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

/*********************************************************************
*
* 2014/06/28 Saar
*        BLASTEST               : OK
*        CTEST                  : OK
*        TEST                   : OK
*
*
* 2013/10/30 Saar
*
* Parameter:
*       UNROLL_M        2
*       UNROLL_N        2
*       ZGEMM_P         384
*       ZGEMM_Q         168
*       A_PR1           512
*       B_PR1           256
*
* Performance at m x n on AMD 8320 (ACML-Version: 5.3.1):
* 
* 3456x3456     82.4    GFLOPS with 8 threads on 4 modules (ACML:  76.3 ) (BULLDOZER:  81.0 )
* 3456x3456     79.9    GFLOPS with 4 threads on 4 modules (ACML:  69.9 ) (BULLDOZER:  74.6 )
* 3456x3456     40.4    GFLOPS with 2 threads on 2 modules (ACML:  35.8 ) (BULLDOZER:  37.9 )
* 3456x3456     20.3    GFLOPS with 1 threads on 1 modules (ACML:  18.1 ) (BULLDOZER:  19.2 )
*
* Performance at m x n on AMD 6380  (ACML-Version: 5.3.1):
* 
* 6912x6912    227.5    GFLOPS with 32 threads on 16 modules (ACML: 166.3 ) (BULLDOZER: 228.5 )
* 6912x6912    211.6    GFLOPS with 16 threads on 16 modules (ACML: 169.5 ) (BULLDOZER: 204.3 )
* 6912x6912    123.5    GFLOPS with  8 threads on  8 modules (ACML:  92.7 ) (BULLDOZER: 117.0 )
* 3456x3456     64.1    GFLOPS with  4 threads on  4 modules (ACML:  49.1 ) (BULLDOZER:  61.7 )
* 3456x3456     33.4    GFLOPS with  2 threads on  2 modules (ACML:  28.1 ) (BULLDOZER:  30.9 )
* 3456x3456     17.0    GFLOPS with  1 threads on  1 modules (ACML:  15.2 ) (BULLDOZER:  15.7 )
*
*********************************************************************/


#define ASSEMBLER
#include "common.h"
 
#define OLD_M	%rdi
#define OLD_N	%rsi
#define M	%r13
#define J	%r14
#define OLD_K	%rdx

#define A	%rcx
#define B	%r8
#define C	%r9
#define LDC	%r10
	
#define I	%r11
#define AO	%rdi
#define BO	%rsi
#define	CO1	%r15
#define K	%r12
#define BI	%rbp
#define	SP	%rbx

#define BO1	%rdi
#define BO2	%r15

#ifndef WINDOWS_ABI

#define STACKSIZE 96

#else

#define STACKSIZE 320

#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)

#endif

#define L_BUFFER_SIZE 256*8*4

#define Ndiv6	 24(%rsp)
#define Nmod6	 32(%rsp)
#define N	 40(%rsp)
#define ALPHA_R  48(%rsp)
#define ALPHA_I  56(%rsp)
#define OFFSET   64(%rsp)
#define KK       72(%rsp)
#define KKK      80(%rsp)
#define BUFFER1	           128(%rsp)

#if defined(OS_WINDOWS)
#if   L_BUFFER_SIZE > 16384
#define STACK_TOUCH \
        movl    $0,  4096 * 4(%rsp);\
        movl    $0,  4096 * 3(%rsp);\
        movl    $0,  4096 * 2(%rsp);\
        movl    $0,  4096 * 1(%rsp);
#elif L_BUFFER_SIZE > 12288
#define STACK_TOUCH \
        movl    $0,  4096 * 3(%rsp);\
        movl    $0,  4096 * 2(%rsp);\
        movl    $0,  4096 * 1(%rsp);
#elif L_BUFFER_SIZE > 8192
#define STACK_TOUCH \
        movl    $0,  4096 * 2(%rsp);\
        movl    $0,  4096 * 1(%rsp);
#elif L_BUFFER_SIZE > 4096
#define STACK_TOUCH \
        movl    $0,  4096 * 1(%rsp);
#else
#define STACK_TOUCH
#endif
#else
#define STACK_TOUCH
#endif


#if   defined(NN) || defined(NT) || defined(TN) || defined(TT)
#define VFMADD_R    vfmaddpd
#define VFMADD_I    vfmaddpd
#elif defined(RN) || defined(RT) || defined(CN) || defined(CT)
#define VFMADD_R    vfnmaddpd
#define VFMADD_I    vfmaddpd
#elif defined(NR) || defined(NC) || defined(TR) || defined(TC)
#define VFMADD_R    vfmaddpd
#define VFMADD_I    vfnmaddpd
#else
#define VFMADD_R    vfnmaddpd
#define VFMADD_I    vfnmaddpd
#endif


#define	A_PR1	512
#define	B_PR1	256

#define KERNEL2x2_1(xx) \
        prefetcht0      A_PR1(AO,%rax,SIZE)        ;\
        vmovups          -8 * SIZE(AO, %rax, SIZE), %xmm0 ;\
        vmovddup         -8 * SIZE(BO, BI, SIZE), %xmm4 ;\
        VFMADD_R        %xmm8,%xmm4,%xmm0,%xmm8 ;\
        vmovups          -6 * SIZE(AO, %rax, SIZE), %xmm1 ;\
        VFMADD_R        %xmm12,%xmm4,%xmm1,%xmm12 ;\
        vmovddup         -7 * SIZE(BO, BI, SIZE), %xmm5 ;\
        VFMADD_I        %xmm9,%xmm5,%xmm0,%xmm9 ;\
        VFMADD_I        %xmm13,%xmm5,%xmm1,%xmm13 ;\
        vmovddup         -6 * SIZE(BO, BI, SIZE), %xmm6 ;\
        VFMADD_R        %xmm10,%xmm6,%xmm0,%xmm10 ;\
        VFMADD_R        %xmm14,%xmm6,%xmm1,%xmm14 ;\
        vmovddup         -5 * SIZE(BO, BI, SIZE), %xmm7 ;\
        VFMADD_I        %xmm11,%xmm7,%xmm0,%xmm11 ;\
        VFMADD_I        %xmm15,%xmm7,%xmm1,%xmm15 ;\

#define KERNEL2x2_2(xx) \
        vmovups          -4 * SIZE(AO, %rax, SIZE), %xmm0 ;\
        vmovddup         -4 * SIZE(BO, BI, SIZE), %xmm4 ;\
        VFMADD_R        %xmm8,%xmm4,%xmm0,%xmm8 ;\
        vmovups          -2 * SIZE(AO, %rax, SIZE), %xmm1 ;\
        VFMADD_R        %xmm12,%xmm4,%xmm1,%xmm12 ;\
        vmovddup         -3 * SIZE(BO, BI, SIZE), %xmm5 ;\
        VFMADD_I        %xmm9,%xmm5,%xmm0,%xmm9 ;\
        VFMADD_I        %xmm13,%xmm5,%xmm1,%xmm13 ;\
        vmovddup         -2 * SIZE(BO, BI, SIZE), %xmm6 ;\
        VFMADD_R        %xmm10,%xmm6,%xmm0,%xmm10 ;\
        VFMADD_R        %xmm14,%xmm6,%xmm1,%xmm14 ;\
        vmovddup         -1 * SIZE(BO, BI, SIZE), %xmm7 ;\
        VFMADD_I        %xmm11,%xmm7,%xmm0,%xmm11 ;\
        VFMADD_I        %xmm15,%xmm7,%xmm1,%xmm15 ;\

#define KERNEL2x2_3(xx) \
	prefetcht0      A_PR1+64(AO,%rax,SIZE)     ;\
        vmovups           0 * SIZE(AO, %rax, SIZE), %xmm0 ;\
        vmovddup          0 * SIZE(BO, BI, SIZE), %xmm4 ;\
        VFMADD_R        %xmm8,%xmm4,%xmm0,%xmm8 ;\
        vmovups           2 * SIZE(AO, %rax, SIZE), %xmm1 ;\
        VFMADD_R        %xmm12,%xmm4,%xmm1,%xmm12 ;\
        vmovddup          1 * SIZE(BO, BI, SIZE), %xmm5 ;\
        VFMADD_I        %xmm9,%xmm5,%xmm0,%xmm9 ;\
        VFMADD_I        %xmm13,%xmm5,%xmm1,%xmm13 ;\
        vmovddup          2 * SIZE(BO, BI, SIZE), %xmm6 ;\
        VFMADD_R        %xmm10,%xmm6,%xmm0,%xmm10 ;\
        VFMADD_R        %xmm14,%xmm6,%xmm1,%xmm14 ;\
        vmovddup          3 * SIZE(BO, BI, SIZE), %xmm7 ;\
        VFMADD_I        %xmm11,%xmm7,%xmm0,%xmm11 ;\
        VFMADD_I        %xmm15,%xmm7,%xmm1,%xmm15 ;\

#define KERNEL2x2_4(xx) \
        vmovups           4 * SIZE(AO, %rax, SIZE), %xmm0 ;\
        vmovddup          4 * SIZE(BO, BI, SIZE), %xmm4 ;\
        VFMADD_R        %xmm8,%xmm4,%xmm0,%xmm8 ;\
        vmovups           6 * SIZE(AO, %rax, SIZE), %xmm1 ;\
        VFMADD_R        %xmm12,%xmm4,%xmm1,%xmm12 ;\
        vmovddup          5 * SIZE(BO, BI, SIZE), %xmm5 ;\
        VFMADD_I        %xmm9,%xmm5,%xmm0,%xmm9 ;\
        VFMADD_I        %xmm13,%xmm5,%xmm1,%xmm13 ;\
        vmovddup          6 * SIZE(BO, BI, SIZE), %xmm6 ;\
        VFMADD_R        %xmm10,%xmm6,%xmm0,%xmm10 ;\
        VFMADD_R        %xmm14,%xmm6,%xmm1,%xmm14 ;\
        vmovddup          7 * SIZE(BO, BI, SIZE), %xmm7 ;\
        VFMADD_I        %xmm11,%xmm7,%xmm0,%xmm11 ;\
        VFMADD_I        %xmm15,%xmm7,%xmm1,%xmm15 ;\
        addq    $16, BI                            ;\
        addq    $16, %rax                          ;\


#define KERNEL2x2_SUB(xx) \
        vmovups          -8 * SIZE(AO, %rax, SIZE), %xmm0 ;\
        vmovddup         -8 * SIZE(BO, BI, SIZE), %xmm4 ;\
        VFMADD_R        %xmm8,%xmm4,%xmm0,%xmm8 ;\
        vmovups          -6 * SIZE(AO, %rax, SIZE), %xmm1 ;\
        VFMADD_R        %xmm12,%xmm4,%xmm1,%xmm12 ;\
        vmovddup         -7 * SIZE(BO, BI, SIZE), %xmm5 ;\
        VFMADD_I        %xmm9,%xmm5,%xmm0,%xmm9 ;\
        VFMADD_I        %xmm13,%xmm5,%xmm1,%xmm13 ;\
        vmovddup         -6 * SIZE(BO, BI, SIZE), %xmm6 ;\
        VFMADD_R        %xmm10,%xmm6,%xmm0,%xmm10 ;\
        VFMADD_R        %xmm14,%xmm6,%xmm1,%xmm14 ;\
        vmovddup         -5 * SIZE(BO, BI, SIZE), %xmm7 ;\
        VFMADD_I        %xmm11,%xmm7,%xmm0,%xmm11 ;\
        VFMADD_I        %xmm15,%xmm7,%xmm1,%xmm15 ;\
        addq    $4, BI                            ;\
        addq    $4, %rax                          ;\

/************************************************************************************************/

#define KERNEL1x2_1(xx) \
        prefetcht0      A_PR1(AO,%rax,SIZE)        ;\
        vmovups          -8 * SIZE(AO, %rax, SIZE), %xmm0 ;\
        vmovddup         -8 * SIZE(BO, BI, SIZE), %xmm4 ;\
        VFMADD_R        %xmm8,%xmm4,%xmm0,%xmm8 ;\
        vmovddup         -7 * SIZE(BO, BI, SIZE), %xmm5 ;\
        VFMADD_I        %xmm9,%xmm5,%xmm0,%xmm9 ;\
        vmovddup         -6 * SIZE(BO, BI, SIZE), %xmm6 ;\
        VFMADD_R        %xmm10,%xmm6,%xmm0,%xmm10 ;\
        vmovddup         -5 * SIZE(BO, BI, SIZE), %xmm7 ;\
        VFMADD_I        %xmm11,%xmm7,%xmm0,%xmm11 ;\

#define KERNEL1x2_2(xx) \
        vmovups          -6 * SIZE(AO, %rax, SIZE), %xmm0 ;\
        vmovddup         -4 * SIZE(BO, BI, SIZE), %xmm4 ;\
        VFMADD_R        %xmm8,%xmm4,%xmm0,%xmm8 ;\
        VFMADD_R        %xmm12,%xmm4,%xmm1,%xmm12 ;\
        vmovddup         -3 * SIZE(BO, BI, SIZE), %xmm5 ;\
        VFMADD_I        %xmm9,%xmm5,%xmm0,%xmm9 ;\
        vmovddup         -2 * SIZE(BO, BI, SIZE), %xmm6 ;\
        VFMADD_R        %xmm10,%xmm6,%xmm0,%xmm10 ;\
        vmovddup         -1 * SIZE(BO, BI, SIZE), %xmm7 ;\
        VFMADD_I        %xmm11,%xmm7,%xmm0,%xmm11 ;\

#define KERNEL1x2_3(xx) \
        vmovups          -4 * SIZE(AO, %rax, SIZE), %xmm0 ;\
        vmovddup          0 * SIZE(BO, BI, SIZE), %xmm4 ;\
        VFMADD_R        %xmm8,%xmm4,%xmm0,%xmm8 ;\
        vmovddup          1 * SIZE(BO, BI, SIZE), %xmm5 ;\
        VFMADD_I        %xmm9,%xmm5,%xmm0,%xmm9 ;\
        vmovddup          2 * SIZE(BO, BI, SIZE), %xmm6 ;\
        VFMADD_R        %xmm10,%xmm6,%xmm0,%xmm10 ;\
        vmovddup          3 * SIZE(BO, BI, SIZE), %xmm7 ;\
        VFMADD_I        %xmm11,%xmm7,%xmm0,%xmm11 ;\

#define KERNEL1x2_4(xx) \
        vmovups          -2 * SIZE(AO, %rax, SIZE), %xmm0 ;\
        vmovddup          4 * SIZE(BO, BI, SIZE), %xmm4 ;\
        VFMADD_R        %xmm8,%xmm4,%xmm0,%xmm8 ;\
        vmovddup          5 * SIZE(BO, BI, SIZE), %xmm5 ;\
        VFMADD_I        %xmm9,%xmm5,%xmm0,%xmm9 ;\
        vmovddup          6 * SIZE(BO, BI, SIZE), %xmm6 ;\
        VFMADD_R        %xmm10,%xmm6,%xmm0,%xmm10 ;\
        vmovddup          7 * SIZE(BO, BI, SIZE), %xmm7 ;\
        VFMADD_I        %xmm11,%xmm7,%xmm0,%xmm11 ;\
        addq    $16, BI                            ;\
        addq    $8 , %rax                          ;\


#define KERNEL1x2_SUB(xx) \
        vmovups          -8 * SIZE(AO, %rax, SIZE), %xmm0 ;\
        vmovddup         -8 * SIZE(BO, BI, SIZE), %xmm4 ;\
        VFMADD_R        %xmm8,%xmm4,%xmm0,%xmm8 ;\
        vmovddup         -7 * SIZE(BO, BI, SIZE), %xmm5 ;\
        VFMADD_I        %xmm9,%xmm5,%xmm0,%xmm9 ;\
        vmovddup         -6 * SIZE(BO, BI, SIZE), %xmm6 ;\
        VFMADD_R        %xmm10,%xmm6,%xmm0,%xmm10 ;\
        vmovddup         -5 * SIZE(BO, BI, SIZE), %xmm7 ;\
        VFMADD_I        %xmm11,%xmm7,%xmm0,%xmm11 ;\
        addq    $4, BI                            ;\
        addq    $2, %rax                          ;\

/************************************************************************************************/

#define KERNEL2x1_1(xx) \
        prefetcht0      A_PR1(AO,%rax,SIZE)        ;\
        vmovups          -8 * SIZE(AO, %rax, SIZE), %xmm0 ;\
        vmovddup         -4 * SIZE(BO, BI, SIZE), %xmm4 ;\
        VFMADD_R        %xmm8,%xmm4,%xmm0,%xmm8 ;\
        vmovups          -6 * SIZE(AO, %rax, SIZE), %xmm1 ;\
        VFMADD_R        %xmm12,%xmm4,%xmm1,%xmm12 ;\
        vmovddup         -3 * SIZE(BO, BI, SIZE), %xmm5 ;\
        VFMADD_I        %xmm9,%xmm5,%xmm0,%xmm9 ;\
        VFMADD_I        %xmm13,%xmm5,%xmm1,%xmm13 ;\

#define KERNEL2x1_2(xx) \
        vmovups          -4 * SIZE(AO, %rax, SIZE), %xmm0 ;\
        vmovddup         -2 * SIZE(BO, BI, SIZE), %xmm4 ;\
        VFMADD_R        %xmm8,%xmm4,%xmm0,%xmm8 ;\
        vmovups          -2 * SIZE(AO, %rax, SIZE), %xmm1 ;\
        VFMADD_R        %xmm12,%xmm4,%xmm1,%xmm12 ;\
        vmovddup         -1 * SIZE(BO, BI, SIZE), %xmm5 ;\
        VFMADD_I        %xmm9,%xmm5,%xmm0,%xmm9 ;\
        VFMADD_I        %xmm13,%xmm5,%xmm1,%xmm13 ;\

#define KERNEL2x1_3(xx) \
	prefetcht0      A_PR1+64(AO,%rax,SIZE)     ;\
        vmovups           0 * SIZE(AO, %rax, SIZE), %xmm0 ;\
        vmovddup          0 * SIZE(BO, BI, SIZE), %xmm4 ;\
        VFMADD_R        %xmm8,%xmm4,%xmm0,%xmm8 ;\
        vmovups           2 * SIZE(AO, %rax, SIZE), %xmm1 ;\
        VFMADD_R        %xmm12,%xmm4,%xmm1,%xmm12 ;\
        vmovddup          1 * SIZE(BO, BI, SIZE), %xmm5 ;\
        VFMADD_I        %xmm9,%xmm5,%xmm0,%xmm9 ;\
        VFMADD_I        %xmm13,%xmm5,%xmm1,%xmm13 ;\

#define KERNEL2x1_4(xx) \
        vmovups           4 * SIZE(AO, %rax, SIZE), %xmm0 ;\
        vmovddup          2 * SIZE(BO, BI, SIZE), %xmm4 ;\
        VFMADD_R        %xmm8,%xmm4,%xmm0,%xmm8 ;\
        vmovups           6 * SIZE(AO, %rax, SIZE), %xmm1 ;\
        VFMADD_R        %xmm12,%xmm4,%xmm1,%xmm12 ;\
        vmovddup          3 * SIZE(BO, BI, SIZE), %xmm5 ;\
        VFMADD_I        %xmm9,%xmm5,%xmm0,%xmm9 ;\
        VFMADD_I        %xmm13,%xmm5,%xmm1,%xmm13 ;\
        addq    $8, BI                            ;\
        addq    $16, %rax                          ;\


#define KERNEL2x1_SUB(xx) \
        vmovups          -8 * SIZE(AO, %rax, SIZE), %xmm0 ;\
        vmovddup         -4 * SIZE(BO, BI, SIZE), %xmm4 ;\
        VFMADD_R        %xmm8,%xmm4,%xmm0,%xmm8 ;\
        vmovups          -6 * SIZE(AO, %rax, SIZE), %xmm1 ;\
        VFMADD_R        %xmm12,%xmm4,%xmm1,%xmm12 ;\
        vmovddup         -3 * SIZE(BO, BI, SIZE), %xmm5 ;\
        VFMADD_I        %xmm9,%xmm5,%xmm0,%xmm9 ;\
        VFMADD_I        %xmm13,%xmm5,%xmm1,%xmm13 ;\
        addq    $2, BI                            ;\
        addq    $4, %rax                          ;\


/************************************************************************************************/

#define KERNEL1x1_1(xx) \
        prefetcht0      A_PR1(AO,%rax,SIZE)        ;\
        vmovups          -8 * SIZE(AO, %rax, SIZE), %xmm0 ;\
        vmovddup         -4 * SIZE(BO, BI, SIZE), %xmm4 ;\
        VFMADD_R        %xmm8,%xmm4,%xmm0,%xmm8 ;\
        vmovddup         -3 * SIZE(BO, BI, SIZE), %xmm5 ;\
        VFMADD_I        %xmm9,%xmm5,%xmm0,%xmm9 ;\

#define KERNEL1x1_2(xx) \
        vmovups          -6 * SIZE(AO, %rax, SIZE), %xmm0 ;\
        vmovddup         -2 * SIZE(BO, BI, SIZE), %xmm4 ;\
        VFMADD_R        %xmm8,%xmm4,%xmm0,%xmm8 ;\
        vmovddup         -1 * SIZE(BO, BI, SIZE), %xmm5 ;\
        VFMADD_I        %xmm9,%xmm5,%xmm0,%xmm9 ;\

#define KERNEL1x1_3(xx) \
        vmovups          -4 * SIZE(AO, %rax, SIZE), %xmm0 ;\
        vmovddup          0 * SIZE(BO, BI, SIZE), %xmm4 ;\
        VFMADD_R        %xmm8,%xmm4,%xmm0,%xmm8 ;\
        vmovddup          1 * SIZE(BO, BI, SIZE), %xmm5 ;\
        VFMADD_I        %xmm9,%xmm5,%xmm0,%xmm9 ;\

#define KERNEL1x1_4(xx) \
        vmovups          -2 * SIZE(AO, %rax, SIZE), %xmm0 ;\
        vmovddup          2 * SIZE(BO, BI, SIZE), %xmm4 ;\
        VFMADD_R        %xmm8,%xmm4,%xmm0,%xmm8 ;\
        vmovddup          3 * SIZE(BO, BI, SIZE), %xmm5 ;\
        VFMADD_I        %xmm9,%xmm5,%xmm0,%xmm9 ;\
        addq    $8, BI                            ;\
        addq    $8, %rax                          ;\


#define KERNEL1x1_SUB(xx) \
        vmovups          -8 * SIZE(AO, %rax, SIZE), %xmm0 ;\
        vmovddup         -4 * SIZE(BO, BI, SIZE), %xmm4 ;\
        VFMADD_R        %xmm8,%xmm4,%xmm0,%xmm8 ;\
        vmovddup         -3 * SIZE(BO, BI, SIZE), %xmm5 ;\
        VFMADD_I        %xmm9,%xmm5,%xmm0,%xmm9 ;\
        addq    $2, BI                            ;\
        addq    $2, %rax                          ;\


/************************************************************************************************/




	PROLOGUE
	PROFCODE
	
	subq	$STACKSIZE, %rsp
	movq	%rbx,   (%rsp)
	movq	%rbp,  8(%rsp)
	movq	%r12, 16(%rsp)
	movq	%r13, 24(%rsp)
	movq	%r14, 32(%rsp)
	movq	%r15, 40(%rsp)

	vzeroupper

#ifdef WINDOWS_ABI
	movq	%rdi,    48(%rsp)
	movq	%rsi,    56(%rsp)
	vmovups	%xmm6,   64(%rsp)
	vmovups	%xmm7,   80(%rsp)
	vmovups	%xmm8,   96(%rsp)
	vmovups	%xmm9,  112(%rsp)
	vmovups	%xmm10, 128(%rsp)
	vmovups	%xmm11, 144(%rsp)
	vmovups	%xmm12, 160(%rsp)
	vmovups	%xmm13, 176(%rsp)
	vmovups	%xmm14, 192(%rsp)
	vmovups	%xmm15, 208(%rsp)

	movq	ARG1,      OLD_M
	movq	ARG2,      OLD_N
	movq	ARG3,      OLD_K
	movq	OLD_A,     A
	movq	OLD_B,     B
	movq	OLD_C,     C
	movq	OLD_LDC,   LDC
#ifdef TRMMKERNEL
	vmovsd	OLD_OFFSET, %xmm12
#endif
	vmovaps	%xmm3, %xmm0
	vmovsd   OLD_ALPHA_I, %xmm1

#else
	movq	STACKSIZE +  8(%rsp), LDC
#ifdef TRMMKERNEL
	vmovsd	STACKSIZE + 16(%rsp), %xmm12
#endif

#endif

	movq    %rsp, SP      # save old stack
        subq    $128 + L_BUFFER_SIZE, %rsp
        andq    $-4096, %rsp    # align stack

        STACK_TOUCH

	cmpq	$0, OLD_M
	je	.L999

	cmpq	$0, OLD_N
	je	.L999

	cmpq	$0, OLD_K
	je	.L999

	movq	OLD_M, M
	movq	OLD_N, N
	movq	OLD_K, K

	vmovsd	 %xmm0, ALPHA_R
	vmovsd	 %xmm1, ALPHA_I

	salq	$ZBASE_SHIFT, LDC

	movq    N, %rax
        xorq    %rdx, %rdx
        movq    $2,  %rdi
        divq    %rdi                    //    N / 2
        movq    %rax, Ndiv6             //    N / 2
        movq    %rdx, Nmod6             //    N % 2

	

#ifdef TRMMKERNEL
	vmovsd	%xmm12, OFFSET
	vmovsd	%xmm12, KK
#ifndef LEFT
	negq	KK
#endif	
#endif

.L2_0:

	movq	Ndiv6,  J
	cmpq	$0, J
	je	.L1_0
	ALIGN_4



.L2_01:
	// copy to sub buffer
	movq	B, BO1
	leaq    BUFFER1, BO		// first buffer to BO
	movq	K, %rax
	ALIGN_4

.L2_02b:

	vmovups		(BO1), %xmm0
	vmovups	2 * SIZE(BO1), %xmm1
	vmovups	%xmm0,       (BO)
	vmovups	%xmm1, 2 * SIZE(BO)
	addq	$4*SIZE,BO1
	addq	$4*SIZE,BO
	decq	%rax
	jnz	.L2_02b

.L2_02c:

	movq	BO1, B			// next offset of B

.L2_10:
	movq	C, CO1
	leaq	(C, LDC, 2), C		// c += 2 * ldc

#if defined(TRMMKERNEL) && defined(LEFT)
        movq    OFFSET, %rax
        movq    %rax, KK
#endif
	
	movq	A, AO		 	// aoffset = a
	addq	$8 * SIZE, AO

	movq	M,  I
	sarq	$1, I			// i = (m >> 1)
	je	.L2_40

	ALIGN_4

.L2_11:

#if !defined(TRMMKERNEL) || \
        (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
        (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
	leaq	BUFFER1, BO		// first buffer to BO
	addq	$8 * SIZE, BO
#else
        movq    KK, %rax
	leaq	BUFFER1, BO			// first buffer to BO
	addq	$8 * SIZE, BO
	movq    %rax, BI                        //  Index for BO
        leaq    (,BI,4), BI                     //  BI = BI * 4 ; number of values
        leaq    (BO, BI, SIZE), BO
	salq	$2, %rax			// rax = rax * 4 ; number of values
        leaq    (AO, %rax, SIZE), AO
#endif

	vzeroall

#ifndef TRMMKERNEL
        movq    K, %rax
#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
        movq    K, %rax
        subq    KK, %rax
        movq    %rax, KKK
#else
        movq    KK, %rax
#ifdef LEFT
        addq    $2, %rax        // number of values in AO
#else
        addq    $2, %rax        // number of values in BO
#endif
        movq    %rax, KKK
#endif


	andq	$-8, %rax			//  K = K - ( K % 8 )
	je	.L2_16
	movq    %rax, BI                        //  Index for BO
        leaq    ( ,BI,4), BI                    //  BI = BI * 4 ; number of values

	salq	$2, %rax			// rax = rax * 4 ; number of values
	leaq	(AO, %rax, SIZE), AO
	leaq	(BO, BI, SIZE), BO
	negq	BI
	negq	%rax
	ALIGN_4

.L2_12:

	prefetcht0	B_PR1(BO,BI,SIZE)
	KERNEL2x2_1(xxx)
	KERNEL2x2_2(xxx)
	prefetcht0	B_PR1+64(BO,BI,SIZE)
	KERNEL2x2_3(xxx)
	KERNEL2x2_4(xxx)

	prefetcht0	B_PR1(BO,BI,SIZE)
	KERNEL2x2_1(xxx)
	KERNEL2x2_2(xxx)
	prefetcht0	B_PR1+64(BO,BI,SIZE)
	KERNEL2x2_3(xxx)
	KERNEL2x2_4(xxx)

	je	.L2_16

	prefetcht0	B_PR1(BO,BI,SIZE)
	KERNEL2x2_1(xxx)
	KERNEL2x2_2(xxx)
	prefetcht0	B_PR1+64(BO,BI,SIZE)
	KERNEL2x2_3(xxx)
	KERNEL2x2_4(xxx)

	prefetcht0	B_PR1(BO,BI,SIZE)
	KERNEL2x2_1(xxx)
	KERNEL2x2_2(xxx)
	prefetcht0	B_PR1+64(BO,BI,SIZE)
	KERNEL2x2_3(xxx)
	KERNEL2x2_4(xxx)

	je	.L2_16

	jmp	.L2_12
	ALIGN_4

.L2_16:
#ifndef TRMMKERNEL
        movq    K, %rax
#else
        movq    KKK, %rax
#endif

	andq	$7, %rax		# if (k & 1)
	je .L2_19

	movq    %rax, BI                        //  Index for BO
        leaq    ( ,BI,4), BI                    //  BI = BI * 4 ; number of values

	salq	$2, %rax			// rax = rax * 4 ; number of values
	leaq	(AO, %rax, SIZE), AO
	leaq	(BO, BI, SIZE), BO
	negq	BI
	negq	%rax
	ALIGN_4

.L2_17:

	KERNEL2x2_SUB(xxx)
	jl	.L2_17
	ALIGN_4


.L2_19:

	vmovddup	ALPHA_R, %xmm0
	vmovddup	ALPHA_I, %xmm1

	// swap high and low 64 bytes
        vshufpd $0x01, %xmm9 , %xmm9, %xmm9
        vshufpd $0x01, %xmm11, %xmm11, %xmm11
        vshufpd $0x01, %xmm13, %xmm13, %xmm13
        vshufpd $0x01, %xmm15, %xmm15, %xmm15

#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
    defined(NR) || defined(NC) || defined(TR) || defined(TC)

        vaddsubpd %xmm9, %xmm8 , %xmm8
        vaddsubpd %xmm11,%xmm10, %xmm10
        vaddsubpd %xmm13,%xmm12, %xmm12
        vaddsubpd %xmm15,%xmm14, %xmm14

        vshufpd $0x01, %xmm8 , %xmm8, %xmm9
        vshufpd $0x01, %xmm10, %xmm10, %xmm11
        vshufpd $0x01, %xmm12, %xmm12, %xmm13
        vshufpd $0x01, %xmm14, %xmm14, %xmm15

#else
        vaddsubpd %xmm8,  %xmm9 ,%xmm9
        vaddsubpd %xmm10, %xmm11,%xmm11
        vaddsubpd %xmm12, %xmm13,%xmm13
        vaddsubpd %xmm14, %xmm15,%xmm15

        vmovapd   %xmm9,  %xmm8
        vmovapd   %xmm11, %xmm10
        vmovapd   %xmm13, %xmm12
        vmovapd   %xmm15, %xmm14

	// swap high and low 64 bytes
        vshufpd $0x01, %xmm9 , %xmm9, %xmm9
        vshufpd $0x01, %xmm11, %xmm11, %xmm11
        vshufpd $0x01, %xmm13, %xmm13, %xmm13
        vshufpd $0x01, %xmm15, %xmm15, %xmm15

#endif

	// multiply with ALPHA_R
        vmulpd  %xmm8 , %xmm0, %xmm8
        vmulpd  %xmm10, %xmm0, %xmm10
        vmulpd  %xmm12, %xmm0, %xmm12
        vmulpd  %xmm14, %xmm0, %xmm14

	// multiply with ALPHA_I
        vmulpd  %xmm9 , %xmm1, %xmm9
        vmulpd  %xmm11, %xmm1, %xmm11
        vmulpd  %xmm13, %xmm1, %xmm13
        vmulpd  %xmm15, %xmm1, %xmm15

	vaddsubpd %xmm9, %xmm8 , %xmm8
        vaddsubpd %xmm11,%xmm10, %xmm10
        vaddsubpd %xmm13,%xmm12, %xmm12
        vaddsubpd %xmm15,%xmm14, %xmm14



#ifndef TRMMKERNEL

	vaddpd 	 	(CO1), %xmm8 , %xmm8
	vaddpd  2 * SIZE(CO1), %xmm12, %xmm12

	vaddpd 	 	(CO1, LDC), %xmm10, %xmm10
	vaddpd  2 * SIZE(CO1, LDC), %xmm14, %xmm14

#endif

	vmovups	%xmm8 ,  	(CO1)
	vmovups	%xmm12 , 2 * SIZE(CO1)

	vmovups	%xmm10 ,  	(CO1, LDC)
	vmovups	%xmm14 , 2 * SIZE(CO1, LDC)

#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
        movq    K, %rax
        subq    KKK, %rax
	movq    %rax, BI                        //  Index for BO
        leaq    ( ,BI,4), BI                    //  BI = BI * 4 ; number of values
        leaq    (BO, BI, SIZE), BO
	salq	$2, %rax			// rax = rax * 4 ; number of values
        leaq    (AO, %rax, SIZE), AO
#endif


#if defined(TRMMKERNEL) && defined(LEFT)
        addq    $2, KK
#endif

	addq	$4 * SIZE, CO1		# coffset += 4
	decq	I			# i --
	jg	.L2_11
	ALIGN_4	


/**************************************************************************
* Rest of M 
***************************************************************************/
.L2_40:
	testq	$1, M		
	jz	.L2_60		// to next 2 lines of N

	ALIGN_4

.L2_41:

#if !defined(TRMMKERNEL) || \
        (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
        (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
	leaq	BUFFER1, BO		// first buffer to BO
	addq	$8 * SIZE, BO
#else
        movq    KK, %rax
	leaq	BUFFER1, BO			// first buffer to BO
	addq	$8 * SIZE, BO
	movq    %rax, BI                        //  Index for BO
        leaq    (,BI,4), BI                     //  BI = BI * 4 ; number of values
        leaq    (BO, BI, SIZE), BO
	salq	$1, %rax			// rax = rax * 2 ; number of values
        leaq    (AO, %rax, SIZE), AO
#endif

	vzeroall

#ifndef TRMMKERNEL
        movq    K, %rax
#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
        movq    K, %rax
        subq    KK, %rax
        movq    %rax, KKK
#else
        movq    KK, %rax
#ifdef LEFT
        addq    $1, %rax        // number of values in AO
#else
        addq    $2, %rax        // number of values in BO
#endif
        movq    %rax, KKK
#endif


	andq	$-8, %rax			//  K = K - ( K % 8 )
	je	.L2_46
	movq    %rax, BI                        //  Index for BO
        leaq    ( ,BI,4), BI                    //  BI = BI * 4 ; number of values

	salq	$1, %rax			// rax = rax * 2 ; number of values
	leaq	(AO, %rax, SIZE), AO
	leaq	(BO, BI, SIZE), BO
	negq	BI
	negq	%rax
	ALIGN_4

.L2_42:

	prefetcht0	B_PR1(BO,BI,SIZE)
	KERNEL1x2_1(xxx)
	KERNEL1x2_2(xxx)
	prefetcht0	B_PR1+64(BO,BI,SIZE)
	KERNEL1x2_3(xxx)
	KERNEL1x2_4(xxx)

	prefetcht0	B_PR1(BO,BI,SIZE)
	KERNEL1x2_1(xxx)
	KERNEL1x2_2(xxx)
	prefetcht0	B_PR1+64(BO,BI,SIZE)
	KERNEL1x2_3(xxx)
	KERNEL1x2_4(xxx)

	je	.L2_46

	prefetcht0	B_PR1(BO,BI,SIZE)
	KERNEL1x2_1(xxx)
	KERNEL1x2_2(xxx)
	prefetcht0	B_PR1+64(BO,BI,SIZE)
	KERNEL1x2_3(xxx)
	KERNEL1x2_4(xxx)

	prefetcht0	B_PR1(BO,BI,SIZE)
	KERNEL1x2_1(xxx)
	KERNEL1x2_2(xxx)
	prefetcht0	B_PR1+64(BO,BI,SIZE)
	KERNEL1x2_3(xxx)
	KERNEL1x2_4(xxx)

	je	.L2_46

	jmp	.L2_42
	ALIGN_4

.L2_46:
#ifndef TRMMKERNEL
        movq    K, %rax
#else
        movq    KKK, %rax
#endif

	andq	$7, %rax		# if (k & 1)
	je .L2_49

	movq    %rax, BI                        //  Index for BO
        leaq    ( ,BI,4), BI                    //  BI = BI * 4 ; number of values

	salq	$1, %rax			// rax = rax * 2 ; number of values
	leaq	(AO, %rax, SIZE), AO
	leaq	(BO, BI, SIZE), BO
	negq	BI
	negq	%rax
	ALIGN_4

.L2_47:

	KERNEL1x2_SUB(xxx)
	jl	.L2_47
	ALIGN_4


.L2_49:

	vmovddup	ALPHA_R, %xmm0
	vmovddup	ALPHA_I, %xmm1

	// swap high and low 64 bytes
        vshufpd $0x01, %xmm9 , %xmm9, %xmm9
        vshufpd $0x01, %xmm11, %xmm11, %xmm11

#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
    defined(NR) || defined(NC) || defined(TR) || defined(TC)

        vaddsubpd %xmm9, %xmm8 , %xmm8
        vaddsubpd %xmm11,%xmm10, %xmm10

        vshufpd $0x01, %xmm8 , %xmm8, %xmm9
        vshufpd $0x01, %xmm10, %xmm10, %xmm11

#else
        vaddsubpd %xmm8, %xmm9, %xmm9
        vaddsubpd %xmm10,%xmm11, %xmm11

        vmovapd   %xmm9,  %xmm8
        vmovapd   %xmm11, %xmm10

	// swap high and low 64 bytes
        vshufpd $0x01, %xmm9 , %xmm9, %xmm9
        vshufpd $0x01, %xmm11, %xmm11, %xmm11

#endif

	// multiply with ALPHA_R
        vmulpd  %xmm8 , %xmm0, %xmm8
        vmulpd  %xmm10, %xmm0, %xmm10

	// multiply with ALPHA_I
        vmulpd  %xmm9 , %xmm1, %xmm9
        vmulpd  %xmm11, %xmm1, %xmm11

	vaddsubpd %xmm9, %xmm8 , %xmm8
        vaddsubpd %xmm11,%xmm10, %xmm10



#ifndef TRMMKERNEL

	vaddpd 	 	(CO1), %xmm8 , %xmm8
	vaddpd 	 	(CO1, LDC), %xmm10, %xmm10

#endif

	vmovups	%xmm8 ,  	(CO1)
	vmovups	%xmm10 ,  	(CO1, LDC)

#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
        movq    K, %rax
        subq    KKK, %rax
	movq    %rax, BI                        //  Index for BO
        leaq    ( ,BI,4), BI                    //  BI = BI * 4 ; number of values
        leaq    (BO, BI, SIZE), BO
	salq	$1, %rax			// rax = rax * 2 ; number of values
        leaq    (AO, %rax, SIZE), AO
#endif


#if defined(TRMMKERNEL) && defined(LEFT)
        addq    $1, KK
#endif

	addq	$2 * SIZE, CO1		# coffset += 2
	ALIGN_4	



	
.L2_60:
#if defined(TRMMKERNEL) && !defined(LEFT)
        addq    $2, KK
#endif

	decq	J			// j --
	jg	.L2_01			// next 2 lines of N



.L1_0:

/************************************************************************************************
* Loop for Nmod6 % 2 > 0
*************************************************************************************************/

	movq	Nmod6, J		
	andq	$1, J			// j % 2
	je	.L999
	ALIGN_4

.L1_01:
	// copy to sub buffer
	movq	B, BO1
	leaq    BUFFER1, BO		// first buffer to BO
	movq	K, %rax
	ALIGN_4

.L1_02b:

	vmovups		(BO1), %xmm0
	vmovups	%xmm0,       (BO)
	addq	$2*SIZE,BO1
	addq	$2*SIZE,BO
	decq	%rax
	jnz	.L1_02b

.L1_02c:

	movq	BO1, B			// next offset of B

.L1_10:
	movq	C, CO1
	leaq	(C, LDC, 1), C		// c += 1 * ldc

#if defined(TRMMKERNEL) && defined(LEFT)
        movq    OFFSET, %rax
        movq    %rax, KK
#endif
	
	movq	A, AO		 	// aoffset = a
	addq	$8 * SIZE, AO

	movq	M,  I
	sarq	$1, I			// i = (m >> 1)
	je	.L1_40

	ALIGN_4

.L1_11:

#if !defined(TRMMKERNEL) || \
        (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
        (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
	leaq	BUFFER1, BO		// first buffer to BO
	addq	$4 * SIZE, BO
#else
        movq    KK, %rax
	leaq	BUFFER1, BO			// first buffer to BO
	addq	$4 * SIZE, BO
	movq    %rax, BI                        //  Index for BO
        leaq    (,BI,2), BI                     //  BI = BI * 2 ; number of values
        leaq    (BO, BI, SIZE), BO
	salq	$2, %rax			// rax = rax * 4 ; number of values
        leaq    (AO, %rax, SIZE), AO
#endif

	vzeroall

#ifndef TRMMKERNEL
        movq    K, %rax
#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
        movq    K, %rax
        subq    KK, %rax
        movq    %rax, KKK
#else
        movq    KK, %rax
#ifdef LEFT
        addq    $2, %rax        // number of values in AO
#else
        addq    $1, %rax        // number of values in BO
#endif
        movq    %rax, KKK
#endif


	andq	$-8, %rax			//  K = K - ( K % 8 )
	je	.L1_16
	movq    %rax, BI                        //  Index for BO
        leaq    ( ,BI,2), BI                    //  BI = BI * 2 ; number of values

	salq	$2, %rax			// rax = rax * 4 ; number of values
	leaq	(AO, %rax, SIZE), AO
	leaq	(BO, BI, SIZE), BO
	negq	BI
	negq	%rax
	ALIGN_4

.L1_12:

	prefetcht0	B_PR1(BO,BI,SIZE)
	KERNEL2x1_1(xxx)
	KERNEL2x1_2(xxx)
	KERNEL2x1_3(xxx)
	KERNEL2x1_4(xxx)

	prefetcht0	B_PR1(BO,BI,SIZE)
	KERNEL2x1_1(xxx)
	KERNEL2x1_2(xxx)
	KERNEL2x1_3(xxx)
	KERNEL2x1_4(xxx)

	je	.L1_16

	prefetcht0	B_PR1(BO,BI,SIZE)
	KERNEL2x1_1(xxx)
	KERNEL2x1_2(xxx)
	KERNEL2x1_3(xxx)
	KERNEL2x1_4(xxx)

	prefetcht0	B_PR1(BO,BI,SIZE)
	KERNEL2x1_1(xxx)
	KERNEL2x1_2(xxx)
	KERNEL2x1_3(xxx)
	KERNEL2x1_4(xxx)

	je	.L1_16

	jmp	.L1_12
	ALIGN_4

.L1_16:
#ifndef TRMMKERNEL
        movq    K, %rax
#else
        movq    KKK, %rax
#endif

	andq	$7, %rax		# if (k & 1)
	je .L1_19

	movq    %rax, BI                        //  Index for BO
        leaq    ( ,BI,2), BI                    //  BI = BI * 2 ; number of values

	salq	$2, %rax			// rax = rax * 4 ; number of values
	leaq	(AO, %rax, SIZE), AO
	leaq	(BO, BI, SIZE), BO
	negq	BI
	negq	%rax
	ALIGN_4

.L1_17:

	KERNEL2x1_SUB(xxx)
	jl	.L1_17
	ALIGN_4


.L1_19:

	vmovddup	ALPHA_R, %xmm0
	vmovddup	ALPHA_I, %xmm1

	// swap high and low 64 bytes
        vshufpd $0x01, %xmm9 , %xmm9, %xmm9
        vshufpd $0x01, %xmm13, %xmm13, %xmm13

#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
    defined(NR) || defined(NC) || defined(TR) || defined(TC)

        vaddsubpd %xmm9, %xmm8  , %xmm8
        vaddsubpd %xmm13,%xmm12 , %xmm12

        vshufpd $0x01, %xmm8 , %xmm8, %xmm9
        vshufpd $0x01, %xmm12, %xmm12, %xmm13

#else
        vaddsubpd %xmm8, %xmm9 , %xmm9
        vaddsubpd %xmm12,%xmm13, %xmm13

        vmovapd   %xmm9,  %xmm8
        vmovapd   %xmm13, %xmm12

	// swap high and low 64 bytes
        vshufpd $0x01, %xmm9 , %xmm9, %xmm9
        vshufpd $0x01, %xmm13, %xmm13, %xmm13

#endif

	// multiply with ALPHA_R
        vmulpd  %xmm8 , %xmm0, %xmm8
        vmulpd  %xmm12, %xmm0, %xmm12

	// multiply with ALPHA_I
        vmulpd  %xmm9 , %xmm1, %xmm9
        vmulpd  %xmm13, %xmm1, %xmm13

	vaddsubpd %xmm9,  %xmm8 , %xmm8
        vaddsubpd %xmm13, %xmm12, %xmm12



#ifndef TRMMKERNEL

	vaddpd 	 	(CO1), %xmm8 , %xmm8
	vaddpd  2 * SIZE(CO1), %xmm12, %xmm12

#endif

	vmovups	%xmm8 ,  	(CO1)
	vmovups	%xmm12 , 2 * SIZE(CO1)


#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
        movq    K, %rax
        subq    KKK, %rax
	movq    %rax, BI                        //  Index for BO
        leaq    ( ,BI,2), BI                    //  BI = BI * 2 ; number of values
        leaq    (BO, BI, SIZE), BO
	salq	$2, %rax			// rax = rax * 4 ; number of values
        leaq    (AO, %rax, SIZE), AO
#endif


#if defined(TRMMKERNEL) && defined(LEFT)
        addq    $2, KK
#endif

	addq	$4 * SIZE, CO1		# coffset += 4
	decq	I			# i --
	jg	.L1_11
	ALIGN_4	


/**************************************************************************
* Rest of M 
***************************************************************************/
.L1_40:
	testq	$1, M		
	jz	.L999

	ALIGN_4

.L1_41:

#if !defined(TRMMKERNEL) || \
        (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
        (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
	leaq	BUFFER1, BO		// first buffer to BO
	addq	$4 * SIZE, BO
#else
        movq    KK, %rax
	leaq	BUFFER1, BO			// first buffer to BO
	addq	$4 * SIZE, BO
	movq    %rax, BI                        //  Index for BO
        leaq    (,BI,2), BI                     //  BI = BI * 2 ; number of values
        leaq    (BO, BI, SIZE), BO
	salq	$1, %rax			// rax = rax * 2 ; number of values
        leaq    (AO, %rax, SIZE), AO
#endif

	vzeroall

#ifndef TRMMKERNEL
        movq    K, %rax
#elif (defined(LEFT) && !defined(TRANSA)) || (!defined(LEFT) && defined(TRANSA))
        movq    K, %rax
        subq    KK, %rax
        movq    %rax, KKK
#else
        movq    KK, %rax
#ifdef LEFT
        addq    $1, %rax        // number of values in AO
#else
        addq    $1, %rax        // number of values in BO
#endif
        movq    %rax, KKK
#endif


	andq	$-8, %rax			//  K = K - ( K % 8 )
	je	.L1_46
	movq    %rax, BI                        //  Index for BO
        leaq    ( ,BI,2), BI                    //  BI = BI * 2 ; number of values

	salq	$1, %rax			// rax = rax * 2 ; number of values
	leaq	(AO, %rax, SIZE), AO
	leaq	(BO, BI, SIZE), BO
	negq	BI
	negq	%rax
	ALIGN_4

.L1_42:

	prefetcht0	B_PR1(BO,BI,SIZE)
	KERNEL1x1_1(xxx)
	KERNEL1x1_2(xxx)
	KERNEL1x1_3(xxx)
	KERNEL1x1_4(xxx)

	prefetcht0	B_PR1(BO,BI,SIZE)
	KERNEL1x1_1(xxx)
	KERNEL1x1_2(xxx)
	KERNEL1x1_3(xxx)
	KERNEL1x1_4(xxx)

	je	.L1_46

	prefetcht0	B_PR1(BO,BI,SIZE)
	KERNEL1x1_1(xxx)
	KERNEL1x1_2(xxx)
	KERNEL1x1_3(xxx)
	KERNEL1x1_4(xxx)

	prefetcht0	B_PR1(BO,BI,SIZE)
	KERNEL1x1_1(xxx)
	KERNEL1x1_2(xxx)
	KERNEL1x1_3(xxx)
	KERNEL1x1_4(xxx)

	je	.L1_46

	jmp	.L1_42
	ALIGN_4

.L1_46:
#ifndef TRMMKERNEL
        movq    K, %rax
#else
        movq    KKK, %rax
#endif

	andq	$7, %rax		# if (k & 1)
	je .L1_49

	movq    %rax, BI                        //  Index for BO
        leaq    ( ,BI,2), BI                    //  BI = BI * 2 ; number of values

	salq	$1, %rax			// rax = rax * 2 ; number of values
	leaq	(AO, %rax, SIZE), AO
	leaq	(BO, BI, SIZE), BO
	negq	BI
	negq	%rax
	ALIGN_4

.L1_47:

	KERNEL1x1_SUB(xxx)
	jl	.L1_47
	ALIGN_4


.L1_49:

	vmovddup	ALPHA_R, %xmm0
	vmovddup	ALPHA_I, %xmm1

	// swap high and low 64 bytes
        vshufpd $0x01, %xmm9 , %xmm9, %xmm9

#if defined(NN) || defined(NT) || defined(TN) || defined(TT) || \
    defined(NR) || defined(NC) || defined(TR) || defined(TC)

        vaddsubpd %xmm9, %xmm8,  %xmm8

        vshufpd $0x01, %xmm8 , %xmm8, %xmm9

#else
        vaddsubpd %xmm8, %xmm9,  %xmm9

        vmovapd   %xmm9,  %xmm8

	// swap high and low 64 bytes
        vshufpd $0x01, %xmm9 , %xmm9, %xmm9

#endif

	// multiply with ALPHA_R
        vmulpd  %xmm8 , %xmm0, %xmm8

	// multiply with ALPHA_I
        vmulpd  %xmm9 , %xmm1, %xmm9

	vaddsubpd %xmm9 ,%xmm8,  %xmm8



#ifndef TRMMKERNEL

	vaddpd 	 	(CO1), %xmm8 , %xmm8

#endif

	vmovups	%xmm8 ,  	(CO1)

#if (defined(TRMMKERNEL) &&  defined(LEFT) &&  defined(TRANSA)) || \
    (defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
        movq    K, %rax
        subq    KKK, %rax
	movq    %rax, BI                        //  Index for BO
        leaq    ( ,BI,2), BI                    //  BI = BI * 2 ; number of values
        leaq    (BO, BI, SIZE), BO
	salq	$1, %rax			// rax = rax * 2 ; number of values
        leaq    (AO, %rax, SIZE), AO
#endif


#if defined(TRMMKERNEL) && defined(LEFT)
        addq    $1, KK
#endif

	addq	$2 * SIZE, CO1		# coffset += 2
	ALIGN_4	



.L999:
	vzeroupper

	movq   		SP, %rsp
	movq	   (%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
	vmovups	 64(%rsp), %xmm6
	vmovups	 80(%rsp), %xmm7
	vmovups	 96(%rsp), %xmm8
	vmovups	112(%rsp), %xmm9
	vmovups	128(%rsp), %xmm10
	vmovups	144(%rsp), %xmm11
	vmovups	160(%rsp), %xmm12
	vmovups	176(%rsp), %xmm13
	vmovups	192(%rsp), %xmm14
	vmovups	208(%rsp), %xmm15
#endif

	addq	$STACKSIZE, %rsp
	ret

	EPILOGUE