,%%v28,%%v20\n\t"
+    "vfadb   %%v29,%%v29,%%v21\n\t"
+    "vfadb   %%v30,%%v30,%%v22\n\t"
+    "vfadb   %%v31,%%v31,%%v23\n\t"
+    "vl  %%v16, 128(%%r1,%[x])\n\t"
+    "vl  %%v17, 144(%%r1,%[x])\n\t"
+    "vl  %%v18, 160(%%r1,%[x])\n\t"
+    "vl  %%v19, 176(%%r1,%[x])\n\t"
+    "vl  %%v20, 192(%%r1,%[x])\n\t"
+    "vl  %%v21, 208(%%r1,%[x])\n\t"
+    "vl  %%v22, 224(%%r1,%[x])\n\t"
+    "vl  %%v23, 240(%%r1,%[x])\n\t"
+    "vfadb   %%v24,%%v24,%%v16\n\t"
+    "vfadb   %%v25,%%v25,%%v17\n\t"
+    "vfadb   %%v26,%%v26,%%v18\n\t"
+    "vfadb   %%v27,%%v27,%%v19\n\t"
+    "vfadb   %%v28,%%v28,%%v20\n\t"
+    "vfadb   %%v29,%%v29,%%v21\n\t"
+    "vfadb   %%v30,%%v30,%%v22\n\t"
+    "vfadb   %%v31,%%v31,%%v23\n\t"
+    "agfi  %%r1,256\n\t"
+    "brctg %[n],0b\n\t"
+    "vfadb   %%v24,%%v24,%%v25\n\t"
+    "vfadb   %%v24,%%v24,%%v26\n\t"
+    "vfadb   %%v24,%%v24,%%v27\n\t"
+    "vfadb   %%v24,%%v24,%%v28\n\t"
+    "vfadb   %%v24,%%v24,%%v29\n\t"
+    "vfadb   %%v24,%%v24,%%v30\n\t"
+    "vfadb   %%v24,%%v24,%%v31\n\t"
+    "vrepg   %%v25,%%v24,1\n\t"
+    "vfadb   %%v24,%%v24,%%v25\n\t"
+    "vsteg   %%v24,%[asum],0"
+    : [sum] "=Q"(sum),[n] "+&r"(n)
+    : "m"(*(const struct { FLOAT x[n * 2]; } *) x),[x] "a"(x)
+    : "cc", "r1", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23",
+       "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31");
+
+  return sum;
+}
+
+FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x) {
+  BLASLONG i = 0;
+  BLASLONG ip = 0;
+  FLOAT sumf = 0.0;
+  BLASLONG n1;
+  BLASLONG inc_x2;
+
+  if (n <= 0 || inc_x <= 0)
+    return (sumf);
+
+  if (inc_x == 1) {
+
+    n1 = n & -16;
+    if (n1 > 0) {
+
+      sumf = zsum_kernel_16(n1, x);
+      i = n1;
+      ip = 2 * n1;
+    }
+
+    while (i < n) {
+      sumf += x[ip] + x[ip + 1];
+      i++;
+      ip += 2;
+    }
+
+  } else {
+    inc_x2 = 2 * inc_x;
+
+    while (i < n) {
+      sumf += x[ip] + x[ip + 1];
+      ip += inc_x2;
+      i++;
+    }
+
+  }
+  return (sumf);
+}