From f76f9525477785bb452699c07d1985ec14dc2b61 Mon Sep 17 00:00:00 2001
From: wangqian <wangqian10@iscas.ac.cn>
Date: Tue, 19 Jun 2012 16:17:43 +0800
Subject: [PATCH] Refs #83 #53. Adding Intel Sandy Bridge (AVX supported)
 kernel codes for BLAS level 3 functions.

---
 kernel/generic/zgemm_ncopy_4_sandy.c   |  235 ++
 kernel/generic/zgemm_ncopy_8_sandy.c   |  401 +++
 kernel/generic/zgemm_tcopy_4_sandy.c   |  237 ++
 kernel/generic/zgemm_tcopy_8_sandy.c   |  370 ++
 kernel/x86_64/KERNEL.SANDYBRIDGE       |   97 +-
 kernel/x86_64/cgemm_kernel_4x8_sandy.S | 4478 ++++++++++++++++++++++++
 kernel/x86_64/dgemm_kernel_4x8_sandy.S | 3186 +++++++++++++++++
 kernel/x86_64/sgemm_kernel_8x8_sandy.S | 3736 ++++++++++++++++++++
 kernel/x86_64/zgemm_kernel_4x4_sandy.S | 3257 +++++++++++++++++
 param.h                                |  100 +-
 10 files changed, 15982 insertions(+), 115 deletions(-)
 create mode 100644 kernel/generic/zgemm_ncopy_4_sandy.c
 create mode 100644 kernel/generic/zgemm_ncopy_8_sandy.c
 create mode 100644 kernel/generic/zgemm_tcopy_4_sandy.c
 create mode 100644 kernel/generic/zgemm_tcopy_8_sandy.c
 create mode 100644 kernel/x86_64/cgemm_kernel_4x8_sandy.S
 create mode 100644 kernel/x86_64/dgemm_kernel_4x8_sandy.S
 create mode 100644 kernel/x86_64/sgemm_kernel_8x8_sandy.S
 create mode 100644 kernel/x86_64/zgemm_kernel_4x4_sandy.S

diff --git a/kernel/generic/zgemm_ncopy_4_sandy.c b/kernel/generic/zgemm_ncopy_4_sandy.c
new file mode 100644
index 000000000..839bd5939
--- /dev/null
+++ b/kernel/generic/zgemm_ncopy_4_sandy.c
@@ -0,0 +1,235 @@
+/*****************************************************************************
+  Copyright (c) 2011,2012 Lab of Parallel Software and Computational Science,ISCAS
+  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 ISCAS 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 COPYRIGHT OWNER 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 <stdio.h>
+#include "common.h"
+
+int CNAME(BLASLONG row,BLASLONG col,FLOAT* src,BLASLONG srcdim,FLOAT* dest) 
+{
+   BLASLONG i,j;
+   BLASLONG idx=0;
+   BLASLONG ii;
+   FLOAT *src0,*src1,*src2,*src3,*dest0;
+   for (j=0; j<col/4; j+=1) 
+     {
+        src0 = src;
+        src1 = src0+2*srcdim;
+        src2 = src1+2*srcdim;
+        src3 = src2+2*srcdim;
+        src = src3+2*srcdim;
+        dest0 = dest;
+        ii = (row<<3);
+        dest = dest+ii;
+        for (i=0; i<row/4; i+=1) 
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src1[0];
+             dest0[3] = src1[1];
+             dest0[4] = src2[0];
+             dest0[5] = src2[1];
+             dest0[6] = src3[0];
+             dest0[7] = src3[1];
+             dest0[8] = src0[2];
+             dest0[9] = src0[3];
+             dest0[10] = src1[2];
+             dest0[11] = src1[3];
+             dest0[12] = src2[2];
+             dest0[13] = src2[3];
+             dest0[14] = src3[2];
+             dest0[15] = src3[3];
+             dest0[16] = src0[4];
+             dest0[17] = src0[5];
+             dest0[18] = src1[4];
+             dest0[19] = src1[5];
+             dest0[20] = src2[4];
+             dest0[21] = src2[5];
+             dest0[22] = src3[4];
+             dest0[23] = src3[5];
+             dest0[24] = src0[6];
+             dest0[25] = src0[7];
+             dest0[26] = src1[6];
+             dest0[27] = src1[7];
+             dest0[28] = src2[6];
+             dest0[29] = src2[7];
+             dest0[30] = src3[6];
+             dest0[31] = src3[7];
+             src0 = src0+8;
+             src1 = src1+8;
+             src2 = src2+8;
+             src3 = src3+8;
+             ii = (4<<3);
+             dest0 = dest0+ii;
+          }
+        if (row&2)  
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src1[0];
+             dest0[3] = src1[1];
+             dest0[4] = src2[0];
+             dest0[5] = src2[1];
+             dest0[6] = src3[0];
+             dest0[7] = src3[1];
+             dest0[8] = src0[2];
+             dest0[9] = src0[3];
+             dest0[10] = src1[2];
+             dest0[11] = src1[3];
+             dest0[12] = src2[2];
+             dest0[13] = src2[3];
+             dest0[14] = src3[2];
+             dest0[15] = src3[3];
+             src0 = src0+4;
+             src1 = src1+4;
+             src2 = src2+4;
+             src3 = src3+4;
+             ii = (2<<3);
+             dest0 = dest0+ii;
+          }
+        if (row&1)  
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src1[0];
+             dest0[3] = src1[1];
+             dest0[4] = src2[0];
+             dest0[5] = src2[1];
+             dest0[6] = src3[0];
+             dest0[7] = src3[1];
+             src0 = src0+2;
+             src1 = src1+2;
+             src2 = src2+2;
+             src3 = src3+2;
+             ii = (1<<3);
+             dest0 = dest0+ii;
+          }
+     }
+   if (col&2)  
+     {
+        src0 = src;
+        src1 = src0+2*srcdim;
+        src = src1+2*srcdim;
+        dest0 = dest;
+        ii = (row<<2);
+        dest = dest+ii;
+        for (i=0; i<row/4; i+=1) 
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src1[0];
+             dest0[3] = src1[1];
+             dest0[4] = src0[2];
+             dest0[5] = src0[3];
+             dest0[6] = src1[2];
+             dest0[7] = src1[3];
+             dest0[8] = src0[4];
+             dest0[9] = src0[5];
+             dest0[10] = src1[4];
+             dest0[11] = src1[5];
+             dest0[12] = src0[6];
+             dest0[13] = src0[7];
+             dest0[14] = src1[6];
+             dest0[15] = src1[7];
+             src0 = src0+8;
+             src1 = src1+8;
+             ii = (4<<2);
+             dest0 = dest0+ii;
+          }
+        if (row&2)  
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src1[0];
+             dest0[3] = src1[1];
+             dest0[4] = src0[2];
+             dest0[5] = src0[3];
+             dest0[6] = src1[2];
+             dest0[7] = src1[3];
+             src0 = src0+4;
+             src1 = src1+4;
+             ii = (2<<2);
+             dest0 = dest0+ii;
+          }
+        if (row&1)  
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src1[0];
+             dest0[3] = src1[1];
+             src0 = src0+2;
+             src1 = src1+2;
+             ii = (1<<2);
+             dest0 = dest0+ii;
+          }
+     }
+   if (col&1)  
+     {
+        src0 = src;
+        src = src0+2*srcdim;
+        dest0 = dest;
+        ii = (row<<1);
+        dest = dest+ii;
+        for (i=0; i<row/4; i+=1) 
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src0[2];
+             dest0[3] = src0[3];
+             dest0[4] = src0[4];
+             dest0[5] = src0[5];
+             dest0[6] = src0[6];
+             dest0[7] = src0[7];
+             src0 = src0+8;
+             ii = (4<<1);
+             dest0 = dest0+ii;
+          }
+        if (row&2)  
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src0[2];
+             dest0[3] = src0[3];
+             src0 = src0+4;
+             ii = (2<<1);
+             dest0 = dest0+ii;
+          }
+        if (row&1)  
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             src0 = src0+2;
+             ii = (1<<1);
+             dest0 = dest0+ii;
+          }
+     }
+   return 0;
+}
diff --git a/kernel/generic/zgemm_ncopy_8_sandy.c b/kernel/generic/zgemm_ncopy_8_sandy.c
new file mode 100644
index 000000000..ed580a126
--- /dev/null
+++ b/kernel/generic/zgemm_ncopy_8_sandy.c
@@ -0,0 +1,401 @@
+/*****************************************************************************
+  Copyright (c) 2011,2012 Lab of Parallel Software and Computational Science,ISCAS
+  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 ISCAS 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 COPYRIGHT OWNER 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 <stdio.h>
+#include "common.h"
+
+int CNAME(BLASLONG row,BLASLONG col,FLOAT* src,BLASLONG srcdim,FLOAT* dest) 
+{
+   BLASLONG i,j;
+   BLASLONG idx=0;
+   BLASLONG ii;
+   FLOAT *src0,*src1,*src2,*src3,*src4,*src5,*src6,*src7,*dest0;
+   for (j=0; j<col/8; j+=1) 
+     {
+        src0 = src;
+        src1 = src0+2*srcdim;
+        src2 = src1+2*srcdim;
+        src3 = src2+2*srcdim;
+        src4 = src3+2*srcdim;
+        src5 = src4+2*srcdim;
+        src6 = src5+2*srcdim;
+        src7 = src6+2*srcdim;
+        src = src7+2*srcdim;
+        dest0 = dest;
+        ii = (row<<4);
+        dest = dest+ii;
+        for (i=0; i<row/4; i+=1) 
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src1[0];
+             dest0[3] = src1[1];
+             dest0[4] = src2[0];
+             dest0[5] = src2[1];
+             dest0[6] = src3[0];
+             dest0[7] = src3[1];
+             dest0[8] = src4[0];
+             dest0[9] = src4[1];
+             dest0[10] = src5[0];
+             dest0[11] = src5[1];
+             dest0[12] = src6[0];
+             dest0[13] = src6[1];
+             dest0[14] = src7[0];
+             dest0[15] = src7[1];
+             dest0[16] = src0[2];
+             dest0[17] = src0[3];
+             dest0[18] = src1[2];
+             dest0[19] = src1[3];
+             dest0[20] = src2[2];
+             dest0[21] = src2[3];
+             dest0[22] = src3[2];
+             dest0[23] = src3[3];
+             dest0[24] = src4[2];
+             dest0[25] = src4[3];
+             dest0[26] = src5[2];
+             dest0[27] = src5[3];
+             dest0[28] = src6[2];
+             dest0[29] = src6[3];
+             dest0[30] = src7[2];
+             dest0[31] = src7[3];
+             dest0[32] = src0[4];
+             dest0[33] = src0[5];
+             dest0[34] = src1[4];
+             dest0[35] = src1[5];
+             dest0[36] = src2[4];
+             dest0[37] = src2[5];
+             dest0[38] = src3[4];
+             dest0[39] = src3[5];
+             dest0[40] = src4[4];
+             dest0[41] = src4[5];
+             dest0[42] = src5[4];
+             dest0[43] = src5[5];
+             dest0[44] = src6[4];
+             dest0[45] = src6[5];
+             dest0[46] = src7[4];
+             dest0[47] = src7[5];
+             dest0[48] = src0[6];
+             dest0[49] = src0[7];
+             dest0[50] = src1[6];
+             dest0[51] = src1[7];
+             dest0[52] = src2[6];
+             dest0[53] = src2[7];
+             dest0[54] = src3[6];
+             dest0[55] = src3[7];
+             dest0[56] = src4[6];
+             dest0[57] = src4[7];
+             dest0[58] = src5[6];
+             dest0[59] = src5[7];
+             dest0[60] = src6[6];
+             dest0[61] = src6[7];
+             dest0[62] = src7[6];
+             dest0[63] = src7[7];
+             src0 = src0+8;
+             src1 = src1+8;
+             src2 = src2+8;
+             src3 = src3+8;
+             src4 = src4+8;
+             src5 = src5+8;
+             src6 = src6+8;
+             src7 = src7+8;
+             ii = (4<<4);
+             dest0 = dest0+ii;
+          }
+        if (row&2)  
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src1[0];
+             dest0[3] = src1[1];
+             dest0[4] = src2[0];
+             dest0[5] = src2[1];
+             dest0[6] = src3[0];
+             dest0[7] = src3[1];
+             dest0[8] = src4[0];
+             dest0[9] = src4[1];
+             dest0[10] = src5[0];
+             dest0[11] = src5[1];
+             dest0[12] = src6[0];
+             dest0[13] = src6[1];
+             dest0[14] = src7[0];
+             dest0[15] = src7[1];
+             dest0[16] = src0[2];
+             dest0[17] = src0[3];
+             dest0[18] = src1[2];
+             dest0[19] = src1[3];
+             dest0[20] = src2[2];
+             dest0[21] = src2[3];
+             dest0[22] = src3[2];
+             dest0[23] = src3[3];
+             dest0[24] = src4[2];
+             dest0[25] = src4[3];
+             dest0[26] = src5[2];
+             dest0[27] = src5[3];
+             dest0[28] = src6[2];
+             dest0[29] = src6[3];
+             dest0[30] = src7[2];
+             dest0[31] = src7[3];
+             src0 = src0+4;
+             src1 = src1+4;
+             src2 = src2+4;
+             src3 = src3+4;
+             src4 = src4+4;
+             src5 = src5+4;
+             src6 = src6+4;
+             src7 = src7+4;
+             ii = (2<<4);
+             dest0 = dest0+ii;
+          }
+        if (row&1)  
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src1[0];
+             dest0[3] = src1[1];
+             dest0[4] = src2[0];
+             dest0[5] = src2[1];
+             dest0[6] = src3[0];
+             dest0[7] = src3[1];
+             dest0[8] = src4[0];
+             dest0[9] = src4[1];
+             dest0[10] = src5[0];
+             dest0[11] = src5[1];
+             dest0[12] = src6[0];
+             dest0[13] = src6[1];
+             dest0[14] = src7[0];
+             dest0[15] = src7[1];
+             src0 = src0+2;
+             src1 = src1+2;
+             src2 = src2+2;
+             src3 = src3+2;
+             src4 = src4+2;
+             src5 = src5+2;
+             src6 = src6+2;
+             src7 = src7+2;
+             ii = (1<<4);
+             dest0 = dest0+ii;
+          }
+     }
+   if (col&4)  
+     {
+        src0 = src;
+        src1 = src0+2*srcdim;
+        src2 = src1+2*srcdim;
+        src3 = src2+2*srcdim;
+        src = src3+2*srcdim;
+        dest0 = dest;
+        ii = (row<<3);
+        dest = dest+ii;
+        for (i=0; i<row/4; i+=1) 
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src1[0];
+             dest0[3] = src1[1];
+             dest0[4] = src2[0];
+             dest0[5] = src2[1];
+             dest0[6] = src3[0];
+             dest0[7] = src3[1];
+             dest0[8] = src0[2];
+             dest0[9] = src0[3];
+             dest0[10] = src1[2];
+             dest0[11] = src1[3];
+             dest0[12] = src2[2];
+             dest0[13] = src2[3];
+             dest0[14] = src3[2];
+             dest0[15] = src3[3];
+             dest0[16] = src0[4];
+             dest0[17] = src0[5];
+             dest0[18] = src1[4];
+             dest0[19] = src1[5];
+             dest0[20] = src2[4];
+             dest0[21] = src2[5];
+             dest0[22] = src3[4];
+             dest0[23] = src3[5];
+             dest0[24] = src0[6];
+             dest0[25] = src0[7];
+             dest0[26] = src1[6];
+             dest0[27] = src1[7];
+             dest0[28] = src2[6];
+             dest0[29] = src2[7];
+             dest0[30] = src3[6];
+             dest0[31] = src3[7];
+             src0 = src0+8;
+             src1 = src1+8;
+             src2 = src2+8;
+             src3 = src3+8;
+             ii = (4<<3);
+             dest0 = dest0+ii;
+          }
+        if (row&2)  
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src1[0];
+             dest0[3] = src1[1];
+             dest0[4] = src2[0];
+             dest0[5] = src2[1];
+             dest0[6] = src3[0];
+             dest0[7] = src3[1];
+             dest0[8] = src0[2];
+             dest0[9] = src0[3];
+             dest0[10] = src1[2];
+             dest0[11] = src1[3];
+             dest0[12] = src2[2];
+             dest0[13] = src2[3];
+             dest0[14] = src3[2];
+             dest0[15] = src3[3];
+             src0 = src0+4;
+             src1 = src1+4;
+             src2 = src2+4;
+             src3 = src3+4;
+             ii = (2<<3);
+             dest0 = dest0+ii;
+          }
+        if (row&1)  
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src1[0];
+             dest0[3] = src1[1];
+             dest0[4] = src2[0];
+             dest0[5] = src2[1];
+             dest0[6] = src3[0];
+             dest0[7] = src3[1];
+             src0 = src0+2;
+             src1 = src1+2;
+             src2 = src2+2;
+             src3 = src3+2;
+             ii = (1<<3);
+             dest0 = dest0+ii;
+          }
+     }
+   if (col&2)  
+     {
+        src0 = src;
+        src1 = src0+2*srcdim;
+        src = src1+2*srcdim;
+        dest0 = dest;
+        ii = (row<<2);
+        dest = dest+ii;
+        for (i=0; i<row/4; i+=1) 
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src1[0];
+             dest0[3] = src1[1];
+             dest0[4] = src0[2];
+             dest0[5] = src0[3];
+             dest0[6] = src1[2];
+             dest0[7] = src1[3];
+             dest0[8] = src0[4];
+             dest0[9] = src0[5];
+             dest0[10] = src1[4];
+             dest0[11] = src1[5];
+             dest0[12] = src0[6];
+             dest0[13] = src0[7];
+             dest0[14] = src1[6];
+             dest0[15] = src1[7];
+             src0 = src0+8;
+             src1 = src1+8;
+             ii = (4<<2);
+             dest0 = dest0+ii;
+          }
+        if (row&2)  
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src1[0];
+             dest0[3] = src1[1];
+             dest0[4] = src0[2];
+             dest0[5] = src0[3];
+             dest0[6] = src1[2];
+             dest0[7] = src1[3];
+             src0 = src0+4;
+             src1 = src1+4;
+             ii = (2<<2);
+             dest0 = dest0+ii;
+          }
+        if (row&1)  
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src1[0];
+             dest0[3] = src1[1];
+             src0 = src0+2;
+             src1 = src1+2;
+             ii = (1<<2);
+             dest0 = dest0+ii;
+          }
+     }
+   if (col&1)  
+     {
+        src0 = src;
+        src = src0+2*srcdim;
+        dest0 = dest;
+        ii = (row<<1);
+        dest = dest+ii;
+        for (i=0; i<row/4; i+=1) 
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src0[2];
+             dest0[3] = src0[3];
+             dest0[4] = src0[4];
+             dest0[5] = src0[5];
+             dest0[6] = src0[6];
+             dest0[7] = src0[7];
+             src0 = src0+8;
+             ii = (4<<1);
+             dest0 = dest0+ii;
+          }
+        if (row&2)  
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src0[2];
+             dest0[3] = src0[3];
+             src0 = src0+4;
+             ii = (2<<1);
+             dest0 = dest0+ii;
+          }
+        if (row&1)  
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             src0 = src0+2;
+             ii = (1<<1);
+             dest0 = dest0+ii;
+          }
+     }
+   return 0;
+}
diff --git a/kernel/generic/zgemm_tcopy_4_sandy.c b/kernel/generic/zgemm_tcopy_4_sandy.c
new file mode 100644
index 000000000..1ae4a4e68
--- /dev/null
+++ b/kernel/generic/zgemm_tcopy_4_sandy.c
@@ -0,0 +1,237 @@
+/*****************************************************************************
+  Copyright (c) 2011,2012 Lab of Parallel Software and Computational Science,ISCAS
+  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 ISCAS 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 COPYRIGHT OWNER 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 <stdio.h>
+#include "common.h"
+
+int CNAME(BLASLONG row,BLASLONG col,FLOAT* src,BLASLONG srcdim,FLOAT* dest) 
+{
+   BLASLONG i,j;
+   BLASLONG idx=0;
+   BLASLONG ii;
+   FLOAT *src0,*src1,*src2,*src3,*dest0;
+   FLOAT *dest1,*dest2;
+   ii = col&-4;
+   ii = ii*(2*row);
+   dest2 = dest+ii;
+   ii = col&-2;
+   ii = ii*(2*row);
+   dest1 = dest+ii;
+   for (j=0; j<row/4; j+=1) 
+     {
+        src0 = src;
+        src1 = src0+2*srcdim;
+        src2 = src1+2*srcdim;
+        src3 = src2+2*srcdim;
+        src = src3+2*srcdim;
+        dest0 = dest;
+        ii = (4<<3);
+        dest = dest+ii;
+        for (i=0; i<col/4; i+=1) 
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src0[2];
+             dest0[3] = src0[3];
+             dest0[4] = src0[4];
+             dest0[5] = src0[5];
+             dest0[6] = src0[6];
+             dest0[7] = src0[7];
+             dest0[8] = src1[0];
+             dest0[9] = src1[1];
+             dest0[10] = src1[2];
+             dest0[11] = src1[3];
+             dest0[12] = src1[4];
+             dest0[13] = src1[5];
+             dest0[14] = src1[6];
+             dest0[15] = src1[7];
+             dest0[16] = src2[0];
+             dest0[17] = src2[1];
+             dest0[18] = src2[2];
+             dest0[19] = src2[3];
+             dest0[20] = src2[4];
+             dest0[21] = src2[5];
+             dest0[22] = src2[6];
+             dest0[23] = src2[7];
+             dest0[24] = src3[0];
+             dest0[25] = src3[1];
+             dest0[26] = src3[2];
+             dest0[27] = src3[3];
+             dest0[28] = src3[4];
+             dest0[29] = src3[5];
+             dest0[30] = src3[6];
+             dest0[31] = src3[7];
+             src0 = src0+8;
+             src1 = src1+8;
+             src2 = src2+8;
+             src3 = src3+8;
+             ii = (row<<3);
+             dest0 = dest0+ii;
+          }
+        if (col&2)  
+          {
+             dest2[0] = src0[0];
+             dest2[1] = src0[1];
+             dest2[2] = src0[2];
+             dest2[3] = src0[3];
+             dest2[4] = src1[0];
+             dest2[5] = src1[1];
+             dest2[6] = src1[2];
+             dest2[7] = src1[3];
+             dest2[8] = src2[0];
+             dest2[9] = src2[1];
+             dest2[10] = src2[2];
+             dest2[11] = src2[3];
+             dest2[12] = src3[0];
+             dest2[13] = src3[1];
+             dest2[14] = src3[2];
+             dest2[15] = src3[3];
+             src0 = src0+4;
+             src1 = src1+4;
+             src2 = src2+4;
+             src3 = src3+4;
+             dest2 = dest2+16;
+          }
+        if (col&1)  
+          {
+             dest1[0] = src0[0];
+             dest1[1] = src0[1];
+             dest1[2] = src1[0];
+             dest1[3] = src1[1];
+             dest1[4] = src2[0];
+             dest1[5] = src2[1];
+             dest1[6] = src3[0];
+             dest1[7] = src3[1];
+             src0 = src0+2;
+             src1 = src1+2;
+             src2 = src2+2;
+             src3 = src3+2;
+             dest1 = dest1+8;
+          }
+     }
+   if (row&2)  
+     {
+        src0 = src;
+        src1 = src0+2*srcdim;
+        src = src1+2*srcdim;
+        dest0 = dest;
+        ii = (2<<3);
+        dest = dest+ii;
+        for (i=0; i<col/4; i+=1) 
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src0[2];
+             dest0[3] = src0[3];
+             dest0[4] = src0[4];
+             dest0[5] = src0[5];
+             dest0[6] = src0[6];
+             dest0[7] = src0[7];
+             dest0[8] = src1[0];
+             dest0[9] = src1[1];
+             dest0[10] = src1[2];
+             dest0[11] = src1[3];
+             dest0[12] = src1[4];
+             dest0[13] = src1[5];
+             dest0[14] = src1[6];
+             dest0[15] = src1[7];
+             src0 = src0+8;
+             src1 = src1+8;
+             ii = (row<<3);
+             dest0 = dest0+ii;
+          }
+        if (col&2)  
+          {
+             dest2[0] = src0[0];
+             dest2[1] = src0[1];
+             dest2[2] = src0[2];
+             dest2[3] = src0[3];
+             dest2[4] = src1[0];
+             dest2[5] = src1[1];
+             dest2[6] = src1[2];
+             dest2[7] = src1[3];
+             src0 = src0+4;
+             src1 = src1+4;
+             dest2 = dest2+8;
+          }
+        if (col&1)  
+          {
+             dest1[0] = src0[0];
+             dest1[1] = src0[1];
+             dest1[2] = src1[0];
+             dest1[3] = src1[1];
+             src0 = src0+2;
+             src1 = src1+2;
+             dest1 = dest1+4;
+          }
+     }
+   if (row&1)  
+     {
+        src0 = src;
+        src = src0+2*srcdim;
+        dest0 = dest;
+        ii = (1<<3);
+        dest = dest+ii;
+        for (i=0; i<col/4; i+=1) 
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src0[2];
+             dest0[3] = src0[3];
+             dest0[4] = src0[4];
+             dest0[5] = src0[5];
+             dest0[6] = src0[6];
+             dest0[7] = src0[7];
+             src0 = src0+8;
+             ii = (row<<3);
+             dest0 = dest0+ii;
+          }
+        if (col&2)  
+          {
+             dest2[0] = src0[0];
+             dest2[1] = src0[1];
+             dest2[2] = src0[2];
+             dest2[3] = src0[3];
+             src0 = src0+4;
+             dest2 = dest2+4;
+          }
+        if (col&1)  
+          {
+             dest1[0] = src0[0];
+             dest1[1] = src0[1];
+             src0 = src0+2;
+             dest1 = dest1+2;
+          }
+     }
+
+   return 0;
+}
diff --git a/kernel/generic/zgemm_tcopy_8_sandy.c b/kernel/generic/zgemm_tcopy_8_sandy.c
new file mode 100644
index 000000000..b53dd3ec0
--- /dev/null
+++ b/kernel/generic/zgemm_tcopy_8_sandy.c
@@ -0,0 +1,370 @@
+/*****************************************************************************
+  Copyright (c) 2011,2012 Lab of Parallel Software and Computational Science,ISCAS
+  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 ISCAS 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 COPYRIGHT OWNER 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 <stdio.h>
+#include "common.h"
+
+int CNAME(BLASLONG row,BLASLONG col,FLOAT* src,BLASLONG srcdim,FLOAT* dest) 
+{
+   BLASLONG i,j;
+   BLASLONG idx=0;
+   BLASLONG ii;
+   FLOAT *src0,*src1,*src2,*src3,*dest0;
+   FLOAT *dest1,*dest2,*dest4;
+   ii = col&-8;
+   ii = ii*(2*row);
+   dest4 = dest+ii;
+   ii = col&-4;
+   ii = ii*(2*row);
+   dest2 = dest+ii;
+   ii = col&-2;
+   ii = ii*(2*row);
+   dest1 = dest+ii;
+   for (j=0; j<row/4; j+=1) 
+     {
+        src0 = src;
+        src1 = src0+2*srcdim;
+        src2 = src1+2*srcdim;
+        src3 = src2+2*srcdim;
+        src = src3+2*srcdim;
+        dest0 = dest;
+        ii = (4<<4);
+        dest = dest+ii;
+        for (i=0; i<col/8; i+=1) 
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src0[2];
+             dest0[3] = src0[3];
+             dest0[4] = src0[4];
+             dest0[5] = src0[5];
+             dest0[6] = src0[6];
+             dest0[7] = src0[7];
+             dest0[8] = src0[8];
+             dest0[9] = src0[9];
+             dest0[10] = src0[10];
+             dest0[11] = src0[11];
+             dest0[12] = src0[12];
+             dest0[13] = src0[13];
+             dest0[14] = src0[14];
+             dest0[15] = src0[15];
+             dest0[16] = src1[0];
+             dest0[17] = src1[1];
+             dest0[18] = src1[2];
+             dest0[19] = src1[3];
+             dest0[20] = src1[4];
+             dest0[21] = src1[5];
+             dest0[22] = src1[6];
+             dest0[23] = src1[7];
+             dest0[24] = src1[8];
+             dest0[25] = src1[9];
+             dest0[26] = src1[10];
+             dest0[27] = src1[11];
+             dest0[28] = src1[12];
+             dest0[29] = src1[13];
+             dest0[30] = src1[14];
+             dest0[31] = src1[15];
+             dest0[32] = src2[0];
+             dest0[33] = src2[1];
+             dest0[34] = src2[2];
+             dest0[35] = src2[3];
+             dest0[36] = src2[4];
+             dest0[37] = src2[5];
+             dest0[38] = src2[6];
+             dest0[39] = src2[7];
+             dest0[40] = src2[8];
+             dest0[41] = src2[9];
+             dest0[42] = src2[10];
+             dest0[43] = src2[11];
+             dest0[44] = src2[12];
+             dest0[45] = src2[13];
+             dest0[46] = src2[14];
+             dest0[47] = src2[15];
+             dest0[48] = src3[0];
+             dest0[49] = src3[1];
+             dest0[50] = src3[2];
+             dest0[51] = src3[3];
+             dest0[52] = src3[4];
+             dest0[53] = src3[5];
+             dest0[54] = src3[6];
+             dest0[55] = src3[7];
+             dest0[56] = src3[8];
+             dest0[57] = src3[9];
+             dest0[58] = src3[10];
+             dest0[59] = src3[11];
+             dest0[60] = src3[12];
+             dest0[61] = src3[13];
+             dest0[62] = src3[14];
+             dest0[63] = src3[15];
+             src0 = src0+16;
+             src1 = src1+16;
+             src2 = src2+16;
+             src3 = src3+16;
+             ii = (row<<4);
+             dest0 = dest0+ii;
+          }
+        if (col&4)  
+          {
+             dest4[0] = src0[0];
+             dest4[1] = src0[1];
+             dest4[2] = src0[2];
+             dest4[3] = src0[3];
+             dest4[4] = src0[4];
+             dest4[5] = src0[5];
+             dest4[6] = src0[6];
+             dest4[7] = src0[7];
+             dest4[8] = src1[0];
+             dest4[9] = src1[1];
+             dest4[10] = src1[2];
+             dest4[11] = src1[3];
+             dest4[12] = src1[4];
+             dest4[13] = src1[5];
+             dest4[14] = src1[6];
+             dest4[15] = src1[7];
+             dest4[16] = src2[0];
+             dest4[17] = src2[1];
+             dest4[18] = src2[2];
+             dest4[19] = src2[3];
+             dest4[20] = src2[4];
+             dest4[21] = src2[5];
+             dest4[22] = src2[6];
+             dest4[23] = src2[7];
+             dest4[24] = src3[0];
+             dest4[25] = src3[1];
+             dest4[26] = src3[2];
+             dest4[27] = src3[3];
+             dest4[28] = src3[4];
+             dest4[29] = src3[5];
+             dest4[30] = src3[6];
+             dest4[31] = src3[7];
+             src0 = src0+8;
+             src1 = src1+8;
+             src2 = src2+8;
+             src3 = src3+8;
+             dest4 = dest4+32;
+          }
+        if (col&2)  
+          {
+             dest2[0] = src0[0];
+             dest2[1] = src0[1];
+             dest2[2] = src0[2];
+             dest2[3] = src0[3];
+             dest2[4] = src1[0];
+             dest2[5] = src1[1];
+             dest2[6] = src1[2];
+             dest2[7] = src1[3];
+             dest2[8] = src2[0];
+             dest2[9] = src2[1];
+             dest2[10] = src2[2];
+             dest2[11] = src2[3];
+             dest2[12] = src3[0];
+             dest2[13] = src3[1];
+             dest2[14] = src3[2];
+             dest2[15] = src3[3];
+             src0 = src0+4;
+             src1 = src1+4;
+             src2 = src2+4;
+             src3 = src3+4;
+             dest2 = dest2+16;
+          }
+        if (col&1)  
+          {
+             dest1[0] = src0[0];
+             dest1[1] = src0[1];
+             dest1[2] = src1[0];
+             dest1[3] = src1[1];
+             dest1[4] = src2[0];
+             dest1[5] = src2[1];
+             dest1[6] = src3[0];
+             dest1[7] = src3[1];
+             src0 = src0+2;
+             src1 = src1+2;
+             src2 = src2+2;
+             src3 = src3+2;
+             dest1 = dest1+8;
+          }
+     }
+   if (row&2)  
+     {
+        src0 = src;
+        src1 = src0+2*srcdim;
+        src = src1+2*srcdim;
+        dest0 = dest;
+        ii = (2<<4);
+        dest = dest+ii;
+        for (i=0; i<col/8; i+=1) 
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src0[2];
+             dest0[3] = src0[3];
+             dest0[4] = src0[4];
+             dest0[5] = src0[5];
+             dest0[6] = src0[6];
+             dest0[7] = src0[7];
+             dest0[8] = src0[8];
+             dest0[9] = src0[9];
+             dest0[10] = src0[10];
+             dest0[11] = src0[11];
+             dest0[12] = src0[12];
+             dest0[13] = src0[13];
+             dest0[14] = src0[14];
+             dest0[15] = src0[15];
+             dest0[16] = src1[0];
+             dest0[17] = src1[1];
+             dest0[18] = src1[2];
+             dest0[19] = src1[3];
+             dest0[20] = src1[4];
+             dest0[21] = src1[5];
+             dest0[22] = src1[6];
+             dest0[23] = src1[7];
+             dest0[24] = src1[8];
+             dest0[25] = src1[9];
+             dest0[26] = src1[10];
+             dest0[27] = src1[11];
+             dest0[28] = src1[12];
+             dest0[29] = src1[13];
+             dest0[30] = src1[14];
+             dest0[31] = src1[15];
+             src0 = src0+16;
+             src1 = src1+16;
+             ii = (row<<4);
+             dest0 = dest0+ii;
+          }
+        if (col&4)  
+          {
+             dest4[0] = src0[0];
+             dest4[1] = src0[1];
+             dest4[2] = src0[2];
+             dest4[3] = src0[3];
+             dest4[4] = src0[4];
+             dest4[5] = src0[5];
+             dest4[6] = src0[6];
+             dest4[7] = src0[7];
+             dest4[8] = src1[0];
+             dest4[9] = src1[1];
+             dest4[10] = src1[2];
+             dest4[11] = src1[3];
+             dest4[12] = src1[4];
+             dest4[13] = src1[5];
+             dest4[14] = src1[6];
+             dest4[15] = src1[7];
+             src0 = src0+8;
+             src1 = src1+8;
+             dest4 = dest4+16;
+          }
+        if (col&2)  
+          {
+             dest2[0] = src0[0];
+             dest2[1] = src0[1];
+             dest2[2] = src0[2];
+             dest2[3] = src0[3];
+             dest2[4] = src1[0];
+             dest2[5] = src1[1];
+             dest2[6] = src1[2];
+             dest2[7] = src1[3];
+             src0 = src0+4;
+             src1 = src1+4;
+             dest2 = dest2+8;
+          }
+        if (col&1)  
+          {
+             dest1[0] = src0[0];
+             dest1[1] = src0[1];
+             dest1[2] = src1[0];
+             dest1[3] = src1[1];
+             src0 = src0+2;
+             src1 = src1+2;
+             dest1 = dest1+4;
+          }
+     }
+   if (row&1)  
+     {
+        src0 = src;
+        src = src0+2*srcdim;
+        dest0 = dest;
+        ii = (1<<4);
+        dest = dest+ii;
+        for (i=0; i<col/8; i+=1) 
+          {
+             dest0[0] = src0[0];
+             dest0[1] = src0[1];
+             dest0[2] = src0[2];
+             dest0[3] = src0[3];
+             dest0[4] = src0[4];
+             dest0[5] = src0[5];
+             dest0[6] = src0[6];
+             dest0[7] = src0[7];
+             dest0[8] = src0[8];
+             dest0[9] = src0[9];
+             dest0[10] = src0[10];
+             dest0[11] = src0[11];
+             dest0[12] = src0[12];
+             dest0[13] = src0[13];
+             dest0[14] = src0[14];
+             dest0[15] = src0[15];
+             src0 = src0+16;
+             ii = (row<<4);
+             dest0 = dest0+ii;
+          }
+        if (col&4)  
+          {
+             dest4[0] = src0[0];
+             dest4[1] = src0[1];
+             dest4[2] = src0[2];
+             dest4[3] = src0[3];
+             dest4[4] = src0[4];
+             dest4[5] = src0[5];
+             dest4[6] = src0[6];
+             dest4[7] = src0[7];
+             src0 = src0+8;
+             dest4 = dest4+8;
+          }
+        if (col&2)  
+          {
+             dest2[0] = src0[0];
+             dest2[1] = src0[1];
+             dest2[2] = src0[2];
+             dest2[3] = src0[3];
+             src0 = src0+4;
+             dest2 = dest2+4;
+          }
+        if (col&1)  
+          {
+             dest1[0] = src0[0];
+             dest1[1] = src0[1];
+             src0 = src0+2;
+             dest1 = dest1+2;
+          }
+     }
+   return 0;
+}
diff --git a/kernel/x86_64/KERNEL.SANDYBRIDGE b/kernel/x86_64/KERNEL.SANDYBRIDGE
index 58a883243..c321be752 100644
--- a/kernel/x86_64/KERNEL.SANDYBRIDGE
+++ b/kernel/x86_64/KERNEL.SANDYBRIDGE
@@ -1,59 +1,84 @@
-SGEMMKERNEL    =  gemm_kernel_4x8_nehalem.S
-SGEMMINCOPY    =  gemm_ncopy_4.S
-SGEMMITCOPY    =  gemm_tcopy_4.S
+SGEMMKERNEL    =  sgemm_kernel_8x8_sandy.S
+SGEMMINCOPY    =  
+SGEMMITCOPY    =  
 SGEMMONCOPY    =  ../generic/gemm_ncopy_8.c
 SGEMMOTCOPY    =  ../generic/gemm_tcopy_8.c
-SGEMMINCOPYOBJ =  sgemm_incopy$(TSUFFIX).$(SUFFIX) 
-SGEMMITCOPYOBJ =  sgemm_itcopy$(TSUFFIX).$(SUFFIX) 
+SGEMMINCOPYOBJ =   
+SGEMMITCOPYOBJ =   
 SGEMMONCOPYOBJ =  sgemm_oncopy$(TSUFFIX).$(SUFFIX)
 SGEMMOTCOPYOBJ =  sgemm_otcopy$(TSUFFIX).$(SUFFIX)
-DGEMMKERNEL    =  gemm_kernel_2x8_nehalem.S
-DGEMMINCOPY    =  dgemm_ncopy_2.S
-DGEMMITCOPY    =  dgemm_tcopy_2.S
-DGEMMONCOPY    =  dgemm_ncopy_8.S
-DGEMMOTCOPY    =  dgemm_tcopy_8.S
+DGEMMKERNEL    =  dgemm_kernel_4x8_sandy.S
+DGEMMINCOPY    =  ../generic/gemm_ncopy_8.c
+DGEMMITCOPY    =  ../generic/gemm_tcopy_8.c
+#DGEMMONCOPY    =  gemm_ncopy_4.S
+DGEMMONCOPY    =  ../generic/gemm_ncopy_4.c
+DGEMMOTCOPY    =  ../generic/gemm_tcopy_4.c
+#DGEMMOTCOPY    =  gemm_tcopy_4.S
 DGEMMINCOPYOBJ =  dgemm_incopy$(TSUFFIX).$(SUFFIX)
 DGEMMITCOPYOBJ =  dgemm_itcopy$(TSUFFIX).$(SUFFIX)
 DGEMMONCOPYOBJ =  dgemm_oncopy$(TSUFFIX).$(SUFFIX)
 DGEMMOTCOPYOBJ =  dgemm_otcopy$(TSUFFIX).$(SUFFIX)
-CGEMMKERNEL    =  zgemm_kernel_2x4_nehalem.S
-CGEMMINCOPY    =  zgemm_ncopy_2.S
-CGEMMITCOPY    =  zgemm_tcopy_2.S
-CGEMMONCOPY    =  ../generic/zgemm_ncopy_4.c
-CGEMMOTCOPY    =  ../generic/zgemm_tcopy_4.c
+#CGEMMKERNEL    =  zgemm_kernel_2x4_nehalem.S
+CGEMMKERNEL    =  cgemm_kernel_4x8_sandy.S
+CGEMMINCOPY    =  ../generic/zgemm_ncopy_8_sandy.c
+CGEMMITCOPY    =  ../generic/zgemm_tcopy_8_sandy.c
+CGEMMONCOPY    =  ../generic/zgemm_ncopy_4_sandy.c
+CGEMMOTCOPY    =  ../generic/zgemm_tcopy_4_sandy.c
 CGEMMINCOPYOBJ =  cgemm_incopy$(TSUFFIX).$(SUFFIX)
 CGEMMITCOPYOBJ =  cgemm_itcopy$(TSUFFIX).$(SUFFIX)
 CGEMMONCOPYOBJ =  cgemm_oncopy$(TSUFFIX).$(SUFFIX)
 CGEMMOTCOPYOBJ =  cgemm_otcopy$(TSUFFIX).$(SUFFIX)
-ZGEMMKERNEL    =  zgemm_kernel_1x4_nehalem.S
-ZGEMMINCOPY    =  zgemm_ncopy_1.S
-ZGEMMITCOPY    =  zgemm_tcopy_1.S
+#ZGEMMKERNEL    =  zgemm_kernel_1x4_nehalem.S
+ZGEMMKERNEL    =  zgemm_kernel_4x4_sandy.S
+ZGEMMINCOPY    =  
+ZGEMMITCOPY    = 
 ZGEMMONCOPY    =  ../generic/zgemm_ncopy_4.c
 ZGEMMOTCOPY    =  ../generic/zgemm_tcopy_4.c
-ZGEMMINCOPYOBJ =  zgemm_incopy$(TSUFFIX).$(SUFFIX)
-ZGEMMITCOPYOBJ =  zgemm_itcopy$(TSUFFIX).$(SUFFIX)
+ZGEMMINCOPYOBJ =  
+ZGEMMITCOPYOBJ =  
 ZGEMMONCOPYOBJ =  zgemm_oncopy$(TSUFFIX).$(SUFFIX)
 ZGEMMOTCOPYOBJ =  zgemm_otcopy$(TSUFFIX).$(SUFFIX)
 
-STRSMKERNEL_LN	=  trsm_kernel_LN_4x8_nehalem.S
-STRSMKERNEL_LT	=  trsm_kernel_LT_4x8_nehalem.S
-STRSMKERNEL_RN	=  trsm_kernel_LT_4x8_nehalem.S
-STRSMKERNEL_RT	=  trsm_kernel_RT_4x8_nehalem.S
+#STRSMKERNEL_LN	=  trsm_kernel_LN_4x8_nehalem.S
+#STRSMKERNEL_LT	=  trsm_kernel_LT_4x8_nehalem.S
+#STRSMKERNEL_RN	=  trsm_kernel_LT_4x8_nehalem.S
+#STRSMKERNEL_RT	=  trsm_kernel_RT_4x8_nehalem.S
 
-DTRSMKERNEL_LN	=  trsm_kernel_LN_2x8_nehalem.S
-DTRSMKERNEL_LT	=  trsm_kernel_LT_2x8_nehalem.S
-DTRSMKERNEL_RN	=  trsm_kernel_LT_2x8_nehalem.S
-DTRSMKERNEL_RT	=  trsm_kernel_RT_2x8_nehalem.S
+#DTRSMKERNEL_LN	=  trsm_kernel_LN_2x8_nehalem.S
+#DTRSMKERNEL_LT	=  trsm_kernel_LT_2x8_nehalem.S
+#DTRSMKERNEL_RN	=  trsm_kernel_LT_2x8_nehalem.S
+#DTRSMKERNEL_RT	=  trsm_kernel_RT_2x8_nehalem.S
+
+#CTRSMKERNEL_LN	=  ztrsm_kernel_LN_2x4_nehalem.S
+#CTRSMKERNEL_LT	=  ztrsm_kernel_LT_2x4_nehalem.S
+#CTRSMKERNEL_RN	=  ztrsm_kernel_LT_2x4_nehalem.S
+#CTRSMKERNEL_RT	=  ztrsm_kernel_RT_2x4_nehalem.S
+
+#ZTRSMKERNEL_LN	=  ztrsm_kernel_LT_1x4_nehalem.S
+#ZTRSMKERNEL_LT	=  ztrsm_kernel_LT_1x4_nehalem.S
+#ZTRSMKERNEL_RN	=  ztrsm_kernel_LT_1x4_nehalem.S
+#ZTRSMKERNEL_RT	=  ztrsm_kernel_RT_1x4_nehalem.S
+STRSMKERNEL_LN	=  ../generic/trsm_kernel_LN.c
+STRSMKERNEL_LT	=  ../generic/trsm_kernel_LT.c
+STRSMKERNEL_RN	=  ../generic/trsm_kernel_RN.c
+STRSMKERNEL_RT	=  ../generic/trsm_kernel_RT.c
+
+DTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
+DTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
+DTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
+DTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
+
+CTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
+CTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
+CTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
+CTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
+
+ZTRSMKERNEL_LN	= ../generic/trsm_kernel_LN.c
+ZTRSMKERNEL_LT	= ../generic/trsm_kernel_LT.c
+ZTRSMKERNEL_RN	= ../generic/trsm_kernel_RN.c
+ZTRSMKERNEL_RT	= ../generic/trsm_kernel_RT.c
 
-CTRSMKERNEL_LN	=  ztrsm_kernel_LN_2x4_nehalem.S
-CTRSMKERNEL_LT	=  ztrsm_kernel_LT_2x4_nehalem.S
-CTRSMKERNEL_RN	=  ztrsm_kernel_LT_2x4_nehalem.S
-CTRSMKERNEL_RT	=  ztrsm_kernel_RT_2x4_nehalem.S
 
-ZTRSMKERNEL_LN	=  ztrsm_kernel_LT_1x4_nehalem.S
-ZTRSMKERNEL_LT	=  ztrsm_kernel_LT_1x4_nehalem.S
-ZTRSMKERNEL_RN	=  ztrsm_kernel_LT_1x4_nehalem.S
-ZTRSMKERNEL_RT	=  ztrsm_kernel_RT_1x4_nehalem.S
 
 CGEMM3MKERNEL    =  zgemm3m_kernel_4x8_nehalem.S
 ZGEMM3MKERNEL    =  zgemm3m_kernel_2x8_nehalem.S
diff --git a/kernel/x86_64/cgemm_kernel_4x8_sandy.S b/kernel/x86_64/cgemm_kernel_4x8_sandy.S
new file mode 100644
index 000000000..2b4e4dc64
--- /dev/null
+++ b/kernel/x86_64/cgemm_kernel_4x8_sandy.S
@@ -0,0 +1,4478 @@
+/*****************************************************************************
+  Copyright (c) 2011,2012 Lab of Parallel Software and Computational Science,ISCAS
+  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 ISCAS 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 COPYRIGHT OWNER 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	old_bm		%rdi
+#define old_bn		%rsi
+#define	old_bk		%rdx
+
+#define bm	%r13
+#define bn	%r14
+#define bk	%r15
+
+#define ALPHA	%xmm0
+#define ba		%rcx
+#define bb		%r8
+#define C		%r9
+#define ldc		%r10
+
+#define i	%r11
+#define k	%rax
+
+#define ptrba	%rdi
+#define ptrbb	%rsi
+#define C0		%rbx
+#define C1		%rbp
+
+#define prebb	%r12
+
+#ifndef WINDOWS_ABI
+
+#define STACKSIZE 128 
+
+#define old_ldc	8+STACKSIZE(%rsp)
+#define old_offset	16+STACKSIZE(%rsp)
+
+#define MEMALPHA_R	48(%rsp)
+#define MEMALPHA_I	56(%rsp)
+#define j	64(%rsp)
+#define OFFSET	72(%rsp)
+#define kk	80(%rsp)
+#define kkk	88(%rsp)
+
+#else
+
+#define STACKSIZE 512
+
+#define OLD_ALPHA_I	40 + STACKSIZE(%rsp)
+#define OLD_A		48 + STACKSIZE(%rsp)
+#define OLD_B		56 + STACKSIZE(%rsp)
+#define OLD_C		64 + STACKSIZE(%rsp)
+#define old_ldc		72 + STACKSIZE(%rsp)
+#define old_offset	80 + STACKSIZE(%rsp)
+
+#define MEMALPHA_R	  224(%rsp)
+#define MEMALPHA_I	  232(%rsp)
+#define j	  240(%rsp)
+#define OFFSET	  248(%rsp)
+#define kk	  256(%rsp)
+#define kkk	  264(%rsp)
+
+#endif
+
+#define PREFETCH0	prefetcht0
+#define PREFETCH1	prefetcht0
+#define PREFETCH2	prefetcht0
+#define PRESIZE	64
+
+#define xvec0	%xmm0
+#define xvec1	%xmm1
+#define xvec2	%xmm2
+#define xvec3	%xmm3
+#define xvec4	%xmm4
+#define xvec5	%xmm5
+#define xvec6	%xmm6
+#define xvec7	%xmm7
+#define xvec8	%xmm8
+#define xvec9	%xmm9
+#define xvec10	%xmm10
+#define xvec11	%xmm11
+#define xvec12	%xmm12
+#define xvec13	%xmm13
+#define xvec14	%xmm14
+#define xvec15	%xmm15
+
+#define yvec0	%ymm0
+#define yvec1	%ymm1
+#define yvec2	%ymm2
+#define yvec3	%ymm3
+#define yvec4	%ymm4
+#define yvec5	%ymm5
+#define yvec6	%ymm6
+#define yvec7	%ymm7
+#define yvec8	%ymm8
+#define yvec9	%ymm9
+#define yvec10	%ymm10
+#define yvec11	%ymm11
+#define yvec12	%ymm12
+#define yvec13	%ymm13
+#define yvec14	%ymm14
+#define yvec15	%ymm15
+
+#define LEAQ	leaq
+#define ADDQ	addq
+#define MULQ	imulq
+#define SARQ	sarq
+#define SALQ	salq
+#define ANDQ	andq
+#define SUBQ	subq
+#define DECQ	decq
+#define JG		jg
+#define JLE		jle
+#define TEST	testq
+#define OR		orq
+#define JNE		jne
+#define JMP		jmp
+#define NOP		
+#define XOR		xorpd
+#define MOVQ	movq 
+
+#define XOR_SY	vxorps
+#define XOR_DY	vxorpd
+#define XOR_SX	xorps
+#define XOR_DX	xorpd
+
+#define LD_SY	vmovaps
+#define LD_DY	vmovapd
+#define LD_SX	movaps
+#define LD_DX	movapd
+#define LDL_SX	movlps
+#define LDL_SY	vmovlps
+#define LDH_SX	movhps
+#define LDH_SY	vmovhps
+
+#define ST_SY	vmovaps
+#define ST_DY	vmovapd
+#define ST_SX	movaps
+#define ST_DX	movapd
+#define STL_SX	movlps
+#define STL_SY	vmovlps
+#define STH_SX	movhps
+#define STH_SY	vmovhps
+
+#define EDUP_SY	vmovsldup
+#define ODUP_SY	vmovshdup
+#define EDUP_SX	movsldup
+#define ODUP_SX	movshdup
+#define EDUP_DY	vmovddup
+
+#define ADD_SY	vaddps 
+#define ADD_DY	vaddpd
+#define ADD_SX	addps
+#define ADD_DX	addpd
+#define SUB_DY	vsubpd
+#define SUB_SY	vsubps
+#define SUB_DX	subpd
+#define SUB_SX	subps
+
+#define ADDSUB_DY	vaddsubpd
+#define ADDSUB_DX	addsubpd
+#define ADDSUB_SY	vaddsubps
+#define ADDSUB_SX	addsubps
+
+#define MUL_SY	vmulps
+#define MUL_DY	vmulpd
+#define MUL_SX	mulps
+#define MUL_DX	mulpd
+
+#define SHUF_SY	vperm2f128
+#define SHUF_DY	vperm2f128
+#define SHUF_DX	pshufd
+#define SHUF_SX	pshufd
+
+#define VPERMILP_SY	vpermilps
+#define VPERMILP_SX	vpermilps
+#define VPERMILP_DY	vpermilpd
+
+#define BROAD_SY vbroadcastss
+#define BROAD_DY vbroadcastsd
+#define BROAD_SX vbroadcastss  
+#define BROAD_DX movddup
+
+#define MOV_SY vmovaps
+#define MOV_DY vmovapd
+#define MOV_SX movaps
+#define MOV_DX movapd
+
+#define REVS_SY	vshufps
+#define REVS_DY vshufpd
+#define REVS_SX shufps
+#define REVS_DX	movsd
+
+#define EXTRA_SY vextractf128
+#define EXTRA_DY vextractf128
+
+#if   defined(NN) || defined(NT) || defined(TN) || defined(TT)
+#define ADD1_SY ADD_SY
+#define ADD2_SY	ADDSUB_SY
+#define ADD1_SX	ADD_SX
+#define ADD2_SX	ADDSUB_SX
+#elif defined(NR) || defined(NC) || defined(TR) || defined(TC)
+#define ADD1_SY SUB_SY
+#define ADD2_SY	ADDSUB_SY
+#define ADD1_SX	SUB_SX
+#define ADD2_SX	ADDSUB_SX
+#elif  defined(RN) || defined(RT) || defined(CN) || defined(CT)
+#define ADD1_SY SUB_SY
+#define ADD2_SY	ADDSUB_SY
+#define ADD1_SX	SUB_SX
+#define ADD2_SX	ADDSUB_SX
+#else 
+#define ADD1_SY ADD_SY
+#define ADD2_SY	ADDSUB_SY
+#define ADD1_SX	ADD_SX
+#define ADD2_SX	ADDSUB_SX
+#endif
+
+PROLOGUE
+
+subq    $STACKSIZE, %rsp;
+movq    %rbx,  0(%rsp);
+movq    %rbp,  8(%rsp);
+movq    %r12, 16(%rsp);
+movq    %r13, 24(%rsp);
+movq    %r14, 32(%rsp);
+movq    %r15, 40(%rsp);
+
+#ifdef WINDOWS_ABI
+	movq	%rdi,    48(%rsp)
+	movq	%rsi,    56(%rsp)
+	movups	%xmm6,   64(%rsp)
+	movups	%xmm7,   80(%rsp)
+	movups	%xmm8,   96(%rsp)
+	movups	%xmm9,  112(%rsp)
+	movups	%xmm10, 128(%rsp)
+	movups	%xmm11, 144(%rsp)
+	movups	%xmm12, 160(%rsp)
+	movups	%xmm13, 176(%rsp)
+	movups	%xmm14, 192(%rsp)
+	movups	%xmm15, 208(%rsp)
+
+	movq	ARG1,      old_bm
+	movq	ARG2,      old_bn
+	movq	ARG3,      old_bk
+	movq	OLD_A,     ba
+	movq	OLD_B,     bb
+	movq	OLD_C,     C
+	movq	old_ldc,   ldc
+#ifdef TRMMKERNEL
+	movq	old_offset, %r11
+#endif
+	movaps	%xmm3, %xmm0
+	movsd	OLD_ALPHA_I, %xmm1
+#else
+movq	old_ldc, ldc
+#ifdef	TRMMKERNEL
+movq	old_offset, %r11;
+#endif
+#endif
+
+vmovlps	%xmm0, MEMALPHA_R
+vmovlps	%xmm1, MEMALPHA_I
+movq	old_bm,	bm
+movq	old_bn, bn
+movq	old_bk, bk
+salq	$ZBASE_SHIFT, ldc
+#ifdef	TRMMKERNEL
+movq	%r11, OFFSET
+#ifndef	LEFT
+negq	%r11;
+#endif
+movq	%r11, kk;
+#endif
+
+MOVQ bn,j;
+SARQ $2,j;							# Rn = 4
+JLE .L0_loopE;
+.align 32;
+.L0_bodyB:;
+#if defined(TRMMKERNEL) && defined(LEFT)
+MOVQ OFFSET, %rax;
+MOVQ %rax, kk;
+#endif
+MOVQ C,C0;
+LEAQ (C,ldc,2),C1;
+MOVQ bk, k;
+SALQ $5, k;
+LEAQ (bb, k, 1), prebb;				# Rn=4, SIZE=4 COMPLEX=2
+MOVQ ba,ptrba;
+MOVQ bm,i;
+SARQ $3,i;							# Rm = 8
+JLE .L1_loopE;
+.align 32;
+.L1_bodyB:;
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb,ptrbb;
+#else
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+LEAQ (ptrba, %rax, 8), ptrba;
+LEAQ (ptrbb, %rax, 4), ptrbb;
+#endif
+# Initial results register 
+PREFETCH0	0*SIZE(prebb);
+XOR_SY	yvec15, yvec15, yvec15;
+PREFETCH0	16*SIZE(prebb);
+ADDQ	$32*SIZE, prebb;
+XOR_SY	yvec14, yvec14, yvec14;
+PREFETCH2	3*SIZE(C0);
+XOR_SY	yvec13, yvec13, yvec13;
+PREFETCH2	3*SIZE(C0, ldc, 1);
+XOR_SY	yvec12, yvec12, yvec12;
+PREFETCH2	3*SIZE(C1);
+EDUP_SY 0*SIZE(ptrbb), yvec2;		# Br0, Br1, Br2, Br3
+PREFETCH2	3*SIZE(C1, ldc, 1);
+XOR_SY	yvec11, yvec11, yvec11;
+XOR_SY	yvec10, yvec10, yvec10;
+LD_SY 	0*SIZE(ptrba), yvec0;			# Ar0, Ai0, Ar1, Ai1..
+XOR_SY	yvec9, yvec9, yvec9;
+XOR_SY	yvec8, yvec8, yvec8;
+VPERMILP_SY	$0x4e, yvec2, yvec3;	# Br2, Br3, Br0, Br1
+#ifndef	TRMMKERNEL
+MOVQ bk,k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else
+MOVQ kk, %rax;
+#ifdef	LEFT
+ADDQ $8, %rax;
+#else 
+ADDQ $4, %rax;
+#endif
+MOVQ %rax, kkk;
+#endif
+SARQ $2,k;							# Unroll 4 times
+JLE .L2_loopE;
+.align 32;
+.L2_bodyB:;
+# Computing kernel
+
+######### Unroll 1 ##################
+PREFETCH0	PRESIZE*SIZE(ptrba);
+LD_SY 8*SIZE(ptrba), yvec1;			# Ar4, Ai4, Ar5, Ai5..
+MUL_SY	yvec0, yvec2, yvec6;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;	# Br1, Br0, Br3, Br2
+MUL_SY	yvec0, yvec3, yvec7;
+SHUF_SY $0x03, yvec3, yvec3, yvec5;	# Br3, Br2, Br1, Br0
+ADD1_SY	yvec6, yvec15, yvec15;
+ADD1_SY	yvec7, yvec13, yvec13;
+
+MUL_SY	yvec1, yvec2, yvec6;
+ODUP_SY	0*SIZE(ptrbb), yvec2;		# Bi0, Bi1, Bi2, Bi3
+MUL_SY	yvec1, yvec3, yvec7;
+VPERMILP_SY	$0x4e, yvec2, yvec3;	# Bi2, Bi3, Bi0, Bi1
+ADD1_SY	yvec6, yvec14, yvec14;
+ADD1_SY	yvec7, yvec12, yvec12;
+
+MUL_SY	yvec0, yvec4, yvec6;	
+MUL_SY 	yvec0, yvec5, yvec7;
+VPERMILP_SY $0xb1, yvec0, yvec0;	# Ai0, Ar0, Ai1, Ar1..
+ADD1_SY	yvec6, yvec11, yvec11;
+ADD1_SY	yvec7, yvec9, yvec9;
+
+MUL_SY	yvec1, yvec4, yvec6;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;	# Br1, Br0, Br3, Br2
+MUL_SY	yvec1, yvec5, yvec7;
+SHUF_SY $0x03, yvec3, yvec3, yvec5;	# Br3, Br2, Br1, Br0
+ADD1_SY	yvec6, yvec10, yvec10;
+ADD1_SY	yvec7, yvec8, yvec8;
+
+VPERMILP_SY $0xb1, yvec1, yvec1;
+MUL_SY	yvec0, yvec2, yvec6;
+MUL_SY	yvec0, yvec3, yvec7;
+ADD2_SY	yvec6, yvec15, yvec15;
+ADD2_SY	yvec7, yvec13, yvec13;
+
+MUL_SY	yvec1, yvec2, yvec6;
+EDUP_SY	8*SIZE(ptrbb), yvec2;
+MUL_SY	yvec1, yvec3, yvec7;
+VPERMILP_SY $0x4e, yvec2, yvec3;
+ADD2_SY	yvec6, yvec14, yvec14;
+ADD2_SY	yvec7, yvec12, yvec12;
+
+MUL_SY	yvec0, yvec4, yvec6;
+MUL_SY	yvec0, yvec5, yvec7;
+LD_SY	16*SIZE(ptrba), yvec0;
+ADD2_SY	yvec6, yvec11, yvec11;
+ADD2_SY	yvec7, yvec9, yvec9;
+
+MUL_SY	yvec1, yvec4, yvec6;
+MUL_SY	yvec1, yvec5, yvec7;
+ADD2_SY	yvec6, yvec10, yvec10;
+ADD2_SY	yvec7, yvec8, yvec8;
+
+######### Unroll 2 ##################
+PREFETCH0	(PRESIZE+16)*SIZE(ptrba);
+LD_SY 	24*SIZE(ptrba), yvec1;			# Ar4, Ai4, Ar5, Ai5..
+MUL_SY	yvec0, yvec2, yvec6;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;	# Br1, Br0, Br3, Br2
+MUL_SY	yvec0, yvec3, yvec7;
+SHUF_SY $0x03, yvec3, yvec3, yvec5;	# Br3, Br2, Br1, Br0
+ADD1_SY	yvec6, yvec15, yvec15;
+ADD1_SY	yvec7, yvec13, yvec13;
+
+MUL_SY	yvec1, yvec2, yvec6;
+ODUP_SY	8*SIZE(ptrbb), yvec2;		# Bi0, Bi1, Bi2, Bi3
+MUL_SY	yvec1, yvec3, yvec7;
+VPERMILP_SY	$0x4e, yvec2, yvec3;	# Bi2, Bi3, Bi0, Bi1
+ADD1_SY	yvec6, yvec14, yvec14;
+ADD1_SY	yvec7, yvec12, yvec12;
+
+MUL_SY	yvec0, yvec4, yvec6;	
+MUL_SY 	yvec0, yvec5, yvec7;
+VPERMILP_SY $0xb1, yvec0, yvec0;	# Ai0, Ar0, Ai1, Ar1..
+ADD1_SY	yvec6, yvec11, yvec11;
+ADD1_SY	yvec7, yvec9, yvec9;
+
+MUL_SY	yvec1, yvec4, yvec6;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;	# Br1, Br0, Br3, Br2
+MUL_SY	yvec1, yvec5, yvec7;
+SHUF_SY $0x03, yvec3, yvec3, yvec5;	# Br3, Br2, Br1, Br0
+ADD1_SY	yvec6, yvec10, yvec10;
+ADD1_SY	yvec7, yvec8, yvec8;
+
+VPERMILP_SY $0xb1, yvec1, yvec1;
+MUL_SY	yvec0, yvec2, yvec6;
+MUL_SY	yvec0, yvec3, yvec7;
+ADD2_SY	yvec6, yvec15, yvec15;
+ADD2_SY	yvec7, yvec13, yvec13;
+
+MUL_SY	yvec1, yvec2, yvec6;
+EDUP_SY	16*SIZE(ptrbb), yvec2;
+MUL_SY	yvec1, yvec3, yvec7;
+VPERMILP_SY $0x4e, yvec2, yvec3;
+ADD2_SY	yvec6, yvec14, yvec14;
+ADD2_SY	yvec7, yvec12, yvec12;
+
+MUL_SY	yvec0, yvec4, yvec6;
+MUL_SY	yvec0, yvec5, yvec7;
+LD_SY	32*SIZE(ptrba), yvec0;
+ADD2_SY	yvec6, yvec11, yvec11;
+ADD2_SY	yvec7, yvec9, yvec9;
+
+MUL_SY	yvec1, yvec4, yvec6;
+MUL_SY	yvec1, yvec5, yvec7;
+ADD2_SY	yvec6, yvec10, yvec10;
+ADD2_SY	yvec7, yvec8, yvec8;
+
+######### Unroll 3 ##################
+PREFETCH0	(PRESIZE+32)*SIZE(ptrba);
+LD_SY 	40*SIZE(ptrba), yvec1;			# Ar4, Ai4, Ar5, Ai5..
+MUL_SY	yvec0, yvec2, yvec6;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;	# Br1, Br0, Br3, Br2
+MUL_SY	yvec0, yvec3, yvec7;
+SHUF_SY $0x03, yvec3, yvec3, yvec5;	# Br3, Br2, Br1, Br0
+ADD1_SY	yvec6, yvec15, yvec15;
+ADD1_SY	yvec7, yvec13, yvec13;
+
+MUL_SY	yvec1, yvec2, yvec6;
+ODUP_SY	16*SIZE(ptrbb), yvec2;		# Bi0, Bi1, Bi2, Bi3
+MUL_SY	yvec1, yvec3, yvec7;
+VPERMILP_SY	$0x4e, yvec2, yvec3;	# Bi2, Bi3, Bi0, Bi1
+ADD1_SY	yvec6, yvec14, yvec14;
+ADD1_SY	yvec7, yvec12, yvec12;
+
+MUL_SY	yvec0, yvec4, yvec6;	
+MUL_SY 	yvec0, yvec5, yvec7;
+VPERMILP_SY $0xb1, yvec0, yvec0;	# Ai0, Ar0, Ai1, Ar1..
+ADD1_SY	yvec6, yvec11, yvec11;
+ADD1_SY	yvec7, yvec9, yvec9;
+
+MUL_SY	yvec1, yvec4, yvec6;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;	# Br1, Br0, Br3, Br2
+MUL_SY	yvec1, yvec5, yvec7;
+SHUF_SY $0x03, yvec3, yvec3, yvec5;	# Br3, Br2, Br1, Br0
+ADD1_SY	yvec6, yvec10, yvec10;
+ADD1_SY	yvec7, yvec8, yvec8;
+
+VPERMILP_SY $0xb1, yvec1, yvec1;
+MUL_SY	yvec0, yvec2, yvec6;
+MUL_SY	yvec0, yvec3, yvec7;
+ADD2_SY	yvec6, yvec15, yvec15;
+ADD2_SY	yvec7, yvec13, yvec13;
+
+MUL_SY	yvec1, yvec2, yvec6;
+EDUP_SY	24*SIZE(ptrbb), yvec2;
+MUL_SY	yvec1, yvec3, yvec7;
+VPERMILP_SY $0x4e, yvec2, yvec3;
+ADD2_SY	yvec6, yvec14, yvec14;
+ADD2_SY	yvec7, yvec12, yvec12;
+
+MUL_SY	yvec0, yvec4, yvec6;
+MUL_SY	yvec0, yvec5, yvec7;
+LD_SY	48*SIZE(ptrba), yvec0;
+ADD2_SY	yvec6, yvec11, yvec11;
+ADD2_SY	yvec7, yvec9, yvec9;
+
+MUL_SY	yvec1, yvec4, yvec6;
+MUL_SY	yvec1, yvec5, yvec7;
+ADD2_SY	yvec6, yvec10, yvec10;
+ADD2_SY	yvec7, yvec8, yvec8;
+
+######### Unroll 4 ##################
+PREFETCH0	(PRESIZE+48)*SIZE(ptrba);
+LD_SY 	56*SIZE(ptrba), yvec1;			# Ar4, Ai4, Ar5, Ai5..
+MUL_SY	yvec0, yvec2, yvec6;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;	# Br1, Br0, Br3, Br2
+MUL_SY	yvec0, yvec3, yvec7;
+SHUF_SY $0x03, yvec3, yvec3, yvec5;	# Br3, Br2, Br1, Br0
+ADDQ	$64*SIZE, ptrba;
+ADD1_SY	yvec6, yvec15, yvec15;
+ADD1_SY	yvec7, yvec13, yvec13;
+
+MUL_SY	yvec1, yvec2, yvec6;
+ODUP_SY	24*SIZE(ptrbb), yvec2;		# Bi0, Bi1, Bi2, Bi3
+MUL_SY	yvec1, yvec3, yvec7;
+VPERMILP_SY	$0x4e, yvec2, yvec3;	# Bi2, Bi3, Bi0, Bi1
+ADDQ	$32*SIZE, ptrbb;
+ADD1_SY	yvec6, yvec14, yvec14;
+ADD1_SY	yvec7, yvec12, yvec12;
+
+MUL_SY	yvec0, yvec4, yvec6;	
+MUL_SY 	yvec0, yvec5, yvec7;
+VPERMILP_SY $0xb1, yvec0, yvec0;	# Ai0, Ar0, Ai1, Ar1..
+ADD1_SY	yvec6, yvec11, yvec11;
+ADD1_SY	yvec7, yvec9, yvec9;
+
+MUL_SY	yvec1, yvec4, yvec6;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;	# Br1, Br0, Br3, Br2
+MUL_SY	yvec1, yvec5, yvec7;
+SHUF_SY $0x03, yvec3, yvec3, yvec5;	# Br3, Br2, Br1, Br0
+ADD1_SY	yvec6, yvec10, yvec10;
+ADD1_SY	yvec7, yvec8, yvec8;
+
+VPERMILP_SY $0xb1, yvec1, yvec1;
+MUL_SY	yvec0, yvec2, yvec6;
+MUL_SY	yvec0, yvec3, yvec7;
+ADD2_SY	yvec6, yvec15, yvec15;
+ADD2_SY	yvec7, yvec13, yvec13;
+
+MUL_SY	yvec1, yvec2, yvec6;
+EDUP_SY	0*SIZE(ptrbb), yvec2;
+MUL_SY	yvec1, yvec3, yvec7;
+VPERMILP_SY $0x4e, yvec2, yvec3;
+ADD2_SY	yvec6, yvec14, yvec14;
+ADD2_SY	yvec7, yvec12, yvec12;
+
+MUL_SY	yvec0, yvec4, yvec6;
+MUL_SY	yvec0, yvec5, yvec7;
+LD_SY	0*SIZE(ptrba), yvec0;
+ADD2_SY	yvec6, yvec11, yvec11;
+ADD2_SY	yvec7, yvec9, yvec9;
+
+MUL_SY	yvec1, yvec4, yvec6;
+MUL_SY	yvec1, yvec5, yvec7;
+ADD2_SY	yvec6, yvec10, yvec10;
+ADD2_SY	yvec7, yvec8, yvec8;
+.L2_bodyE:;
+DECQ k;
+JG .L2_bodyB;
+.align	64;
+.L2_loopE:;
+#ifndef	TRMMKERNEL
+TEST	$2, bk;
+#else 
+TEST 	$2, kkk;
+#endif
+JLE		.L3_loopE;
+.align	64
+.L3_loopB:
+######### Unroll 1 ##################
+PREFETCH0	PRESIZE*SIZE(ptrba)
+LD_SY 8*SIZE(ptrba), yvec1;			# Ar4, Ai4, Ar5, Ai5..
+MUL_SY	yvec0, yvec2, yvec6;
+MUL_SY	yvec0, yvec3, yvec7;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;	# Br1, Br0, Br3, Br2
+ADD1_SY	yvec6, yvec15, yvec15;
+ADD1_SY	yvec7, yvec13, yvec13;
+
+MUL_SY	yvec1, yvec2, yvec6;
+MUL_SY	yvec1, yvec3, yvec7;
+SHUF_SY $0x03, yvec3, yvec3, yvec5;	# Br3, Br2, Br1, Br0
+ADD1_SY	yvec6, yvec14, yvec14;
+ADD1_SY	yvec7, yvec12, yvec12;
+
+ODUP_SY	0*SIZE(ptrbb), yvec2;		# Bi0, Bi1, Bi2, Bi3
+MUL_SY	yvec0, yvec4, yvec6;	
+MUL_SY 	yvec0, yvec5, yvec7;
+VPERMILP_SY	$0x4e, yvec2, yvec3;	# Bi2, Bi3, Bi0, Bi1
+ADD1_SY	yvec6, yvec11, yvec11;
+ADD1_SY	yvec7, yvec9, yvec9;
+
+MUL_SY	yvec1, yvec4, yvec6;
+MUL_SY	yvec1, yvec5, yvec7;
+VPERMILP_SY $0xb1, yvec0, yvec0;	# Ai0, Ar0, Ai1, Ar1..
+ADD1_SY	yvec6, yvec10, yvec10;
+ADD1_SY	yvec7, yvec8, yvec8;
+
+VPERMILP_SY $0xb1, yvec1, yvec1;
+MUL_SY	yvec0, yvec2, yvec6;
+MUL_SY	yvec0, yvec3, yvec7;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;	# Br1, Br0, Br3, Br2
+ADD2_SY	yvec6, yvec15, yvec15;
+ADD2_SY	yvec7, yvec13, yvec13;
+
+MUL_SY	yvec1, yvec2, yvec6;
+MUL_SY	yvec1, yvec3, yvec7;
+SHUF_SY $0x03, yvec3, yvec3, yvec5;	# Br3, Br2, Br1, Br0
+ADD2_SY	yvec6, yvec14, yvec14;
+ADD2_SY	yvec7, yvec12, yvec12;
+
+EDUP_SY	8*SIZE(ptrbb), yvec2;
+MUL_SY	yvec0, yvec4, yvec6;
+MUL_SY	yvec0, yvec5, yvec7;
+VPERMILP_SY $0x4e, yvec2, yvec3;
+ADD2_SY	yvec6, yvec11, yvec11;
+ADD2_SY	yvec7, yvec9, yvec9;
+
+LD_SY	16*SIZE(ptrba), yvec0;
+MUL_SY	yvec1, yvec4, yvec6;
+MUL_SY	yvec1, yvec5, yvec7;
+ADD2_SY	yvec6, yvec10, yvec10;
+ADD2_SY	yvec7, yvec8, yvec8;
+
+######### Unroll 2 ##################
+PREFETCH0	(PRESIZE+16)*SIZE(ptrba)
+LD_SY 	24*SIZE(ptrba), yvec1;			# Ar4, Ai4, Ar5, Ai5..
+MUL_SY	yvec0, yvec2, yvec6;
+MUL_SY	yvec0, yvec3, yvec7;
+ADDQ	$32*SIZE, ptrba
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;	# Br1, Br0, Br3, Br2
+ADD1_SY	yvec6, yvec15, yvec15;
+ADD1_SY	yvec7, yvec13, yvec13;
+
+MUL_SY	yvec1, yvec2, yvec6;
+MUL_SY	yvec1, yvec3, yvec7;
+SHUF_SY $0x03, yvec3, yvec3, yvec5;	# Br3, Br2, Br1, Br0
+ADD1_SY	yvec6, yvec14, yvec14;
+ADD1_SY	yvec7, yvec12, yvec12;
+
+ODUP_SY	8*SIZE(ptrbb), yvec2;		# Bi0, Bi1, Bi2, Bi3
+MUL_SY	yvec0, yvec4, yvec6;	
+MUL_SY 	yvec0, yvec5, yvec7;
+ADDQ	$16*SIZE, ptrbb;
+VPERMILP_SY	$0x4e, yvec2, yvec3;	# Bi2, Bi3, Bi0, Bi1
+ADD1_SY	yvec6, yvec11, yvec11;
+ADD1_SY	yvec7, yvec9, yvec9;
+
+MUL_SY	yvec1, yvec4, yvec6;
+MUL_SY	yvec1, yvec5, yvec7;
+VPERMILP_SY $0xb1, yvec0, yvec0;	# Ai0, Ar0, Ai1, Ar1..
+ADD1_SY	yvec6, yvec10, yvec10;
+ADD1_SY	yvec7, yvec8, yvec8;
+
+VPERMILP_SY $0xb1, yvec1, yvec1;
+MUL_SY	yvec0, yvec2, yvec6;
+MUL_SY	yvec0, yvec3, yvec7;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;	# Br1, Br0, Br3, Br2
+ADD2_SY	yvec6, yvec15, yvec15;
+ADD2_SY	yvec7, yvec13, yvec13;
+
+MUL_SY	yvec1, yvec2, yvec6;
+MUL_SY	yvec1, yvec3, yvec7;
+SHUF_SY $0x03, yvec3, yvec3, yvec5;	# Br3, Br2, Br1, Br0
+ADD2_SY	yvec6, yvec14, yvec14;
+ADD2_SY	yvec7, yvec12, yvec12;
+
+EDUP_SY	0*SIZE(ptrbb), yvec2;
+MUL_SY	yvec0, yvec4, yvec6;
+MUL_SY	yvec0, yvec5, yvec7;
+VPERMILP_SY $0x4e, yvec2, yvec3;
+ADD2_SY	yvec6, yvec11, yvec11;
+ADD2_SY	yvec7, yvec9, yvec9;
+
+LD_SY	0*SIZE(ptrba), yvec0;
+MUL_SY	yvec1, yvec4, yvec6;
+MUL_SY	yvec1, yvec5, yvec7;
+ADD2_SY	yvec6, yvec10, yvec10;
+ADD2_SY	yvec7, yvec8, yvec8;
+.L3_loopE:
+#ifndef	TRMMKERNEL
+TEST	$1, bk;
+#else
+TEST 	$1, kkk;
+#endif
+JLE		.L4_loopE;
+.align	64
+.L4_loopB:;
+######### Unroll 1 ##################
+PREFETCH0	PRESIZE*SIZE(ptrba)
+LD_SY	8*SIZE(ptrba), yvec1;			# Ar4, Ai4, Ar5, Ai5..
+MUL_SY	yvec0, yvec2, yvec6;
+MUL_SY	yvec0, yvec3, yvec7;
+ADDQ	$16*SIZE, ptrba;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;	# Br1, Br0, Br3, Br2
+ADD1_SY	yvec6, yvec15, yvec15;
+ADD1_SY	yvec7, yvec13, yvec13;
+
+MUL_SY	yvec1, yvec2, yvec6;
+MUL_SY	yvec1, yvec3, yvec7;
+SHUF_SY $0x03, yvec3, yvec3, yvec5;	# Br3, Br2, Br1, Br0
+ADD1_SY	yvec6, yvec14, yvec14;
+ADD1_SY	yvec7, yvec12, yvec12;
+
+ODUP_SY	0*SIZE(ptrbb), yvec2;		# Bi0, Bi1, Bi2, Bi3
+MUL_SY	yvec0, yvec4, yvec6;	
+MUL_SY 	yvec0, yvec5, yvec7;
+ADDQ	$8*SIZE, ptrbb;
+VPERMILP_SY	$0x4e, yvec2, yvec3;	# Bi2, Bi3, Bi0, Bi1
+ADD1_SY	yvec6, yvec11, yvec11;
+ADD1_SY	yvec7, yvec9, yvec9;
+
+MUL_SY	yvec1, yvec4, yvec6;
+MUL_SY	yvec1, yvec5, yvec7;
+VPERMILP_SY $0xb1, yvec0, yvec0;	# Ai0, Ar0, Ai1, Ar1..
+ADD1_SY	yvec6, yvec10, yvec10;
+ADD1_SY	yvec7, yvec8, yvec8;
+
+VPERMILP_SY $0xb1, yvec1, yvec1;
+MUL_SY	yvec0, yvec2, yvec6;
+MUL_SY	yvec0, yvec3, yvec7;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;	# Br1, Br0, Br3, Br2
+ADD2_SY	yvec6, yvec15, yvec15;
+ADD2_SY	yvec7, yvec13, yvec13;
+
+MUL_SY	yvec1, yvec2, yvec6;
+ADD2_SY	yvec6, yvec14, yvec14;
+SHUF_SY $0x03, yvec3, yvec3, yvec5;	# Br3, Br2, Br1, Br0
+MUL_SY	yvec1, yvec3, yvec7;
+ADD2_SY	yvec7, yvec12, yvec12;
+
+MUL_SY	yvec0, yvec4, yvec6;
+MUL_SY	yvec0, yvec5, yvec7;
+VPERMILP_SY $0x4e, yvec2, yvec3;
+ADD2_SY	yvec6, yvec11, yvec11;
+ADD2_SY	yvec7, yvec9, yvec9;
+
+MUL_SY	yvec1, yvec4, yvec6;
+MUL_SY	yvec1, yvec5, yvec7;
+ADD2_SY	yvec6, yvec10, yvec10;
+ADD2_SY	yvec7, yvec8, yvec8;
+
+.L4_loopE:;
+#### Handle ####
+XOR_SY	yvec7, yvec7, yvec7;
+#if  defined(RN) || defined(RT) || defined(CN) || defined(CT)
+ADDSUB_SY	yvec15, yvec7, yvec15;
+ADDSUB_SY	yvec14, yvec7, yvec14;
+ADDSUB_SY	yvec13, yvec7, yvec13;
+ADDSUB_SY	yvec12, yvec7, yvec12;
+ADDSUB_SY	yvec11, yvec7, yvec11;
+ADDSUB_SY	yvec10, yvec7, yvec10;
+ADDSUB_SY	yvec9, yvec7, yvec9;
+ADDSUB_SY	yvec8, yvec7, yvec8;
+#elif defined(NR) || defined(NC) || defined(TR) || defined(TC)
+SUB_SY	yvec15, yvec7, yvec15;
+SUB_SY	yvec14, yvec7, yvec14;
+SUB_SY	yvec13, yvec7, yvec13;
+SUB_SY	yvec12, yvec7, yvec12;
+SUB_SY	yvec11, yvec7, yvec11;
+SUB_SY	yvec10, yvec7, yvec10;
+SUB_SY	yvec9, yvec7, yvec9;
+SUB_SY	yvec8, yvec7, yvec8;
+#elif defined(RR) || defined(RC) || defined(CR) || defined(CC)
+VPERMILP_SY $0xb1, yvec15, yvec15;
+VPERMILP_SY $0xb1, yvec14, yvec14;
+VPERMILP_SY $0xb1, yvec13, yvec13;
+VPERMILP_SY $0xb1, yvec12, yvec12;
+VPERMILP_SY $0xb1, yvec11, yvec11;
+VPERMILP_SY $0xb1, yvec10, yvec10;
+VPERMILP_SY $0xb1, yvec9, yvec9;
+VPERMILP_SY $0xb1, yvec8, yvec8;
+ADDSUB_SY	yvec15, yvec7, yvec15;
+ADDSUB_SY	yvec14, yvec7, yvec14;
+ADDSUB_SY	yvec13, yvec7, yvec13;
+ADDSUB_SY	yvec12, yvec7, yvec12;
+ADDSUB_SY	yvec11, yvec7, yvec11;
+ADDSUB_SY	yvec10, yvec7, yvec10;
+ADDSUB_SY	yvec9, yvec7, yvec9;
+ADDSUB_SY	yvec8, yvec7, yvec8;
+VPERMILP_SY $0xb1, yvec15, yvec15;
+VPERMILP_SY $0xb1, yvec14, yvec14;
+VPERMILP_SY $0xb1, yvec13, yvec13;
+VPERMILP_SY $0xb1, yvec12, yvec12;
+VPERMILP_SY $0xb1, yvec11, yvec11;
+VPERMILP_SY $0xb1, yvec10, yvec10;
+VPERMILP_SY $0xb1, yvec9, yvec9;
+VPERMILP_SY $0xb1, yvec8, yvec8;
+#endif
+##### Load Alpha ####
+BROAD_SY MEMALPHA_R,yvec7;
+BROAD_SY MEMALPHA_I,yvec6;
+##### Multiply Alpha ####
+VPERMILP_SY $0xb1,yvec15, yvec5;
+MUL_SY	yvec15, yvec7, yvec15;
+MUL_SY	yvec5, yvec6, yvec5;
+ADDSUB_SY	yvec5, yvec15, yvec15;
+VPERMILP_SY $0xb1,yvec14, yvec4;
+MUL_SY	yvec14, yvec7, yvec14;
+MUL_SY	yvec4, yvec6, yvec4;
+ADDSUB_SY	yvec4, yvec14, yvec14;
+VPERMILP_SY $0xb1,yvec13, yvec3;
+MUL_SY	yvec13, yvec7, yvec13;
+MUL_SY	yvec3, yvec6, yvec3;
+ADDSUB_SY	yvec3, yvec13, yvec13;
+VPERMILP_SY $0xb1,yvec12, yvec2;
+MUL_SY	yvec12, yvec7, yvec12;
+MUL_SY	yvec2, yvec6, yvec2;
+ADDSUB_SY	yvec2, yvec12, yvec12;
+VPERMILP_SY $0xb1,yvec11, yvec1;
+MUL_SY	yvec11, yvec7, yvec11;
+MUL_SY	yvec1, yvec6, yvec1;
+ADDSUB_SY	yvec1, yvec11, yvec11;
+VPERMILP_SY $0xb1,yvec10, yvec0;
+MUL_SY	yvec10, yvec7, yvec10;
+MUL_SY	yvec0, yvec6, yvec0;
+ADDSUB_SY	yvec0, yvec10, yvec10;
+VPERMILP_SY $0xb1,yvec9, yvec5;
+MUL_SY	yvec9, yvec7, yvec9;
+MUL_SY	yvec5, yvec6, yvec5;
+ADDSUB_SY	yvec5, yvec9, yvec9;
+VPERMILP_SY $0xb1,yvec8, yvec4;
+MUL_SY	yvec8, yvec7, yvec8;
+MUL_SY	yvec4, yvec6, yvec4;
+ADDSUB_SY	yvec4, yvec8, yvec8;
+#### Shuffle Results ####
+MOV_SY yvec15,yvec7;
+REVS_SY $0xe4,yvec13,yvec15,yvec15;
+REVS_SY $0xe4,yvec7,yvec13,yvec13;
+MOV_SY yvec14,yvec7;
+REVS_SY $0xe4,yvec12,yvec14,yvec14;
+REVS_SY $0xe4,yvec7,yvec12,yvec12;
+MOV_SY yvec11,yvec7;
+REVS_SY $0xe4,yvec9,yvec11,yvec11;
+REVS_SY $0xe4,yvec7,yvec9,yvec9;
+MOV_SY yvec10,yvec7;
+REVS_SY $0xe4,yvec8,yvec10,yvec10;
+REVS_SY $0xe4,yvec7,yvec8,yvec8;
+#### Store Back ####
+#### Testing alignment ####
+MOVQ	C0, %rax;
+OR		ldc, %rax;
+TEST 	$15, %rax;
+JNE		.L4_loopEx;
+.align 32
+EXTRA_SY $1,yvec15,xvec7;
+EXTRA_SY $1,yvec14,xvec6;
+EXTRA_SY $1,yvec13,xvec5;
+EXTRA_SY $1,yvec12,xvec4;
+EXTRA_SY $1,yvec11,xvec3;
+EXTRA_SY $1,yvec10,xvec2;
+EXTRA_SY $1,yvec9,xvec1;
+EXTRA_SY $1,yvec8,xvec0;
+#ifndef	TRMMKERNEL
+ADD_SY 0*SIZE(C0),xvec15, xvec15;
+ADD_SY 4*SIZE(C1),xvec7, xvec7;
+ADD_SY 8*SIZE(C0),xvec14, xvec14;
+ADD_SY 12*SIZE(C1),xvec6, xvec6;
+ADD_SY 0*SIZE(C0,ldc,1),xvec13, xvec13;
+ADD_SY 4*SIZE(C1,ldc,1),xvec5, xvec5;
+ADD_SY 8*SIZE(C0,ldc,1),xvec12, xvec12;
+ADD_SY 12*SIZE(C1,ldc,1),xvec4, xvec4;
+ADD_SY 0*SIZE(C1),xvec11, xvec11;
+ADD_SY 4*SIZE(C0),xvec3, xvec3;
+ADD_SY 8*SIZE(C1),xvec10, xvec10;
+ADD_SY 12*SIZE(C0),xvec2, xvec2;
+ADD_SY 0*SIZE(C1,ldc,1),xvec9, xvec9;
+ADD_SY 4*SIZE(C0,ldc,1),xvec1, xvec1;
+ADD_SY 8*SIZE(C1,ldc,1),xvec8, xvec8;
+ADD_SY 12*SIZE(C0,ldc,1),xvec0, xvec0;
+#endif
+ST_SY xvec15,0*SIZE(C0);
+ST_SY xvec7,4*SIZE(C1);
+ST_SY xvec14,8*SIZE(C0);
+ST_SY xvec6,12*SIZE(C1);
+ST_SY xvec13,0*SIZE(C0,ldc,1);
+ST_SY xvec5,4*SIZE(C1,ldc,1);
+ST_SY xvec12,8*SIZE(C0,ldc,1);
+ST_SY xvec4,12*SIZE(C1,ldc,1);
+ST_SY xvec11,0*SIZE(C1);
+ST_SY xvec3,4*SIZE(C0);
+ST_SY xvec10,8*SIZE(C1);
+ST_SY xvec2,12*SIZE(C0);
+ST_SY xvec9,0*SIZE(C1,ldc,1);
+ST_SY xvec1,4*SIZE(C0,ldc,1);
+ST_SY xvec8,8*SIZE(C1,ldc,1);
+ST_SY xvec0,12*SIZE(C0,ldc,1);
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA))||(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kkk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+LEAQ (ptrba, %rax, 8), ptrba;
+LEAQ (ptrbb, %rax, 4), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT) 
+ADDQ	$8, kk;
+#endif
+ADDQ $16*SIZE,C0;
+ADDQ $16*SIZE,C1;
+.L1_bodyE:;
+DECQ i;
+JG .L1_bodyB;
+JMP .L1_loopE;
+.align 32
+.L4_loopEx:
+EXTRA_SY $1, yvec15, xvec7;
+#ifndef	TRMMKERNEL
+LDL_SY	0*SIZE(C0), xvec6, xvec6;
+LDH_SY	2*SIZE(C0), xvec6, xvec6;
+ADD_SY	xvec6, xvec15, xvec15;
+#endif
+STL_SY	xvec15, 0*SIZE(C0);
+STH_SY	xvec15, 2*SIZE(C0);
+#ifndef	TRMMKERNEL
+LDL_SY	4*SIZE(C1), xvec5, xvec5;
+LDH_SY	6*SIZE(C1), xvec5, xvec5;
+ADD_SY	xvec5, xvec7, xvec7;
+#endif
+STL_SY	xvec7, 4*SIZE(C1);
+STH_SY	xvec7, 6*SIZE(C1);
+
+EXTRA_SY $1, yvec14, xvec6;
+#ifndef	TRMMKERNEL
+LDL_SY	8*SIZE(C0), xvec5, xvec5;
+LDH_SY	10*SIZE(C0), xvec5, xvec5;
+ADD_SY	xvec5, xvec14, xvec14;
+#endif
+STL_SY	xvec14, 8*SIZE(C0);
+STH_SY	xvec14, 10*SIZE(C0);
+#ifndef	TRMMKERNEL
+LDL_SY	12*SIZE(C1), xvec4, xvec4;
+LDH_SY	14*SIZE(C1), xvec4, xvec4;
+ADD_SY	xvec4, xvec6, xvec6;
+#endif
+STL_SY	xvec6, 12*SIZE(C1);
+STH_SY	xvec6, 14*SIZE(C1);
+
+EXTRA_SY $1, yvec13, xvec5;
+#ifndef	TRMMKERNEL
+LDL_SY	0*SIZE(C0, ldc, 1), xvec4, xvec4;
+LDH_SY	2*SIZE(C0, ldc, 1), xvec4, xvec4;
+ADD_SY	xvec4, xvec13, xvec13;
+#endif
+STL_SY	xvec13, 0*SIZE(C0, ldc, 1);
+STH_SY	xvec13, 2*SIZE(C0, ldc, 1);
+#ifndef	TRMMKERNEL
+LDL_SY	4*SIZE(C1, ldc, 1), xvec3, xvec3;
+LDH_SY	6*SIZE(C1, ldc, 1), xvec3, xvec3;
+ADD_SY	xvec3, xvec5, xvec5;
+#endif
+STL_SY	xvec5, 4*SIZE(C1, ldc, 1);
+STH_SX	xvec5, 6*SIZE(C1, ldc, 1);
+
+EXTRA_SY $1, yvec12, xvec4;
+#ifndef	TRMMKERNEL
+LDL_SY	8*SIZE(C0, ldc, 1), xvec3, xvec3;
+LDH_SY	10*SIZE(C0, ldc, 1), xvec3, xvec3;
+ADD_SY	xvec3, xvec12, xvec12;
+#endif
+STL_SY	xvec12, 8*SIZE(C0, ldc, 1);
+STH_SY	xvec12, 10*SIZE(C0, ldc, 1);
+#ifndef	TRMMKERNEL
+LDL_SY	12*SIZE(C1, ldc, 1), xvec2, xvec2;
+LDH_SY	14*SIZE(C1, ldc, 1), xvec2, xvec2;
+ADD_SY	xvec2, xvec4, xvec4;
+#endif
+STL_SY	xvec4, 12*SIZE(C1, ldc, 1);
+STH_SY	xvec4, 14*SIZE(C1, ldc, 1);
+
+EXTRA_SY $1, yvec11, xvec3;
+#ifndef	TRMMKERNEL
+LDL_SY	0*SIZE(C1), xvec2, xvec2;
+LDH_SY	2*SIZE(C1), xvec2, xvec2;
+ADD_SY	xvec2, xvec11, xvec11;
+#endif
+STL_SY	xvec11, 0*SIZE(C1);
+STH_SY	xvec11, 2*SIZE(C1);
+#ifndef	TRMMKERNEL
+LDL_SY	4*SIZE(C0), xvec1, xvec1;
+LDH_SY	6*SIZE(C0), xvec1, xvec1;
+ADD_SY	xvec1, xvec3, xvec3;
+#endif
+STL_SY	xvec3, 4*SIZE(C0);
+STH_SY	xvec3, 6*SIZE(C0);
+
+EXTRA_SY $1, yvec10, xvec2;
+#ifndef	TRMMKERNEL
+LDL_SY	8*SIZE(C1), xvec1, xvec1;
+LDH_SY	10*SIZE(C1), xvec1, xvec1;
+ADD_SY	xvec1, xvec10, xvec10;
+#endif
+STL_SY	xvec10, 8*SIZE(C1);
+STH_SY	xvec10, 10*SIZE(C1);
+#ifndef	TRMMKERNEL
+LDL_SY	12*SIZE(C0), xvec0, xvec0;
+LDH_SY	14*SIZE(C0), xvec0, xvec0;
+ADD_SY	xvec0, xvec2, xvec2;
+#endif
+STL_SY	xvec2, 12*SIZE(C0);
+STH_SY	xvec2, 14*SIZE(C0);
+
+EXTRA_SY $1, yvec9, xvec1;
+#ifndef	TRMMKERNEL
+LDL_SY	0*SIZE(C1, ldc, 1), xvec7, xvec7;
+LDH_SY	2*SIZE(C1, ldc, 1), xvec7, xvec7;
+ADD_SY	xvec7, xvec9, xvec9;
+#endif
+STL_SY	xvec9, 0*SIZE(C1, ldc, 1);
+STH_SY	xvec9, 2*SIZE(C1, ldc, 1);
+#ifndef	TRMMKERNEL
+LDL_SY	4*SIZE(C0, ldc, 1), xvec6, xvec6;
+LDH_SY	6*SIZE(C0, ldc, 1), xvec6, xvec6;
+ADD_SY	xvec6, xvec1, xvec1;
+#endif
+STL_SY	xvec1, 4*SIZE(C0, ldc, 1);
+STH_SY	xvec1, 6*SIZE(C0, ldc, 1);
+
+EXTRA_SY $1, yvec8, xvec0;
+#ifndef	TRMMKERNEL
+LDL_SY	8*SIZE(C1, ldc, 1), xvec6, xvec6;
+LDH_SY	10*SIZE(C1, ldc, 1), xvec6, xvec6;
+ADD_SY	xvec6, xvec8, xvec8;
+#endif
+STL_SY	xvec8, 8*SIZE(C1, ldc, 1);
+STH_SY	xvec8, 10*SIZE(C1, ldc, 1);
+#ifndef	TRMMKERNEL
+LDL_SY	12*SIZE(C0, ldc, 1), xvec5, xvec5;
+LDH_SY	14*SIZE(C0, ldc, 1), xvec5, xvec5;
+ADD_SY	xvec5, xvec0, xvec0;
+#endif
+STL_SY	xvec0, 12*SIZE(C0, ldc, 1);
+STH_SY	xvec0, 14*SIZE(C0, ldc, 1);
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA))||(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kkk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+LEAQ (ptrba, %rax, 8), ptrba;
+LEAQ (ptrbb, %rax, 4), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT) 
+ADDQ	$8, kk;
+#endif
+ADDQ	$16*SIZE, C0;
+ADDQ	$16*SIZE, C1;
+DECQ	i;
+JG	.L1_bodyB;
+.align 32;
+.L1_loopE:;
+TEST $4, bm;
+JLE .L5_loopE;
+.align 32
+.L5_bodyB:
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb,ptrbb;
+#else
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+LEAQ (ptrba, %rax, 4), ptrba;
+LEAQ (ptrbb, %rax, 4), ptrbb;
+#endif
+XOR_SY	yvec15, yvec15, yvec15;
+XOR_SY	yvec13, yvec13, yvec13;
+XOR_SY	yvec11, yvec11, yvec11;
+XOR_SY	yvec9, yvec9, yvec9;
+#ifndef	TRMMKERNEL
+MOVQ bk,k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else
+MOVQ kk, %rax;
+#ifdef	LEFT
+ADDQ $4, %rax;
+#else 
+ADDQ $4, %rax;
+#endif
+MOVQ %rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L8_loopE;
+.align 32
+.L8_bodyB:
+#### Unroll times 1 ####
+LD_SY	0*SIZE(ptrba), yvec0;
+VPERMILP_SY	$0xb1, yvec0, yvec1;
+EDUP_SY	0*SIZE(ptrbb), yvec2;
+VPERMILP_SY $0x4e, yvec2, yvec3;
+MUL_SY	yvec0, yvec2, yvec6;
+ADD1_SY	yvec6, yvec15, yvec15;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;
+MUL_SY	yvec0, yvec3, yvec7;
+ADD1_SY	yvec7, yvec13, yvec13;
+
+ODUP_SY 0*SIZE(ptrbb), yvec2;
+SHUF_SY	$0x03, yvec3, yvec3, yvec5;
+MUL_SY	yvec0, yvec4, yvec6;
+ADD1_SY	yvec6, yvec11, yvec11;
+VPERMILP_SY $0x4e, yvec2, yvec3;
+MUL_SY	yvec0, yvec5, yvec7;
+ADD1_SY	yvec7, yvec9, yvec9;
+
+MUL_SY	yvec1, yvec2, yvec6;
+ADD2_SY	yvec6, yvec15, yvec15;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;
+MUL_SY	yvec1, yvec3, yvec7;
+ADD2_SY	yvec7, yvec13, yvec13;
+
+SHUF_SY $0x03, yvec3, yvec3, yvec5;
+MUL_SY	yvec1, yvec4, yvec6;
+ADD2_SY	yvec6, yvec11, yvec11;
+MUL_SY	yvec1, yvec5, yvec7;
+ADD2_SY	yvec7, yvec9, yvec9;
+
+#### Unroll time 2 ####
+LD_SY	8*SIZE(ptrba), yvec0;
+VPERMILP_SY	$0xb1, yvec0, yvec1;
+EDUP_SY	8*SIZE(ptrbb), yvec2;
+VPERMILP_SY $0x4e, yvec2, yvec3;
+MUL_SY	yvec0, yvec2, yvec6;
+ADD1_SY	yvec6, yvec15, yvec15;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;
+MUL_SY	yvec0, yvec3, yvec7;
+ADD1_SY	yvec7, yvec13, yvec13;
+
+ODUP_SY 8*SIZE(ptrbb), yvec2;
+SHUF_SY	$0x03, yvec3, yvec3, yvec5;
+MUL_SY	yvec0, yvec4, yvec6;
+ADD1_SY	yvec6, yvec11, yvec11;
+VPERMILP_SY $0x4e, yvec2, yvec3;
+MUL_SY	yvec0, yvec5, yvec7;
+ADD1_SY	yvec7, yvec9, yvec9;
+
+MUL_SY	yvec1, yvec2, yvec6;
+ADD2_SY	yvec6, yvec15, yvec15;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;
+MUL_SY	yvec1, yvec3, yvec7;
+ADD2_SY	yvec7, yvec13, yvec13;
+
+SHUF_SY $0x03, yvec3, yvec3, yvec5;
+MUL_SY	yvec1, yvec4, yvec6;
+ADD2_SY	yvec6, yvec11, yvec11;
+MUL_SY	yvec1, yvec5, yvec7;
+ADD2_SY	yvec7, yvec9, yvec9;
+
+#### Unroll time 3 ####
+LD_SY	16*SIZE(ptrba), yvec0;
+VPERMILP_SY	$0xb1, yvec0, yvec1;
+EDUP_SY	16*SIZE(ptrbb), yvec2;
+VPERMILP_SY $0x4e, yvec2, yvec3;
+MUL_SY	yvec0, yvec2, yvec6;
+ADD1_SY	yvec6, yvec15, yvec15;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;
+MUL_SY	yvec0, yvec3, yvec7;
+ADD1_SY	yvec7, yvec13, yvec13;
+
+ODUP_SY 16*SIZE(ptrbb), yvec2;
+SHUF_SY	$0x03, yvec3, yvec3, yvec5;
+MUL_SY	yvec0, yvec4, yvec6;
+ADD1_SY	yvec6, yvec11, yvec11;
+VPERMILP_SY $0x4e, yvec2, yvec3;
+MUL_SY	yvec0, yvec5, yvec7;
+ADD1_SY	yvec7, yvec9, yvec9;
+
+MUL_SY	yvec1, yvec2, yvec6;
+ADD2_SY	yvec6, yvec15, yvec15;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;
+MUL_SY	yvec1, yvec3, yvec7;
+ADD2_SY	yvec7, yvec13, yvec13;
+
+SHUF_SY $0x03, yvec3, yvec3, yvec5;
+MUL_SY	yvec1, yvec4, yvec6;
+ADD2_SY	yvec6, yvec11, yvec11;
+MUL_SY	yvec1, yvec5, yvec7;
+ADD2_SY	yvec7, yvec9, yvec9;
+
+#### Unroll time 3 ####
+LD_SY	24*SIZE(ptrba), yvec0;
+VPERMILP_SY	$0xb1, yvec0, yvec1;
+EDUP_SY	24*SIZE(ptrbb), yvec2;
+VPERMILP_SY $0x4e, yvec2, yvec3;
+MUL_SY	yvec0, yvec2, yvec6;
+ADD1_SY	yvec6, yvec15, yvec15;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;
+MUL_SY	yvec0, yvec3, yvec7;
+ADD1_SY	yvec7, yvec13, yvec13;
+
+ODUP_SY 24*SIZE(ptrbb), yvec2;
+SHUF_SY	$0x03, yvec3, yvec3, yvec5;
+MUL_SY	yvec0, yvec4, yvec6;
+ADD1_SY	yvec6, yvec11, yvec11;
+VPERMILP_SY $0x4e, yvec2, yvec3;
+MUL_SY	yvec0, yvec5, yvec7;
+ADD1_SY	yvec7, yvec9, yvec9;
+
+MUL_SY	yvec1, yvec2, yvec6;
+ADD2_SY	yvec6, yvec15, yvec15;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;
+MUL_SY	yvec1, yvec3, yvec7;
+ADD2_SY	yvec7, yvec13, yvec13;
+
+SHUF_SY $0x03, yvec3, yvec3, yvec5;
+MUL_SY	yvec1, yvec4, yvec6;
+ADD2_SY	yvec6, yvec11, yvec11;
+MUL_SY	yvec1, yvec5, yvec7;
+ADD2_SY	yvec7, yvec9, yvec9;
+ADDQ	$32*SIZE, ptrba;
+ADDQ	$32*SIZE, ptrbb;
+DECQ	k;
+JG .L8_bodyB;
+.align 32
+.L8_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else 
+TEST $2, kkk;
+#endif
+JLE .L9_loopE;
+.align 32
+.L9_bodyB:
+#### Unroll times 1 ####
+LD_SY	0*SIZE(ptrba), yvec0;
+VPERMILP_SY	$0xb1, yvec0, yvec1;
+EDUP_SY	0*SIZE(ptrbb), yvec2;
+VPERMILP_SY $0x4e, yvec2, yvec3;
+MUL_SY	yvec0, yvec2, yvec6;
+ADD1_SY	yvec6, yvec15, yvec15;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;
+MUL_SY	yvec0, yvec3, yvec7;
+ADD1_SY	yvec7, yvec13, yvec13;
+
+ODUP_SY 0*SIZE(ptrbb), yvec2;
+SHUF_SY	$0x03, yvec3, yvec3, yvec5;
+MUL_SY	yvec0, yvec4, yvec6;
+ADD1_SY	yvec6, yvec11, yvec11;
+VPERMILP_SY $0x4e, yvec2, yvec3;
+MUL_SY	yvec0, yvec5, yvec7;
+ADD1_SY	yvec7, yvec9, yvec9;
+
+MUL_SY	yvec1, yvec2, yvec6;
+ADD2_SY	yvec6, yvec15, yvec15;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;
+MUL_SY	yvec1, yvec3, yvec7;
+ADD2_SY	yvec7, yvec13, yvec13;
+
+SHUF_SY $0x03, yvec3, yvec3, yvec5;
+MUL_SY	yvec1, yvec4, yvec6;
+ADD2_SY	yvec6, yvec11, yvec11;
+MUL_SY	yvec1, yvec5, yvec7;
+ADD2_SY	yvec7, yvec9, yvec9;
+
+#### Unroll time 2 ####
+LD_SY	8*SIZE(ptrba), yvec0;
+VPERMILP_SY	$0xb1, yvec0, yvec1;
+EDUP_SY	8*SIZE(ptrbb), yvec2;
+VPERMILP_SY $0x4e, yvec2, yvec3;
+MUL_SY	yvec0, yvec2, yvec6;
+ADD1_SY	yvec6, yvec15, yvec15;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;
+MUL_SY	yvec0, yvec3, yvec7;
+ADD1_SY	yvec7, yvec13, yvec13;
+
+ODUP_SY 8*SIZE(ptrbb), yvec2;
+SHUF_SY	$0x03, yvec3, yvec3, yvec5;
+MUL_SY	yvec0, yvec4, yvec6;
+ADD1_SY	yvec6, yvec11, yvec11;
+VPERMILP_SY $0x4e, yvec2, yvec3;
+MUL_SY	yvec0, yvec5, yvec7;
+ADD1_SY	yvec7, yvec9, yvec9;
+
+MUL_SY	yvec1, yvec2, yvec6;
+ADD2_SY	yvec6, yvec15, yvec15;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;
+MUL_SY	yvec1, yvec3, yvec7;
+ADD2_SY	yvec7, yvec13, yvec13;
+
+SHUF_SY $0x03, yvec3, yvec3, yvec5;
+MUL_SY	yvec1, yvec4, yvec6;
+ADD2_SY	yvec6, yvec11, yvec11;
+MUL_SY	yvec1, yvec5, yvec7;
+ADD2_SY	yvec7, yvec9, yvec9;
+ADDQ	$16*SIZE, ptrba;
+ADDQ	$16*SIZE, ptrbb;
+
+.L9_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else
+TEST $1, kkk;
+#endif
+JLE .L10_loopE;
+.align 32
+.L10_bodyB:
+#### Unroll times 1 ####
+LD_SY	0*SIZE(ptrba), yvec0;
+VPERMILP_SY	$0xb1, yvec0, yvec1;
+EDUP_SY	0*SIZE(ptrbb), yvec2;
+VPERMILP_SY $0x4e, yvec2, yvec3;
+MUL_SY	yvec0, yvec2, yvec6;
+ADD1_SY	yvec6, yvec15, yvec15;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;
+MUL_SY	yvec0, yvec3, yvec7;
+ADD1_SY	yvec7, yvec13, yvec13;
+
+ODUP_SY 0*SIZE(ptrbb), yvec2;
+SHUF_SY	$0x03, yvec3, yvec3, yvec5;
+MUL_SY	yvec0, yvec4, yvec6;
+ADD1_SY	yvec6, yvec11, yvec11;
+VPERMILP_SY $0x4e, yvec2, yvec3;
+MUL_SY	yvec0, yvec5, yvec7;
+ADD1_SY	yvec7, yvec9, yvec9;
+
+MUL_SY	yvec1, yvec2, yvec6;
+ADD2_SY	yvec6, yvec15, yvec15;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;
+MUL_SY	yvec1, yvec3, yvec7;
+ADD2_SY	yvec7, yvec13, yvec13;
+
+SHUF_SY $0x03, yvec3, yvec3, yvec5;
+MUL_SY	yvec1, yvec4, yvec6;
+ADD2_SY	yvec6, yvec11, yvec11;
+MUL_SY	yvec1, yvec5, yvec7;
+ADD2_SY	yvec7, yvec9, yvec9;
+ADDQ	$8*SIZE, ptrba;
+ADDQ	$8*SIZE, ptrbb;
+
+.L10_loopE:
+#### Handle ####
+XOR_SY	yvec7, yvec7, yvec7;
+#if  defined(RN) || defined(RT) || defined(CN) || defined(CT)
+ADDSUB_SY	yvec15, yvec7, yvec15;
+ADDSUB_SY	yvec13, yvec7, yvec13;
+ADDSUB_SY	yvec11, yvec7, yvec11;
+ADDSUB_SY	yvec9, yvec7, yvec9;
+#elif defined(NR) || defined(NC) || defined(TR) || defined(TC)
+SUB_SY	yvec15, yvec7, yvec15;
+SUB_SY	yvec13, yvec7, yvec13;
+SUB_SY	yvec11, yvec7, yvec11;
+SUB_SY	yvec9, yvec7, yvec9;
+#elif defined(RR) || defined(RC) || defined(CR) || defined(CC)
+VPERMILP_SY $0xb1, yvec15, yvec15;
+VPERMILP_SY $0xb1, yvec13, yvec13;
+VPERMILP_SY $0xb1, yvec11, yvec11;
+VPERMILP_SY $0xb1, yvec9, yvec9;
+ADDSUB_SY	yvec15, yvec7, yvec15;
+ADDSUB_SY	yvec13, yvec7, yvec13;
+ADDSUB_SY	yvec11, yvec7, yvec11;
+ADDSUB_SY	yvec9, yvec7, yvec9;
+VPERMILP_SY $0xb1, yvec15, yvec15;
+VPERMILP_SY $0xb1, yvec13, yvec13;
+VPERMILP_SY $0xb1, yvec11, yvec11;
+VPERMILP_SY $0xb1, yvec9, yvec9;
+#endif
+##### Load Alpha ####
+BROAD_SY MEMALPHA_R,yvec7;
+BROAD_SY MEMALPHA_I,yvec6;
+##### Multiply Alpha ####
+VPERMILP_SY $0xb1,yvec15, yvec5;
+MUL_SY	yvec15, yvec7, yvec15;
+MUL_SY	yvec5, yvec6, yvec5;
+ADDSUB_SY	yvec5, yvec15, yvec15;
+VPERMILP_SY $0xb1,yvec13, yvec3;
+MUL_SY	yvec13, yvec7, yvec13;
+MUL_SY	yvec3, yvec6, yvec3;
+ADDSUB_SY	yvec3, yvec13, yvec13;
+VPERMILP_SY $0xb1,yvec11, yvec1;
+MUL_SY	yvec11, yvec7, yvec11;
+MUL_SY	yvec1, yvec6, yvec1;
+ADDSUB_SY	yvec1, yvec11, yvec11;
+VPERMILP_SY $0xb1,yvec9, yvec5;
+MUL_SY	yvec9, yvec7, yvec9;
+MUL_SY	yvec5, yvec6, yvec5;
+ADDSUB_SY	yvec5, yvec9, yvec9;
+#### Writing back ####
+#### Shuffle Results ####
+MOV_SY yvec15,yvec7;
+REVS_SY $0xe4,yvec13,yvec15,yvec15;
+REVS_SY $0xe4,yvec7,yvec13,yvec13;
+MOV_SY yvec11,yvec7;
+REVS_SY $0xe4,yvec9,yvec11,yvec11;
+REVS_SY $0xe4,yvec7,yvec9,yvec9;
+#### Writing back ####
+EXTRA_SY $1, yvec15, xvec7;
+#ifndef	TRMMKERNEL
+LDL_SX	0*SIZE(C0), xvec6;
+LDH_SX	2*SIZE(C0), xvec6;
+ADD_SX	xvec6, xvec15;
+#endif
+STL_SX	xvec15, 0*SIZE(C0);
+STH_SX	xvec15, 2*SIZE(C0);
+#ifndef	TRMMKERNEL
+LDL_SX	4*SIZE(C1), xvec4;
+LDH_SX	6*SIZE(C1), xvec4;
+ADD_SX	xvec4, xvec7;
+#endif
+STL_SX	xvec7, 4*SIZE(C1);
+STH_SX	xvec7, 6*SIZE(C1);
+
+EXTRA_SY $1, yvec13, xvec5;
+#ifndef	TRMMKERNEL
+LDL_SX	0*SIZE(C0, ldc, 1), xvec4;
+LDH_SX	2*SIZE(C0, ldc, 1), xvec4;
+ADD_SX	xvec4, xvec13;
+#endif
+STL_SX	xvec13, 0*SIZE(C0, ldc, 1);
+STH_SX	xvec13, 2*SIZE(C0, ldc, 1);
+#ifndef	TRMMKERNEL
+LDL_SX	4*SIZE(C1, ldc, 1), xvec2;
+LDH_SX	6*SIZE(C1, ldc, 1), xvec2;
+ADD_SX	xvec2, xvec5;
+#endif
+STL_SX	xvec5, 4*SIZE(C1, ldc, 1);
+STH_SX	xvec5, 6*SIZE(C1, ldc, 1);
+
+EXTRA_SY $1, yvec11, xvec3;
+#ifndef	TRMMKERNEL
+LDL_SX	0*SIZE(C1), xvec2;
+LDH_SX	2*SIZE(C1), xvec2;
+ADD_SX	xvec2, xvec11;
+#endif
+STL_SX	xvec11, 0*SIZE(C1);
+STH_SX	xvec11, 2*SIZE(C1);
+#ifndef	TRMMKERNEL
+LDL_SX	4*SIZE(C0), xvec0;
+LDH_SX	6*SIZE(C0), xvec0;
+ADD_SX	xvec0, xvec3;
+#endif
+STL_SX	xvec3, 4*SIZE(C0);
+STH_SX	xvec3, 6*SIZE(C0);
+
+EXTRA_SY $1, yvec9, xvec1;
+#ifndef	TRMMKERNEL
+LDL_SX	0*SIZE(C1, ldc, 1), xvec0;
+LDH_SX	2*SIZE(C1, ldc, 1), xvec0;
+ADD_SX	xvec0, xvec9;
+#endif
+STL_SX	xvec9, 0*SIZE(C1, ldc, 1);
+STH_SX	xvec9, 2*SIZE(C1, ldc, 1);
+#ifndef	TRMMKERNEL
+LDL_SX	4*SIZE(C0, ldc, 1), xvec6;
+LDH_SX	6*SIZE(C0, ldc, 1), xvec6;
+ADD_SX	xvec6, xvec1;
+#endif
+STL_SX	xvec1, 4*SIZE(C0, ldc, 1);
+STH_SX	xvec1, 6*SIZE(C0, ldc, 1);
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA))||(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kkk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+LEAQ (ptrba, %rax, 4), ptrba;
+LEAQ (ptrbb, %rax, 4), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT) 
+ADDQ	$4, kk;
+#endif
+
+ADDQ	$8*SIZE, C0;
+ADDQ	$8*SIZE, C1;
+
+.L5_loopE:
+TEST $2, bm;
+JLE .L6_loopE;
+.align 32
+.L6_bodyB:
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb,ptrbb;
+#else
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+LEAQ (ptrba, %rax, 2), ptrba;
+LEAQ (ptrbb, %rax, 4), ptrbb;
+#endif
+#### Initial Results Register ####
+XOR_SY	yvec15, yvec15, yvec15;
+XOR_SY	yvec14, yvec14, yvec14;
+XOR_SY	yvec13, yvec13, yvec13;
+XOR_SY	yvec12, yvec12, yvec12;
+#ifndef	TRMMKERNEL
+MOVQ bk,k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else
+MOVQ kk, %rax;
+#ifdef	LEFT
+ADDQ $2, %rax;
+#else 
+ADDQ $4, %rax;
+#endif
+MOVQ %rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L11_loopE;
+.align 32
+.L11_bodyB:
+LD_SX	0*SIZE(ptrba), xvec0;	# ar1, ai1, ar2, ai2
+EDUP_SX	0*SIZE(ptrbb), xvec2;	# br1, br1, br2, br2
+SHUF_SX	$0x4e, xvec2, xvec3;	# br3, br3, br4, br4
+MUL_SX	xvec0, xvec2;
+ADD1_SX	xvec2, xvec15;
+MUL_SX	xvec0, xvec3;
+ADD1_SX	xvec3, xvec14;
+
+EDUP_SX	4*SIZE(ptrbb), xvec4;
+SHUF_SX	$0x4e, xvec4, xvec5;
+MUL_SX	xvec0, xvec4;
+ADD1_SX	xvec4, xvec13;
+MUL_SX	xvec0, xvec5;
+ADD1_SX	xvec5, xvec12;
+
+SHUF_SX $0xb1, xvec0, xvec1;
+ODUP_SX 0*SIZE(ptrbb), xvec2;
+SHUF_SX	$0x4e, xvec2, xvec3;
+MUL_SX	xvec1, xvec2;
+ADD2_SX	xvec2, xvec15;
+MUL_SX	xvec1, xvec3;
+ADD2_SX	xvec3, xvec14;
+
+ODUP_SX	4*SIZE(ptrbb), xvec4;
+SHUF_SX	$0x4e, xvec4, xvec5;
+MUL_SX	xvec1, xvec4;
+ADD2_SX	xvec4, xvec13;
+MUL_SX	xvec1, xvec5;
+ADD2_SX	xvec5, xvec12;
+
+LD_SX	4*SIZE(ptrba), xvec0;	# ar1, ai1, ar2, ai2
+EDUP_SX	8*SIZE(ptrbb), xvec2;	# br1, br1, br2, br2
+SHUF_SX	$0x4e, xvec2, xvec3;	# br3, br3, br4, br4
+MUL_SX	xvec0, xvec2;
+ADD1_SX	xvec2, xvec15;
+MUL_SX	xvec0, xvec3;
+ADD1_SX	xvec3, xvec14;
+
+EDUP_SX	12*SIZE(ptrbb), xvec4;
+SHUF_SX	$0x4e, xvec4, xvec5;
+MUL_SX	xvec0, xvec4;
+ADD1_SX	xvec4, xvec13;
+MUL_SX	xvec0, xvec5;
+ADD1_SX	xvec5, xvec12;
+
+SHUF_SX $0xb1, xvec0, xvec1;
+ODUP_SX 8*SIZE(ptrbb), xvec2;
+SHUF_SX	$0x4e, xvec2, xvec3;
+MUL_SX	xvec1, xvec2;
+ADD2_SX	xvec2, xvec15;
+MUL_SX	xvec1, xvec3;
+ADD2_SX	xvec3, xvec14;
+
+ODUP_SX	12*SIZE(ptrbb), xvec4;
+SHUF_SX	$0x4e, xvec4, xvec5;
+MUL_SX	xvec1, xvec4;
+ADD2_SX	xvec4, xvec13;
+MUL_SX	xvec1, xvec5;
+ADD2_SX	xvec5, xvec12;
+
+LD_SX	8*SIZE(ptrba), xvec0;	# ar1, ai1, ar2, ai2
+EDUP_SX	16*SIZE(ptrbb), xvec2;	# br1, br1, br2, br2
+SHUF_SX	$0x4e, xvec2, xvec3;	# br3, br3, br4, br4
+MUL_SX	xvec0, xvec2;
+ADD1_SX	xvec2, xvec15;
+MUL_SX	xvec0, xvec3;
+ADD1_SX	xvec3, xvec14;
+
+EDUP_SX	20*SIZE(ptrbb), xvec4;
+SHUF_SX	$0x4e, xvec4, xvec5;
+MUL_SX	xvec0, xvec4;
+ADD1_SX	xvec4, xvec13;
+MUL_SX	xvec0, xvec5;
+ADD1_SX	xvec5, xvec12;
+
+SHUF_SX $0xb1, xvec0, xvec1;
+ODUP_SX 16*SIZE(ptrbb), xvec2;
+SHUF_SX	$0x4e, xvec2, xvec3;
+MUL_SX	xvec1, xvec2;
+ADD2_SX	xvec2, xvec15;
+MUL_SX	xvec1, xvec3;
+ADD2_SX	xvec3, xvec14;
+
+ODUP_SX	20*SIZE(ptrbb), xvec4;
+SHUF_SX	$0x4e, xvec4, xvec5;
+MUL_SX	xvec1, xvec4;
+ADD2_SX	xvec4, xvec13;
+MUL_SX	xvec1, xvec5;
+ADD2_SX	xvec5, xvec12;
+
+LD_SX	12*SIZE(ptrba), xvec0;	# ar1, ai1, ar2, ai2
+EDUP_SX	24*SIZE(ptrbb), xvec2;	# br1, br1, br2, br2
+SHUF_SX	$0x4e, xvec2, xvec3;	# br3, br3, br4, br4
+MUL_SX	xvec0, xvec2;
+ADD1_SX	xvec2, xvec15;
+MUL_SX	xvec0, xvec3;
+ADD1_SX	xvec3, xvec14;
+
+EDUP_SX	28*SIZE(ptrbb), xvec4;
+SHUF_SX	$0x4e, xvec4, xvec5;
+MUL_SX	xvec0, xvec4;
+ADD1_SX	xvec4, xvec13;
+MUL_SX	xvec0, xvec5;
+ADD1_SX	xvec5, xvec12;
+
+SHUF_SX $0xb1, xvec0, xvec1;
+ODUP_SX 24*SIZE(ptrbb), xvec2;
+SHUF_SX	$0x4e, xvec2, xvec3;
+MUL_SX	xvec1, xvec2;
+ADD2_SX	xvec2, xvec15;
+MUL_SX	xvec1, xvec3;
+ADD2_SX	xvec3, xvec14;
+
+ODUP_SX	28*SIZE(ptrbb), xvec4;
+SHUF_SX	$0x4e, xvec4, xvec5;
+MUL_SX	xvec1, xvec4;
+ADD2_SX	xvec4, xvec13;
+MUL_SX	xvec1, xvec5;
+ADD2_SX	xvec5, xvec12;
+ADDQ	$16*SIZE, ptrba;
+ADDQ	$32*SIZE, ptrbb;
+DECQ	k;
+JG .L11_bodyB;
+.align 32
+.L11_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else
+TEST $2, kkk;
+#endif
+JLE .L12_loopE;
+.align 32
+.L12_bodyB:
+LD_SX	0*SIZE(ptrba), xvec0;	# ar1, ai1, ar2, ai2
+EDUP_SX	0*SIZE(ptrbb), xvec2;	# br1, br1, br2, br2
+SHUF_SX	$0x4e, xvec2, xvec3;	# br3, br3, br4, br4
+MUL_SX	xvec0, xvec2;
+ADD1_SX	xvec2, xvec15;
+MUL_SX	xvec0, xvec3;
+ADD1_SX	xvec3, xvec14;
+
+EDUP_SX	4*SIZE(ptrbb), xvec4;
+SHUF_SX	$0x4e, xvec4, xvec5;
+MUL_SX	xvec0, xvec4;
+ADD1_SX	xvec4, xvec13;
+MUL_SX	xvec0, xvec5;
+ADD1_SX	xvec5, xvec12;
+
+SHUF_SX $0xb1, xvec0, xvec1;
+ODUP_SX 0*SIZE(ptrbb), xvec2;
+SHUF_SX	$0x4e, xvec2, xvec3;
+MUL_SX	xvec1, xvec2;
+ADD2_SX	xvec2, xvec15;
+MUL_SX	xvec1, xvec3;
+ADD2_SX	xvec3, xvec14;
+
+ODUP_SX	4*SIZE(ptrbb), xvec4;
+SHUF_SX	$0x4e, xvec4, xvec5;
+MUL_SX	xvec1, xvec4;
+ADD2_SX	xvec4, xvec13;
+MUL_SX	xvec1, xvec5;
+ADD2_SX	xvec5, xvec12;
+
+LD_SX	4*SIZE(ptrba), xvec0;	# ar1, ai1, ar2, ai2
+EDUP_SX	8*SIZE(ptrbb), xvec2;	# br1, br1, br2, br2
+SHUF_SX	$0x4e, xvec2, xvec3;	# br3, br3, br4, br4
+MUL_SX	xvec0, xvec2;
+ADD1_SX	xvec2, xvec15;
+MUL_SX	xvec0, xvec3;
+ADD1_SX	xvec3, xvec14;
+
+EDUP_SX	12*SIZE(ptrbb), xvec4;
+SHUF_SX	$0x4e, xvec4, xvec5;
+MUL_SX	xvec0, xvec4;
+ADD1_SX	xvec4, xvec13;
+MUL_SX	xvec0, xvec5;
+ADD1_SX	xvec5, xvec12;
+
+SHUF_SX $0xb1, xvec0, xvec1;
+ODUP_SX 8*SIZE(ptrbb), xvec2;
+SHUF_SX	$0x4e, xvec2, xvec3;
+MUL_SX	xvec1, xvec2;
+ADD2_SX	xvec2, xvec15;
+MUL_SX	xvec1, xvec3;
+ADD2_SX	xvec3, xvec14;
+
+ODUP_SX	12*SIZE(ptrbb), xvec4;
+SHUF_SX	$0x4e, xvec4, xvec5;
+MUL_SX	xvec1, xvec4;
+ADD2_SX	xvec4, xvec13;
+MUL_SX	xvec1, xvec5;
+ADD2_SX	xvec5, xvec12;
+ADDQ	$8*SIZE, ptrba;
+ADDQ	$16*SIZE, ptrbb;
+
+.L12_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else
+TEST $1, kkk;
+#endif
+JLE .L13_loopE;
+.align 32
+.L13_bodyB:
+LD_SX	0*SIZE(ptrba), xvec0;	# ar1, ai1, ar2, ai2
+EDUP_SX	0*SIZE(ptrbb), xvec2;	# br1, br1, br2, br2
+SHUF_SX	$0x4e, xvec2, xvec3;	# br3, br3, br4, br4
+MUL_SX	xvec0, xvec2;
+ADD1_SX	xvec2, xvec15;
+MUL_SX	xvec0, xvec3;
+ADD1_SX	xvec3, xvec14;
+
+EDUP_SX	4*SIZE(ptrbb), xvec4;
+SHUF_SX	$0x4e, xvec4, xvec5;
+MUL_SX	xvec0, xvec4;
+ADD1_SX	xvec4, xvec13;
+MUL_SX	xvec0, xvec5;
+ADD1_SX	xvec5, xvec12;
+
+SHUF_SX $0xb1, xvec0, xvec1;
+ODUP_SX 0*SIZE(ptrbb), xvec2;
+SHUF_SX	$0x4e, xvec2, xvec3;
+MUL_SX	xvec1, xvec2;
+ADD2_SX	xvec2, xvec15;
+MUL_SX	xvec1, xvec3;
+ADD2_SX	xvec3, xvec14;
+
+ODUP_SX	4*SIZE(ptrbb), xvec4;
+SHUF_SX	$0x4e, xvec4, xvec5;
+MUL_SX	xvec1, xvec4;
+ADD2_SX	xvec4, xvec13;
+MUL_SX	xvec1, xvec5;
+ADD2_SX	xvec5, xvec12;
+ADDQ	$4*SIZE, ptrba;
+ADDQ	$8*SIZE, ptrbb;
+
+.L13_loopE:
+#### Handle ####
+#if  defined(RN) || defined(RT) || defined(CN) || defined(CT)
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec15, xvec7;
+MOV_SX		xvec7, xvec15;
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec14, xvec7;
+MOV_SX		xvec7, xvec14;
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec13, xvec7;
+MOV_SX		xvec7, xvec13;
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec12, xvec7;
+MOV_SX		xvec7, xvec12;
+#elif defined(NR) || defined(NC) || defined(TR) || defined(TC)
+XOR_SY	yvec7, yvec7, yvec7;
+SUB_SX	xvec15, xvec7;
+MOV_SX	xvec7, xvec15;
+XOR_SY	yvec7, yvec7, yvec7;
+SUB_SX	xvec14, xvec7;
+MOV_SX	xvec7, xvec14;
+XOR_SY	yvec7, yvec7, yvec7;
+SUB_SX	xvec13, xvec7;
+MOV_SX	xvec7, xvec13;
+XOR_SY	yvec7, yvec7, yvec7;
+SUB_SX	xvec12, xvec7;
+MOV_SX	xvec7, xvec12;
+#elif defined(RR) || defined(RC) || defined(CR) || defined(CC)
+SHUF_SX $0xb1, xvec15, xvec15;
+SHUF_SX $0xb1, xvec14, xvec14;
+SHUF_SX $0xb1, xvec13, xvec13;
+SHUF_SX $0xb1, xvec12, xvec12;
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec15, xvec7;
+MOV_SX		xvec7, xvec15;
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec14, xvec7;
+MOV_SX		xvec7, xvec14;
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec13, xvec7;
+MOV_SX		xvec7, xvec13;
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec12, xvec7;
+MOV_SX		xvec7, xvec12;
+SHUF_SX $0xb1, xvec15, xvec15;
+SHUF_SX $0xb1, xvec14, xvec14;
+SHUF_SX $0xb1, xvec13, xvec13;
+SHUF_SX $0xb1, xvec12, xvec12;
+#endif
+##### Load Alpha ####
+BROAD_SX MEMALPHA_R,xvec7;
+BROAD_SX MEMALPHA_I,xvec6;
+##### Multiply Alpha ####
+VPERMILP_SX $0xb1,xvec15, xvec5;
+MUL_SX	xvec7, xvec15;
+MUL_SX	xvec6, xvec5;
+ADDSUB_SX	xvec5, xvec15;
+VPERMILP_SX $0xb1,xvec14, xvec4;
+MUL_SX	xvec7, xvec14;
+MUL_SX	xvec6, xvec4;
+ADDSUB_SX	xvec4, xvec14;
+VPERMILP_SX $0xb1,xvec13, xvec3;
+MUL_SX	xvec7, xvec13;
+MUL_SX	xvec6, xvec3;
+ADDSUB_SX	xvec3, xvec13;
+VPERMILP_SX $0xb1,xvec12, xvec2;
+MUL_SX	xvec7, xvec12;
+MUL_SX	xvec6, xvec2;
+ADDSUB_SX	xvec2, xvec12;
+#### Writing back ####
+#ifndef	TRMMKERNEL
+LDL_SX	0*SIZE(C0), xvec0;
+LDH_SX	2*SIZE(C0, ldc, 1), xvec0;
+LDL_SX	0*SIZE(C0, ldc, 1), xvec1;
+LDH_SX	2*SIZE(C0), xvec1;
+LDL_SX	0*SIZE(C1), xvec2;
+LDH_SX	2*SIZE(C1, ldc, 1), xvec2;
+LDL_SX	0*SIZE(C1, ldc, 1), xvec3;
+LDH_SX	2*SIZE(C1), xvec3;
+ADD_SX	xvec0, xvec15;
+ADD_SX	xvec1, xvec14;
+ADD_SX	xvec2, xvec13;
+ADD_SX	xvec3, xvec12;
+#endif
+STL_SX	xvec15, 0*SIZE(C0);
+STH_SX	xvec15, 2*SIZE(C0, ldc, 1);
+STL_SX	xvec14, 0*SIZE(C0, ldc, 1);
+STH_SX	xvec14, 2*SIZE(C0);
+STL_SX	xvec13, 0*SIZE(C1);
+STH_SX	xvec13, 2*SIZE(C1, ldc, 1);
+STL_SX	xvec12, 0*SIZE(C1, ldc, 1);
+STH_SX	xvec12, 2*SIZE(C1);
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA))||(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kkk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+LEAQ (ptrba, %rax, 2), ptrba;
+LEAQ (ptrbb, %rax, 4), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT) 
+ADDQ	$2, kk;
+#endif
+
+ADDQ	$4*SIZE, C0;
+ADDQ	$4*SIZE, C1;
+
+.L6_loopE:
+TEST $1, bm;
+JLE .L7_loopE;
+.align 32
+.L7_bodyB:
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb,ptrbb;
+#else
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+ADDQ %rax, ptrba;
+LEAQ (ptrbb, %rax, 4), ptrbb;
+#endif
+XOR_SY	yvec15, yvec15, yvec15;
+XOR_SY	yvec14, yvec14, yvec14;
+#ifndef	TRMMKERNEL
+MOVQ bk,k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else
+MOVQ kk, %rax;
+#ifdef	LEFT
+ADDQ $1, %rax;
+#else 
+ADDQ $4, %rax;
+#endif
+MOVQ %rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L14_loopE;
+.align 32
+.L14_bodyB:
+BROAD_SX 0*SIZE(ptrba), xvec0;
+LD_SX	0*SIZE(ptrbb), xvec2;
+SHUF_SX	$0xb1, xvec2, xvec3;
+MUL_SX	xvec0, xvec2;
+ADD1_SX	xvec2, xvec15;
+
+LD_SX	4*SIZE(ptrbb), xvec4;
+SHUF_SX	$0xb1, xvec4, xvec5;
+MUL_SX	xvec0, xvec4;
+ADD1_SX	xvec4, xvec14;
+
+BROAD_SX 1*SIZE(ptrba), xvec1;
+MUL_SX	xvec1, xvec3;
+ADD2_SX	xvec3, xvec15;
+MUL_SX	xvec1, xvec5;
+ADD2_SX	xvec5, xvec14;
+
+BROAD_SX 2*SIZE(ptrba), xvec0;
+LD_SX	8*SIZE(ptrbb), xvec2;
+SHUF_SX	$0xb1, xvec2, xvec3;
+MUL_SX	xvec0, xvec2;
+ADD1_SX	xvec2, xvec15;
+
+LD_SX	12*SIZE(ptrbb), xvec4;
+SHUF_SX	$0xb1, xvec4, xvec5;
+MUL_SX	xvec0, xvec4;
+ADD1_SX	xvec4, xvec14;
+
+BROAD_SX 3*SIZE(ptrba), xvec1;
+MUL_SX	xvec1, xvec3;
+ADD2_SX	xvec3, xvec15;
+MUL_SX	xvec1, xvec5;
+ADD2_SX	xvec5, xvec14;
+
+BROAD_SX 4*SIZE(ptrba), xvec0;
+LD_SX	16*SIZE(ptrbb), xvec2;
+SHUF_SX	$0xb1, xvec2, xvec3;
+MUL_SX	xvec0, xvec2;
+ADD1_SX	xvec2, xvec15;
+
+LD_SX	20*SIZE(ptrbb), xvec4;
+SHUF_SX	$0xb1, xvec4, xvec5;
+MUL_SX	xvec0, xvec4;
+ADD1_SX	xvec4, xvec14;
+
+BROAD_SX 5*SIZE(ptrba), xvec1;
+MUL_SX	xvec1, xvec3;
+ADD2_SX	xvec3, xvec15;
+MUL_SX	xvec1, xvec5;
+ADD2_SX	xvec5, xvec14;
+
+BROAD_SX 6*SIZE(ptrba), xvec0;
+LD_SX	24*SIZE(ptrbb), xvec2;
+SHUF_SX	$0xb1, xvec2, xvec3;
+MUL_SX	xvec0, xvec2;
+ADD1_SX	xvec2, xvec15;
+
+LD_SX	28*SIZE(ptrbb), xvec4;
+SHUF_SX	$0xb1, xvec4, xvec5;
+MUL_SX	xvec0, xvec4;
+ADD1_SX	xvec4, xvec14;
+
+BROAD_SX 7*SIZE(ptrba), xvec1;
+MUL_SX	xvec1, xvec3;
+ADD2_SX	xvec3, xvec15;
+MUL_SX	xvec1, xvec5;
+ADD2_SX	xvec5, xvec14;
+ADDQ	$8*SIZE, ptrba;
+ADDQ	$32*SIZE, ptrbb;
+DECQ	k;
+JG	.L14_bodyB;
+.align 32
+.L14_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else
+TEST $2, kkk;
+#endif
+JLE .L15_loopE;
+.align 32
+.L15_bodyB:
+BROAD_SX 0*SIZE(ptrba), xvec0;
+LD_SX	0*SIZE(ptrbb), xvec2;
+SHUF_SX	$0xb1, xvec2, xvec3;
+MUL_SX	xvec0, xvec2;
+ADD1_SX	xvec2, xvec15;
+
+LD_SX	4*SIZE(ptrbb), xvec4;
+SHUF_SX	$0xb1, xvec4, xvec5;
+MUL_SX	xvec0, xvec4;
+ADD1_SX	xvec4, xvec14;
+
+BROAD_SX 1*SIZE(ptrba), xvec1;
+MUL_SX	xvec1, xvec3;
+ADD2_SX	xvec3, xvec15;
+MUL_SX	xvec1, xvec5;
+ADD2_SX	xvec5, xvec14;
+
+BROAD_SX 2*SIZE(ptrba), xvec0;
+LD_SX	8*SIZE(ptrbb), xvec2;
+SHUF_SX	$0xb1, xvec2, xvec3;
+MUL_SX	xvec0, xvec2;
+ADD1_SX	xvec2, xvec15;
+
+LD_SX	12*SIZE(ptrbb), xvec4;
+SHUF_SX	$0xb1, xvec4, xvec5;
+MUL_SX	xvec0, xvec4;
+ADD1_SX	xvec4, xvec14;
+
+BROAD_SX 3*SIZE(ptrba), xvec1;
+MUL_SX	xvec1, xvec3;
+ADD2_SX	xvec3, xvec15;
+MUL_SX	xvec1, xvec5;
+ADD2_SX	xvec5, xvec14;
+ADDQ	$4*SIZE, ptrba;
+ADDQ	$16*SIZE, ptrbb;
+
+.L15_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else
+TEST $1, kkk;
+#endif
+JLE .L16_loopE;
+.align 32
+.L16_bodyB:
+BROAD_SX 0*SIZE(ptrba), xvec0;
+LD_SX	0*SIZE(ptrbb), xvec2;
+SHUF_SX	$0xb1, xvec2, xvec3;
+MUL_SX	xvec0, xvec2;
+ADD1_SX	xvec2, xvec15;
+
+LD_SX	4*SIZE(ptrbb), xvec4;
+SHUF_SX	$0xb1, xvec4, xvec5;
+MUL_SX	xvec0, xvec4;
+ADD1_SX	xvec4, xvec14;
+
+BROAD_SX 1*SIZE(ptrba), xvec1;
+MUL_SX	xvec1, xvec3;
+ADD2_SX	xvec3, xvec15;
+MUL_SX	xvec1, xvec5;
+ADD2_SX	xvec5, xvec14;
+ADDQ	$2*SIZE, ptrba;
+ADDQ	$8*SIZE, ptrbb;
+
+.L16_loopE:
+#### Handle ####
+#if defined(NR) || defined(NC) || defined(TR) || defined(TC)
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec15, xvec7;
+MOV_SX		xvec7, xvec15;
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec14, xvec7;
+MOV_SX		xvec7, xvec14;
+#elif  defined(RN) || defined(RT) || defined(CN) || defined(CT)
+XOR_SY	yvec7, yvec7, yvec7;
+SUB_SX	xvec15, xvec7;
+MOV_SX	xvec7, xvec15;
+XOR_SY	yvec7, yvec7, yvec7;
+SUB_SX	xvec14, xvec7;
+MOV_SX	xvec7, xvec14;
+#elif defined(RR) || defined(RC) || defined(CR) || defined(CC)
+SHUF_SX $0xb1, xvec15, xvec15;
+SHUF_SX $0xb1, xvec14, xvec14;
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec15, xvec7;
+MOV_SX		xvec7, xvec15;
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec14, xvec7;
+MOV_SX		xvec7, xvec14;
+SHUF_SX $0xb1, xvec15, xvec15;
+SHUF_SX $0xb1, xvec14, xvec14;
+#endif
+##### Load Alpha ####
+BROAD_SX MEMALPHA_R,xvec7;
+BROAD_SX MEMALPHA_I,xvec6;
+##### Multiply Alpha ####
+VPERMILP_SX $0xb1,xvec15, xvec5;
+MUL_SX	xvec7, xvec15;
+MUL_SX	xvec6, xvec5;
+ADDSUB_SX	xvec5, xvec15;
+VPERMILP_SX $0xb1,xvec14, xvec4;
+MUL_SX	xvec7, xvec14;
+MUL_SX	xvec6, xvec4;
+ADDSUB_SX	xvec4, xvec14;
+#### Writing back ####
+#ifndef	TRMMKERNEL
+LDL_SX	0*SIZE(C0), xvec0;
+LDH_SX	0*SIZE(C0, ldc, 1), xvec0;
+LDL_SX	0*SIZE(C1), xvec1;
+LDH_SX	0*SIZE(C1, ldc, 1), xvec1;
+ADD_SX	xvec0, xvec15;
+ADD_SX	xvec1, xvec14;
+#endif
+STL_SX	xvec15, 0*SIZE(C0);
+STH_SX	xvec15, 0*SIZE(C0, ldc, 1);
+STL_SX	xvec14, 0*SIZE(C1);
+STH_SX	xvec14, 0*SIZE(C1, ldc, 1);
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA))||(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kkk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+ADDQ %rax, ptrba;
+LEAQ (ptrbb, %rax, 4), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT) 
+ADDQ	$1, kk;
+#endif
+
+ADDQ	$2*SIZE, C0;
+ADDQ	$2*SIZE, C1;
+.L7_loopE:
+#if defined(TRMMKERNEL) && !defined(LEFT)
+ADDQ	$4, kk;
+#endif
+MOVQ bk,k;
+SALQ $5,k;
+ADDQ k,bb;
+LEAQ (C,ldc,4),C;
+.L0_bodyE:;
+DECQ j;
+JG .L0_bodyB;
+.align 32;
+.L0_loopE:;
+TEST $2, bn;
+JLE .L20_loopE;
+.align 32
+.L20_bodyB:
+#if defined(TRMMKERNEL) && defined(LEFT)
+MOVQ OFFSET, %rax;
+MOVQ %rax, kk;
+#endif
+MOVQ C, C0;
+LEAQ (C, ldc, 1), C1;
+MOVQ ba, ptrba;
+MOVQ bm, i;
+SARQ $3, i;
+JLE .L21_loopE;
+.align 32
+.L21_bodyB:
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb,ptrbb;
+#else
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+LEAQ (ptrba, %rax, 8), ptrba;
+LEAQ (ptrbb, %rax, 2), ptrbb;
+#endif
+XOR_SY	yvec15, yvec15, yvec15;
+XOR_SY	yvec14, yvec14, yvec14;
+XOR_SY	yvec13, yvec13, yvec13;
+XOR_SY	yvec12, yvec12, yvec12;
+XOR_SY	yvec11, yvec11, yvec11;
+XOR_SY	yvec10, yvec10, yvec10;
+XOR_SY	yvec9, yvec9, yvec9;
+XOR_SY	yvec8, yvec8, yvec8;
+#ifndef	TRMMKERNEL
+MOVQ bk,k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else
+MOVQ kk, %rax;
+#ifdef	LEFT
+ADDQ $8, %rax;
+#else 
+ADDQ $2, %rax;
+#endif
+MOVQ %rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L211_loopE;
+.align 32
+.L211_bodyB:
+EDUP_SX	0*SIZE(ptrbb), xvec4;
+ODUP_SX	0*SIZE(ptrbb), xvec5;
+SHUF_SX	$0x4e, xvec4, xvec6;
+SHUF_SX	$0x4e, xvec5, xvec7;
+
+LD_SX	0*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec15;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec11;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX	xvec2, xvec15;
+MUL_SX	xvec7, xvec3;
+ADD2_SX	xvec3, xvec11;
+
+LD_SX	4*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec14;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec10;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX xvec2, xvec14;
+MUL_SX	xvec7, xvec3;
+ADD2_SX xvec3, xvec10;
+
+LD_SX	8*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec13;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec9;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX xvec2, xvec13;
+MUL_SX	xvec7, xvec3;
+ADD2_SX xvec3, xvec9;
+
+LD_SX	12*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec12;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec8;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX xvec2, xvec12;
+MUL_SX	xvec7, xvec3;
+ADD2_SX xvec3, xvec8;
+
+EDUP_SX	4*SIZE(ptrbb), xvec4;
+ODUP_SX	4*SIZE(ptrbb), xvec5;
+SHUF_SX	$0x4e, xvec4, xvec6;
+SHUF_SX	$0x4e, xvec5, xvec7;
+
+LD_SX	16*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec15;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec11;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX	xvec2, xvec15;
+MUL_SX	xvec7, xvec3;
+ADD2_SX	xvec3, xvec11;
+
+LD_SX	20*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec14;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec10;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX xvec2, xvec14;
+MUL_SX	xvec7, xvec3;
+ADD2_SX xvec3, xvec10;
+
+LD_SX	24*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec13;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec9;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX xvec2, xvec13;
+MUL_SX	xvec7, xvec3;
+ADD2_SX xvec3, xvec9;
+
+LD_SX	28*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec12;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec8;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX xvec2, xvec12;
+MUL_SX	xvec7, xvec3;
+ADD2_SX xvec3, xvec8;
+
+EDUP_SX	8*SIZE(ptrbb), xvec4;
+ODUP_SX	8*SIZE(ptrbb), xvec5;
+SHUF_SX	$0x4e, xvec4, xvec6;
+SHUF_SX	$0x4e, xvec5, xvec7;
+
+LD_SX	32*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec15;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec11;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX	xvec2, xvec15;
+MUL_SX	xvec7, xvec3;
+ADD2_SX	xvec3, xvec11;
+
+LD_SX	36*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec14;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec10;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX xvec2, xvec14;
+MUL_SX	xvec7, xvec3;
+ADD2_SX xvec3, xvec10;
+
+LD_SX	40*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec13;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec9;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX xvec2, xvec13;
+MUL_SX	xvec7, xvec3;
+ADD2_SX xvec3, xvec9;
+
+LD_SX	44*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec12;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec8;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX xvec2, xvec12;
+MUL_SX	xvec7, xvec3;
+ADD2_SX xvec3, xvec8;
+
+EDUP_SX	12*SIZE(ptrbb), xvec4;
+ODUP_SX	12*SIZE(ptrbb), xvec5;
+SHUF_SX	$0x4e, xvec4, xvec6;
+SHUF_SX	$0x4e, xvec5, xvec7;
+
+LD_SX	48*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec15;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec11;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX	xvec2, xvec15;
+MUL_SX	xvec7, xvec3;
+ADD2_SX	xvec3, xvec11;
+
+LD_SX	52*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec14;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec10;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX xvec2, xvec14;
+MUL_SX	xvec7, xvec3;
+ADD2_SX xvec3, xvec10;
+
+LD_SX	56*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec13;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec9;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX xvec2, xvec13;
+MUL_SX	xvec7, xvec3;
+ADD2_SX xvec3, xvec9;
+
+LD_SX	60*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec12;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec8;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX xvec2, xvec12;
+MUL_SX	xvec7, xvec3;
+ADD2_SX xvec3, xvec8;
+ADDQ	$64*SIZE, ptrba;
+ADDQ	$16*SIZE, ptrbb;
+DECQ 	k;
+JG	.L211_bodyB;
+.align 32
+.L211_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else
+TEST $2, kkk;
+#endif
+JLE .L212_loopE;
+.align 32
+.L212_bodyB:
+EDUP_SX	0*SIZE(ptrbb), xvec4;
+ODUP_SX	0*SIZE(ptrbb), xvec5;
+SHUF_SX	$0x4e, xvec4, xvec6;
+SHUF_SX	$0x4e, xvec5, xvec7;
+
+LD_SX	0*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec15;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec11;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX	xvec2, xvec15;
+MUL_SX	xvec7, xvec3;
+ADD2_SX	xvec3, xvec11;
+
+LD_SX	4*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec14;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec10;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX xvec2, xvec14;
+MUL_SX	xvec7, xvec3;
+ADD2_SX xvec3, xvec10;
+
+LD_SX	8*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec13;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec9;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX xvec2, xvec13;
+MUL_SX	xvec7, xvec3;
+ADD2_SX xvec3, xvec9;
+
+LD_SX	12*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec12;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec8;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX xvec2, xvec12;
+MUL_SX	xvec7, xvec3;
+ADD2_SX xvec3, xvec8;
+
+EDUP_SX	4*SIZE(ptrbb), xvec4;
+ODUP_SX	4*SIZE(ptrbb), xvec5;
+SHUF_SX	$0x4e, xvec4, xvec6;
+SHUF_SX	$0x4e, xvec5, xvec7;
+
+LD_SX	16*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec15;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec11;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX	xvec2, xvec15;
+MUL_SX	xvec7, xvec3;
+ADD2_SX	xvec3, xvec11;
+
+LD_SX	20*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec14;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec10;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX xvec2, xvec14;
+MUL_SX	xvec7, xvec3;
+ADD2_SX xvec3, xvec10;
+
+LD_SX	24*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec13;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec9;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX xvec2, xvec13;
+MUL_SX	xvec7, xvec3;
+ADD2_SX xvec3, xvec9;
+
+LD_SX	28*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec12;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec8;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX xvec2, xvec12;
+MUL_SX	xvec7, xvec3;
+ADD2_SX xvec3, xvec8;
+ADDQ	$32*SIZE, ptrba;
+ADDQ	$8*SIZE, ptrbb;
+
+.L212_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else
+TEST $1, kkk;
+#endif
+JLE .L213_loopE;
+.align 32
+.L213_bodyB:
+EDUP_SX	0*SIZE(ptrbb), xvec4;
+ODUP_SX	0*SIZE(ptrbb), xvec5;
+SHUF_SX	$0x4e, xvec4, xvec6;
+SHUF_SX	$0x4e, xvec5, xvec7;
+
+LD_SX	0*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec15;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec11;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX	xvec2, xvec15;
+MUL_SX	xvec7, xvec3;
+ADD2_SX	xvec3, xvec11;
+
+LD_SX	4*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec14;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec10;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX xvec2, xvec14;
+MUL_SX	xvec7, xvec3;
+ADD2_SX xvec3, xvec10;
+
+LD_SX	8*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec13;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec9;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX xvec2, xvec13;
+MUL_SX	xvec7, xvec3;
+ADD2_SX xvec3, xvec9;
+
+LD_SX	12*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec12;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec8;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX xvec2, xvec12;
+MUL_SX	xvec7, xvec3;
+ADD2_SX xvec3, xvec8;
+ADDQ	$16*SIZE, ptrba;
+ADDQ	$4*SIZE, ptrbb
+
+.L213_loopE:
+#### Handle ####
+#if  defined(RN) || defined(RT) || defined(CN) || defined(CT)
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec15, xvec7;
+MOV_SX		xvec7, xvec15;
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec14, xvec7;
+MOV_SX		xvec7, xvec14;
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec13, xvec7;
+MOV_SX		xvec7, xvec13;
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec12, xvec7;
+MOV_SX		xvec7, xvec12;
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec11, xvec7;
+MOV_SX		xvec7, xvec11;
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec10, xvec7;
+MOV_SX		xvec7, xvec10;
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec9, xvec7;
+MOV_SX		xvec7, xvec9;
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec8, xvec7;
+MOV_SX		xvec7, xvec8;
+#elif defined(NR) || defined(NC) || defined(TR) || defined(TC)
+XOR_SY	yvec7, yvec7, yvec7;
+SUB_SX	xvec15, xvec7;
+MOV_SX	xvec7, xvec15;
+XOR_SY	yvec7, yvec7, yvec7;
+SUB_SX	xvec14, xvec7;
+MOV_SX	xvec7, xvec14;
+XOR_SY	yvec7, yvec7, yvec7;
+SUB_SX	xvec13, xvec7;
+MOV_SX	xvec7, xvec13;
+XOR_SY	yvec7, yvec7, yvec7;
+SUB_SX	xvec12, xvec7;
+MOV_SX	xvec7, xvec12;
+XOR_SY	yvec7, yvec7, yvec7;
+SUB_SX	xvec11, xvec7;
+MOV_SX	xvec7, xvec11;
+XOR_SY	yvec7, yvec7, yvec7;
+SUB_SX	xvec10, xvec7;
+MOV_SX	xvec7, xvec10;
+XOR_SY	yvec7, yvec7, yvec7;
+SUB_SX	xvec9, xvec7;
+MOV_SX	xvec7, xvec9;
+XOR_SY	yvec7, yvec7, yvec7;
+SUB_SX	xvec8, xvec7;
+MOV_SX	xvec7, xvec8;
+#elif defined(RR) || defined(RC) || defined(CR) || defined(CC)
+SHUF_SX $0xb1, xvec15, xvec15;
+SHUF_SX $0xb1, xvec14, xvec14;
+SHUF_SX $0xb1, xvec13, xvec13;
+SHUF_SX $0xb1, xvec12, xvec12;
+SHUF_SX $0xb1, xvec11, xvec11;
+SHUF_SX $0xb1, xvec10, xvec10;
+SHUF_SX $0xb1, xvec9, xvec9;
+SHUF_SX $0xb1, xvec8, xvec8;
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec15, xvec7;
+MOV_SX		xvec7, xvec15;
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec14, xvec7;
+MOV_SX		xvec7, xvec14;
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec13, xvec7;
+MOV_SX		xvec7, xvec13;
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec12, xvec7;
+MOV_SX		xvec7, xvec12;
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec11, xvec7;
+MOV_SX		xvec7, xvec11;
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec10, xvec7;
+MOV_SX		xvec7, xvec10;
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec9, xvec7;
+MOV_SX		xvec7, xvec9;
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec8, xvec7;
+MOV_SX		xvec7, xvec8;
+SHUF_SX $0xb1, xvec15, xvec15;
+SHUF_SX $0xb1, xvec14, xvec14;
+SHUF_SX $0xb1, xvec13, xvec13;
+SHUF_SX $0xb1, xvec12, xvec12;
+SHUF_SX $0xb1, xvec11, xvec11;
+SHUF_SX $0xb1, xvec10, xvec10;
+SHUF_SX $0xb1, xvec9, xvec9;
+SHUF_SX	$0xb1, xvec8, xvec8;
+#endif
+#### Mulitply Alpha ####
+BROAD_SX MEMALPHA_R, xvec7;
+BROAD_SX MEMALPHA_I, xvec6;
+#### Writng back ####
+VPERMILP_SX $0xb1,xvec15, xvec5;
+MUL_SX	xvec7, xvec15;
+MUL_SX	xvec6, xvec5;
+ADDSUB_SX	xvec5, xvec15;
+VPERMILP_SX $0xb1,xvec14, xvec4;
+MUL_SX	xvec7, xvec14;
+MUL_SX	xvec6, xvec4;
+ADDSUB_SX	xvec4, xvec14;
+VPERMILP_SX $0xb1,xvec13, xvec3;
+MUL_SX	xvec7, xvec13;
+MUL_SX	xvec6, xvec3;
+ADDSUB_SX	xvec3, xvec13;
+VPERMILP_SX $0xb1,xvec12, xvec2;
+MUL_SX	xvec7, xvec12;
+MUL_SX	xvec6, xvec2;
+ADDSUB_SX	xvec2, xvec12;
+VPERMILP_SX $0xb1,xvec11, xvec1;
+MUL_SX	xvec7, xvec11;
+MUL_SX	xvec6, xvec1;
+ADDSUB_SX	xvec1, xvec11;
+VPERMILP_SX $0xb1,xvec10, xvec0;
+MUL_SX	xvec7, xvec10;
+MUL_SX	xvec6, xvec0;
+ADDSUB_SX	xvec0, xvec10;
+VPERMILP_SX $0xb1,xvec9, xvec5;
+MUL_SX	xvec7, xvec9;
+MUL_SX	xvec6, xvec5;
+ADDSUB_SX	xvec5, xvec9;
+VPERMILP_SX $0xb1,xvec8, xvec4;
+MUL_SX	xvec7, xvec8;
+MUL_SX	xvec6, xvec4;
+ADDSUB_SX	xvec4, xvec8;
+#ifndef	TRMMKERNEL
+LDL_SX	0*SIZE(C0), xvec0;
+LDH_SX	2*SIZE(C1), xvec0;
+LDL_SX	4*SIZE(C0), xvec1;
+LDH_SX	6*SIZE(C1), xvec1;
+LDL_SX	8*SIZE(C0), xvec2;
+LDH_SX	10*SIZE(C1), xvec2;
+LDL_SX	12*SIZE(C0), xvec3;
+LDH_SX	14*SIZE(C1), xvec3;
+ADD_SX	xvec0, xvec15;
+ADD_SX	xvec1, xvec14;
+ADD_SX	xvec2, xvec13;
+ADD_SX	xvec3, xvec12;
+#endif
+STL_SX	xvec15, 0*SIZE(C0);
+STH_SX	xvec15, 2*SIZE(C1);
+STL_SX	xvec14, 4*SIZE(C0);
+STH_SX	xvec14, 6*SIZE(C1);
+STL_SX	xvec13, 8*SIZE(C0);
+STH_SX	xvec13, 10*SIZE(C1);
+STL_SX	xvec12, 12*SIZE(C0);
+STH_SX	xvec12, 14*SIZE(C1);
+#ifndef	TRMMKERNEL
+LDL_SX	0*SIZE(C1), xvec4;
+LDH_SX	2*SIZE(C0), xvec4;
+LDL_SX	4*SIZE(C1), xvec5;
+LDH_SX	6*SIZE(C0), xvec5;
+LDL_SX	8*SIZE(C1), xvec6;
+LDH_SX	10*SIZE(C0), xvec6;
+LDL_SX	12*SIZE(C1), xvec7;
+LDH_SX	14*SIZE(C0), xvec7;
+ADD_SX	xvec4, xvec11;
+ADD_SX	xvec5, xvec10;
+ADD_SX	xvec6, xvec9;
+ADD_SX	xvec7, xvec8;
+#endif
+STL_SX	xvec11, 0*SIZE(C1);
+STH_SX	xvec11, 2*SIZE(C0);
+STL_SX	xvec10, 4*SIZE(C1);
+STH_SX	xvec10, 6*SIZE(C0);
+STL_SX	xvec9, 8*SIZE(C1);
+STH_SX	xvec9, 10*SIZE(C0);
+STL_SX	xvec8, 12*SIZE(C1);
+STH_SX	xvec8, 14*SIZE(C0);
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA))||(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kkk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+LEAQ (ptrba, %rax, 8), ptrba;
+LEAQ (ptrbb, %rax, 2), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT) 
+ADDQ	$8, kk;
+#endif
+
+ADDQ	$16*SIZE, C0;
+ADDQ	$16*SIZE, C1;
+DECQ	i;
+JG .L21_bodyB;
+.align 32
+.L21_loopE:
+TEST $4, bm;
+JLE .L22_loopE;
+.align 32
+.L22_bodyB:
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb,ptrbb;
+#else
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+LEAQ (ptrba, %rax, 4), ptrba;
+LEAQ (ptrbb, %rax, 2), ptrbb;
+#endif
+XOR_SY	yvec15, yvec15, yvec15;
+XOR_SY	yvec14, yvec14, yvec14;
+XOR_SY	yvec11, yvec11, yvec11;
+XOR_SY	yvec10, yvec10, yvec10;
+#ifndef	TRMMKERNEL
+MOVQ bk,k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else
+MOVQ kk, %rax;
+#ifdef	LEFT
+ADDQ $4, %rax;
+#else 
+ADDQ $2, %rax;
+#endif
+MOVQ %rax, kkk;
+#endif
+
+SARQ $2, k;
+JLE .L221_loopE;
+.align 32
+.L221_bodyB:
+EDUP_SX	0*SIZE(ptrbb), xvec4;
+ODUP_SX	0*SIZE(ptrbb), xvec5;
+SHUF_SX	$0x4e, xvec4, xvec6;
+SHUF_SX	$0x4e, xvec5, xvec7;
+
+LD_SX	0*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec15;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec11;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX	xvec2, xvec15;
+MUL_SX	xvec7, xvec3;
+ADD2_SX	xvec3, xvec11;
+
+LD_SX	4*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec14;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec10;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX xvec2, xvec14;
+MUL_SX	xvec7, xvec3;
+ADD2_SX xvec3, xvec10;
+
+#### Unroll 2 #####
+EDUP_SX	4*SIZE(ptrbb), xvec4;
+ODUP_SX	4*SIZE(ptrbb), xvec5;
+SHUF_SX	$0x4e, xvec4, xvec6;
+SHUF_SX	$0x4e, xvec5, xvec7;
+
+LD_SX	8*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec15;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec11;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX	xvec2, xvec15;
+MUL_SX	xvec7, xvec3;
+ADD2_SX	xvec3, xvec11;
+
+LD_SX	12*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec14;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec10;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX xvec2, xvec14;
+MUL_SX	xvec7, xvec3;
+ADD2_SX xvec3, xvec10;
+
+#### Unroll 3 ####
+EDUP_SX	8*SIZE(ptrbb), xvec4;
+ODUP_SX	8*SIZE(ptrbb), xvec5;
+SHUF_SX	$0x4e, xvec4, xvec6;
+SHUF_SX	$0x4e, xvec5, xvec7;
+
+LD_SX	16*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec15;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec11;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX	xvec2, xvec15;
+MUL_SX	xvec7, xvec3;
+ADD2_SX	xvec3, xvec11;
+
+LD_SX	20*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec14;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec10;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX xvec2, xvec14;
+MUL_SX	xvec7, xvec3;
+ADD2_SX xvec3, xvec10;
+
+#### Unroll 4 ####
+EDUP_SX	12*SIZE(ptrbb), xvec4;
+ODUP_SX	12*SIZE(ptrbb), xvec5;
+SHUF_SX	$0x4e, xvec4, xvec6;
+SHUF_SX	$0x4e, xvec5, xvec7;
+
+LD_SX	24*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec15;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec11;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX	xvec2, xvec15;
+MUL_SX	xvec7, xvec3;
+ADD2_SX	xvec3, xvec11;
+
+LD_SX	28*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec14;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec10;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX xvec2, xvec14;
+MUL_SX	xvec7, xvec3;
+ADD2_SX xvec3, xvec10;
+ADDQ	$32*SIZE, ptrba;
+ADDQ	$16*SIZE, ptrbb;
+DECQ	k;
+JG	.L221_bodyB;
+.align 32
+.L221_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else 
+TEST $2, kkk;
+#endif
+JLE .L222_loopE;
+.align 32
+.L222_bodyB:
+EDUP_SX	0*SIZE(ptrbb), xvec4;
+ODUP_SX	0*SIZE(ptrbb), xvec5;
+SHUF_SX	$0x4e, xvec4, xvec6;
+SHUF_SX	$0x4e, xvec5, xvec7;
+
+LD_SX	0*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec15;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec11;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX	xvec2, xvec15;
+MUL_SX	xvec7, xvec3;
+ADD2_SX	xvec3, xvec11;
+
+LD_SX	4*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec14;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec10;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX xvec2, xvec14;
+MUL_SX	xvec7, xvec3;
+ADD2_SX xvec3, xvec10;
+
+#### Unroll 2 #####
+EDUP_SX	4*SIZE(ptrbb), xvec4;
+ODUP_SX	4*SIZE(ptrbb), xvec5;
+SHUF_SX	$0x4e, xvec4, xvec6;
+SHUF_SX	$0x4e, xvec5, xvec7;
+
+LD_SX	8*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec15;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec11;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX	xvec2, xvec15;
+MUL_SX	xvec7, xvec3;
+ADD2_SX	xvec3, xvec11;
+
+LD_SX	12*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec14;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec10;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX xvec2, xvec14;
+MUL_SX	xvec7, xvec3;
+ADD2_SX xvec3, xvec10;
+ADDQ	$16*SIZE, ptrba;
+ADDQ	$8*SIZE, ptrbb;
+
+
+.L222_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else
+TEST $1, kkk;
+#endif
+JLE .L223_loopE;
+.align 32
+.L223_bodyB:
+EDUP_SX	0*SIZE(ptrbb), xvec4;
+ODUP_SX	0*SIZE(ptrbb), xvec5;
+SHUF_SX	$0x4e, xvec4, xvec6;
+SHUF_SX	$0x4e, xvec5, xvec7;
+
+LD_SX	0*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec15;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec11;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX	xvec2, xvec15;
+MUL_SX	xvec7, xvec3;
+ADD2_SX	xvec3, xvec11;
+
+LD_SX	4*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec14;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec10;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX xvec2, xvec14;
+MUL_SX	xvec7, xvec3;
+ADD2_SX xvec3, xvec10;
+ADDQ	$8*SIZE, ptrba;
+ADDQ	$4*SIZE, ptrbb;
+
+.L223_loopE:
+#### Handle ####
+#if  defined(RN) || defined(RT) || defined(CN) || defined(CT)
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec15, xvec7;
+MOV_SX		xvec7, xvec15;
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec14, xvec7;
+MOV_SX		xvec7, xvec14;
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec11, xvec7;
+MOV_SX		xvec7, xvec11;
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec10, xvec7;
+MOV_SX		xvec7, xvec10;
+#elif defined(NR) || defined(NC) || defined(TR) || defined(TC)
+XOR_SY	yvec7, yvec7, yvec7;
+SUB_SX	xvec15, xvec7;
+MOV_SX	xvec7, xvec15;
+XOR_SY	yvec7, yvec7, yvec7;
+SUB_SX	xvec14, xvec7;
+MOV_SX	xvec7, xvec14;
+XOR_SY	yvec7, yvec7, yvec7;
+SUB_SX	xvec11, xvec7;
+MOV_SX	xvec7, xvec11;
+XOR_SY	yvec7, yvec7, yvec7;
+SUB_SX	xvec10, xvec7;
+MOV_SX	xvec7, xvec10;
+#elif defined(RR) || defined(RC) || defined(CR) || defined(CC)
+SHUF_SX $0xb1, xvec15, xvec15;
+SHUF_SX $0xb1, xvec14, xvec14;
+SHUF_SX $0xb1, xvec11, xvec11;
+SHUF_SX $0xb1, xvec10, xvec10;
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec15, xvec7;
+MOV_SX		xvec7, xvec15;
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec14, xvec7;
+MOV_SX		xvec7, xvec14;
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec11, xvec7;
+MOV_SX		xvec7, xvec11;
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec10, xvec7;
+MOV_SX		xvec7, xvec10;
+SHUF_SX $0xb1, xvec15, xvec15;
+SHUF_SX $0xb1, xvec14, xvec14;
+SHUF_SX $0xb1, xvec11, xvec11;
+SHUF_SX $0xb1, xvec10, xvec10;
+#endif
+#### Mulitply Alpha ####
+BROAD_SX MEMALPHA_R, xvec7;
+BROAD_SX MEMALPHA_I, xvec6;
+#### Writng back ####
+VPERMILP_SX $0xb1,xvec15, xvec5;
+MUL_SX	xvec7, xvec15;
+MUL_SX	xvec6, xvec5;
+ADDSUB_SX	xvec5, xvec15;
+VPERMILP_SX $0xb1,xvec14, xvec4;
+MUL_SX	xvec7, xvec14;
+MUL_SX	xvec6, xvec4;
+ADDSUB_SX	xvec4, xvec14;
+VPERMILP_SX $0xb1,xvec11, xvec1;
+MUL_SX	xvec7, xvec11;
+MUL_SX	xvec6, xvec1;
+ADDSUB_SX	xvec1, xvec11;
+VPERMILP_SX $0xb1,xvec10, xvec0;
+MUL_SX	xvec7, xvec10;
+MUL_SX	xvec6, xvec0;
+ADDSUB_SX	xvec0, xvec10;
+#ifndef	TRMMKERNEL
+LDL_SX	0*SIZE(C0), xvec0;
+LDH_SX	2*SIZE(C1), xvec0;
+LDL_SX	4*SIZE(C0), xvec1;
+LDH_SX	6*SIZE(C1), xvec1;
+ADD_SX	xvec0, xvec15;
+ADD_SX	xvec1, xvec14;
+#endif
+STL_SX	xvec15, 0*SIZE(C0);
+STH_SX	xvec15, 2*SIZE(C1);
+STL_SX	xvec14, 4*SIZE(C0);
+STH_SX	xvec14, 6*SIZE(C1);
+#ifndef	TRMMKERNEL
+LDL_SX	0*SIZE(C1), xvec4;
+LDH_SX	2*SIZE(C0), xvec4;
+LDL_SX	4*SIZE(C1), xvec5;
+LDH_SX	6*SIZE(C0), xvec5;
+ADD_SX	xvec4, xvec11;
+ADD_SX	xvec5, xvec10;
+#endif
+STL_SX	xvec11, 0*SIZE(C1);
+STH_SX	xvec11, 2*SIZE(C0);
+STL_SX	xvec10, 4*SIZE(C1);
+STH_SX	xvec10, 6*SIZE(C0);
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA))||(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kkk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+LEAQ (ptrba, %rax, 4), ptrba;
+LEAQ (ptrbb, %rax, 2), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT) 
+ADDQ	$4, kk;
+#endif
+
+ADDQ	$8*SIZE, C0;
+ADDQ	$8*SIZE, C1;
+
+.L22_loopE:
+TEST $2, bm;
+JLE .L23_loopE;
+.align 32
+.L23_bodyB:
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb,ptrbb;
+#else
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+LEAQ (ptrba, %rax, 2), ptrba;
+LEAQ (ptrbb, %rax, 2), ptrbb;
+#endif
+XOR_SY	yvec15, yvec15, yvec15;
+XOR_SY	yvec11, yvec11, yvec11;
+#ifndef	TRMMKERNEL
+MOVQ bk,k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else
+MOVQ kk, %rax;
+#ifdef	LEFT
+ADDQ $2, %rax;
+#else 
+ADDQ $2, %rax;
+#endif
+MOVQ %rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L231_loopE;
+.align 32
+.L231_bodyB:
+EDUP_SX	0*SIZE(ptrbb), xvec4;
+ODUP_SX	0*SIZE(ptrbb), xvec5;
+SHUF_SX	$0x4e, xvec4, xvec6;
+SHUF_SX	$0x4e, xvec5, xvec7;
+
+LD_SX	0*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec15;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec11;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX	xvec2, xvec15;
+MUL_SX	xvec7, xvec3;
+ADD2_SX	xvec3, xvec11;
+
+#### Unroll 2 #####
+EDUP_SX	4*SIZE(ptrbb), xvec4;
+ODUP_SX	4*SIZE(ptrbb), xvec5;
+SHUF_SX	$0x4e, xvec4, xvec6;
+SHUF_SX	$0x4e, xvec5, xvec7;
+
+LD_SX	4*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec15;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec11;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX	xvec2, xvec15;
+MUL_SX	xvec7, xvec3;
+ADD2_SX	xvec3, xvec11;
+
+#### Unroll 3 ####
+EDUP_SX	8*SIZE(ptrbb), xvec4;
+ODUP_SX	8*SIZE(ptrbb), xvec5;
+SHUF_SX	$0x4e, xvec4, xvec6;
+SHUF_SX	$0x4e, xvec5, xvec7;
+
+LD_SX	8*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec15;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec11;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX	xvec2, xvec15;
+MUL_SX	xvec7, xvec3;
+ADD2_SX	xvec3, xvec11;
+
+#### Unroll 4 ####
+EDUP_SX	12*SIZE(ptrbb), xvec4;
+ODUP_SX	12*SIZE(ptrbb), xvec5;
+SHUF_SX	$0x4e, xvec4, xvec6;
+SHUF_SX	$0x4e, xvec5, xvec7;
+
+LD_SX	12*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec15;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec11;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX	xvec2, xvec15;
+MUL_SX	xvec7, xvec3;
+ADD2_SX	xvec3, xvec11;
+ADDQ	$16*SIZE, ptrba;
+ADDQ	$16*SIZE, ptrbb;
+DECQ	k;
+JG	.L231_bodyB;
+.align 32
+.L231_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else
+TEST $2, kkk;
+#endif
+JLE .L232_loopE;
+.align 32 
+.L232_bodyB:
+EDUP_SX	0*SIZE(ptrbb), xvec4;
+ODUP_SX	0*SIZE(ptrbb), xvec5;
+SHUF_SX	$0x4e, xvec4, xvec6;
+SHUF_SX	$0x4e, xvec5, xvec7;
+
+LD_SX	0*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec15;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec11;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX	xvec2, xvec15;
+MUL_SX	xvec7, xvec3;
+ADD2_SX	xvec3, xvec11;
+
+#### Unroll 2 #####
+EDUP_SX	4*SIZE(ptrbb), xvec4;
+ODUP_SX	4*SIZE(ptrbb), xvec5;
+SHUF_SX	$0x4e, xvec4, xvec6;
+SHUF_SX	$0x4e, xvec5, xvec7;
+
+LD_SX	4*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec15;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec11;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX	xvec2, xvec15;
+MUL_SX	xvec7, xvec3;
+ADD2_SX	xvec3, xvec11;
+ADDQ	$8*SIZE, ptrba;
+ADDQ	$8*SIZE, ptrbb;
+
+.L232_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else
+TEST $1, kkk;
+#endif
+JLE .L233_loopE;
+.align 32
+.L233_bodyB:
+EDUP_SX	0*SIZE(ptrbb), xvec4;
+ODUP_SX	0*SIZE(ptrbb), xvec5;
+SHUF_SX	$0x4e, xvec4, xvec6;
+SHUF_SX	$0x4e, xvec5, xvec7;
+
+LD_SX	0*SIZE(ptrba), xvec0;
+MOV_SX	xvec0, xvec1;
+MUL_SX	xvec4, xvec0;
+ADD1_SX	xvec0, xvec15;
+SHUF_SX	$0xb1, xvec1, xvec2;
+MUL_SX	xvec6, xvec1;
+ADD1_SX	xvec1, xvec11;
+
+MOV_SX	xvec2, xvec3;
+MUL_SX	xvec5, xvec2;
+ADD2_SX	xvec2, xvec15;
+MUL_SX	xvec7, xvec3;
+ADD2_SX	xvec3, xvec11;
+ADDQ	$4*SIZE, ptrba;
+ADDQ	$4*SIZE, ptrbb;
+
+.L233_loopE:
+#### Handle ####
+#if  defined(RN) || defined(RT) || defined(CN) || defined(CT)
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec15, xvec7;
+MOV_SX		xvec7, xvec15;
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec11, xvec7;
+MOV_SX		xvec7, xvec11;
+#elif defined(NR) || defined(NC) || defined(TR) || defined(TC)
+XOR_SY	yvec7, yvec7, yvec7;
+SUB_SX	xvec15, xvec7;
+MOV_SX	xvec7, xvec15;
+XOR_SY	yvec7, yvec7, yvec7;
+SUB_SX	xvec11, xvec7;
+MOV_SX	xvec7, xvec11;
+#elif defined(RR) || defined(RC) || defined(CR) || defined(CC)
+SHUF_SX $0xb1, xvec15, xvec15;
+SHUF_SX $0xb1, xvec11, xvec11;
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec15, xvec7;
+MOV_SX		xvec7, xvec15;
+XOR_SY	yvec7, yvec7, yvec7;
+ADDSUB_SX	xvec11, xvec7;
+MOV_SX		xvec7, xvec11;
+SHUF_SX $0xb1, xvec15, xvec15;
+SHUF_SX $0xb1, xvec11, xvec11;
+#endif
+#### Mulitply Alpha ####
+BROAD_SX MEMALPHA_R, xvec7;
+BROAD_SX MEMALPHA_I, xvec6;
+#### Writng back ####
+VPERMILP_SX $0xb1,xvec15, xvec5;
+MUL_SX	xvec7, xvec15;
+MUL_SX	xvec6, xvec5;
+ADDSUB_SX	xvec5, xvec15;
+VPERMILP_SX $0xb1,xvec11, xvec1;
+MUL_SX	xvec7, xvec11;
+MUL_SX	xvec6, xvec1;
+ADDSUB_SX	xvec1, xvec11;
+#ifndef	TRMMKERNEL
+LDL_SX	0*SIZE(C0), xvec0;
+LDH_SX	2*SIZE(C1), xvec0;
+ADD_SX	xvec0, xvec15;
+#endif
+STL_SX	xvec15, 0*SIZE(C0);
+STH_SX	xvec15, 2*SIZE(C1);
+#ifndef	TRMMKERNEL
+LDL_SX	0*SIZE(C1), xvec4;
+LDH_SX	2*SIZE(C0), xvec4;
+ADD_SX	xvec4, xvec11;
+#endif
+STL_SX	xvec11, 0*SIZE(C1);
+STH_SX	xvec11, 2*SIZE(C0);
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA))||(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kkk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+LEAQ (ptrba, %rax, 2), ptrba;
+LEAQ (ptrbb, %rax, 2), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT) 
+ADDQ	$2, kk;
+#endif
+
+ADDQ	$4*SIZE, C0;
+ADDQ	$4*SIZE, C1;
+
+.L23_loopE:
+TEST $1, bm;
+JLE .L24_loopE;
+.align 32
+.L24_bodyB:
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb,ptrbb;
+#else
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+ADDQ %rax, ptrba;
+LEAQ (ptrbb, %rax, 2), ptrbb;
+#endif
+XOR_SY	yvec15, yvec15, yvec15;
+#ifndef	TRMMKERNEL
+MOVQ bk,k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else
+MOVQ kk, %rax;
+#ifdef	LEFT
+ADDQ $1, %rax;
+#else 
+ADDQ $2, %rax;
+#endif
+MOVQ %rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L241_loopE;
+.align 32
+.L241_bodyB:
+BROAD_SX 0*SIZE(ptrba), xvec0;
+LD_SX	0*SIZE(ptrbb), xvec2;
+SHUF_SX	$0xb1, xvec2, xvec3;
+MUL_SX	xvec0, xvec2;
+ADD1_SX	xvec2, xvec15;
+
+BROAD_SX 1*SIZE(ptrba), xvec1;
+MUL_SX	xvec1, xvec3;
+ADD2_SX	xvec3, xvec15;
+
+BROAD_SX 2*SIZE(ptrba), xvec0;
+LD_SX	4*SIZE(ptrbb), xvec2;
+SHUF_SX	$0xb1, xvec2, xvec3;
+MUL_SX	xvec0, xvec2;
+ADD1_SX	xvec2, xvec15;
+
+BROAD_SX 3*SIZE(ptrba), xvec1;
+MUL_SX	xvec1, xvec3;
+ADD2_SX	xvec3, xvec15;
+
+BROAD_SX 4*SIZE(ptrba), xvec0;
+LD_SX	8*SIZE(ptrbb), xvec2;
+SHUF_SX	$0xb1, xvec2, xvec3;
+MUL_SX	xvec0, xvec2;
+ADD1_SX	xvec2, xvec15;
+
+BROAD_SX 5*SIZE(ptrba), xvec1;
+MUL_SX	xvec1, xvec3;
+ADD2_SX	xvec3, xvec15;
+
+BROAD_SX 6*SIZE(ptrba), xvec0;
+LD_SX	12*SIZE(ptrbb), xvec2;
+SHUF_SX	$0xb1, xvec2, xvec3;
+MUL_SX	xvec0, xvec2;
+ADD1_SX	xvec2, xvec15;
+
+BROAD_SX 7*SIZE(ptrba), xvec1;
+MUL_SX	xvec1, xvec3;
+ADD2_SX	xvec3, xvec15;
+ADDQ	$8*SIZE, ptrba;
+ADDQ	$16*SIZE, ptrbb;
+DECQ	k;
+JG	.L241_bodyB;
+.align
+.L241_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else
+TEST $2, kkk;
+#endif
+JLE .L242_loopE;
+.align 32
+.L242_bodyB:
+BROAD_SX 0*SIZE(ptrba), xvec0;
+LD_SX	0*SIZE(ptrbb), xvec2;
+SHUF_SX	$0xb1, xvec2, xvec3;
+MUL_SX	xvec0, xvec2;
+ADD1_SX	xvec2, xvec15;
+
+BROAD_SX 1*SIZE(ptrba), xvec1;
+MUL_SX	xvec1, xvec3;
+ADD2_SX	xvec3, xvec15;
+
+BROAD_SX 2*SIZE(ptrba), xvec0;
+LD_SX	4*SIZE(ptrbb), xvec2;
+SHUF_SX	$0xb1, xvec2, xvec3;
+MUL_SX	xvec0, xvec2;
+ADD1_SX	xvec2, xvec15;
+
+BROAD_SX 3*SIZE(ptrba), xvec1;
+MUL_SX	xvec1, xvec3;
+ADD2_SX	xvec3, xvec15;
+ADDQ	$4*SIZE, ptrba;
+ADDQ	$8*SIZE, ptrbb;
+
+.L242_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else
+TEST $1, kkk;
+#endif
+JLE .L243_loopE;
+.align 32
+.L243_bodyB:
+BROAD_SX 0*SIZE(ptrba), xvec0;
+LD_SX	0*SIZE(ptrbb), xvec2;
+SHUF_SX	$0xb1, xvec2, xvec3;
+MUL_SX	xvec0, xvec2;
+ADD1_SX	xvec2, xvec15;
+
+BROAD_SX 1*SIZE(ptrba), xvec1;
+MUL_SX	xvec1, xvec3;
+ADD2_SX	xvec3, xvec15;
+ADDQ	$2*SIZE, ptrba;
+ADDQ	$4*SIZE, ptrbb;
+
+.L243_loopE:
+#### Handle ####
+XOR_SY	yvec7, yvec7, yvec7;
+#if defined(NR) || defined(NC) || defined(TR) || defined(TC)
+ADDSUB_SX	xvec15, xvec7;
+MOV_SX		xvec7, xvec15;
+#elif  defined(RN) || defined(RT) || defined(CN) || defined(CT)
+SUB_SX	xvec15, xvec7;
+MOV_SX	xvec7, xvec15;
+#elif defined(RR) || defined(RC) || defined(CR) || defined(CC)
+SHUF_SX $0xb1, xvec15, xvec15;
+ADDSUB_SX	xvec15, xvec7;
+MOV_SX		xvec7, xvec15;
+SHUF_SX $0xb1, xvec15, xvec15;
+#endif
+##### Load Alpha ####
+BROAD_SX MEMALPHA_R,xvec7;
+BROAD_SX MEMALPHA_I,xvec6;
+##### Multiply Alpha ####
+VPERMILP_SX $0xb1,xvec15, xvec5;
+MUL_SX	xvec7, xvec15;
+MUL_SX	xvec6, xvec5;
+ADDSUB_SX	xvec5, xvec15;
+#### Writing back ####
+#ifndef	TRMMKERNEL
+LDL_SX	0*SIZE(C0), xvec0;
+LDH_SX	0*SIZE(C1), xvec0;
+ADD_SX	xvec0, xvec15;
+#endif
+STL_SX	xvec15, 0*SIZE(C0);
+STH_SX	xvec15, 0*SIZE(C1);
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA))||(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kkk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+ADDQ %rax, ptrba;
+LEAQ (ptrbb, %rax, 2), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT) 
+ADDQ	$1, kk;
+#endif
+ADDQ	$2*SIZE, C0;
+ADDQ	$2*SIZE, C1;
+.L24_loopE:
+#if defined(TRMMKERNEL) && !defined(LEFT)
+ADDQ	$2, kk;
+#endif
+MOVQ bk, k;
+SALQ $4, k;
+ADDQ k, bb;
+LEAQ (C, ldc, 2), C;
+.L20_loopE:
+TEST $1, bn;
+JLE .L30_loopE;
+.align 32
+.L30_bodyB:
+#if defined(TRMMKERNEL) && defined(LEFT)
+MOVQ OFFSET, %rax;
+MOVQ %rax, kk;
+#endif
+MOVQ C, C0;
+MOVQ ba, ptrba;
+MOVQ bm, i;
+SARQ $3, i;
+JLE .L31_loopE;
+.align 32
+.L31_bodyB:
+MOVQ bb, ptrbb;
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb,ptrbb;
+#else
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+LEAQ (ptrba, %rax, 8), ptrba;
+ADDQ %rax, ptrbb;
+#endif
+XOR_SY	yvec15, yvec15, yvec15;
+XOR_SY	yvec14, yvec14, yvec14;
+MOVQ bk, k;
+#ifndef	TRMMKERNEL
+MOVQ bk,k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else
+MOVQ kk, %rax;
+#ifdef	LEFT
+ADDQ $8, %rax;
+#else 
+ADDQ $1, %rax;
+#endif
+MOVQ %rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L311_loopE;
+.align 32
+.L311_bodyB:
+#### Unroll 1 ####
+LD_SY	0*SIZE(ptrba), yvec0;
+LD_SY	8*SIZE(ptrba), yvec1;
+BROAD_SY 0*SIZE(ptrbb), yvec2;
+MUL_SY	yvec0, yvec2, yvec6;
+ADD1_SY	yvec6, yvec15, yvec15;
+MUL_SY	yvec1, yvec2, yvec7;
+ADD1_SY	yvec7, yvec14, yvec14;
+
+BROAD_SY 1*SIZE(ptrbb), yvec3;
+VPERMILP_SY $0xb1, yvec0, yvec4;
+VPERMILP_SY	$0xb1, yvec1, yvec5;
+MUL_SY	yvec4, yvec3, yvec6;
+ADD2_SY yvec6, yvec15, yvec15;
+MUL_SY	yvec5, yvec3, yvec7;
+ADD2_SY yvec7, yvec14, yvec14;
+
+#### Unroll 2 ####
+LD_SY	16*SIZE(ptrba), yvec0;
+LD_SY	24*SIZE(ptrba), yvec1;
+BROAD_SY 2*SIZE(ptrbb), yvec2;
+MUL_SY	yvec0, yvec2, yvec6;
+ADD1_SY	yvec6, yvec15, yvec15;
+MUL_SY	yvec1, yvec2, yvec7;
+ADD1_SY	yvec7, yvec14, yvec14;
+
+BROAD_SY 3*SIZE(ptrbb), yvec3;
+VPERMILP_SY $0xb1, yvec0, yvec4;
+VPERMILP_SY	$0xb1, yvec1, yvec5;
+MUL_SY	yvec4, yvec3, yvec6;
+ADD2_SY yvec6, yvec15, yvec15;
+MUL_SY	yvec5, yvec3, yvec7;
+ADD2_SY yvec7, yvec14, yvec14;
+
+#### Unroll 3 ####
+LD_SY	32*SIZE(ptrba), yvec0;
+LD_SY	40*SIZE(ptrba), yvec1;
+BROAD_SY 4*SIZE(ptrbb), yvec2;
+MUL_SY	yvec0, yvec2, yvec6;
+ADD1_SY	yvec6, yvec15, yvec15;
+MUL_SY	yvec1, yvec2, yvec7;
+ADD1_SY	yvec7, yvec14, yvec14;
+
+BROAD_SY 5*SIZE(ptrbb), yvec3;
+VPERMILP_SY $0xb1, yvec0, yvec4;
+VPERMILP_SY	$0xb1, yvec1, yvec5;
+MUL_SY	yvec4, yvec3, yvec6;
+ADD2_SY yvec6, yvec15, yvec15;
+MUL_SY	yvec5, yvec3, yvec7;
+ADD2_SY yvec7, yvec14, yvec14;
+
+#### Unroll 4 ####
+LD_SY	48*SIZE(ptrba), yvec0;
+LD_SY	56*SIZE(ptrba), yvec1;
+BROAD_SY 6*SIZE(ptrbb), yvec2;
+MUL_SY	yvec0, yvec2, yvec6;
+ADD1_SY	yvec6, yvec15, yvec15;
+MUL_SY	yvec1, yvec2, yvec7;
+ADD1_SY	yvec7, yvec14, yvec14;
+
+BROAD_SY 7*SIZE(ptrbb), yvec3;
+VPERMILP_SY $0xb1, yvec0, yvec4;
+VPERMILP_SY	$0xb1, yvec1, yvec5;
+MUL_SY	yvec4, yvec3, yvec6;
+ADD2_SY yvec6, yvec15, yvec15;
+MUL_SY	yvec5, yvec3, yvec7;
+ADD2_SY yvec7, yvec14, yvec14;
+ADDQ	$64*SIZE, ptrba;
+ADDQ	$8*SIZE, ptrbb;
+DECQ	k;
+JG	.L311_bodyB;
+.align 32
+.L311_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else
+TEST $2, kkk;
+#endif
+JLE .L312_loopE;
+.align 32
+.L312_bodyB:
+#### Unroll 1 ####
+LD_SY	0*SIZE(ptrba), yvec0;
+LD_SY	8*SIZE(ptrba), yvec1;
+BROAD_SY 0*SIZE(ptrbb), yvec2;
+MUL_SY	yvec0, yvec2, yvec6;
+ADD1_SY	yvec6, yvec15, yvec15;
+MUL_SY	yvec1, yvec2, yvec7;
+ADD1_SY	yvec7, yvec14, yvec14;
+
+BROAD_SY 1*SIZE(ptrbb), yvec3;
+VPERMILP_SY $0xb1, yvec0, yvec4;
+VPERMILP_SY	$0xb1, yvec1, yvec5;
+MUL_SY	yvec4, yvec3, yvec6;
+ADD2_SY yvec6, yvec15, yvec15;
+MUL_SY	yvec5, yvec3, yvec7;
+ADD2_SY yvec7, yvec14, yvec14;
+
+#### Unroll 2 ####
+LD_SY	16*SIZE(ptrba), yvec0;
+LD_SY	24*SIZE(ptrba), yvec1;
+BROAD_SY 2*SIZE(ptrbb), yvec2;
+MUL_SY	yvec0, yvec2, yvec6;
+ADD1_SY	yvec6, yvec15, yvec15;
+MUL_SY	yvec1, yvec2, yvec7;
+ADD1_SY	yvec7, yvec14, yvec14;
+
+BROAD_SY 3*SIZE(ptrbb), yvec3;
+VPERMILP_SY $0xb1, yvec0, yvec4;
+VPERMILP_SY	$0xb1, yvec1, yvec5;
+MUL_SY	yvec4, yvec3, yvec6;
+ADD2_SY yvec6, yvec15, yvec15;
+MUL_SY	yvec5, yvec3, yvec7;
+ADD2_SY yvec7, yvec14, yvec14;
+ADDQ	$32*SIZE, ptrba;
+ADDQ	$4*SIZE, ptrbb;
+
+.L312_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else
+TEST $1, kkk;
+#endif
+JLE .L313_loopE;
+.align 32
+.L313_bodyB:
+#### Unroll 1 ####
+LD_SY	0*SIZE(ptrba), yvec0;
+LD_SY	8*SIZE(ptrba), yvec1;
+BROAD_SY 0*SIZE(ptrbb), yvec2;
+MUL_SY	yvec0, yvec2, yvec6;
+ADD1_SY	yvec6, yvec15, yvec15;
+MUL_SY	yvec1, yvec2, yvec7;
+ADD1_SY	yvec7, yvec14, yvec14;
+
+BROAD_SY 1*SIZE(ptrbb), yvec3;
+VPERMILP_SY $0xb1, yvec0, yvec4;
+VPERMILP_SY	$0xb1, yvec1, yvec5;
+MUL_SY	yvec4, yvec3, yvec6;
+ADD2_SY yvec6, yvec15, yvec15;
+MUL_SY	yvec5, yvec3, yvec7;
+ADD2_SY yvec7, yvec14, yvec14;
+ADDQ	$16*SIZE, ptrba;
+ADDQ	$2*SIZE, ptrbb;
+
+.L313_loopE:
+#### Handle ####
+XOR_SY	yvec7, yvec7, yvec7;
+#if  defined(RN) || defined(RT) || defined(CN) || defined(CT)
+ADDSUB_SY	yvec15, yvec7, yvec15;
+ADDSUB_SY	yvec14, yvec7, yvec14;
+#elif defined(NR) || defined(NC) || defined(TR) || defined(TC)
+SUB_SY	yvec15, yvec7, yvec15;
+SUB_SY	yvec14, yvec7, yvec14;
+#elif defined(RR) || defined(RC) || defined(CR) || defined(CC)
+VPERMILP_SY $0xb1, yvec15, yvec15;
+VPERMILP_SY $0xb1, yvec14, yvec14;
+ADDSUB_SY	yvec15, yvec7, yvec15;
+ADDSUB_SY	yvec14, yvec7, yvec14;
+VPERMILP_SY $0xb1, yvec15, yvec15;
+VPERMILP_SY $0xb1, yvec14, yvec14;
+#endif
+##### Load Alpha ####
+BROAD_SY MEMALPHA_R,yvec7;
+BROAD_SY MEMALPHA_I,yvec6;
+##### Multiply Alpha ####
+VPERMILP_SY $0xb1,yvec15, yvec5;
+MUL_SY	yvec15, yvec7, yvec15;
+MUL_SY	yvec5, yvec6, yvec5;
+ADDSUB_SY	yvec5, yvec15, yvec15;
+VPERMILP_SY $0xb1,yvec14, yvec4;
+MUL_SY	yvec14, yvec7, yvec14;
+MUL_SY	yvec4, yvec6, yvec4;
+ADDSUB_SY	yvec4, yvec14, yvec14;
+#### Writing back ####
+EXTRA_SY $1, yvec15, xvec7;
+EXTRA_SY $1, yvec14, xvec6;
+#ifndef	TRMMKERNEL
+LDL_SX	0*SIZE(C0), xvec0;
+LDH_SX	2*SIZE(C0), xvec0;
+LDL_SX	4*SIZE(C0), xvec1;
+LDH_SX	6*SIZE(C0), xvec1;
+LDL_SX	8*SIZE(C0), xvec2;
+LDH_SX	10*SIZE(C0), xvec2;
+LDL_SX	12*SIZE(C0), xvec3;
+LDH_SX	14*SIZE(C0), xvec3;
+ADD_SX	xvec0, xvec15;
+ADD_SX	xvec1, xvec7;
+ADD_SX	xvec2, xvec14;
+ADD_SX	xvec3, xvec6;
+#endif
+STL_SX	xvec15, 0*SIZE(C0);
+STH_SX	xvec15, 2*SIZE(C0);
+STL_SX	xvec7, 4*SIZE(C0);
+STH_SX	xvec7, 6*SIZE(C0);
+STL_SX	xvec14, 8*SIZE(C0);
+STH_SX	xvec14,	10*SIZE(C0);
+STL_SX	xvec6, 12*SIZE(C0);
+STH_SX	xvec6, 14*SIZE(C0);
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA))||(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kkk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+LEAQ (ptrba, %rax, 8), ptrba;
+ADDQ %rax, ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT) 
+ADDQ	$8, kk;
+#endif
+ADDQ	$16*SIZE, C0;
+DECQ	i;
+JG .L31_bodyB;
+.align 32
+.L31_loopE:
+TEST $4, bm;
+JLE .L32_loopE;
+.align 32
+.L32_bodyB:
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb,ptrbb;
+#else
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+LEAQ (ptrba, %rax, 4), ptrba;
+ADDQ %rax, ptrbb;
+#endif
+XOR_SY 	yvec15, yvec15, yvec15;
+#ifndef	TRMMKERNEL
+MOVQ bk,k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else
+MOVQ kk, %rax;
+#ifdef	LEFT
+ADDQ $4, %rax;
+#else 
+ADDQ $1, %rax;
+#endif
+MOVQ %rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L321_loopE;
+.align 32
+.L321_bodyB:
+#### Unroll 1 ####
+LD_SY	0*SIZE(ptrba), yvec0;
+BROAD_SY 0*SIZE(ptrbb), yvec2;
+MUL_SY	yvec0, yvec2, yvec6;
+ADD1_SY	yvec6, yvec15, yvec15;
+
+BROAD_SY 1*SIZE(ptrbb), yvec3;
+VPERMILP_SY $0xb1, yvec0, yvec4;
+MUL_SY	yvec4, yvec3, yvec6;
+ADD2_SY yvec6, yvec15, yvec15;
+
+#### Unroll 2 ####
+LD_SY	8*SIZE(ptrba), yvec0;
+BROAD_SY 2*SIZE(ptrbb), yvec2;
+MUL_SY	yvec0, yvec2, yvec6;
+ADD1_SY	yvec6, yvec15, yvec15;
+
+BROAD_SY 3*SIZE(ptrbb), yvec3;
+VPERMILP_SY $0xb1, yvec0, yvec4;
+MUL_SY	yvec4, yvec3, yvec6;
+ADD2_SY yvec6, yvec15, yvec15;
+
+#### Unroll 3 ####
+LD_SY	16*SIZE(ptrba), yvec0;
+BROAD_SY 4*SIZE(ptrbb), yvec2;
+MUL_SY	yvec0, yvec2, yvec6;
+ADD1_SY	yvec6, yvec15, yvec15;
+
+BROAD_SY 5*SIZE(ptrbb), yvec3;
+VPERMILP_SY $0xb1, yvec0, yvec4;
+MUL_SY	yvec4, yvec3, yvec6;
+ADD2_SY yvec6, yvec15, yvec15;
+
+#### Unroll 4 ####
+LD_SY	24*SIZE(ptrba), yvec0;
+BROAD_SY 6*SIZE(ptrbb), yvec2;
+MUL_SY	yvec0, yvec2, yvec6;
+ADD1_SY	yvec6, yvec15, yvec15;
+
+BROAD_SY 7*SIZE(ptrbb), yvec3;
+VPERMILP_SY $0xb1, yvec0, yvec4;
+MUL_SY	yvec4, yvec3, yvec6;
+ADD2_SY yvec6, yvec15, yvec15;
+ADDQ	$32*SIZE, ptrba;
+ADDQ	$8*SIZE, ptrbb;
+DECQ	k;
+JG .L321_bodyB;
+.align 32
+.L321_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else
+TEST $2, kkk;
+#endif
+JLE .L322_loopE;
+.align 32
+.L322_bodyB:
+#### Unroll 1 ####
+LD_SY	0*SIZE(ptrba), yvec0;
+BROAD_SY 0*SIZE(ptrbb), yvec2;
+MUL_SY	yvec0, yvec2, yvec6;
+ADD1_SY	yvec6, yvec15, yvec15;
+
+BROAD_SY 1*SIZE(ptrbb), yvec3;
+VPERMILP_SY $0xb1, yvec0, yvec4;
+MUL_SY	yvec4, yvec3, yvec6;
+ADD2_SY yvec6, yvec15, yvec15;
+
+#### Unroll 2 ####
+LD_SY	8*SIZE(ptrba), yvec0;
+BROAD_SY 2*SIZE(ptrbb), yvec2;
+MUL_SY	yvec0, yvec2, yvec6;
+ADD1_SY	yvec6, yvec15, yvec15;
+
+BROAD_SY 3*SIZE(ptrbb), yvec3;
+VPERMILP_SY $0xb1, yvec0, yvec4;
+MUL_SY	yvec4, yvec3, yvec6;
+ADD2_SY yvec6, yvec15, yvec15;
+ADDQ	$16*SIZE, ptrba;
+ADDQ	$4*SIZE, ptrbb;
+
+.L322_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else
+TEST $1, kkk;
+#endif
+JLE	.L323_loopE;
+.align 32
+.L323_bodyB:
+#### Unroll 1 ####
+LD_SY	0*SIZE(ptrba), yvec0;
+BROAD_SY 0*SIZE(ptrbb), yvec2;
+MUL_SY	yvec0, yvec2, yvec6;
+ADD1_SY	yvec6, yvec15, yvec15;
+
+BROAD_SY 1*SIZE(ptrbb), yvec3;
+VPERMILP_SY $0xb1, yvec0, yvec4;
+MUL_SY	yvec4, yvec3, yvec6;
+ADD2_SY yvec6, yvec15, yvec15;
+ADDQ	$8*SIZE, ptrba;
+ADDQ	$2*SIZE, ptrbb;
+
+.L323_loopE:
+#### Handle ####
+XOR_SY	yvec7, yvec7, yvec7;
+#if  defined(RN) || defined(RT) || defined(CN) || defined(CT)
+ADDSUB_SY	yvec15, yvec7, yvec15;
+#elif defined(NR) || defined(NC) || defined(TR) || defined(TC)
+SUB_SY	yvec15, yvec7, yvec15;
+#elif defined(RR) || defined(RC) || defined(CR) || defined(CC)
+VPERMILP_SY $0xb1, yvec15, yvec15;
+ADDSUB_SY	yvec15, yvec7, yvec15;
+VPERMILP_SY $0xb1, yvec15, yvec15;
+#endif
+##### Load Alpha ####
+BROAD_SY MEMALPHA_R,yvec7;
+BROAD_SY MEMALPHA_I,yvec6;
+##### Multiply Alpha ####
+VPERMILP_SY $0xb1,yvec15, yvec5;
+MUL_SY	yvec15, yvec7, yvec15;
+MUL_SY	yvec5, yvec6, yvec5;
+ADDSUB_SY	yvec5, yvec15, yvec15;
+#### Writing back ####
+EXTRA_SY $1, yvec15, xvec7;
+#ifndef	TRMMKERNEL
+LDL_SX	0*SIZE(C0), xvec0;
+LDH_SX	2*SIZE(C0), xvec0;
+LDL_SX	4*SIZE(C0), xvec1;
+LDH_SX	6*SIZE(C0), xvec1;
+ADD_SX	xvec0, xvec15;
+ADD_SX	xvec1, xvec7;
+#endif
+STL_SX	xvec15, 0*SIZE(C0);
+STH_SX	xvec15, 2*SIZE(C0);
+STL_SX	xvec7, 4*SIZE(C0);
+STH_SX	xvec7, 6*SIZE(C0);
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA))||(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kkk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+LEAQ (ptrba, %rax, 4), ptrba;
+ADDQ %rax, ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT) 
+ADDQ	$4, kk;
+#endif
+ADDQ	$8*SIZE, C0;
+
+.L32_loopE:
+TEST $2, bm;
+JLE .L33_loopE;
+.align 32
+.L33_bodyB:
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb,ptrbb;
+#else
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+LEAQ (ptrba, %rax, 2), ptrba;
+ADDQ %rax, ptrbb;
+#endif
+XOR_SY	yvec15, yvec15, yvec15;
+#ifndef	TRMMKERNEL
+MOVQ bk,k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else
+MOVQ kk, %rax;
+#ifdef	LEFT
+ADDQ $2, %rax;
+#else 
+ADDQ $1, %rax;
+#endif
+MOVQ %rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L331_loopE;
+.align 32
+.L331_bodyB:
+#### Unroll 1 ####
+LD_SX	0*SIZE(ptrba), xvec0;
+BROAD_SX 0*SIZE(ptrbb), xvec2;
+MUL_SX	xvec0, xvec2;
+ADD1_SX	xvec2, xvec15;
+SHUF_SX	$0xb1, xvec0, xvec1;
+BROAD_SX 1*SIZE(ptrbb), xvec3;
+MUL_SX	xvec1, xvec3;
+ADD2_SX xvec3, xvec15;
+
+#### Unroll 2 ####
+LD_SX	4*SIZE(ptrba), xvec0;
+BROAD_SX 2*SIZE(ptrbb), xvec2;
+MUL_SX	xvec0, xvec2;
+ADD1_SX	xvec2, xvec15;
+SHUF_SX	$0xb1, xvec0, xvec1;
+BROAD_SX 3*SIZE(ptrbb), xvec3;
+MUL_SX	xvec1, xvec3;
+ADD2_SX xvec3, xvec15;
+
+#### Unroll 3 ####
+LD_SX	8*SIZE(ptrba), xvec0;
+BROAD_SX 4*SIZE(ptrbb), xvec2;
+MUL_SX	xvec0, xvec2;
+ADD1_SX	xvec2, xvec15;
+SHUF_SX	$0xb1, xvec0, xvec1;
+BROAD_SX 5*SIZE(ptrbb), xvec3;
+MUL_SX	xvec1, xvec3;
+ADD2_SX xvec3, xvec15;
+
+#### Unroll 4 ####
+LD_SX	12*SIZE(ptrba), xvec0;
+BROAD_SX 6*SIZE(ptrbb), xvec2;
+MUL_SX	xvec0, xvec2;
+ADD1_SX	xvec2, xvec15;
+SHUF_SX	$0xb1, xvec0, xvec1;
+BROAD_SX 7*SIZE(ptrbb), xvec3;
+MUL_SX	xvec1, xvec3;
+ADD2_SX xvec3, xvec15;
+ADDQ	$16*SIZE, ptrba;
+ADDQ	$8*SIZE, ptrbb;
+DECQ	k;
+JG 	.L331_bodyB;
+.align 32
+.L331_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else
+TEST $2, kkk;
+#endif
+JLE .L332_loopE;
+.align 32
+.L332_bodyB:
+#### Unroll 1 ####
+LD_SX	0*SIZE(ptrba), xvec0;
+BROAD_SX 0*SIZE(ptrbb), xvec2;
+MUL_SX	xvec0, xvec2;
+ADD1_SX	xvec2, xvec15;
+SHUF_SX	$0xb1, xvec0, xvec1;
+BROAD_SX 1*SIZE(ptrbb), xvec3;
+MUL_SX	xvec1, xvec3;
+ADD2_SX xvec3, xvec15;
+
+#### Unroll 2 ####
+LD_SX	4*SIZE(ptrba), xvec0;
+BROAD_SX 2*SIZE(ptrbb), xvec2;
+MUL_SX	xvec0, xvec2;
+ADD1_SX	xvec2, xvec15;
+SHUF_SX	$0xb1, xvec0, xvec1;
+BROAD_SX 3*SIZE(ptrbb), xvec3;
+MUL_SX	xvec1, xvec3;
+ADD2_SX xvec3, xvec15;
+ADDQ	$8*SIZE, ptrba;
+ADDQ	$4*SIZE, ptrbb;
+
+.L332_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else
+TEST $1, kkk;
+#endif
+JLE .L333_loopE;
+.align 32
+.L333_bodyB:
+#### Unroll 1 ####
+LD_SX	0*SIZE(ptrba), xvec0;
+BROAD_SX 0*SIZE(ptrbb), xvec2;
+MUL_SX	xvec0, xvec2;
+ADD1_SX	xvec2, xvec15;
+SHUF_SX	$0xb1, xvec0, xvec1;
+BROAD_SX 1*SIZE(ptrbb), xvec3;
+MUL_SX	xvec1, xvec3;
+ADD2_SX xvec3, xvec15;
+ADDQ	$4*SIZE, ptrba;
+ADDQ	$2*SIZE, ptrbb;
+
+.L333_loopE:
+#### Handle ####
+XOR_SY	yvec7, yvec7, yvec7;
+#if  defined(RN) || defined(RT) || defined(CN) || defined(CT)
+ADDSUB_SX	xvec15, xvec7;
+MOV_SX		xvec7, xvec15;
+#elif defined(NR) || defined(NC) || defined(TR) || defined(TC)
+SUB_SX	xvec15, xvec7;
+MOV_SX	xvec7, xvec15;
+#elif defined(RR) || defined(RC) || defined(CR) || defined(CC)
+SHUF_SX $0xb1, xvec15, xvec15;
+ADDSUB_SX	xvec15, xvec7;
+MOV_SX		xvec7, xvec15;
+SHUF_SX $0xb1, xvec15, xvec15;
+#endif
+#### Mulitply Alpha ####
+BROAD_SX MEMALPHA_R, xvec7;
+BROAD_SX MEMALPHA_I, xvec6;
+#### Writng back ####
+VPERMILP_SX $0xb1,xvec15, xvec5;
+MUL_SX	xvec7, xvec15;
+MUL_SX	xvec6, xvec5;
+ADDSUB_SX	xvec5, xvec15;
+#ifndef	TRMMKERNEL
+LDL_SX	0*SIZE(C0), xvec0;
+LDH_SX	2*SIZE(C0), xvec0;
+ADD_SX	xvec0, xvec15;
+#endif
+STL_SX	xvec15, 0*SIZE(C0);
+STH_SX	xvec15, 2*SIZE(C0);
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA))||(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kkk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+LEAQ (ptrba, %rax, 2), ptrba;
+ADDQ %rax, ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT) 
+ADDQ	$2, kk;
+#endif
+ADDQ	$4*SIZE, C0;
+
+.L33_loopE:
+TEST $1, bm;
+JLE .L34_loopE;
+.align 32
+.L34_bodyB:
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb,ptrbb;
+#else
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+ADDQ %rax, ptrba;
+ADDQ %rax, ptrbb;
+#endif
+XOR_SY	yvec15, yvec15, yvec15;
+#ifndef	TRMMKERNEL
+MOVQ bk,k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else
+MOVQ kk, %rax;
+#ifdef	LEFT
+ADDQ $1, %rax;
+#else 
+ADDQ $1, %rax;
+#endif
+MOVQ %rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L341_loopE;
+.align 32
+.L341_bodyB:
+LD_SX	0*SIZE(ptrba), xvec0;
+LD_SX	0*SIZE(ptrbb), xvec2;
+SHUF_SX	$0xa0, xvec2, xvec3;
+MUL_SX	xvec0, xvec3;
+ADD1_SX	xvec3, xvec15;
+SHUF_SX	$0xb1, xvec0, xvec1;
+SHUF_SX	$0xf5, xvec2, xvec4;
+MUL_SX	xvec1, xvec4;
+ADD2_SX xvec4, xvec15;
+
+LD_SX	4*SIZE(ptrba), xvec0;
+LD_SX	4*SIZE(ptrbb), xvec2;
+SHUF_SX	$0xa0, xvec2, xvec3;
+MUL_SX	xvec0, xvec3;
+ADD1_SX	xvec3, xvec15;
+SHUF_SX	$0xb1, xvec0, xvec1;
+SHUF_SX	$0xf5, xvec2, xvec4;
+MUL_SX	xvec1, xvec4;
+ADD2_SX xvec4, xvec15;
+ADDQ	$8*SIZE, ptrba;
+ADDQ	$8*SIZE, ptrbb;
+DECQ	k;
+JG .L341_bodyB;
+.align 32
+.L341_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else
+TEST $2, kkk;
+#endif
+JLE .L342_loopE;
+.align 32
+.L342_bodyB:
+LD_SX	0*SIZE(ptrba), xvec0;
+LD_SX	0*SIZE(ptrbb), xvec2;
+SHUF_SX	$0xa0, xvec2, xvec3;
+MUL_SX	xvec0, xvec3;
+ADD1_SX	xvec3, xvec15;
+SHUF_SX	$0xb1, xvec0, xvec1;
+SHUF_SX	$0xf5, xvec2, xvec4;
+MUL_SX	xvec1, xvec4;
+ADD2_SX xvec4, xvec15;
+ADDQ	$4*SIZE, ptrba;
+ADDQ	$4*SIZE, ptrbb;
+
+.L342_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else
+TEST $1, kkk;
+#endif
+JLE .L343_loopE;
+.align 32
+.L343_bodyB:
+XOR_SY	yvec0, yvec0, yvec0;
+XOR_SY	yvec2, yvec2, yvec2;
+LDL_SX	0*SIZE(ptrba), xvec0;
+LDL_SX	0*SIZE(ptrbb), xvec2;
+SHUF_SX	$0xe0, xvec2, xvec3;
+MUL_SX	xvec0, xvec3;
+ADD1_SX	xvec3, xvec15;
+SHUF_SX	$0xe1, xvec0, xvec1;
+SHUF_SX	$0xe5, xvec2, xvec4;
+MUL_SX	xvec1, xvec4;
+ADD2_SX	xvec4, xvec15;
+ADDQ	$2*SIZE, ptrba;
+ADDQ	$2*SIZE, ptrbb;
+
+.L343_loopE:
+#### Handle ####
+XOR_SY	yvec7, yvec7, yvec7;
+#if  defined(RN) || defined(RT) || defined(CN) || defined(CT)
+ADDSUB_SX	xvec15, xvec7;
+MOV_SX		xvec7, xvec15;
+#elif defined(NR) || defined(NC) || defined(TR) || defined(TC)
+SUB_SX	xvec15, xvec7;
+MOV_SX	xvec7, xvec15;
+#elif defined(RR) || defined(RC) || defined(CR) || defined(CC)
+SHUF_SX $0xb1, xvec15, xvec15;
+ADDSUB_SX	xvec15, xvec7;
+MOV_SX		xvec7, xvec15;
+SHUF_SX $0xb1, xvec15, xvec15;
+#endif
+BROAD_SX MEMALPHA_R, xvec7;
+BROAD_SX MEMALPHA_I, xvec6;
+VPERMILP_SX	$0xb1, xvec15, xvec5;
+MUL_SX	xvec7, xvec15;
+MUL_SX	xvec6, xvec5;
+ADDSUB_SX	xvec5, xvec15;
+SHUF_SX	$0x44, xvec15, xvec14;
+SHUF_SX	$0xee, xvec15, xvec13;
+ADD_SX	xvec13, xvec14;
+#ifndef	TRMMKERNEL
+LDL_SX	0*SIZE(C0), xvec0;
+ADD_SX	xvec0, xvec14;
+#endif
+STL_SX	xvec14, 0*SIZE(C0);
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA))||(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kkk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+ADDQ %rax, ptrba;
+ADDQ %rax, ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT) 
+ADDQ	$1, kk;
+#endif
+ADDQ	$2*SIZE, C0;
+
+.L34_loopE:
+#if defined(TRMMKERNEL) && !defined(LEFT)
+ADDQ	$1, kk;
+#endif
+
+MOVQ bk, k;
+SALQ $3, k;
+ADDQ k, bb;
+ADDQ ldc, C;
+.L30_loopE:
+movq      0(%rsp), %rbx;
+movq      8(%rsp), %rbp;
+movq     16(%rsp), %r12;
+movq     24(%rsp), %r13;
+movq     32(%rsp), %r14;
+movq     40(%rsp), %r15;
+
+#ifdef WINDOWS_ABI
+	movq	 48(%rsp), %rdi
+	movq	 56(%rsp), %rsi
+	movups	 64(%rsp), %xmm6
+	movups	 80(%rsp), %xmm7
+	movups	 96(%rsp), %xmm8
+	movups	112(%rsp), %xmm9
+	movups	128(%rsp), %xmm10
+	movups	144(%rsp), %xmm11
+	movups	160(%rsp), %xmm12
+	movups	176(%rsp), %xmm13
+	movups	192(%rsp), %xmm14
+	movups	208(%rsp), %xmm15
+#endif
+
+addq    $STACKSIZE, %rsp;
+ret
+
+EPILOGUE
diff --git a/kernel/x86_64/dgemm_kernel_4x8_sandy.S b/kernel/x86_64/dgemm_kernel_4x8_sandy.S
new file mode 100644
index 000000000..fea5ecb4a
--- /dev/null
+++ b/kernel/x86_64/dgemm_kernel_4x8_sandy.S
@@ -0,0 +1,3186 @@
+/*****************************************************************************
+  Copyright (c) 2011,2012 Lab of Parallel Software and Computational Science,ISCAS
+  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 ISCAS 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 COPYRIGHT OWNER 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 old_bm %rdi
+#define old_bn %rsi
+#define old_bk %rdx
+
+#define bm %r13
+#define bn %r14
+#define bk %r15
+
+#define ALPHA %xmm0
+#define ba %rcx
+#define bb %r8
+#define C %r9
+#define ldc %r10
+
+#define i %r11
+#define k %rax
+
+#define ptrba %rdi
+#define ptrbb %rsi
+#define C0 %rbx
+#define C1 %rbp
+
+#define prebb %r12
+
+#ifndef WINDOWS_ABI
+
+#define STACKSIZE 128
+
+#define old_ldc 8+STACKSIZE(%rsp)
+#define old_offset 16+STACKSIZE(%rsp)
+#define MEMALPHA 48(%rsp)
+#define j 56(%rsp)
+#define OFFSET 64(%rsp)
+#define kk 72(%rsp)
+#define kkk 80(%rsp)
+
+#else
+
+#define STACKSIZE 512
+
+#define OLD_A		40 + STACKSIZE(%rsp)
+#define OLD_B		48 + STACKSIZE(%rsp)
+#define OLD_C		56 + STACKSIZE(%rsp)
+#define old_ldc		64 + STACKSIZE(%rsp)
+#define old_offset	72 + STACKSIZE(%rsp)
+
+#define MEMALPHA	  224(%rsp)
+#define j	  232(%rsp)
+#define OFFSET	  240(%rsp)
+#define kk	  248(%rsp)
+#define kkk	  256(%rsp)
+
+#endif
+
+#define PREFETCH0 prefetcht0
+#define PREFETCH1 prefetcht0
+#define PREFETCH2 prefetcht2
+
+#define xvec0	%xmm0
+#define xvec1	%xmm1
+#define xvec2	%xmm2
+#define xvec3	%xmm3
+#define xvec4	%xmm4
+#define xvec5	%xmm5
+#define xvec6	%xmm6
+#define xvec7	%xmm7
+#define xvec8	%xmm8
+#define xvec9	%xmm9
+#define xvec10	%xmm10
+#define xvec11	%xmm11
+#define xvec12	%xmm12
+#define xvec13	%xmm13
+#define xvec14	%xmm14
+#define xvec15	%xmm15
+
+#define yvec0	%ymm0
+#define yvec1	%ymm1
+#define yvec2	%ymm2
+#define yvec3	%ymm3
+#define yvec4	%ymm4
+#define yvec5	%ymm5
+#define yvec6	%ymm6
+#define yvec7	%ymm7
+#define yvec8	%ymm8
+#define yvec9	%ymm9
+#define yvec10	%ymm10
+#define yvec11	%ymm11
+#define yvec12	%ymm12
+#define yvec13	%ymm13
+#define yvec14	%ymm14
+#define yvec15	%ymm15
+
+#define LEAQ	leaq
+#define ADDQ	addq
+#define MULQ	imulq
+#define SARQ	sarq
+#define SALQ	salq
+#define ANDQ	andq
+#define SUBQ	subq
+#define DECQ	decq
+#define JG		jg
+#define JLE		jle
+#define TEST	testq
+#define OR		orq
+#define JNE		jne
+#define NOP		
+#define XOR		xorpd
+#define MOVQ	movq 
+
+#define XOR_SY	vxorps
+#define XOR_DY	vxorpd
+#define XOR_SX	xorps
+#define XOR_DX	xorpd
+
+#define LD_SY	vmovaps
+#define LD_DY	vmovapd
+#define LD_SX	movaps
+#define LD_DX	movapd
+#define LDL_DX	movlpd
+#define LDL_DY	vmovlpd
+#define LDH_DX	movhpd
+#define LDH_DY	vmovhpd
+
+#define ST_SY	vmovaps
+#define ST_DY	vmovapd
+#define ST_SX	movaps
+#define ST_DX	movapd
+#define STL_DX	movlpd
+#define STH_DX	movhpd
+
+#define EDUP_SY	vmovsldup
+#define ODUP_SY	vmovshdup
+#define EDUP_DY	vmovddup
+
+#define ADD_SY	vaddps 
+#define ADD_DY	vaddpd
+#define ADD_SX	addps
+#define ADD_DX	addpd
+
+#define ADD1_DY	vaddpd
+#define ADD2_DY	vaddpd
+#define ADDSUB_DY	vaddsubpd
+#define ADDSUB_SY	vaddsubps
+
+#define MUL_SY	vmulps
+#define MUL_DY	vmulpd
+#define MUL_SX	mulps
+#define MUL_DX	mulpd
+
+#define SHUF_SY	vperm2f128
+#define SHUF_DY	vperm2f128
+#define SHUF_DX	pshufd
+
+#define VPERMILP_SY	vpermilps
+#define VPERMILP_DY	vpermilpd
+
+#define BROAD_SY vbroadcastss
+#define BROAD_DY vbroadcastsd
+#define BROAD_SX 
+#define BROAD_DX movddup
+
+#define MOV_SY vmovaps
+#define MOV_DY vmovapd
+#define MOV_SX movaps
+#define MOV_DX movapd
+
+#define REVS_SY	vshufps
+#define REVS_DY vshufpd
+#define REVS_SX shufps
+#define REVS_DX	movsd
+
+#define EXTRA_SY vextractf128
+#define EXTRA_DY vextractf128
+
+PROLOGUE
+
+subq    $STACKSIZE, %rsp;
+movq    %rbx,  0(%rsp);
+movq    %rbp,  8(%rsp);
+movq    %r12, 16(%rsp);
+movq    %r13, 24(%rsp);
+movq    %r14, 32(%rsp);
+movq    %r15, 40(%rsp);
+
+#ifdef WINDOWS_ABI
+	movq	%rdi,    48(%rsp)
+	movq	%rsi,    56(%rsp)
+	movups	%xmm6,   64(%rsp)
+	movups	%xmm7,   80(%rsp)
+	movups	%xmm8,   96(%rsp)
+	movups	%xmm9,  112(%rsp)
+	movups	%xmm10, 128(%rsp)
+	movups	%xmm11, 144(%rsp)
+	movups	%xmm12, 160(%rsp)
+	movups	%xmm13, 176(%rsp)
+	movups	%xmm14, 192(%rsp)
+	movups	%xmm15, 208(%rsp)
+
+	movq	ARG1,      old_bm
+	movq	ARG2,      old_bn
+	movq	ARG3,      old_bk
+	movq	OLD_A,     ba 
+	movq	OLD_B,     bb
+	movq	OLD_C,     C
+	movq	old_ldc,   ldc
+#ifdef TRMMKERNEL
+	movq	old_offset, %r11
+#endif
+#else
+
+movq	old_ldc, ldc
+#ifdef	TRMMKERNEL
+movq	old_offset, %r11
+#endif
+#endif
+
+vmovlps	ALPHA, MEMALPHA
+movq	old_bm,	bm
+movq	old_bn, bn
+movq	old_bk, bk
+leaq	(, ldc, SIZE), ldc
+#ifdef	TRMMKERNEL
+movq	%r11, OFFSET
+#ifndef	LEFT
+negq	%r11;
+#endif
+movq	%r11, kk
+#endif
+
+MOVQ bn,j;
+SARQ $2,j;							# Rn = 4
+JLE .L0_loopE;
+.align 32;
+.L0_bodyB:;
+#if defined(TRMMKERNEL) && defined(LEFT)
+MOVQ OFFSET, %rax;
+MOVQ %rax, kk;
+#endif
+
+MOVQ C,C0;
+LEAQ (C,ldc,2),C1;
+MOVQ bk, k;
+SALQ $5, k;
+LEAQ (bb, k, 1), prebb;
+MOVQ ba,ptrba;
+MOVQ bm,i;
+SARQ $3,i;							# Rm = 8
+JLE .L1_loopE;
+.align 32;
+.L1_bodyB:;
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb, ptrbb;
+#else 
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+LEAQ (, %rax, SIZE), %rax;
+LEAQ (ptrba, %rax, 8), ptrba;
+LEAQ (ptrbb, %rax, 4), ptrbb;
+#endif
+#### Initial Results Register ####
+PREFETCH2	0*SIZE(prebb);
+XOR_DY yvec15, yvec15, yvec15;
+PREFETCH2	8*SIZE(prebb);
+XOR_DY yvec14, yvec14, yvec14;
+XOR_DY yvec13, yvec13, yvec13;
+ADDQ	$16*SIZE, prebb
+XOR_DY yvec12, yvec12, yvec12;
+PREFETCH0	3*SIZE(C0)
+LD_DY	0*SIZE(ptrbb), yvec2;
+PREFETCH0	3*SIZE(C0, ldc, 1)
+XOR_DY yvec11, yvec11, yvec11;
+PREFETCH0	3*SIZE(C1)
+XOR_DY yvec10, yvec10, yvec10;
+PREFETCH0	3*SIZE(C1, ldc, 1)
+LD_DY	0*SIZE(ptrba), yvec0;
+XOR_DY yvec9, yvec9, yvec9;
+XOR_DY yvec8, yvec8, yvec8;
+VPERMILP_DY	$0x05, yvec2, yvec3;
+#ifndef	TRMMKERNEL
+MOVQ bk,k;
+#elif (defined(LEFT) && !defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else
+MOVQ kk, %rax;
+#ifdef LEFT
+ADDQ $8, %rax;
+#else 
+ADDQ $4, %rax;
+#endif
+MOVQ %rax, kkk;
+#endif
+SARQ $2,k;
+JLE .L2_loopE;
+.align 32;
+.L2_bodyB:;
+# Computing kernel
+
+#### Unroll times 1 #### 
+LD_DY	4*SIZE(ptrba), yvec1;
+MUL_DY	yvec0, yvec2, yvec6;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+MUL_DY	yvec0, yvec3, yvec7;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+ADD_DY	yvec15, yvec6, yvec15;
+ADD_DY	yvec13, yvec7, yvec13;
+
+PREFETCH0	64*SIZE(ptrba)
+MUL_DY	yvec1, yvec2, yvec6;
+LD_DY	4*SIZE(ptrbb), yvec2;
+MUL_DY	yvec1, yvec3, yvec7;
+VPERMILP_DY	$0x05, yvec2, yvec3;
+ADD_DY	yvec14, yvec6, yvec14;
+ADD_DY	yvec12, yvec7, yvec12;
+
+MUL_DY	yvec0, yvec4, yvec6;
+MUL_DY	yvec0, yvec5, yvec7;
+LD_DY	8*SIZE(ptrba), yvec0;
+ADD_DY	yvec11, yvec6, yvec11;
+ADD_DY	yvec9, yvec7, yvec9;
+
+MUL_DY	yvec1, yvec4, yvec6;
+MUL_DY	yvec1, yvec5, yvec7;
+ADD_DY	yvec10, yvec6, yvec10;
+ADD_DY	yvec8, yvec7, yvec8;
+
+#### Unroll times 2 ####
+LD_DY	12*SIZE(ptrba), yvec1;
+MUL_DY	yvec0, yvec2, yvec6;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+MUL_DY	yvec0, yvec3, yvec7;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+ADD_DY	yvec15, yvec6, yvec15;
+ADD_DY	yvec13, yvec7, yvec13;
+
+PREFETCH0	72*SIZE(ptrba)
+MUL_DY	yvec1, yvec2, yvec6;
+LD_DY	8*SIZE(ptrbb), yvec2;
+MUL_DY	yvec1, yvec3, yvec7;
+VPERMILP_DY	$0x05, yvec2, yvec3;
+ADD_DY	yvec14, yvec6, yvec14;
+ADD_DY	yvec12, yvec7, yvec12;
+
+MUL_DY	yvec0, yvec4, yvec6;
+MUL_DY	yvec0, yvec5, yvec7;
+LD_DY	16*SIZE(ptrba), yvec0;
+ADD_DY	yvec11, yvec6, yvec11;
+ADD_DY	yvec9, yvec7, yvec9;
+
+MUL_DY	yvec1, yvec4, yvec6;
+MUL_DY	yvec1, yvec5, yvec7;
+ADD_DY	yvec10, yvec6, yvec10;
+ADD_DY	yvec8, yvec7, yvec8;
+
+#### Unroll times 3 #### 
+LD_DY	20*SIZE(ptrba), yvec1;
+MUL_DY	yvec0, yvec2, yvec6;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+MUL_DY	yvec0, yvec3, yvec7;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+ADD_DY	yvec15, yvec6, yvec15;
+ADD_DY	yvec13, yvec7, yvec13;
+
+PREFETCH0	80*SIZE(ptrba)
+MUL_DY	yvec1, yvec2, yvec6;
+LD_DY	12*SIZE(ptrbb), yvec2;
+ADDQ	$16*SIZE, ptrbb;
+MUL_DY	yvec1, yvec3, yvec7;
+VPERMILP_DY	$0x05, yvec2, yvec3;
+ADD_DY	yvec14, yvec6, yvec14;
+ADD_DY	yvec12, yvec7, yvec12;
+
+MUL_DY	yvec0, yvec4, yvec6;
+MUL_DY	yvec0, yvec5, yvec7;
+LD_DY	24*SIZE(ptrba), yvec0;
+ADD_DY	yvec11, yvec6, yvec11;
+ADD_DY	yvec9, yvec7, yvec9;
+
+MUL_DY	yvec1, yvec4, yvec6;
+MUL_DY	yvec1, yvec5, yvec7;
+ADD_DY	yvec10, yvec6, yvec10;
+ADD_DY	yvec8, yvec7, yvec8;
+
+#### Unroll times 4 ####
+LD_DY	28*SIZE(ptrba), yvec1;
+MUL_DY	yvec0, yvec2, yvec6;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+MUL_DY	yvec0, yvec3, yvec7;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+ADDQ	$32*SIZE, ptrba;
+ADD_DY	yvec15, yvec6, yvec15;
+ADD_DY	yvec13, yvec7, yvec13;
+
+PREFETCH0	88*SIZE(ptrba)
+MUL_DY	yvec1, yvec2, yvec6;
+LD_DY	0*SIZE(ptrbb), yvec2;
+MUL_DY	yvec1, yvec3, yvec7;
+VPERMILP_DY	$0x05, yvec2, yvec3;
+ADD_DY	yvec14, yvec6, yvec14;
+ADD_DY	yvec12, yvec7, yvec12;
+
+MUL_DY	yvec0, yvec4, yvec6;
+MUL_DY	yvec0, yvec5, yvec7;
+LD_DY	0*SIZE(ptrba), yvec0;
+ADD_DY	yvec11, yvec6, yvec11;
+ADD_DY	yvec9, yvec7, yvec9;
+
+MUL_DY	yvec1, yvec4, yvec6;
+MUL_DY	yvec1, yvec5, yvec7;
+ADD_DY	yvec10, yvec6, yvec10;
+ADD_DY	yvec8, yvec7, yvec8;
+.L2_bodyE:;
+DECQ k;
+JG .L2_bodyB;
+.align	64;
+.L2_loopE:;
+PREFETCH2	0*SIZE(prebb);
+ADDQ	$8*SIZE, prebb;
+#ifndef	TRMMKERNEL
+TEST	$2, bk;
+#else 
+MOVQ	kkk, %rax;
+TEST 	$2, %rax;
+#endif
+JLE		.L3_loopE;
+.align	64
+.L3_bodyB:
+#### Unroll times 1 #### 
+PREFETCH0	64*SIZE(ptrba)
+LD_DY	4*SIZE(ptrba), yvec1;
+MUL_DY	yvec0, yvec2, yvec6;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+MUL_DY	yvec0, yvec3, yvec7;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+ADD_DY	yvec15, yvec6, yvec15;
+ADD_DY	yvec13, yvec7, yvec13;
+
+MUL_DY	yvec1, yvec2, yvec6;
+LD_DY	4*SIZE(ptrbb), yvec2;
+ADDQ	$8*SIZE, ptrbb;
+MUL_DY	yvec1, yvec3, yvec7;
+VPERMILP_DY	$0x05, yvec2, yvec3;
+ADD_DY	yvec14, yvec6, yvec14;
+ADD_DY	yvec12, yvec7, yvec12;
+
+MUL_DY	yvec0, yvec4, yvec6;
+MUL_DY	yvec0, yvec5, yvec7;
+LD_DY	8*SIZE(ptrba), yvec0;
+ADD_DY	yvec11, yvec6, yvec11;
+ADD_DY	yvec9, yvec7, yvec9;
+
+MUL_DY	yvec1, yvec4, yvec6;
+MUL_DY	yvec1, yvec5, yvec7;
+ADD_DY	yvec10, yvec6, yvec10;
+ADD_DY	yvec8, yvec7, yvec8;
+
+#### Unroll times 2 ####
+PREFETCH0	72*SIZE(ptrba)
+LD_DY	12*SIZE(ptrba), yvec1;
+MUL_DY	yvec0, yvec2, yvec6;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+MUL_DY	yvec0, yvec3, yvec7;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+ADDQ	$16*SIZE, ptrba;
+ADD_DY	yvec15, yvec6, yvec15;
+ADD_DY	yvec13, yvec7, yvec13;
+
+MUL_DY	yvec1, yvec2, yvec6;
+LD_DY	0*SIZE(ptrbb), yvec2;
+MUL_DY	yvec1, yvec3, yvec7;
+VPERMILP_DY	$0x05, yvec2, yvec3;
+ADD_DY	yvec14, yvec6, yvec14;
+ADD_DY	yvec12, yvec7, yvec12;
+
+MUL_DY	yvec0, yvec4, yvec6;
+MUL_DY	yvec0, yvec5, yvec7;
+LD_DY	0*SIZE(ptrba), yvec0;
+ADD_DY	yvec11, yvec6, yvec11;
+ADD_DY	yvec9, yvec7, yvec9;
+
+MUL_DY	yvec1, yvec4, yvec6;
+MUL_DY	yvec1, yvec5, yvec7;
+ADD_DY	yvec10, yvec6, yvec10;
+ADD_DY	yvec8, yvec7, yvec8;
+
+.L3_loopE:
+PREFETCH2	0*SIZE(prebb);
+ADDQ	$8*SIZE, prebb
+#ifndef	TRMMKERNEL
+TEST	$1, bk;
+#else 
+MOVQ	kkk, %rax;
+TEST 	$1, %rax;
+#endif
+JLE		.L4_loopE;
+.align	64
+.L4_bodyB:;
+#### Unroll times 1 #### 
+PREFETCH0	64*SIZE(ptrba)
+LD_DY	4*SIZE(ptrba), yvec1;
+MUL_DY	yvec0, yvec2, yvec6;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+MUL_DY	yvec0, yvec3, yvec7;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+ADDQ	$8*SIZE, ptrba;
+ADD_DY	yvec15, yvec6, yvec15;
+ADD_DY	yvec13, yvec7, yvec13;
+
+MUL_DY	yvec1, yvec2, yvec6;
+MUL_DY	yvec1, yvec3, yvec7;
+ADDQ	$4*SIZE, ptrbb;
+ADD_DY	yvec14, yvec6, yvec14;
+ADD_DY	yvec12, yvec7, yvec12;
+
+MUL_DY	yvec0, yvec4, yvec6;
+MUL_DY	yvec0, yvec5, yvec7;
+ADD_DY	yvec11, yvec6, yvec11;
+ADD_DY	yvec9, yvec7, yvec9;
+
+MUL_DY	yvec1, yvec4, yvec6;
+MUL_DY	yvec1, yvec5, yvec7;
+ADD_DY	yvec10, yvec6, yvec10;
+ADD_DY	yvec8, yvec7, yvec8;
+
+.L4_loopE:;
+#### Load Alpha ####
+BROAD_DY MEMALPHA,yvec7;
+#### Multiply Alpha ####
+MUL_DY yvec7,yvec15,yvec15;
+MUL_DY yvec7,yvec14,yvec14;
+MUL_DY yvec7,yvec13,yvec13;
+MUL_DY yvec7,yvec12,yvec12;
+MUL_DY yvec7,yvec11,yvec11;
+MUL_DY yvec7,yvec10,yvec10;
+MUL_DY yvec7,yvec9,yvec9;
+MUL_DY yvec7,yvec8,yvec8;
+#### Reverse the Results ####
+MOV_DY yvec15,yvec7;
+REVS_DY $0x0a,yvec13,yvec15,yvec15;
+REVS_DY $0x0a,yvec7,yvec13,yvec13;
+MOV_DY yvec14,yvec7;
+REVS_DY $0x0a,yvec12,yvec14,yvec14;
+REVS_DY $0x0a,yvec7,yvec12,yvec12;
+MOV_DY yvec11,yvec7;
+REVS_DY $0x0a,yvec9,yvec11,yvec11;
+REVS_DY $0x0a,yvec7,yvec9,yvec9;
+MOV_DY yvec10,yvec7;
+REVS_DY $0x0a,yvec8,yvec10,yvec10;
+REVS_DY $0x0a,yvec7,yvec8,yvec8;
+#### Testing alignment ####
+MOVQ	C0, %rax;
+OR		ldc, %rax;
+TEST	$15, %rax;
+JNE		.L4_loopEx;					# Unalign part write back
+.align 32
+#### Writing Back ####
+EXTRA_DY $1,yvec15,xvec7;
+EXTRA_DY $1,yvec14,xvec6;
+EXTRA_DY $1,yvec13,xvec5;
+EXTRA_DY $1,yvec12,xvec4;
+EXTRA_DY $1,yvec11,xvec3;
+EXTRA_DY $1,yvec10,xvec2;
+EXTRA_DY $1,yvec9,xvec1;
+EXTRA_DY $1,yvec8,xvec0;
+#ifndef	TRMMKERNEL
+ADD_DY 0*SIZE(C0),xvec15,xvec15;
+ADD_DY 2*SIZE(C1),xvec7,xvec7;
+ADD_DY 4*SIZE(C0),xvec14,xvec14;
+ADD_DY 6*SIZE(C1),xvec6,xvec6;
+ADD_DY 0*SIZE(C0,ldc,1),xvec13,xvec13;
+ADD_DY 2*SIZE(C1,ldc,1),xvec5,xvec5;
+ADD_DY 4*SIZE(C0,ldc,1),xvec12,xvec12;
+ADD_DY 6*SIZE(C1,ldc,1),xvec4,xvec4;
+ADD_DY 0*SIZE(C1),xvec11,xvec11;
+ADD_DY 2*SIZE(C0),xvec3,xvec3;
+ADD_DY 4*SIZE(C1),xvec10,xvec10;
+ADD_DY 6*SIZE(C0),xvec2,xvec2;
+ADD_DY 0*SIZE(C1,ldc,1),xvec9,xvec9;
+ADD_DY 2*SIZE(C0,ldc,1),xvec1,xvec1;
+ADD_DY 4*SIZE(C1,ldc,1),xvec8,xvec8;
+ADD_DY 6*SIZE(C0,ldc,1),xvec0,xvec0;
+#endif
+ST_DY xvec15, 0*SIZE(C0);
+ST_DY xvec7, 2*SIZE(C1);
+ST_DY xvec14, 4*SIZE(C0);
+ST_DY xvec6, 6*SIZE(C1);
+ST_DY xvec13, 0*SIZE(C0,ldc,1);
+ST_DY xvec5, 2*SIZE(C1,ldc,1);
+ST_DY xvec12, 4*SIZE(C0,ldc,1);
+ST_DY xvec4, 6*SIZE(C1,ldc,1);
+ST_DY xvec11, 0*SIZE(C1);
+ST_DY xvec3, 2*SIZE(C0);
+ST_DY xvec10, 4*SIZE(C1);
+ST_DY xvec2, 6*SIZE(C0);
+ST_DY xvec9, 0*SIZE(C1,ldc,1);
+ST_DY xvec1, 2*SIZE(C0,ldc,1);
+ST_DY xvec8, 4*SIZE(C1,ldc,1);
+ST_DY xvec0, 6*SIZE(C0,ldc,1);
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) ||(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kkk, %rax;
+LEAQ 	(, %rax, SIZE), %rax;
+LEAQ	(ptrba, %rax, 8), ptrba;
+LEAQ 	(ptrbb, %rax, 4), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT)
+ADDQ	$8, kk
+#endif
+ADDQ $8*SIZE,C0;
+ADDQ $8*SIZE,C1;
+.L1_bodyE:;
+DECQ i;
+JG .L1_bodyB;
+JMP	.L1_loopE;
+.align 32;
+.L4_loopEx:;
+EXTRA_DY $1, yvec15, xvec7;
+#ifndef	TRMMKERNEL
+LDL_DY	0*SIZE(C0), xvec6, xvec6;
+LDH_DY	1*SIZE(C0), xvec6, xvec6;
+ADD_DY	xvec6, xvec15, xvec15;
+LDL_DY	2*SIZE(C1), xvec5, xvec5;
+LDH_DY	3*SIZE(C1), xvec5, xvec5;
+ADD_DY	xvec5, xvec7, xvec7;
+#endif
+STL_DX	xvec15, 0*SIZE(C0);
+STH_DX	xvec15, 1*SIZE(C0);
+STL_DX	xvec7, 2*SIZE(C1);
+STH_DX	xvec7, 3*SIZE(C1);
+
+EXTRA_DY $1, yvec14, xvec4;
+#ifndef	TRMMKERNEL
+LDL_DY	4*SIZE(C0), xvec3, xvec3;
+LDH_DY	5*SIZE(C0), xvec3, xvec3;
+ADD_DY	xvec3, xvec14, xvec14;
+LDL_DY	6*SIZE(C1), xvec2, xvec2;
+LDH_DY	7*SIZE(C1), xvec2, xvec2;
+ADD_DY	xvec2, xvec4, xvec4;
+#endif
+STL_DX	xvec14, 4*SIZE(C0);
+STH_DX	xvec14, 5*SIZE(C0);
+STL_DX	xvec4, 6*SIZE(C1);
+STH_DX	xvec4, 7*SIZE(C1);
+
+EXTRA_DY $1, yvec13, xvec7;
+#ifndef	TRMMKERNEL
+LDL_DY	0*SIZE(C0, ldc, 1), xvec6, xvec6;
+LDH_DY	1*SIZE(C0, ldc, 1), xvec6, xvec6;
+ADD_DY	xvec6, xvec13, xvec13;
+LDL_DY	2*SIZE(C1, ldc, 1), xvec5, xvec5;
+LDH_DY	3*SIZE(C1, ldc, 1), xvec5, xvec5;
+ADD_DY	xvec5, xvec7, xvec7;
+#endif
+STL_DX	xvec13, 0*SIZE(C0, ldc, 1);
+STH_DX	xvec13, 1*SIZE(C0, ldc, 1);
+STL_DX	xvec7, 2*SIZE(C1, ldc, 1);
+STH_DX	xvec7, 3*SIZE(C1, ldc, 1);
+
+EXTRA_DY $1, yvec12, xvec4;
+#ifndef	TRMMKERNEL
+LDL_DY	4*SIZE(C0, ldc, 1), xvec3, xvec3;
+LDH_DY	5*SIZE(C0, ldc, 1), xvec3, xvec3;
+ADD_DY	xvec3, xvec12, xvec12;
+LDL_DY	6*SIZE(C1, ldc, 1), xvec2, xvec2;
+LDH_DY	7*SIZE(C1, ldc, 1), xvec2, xvec2;
+ADD_DY	xvec2, xvec4, xvec4;
+#endif
+STL_DX	xvec12, 4*SIZE(C0, ldc, 1);
+STH_DX	xvec12, 5*SIZE(C0, ldc ,1);
+STL_DX	xvec4, 6*SIZE(C1, ldc, 1);
+STH_DX	xvec4, 7*SIZE(C1, ldc, 1);
+
+EXTRA_DY $1, yvec11, xvec7;
+#ifndef	TRMMKERNEL
+LDL_DY	0*SIZE(C1), xvec6, xvec6;
+LDH_DY	1*SIZE(C1), xvec6, xvec6;
+ADD_DY	xvec6, xvec11, xvec11;
+LDL_DY	2*SIZE(C0), xvec5, xvec5;
+LDH_DY	3*SIZE(C0), xvec5, xvec5;
+ADD_DY	xvec5, xvec7, xvec7;
+#endif
+STL_DX	xvec11, 0*SIZE(C1);
+STH_DX	xvec11, 1*SIZE(C1);
+STL_DX	xvec7, 2*SIZE(C0);
+STH_DX	xvec7, 3*SIZE(C0);
+
+EXTRA_DY $1, yvec10, xvec4;
+#ifndef	TRMMKERNEL
+LDL_DY	4*SIZE(C1), xvec3, xvec3;
+LDH_DY	5*SIZE(C1), xvec3, xvec3;
+ADD_DY	xvec3, xvec10, xvec10;
+LDL_DY	6*SIZE(C0), xvec2, xvec2;
+LDH_DY	7*SIZE(C0), xvec2, xvec2;
+ADD_DY	xvec2, xvec4, xvec4;
+#endif
+STL_DX	xvec10, 4*SIZE(C1);
+STH_DX	xvec10, 5*SIZE(C1);
+STL_DX	xvec4, 6*SIZE(C0);
+STH_DX	xvec4, 7*SIZE(C0);
+
+EXTRA_DY $1, yvec9, xvec7;
+#ifndef	TRMMKERNEL
+LDL_DY	0*SIZE(C1, ldc, 1), xvec6, xvec6;
+LDH_DY	1*SIZE(C1, ldc, 1), xvec6, xvec6;
+ADD_DY	xvec6, xvec9, xvec9;
+LDL_DY	2*SIZE(C0, ldc, 1), xvec5, xvec5;
+LDH_DY	3*SIZE(C0, ldc ,1), xvec5, xvec5;
+ADD_DY	xvec5, xvec7, xvec7;
+#endif
+STL_DX	xvec9, 0*SIZE(C1, ldc, 1);
+STH_DX	xvec9, 1*SIZE(C1, ldc, 1);
+STL_DX	xvec7, 2*SIZE(C0, ldc, 1);
+STH_DX	xvec7, 3*SIZE(C0, ldc, 1);
+
+EXTRA_DY $1, yvec8, xvec4;
+#ifndef	TRMMKERNEL
+LDL_DY	4*SIZE(C1, ldc, 1), xvec3, xvec3;
+LDH_DY	5*SIZE(C1, ldc, 1), xvec3, xvec3;
+ADD_DY	xvec3, xvec8, xvec8;
+LDL_DY	6*SIZE(C0, ldc, 1), xvec2, xvec2;
+LDH_DY	7*SIZE(C0, ldc, 1), xvec2, xvec2;
+ADD_DY	xvec2, xvec4, xvec4;
+#endif
+STL_DX	xvec8, 4*SIZE(C1, ldc, 1);
+STH_DX	xvec8, 5*SIZE(C1, ldc, 1);
+STL_DX	xvec4, 6*SIZE(C0, ldc, 1);
+STH_DX	xvec4, 7*SIZE(C0, ldc, 1);
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) ||(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kkk, %rax;
+LEAQ 	(, %rax, SIZE), %rax;
+LEAQ	(ptrba, %rax, 8), ptrba;
+LEAQ 	(ptrbb, %rax, 4), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT)
+ADDQ	$8, kk
+#endif
+
+ADDQ	$8*SIZE, C0;
+ADDQ	$8*SIZE, C1;
+DECQ i;
+JG .L1_bodyB;
+.align 32
+.L1_loopE:;
+TEST $4, bm;							# Rm = 4
+JLE	.L5_loopE;
+.align 32
+.L5_bodyB:;
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb, ptrbb;						
+#else 
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+LEAQ (, %rax, SIZE), %rax;
+LEAQ (ptrba, %rax, 4), ptrba;
+LEAQ (ptrbb, %rax, 4), ptrbb;
+#endif
+#### Initial Results Register ####
+XOR_DY	yvec15, yvec15, yvec15;
+XOR_DY 	yvec13, yvec13, yvec13;
+LD_DY	0*SIZE(ptrbb), yvec2;
+XOR_DY	yvec11, yvec11, yvec11;
+XOR_DY	yvec9, yvec9, yvec9;
+LD_DY	0*SIZE(ptrba), yvec0;
+VPERMILP_DY	$0x05, yvec2, yvec3;
+#ifndef	TRMMKERNEL
+MOVQ bk, k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else
+MOVQ kk, %rax;
+#ifdef	LEFT
+ADDQ $4, %rax;
+#else 
+ADDQ $4, %rax;
+#endif
+MOVQ %rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L6_loopE;
+.align 32;
+.L6_bodyB:;
+# Computing kernel
+
+#### Untoll time 1 ####
+LD_DY	4*SIZE(ptrba), yvec1;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD_DY	yvec15, yvec6, yvec15;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+MUL_DY	yvec0, yvec3, yvec7;
+ADD_DY	yvec13, yvec7, yvec13;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+
+LD_DY	4*SIZE(ptrbb), yvec2;
+MUL_DY	yvec0, yvec4, yvec6;
+ADD_DY	yvec11, yvec6, yvec11;
+VPERMILP_DY	$0x05, yvec2, yvec3;
+MUL_DY	yvec0, yvec5, yvec7;
+ADD_DY	yvec9, yvec7, yvec9;
+
+#### Untoll time 2 ####
+LD_DY	8*SIZE(ptrba), yvec0;
+MUL_DY	yvec1, yvec2, yvec6;
+ADD_DY	yvec15, yvec6, yvec15;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+MUL_DY	yvec1, yvec3, yvec7;
+ADD_DY	yvec13, yvec7, yvec13;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+
+LD_DY	8*SIZE(ptrbb), yvec2;
+MUL_DY	yvec1, yvec4, yvec6;
+ADD_DY	yvec11, yvec6, yvec11;
+VPERMILP_DY	$0x05, yvec2, yvec3;
+MUL_DY	yvec1, yvec5, yvec7;
+ADD_DY	yvec9, yvec7, yvec9;
+
+#### Untoll time 3 ####
+LD_DY	12*SIZE(ptrba), yvec1;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD_DY	yvec15, yvec6, yvec15;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+ADDQ	$16*SIZE, ptrba;
+MUL_DY	yvec0, yvec3, yvec7;
+ADD_DY	yvec13, yvec7, yvec13;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+
+LD_DY	12*SIZE(ptrbb), yvec2;
+MUL_DY	yvec0, yvec4, yvec6;
+ADD_DY	yvec11, yvec6, yvec11;
+VPERMILP_DY	$0x05, yvec2, yvec3;
+ADDQ	$16*SIZE, ptrbb;
+MUL_DY	yvec0, yvec5, yvec7;
+ADD_DY	yvec9, yvec7, yvec9;
+
+#### Untoll time 4 ####
+LD_DY	0*SIZE(ptrba), yvec0;
+MUL_DY	yvec1, yvec2, yvec6;
+ADD_DY	yvec15, yvec6, yvec15;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+MUL_DY	yvec1, yvec3, yvec7;
+ADD_DY	yvec13, yvec7, yvec13;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+
+LD_DY	0*SIZE(ptrbb), yvec2;
+MUL_DY	yvec1, yvec4, yvec6;
+ADD_DY	yvec11, yvec6, yvec11;
+VPERMILP_DY	$0x05, yvec2, yvec3;
+MUL_DY	yvec1, yvec5, yvec7;
+ADD_DY	yvec9, yvec7, yvec9;
+DECQ k;
+JG .L6_bodyB;
+.align 32
+.L6_loopE:;
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else
+MOVQ kkk, %rax;
+TEST $2, %rax;
+#endif
+JLE .L7_loopE;
+.align 32
+.L7_bodyB:;
+#### Untoll time 1 ####
+LD_DY	4*SIZE(ptrba), yvec1;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD_DY	yvec15, yvec6, yvec15;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+ADDQ	$8*SIZE, ptrba;
+MUL_DY	yvec0, yvec3, yvec7;
+ADD_DY	yvec13, yvec7, yvec13;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+
+LD_DY	4*SIZE(ptrbb), yvec2;
+MUL_DY	yvec0, yvec4, yvec6;
+ADD_DY	yvec11, yvec6, yvec11;
+VPERMILP_DY	$0x05, yvec2, yvec3;
+ADDQ	$8*SIZE, ptrbb;
+MUL_DY	yvec0, yvec5, yvec7;
+ADD_DY	yvec9, yvec7, yvec9;
+
+#### Untoll time 2 ####
+LD_DY	0*SIZE(ptrba), yvec0;
+MUL_DY	yvec1, yvec2, yvec6;
+ADD_DY	yvec15, yvec6, yvec15;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+MUL_DY	yvec1, yvec3, yvec7;
+ADD_DY	yvec13, yvec7, yvec13;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+
+LD_DY	0*SIZE(ptrbb), yvec2;
+MUL_DY	yvec1, yvec4, yvec6;
+ADD_DY	yvec11, yvec6, yvec11;
+VPERMILP_DY	$0x05, yvec2, yvec3;
+MUL_DY	yvec1, yvec5, yvec7;
+ADD_DY	yvec9, yvec7, yvec9;
+
+.L7_loopE:;
+#ifndef	TRMMKERNEL
+TEST $1, bk
+#else 
+MOVQ kkk, %rax;
+TEST $1, %rax;
+#endif
+JLE .L8_loopE;
+.align 32
+.L8_bodyB:;
+#### Untoll time 1 ####
+MUL_DY	yvec0, yvec2, yvec6;
+ADD_DY	yvec15, yvec6, yvec15;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+ADDQ	$4*SIZE, ptrba;
+MUL_DY	yvec0, yvec3, yvec7;
+ADD_DY	yvec13, yvec7, yvec13;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+
+MUL_DY	yvec0, yvec4, yvec6;
+ADD_DY	yvec11, yvec6, yvec11;
+ADDQ	$4*SIZE, ptrbb;
+MUL_DY	yvec0, yvec5, yvec7;
+ADD_DY	yvec9, yvec7, yvec9;
+
+.L8_loopE:;
+#### Load Alpha ####
+BROAD_DY	MEMALPHA, yvec7;
+#### Multiply Alpha ####
+MUL_DY yvec7,yvec15,yvec15;
+MUL_DY yvec7,yvec13,yvec13;
+MUL_DY yvec7,yvec11,yvec11;
+MUL_DY yvec7,yvec9,yvec9;
+#### Reverse the Results ####
+MOV_DY	yvec15, yvec7;
+REVS_DY $0x0a,yvec13,yvec15,yvec15;
+REVS_DY $0x0a,yvec7,yvec13,yvec13;
+MOV_DY yvec11,yvec7;
+REVS_DY $0x0a,yvec9,yvec11,yvec11;
+REVS_DY $0x0a,yvec7,yvec9,yvec9;
+#### Testing alignment ####
+MOVQ	C0, %rax;
+OR		ldc, %rax;
+TEST	$15, %rax;
+JNE		.L8_loopEx;					# Unalign part write back
+.align 32
+#### Writing Back ####
+EXTRA_DY $1,yvec15,xvec7;
+EXTRA_DY $1,yvec13,xvec5;
+EXTRA_DY $1,yvec11,xvec3;
+EXTRA_DY $1,yvec9,xvec1;
+#ifndef	TRMMKERNEL
+ADD_DX 0*SIZE(C0),xvec15;
+ADD_DX 2*SIZE(C1),xvec7;
+ADD_DX 0*SIZE(C0,ldc,1),xvec13;
+ADD_DX 2*SIZE(C1,ldc,1),xvec5;
+ADD_DX 0*SIZE(C1),xvec11;
+ADD_DX 2*SIZE(C0),xvec3;
+ADD_DX 0*SIZE(C1,ldc,1),xvec9;
+ADD_DX 2*SIZE(C0,ldc,1),xvec1;
+#endif
+ST_DX xvec15, 0*SIZE(C0);
+ST_DX xvec7, 2*SIZE(C1);
+ST_DX xvec13, 0*SIZE(C0,ldc,1);
+ST_DX xvec5, 2*SIZE(C1,ldc,1);
+ST_DX xvec11, 0*SIZE(C1);
+ST_DX xvec3, 2*SIZE(C0);
+ST_DX xvec9, 0*SIZE(C1,ldc,1);
+ST_DX xvec1, 2*SIZE(C0,ldc,1);
+#if	(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kkk, %rax;
+LEAQ	(, %rax, SIZE), %rax;
+LEAQ	(ptrba, %rax, 4), ptrba;
+LEAQ	(ptrbb, %rax, 4), ptrbb;
+#endif
+#if	defined(TRMMKERNEL)&&defined(LEFT)
+ADDQ	$4, kk
+#endif
+ADDQ $4*SIZE, C0;
+ADDQ $4*SIZE, C1;
+JMP	.L5_loopE;
+.align 32
+.L8_loopEx:;
+EXTRA_DY $1,yvec15,xvec7;
+EXTRA_DY $1,yvec13,xvec5;
+EXTRA_DY $1,yvec11,xvec3;
+EXTRA_DY $1,yvec9,xvec1;
+#ifndef	TRMMKERNEL
+LDL_DX	0*SIZE(C0), xvec14;
+LDH_DX	1*SIZE(C0), xvec14;
+LDL_DX	0*SIZE(C0, ldc, 1), xvec12;
+LDH_DX	1*SIZE(C0, ldc, 1), xvec12;
+LDL_DX	0*SIZE(C1), xvec10;
+LDH_DX	1*SIZE(C1), xvec10;
+LDL_DX	0*SIZE(C1, ldc, 1), xvec8;
+LDH_DX	1*SIZE(C1, ldc, 1), xvec8;
+ADD_DX	xvec14, xvec15;
+ADD_DX	xvec12, xvec13;
+ADD_DX	xvec10, xvec11;
+ADD_DX	xvec8, xvec9;
+#endif
+STL_DX	xvec15, 0*SIZE(C0);
+STH_DX	xvec15, 1*SIZE(C0);
+STL_DX	xvec13, 0*SIZE(C0, ldc, 1);
+STH_DX	xvec13, 1*SIZE(C0, ldc, 1);
+STL_DX	xvec11, 0*SIZE(C1);
+STH_DX	xvec11, 1*SIZE(C1);
+STL_DX	xvec9,	0*SIZE(C1, ldc, 1);
+STH_DX	xvec9, 	1*SIZE(C1, ldc, 1);
+#ifndef	TRMMKERNEL
+LDL_DX	2*SIZE(C0), xvec0;
+LDH_DX	3*SIZE(C0), xvec0;
+LDL_DX	2*SIZE(C0, ldc, 1), xvec2;
+LDH_DX	3*SIZE(C0, ldc, 1), xvec2;
+LDL_DX	2*SIZE(C1), xvec4;
+LDH_DX	3*SIZE(C1), xvec4;
+LDL_DX	2*SIZE(C1, ldc, 1), xvec6;
+LDH_DX	3*SIZE(C1, ldc, 1), xvec6;
+ADD_DX	xvec0, xvec3;
+ADD_DX	xvec2, xvec1;
+ADD_DX	xvec4, xvec7;
+ADD_DX	xvec6, xvec5;
+#endif
+STL_DX	xvec3, 2*SIZE(C0);
+STH_DX	xvec3, 3*SIZE(C0);
+STL_DX	xvec1, 2*SIZE(C0, ldc, 1);
+STH_DX	xvec1, 3*SIZE(C0, ldc, 1);
+STL_DX	xvec7, 2*SIZE(C1);
+STH_DX	xvec7, 3*SIZE(C1);
+STL_DX	xvec5, 2*SIZE(C1, ldc, 1);
+STH_DX	xvec5, 3*SIZE(C1, ldc, 1);
+#if	(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kkk, %rax;
+LEAQ	(, %rax, SIZE), %rax;
+LEAQ	(ptrba, %rax, 4), ptrba;
+LEAQ	(ptrbb, %rax, 4), ptrbb;
+#endif
+#if	defined(TRMMKERNEL)&&defined(LEFT)
+ADDQ	$4, kk
+#endif
+
+ADDQ	$4*SIZE, C0;
+ADDQ	$4*SIZE, C1;
+.L5_loopE:;
+TEST $2, bm;
+JLE .L9_loopE;
+.align 32
+.L9_bodyB:;
+#if	!defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb, ptrbb;						
+#else 
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+LEAQ (, %rax, SIZE), %rax;
+LEAQ (ptrba, %rax, 2), ptrba;
+LEAQ (ptrbb, %rax, 4), ptrbb
+#endif
+#### Initial Results Register ####
+LD_DX	0*SIZE(ptrbb), xvec2;
+XOR_DY yvec15, yvec15, yvec15;
+LD_DX	2*SIZE(ptrbb), xvec3;
+XOR_DY yvec13, yvec13, yvec13;
+LD_DX	0*SIZE(ptrba), xvec0;
+XOR_DY yvec11, yvec11, yvec11;
+SHUF_DX	$0x4e, xvec2, xvec4;
+XOR_DY yvec9, yvec9, yvec9;
+#ifndef	TRMMKERNEL
+MOVQ bk, k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ 	kk, %rax;
+MOVQ 	%rax, kkk;
+#else 
+MOVQ	kk, %rax;
+#ifdef	LEFT
+ADDQ	$2, %rax;
+#else
+ADDQ	$4, %rax;
+#endif
+MOVQ	%rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L10_loopE;
+.align 32;
+.L10_bodyB:;
+# Computing kernel 
+
+##### Unroll time 1 ####
+LD_DX	4*SIZE(ptrbb), xvec6;
+SHUF_DX	$0x4e, xvec3, xvec5;
+MUL_DX	xvec0, xvec2;
+ADD_DX	xvec2, xvec15;
+
+LD_DX	6*SIZE(ptrbb), xvec7;
+MUL_DX	xvec0, xvec3;
+ADD_DX	xvec3, xvec11;
+
+LD_DX	2*SIZE(ptrba), xvec1;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec13;
+SHUF_DX	$0x4e, xvec6, xvec4;
+MUL_DX	xvec0, xvec5;
+ADD_DX	xvec5, xvec9;
+
+#### Unroll time 2 ####
+LD_DX	8*SIZE(ptrbb), xvec2;
+SHUF_DX	$0x4e, xvec7, xvec5;
+MUL_DX	xvec1, xvec6;
+ADD_DX	xvec6, xvec15;
+
+LD_DX	10*SIZE(ptrbb), xvec3;
+MUL_DX 	xvec1, xvec7;
+ADD_DX	xvec7, xvec11;
+
+LD_DX	4*SIZE(ptrba), xvec0;
+MUL_DX	xvec1, xvec4;
+ADD_DX	xvec4, xvec13;
+SHUF_DX	$0x4e, xvec2, xvec4;
+MUL_DX	xvec1, xvec5;
+ADD_DX	xvec5, xvec9;
+
+##### Unroll time 3 ####
+LD_DX	12*SIZE(ptrbb), xvec6;
+SHUF_DX	$0x4e, xvec3, xvec5;
+MUL_DX	xvec0, xvec2;
+ADD_DX	xvec2, xvec15;
+
+LD_DX	14*SIZE(ptrbb), xvec7;
+MUL_DX	xvec0, xvec3;
+ADD_DX	xvec3, xvec11;
+ADDQ	$16*SIZE, ptrbb;
+
+LD_DX	6*SIZE(ptrba), xvec1;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec13;
+SHUF_DX	$0x4e, xvec6, xvec4;
+ADDQ	$8*SIZE, ptrba;
+MUL_DX	xvec0, xvec5;
+ADD_DX	xvec5, xvec9;
+
+#### Unroll time 4 ####
+LD_DX	0*SIZE(ptrbb), xvec2;
+SHUF_DX	$0x4e, xvec7, xvec5;
+MUL_DX	xvec1, xvec6;
+ADD_DX	xvec6, xvec15;
+
+LD_DX	2*SIZE(ptrbb), xvec3;
+MUL_DX 	xvec1, xvec7;
+ADD_DX	xvec7, xvec11;
+
+LD_DX	0*SIZE(ptrba), xvec0;
+MUL_DX	xvec1, xvec4;
+ADD_DX	xvec4, xvec13;
+SHUF_DX	$0x4e, xvec2, xvec4;
+MUL_DX	xvec1, xvec5;
+ADD_DX	xvec5, xvec9;
+DECQ k;
+JG .L10_bodyB;
+.align 32
+.L10_loopE:;
+#ifndef	TRMMKERNEL
+TEST $2, bk
+#else 
+MOVQ kkk, %rax;
+TEST $2, %rax;
+#endif
+JLE .L11_loopE;
+.align 32
+.L11_bodyB:;
+##### Unroll time 1 ####
+LD_DX	4*SIZE(ptrbb), xvec6;
+SHUF_DX	$0x4e, xvec3, xvec5;
+MUL_DX	xvec0, xvec2;
+ADD_DX	xvec2, xvec15;
+
+LD_DX	6*SIZE(ptrbb), xvec7;
+MUL_DX	xvec0, xvec3;
+ADD_DX	xvec3, xvec11;
+ADDQ	$8*SIZE, ptrbb;
+
+LD_DX	2*SIZE(ptrba), xvec1;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec13;
+SHUF_DX	$0x4e, xvec6, xvec4;
+ADDQ	$4*SIZE, ptrba;
+
+MUL_DX	xvec0, xvec5;
+ADD_DX	xvec5, xvec9;
+
+#### Unroll time 2 ####
+LD_DX	0*SIZE(ptrbb), xvec2;
+SHUF_DX	$0x4e, xvec7, xvec5;
+MUL_DX	xvec1, xvec6;
+ADD_DX	xvec6, xvec15;
+
+LD_DX	2*SIZE(ptrbb), xvec3;
+MUL_DX 	xvec1, xvec7;
+ADD_DX	xvec7, xvec11;
+
+LD_DX	0*SIZE(ptrba), xvec0;
+MUL_DX	xvec1, xvec4;
+ADD_DX	xvec4, xvec13;
+SHUF_DX	$0x4e, xvec2, xvec4;
+MUL_DX	xvec1, xvec5;
+ADD_DX	xvec5, xvec9;
+
+.L11_loopE:;
+#ifndef	TRMMKERNEL
+TEST $1, bk
+#else
+MOVQ kkk, %rax;
+TEST $1, %rax;
+#endif
+JLE .L12_loopE;
+.align 32
+.L12_bodyB:;
+SHUF_DX	$0x4e, xvec3, xvec5;
+MUL_DX	xvec0, xvec2;
+ADD_DX	xvec2, xvec15;
+ADDQ	$4*SIZE, ptrbb;
+
+MUL_DX	xvec0, xvec3;
+ADD_DX	xvec3, xvec11;
+ADDQ	$2*SIZE, ptrba;
+
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec13;
+
+MUL_DX	xvec0, xvec5;
+ADD_DX	xvec5, xvec9;
+
+.L12_loopE:;
+#### Load Alpha ####
+BROAD_DX MEMALPHA, xvec7;
+#### Multiply Alpha ####
+MUL_DX	xvec7, xvec15;
+MUL_DX	xvec7, xvec13;
+MUL_DX	xvec7, xvec11;
+MUL_DX	xvec7, xvec9;
+#### Reverse the Results ####
+MOV_DX	xvec15, xvec6;
+REVS_DX	xvec13, xvec15;
+REVS_DX xvec6, xvec13;
+MOV_DX	xvec11, xvec6;
+REVS_DX	xvec9, xvec11;
+REVS_DX	xvec6, xvec9;
+#### Testing Alignment ####
+MOVQ	C0, %rax;
+OR		ldc, %rax;
+TEST	$15, %rax;
+JNE		.L12_loopEx;
+.align 32
+#### Writing Back ####
+#ifndef	TRMMKERNEL
+ADD_DX	0*SIZE(C0), xvec13;
+ADD_DX	0*SIZE(C0, ldc, 1), xvec15;
+ADD_DX	0*SIZE(C1), xvec9;
+ADD_DX	0*SIZE(C1, ldc, 1), xvec11;
+#endif
+ST_DX	xvec13, 0*SIZE(C0);
+ST_DX	xvec15, 0*SIZE(C0, ldc, 1);
+ST_DX	xvec9, 0*SIZE(C1);
+ST_DX	xvec11, 0*SIZE(C1, ldc, 1);
+#if (defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kkk, %rax;
+LEAQ	(,%rax, SIZE), %rax;
+LEAQ	(ptrba, %rax, 2), ptrba;
+LEAQ	(ptrbb, %rax, 4), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT)
+ADDQ	$2, kk
+#endif
+ADDQ	$2*SIZE, C0
+ADDQ	$2*SIZE, C1
+JMP		.L9_loopE;
+.align 32
+.L12_loopEx:
+#ifndef	TRMMKERNEL
+LDL_DX	0*SIZE(C0), xvec14;
+LDH_DX	1*SIZE(C0), xvec14;
+LDL_DX	0*SIZE(C0, ldc, 1), xvec12;
+LDH_DX	1*SIZE(C0, ldc, 1), xvec12;
+LDL_DX	0*SIZE(C1), xvec10;
+LDH_DX	1*SIZE(C1), xvec10;
+LDL_DX	0*SIZE(C1, ldc, 1), xvec8;
+LDH_DX	1*SIZE(C1, ldc, 1), xvec8;
+ADD_DX	xvec14, xvec13;
+ADD_DX	xvec12, xvec15;
+ADD_DX	xvec10, xvec9;
+ADD_DX	xvec8, xvec11;
+#endif
+STL_DX	xvec13, 0*SIZE(C0);
+STH_DX	xvec13, 1*SIZE(C0);
+STL_DX	xvec15, 0*SIZE(C0, ldc, 1);
+STH_DX	xvec15, 1*SIZE(C0, ldc, 1);
+STL_DX	xvec9,	0*SIZE(C1);
+STH_DX	xvec9,	1*SIZE(C1);
+STL_DX	xvec11,	0*SIZE(C1, ldc, 1);
+STH_DX	xvec11,	1*SIZE(C1, ldc, 1);
+#if (defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kkk, %rax;
+LEAQ	(,%rax, SIZE), %rax;
+LEAQ	(ptrba, %rax, 2), ptrba;
+LEAQ	(ptrbb, %rax, 4), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT)
+ADDQ	$2, kk
+#endif
+ADDQ	$2*SIZE, C0;
+ADDQ	$2*SIZE, C1;
+.L9_loopE:;
+TEST $1, bm
+JLE	.L13_loopE;
+.align 32
+.L13_bodyB:;
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb, ptrbb;
+#else 
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+LEAQ (,%rax, SIZE), %rax;
+ADDQ	%rax, ptrba;
+LEAQ (ptrbb, %rax, 4), ptrbb;
+#endif
+#### Initial Results Register ####
+XOR_DY	yvec15, yvec15, yvec15;
+#ifndef	TRMMKERNEL
+MOVQ bk, k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kk, %rax;
+MOVQ	%rax, kkk;
+#else
+MOVQ	kk, %rax;
+#ifdef	LEFT
+ADDQ	$1, %rax;
+#else 
+ADDQ	$4, %rax;
+#endif
+MOVQ	%rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L14_loopE;
+.align 32
+.L14_bodyB:;
+BROAD_DY 0*SIZE(ptrba), yvec0;
+LD_DY	0*SIZE(ptrbb), yvec2;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD_DY	yvec15, yvec6, yvec15;
+
+BROAD_DY 1*SIZE(ptrba), yvec1;
+LD_DY	4*SIZE(ptrbb), yvec3;
+MUL_DY	yvec1, yvec3, yvec7;
+ADD_DY	yvec15, yvec7, yvec15;
+
+BROAD_DY 2*SIZE(ptrba), yvec0;
+LD_DY	8*SIZE(ptrbb), yvec2;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD_DY	yvec15, yvec6, yvec15;
+
+BROAD_DY 3*SIZE(ptrba), yvec1;
+LD_DY	12*SIZE(ptrbb), yvec3;
+MUL_DY	yvec1, yvec3, yvec7;
+ADD_DY	yvec15, yvec7, yvec15;
+ADDQ	$4*SIZE, ptrba;
+ADDQ	$16*SIZE, ptrbb;
+DECQ	k;
+JG	.L14_bodyB;
+.align 32
+.L14_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else
+MOVQ kkk, %rax;
+TEST $2, %rax;
+#endif
+JLE .L15_loopE;
+.align 32
+.L15_bodyB:
+BROAD_DY 0*SIZE(ptrba), yvec0;
+LD_DY	0*SIZE(ptrbb), yvec2;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD_DY	yvec15, yvec6, yvec15;
+
+BROAD_DY 1*SIZE(ptrba), yvec1;
+LD_DY	4*SIZE(ptrbb), yvec3;
+MUL_DY	yvec1, yvec3, yvec7;
+ADD_DY	yvec15, yvec7, yvec15;
+ADDQ	$2*SIZE, ptrba;
+ADDQ	$8*SIZE, ptrbb;
+.L15_loopE:;
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else 
+MOVQ kkk, %rax;
+TEST $1, %rax;
+#endif
+JLE .L16_loopE;
+.align 32
+.L16_bodyB:;
+BROAD_DY 0*SIZE(ptrba), yvec0;
+LD_DY	0*SIZE(ptrbb), yvec2;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD_DY	yvec15, yvec6, yvec15;
+ADDQ	$1*SIZE, ptrba;
+ADDQ	$4*SIZE, ptrbb;
+
+.L16_loopE:
+#### Load Alpha #### 
+BROAD_DY MEMALPHA, yvec7;
+#### Multiply Alpha ####
+MUL_DY	yvec15, yvec7, yvec15;
+#### Writing Back ####
+EXTRA_DY $1, yvec15, xvec7;
+#ifndef	TRMMKERNEL
+LDL_DX	0*SIZE(C0), xvec0;
+LDH_DX	0*SIZE(C0, ldc, 1), xvec0;
+LDL_DX	0*SIZE(C1), xvec1;
+LDH_DX	0*SIZE(C1, ldc, 1), xvec1;
+ADD_DX	xvec0, xvec15;
+ADD_DX	xvec1, xvec7;
+#endif
+STL_DX	xvec15, 0*SIZE(C0);
+STH_DX	xvec15, 0*SIZE(C0, ldc, 1);
+STL_DX	xvec7, 0*SIZE(C1);
+STH_DX	xvec7, 0*SIZE(C1, ldc, 1);
+#if (defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kkk, %rax;
+LEAQ	(,%rax, SIZE), %rax;
+ADDQ	%rax, ptrba;
+LEAQ	(ptrbb, %rax, 4), ptrbb;
+#endif
+#if defined(TRMMKERNEL)&&defined(LEFT)
+ADDQ	$1, kk
+#endif
+ADDQ	$1*SIZE, C0
+ADDQ	$1*SIZE, C1
+.L13_loopE:;
+#if	defined(TRMMKERNEL)&&!defined(LEFT)
+ADDQ	$4, kk
+#endif
+MOVQ bk,k;
+SALQ $5,k;
+ADDQ k,bb;
+LEAQ (C,ldc,4),C;
+.L0_bodyE:;
+DECQ j;
+JG .L0_bodyB;
+.align 32;
+.L0_loopE:;
+TEST $2, bn;
+JLE .L20_loopE;
+.align 32;
+.L20_loopB:;
+#if defined(TRMMKERNEL) && defined(LEFT)
+MOVQ	OFFSET, %rax;
+MOVQ	%rax, kk
+#endif
+MOVQ C, C0;
+LEAQ (C, ldc, 1), C1;
+MOVQ ba, ptrba;
+MOVQ bm, i;
+SARQ $3, i;					# Rm = 8
+JLE	.L21_loopE;
+.align 32;
+.L21_bodyB:;
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb, ptrbb;
+#else 
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+LEAQ (, %rax, SIZE), %rax;
+LEAQ (ptrba, %rax, 8), ptrba;
+LEAQ (ptrbb, %rax, 2), ptrbb;
+#endif
+#### Initial Results Register ####
+XOR_DY yvec15, yvec15, yvec15;
+XOR_DY yvec14, yvec14, yvec14;
+XOR_DY yvec13, yvec13, yvec13;
+XOR_DY yvec12, yvec12, yvec12;
+XOR_DY yvec11, yvec11, yvec11;
+XOR_DY yvec10, yvec10, yvec10;
+XOR_DY yvec9, yvec9, yvec9;
+XOR_DY yvec8, yvec8, yvec8;
+#ifndef	TRMMKERNEL
+MOVQ bk, k;
+#elif (defined(LEFT) && !defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kk, %rax;
+MOVQ	%rax, kkk;
+#else 
+MOVQ	kk, %rax;
+#ifdef	LEFT
+ADDQ	$8, %rax;
+#else 
+ADDQ	$2, %rax;
+#endif
+MOVQ	%rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L211_loopE;
+.align 32;
+.L211_bodyB:
+# Computing kernel 
+#### Unroll time 1 ####
+LD_DX	0*SIZE(ptrba), xvec0;
+LD_DX	0*SIZE(ptrbb), xvec4;
+MOV_DX	xvec4, xvec5;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec15;
+
+LD_DX	2*SIZE(ptrba), xvec1;
+MOV_DX	xvec5, xvec6;
+MUL_DX	xvec1, xvec5;
+ADD_DX	xvec5, xvec14;
+
+LD_DX	4*SIZE(ptrba), xvec2;
+MOV_DX	xvec6, xvec7;
+MUL_DX	xvec2, xvec6;
+ADD_DX	xvec6, xvec13;
+
+LD_DX	6*SIZE(ptrba), xvec3;
+SHUF_DX	$0x4e, xvec7, xvec4;
+MUL_DX	xvec3, xvec7;
+ADD_DX	xvec7, xvec12;
+
+MOV_DX	xvec4, xvec5;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec11;
+
+MOV_DX	xvec5, xvec6;
+MUL_DX	xvec1, xvec5;
+ADD_DX	xvec5, xvec10;
+
+MOV_DX	xvec6, xvec7;
+MUL_DX	xvec2, xvec6;
+ADD_DX	xvec6, xvec9;
+
+MUL_DX	xvec3, xvec7;
+ADD_DX 	xvec7, xvec8;
+
+#### Unroll time 2 ####
+LD_DX	8*SIZE(ptrba), xvec0;
+LD_DX	2*SIZE(ptrbb), xvec4;
+MOV_DX	xvec4, xvec5;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec15;
+
+LD_DX	10*SIZE(ptrba), xvec1;
+MOV_DX	xvec5, xvec6;
+MUL_DX	xvec1, xvec5;
+ADD_DX	xvec5, xvec14;
+
+LD_DX	12*SIZE(ptrba), xvec2;
+MOV_DX	xvec6, xvec7;
+MUL_DX	xvec2, xvec6;
+ADD_DX	xvec6, xvec13;
+
+LD_DX	14*SIZE(ptrba), xvec3;
+SHUF_DX	$0x4e, xvec7, xvec4;
+MUL_DX	xvec3, xvec7;
+ADD_DX	xvec7, xvec12;
+
+MOV_DX	xvec4, xvec5;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec11;
+
+MOV_DX	xvec5, xvec6;
+MUL_DX	xvec1, xvec5;
+ADD_DX	xvec5, xvec10;
+
+MOV_DX	xvec6, xvec7;
+MUL_DX	xvec2, xvec6;
+ADD_DX	xvec6, xvec9;
+
+MUL_DX	xvec3, xvec7;
+ADD_DX 	xvec7, xvec8;
+
+#### Unroll time 3 ####
+LD_DX	16*SIZE(ptrba), xvec0;
+LD_DX	4*SIZE(ptrbb), xvec4;
+MOV_DX	xvec4, xvec5;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec15;
+
+LD_DX	18*SIZE(ptrba), xvec1;
+MOV_DX	xvec5, xvec6;
+MUL_DX	xvec1, xvec5;
+ADD_DX	xvec5, xvec14;
+
+LD_DX	20*SIZE(ptrba), xvec2;
+MOV_DX	xvec6, xvec7;
+MUL_DX	xvec2, xvec6;
+ADD_DX	xvec6, xvec13;
+
+LD_DX	22*SIZE(ptrba), xvec3;
+SHUF_DX	$0x4e, xvec7, xvec4;
+MUL_DX	xvec3, xvec7;
+ADD_DX	xvec7, xvec12;
+
+MOV_DX	xvec4, xvec5;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec11;
+
+MOV_DX	xvec5, xvec6;
+MUL_DX	xvec1, xvec5;
+ADD_DX	xvec5, xvec10;
+
+MOV_DX	xvec6, xvec7;
+MUL_DX	xvec2, xvec6;
+ADD_DX	xvec6, xvec9;
+
+MUL_DX	xvec3, xvec7;
+ADD_DX 	xvec7, xvec8;
+
+#### Unroll time 4 ####
+LD_DX	24*SIZE(ptrba), xvec0;
+LD_DX	6*SIZE(ptrbb), xvec4;
+MOV_DX	xvec4, xvec5;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec15;
+ADDQ	$8*SIZE, ptrbb;
+
+LD_DX	26*SIZE(ptrba), xvec1;
+MOV_DX	xvec5, xvec6;
+MUL_DX	xvec1, xvec5;
+ADD_DX	xvec5, xvec14;
+
+LD_DX	28*SIZE(ptrba), xvec2;
+MOV_DX	xvec6, xvec7;
+MUL_DX	xvec2, xvec6;
+ADD_DX	xvec6, xvec13;
+
+LD_DX	30*SIZE(ptrba), xvec3;
+SHUF_DX	$0x4e, xvec7, xvec4;
+MUL_DX	xvec3, xvec7;
+ADD_DX	xvec7, xvec12;
+ADDQ	$32*SIZE, ptrba;
+
+MOV_DX	xvec4, xvec5;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec11;
+
+MOV_DX	xvec5, xvec6;
+MUL_DX	xvec1, xvec5;
+ADD_DX	xvec5, xvec10;
+
+MOV_DX	xvec6, xvec7;
+MUL_DX	xvec2, xvec6;
+ADD_DX	xvec6, xvec9;
+
+MUL_DX	xvec3, xvec7;
+ADD_DX 	xvec7, xvec8;
+DECQ	k;
+JG	.L211_bodyB;
+.align 32
+.L211_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else 
+MOVQ kkk, %rax;
+TEST $2, %rax;
+#endif
+JLE .L212_loopE;
+.align 32;
+.L212_bodyB:
+# Computing kernel 
+#### Unroll time 1 ####
+LD_DX	0*SIZE(ptrba), xvec0;
+LD_DX	0*SIZE(ptrbb), xvec4;
+MOV_DX	xvec4, xvec5;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec15;
+
+LD_DX	2*SIZE(ptrba), xvec1;
+MOV_DX	xvec5, xvec6;
+MUL_DX	xvec1, xvec5;
+ADD_DX	xvec5, xvec14;
+
+LD_DX	4*SIZE(ptrba), xvec2;
+MOV_DX	xvec6, xvec7;
+MUL_DX	xvec2, xvec6;
+ADD_DX	xvec6, xvec13;
+
+LD_DX	6*SIZE(ptrba), xvec3;
+SHUF_DX	$0x4e, xvec7, xvec4;
+MUL_DX	xvec3, xvec7;
+ADD_DX	xvec7, xvec12;
+
+MOV_DX	xvec4, xvec5;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec11;
+
+MOV_DX	xvec5, xvec6;
+MUL_DX	xvec1, xvec5;
+ADD_DX	xvec5, xvec10;
+
+MOV_DX	xvec6, xvec7;
+MUL_DX	xvec2, xvec6;
+ADD_DX	xvec6, xvec9;
+
+MUL_DX	xvec3, xvec7;
+ADD_DX 	xvec7, xvec8;
+
+#### Unroll time 2 ####
+LD_DX	8*SIZE(ptrba), xvec0;
+LD_DX	2*SIZE(ptrbb), xvec4;
+MOV_DX	xvec4, xvec5;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec15;
+ADDQ	$4*SIZE, ptrbb;
+
+LD_DX	10*SIZE(ptrba), xvec1;
+MOV_DX	xvec5, xvec6;
+MUL_DX	xvec1, xvec5;
+ADD_DX	xvec5, xvec14;
+
+LD_DX	12*SIZE(ptrba), xvec2;
+MOV_DX	xvec6, xvec7;
+MUL_DX	xvec2, xvec6;
+ADD_DX	xvec6, xvec13;
+
+LD_DX	14*SIZE(ptrba), xvec3;
+SHUF_DX	$0x4e, xvec7, xvec4;
+MUL_DX	xvec3, xvec7;
+ADD_DX	xvec7, xvec12;
+ADDQ	$16*SIZE, ptrba;
+
+MOV_DX	xvec4, xvec5;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec11;
+
+MOV_DX	xvec5, xvec6;
+MUL_DX	xvec1, xvec5;
+ADD_DX	xvec5, xvec10;
+
+MOV_DX	xvec6, xvec7;
+MUL_DX	xvec2, xvec6;
+ADD_DX	xvec6, xvec9;
+
+MUL_DX	xvec3, xvec7;
+ADD_DX 	xvec7, xvec8;
+
+.L212_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else 
+MOVQ kkk, %rax;
+TEST $1, %rax;
+#endif
+JLE .L213_loopE;
+.align 32
+.L213_bodyB:
+#### Unroll time 1 ####
+LD_DX	0*SIZE(ptrba), xvec0;
+LD_DX	0*SIZE(ptrbb), xvec4;
+MOV_DX	xvec4, xvec5;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec15;
+ADDQ	$2*SIZE, ptrbb;
+
+LD_DX	2*SIZE(ptrba), xvec1;
+MOV_DX	xvec5, xvec6;
+MUL_DX	xvec1, xvec5;
+ADD_DX	xvec5, xvec14;
+
+LD_DX	4*SIZE(ptrba), xvec2;
+MOV_DX	xvec6, xvec7;
+MUL_DX	xvec2, xvec6;
+ADD_DX	xvec6, xvec13;
+
+LD_DX	6*SIZE(ptrba), xvec3;
+SHUF_DX	$0x4e, xvec7, xvec4;
+MUL_DX	xvec3, xvec7;
+ADD_DX	xvec7, xvec12;
+ADDQ	$8*SIZE, ptrba;
+
+MOV_DX	xvec4, xvec5;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec11;
+
+MOV_DX	xvec5, xvec6;
+MUL_DX	xvec1, xvec5;
+ADD_DX	xvec5, xvec10;
+
+MOV_DX	xvec6, xvec7;
+MUL_DX	xvec2, xvec6;
+ADD_DX	xvec6, xvec9;
+
+MUL_DX	xvec3, xvec7;
+ADD_DX 	xvec7, xvec8;
+
+.L213_loopE:
+#### Multiply Alpha ####
+BROAD_DX MEMALPHA, xvec7;
+MUL_DX	xvec7, xvec15;
+MUL_DX	xvec7, xvec14;
+MUL_DX	xvec7, xvec13;
+MUL_DX	xvec7, xvec12;
+MUL_DX	xvec7, xvec11;
+MUL_DX	xvec7, xvec10;
+MUL_DX	xvec7, xvec9;
+MUL_DX	xvec7, xvec8;
+#### Reverse #####
+MOV_DX	xvec15, xvec6;
+REVS_DX	xvec11, xvec15;
+REVS_DX	xvec6, xvec11;
+MOV_DX	xvec14, xvec6;
+REVS_DX	xvec10, xvec14;
+REVS_DX	xvec6, xvec10;
+MOV_DX	xvec13, xvec6;
+REVS_DX	xvec9, xvec13;
+REVS_DX	xvec6, xvec9;
+MOV_DX	xvec12, xvec6;
+REVS_DX	xvec8, xvec12;
+REVS_DX	xvec6, xvec8;
+#### Testing Alignment ####
+MOVQ	C0, %rax;
+OR		ldc, %rax;
+TEST	$15, %rax;
+JNE		.L213_loopEx;
+.align 32
+#### Writing Back ####
+#ifndef	TRMMKERNEL
+ADD_DX	0*SIZE(C0), xvec11;
+ADD_DX	2*SIZE(C0), xvec10;
+ADD_DX	4*SIZE(C0), xvec9;
+ADD_DX	6*SIZE(C0), xvec8;
+ADD_DX	0*SIZE(C1), xvec15;
+ADD_DX	2*SIZE(C1), xvec14;
+ADD_DX	4*SIZE(C1), xvec13;
+ADD_DX	6*SIZE(C1), xvec12;
+#endif
+ST_DX	xvec11,	0*SIZE(C0);
+ST_DX	xvec10,	2*SIZE(C0);
+ST_DX	xvec9,	4*SIZE(C0);
+ST_DX	xvec8,	6*SIZE(C0);
+ST_DX	xvec15, 0*SIZE(C1);
+ST_DX	xvec14, 2*SIZE(C1);
+ST_DX	xvec13, 4*SIZE(C1);
+ST_DX	xvec12, 6*SIZE(C1);
+#if (defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kkk, %rax;
+LEAQ	(,%rax, SIZE), %rax;
+LEAQ 	(ptrba, %rax, 8), ptrba;
+LEAQ	(ptrbb, %rax, 2), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT)
+ADDQ	$8, kk
+#endif
+ADDQ	$8*SIZE, C0;
+ADDQ	$8*SIZE, C1;
+DECQ i;
+JG	.L21_bodyB;
+JMP	.L21_loopE;
+.align 32
+.L213_loopEx:;
+#ifndef	TRMMKERNEL
+LDL_DX	0*SIZE(C0), xvec0;
+LDH_DX	1*SIZE(C0), xvec0;
+LDL_DX	2*SIZE(C0), xvec1;
+LDH_DX	3*SIZE(C0), xvec1;
+LDL_DX	4*SIZE(C0), xvec2;
+LDH_DX	5*SIZE(C0), xvec2;
+LDL_DX	6*SIZE(C0), xvec3;
+LDH_DX	7*SIZE(C0), xvec3;
+ADD_DX	xvec0, xvec11;
+ADD_DX	xvec1, xvec10;
+ADD_DX	xvec2, xvec9;
+ADD_DX	xvec3, xvec8;
+#endif
+STL_DX	xvec11, 0*SIZE(C0);
+STH_DX	xvec11, 1*SIZE(C0);
+STL_DX	xvec10, 2*SIZE(C0);
+STH_DX	xvec10, 3*SIZE(C0);
+STL_DX	xvec9, 4*SIZE(C0);
+STH_DX	xvec9, 5*SIZE(C0);
+STL_DX	xvec8, 6*SIZE(C0);
+STH_DX	xvec8, 7*SIZE(C0);
+#ifndef	TRMMKERNEL
+LDL_DX	0*SIZE(C1), xvec4;
+LDH_DX	1*SIZE(C1), xvec4;
+LDL_DX	2*SIZE(C1), xvec5;
+LDH_DX	3*SIZE(C1), xvec5;
+LDL_DX	4*SIZE(C1), xvec6;
+LDH_DX	5*SIZE(C1), xvec6;
+LDL_DX	6*SIZE(C1), xvec7;
+LDH_DX	7*SIZE(C1), xvec7;
+ADD_DX	xvec4, xvec15;
+ADD_DX	xvec5, xvec14;
+ADD_DX	xvec6, xvec13;
+ADD_DX	xvec7, xvec12;
+#endif
+STL_DX	xvec15, 0*SIZE(C1);
+STH_DX	xvec15, 1*SIZE(C1);
+STL_DX	xvec14, 2*SIZE(C1);
+STH_DX	xvec14, 3*SIZE(C1);
+STL_DX	xvec13, 4*SIZE(C1);
+STH_DX	xvec13, 5*SIZE(C1);
+STL_DX	xvec12, 6*SIZE(C1);
+STH_DX	xvec12, 7*SIZE(C1);
+#if (defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kkk, %rax;
+LEAQ	(,%rax, SIZE), %rax;
+LEAQ 	(ptrba, %rax, 8), ptrba;
+LEAQ	(ptrbb, %rax, 2), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT)
+ADDQ	$8, kk
+#endif
+ADDQ	$8*SIZE, C0;
+ADDQ	$8*SIZE, C1;
+DECQ	i;
+JG	.L21_bodyB;
+.L21_loopE:;
+TEST $4, bm;				# Rm = 4
+JLE .L22_loopE;
+.align 32;
+.L22_bodyB:;
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb, ptrbb;
+#else 
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+LEAQ	(,%rax, SIZE), %rax;
+LEAQ	(ptrba, %rax, 4), ptrba;
+LEAQ	(ptrbb, %rax, 2), ptrbb;
+#endif
+#### Initial Results Register ####
+XOR_DY yvec15, yvec15, yvec15;
+XOR_DY yvec14, yvec14, yvec14;
+XOR_DY yvec11, yvec11, yvec11;
+XOR_DY yvec10, yvec10, yvec10;
+#ifndef	TRMMKERNEL
+MOVQ bk, k;
+#elif (defined(LEFT) && !defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kk, %rax;
+MOVQ	%rax, kkk;
+#else 
+MOVQ	kk, %rax;
+#ifdef	LEFT
+ADDQ	$4, %rax;
+#else 
+ADDQ	$2, %rax;
+#endif
+MOVQ	%rax, kkk;
+#endif
+SARQ $2, k;
+JLE	.L221_loopE;
+.align 32
+.L221_bodyB:;
+# Computing kernel 
+#### Unroll time 1 ####
+LD_DX	0*SIZE(ptrba), xvec0;
+LD_DX	0*SIZE(ptrbb), xvec4;
+MOV_DX	xvec4, xvec5;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec15;
+
+LD_DX	2*SIZE(ptrba), xvec1;
+SHUF_DX	$0x4e, xvec5, xvec4;
+MUL_DX	xvec1, xvec5;
+ADD_DX	xvec5, xvec14;
+
+MOV_DX	xvec4, xvec5;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec11;
+
+MUL_DX	xvec1, xvec5;
+ADD_DX	xvec5, xvec10;
+
+#### Unroll time 2 ####
+LD_DX	4*SIZE(ptrba), xvec0;
+LD_DX	2*SIZE(ptrbb), xvec4;
+MOV_DX	xvec4, xvec5;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec15;
+
+LD_DX	6*SIZE(ptrba), xvec1;
+SHUF_DX	$0x4e, xvec5, xvec4;
+MUL_DX	xvec1, xvec5;
+ADD_DX	xvec5, xvec14;
+
+MOV_DX	xvec4, xvec5;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec11;
+
+MUL_DX	xvec1, xvec5;
+ADD_DX	xvec5, xvec10;
+
+#### Unroll time 3 ####
+LD_DX	8*SIZE(ptrba), xvec0;
+LD_DX	4*SIZE(ptrbb), xvec4;
+MOV_DX	xvec4, xvec5;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec15;
+
+LD_DX	10*SIZE(ptrba), xvec1;
+SHUF_DX	$0x4e, xvec5, xvec4;
+MUL_DX	xvec1, xvec5;
+ADD_DX	xvec5, xvec14;
+
+MOV_DX	xvec4, xvec5;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec11;
+
+MUL_DX	xvec1, xvec5;
+ADD_DX	xvec5, xvec10;
+
+#### Unroll time 4 ####
+LD_DX	12*SIZE(ptrba), xvec0;
+LD_DX	6*SIZE(ptrbb), xvec4;
+MOV_DX	xvec4, xvec5;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec15;
+ADDQ	$8*SIZE, ptrbb;
+
+LD_DX	14*SIZE(ptrba), xvec1;
+SHUF_DX	$0x4e, xvec5, xvec4;
+MUL_DX	xvec1, xvec5;
+ADD_DX	xvec5, xvec14;
+ADDQ	$16*SIZE, ptrba;
+
+MOV_DX	xvec4, xvec5;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec11;
+
+MUL_DX	xvec1, xvec5;
+ADD_DX	xvec5, xvec10;
+DECQ	k;
+JG	.L221_bodyB;
+.align 32
+.L221_loopE:;
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else 
+MOVQ kkk, %rax;
+TEST $2, %rax;
+#endif
+JLE .L222_loopE;
+.align 32
+.L222_bodyB:
+#### Unroll time 1 ####
+LD_DX	0*SIZE(ptrba), xvec0;
+LD_DX	0*SIZE(ptrbb), xvec4;
+MOV_DX	xvec4, xvec5;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec15;
+
+LD_DX	2*SIZE(ptrba), xvec1;
+SHUF_DX	$0x4e, xvec5, xvec4;
+MUL_DX	xvec1, xvec5;
+ADD_DX	xvec5, xvec14;
+
+MOV_DX	xvec4, xvec5;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec11;
+
+MUL_DX	xvec1, xvec5;
+ADD_DX	xvec5, xvec10;
+
+#### Unroll time 2 ####
+LD_DX	4*SIZE(ptrba), xvec0;
+LD_DX	2*SIZE(ptrbb), xvec4;
+MOV_DX	xvec4, xvec5;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec15;
+ADDQ	$4*SIZE, ptrbb;
+
+LD_DX	6*SIZE(ptrba), xvec1;
+SHUF_DX	$0x4e, xvec5, xvec4;
+MUL_DX	xvec1, xvec5;
+ADD_DX	xvec5, xvec14;
+ADDQ	$8*SIZE, ptrba;
+MOV_DX	xvec4, xvec5;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec11;
+
+MUL_DX	xvec1, xvec5;
+ADD_DX	xvec5, xvec10;
+
+.L222_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk
+#else 
+MOVQ kkk, %rax;
+TEST $1, %rax;
+#endif
+JLE .L223_loopE;
+.align 32
+.L223_bodyB:
+#### Unroll time 1 ####
+LD_DX	0*SIZE(ptrba), xvec0;
+LD_DX	0*SIZE(ptrbb), xvec4;
+MOV_DX	xvec4, xvec5;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec15;
+ADDQ	$2*SIZE, ptrbb;
+
+LD_DX	2*SIZE(ptrba), xvec1;
+SHUF_DX	$0x4e, xvec5, xvec4;
+MUL_DX	xvec1, xvec5;
+ADD_DX	xvec5, xvec14;
+ADDQ	$4*SIZE, ptrba;
+
+MOV_DX	xvec4, xvec5;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec11;
+
+MUL_DX	xvec1, xvec5;
+ADD_DX	xvec5, xvec10;
+
+.L223_loopE:
+#### Multiply Alpha ####
+BROAD_DX MEMALPHA, xvec7;
+MUL_DX	xvec7, xvec15;
+MUL_DX	xvec7, xvec14;
+MUL_DX	xvec7, xvec11;
+MUL_DX	xvec7, xvec10;
+#### Reverse #####
+MOV_DX	xvec15, xvec6;
+REVS_DX	xvec11, xvec15;
+REVS_DX	xvec6, xvec11;
+MOV_DX	xvec14, xvec6;
+REVS_DX	xvec10, xvec14;
+REVS_DX	xvec6, xvec10;
+#### Testing Alignment ####
+MOVQ	C0, %rax;
+OR		ldc, %rax;
+TEST	$15, %rax;
+JNE		.L223_loopEx;
+.align 32
+#### Writing Back ####
+#ifndef	TRMMKERNEL
+ADD_DX	0*SIZE(C0), xvec11;
+ADD_DX	2*SIZE(C0), xvec10;
+ADD_DX	0*SIZE(C1), xvec15;
+ADD_DX	2*SIZE(C1), xvec14;
+#endif
+ST_DX	xvec11,	0*SIZE(C0);
+ST_DX	xvec10,	2*SIZE(C0);
+ST_DX	xvec15, 0*SIZE(C1);
+ST_DX	xvec14, 2*SIZE(C1);
+#if (defined(TRMMKERNEL)&& defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&& !defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kkk, %rax;
+LEAQ	(,%rax, SIZE), %rax;
+LEAQ	(ptrba, %rax, 4), ptrba;
+LEAQ	(ptrbb, %rax, 2), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT)
+ADDQ	$4, kk
+#endif
+ADDQ	$4*SIZE, C0;
+ADDQ	$4*SIZE, C1;
+JMP	.L22_loopE;
+.align 32
+.L223_loopEx:;
+#ifndef	TRMMKERNEL
+LDL_DX	0*SIZE(C0), xvec0;
+LDH_DX	1*SIZE(C0), xvec0;
+LDL_DX	2*SIZE(C0), xvec1;
+LDH_DX	3*SIZE(C0), xvec1;
+ADD_DX	xvec0, xvec11;
+ADD_DX	xvec1, xvec10;
+#endif
+STL_DX	xvec11, 0*SIZE(C0);
+STH_DX	xvec11, 1*SIZE(C0);
+STL_DX	xvec10, 2*SIZE(C0);
+STH_DX	xvec10, 3*SIZE(C0);
+#ifndef	TRMMKERNEL
+LDL_DX	0*SIZE(C1), xvec4;
+LDH_DX	1*SIZE(C1), xvec4;
+LDL_DX	2*SIZE(C1), xvec5;
+LDH_DX	3*SIZE(C1), xvec5;
+ADD_DX	xvec4, xvec15;
+ADD_DX	xvec5, xvec14;
+#endif
+STL_DX	xvec15, 0*SIZE(C1);
+STH_DX	xvec15, 1*SIZE(C1);
+STL_DX	xvec14, 2*SIZE(C1);
+STH_DX	xvec14, 3*SIZE(C1);
+#if (defined(TRMMKERNEL)&& defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&& !defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kkk, %rax;
+LEAQ	(,%rax, SIZE), %rax;
+LEAQ	(ptrba, %rax, 4), ptrba;
+LEAQ	(ptrbb, %rax, 2), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT)
+ADDQ	$4, kk
+#endif
+ADDQ	$4*SIZE, C0;
+ADDQ	$4*SIZE, C1;
+.L22_loopE:;
+TEST $2, bm;				# Rm = 2
+JLE .L23_loopE;
+.align 32;
+.L23_bodyB:
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb, ptrbb;
+#else 
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+LEAQ	(,%rax, SIZE), %rax;
+LEAQ	(ptrba, %rax, 2), ptrba;
+LEAQ	(ptrbb, %rax, 2), ptrbb;
+#endif
+XOR_DY yvec15, yvec15, yvec15;
+XOR_DY yvec11, yvec11, yvec11;
+#ifndef	TRMMKERNEL
+MOVQ bk, k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kk, %rax;
+MOVQ	%rax, kkk;
+#else 
+MOVQ	kk, %rax;
+#ifdef	LEFT
+ADDQ	$2, %rax;
+#else 
+ADDQ	$2, %rax;
+#endif
+MOVQ	%rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L231_loopE;
+.align 32
+.L231_bodyB:
+# Computing kernel 
+#### Unroll time 1 ####
+LD_DX	0*SIZE(ptrba), xvec0;
+LD_DX	0*SIZE(ptrbb), xvec4;
+SHUF_DX $0x4e, xvec4, xvec5;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec15;
+
+MUL_DX	xvec0, xvec5;
+ADD_DX	xvec5, xvec11;
+#### Unroll time 2 ####
+LD_DX	2*SIZE(ptrba), xvec0;
+LD_DX	2*SIZE(ptrbb), xvec4;
+SHUF_DX $0x4e, xvec4, xvec5;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec15;
+
+MUL_DX	xvec0, xvec5;
+ADD_DX	xvec5, xvec11;
+#### Unroll time 3 ####
+LD_DX	4*SIZE(ptrba), xvec0;
+LD_DX	4*SIZE(ptrbb), xvec4;
+SHUF_DX $0x4e, xvec4, xvec5;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec15;
+
+MUL_DX	xvec0, xvec5;
+ADD_DX	xvec5, xvec11;
+#### Unroll time 4 ####
+LD_DX	6*SIZE(ptrba), xvec0;
+LD_DX	6*SIZE(ptrbb), xvec4;
+SHUF_DX $0x4e, xvec4, xvec5;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec15;
+ADDQ	$8*SIZE, ptrba;
+MUL_DX	xvec0, xvec5;
+ADD_DX	xvec5, xvec11;
+ADDQ	$8*SIZE, ptrbb;
+DECQ k;
+JG .L231_bodyB;
+.align 32
+.L231_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else 
+MOVQ kkk, %rax;
+TEST $2, %rax;
+#endif
+JLE .L232_loopE;
+.align 32
+.L232_bodyB:
+#### Unroll time 1 ####
+LD_DX	0*SIZE(ptrba), xvec0;
+LD_DX	0*SIZE(ptrbb), xvec4;
+SHUF_DX $0x4e, xvec4, xvec5;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec15;
+
+MUL_DX	xvec0, xvec5;
+ADD_DX	xvec5, xvec11;
+#### Unroll time 2 ####
+LD_DX	2*SIZE(ptrba), xvec0;
+LD_DX	2*SIZE(ptrbb), xvec4;
+SHUF_DX $0x4e, xvec4, xvec5;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec15;
+ADDQ	$4*SIZE, ptrba;
+MUL_DX	xvec0, xvec5;
+ADD_DX	xvec5, xvec11;
+ADDQ	$4*SIZE, ptrbb;
+.L232_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else 
+MOVQ kkk, %rax;
+TEST $1, %rax;
+#endif
+JLE .L233_loopE;
+.align 32
+.L233_bodyB:
+#### Unroll time 1 ####
+LD_DX	0*SIZE(ptrba), xvec0;
+LD_DX	0*SIZE(ptrbb), xvec4;
+SHUF_DX $0x4e, xvec4, xvec5;
+MUL_DX	xvec0, xvec4;
+ADD_DX	xvec4, xvec15;
+ADDQ	$2*SIZE, ptrba;
+MUL_DX	xvec0, xvec5;
+ADD_DX	xvec5, xvec11;
+ADDQ	$2*SIZE, ptrbb;
+.L233_loopE:
+#### Multiply Alpha ####
+BROAD_DX MEMALPHA, xvec7;
+MUL_DX	xvec7, xvec15;
+MUL_DX	xvec7, xvec11;
+#### Reverse #####
+MOV_DX	xvec15, xvec6;
+REVS_DX	xvec11, xvec15;
+REVS_DX	xvec6, xvec11;
+#### Testing Alignment ####
+MOVQ	C0, %rax;
+OR		ldc, %rax;
+TEST	$15, %rax;
+JNE		.L233_loopEx;
+.align 32
+#### Writing Back ####
+#ifndef	TRMMKERNEL
+ADD_DX	0*SIZE(C0), xvec11;
+ADD_DX	0*SIZE(C1), xvec15;
+#endif
+ST_DX	xvec11,	0*SIZE(C0);
+ST_DX	xvec15, 0*SIZE(C1);
+#if (defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kkk, %rax;
+LEAQ	(,%rax, SIZE), %rax;
+LEAQ	(ptrba, %rax, 2), ptrba;
+LEAQ	(ptrbb, %rax, 2), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT)
+ADDQ	$2, kk;
+#endif
+ADDQ	$2*SIZE, C0;
+ADDQ	$2*SIZE, C1;
+JMP	.L23_loopE;
+.align 32
+.L233_loopEx:;
+#ifndef	TRMMKERNEL
+LDL_DX	0*SIZE(C0), xvec0;
+LDH_DX	1*SIZE(C0), xvec0;
+ADD_DX	xvec0, xvec11;
+#endif
+STL_DX	xvec11, 0*SIZE(C0);
+STH_DX	xvec11, 1*SIZE(C0);
+#ifndef	TRMMKERNEL
+LDL_DX	0*SIZE(C1), xvec4;
+LDH_DX	1*SIZE(C1), xvec4;
+ADD_DX	xvec4, xvec15;
+#endif
+STL_DX	xvec15, 0*SIZE(C1);
+STH_DX	xvec15, 1*SIZE(C1);
+#if (defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kkk, %rax;
+LEAQ	(,%rax, SIZE), %rax;
+LEAQ	(ptrba, %rax, 2), ptrba;
+LEAQ	(ptrbb, %rax, 2), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT)
+ADDQ	$2, kk;
+#endif
+ADDQ	$2*SIZE, C0;
+ADDQ	$2*SIZE, C1;
+.L23_loopE:
+TEST $1, bm;				# Rm = 1
+JLE .L24_loopE;
+.align 32;
+.L24_bodyB:
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb, ptrbb;
+#else 
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+LEAQ	(, %rax, SIZE), %rax;
+ADDQ	%rax, ptrba;
+LEAQ	(ptrbb, %rax, 2), ptrbb;
+#endif
+XOR_DY	yvec15, yvec15, yvec15;
+#ifndef	TRMMKERNEL
+MOVQ bk, k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else 
+MOVQ kk, %rax;
+#ifdef	LEFT
+ADDQ	$1, %rax;
+#else 
+ADDQ	$2, %rax;
+#endif	
+MOVQ	%rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L241_loopE;
+.align 32
+.L241_bodyB:
+BROAD_DX 0*SIZE(ptrba), xvec0;
+LD_DX	0*SIZE(ptrbb), xvec2;
+MUL_DX	xvec0, xvec2;
+ADD_DX	xvec2, xvec15;
+
+BROAD_DX 1*SIZE(ptrba), xvec1;
+LD_DX	2*SIZE(ptrbb), xvec3;
+MUL_DX	xvec1, xvec3;
+ADD_DX	xvec3, xvec15;
+
+BROAD_DX 2*SIZE(ptrba), xvec0;
+LD_DX	4*SIZE(ptrbb), xvec2;
+MUL_DX	xvec0, xvec2;
+ADD_DX	xvec2, xvec15;
+
+BROAD_DX 3*SIZE(ptrba), xvec1;
+LD_DX	6*SIZE(ptrbb), xvec3;
+MUL_DX	xvec1, xvec3;
+ADD_DX	xvec3, xvec15;
+ADDQ	$4*SIZE, ptrba;
+ADDQ	$8*SIZE, ptrbb;
+DECQ k;
+JG .L241_bodyB;
+.align 32
+.L241_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else 
+MOVQ kkk, %rax;
+TEST $2, %rax;
+#endif
+JLE .L242_loopE;
+.align 32
+.L242_bodyB:
+BROAD_DX 0*SIZE(ptrba), xvec0;
+LD_DX	0*SIZE(ptrbb), xvec2;
+MUL_DX	xvec0, xvec2;
+ADD_DX	xvec2, xvec15;
+
+BROAD_DX 1*SIZE(ptrba), xvec1;
+LD_DX	2*SIZE(ptrbb), xvec3;
+MUL_DX	xvec1, xvec3;
+ADD_DX	xvec3, xvec15;
+ADDQ	$2*SIZE, ptrba;
+ADDQ	$4*SIZE, ptrbb;
+.L242_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else 
+MOVQ kkk, %rax;
+TEST $1, %rax;
+#endif
+JLE .L243_loopE;
+.align 32
+.L243_bodyB:
+BROAD_DX 0*SIZE(ptrba), xvec0;
+LD_DX	0*SIZE(ptrbb), xvec2;
+MUL_DX	xvec0, xvec2;
+ADD_DX	xvec2, xvec15;
+ADDQ	$1*SIZE, ptrba;
+ADDQ	$2*SIZE, ptrbb;
+
+.L243_loopE:
+BROAD_DX MEMALPHA, xvec7;
+MUL_DX	xvec7, xvec15;
+#ifndef	TRMMKERNEL
+LDL_DX	0*SIZE(C0), xvec0;
+LDH_DX 	0*SIZE(C1), xvec0;
+ADD_DX	xvec0, xvec15;
+#endif
+STL_DX	xvec15, 0*SIZE(C0);
+STH_DX	xvec15, 0*SIZE(C1);
+#if (defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kkk, %rax;
+LEAQ	(,%rax, SIZE), %rax;
+ADDQ	%rax, ptrba;
+LEAQ	(ptrbb, %rax, 2), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT)
+ADDQ	$1, kk;
+#endif
+ADDQ	$1*SIZE, C0;
+ADDQ	$1*SIZE, C1;
+.L24_loopE:
+#if defined(TRMMKERNEL) && !defined(LEFT)
+ADDQ	$2, kk;
+#endif
+MOVQ bk, k;
+SALQ $4, k;
+ADDQ k, bb;
+LEAQ (C, ldc, 2), C;
+.L20_loopE:;
+TEST $1, bn;					# Rn = 1
+JLE .L30_loopE;
+.align 32
+.L30_bodyB:
+#if defined(TRMMKERNEL)&&defined(LEFT)
+MOVQ	OFFSET, %rax;
+MOVQ	%rax, kk;
+#endif
+MOVQ C, C0;
+MOVQ ba, ptrba;
+MOVQ bm, i;
+SARQ $3, i;
+JLE .L31_loopE;
+.align 32
+.L31_bodyB:
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb, ptrbb;
+#else 
+MOVQ bb, ptrbb;
+MOVQ kk, %rax
+LEAQ (, %rax, SIZE), %rax;
+LEAQ (ptrba, %rax, 8), ptrba;
+ADDQ	%rax, ptrbb;
+#endif
+#### Initial Results Register ####
+XOR_DY	yvec15, yvec15, yvec15;
+XOR_DY	yvec14, yvec14, yvec14;
+#ifndef	TRMMKERNEL
+MOVQ bk, k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else 
+MOVQ kk, %rax;
+#ifdef	LEFT
+ADDQ $8, %rax;
+#else 
+ADDQ $1, %rax;
+#endif
+MOVQ %rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L311_loopE;
+.align 32
+.L311_bodyB:
+#### Unroll time 1 ####
+LD_DY	0*SIZE(ptrba), yvec0;
+LD_DY	4*SIZE(ptrba), yvec1;
+BROAD_DY 0*SIZE(ptrbb), yvec2;
+MUL_DY	yvec2, yvec0, yvec0;
+ADD_DY	yvec0, yvec15, yvec15;
+MUL_DY	yvec2, yvec1, yvec1;
+ADD_DY	yvec1, yvec14, yvec14;
+
+#### Unroll time 2 ####
+LD_DY	8*SIZE(ptrba), yvec3;
+LD_DY	12*SIZE(ptrba), yvec4;
+BROAD_DY 1*SIZE(ptrbb), yvec5;
+MUL_DY	yvec5, yvec3, yvec3;
+ADD_DY	yvec3, yvec15, yvec15;
+MUL_DY	yvec5, yvec4, yvec4
+ADD_DY	yvec4, yvec14, yvec14;
+
+#### Unroll time 3 ####
+LD_DY	16*SIZE(ptrba), yvec0;
+LD_DY	20*SIZE(ptrba), yvec1;
+BROAD_DY 2*SIZE(ptrbb), yvec2;
+MUL_DY	yvec2, yvec0, yvec0;
+ADD_DY	yvec0, yvec15, yvec15;
+MUL_DY	yvec2, yvec1, yvec1;
+ADD_DY	yvec1, yvec14, yvec14;
+
+#### Unroll time 2 ####
+LD_DY	24*SIZE(ptrba), yvec3;
+LD_DY	28*SIZE(ptrba), yvec4;
+BROAD_DY 3*SIZE(ptrbb), yvec5;
+MUL_DY	yvec5, yvec3, yvec3;
+ADD_DY	yvec3, yvec15, yvec15;
+ADDQ	$32*SIZE, ptrba;
+MUL_DY	yvec5, yvec4, yvec4;
+ADD_DY	yvec4, yvec14, yvec14;
+ADDQ	$4*SIZE, ptrbb;
+DECQ	k;
+JG	.L311_bodyB;
+.align 32
+.L311_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else 
+MOVQ kkk, %rax;
+TEST $2, %rax;
+#endif
+JLE .L312_loopE;
+.align 32
+.L312_bodyB:
+#### Unroll time 1 ####
+LD_DY	0*SIZE(ptrba), yvec0;
+LD_DY	4*SIZE(ptrba), yvec1;
+BROAD_DY 0*SIZE(ptrbb), yvec2;
+MUL_DY	yvec2, yvec0, yvec0;
+ADD_DY	yvec0, yvec15, yvec15;
+MUL_DY	yvec2, yvec1, yvec1;
+ADD_DY	yvec1, yvec14, yvec14;
+
+#### Unroll time 2 ####
+LD_DY	8*SIZE(ptrba), yvec3;
+LD_DY	12*SIZE(ptrba), yvec4;
+BROAD_DY 1*SIZE(ptrbb), yvec5;
+MUL_DY	yvec5, yvec3, yvec3;
+ADD_DY	yvec3, yvec15, yvec15;
+ADDQ	$16*SIZE, ptrba;
+MUL_DY	yvec5, yvec4, yvec4
+ADD_DY	yvec4, yvec14, yvec14;
+ADDQ	$2*SIZE, ptrbb;
+
+.L312_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else
+MOVQ kkk, %rax;
+TEST $1, %rax;
+#endif
+JLE .L313_loopE;
+.align 32
+.L313_bodyB:
+#### Unroll time 1 ####
+LD_DY	0*SIZE(ptrba), yvec0;
+LD_DY	4*SIZE(ptrba), yvec1;
+BROAD_DY 0*SIZE(ptrbb), yvec2;
+MUL_DY	yvec2, yvec0, yvec0;
+ADD_DY	yvec0, yvec15, yvec15;
+ADDQ	$8*SIZE, ptrba;
+MUL_DY	yvec2, yvec1, yvec1;
+ADD_DY	yvec1, yvec14, yvec14;
+ADDQ	$1*SIZE, ptrbb;
+
+.L313_loopE:
+#### Multiply Alpha ####
+BROAD_DY MEMALPHA, yvec7;
+MUL_DY	yvec7, yvec15, yvec15;
+MUL_DY	yvec7, yvec14, yvec14;
+#### Testing Alignment ####
+MOVQ	C0, %rax;
+OR		ldc, %rax;
+TEST 	$15, %rax;
+JNE		.L313_loopEx;
+.align 32
+#### Writing Back ####
+EXTRA_DY $1, yvec15, xvec13;
+EXTRA_DY $1, yvec14, xvec12;
+#ifndef	TRMMKERNEL
+ADD_DX	0*SIZE(C0), xvec15;
+ADD_DX	2*SIZE(C0), xvec13;
+ADD_DX	4*SIZE(C0), xvec14;
+ADD_DX	6*SIZE(C0), xvec12;
+#endif
+ST_DX	xvec15, 0*SIZE(C0);
+ST_DX	xvec13, 2*SIZE(C0);
+ST_DX	xvec14,	4*SIZE(C0);
+ST_DX	xvec12, 6*SIZE(C0);
+#if (defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kkk, %rax;
+LEAQ (,%rax, SIZE), %rax;
+LEAQ (ptrba, %rax, 8), ptrba;
+ADDQ %rax, ptrbb;
+#endif
+#if defined(TRMMKERNEL)&&defined(LEFT)
+ADDQ	$8, kk;
+#endif
+ADDQ	$8*SIZE, C0;
+DECQ	i;
+JG	.L31_bodyB;
+JMP .L31_loopE;
+.align 32
+.L313_loopEx:
+EXTRA_DY $1, yvec15, xvec13;
+EXTRA_DY $1, yvec14, xvec12;
+#ifndef	TRMMKERNEL
+LDL_DX	0*SIZE(C0), xvec11;
+LDH_DX	1*SIZE(C0), xvec11;
+LDL_DX	2*SIZE(C0), xvec10;
+LDH_DX	3*SIZE(C0), xvec10;
+LDL_DX	4*SIZE(C0), xvec9;
+LDH_DX	5*SIZE(C0), xvec9;
+LDL_DX	6*SIZE(C0), xvec8;
+LDH_DX	7*SIZE(C0), xvec8;
+ADD_DX	xvec11, xvec15;
+ADD_DX	xvec10, xvec13;
+ADD_DX	xvec9, xvec14;
+ADD_DX	xvec8, xvec12;
+#endif
+STL_DX	xvec15, 0*SIZE(C0);
+STH_DX	xvec15, 1*SIZE(C0);
+STL_DX	xvec13, 2*SIZE(C0);
+STH_DX	xvec13, 3*SIZE(C0);
+STL_DX	xvec14, 4*SIZE(C0);
+STH_DX	xvec14, 5*SIZE(C0);
+STL_DX	xvec12, 6*SIZE(C0);
+STH_DX	xvec12, 7*SIZE(C0);
+#if (defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kkk, %rax;
+LEAQ (,%rax, SIZE), %rax;
+LEAQ (ptrba, %rax, 8), ptrba;
+ADDQ %rax, ptrbb;
+#endif
+#if defined(TRMMKERNEL)&&defined(LEFT)
+ADDQ	$8, kk;
+#endif
+ADDQ	$8*SIZE, C0;
+DECQ	i;
+JG	.L31_bodyB;
+.L31_loopE:
+TEST $4, bm
+JLE .L32_loopE;
+.align 32
+.L32_bodyB:
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb, ptrbb;
+#else
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+LEAQ (,%rax, SIZE), %rax;
+LEAQ (ptrba, %rax, 4), ptrba;
+ADDQ	%rax, ptrbb;
+#endif
+#### Initial Results Register ####
+XOR_DY	yvec15, yvec15, yvec15;
+#ifndef	TRMMKERNEL
+MOVQ bk, k;
+#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else 
+MOVQ kk, %rax;
+#ifdef	LEFT
+ADDQ $4, %rax;
+#else 
+ADDQ $1, %rax;
+#endif
+MOVQ %rax, kkk
+#endif
+SARQ $2, k;
+JLE .L321_loopE;
+.align 32
+.L321_bodyB:
+LD_DY	0*SIZE(ptrba), yvec0;
+BROAD_DY 0*SIZE(ptrbb), yvec1;
+MUL_DY	yvec0, yvec1, yvec1;
+ADD_DY	yvec1, yvec15, yvec15;
+
+LD_DY	4*SIZE(ptrba), yvec2;
+BROAD_DY 1*SIZE(ptrbb), yvec3;
+MUL_DY	yvec2, yvec3, yvec3;
+ADD_DY	yvec3, yvec15, yvec15;
+
+LD_DY	8*SIZE(ptrba), yvec4;
+BROAD_DY 2*SIZE(ptrbb), yvec5;
+MUL_DY	yvec4, yvec5, yvec5;
+ADD_DY	yvec5, yvec15, yvec15;
+
+LD_DY	12*SIZE(ptrba), yvec6;
+BROAD_DY 3*SIZE(ptrbb), yvec7;
+MUL_DY	yvec6, yvec7, yvec7;
+ADD_DY	yvec7, yvec15, yvec15;
+ADDQ	$16*SIZE, ptrba;
+ADDQ	$4*SIZE, ptrbb;
+DECQ	k;
+JG .L321_bodyB;
+.align 32
+.L321_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else 
+MOVQ kkk, %rax;
+TEST $2, %rax;
+#endif
+JLE .L322_loopE;
+.align 32
+.L322_bodyB:
+LD_DY	0*SIZE(ptrba), yvec0;
+BROAD_DY 0*SIZE(ptrbb), yvec1;
+MUL_DY	yvec0, yvec1, yvec1;
+ADD_DY	yvec1, yvec15, yvec15;
+
+LD_DY	4*SIZE(ptrba), yvec2;
+BROAD_DY 1*SIZE(ptrbb), yvec3;
+MUL_DY	yvec2, yvec3, yvec3;
+ADD_DY	yvec3, yvec15, yvec15;
+ADDQ	$8*SIZE, ptrba;
+ADDQ	$2*SIZE, ptrbb;
+
+.L322_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else 
+MOVQ kkk, %rax;
+TEST $1, %rax;
+#endif
+JLE .L323_loopE;
+.align 32
+.L323_bodyB:
+LD_DY	0*SIZE(ptrba), yvec0;
+BROAD_DY 0*SIZE(ptrbb), yvec1;
+MUL_DY	yvec0, yvec1, yvec1;
+ADD_DY	yvec1, yvec15, yvec15;
+ADDQ	$4*SIZE, ptrba;
+ADDQ	$1*SIZE, ptrbb;
+
+.L323_loopE:
+#### Multiply Alpha ####
+BROAD_DY MEMALPHA, yvec7;
+MUL_DY	yvec7, yvec15, yvec15;
+#### Testing Alignment ####
+MOVQ	C0, %rax;
+OR		ldc, %rax;
+TEST	$15, %rax;
+JNE	.L323_loopEx;
+.align 32
+#### Writing Back ####
+EXTRA_DY $1, yvec15, xvec14;
+#ifndef	TRMMKERNEL
+ADD_DX	0*SIZE(C0), xvec15;
+ADD_DX	2*SIZE(C0), xvec14;
+#endif
+ST_DX	xvec15, 0*SIZE(C0);
+ST_DX	xvec14, 2*SIZE(C0);
+#if (defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kkk, %rax;
+LEAQ (, %rax, SIZE), %rax;
+LEAQ (ptrba, %rax, 4), ptrba;
+ADDQ %rax, ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT)
+ADDQ $4, kk
+#endif
+ADDQ	$4*SIZE, C0;
+JMP .L32_loopE;
+.align 32
+.L323_loopEx:
+#### Writing Back ####
+EXTRA_DY $1, yvec15, xvec14;
+#ifndef	TRMMKERNEL
+LDL_DX	0*SIZE(C0), xvec13;
+LDH_DX	1*SIZE(C0), xvec13;
+LDL_DX	2*SIZE(C0), xvec12;
+LDH_DX	3*SIZE(C0), xvec12;
+ADD_DX	xvec13, xvec15;
+ADD_DX	xvec12, xvec14;
+#endif
+STL_DX	xvec15, 0*SIZE(C0);
+STH_DX	xvec15, 1*SIZE(C0);
+STL_DX	xvec14, 2*SIZE(C0);
+STH_DX	xvec14, 3*SIZE(C0);
+#if (defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kkk, %rax;
+LEAQ (, %rax, SIZE), %rax;
+LEAQ (ptrba, %rax, 4), ptrba;
+ADDQ %rax, ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT)
+ADDQ $4, kk
+#endif
+ADDQ	$4*SIZE, C0;
+.L32_loopE:
+TEST $2, bm
+JLE .L33_loopE;
+.align 32
+.L33_bodyB:
+#if !defined(TRMMKERNEL) || (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+MOVQ bb, ptrbb;
+#else 
+MOVQ bb, ptrbb;
+MOVQ kk, %rax
+LEAQ (, %rax, SIZE), %rax
+LEAQ (ptrba, %rax, 2), ptrba
+ADDQ %rax, ptrbb;
+#endif 
+#### Initial Result ####
+XOR_DY	yvec15, yvec15, yvec15;
+#ifndef	TRMMKERNEL
+MOVQ bk, k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else 
+MOVQ kk, %rax;
+#ifdef	LEFT
+ADDQ $2, %rax;
+#else
+ADDQ $1, %rax;
+#endif
+MOVQ %rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L331_loopE;
+.align 32
+.L331_bodyB:
+LD_DX	0*SIZE(ptrba), xvec0;
+BROAD_DX 0*SIZE(ptrbb), xvec2;
+MUL_DX	xvec0, xvec2;
+ADD_DX	xvec2, xvec15;
+
+LD_DX	2*SIZE(ptrba), xvec1;
+BROAD_DX 1*SIZE(ptrbb), xvec3;
+MUL_DX	xvec1, xvec3;
+ADD_DX	xvec3, xvec15;
+
+LD_DX	4*SIZE(ptrba), xvec4;
+BROAD_DX 2*SIZE(ptrbb), xvec5;
+MUL_DX	xvec4, xvec5;
+ADD_DX	xvec5, xvec15;
+
+LD_DX	6*SIZE(ptrba), xvec6;
+BROAD_DX 3*SIZE(ptrbb), xvec7;
+MUL_DX	xvec6, xvec7;
+ADD_DX	xvec7, xvec15;
+ADDQ	$8*SIZE, ptrba;
+ADDQ	$4*SIZE, ptrbb;
+DECQ	k;
+JG	.L331_bodyB;
+.align 32
+.L331_loopE:
+#ifndef	TRMMKERNEL
+TEST $2,bk;
+#else 
+MOVQ kkk, %rax;
+TEST $2, %rax
+#endif
+JLE .L332_loopE;
+.align 32
+.L332_bodyB:
+LD_DX	0*SIZE(ptrba), xvec0;
+BROAD_DX 0*SIZE(ptrbb), xvec2;
+MUL_DX	xvec0, xvec2;
+ADD_DX	xvec2, xvec15;
+
+LD_DX	2*SIZE(ptrba), xvec1;
+BROAD_DX 1*SIZE(ptrbb), xvec3;
+MUL_DX	xvec1, xvec3;
+ADD_DX	xvec3, xvec15;
+ADDQ	$4*SIZE, ptrba;
+ADDQ	$2*SIZE, ptrbb;
+.L332_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else 
+MOVQ kkk, %rax;
+TEST $1, %rax;
+#endif
+JLE .L333_loopE;
+.align 32
+.L333_bodyB:
+LD_DX	0*SIZE(ptrba), xvec0;
+BROAD_DX 0*SIZE(ptrbb), xvec2;
+MUL_DX	xvec0, xvec2;
+ADD_DX	xvec2, xvec15;
+ADDQ	$2*SIZE, ptrba;
+ADDQ	$1*SIZE, ptrbb;
+.L333_loopE:
+#### Multiply Alpha ####
+BROAD_DX MEMALPHA, xvec7;
+MUL_DX	xvec7, xvec15;
+#ifndef	TRMMKERNEL
+LDL_DX	0*SIZE(C0), xvec14;
+LDH_DX	1*SIZE(C0), xvec14;
+ADD_DX	xvec14, xvec15;
+#endif
+STL_DX	xvec15, 0*SIZE(C0);
+STH_DX	xvec15, 1*SIZE(C0);
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) ||(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kkk, %rax;
+LEAQ (,%rax, SIZE), %rax;
+LEAQ (ptrba, %rax, 2), ptrba;
+ADDQ %rax, ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT)
+addq $2, kk
+#endif
+ADDQ	$2*SIZE, C0;
+.L33_loopE:
+TEST $1, bm
+JLE .L34_loopE;
+.align 32
+.L34_bodyB:
+#if !defined(TRMMKERNEL) || (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA)) 
+MOVQ bb, ptrbb;
+#else 
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+LEAQ (, %rax, SIZE), %rax;
+ADDQ %rax, ptrba;
+ADDQ %rax, ptrbb;
+#endif
+XOR_DY	yvec15, yvec15, yvec15;
+#ifndef	TRMMKERNEL
+MOVQ bk, k;
+#elif (defined(LEFT)&& !defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else 
+MOVQ kk, %rax;
+#ifdef	LEFT
+ADDQ $1, %rax;
+#else
+ADDQ $1, %rax;
+#endif
+MOVQ %rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L341_loopE;
+.align 32
+.L341_bodyB:
+movsd	0*SIZE(ptrba), xvec0;
+movsd 	0*SIZE(ptrbb), xvec1;
+mulsd	xvec0, xvec1;
+addsd	xvec1, xvec15;
+
+movsd	1*SIZE(ptrba), xvec0;
+movsd 	1*SIZE(ptrbb), xvec1;
+mulsd	xvec0, xvec1;
+addsd	xvec1, xvec15;
+
+movsd	2*SIZE(ptrba), xvec0;
+movsd 	2*SIZE(ptrbb), xvec1;
+mulsd	xvec0, xvec1;
+addsd	xvec1, xvec15;
+
+movsd	3*SIZE(ptrba), xvec0;
+movsd 	3*SIZE(ptrbb), xvec1;
+mulsd	xvec0, xvec1;
+addsd	xvec1, xvec15;
+addq	$4*SIZE, ptrba;
+addq	$4*SIZE, ptrbb;
+decq	k;
+JG	.L341_bodyB;
+.align 32
+.L341_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else 
+MOVQ kkk, %rax;
+TEST $2, %rax;
+#endif
+JLE .L342_loopE;
+.align 32
+.L342_bodyB:
+movsd	0*SIZE(ptrba), xvec0;
+movsd 	0*SIZE(ptrbb), xvec1;
+mulsd	xvec0, xvec1;
+addsd	xvec1, xvec15;
+
+movsd	1*SIZE(ptrba), xvec0;
+movsd 	1*SIZE(ptrbb), xvec1;
+mulsd	xvec0, xvec1;
+addsd	xvec1, xvec15;
+addq	$2*SIZE, ptrba;
+addq	$2*SIZE, ptrbb;
+
+.L342_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk
+#else
+MOVQ kkk, %rax;
+TEST $1, %rax;
+#endif
+JLE .L343_loopE;
+.align 32
+.L343_bodyB:
+movsd	0*SIZE(ptrba), xvec0;
+movsd 	0*SIZE(ptrbb), xvec1;
+mulsd	xvec0, xvec1;
+addsd	xvec1, xvec15;
+addq	$1*SIZE, ptrba;
+addq	$1*SIZE, ptrbb;
+
+.L343_loopE:
+#### Writing Back ####
+movsd	MEMALPHA, xvec7;
+mulsd	xvec7, xvec15;
+#ifndef	TRMMKERNEL
+movsd	0*SIZE(C0), xvec0;
+addsd	xvec0, xvec15;
+#endif
+movsd	xvec15, 0*SIZE(C0);
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) ||(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kkk, %rax;
+LEAQ (,%rax, SIZE), %rax;
+ADDQ %rax, ptrba;
+ADDQ %rax, ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT)
+addq $1, kk
+#endif
+addq	$1*SIZE, C0;
+.L34_loopE:
+MOVQ bk, k
+SALQ $3, k;
+ADDQ k, bb;
+LEAQ (C, ldc, 1), C;
+
+.L30_loopE:
+movq      0(%rsp), %rbx;
+movq      8(%rsp), %rbp;
+movq     16(%rsp), %r12;
+movq     24(%rsp), %r13;
+movq     32(%rsp), %r14;
+movq     40(%rsp), %r15;
+#ifdef WINDOWS_ABI
+	movq	 48(%rsp), %rdi
+	movq	 56(%rsp), %rsi
+	movups	 64(%rsp), %xmm6
+	movups	 80(%rsp), %xmm7
+	movups	 96(%rsp), %xmm8
+	movups	112(%rsp), %xmm9
+	movups	128(%rsp), %xmm10
+	movups	144(%rsp), %xmm11
+	movups	160(%rsp), %xmm12
+	movups	176(%rsp), %xmm13
+	movups	192(%rsp), %xmm14
+	movups	208(%rsp), %xmm15
+#endif
+addq    $STACKSIZE, %rsp;
+ret
+
+EPILOGUE
diff --git a/kernel/x86_64/sgemm_kernel_8x8_sandy.S b/kernel/x86_64/sgemm_kernel_8x8_sandy.S
new file mode 100644
index 000000000..44f8f1802
--- /dev/null
+++ b/kernel/x86_64/sgemm_kernel_8x8_sandy.S
@@ -0,0 +1,3736 @@
+/*****************************************************************************
+  Copyright (c) 2011,2012 Lab of Parallel Software and Computational Science,ISCAS
+  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 ISCAS 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 COPYRIGHT OWNER 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 old_bm %rdi
+#define old_bn %rsi
+#define old_bk %rdx
+
+#define bm %r13
+#define bn %r14
+#define bk %r15
+
+#define ALPHA %xmm0
+#define ba %rcx
+#define bb %r8
+#define C %r9
+#define ldc %r10
+
+#define i %r11
+#define k %rax
+
+#define ptrba %rdi
+#define ptrbb %rsi
+#define C0 %rbx
+#define C1 %rbp
+
+#define prebb %r12
+
+#ifndef WINDOWS_ABI
+
+#define STACKSIZE 128
+
+#define old_ldc 8+STACKSIZE(%rsp)
+#define old_offset 16+STACKSIZE(%rsp)
+#define MEMALPHA 48(%rsp)
+#define j 56(%rsp)
+#define OFFSET 64(%rsp)
+#define kk 72(%rsp)
+#define kkk 80(%rsp)
+
+#else
+
+#define STACKSIZE 512
+
+#define OLD_A		40 + STACKSIZE(%rsp)
+#define OLD_B		48 + STACKSIZE(%rsp)
+#define OLD_C		56 + STACKSIZE(%rsp)
+#define old_ldc		64 + STACKSIZE(%rsp)
+#define old_offset	72 + STACKSIZE(%rsp)
+
+#define MEMALPHA	  224(%rsp)
+#define j	  232(%rsp)
+#define OFFSET	  240(%rsp)
+#define kk	  248(%rsp)
+#define kkk	  256(%rsp)
+
+#endif
+
+#define PREFETCH0 prefetcht0
+#define PREFETCH1 prefetcht0
+#define PREFETCH2 prefetcht2
+#define PRESIZE	80 
+
+#define xvec0	%xmm0
+#define xvec1	%xmm1
+#define xvec2	%xmm2
+#define xvec3	%xmm3
+#define xvec4	%xmm4
+#define xvec5	%xmm5
+#define xvec6	%xmm6
+#define xvec7	%xmm7
+#define xvec8	%xmm8
+#define xvec9	%xmm9
+#define xvec10	%xmm10
+#define xvec11	%xmm11
+#define xvec12	%xmm12
+#define xvec13	%xmm13
+#define xvec14	%xmm14
+#define xvec15	%xmm15
+
+#define yvec0	%ymm0
+#define yvec1	%ymm1
+#define yvec2	%ymm2
+#define yvec3	%ymm3
+#define yvec4	%ymm4
+#define yvec5	%ymm5
+#define yvec6	%ymm6
+#define yvec7	%ymm7
+#define yvec8	%ymm8
+#define yvec9	%ymm9
+#define yvec10	%ymm10
+#define yvec11	%ymm11
+#define yvec12	%ymm12
+#define yvec13	%ymm13
+#define yvec14	%ymm14
+#define yvec15	%ymm15
+
+#define LEAQ	leaq
+#define ADDQ	addq
+#define MULQ	imulq
+#define SARQ	sarq
+#define SALQ	salq
+#define ANDQ	andq
+#define SUBQ	subq
+#define DECQ	decq
+#define JG		jg
+#define JLE		jle
+#define TEST	testq
+#define OR		orq
+#define JNE		jne
+#define JMP		jmp
+#define NOP		
+#define XOR		xorpd
+#define MOVQ	movq 
+
+#define XOR_SY	vxorps
+#define XOR_DY	vxorpd
+#define XOR_SX	xorps
+#define XOR_DX	xorpd
+
+#define LD_SY	vmovaps
+#define LD_DY	vmovapd
+#define LD_SX	movaps
+#define LD_DX	movapd
+#define LDL_SX	movlps
+#define LDL_SY	vmovlps
+#define LDH_SX	movhps
+#define LDH_SY	vmovhps
+
+#define ST_SY	vmovaps
+#define ST_DY	vmovapd
+#define ST_SX	movaps
+#define ST_DX	movapd
+#define STL_SX	movlps
+#define STL_SY	vmovlps
+#define STH_SX	movhps
+#define STH_SY	vmovhps
+
+#define EDUP_SY	vmovsldup
+#define ODUP_SY	vmovshdup
+#define EDUP_SX	movsldup
+#define ODUP_SX	movshdup
+#define EDUP_DY	vmovddup
+
+#define ADD_SY	vaddps 
+#define ADD_DY	vaddpd
+#define ADD_SX	addps
+#define ADD_DX	addpd
+
+#define ADD1_DY	vaddpd
+#define ADD2_DY	vaddpd
+#define ADDSUB_DY	vaddsubpd
+#define ADDSUB_SY	vaddsubps
+
+#define MUL_SY	vmulps
+#define MUL_DY	vmulpd
+#define MUL_SX	mulps
+#define MUL_DX	mulpd
+
+#define SHUF_SY	vperm2f128
+#define SHUF_DY	vperm2f128
+#define SHUF_DX	pshufd
+#define SHUF_SX	pshufd
+
+#define VPERMILP_SY	vpermilps
+#define VPERMILP_SX	vpermilps
+#define VPERMILP_DY	vpermilpd
+
+#define BROAD_SY vbroadcastss
+#define BROAD_DY vbroadcastsd
+#define BROAD_SX vbroadcastss  
+#define BROAD_DX movddup
+
+#define MOV_SY vmovaps
+#define MOV_DY vmovapd
+#define MOV_SX movaps
+#define MOV_DX movapd
+
+#define REVS_SY	vshufps
+#define REVS_DY vshufpd
+#define REVS_SX shufps
+#define REVS_DX	movsd
+
+#define EXTRA_SY vextractf128
+#define EXTRA_DY vextractf128
+
+
+PROLOGUE
+
+subq    $STACKSIZE, %rsp;
+movq    %rbx,  0(%rsp);
+movq    %rbp,  8(%rsp);
+movq    %r12, 16(%rsp);
+movq    %r13, 24(%rsp);
+movq    %r14, 32(%rsp);
+movq    %r15, 40(%rsp);
+
+#ifdef WINDOWS_ABI
+	movq	%rdi,    48(%rsp)
+	movq	%rsi,    56(%rsp)
+	movups	%xmm6,   64(%rsp)
+	movups	%xmm7,   80(%rsp)
+	movups	%xmm8,   96(%rsp)
+	movups	%xmm9,  112(%rsp)
+	movups	%xmm10, 128(%rsp)
+	movups	%xmm11, 144(%rsp)
+	movups	%xmm12, 160(%rsp)
+	movups	%xmm13, 176(%rsp)
+	movups	%xmm14, 192(%rsp)
+	movups	%xmm15, 208(%rsp)
+
+	movq	ARG1,      old_bm
+	movq	ARG2,      old_bn
+	movq	ARG3,      old_bk
+	movq	OLD_A,     ba 
+	movq	OLD_B,     bb
+	movq	OLD_C,     C
+	movq	old_ldc,   ldc
+#ifdef TRMMKERNEL
+	movq	old_offset, %r11
+#endif
+#else
+
+movq	old_ldc, ldc
+#ifdef	TRMMKERNEL
+movq	old_offset, %r11
+#endif
+#endif
+
+vmovlps	ALPHA, MEMALPHA
+movq	old_bm,	bm
+movq	old_bn, bn
+movq	old_bk, bk
+leaq	(, ldc, SIZE), ldc
+#ifdef	TRMMKERNEL
+movq	%r11, OFFSET
+#ifndef	LEFT
+negq	%r11;
+#endif
+movq	%r11, kk
+#endif
+MOVQ bn,j;
+SARQ $3,j;
+JLE .L0_loopE;
+.align 16;
+.L0_bodyB:;
+#if defined(TRMMKERNEL) && defined(LEFT)
+MOVQ OFFSET, %rax;
+MOVQ %rax, kk;
+#endif
+
+MOVQ C,C0;
+LEAQ (C,ldc,4),C1;
+MOVQ bk, k;
+SALQ $5, k;
+LEAQ (bb, k, 1), prebb;
+MOVQ ba,ptrba;
+MOVQ bm,i;
+SARQ $3,i;
+JLE .L1_loopE;
+.align 16;
+.L1_bodyB:;
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb, ptrbb;
+#else 
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+LEAQ (, %rax, SIZE), %rax;
+LEAQ (ptrba, %rax, 8), ptrba;
+LEAQ (ptrbb, %rax, 8), ptrbb;
+#endif
+#### Initial Results Register ####
+XOR_SY yvec15, yvec15, yvec15;
+PREFETCH0	0*SIZE(prebb);
+XOR_SY yvec14, yvec14, yvec14;
+PREFETCH0	16*SIZE(prebb);
+XOR_SY yvec13, yvec13, yvec13;
+PREFETCH0	32*SIZE(prebb);
+XOR_SY yvec12, yvec12, yvec12;
+ADDQ	$48*SIZE, prebb;
+EDUP_SY	0*SIZE(ptrbb), yvec2;
+LEAQ	(ldc, ldc, 2), %rax;
+PREFETCH2	7*SIZE(C0);
+PREFETCH2	7*SIZE(C1);
+XOR_SY yvec11, yvec11, yvec11;
+XOR_SY yvec10, yvec10, yvec10;
+PREFETCH2	7*SIZE(C0, ldc, 1);
+PREFETCH2	7*SIZE(C1, ldc, 1);
+LD_SY	0*SIZE(ptrba), yvec0;
+XOR_SY yvec9, yvec9, yvec9;
+PREFETCH2	7*SIZE(C0, ldc, 2);
+PREFETCH2	7*SIZE(C1, ldc, 2);
+XOR_SY yvec8, yvec8, yvec8;
+VPERMILP_SY	$0x4e, yvec2, yvec3;
+PREFETCH2	7*SIZE(C0, %rax, 1);
+PREFETCH2	7*SIZE(C1, %rax, 1);
+#ifndef	TRMMKERNEL
+MOVQ bk,k;
+#elif (defined(LEFT) && !defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else
+MOVQ kk, %rax;
+#ifdef LEFT
+ADDQ $8, %rax;
+#else 
+ADDQ $8, %rax;
+#endif
+MOVQ %rax, kkk;
+#endif
+SARQ $2,k;
+JLE .L2_loopE;
+.align 16;
+.L2_bodyB:;
+# Computing kernel
+
+#### Unroll times 1 #### 
+PREFETCH0	PRESIZE*SIZE(ptrba);
+MUL_SY	yvec0, yvec2, yvec6;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;
+ODUP_SY	0*SIZE(ptrbb), yvec2
+MUL_SY	yvec0, yvec3, yvec7;
+SHUF_SY	$0x03, yvec3, yvec3, yvec5
+ADD_SY	yvec15, yvec6, yvec15
+ADD_SY	yvec13, yvec7, yvec13;
+
+LD_SY	8*SIZE(ptrba), yvec1;
+VPERMILP_SY	$0x4e, yvec2, yvec3;
+MUL_SY	yvec0, yvec4, yvec6;
+MUL_SY	yvec0, yvec5, yvec7;
+ADD_SY	yvec11, yvec6, yvec11;
+ADD_SY	yvec9, yvec7, yvec9;
+
+MUL_SY	yvec0, yvec2, yvec6;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;
+EDUP_SY	8*SIZE(ptrbb), yvec2;
+MUL_SY	yvec0, yvec3, yvec7;
+SHUF_SY	$0x03, yvec3, yvec3, yvec5;
+ADD_SY	yvec14, yvec6, yvec14;
+ADD_SY	yvec12, yvec7, yvec12;
+
+VPERMILP_SY	$0x4e, yvec2, yvec3;
+MUL_SY	yvec0, yvec4, yvec6;
+MUL_SY	yvec0, yvec5, yvec7;
+ADD_SY	yvec10, yvec6, yvec10;
+ADD_SY	yvec8, yvec7, yvec8;
+
+#### Unroll times 2 #### 
+MUL_SY	yvec1, yvec2, yvec6;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;
+ODUP_SY	8*SIZE(ptrbb), yvec2
+MUL_SY	yvec1, yvec3, yvec7;
+SHUF_SY	$0x03, yvec3, yvec3, yvec5
+ADD_SY	yvec15, yvec6, yvec15
+ADD_SY	yvec13, yvec7, yvec13;
+
+LD_SY	16*SIZE(ptrba), yvec0;
+VPERMILP_SY	$0x4e, yvec2, yvec3;
+MUL_SY	yvec1, yvec4, yvec6;
+MUL_SY	yvec1, yvec5, yvec7;
+ADD_SY	yvec11, yvec6, yvec11;
+ADD_SY	yvec9, yvec7, yvec9;
+
+MUL_SY	yvec1, yvec2, yvec6;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;
+EDUP_SY	16*SIZE(ptrbb), yvec2;
+MUL_SY	yvec1, yvec3, yvec7;
+SHUF_SY	$0x03, yvec3, yvec3, yvec5;
+ADD_SY	yvec14, yvec6, yvec14;
+ADD_SY	yvec12, yvec7, yvec12;
+
+VPERMILP_SY	$0x4e, yvec2, yvec3;
+MUL_SY	yvec1, yvec4, yvec6;
+MUL_SY	yvec1, yvec5, yvec7;
+ADD_SY	yvec10, yvec6, yvec10;
+ADD_SY	yvec8, yvec7, yvec8;
+
+#### Unroll times 3 #### 
+PREFETCH0	(PRESIZE+16)*SIZE(ptrba);
+MUL_SY	yvec0, yvec2, yvec6;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;
+ODUP_SY	16*SIZE(ptrbb), yvec2
+MUL_SY	yvec0, yvec3, yvec7;
+SHUF_SY	$0x03, yvec3, yvec3, yvec5
+ADD_SY	yvec15, yvec6, yvec15
+ADD_SY	yvec13, yvec7, yvec13;
+
+LD_SY	24*SIZE(ptrba), yvec1;
+VPERMILP_SY	$0x4e, yvec2, yvec3;
+MUL_SY	yvec0, yvec4, yvec6;
+MUL_SY	yvec0, yvec5, yvec7;
+ADD_SY	yvec11, yvec6, yvec11;
+ADD_SY	yvec9, yvec7, yvec9;
+ADDQ	$32*SIZE, ptrba;
+
+MUL_SY	yvec0, yvec2, yvec6;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;
+MUL_SY	yvec0, yvec3, yvec7;
+SHUF_SY	$0x03, yvec3, yvec3, yvec5;
+EDUP_SY	24*SIZE(ptrbb), yvec2;
+ADD_SY	yvec14, yvec6, yvec14;
+ADD_SY	yvec12, yvec7, yvec12;
+
+VPERMILP_SY	$0x4e, yvec2, yvec3;
+MUL_SY	yvec0, yvec4, yvec6;
+MUL_SY	yvec0, yvec5, yvec7;
+ADD_SY	yvec10, yvec6, yvec10;
+ADD_SY	yvec8, yvec7, yvec8;
+
+#### Unroll times 4 #### 
+MUL_SY	yvec1, yvec2, yvec6;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;
+ODUP_SY	24*SIZE(ptrbb), yvec2
+MUL_SY	yvec1, yvec3, yvec7;
+SHUF_SY	$0x03, yvec3, yvec3, yvec5
+ADDQ	$32*SIZE, ptrbb;
+ADD_SY	yvec15, yvec6, yvec15
+ADD_SY	yvec13, yvec7, yvec13;
+
+LD_SY	0*SIZE(ptrba), yvec0;
+VPERMILP_SY	$0x4e, yvec2, yvec3;
+MUL_SY	yvec1, yvec4, yvec6;
+MUL_SY	yvec1, yvec5, yvec7;
+ADD_SY	yvec11, yvec6, yvec11;
+ADD_SY	yvec9, yvec7, yvec9;
+
+MUL_SY	yvec1, yvec2, yvec6;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;
+EDUP_SY	0*SIZE(ptrbb), yvec2;
+MUL_SY	yvec1, yvec3, yvec7;
+SHUF_SY	$0x03, yvec3, yvec3, yvec5;
+ADD_SY	yvec14, yvec6, yvec14;
+ADD_SY	yvec12, yvec7, yvec12;
+
+VPERMILP_SY	$0x4e, yvec2, yvec3;
+MUL_SY	yvec1, yvec4, yvec6;
+MUL_SY	yvec1, yvec5, yvec7;
+ADD_SY	yvec10, yvec6, yvec10;
+ADD_SY	yvec8, yvec7, yvec8;
+.L2_bodyE:;
+DECQ k;
+JG .L2_bodyB;
+.align	64;
+.L2_loopE:;
+#ifndef	TRMMKERNEL
+TEST	$2, bk;
+#else
+TEST	$2, kkk;
+#endif
+JLE		.L3_loopE;
+.align	64
+.L3_loobB:
+#### Unroll times 1 #### 
+MUL_SY	yvec0, yvec2, yvec6;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;
+ODUP_SY	0*SIZE(ptrbb), yvec2
+MUL_SY	yvec0, yvec3, yvec7;
+SHUF_SY	$0x03, yvec3, yvec3, yvec5
+ADD_SY	yvec15, yvec6, yvec15
+ADD_SY	yvec13, yvec7, yvec13;
+
+LD_SY	8*SIZE(ptrba), yvec1;
+VPERMILP_SY	$0x4e, yvec2, yvec3;
+MUL_SY	yvec0, yvec4, yvec6;
+MUL_SY	yvec0, yvec5, yvec7;
+ADDQ	$16*SIZE, ptrba;
+ADD_SY	yvec11, yvec6, yvec11;
+ADD_SY	yvec9, yvec7, yvec9;
+
+MUL_SY	yvec0, yvec2, yvec6;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;
+EDUP_SY	8*SIZE(ptrbb), yvec2;
+MUL_SY	yvec0, yvec3, yvec7;
+SHUF_SY	$0x03, yvec3, yvec3, yvec5;
+ADD_SY	yvec14, yvec6, yvec14;
+ADD_SY	yvec12, yvec7, yvec12;
+
+VPERMILP_SY	$0x4e, yvec2, yvec3;
+MUL_SY	yvec0, yvec4, yvec6;
+MUL_SY	yvec0, yvec5, yvec7;
+ADD_SY	yvec10, yvec6, yvec10;
+ADD_SY	yvec8, yvec7, yvec8;
+
+#### Unroll times 2 #### 
+MUL_SY	yvec1, yvec2, yvec6;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;
+ODUP_SY	8*SIZE(ptrbb), yvec2
+MUL_SY	yvec1, yvec3, yvec7;
+SHUF_SY	$0x03, yvec3, yvec3, yvec5
+ADDQ	$16*SIZE, ptrbb
+ADD_SY	yvec15, yvec6, yvec15
+ADD_SY	yvec13, yvec7, yvec13;
+
+LD_SY	0*SIZE(ptrba), yvec0;
+VPERMILP_SY	$0x4e, yvec2, yvec3;
+MUL_SY	yvec1, yvec4, yvec6;
+MUL_SY	yvec1, yvec5, yvec7;
+ADD_SY	yvec11, yvec6, yvec11;
+ADD_SY	yvec9, yvec7, yvec9;
+
+MUL_SY	yvec1, yvec2, yvec6;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;
+EDUP_SY	0*SIZE(ptrbb), yvec2;
+MUL_SY	yvec1, yvec3, yvec7;
+SHUF_SY	$0x03, yvec3, yvec3, yvec5;
+ADD_SY	yvec14, yvec6, yvec14;
+ADD_SY	yvec12, yvec7, yvec12;
+
+VPERMILP_SY	$0x4e, yvec2, yvec3;
+MUL_SY	yvec1, yvec4, yvec6;
+MUL_SY	yvec1, yvec5, yvec7;
+ADD_SY	yvec10, yvec6, yvec10;
+ADD_SY	yvec8, yvec7, yvec8;
+.L3_loopE:
+#ifndef	TRMMKERNEL
+TEST	$1, bk;
+#else
+TEST	$1, kkk;
+#endif
+JLE		.L4_loopE;
+.align	64
+.L4_loopB:;
+#### Unroll times 1 #### 
+MUL_SY	yvec0, yvec2, yvec6;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;
+ODUP_SY	0*SIZE(ptrbb), yvec2
+MUL_SY	yvec0, yvec3, yvec7;
+SHUF_SY	$0x03, yvec3, yvec3, yvec5
+ADDQ	$8*SIZE, ptrba;
+ADD_SY	yvec15, yvec6, yvec15
+ADD_SY	yvec13, yvec7, yvec13;
+
+VPERMILP_SY	$0x4e, yvec2, yvec3;
+MUL_SY	yvec0, yvec4, yvec6;
+MUL_SY	yvec0, yvec5, yvec7;
+ADDQ	$8*SIZE, ptrbb;
+ADD_SY	yvec11, yvec6, yvec11;
+ADD_SY	yvec9, yvec7, yvec9;
+
+MUL_SY	yvec0, yvec2, yvec6;
+SHUF_SY	$0x03, yvec2, yvec2, yvec4;
+MUL_SY	yvec0, yvec3, yvec7;
+SHUF_SY	$0x03, yvec3, yvec3, yvec5;
+ADD_SY	yvec14, yvec6, yvec14;
+ADD_SY	yvec12, yvec7, yvec12;
+
+MUL_SY	yvec0, yvec4, yvec6;
+MUL_SY	yvec0, yvec5, yvec7;
+ADD_SY	yvec10, yvec6, yvec10;
+ADD_SY	yvec8, yvec7, yvec8;
+
+.L4_loopE:;
+#### Load Alpha ####
+BROAD_SY MEMALPHA,yvec7;
+MUL_SY yvec7,yvec15,yvec15;
+MUL_SY yvec7,yvec14,yvec14;
+MUL_SY yvec7,yvec13,yvec13;
+MUL_SY yvec7,yvec12,yvec12;
+MUL_SY yvec7,yvec11,yvec11;
+MUL_SY yvec7,yvec10,yvec10;
+MUL_SY yvec7,yvec9,yvec9;
+MUL_SY yvec7,yvec8,yvec8;
+MOV_SY yvec15,yvec7;
+REVS_SY $0xe4,yvec13,yvec15,yvec15;
+REVS_SY $0xe4,yvec7,yvec13,yvec13;
+MOV_SY yvec14,yvec7;
+REVS_SY $0xe4,yvec12,yvec14,yvec14;
+REVS_SY $0xe4,yvec7,yvec12,yvec12;
+MOV_SY yvec11,yvec7;
+REVS_SY $0xe4,yvec9,yvec11,yvec11;
+REVS_SY $0xe4,yvec7,yvec9,yvec9;
+MOV_SY yvec10,yvec7;
+REVS_SY $0xe4,yvec8,yvec10,yvec10;
+REVS_SY $0xe4,yvec7,yvec8,yvec8;
+##### Testing alignment #####
+MOVQ	C0, %rax;
+OR		ldc, %rax;
+TEST	$15, %rax;
+JNE		.L4_loopEx;
+.align 16
+LEAQ (ldc,ldc,2),%rax;
+EXTRA_SY $1,yvec15,xvec7;
+EXTRA_SY $1,yvec14,xvec6;
+EXTRA_SY $1,yvec13,xvec5;
+EXTRA_SY $1,yvec12,xvec4;
+EXTRA_SY $1,yvec11,xvec3;
+EXTRA_SY $1,yvec10,xvec2;
+EXTRA_SY $1,yvec9,xvec1;
+EXTRA_SY $1,yvec8,xvec0;
+#ifndef	TRMMKERNEL
+ADD_SY 0*SIZE(C0), xvec15, xvec15;
+ADD_SY 4*SIZE(C1), xvec7, xvec7;
+ADD_SY 0*SIZE(C0,ldc,1), xvec14, xvec14;
+ADD_SY 4*SIZE(C1,ldc,1), xvec6, xvec6;
+ADD_SY 0*SIZE(C0,ldc,2), xvec13, xvec13;
+ADD_SY 4*SIZE(C1,ldc,2), xvec5, xvec5;
+ADD_SY 0*SIZE(C0,%rax,1), xvec12, xvec12;
+ADD_SY 4*SIZE(C1,%rax,1), xvec4, xvec4;
+ADD_SY 0*SIZE(C1), xvec11, xvec11;
+ADD_SY 4*SIZE(C0), xvec3, xvec3;
+ADD_SY 0*SIZE(C1,ldc,1), xvec10, xvec10;
+ADD_SY 4*SIZE(C0,ldc,1), xvec2, xvec2;
+ADD_SY 0*SIZE(C1,ldc,2), xvec9, xvec9;
+ADD_SY 4*SIZE(C0,ldc,2), xvec1, xvec1;
+ADD_SY 0*SIZE(C1,%rax,1), xvec8, xvec8;
+ADD_SY 4*SIZE(C0,%rax,1), xvec0, xvec0;
+#endif
+ST_SY xvec15,0*SIZE(C0);
+ST_SY xvec7,4*SIZE(C1);
+ST_SY xvec14,0*SIZE(C0,ldc,1);
+ST_SY xvec6,4*SIZE(C1,ldc,1);
+ST_SY xvec13,0*SIZE(C0,ldc,2);
+ST_SY xvec5,4*SIZE(C1,ldc,2);
+ST_SY xvec12,0*SIZE(C0,%rax,1);
+ST_SY xvec4,4*SIZE(C1,%rax,1);
+ST_SY xvec11,0*SIZE(C1);
+ST_SY xvec3,4*SIZE(C0);
+ST_SY xvec10,0*SIZE(C1,ldc,1);
+ST_SY xvec2,4*SIZE(C0,ldc,1);
+ST_SY xvec9,0*SIZE(C1,ldc,2);
+ST_SY xvec1,4*SIZE(C0,ldc,2);
+ST_SY xvec8,0*SIZE(C1,%rax,1);
+ST_SY xvec0,4*SIZE(C0,%rax,1);
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) ||(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kkk, %rax;
+LEAQ 	(, %rax, SIZE), %rax;
+LEAQ	(ptrba, %rax, 8), ptrba;
+LEAQ 	(ptrbb, %rax, 8), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT)
+ADDQ	$8, kk
+#endif
+ADDQ $8*SIZE,C0;
+ADDQ $8*SIZE,C1;
+.L1_bodyE:;
+DECQ i;
+JG .L1_bodyB;
+JMP .L1_loopE;
+.align 16;
+.L4_loopEx:
+LEAQ (ldc,ldc,2),%rax;
+EXTRA_SY $1, yvec15, xvec7;
+#ifndef	TRMMKERNEL
+LDL_SY	0*SIZE(C0), xvec6, xvec6;
+LDH_SY	2*SIZE(C0), xvec6, xvec6;
+ADD_SY	xvec6, xvec15, xvec15;
+#endif
+STL_SY	xvec15, 0*SIZE(C0);
+STH_SY	xvec15, 2*SIZE(C0);
+#ifndef	TRMMKERNEL
+LDL_SY	4*SIZE(C1), xvec5, xvec5;
+LDH_SY	6*SIZE(C1), xvec5, xvec5;
+ADD_SY	xvec5, xvec7, xvec7;
+#endif
+STL_SY	xvec7, 4*SIZE(C1);
+STH_SY	xvec7, 6*SIZE(C1);
+
+EXTRA_SY $1, yvec14, xvec6;
+#ifndef	TRMMKERNEL
+LDL_SY	0*SIZE(C0, ldc, 1), xvec5, xvec5;
+LDH_SY	2*SIZE(C0, ldc, 1), xvec5, xvec5;
+ADD_SY	xvec5, xvec14, xvec14;
+#endif
+STL_SY	xvec14, 0*SIZE(C0, ldc, 1);
+STH_SY	xvec14, 2*SIZE(C0, ldc, 1);
+#ifndef	TRMMKERNEL
+LDL_SY	4*SIZE(C1, ldc, 1), xvec4, xvec4;
+LDH_SY	6*SIZE(C1, ldc, 1), xvec4, xvec4;
+ADD_SY	xvec4, xvec6, xvec6;
+#endif
+STL_SY	xvec6, 4*SIZE(C1, ldc, 1);
+STH_SY	xvec6, 6*SIZE(C1, ldc, 1);
+
+EXTRA_SY $1, yvec13, xvec5;
+#ifndef	TRMMKERNEL
+LDL_SY	0*SIZE(C0, ldc, 2), xvec4, xvec4;
+LDH_SY	2*SIZE(C0, ldc, 2), xvec4, xvec4;
+ADD_SY	xvec4, xvec13, xvec13;
+#endif
+STL_SY	xvec13, 0*SIZE(C0, ldc, 2);
+STH_SY	xvec13, 2*SIZE(C0, ldc, 2);
+#ifndef	TRMMKERNEL
+LDL_SY	4*SIZE(C1, ldc, 2), xvec3, xvec3;
+LDH_SY	6*SIZE(C1, ldc, 2), xvec3, xvec3;
+ADD_SY	xvec3, xvec5, xvec5;
+#endif
+STL_SY	xvec5, 4*SIZE(C1, ldc, 2);
+STH_SY	xvec5, 6*SIZE(C1, ldc, 2);
+
+EXTRA_SY $1, yvec12, xvec4;
+#ifndef	TRMMKERNEL
+LDL_SY	0*SIZE(C0, %rax, 1), xvec3, xvec3;
+LDH_SY	2*SIZE(C0, %rax, 1), xvec3, xvec3;
+ADD_SY	xvec3, xvec12, xvec12;
+#endif
+STL_SY	xvec12, 0*SIZE(C0, %rax, 1);
+STH_SY	xvec12, 2*SIZE(C0, %rax, 1);
+#ifndef	TRMMKERNEL
+LDL_SY	4*SIZE(C1, %rax, 1), xvec2, xvec2;
+LDH_SY	6*SIZE(C1, %rax, 1), xvec2, xvec2;
+ADD_SY	xvec2, xvec4, xvec4;
+#endif
+STL_SY	xvec4, 4*SIZE(C1, %rax, 1);
+STH_SY	xvec4, 6*SIZE(C1, %rax, 1);
+
+EXTRA_SY $1, yvec11, xvec3;
+#ifndef	TRMMKERNEL
+LDL_SY	0*SIZE(C1), xvec2, xvec2;
+LDH_SY	2*SIZE(C1), xvec2, xvec2;
+ADD_SY	xvec2, xvec11, xvec11;
+#endif
+STL_SY	xvec11, 0*SIZE(C1);
+STH_SY	xvec11, 2*SIZE(C1);
+#ifndef	TRMMKERNEL
+LDL_SY	4*SIZE(C0), xvec1, xvec1;
+LDH_SY	6*SIZE(C0), xvec1, xvec1;
+ADD_SY	xvec1, xvec3, xvec3;
+#endif
+STL_SY	xvec3, 4*SIZE(C0);
+STH_SY	xvec3, 6*SIZE(C0);
+
+EXTRA_SY $1, yvec10, xvec2;
+#ifndef	TRMMKERNEL
+LDL_SY	0*SIZE(C1, ldc, 1), xvec1, xvec1;
+LDH_SY	2*SIZE(C1, ldc, 1), xvec1, xvec1;
+ADD_SY	xvec1, xvec10, xvec10;
+#endif
+STL_SY	xvec10, 0*SIZE(C1, ldc, 1);
+STH_SY	xvec10, 2*SIZE(C1, ldc, 1);
+#ifndef	TRMMKERNEL
+LDL_SY	4*SIZE(C0, ldc, 1), xvec0, xvec0;
+LDH_SY	6*SIZE(C0, ldc, 1), xvec0, xvec0;
+ADD_SY	xvec0, xvec2, xvec2;
+#endif
+STL_SY	xvec2, 4*SIZE(C0, ldc, 1);
+STH_SY	xvec2, 6*SIZE(C0, ldc, 1);
+
+EXTRA_SY $1, yvec9, xvec1;
+#ifndef	TRMMKERNEL
+LDL_SY	0*SIZE(C1, ldc, 2), xvec0, xvec0;
+LDH_SY	2*SIZE(C1, ldc, 2), xvec0, xvec0;
+ADD_SY	xvec0, xvec9, xvec9;
+#endif
+STL_SY	xvec9, 0*SIZE(C1, ldc, 2);
+STH_SY	xvec9, 2*SIZE(C1, ldc, 2);
+#ifndef	TRMMKERNEL
+LDL_SY	4*SIZE(C0, ldc, 2), xvec7, xvec7;
+LDH_SY	6*SIZE(C0, ldc, 2), xvec7, xvec7;
+ADD_SY	xvec7, xvec1, xvec1;
+#endif
+STL_SY	xvec1, 4*SIZE(C0, ldc, 2);
+STH_SY	xvec1, 6*SIZE(C0, ldc, 2);
+
+EXTRA_SY $1, yvec8, xvec0;
+#ifndef	TRMMKERNEL
+LDL_SY	0*SIZE(C1, %rax, 1), xvec6, xvec6;
+LDH_SY	2*SIZE(C1, %rax, 1), xvec6, xvec6;
+ADD_SY	xvec6, xvec8, xvec8;
+#endif
+STL_SY	xvec8, 0*SIZE(C1, %rax, 1);
+STH_SY	xvec8, 2*SIZE(C1, %rax, 1);
+#ifndef	TRMMKERNEL
+LDL_SY	4*SIZE(C0, %rax, 1), xvec5, xvec5;
+LDH_SY	6*SIZE(C0, %rax, 1), xvec5, xvec5;
+ADD_SY	xvec5, xvec0, xvec0;
+#endif
+STL_SY	xvec0, 4*SIZE(C0, %rax, 1);
+STH_SY	xvec0, 6*SIZE(C0, %rax, 1);
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) ||(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kkk, %rax;
+LEAQ 	(, %rax, SIZE), %rax;
+LEAQ	(ptrba, %rax, 8), ptrba;
+LEAQ 	(ptrbb, %rax, 8), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT)
+ADDQ	$8, kk
+#endif
+ADDQ	$8*SIZE, C0;
+ADDQ	$8*SIZE, C1;
+DECQ i;
+JG .L1_bodyB;
+.align 16
+.L1_loopE:;
+TEST $4, bm;
+JLE	.L5_loopE;
+.align 16
+.L5_bodyB:
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb, ptrbb;						
+#else 
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+LEAQ (, %rax, SIZE), %rax;
+LEAQ (ptrba, %rax, 4), ptrba;
+LEAQ (ptrbb, %rax, 8), ptrbb;
+#endif
+#### Initial Results Register ####
+XOR_SY	yvec15, yvec15, yvec15;
+XOR_SY	yvec14, yvec14, yvec14;
+XOR_SY	yvec13, yvec13, yvec13;
+XOR_SY	yvec12, yvec12, yvec12;
+LD_SX	0*SIZE(ptrba), xvec0;
+XOR_SY	yvec11, yvec11, yvec11;
+XOR_SY	yvec10, yvec10, yvec10;
+EDUP_SX	0*SIZE(ptrbb), xvec2;
+XOR_SY	yvec9, yvec9, yvec9;
+XOR_SY	yvec8, yvec8, yvec8;
+ODUP_SX	0*SIZE(ptrbb), xvec3;
+#ifndef	TRMMKERNEL
+MOVQ bk, k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else
+MOVQ kk, %rax;
+#ifdef	LEFT
+ADDQ $4, %rax;
+#else 
+ADDQ $8, %rax;
+#endif
+MOVQ %rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L8_loopE;
+.align 16
+.L8_bodyB:
+
+#### Unroll time 1 ####
+SHUF_SX	$0x4e, xvec2, xvec4;
+MUL_SX	xvec0, xvec2;
+ADD_SX	xvec2, xvec15;
+SHUF_SX	$0x4e, xvec3, xvec5;
+MUL_SX	xvec0, xvec3;
+ADD_SX	xvec3, xvec14;
+
+EDUP_SX	4*SIZE(ptrbb), xvec2;
+MUL_SX	xvec0, xvec4;
+ADD_SX	xvec4, xvec13;
+ODUP_SX 4*SIZE(ptrbb), xvec3;
+MUL_SX	xvec0, xvec5;
+ADD_SX	xvec5, xvec12;
+
+LD_SX	4*SIZE(ptrba), xvec1;
+SHUF_SX	$0x4e, xvec2, xvec4;
+MUL_SX	xvec0, xvec2;
+ADD_SX	xvec2, xvec11;
+SHUF_SX	$0x4e, xvec3, xvec5;
+MUL_SX	xvec0, xvec3;
+ADD_SX	xvec3, xvec10;
+
+EDUP_SX	8*SIZE(ptrbb), xvec2;
+MUL_SX	xvec0, xvec4;
+ADD_SX	xvec4, xvec9;
+ODUP_SX 8*SIZE(ptrbb), xvec3;
+MUL_SX	xvec0, xvec5;
+ADD_SX	xvec5, xvec8;
+
+#### Unroll time 2 #### 
+SHUF_SX	$0x4e, xvec2, xvec4;
+MUL_SX	xvec1, xvec2;
+ADD_SX	xvec2, xvec15;
+SHUF_SX	$0x4e, xvec3, xvec5;
+MUL_SX	xvec1, xvec3;
+ADD_SX	xvec3, xvec14;
+
+EDUP_SX	12*SIZE(ptrbb), xvec2;
+MUL_SX	xvec1, xvec4;
+ADD_SX	xvec4, xvec13;
+ODUP_SX	12*SIZE(ptrbb), xvec3;
+MUL_SX	xvec1, xvec5;
+ADD_SX	xvec5, xvec12;
+
+LD_SX	8*SIZE(ptrba), xvec0;
+SHUF_SX	$0x4e, xvec2, xvec4;
+MUL_SX	xvec1, xvec2;
+ADD_SX	xvec2, xvec11;
+SHUF_SX	$0x4e, xvec3, xvec5;
+MUL_SX	xvec1, xvec3;
+ADD_SX	xvec3, xvec10;
+
+EDUP_SX	16*SIZE(ptrbb), xvec2;
+MUL_SX	xvec1, xvec4;
+ADD_SX	xvec4, xvec9;
+ODUP_SX	16*SIZE(ptrbb), xvec3;
+MUL_SX	xvec1, xvec5;
+ADD_SX	xvec5, xvec8;
+
+#### Unroll time 3 #### 
+SHUF_SX	$0x4e, xvec2, xvec4;
+MUL_SX	xvec0, xvec2;
+ADD_SX	xvec2, xvec15;
+SHUF_SX	$0x4e, xvec3, xvec5;
+MUL_SX	xvec0, xvec3;
+ADD_SX	xvec3, xvec14;
+
+EDUP_SX	20*SIZE(ptrbb), xvec2;
+MUL_SX	xvec0, xvec4;
+ADD_SX	xvec4, xvec13;
+ODUP_SX	20*SIZE(ptrbb), xvec3;
+MUL_SX	xvec0, xvec5;
+ADD_SX	xvec5, xvec12;
+
+LD_SX	12*SIZE(ptrba), xvec1;
+SHUF_SX	$0x4e, xvec2, xvec4;
+MUL_SX	xvec0, xvec2;
+ADD_SX	xvec2, xvec11;
+SHUF_SX	$0x4e, xvec3, xvec5;
+MUL_SX	xvec0, xvec3;
+ADD_SX	xvec3, xvec10;
+
+EDUP_SX	24*SIZE(ptrbb), xvec2;
+MUL_SX	xvec0, xvec4;
+ADD_SX	xvec4, xvec9;
+ODUP_SX	24*SIZE(ptrbb), xvec3;
+MUL_SX	xvec0, xvec5;
+ADD_SX	xvec5, xvec8;
+ADDQ	$16*SIZE, ptrba;
+
+#### Unroll time 4 #### 
+SHUF_SX	$0x4e, xvec2, xvec4;
+MUL_SX	xvec1, xvec2;
+ADD_SX	xvec2, xvec15;
+SHUF_SX	$0x4e, xvec3, xvec5;
+MUL_SX	xvec1, xvec3;
+ADD_SX	xvec3, xvec14;
+
+EDUP_SX	28*SIZE(ptrbb), xvec2;
+MUL_SX	xvec1, xvec4;
+ADD_SX	xvec4, xvec13;
+ODUP_SX	28*SIZE(ptrbb), xvec3;
+MUL_SX	xvec1, xvec5;
+ADD_SX	xvec5, xvec12;
+ADDQ	$32*SIZE, ptrbb;
+
+LD_SX	0*SIZE(ptrba), xvec0;
+SHUF_SX	$0x4e, xvec2, xvec4;
+MUL_SX	xvec1, xvec2;
+ADD_SX	xvec2, xvec11;
+SHUF_SX	$0x4e, xvec3, xvec5;
+MUL_SX	xvec1, xvec3;
+ADD_SX	xvec3, xvec10;
+
+EDUP_SX	0*SIZE(ptrbb), xvec2;
+MUL_SX	xvec1, xvec4;
+ADD_SX	xvec4, xvec9;
+ODUP_SX	0*SIZE(ptrbb), xvec3;
+MUL_SX	xvec1, xvec5;
+ADD_SX	xvec5, xvec8;
+DECQ k;
+JG .L8_bodyB;
+.align 16
+.L8_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else
+TEST $2, kkk;
+#endif
+JLE .L9_loopE;
+.align 16
+.L9_bodyB:
+#### Unroll time 1 ####
+SHUF_SX	$0x4e, xvec2, xvec4;
+MUL_SX	xvec0, xvec2;
+ADD_SX	xvec2, xvec15;
+SHUF_SX	$0x4e, xvec3, xvec5;
+MUL_SX	xvec0, xvec3;
+ADD_SX	xvec3, xvec14;
+
+EDUP_SX	4*SIZE(ptrbb), xvec2;
+MUL_SX	xvec0, xvec4;
+ADD_SX	xvec4, xvec13;
+ODUP_SX	4*SIZE(ptrbb), xvec3;
+MUL_SX	xvec0, xvec5;
+ADD_SX	xvec5, xvec12;
+
+LD_SX	4*SIZE(ptrba), xvec1;
+SHUF_SX	$0x4e, xvec2, xvec4;
+MUL_SX	xvec0, xvec2;
+ADD_SX	xvec2, xvec11;
+SHUF_SX	$0x4e, xvec3, xvec5;
+MUL_SX	xvec0, xvec3;
+ADD_SX	xvec3, xvec10;
+
+EDUP_SX	8*SIZE(ptrbb), xvec2;
+MUL_SX	xvec0, xvec4;
+ADD_SX	xvec4, xvec9;
+ODUP_SX	8*SIZE(ptrbb), xvec3;
+MUL_SX	xvec0, xvec5;
+ADD_SX	xvec5, xvec8;
+
+#### Unroll time 2 #### 
+ADDQ	$8*SIZE, ptrba;
+SHUF_SX	$0x4e, xvec2, xvec4;
+MUL_SX	xvec1, xvec2;
+ADD_SX	xvec2, xvec15;
+SHUF_SX	$0x4e, xvec3, xvec5;
+MUL_SX	xvec1, xvec3;
+ADD_SX	xvec3, xvec14;
+
+EDUP_SX	12*SIZE(ptrbb), xvec2;
+MUL_SX	xvec1, xvec4;
+ADD_SX	xvec4, xvec13;
+ODUP_SX	12*SIZE(ptrbb), xvec3;
+MUL_SX	xvec1, xvec5;
+ADD_SX	xvec5, xvec12;
+ADDQ	$16*SIZE, ptrbb;
+
+LD_SX	0*SIZE(ptrba), xvec0;
+SHUF_SX	$0x4e, xvec2, xvec4;
+MUL_SX	xvec1, xvec2;
+ADD_SX	xvec2, xvec11;
+SHUF_SX	$0x4e, xvec3, xvec5;
+MUL_SX	xvec1, xvec3;
+ADD_SX	xvec3, xvec10;
+
+EDUP_SX	0*SIZE(ptrbb), xvec2;
+MUL_SX	xvec1, xvec4;
+ADD_SX	xvec4, xvec9;
+ODUP_SX	0*SIZE(ptrbb), xvec3;
+MUL_SX	xvec1, xvec5;
+ADD_SX	xvec5, xvec8;
+
+.L9_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else
+TEST $1, kkk;
+#endif
+JLE .L10_loopE;
+.align 16
+.L10_bodyB:
+#### Unroll time 1 ####
+SHUF_SX	$0x4e, xvec2, xvec4;
+MUL_SX	xvec0, xvec2;
+ADD_SX	xvec2, xvec15;
+SHUF_SX	$0x4e, xvec3, xvec5;
+MUL_SX	xvec0, xvec3;
+ADD_SX	xvec3, xvec14;
+ADDQ	$4*SIZE, ptrba;
+
+EDUP_SX	4*SIZE(ptrbb), xvec2;
+MUL_SX	xvec0, xvec4;
+ADD_SX	xvec4, xvec13;
+ODUP_SX 4*SIZE(ptrbb), xvec3;
+MUL_SX	xvec0, xvec5;
+ADD_SX	xvec5, xvec12;
+ADDQ	$8*SIZE, ptrbb;
+
+SHUF_SX	$0x4e, xvec2, xvec4;
+MUL_SX	xvec0, xvec2;
+ADD_SX	xvec2, xvec11;
+SHUF_SX	$0x4e, xvec3, xvec5;
+MUL_SX	xvec0, xvec3;
+ADD_SX	xvec3, xvec10;
+
+MUL_SX	xvec0, xvec4;
+ADD_SX	xvec4, xvec9;
+MUL_SX	xvec0, xvec5;
+ADD_SX	xvec5, xvec8;
+
+.L10_loopE:
+#### Multiply Alpha ####
+BROAD_SX MEMALPHA, xvec7;
+MUL_SX	xvec7, xvec15;
+MUL_SX	xvec7, xvec14;
+MUL_SX	xvec7, xvec13;
+MUL_SX	xvec7, xvec12;
+MUL_SX	xvec7, xvec11;
+MUL_SX	xvec7, xvec10;
+MUL_SX	xvec7, xvec9;
+MUL_SX	xvec7, xvec8;
+#### Reverse Result ####
+MOV_SX	xvec15, xvec7;
+REVS_SX	$0xe4, xvec13, xvec15;
+REVS_SX	$0xe4, xvec7, xvec13;
+MOV_SX	xvec14, xvec7;
+REVS_SX	$0xe4, xvec12, xvec14;
+REVS_SX	$0xe4, xvec7, xvec12;
+MOV_SX	xvec11, xvec7;
+REVS_SX	$0xe4, xvec9, xvec11;
+REVS_SX	$0xe4, xvec7, xvec9;
+MOV_SX	xvec10, xvec7;
+REVS_SX	$0xe4, xvec8, xvec10;
+REVS_SX	$0xe4, xvec7, xvec8;
+#### Testing Alignment #### 
+MOVQ	C0, %rax;
+OR		ldc, %rax;
+TEST	$15, %rax;
+JNE		.L10_loopEx;
+.align 16
+LEAQ (ldc,ldc,2),%rax;
+#ifndef	TRMMKERNEL
+ADD_SX	0*SIZE(C0), xvec15;
+ADD_SX	0*SIZE(C0, ldc, 1), xvec14;
+ADD_SX	0*SIZE(C0, ldc, 2), xvec13;
+ADD_SX	0*SIZE(C0, %rax, 1), xvec12;
+ADD_SX	0*SIZE(C1), xvec11;
+ADD_SX	0*SIZE(C1, ldc, 1), xvec10;
+ADD_SX	0*SIZE(C1, ldc, 2), xvec9;
+ADD_SX	0*SIZE(C1, %rax, 1), xvec8;
+#endif
+ST_SX	xvec15, 0*SIZE(C0);
+ST_SX	xvec14, 0*SIZE(C0, ldc, 1);
+ST_SX	xvec13, 0*SIZE(C0, ldc, 2);
+ST_SX	xvec12, 0*SIZE(C0, %rax, 1);
+ST_SX	xvec11, 0*SIZE(C1);
+ST_SX	xvec10, 0*SIZE(C1, ldc, 1);
+ST_SX	xvec9, 0*SIZE(C1, ldc, 2);
+ST_SX	xvec8, 0*SIZE(C1, %rax, 1);
+#if	(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kkk, %rax;
+LEAQ	(, %rax, SIZE), %rax;
+LEAQ	(ptrba, %rax, 4), ptrba;
+LEAQ	(ptrbb, %rax, 8), ptrbb;
+#endif
+#if	defined(TRMMKERNEL)&&defined(LEFT)
+ADDQ	$4, kk
+#endif
+ADDQ	$4*SIZE, C0;
+ADDQ	$4*SIZE, C1;
+JMP	.L5_loopE;
+.align 16
+.L10_loopEx:
+LEAQ (ldc,ldc,2),%rax;
+#ifndef	TRMMKERNEL
+LDL_SX	0*SIZE(C0), xvec7;
+LDH_SX	2*SIZE(C0), xvec7;
+LDL_SX	0*SIZE(C0, ldc, 1), xvec6;
+LDH_SX	2*SIZE(C0, ldc, 1), xvec6;
+LDL_SX	0*SIZE(C0, ldc, 2), xvec5;
+LDH_SX	2*SIZE(C0, ldc, 2), xvec5;
+LDL_SX	0*SIZE(C0, %rax, 1), xvec4;
+LDH_SX	2*SIZE(C0, %rax, 1), xvec4;
+LDL_SX	0*SIZE(C1), xvec3;
+LDH_SX	2*SIZE(C1), xvec3;
+LDL_SX	0*SIZE(C1, ldc, 1), xvec2;
+LDH_SX	2*SIZE(C1, ldc, 1), xvec2;
+LDL_SX	0*SIZE(C1, ldc, 2), xvec1;
+LDH_SX	2*SIZE(C1, ldc, 2), xvec1;
+LDL_SX	0*SIZE(C1, %rax, 1), xvec0;
+LDH_SX	2*SIZE(C1, %rax, 1), xvec0;
+ADD_SX	xvec7, xvec15;
+ADD_SX	xvec6, xvec14;
+ADD_SX	xvec5, xvec13;
+ADD_SX	xvec4, xvec12;
+ADD_SX	xvec3, xvec11;
+ADD_SX	xvec2, xvec10;
+ADD_SX	xvec1, xvec9;
+ADD_SX	xvec0, xvec8;
+#endif
+STL_SX	xvec15, 0*SIZE(C0);
+STH_SX	xvec15, 2*SIZE(C0);
+STL_SX	xvec14, 0*SIZE(C0, ldc, 1);
+STH_SX	xvec14, 2*SIZE(C0, ldc, 1);
+STL_SX	xvec13, 0*SIZE(C0, ldc, 2);
+STH_SX	xvec13, 2*SIZE(C0, ldc, 2);
+STL_SX 	xvec12, 0*SIZE(C0, %rax, 1);
+STH_SX	xvec12, 2*SIZE(C0, %rax, 1);
+STL_SX	xvec11, 0*SIZE(C1);
+STH_SX	xvec11, 2*SIZE(C1);
+STL_SX	xvec10, 0*SIZE(C1, ldc, 1);
+STH_SX	xvec10, 2*SIZE(C1, ldc, 1);
+STL_SX	xvec9, 0*SIZE(C1, ldc, 2);
+STH_SX	xvec9, 2*SIZE(C1, ldc, 2);
+STL_SX	xvec8, 0*SIZE(C1, %rax, 1);
+STH_SX	xvec8, 2*SIZE(C1, %rax, 1);
+#if	(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kkk, %rax;
+LEAQ	(, %rax, SIZE), %rax;
+LEAQ	(ptrba, %rax, 4), ptrba;
+LEAQ	(ptrbb, %rax, 8), ptrbb;
+#endif
+#if	defined(TRMMKERNEL)&&defined(LEFT)
+ADDQ	$4, kk
+#endif
+ADDQ	$4*SIZE, C0;
+ADDQ	$4*SIZE, C1;
+.L5_loopE:
+TEST $2, bm;
+JLE	.L6_loopE;
+.align 16
+.L6_bodyB:
+#if	!defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb, ptrbb;						
+#else 
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+LEAQ (, %rax, SIZE), %rax;
+LEAQ (ptrba, %rax, 2), ptrba;
+LEAQ (ptrbb, %rax, 8), ptrbb
+#endif
+#### Initial Results Register ####
+XOR_SY	yvec15, yvec15, yvec15;
+XOR_SY	yvec14, yvec14, yvec14;
+XOR_SY	yvec13, yvec13, yvec13;
+XOR_SY	yvec12, yvec12, yvec12;
+MOVQ bk, k;
+#ifndef	TRMMKERNEL
+MOVQ bk, k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ 	kk, %rax;
+MOVQ 	%rax, kkk;
+#else 
+MOVQ	kk, %rax;
+#ifdef	LEFT
+ADDQ	$2, %rax;
+#else
+ADDQ	$8, %rax;
+#endif
+MOVQ	%rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L11_loopE;
+.align 16
+.L11_bodyB:
+#### Computing kernel
+LD_SX	0*SIZE(ptrba), xvec0;	# a1, a2, a3, a4
+SHUF_SX	$0x44, xvec0, xvec1;	# a1, a2, a1, a2
+EDUP_SX	0*SIZE(ptrbb), xvec2;
+ODUP_SX	0*SIZE(ptrbb), xvec3;
+MUL_SX	xvec1, xvec2;
+ADD_SX	xvec2, xvec15;
+MUL_SX	xvec1, xvec3;
+ADD_SX	xvec3, xvec14;
+
+EDUP_SX	4*SIZE(ptrbb), xvec4;
+ODUP_SX	4*SIZE(ptrbb), xvec5;
+MUL_SX	xvec1, xvec4;
+ADD_SX	xvec4, xvec13;
+MUL_SX	xvec1, xvec5;
+ADD_SX	xvec5, xvec12;
+
+SHUF_SX	$0xee, xvec0, xvec6;
+EDUP_SX	8*SIZE(ptrbb), xvec2;
+ODUP_SX	8*SIZE(ptrbb), xvec3;
+MUL_SX	xvec6, xvec2;
+ADD_SX	xvec2, xvec15;
+MUL_SX	xvec6, xvec3;
+ADD_SX	xvec3, xvec14;
+
+EDUP_SX	12*SIZE(ptrbb), xvec4;
+ODUP_SX	12*SIZE(ptrbb), xvec5;
+MUL_SX	xvec6, xvec4;
+ADD_SX	xvec4, xvec13;
+MUL_SX	xvec6, xvec5;
+ADD_SX	xvec5, xvec12;
+
+LD_SX	4*SIZE(ptrba), xvec0;
+SHUF_SX	$0x44, xvec0, xvec1;
+EDUP_SX	16*SIZE(ptrbb), xvec2;
+ODUP_SX	16*SIZE(ptrbb), xvec3;
+MUL_SX	xvec1, xvec2;
+ADD_SX	xvec2, xvec15;
+MUL_SX	xvec1, xvec3;
+ADD_SX	xvec3, xvec14;
+
+EDUP_SX	20*SIZE(ptrbb), xvec4;
+ODUP_SX	20*SIZE(ptrbb), xvec5;
+MUL_SX	xvec1, xvec4;
+ADD_SX	xvec4, xvec13;
+MUL_SX	xvec1, xvec5;
+ADD_SX	xvec5, xvec12;
+
+SHUF_SX	$0xee, xvec0, xvec6;
+EDUP_SX	24*SIZE(ptrbb), xvec2;
+ODUP_SX	24*SIZE(ptrbb), xvec3;
+MUL_SX	xvec6, xvec2;
+ADD_SX	xvec2, xvec15;
+MUL_SX	xvec6, xvec3;
+ADD_SX	xvec3, xvec14;
+
+EDUP_SX	28*SIZE(ptrbb), xvec4;
+ODUP_SX	28*SIZE(ptrbb), xvec5;
+MUL_SX	xvec6, xvec4;
+ADD_SX	xvec4, xvec13;
+MUL_SX	xvec6, xvec5;
+ADD_SX	xvec5, xvec12;
+
+ADDQ	$8*SIZE, ptrba;
+ADDQ	$32*SIZE, ptrbb;
+DECQ	k;
+JG .L11_bodyB;
+.align 16
+.L11_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else
+TEST $2, kkk;
+#endif
+JLE .L12_loopE;
+.align 16
+.L12_bodyB:
+LD_SX	0*SIZE(ptrba), xvec0;	# a1, a2, a3, a4
+SHUF_SX	$0x44, xvec0, xvec1;	# a1, a2, a1, a2
+EDUP_SX	0*SIZE(ptrbb), xvec2;
+ODUP_SX	0*SIZE(ptrbb), xvec3;
+MUL_SX	xvec1, xvec2;
+ADD_SX	xvec2, xvec15;
+MUL_SX	xvec1, xvec3;
+ADD_SX	xvec3, xvec14;
+
+EDUP_SX	4*SIZE(ptrbb), xvec4;
+ODUP_SX	4*SIZE(ptrbb), xvec5;
+MUL_SX	xvec1, xvec4;
+ADD_SX	xvec4, xvec13;
+MUL_SX	xvec1, xvec5;
+ADD_SX	xvec5, xvec12;
+
+SHUF_SX	$0xee, xvec0, xvec6;
+EDUP_SX	8*SIZE(ptrbb), xvec2;
+ODUP_SX	8*SIZE(ptrbb), xvec3;
+MUL_SX	xvec6, xvec2;
+ADD_SX	xvec2, xvec15;
+MUL_SX	xvec6, xvec3;
+ADD_SX	xvec3, xvec14;
+
+EDUP_SX	12*SIZE(ptrbb), xvec4;
+ODUP_SX	12*SIZE(ptrbb), xvec5;
+MUL_SX	xvec6, xvec4;
+ADD_SX	xvec4, xvec13;
+MUL_SX	xvec6, xvec5;
+ADD_SX	xvec5, xvec12;
+ADDQ	$4*SIZE, ptrba;
+ADDQ	$16*SIZE, ptrbb;
+
+.L12_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else
+TEST $1, kkk;
+#endif
+JLE .L13_loopE;
+.align 16
+.L13_bodyB:
+LD_SX	0*SIZE(ptrba), xvec0;	# a1, a2, a3, a4
+SHUF_SX	$0x44, xvec0, xvec1;	# a1, a2, a1, a2
+EDUP_SX	0*SIZE(ptrbb), xvec2;
+ODUP_SX	0*SIZE(ptrbb), xvec3;
+MUL_SX	xvec1, xvec2;
+ADD_SX	xvec2, xvec15;
+MUL_SX	xvec1, xvec3;
+ADD_SX	xvec3, xvec14;
+
+EDUP_SX	4*SIZE(ptrbb), xvec4;
+ODUP_SX	4*SIZE(ptrbb), xvec5;
+MUL_SX	xvec1, xvec4;
+ADD_SX	xvec4, xvec13;
+MUL_SX	xvec1, xvec5;
+ADD_SX	xvec5, xvec12;
+ADDQ	$2*SIZE, ptrba;
+ADDQ	$8*SIZE, ptrbb;
+.L13_loopE:
+LEAQ (ldc,ldc,2),%rax;
+#### Multiply Alpha ####
+BROAD_SX MEMALPHA, xvec7;
+MUL_SX	xvec7, xvec15;
+MUL_SX	xvec7, xvec14;
+MUL_SX	xvec7, xvec13;
+MUL_SX	xvec7, xvec12;
+#ifndef	TRMMKERNEL
+LDL_SX	0*SIZE(C0), xvec11;
+LDH_SX	0*SIZE(C0, ldc, 2), xvec11;
+LDL_SX	0*SIZE(C0, ldc, 1), xvec10;
+LDH_SX	0*SIZE(C0, %rax, 1), xvec10;
+LDL_SX	0*SIZE(C1), xvec9;
+LDH_SX	0*SIZE(C1, ldc, 2), xvec9;
+LDL_SX	0*SIZE(C1, ldc, 1), xvec8;
+LDH_SX	0*SIZE(C1, %rax, 1), xvec8;
+ADD_SX	xvec11, xvec15;
+ADD_SX	xvec10, xvec14;
+ADD_SX	xvec9, xvec13;
+ADD_SX	xvec8, xvec12;
+#endif
+STL_SX	xvec15, 0*SIZE(C0);
+STH_SX	xvec15, 0*SIZE(C0, ldc, 2);
+STL_SX	xvec14, 0*SIZE(C0, ldc, 1);
+STH_SX	xvec14, 0*SIZE(C0, %rax, 1);
+STL_SX	xvec13, 0*SIZE(C1);
+STH_SX	xvec13, 0*SIZE(C1, ldc, 2);
+STL_SX	xvec12, 0*SIZE(C1, ldc, 1);
+STH_SX	xvec12, 0*SIZE(C1, %rax, 1);
+#if (defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kkk, %rax;
+LEAQ	(,%rax, SIZE), %rax;
+LEAQ	(ptrba, %rax, 2), ptrba;
+LEAQ	(ptrbb, %rax, 8), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT)
+ADDQ	$2, kk
+#endif
+ADDQ	$2*SIZE, C0;
+ADDQ	$2*SIZE, C1;
+#### Writing Back ####
+.L6_loopE:
+TEST $1, bm;
+JLE .L7_loopE;
+.align 16
+.L7_bodyB:
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb, ptrbb;
+#else 
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+LEAQ (,%rax, SIZE), %rax;
+ADDQ	%rax, ptrba;
+LEAQ (ptrbb, %rax, 8), ptrbb;
+#endif
+#### intitial ####
+XOR_SY	yvec15, yvec15, yvec15;
+XOR_SY	yvec14, yvec14, yvec14;
+MOVQ bk, k;
+#ifndef	TRMMKERNEL
+MOVQ bk, k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kk, %rax;
+MOVQ	%rax, kkk;
+#else
+MOVQ	kk, %rax;
+#ifdef	LEFT
+ADDQ	$1, %rax;
+#else 
+ADDQ	$8, %rax;
+#endif
+MOVQ	%rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L14_loopE;
+.align 16
+.L14_bodyB:
+BROAD_SX 0*SIZE(ptrba), xvec0;
+LD_SX	0*SIZE(ptrbb), xvec2;
+MUL_SX	xvec0, xvec2;
+ADD_SX	xvec2, xvec15;
+LD_SX	4*SIZE(ptrbb), xvec3;
+MUL_SX	xvec0, xvec3;
+ADD_SX	xvec3, xvec14;
+
+BROAD_SX 1*SIZE(ptrba), xvec1;
+LD_SX	8*SIZE(ptrbb), xvec4;
+MUL_SX	xvec1, xvec4;
+ADD_SX	xvec4, xvec15;
+LD_SX	12*SIZE(ptrbb), xvec5;
+MUL_SX	xvec1, xvec5;
+ADD_SX	xvec5, xvec14;
+
+BROAD_SX 2*SIZE(ptrba), xvec0;
+LD_SX	16*SIZE(ptrbb), xvec2;
+MUL_SX	xvec0, xvec2;
+ADD_SX	xvec2, xvec15;
+LD_SX	20*SIZE(ptrbb), xvec3;
+MUL_SX	xvec0, xvec3;
+ADD_SX	xvec3, xvec14;
+
+BROAD_SX 3*SIZE(ptrba), xvec1;
+LD_SX	24*SIZE(ptrbb), xvec4;
+MUL_SX	xvec1, xvec4;
+ADD_SX	xvec4, xvec15;
+LD_SX	28*SIZE(ptrbb), xvec5;
+MUL_SX	xvec1, xvec5;
+ADD_SX	xvec5, xvec14;
+
+ADDQ	$4*SIZE, ptrba;
+ADDQ	$32*SIZE, ptrbb;
+DECQ	k;
+JG .L14_bodyB;
+.align 16
+.L14_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else
+TEST $2, kkk;
+#endif
+JLE .L15_loopE;
+.align 16
+.L15_bodyB:
+BROAD_SX 0*SIZE(ptrba), xvec0;
+LD_SX	0*SIZE(ptrbb), xvec2;
+MUL_SX	xvec0, xvec2;
+ADD_SX	xvec2, xvec15;
+LD_SX	4*SIZE(ptrbb), xvec3;
+MUL_SX	xvec0, xvec3;
+ADD_SX	xvec3, xvec14;
+
+BROAD_SX 1*SIZE(ptrba), xvec1;
+LD_SX	8*SIZE(ptrbb), xvec4;
+MUL_SX	xvec1, xvec4;
+ADD_SX	xvec4, xvec15;
+LD_SX	12*SIZE(ptrbb), xvec5;
+MUL_SX	xvec1, xvec5;
+ADD_SX	xvec5, xvec14;
+ADDQ	$2*SIZE, ptrba;
+ADDQ	$16*SIZE, ptrbb;
+
+.L15_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else
+TEST $1, kkk;
+#endif
+JLE	.L16_loopE;
+.align 16
+.L16_bodyB:
+BROAD_SX 0*SIZE(ptrba), xvec0;
+LD_SX	0*SIZE(ptrbb), xvec2;
+MUL_SX	xvec0, xvec2;
+ADD_SX	xvec2, xvec15;
+LD_SX	4*SIZE(ptrbb), xvec3;
+MUL_SX	xvec0, xvec3;
+ADD_SX	xvec3, xvec14;
+ADDQ	$1, ptrba;
+ADDQ	$4, ptrbb;
+
+.L16_loopE:
+BROAD_SX MEMALPHA, xvec7;
+MUL_SX	xvec7, xvec15;
+MUL_SX	xvec7, xvec14;
+
+LEAQ (ldc,ldc,2),%rax;
+SHUF_SX	$0xff, xvec15, xvec13;
+SHUF_SX	$0xaa, xvec15, xvec12;
+SHUF_SX	$0x55, xvec15, xvec11;
+SHUF_SX	$0x00, xvec15, xvec10;
+
+#ifndef	TRMMKERNEL
+addss	0*SIZE(C0), xvec10;
+addss	0*SIZE(C0, ldc, 1), xvec11;
+addss	0*SIZE(C0, ldc, 2), xvec12;
+addss	0*SIZE(C0, %rax, 1), xvec13;
+#endif
+movss	xvec10, 0*SIZE(C0);
+movss	xvec11, 0*SIZE(C0, ldc, 1);
+movss	xvec12, 0*SIZE(C0, ldc, 2);
+movss	xvec13, 0*SIZE(C0, %rax, 1);
+
+SHUF_SX	$0xff, xvec14, xvec9;
+SHUF_SX	$0xaa, xvec14, xvec8;
+SHUF_SX	$0x55, xvec14, xvec7;
+SHUF_SX	$0x00, xvec14, xvec6;
+
+#ifndef	TRMMKERNEL
+addss	0*SIZE(C1), xvec6;
+addss	0*SIZE(C1, ldc, 1), xvec7;
+addss	0*SIZE(C1, ldc, 2), xvec8;
+addss	0*SIZE(C1, %rax, 1), xvec9;
+#endif
+movss	xvec6, 0*SIZE(C1);
+movss	xvec7, 0*SIZE(C1, ldc, 1);
+movss	xvec8, 0*SIZE(C1, ldc, 2);
+movss	xvec9, 0*SIZE(C1, %rax, 1);
+#if (defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kkk, %rax;
+LEAQ	(,%rax, SIZE), %rax;
+ADDQ	%rax, ptrba;
+LEAQ	(ptrbb, %rax, 8), ptrbb;
+#endif
+#if defined(TRMMKERNEL)&&defined(LEFT)
+ADDQ	$1, kk
+#endif
+ADDQ	$1*SIZE, C0;
+ADDQ	$1*SIZE, C1;
+#### Writing Back ####
+.L7_loopE:
+#if	defined(TRMMKERNEL)&&!defined(LEFT)
+ADDQ	$8, kk
+#endif
+MOVQ bk,k;
+SALQ $5,k;
+ADDQ k,bb;
+LEAQ (C,ldc,8),C;
+.L0_bodyE:;
+DECQ j;
+JG .L0_bodyB;
+.align 16;
+.L0_loopE:;
+TEST $4, bn;					# Rn = 4
+JLE	.L20_loopE;
+.align 16;
+.L20_bodyB:
+#if defined(TRMMKERNEL) && defined(LEFT)
+MOVQ OFFSET, %rax;
+MOVQ %rax, kk;
+#endif
+
+MOVQ C, C0;
+LEAQ (C, ldc, 2), C1;
+MOVQ ba, ptrba;
+MOVQ bm, i;
+SARQ $3, i;
+JLE .L21_loopE;
+.align 16
+.L21_bodyB:
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb, ptrbb;
+#else 
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+LEAQ (, %rax, SIZE), %rax;
+LEAQ (ptrba, %rax, 8), ptrba;
+LEAQ (ptrbb, %rax, 4), ptrbb;
+#endif
+#### Initial ####
+XOR_SY yvec15, yvec15, yvec15;
+XOR_SY yvec14, yvec14, yvec14;
+XOR_SY yvec13, yvec13, yvec13;
+XOR_SY yvec12, yvec12, yvec12;
+EDUP_SX	0*SIZE(ptrbb), xvec2;
+XOR_SY yvec11, yvec11, yvec11;
+XOR_SY yvec10, yvec10, yvec10;
+LD_SX	0*SIZE(ptrba), xvec0;
+XOR_SY yvec9, yvec9, yvec9;
+XOR_SY yvec8, yvec8, yvec8;
+LD_SX	4*SIZE(ptrba), xvec1;
+#ifndef	TRMMKERNEL
+MOVQ bk,k;
+#elif (defined(LEFT) && !defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else
+MOVQ kk, %rax;
+#ifdef LEFT
+ADDQ $8, %rax;
+#else 
+ADDQ $4, %rax;
+#endif
+MOVQ %rax, kkk;
+#endif
+SARQ $2,k;
+JLE .L211_loopE;
+.align 16
+.L211_bodyB:
+#### Unroll time 1 ####
+ODUP_SX	0*SIZE(ptrbb), xvec3;
+SHUF_SX	$0x4e, xvec2, xvec4;
+MOV_SX	xvec2, xvec6;
+MUL_SX	xvec0, xvec2;
+ADD_SX	xvec2, xvec15;
+MUL_SX	xvec1, xvec6;
+ADD_SX	xvec6, xvec14;
+
+SHUF_SX	$0x4e, xvec3, xvec5;
+MOV_SX	xvec3, xvec7;
+MUL_SX	xvec0, xvec3;
+ADD_SX	xvec3, xvec13;
+MUL_SX	xvec1, xvec7;
+ADD_SX	xvec7, xvec12;
+
+EDUP_SX	4*SIZE(ptrbb), xvec2;
+MOV_SX	xvec4, xvec6;
+MUL_SX	xvec0, xvec4;
+ADD_SX	xvec4, xvec11;
+MUL_SX	xvec1, xvec6;
+ADD_SX	xvec6, xvec10;
+
+MOV_SX	xvec5, xvec7;
+MUL_SX	xvec0, xvec5;
+ADD_SX	xvec5, xvec9;
+LD_SX	8*SIZE(ptrba), xvec0;
+MUL_SX	xvec1, xvec7;
+ADD_SX	xvec7, xvec8;
+LD_SX	12*SIZE(ptrba), xvec1;
+
+#### Unroll time 2 ####
+ODUP_SX	4*SIZE(ptrbb), xvec3;
+SHUF_SX	$0x4e, xvec2, xvec4;
+MOV_SX	xvec2, xvec6;
+MUL_SX	xvec0, xvec2;
+ADD_SX	xvec2, xvec15;
+MUL_SX	xvec1, xvec6;
+ADD_SX	xvec6, xvec14;
+
+SHUF_SX	$0x4e, xvec3, xvec5;
+MOV_SX	xvec3, xvec7;
+MUL_SX	xvec0, xvec3;
+ADD_SX	xvec3, xvec13;
+MUL_SX	xvec1, xvec7;
+ADD_SX	xvec7, xvec12;
+
+EDUP_SX	8*SIZE(ptrbb), xvec2;
+MOV_SX	xvec4, xvec6;
+MUL_SX	xvec0, xvec4;
+ADD_SX	xvec4, xvec11;
+MUL_SX	xvec1, xvec6;
+ADD_SX	xvec6, xvec10;
+
+MOV_SX	xvec5, xvec7;
+MUL_SX	xvec0, xvec5;
+ADD_SX	xvec5, xvec9;
+LD_SX	16*SIZE(ptrba), xvec0;
+MUL_SX	xvec1, xvec7;
+ADD_SX	xvec7, xvec8;
+LD_SX	20*SIZE(ptrba), xvec1;
+
+#### Unroll time 3 ####
+ODUP_SX	8*SIZE(ptrbb), xvec3;
+SHUF_SX	$0x4e, xvec2, xvec4;
+MOV_SX	xvec2, xvec6;
+MUL_SX	xvec0, xvec2;
+ADD_SX	xvec2, xvec15;
+MUL_SX	xvec1, xvec6;
+ADD_SX	xvec6, xvec14;
+
+SHUF_SX	$0x4e, xvec3, xvec5;
+MOV_SX	xvec3, xvec7;
+MUL_SX	xvec0, xvec3;
+ADD_SX	xvec3, xvec13;
+MUL_SX	xvec1, xvec7;
+ADD_SX	xvec7, xvec12;
+
+EDUP_SX	12*SIZE(ptrbb), xvec2;
+MOV_SX	xvec4, xvec6;
+ADDQ	$16*SIZE, ptrbb;
+
+MUL_SX	xvec0, xvec4;
+ADD_SX	xvec4, xvec11;
+MUL_SX	xvec1, xvec6;
+ADD_SX	xvec6, xvec10;
+
+MOV_SX	xvec5, xvec7;
+MUL_SX	xvec0, xvec5;
+ADD_SX	xvec5, xvec9;
+LD_SX	24*SIZE(ptrba), xvec0;
+
+MUL_SX	xvec1, xvec7;
+ADD_SX	xvec7, xvec8;
+LD_SX	28*SIZE(ptrba), xvec1;
+ADDQ	$32*SIZE, ptrba;
+
+#### Unroll time 4 ####
+ODUP_SX	-4*SIZE(ptrbb), xvec3;
+SHUF_SX	$0x4e, xvec2, xvec4;
+MOV_SX	xvec2, xvec6;
+MUL_SX	xvec0, xvec2;
+ADD_SX	xvec2, xvec15;
+MUL_SX	xvec1, xvec6;
+ADD_SX	xvec6, xvec14;
+
+SHUF_SX	$0x4e, xvec3, xvec5;
+MOV_SX	xvec3, xvec7;
+MUL_SX	xvec0, xvec3;
+ADD_SX	xvec3, xvec13;
+MUL_SX	xvec1, xvec7;
+ADD_SX	xvec7, xvec12;
+
+EDUP_SX	0*SIZE(ptrbb), xvec2;
+MOV_SX	xvec4, xvec6;
+MUL_SX	xvec0, xvec4;
+ADD_SX	xvec4, xvec11;
+MUL_SX	xvec1, xvec6;
+ADD_SX	xvec6, xvec10;
+
+MOV_SX	xvec5, xvec7;
+MUL_SX	xvec0, xvec5;
+ADD_SX	xvec5, xvec9;
+LD_SX	0*SIZE(ptrba), xvec0;
+
+MUL_SX	xvec1, xvec7;
+ADD_SX	xvec7, xvec8;
+LD_SX	4*SIZE(ptrba), xvec1;
+DECQ	k;
+JG	.L211_bodyB;
+.align 16
+.L211_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk
+#else
+TEST $2, kkk;
+#endif
+JLE .L212_loopE;
+.align 16
+.L212_bodyB:
+#### Unroll time 1 ####
+ODUP_SX	0*SIZE(ptrbb), xvec3;
+SHUF_SX	$0x4e, xvec2, xvec4;
+MOV_SX	xvec2, xvec6;
+MUL_SX	xvec0, xvec2;
+ADD_SX	xvec2, xvec15;
+MUL_SX	xvec1, xvec6;
+ADD_SX	xvec6, xvec14;
+
+SHUF_SX	$0x4e, xvec3, xvec5;
+MOV_SX	xvec3, xvec7;
+MUL_SX	xvec0, xvec3;
+ADD_SX	xvec3, xvec13;
+MUL_SX	xvec1, xvec7;
+ADD_SX	xvec7, xvec12;
+
+EDUP_SX	4*SIZE(ptrbb), xvec2;
+MOV_SX	xvec4, xvec6;
+MUL_SX	xvec0, xvec4;
+ADD_SX	xvec4, xvec11;
+ADDQ	$8*SIZE, ptrbb;
+
+MUL_SX	xvec1, xvec6;
+ADD_SX	xvec6, xvec10;
+MOV_SX	xvec5, xvec7;
+MUL_SX	xvec0, xvec5;
+ADD_SX	xvec5, xvec9;
+LD_SX	8*SIZE(ptrba), xvec0;
+
+MUL_SX	xvec1, xvec7;
+ADD_SX	xvec7, xvec8;
+LD_SX	12*SIZE(ptrba), xvec1;
+ADDQ	$16*SIZE, ptrba;
+
+#### Unroll time 2 ####
+ODUP_SX	-4*SIZE(ptrbb), xvec3;
+SHUF_SX	$0x4e, xvec2, xvec4;
+MOV_SX	xvec2, xvec6;
+MUL_SX	xvec0, xvec2;
+ADD_SX	xvec2, xvec15;
+MUL_SX	xvec1, xvec6;
+ADD_SX	xvec6, xvec14;
+
+SHUF_SX	$0x4e, xvec3, xvec5;
+MOV_SX	xvec3, xvec7;
+MUL_SX	xvec0, xvec3;
+ADD_SX	xvec3, xvec13;
+MUL_SX	xvec1, xvec7;
+ADD_SX	xvec7, xvec12;
+
+EDUP_SX	0*SIZE(ptrbb), xvec2;
+MOV_SX	xvec4, xvec6;
+MUL_SX	xvec0, xvec4;
+ADD_SX	xvec4, xvec11;
+MUL_SX	xvec1, xvec6;
+ADD_SX	xvec6, xvec10;
+
+MOV_SX	xvec5, xvec7;
+MUL_SX	xvec0, xvec5;
+ADD_SX	xvec5, xvec9;
+LD_SX	0*SIZE(ptrba), xvec0;
+MUL_SX	xvec1, xvec7;
+ADD_SX	xvec7, xvec8;
+LD_SX	4*SIZE(ptrba), xvec1;
+
+.L212_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else
+TEST $1, kkk;
+#endif
+JLE .L213_loopE;
+.align 16
+.L213_bodyB:
+ODUP_SX	0*SIZE(ptrbb), xvec3;
+SHUF_SX	$0x4e, xvec2, xvec4;
+MOV_SX	xvec2, xvec6;
+MUL_SX	xvec0, xvec2;
+ADD_SX	xvec2, xvec15;
+ADDQ	$4*SIZE, ptrbb;
+
+SHUF_SX	$0x4e, xvec3, xvec5;
+MUL_SX	xvec1, xvec6;
+ADD_SX	xvec6, xvec14;
+MOV_SX	xvec3, xvec7;
+MUL_SX	xvec0, xvec3;
+ADD_SX	xvec3, xvec13;
+
+MUL_SX	xvec1, xvec7;
+ADD_SX	xvec7, xvec12;
+MOV_SX	xvec4, xvec6;
+ADDQ	$8*SIZE, ptrba;
+
+MUL_SX	xvec0, xvec4;
+ADD_SX	xvec4, xvec11;
+MUL_SX	xvec1, xvec6;
+ADD_SX	xvec6, xvec10;
+
+MOV_SX	xvec5, xvec7;
+MUL_SX	xvec0, xvec5;
+ADD_SX	xvec5, xvec9;
+MUL_SX	xvec1, xvec7;
+ADD_SX	xvec7, xvec8;
+
+.L213_loopE:
+#### Multiply Alpha ####
+BROAD_SX MEMALPHA, xvec7;
+MUL_SX	xvec7, xvec15;
+MUL_SX	xvec7, xvec14;
+MUL_SX	xvec7, xvec13;
+MUL_SX	xvec7, xvec12;
+MUL_SX	xvec7, xvec11;
+MUL_SX	xvec7, xvec10;
+MUL_SX	xvec7, xvec9;
+MUL_SX	xvec7, xvec8;
+#### Writing Back #### 
+#ifndef	TRMMKERNEL
+LDL_SX	0*SIZE(C0), xvec0;
+LDH_SX	2*SIZE(C1), xvec0;
+LDL_SX	4*SIZE(C0), xvec1;
+LDH_SX	6*SIZE(C1), xvec1;
+LDL_SX	0*SIZE(C0, ldc, 1), xvec2;
+LDH_SX	2*SIZE(C1, ldc, 1), xvec2;
+LDL_SX	4*SIZE(C0, ldc, 1), xvec3;
+LDH_SX	6*SIZE(C1, ldc, 1), xvec3;
+LDL_SX	0*SIZE(C1), xvec4;
+LDH_SX	2*SIZE(C0), xvec4;
+LDL_SX	4*SIZE(C1), xvec5;
+LDH_SX	6*SIZE(C0), xvec5;
+LDL_SX	0*SIZE(C1, ldc, 1), xvec6;
+LDH_SX	2*SIZE(C0, ldc, 1), xvec6;
+LDL_SX	4*SIZE(C1, ldc, 1), xvec7;
+LDH_SX	6*SIZE(C0, ldc, 1), xvec7;
+ADD_SX	xvec0, xvec15;
+ADD_SX	xvec1, xvec14;
+ADD_SX	xvec2, xvec13;
+ADD_SX	xvec3, xvec12;
+ADD_SX	xvec4, xvec11;
+ADD_SX	xvec5, xvec10;
+ADD_SX	xvec6, xvec9;
+ADD_SX	xvec7, xvec8;
+#endif
+STL_SX	xvec15, 0*SIZE(C0);
+STH_SX	xvec15, 2*SIZE(C1);
+STL_SX	xvec14, 4*SIZE(C0);
+STH_SX	xvec14, 6*SIZE(C1);
+STL_SX	xvec13, 0*SIZE(C0, ldc, 1);
+STH_SX	xvec13, 2*SIZE(C1, ldc, 1);
+STL_SX	xvec12, 4*SIZE(C0, ldc, 1);
+STH_SX	xvec12, 6*SIZE(C1, ldc, 1);
+STL_SX	xvec11, 0*SIZE(C1);
+STH_SX	xvec11, 2*SIZE(C0);
+STL_SX	xvec10, 4*SIZE(C1);
+STH_SX	xvec10, 6*SIZE(C0);
+STL_SX	xvec9, 0*SIZE(C1, ldc, 1);
+STH_SX	xvec9, 2*SIZE(C0, ldc, 1);
+STL_SX	xvec8, 4*SIZE(C1, ldc, 1);
+STH_SX	xvec8, 6*SIZE(C0, ldc, 1);
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) ||(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kkk, %rax;
+LEAQ 	(, %rax, SIZE), %rax;
+LEAQ	(ptrba, %rax, 8), ptrba;
+LEAQ 	(ptrbb, %rax, 4), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT)
+ADDQ	$8, kk
+#endif
+ADDQ	$8*SIZE, C0;
+ADDQ	$8*SIZE, C1;
+DECQ	i;
+JG	.L21_bodyB;
+.align 16
+.L21_loopE:
+TEST $4, bm;
+JLE .L22_loopE;
+.align 16
+.L22_bodyB:
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb, ptrbb;						
+#else 
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+LEAQ (, %rax, SIZE), %rax;
+LEAQ (ptrba, %rax, 4), ptrba;
+LEAQ (ptrbb, %rax, 4), ptrbb;
+#endif
+#### Initial Results ####
+XOR_SY	yvec15, yvec15, yvec15;
+XOR_SY	yvec14, yvec14, yvec14;
+XOR_SY	yvec13, yvec13, yvec13;
+XOR_SY	yvec12, yvec12, yvec12;
+#ifndef	TRMMKERNEL
+MOVQ bk, k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else
+MOVQ kk, %rax;
+#ifdef	LEFT
+ADDQ $4, %rax;
+#else 
+ADDQ $4, %rax;
+#endif
+MOVQ %rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L221_loopE;
+.align 16
+.L221_bodyB:
+LD_SX	0*SIZE(ptrba), xvec0;
+EDUP_SX	0*SIZE(ptrbb), xvec2;
+ODUP_SX	0*SIZE(ptrbb), xvec3;
+
+SHUF_SX	$0x4e, xvec2, xvec4;
+MUL_SX	xvec0, xvec2;
+ADD_SX	xvec2, xvec15;
+SHUF_SX	$0x4e, xvec3, xvec5;
+MUL_SX	xvec0, xvec3;
+ADD_SX	xvec3, xvec14;
+
+MUL_SX	xvec0, xvec4;
+ADD_SX	xvec4, xvec13;
+MUL_SX	xvec0, xvec5;
+ADD_SX	xvec5, xvec12;
+
+LD_SX	4*SIZE(ptrba), xvec1;
+EDUP_SX	4*SIZE(ptrbb), xvec2;
+ODUP_SX	4*SIZE(ptrbb), xvec3;
+
+SHUF_SX	$0x4e, xvec2, xvec4;
+MUL_SX	xvec1, xvec2;
+ADD_SX	xvec2, xvec15;
+SHUF_SX	$0x4e, xvec3, xvec5;
+MUL_SX	xvec1, xvec3;
+ADD_SX	xvec3, xvec14;
+
+MUL_SX	xvec1, xvec4;
+ADD_SX	xvec4, xvec13;
+MUL_SX	xvec1, xvec5;
+ADD_SX	xvec5, xvec12;
+
+LD_SX	8*SIZE(ptrba), xvec0;
+EDUP_SX	8*SIZE(ptrbb), xvec2;
+ODUP_SX	8*SIZE(ptrbb), xvec3;
+
+SHUF_SX	$0x4e, xvec2, xvec4;
+MUL_SX	xvec0, xvec2;
+ADD_SX	xvec2, xvec15;
+SHUF_SX	$0x4e, xvec3, xvec5;
+MUL_SX	xvec0, xvec3;
+ADD_SX	xvec3, xvec14;
+
+MUL_SX	xvec0, xvec4;
+ADD_SX	xvec4, xvec13;
+MUL_SX	xvec0, xvec5;
+ADD_SX	xvec5, xvec12;
+
+LD_SX	12*SIZE(ptrba), xvec1;
+EDUP_SX	12*SIZE(ptrbb), xvec2;
+ODUP_SX	12*SIZE(ptrbb), xvec3;
+
+SHUF_SX	$0x4e, xvec2, xvec4;
+MUL_SX	xvec1, xvec2;
+ADD_SX	xvec2, xvec15
+SHUF_SX	$0x4e, xvec3, xvec5;
+MUL_SX	xvec1, xvec3;
+ADD_SX	xvec3, xvec14;
+
+MUL_SX	xvec1, xvec4;
+ADD_SX	xvec4, xvec13;
+MUL_SX	xvec1, xvec5;
+ADD_SX	xvec5, xvec12;
+ADDQ	$16*SIZE, ptrba;
+ADDQ	$16*SIZE, ptrbb;
+
+DECQ	k;
+JG	.L221_bodyB;
+.align 16
+.L221_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else
+TEST $2, kkk;
+#endif
+JLE	.L222_loopE;
+.align 16
+.L222_bodyB:
+LD_SX	0*SIZE(ptrba), xvec0;
+EDUP_SX	0*SIZE(ptrbb), xvec2;
+ODUP_SX	0*SIZE(ptrbb), xvec3;
+
+SHUF_SX	$0x4e, xvec2, xvec4;
+MUL_SX	xvec0, xvec2;
+ADD_SX	xvec2, xvec15;
+SHUF_SX	$0x4e, xvec3, xvec5;
+MUL_SX	xvec0, xvec3;
+ADD_SX	xvec3, xvec14;
+
+MUL_SX	xvec0, xvec4;
+ADD_SX	xvec4, xvec13;
+MUL_SX	xvec0, xvec5;
+ADD_SX	xvec5, xvec12;
+
+LD_SX	4*SIZE(ptrba), xvec1;
+EDUP_SX	4*SIZE(ptrbb), xvec2;
+ODUP_SX	4*SIZE(ptrbb), xvec3;
+
+SHUF_SX	$0x4e, xvec2, xvec4;
+MUL_SX	xvec1, xvec2;
+ADD_SX	xvec2, xvec15;
+SHUF_SX	$0x4e, xvec3, xvec5;
+MUL_SX	xvec1, xvec3;
+ADD_SX	xvec3, xvec14;
+
+MUL_SX	xvec1, xvec4;
+ADD_SX	xvec4, xvec13
+MUL_SX	xvec1, xvec5;
+ADD_SX	xvec5, xvec12;
+ADDQ	$8*SIZE, ptrba;
+ADDQ	$8*SIZE, ptrbb;
+.L222_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else
+TEST $1, kkk;
+#endif
+JLE .L223_loopE;
+.align 16
+.L223_bodyB:
+LD_SX	0*SIZE(ptrba), xvec0;
+EDUP_SX	0*SIZE(ptrbb), xvec2;
+ODUP_SX	0*SIZE(ptrbb), xvec3;
+
+SHUF_SX	$0x4e, xvec2, xvec4;
+MUL_SX	xvec0, xvec2;
+ADD_SX	xvec2, xvec15;
+SHUF_SX	$0x4e, xvec3, xvec5;
+MUL_SX	xvec0, xvec3;
+ADD_SX	xvec3, xvec14;
+
+MUL_SX	xvec0, xvec4;
+ADD_SX	xvec4, xvec13;
+MUL_SX	xvec0, xvec5;
+ADD_SX	xvec5, xvec12;
+ADDQ	$4*SIZE, ptrba;
+ADDQ	$4*SIZE, ptrbb;
+.L223_loopE:
+#### Multiply Alpha ####
+BROAD_SX MEMALPHA, xvec7;
+MUL_SX	xvec7, xvec15;
+MUL_SX	xvec7, xvec14;
+MUL_SX	xvec7, xvec13;
+MUL_SX	xvec7, xvec12;
+#### Writing back ####
+#ifndef	TRMMKERNEL
+LDL_SX	0*SIZE(C0), xvec0;
+LDH_SX	2*SIZE(C1), xvec0;
+LDL_SX	0*SIZE(C0, ldc, 1), xvec1;
+LDH_SX	2*SIZE(C1, ldc, 1), xvec1;
+LDL_SX	0*SIZE(C1), xvec2;
+LDH_SX	2*SIZE(C0), xvec2;
+LDL_SX	0*SIZE(C1, ldc, 1), xvec3;
+LDH_SX	2*SIZE(C0, ldc, 1), xvec3;
+ADD_SX	xvec0, xvec15;
+ADD_SX	xvec1, xvec14;
+ADD_SX	xvec2, xvec13;
+ADD_SX	xvec3, xvec12;
+#endif
+STL_SX	xvec15, 0*SIZE(C0);
+STH_SX	xvec15, 2*SIZE(C1);
+STL_SX	xvec14, 0*SIZE(C0, ldc, 1);
+STH_SX	xvec14, 2*SIZE(C1, ldc, 1);
+STL_SX	xvec13, 0*SIZE(C1);
+STH_SX	xvec13, 2*SIZE(C0);
+STL_SX	xvec12, 0*SIZE(C1, ldc, 1);
+STH_SX	xvec12, 2*SIZE(C0, ldc, 1);
+#if	(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kkk, %rax;
+LEAQ	(, %rax, SIZE), %rax;
+LEAQ	(ptrba, %rax, 4), ptrba;
+LEAQ	(ptrbb, %rax, 4), ptrbb;
+#endif
+#if	defined(TRMMKERNEL)&&defined(LEFT)
+ADDQ	$4, kk
+#endif
+ADDQ	$4*SIZE, C0;
+ADDQ	$4*SIZE, C1;
+.L22_loopE:
+TEST $2, bm;
+JLE .L23_loopE;
+.align 16
+.L23_bodyB:
+#if	!defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb, ptrbb;						
+#else 
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+LEAQ (, %rax, SIZE), %rax;
+LEAQ (ptrba, %rax, 2), ptrba;
+LEAQ (ptrbb, %rax, 4), ptrbb
+#endif
+#### Initial ####
+XOR_SY	yvec15, yvec15, yvec15;
+XOR_SY	yvec14, yvec14, yvec14;
+#ifndef	TRMMKERNEL
+MOVQ bk, k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ 	kk, %rax;
+MOVQ 	%rax, kkk;
+#else 
+MOVQ	kk, %rax;
+#ifdef	LEFT
+ADDQ	$2, %rax;
+#else
+ADDQ	$4, %rax;
+#endif
+MOVQ	%rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L231_loopE;
+.align 16
+.L231_bodyB:
+LD_SX	0*SIZE(ptrba), xvec0;
+EDUP_SX	0*SIZE(ptrbb), xvec4;
+ODUP_SX	0*SIZE(ptrbb), xvec5;
+SHUF_SX	$0x44, xvec0, xvec1;
+MUL_SX	xvec1, xvec4;
+ADD_SX	xvec4, xvec15;
+MUL_SX	xvec1, xvec5;
+ADD_SX	xvec5, xvec14;
+
+SHUF_SX	$0xee, xvec0, xvec2;
+EDUP_SX	4*SIZE(ptrbb), xvec6;
+ODUP_SX	4*SIZE(ptrbb), xvec7;
+MUL_SX	xvec2, xvec6;
+ADD_SX	xvec6, xvec15;
+MUL_SX	xvec2, xvec7;
+ADD_SX	xvec7, xvec14;
+
+LD_SX	4*SIZE(ptrba), xvec0;
+EDUP_SX	8*SIZE(ptrbb), xvec4;
+ODUP_SX	8*SIZE(ptrbb), xvec5;
+SHUF_SX	$0x44, xvec0, xvec1;
+MUL_SX	xvec1, xvec4;
+ADD_SX	xvec4, xvec15;
+MUL_SX	xvec1, xvec5;
+ADD_SX	xvec5, xvec14;
+
+SHUF_SX	$0xee, xvec0, xvec2;
+EDUP_SX	12*SIZE(ptrbb), xvec6;
+ODUP_SX	12*SIZE(ptrbb), xvec7;
+MUL_SX	xvec2, xvec6;
+ADD_SX	xvec6, xvec15;
+MUL_SX	xvec2, xvec7;
+ADD_SX	xvec7, xvec14;
+
+ADDQ	$8*SIZE, ptrba;
+ADDQ	$16*SIZE, ptrbb;
+DECQ	k;
+JG .L231_bodyB;
+.align 16
+.L231_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else
+TEST $2, kkk;
+#endif
+JLE .L232_loopE;
+.align 16
+.L232_bodyB:
+LD_SX	0*SIZE(ptrba), xvec0;
+EDUP_SX	0*SIZE(ptrbb), xvec4;
+ODUP_SX	0*SIZE(ptrbb), xvec5;
+SHUF_SX	$0x44, xvec0, xvec1;
+MUL_SX	xvec1, xvec4;
+ADD_SX	xvec4, xvec15;
+MUL_SX	xvec1, xvec5;
+ADD_SX	xvec5, xvec14;
+
+SHUF_SX	$0xee, xvec0, xvec2;
+EDUP_SX	4*SIZE(ptrbb), xvec6;
+ODUP_SX	4*SIZE(ptrbb), xvec7;
+MUL_SX	xvec2, xvec6;
+ADD_SX	xvec6, xvec15;
+MUL_SX	xvec2, xvec7;
+ADD_SX	xvec7, xvec14;
+
+ADDQ	$4*SIZE, ptrba;
+ADDQ	$8*SIZE, ptrbb;
+.L232_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else
+TEST $1, kkk;
+#endif
+JLE .L233_loopE;
+.align 16
+.L233_bodyB:
+LD_SX	0*SIZE(ptrba), xvec0;
+EDUP_SX	0*SIZE(ptrbb), xvec4;
+ODUP_SX	0*SIZE(ptrbb), xvec5;
+SHUF_SX	$0x44, xvec0, xvec1;
+MUL_SX	xvec1, xvec4;
+ADD_SX	xvec4, xvec15;
+MUL_SX	xvec1, xvec5;
+ADD_SX	xvec5, xvec14;
+
+ADDQ	$2*SIZE, ptrba;
+ADDQ	$4*SIZE, ptrbb;
+.L233_loopE:
+#### Multiply Alpha ####
+BROAD_SY MEMALPHA, yvec7;
+MUL_SY	xvec7, xvec15, xvec15;
+MUL_SY	xvec7, xvec14, xvec14;
+#### Writing Back ####
+SHUF_SX	$0xee, xvec15, xvec13;
+SHUF_SX	$0xee, xvec14, xvec12;
+#ifndef	TRMMKERNEL
+ADD_SY	0*SIZE(C0), xvec15, xvec15;
+ADD_SY	0*SIZE(C0, ldc, 1), xvec14, xvec14;
+ADD_SY	0*SIZE(C1), xvec13, xvec13;
+ADD_SY	0*SIZE(C1, ldc, 1), xvec12, xvec12;
+#endif
+STL_SY	xvec15, 0*SIZE(C0);
+STL_SY	xvec14, 0*SIZE(C0, ldc, 1);
+STL_SY	xvec13, 0*SIZE(C1);
+STL_SY	xvec12, 0*SIZE(C1, ldc, 1);
+#if (defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kkk, %rax;
+LEAQ	(,%rax, SIZE), %rax;
+LEAQ	(ptrba, %rax, 2), ptrba;
+LEAQ	(ptrbb, %rax, 4), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT)
+ADDQ	$2, kk
+#endif
+ADDQ	$2*SIZE, C0;
+ADDQ	$2*SIZE, C1;
+.L23_loopE:
+TEST $1, bm;
+JLE .L24_loopE;
+.align 16
+.L24_bodyB:
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb, ptrbb;
+#else 
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+LEAQ (,%rax, SIZE), %rax;
+ADDQ	%rax, ptrba;
+LEAQ (ptrbb, %rax, 4), ptrbb;
+#endif
+#### Initial ####
+XOR_SY	yvec15, yvec15, yvec15;
+#ifndef	TRMMKERNEL
+MOVQ bk, k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kk, %rax;
+MOVQ	%rax, kkk;
+#else
+MOVQ	kk, %rax;
+#ifdef	LEFT
+ADDQ	$1, %rax;
+#else 
+ADDQ	$4, %rax;
+#endif
+MOVQ	%rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L241_loopE;
+.align 16
+.L241_bodyB:
+BROAD_SX 0*SIZE(ptrba), xvec0;
+LD_SX	0*SIZE(ptrbb), xvec1;
+MUL_SX	xvec0, xvec1;
+ADD_SX	xvec1, xvec15;
+
+BROAD_SX 1*SIZE(ptrba), xvec2;
+LD_SX	4*SIZE(ptrbb), xvec3;
+MUL_SX	xvec2, xvec3;
+ADD_SX	xvec3, xvec15;
+
+BROAD_SX 2*SIZE(ptrba), xvec4;
+LD_SX	8*SIZE(ptrbb), xvec5;
+MUL_SX	xvec4, xvec5;
+ADD_SX	xvec5, xvec15;
+
+BROAD_SX 3*SIZE(ptrba), xvec6;
+LD_SX	12*SIZE(ptrbb), xvec7;
+MUL_SX	xvec6, xvec7;
+ADD_SX	xvec7, xvec15;
+ADDQ	$4*SIZE, ptrba;
+ADDQ	$16*SIZE, ptrbb;
+DECQ	k;
+JG .L241_bodyB;
+.align
+.L241_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else
+TEST $2, kkk;
+#endif
+JLE	.L242_loopE;
+.align 16
+.L242_bodyB:
+BROAD_SX 0*SIZE(ptrba), xvec0;
+LD_SX	0*SIZE(ptrbb), xvec1;
+MUL_SX	xvec0, xvec1;
+ADD_SX	xvec1, xvec15;
+
+BROAD_SX 1*SIZE(ptrba), xvec2;
+LD_SX	4*SIZE(ptrbb), xvec3;
+MUL_SX	xvec2, xvec3;
+ADD_SX	xvec3, xvec15;
+ADDQ	$2*SIZE, ptrba;
+ADDQ	$8*SIZE, ptrbb;
+
+.L242_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else
+TEST $1, kkk;
+#endif
+JLE .L243_loopE;
+.align 16;
+.L243_bodyB:
+BROAD_SX 0*SIZE(ptrba), xvec0;
+LD_SX	0*SIZE(ptrbb), xvec1;
+MUL_SX	xvec0, xvec1;
+ADD_SX	xvec1, xvec15;
+ADDQ	$1*SIZE, ptrba;
+ADDQ	$4*SIZE, ptrbb;
+.L243_loopE:
+#### Multiply Alpha ####
+BROAD_SX MEMALPHA, xvec7;
+MUL_SX	xvec7, xvec15;
+SHUF_SX $0xff, xvec15, xvec14;
+SHUF_SX $0xaa, xvec15, xvec13;
+SHUF_SX $0x55, xvec15, xvec12;
+SHUF_SX $0x00, xvec15, xvec11;
+
+#ifndef	TRMMKERNEL
+addss	0*SIZE(C0), xvec11;
+addss	0*SIZE(C0, ldc, 1), xvec12;
+addss	0*SIZE(C1), xvec13;
+addss	0*SIZE(C1, ldc, 1), xvec14;
+#endif
+
+movss	xvec11, 0*SIZE(C0);
+movss	xvec12, 0*SIZE(C0, ldc, 1);
+movss	xvec13, 0*SIZE(C1);
+movss	xvec14, 0*SIZE(C1, ldc, 1);
+#if (defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kkk, %rax;
+LEAQ	(,%rax, SIZE), %rax;
+ADDQ	%rax, ptrba;
+LEAQ	(ptrbb, %rax, 4), ptrbb;
+#endif
+#if defined(TRMMKERNEL)&&defined(LEFT)
+ADDQ	$1, kk
+#endif
+ADDQ	$1*SIZE, C0;
+ADDQ	$1*SIZE, C1;
+.L24_loopE:
+#if	defined(TRMMKERNEL)&&!defined(LEFT)
+ADDQ	$4, kk
+#endif
+MOVQ bk, k;
+SALQ $4, k;
+ADDQ k, bb;
+LEAQ (C, ldc, 4), C;
+.L20_loopE:
+TEST $2, bn;
+JLE	.L30_loopE;
+.align 16
+.L30_bodyB:
+#if defined(TRMMKERNEL) && defined(LEFT)
+MOVQ	OFFSET, %rax;
+MOVQ	%rax, kk
+#endif
+MOVQ C, C0;
+LEAQ (C, ldc, 1), C1;
+MOVQ ba, ptrba;
+MOVQ bm, i;
+SARQ $3, i;
+JLE .L31_loopE;
+.align 16
+.L31_bodyB:
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb, ptrbb;
+#else 
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+LEAQ (, %rax, SIZE), %rax;
+LEAQ (ptrba, %rax, 8), ptrba;
+LEAQ (ptrbb, %rax, 2), ptrbb;
+#endif
+#### Initial ####
+XOR_SY	yvec15, yvec15, yvec15;
+XOR_SY	yvec14, yvec14, yvec14;
+XOR_SY	yvec13, yvec13, yvec13;
+XOR_SY	yvec12, yvec12, yvec12;
+#ifndef	TRMMKERNEL
+MOVQ bk, k;
+#elif (defined(LEFT) && !defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kk, %rax;
+MOVQ	%rax, kkk;
+#else 
+MOVQ	kk, %rax;
+#ifdef	LEFT
+ADDQ	$8, %rax;
+#else 
+ADDQ	$2, %rax;
+#endif
+MOVQ	%rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L311_loopE;
+.align 16
+.L311_bodyB:
+LD_SX	0*SIZE(ptrbb), xvec2;
+SHUF_SX $0x50, xvec2, xvec3;
+LD_SX	0*SIZE(ptrba), xvec0;
+LD_SX	4*SIZE(ptrba), xvec1;
+
+MOV_SX	xvec3, xvec4;
+MUL_SX	xvec0, xvec3;
+ADD_SX	xvec3, xvec15;
+SHUF_SX $0x4e, xvec4, xvec5;
+MUL_SX	xvec1, xvec4;
+ADD_SX	xvec4, xvec14;
+
+MOV_SX	xvec5, xvec6;
+MUL_SX	xvec0, xvec5;
+ADD_SX	xvec5, xvec13;
+MUL_SX	xvec1, xvec6;
+ADD_SX	xvec6, xvec12;
+
+SHUF_SX	$0xfa, xvec2, xvec3;
+LD_SX	8*SIZE(ptrba), xvec0;
+LD_SX	12*SIZE(ptrba), xvec1;
+
+MOV_SX	xvec3, xvec4;
+MUL_SX	xvec0, xvec3;
+ADD_SX	xvec3, xvec15;
+SHUF_SX	$0x4e, xvec4, xvec5;
+MUL_SX	xvec1, xvec4;
+ADD_SX	xvec4, xvec14;
+
+MOV_SX	xvec5, xvec6;
+MUL_SX	xvec0, xvec5;
+ADD_SX	xvec5, xvec13;
+MUL_SX	xvec1, xvec6;
+ADD_SX	xvec6, xvec12;
+
+LD_SX	4*SIZE(ptrbb), xvec2;
+SHUF_SX	$0x50, xvec2, xvec3;
+LD_SX	16*SIZE(ptrba), xvec0;
+LD_SX	20*SIZE(ptrba), xvec1;
+
+MOV_SX	xvec3, xvec4;
+MUL_SX	xvec0, xvec3;
+ADD_SX	xvec3, xvec15;
+SHUF_SX	$0x4e, xvec4, xvec5;
+MUL_SX	xvec1, xvec4;
+ADD_SX	xvec4, xvec14;
+
+MOV_SX	xvec5, xvec6;
+MUL_SX	xvec0, xvec5;
+ADD_SX	xvec5, xvec13;
+MUL_SX	xvec1, xvec6;
+ADD_SX	xvec6, xvec12;
+
+SHUF_SX $0xfa, xvec2, xvec3;
+LD_SX	24*SIZE(ptrba), xvec0;
+LD_SX	28*SIZE(ptrba), xvec1;
+
+MOV_SX	xvec3, xvec4;
+MUL_SX	xvec0, xvec3;
+ADD_SX	xvec3, xvec15;
+SHUF_SX	$0x4e, xvec4, xvec5;
+MUL_SX	xvec1, xvec4;
+ADD_SX	xvec4, xvec14;
+
+MOV_SX	xvec5, xvec6;
+MUL_SX	xvec0, xvec5;
+ADD_SX	xvec5, xvec13;
+MUL_SX	xvec1, xvec6;
+ADD_SX	xvec6, xvec12;
+
+ADDQ	$32*SIZE, ptrba;
+ADDQ	$8*SIZE, ptrbb;
+DECQ	k;
+JG .L311_bodyB;
+.align 16
+.L311_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else
+TEST $2, kkk;
+#endif
+JLE .L312_loopE;
+.align 16
+.L312_bodyB:
+LD_SX	0*SIZE(ptrbb), xvec2;
+SHUF_SX $0x50, xvec2, xvec3;
+LD_SX	0*SIZE(ptrba), xvec0;
+LD_SX	4*SIZE(ptrba), xvec1;
+
+MOV_SX	xvec3, xvec4;
+MUL_SX	xvec0, xvec3;
+ADD_SX	xvec3, xvec15;
+SHUF_SX $0x4e, xvec4, xvec5;
+MUL_SX	xvec1, xvec4;
+ADD_SX	xvec4, xvec14;
+
+MOV_SX	xvec5, xvec6;
+MUL_SX	xvec0, xvec5;
+ADD_SX	xvec5, xvec13;
+MUL_SX	xvec1, xvec6;
+ADD_SX	xvec6, xvec12;
+
+SHUF_SX	$0xfa, xvec2, xvec3;
+LD_SX	8*SIZE(ptrba), xvec0;
+LD_SX	12*SIZE(ptrba), xvec1;
+
+MOV_SX	xvec3, xvec4;
+MUL_SX	xvec0, xvec3;
+ADD_SX	xvec3, xvec15;
+SHUF_SX	$0x4e, xvec4, xvec5;
+MUL_SX	xvec1, xvec4;
+ADD_SX	xvec4, xvec14;
+
+MOV_SX	xvec5, xvec6;
+MUL_SX	xvec0, xvec5;
+ADD_SX	xvec5, xvec13;
+MUL_SX	xvec1, xvec6;
+ADD_SX	xvec6, xvec12;
+ADDQ	$16*SIZE, ptrba;
+ADDQ	$4*SIZE, ptrbb;
+
+.L312_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else
+TEST $1, kkk;
+#endif
+JLE .L313_loopE;
+.align 16
+.L313_bodyB:
+LD_SX	0*SIZE(ptrbb), xvec2;
+SHUF_SX $0x50, xvec2, xvec3;
+LD_SX	0*SIZE(ptrba), xvec0;
+LD_SX	4*SIZE(ptrba), xvec1;
+
+MOV_SX	xvec3, xvec4;
+MUL_SX	xvec0, xvec3;
+ADD_SX	xvec3, xvec15;
+SHUF_SX $0x4e, xvec4, xvec5;
+MUL_SX	xvec1, xvec4;
+ADD_SX	xvec4, xvec14;
+
+MOV_SX	xvec5, xvec6;
+MUL_SX	xvec0, xvec5;
+ADD_SX	xvec5, xvec13;
+MUL_SX	xvec1, xvec6;
+ADD_SX	xvec6, xvec12;
+ADDQ	$8*SIZE, ptrba;
+ADDQ	$2*SIZE, ptrbb;
+
+.L313_loopE:
+BROAD_SX MEMALPHA, xvec7;
+MUL_SX	xvec7, xvec15;
+MUL_SX	xvec7, xvec14;
+MUL_SX	xvec7, xvec13;
+MUL_SX	xvec7, xvec12;
+#### Writing Back ####
+#ifndef	TRMMKERNEL
+LDL_SX	0*SIZE(C0), xvec0;
+LDH_SX	2*SIZE(C1), xvec0;
+LDL_SX	4*SIZE(C0), xvec1;
+LDH_SX	6*SIZE(C1), xvec1;
+LDL_SX	0*SIZE(C1), xvec2;
+LDH_SX	2*SIZE(C0), xvec2;
+LDL_SX	4*SIZE(C1), xvec3;
+LDH_SX	6*SIZE(C0), xvec3;
+ADD_SX	xvec0, xvec15;
+ADD_SX	xvec1, xvec14;
+ADD_SX	xvec2, xvec13;
+ADD_SX	xvec3, xvec12;
+#endif
+STL_SX	xvec15, 0*SIZE(C0);
+STH_SX	xvec15, 2*SIZE(C1);
+STL_SX	xvec14, 4*SIZE(C0);
+STH_SX	xvec14, 6*SIZE(C1);
+STL_SX	xvec13, 0*SIZE(C1);
+STH_SX	xvec13, 2*SIZE(C0);
+STL_SX	xvec12, 4*SIZE(C1);
+STH_SX	xvec12, 6*SIZE(C0);
+#if (defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kkk, %rax;
+LEAQ	(,%rax, SIZE), %rax;
+LEAQ 	(ptrba, %rax, 8), ptrba;
+LEAQ	(ptrbb, %rax, 2), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT)
+ADDQ	$8, kk
+#endif
+ADDQ	$8*SIZE, C0;
+ADDQ	$8*SIZE, C1;
+DECQ	i;
+JG .L31_bodyB;
+.align 16
+.L31_loopE:
+TEST $4, bm;
+JLE .L32_loopE;
+.align 16
+.L32_bodyB:
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb, ptrbb;
+#else 
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+LEAQ	(,%rax, SIZE), %rax;
+LEAQ	(ptrba, %rax, 4), ptrba;
+LEAQ	(ptrbb, %rax, 2), ptrbb;
+#endif
+#### Initial ####
+XOR_SY	yvec15, yvec15, yvec15;
+XOR_SY	yvec14, yvec14, yvec14;
+#ifndef	TRMMKERNEL
+MOVQ bk, k;
+#elif (defined(LEFT) && !defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kk, %rax;
+MOVQ	%rax, kkk;
+#else 
+MOVQ	kk, %rax;
+#ifdef	LEFT
+ADDQ	$4, %rax;
+#else 
+ADDQ	$2, %rax;
+#endif
+MOVQ	%rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L321_loopE;
+.align 16
+.L321_bodyB:
+LD_SX	0*SIZE(ptrba), xvec0;
+LD_SX	0*SIZE(ptrbb), xvec2;
+SHUF_SX	$0x50, xvec2, xvec3;
+SHUF_SX $0x05, xvec2, xvec4;
+MUL_SX	xvec0, xvec3;
+ADD_SX	xvec3, xvec15;
+MUL_SX	xvec0, xvec4;
+ADD_SX	xvec4, xvec14;
+
+LD_SX	4*SIZE(ptrba), xvec0;
+SHUF_SX	$0xfa, xvec2, xvec5;
+SHUF_SX	$0xaf, xvec2, xvec6;
+MUL_SX	xvec0, xvec5;
+ADD_SX	xvec5, xvec15;
+MUL_SX	xvec0, xvec6;
+ADD_SX	xvec6, xvec14;
+
+LD_SX	8*SIZE(ptrba), xvec0;
+LD_SX	4*SIZE(ptrbb), xvec2;
+SHUF_SX	$0x50, xvec2, xvec3;
+SHUF_SX	$0x05, xvec2, xvec4;
+MUL_SX	xvec0, xvec3;
+ADD_SX	xvec3, xvec15;
+MUL_SX	xvec0, xvec4;
+ADD_SX	xvec4, xvec14;
+
+LD_SX	12*SIZE(ptrba), xvec0;
+SHUF_SX	$0xfa, xvec2, xvec5;
+SHUF_SX	$0xaf, xvec2, xvec6;
+MUL_SX	xvec0, xvec5;
+ADD_SX	xvec5, xvec15;
+MUL_SX	xvec0, xvec6;
+ADD_SX	xvec6, xvec14;
+
+ADDQ	$16*SIZE, ptrba;
+ADDQ	$8*SIZE, ptrbb;
+DECQ	k;
+JG .L321_bodyB;
+.align 16
+.L321_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else
+TEST $2, kkk;
+#endif
+JLE .L322_loopE;
+.align 16
+.L322_bodyB:
+LD_SX	0*SIZE(ptrba), xvec0;
+LD_SX	0*SIZE(ptrbb), xvec2;
+SHUF_SX	$0x50, xvec2, xvec3;
+SHUF_SX $0x05, xvec2, xvec4;
+MUL_SX	xvec0, xvec3;
+ADD_SX	xvec3, xvec15;
+MUL_SX	xvec0, xvec4;
+ADD_SX	xvec4, xvec14;
+
+LD_SX	4*SIZE(ptrba), xvec0;
+SHUF_SX	$0xfa, xvec2, xvec5;
+SHUF_SX	$0xaf, xvec2, xvec6;
+MUL_SX	xvec0, xvec5;
+ADD_SX	xvec5, xvec15;
+MUL_SX	xvec0, xvec6;
+ADD_SX	xvec6, xvec14;
+ADDQ	$8*SIZE, ptrba;
+ADDQ	$4*SIZE, ptrbb;
+
+.L322_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else
+TEST $1, kkk;
+#endif
+JLE .L323_loopE;
+.align 16
+.L323_bodyB:
+LD_SX	0*SIZE(ptrba), xvec0;
+LD_SX	0*SIZE(ptrbb), xvec2;
+SHUF_SX	$0x50, xvec2, xvec3;
+SHUF_SX $0x05, xvec2, xvec4;
+MUL_SX	xvec0, xvec3;
+ADD_SX	xvec3, xvec15;
+MUL_SX	xvec0, xvec4;
+ADD_SX	xvec4, xvec14;
+ADDQ	$4*SIZE, ptrba;
+ADDQ	$2*SIZE, ptrbb;
+
+.L323_loopE:
+BROAD_SX MEMALPHA, xvec7;
+MUL_SX	xvec7, xvec15;
+MUL_SX	xvec7, xvec14;
+#### Writing back ####
+#ifndef	TRMMKERNEL
+LDL_SX	0*SIZE(C0), xvec0;
+LDH_SX	2*SIZE(C1), xvec0;
+LDL_SX	0*SIZE(C1), xvec1;
+LDH_SX	2*SIZE(C0), xvec1;
+ADD_SX	xvec0, xvec15;
+ADD_SX	xvec1, xvec14;
+#endif
+STL_SX	xvec15, 0*SIZE(C0);
+STH_SX	xvec15, 2*SIZE(C1);
+STL_SX	xvec14, 0*SIZE(C1);
+STH_SX	xvec14, 2*SIZE(C0);
+#if (defined(TRMMKERNEL)&& defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&& !defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kkk, %rax;
+LEAQ	(,%rax, SIZE), %rax;
+LEAQ	(ptrba, %rax, 4), ptrba;
+LEAQ	(ptrbb, %rax, 2), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT)
+ADDQ	$4, kk
+#endif
+ADDQ	$4*SIZE, C0;
+ADDQ	$4*SIZE, C1;
+.L32_loopE:
+TEST $2, bm;
+JLE .L33_loopE;
+.align 16
+.L33_bodyB:
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb, ptrbb;
+#else 
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+LEAQ	(,%rax, SIZE), %rax;
+LEAQ	(ptrba, %rax, 2), ptrba;
+LEAQ	(ptrbb, %rax, 2), ptrbb;
+#endif
+#### Initial ####
+XOR_SY	yvec15, yvec15, yvec15;
+XOR_SY	yvec14, yvec14, yvec14;
+XOR_SY	yvec13, yvec13, yvec13;
+XOR_SY	yvec12, yvec12, yvec12;
+#ifndef	TRMMKERNEL
+MOVQ bk, k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kk, %rax;
+MOVQ	%rax, kkk;
+#else 
+MOVQ	kk, %rax;
+#ifdef	LEFT
+ADDQ	$2, %rax;
+#else 
+ADDQ	$2, %rax;
+#endif
+MOVQ	%rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L331_loopE;
+.align 16
+.L331_bodyB:
+LD_SX	0*SIZE(ptrba), xvec0;		# a0, a1, a2, a3
+EDUP_SX	0*SIZE(ptrbb), xvec2;		# b0, b0, b2, b2
+ODUP_SX	0*SIZE(ptrbb), xvec3;		# b1, b1, b3, b3
+
+MUL_SX	xvec0, xvec2;				# c00, c10
+ADD_SX	xvec2, xvec15;
+MUL_SX	xvec0, xvec3;				# C01, c11
+ADD_SX	xvec3, xvec14;
+
+LD_SX	4*SIZE(ptrba), xvec0;
+EDUP_SX	4*SIZE(ptrbb), xvec2;
+ODUP_SX	4*SIZE(ptrbb), xvec3;
+
+MUL_SX	xvec0, xvec2;
+ADD_SX	xvec2, xvec15;
+MUL_SX	xvec0, xvec3;
+ADD_SX	xvec3, xvec14;
+ADDQ	$8*SIZE, ptrba;
+ADDQ	$8*SIZE, ptrbb;
+DECQ	k;
+JG .L331_bodyB;
+.align 16
+.L331_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else
+TEST $2, kkk;
+#endif
+JLE .L332_loopE;
+.align 16
+.L332_bodyB:
+LD_SX	0*SIZE(ptrba), xvec0;		# a0, a1, a2, a3
+EDUP_SX	0*SIZE(ptrbb), xvec2;		# b0, b0, b2, b2
+ODUP_SX	0*SIZE(ptrbb), xvec3;		# b1, b1, b3, b3
+
+MUL_SX	xvec0, xvec2;				# c00, c10
+ADD_SX	xvec2, xvec15;
+MUL_SX	xvec0, xvec3;				# C01, c11
+ADD_SX	xvec3, xvec14;
+
+ADDQ	$4*SIZE, ptrba;
+ADDQ	$4*SIZE, ptrbb;
+
+.L332_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else
+TEST $1, kkk;
+#endif
+JLE .L333_loopE;
+.align 16
+.L333_bodyB:
+movss	0*SIZE(ptrba), xvec0;
+movss	1*SIZE(ptrba), xvec1;
+movss	0*SIZE(ptrbb), xvec2;
+XOR_SY	yvec3, yvec3, yvec3;
+movss	xvec2, xvec3;
+mulss	xvec0, xvec2;
+addss	xvec2, xvec15;
+mulss	xvec1, xvec3;
+SHUF_SX	$0xe1, xvec3, xvec4;
+ADD_SX	xvec4, xvec15;
+
+movss	1*SIZE(ptrbb), xvec5;
+XOR_SY	yvec6, yvec6, yvec6;
+movss	xvec5, xvec6;
+mulss	xvec0, xvec5;
+addss	xvec5, xvec14;
+mulss	xvec1, xvec6;
+SHUF_SX	$0xe1, xvec6, xvec7;
+ADD_SX	xvec7, xvec14
+
+ADDQ	$2*SIZE, ptrba;
+ADDQ	$2*SIZE, ptrbb;
+.L333_loopE:
+BROAD_SX MEMALPHA, xvec7;
+MUL_SX	xvec7, xvec15;
+MUL_SX	xvec7, xvec14;
+SHUF_SX	$0xee, xvec15, xvec13;
+SHUF_SX	$0xee, xvec14, xvec12;
+SHUF_SX	$0x44, xvec15, xvec11;
+SHUF_SX	$0x44, xvec14, xvec10;
+ADD_SX	xvec13, xvec11;
+ADD_SX	xvec12, xvec10;
+
+#ifndef	TRMMKERNEL
+LDL_SX	0*SIZE(C0), xvec0;
+LDL_SX	0*SIZE(C1), xvec1;
+ADD_SX	xvec0, xvec11;
+ADD_SX	xvec1, xvec10;
+#endif
+STL_SX	xvec11, 0*SIZE(C0);
+STL_SX	xvec10, 0*SIZE(C1);
+#if (defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kkk, %rax;
+LEAQ	(,%rax, SIZE), %rax;
+LEAQ	(ptrba, %rax, 2), ptrba;
+LEAQ	(ptrbb, %rax, 2), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT)
+ADDQ	$2, kk;
+#endif
+ADDQ	$2*SIZE, C0;
+ADDQ	$2*SIZE, C1;
+#### Writing Back ####
+.L33_loopE:
+TEST $1, bm;
+JLE .L34_loopE;
+.align 16
+.L34_bodyB:
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb, ptrbb;
+#else 
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+LEAQ	(, %rax, SIZE), %rax;
+ADDQ	%rax, ptrba;
+LEAQ	(ptrbb, %rax, 2), ptrbb;
+#endif
+#### Initial ####
+XOR_SY	yvec15, yvec15, yvec15;
+XOR_SY	yvec14, yvec14, yvec14;
+#ifndef	TRMMKERNEL
+MOVQ bk, k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else 
+MOVQ kk, %rax;
+#ifdef	LEFT
+ADDQ	$1, %rax;
+#else 
+ADDQ	$2, %rax;
+#endif	
+MOVQ	%rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L341_loopE;
+.align 16
+.L341_bodyB:
+movss	0*SIZE(ptrba), xvec0;
+movss	0*SIZE(ptrbb), xvec1;
+mulss	xvec0, xvec1;
+addss	xvec1, xvec15;
+
+movss	1*SIZE(ptrbb), xvec2;
+mulss	xvec0, xvec2;
+addss	xvec2, xvec14;
+
+movss	1*SIZE(ptrba), xvec0;
+movss	2*SIZE(ptrbb), xvec1;
+mulss	xvec0, xvec1;
+addss	xvec1, xvec15;
+
+movss	3*SIZE(ptrbb), xvec2;
+mulss	xvec0, xvec2;
+addss	xvec2, xvec14;
+
+movss	2*SIZE(ptrba), xvec0;
+movss	4*SIZE(ptrbb), xvec1;
+mulss	xvec0, xvec1;
+addss	xvec1, xvec15;
+
+movss	5*SIZE(ptrbb), xvec2;
+mulss	xvec0, xvec2;
+addss	xvec2, xvec14;
+
+movss	3*SIZE(ptrba), xvec0;
+movss	6*SIZE(ptrbb), xvec1;
+mulss	xvec0, xvec1;
+addss	xvec1, xvec15;
+
+movss	7*SIZE(ptrbb), xvec2;
+mulss	xvec0, xvec2;
+addss	xvec2, xvec14;
+
+addq	$4*SIZE, ptrba;
+addq	$8*SIZE, ptrbb;
+decq	k;
+jg	.L341_bodyB;
+.align 16
+.L341_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else
+TEST $2, kkk;
+#endif
+JLE .L342_loopE;
+.align 16
+.L342_bodyB:
+movss	0*SIZE(ptrba), xvec0;
+movss	0*SIZE(ptrbb), xvec1;
+mulss	xvec0, xvec1;
+addss	xvec1, xvec15;
+
+movss	1*SIZE(ptrbb), xvec2;
+mulss	xvec0, xvec2;
+addss	xvec2, xvec14;
+
+movss	1*SIZE(ptrba), xvec0;
+movss	2*SIZE(ptrbb), xvec1;
+mulss	xvec0, xvec1;
+addss	xvec1, xvec15;
+
+movss	3*SIZE(ptrbb), xvec2;
+mulss	xvec0, xvec2;
+addss	xvec2, xvec14;
+addq	$2*SIZE, ptrba;
+addq	$4*SIZE, ptrbb;
+.L342_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else
+TEST $1, kkk;
+#endif
+JLE .L343_loopE;
+.align 16
+.L343_bodyB:
+movss	0*SIZE(ptrba), xvec0;
+movss	0*SIZE(ptrbb), xvec1;
+mulss	xvec0, xvec1;
+addss	xvec1, xvec15;
+
+movss	1*SIZE(ptrbb), xvec2;
+mulss	xvec0, xvec2;
+addss	xvec2, xvec14;
+addq	$1*SIZE, ptrba;
+addq	$2*SIZE, ptrbb
+
+.L343_loopE:
+#### Writing back #### 
+movss	MEMALPHA, xvec7;
+mulss	xvec7, xvec15;
+mulss	xvec7, xvec14;
+movss	0*SIZE(C0), xvec0;
+movss	0*SIZE(C1), xvec1;
+#ifndef	TRMMKERNEL
+addss	xvec0, xvec15;
+addss	xvec1, xvec14;
+#endif
+movss	xvec15, 0*SIZE(C0);
+movss	xvec14, 0*SIZE(C1);
+#if (defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ	bk, %rax;
+SUBQ	kkk, %rax;
+LEAQ	(,%rax, SIZE), %rax;
+ADDQ	%rax, ptrba;
+LEAQ	(ptrbb, %rax, 2), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT)
+ADDQ	$1, kk;
+#endif
+addq	$1*SIZE, C0;
+addq	$1*SIZE, C1;
+.L34_loopE:
+#if defined(TRMMKERNEL) && !defined(LEFT)
+ADDQ	$2, kk;
+#endif
+MOVQ bk, k;
+SALQ $3, k;
+ADDQ k, bb;
+LEAQ (C, ldc, 2), C;
+.L30_loopE:
+TEST $1, bn;
+JLE .L40_loopE;
+.align 16
+.L40_bodyB:
+#if defined(TRMMKERNEL)&&defined(LEFT)
+MOVQ	OFFSET, %rax;
+MOVQ	%rax, kk;
+#endif
+MOVQ C, C0;
+MOVQ ba, ptrba;
+MOVQ bm, i;
+SARQ $3, i;
+JLE .L41_loopE;
+.align 16
+.L41_bodyB:
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb, ptrbb;
+#else 
+MOVQ bb, ptrbb;
+MOVQ kk, %rax
+LEAQ (, %rax, SIZE), %rax;
+LEAQ (ptrba, %rax, 8), ptrba;
+ADDQ	%rax, ptrbb;
+#endif
+#### initial ####
+XOR_SY	yvec15, yvec15, yvec15;
+#ifndef	TRMMKERNEL
+MOVQ bk, k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else 
+MOVQ kk, %rax;
+#ifdef	LEFT
+ADDQ $8, %rax;
+#else 
+ADDQ $1, %rax;
+#endif
+MOVQ %rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L411_loopE;
+.align 16
+.L411_bodyB:
+LD_SY		0*SIZE(ptrba), yvec0;
+BROAD_SY	0*SIZE(ptrbb), yvec1;
+MUL_SY	yvec0, yvec1, yvec2;
+ADD_SY	yvec2, yvec15, yvec15;
+
+LD_SY		8*SIZE(ptrba), yvec0;
+BROAD_SY	1*SIZE(ptrbb), yvec1;
+MUL_SY	yvec0, yvec1, yvec2;
+ADD_SY	yvec2, yvec15, yvec15;
+
+LD_SY		16*SIZE(ptrba), yvec0;
+BROAD_SY	2*SIZE(ptrbb), yvec1;
+MUL_SY	yvec0, yvec1, yvec2;
+ADD_SY	yvec2, yvec15, yvec15;
+
+LD_SY		24*SIZE(ptrba), yvec0;
+BROAD_SY	3*SIZE(ptrbb), yvec1;
+MUL_SY	yvec0, yvec1, yvec2;
+ADD_SY	yvec2, yvec15, yvec15;
+
+ADDQ	$32*SIZE, ptrba;
+ADDQ	$4*SIZE, ptrbb;
+DECQ	k;
+JG .L411_bodyB;
+.align 16
+.L411_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else
+TEST $2, kkk;
+#endif
+JLE .L412_loopE;
+.align 16
+.L412_bodyB:
+LD_SY		0*SIZE(ptrba), yvec0;
+BROAD_SY	0*SIZE(ptrbb), yvec1;
+MUL_SY	yvec0, yvec1, yvec2;
+ADD_SY	yvec2, yvec15, yvec15;
+
+LD_SY		8*SIZE(ptrba), yvec0;
+BROAD_SY	1*SIZE(ptrbb), yvec1;
+MUL_SY	yvec0, yvec1, yvec2;
+ADD_SY	yvec2, yvec15, yvec15;
+
+ADDQ	$16*SIZE, ptrba;
+ADDQ	$2*SIZE, ptrbb;
+.L412_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else
+TEST $1, kkk;
+#endif
+JLE .L413_loopE;
+.align 16
+.L413_bodyB:
+LD_SY		0*SIZE(ptrba), yvec0;
+BROAD_SY	0*SIZE(ptrbb), yvec1;
+MUL_SY	yvec0, yvec1, yvec2;
+ADD_SY	yvec2, yvec15, yvec15;
+ADDQ	$8*SIZE, ptrba;
+ADDQ	$1*SIZE, ptrbb;
+.L413_loopE:
+#### Writing ####
+BROAD_SY MEMALPHA, yvec7;
+MUL_SY	yvec7, yvec15, yvec15;
+EXTRA_SY $1, yvec15, xvec14;
+SHUF_SX	$0x44, xvec15, xvec13;
+SHUF_SX	$0xee, xvec15, xvec12;
+SHUF_SX	$0x44, xvec14, xvec11;
+SHUF_SX	$0xee, xvec14, xvec10;
+#ifndef	TRMMKERNEL
+LDL_SX	0*SIZE(C0), xvec0;
+LDL_SX	2*SIZE(C0), xvec1;
+LDL_SX	4*SIZE(C0), xvec2;
+LDL_SX	6*SIZE(C0), xvec3;
+ADD_SX	xvec0, xvec13;
+ADD_SX	xvec1, xvec12;
+ADD_SX	xvec2, xvec11;
+ADD_SX	xvec3, xvec10;
+#endif
+STL_SX	xvec13, 0*SIZE(C0);
+STL_SX	xvec12, 2*SIZE(C0);
+STL_SX	xvec11, 4*SIZE(C0);
+STL_SX	xvec10, 6*SIZE(C0);
+#if (defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kkk, %rax;
+LEAQ (,%rax, SIZE), %rax;
+LEAQ (ptrba, %rax, 8), ptrba;
+ADDQ %rax, ptrbb;
+#endif
+#if defined(TRMMKERNEL)&&defined(LEFT)
+ADDQ	$8, kk;
+#endif
+ADDQ	$8*SIZE, C0;
+DECQ	i;
+JG	.L41_bodyB;
+.align 16
+.L41_loopE:
+TEST $4, bm;
+JLE .L42_loopE;
+.align 16
+.L42_bodyB:
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb, ptrbb;
+#else
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+LEAQ (,%rax, SIZE), %rax;
+LEAQ (ptrba, %rax, 4), ptrba;
+ADDQ	%rax, ptrbb;
+#endif
+XOR_SY	yvec15, yvec15, yvec15;
+#ifndef	TRMMKERNEL
+MOVQ bk, k;
+#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else 
+MOVQ kk, %rax;
+#ifdef	LEFT
+ADDQ $4, %rax;
+#else 
+ADDQ $1, %rax;
+#endif
+MOVQ %rax, kkk
+#endif
+SARQ $2, k;
+JLE .L421_loopE;
+.align 16
+.L421_bodyB:
+LD_SX	0*SIZE(ptrba), xvec0;
+BROAD_SX 0*SIZE(ptrbb), xvec1;
+MUL_SX	xvec0, xvec1;
+ADD_SX	xvec1, xvec15;
+
+LD_SX	4*SIZE(ptrba), xvec0;
+BROAD_SX 1*SIZE(ptrbb), xvec1;
+MUL_SX	xvec0, xvec1;
+ADD_SX	xvec1, xvec15;
+
+LD_SX	8*SIZE(ptrba), xvec0;
+BROAD_SX 2*SIZE(ptrbb), xvec1;
+MUL_SX	xvec0, xvec1;
+ADD_SX	xvec1, xvec15;
+
+LD_SX	12*SIZE(ptrba), xvec0;
+BROAD_SX 3*SIZE(ptrbb), xvec1;
+MUL_SX	xvec0, xvec1;
+ADD_SX	xvec1, xvec15;
+ADDQ	$16*SIZE, ptrba;
+ADDQ	$4*SIZE, ptrbb;
+DECQ 	k;
+JG .L421_bodyB;
+.align 16
+.L421_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else
+TEST $2, kkk;
+#endif
+JLE .L422_loopE;
+.align 16
+.L422_bodyB:
+LD_SX	0*SIZE(ptrba), xvec0;
+BROAD_SX 0*SIZE(ptrbb), xvec1;
+MUL_SX	xvec0, xvec1;
+ADD_SX	xvec1, xvec15;
+
+LD_SX	4*SIZE(ptrba), xvec0;
+BROAD_SX 1*SIZE(ptrbb), xvec1;
+MUL_SX	xvec0, xvec1;
+ADD_SX	xvec1, xvec15;
+
+ADDQ	$8*SIZE, ptrba;
+ADDQ	$2*SIZE, ptrbb;
+.L422_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else
+TEST $1, kkk;
+#endif
+JLE .L423_loopE;
+.align 16
+.L423_bodyB:
+LD_SX	0*SIZE(ptrba), xvec0;
+BROAD_SX 0*SIZE(ptrbb), xvec1;
+MUL_SX	xvec0, xvec1;
+ADD_SX	xvec1, xvec15;
+ADDQ	$4*SIZE, ptrba;
+ADDQ	$1*SIZE, ptrbb;
+
+.L423_loopE:
+#### Writing back ####
+BROAD_SX MEMALPHA, xvec7;
+MUL_SX	xvec7, xvec15;
+#ifndef	TRMMKERNEL
+LDL_SX	0*SIZE(C0), xvec0;
+LDH_SX	2*SIZE(C0), xvec0;
+ADD_SX	xvec0, xvec15;
+#endif
+STL_SX	xvec15, 0*SIZE(C0);
+STH_SX	xvec15, 2*SIZE(C0);
+#if (defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kkk, %rax;
+LEAQ (, %rax, SIZE), %rax;
+LEAQ (ptrba, %rax, 4), ptrba;
+ADDQ %rax, ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT)
+ADDQ $4, kk
+#endif
+ADDQ	$4*SIZE, C0;
+
+.L42_loopE:
+TEST $2, bm;
+JLE .L43_loopE;
+.align 16
+.L43_bodyB:
+#if !defined(TRMMKERNEL) || (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+MOVQ bb, ptrbb;
+#else 
+MOVQ bb, ptrbb;
+MOVQ kk, %rax
+LEAQ (, %rax, SIZE), %rax
+LEAQ (ptrba, %rax, 2), ptrba
+ADDQ %rax, ptrbb;
+#endif 
+XOR_SY	yvec15, yvec15, yvec15;
+XOR_SY	yvec14, yvec14, yvec14;
+#ifndef	TRMMKERNEL
+MOVQ bk, k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else 
+MOVQ kk, %rax;
+#ifdef	LEFT
+ADDQ $2, %rax;
+#else
+ADDQ $1, %rax;
+#endif
+MOVQ %rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L431_loopE;
+.align 16
+.L431_bodyB:
+movss	0*SIZE(ptrba), xvec0;
+movss	1*SIZE(ptrba), xvec1;
+movss	0*SIZE(ptrbb), xvec2;
+mulss	xvec2, xvec0;
+addss	xvec0, xvec15;
+mulss	xvec2, xvec1;
+addss	xvec1, xvec14;
+
+movss	2*SIZE(ptrba), xvec3;
+movss	3*SIZE(ptrba), xvec4;
+movss	1*SIZE(ptrbb), xvec5;
+mulss	xvec5, xvec3;
+addss	xvec3, xvec15;
+mulss	xvec5, xvec4;
+addss	xvec4, xvec14;
+
+movss	4*SIZE(ptrba), xvec0;
+movss	5*SIZE(ptrba), xvec1;
+movss	2*SIZE(ptrbb), xvec2;
+mulss	xvec2, xvec0;
+addss	xvec0, xvec15;
+mulss	xvec2, xvec1;
+addss	xvec1, xvec14;
+
+movss	6*SIZE(ptrba), xvec3;
+movss	7*SIZE(ptrba), xvec4;
+movss	3*SIZE(ptrbb), xvec5;
+mulss	xvec5, xvec3;
+addss	xvec3, xvec15;
+mulss	xvec5, xvec4;
+addss	xvec4, xvec14;
+addq	$8*SIZE, ptrba;
+addq	$4*SIZE, ptrbb;
+decq	k;
+JG .L431_bodyB;
+.align 16
+.L431_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else
+TEST $2, kkk;
+#endif
+JLE .L432_loopE;
+.align 16
+.L432_bodyB:
+movss	0*SIZE(ptrba), xvec0;
+movss	1*SIZE(ptrba), xvec1;
+movss	0*SIZE(ptrbb), xvec2;
+mulss	xvec2, xvec0;
+addss	xvec0, xvec15;
+mulss	xvec2, xvec1;
+addss	xvec1, xvec14;
+
+movss	2*SIZE(ptrba), xvec3;
+movss	3*SIZE(ptrba), xvec4;
+movss	1*SIZE(ptrbb), xvec5;
+mulss	xvec5, xvec3;
+addss	xvec3, xvec15;
+mulss	xvec5, xvec4;
+addss	xvec4, xvec14;
+addq	$4*SIZE, ptrba;
+addq	$2*SIZE, ptrbb;
+
+.L432_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else
+TEST $1, kkk;
+#endif
+JLE .L433_loopE;
+.align 16
+.L433_bodyB:
+movss	0*SIZE(ptrba), xvec0;
+movss	1*SIZE(ptrba), xvec1;
+movss	0*SIZE(ptrbb), xvec2;
+mulss	xvec2, xvec0;
+addss	xvec0, xvec15;
+mulss	xvec2, xvec1;
+addss	xvec1, xvec14;
+addq	$2*SIZE, ptrba;
+addq	$1*SIZE, ptrbb;
+
+.L433_loopE:
+#### Writing Back ####
+movss	MEMALPHA, xvec7;
+mulss	xvec7, xvec15;
+mulss	xvec7, xvec14;
+
+#ifndef	TRMMKERNEL
+addss	0*SIZE(C0), xvec15;
+addss	1*SIZE(C0), xvec14;
+#endif
+movss	xvec15, 0*SIZE(C0);
+movss	xvec14, 1*SIZE(C0);
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) ||(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kkk, %rax;
+LEAQ (,%rax, SIZE), %rax;
+LEAQ (ptrba, %rax, 2), ptrba;
+ADDQ %rax, ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT)
+addq $2, kk
+#endif
+addq	$2*SIZE, C0;
+
+.L43_loopE:
+TEST $1, bm;
+JLE .L44_loopE;
+.align 16
+.L44_bodyB:
+#if !defined(TRMMKERNEL) || (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA)) 
+MOVQ bb, ptrbb;
+#else 
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+LEAQ (, %rax, SIZE), %rax;
+ADDQ %rax, ptrba;
+ADDQ %rax, ptrbb;
+#endif
+XOR_SY	yvec15, yvec15, yvec15;
+#ifndef	TRMMKERNEL
+MOVQ bk, k;
+#elif (defined(LEFT)&& !defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else 
+MOVQ kk, %rax;
+#ifdef	LEFT
+ADDQ $1, %rax;
+#else
+ADDQ $1, %rax;
+#endif
+MOVQ %rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L441_loopE;
+.align 16
+.L441_bodyB:
+movss	0*SIZE(ptrba), xvec0;
+movss	0*SIZE(ptrbb), xvec1;
+mulss	xvec0, xvec1;
+addss	xvec1, xvec15;
+
+movss	1*SIZE(ptrba), xvec0;
+movss	1*SIZE(ptrbb), xvec1;
+mulss	xvec0, xvec1;
+addss	xvec1, xvec15;
+
+movss	2*SIZE(ptrba), xvec0;
+movss	2*SIZE(ptrbb), xvec1;
+mulss	xvec0, xvec1;
+addss	xvec1, xvec15;
+
+movss	3*SIZE(ptrba), xvec0;
+movss	3*SIZE(ptrbb), xvec1;
+mulss	xvec0, xvec1;
+addss	xvec1, xvec15;
+addq	$4*SIZE, ptrba;
+addq	$4*SIZE, ptrbb;
+decq	k;
+JG .L441_bodyB;
+.align 16
+.L441_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else
+TEST $2, kkk;
+#endif
+JLE .L442_loopE;
+.align 16
+.L442_bodyB:
+movss	0*SIZE(ptrba), xvec0;
+movss	0*SIZE(ptrbb), xvec1;
+mulss	xvec0, xvec1;
+addss	xvec1, xvec15;
+
+movss	1*SIZE(ptrba), xvec0;
+movss	1*SIZE(ptrbb), xvec1;
+mulss	xvec0, xvec1;
+addss	xvec1, xvec15;
+addq	$2*SIZE, ptrba;
+addq	$2*SIZE, ptrbb;
+
+.L442_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else
+TEST $1, kkk;
+#endif
+JLE .L443_loopE;
+.align 16
+.L443_bodyB:
+movss	0*SIZE(ptrba), xvec0;
+movss	0*SIZE(ptrbb), xvec1;
+mulss	xvec0, xvec1;
+addss	xvec1, xvec15;
+addq	$1*SIZE, ptrba;
+addq	$1*SIZE, ptrbb;
+
+.L443_loopE:
+#### Writing Back ####
+movss	MEMALPHA, xvec7;
+mulss	xvec7, xvec15;
+#ifndef	TRMMKERNEL
+addss	0*SIZE(C0), xvec15;
+#endif
+movss	xvec15, 0*SIZE(C0);
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) ||(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kkk, %rax;
+LEAQ (,%rax, SIZE), %rax;
+ADDQ %rax, ptrba;
+ADDQ %rax, ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT)
+addq $1, kk
+#endif
+addq	$1*SIZE, C0;
+
+.L44_loopE:
+MOV	bk, k;
+SALQ $2, k;
+ADDQ k, bb;
+ADDQ ldc, C;
+.L40_loopE:
+movq      0(%rsp), %rbx;
+movq      8(%rsp), %rbp;
+movq     16(%rsp), %r12;
+movq     24(%rsp), %r13;
+movq     32(%rsp), %r14;
+movq     40(%rsp), %r15;
+#ifdef WINDOWS_ABI
+	movq	 48(%rsp), %rdi
+	movq	 56(%rsp), %rsi
+	movups	 64(%rsp), %xmm6
+	movups	 80(%rsp), %xmm7
+	movups	 96(%rsp), %xmm8
+	movups	112(%rsp), %xmm9
+	movups	128(%rsp), %xmm10
+	movups	144(%rsp), %xmm11
+	movups	160(%rsp), %xmm12
+	movups	176(%rsp), %xmm13
+	movups	192(%rsp), %xmm14
+	movups	208(%rsp), %xmm15
+#endif
+addq    $STACKSIZE, %rsp;
+ret
+
+EPILOGUE
diff --git a/kernel/x86_64/zgemm_kernel_4x4_sandy.S b/kernel/x86_64/zgemm_kernel_4x4_sandy.S
new file mode 100644
index 000000000..34abbb529
--- /dev/null
+++ b/kernel/x86_64/zgemm_kernel_4x4_sandy.S
@@ -0,0 +1,3257 @@
+/*****************************************************************************
+  Copyright (c) 2011,2012 Lab of Parallel Software and Computational Science,ISCAS
+  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 ISCAS 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 COPYRIGHT OWNER 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	old_bm		%rdi
+#define old_bn		%rsi
+#define	old_bk		%rdx
+
+#define bm	%r13
+#define bn	%r14
+#define bk	%r15
+
+#define ALPHA	%xmm0
+#define ba		%rcx
+#define bb		%r8
+#define C		%r9
+#define ldc		%r10
+
+#define i	%r11
+#define k	%rax
+
+#define ptrba	%rdi
+#define ptrbb	%rsi
+#define C0		%rbx
+#define C1		%rbp
+
+#define prebb	%r12
+
+#ifndef WINDOWS_ABI
+
+#define STACKSIZE 128 
+
+#define old_ldc	8+STACKSIZE(%rsp)
+#define old_offset 16+STACKSIZE(%rsp)
+
+#define MEMALPHA_R	48(%rsp)
+#define MEMALPHA_I	56(%rsp)
+#define j	64(%rsp)
+#define OFFSET	72(%rsp)
+#define kk	80(%rsp)
+#define kkk	88(%rsp)
+
+#else
+#define STACKSIZE 512
+
+#define OLD_ALPHA_I	40 + STACKSIZE(%rsp)
+#define OLD_A		48 + STACKSIZE(%rsp)
+#define OLD_B		56 + STACKSIZE(%rsp)
+#define OLD_C		64 + STACKSIZE(%rsp)
+#define old_ldc		72 + STACKSIZE(%rsp)
+#define old_offset	80 + STACKSIZE(%rsp)
+
+#define MEMALPHA_R	  224(%rsp)
+#define MEMALPHA_I	  232(%rsp)
+#define j	  240(%rsp)
+#define OFFSET	  248(%rsp)
+#define kk	  256(%rsp)
+#define kkk	  264(%rsp)
+
+#endif
+
+#define PREFETCH0	prefetcht0
+#define PREFETCH1	prefetcht0
+#define PREFETCH2	prefetcht0
+#define PRESIZE	64
+
+#define xvec0	%xmm0
+#define xvec1	%xmm1
+#define xvec2	%xmm2
+#define xvec3	%xmm3
+#define xvec4	%xmm4
+#define xvec5	%xmm5
+#define xvec6	%xmm6
+#define xvec7	%xmm7
+#define xvec8	%xmm8
+#define xvec9	%xmm9
+#define xvec10	%xmm10
+#define xvec11	%xmm11
+#define xvec12	%xmm12
+#define xvec13	%xmm13
+#define xvec14	%xmm14
+#define xvec15	%xmm15
+
+#define yvec0	%ymm0
+#define yvec1	%ymm1
+#define yvec2	%ymm2
+#define yvec3	%ymm3
+#define yvec4	%ymm4
+#define yvec5	%ymm5
+#define yvec6	%ymm6
+#define yvec7	%ymm7
+#define yvec8	%ymm8
+#define yvec9	%ymm9
+#define yvec10	%ymm10
+#define yvec11	%ymm11
+#define yvec12	%ymm12
+#define yvec13	%ymm13
+#define yvec14	%ymm14
+#define yvec15	%ymm15
+
+#define LEAQ	leaq
+#define ADDQ	addq
+#define MULQ	imulq
+#define SARQ	sarq
+#define SALQ	salq
+#define ANDQ	andq
+#define SUBQ	subq
+#define DECQ	decq
+#define JG		jg
+#define JLE		jle
+#define TEST	testq
+#define OR		orq
+#define JNE		jne
+#define JMP		jmp
+#define NOP		
+#define XOR		xorpd
+
+#define XOR_SY	vxorps
+#define XOR_DY	vxorpd
+#define XOR_SX	xorps
+#define XOR_DX	xorpd
+
+#define LD_SY	vmovaps
+#define LD_DY	vmovapd
+#define LD_SX	movaps
+#define LD_DX	movapd
+#define LDL_DY	vmovlpd
+#define LDL_DX	movlpd
+#define LDH_DY	vmovhpd
+#define LDH_DX	movhpd
+
+#define ST_SY	vmovaps
+#define ST_DY	vmovapd
+#define ST_SX	movaps
+#define ST_DX	movapd
+#define STL_DY	vmovlpd
+#define STL_DX	movlpd
+#define STH_DY	vmovhpd
+#define STH_DX	movhpd
+
+#define EDUP_SY	vmovsldup
+#define ODUP_SY	vmovshdup
+#define EDUP_SX	movsldup
+#define ODUP_SX	movshdup
+#define EDUP_DY	vmovddup
+
+#define ADD_SY	vaddps 
+#define ADD_DY	vaddpd
+#define ADD_SX	addps
+#define ADD_DX	addpd
+#define SUB_DY	vsubpd
+#define SUB_DX	subpd
+
+#define ADDSUB_DY	vaddsubpd
+#define ADDSUB_DX	addsubpd
+#define ADDSUB_SY	vaddsubps
+
+#define MUL_SY	vmulps
+#define MUL_DY	vmulpd
+#define MUL_SX	mulps
+#define MUL_DX	mulpd
+
+#define SHUF_SY	vperm2f128
+#define SHUF_DY	vperm2f128
+#define SHUF_DX	pshufd
+#define SHUF_SX	pshufd
+
+#define VPERMILP_SY	vpermilps
+#define VPERMILP_SX	vpermilps
+#define VPERMILP_DY	vpermilpd
+
+#define BROAD_SY vbroadcastss
+#define BROAD_DY vbroadcastsd
+#define BROAD_SX vbroadcastss  
+#define BROAD_DX movddup
+
+#define MOV_SY vmovaps
+#define MOV_DY vmovapd
+#define MOV_SX movaps
+#define MOV_DX movapd
+
+#define REVS_SY	vshufps
+#define REVS_DY vshufpd
+#define REVS_SX shufps
+#define REVS_DX	movsd
+
+#define EXTRA_DY vextractf128
+
+
+#if   defined(NN) || defined(NT) || defined(TN) || defined(TT)
+#define ADD1_DX ADD_DX
+#define ADD1_DY	ADD_DY
+#define ADD2_DY	ADDSUB_DY
+#elif defined(NR) || defined(NC) || defined(TR) || defined(TC)
+#define ADD1_DX	SUB_DX
+#define ADD1_DY SUB_DY
+#define ADD2_DY	ADDSUB_DY
+#elif  defined(RN) || defined(RT) || defined(CN) || defined(CT)
+#define ADD1_DX	SUB_DX
+#define ADD1_DY SUB_DY
+#define ADD2_DY ADDSUB_DY
+#else 
+#define ADD1_DX	ADD_DX
+#define ADD1_DY ADD_DY
+#define ADD2_DY ADDSUB_DY
+#endif
+
+PROLOGUE
+
+subq    $STACKSIZE, %rsp;
+movq    %rbx,  0(%rsp);
+movq    %rbp,  8(%rsp);
+movq    %r12, 16(%rsp);
+movq    %r13, 24(%rsp);
+movq    %r14, 32(%rsp);
+movq    %r15, 40(%rsp);
+
+#ifdef WINDOWS_ABI
+	movq	%rdi,    48(%rsp)
+	movq	%rsi,    56(%rsp)
+	movups	%xmm6,   64(%rsp)
+	movups	%xmm7,   80(%rsp)
+	movups	%xmm8,   96(%rsp)
+	movups	%xmm9,  112(%rsp)
+	movups	%xmm10, 128(%rsp)
+	movups	%xmm11, 144(%rsp)
+	movups	%xmm12, 160(%rsp)
+	movups	%xmm13, 176(%rsp)
+	movups	%xmm14, 192(%rsp)
+	movups	%xmm15, 208(%rsp)
+
+	movq	ARG1,      old_bm
+	movq	ARG2,      old_bn
+	movq	ARG3,      old_bk
+	movq	OLD_A,     ba
+	movq	OLD_B,     bb
+	movq	OLD_C,     C
+	movq	old_ldc,   ldc
+#ifdef TRMMKERNEL
+	movq	old_offset, %r11
+#endif
+	movaps	%xmm3, %xmm0
+	movsd	OLD_ALPHA_I, %xmm1
+#else
+
+movq	old_ldc, ldc
+#ifdef	TRMMKERNEL
+movq	old_offset, %r11;
+#endif
+#endif
+
+vmovlps	%xmm0, MEMALPHA_R
+vmovlps	%xmm1, MEMALPHA_I
+movq	old_bm,	bm
+movq	old_bn, bn
+movq	old_bk, bk
+salq	$ZBASE_SHIFT, ldc
+#ifdef	TRMMKERNEL
+movq	%r11, OFFSET
+#ifndef	LEFT
+negq	%r11;
+#endif
+movq	%r11, kk;
+#endif
+
+MOVQ bn,j;
+SARQ $2,j;						# Rn = 4
+JLE .L0_loopE;
+.align 32;
+.L0_bodyB:;
+#if defined(TRMMKERNEL) && defined(LEFT)
+MOVQ OFFSET, %rax;
+MOVQ %rax, kk;
+#endif
+MOVQ C,C0;
+LEAQ (C,ldc,2),C1;
+MOVQ bk, k;
+SALQ $6, k;
+LEAQ (bb, k, 1), prebb;			# Rn=4 SIZE=8 COMPLEX=2
+MOVQ ba,ptrba;
+MOVQ bm,i;
+SARQ $2,i;						# Rm = 4 
+JLE .L1_loopE;
+.align 32;
+.L1_bodyB:;
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb,ptrbb;
+#else
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+LEAQ (ptrba, %rax, 4), ptrba;
+LEAQ (ptrbb, %rax, 4), ptrbb;
+#endif
+
+PREFETCH0	0*SIZE(prebb);
+PREFETCH0	8*SIZE(prebb);
+PREFETCH0	16*SIZE(prebb)
+ADDQ	$24*SIZE, prebb;
+# Initial Results Register
+XOR_DY	yvec15, yvec15, yvec15;
+XOR_DY	yvec14, yvec14, yvec14;
+EDUP_DY	0*SIZE(ptrbb), yvec2;		# Br1, Br1, Br2, Br2
+XOR_DY	yvec13, yvec13, yvec13;
+XOR_DY	yvec12, yvec12, yvec12;
+EDUP_DY	4*SIZE(ptrbb), yvec3;		# Br3, Br3, Br4, Br4
+PREFETCH2	3*SIZE(C0);
+PREFETCH2	3*SIZE(C1);
+XOR_DY	yvec11, yvec11, yvec11;
+XOR_DY	yvec10, yvec10, yvec10;
+LD_DY	0*SIZE(ptrba), yvec0;		# Ar1, Ai1, Ar2, Ai2
+PREFETCH2	7*SIZE(C0, ldc, 1);
+PREFETCH2	7*SIZE(C1, ldc, 1);
+XOR_DY	yvec9, yvec9, yvec9;
+XOR_DY	yvec8, yvec8, yvec8;
+#ifndef	TRMMKERNEL
+MOVQ bk,k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else
+MOVQ kk, %rax;
+#ifdef	LEFT
+ADDQ $4, %rax;
+#else 
+ADDQ $4, %rax;
+#endif
+MOVQ %rax, kkk;
+#endif
+SARQ $2,k;						# Unroll 4 times
+JLE .L2_loopE;
+.align 32;
+.L2_bodyB:;
+#### Computing kernel ####
+
+#### Unroll time 1 ####
+LD_DY	4*SIZE(ptrba), yvec1;
+MUL_DY	yvec0, yvec2, yvec6;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;	# Br2, Br2, Br1, Br1
+MUL_DY	yvec0, yvec3, yvec7;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;	# Br4, Br4, Br3, Br3
+ADD1_DY	yvec6, yvec15, yvec15;
+ADD1_DY	yvec7, yvec11, yvec11;
+
+PREFETCH0	PRESIZE*SIZE(ptrba);
+MUL_DY	yvec1, yvec2, yvec6;
+EDUP_DY	1*SIZE(ptrbb), yvec2;		# Bi1, Bi1, Bi2, Bi2
+MUL_DY	yvec1, yvec3, yvec7;
+EDUP_DY	5*SIZE(ptrbb), yvec3;		# Bi3, Bi3, Bi4, Bi4
+ADD1_DY	yvec6, yvec14, yvec14;
+ADD1_DY	yvec7, yvec10, yvec10;
+
+MUL_DY	yvec0, yvec4, yvec6;
+MUL_DY	yvec0, yvec5, yvec7;
+VPERMILP_DY	$0x05, yvec0, yvec0;	# Ai1, Ar1, Ai2, Ar2
+ADD1_DY	yvec6, yvec13, yvec13;
+ADD1_DY	yvec7, yvec9, yvec9;
+
+MUL_DY	yvec1, yvec4, yvec6;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;	# Bi2, Bi2, Bi1, Bi1
+MUL_DY	yvec1, yvec5, yvec7;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;	# Bi4, Bi4, Bi3, Bi3
+ADD1_DY	yvec6, yvec12, yvec12;
+ADD1_DY	yvec7, yvec8, yvec8;
+
+VPERMILP_DY	$0x05, yvec1, yvec1;	# Ai3, Ar3, Ai4, Ar4
+MUL_DY	yvec0, yvec2, yvec6;
+MUL_DY	yvec0, yvec3, yvec7;
+ADD2_DY	yvec6, yvec15, yvec15;
+ADD2_DY	yvec7, yvec11, yvec11;
+
+MUL_DY	yvec1, yvec2, yvec6;
+EDUP_DY	8*SIZE(ptrbb), yvec2;
+MUL_DY	yvec1, yvec3, yvec7;
+EDUP_DY	12*SIZE(ptrbb), yvec3;
+ADD2_DY	yvec6, yvec14, yvec14;
+ADD2_DY	yvec7, yvec10, yvec10;
+
+MUL_DY	yvec0, yvec4, yvec6;
+MUL_DY	yvec0, yvec5, yvec7;
+LD_DY	8*SIZE(ptrba), yvec0;
+ADD2_DY	yvec6, yvec13, yvec13;
+ADD2_DY	yvec7, yvec9, yvec9;
+
+MUL_DY	yvec1, yvec4, yvec6;
+MUL_DY	yvec1, yvec5, yvec7;
+ADD2_DY	yvec6, yvec12, yvec12;
+ADD2_DY	yvec7, yvec8, yvec8;
+
+#### Unroll time 2 ####
+LD_DY	12*SIZE(ptrba), yvec1;
+MUL_DY	yvec0, yvec2, yvec6;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+MUL_DY	yvec0, yvec3, yvec7;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;	# Br4, Br4, Br3, Br3
+ADD1_DY	yvec6, yvec15, yvec15;
+ADD1_DY	yvec7, yvec11, yvec11;
+
+PREFETCH0	(PRESIZE+8)*SIZE(ptrba);
+MUL_DY	yvec1, yvec2, yvec6;
+EDUP_DY	9*SIZE(ptrbb), yvec2;		# Bi1, Bi1, Bi2, Bi2
+MUL_DY	yvec1, yvec3, yvec7;
+EDUP_DY	13*SIZE(ptrbb), yvec3;		# Bi3, Bi3, Bi4, Bi4
+ADD1_DY	yvec6, yvec14, yvec14;
+ADD1_DY	yvec7, yvec10, yvec10;
+
+MUL_DY	yvec0, yvec4, yvec6;
+MUL_DY	yvec0, yvec5, yvec7;
+VPERMILP_DY	$0x05, yvec0, yvec0;	# Ai1, Ar1, Ai2, Ar2
+ADD1_DY	yvec6, yvec13, yvec13;
+ADD1_DY	yvec7, yvec9, yvec9;
+
+MUL_DY	yvec1, yvec4, yvec6;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;	# Bi2, Bi2, Bi1, Bi1
+MUL_DY	yvec1, yvec5, yvec7;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;	# Bi4, Bi4, Bi3, Bi3
+ADD1_DY	yvec6, yvec12, yvec12;
+ADD1_DY	yvec7, yvec8, yvec8;
+
+VPERMILP_DY	$0x05, yvec1, yvec1;	# Ai3, Ar3, Ai4, Ar4
+MUL_DY	yvec0, yvec2, yvec6;
+MUL_DY	yvec0, yvec3, yvec7;
+ADD2_DY	yvec6, yvec15, yvec15;
+ADD2_DY	yvec7, yvec11, yvec11;
+
+MUL_DY	yvec1, yvec2, yvec6;
+EDUP_DY	16*SIZE(ptrbb), yvec2;
+MUL_DY	yvec1, yvec3, yvec7;
+EDUP_DY	20*SIZE(ptrbb), yvec3;
+ADD2_DY	yvec6, yvec14, yvec14;
+ADD2_DY	yvec7, yvec10, yvec10;
+
+MUL_DY	yvec0, yvec4, yvec6;
+MUL_DY	yvec0, yvec5, yvec7;
+LD_DY	16*SIZE(ptrba), yvec0;
+ADD2_DY	yvec6, yvec13, yvec13;
+ADD2_DY	yvec7, yvec9, yvec9;
+
+MUL_DY	yvec1, yvec4, yvec6;
+MUL_DY	yvec1, yvec5, yvec7;
+ADD2_DY	yvec6, yvec12, yvec12;
+ADD2_DY	yvec7, yvec8, yvec8;
+
+#### Unroll time 3 ####
+LD_DY	20*SIZE(ptrba), yvec1;
+MUL_DY	yvec0, yvec2, yvec6;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+MUL_DY	yvec0, yvec3, yvec7;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;	# Br4, Br4, Br3, Br3
+ADD1_DY	yvec6, yvec15, yvec15;
+ADD1_DY	yvec7, yvec11, yvec11;
+
+PREFETCH0	(PRESIZE+16)*SIZE(ptrba);
+MUL_DY	yvec1, yvec2, yvec6;
+EDUP_DY	17*SIZE(ptrbb), yvec2;		# Bi1, Bi1, Bi2, Bi2
+MUL_DY	yvec1, yvec3, yvec7;
+EDUP_DY	21*SIZE(ptrbb), yvec3;		# Bi3, Bi3, Bi4, Bi4
+ADD1_DY	yvec6, yvec14, yvec14;
+ADD1_DY	yvec7, yvec10, yvec10;
+
+MUL_DY	yvec0, yvec4, yvec6;
+MUL_DY	yvec0, yvec5, yvec7;
+VPERMILP_DY	$0x05, yvec0, yvec0;	# Ai1, Ar1, Ai2, Ar2
+ADD1_DY	yvec6, yvec13, yvec13;
+ADD1_DY	yvec7, yvec9, yvec9;
+
+MUL_DY	yvec1, yvec4, yvec6;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;	# Bi2, Bi2, Bi1, Bi1
+MUL_DY	yvec1, yvec5, yvec7;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;	# Bi4, Bi4, Bi3, Bi3
+ADD1_DY	yvec6, yvec12, yvec12;
+ADD1_DY	yvec7, yvec8, yvec8;
+
+VPERMILP_DY	$0x05, yvec1, yvec1;	# Ai3, Ar3, Ai4, Ar4
+MUL_DY	yvec0, yvec2, yvec6;
+MUL_DY	yvec0, yvec3, yvec7;
+ADD2_DY	yvec6, yvec15, yvec15;
+ADD2_DY	yvec7, yvec11, yvec11;
+
+MUL_DY	yvec1, yvec2, yvec6;
+EDUP_DY	24*SIZE(ptrbb), yvec2;
+MUL_DY	yvec1, yvec3, yvec7;
+EDUP_DY	28*SIZE(ptrbb), yvec3;
+ADD2_DY	yvec6, yvec14, yvec14;
+ADD2_DY	yvec7, yvec10, yvec10;
+
+MUL_DY	yvec0, yvec4, yvec6;
+MUL_DY	yvec0, yvec5, yvec7;
+LD_DY	24*SIZE(ptrba), yvec0;
+ADD2_DY	yvec6, yvec13, yvec13;
+ADD2_DY	yvec7, yvec9, yvec9;
+
+MUL_DY	yvec1, yvec4, yvec6;
+MUL_DY	yvec1, yvec5, yvec7;
+ADD2_DY	yvec6, yvec12, yvec12;
+ADD2_DY	yvec7, yvec8, yvec8;
+
+#### Unroll time 4 ####
+LD_DY	28*SIZE(ptrba), yvec1;
+MUL_DY	yvec0, yvec2, yvec6;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+MUL_DY	yvec0, yvec3, yvec7;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;	# Br4, Br4, Br3, Br3
+ADDQ	$32*SIZE, ptrba;
+ADD1_DY	yvec6, yvec15, yvec15;
+ADD1_DY	yvec7, yvec11, yvec11;
+
+PREFETCH0	(PRESIZE+24)*SIZE(ptrba);
+MUL_DY	yvec1, yvec2, yvec6;
+EDUP_DY	25*SIZE(ptrbb), yvec2;		# Bi1, Bi1, Bi2, Bi2
+MUL_DY	yvec1, yvec3, yvec7;
+EDUP_DY	29*SIZE(ptrbb), yvec3;		# Bi3, Bi3, Bi4, Bi4
+ADD1_DY	yvec6, yvec14, yvec14;
+ADD1_DY	yvec7, yvec10, yvec10;
+
+MUL_DY	yvec0, yvec4, yvec6;
+MUL_DY	yvec0, yvec5, yvec7;
+VPERMILP_DY	$0x05, yvec0, yvec0;	# Ai1, Ar1, Ai2, Ar2
+ADDQ	$32*SIZE, ptrbb;
+ADD1_DY	yvec6, yvec13, yvec13;
+ADD1_DY	yvec7, yvec9, yvec9;
+
+MUL_DY	yvec1, yvec4, yvec6;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;	# Bi2, Bi2, Bi1, Bi1
+MUL_DY	yvec1, yvec5, yvec7;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;	# Bi4, Bi4, Bi3, Bi3
+ADD1_DY	yvec6, yvec12, yvec12;
+ADD1_DY	yvec7, yvec8, yvec8;
+
+VPERMILP_DY	$0x05, yvec1, yvec1;	# Ai3, Ar3, Ai4, Ar4
+MUL_DY	yvec0, yvec2, yvec6;
+MUL_DY	yvec0, yvec3, yvec7;
+ADD2_DY	yvec6, yvec15, yvec15;
+ADD2_DY	yvec7, yvec11, yvec11;
+
+MUL_DY	yvec1, yvec2, yvec6;
+EDUP_DY	0*SIZE(ptrbb), yvec2;
+MUL_DY	yvec1, yvec3, yvec7;
+EDUP_DY	4*SIZE(ptrbb), yvec3;
+ADD2_DY	yvec6, yvec14, yvec14;
+ADD2_DY	yvec7, yvec10, yvec10;
+
+MUL_DY	yvec0, yvec4, yvec6;
+MUL_DY	yvec0, yvec5, yvec7;
+LD_DY	0*SIZE(ptrba), yvec0;
+ADD2_DY	yvec6, yvec13, yvec13;
+ADD2_DY	yvec7, yvec9, yvec9;
+
+MUL_DY	yvec1, yvec4, yvec6;
+MUL_DY	yvec1, yvec5, yvec7;
+ADD2_DY	yvec6, yvec12, yvec12;
+ADD2_DY	yvec7, yvec8, yvec8;
+DECQ k;
+JG .L2_bodyB;
+.align	64;
+.L2_loopE:;
+#ifndef	TRMMKERNEL
+TEST	$2, bk;
+#else
+TEST 	$2, kkk;
+#endif
+JLE		.L3_loopE;
+.align	64
+.L3_bodyB:
+#### Unroll time 1 ####
+LD_DY	4*SIZE(ptrba), yvec1;
+MUL_DY	yvec0, yvec2, yvec6;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;	# Br2, Br2, Br1, Br1
+MUL_DY	yvec0, yvec3, yvec7;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;	# Br4, Br4, Br3, Br3
+ADD1_DY	yvec6, yvec15, yvec15;
+ADD1_DY	yvec7, yvec11, yvec11;
+
+PREFETCH0	PRESIZE*SIZE(ptrba);
+MUL_DY	yvec1, yvec2, yvec6;
+EDUP_DY	1*SIZE(ptrbb), yvec2;		# Bi1, Bi1, Bi2, Bi2
+MUL_DY	yvec1, yvec3, yvec7;
+EDUP_DY	5*SIZE(ptrbb), yvec3;		# Bi3, Bi3, Bi4, Bi4
+ADD1_DY	yvec6, yvec14, yvec14;
+ADD1_DY	yvec7, yvec10, yvec10;
+
+MUL_DY	yvec0, yvec4, yvec6;
+MUL_DY	yvec0, yvec5, yvec7;
+VPERMILP_DY	$0x05, yvec0, yvec0;	# Ai1, Ar1, Ai2, Ar2
+ADD1_DY	yvec6, yvec13, yvec13;
+ADD1_DY	yvec7, yvec9, yvec9;
+
+MUL_DY	yvec1, yvec4, yvec6;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;	# Bi2, Bi2, Bi1, Bi1
+MUL_DY	yvec1, yvec5, yvec7;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;	# Bi4, Bi4, Bi3, Bi3
+ADD1_DY	yvec6, yvec12, yvec12;
+ADD1_DY	yvec7, yvec8, yvec8;
+
+VPERMILP_DY	$0x05, yvec1, yvec1;	# Ai3, Ar3, Ai4, Ar4
+MUL_DY	yvec0, yvec2, yvec6;
+MUL_DY	yvec0, yvec3, yvec7;
+ADD2_DY	yvec6, yvec15, yvec15;
+ADD2_DY	yvec7, yvec11, yvec11;
+
+MUL_DY	yvec1, yvec2, yvec6;
+EDUP_DY	8*SIZE(ptrbb), yvec2;
+MUL_DY	yvec1, yvec3, yvec7;
+EDUP_DY	12*SIZE(ptrbb), yvec3;
+ADD2_DY	yvec6, yvec14, yvec14;
+ADD2_DY	yvec7, yvec10, yvec10;
+
+MUL_DY	yvec0, yvec4, yvec6;
+MUL_DY	yvec0, yvec5, yvec7;
+LD_DY	8*SIZE(ptrba), yvec0;
+ADD2_DY	yvec6, yvec13, yvec13;
+ADD2_DY	yvec7, yvec9, yvec9;
+
+MUL_DY	yvec1, yvec4, yvec6;
+MUL_DY	yvec1, yvec5, yvec7;
+ADD2_DY	yvec6, yvec12, yvec12;
+ADD2_DY	yvec7, yvec8, yvec8;
+
+#### Unroll time 2 ####
+LD_DY	12*SIZE(ptrba), yvec1;
+MUL_DY	yvec0, yvec2, yvec6;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+MUL_DY	yvec0, yvec3, yvec7;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;	# Br4, Br4, Br3, Br3
+ADDQ	$16*SIZE, ptrba
+ADD1_DY	yvec6, yvec15, yvec15;
+ADD1_DY	yvec7, yvec11, yvec11;
+
+PREFETCH0	(PRESIZE+8)*SIZE(ptrba);
+MUL_DY	yvec1, yvec2, yvec6;
+EDUP_DY	9*SIZE(ptrbb), yvec2;		# Bi1, Bi1, Bi2, Bi2
+MUL_DY	yvec1, yvec3, yvec7;
+EDUP_DY	13*SIZE(ptrbb), yvec3;		# Bi3, Bi3, Bi4, Bi4
+ADD1_DY	yvec6, yvec14, yvec14;
+ADD1_DY	yvec7, yvec10, yvec10;
+
+MUL_DY	yvec0, yvec4, yvec6;
+MUL_DY	yvec0, yvec5, yvec7;
+VPERMILP_DY	$0x05, yvec0, yvec0;	# Ai1, Ar1, Ai2, Ar2
+ADDQ	$16*SIZE, ptrbb
+ADD1_DY	yvec6, yvec13, yvec13;
+ADD1_DY	yvec7, yvec9, yvec9;
+
+MUL_DY	yvec1, yvec4, yvec6;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;	# Bi2, Bi2, Bi1, Bi1
+MUL_DY	yvec1, yvec5, yvec7;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;	# Bi4, Bi4, Bi3, Bi3
+ADD1_DY	yvec6, yvec12, yvec12;
+ADD1_DY	yvec7, yvec8, yvec8;
+
+VPERMILP_DY	$0x05, yvec1, yvec1;	# Ai3, Ar3, Ai4, Ar4
+MUL_DY	yvec0, yvec2, yvec6;
+MUL_DY	yvec0, yvec3, yvec7;
+ADD2_DY	yvec6, yvec15, yvec15;
+ADD2_DY	yvec7, yvec11, yvec11;
+
+MUL_DY	yvec1, yvec2, yvec6;
+EDUP_DY	0*SIZE(ptrbb), yvec2;
+MUL_DY	yvec1, yvec3, yvec7;
+EDUP_DY	4*SIZE(ptrbb), yvec3;
+ADD2_DY	yvec6, yvec14, yvec14;
+ADD2_DY	yvec7, yvec10, yvec10;
+
+MUL_DY	yvec0, yvec4, yvec6;
+MUL_DY	yvec0, yvec5, yvec7;
+LD_DY	0*SIZE(ptrba), yvec0;
+ADD2_DY	yvec6, yvec13, yvec13;
+ADD2_DY	yvec7, yvec9, yvec9;
+
+MUL_DY	yvec1, yvec4, yvec6;
+MUL_DY	yvec1, yvec5, yvec7;
+ADD2_DY	yvec6, yvec12, yvec12;
+ADD2_DY	yvec7, yvec8, yvec8;
+.L3_loopE:;
+#ifndef	TRMMKERNEL
+TEST	$1, bk;
+#else
+TEST 	$1, kkk;
+#endif
+JLE		.L4_loopE;
+.align	64
+.L4_loopB:;
+#### Unroll time 1 ####
+PREFETCH0	PRESIZE*SIZE(ptrba);
+LD_DY	4*SIZE(ptrba), yvec1;
+MUL_DY	yvec0, yvec2, yvec6;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+MUL_DY	yvec0, yvec3, yvec7;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;	# Br4, Br4, Br3, Br3
+ADDQ	$8*SIZE, ptrba;
+ADD1_DY	yvec6, yvec15, yvec15;
+ADD1_DY	yvec7, yvec11, yvec11;
+
+MUL_DY	yvec1, yvec2, yvec6;
+EDUP_DY	1*SIZE(ptrbb), yvec2;		# Bi1, Bi1, Bi2, Bi2
+MUL_DY	yvec1, yvec3, yvec7;
+EDUP_DY	5*SIZE(ptrbb), yvec3;		# Bi3, Bi3, Bi4, Bi4
+ADD1_DY	yvec6, yvec14, yvec14;
+ADD1_DY	yvec7, yvec10, yvec10;
+
+MUL_DY	yvec0, yvec4, yvec6;
+MUL_DY	yvec0, yvec5, yvec7;
+VPERMILP_DY	$0x05, yvec0, yvec0;	# Ai1, Ar1, Ai2, Ar2
+ADDQ	$8*SIZE, ptrbb;
+ADD1_DY	yvec6, yvec13, yvec13;
+ADD1_DY	yvec7, yvec9, yvec9;
+
+MUL_DY	yvec1, yvec4, yvec6;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;	# Bi2, Bi2, Bi1, Bi1
+MUL_DY	yvec1, yvec5, yvec7;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;	# Bi4, Bi4, Bi3, Bi3
+ADD1_DY	yvec6, yvec12, yvec12;
+ADD1_DY	yvec7, yvec8, yvec8;
+
+VPERMILP_DY	$0x05, yvec1, yvec1;	# Ai3, Ar3, Ai4, Ar4
+MUL_DY	yvec0, yvec2, yvec6;
+MUL_DY	yvec0, yvec3, yvec7;
+ADD2_DY	yvec6, yvec15, yvec15;
+ADD2_DY	yvec7, yvec11, yvec11;
+
+MUL_DY	yvec1, yvec2, yvec6;
+MUL_DY	yvec1, yvec3, yvec7;
+ADD2_DY	yvec6, yvec14, yvec14;
+ADD2_DY	yvec7, yvec10, yvec10;
+
+MUL_DY	yvec0, yvec4, yvec6;
+MUL_DY	yvec0, yvec5, yvec7;
+ADD2_DY	yvec6, yvec13, yvec13;
+ADD2_DY	yvec7, yvec9, yvec9;
+
+MUL_DY	yvec1, yvec4, yvec6;
+MUL_DY	yvec1, yvec5, yvec7;
+ADD2_DY	yvec6, yvec12, yvec12;
+ADD2_DY	yvec7, yvec8, yvec8;
+.L4_loopE:;
+#### Handle ####
+XOR_DY	yvec7, yvec7, yvec7;
+#if  defined(RN) || defined(RT) || defined(CN) || defined(CT)
+ADDSUB_DY	yvec15, yvec7, yvec15;
+ADDSUB_DY	yvec14, yvec7, yvec14;
+ADDSUB_DY	yvec13, yvec7, yvec13;
+ADDSUB_DY	yvec12, yvec7, yvec12;
+ADDSUB_DY	yvec11, yvec7, yvec11;
+ADDSUB_DY	yvec10, yvec7, yvec10;
+ADDSUB_DY	yvec9, yvec7, yvec9;
+ADDSUB_DY	yvec8, yvec7, yvec8;
+#elif defined(NR) || defined(NC) || defined(TR) || defined(TC)
+SUB_DY	yvec15, yvec7, yvec15;
+SUB_DY	yvec14, yvec7, yvec14;
+SUB_DY	yvec13, yvec7, yvec13;
+SUB_DY	yvec12, yvec7, yvec12;
+SUB_DY	yvec11, yvec7, yvec11;
+SUB_DY	yvec10, yvec7, yvec10;
+SUB_DY	yvec9, yvec7, yvec9;
+SUB_DY	yvec8, yvec7, yvec8;
+#elif defined(RR) || defined(RC) || defined(CR) || defined(CC) 
+VPERMILP_DY $0x05, yvec15, yvec15;
+VPERMILP_DY $0x05, yvec14, yvec14;
+VPERMILP_DY $0x05, yvec13, yvec13;
+VPERMILP_DY $0x05, yvec12, yvec12;
+VPERMILP_DY $0x05, yvec11, yvec11;
+VPERMILP_DY $0x05, yvec10, yvec10;
+VPERMILP_DY $0x05, yvec9, yvec9;
+VPERMILP_DY $0x05, yvec8, yvec8;
+ADDSUB_DY	yvec15, yvec7, yvec15;
+ADDSUB_DY	yvec14, yvec7, yvec14;
+ADDSUB_DY	yvec13, yvec7, yvec13;
+ADDSUB_DY	yvec12, yvec7, yvec12;
+ADDSUB_DY	yvec11, yvec7, yvec11;
+ADDSUB_DY	yvec10, yvec7, yvec10;
+ADDSUB_DY	yvec9, yvec7, yvec9;
+ADDSUB_DY	yvec8, yvec7, yvec8;
+VPERMILP_DY $0x05, yvec15, yvec15;
+VPERMILP_DY $0x05, yvec14, yvec14;
+VPERMILP_DY $0x05, yvec13, yvec13;
+VPERMILP_DY $0x05, yvec12, yvec12;
+VPERMILP_DY $0x05, yvec11, yvec11;
+VPERMILP_DY $0x05, yvec10, yvec10;
+VPERMILP_DY $0x05, yvec9, yvec9;
+VPERMILP_DY $0x05, yvec8, yvec8;
+#endif
+#### Load Alpha ####
+BROAD_DY MEMALPHA_R,yvec7;
+BROAD_DY MEMALPHA_I,yvec6;
+#### Multiply Alpha	####
+VPERMILP_DY	$0x05, yvec15, yvec5;
+MUL_DY	yvec7, yvec15, yvec15;
+MUL_DY	yvec6, yvec5, yvec5;
+ADDSUB_DY	yvec5, yvec15, yvec15;
+VPERMILP_DY	$0x05, yvec14, yvec4;
+MUL_DY 	yvec7, yvec14, yvec14;
+MUL_DY 	yvec6, yvec4, yvec4;
+ADDSUB_DY 	yvec4, yvec14, yvec14;
+VPERMILP_DY	$0x05, yvec13, yvec3;
+MUL_DY	yvec7, yvec13, yvec13;
+MUL_DY 	yvec6, yvec3, yvec3;
+ADDSUB_DY	yvec3, yvec13, yvec13;
+VPERMILP_DY	$0x05,yvec12, yvec2;
+MUL_DY 	yvec7, yvec12, yvec12;
+MUL_DY 	yvec6, yvec2, yvec2;
+ADDSUB_DY 	yvec2, yvec12, yvec12;
+VPERMILP_DY $0x05, yvec11, yvec1;
+MUL_DY 	yvec7, yvec11, yvec11;
+MUL_DY 	yvec6, yvec1, yvec1;
+ADDSUB_DY 	yvec1, yvec11, yvec11;
+VPERMILP_DY $0x05,yvec10, yvec0;
+MUL_DY 	yvec7, yvec10, yvec10;
+MUL_DY 	yvec6, yvec0, yvec0;
+ADDSUB_DY 	yvec0, yvec10, yvec10;
+VPERMILP_DY $0x05, yvec9, yvec5;
+MUL_DY 	yvec7, yvec9, yvec9;
+MUL_DY 	yvec6, yvec5, yvec5;
+ADDSUB_DY 	yvec5, yvec9, yvec9;
+VPERMILP_DY $0x05, yvec8, yvec4;
+MUL_DY 	yvec7, yvec8, yvec8;
+MUL_DY 	yvec6, yvec4, yvec4;
+ADDSUB_DY	yvec4, yvec8, yvec8;
+#### Testing Alignment ####
+MOVQ	C0, %rax;
+OR		ldc, %rax;
+TEST 	$15, %rax;
+JNE		.L4_loopEx;
+.align 32
+#### Store Back ####
+EXTRA_DY $1,yvec15,xvec7;
+EXTRA_DY $1,yvec14,xvec6;
+EXTRA_DY $1,yvec13,xvec5;
+EXTRA_DY $1,yvec12,xvec4;
+EXTRA_DY $1,yvec11,xvec3;
+EXTRA_DY $1,yvec10,xvec2;
+EXTRA_DY $1,yvec9,xvec1;
+EXTRA_DY $1,yvec8,xvec0;
+#ifndef	TRMMKERNEL
+ADD_DY 0*SIZE(C0),xvec15, xvec15;
+ADD_DY 2*SIZE(C0,ldc,1), xvec7, xvec7;
+ADD_DY 4*SIZE(C0),xvec14, xvec14;
+ADD_DY 6*SIZE(C0,ldc,1),xvec6, xvec6;
+ADD_DY 0*SIZE(C0,ldc,1),xvec13, xvec13;
+ADD_DY 2*SIZE(C0),xvec5, xvec5;
+ADD_DY 4*SIZE(C0,ldc,1),xvec12, xvec12;
+ADD_DY 6*SIZE(C0),xvec4, xvec4;
+ADD_DY 0*SIZE(C1),xvec11, xvec11;
+ADD_DY 2*SIZE(C1,ldc,1),xvec3, xvec3;
+ADD_DY 4*SIZE(C1),xvec10, xvec10;
+ADD_DY 6*SIZE(C1,ldc,1),xvec2, xvec2;
+ADD_DY 0*SIZE(C1,ldc,1),xvec9, xvec9;
+ADD_DY 2*SIZE(C1),xvec1, xvec1;
+ADD_DY 4*SIZE(C1,ldc,1),xvec8, xvec8;
+ADD_DY 6*SIZE(C1),xvec0, xvec0;
+#endif
+ST_DY xvec15,0*SIZE(C0);
+ST_DY xvec7,2*SIZE(C0,ldc,1);
+ST_DY xvec14,4*SIZE(C0);
+ST_DY xvec6,6*SIZE(C0,ldc,1);
+ST_DY xvec13,0*SIZE(C0,ldc,1);
+ST_DY xvec5,2*SIZE(C0);
+ST_DY xvec12,4*SIZE(C0,ldc,1);
+ST_DY xvec4,6*SIZE(C0);
+ST_DY xvec11,0*SIZE(C1);
+ST_DY xvec3,2*SIZE(C1,ldc,1);
+ST_DY xvec10,4*SIZE(C1);
+ST_DY xvec2,6*SIZE(C1,ldc,1);
+ST_DY xvec9,0*SIZE(C1,ldc,1);
+ST_DY xvec1,2*SIZE(C1);
+ST_DY xvec8,4*SIZE(C1,ldc,1);
+ST_DY xvec0,6*SIZE(C1);
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA))||(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kkk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+LEAQ (ptrba, %rax, 4), ptrba;
+LEAQ (ptrbb, %rax, 4), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT) 
+ADDQ	$4, kk;
+#endif
+ADDQ $8*SIZE,C0;
+ADDQ $8*SIZE,C1;
+.L1_bodyE:;
+DECQ i;
+JG .L1_bodyB;
+JMP	.L1_loopE;
+.align 32
+.L4_loopEx:
+EXTRA_DY $1, yvec15, xvec7;
+EXTRA_DY $1, yvec14, xvec6;
+#ifndef	TRMMKERNEL
+LDL_DY	0*SIZE(C0), xvec0, xvec0;
+LDH_DY	1*SIZE(C0), xvec0, xvec0;
+LDL_DY	2*SIZE(C0, ldc, 1), xvec1, xvec1;
+LDH_DY	3*SIZE(C0, ldc, 1), xvec1, xvec1;
+LDL_DY	4*SIZE(C0), xvec2, xvec2;
+LDH_DY	5*SIZE(C0), xvec2, xvec2;
+LDL_DY	6*SIZE(C0, ldc, 1), xvec3, xvec3;
+LDH_DY	7*SIZE(C0, ldc, 1), xvec3, xvec3;
+ADD_DY	xvec0, xvec15, xvec15;
+ADD_DY	xvec1, xvec7, xvec7;
+ADD_DY	xvec2, xvec14, xvec14;
+ADD_DY	xvec3, xvec6, xvec6;
+#endif
+STL_DY	xvec15, 0*SIZE(C0);
+STH_DY	xvec15, 1*SIZE(C0);
+STL_DY	xvec7, 2*SIZE(C0, ldc, 1);
+STH_DY	xvec7, 3*SIZE(C0, ldc, 1);
+STL_DY	xvec14, 4*SIZE(C0);
+STH_DY	xvec14, 5*SIZE(C0);
+STL_DY	xvec6, 6*SIZE(C0, ldc, 1);
+STH_DY	xvec6, 7*SIZE(C0, ldc, 1);
+EXTRA_DY $1, yvec13, xvec5;
+EXTRA_DY $1, yvec12, xvec4;
+#ifndef	TRMMKERNEL
+LDL_DY	0*SIZE(C0, ldc, 1), xvec3, xvec3;
+LDH_DY	1*SIZE(C0, ldc, 1), xvec3, xvec3;
+LDL_DY	2*SIZE(C0), xvec2, xvec2;
+LDH_DY	3*SIZE(C0), xvec2, xvec2;
+LDL_DY	4*SIZE(C0, ldc, 1), xvec1, xvec1;
+LDH_DY	5*SIZE(C0, ldc, 1), xvec1, xvec1;
+LDL_DY	6*SIZE(C0), xvec0, xvec0;
+LDH_DY	7*SIZE(C0), xvec0, xvec0;
+ADD_DY	xvec3, xvec13, xvec13;
+ADD_DY	xvec2, xvec5, xvec5;
+ADD_DY	xvec1, xvec12, xvec12;
+ADD_DY	xvec0, xvec4, xvec4;
+#endif
+STL_DY	xvec13, 0*SIZE(C0, ldc, 1);
+STH_DY	xvec13, 1*SIZE(C0, ldc, 1);
+STL_DY	xvec5, 2*SIZE(C0);
+STH_DY	xvec5, 3*SIZE(C0);
+STL_DY	xvec12, 4*SIZE(C0, ldc, 1);
+STH_DY	xvec12, 5*SIZE(C0, ldc, 1);
+STL_DY	xvec4, 6*SIZE(C0);
+STH_DY	xvec4, 7*SIZE(C0);
+EXTRA_DY $1, yvec11, xvec3;
+EXTRA_DY $1, yvec10, xvec2;
+#ifndef	TRMMKERNEL
+LDL_DY	0*SIZE(C1), xvec7, xvec7;
+LDH_DY	1*SIZE(C1), xvec7, xvec7;
+LDL_DY	2*SIZE(C1, ldc, 1), xvec6, xvec6;
+LDH_DY	3*SIZE(C1, ldc, 1), xvec6, xvec6;
+LDL_DY	4*SIZE(C1), xvec5, xvec5;
+LDH_DY	5*SIZE(C1), xvec5, xvec5;
+LDL_DY	6*SIZE(C1, ldc, 1), xvec4, xvec4;
+LDH_DY	7*SIZE(C1, ldc, 1), xvec4, xvec4;
+ADD_DY	xvec7, xvec11, xvec11;
+ADD_DY	xvec6, xvec3, xvec3;
+ADD_DY	xvec5, xvec10, xvec10;
+ADD_DY	xvec4, xvec2, xvec2;
+#endif
+STL_DY	xvec11, 0*SIZE(C1);
+STH_DY	xvec11, 1*SIZE(C1);
+STL_DY	xvec3, 2*SIZE(C1, ldc, 1);
+STH_DY	xvec3, 3*SIZE(C1, ldc, 1);
+STL_DY	xvec10, 4*SIZE(C1);
+STH_DY	xvec10, 5*SIZE(C1);
+STL_DY	xvec2, 6*SIZE(C1, ldc, 1);
+STH_DY	xvec2, 7*SIZE(C1, ldc, 1);
+EXTRA_DY $1, yvec9, xvec1;
+EXTRA_DY $1, yvec8, xvec0;
+#ifndef	TRMMKERNEL
+LDL_DY	0*SIZE(C1, ldc, 1), xvec5, xvec5;
+LDH_DY	1*SIZE(C1, ldc, 1), xvec5, xvec5;
+LDL_DY	2*SIZE(C1), xvec4, xvec4;
+LDH_DY	3*SIZE(C1), xvec4, xvec4;
+LDL_DY	4*SIZE(C1, ldc, 1), xvec3, xvec3;
+LDH_DY	5*SIZE(C1, ldc, 1), xvec3, xvec3;
+LDL_DY	6*SIZE(C1), xvec2, xvec2;
+LDH_DY	7*SIZE(C1), xvec2, xvec2;
+ADD_DY	xvec5, xvec9, xvec9;
+ADD_DY	xvec4, xvec1, xvec1;
+ADD_DY	xvec3, xvec8, xvec8;
+ADD_DY	xvec2, xvec0, xvec0;
+#endif
+STL_DY	xvec9, 0*SIZE(C1, ldc, 1);
+STH_DY	xvec9, 1*SIZE(C1, ldc, 1);
+STL_DY	xvec1, 2*SIZE(C1);
+STH_DY	xvec1, 3*SIZE(C1);
+STL_DY	xvec8, 4*SIZE(C1, ldc, 1);
+STH_DY	xvec8, 5*SIZE(C1, ldc, 1);
+STL_DY	xvec0, 6*SIZE(C1);
+STH_DY	xvec0, 7*SIZE(C1);
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA))||(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kkk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+LEAQ (ptrba, %rax, 4), ptrba;
+LEAQ (ptrbb, %rax, 4), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT) 
+ADDQ	$4, kk;
+#endif
+ADDQ	$8*SIZE, C0;
+ADDQ	$8*SIZE, C1;
+DECQ	i;
+JG .L1_bodyB;
+.align 32;
+.L1_loopE:;
+TEST $2, bm;
+JLE .L5_loopE;
+.align 32
+.L5_bodyB:
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb,ptrbb;
+#else
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+LEAQ (ptrba, %rax, 2), ptrba;
+LEAQ (ptrbb, %rax, 4), ptrbb;
+#endif
+XOR_DY	yvec15, yvec15, yvec15;
+XOR_DY	yvec14, yvec14, yvec14;
+XOR_DY	yvec13, yvec13, yvec13;
+XOR_DY	yvec12, yvec12, yvec12;
+#ifndef	TRMMKERNEL
+MOVQ bk,k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else
+MOVQ kk, %rax;
+#ifdef	LEFT
+ADDQ $2, %rax;
+#else 
+ADDQ $4, %rax;
+#endif
+MOVQ %rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L7_loopE;
+.align 32
+.L7_bodyB:
+#### Compute kernel ####
+#### Unroll times 1 ####
+LD_DY	0*SIZE(ptrba), yvec0;
+EDUP_DY	0*SIZE(ptrbb), yvec2;
+EDUP_DY	4*SIZE(ptrbb), yvec3;
+
+MUL_DY	yvec0, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+MUL_DY	yvec0, yvec3, yvec7;
+ADD1_DY	yvec7, yvec14, yvec14;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+
+MUL_DY	yvec0, yvec4, yvec6;
+ADD1_DY	yvec6, yvec13, yvec13;
+EDUP_DY	1*SIZE(ptrbb), yvec2;
+MUL_DY	yvec0, yvec5, yvec7;
+ADD1_DY	yvec7 ,yvec12, yvec12;
+EDUP_DY	5*SIZE(ptrbb), yvec3
+
+VPERMILP_DY	$0x05, yvec0, yvec0;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+MUL_DY	yvec0, yvec3, yvec7;
+ADD2_DY	yvec7, yvec14, yvec14;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+
+MUL_DY	yvec0, yvec4, yvec6;
+ADD2_DY	yvec6, yvec13, yvec13;
+MUL_DY	yvec0, yvec5, yvec7;
+ADD2_DY	yvec7, yvec12, yvec12;
+
+#### Unroll time 2 ####
+LD_DY	4*SIZE(ptrba), yvec0;
+EDUP_DY	8*SIZE(ptrbb), yvec2;
+EDUP_DY	12*SIZE(ptrbb), yvec3;
+
+MUL_DY	yvec0, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+MUL_DY	yvec0, yvec3, yvec7;
+ADD1_DY	yvec7, yvec14, yvec14;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+
+MUL_DY	yvec0, yvec4, yvec6;
+ADD1_DY	yvec6, yvec13, yvec13;
+EDUP_DY	9*SIZE(ptrbb), yvec2;
+MUL_DY	yvec0, yvec5, yvec7;
+ADD1_DY	yvec7 ,yvec12, yvec12;
+EDUP_DY	13*SIZE(ptrbb), yvec3
+
+VPERMILP_DY	$0x05, yvec0, yvec0;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+MUL_DY	yvec0, yvec3, yvec7;
+ADD2_DY	yvec7, yvec14, yvec14;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+
+MUL_DY	yvec0, yvec4, yvec6;
+ADD2_DY	yvec6, yvec13, yvec13;
+MUL_DY	yvec0, yvec5, yvec7;
+ADD2_DY	yvec7, yvec12, yvec12;
+
+#### Unroll time 3 ####
+LD_DY	8*SIZE(ptrba), yvec0;
+EDUP_DY	16*SIZE(ptrbb), yvec2;
+EDUP_DY	20*SIZE(ptrbb), yvec3;
+
+MUL_DY	yvec0, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+MUL_DY	yvec0, yvec3, yvec7;
+ADD1_DY	yvec7, yvec14, yvec14;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+
+MUL_DY	yvec0, yvec4, yvec6;
+ADD1_DY	yvec6, yvec13, yvec13;
+EDUP_DY	17*SIZE(ptrbb), yvec2;
+MUL_DY	yvec0, yvec5, yvec7;
+ADD1_DY	yvec7 ,yvec12, yvec12;
+EDUP_DY	21*SIZE(ptrbb), yvec3
+
+VPERMILP_DY	$0x05, yvec0, yvec0;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+MUL_DY	yvec0, yvec3, yvec7;
+ADD2_DY	yvec7, yvec14, yvec14;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+
+MUL_DY	yvec0, yvec4, yvec6;
+ADD2_DY	yvec6, yvec13, yvec13;
+MUL_DY	yvec0, yvec5, yvec7;
+ADD2_DY	yvec7, yvec12, yvec12;
+
+#### Unroll time 4 ####
+LD_DY	12*SIZE(ptrba), yvec0;
+EDUP_DY	24*SIZE(ptrbb), yvec2;
+EDUP_DY	28*SIZE(ptrbb), yvec3;
+
+MUL_DY	yvec0, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+MUL_DY	yvec0, yvec3, yvec7;
+ADD1_DY	yvec7, yvec14, yvec14;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+
+MUL_DY	yvec0, yvec4, yvec6;
+ADD1_DY	yvec6, yvec13, yvec13;
+EDUP_DY	25*SIZE(ptrbb), yvec2;
+MUL_DY	yvec0, yvec5, yvec7;
+ADD1_DY	yvec7 ,yvec12, yvec12;
+EDUP_DY	29*SIZE(ptrbb), yvec3
+
+VPERMILP_DY	$0x05, yvec0, yvec0;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+MUL_DY	yvec0, yvec3, yvec7;
+ADD2_DY	yvec7, yvec14, yvec14;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+
+MUL_DY	yvec0, yvec4, yvec6;
+ADD2_DY	yvec6, yvec13, yvec13;
+ADDQ	$16*SIZE, ptrba;
+MUL_DY	yvec0, yvec5, yvec7;
+ADD2_DY	yvec7, yvec12, yvec12;
+ADDQ	$32*SIZE, ptrbb;
+DECQ	k;
+JG .L7_bodyB;
+.align 32
+.L7_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else
+TEST $2, kkk;
+#endif
+JLE .L8_loopE;
+.align 32
+.L8_bodyB:
+#### Unroll times 1 ####
+LD_DY	0*SIZE(ptrba), yvec0;
+EDUP_DY	0*SIZE(ptrbb), yvec2;
+EDUP_DY	4*SIZE(ptrbb), yvec3;
+
+MUL_DY	yvec0, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+MUL_DY	yvec0, yvec3, yvec7;
+ADD1_DY	yvec7, yvec14, yvec14;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+
+MUL_DY	yvec0, yvec4, yvec6;
+ADD1_DY	yvec6, yvec13, yvec13;
+EDUP_DY	1*SIZE(ptrbb), yvec2;
+MUL_DY	yvec0, yvec5, yvec7;
+ADD1_DY	yvec7 ,yvec12, yvec12;
+EDUP_DY	5*SIZE(ptrbb), yvec3
+
+VPERMILP_DY	$0x05, yvec0, yvec0;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+MUL_DY	yvec0, yvec3, yvec7;
+ADD2_DY	yvec7, yvec14, yvec14;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+
+MUL_DY	yvec0, yvec4, yvec6;
+ADD2_DY	yvec6, yvec13, yvec13;
+MUL_DY	yvec0, yvec5, yvec7;
+ADD2_DY	yvec7, yvec12, yvec12;
+
+#### Unroll time 2 ####
+LD_DY	4*SIZE(ptrba), yvec0;
+EDUP_DY	8*SIZE(ptrbb), yvec2;
+EDUP_DY	12*SIZE(ptrbb), yvec3;
+
+MUL_DY	yvec0, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+MUL_DY	yvec0, yvec3, yvec7;
+ADD1_DY	yvec7, yvec14, yvec14;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+
+MUL_DY	yvec0, yvec4, yvec6;
+ADD1_DY	yvec6, yvec13, yvec13;
+EDUP_DY	9*SIZE(ptrbb), yvec2;
+MUL_DY	yvec0, yvec5, yvec7;
+ADD1_DY	yvec7 ,yvec12, yvec12;
+EDUP_DY	13*SIZE(ptrbb), yvec3
+
+VPERMILP_DY	$0x05, yvec0, yvec0;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+MUL_DY	yvec0, yvec3, yvec7;
+ADD2_DY	yvec7, yvec14, yvec14;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+
+MUL_DY	yvec0, yvec4, yvec6;
+ADD2_DY	yvec6, yvec13, yvec13;
+ADDQ	$8*SIZE, ptrba;
+MUL_DY	yvec0, yvec5, yvec7;
+ADD2_DY	yvec7, yvec12, yvec12;
+ADDQ	$16*SIZE, ptrbb;
+.L8_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else
+TEST $1, kkk;
+#endif
+JLE .L9_loopE;
+.align 32
+.L9_bodyB:
+#### Unroll times 1 ####
+LD_DY	0*SIZE(ptrba), yvec0;
+EDUP_DY	0*SIZE(ptrbb), yvec2;
+EDUP_DY	4*SIZE(ptrbb), yvec3;
+
+MUL_DY	yvec0, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+MUL_DY	yvec0, yvec3, yvec7;
+ADD1_DY	yvec7, yvec14, yvec14;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+
+MUL_DY	yvec0, yvec4, yvec6;
+ADD1_DY	yvec6, yvec13, yvec13;
+EDUP_DY	1*SIZE(ptrbb), yvec2;
+MUL_DY	yvec0, yvec5, yvec7;
+ADD1_DY	yvec7 ,yvec12, yvec12;
+EDUP_DY	5*SIZE(ptrbb), yvec3
+
+VPERMILP_DY	$0x05, yvec0, yvec0;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+MUL_DY	yvec0, yvec3, yvec7;
+ADD2_DY	yvec7, yvec14, yvec14;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+
+MUL_DY	yvec0, yvec4, yvec6;
+ADD2_DY	yvec6, yvec13, yvec13;
+MUL_DY	yvec0, yvec5, yvec7;
+ADD2_DY	yvec7, yvec12, yvec12;
+ADDQ	$4*SIZE, ptrba;
+ADDQ	$8*SIZE, ptrbb;
+
+.L9_loopE:
+#### Handle ####
+XOR_DY	yvec7, yvec7, yvec7;
+#if  defined(RN) || defined(RT) || defined(CN) || defined(CT)
+ADDSUB_DY	yvec15, yvec7, yvec15;
+ADDSUB_DY	yvec14, yvec7, yvec14;
+ADDSUB_DY	yvec13, yvec7, yvec13;
+ADDSUB_DY	yvec12, yvec7, yvec12;
+#elif defined(NR) || defined(NC) || defined(TR) || defined(TC)
+SUB_DY	yvec15, yvec7, yvec15;
+SUB_DY	yvec14, yvec7, yvec14;
+SUB_DY	yvec13, yvec7, yvec13;
+SUB_DY	yvec12, yvec7, yvec12;
+#elif defined(RR) || defined(RC) || defined(CR) || defined(CC) 
+VPERMILP_DY $0x05, yvec15, yvec15;
+VPERMILP_DY $0x05, yvec14, yvec14;
+VPERMILP_DY $0x05, yvec13, yvec13;
+VPERMILP_DY $0x05, yvec12, yvec12;
+ADDSUB_DY	yvec15, yvec7, yvec15;
+ADDSUB_DY	yvec14, yvec7, yvec14;
+ADDSUB_DY	yvec13, yvec7, yvec13;
+ADDSUB_DY	yvec12, yvec7, yvec12;
+VPERMILP_DY $0x05, yvec15, yvec15;
+VPERMILP_DY $0x05, yvec14, yvec14;
+VPERMILP_DY $0x05, yvec13, yvec13;
+VPERMILP_DY $0x05, yvec12, yvec12;
+#endif
+#### Load Alpha ####
+BROAD_DY	MEMALPHA_R, yvec7;
+BROAD_DY	MEMALPHA_I, yvec6;
+#### Multiply Alpha ####
+VPERMILP_DY	$0x05, yvec15, yvec5;
+MUL_DY	yvec7, yvec15, yvec15;
+MUL_DY	yvec6, yvec5, yvec5;
+ADD2_DY	yvec5, yvec15, yvec15;
+VPERMILP_DY	$0x05, yvec14, yvec4;
+MUL_DY 	yvec7, yvec14, yvec14;
+MUL_DY 	yvec6, yvec4, yvec4;
+ADD2_DY 	yvec4, yvec14, yvec14;
+VPERMILP_DY	$0x05, yvec13, yvec3;
+MUL_DY	yvec7, yvec13, yvec13;
+MUL_DY 	yvec6, yvec3, yvec3;
+ADD2_DY	yvec3, yvec13, yvec13;
+VPERMILP_DY	$0x05,yvec12, yvec2;
+MUL_DY 	yvec7, yvec12, yvec12;
+MUL_DY 	yvec6, yvec2, yvec2;
+ADD2_DY 	yvec2, yvec12, yvec12;
+#### Testing Alignment ####
+MOVQ	C0, %rax;
+OR		ldc, %rax;
+TEST	$15, %rax;
+JNE		.L9_loopEx;
+.align 32
+#### Writing back ####
+EXTRA_DY $1, yvec15, xvec7;
+EXTRA_DY $1, yvec14, xvec6;
+EXTRA_DY $1, yvec13, xvec5;
+EXTRA_DY $1, yvec12, xvec4;
+#ifndef	TRMMKERNEL
+ADD_DX	0*SIZE(C0), xvec15;
+ADD_DX	2*SIZE(C0, ldc, 1), xvec7;
+ADD_DX	0*SIZE(C0, ldc, 1), xvec13;
+ADD_DX	2*SIZE(C0), xvec5;
+ADD_DX	0*SIZE(C1), xvec14;
+ADD_DX	2*SIZE(C1, ldc, 1), xvec6;
+ADD_DX	0*SIZE(C1, ldc, 1), xvec12;
+ADD_DX	2*SIZE(C1), xvec4;
+#endif
+ST_DX	xvec15, 0*SIZE(C0);
+ST_DX	xvec7, 2*SIZE(C0, ldc, 1);
+ST_DX	xvec13, 0*SIZE(C0, ldc, 1);
+ST_DX	xvec5, 2*SIZE(C0);
+ST_DX	xvec14, 0*SIZE(C1);
+ST_DX	xvec6, 2*SIZE(C1, ldc, 1);
+ST_DX	xvec12, 0*SIZE(C1, ldc, 1);
+ST_DX	xvec4, 2*SIZE(C1);
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA))||(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kkk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+LEAQ (ptrba, %rax, 2), ptrba;
+LEAQ (ptrbb, %rax, 4), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT) 
+ADDQ	$2, kk;
+#endif
+ADDQ	$4*SIZE, C0;
+ADDQ	$4*SIZE, C1;
+JMP .L5_loopE;
+.align 32
+.L9_loopEx:
+EXTRA_DY $1, yvec15, xvec7;
+EXTRA_DY $1, yvec14, xvec6;
+EXTRA_DY $1, yvec13, xvec5;
+EXTRA_DY $2, yvec12, xvec4;
+#ifndef	TRMMKERNEL
+LDL_DX	0*SIZE(C0), xvec0;
+LDH_DX	1*SIZE(C0), xvec0;
+LDL_DX	2*SIZE(C0, ldc, 1), xvec1;
+LDH_DX	3*SIZE(C0, ldc, 1), xvec1;
+LDL_DX	0*SIZE(C0, ldc, 1), xvec2;
+LDH_DX	1*SIZE(C0, ldc, 1), xvec2;
+LDL_DX	2*SIZE(C0), xvec3;
+LDH_DX	3*SIZE(C0), xvec3;
+ADD_DX	xvec0, xvec15;
+ADD_DX	xvec1, xvec7;
+ADD_DX	xvec2, xvec13;
+ADD_DX	xvec3, xvec5;
+#endif
+STL_DX	xvec15, 0*SIZE(C0);
+STH_DX	xvec15, 1*SIZE(C0);
+STL_DX	xvec7, 2*SIZE(C0, ldc, 1);
+STH_DX	xvec7, 3*SIZE(C0, ldc, 1);
+STL_DX	xvec13, 0*SIZE(C0, ldc, 1);
+STH_DX	xvec13, 1*SIZE(C0, ldc, 1);
+STL_DX	xvec6, 2*SIZE(C0);
+STH_DX	xvec6, 3*SIZE(C0);
+#ifndef	TRMMKERNEL
+LDL_DX	0*SIZE(C1), xvec0;
+LDH_DX	1*SIZE(C1), xvec0;
+LDL_DX	2*SIZE(C1, ldc, 1), xvec1;
+LDH_DX	3*SIZE(C1, ldc, 1), xvec1;
+LDL_DX	0*SIZE(C1, ldc, 1), xvec2;
+LDH_DX	1*SIZE(C1, ldc, 1), xvec2;
+LDL_DX	2*SIZE(C1), xvec3;
+LDH_DX	3*SIZE(C1), xvec3;
+ADD_DX	xvec0, xvec14;
+ADD_DX	xvec1, xvec6;
+ADD_DX	xvec2, xvec12;
+ADD_DX	xvec3, xvec4;
+#endif
+STL_DX	xvec14, 0*SIZE(C1);
+STH_DX	xvec14, 1*SIZE(C1);
+STL_DX	xvec6, 2*SIZE(C1, ldc, 1);
+STH_DX	xvec6, 3*SIZE(C1, ldc, 1);
+STL_DX	xvec12, 0*SIZE(C1, ldc, 1);
+STH_DX	xvec12, 1*SIZE(C1, ldc, 1);
+STL_DX	xvec4, 2*SIZE(C1);
+STH_DX	xvec4, 3*SIZE(C1);
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA))||(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kkk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+LEAQ (ptrba, %rax, 2), ptrba;
+LEAQ (ptrbb, %rax, 4), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT) 
+ADDQ	$2, kk;
+#endif
+ADDQ	$4*SIZE, C0;
+ADDQ	$4*SIZE, C1;
+.L5_loopE:
+TEST $1, bm;
+JLE .L6_loopE;
+.align 32
+.L6_bodyB:
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb,ptrbb;
+#else
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+ADDQ %rax, ptrba;
+LEAQ (ptrbb, %rax, 4), ptrbb;
+#endif
+XOR_DY	yvec15, yvec15, yvec15;
+XOR_DY	yvec14, yvec14, yvec14;
+#ifndef	TRMMKERNEL
+MOVQ bk,k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else
+MOVQ kk, %rax;
+#ifdef	LEFT
+ADDQ $1, %rax;
+#else 
+ADDQ $4, %rax;
+#endif
+MOVQ %rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L10_loopE;
+.align 32
+.L10_bodyB:
+LD_DY	0*SIZE(ptrba), yvec0;	#### A1r A1i A2r A2i
+EDUP_DY	0*SIZE(ptrbb), yvec2;
+EDUP_DY	4*SIZE(ptrbb), yvec3;
+
+SHUF_DY	$0x20, yvec0, yvec0, yvec1;
+MUL_DY	yvec1, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+MUL_DY	yvec1, yvec3, yvec7;
+ADD1_DY	yvec7, yvec14, yvec14;
+
+VPERMILP_DY	$0x05, yvec1, yvec4;
+EDUP_DY	1*SIZE(ptrbb), yvec2;
+EDUP_DY	5*SIZE(ptrbb), yvec3;
+MUL_DY	yvec4, yvec2, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+MUL_DY	yvec4, yvec3, yvec7;
+ADD2_DY	yvec7, yvec14, yvec14;
+
+SHUF_DY	$0x31, yvec0, yvec0, yvec1;
+EDUP_DY	8*SIZE(ptrbb), yvec2;
+EDUP_DY	12*SIZE(ptrbb), yvec3;
+
+MUL_DY	yvec1, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+MUL_DY	yvec1, yvec3, yvec7;
+ADD1_DY	yvec7, yvec14, yvec14;
+
+VPERMILP_DY	$0x05, yvec1, yvec4;
+EDUP_DY	9*SIZE(ptrbb), yvec2;
+EDUP_DY	13*SIZE(ptrbb), yvec3;
+MUL_DY	yvec4, yvec2, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+MUL_DY	yvec4, yvec3, yvec7;
+ADD2_DY	yvec7, yvec14, yvec14;
+
+LD_DY	4*SIZE(ptrba), yvec0;
+EDUP_DY	16*SIZE(ptrbb), yvec2;
+EDUP_DY	20*SIZE(ptrbb), yvec3;
+
+SHUF_DY	$0x20, yvec0, yvec0, yvec1;
+MUL_DY	yvec1, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+MUL_DY	yvec1, yvec3, yvec7;
+ADD1_DY	yvec7, yvec14, yvec14;
+
+VPERMILP_DY	$0x05, yvec1, yvec4;
+EDUP_DY	17*SIZE(ptrbb), yvec2;
+EDUP_DY	21*SIZE(ptrbb), yvec3;
+MUL_DY	yvec4, yvec2, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+MUL_DY	yvec4, yvec3, yvec7;
+ADD2_DY	yvec7, yvec14, yvec14;
+
+SHUF_DY	$0x31, yvec0, yvec0, yvec1;
+EDUP_DY	24*SIZE(ptrbb), yvec2;
+EDUP_DY	28*SIZE(ptrbb), yvec3;
+MUL_DY	yvec1, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+MUL_DY	yvec1, yvec3, yvec7;
+ADD1_DY	yvec7, yvec14, yvec14;
+
+VPERMILP_DY	$0x05, yvec1, yvec4;
+EDUP_DY	25*SIZE(ptrbb), yvec2;
+EDUP_DY	29*SIZE(ptrbb), yvec3;
+MUL_DY	yvec4, yvec2, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+MUL_DY	yvec4, yvec3, yvec7;
+ADD2_DY	yvec7, yvec14, yvec14
+ADDQ	$8*SIZE, ptrba;
+ADDQ	$32*SIZE, ptrbb;
+DECQ	k;
+JG .L10_bodyB;
+.align 32
+.L10_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else
+TEST $2, kkk;
+#endif
+JLE .L11_loopE;
+.align 32
+.L11_bodyB:
+LD_DY	0*SIZE(ptrba), yvec0;	#### A1r A1i A2r A2i
+EDUP_DY	0*SIZE(ptrbb), yvec2;
+EDUP_DY	4*SIZE(ptrbb), yvec3;
+
+SHUF_DY	$0x20, yvec0, yvec0, yvec1;
+MUL_DY	yvec1, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+MUL_DY	yvec1, yvec3, yvec7;
+ADD1_DY	yvec7, yvec14, yvec14;
+
+VPERMILP_DY	$0x05, yvec1, yvec4;
+EDUP_DY	1*SIZE(ptrbb), yvec2;
+EDUP_DY	5*SIZE(ptrbb), yvec3;
+MUL_DY	yvec4, yvec2, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+MUL_DY	yvec4, yvec3, yvec7;
+ADD2_DY	yvec7, yvec14, yvec14;
+
+SHUF_DY	$0x31, yvec0, yvec0, yvec1;
+EDUP_DY	8*SIZE(ptrbb), yvec2;
+EDUP_DY	12*SIZE(ptrbb), yvec3;
+
+MUL_DY	yvec1, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+MUL_DY	yvec1, yvec3, yvec7;
+ADD1_DY	yvec7, yvec14, yvec14;
+
+VPERMILP_DY	$0x05, yvec1, yvec4;
+EDUP_DY	9*SIZE(ptrbb), yvec2;
+EDUP_DY	13*SIZE(ptrbb), yvec3;
+MUL_DY	yvec4, yvec2, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+MUL_DY	yvec4, yvec3, yvec7;
+ADD2_DY	yvec7, yvec14, yvec14;
+ADDQ	$4*SIZE, ptrba;
+ADDQ	$16*SIZE, ptrbb;
+
+.L11_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else
+TEST $1, kkk;
+#endif
+JLE .L12_loopE;
+.align 32
+.L12_bodyB:
+LD_DY	0*SIZE(ptrba), yvec0;	#### A1r A1i A2r A2i
+EDUP_DY	0*SIZE(ptrbb), yvec2;
+EDUP_DY	4*SIZE(ptrbb), yvec3;
+
+SHUF_DY	$0x20, yvec0, yvec0, yvec1;
+MUL_DY	yvec1, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+MUL_DY	yvec1, yvec3, yvec7;
+ADD1_DY	yvec7, yvec14, yvec14;
+
+VPERMILP_DY	$0x05, yvec1, yvec4;
+EDUP_DY	1*SIZE(ptrbb), yvec2;
+EDUP_DY	5*SIZE(ptrbb), yvec3;
+MUL_DY	yvec4, yvec2, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+MUL_DY	yvec4, yvec3, yvec7;
+ADD2_DY	yvec7, yvec14, yvec14;
+ADDQ	$2*SIZE, ptrba;
+ADDQ	$8*SIZE, ptrbb;
+
+.L12_loopE:
+#### Handle ####
+XOR_DY	yvec7, yvec7, yvec7;
+#if  defined(RN) || defined(RT) || defined(CN) || defined(CT)
+ADDSUB_DY	yvec15, yvec7, yvec15;
+ADDSUB_DY	yvec14, yvec7, yvec14;
+#elif defined(NR) || defined(NC) || defined(TR) || defined(TC)
+SUB_DY	yvec15, yvec7, yvec15;
+SUB_DY	yvec14, yvec7, yvec14;
+#elif defined(RR) || defined(RC) || defined(CR) || defined(CC) 
+VPERMILP_DY $0x05, yvec15, yvec15;
+VPERMILP_DY $0x05, yvec14, yvec14;
+ADDSUB_DY	yvec15, yvec7, yvec15;
+ADDSUB_DY	yvec14, yvec7, yvec14;
+VPERMILP_DY $0x05, yvec15, yvec15;
+VPERMILP_DY $0x05, yvec14, yvec14;
+#endif
+#### Multiply Alpha	####
+BROAD_DY MEMALPHA_R, yvec7;
+BROAD_DY MEMALPHA_I, yvec6;
+VPERMILP_DY	$0x05, yvec15, yvec5;
+MUL_DY	yvec7, yvec15, yvec15;
+MUL_DY	yvec6, yvec5, yvec5;
+ADD2_DY	yvec5, yvec15, yvec15;
+VPERMILP_DY	$0x05, yvec14, yvec4;
+MUL_DY 	yvec7, yvec14, yvec14;
+MUL_DY 	yvec6, yvec4, yvec4;
+ADD2_DY 	yvec4, yvec14, yvec14;
+#### Writing Back ####
+EXTRA_DY $1, yvec15, xvec7;
+EXTRA_DY $1, yvec14, xvec6;
+#ifndef	TRMMKERNEL
+LDL_DX	0*SIZE(C0), xvec0;
+LDH_DX	1*SIZE(C0), xvec0;
+LDL_DX	0*SIZE(C0, ldc, 1), xvec1;
+LDH_DX	1*SIZE(C0, ldc, 1), xvec1;
+LDL_DX	0*SIZE(C1), xvec2;
+LDH_DX	1*SIZE(C1), xvec2;
+LDL_DX	0*SIZE(C1, ldc, 1), xvec3;
+LDH_DX	1*SIZE(C1, ldc, 1), xvec3;
+ADD_DX	xvec0, xvec15;
+ADD_DX	xvec1, xvec7;
+ADD_DX	xvec2, xvec14;
+ADD_DX	xvec3, xvec6;
+#endif 
+STL_DX	xvec15, 0*SIZE(C0);
+STH_DX	xvec15, 1*SIZE(C0);
+STL_DX	xvec7, 0*SIZE(C0, ldc, 1);
+STH_DX	xvec7, 1*SIZE(C0, ldc, 1);
+STL_DX	xvec14, 0*SIZE(C1);
+STH_DX	xvec14, 1*SIZE(C1);
+STL_DX	xvec6, 0*SIZE(C1, ldc, 1);
+STH_DX	xvec6, 1*SIZE(C1, ldc, 1);
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA))||(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kkk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+ADDQ %rax, ptrba;
+LEAQ (ptrbb, %rax, 4), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT) 
+ADDQ	$1, kk;
+#endif
+ADDQ	$2*SIZE, C0;
+ADDQ	$2*SIZE, C1;
+.L6_loopE:
+#if defined(TRMMKERNEL) && !defined(LEFT)
+ADDQ	$4, kk;
+#endif
+MOVQ bk,k;
+SALQ $6,k;
+ADDQ k,bb;
+LEAQ (C,ldc,4),C;
+.L0_bodyE:;
+DECQ j;
+JG .L0_bodyB;
+.align 32;
+.L0_loopE:;
+TEST $2, bn;
+JLE	.L20_loopE;
+.align 32
+.L20_bodyB:
+#if defined(TRMMKERNEL) && defined(LEFT)
+MOVQ OFFSET, %rax;
+MOVQ %rax, kk;
+#endif
+MOVQ C, C0;
+LEAQ (C, ldc, 1), C1;
+MOVQ ba, ptrba;
+MOVQ bm, i;
+SARQ $2, i;
+JLE .L21_loopE;
+.align 32
+.L21_bodyB:
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb,ptrbb;
+#else
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+LEAQ (ptrba, %rax, 4), ptrba;
+LEAQ (ptrbb, %rax, 2), ptrbb;
+#endif
+XOR_DY yvec15, yvec15, yvec15;
+XOR_DY yvec14, yvec14, yvec14;
+XOR_DY yvec13, yvec13, yvec13;
+XOR_DY yvec12, yvec12, yvec12;
+#ifndef	TRMMKERNEL
+MOVQ bk,k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else
+MOVQ kk, %rax;
+#ifdef	LEFT
+ADDQ $4, %rax;
+#else 
+ADDQ $2, %rax;
+#endif
+MOVQ %rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L211_loopE;
+.align 32
+.L211_bodyB:
+#### Unroll time 1 ####
+EDUP_DY	0*SIZE(ptrbb), yvec2;
+LD_DY	0*SIZE(ptrba), yvec0;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+LD_DY	4*SIZE(ptrba), yvec1;
+MUL_DY	yvec1, yvec2, yvec7;
+ADD1_DY	yvec7, yvec14, yvec14;
+
+EDUP_DY	1*SIZE(ptrbb), yvec3;
+MUL_DY	yvec0, yvec4, yvec6;
+ADD1_DY	yvec6, yvec13, yvec13;
+VPERMILP_DY	$0x05, yvec0, yvec0;
+MUL_DY	yvec1, yvec4, yvec7;
+ADD1_DY 	yvec7, yvec12, yvec12;
+VPERMILP_DY	$0x05, yvec1, yvec1;
+
+MUL_DY	yvec0, yvec3, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+MUL_DY	yvec1, yvec3, yvec7;
+ADD2_DY	yvec7, yvec14, yvec14;
+
+MUL_DY	yvec0, yvec5, yvec6;
+ADD2_DY	yvec6, yvec13, yvec13;
+MUL_DY	yvec1, yvec5, yvec7;
+ADD2_DY	yvec7, yvec12, yvec12;
+
+#### Unroll time 2 ####
+EDUP_DY	4*SIZE(ptrbb), yvec2;
+LD_DY	8*SIZE(ptrba), yvec0;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+LD_DY	12*SIZE(ptrba), yvec1;
+MUL_DY	yvec1, yvec2, yvec7;
+ADD1_DY	yvec7, yvec14, yvec14;
+
+EDUP_DY	5*SIZE(ptrbb), yvec3;
+MUL_DY	yvec0, yvec4, yvec6;
+ADD1_DY	yvec6, yvec13, yvec13;
+VPERMILP_DY	$0x05, yvec0, yvec0;
+MUL_DY	yvec1, yvec4, yvec7;
+ADD1_DY 	yvec7, yvec12, yvec12;
+VPERMILP_DY	$0x05, yvec1, yvec1;
+
+MUL_DY	yvec0, yvec3, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+MUL_DY	yvec1, yvec3, yvec7;
+ADD2_DY	yvec7, yvec14, yvec14;
+
+MUL_DY	yvec0, yvec5, yvec6;
+ADD2_DY	yvec6, yvec13, yvec13;
+MUL_DY	yvec1, yvec5, yvec7;
+ADD2_DY	yvec7, yvec12, yvec12;
+
+#### Unroll time 3 ####
+EDUP_DY	8*SIZE(ptrbb), yvec2;
+LD_DY	16*SIZE(ptrba), yvec0;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+LD_DY	20*SIZE(ptrba), yvec1;
+MUL_DY	yvec1, yvec2, yvec7;
+ADD1_DY	yvec7, yvec14, yvec14;
+
+EDUP_DY	9*SIZE(ptrbb), yvec3;
+MUL_DY	yvec0, yvec4, yvec6;
+ADD1_DY	yvec6, yvec13, yvec13;
+VPERMILP_DY	$0x05, yvec0, yvec0;
+MUL_DY	yvec1, yvec4, yvec7;
+ADD1_DY 	yvec7, yvec12, yvec12;
+VPERMILP_DY	$0x05, yvec1, yvec1;
+
+MUL_DY	yvec0, yvec3, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+MUL_DY	yvec1, yvec3, yvec7;
+ADD2_DY	yvec7, yvec14, yvec14;
+
+MUL_DY	yvec0, yvec5, yvec6;
+ADD2_DY	yvec6, yvec13, yvec13;
+MUL_DY	yvec1, yvec5, yvec7;
+ADD2_DY	yvec7, yvec12, yvec12;
+
+#### Unroll time 4 ####
+EDUP_DY	12*SIZE(ptrbb), yvec2;
+LD_DY	24*SIZE(ptrba), yvec0;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+LD_DY	28*SIZE(ptrba), yvec1;
+MUL_DY	yvec1, yvec2, yvec7;
+ADD1_DY	yvec7, yvec14, yvec14;
+
+EDUP_DY	13*SIZE(ptrbb), yvec3;
+MUL_DY	yvec0, yvec4, yvec6;
+ADD1_DY	yvec6, yvec13, yvec13;
+VPERMILP_DY	$0x05, yvec0, yvec0;
+MUL_DY	yvec1, yvec4, yvec7;
+ADD1_DY 	yvec7, yvec12, yvec12;
+VPERMILP_DY	$0x05, yvec1, yvec1;
+
+MUL_DY	yvec0, yvec3, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+MUL_DY	yvec1, yvec3, yvec7;
+ADD2_DY	yvec7, yvec14, yvec14;
+ADDQ	$16*SIZE, ptrbb;
+
+MUL_DY	yvec0, yvec5, yvec6;
+ADD2_DY	yvec6, yvec13, yvec13;
+MUL_DY	yvec1, yvec5, yvec7;
+ADD2_DY	yvec7, yvec12, yvec12;
+ADDQ	$32*SIZE, ptrba;
+DECQ	k;
+JG .L211_bodyB;
+.align 32
+.L211_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else
+TEST $2, kkk;
+#endif
+JLE .L212_loopE;
+.align 32
+.L212_bodyB:
+#### Unroll time 1 ####
+EDUP_DY	0*SIZE(ptrbb), yvec2;
+LD_DY	0*SIZE(ptrba), yvec0;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+LD_DY	4*SIZE(ptrba), yvec1;
+MUL_DY	yvec1, yvec2, yvec7;
+ADD1_DY	yvec7, yvec14, yvec14;
+
+EDUP_DY	1*SIZE(ptrbb), yvec3;
+MUL_DY	yvec0, yvec4, yvec6;
+ADD1_DY	yvec6, yvec13, yvec13;
+VPERMILP_DY	$0x05, yvec0, yvec0;
+MUL_DY	yvec1, yvec4, yvec7;
+ADD1_DY 	yvec7, yvec12, yvec12;
+VPERMILP_DY	$0x05, yvec1, yvec1;
+
+MUL_DY	yvec0, yvec3, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+MUL_DY	yvec1, yvec3, yvec7;
+ADD2_DY	yvec7, yvec14, yvec14;
+
+MUL_DY	yvec0, yvec5, yvec6;
+ADD2_DY	yvec6, yvec13, yvec13;
+MUL_DY	yvec1, yvec5, yvec7;
+ADD2_DY	yvec7, yvec12, yvec12;
+
+#### Unroll time 2 ####
+EDUP_DY	4*SIZE(ptrbb), yvec2;
+LD_DY	8*SIZE(ptrba), yvec0;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+LD_DY	12*SIZE(ptrba), yvec1;
+MUL_DY	yvec1, yvec2, yvec7;
+ADD1_DY	yvec7, yvec14, yvec14;
+
+EDUP_DY	5*SIZE(ptrbb), yvec3;
+MUL_DY	yvec0, yvec4, yvec6;
+ADD1_DY	yvec6, yvec13, yvec13;
+VPERMILP_DY	$0x05, yvec0, yvec0;
+MUL_DY	yvec1, yvec4, yvec7;
+ADD1_DY 	yvec7, yvec12, yvec12;
+VPERMILP_DY	$0x05, yvec1, yvec1;
+
+MUL_DY	yvec0, yvec3, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+MUL_DY	yvec1, yvec3, yvec7;
+ADD2_DY	yvec7, yvec14, yvec14;
+
+MUL_DY	yvec0, yvec5, yvec6;
+ADD2_DY	yvec6, yvec13, yvec13;
+MUL_DY	yvec1, yvec5, yvec7;
+ADD2_DY	yvec7, yvec12, yvec12;
+
+ADDQ	$8*SIZE, ptrbb;
+ADDQ	$16*SIZE, ptrba;
+
+.L212_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else
+TEST $1, kkk;
+#endif
+JLE .L213_loopE;
+.align 32
+.L213_bodyB:
+#### Unroll time 1 ####
+EDUP_DY	0*SIZE(ptrbb), yvec2;
+LD_DY	0*SIZE(ptrba), yvec0;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+LD_DY	4*SIZE(ptrba), yvec1;
+MUL_DY	yvec1, yvec2, yvec7;
+ADD1_DY	yvec7, yvec14, yvec14;
+
+EDUP_DY	1*SIZE(ptrbb), yvec3;
+MUL_DY	yvec0, yvec4, yvec6;
+ADD1_DY	yvec6, yvec13, yvec13;
+VPERMILP_DY	$0x05, yvec0, yvec0;
+MUL_DY	yvec1, yvec4, yvec7;
+ADD1_DY 	yvec7, yvec12, yvec12;
+VPERMILP_DY	$0x05, yvec1, yvec1;
+
+MUL_DY	yvec0, yvec3, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+MUL_DY	yvec1, yvec3, yvec7;
+ADD2_DY	yvec7, yvec14, yvec14;
+
+MUL_DY	yvec0, yvec5, yvec6;
+ADD2_DY	yvec6, yvec13, yvec13;
+MUL_DY	yvec1, yvec5, yvec7;
+ADD2_DY	yvec7, yvec12, yvec12;
+ADDQ	$4*SIZE, ptrbb;
+ADDQ	$8*SIZE, ptrba;
+
+.L213_loopE:
+#### Handle ####
+XOR_DY	yvec7, yvec7, yvec7;
+#if  defined(RN) || defined(RT) || defined(CN) || defined(CT)
+ADDSUB_DY	yvec15, yvec7, yvec15;
+ADDSUB_DY	yvec14, yvec7, yvec14;
+ADDSUB_DY	yvec13, yvec7, yvec13;
+ADDSUB_DY	yvec12, yvec7, yvec12;
+#elif defined(NR) || defined(NC) || defined(TR) || defined(TC)
+SUB_DY	yvec15, yvec7, yvec15;
+SUB_DY	yvec14, yvec7, yvec14;
+SUB_DY	yvec13, yvec7, yvec13;
+SUB_DY	yvec12, yvec7, yvec12;
+#elif defined(RR) || defined(RC) || defined(CR) || defined(CC) 
+VPERMILP_DY $0x05, yvec15, yvec15;
+VPERMILP_DY $0x05, yvec14, yvec14;
+VPERMILP_DY $0x05, yvec13, yvec13;
+VPERMILP_DY $0x05, yvec12, yvec12;
+ADDSUB_DY	yvec15, yvec7, yvec15;
+ADDSUB_DY	yvec14, yvec7, yvec14;
+ADDSUB_DY	yvec13, yvec7, yvec13;
+ADDSUB_DY	yvec12, yvec7, yvec12;
+VPERMILP_DY $0x05, yvec15, yvec15;
+VPERMILP_DY $0x05, yvec14, yvec14;
+VPERMILP_DY $0x05, yvec13, yvec13;
+VPERMILP_DY $0x05, yvec12, yvec12;
+#endif
+#### Load Alpha ####
+BROAD_DY MEMALPHA_R,yvec7;
+BROAD_DY MEMALPHA_I,yvec6;
+#### Multiply Alpha	####
+VPERMILP_DY	$0x05, yvec15, yvec5;
+MUL_DY	yvec7, yvec15, yvec15;
+MUL_DY	yvec6, yvec5, yvec5;
+ADD2_DY	yvec5, yvec15, yvec15;
+VPERMILP_DY	$0x05, yvec14, yvec4;
+MUL_DY 	yvec7, yvec14, yvec14;
+MUL_DY 	yvec6, yvec4, yvec4;
+ADD2_DY 	yvec4, yvec14, yvec14;
+VPERMILP_DY	$0x05, yvec13, yvec3;
+MUL_DY	yvec7, yvec13, yvec13;
+MUL_DY 	yvec6, yvec3, yvec3;
+ADD2_DY	yvec3, yvec13, yvec13;
+VPERMILP_DY	$0x05,yvec12, yvec2;
+MUL_DY 	yvec7, yvec12, yvec12;
+MUL_DY 	yvec6, yvec2, yvec2;
+ADD2_DY 	yvec2, yvec12, yvec12;
+EXTRA_DY $1, yvec15, xvec7;
+EXTRA_DY $1, yvec14, xvec6;
+EXTRA_DY $1, yvec13, xvec5;
+EXTRA_DY $1, yvec12, xvec4;
+#### Testing Alignment #### 
+MOVQ	C0, %rax;
+OR		ldc, %rax;
+TEST 	$15, %rax;
+JNE		.L213_loopEx;
+.align 32
+#### Writing back ####
+#ifndef	TRMMKERNEL
+ADD_DX 0*SIZE(C0),xvec15;
+ADD_DX 2*SIZE(C1),xvec7;
+ADD_DX 4*SIZE(C0),xvec14;
+ADD_DX 6*SIZE(C1),xvec6;
+ADD_DX 0*SIZE(C1),xvec13;
+ADD_DX 2*SIZE(C0),xvec5;
+ADD_DX 4*SIZE(C1),xvec12;
+ADD_DX 6*SIZE(C0),xvec4;
+#endif
+ST_DX xvec15,0*SIZE(C0);
+ST_DX xvec7,2*SIZE(C1);
+ST_DX xvec14,4*SIZE(C0);
+ST_DX xvec6,6*SIZE(C1);
+ST_DX xvec13,0*SIZE(C1);
+ST_DX xvec5,2*SIZE(C0);
+ST_DX xvec12,4*SIZE(C1);
+ST_DX xvec4,6*SIZE(C0);
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA))||(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kkk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+LEAQ (ptrba, %rax, 4), ptrba;
+LEAQ (ptrbb, %rax, 2), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT) 
+ADDQ	$4, kk;
+#endif
+ADDQ $8*SIZE, C0;
+ADDQ $8*SIZE, C1;
+DECQ	i;
+JG .L21_bodyB;
+JMP .L21_loopE;
+.align 32
+.L213_loopEx:
+#ifndef	TRMMKERNEL
+LDL_DX	0*SIZE(C0), xvec0;
+LDH_DX	1*SIZE(C0), xvec0;
+LDL_DX	2*SIZE(C1), xvec1;
+LDH_DX	3*SIZE(C1), xvec1;
+LDL_DX	4*SIZE(C0), xvec2;
+LDH_DX	5*SIZE(C0), xvec2;
+LDL_DX	6*SIZE(C1), xvec3;
+LDH_DX	7*SIZE(C1), xvec3;
+ADD_DX	xvec0, xvec15;
+ADD_DX	xvec1, xvec7;
+ADD_DX	xvec2, xvec14;
+ADD_DX	xvec3, xvec6;
+#endif
+STL_DX	xvec15, 0*SIZE(C0);
+STH_DX	xvec15, 1*SIZE(C0);
+STL_DX	xvec7, 2*SIZE(C1);
+STH_DX	xvec7, 3*SIZE(C1);
+STL_DX	xvec14, 4*SIZE(C0);
+STH_DX	xvec14, 5*SIZE(C0);
+STL_DX	xvec6, 6*SIZE(C1);
+STH_DX	xvec6, 7*SIZE(C1);
+#ifndef	TRMMKERNEL
+LDL_DX	0*SIZE(C1), xvec3;
+LDH_DX	1*SIZE(C1), xvec3;
+LDL_DX	2*SIZE(C0), xvec2;
+LDH_DX	3*SIZE(C0), xvec2;
+LDL_DX	4*SIZE(C1), xvec1;
+LDH_DX	5*SIZE(C1), xvec1;
+LDL_DX	6*SIZE(C0), xvec0;
+LDH_DX	7*SIZE(C0), xvec0;
+ADD_DX	xvec3, xvec13;
+ADD_DX	xvec2, xvec5;
+ADD_DX	xvec1, xvec12;
+ADD_DX	xvec0, xvec4;
+#endif
+STL_DX	xvec13, 0*SIZE(C1);
+STH_DX	xvec13, 1*SIZE(C1);
+STL_DX	xvec5, 2*SIZE(C0);
+STH_DX	xvec5, 3*SIZE(C0);
+STL_DX	xvec12, 4*SIZE(C1);
+STH_DX	xvec12, 5*SIZE(C1);
+STL_DX	xvec4, 6*SIZE(C0);
+STH_DX	xvec4, 7*SIZE(C0);
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA))||(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kkk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+LEAQ (ptrba, %rax, 4), ptrba;
+LEAQ (ptrbb, %rax, 2), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT) 
+ADDQ	$4, kk;
+#endif
+ADDQ	$8*SIZE, C0;
+ADDQ	$8*SIZE, C1;
+DECQ	i;
+JG .L21_bodyB;
+.align 32
+.L21_loopE:
+TEST $2, bm;
+JLE .L22_loopE;
+.align 32
+.L22_bodyB:
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb,ptrbb;
+#else
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+LEAQ (ptrba, %rax, 2), ptrba;
+LEAQ (ptrbb, %rax, 2), ptrbb;
+#endif
+XOR_DY	yvec15, yvec15, yvec15;
+XOR_DY	yvec14, yvec14, yvec13;
+#ifndef	TRMMKERNEL
+MOVQ bk,k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else
+MOVQ kk, %rax;
+#ifdef	LEFT
+ADDQ $2, %rax;
+#else 
+ADDQ $2, %rax;
+#endif
+MOVQ %rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L221_loopE;
+.align 32
+.L221_bodyB:
+#### Unroll time 1 ####
+EDUP_DY	0*SIZE(ptrbb), yvec2;
+LD_DY	0*SIZE(ptrba), yvec0;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+
+EDUP_DY	1*SIZE(ptrbb), yvec3;
+MUL_DY	yvec0, yvec4, yvec6;
+ADD1_DY	yvec6, yvec13, yvec13;
+VPERMILP_DY	$0x05, yvec0, yvec0;
+
+MUL_DY	yvec0, yvec3, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+
+MUL_DY	yvec0, yvec5, yvec6;
+ADD2_DY	yvec6, yvec13, yvec13;
+
+#### Unroll time 2 ####
+EDUP_DY	4*SIZE(ptrbb), yvec2;
+LD_DY	4*SIZE(ptrba), yvec0;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+
+EDUP_DY	5*SIZE(ptrbb), yvec3;
+MUL_DY	yvec0, yvec4, yvec6;
+ADD1_DY	yvec6, yvec13, yvec13;
+VPERMILP_DY	$0x05, yvec0, yvec0;
+
+MUL_DY	yvec0, yvec3, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+
+MUL_DY	yvec0, yvec5, yvec6;
+ADD2_DY	yvec6, yvec13, yvec13;
+
+#### Unroll time 3 ####
+EDUP_DY	8*SIZE(ptrbb), yvec2;
+LD_DY	8*SIZE(ptrba), yvec0;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+
+EDUP_DY	9*SIZE(ptrbb), yvec3;
+MUL_DY	yvec0, yvec4, yvec6;
+ADD1_DY	yvec6, yvec13, yvec13;
+VPERMILP_DY	$0x05, yvec0, yvec0;
+
+MUL_DY	yvec0, yvec3, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+
+MUL_DY	yvec0, yvec5, yvec6;
+ADD2_DY	yvec6, yvec13, yvec13;
+
+#### Unroll time 4 ####
+EDUP_DY	12*SIZE(ptrbb), yvec2;
+LD_DY	12*SIZE(ptrba), yvec0;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+
+EDUP_DY	13*SIZE(ptrbb), yvec3;
+MUL_DY	yvec0, yvec4, yvec6;
+ADD1_DY	yvec6, yvec13, yvec13;
+VPERMILP_DY	$0x05, yvec0, yvec0;
+
+MUL_DY	yvec0, yvec3, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+ADDQ	$16*SIZE, ptrbb;
+
+MUL_DY	yvec0, yvec5, yvec6;
+ADD2_DY	yvec6, yvec13, yvec13;
+ADDQ	$16*SIZE, ptrba;
+DECQ	k;
+JG .L221_bodyB;
+.align 32
+.L221_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else
+TEST $2, kkk;
+#endif
+JLE .L222_loopE;
+.align 32
+.L222_bodyB:
+#### Unroll time 1 ####
+EDUP_DY	0*SIZE(ptrbb), yvec2;
+LD_DY	0*SIZE(ptrba), yvec0;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+
+EDUP_DY	1*SIZE(ptrbb), yvec3;
+MUL_DY	yvec0, yvec4, yvec6;
+ADD1_DY	yvec6, yvec13, yvec13;
+VPERMILP_DY	$0x05, yvec0, yvec0;
+
+MUL_DY	yvec0, yvec3, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+
+MUL_DY	yvec0, yvec5, yvec6;
+ADD2_DY	yvec6, yvec13, yvec13;
+
+#### Unroll time 2 ####
+EDUP_DY	4*SIZE(ptrbb), yvec2;
+LD_DY	4*SIZE(ptrba), yvec0;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+
+EDUP_DY	5*SIZE(ptrbb), yvec3;
+MUL_DY	yvec0, yvec4, yvec6;
+ADD1_DY	yvec6, yvec13, yvec13;
+VPERMILP_DY	$0x05, yvec0, yvec0;
+
+MUL_DY	yvec0, yvec3, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+
+MUL_DY	yvec0, yvec5, yvec6;
+ADD2_DY	yvec6, yvec13, yvec13;
+ADDQ	$8*SIZE, ptrba;
+ADDQ	$8*SIZE, ptrbb;
+
+.L222_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else
+TEST $1, kkk;
+#endif
+JLE .L223_loopE;
+.align 32
+.L223_bodyB:
+#### Unroll time 1 ####
+EDUP_DY	0*SIZE(ptrbb), yvec2;
+LD_DY	0*SIZE(ptrba), yvec0;
+SHUF_DY	$0x03, yvec2, yvec2, yvec4;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+
+EDUP_DY	1*SIZE(ptrbb), yvec3;
+MUL_DY	yvec0, yvec4, yvec6;
+ADD1_DY	yvec6, yvec13, yvec13;
+VPERMILP_DY	$0x05, yvec0, yvec0;
+
+MUL_DY	yvec0, yvec3, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+SHUF_DY	$0x03, yvec3, yvec3, yvec5;
+
+MUL_DY	yvec0, yvec5, yvec6;
+ADD2_DY	yvec6, yvec13, yvec13;
+ADDQ	$4*SIZE, ptrba;
+ADDQ	$4*SIZE, ptrbb;
+
+.L223_loopE:
+#### Handle ####
+XOR_DY	yvec7, yvec7, yvec7;
+#if  defined(RN) || defined(RT) || defined(CN) || defined(CT)
+ADDSUB_DY	yvec15, yvec7, yvec15;
+ADDSUB_DY	yvec13, yvec7, yvec13;
+#elif defined(NR) || defined(NC) || defined(TR) || defined(TC)
+SUB_DY	yvec15, yvec7, yvec15;
+SUB_DY	yvec13, yvec7, yvec13;
+#elif defined(RR) || defined(RC) || defined(CR) || defined(CC) 
+VPERMILP_DY $0x05, yvec15, yvec15;
+VPERMILP_DY $0x05, yvec13, yvec13;
+ADDSUB_DY	yvec15, yvec7, yvec15;
+ADDSUB_DY	yvec13, yvec7, yvec13;
+VPERMILP_DY $0x05, yvec15, yvec15;
+VPERMILP_DY $0x05, yvec13, yvec13;
+#endif
+
+#### Load Alpha ####
+BROAD_DY MEMALPHA_R,yvec7;
+BROAD_DY MEMALPHA_I,yvec6;
+#### Multiply Alpha	####
+VPERMILP_DY	$0x05, yvec15, yvec5;
+MUL_DY	yvec7, yvec15, yvec15;
+MUL_DY	yvec6, yvec5, yvec5;
+ADD2_DY	yvec5, yvec15, yvec15;
+VPERMILP_DY	$0x05, yvec13, yvec3;
+MUL_DY	yvec7, yvec13, yvec13;
+MUL_DY 	yvec6, yvec3, yvec3;
+ADD2_DY	yvec3, yvec13, yvec13;
+EXTRA_DY $1, yvec15, xvec7;
+EXTRA_DY $1, yvec13, xvec5;
+#### Write back ####
+#ifndef	TRMMKERNEL
+LDL_DX	0*SIZE(C0), xvec0;
+LDH_DX	1*SIZE(C0), xvec0;
+LDL_DX	2*SIZE(C1), xvec1;
+LDH_DX	3*SIZE(C1), xvec1;
+LDL_DX	0*SIZE(C1), xvec2;
+LDH_DX	1*SIZE(C1), xvec2;
+LDL_DX	2*SIZE(C0), xvec3;
+LDH_DX	3*SIZE(C0), xvec3;
+ADD_DX	xvec0, xvec15;
+ADD_DX	xvec1, xvec7;
+ADD_DX	xvec2, xvec13;
+ADD_DX	xvec3, xvec5;
+#endif
+STL_DX	xvec15, 0*SIZE(C0);
+STH_DX	xvec15, 1*SIZE(C0);
+STL_DX	xvec7, 2*SIZE(C1);
+STH_DX	xvec7, 3*SIZE(C1);
+STL_DX	xvec13, 0*SIZE(C1);
+STH_DX	xvec13, 1*SIZE(C1);
+STL_DX	xvec5, 2*SIZE(C0);
+STH_DX	xvec5, 3*SIZE(C0);
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA))||(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kkk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+LEAQ (ptrba, %rax, 2), ptrba;
+LEAQ (ptrbb, %rax, 2), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT) 
+ADDQ	$2, kk;
+#endif
+ADDQ	$4*SIZE, C0;
+ADDQ	$4*SIZE, C1;
+
+.L22_loopE:
+TEST $1, bm;
+JLE .L23_loopE;
+.align 32
+.L23_bodyB:
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb,ptrbb;
+#else
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+ADDQ %rax, ptrba;
+LEAQ (ptrbb, %rax, 2), ptrbb;
+#endif
+XOR_DY	yvec15, yvec15, yvec15;
+#ifndef	TRMMKERNEL
+MOVQ bk,k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else
+MOVQ kk, %rax;
+#ifdef	LEFT
+ADDQ $1, %rax;
+#else 
+ADDQ $2, %rax;
+#endif
+MOVQ %rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L231_loopE;
+.align 32
+.L231_bodyB:
+LD_DY	0*SIZE(ptrba), yvec0;	#### A1r A1i A2r A2i
+EDUP_DY	0*SIZE(ptrbb), yvec2;
+
+SHUF_DY	$0x20, yvec0, yvec0, yvec1;
+MUL_DY	yvec1, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+
+VPERMILP_DY	$0x05, yvec1, yvec4;
+EDUP_DY	1*SIZE(ptrbb), yvec2;
+MUL_DY	yvec4, yvec2, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+
+SHUF_DY	$0x31, yvec0, yvec0, yvec1;
+EDUP_DY	4*SIZE(ptrbb), yvec2;
+
+MUL_DY	yvec1, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+
+VPERMILP_DY	$0x05, yvec1, yvec4;
+EDUP_DY	5*SIZE(ptrbb), yvec2;
+MUL_DY	yvec4, yvec2, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+
+LD_DY	4*SIZE(ptrba), yvec0;
+EDUP_DY	8*SIZE(ptrbb), yvec2;
+
+SHUF_DY	$0x20, yvec0, yvec0, yvec1;
+MUL_DY	yvec1, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+
+VPERMILP_DY	$0x05, yvec1, yvec4;
+EDUP_DY	9*SIZE(ptrbb), yvec2;
+MUL_DY	yvec4, yvec2, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+
+SHUF_DY	$0x31, yvec0, yvec0, yvec1;
+EDUP_DY	12*SIZE(ptrbb), yvec2;
+MUL_DY	yvec1, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+
+VPERMILP_DY	$0x05, yvec1, yvec4;
+EDUP_DY	13*SIZE(ptrbb), yvec2;
+MUL_DY	yvec4, yvec2, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+ADDQ	$8*SIZE, ptrba;
+ADDQ	$16*SIZE, ptrbb;
+DECQ	k;
+JG .L231_bodyB;
+.align 32
+.L231_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else
+TEST $2, kkk;
+#endif
+JLE .L232_loopE;
+.align 32
+.L232_bodyB:
+LD_DY	0*SIZE(ptrba), yvec0;	#### A1r A1i A2r A2i
+EDUP_DY	0*SIZE(ptrbb), yvec2;
+
+SHUF_DY	$0x20, yvec0, yvec0, yvec1;
+MUL_DY	yvec1, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+
+VPERMILP_DY	$0x05, yvec1, yvec4;
+EDUP_DY	1*SIZE(ptrbb), yvec2;
+MUL_DY	yvec4, yvec2, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+
+SHUF_DY	$0x31, yvec0, yvec0, yvec1;
+EDUP_DY	4*SIZE(ptrbb), yvec2;
+
+MUL_DY	yvec1, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+
+VPERMILP_DY	$0x05, yvec1, yvec4;
+EDUP_DY	5*SIZE(ptrbb), yvec2;
+MUL_DY	yvec4, yvec2, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+ADDQ	$4*SIZE, ptrba;
+ADDQ	$8*SIZE, ptrbb;
+
+.L232_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else
+TEST $1, kkk;
+#endif
+JLE .L233_loopE;
+.align 32
+.L233_bodyB:
+LD_DY	0*SIZE(ptrba), yvec0;	#### A1r A1i A2r A2i
+EDUP_DY	0*SIZE(ptrbb), yvec2;
+
+SHUF_DY	$0x20, yvec0, yvec0, yvec1;
+MUL_DY	yvec1, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+
+VPERMILP_DY	$0x05, yvec1, yvec4;
+EDUP_DY	1*SIZE(ptrbb), yvec2;
+MUL_DY	yvec4, yvec2, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+ADDQ	$2*SIZE, ptrba;
+ADDQ	$4*SIZE, ptrbb;
+
+.L233_loopE:
+#### Handle ####
+XOR_DY	yvec7, yvec7, yvec7;
+#if  defined(RN) || defined(RT) || defined(CN) || defined(CT)
+ADDSUB_DY	yvec15, yvec7, yvec15;
+#elif defined(NR) || defined(NC) || defined(TR) || defined(TC)
+SUB_DY	yvec15, yvec7, yvec15;
+#elif defined(RR) || defined(RC) || defined(CR) || defined(CC) 
+VPERMILP_DY $0x05, yvec15, yvec15;
+ADDSUB_DY	yvec15, yvec7, yvec15;
+VPERMILP_DY $0x05, yvec15, yvec15;
+#endif
+
+#### Multiply Alpha	####
+BROAD_DY MEMALPHA_R, yvec7;
+BROAD_DY MEMALPHA_I, yvec6;
+#### Writing Back ####
+VPERMILP_DY	$0x05, yvec15, yvec5;
+MUL_DY	yvec7, yvec15, yvec15;
+MUL_DY	yvec6, yvec5, yvec5;
+ADD2_DY	yvec5, yvec15, yvec15;
+EXTRA_DY $1, yvec15, xvec7;
+#### Writing Back ####
+#ifndef	TRMMKERNEL
+LDL_DX	0*SIZE(C0), xvec0;
+LDH_DX	1*SIZE(C0), xvec0;
+LDL_DX	0*SIZE(C1), xvec1;
+LDH_DX	1*SIZE(C1), xvec1;
+ADD_DX	xvec0, xvec15;
+ADD_DX	xvec1, xvec7;
+#endif
+STL_DX	xvec15, 0*SIZE(C0);
+STH_DX	xvec15, 1*SIZE(C0);
+STL_DX	xvec7, 0*SIZE(C1);
+STH_DX	xvec7, 1*SIZE(C1);
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA))||(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kkk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+ADDQ %rax, ptrba;
+LEAQ (ptrbb, %rax, 2), ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT) 
+ADDQ	$1, kk;
+#endif
+ADDQ	$2*SIZE, C0;
+ADDQ	$2*SIZE, C0;
+.L23_loopE:
+#if defined(TRMMKERNEL) && !defined(LEFT)
+ADDQ	$2, kk;
+#endif
+MOVQ bk, k;
+SALQ $5, k;
+ADDQ k, bb;
+LEAQ (C, ldc, 2), C;
+.L20_loopE:
+TEST $1, bn;
+JLE .L30_loopE;
+.align 32
+.L30_bodyB:
+#if defined(TRMMKERNEL) && defined(LEFT)
+MOVQ OFFSET, %rax;
+MOVQ %rax, kk;
+#endif
+MOVQ ba, ptrba;
+MOVQ C, C0;
+MOVQ bm, i;
+SARQ $2, i;
+JLE .L31_loopE;
+.align 32
+.L31_bodyB:
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb,ptrbb;
+#else
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+LEAQ (ptrba, %rax, 4), ptrba;
+ADDQ %rax, ptrbb;
+#endif
+XOR_DY	yvec15, yvec15, yvec15;
+XOR_DY	yvec14, yvec14, yvec14;
+#ifndef	TRMMKERNEL
+MOVQ bk,k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else
+MOVQ kk, %rax;
+#ifdef	LEFT
+ADDQ $4, %rax;
+#else 
+ADDQ $1, %rax;
+#endif
+MOVQ %rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L311_loopE;
+.align 32
+.L311_bodyB:
+LD_DY	0*SIZE(ptrba), yvec0;
+BROAD_DY 0*SIZE(ptrbb), yvec2;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+
+LD_DY	4*SIZE(ptrba), yvec1;
+MUL_DY	yvec1, yvec2, yvec7;
+ADD1_DY	yvec7, yvec14, yvec14;
+
+VPERMILP_DY	$0x05, yvec0, yvec4;
+BROAD_DY 1*SIZE(ptrbb), yvec3;
+MUL_DY	yvec4, yvec3, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+
+VPERMILP_DY	$0x05, yvec1, yvec5;
+MUL_DY	yvec5, yvec3, yvec7;
+ADD2_DY	yvec7, yvec14, yvec14;
+
+LD_DY	8*SIZE(ptrba), yvec0;
+BROAD_DY 2*SIZE(ptrbb), yvec2;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+
+LD_DY	12*SIZE(ptrba), yvec1;
+MUL_DY	yvec1, yvec2, yvec7;
+ADD1_DY	yvec7, yvec14, yvec14;
+
+VPERMILP_DY	$0x05, yvec0, yvec4;
+BROAD_DY 3*SIZE(ptrbb), yvec3;
+MUL_DY	yvec4, yvec3, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+
+VPERMILP_DY	$0x05, yvec1, yvec5;
+MUL_DY	yvec5, yvec3, yvec7;
+ADD2_DY	yvec7, yvec14, yvec14;
+
+LD_DY	16*SIZE(ptrba), yvec0;
+BROAD_DY 4*SIZE(ptrbb), yvec2;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+
+LD_DY	20*SIZE(ptrba), yvec1;
+MUL_DY	yvec1, yvec2, yvec7;
+ADD1_DY	yvec7, yvec14, yvec14;
+
+VPERMILP_DY	$0x05, yvec0, yvec4;
+BROAD_DY	5*SIZE(ptrbb), yvec3;
+MUL_DY	yvec4, yvec3, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+
+VPERMILP_DY	$0x05, yvec1, yvec5;
+MUL_DY	yvec5, yvec3, yvec7;
+ADD2_DY	yvec7, yvec14, yvec14;
+
+LD_DY	24*SIZE(ptrba), yvec0;
+BROAD_DY 6*SIZE(ptrbb), yvec2;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+
+LD_DY	28*SIZE(ptrba), yvec1;
+MUL_DY	yvec1, yvec2, yvec7;
+ADD1_DY	yvec7, yvec14, yvec14;
+
+VPERMILP_DY	$0x05, yvec0, yvec4;
+BROAD_DY 7*SIZE(ptrbb), yvec3;
+MUL_DY	yvec4, yvec3, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+
+VPERMILP_DY	$0x05, yvec1, yvec5;
+MUL_DY	yvec5, yvec3, yvec7;
+ADD2_DY	yvec7, yvec14, yvec14;
+ADDQ	$32*SIZE, ptrba;
+ADDQ	$8*SIZE, ptrbb;
+DECQ	k;
+JG .L311_bodyB;
+.align 32
+.L311_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else
+TEST $2, kkk;
+#endif
+JLE .L312_loopE;
+.align 32
+.L312_bodyB:
+LD_DY	0*SIZE(ptrba), yvec0;
+BROAD_DY 0*SIZE(ptrbb), yvec2;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+
+LD_DY	4*SIZE(ptrba), yvec1;
+MUL_DY	yvec1, yvec2, yvec7;
+ADD1_DY	yvec7, yvec14, yvec14;
+
+VPERMILP_DY	$0x05, yvec0, yvec4;
+BROAD_DY 1*SIZE(ptrbb), yvec3;
+MUL_DY	yvec4, yvec3, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+
+VPERMILP_DY	$0x05, yvec1, yvec5;
+MUL_DY	yvec5, yvec3, yvec7;
+ADD2_DY	yvec7, yvec14, yvec14;
+
+LD_DY	8*SIZE(ptrba), yvec0;
+BROAD_DY 2*SIZE(ptrbb), yvec2;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+
+LD_DY	12*SIZE(ptrba), yvec1;
+MUL_DY	yvec1, yvec2, yvec7;
+ADD1_DY	yvec7, yvec14, yvec14;
+
+VPERMILP_DY	$0x05, yvec0, yvec4;
+BROAD_DY 3*SIZE(ptrbb), yvec3;
+MUL_DY	yvec4, yvec3, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+
+VPERMILP_DY	$0x05, yvec1, yvec5;
+MUL_DY	yvec5, yvec3, yvec7;
+ADD2_DY	yvec7, yvec14, yvec14;
+ADDQ	$16*SIZE, ptrba;
+ADDQ	$4*SIZE, ptrbb;
+
+.L312_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else
+TEST $1, kkk;
+#endif
+JLE .L313_loopE;
+.align 32
+.L313_bodyB:
+LD_DY	0*SIZE(ptrba), yvec0;
+BROAD_DY 0*SIZE(ptrbb), yvec2;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+
+LD_DY	4*SIZE(ptrba), yvec1;
+MUL_DY	yvec1, yvec2, yvec7;
+ADD1_DY	yvec7, yvec14, yvec14;
+
+VPERMILP_DY	$0x05, yvec0, yvec4;
+BROAD_DY 1*SIZE(ptrbb), yvec3;
+MUL_DY	yvec4, yvec3, yvec6;
+ADD2_DY	yvec6, yvec15, yvec15;
+
+VPERMILP_DY	$0x05, yvec1, yvec5;
+MUL_DY	yvec5, yvec3, yvec7;
+ADD2_DY	yvec7, yvec14, yvec14;
+ADDQ	$8*SIZE, ptrba;
+ADDQ	$2*SIZE, ptrbb;
+
+.L313_loopE:
+#### Handle ####
+XOR_DY	yvec7, yvec7, yvec7;
+#if  defined(RN) || defined(RT) || defined(CN) || defined(CT)
+ADDSUB_DY	yvec15, yvec7, yvec15;
+ADDSUB_DY	yvec14, yvec7, yvec14;
+#elif defined(NR) || defined(NC) || defined(TR) || defined(TC)
+SUB_DY	yvec15, yvec7, yvec15;
+SUB_DY	yvec14, yvec7, yvec14;
+#elif defined(RR) || defined(RC) || defined(CR) || defined(CC) 
+VPERMILP_DY $0x05, yvec15, yvec15;
+VPERMILP_DY $0x05, yvec14, yvec14;
+ADDSUB_DY	yvec15, yvec7, yvec15;
+ADDSUB_DY	yvec14, yvec7, yvec14;
+VPERMILP_DY $0x05, yvec15, yvec15;
+VPERMILP_DY $0x05, yvec14, yvec14;
+#endif
+
+#### Load Alpha ####
+BROAD_DY MEMALPHA_R,yvec7;
+BROAD_DY MEMALPHA_I,yvec6;
+#### Multiply Alpha	####
+VPERMILP_DY	$0x05, yvec15, yvec5;
+MUL_DY	yvec7, yvec15, yvec15;
+MUL_DY	yvec6, yvec5, yvec5;
+ADD2_DY	yvec5, yvec15, yvec15;
+VPERMILP_DY	$0x05, yvec14, yvec4;
+MUL_DY 	yvec7, yvec14, yvec14;
+MUL_DY 	yvec6, yvec4, yvec4;
+ADD2_DY	yvec4, yvec14, yvec14;
+EXTRA_DY $1, yvec15, xvec7;
+EXTRA_DY $1, yvec14, xvec6;
+#### Writing Back ####
+#ifndef	TRMMKERNEL
+LDL_DX	0*SIZE(C0), xvec0;
+LDH_DX	1*SIZE(C0), xvec0;
+LDL_DX	2*SIZE(C0), xvec1;
+LDH_DX	3*SIZE(C0), xvec1;
+LDL_DX	4*SIZE(C0), xvec2;
+LDH_DX	5*SIZE(C0), xvec2;
+LDL_DX	6*SIZE(C0), xvec3;
+LDH_DX	7*SIZE(C0), xvec3;
+ADD_DX	xvec0, xvec15;
+ADD_DX	xvec1, xvec7;
+ADD_DX	xvec2, xvec14;
+ADD_DX	xvec3, xvec6;
+#endif
+STL_DX	xvec15, 0*SIZE(C0);
+STH_DX	xvec15, 1*SIZE(C0);
+STL_DX	xvec7, 2*SIZE(C0);
+STH_DX	xvec7, 3*SIZE(C0);
+STL_DX	xvec14, 4*SIZE(C0);
+STH_DX	xvec14, 5*SIZE(C0);
+STL_DX	xvec6, 6*SIZE(C0);
+STH_DX	xvec6, 7*SIZE(C0);
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA))||(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kkk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+LEAQ (ptrba, %rax, 4), ptrba;
+ADDQ %rax, ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT) 
+ADDQ	$4, kk;
+#endif
+ADDQ	$8*SIZE, C0;
+DECQ	i;
+JG .L31_bodyB;
+.align 32
+.L31_loopE:
+TEST $2, bm;
+JLE .L32_loopE;
+.align 32
+.L32_bodyB:
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb,ptrbb;
+#else
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+LEAQ (ptrba, %rax, 2), ptrba;
+ADDQ %rax, ptrbb;
+#endif
+XOR_DY yvec15, yvec15, yvec15;
+#ifndef	TRMMKERNEL
+MOVQ bk,k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else
+MOVQ kk, %rax;
+#ifdef	LEFT
+ADDQ $2, %rax;
+#else 
+ADDQ $1, %rax;
+#endif
+MOVQ %rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L321_loopE;
+.align 32
+.L321_bodyB:
+LD_DY	0*SIZE(ptrba), yvec0;
+BROAD_DY 0*SIZE(ptrbb), yvec2;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+VPERMILP_DY	$0x05, yvec0, yvec1;
+BROAD_DY 1*SIZE(ptrbb), yvec3;
+MUL_DY	yvec1, yvec3, yvec7;
+ADD2_DY	yvec7, yvec15, yvec15;
+
+LD_DY	4*SIZE(ptrba), yvec0;
+BROAD_DY 2*SIZE(ptrbb), yvec2;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+VPERMILP_DY	$0x05, yvec0, yvec1;
+BROAD_DY 3*SIZE(ptrbb), yvec3;
+MUL_DY	yvec1, yvec3, yvec7;
+ADD2_DY	yvec7, yvec15, yvec15;
+
+LD_DY	8*SIZE(ptrba), yvec0;
+BROAD_DY 4*SIZE(ptrbb), yvec2;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+VPERMILP_DY	$0x05, yvec0, yvec1;
+BROAD_DY 5*SIZE(ptrbb), yvec3;
+MUL_DY	yvec1, yvec3, yvec7;
+ADD2_DY	yvec7, yvec15, yvec15;
+
+LD_DY	12*SIZE(ptrba), yvec0;
+BROAD_DY 6*SIZE(ptrbb), yvec2;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+VPERMILP_DY	$0x05, yvec0, yvec1;
+BROAD_DY 7*SIZE(ptrbb), yvec3;
+MUL_DY	yvec1, yvec3, yvec7;
+ADD2_DY	yvec7, yvec15, yvec15;
+ADDQ	$16*SIZE, ptrba;
+ADDQ	$8*SIZE, ptrbb;
+DECQ	k;
+JG .L321_bodyB;
+.align 32
+.L321_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else
+TEST $2, kkk;
+#endif
+JLE .L322_loopE;
+.align 32
+.L322_bodyB:
+LD_DY	0*SIZE(ptrba), yvec0;
+BROAD_DY 0*SIZE(ptrbb), yvec2;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+VPERMILP_DY	$0x05, yvec0, yvec1;
+BROAD_DY 1*SIZE(ptrbb), yvec3;
+MUL_DY	yvec1, yvec3, yvec7;
+ADD2_DY	yvec7, yvec15, yvec15;
+
+LD_DY	4*SIZE(ptrba), yvec0;
+BROAD_DY 2*SIZE(ptrbb), yvec2;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+VPERMILP_DY	$0x05, yvec0, yvec1;
+BROAD_DY 3*SIZE(ptrbb), yvec3;
+MUL_DY	yvec1, yvec3, yvec7;
+ADD2_DY	yvec7, yvec15, yvec15;
+ADDQ	$8*SIZE, ptrba;
+ADDQ	$4*SIZE, ptrbb;
+
+.L322_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else
+TEST $1, kkk;
+#endif
+JLE .L323_loopE;
+.align 32
+.L323_bodyB:
+LD_DY	0*SIZE(ptrba), yvec0;
+BROAD_DY 0*SIZE(ptrbb), yvec2;
+MUL_DY	yvec0, yvec2, yvec6;
+ADD1_DY	yvec6, yvec15, yvec15;
+VPERMILP_DY	$0x05, yvec0, yvec1;
+BROAD_DY 1*SIZE(ptrbb), yvec3;
+MUL_DY	yvec1, yvec3, yvec7;
+ADD2_DY	yvec7, yvec15, yvec15;
+ADDQ	$4*SIZE, ptrba;
+ADDQ	$2*SIZE, ptrbb;
+.L323_loopE:
+#### Handle ####
+XOR_DY	yvec7, yvec7, yvec7;
+#if  defined(RN) || defined(RT) || defined(CN) || defined(CT)
+ADDSUB_DY	yvec15, yvec7, yvec15;
+#elif defined(NR) || defined(NC) || defined(TR) || defined(TC)
+SUB_DY	yvec15, yvec7, yvec15;
+#elif defined(RR) || defined(RC) || defined(CR) || defined(CC) 
+VPERMILP_DY $0x05, yvec15, yvec15;
+ADDSUB_DY	yvec15, yvec7, yvec15;
+VPERMILP_DY $0x05, yvec15, yvec15;
+#endif
+
+#### Load Alpha ####
+BROAD_DY MEMALPHA_R,yvec7;
+BROAD_DY MEMALPHA_I,yvec6;
+#### Multiply Alpha	####
+VPERMILP_DY	$0x05, yvec15, yvec5;
+MUL_DY	yvec7, yvec15, yvec15;
+MUL_DY	yvec6, yvec5, yvec5;
+ADD2_DY	yvec5, yvec15, yvec15;
+EXTRA_DY $1, yvec15, xvec7;
+#### Writing Back ####
+#ifndef	TRMMKERNEL
+LDL_DX	0*SIZE(C0), xvec0;
+LDH_DX	1*SIZE(C0), xvec0;
+LDL_DX	2*SIZE(C0), xvec1;
+LDH_DX	3*SIZE(C0), xvec1;
+ADD_DX	xvec0, xvec15;
+ADD_DX	xvec1, xvec7;
+#endif
+STL_DX	xvec15, 0*SIZE(C0);
+STH_DX	xvec15, 1*SIZE(C0);
+STL_DX	xvec7, 2*SIZE(C0);
+STH_DX	xvec7, 3*SIZE(C0);
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA))||(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kkk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+LEAQ (ptrba, %rax, 2), ptrba;
+ADDQ %rax, ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT) 
+ADDQ	$2, kk;
+#endif
+ADDQ	$4*SIZE, C0;
+.L32_loopE:
+TEST $1, bm;
+JLE .L33_loopE;
+.align 32
+.L33_bodyB:
+#if !defined(TRMMKERNEL)||(defined(TRMMKERNEL)&&defined(LEFT)&&defined(TRANSA))||(defined(TRMMKERNEL)&&!defined(LEFT)&&!defined(TRANSA))
+MOVQ bb,ptrbb;
+#else
+MOVQ bb, ptrbb;
+MOVQ kk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+ADDQ %rax, ptrba;
+ADDQ %rax, ptrbb;
+#endif
+XOR_DY yvec15, yvec15, yvec15;
+#ifndef	TRMMKERNEL
+MOVQ bk,k;
+#elif (defined(LEFT)&&!defined(TRANSA))||(!defined(LEFT)&&defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kk, %rax;
+MOVQ %rax, kkk;
+#else
+MOVQ kk, %rax;
+#ifdef	LEFT
+ADDQ $1, %rax;
+#else 
+ADDQ $1, %rax;
+#endif
+MOVQ %rax, kkk;
+#endif
+SARQ $2, k;
+JLE .L331_loopE;
+.align 32
+.L331_bodyB:
+LD_DX	0*SIZE(ptrba), xvec0;
+BROAD_DX 0*SIZE(ptrbb), xvec2;
+MUL_DX	xvec0, xvec2;
+ADD1_DX	xvec2, xvec15;
+
+SHUF_DX	$0x4e, xvec0, xvec1;
+BROAD_DX 1*SIZE(ptrbb), xvec3;
+MUL_DX	xvec1, xvec3;
+ADDSUB_DX xvec3, xvec15;
+
+LD_DX	2*SIZE(ptrba), xvec0;
+BROAD_DX 2*SIZE(ptrbb), xvec2;
+MUL_DX	xvec0, xvec2;
+ADD1_DX	xvec2, xvec15;
+
+SHUF_DX	$0x4e, xvec0, xvec1;
+BROAD_DX 3*SIZE(ptrbb), xvec3;
+MUL_DX	xvec1, xvec3;
+ADDSUB_DX xvec3, xvec15;
+
+LD_DX	4*SIZE(ptrba), xvec0;
+BROAD_DX 4*SIZE(ptrbb), xvec2;
+MUL_DX	xvec0, xvec2;
+ADD1_DX	xvec2, xvec15;
+
+SHUF_DX	$0x4e, xvec0, xvec1;
+BROAD_DX 5*SIZE(ptrbb), xvec3;
+MUL_DX	xvec1, xvec3;
+ADDSUB_DX xvec3, xvec15;
+
+LD_DX	6*SIZE(ptrba), xvec0;
+BROAD_DX 6*SIZE(ptrbb), xvec2;
+MUL_DX	xvec0, xvec2;
+ADD1_DX	xvec2, xvec15;
+
+SHUF_DX	$0x4e, xvec0, xvec1;
+BROAD_DX 7*SIZE(ptrbb), xvec3;
+MUL_DX	xvec1, xvec3;
+ADDSUB_DX xvec3, xvec15;
+ADDQ	$8*SIZE, ptrba;
+ADDQ	$8*SIZE, ptrbb;
+DECQ	k;
+JG .L331_bodyB;
+.align 32
+.L331_loopE:
+#ifndef	TRMMKERNEL
+TEST $2, bk;
+#else
+TEST $2, kkk;
+#endif
+JLE .L332_loopE;
+.align 32
+.L332_bodyB:
+LD_DX	0*SIZE(ptrba), xvec0;
+BROAD_DX 0*SIZE(ptrbb), xvec2;
+MUL_DX	xvec0, xvec2;
+ADD1_DX	xvec2, xvec15;
+
+SHUF_DX	$0x4e, xvec0, xvec1;
+BROAD_DX 1*SIZE(ptrbb), xvec3;
+MUL_DX	xvec1, xvec3;
+ADDSUB_DX xvec3, xvec15;
+
+LD_DX	2*SIZE(ptrba), xvec0;
+BROAD_DX 2*SIZE(ptrbb), xvec2;
+MUL_DX	xvec0, xvec2;
+ADD1_DX	xvec2, xvec15;
+
+SHUF_DX	$0x4e, xvec0, xvec1;
+BROAD_DX 3*SIZE(ptrbb), xvec3;
+MUL_DX	xvec1, xvec3;
+ADDSUB_DX xvec3, xvec15;
+ADDQ	$4*SIZE, ptrba;
+ADDQ	$4*SIZE, ptrbb;
+
+.L332_loopE:
+#ifndef	TRMMKERNEL
+TEST $1, bk;
+#else
+TEST $1, kkk;
+#endif
+JLE .L333_loopE;
+.align 32
+.L333_bodyB:
+LD_DX	0*SIZE(ptrba), xvec0;
+BROAD_DX 0*SIZE(ptrbb), xvec2;
+MUL_DX	xvec0, xvec2;
+ADD1_DX	xvec2, xvec15;
+
+SHUF_DX	$0x4e, xvec0, xvec1;
+BROAD_DX 1*SIZE(ptrbb), xvec3;
+MUL_DX	xvec1, xvec3;
+ADDSUB_DX xvec3, xvec15;
+ADDQ	$2*SIZE, ptrba;
+ADDQ	$2*SIZE, ptrbb;
+
+.L333_loopE:
+#### Handle ####
+XOR_DY	yvec7, yvec7, yvec7;
+#if  defined(RN) || defined(RT) || defined(CN) || defined(CT)
+ADDSUB_DX   xvec15, xvec7;
+MOV_DX	xvec7, xvec15;
+#elif defined(NR) || defined(NC) || defined(TR) || defined(TC)
+SUB_DX  xvec15, xvec7;
+MOV_DX	xvec7, xvec15;
+#elif defined(RR) || defined(RC) || defined(CR) || defined(CC)
+SHUF_DX $0x4e, xvec15, xvec15;
+ADDSUB_DX   xvec15, xvec7;
+MOV_DX	xvec7, xvec15;
+SHUF_DX $0x4e, xvec15, xvec15;
+#endif
+
+#### Load Alpha ####
+BROAD_DX MEMALPHA_R,xvec7;
+BROAD_DX MEMALPHA_I,xvec6;
+#### Multiply Alpha	####
+SHUF_DX	$0x4e, xvec15, xvec5;
+MUL_DX	xvec7, xvec15;
+MUL_DX	xvec6, xvec5;
+ADDSUB_DX	xvec5, xvec15;
+#### Writing back ####
+#ifndef	TRMMKERNEL
+LDL_DX	0*SIZE(C0), xvec0;
+LDH_DX	1*SIZE(C0), xvec0;
+ADD_DX	xvec0, xvec15;
+#endif
+STL_DX	xvec15, 0*SIZE(C0);
+STH_DX	xvec15, 1*SIZE(C0);
+#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA))||(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
+MOVQ bk, %rax;
+SUBQ kkk, %rax;
+SALQ $ZBASE_SHIFT, %rax;
+ADDQ %rax, ptrba;
+ADDQ %rax, ptrbb;
+#endif
+#if defined(TRMMKERNEL) && defined(LEFT) 
+ADDQ	$1, kk;
+#endif
+ADDQ	$2*SIZE, C0;
+.L33_loopE:
+#if defined(TRMMKERNEL) && !defined(LEFT)
+ADDQ	$1, kk;
+#endif
+MOVQ bk, k;
+SALQ $4*SIZE, k;
+ADDQ k, bb;
+LEAQ (C, ldc, 1), C;
+.L30_loopE:
+movq      0(%rsp), %rbx;
+movq      8(%rsp), %rbp;
+movq     16(%rsp), %r12;
+movq     24(%rsp), %r13;
+movq     32(%rsp), %r14;
+movq     40(%rsp), %r15;
+
+#ifdef WINDOWS_ABI
+	movq	 48(%rsp), %rdi
+	movq	 56(%rsp), %rsi
+	movups	 64(%rsp), %xmm6
+	movups	 80(%rsp), %xmm7
+	movups	 96(%rsp), %xmm8
+	movups	112(%rsp), %xmm9
+	movups	128(%rsp), %xmm10
+	movups	144(%rsp), %xmm11
+	movups	160(%rsp), %xmm12
+	movups	176(%rsp), %xmm13
+	movups	192(%rsp), %xmm14
+	movups	208(%rsp), %xmm15
+#endif
+
+
+addq    $STACKSIZE, %rsp;
+ret
+
+EPILOGUE
diff --git a/param.h b/param.h
index 3add52615..db9ec6a3b 100644
--- a/param.h
+++ b/param.h
@@ -1,5 +1,5 @@
 /*****************************************************************************
-Copyright (c) 2011,2012 Lab of Parallel Software and Computational Science,ISCAS
+Copyright (c) 2011, Lab of Parallel Software and Computational Science,ICSAS
 All rights reserved.
 
 Redistribution and use in source and binary forms, with or without
@@ -208,68 +208,6 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #endif
 
-#define SGEMM_DEFAULT_R sgemm_r
-#define QGEMM_DEFAULT_R qgemm_r
-#define DGEMM_DEFAULT_R dgemm_r
-#define CGEMM_DEFAULT_R cgemm_r
-#define ZGEMM_DEFAULT_R zgemm_r
-#define XGEMM_DEFAULT_R xgemm_r
-
-#define SYMV_P	16
-#define HAVE_EXCLUSIVE_CACHE
-
-#define GEMM_THREAD gemm_thread_mn
-
-#endif
-
-#if defined(BOBCAT)
-
-#define SNUMOPT		8
-#define DNUMOPT		4
-
-#define GEMM_DEFAULT_OFFSET_A  64
-#define GEMM_DEFAULT_OFFSET_B 832
-#define GEMM_DEFAULT_ALIGN 0x0fffUL
-
-#define SGEMM_DEFAULT_UNROLL_N 4
-#define DGEMM_DEFAULT_UNROLL_N 4
-#define QGEMM_DEFAULT_UNROLL_N 2
-#define CGEMM_DEFAULT_UNROLL_N 2
-#define ZGEMM_DEFAULT_UNROLL_N 2
-#define XGEMM_DEFAULT_UNROLL_N 1
-
-#ifdef ARCH_X86
-#define SGEMM_DEFAULT_UNROLL_M 4
-#define DGEMM_DEFAULT_UNROLL_M 2
-#define QGEMM_DEFAULT_UNROLL_M 2
-#define CGEMM_DEFAULT_UNROLL_M 2
-#define ZGEMM_DEFAULT_UNROLL_M 1
-#define XGEMM_DEFAULT_UNROLL_M 1
-#else
-#define SGEMM_DEFAULT_UNROLL_M 8
-#define DGEMM_DEFAULT_UNROLL_M 4
-#define QGEMM_DEFAULT_UNROLL_M 2
-#define CGEMM_DEFAULT_UNROLL_M 4
-#define ZGEMM_DEFAULT_UNROLL_M 2
-#define XGEMM_DEFAULT_UNROLL_M 1
-#endif
-
-
-#define SGEMM_DEFAULT_P 448
-#define DGEMM_DEFAULT_P 224
-#define QGEMM_DEFAULT_P 112
-#define CGEMM_DEFAULT_P 224
-#define ZGEMM_DEFAULT_P 112 
-#define XGEMM_DEFAULT_P  56
-
-#define SGEMM_DEFAULT_Q 224
-#define DGEMM_DEFAULT_Q 224
-#define QGEMM_DEFAULT_Q 224
-#define CGEMM_DEFAULT_Q 224
-#define ZGEMM_DEFAULT_Q 224
-#define XGEMM_DEFAULT_Q 224
-
-
 #define SGEMM_DEFAULT_R sgemm_r
 #define QGEMM_DEFAULT_R qgemm_r
 #define DGEMM_DEFAULT_R dgemm_r
@@ -980,7 +918,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 #define SNUMOPT		8
 #define DNUMOPT		4
 
-#define GEMM_DEFAULT_OFFSET_A    32
+#define GEMM_DEFAULT_OFFSET_A	  0
 #define GEMM_DEFAULT_OFFSET_B     0
 #define GEMM_DEFAULT_ALIGN 0x03fffUL
 
@@ -990,57 +928,61 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #ifdef ARCH_X86
 #define SGEMM_DEFAULT_UNROLL_M 4
-#define DGEMM_DEFAULT_UNROLL_M 2
+#define DGEMM_DEFAULT_UNROLL_M 8
 #define QGEMM_DEFAULT_UNROLL_M 2
 #define CGEMM_DEFAULT_UNROLL_M 2
 #define ZGEMM_DEFAULT_UNROLL_M 1
 #define XGEMM_DEFAULT_UNROLL_M 1
 
 #define SGEMM_DEFAULT_UNROLL_N 4
-#define DGEMM_DEFAULT_UNROLL_N 4
+#define DGEMM_DEFAULT_UNROLL_N 8
 #define QGEMM_DEFAULT_UNROLL_N 2
 #define CGEMM_DEFAULT_UNROLL_N 2
 #define ZGEMM_DEFAULT_UNROLL_N 2
 #define XGEMM_DEFAULT_UNROLL_N 1
 #else
-#define SGEMM_DEFAULT_UNROLL_M 4
-#define DGEMM_DEFAULT_UNROLL_M 2
+#define SGEMM_DEFAULT_UNROLL_M 8
+#define DGEMM_DEFAULT_UNROLL_M 8
 #define QGEMM_DEFAULT_UNROLL_M 2
-#define CGEMM_DEFAULT_UNROLL_M 2
-#define ZGEMM_DEFAULT_UNROLL_M 1
+#define CGEMM_DEFAULT_UNROLL_M 8
+#define ZGEMM_DEFAULT_UNROLL_M 4
 #define XGEMM_DEFAULT_UNROLL_M 1
 
 #define SGEMM_DEFAULT_UNROLL_N 8
-#define DGEMM_DEFAULT_UNROLL_N 8
+#define DGEMM_DEFAULT_UNROLL_N 4
 #define QGEMM_DEFAULT_UNROLL_N 2
 #define CGEMM_DEFAULT_UNROLL_N 4
 #define ZGEMM_DEFAULT_UNROLL_N 4
 #define XGEMM_DEFAULT_UNROLL_N 1
 #endif
 
-#define SGEMM_DEFAULT_P 504
+#define SGEMM_DEFAULT_P 512
 #define SGEMM_DEFAULT_R sgemm_r
+//#define SGEMM_DEFAULT_R 1024
 
-#define DGEMM_DEFAULT_P 504
+#define DGEMM_DEFAULT_P 512
 #define DGEMM_DEFAULT_R dgemm_r
+//#define DGEMM_DEFAULT_R 1024
 
 #define QGEMM_DEFAULT_P 504
 #define QGEMM_DEFAULT_R qgemm_r
 
-#define CGEMM_DEFAULT_P 252
-#define CGEMM_DEFAULT_R cgemm_r
+#define CGEMM_DEFAULT_P 128
+//#define CGEMM_DEFAULT_R cgemm_r
+#define CGEMM_DEFAULT_R 1024
 
-#define ZGEMM_DEFAULT_P 252
+#define ZGEMM_DEFAULT_P 512
 #define ZGEMM_DEFAULT_R zgemm_r
+//#define ZGEMM_DEFAULT_R 1024
 
 #define XGEMM_DEFAULT_P 252
 #define XGEMM_DEFAULT_R xgemm_r
 
-#define SGEMM_DEFAULT_Q 512
+#define SGEMM_DEFAULT_Q 256
 #define DGEMM_DEFAULT_Q 256
 #define QGEMM_DEFAULT_Q 128
-#define CGEMM_DEFAULT_Q 512
-#define ZGEMM_DEFAULT_Q 256
+#define CGEMM_DEFAULT_Q 256 
+#define ZGEMM_DEFAULT_Q 192
 #define XGEMM_DEFAULT_Q 128
 
 #define GETRF_FACTOR 0.72