From e7c969e164900ed19461566af1a2201f98bc0a36 Mon Sep 17 00:00:00 2001
From: Werner Saar <wernsaar@googlemail.com>
Date: Sat, 13 Jun 2015 16:16:29 +0200
Subject: [PATCH] added optimized dtrmm_kernel for haswell

---
 kernel/x86_64/KERNEL.HASWELL             |    2 +-
 kernel/x86_64/dtrmm_kernel_4x8_haswell.c | 1546 ++++++++++++++++++++++++++++++
 2 files changed, 1547 insertions(+), 1 deletion(-)
 create mode 100644 kernel/x86_64/dtrmm_kernel_4x8_haswell.c

diff --git a/kernel/x86_64/KERNEL.HASWELL b/kernel/x86_64/KERNEL.HASWELL
index a01dc1a..a4686de 100644
--- a/kernel/x86_64/KERNEL.HASWELL
+++ b/kernel/x86_64/KERNEL.HASWELL
@@ -40,7 +40,7 @@ SGEMMITCOPYOBJ =  sgemm_itcopy$(TSUFFIX).$(SUFFIX)
 SGEMMONCOPYOBJ =  sgemm_oncopy$(TSUFFIX).$(SUFFIX)
 SGEMMOTCOPYOBJ =  sgemm_otcopy$(TSUFFIX).$(SUFFIX)
 
-DTRMMKERNEL    =  ../generic/trmmkernel_4x8.c
+DTRMMKERNEL    =  dtrmm_kernel_4x8_haswell.c
 DGEMMKERNEL    =  dgemm_kernel_4x8_haswell.S
 DGEMMINCOPY    =  ../generic/gemm_ncopy_4.c
 DGEMMITCOPY    =  ../generic/gemm_tcopy_4.c
diff --git a/kernel/x86_64/dtrmm_kernel_4x8_haswell.c b/kernel/x86_64/dtrmm_kernel_4x8_haswell.c
new file mode 100644
index 0000000..504c784
--- /dev/null
+++ b/kernel/x86_64/dtrmm_kernel_4x8_haswell.c
@@ -0,0 +1,1546 @@
+#include "common.h"
+#include <stdbool.h>
+
+
+static void dtrmm_kernel_4x8( BLASLONG n, FLOAT *alpha ,FLOAT *a, FLOAT *b, FLOAT *C0, FLOAT *C1, FLOAT *C2,FLOAT *C3, FLOAT *C4, FLOAT *C5,FLOAT *C6, FLOAT *C7) __attribute__ ((noinline));
+
+static void dtrmm_kernel_4x8( BLASLONG n, FLOAT *alpha ,FLOAT *a, FLOAT *b, FLOAT *C0, FLOAT *C1, FLOAT *C2,FLOAT *C3, FLOAT *C4, FLOAT *C5,FLOAT *C6, FLOAT *C7)
+{
+
+		BLASLONG I = 0;
+		BLASLONG temp1 = n * 8;
+
+		 __asm__  __volatile__
+        	(
+		"	vxorpd	%%ymm4 , %%ymm4 , %%ymm4			\n\t"
+		"	vxorpd	%%ymm5 , %%ymm5 , %%ymm5			\n\t"
+		"	vxorpd	%%ymm6 , %%ymm6 , %%ymm6			\n\t"
+		"	vxorpd	%%ymm7 , %%ymm7 , %%ymm7			\n\t"
+		"	vxorpd	%%ymm8 , %%ymm8 , %%ymm8			\n\t"
+		"	vxorpd	%%ymm9 , %%ymm9 , %%ymm9			\n\t"
+		"	vxorpd	%%ymm10, %%ymm10, %%ymm10			\n\t"
+		"	vxorpd	%%ymm11, %%ymm11, %%ymm11			\n\t"
+
+		"	cmp $0, %1						\n\t"
+		"	jz 2f							\n\t"
+
+		"	.align 16						\n\t"
+		"1:								\n\t"
+		"	vmovups   	(%2,%0,4) , %%ymm0			\n\t"
+		"	vmovups   	(%3,%0,8) , %%ymm1			\n\t"
+		"	vmovups       32(%3,%0,8) , %%ymm2			\n\t"
+
+		"	vfmadd231pd	%%ymm0 , %%ymm1 , %%ymm4		\n\t"
+		"	vfmadd231pd	%%ymm0 , %%ymm2 , %%ymm8		\n\t"
+
+		"	vpermpd         $0xb1  , %%ymm0 , %%ymm0		\n\t"
+		"	vfmadd231pd	%%ymm0 , %%ymm1 , %%ymm5		\n\t"
+		"	vfmadd231pd	%%ymm0 , %%ymm2 , %%ymm9		\n\t"
+
+		"	vpermpd         $0x1b  , %%ymm0 , %%ymm0		\n\t"
+		"	vfmadd231pd	%%ymm0 , %%ymm1 , %%ymm6		\n\t"
+		"	vfmadd231pd	%%ymm0 , %%ymm2 , %%ymm10		\n\t"
+
+		"	vpermpd         $0xb1  , %%ymm0 , %%ymm0		\n\t"
+		"	vfmadd231pd	%%ymm0 , %%ymm1 , %%ymm7		\n\t"
+		"	vfmadd231pd	%%ymm0 , %%ymm2 , %%ymm11		\n\t"
+
+		"	addq	$8 , %0						\n\t"
+		"	cmp	%0 , %1						\n\t"
+		"	jne	1b						\n\t"
+
+		"2:								\n\t"
+
+		"	vbroadcastsd	(%4), %%ymm0				\n\t"
+
+		"	vmulpd		%%ymm0 , %%ymm4 , %%ymm4		\n\t"
+		"	vmulpd		%%ymm0 , %%ymm5 , %%ymm5		\n\t"
+		"	vmulpd		%%ymm0 , %%ymm6 , %%ymm6		\n\t"
+		"	vmulpd		%%ymm0 , %%ymm7 , %%ymm7		\n\t"
+		"	vmulpd		%%ymm0 , %%ymm8 , %%ymm8		\n\t"
+		"	vmulpd		%%ymm0 , %%ymm9 , %%ymm9		\n\t"
+		"	vmulpd		%%ymm0 , %%ymm10, %%ymm10		\n\t"
+		"	vmulpd		%%ymm0 , %%ymm11, %%ymm11		\n\t"
+
+		"	vpermpd 	$0xb1  , %%ymm5 , %%ymm5		\n\t"
+		"	vpermpd 	$0xb1  , %%ymm7 , %%ymm7		\n\t"
+
+		"	vblendpd 	$0x0a  , %%ymm5 , %%ymm4 , %%ymm0	\n\t"
+		"	vblendpd 	$0x05  , %%ymm5 , %%ymm4 , %%ymm1	\n\t"
+		"	vblendpd 	$0x0a  , %%ymm7 , %%ymm6 , %%ymm2	\n\t"
+		"	vblendpd 	$0x05  , %%ymm7 , %%ymm6 , %%ymm3	\n\t"
+
+		"	vpermpd 	$0x1b  , %%ymm2 , %%ymm2		\n\t"
+		"	vpermpd 	$0x1b  , %%ymm3 , %%ymm3		\n\t"
+		"	vpermpd 	$0xb1  , %%ymm2 , %%ymm2		\n\t"
+		"	vpermpd 	$0xb1  , %%ymm3 , %%ymm3		\n\t"
+
+		"	vblendpd 	$0x03  , %%ymm0 , %%ymm2 , %%ymm4	\n\t"
+		"	vblendpd 	$0x03  , %%ymm1 , %%ymm3 , %%ymm5	\n\t"
+		"	vblendpd 	$0x03  , %%ymm2 , %%ymm0 , %%ymm6	\n\t"
+		"	vblendpd 	$0x03  , %%ymm3 , %%ymm1 , %%ymm7	\n\t"
+
+		"	vmovups		%%ymm4 , (%5)				\n\t"
+		"	vmovups		%%ymm5 , (%6)				\n\t"
+		"	vmovups		%%ymm6 , (%7)				\n\t"
+		"	vmovups		%%ymm7 , (%8)				\n\t"
+
+		"	vpermpd 	$0xb1  , %%ymm9 , %%ymm9		\n\t"
+		"	vpermpd 	$0xb1  , %%ymm11, %%ymm11		\n\t"
+
+		"	vblendpd 	$0x0a  , %%ymm9 , %%ymm8 , %%ymm0	\n\t"
+		"	vblendpd 	$0x05  , %%ymm9 , %%ymm8 , %%ymm1	\n\t"
+		"	vblendpd 	$0x0a  , %%ymm11, %%ymm10, %%ymm2	\n\t"
+		"	vblendpd 	$0x05  , %%ymm11, %%ymm10, %%ymm3	\n\t"
+
+		"	vpermpd 	$0x1b  , %%ymm2 , %%ymm2		\n\t"
+		"	vpermpd 	$0x1b  , %%ymm3 , %%ymm3		\n\t"
+		"	vpermpd 	$0xb1  , %%ymm2 , %%ymm2		\n\t"
+		"	vpermpd 	$0xb1  , %%ymm3 , %%ymm3		\n\t"
+
+		"	vblendpd 	$0x03  , %%ymm0 , %%ymm2 , %%ymm4	\n\t"
+		"	vblendpd 	$0x03  , %%ymm1 , %%ymm3 , %%ymm5	\n\t"
+		"	vblendpd 	$0x03  , %%ymm2 , %%ymm0 , %%ymm6	\n\t"
+		"	vblendpd 	$0x03  , %%ymm3 , %%ymm1 , %%ymm7	\n\t"
+
+		"	vmovups		%%ymm4 , (%9)				\n\t"
+		"	vmovups		%%ymm5 , (%10)				\n\t"
+		"	vmovups		%%ymm6 , (%11)				\n\t"
+		"	vmovups		%%ymm7 , (%12)				\n\t"
+
+	        :
+        	:
+		"a" (I),	 // 0
+          	"r" (temp1),     // 1    
+          	"S" (a),         // 2
+          	"D" (b),         // 3
+          	"r" (alpha),     // 4
+		"r" (C0),	 // 5
+		"r" (C1),	 // 6
+		"r" (C2),	 // 7
+		"r" (C3),	 // 8
+		"r" (C4),	 // 9
+		"r" (C5),	 // 10
+		"r" (C6),	 // 11
+		"r" (C7) 	 // 12
+		: "cc",
+          	"%xmm0", "%xmm1", "%xmm2", "%xmm3",
+          	"%xmm4", "%xmm5", "%xmm6", "%xmm7",
+          	"%xmm8", "%xmm9", "%xmm10", "%xmm11",
+          	"%xmm12", "%xmm13", "%xmm14", "%xmm15",
+          	"memory"
+        	);
+
+
+
+
+}
+
+
+
+
+int CNAME(BLASLONG bm,BLASLONG bn,BLASLONG bk,FLOAT alpha,FLOAT* ba,FLOAT* bb,FLOAT* C,BLASLONG ldc ,BLASLONG offset)
+{
+
+   BLASLONG i,j,k;
+   FLOAT *C0,*C1,*C2,*C3,*C4,*C5,*C6,*C7,*ptrba,*ptrbb;
+
+   FLOAT res0_0;
+   FLOAT res0_1;
+   FLOAT res0_2;
+   FLOAT res0_3;
+
+   FLOAT res1_0;
+   FLOAT res1_1;
+   FLOAT res1_2;
+   FLOAT res1_3;
+
+   FLOAT res2_0;
+   FLOAT res2_1;
+   FLOAT res2_2;
+   FLOAT res2_3;
+
+   FLOAT res3_0;
+   FLOAT res3_1;
+   FLOAT res3_2;
+   FLOAT res3_3;
+
+   FLOAT res4_0;
+   FLOAT res4_1;
+   FLOAT res4_2;
+   FLOAT res4_3;
+
+   FLOAT res5_0;
+   FLOAT res5_1;
+   FLOAT res5_2;
+   FLOAT res5_3;
+
+   FLOAT res6_0;
+   FLOAT res6_1;
+   FLOAT res6_2;
+   FLOAT res6_3;
+
+   FLOAT res7_0;
+   FLOAT res7_1;
+   FLOAT res7_2;
+   FLOAT res7_3;
+
+   FLOAT a0;
+   FLOAT a1;
+
+   FLOAT b0;
+   FLOAT b1;
+   FLOAT b2;
+   FLOAT b3;
+   FLOAT b4;
+   FLOAT b5;
+   FLOAT b6;
+   FLOAT b7;
+
+   BLASLONG off, temp ;
+
+   bool left;
+   bool transposed;
+   bool backwards;
+
+#ifdef LEFT
+   left = true;
+#else
+   left = false;
+#endif
+
+#ifdef TRANSA
+   transposed = true;
+#else
+   transposed = false;
+#endif
+
+   backwards = left != transposed;
+
+   if (!left) {
+      off = -offset;
+   }
+
+
+   for (j=0; j<bn/8; j+=1) // do blocks of the Mx8 loops 
+   {
+        C0 = C;
+        C1 = C0+ldc;
+        C2 = C1+ldc;
+        C3 = C2+ldc;
+        C4 = C3+ldc;
+        C5 = C4+ldc;
+        C6 = C5+ldc;
+        C7 = C6+ldc;
+
+
+        if (left) {
+            off = offset;
+        }
+
+        ptrba = ba;
+
+        for (i=0; i<bm/4; i+=1) // do blocks of 4x4
+	{
+
+		ptrbb = bb;
+                if (backwards)
+                {
+		   ptrba += off*4; // number of values in A
+		   ptrbb += off*8; // number of values in B
+                }
+/*
+		res0_0 = 0;
+		res0_1 = 0;
+		res0_2 = 0;
+		res0_3 = 0;
+
+		res1_0 = 0;
+		res1_1 = 0;
+		res1_2 = 0;
+		res1_3 = 0;
+
+		res2_0 = 0;
+		res2_1 = 0;
+		res2_2 = 0;
+		res2_3 = 0;
+
+		res3_0 = 0;
+		res3_1 = 0;
+		res3_2 = 0;
+		res3_3 = 0;
+
+		res4_0 = 0;
+		res4_1 = 0;
+		res4_2 = 0;
+		res4_3 = 0;
+
+		res5_0 = 0;
+		res5_1 = 0;
+		res5_2 = 0;
+		res5_3 = 0;
+
+		res6_0 = 0;
+		res6_1 = 0;
+		res6_2 = 0;
+		res6_3 = 0;
+
+		res7_0 = 0;
+		res7_1 = 0;
+		res7_2 = 0;
+		res7_3 = 0;
+*/
+                temp = backwards ? bk-off :
+                             left ? off + 4 : // number of values in A
+                                    off + 8;  // number of values in B
+
+	        dtrmm_kernel_4x8( temp, &alpha , ptrba, ptrbb, C0, C1, C2, C3, C4, C5, C6, C7);
+
+		ptrba = ptrba + temp * 4;
+		ptrbb = ptrbb + temp * 8;
+
+/*
+		for (k=0; k<temp; k++)
+                {
+			b0 = ptrbb[0];
+			b1 = ptrbb[1];
+			b2 = ptrbb[2];
+			b3 = ptrbb[3];
+			b4 = ptrbb[4];
+			b5 = ptrbb[5];
+			b6 = ptrbb[6];
+			b7 = ptrbb[7];
+
+			a0 = ptrba[0];
+			res0_0 += a0*b0;
+			res1_0 += a0*b1;
+			res2_0 += a0*b2;
+			res3_0 += a0*b3;
+			res4_0 += a0*b4;
+			res5_0 += a0*b5;
+			res6_0 += a0*b6;
+			res7_0 += a0*b7;
+
+			a1 = ptrba[1];
+			res0_1 += a1*b0;
+			res1_1 += a1*b1;
+			res2_1 += a1*b2;
+			res3_1 += a1*b3;
+			res4_1 += a1*b4;
+			res5_1 += a1*b5;
+			res6_1 += a1*b6;
+			res7_1 += a1*b7;
+
+			a0 = ptrba[2];
+			res0_2 += a0*b0;
+			res1_2 += a0*b1;
+			res2_2 += a0*b2;
+			res3_2 += a0*b3;
+			res4_2 += a0*b4;
+			res5_2 += a0*b5;
+			res6_2 += a0*b6;
+			res7_2 += a0*b7;
+
+			a1 = ptrba[3];
+			res0_3 += a1*b0;
+			res1_3 += a1*b1;
+			res2_3 += a1*b2;
+			res3_3 += a1*b3;
+			res4_3 += a1*b4;
+			res5_3 += a1*b5;
+			res6_3 += a1*b6;
+			res7_3 += a1*b7;
+
+			ptrba = ptrba+4;
+			ptrbb = ptrbb+8;
+                }
+
+		res0_0 *= alpha;
+		res0_1 *= alpha;
+		res0_2 *= alpha;
+		res0_3 *= alpha;
+
+		res1_0 *= alpha;
+		res1_1 *= alpha;
+		res1_2 *= alpha;
+		res1_3 *= alpha;
+
+		res2_0 *= alpha;
+		res2_1 *= alpha;
+		res2_2 *= alpha;
+		res2_3 *= alpha;
+
+		res3_0 *= alpha;
+		res3_1 *= alpha;
+		res3_2 *= alpha;
+		res3_3 *= alpha;
+
+		res4_0 *= alpha;
+		res4_1 *= alpha;
+		res4_2 *= alpha;
+		res4_3 *= alpha;
+
+		res5_0 *= alpha;
+		res5_1 *= alpha;
+		res5_2 *= alpha;
+		res5_3 *= alpha;
+
+		res6_0 *= alpha;
+		res6_1 *= alpha;
+		res6_2 *= alpha;
+		res6_3 *= alpha;
+
+		res7_0 *= alpha;
+		res7_1 *= alpha;
+		res7_2 *= alpha;
+		res7_3 *= alpha;
+
+
+		C0[0] = res0_0;
+		C0[1] = res0_1;
+		C0[2] = res0_2;
+		C0[3] = res0_3;
+
+		C1[0] = res1_0;
+		C1[1] = res1_1;
+		C1[2] = res1_2;
+		C1[3] = res1_3;
+
+		C2[0] = res2_0;
+		C2[1] = res2_1;
+		C2[2] = res2_2;
+		C2[3] = res2_3;
+
+		C3[0] = res3_0;
+		C3[1] = res3_1;
+		C3[2] = res3_2;
+		C3[3] = res3_3;
+
+		C4[0] = res4_0;
+		C4[1] = res4_1;
+		C4[2] = res4_2;
+		C4[3] = res4_3;
+
+		C5[0] = res5_0;
+		C5[1] = res5_1;
+		C5[2] = res5_2;
+		C5[3] = res5_3;
+
+		C6[0] = res6_0;
+		C6[1] = res6_1;
+		C6[2] = res6_2;
+		C6[3] = res6_3;
+
+		C7[0] = res7_0;
+		C7[1] = res7_1;
+		C7[2] = res7_2;
+		C7[3] = res7_3;
+*/
+		if (!backwards) {
+                    temp = bk-off;
+                    temp = left ? temp - 4 : // number of values in A
+                                  temp - 8;  // number of values in B
+
+                    ptrba += temp*4; // number of values in A
+		    ptrbb += temp*8; // number of values in B
+                }
+#ifdef LEFT
+		off += 4; // number of values in A
+#endif
+
+		C0 = C0+4;
+		C1 = C1+4;
+		C2 = C2+4;
+		C3 = C3+4;
+		C4 = C4+4;
+		C5 = C5+4;
+		C6 = C6+4;
+		C7 = C7+4;
+
+	}
+
+	if ( bm & 2 ) // do any 2x4 loop
+	{
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+		ptrbb = bb;
+#else
+		ptrba += off*2;
+		ptrbb = bb + off*8;
+#endif
+
+		res0_0 = 0;
+		res0_1 = 0;
+
+		res1_0 = 0;
+		res1_1 = 0;
+
+		res2_0 = 0;
+		res2_1 = 0;
+
+		res3_0 = 0;
+		res3_1 = 0;
+
+		res4_0 = 0;
+		res4_1 = 0;
+
+		res5_0 = 0;
+		res5_1 = 0;
+
+		res6_0 = 0;
+		res6_1 = 0;
+
+		res7_0 = 0;
+		res7_1 = 0;
+
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+		temp = bk-off;
+#elif defined(LEFT)
+		temp = off+2;	// number of values in A
+#else
+		temp = off+8;	// number of values in B
+#endif
+
+		for (k=0; k<temp; k++)
+                {
+			b0 = ptrbb[0];
+			b1 = ptrbb[1];
+			b2 = ptrbb[2];
+			b3 = ptrbb[3];
+			b4 = ptrbb[4];
+			b5 = ptrbb[5];
+			b6 = ptrbb[6];
+			b7 = ptrbb[7];
+
+			a0 = ptrba[0];
+			res0_0 += a0*b0;
+			res1_0 += a0*b1;
+			res2_0 += a0*b2;
+			res3_0 += a0*b3;
+			res4_0 += a0*b4;
+			res5_0 += a0*b5;
+			res6_0 += a0*b6;
+			res7_0 += a0*b7;
+
+			a1 = ptrba[1];
+			res0_1 += a1*b0;
+			res1_1 += a1*b1;
+			res2_1 += a1*b2;
+			res3_1 += a1*b3;
+			res4_1 += a1*b4;
+			res5_1 += a1*b5;
+			res6_1 += a1*b6;
+			res7_1 += a1*b7;
+
+			ptrba = ptrba+2;
+			ptrbb = ptrbb+8;
+                }
+
+		res0_0 *= alpha;
+		res0_1 *= alpha;
+
+		res1_0 *= alpha;
+		res1_1 *= alpha;
+
+		res2_0 *= alpha;
+		res2_1 *= alpha;
+
+		res3_0 *= alpha;
+		res3_1 *= alpha;
+
+		res4_0 *= alpha;
+		res4_1 *= alpha;
+
+		res5_0 *= alpha;
+		res5_1 *= alpha;
+
+		res6_0 *= alpha;
+		res6_1 *= alpha;
+
+		res7_0 *= alpha;
+		res7_1 *= alpha;
+
+		C0[0] = res0_0;
+		C0[1] = res0_1;
+
+		C1[0] = res1_0;
+		C1[1] = res1_1;
+
+		C2[0] = res2_0;
+		C2[1] = res2_1;
+
+		C3[0] = res3_0;
+		C3[1] = res3_1;
+
+		C4[0] = res4_0;
+		C4[1] = res4_1;
+
+		C5[0] = res5_0;
+		C5[1] = res5_1;
+
+		C6[0] = res6_0;
+		C6[1] = res6_1;
+
+		C7[0] = res7_0;
+		C7[1] = res7_1;
+
+
+#if ( defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+		temp = bk - off;
+#ifdef LEFT
+		temp -= 2; // number of values in A
+#else
+		temp -= 8; // number of values in B
+#endif
+		ptrba += temp*2;
+		ptrbb += temp*8;
+#endif
+
+#ifdef LEFT
+		off += 2; // number of values in A
+#endif
+
+		C0 = C0+2;
+		C1 = C1+2;
+		C2 = C2+2;
+		C3 = C3+2;
+		C4 = C4+2;
+		C5 = C5+2;
+		C6 = C6+2;
+		C7 = C7+2;
+
+	}
+
+	if ( bm & 1 ) // do any 1x4 loop
+	{
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+		ptrbb = bb;
+#else
+		ptrba += off*1;
+		ptrbb = bb + off*8;
+#endif
+
+		res0_0 = 0;
+		res1_0 = 0;
+		res2_0 = 0;
+		res3_0 = 0;
+		res4_0 = 0;
+		res5_0 = 0;
+		res6_0 = 0;
+		res7_0 = 0;
+
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+		temp = bk-off;
+#elif defined(LEFT)
+		temp = off+1;	// number of values in A
+#else
+		temp = off+8;	// number of values in B
+#endif
+
+		for (k=0; k<temp; k++)
+                {
+			b0 = ptrbb[0];
+			b1 = ptrbb[1];
+			b2 = ptrbb[2];
+			b3 = ptrbb[3];
+			b4 = ptrbb[4];
+			b5 = ptrbb[5];
+			b6 = ptrbb[6];
+			b7 = ptrbb[7];
+
+			a0 = ptrba[0];
+			res0_0 += a0*b0;
+			res1_0 += a0*b1;
+			res2_0 += a0*b2;
+			res3_0 += a0*b3;
+			res4_0 += a0*b4;
+			res5_0 += a0*b5;
+			res6_0 += a0*b6;
+			res7_0 += a0*b7;
+
+			ptrba = ptrba+1;
+			ptrbb = ptrbb+8;
+                }
+
+		res0_0 *= alpha;
+
+		res1_0 *= alpha;
+
+		res2_0 *= alpha;
+
+		res3_0 *= alpha;
+		res4_0 *= alpha;
+		res5_0 *= alpha;
+		res6_0 *= alpha;
+		res7_0 *= alpha;
+
+		C0[0] = res0_0;
+
+		C1[0] = res1_0;
+
+		C2[0] = res2_0;
+
+		C3[0] = res3_0;
+		C4[0] = res4_0;
+		C5[0] = res5_0;
+		C6[0] = res6_0;
+		C7[0] = res7_0;
+
+
+#if ( defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+		temp = bk - off;
+#ifdef LEFT
+		temp -= 1; // number of values in A
+#else
+		temp -= 8; // number of values in B
+#endif
+		ptrba += temp*1;
+		ptrbb += temp*8;
+#endif
+
+#ifdef LEFT
+		off += 1; // number of values in A
+#endif
+
+		C0 = C0+1;
+		C1 = C1+1;
+		C2 = C2+1;
+		C3 = C3+1;
+		C4 = C4+1;
+		C5 = C5+1;
+		C6 = C6+1;
+		C7 = C7+1;
+
+	}
+
+
+#if defined(TRMMKERNEL) && !defined(LEFT)
+		off += 8;
+#endif
+
+        k = (bk<<3);
+        bb = bb+k;
+        i = (ldc<<3);
+        C = C+i;
+    }
+
+
+
+   for (j=0; j<(bn&4); j+=4) // do blocks of the Mx4 loops 
+   {
+        C0 = C;
+        C1 = C0+ldc;
+        C2 = C1+ldc;
+        C3 = C2+ldc;
+
+
+        if (left) {
+            off = offset;
+        }
+
+        ptrba = ba;
+
+        for (i=0; i<bm/4; i+=1) // do blocks of 4x4
+	{
+
+		ptrbb = bb;
+                if (backwards)
+                {
+		   ptrba += off*4; // number of values in A
+		   ptrbb += off*4; // number of values in B
+                }
+
+		res0_0 = 0;
+		res0_1 = 0;
+		res0_2 = 0;
+		res0_3 = 0;
+
+		res1_0 = 0;
+		res1_1 = 0;
+		res1_2 = 0;
+		res1_3 = 0;
+
+		res2_0 = 0;
+		res2_1 = 0;
+		res2_2 = 0;
+		res2_3 = 0;
+
+		res3_0 = 0;
+		res3_1 = 0;
+		res3_2 = 0;
+		res3_3 = 0;
+
+                temp = backwards ? bk-off :
+                             left ? off + 4 : // number of values in A
+                                    off + 4;  // number of values in B
+
+		for (k=0; k<temp; k++)
+                {
+			b0 = ptrbb[0];
+			b1 = ptrbb[1];
+			b2 = ptrbb[2];
+			b3 = ptrbb[3];
+
+			a0 = ptrba[0];
+			res0_0 += a0*b0;
+			res1_0 += a0*b1;
+			res2_0 += a0*b2;
+			res3_0 += a0*b3;
+
+			a1 = ptrba[1];
+			res0_1 += a1*b0;
+			res1_1 += a1*b1;
+			res2_1 += a1*b2;
+			res3_1 += a1*b3;
+
+			a0 = ptrba[2];
+			res0_2 += a0*b0;
+			res1_2 += a0*b1;
+			res2_2 += a0*b2;
+			res3_2 += a0*b3;
+
+			a1 = ptrba[3];
+			res0_3 += a1*b0;
+			res1_3 += a1*b1;
+			res2_3 += a1*b2;
+			res3_3 += a1*b3;
+
+			ptrba = ptrba+4;
+			ptrbb = ptrbb+4;
+                }
+
+		res0_0 *= alpha;
+		res0_1 *= alpha;
+		res0_2 *= alpha;
+		res0_3 *= alpha;
+
+		res1_0 *= alpha;
+		res1_1 *= alpha;
+		res1_2 *= alpha;
+		res1_3 *= alpha;
+
+		res2_0 *= alpha;
+		res2_1 *= alpha;
+		res2_2 *= alpha;
+		res2_3 *= alpha;
+
+		res3_0 *= alpha;
+		res3_1 *= alpha;
+		res3_2 *= alpha;
+		res3_3 *= alpha;
+
+		C0[0] = res0_0;
+		C0[1] = res0_1;
+		C0[2] = res0_2;
+		C0[3] = res0_3;
+
+		C1[0] = res1_0;
+		C1[1] = res1_1;
+		C1[2] = res1_2;
+		C1[3] = res1_3;
+
+		C2[0] = res2_0;
+		C2[1] = res2_1;
+		C2[2] = res2_2;
+		C2[3] = res2_3;
+
+		C3[0] = res3_0;
+		C3[1] = res3_1;
+		C3[2] = res3_2;
+		C3[3] = res3_3;
+
+		if (!backwards) {
+                    temp = bk-off;
+                    temp = left ? temp - 4 : // number of values in A
+                                  temp - 4;  // number of values in B
+
+                    ptrba += temp*4; // number of values in A
+		    ptrbb += temp*4; // number of values in B
+                }
+#ifdef LEFT
+		off += 4; // number of values in A
+#endif
+
+		C0 = C0+4;
+		C1 = C1+4;
+		C2 = C2+4;
+		C3 = C3+4;
+
+	}
+
+	if ( bm & 2 ) // do any 2x4 loop
+	{
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+		ptrbb = bb;
+#else
+		ptrba += off*2;
+		ptrbb = bb + off*4;
+#endif
+
+		res0_0 = 0;
+		res0_1 = 0;
+
+		res1_0 = 0;
+		res1_1 = 0;
+
+		res2_0 = 0;
+		res2_1 = 0;
+
+		res3_0 = 0;
+		res3_1 = 0;
+
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+		temp = bk-off;
+#elif defined(LEFT)
+		temp = off+2;	// number of values in A
+#else
+		temp = off+4;	// number of values in B
+#endif
+
+		for (k=0; k<temp; k++)
+                {
+			b0 = ptrbb[0];
+			b1 = ptrbb[1];
+			b2 = ptrbb[2];
+			b3 = ptrbb[3];
+
+			a0 = ptrba[0];
+			res0_0 += a0*b0;
+			res1_0 += a0*b1;
+			res2_0 += a0*b2;
+			res3_0 += a0*b3;
+
+			a1 = ptrba[1];
+			res0_1 += a1*b0;
+			res1_1 += a1*b1;
+			res2_1 += a1*b2;
+			res3_1 += a1*b3;
+
+			ptrba = ptrba+2;
+			ptrbb = ptrbb+4;
+                }
+
+		res0_0 *= alpha;
+		res0_1 *= alpha;
+
+		res1_0 *= alpha;
+		res1_1 *= alpha;
+
+		res2_0 *= alpha;
+		res2_1 *= alpha;
+
+		res3_0 *= alpha;
+		res3_1 *= alpha;
+
+		C0[0] = res0_0;
+		C0[1] = res0_1;
+
+		C1[0] = res1_0;
+		C1[1] = res1_1;
+
+		C2[0] = res2_0;
+		C2[1] = res2_1;
+
+		C3[0] = res3_0;
+		C3[1] = res3_1;
+
+
+#if ( defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+		temp = bk - off;
+#ifdef LEFT
+		temp -= 2; // number of values in A
+#else
+		temp -= 4; // number of values in B
+#endif
+		ptrba += temp*2;
+		ptrbb += temp*4;
+#endif
+
+#ifdef LEFT
+		off += 2; // number of values in A
+#endif
+
+		C0 = C0+2;
+		C1 = C1+2;
+		C2 = C2+2;
+		C3 = C3+2;
+
+	}
+
+	if ( bm & 1 ) // do any 1x4 loop
+	{
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+		ptrbb = bb;
+#else
+		ptrba += off*1;
+		ptrbb = bb + off*4;
+#endif
+
+		res0_0 = 0;
+		res1_0 = 0;
+		res2_0 = 0;
+		res3_0 = 0;
+
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+		temp = bk-off;
+#elif defined(LEFT)
+		temp = off+1;	// number of values in A
+#else
+		temp = off+4;	// number of values in B
+#endif
+
+		for (k=0; k<temp; k++)
+                {
+			b0 = ptrbb[0];
+			b1 = ptrbb[1];
+			b2 = ptrbb[2];
+			b3 = ptrbb[3];
+
+			a0 = ptrba[0];
+			res0_0 += a0*b0;
+			res1_0 += a0*b1;
+			res2_0 += a0*b2;
+			res3_0 += a0*b3;
+
+			ptrba = ptrba+1;
+			ptrbb = ptrbb+4;
+                }
+
+		res0_0 *= alpha;
+
+		res1_0 *= alpha;
+
+		res2_0 *= alpha;
+
+		res3_0 *= alpha;
+
+		C0[0] = res0_0;
+
+		C1[0] = res1_0;
+
+		C2[0] = res2_0;
+
+		C3[0] = res3_0;
+
+
+#if ( defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+		temp = bk - off;
+#ifdef LEFT
+		temp -= 1; // number of values in A
+#else
+		temp -= 4; // number of values in B
+#endif
+		ptrba += temp*1;
+		ptrbb += temp*4;
+#endif
+
+#ifdef LEFT
+		off += 1; // number of values in A
+#endif
+
+		C0 = C0+1;
+		C1 = C1+1;
+		C2 = C2+1;
+		C3 = C3+1;
+
+	}
+
+
+#if defined(TRMMKERNEL) && !defined(LEFT)
+		off += 4;
+#endif
+
+        k = (bk<<2);
+        bb = bb+k;
+        i = (ldc<<2);
+        C = C+i;
+    }
+
+
+
+   for (j=0; j<(bn&2); j+=2) // do the Mx2 loops 
+   {
+        C0 = C;
+        C1 = C0+ldc;
+
+#if defined(TRMMKERNEL) && defined(LEFT)
+		off = offset;
+#endif
+
+
+        ptrba = ba;
+
+        for (i=0; i<bm/4; i+=1) // do blocks of 4x2
+	{
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+		ptrbb = bb;
+#else
+		ptrba += off*4;
+		ptrbb = bb + off*2;
+#endif
+
+		res0_0 = 0;
+		res0_1 = 0;
+		res0_2 = 0;
+		res0_3 = 0;
+
+		res1_0 = 0;
+		res1_1 = 0;
+		res1_2 = 0;
+		res1_3 = 0;
+
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+		temp = bk-off;
+#elif defined(LEFT)
+		temp = off+4;	// number of values in A
+#else
+		temp = off+2;	// number of values in B
+#endif
+
+		for (k=0; k<temp; k++)
+                {
+			b0 = ptrbb[0];
+			b1 = ptrbb[1];
+
+			a0 = ptrba[0];
+			res0_0 += a0*b0;
+			res1_0 += a0*b1;
+
+			a1 = ptrba[1];
+			res0_1 += a1*b0;
+			res1_1 += a1*b1;
+
+			a0 = ptrba[2];
+			res0_2 += a0*b0;
+			res1_2 += a0*b1;
+
+			a1 = ptrba[3];
+			res0_3 += a1*b0;
+			res1_3 += a1*b1;
+
+			ptrba = ptrba+4;
+			ptrbb = ptrbb+2;
+                }
+
+		res0_0 *= alpha;
+		res0_1 *= alpha;
+		res0_2 *= alpha;
+		res0_3 *= alpha;
+
+		res1_0 *= alpha;
+		res1_1 *= alpha;
+		res1_2 *= alpha;
+		res1_3 *= alpha;
+
+		C0[0] = res0_0;
+		C0[1] = res0_1;
+		C0[2] = res0_2;
+		C0[3] = res0_3;
+
+		C1[0] = res1_0;
+		C1[1] = res1_1;
+		C1[2] = res1_2;
+		C1[3] = res1_3;
+
+
+#if ( defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+		temp = bk - off;
+#ifdef LEFT
+		temp -= 4; // number of values in A
+#else
+		temp -= 2; // number of values in B
+#endif
+		ptrba += temp*4;
+		ptrbb += temp*2;
+#endif
+
+#ifdef LEFT
+		off += 4; // number of values in A
+#endif
+
+		C0 = C0+4;
+		C1 = C1+4;
+
+	}
+
+	if ( bm & 2 ) // do any 2x2 loop
+	{
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+		ptrbb = bb;
+#else
+		ptrba += off*2;
+		ptrbb = bb + off*2;
+#endif
+
+		res0_0 = 0;
+		res0_1 = 0;
+
+		res1_0 = 0;
+		res1_1 = 0;
+
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+		temp = bk-off;
+#elif defined(LEFT)
+		temp = off+2;	// number of values in A
+#else
+		temp = off+2;	// number of values in B
+#endif
+
+		for (k=0; k<temp; k++)
+                {
+			b0 = ptrbb[0];
+			b1 = ptrbb[1];
+
+			a0 = ptrba[0];
+			res0_0 += a0*b0;
+			res1_0 += a0*b1;
+
+			a1 = ptrba[1];
+			res0_1 += a1*b0;
+			res1_1 += a1*b1;
+
+			ptrba = ptrba+2;
+			ptrbb = ptrbb+2;
+                }
+
+		res0_0 *= alpha;
+		res0_1 *= alpha;
+
+		res1_0 *= alpha;
+		res1_1 *= alpha;
+
+		C0[0] = res0_0;
+		C0[1] = res0_1;
+
+		C1[0] = res1_0;
+		C1[1] = res1_1;
+
+
+#if ( defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+		temp = bk - off;
+#ifdef LEFT
+		temp -= 2; // number of values in A
+#else
+		temp -= 2; // number of values in B
+#endif
+		ptrba += temp*2;
+		ptrbb += temp*2;
+#endif
+
+#ifdef LEFT
+		off += 2; // number of values in A
+#endif
+
+		C0 = C0+2;
+		C1 = C1+2;
+
+	}
+
+	if ( bm & 1 ) // do any 1x2 loop
+	{
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+		ptrbb = bb;
+#else
+		ptrba += off*1;
+		ptrbb = bb + off*2;
+#endif
+
+		res0_0 = 0;
+
+		res1_0 = 0;
+
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+		temp = bk-off;
+#elif defined(LEFT)
+		temp = off+1;	// number of values in A
+#else
+		temp = off+2;	// number of values in B
+#endif
+
+		for (k=0; k<temp; k++)
+                {
+			b0 = ptrbb[0];
+			b1 = ptrbb[1];
+
+			a0 = ptrba[0];
+			res0_0 += a0*b0;
+			res1_0 += a0*b1;
+
+			ptrba = ptrba+1;
+			ptrbb = ptrbb+2;
+                }
+
+		res0_0 *= alpha;
+
+		res1_0 *= alpha;
+
+		C0[0] = res0_0;
+
+		C1[0] = res1_0;
+
+
+#if ( defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+		temp = bk - off;
+#ifdef LEFT
+		temp -= 1; // number of values in A
+#else
+		temp -= 2; // number of values in B
+#endif
+		ptrba += temp*1;
+		ptrbb += temp*2;
+#endif
+
+#ifdef LEFT
+		off += 1; // number of values in A
+#endif
+
+		C0 = C0+1;
+		C1 = C1+1;
+
+	}
+
+
+#if defined(TRMMKERNEL) && !defined(LEFT)
+		off += 2;
+#endif
+
+        k = (bk<<1);
+        bb = bb+k;
+        i = (ldc<<1);
+        C = C+i;
+    }
+
+
+
+
+
+
+
+   for (j=0; j<(bn&1); j+=1) // do the Mx1 loops
+   {
+        C0 = C;
+
+#if defined(TRMMKERNEL) &&  defined(LEFT)
+	off = offset;
+#endif
+
+        ptrba = ba;
+
+        for (i=0; i<bm/4; i+=1) // do blocks of 4x1 loops
+	{
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+		ptrbb = bb;
+#else
+		ptrba += off*4;
+		ptrbb = bb + off*1;
+#endif
+
+		res0_0 = 0;
+		res0_1 = 0;
+		res0_2 = 0;
+		res0_3 = 0;
+
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+		temp = bk-off;
+#elif defined(LEFT)
+		temp = off+4;	// number of values in A
+#else
+		temp = off+1;	// number of values in B
+#endif
+
+		for (k=0; k<temp; k++)
+                {
+			b0 = ptrbb[0];
+
+			a0 = ptrba[0];
+			res0_0 += a0*b0;
+
+			a1 = ptrba[1];
+			res0_1 += a1*b0;
+
+			a0 = ptrba[2];
+			res0_2 += a0*b0;
+
+			a1 = ptrba[3];
+			res0_3 += a1*b0;
+
+			ptrba = ptrba+4;
+			ptrbb = ptrbb+1;
+                }
+
+		res0_0 *= alpha;
+		res0_1 *= alpha;
+		res0_2 *= alpha;
+		res0_3 *= alpha;
+
+		C0[0] = res0_0;
+		C0[1] = res0_1;
+		C0[2] = res0_2;
+		C0[3] = res0_3;
+
+
+#if ( defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+		temp = bk - off;
+#ifdef LEFT
+		temp -= 4; // number of values in A
+#else
+		temp -= 1; // number of values in B
+#endif
+		ptrba += temp*4;
+		ptrbb += temp*1;
+#endif
+
+#ifdef LEFT
+		off += 4; // number of values in A
+#endif
+
+		C0 = C0+4;
+
+	}
+
+	if ( bm & 2 ) // do any 2x1 loop
+	{
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+		ptrbb = bb;
+#else
+		ptrba += off*2;
+		ptrbb = bb + off*1;
+#endif
+
+		res0_0 = 0;
+		res0_1 = 0;
+
+
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+		temp = bk-off;
+#elif defined(LEFT)
+		temp = off+2;	// number of values in A
+#else
+		temp = off+1;	// number of values in B
+#endif
+
+		for (k=0; k<temp; k++)
+                {
+			b0 = ptrbb[0];
+
+			a0 = ptrba[0];
+			res0_0 += a0*b0;
+
+			a1 = ptrba[1];
+			res0_1 += a1*b0;
+
+			ptrba = ptrba+2;
+			ptrbb = ptrbb+1;
+                }
+
+		res0_0 *= alpha;
+		res0_1 *= alpha;
+
+		C0[0] = res0_0;
+		C0[1] = res0_1;
+
+
+#if ( defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+		temp = bk - off;
+#ifdef LEFT
+		temp -= 2; // number of values in A
+#else
+		temp -= 1; // number of values in B
+#endif
+		ptrba += temp*2;
+		ptrbb += temp*1;
+#endif
+
+#ifdef LEFT
+		off += 2; // number of values in A
+#endif
+
+		C0 = C0+2;
+
+	}
+
+	if ( bm & 1 ) // do any 1x1 loop
+	{
+
+#if (defined(LEFT) &&  defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+		ptrbb = bb;
+#else
+		ptrba += off*1;
+		ptrbb = bb + off*1;
+#endif
+
+		res0_0 = 0;
+
+
+#if (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+		temp = bk-off;
+#elif defined(LEFT)
+		temp = off+1;	// number of values in A
+#else
+		temp = off+1;	// number of values in B
+#endif
+
+		for (k=0; k<temp; k++)
+                {
+			b0 = ptrbb[0];
+
+			a0 = ptrba[0];
+			res0_0 += a0*b0;
+
+			ptrba = ptrba+1;
+			ptrbb = ptrbb+1;
+                }
+
+		res0_0 *= alpha;
+
+		C0[0] = res0_0;
+
+
+#if ( defined(LEFT) && defined(TRANSA)) || (!defined(LEFT) && !defined(TRANSA))
+		temp = bk - off;
+#ifdef LEFT
+		temp -= 1; // number of values in A
+#else
+		temp -= 1; // number of values in B
+#endif
+		ptrba += temp*1;
+		ptrbb += temp*1;
+#endif
+
+#ifdef LEFT
+		off += 1; // number of values in A
+#endif
+
+		C0 = C0+1;
+
+	}
+
+
+
+#if defined(TRMMKERNEL) && !defined(LEFT)
+		off += 1;
+#endif
+
+        k = (bk<<0);
+        bb = bb+k;
+        C = C+ldc;
+   }
+   return 0;
+}
-- 
2.7.4