Merge pull request #832 from wernsaar/develop

updated cgemm- and ctrmm-kernel for POWER8

Author: wernsaar

benchmark/Makefile

@@ -33,6 +33,9 @@ LIBMKL = -L$(MKL) -lmkl_intel_lp64 -lmkl_gnu_thread -lmkl_core -lgomp -lpthread
 # Apple vecLib
 LIBVECLIB = -framework Accelerate
 
+ESSL=/opt/ibm/lib
+LIBESSL = -lessl $(ESSL)/libxlomp_ser.so.1 $(ESSL)/libxlf90_r.so.1 $(ESSL)/libxlfmath.so.1 $(ESSL)/libxlsmp.so.1 /opt/ibm/xlC/13.1.2/lib/libxl.a
+
 ifeq ($(OSNAME), WINNT)
 
 goto :: slinpack.goto dlinpack.goto clinpack.goto zlinpack.goto \
@@ -255,7 +258,8 @@ mkl :: slinpack.mkl dlinpack.mkl clinpack.mkl zlinpack.mkl \
 
 endif
 
-
+essl :: sgemm.essl strmm.essl dgemm.essl dtrmm.essl  \
+	cgemm.essl ctrmm.essl zgemm.essl ztrmm.essl
 
 veclib :: slinpack.veclib dlinpack.veclib clinpack.veclib zlinpack.veclib \
        scholesky.veclib dcholesky.veclib ccholesky.veclib zcholesky.veclib \
@@ -443,6 +447,9 @@ sgemm.mkl : sgemm.$(SUFFIX)
 sgemm.veclib : sgemm.$(SUFFIX)
 	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
 
+sgemm.essl : sgemm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
 ##################################### Dgemm ####################################################
 dgemm.goto : dgemm.$(SUFFIX) ../$(LIBNAME)
 	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
@@ -459,6 +466,9 @@ dgemm.mkl : dgemm.$(SUFFIX)
 dgemm.veclib : dgemm.$(SUFFIX)
 	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
 
+dgemm.essl : dgemm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
 ##################################### Cgemm ####################################################
 
 cgemm.goto : cgemm.$(SUFFIX) ../$(LIBNAME)
@@ -476,6 +486,9 @@ cgemm.mkl : cgemm.$(SUFFIX)
 cgemm.veclib : cgemm.$(SUFFIX)
 	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
 
+cgemm.essl : cgemm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
 ##################################### Zgemm ####################################################
 
 zgemm.goto : zgemm.$(SUFFIX) ../$(LIBNAME)
@@ -493,6 +506,9 @@ zgemm.mkl : zgemm.$(SUFFIX)
 zgemm.veclib : zgemm.$(SUFFIX)
 	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
 
+zgemm.essl : zgemm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
 ##################################### Ssymm ####################################################
 ssymm.goto : ssymm.$(SUFFIX) ../$(LIBNAME)
 	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
@@ -575,6 +591,9 @@ strmm.mkl : strmm.$(SUFFIX)
 strmm.veclib : strmm.$(SUFFIX)
 	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
 
+strmm.essl : strmm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
 ##################################### Dtrmm ####################################################
 dtrmm.goto : dtrmm.$(SUFFIX) ../$(LIBNAME)
 	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
@@ -591,6 +610,9 @@ dtrmm.mkl : dtrmm.$(SUFFIX)
 dtrmm.veclib : dtrmm.$(SUFFIX)
 	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
 
+dtrmm.essl : dtrmm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
 ##################################### Ctrmm ####################################################
 
 ctrmm.goto : ctrmm.$(SUFFIX) ../$(LIBNAME)
@@ -608,6 +630,9 @@ ctrmm.mkl : ctrmm.$(SUFFIX)
 ctrmm.veclib : ctrmm.$(SUFFIX)
 	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
 
+ctrmm.essl : ctrmm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
 ##################################### Ztrmm ####################################################
 
 ztrmm.goto : ztrmm.$(SUFFIX) ../$(LIBNAME)
@@ -625,6 +650,9 @@ ztrmm.mkl : ztrmm.$(SUFFIX)
 ztrmm.veclib : ztrmm.$(SUFFIX)
 	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBVECLIB) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
 
+ztrmm.essl : ztrmm.$(SUFFIX)
+	-$(CC) $(CFLAGS) -o $(@F) $^ $(LIBESSL) $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB)
+
 ##################################### Strsm ####################################################
 strsm.goto : strsm.$(SUFFIX) ../$(LIBNAME)
 	$(CC) $(CFLAGS) -o $(@F) $^ $(CEXTRALIB) $(EXTRALIB) $(FEXTRALIB) -lm
@@ -2179,7 +2207,7 @@ smallscaling: smallscaling.c ../$(LIBNAME)
 	$(CC) $(CFLAGS) -o $(@F) $^ $(EXTRALIB) -fopenmp -lm
 
 clean ::
-	@rm -f *.goto *.mkl *.acml *.atlas *.veclib
+	@rm -f *.goto *.mkl *.acml *.atlas *.veclib *.essl
 
 include $(TOPDIR)/Makefile.tail
 

kernel/power/cgemm_kernel_8x4_power8.S

@@ -26,11 +26,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *****************************************************************************/
 
 /**************************************************************************************
-* 2016/03/18 Werner Saar (wernsaar@googlemail.com)
+* 2016/04/03 Werner Saar (wernsaar@googlemail.com)
 * 	 BLASTEST 		: OK
 * 	 CTEST			: OK
 * 	 TEST			: OK
-*	 LAPACK-TEST		: OK
+* 	 LAPACK-TEST		: OK
 **************************************************************************************/
 
 /*********************************************************************/
@@ -130,10 +130,14 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #endif
 
 #define o0	0
-#define alpha_r vs30
-#define alpha_i vs31
 
-#define TBUFFER	r14
+#define alpha_dr vs28
+#define alpha_di vs29
+#define alpha_sr vs30
+#define alpha_si vs31
+
+
+#define NOTUSED	r14
 #define L	r15
 #define o12	r16
 #define o4	r17
@@ -271,21 +275,20 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #include "cgemm_macros_8x4_power8.S"
 
 	cmpwi	cr0, M, 0
-	ble	.L999_H1
+	ble	L999_H1
 	cmpwi	cr0, N, 0
-	ble	.L999_H1
+	ble	L999_H1
 	cmpwi	cr0, K, 0
-	ble	.L999_H1
+	ble	L999_H1
 
 	slwi	LDC, LDC, ZBASE_SHIFT
-	li	PRE, 256 
+	li	PRE, 384 
 	li	o4  , 4
 	li	o8  , 8
 	li	o12 , 12
 	li	o16 , 16
 	li	o32 , 32
 	li	o48 , 48
-	addi	TBUFFER, SP, 360
 
 
 #ifdef __64BIT__
@@ -294,14 +297,31 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 	addi	T1 , SP, 224
 #endif
 
-	lxsspx	alpha_r, 0, T1
-	lxsspx	alpha_i, o8, T1
+	stxsspx vs1,  0, T1
+        lxsspx  alpha_dr, 0, T1
+	stxsspx vs2,  o8  , T1
+        lxsspx  alpha_di, o8, T1
+        addi    T1, SP, 360
+        li      T2, 0
+
+        stw             T2, 0(T1)
+        stw             T2, 4(T1)
+        stw             T2, 8(T1)
+        stxsspx         alpha_dr, o12, T1
+        lxvw4x          alpha_sr, o0 , T1
+        addi            T1, T1, 16
+
+        stw             T2, 0(T1)
+        stw             T2, 4(T1)
+        stw             T2, 8(T1)
+        stxsspx         alpha_di, o12, T1
+        lxvw4x          alpha_si, o0 , T1
 
 	.align 5
 
 #include "cgemm_logic_8x4_power8.S"
 
-.L999:
+L999:
 	addi	r3, 0, 0
 
 	lfd	f14,    0(SP)