/*********************************************************************/
/* Copyright 2009, 2010 The University of Texas at Austin. */
/* All rights reserved. */
/* */
/* Redistribution and use in source and binary forms, with or */
/* without modification, are permitted provided that the following */
/* conditions are met: */
/* */
/* 1. Redistributions of source code must retain the above */
/* copyright notice, this list of conditions and the following */
/* disclaimer. */
/* */
/* 2. Redistributions in binary form must reproduce the above */
/* copyright notice, this list of conditions and the following */
/* disclaimer in the documentation and/or other materials */
/* provided with the distribution. */
/* */
/* THIS SOFTWARE IS PROVIDED BY THE UNIVERSITY OF TEXAS AT */
/* AUSTIN ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, */
/* INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF */
/* MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE */
/* DISCLAIMED. IN NO EVENT SHALL THE UNIVERSITY OF TEXAS AT */
/* AUSTIN OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, */
/* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES */
/* (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE */
/* GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR */
/* BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF */
/* LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT */
/* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT */
/* OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE */
/* POSSIBILITY OF SUCH DAMAGE. */
/* */
/* The views and conclusions contained in the software and */
/* documentation are those of the authors and should not be */
/* interpreted as representing official policies, either expressed */
/* or implied, of The University of Texas at Austin. */
/*********************************************************************/
#define ASSEMBLER
#include "common.h"
#define OLD_M %rdi
#define OLD_N %rsi
#define M %r13
#define N %r14
#define 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 %rbx
#define CO2 %rbp
#define BB %r12
#ifndef WINDOWS_ABI
#define STACKSIZE 64
#define OLD_LDC 8 + STACKSIZE(%rsp)
#define OLD_OFFSET 16 + STACKSIZE(%rsp)
#else
#define STACKSIZE 256
#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 ALPHA 0(%rsp)
#define J 16(%rsp)
#define OFFSET 24(%rsp)
#define KK 32(%rsp)
#define KKK 40(%rsp)
#define BUFFER 128(%rsp)
#define PREFETCH_R (8 * 4 + 0)
#define PREFETCH_W (PREFETCH_R * 2)
#define PREFETCHSIZE (8 * 13 + 5)
#define PREFETCH prefetcht0
#if defined(OS_LINUX) && defined(CORE_BARCELONA)
.align 32768
#endif
PROLOGUE
PROFCODE
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_M
movq ARG2, OLD_N
movq ARG3, K
movq OLD_A, A
movq OLD_B, B
movq OLD_C, C
movq OLD_LDC, LDC
#ifdef TRMMKERNEL
movsd OLD_OFFSET, %xmm12
#endif
movaps %xmm3, %xmm0
movsd OLD_ALPHA_I, %xmm1
#else
movq OLD_LDC, LDC
#ifdef TRMMKERNEL
movsd OLD_OFFSET, %xmm12
#endif
#endif
movq %rsp, %r15 # save old stack
subq $256 + LOCAL_BUFFER_SIZE, %rsp
andq $-4096, %rsp # align stack
STACK_TOUCHING
movsd %xmm0, 0 + ALPHA
movsd %xmm1, 8 + ALPHA
subq $-16 * SIZE, A
subq $-16 * SIZE, B
movq OLD_M, M
movq OLD_N, N
salq $ZBASE_SHIFT, LDC
#ifdef TRMMKERNEL
movsd %xmm12, OFFSET
movsd %xmm12, KK
#ifndef LEFT
negq KK
#endif
#endif
movq N, J
sarq $2, J
NOBRANCH
jle .L40
ALIGN_4
.L01:
/* Copying to Sub Buffer */
leaq 16 * SIZE + BUFFER, BO
#if defined(TRMMKERNEL) && defined(LEFT)
movq OFFSET, %rax
movq %rax, KK
#endif
movq K, %rax
sarq $2, %rax
NOBRANCH
jle .L05
ALIGN_4
.L02:
movapd -16 * SIZE(B), %xmm0
prefetchnta (PREFETCH_R + 0) * SIZE(B)
movapd -14 * SIZE(B), %xmm1
movapd -12 * SIZE(B), %xmm2
movapd -10 * SIZE(B), %xmm3
movapd -8 * SIZE(B), %xmm4
movapd -6 * SIZE(B), %xmm5
movapd -4 * SIZE(B), %xmm6
movapd -2 * SIZE(B), %xmm7
movddup %xmm0, %xmm8
unpckhpd %xmm0, %xmm0
prefetchnta (PREFETCH_R + 8) * SIZE(B)
movddup %xmm1, %xmm9
unpckhpd %xmm1, %xmm1
movddup %xmm2, %xmm10
unpckhpd %xmm2, %xmm2
movddup %xmm3, %xmm11
unpckhpd %xmm3, %xmm3
prefetcht0 (PREFETCH_W + 0) * SIZE(BO)
movddup %xmm4, %xmm12
unpckhpd %xmm4, %xmm4
movddup %xmm5, %xmm13
unpckhpd %xmm5, %xmm5
movddup %xmm6, %xmm14
unpckhpd %xmm6, %xmm6
movddup %xmm7, %xmm15
unpckhpd %xmm7, %xmm7
prefetcht0 (PREFETCH_W + 8) * SIZE(BO)
movapd %xmm8, -16 * SIZE(BO)
movapd %xmm0, -14 * SIZE(BO)
movapd %xmm9, -12 * SIZE(BO)
movapd %xmm1, -10 * SIZE(BO)
movapd %xmm10, -8 * SIZE(BO)
movapd %xmm2, -6 * SIZE(BO)
movapd %xmm11, -4 * SIZE(BO)
movapd %xmm3, -2 * SIZE(BO)
prefetcht0 (PREFETCH_W + 16) * SIZE(BO)
movapd %xmm12, 0 * SIZE(BO)
movapd %xmm4, 2 * SIZE(BO)
movapd %xmm13, 4 * SIZE(BO)
movapd %xmm5, 6 * SIZE(BO)
prefetcht0 (PREFETCH_W + 24) * SIZE(BO)
movapd %xmm14, 8 * SIZE(BO)
movapd %xmm6, 10 * SIZE(BO)
movapd %xmm15, 12 * SIZE(BO)
movapd %xmm7, 14 * SIZE(BO)
subq $-16 * SIZE, B
subq $-32 * SIZE, BO
decq %rax
BRANCH
jne .L02
ALIGN_4
.L05:
movq K, %rax
andq $3, %rax
BRANCH
BRANCH
jle .L10
ALIGN_4
.L06:
movapd -16 * SIZE(B), %xmm0
movapd -14 * SIZE(B), %xmm1
movddup %xmm0, %xmm8
unpckhpd %xmm0, %xmm0
movddup %xmm1, %xmm9
unpckhpd %xmm1, %xmm1
movapd %xmm8, -16 * SIZE(BO)
movapd %xmm0, -14 * SIZE(BO)
movapd %xmm9, -12 * SIZE(BO)
movapd %xmm1, -10 * SIZE(BO)
addq $4 * SIZE, B
addq $8 * SIZE, BO
decq %rax
BRANCH
jne .L06
ALIGN_4
.L10:
leaq (PREFETCH_R + 0) * SIZE(B), BB
movq C, CO1 # coffset1 = c
leaq (C, LDC, 1), CO2 # coffset2 = c + ldc
movq A, AO # aoffset = a
movq M, I
sarq $2, I # i = (m >> 2)
NOBRANCH
jle .L20
ALIGN_4
.L11:
#if !defined(TRMMKERNEL) || \
(defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \
(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
leaq 20 * SIZE + BUFFER, BO
#else
leaq 20 * SIZE + BUFFER, BO
movq KK, %rax
leaq (, %rax, SIZE), %rax
leaq (AO, %rax, 4), AO
leaq (BO, %rax, 8), BO
#endif
movaps -16 * SIZE(AO), %xmm0
movaps -14 * SIZE(AO), %xmm1
movaps -20 * SIZE(BO), %xmm6
movaps -18 * SIZE(BO), %xmm7
prefetcht2 0 * SIZE(BB)
pxor %xmm2, %xmm2
prefetcht0 7 * SIZE(CO1)
pxor %xmm3, %xmm3
pxor %xmm4, %xmm4
prefetcht0 7 * SIZE(CO2)
pxor %xmm5, %xmm5
movapd %xmm2, %xmm8
movapd %xmm2, %xmm9
movapd %xmm2, %xmm10
prefetcht0 7 * SIZE(CO1, LDC, 2)
movapd %xmm2, %xmm11
movapd %xmm2, %xmm12
movapd %xmm2, %xmm13
prefetcht0 7 * SIZE(CO2, LDC, 2)
movapd %xmm2, %xmm14
movapd %xmm2, %xmm15
subq $-16 * SIZE, BB
#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 $4, %rax
#else
addq $4, %rax
#endif
movq %rax, KKK
#endif
sarq $2, %rax
NOBRANCH
jle .L15
ALIGN_4
.L12:
PADDING;
addpd %xmm2, %xmm10
movaps -16 * SIZE(BO), %xmm2
PADDING;
addpd %xmm3, %xmm14
PREFETCH (PREFETCHSIZE + 0) * SIZE(AO)
movaps %xmm6, %xmm3
mulpd %xmm0, %xmm6
mulpd %xmm1, %xmm3
addpd %xmm4, %xmm11
movaps -14 * SIZE(BO), %xmm4
addpd %xmm5, %xmm15
movaps %xmm7, %xmm5
mulpd %xmm0, %xmm7
mulpd %xmm1, %xmm5
addpd %xmm6, %xmm8
movaps -12 * SIZE(BO), %xmm6
addpd %xmm3, %xmm12
movaps %xmm2, %xmm3
mulpd %xmm0, %xmm2
mulpd %xmm1, %xmm3
addpd %xmm7, %xmm9
movaps -10 * SIZE(BO), %xmm7
addpd %xmm5, %xmm13
movaps %xmm4, %xmm5
mulpd %xmm0, %xmm4
movaps -12 * SIZE(AO), %xmm0
mulpd %xmm1, %xmm5
addpd %xmm2, %xmm10
movaps -8 * SIZE(BO), %xmm2
addpd %xmm3, %xmm14
movaps -10 * SIZE(AO), %xmm1
movaps %xmm6, %xmm3
mulpd %xmm0, %xmm6
mulpd %xmm1, %xmm3
addpd %xmm4, %xmm11
movaps -6 * SIZE(BO), %xmm4
addpd %xmm5, %xmm15
movaps %xmm7, %xmm5
mulpd %xmm0, %xmm7
mulpd %xmm1, %xmm5
addpd %xmm6, %xmm8
movaps -4 * SIZE(BO), %xmm6
addpd %xmm3, %xmm12
movaps %xmm2, %xmm3
mulpd %xmm0, %xmm2
mulpd %xmm1, %xmm3
addpd %xmm7, %xmm9
movaps -2 * SIZE(BO), %xmm7
addpd %xmm5, %xmm13
movaps %xmm4, %xmm5
mulpd %xmm0, %xmm4
movaps -8 * SIZE(AO), %xmm0
mulpd %xmm1, %xmm5
addpd %xmm2, %xmm10
movaps 0 * SIZE(BO), %xmm2
addpd %xmm3, %xmm14
movaps -6 * SIZE(AO), %xmm1
movaps %xmm6, %xmm3
mulpd %xmm0, %xmm6
mulpd %xmm1, %xmm3
addpd %xmm4, %xmm11
movaps 2 * SIZE(BO), %xmm4
addpd %xmm5, %xmm15
PREFETCH (PREFETCHSIZE + 8) * SIZE(AO)
movaps %xmm7, %xmm5
mulpd %xmm1, %xmm5
mulpd %xmm0, %xmm7
addpd %xmm6, %xmm8
movaps 4 * SIZE(BO), %xmm6
addpd %xmm3, %xmm12
movaps %xmm2, %xmm3
mulpd %xmm0, %xmm2
mulpd %xmm1, %xmm3
addpd %xmm7, %xmm9
movaps 6 * SIZE(BO), %xmm7
addpd %xmm5, %xmm13
movaps %xmm4, %xmm5
mulpd %xmm0, %xmm4
movaps -4 * SIZE(AO), %xmm0
mulpd %xmm1, %xmm5
addpd %xmm2, %xmm10
movaps 8 * SIZE(BO), %xmm2
addpd %xmm3, %xmm14
movaps -2 * SIZE(AO), %xmm1
movaps %xmm6, %xmm3
mulpd %xmm0, %xmm6
mulpd %xmm1, %xmm3
addpd %xmm4, %xmm11
movaps 10 * SIZE(BO), %xmm4
addpd %xmm5, %xmm15
movaps %xmm7, %xmm5
mulpd %xmm0, %xmm7
mulpd %xmm1, %xmm5
addpd %xmm6, %xmm8
movaps 12 * SIZE(BO), %xmm6
addpd %xmm3, %xmm12
movaps %xmm2, %xmm3
mulpd %xmm0, %xmm2
mulpd %xmm1, %xmm3
subq $-16 * SIZE, AO
addpd %xmm7, %xmm9
movaps 14 * SIZE(BO), %xmm7
addpd %xmm5, %xmm13
subq $-32 * SIZE, BO
movaps %xmm4, %xmm5
mulpd %xmm0, %xmm4
movaps -16 * SIZE(AO), %xmm0
mulpd %xmm1, %xmm5
movaps -14 * SIZE(AO), %xmm1
subq $1, %rax
BRANCH
jg .L12
ALIGN_4
.L15:
prefetcht2 -8 * SIZE(BB)
#ifndef TRMMKERNEL
movq K, %rax
#else
movq KKK, %rax
#endif
andq $3, %rax # if (k & 1)
BRANCH
je .L18
ALIGN_4
.L16:
addpd %xmm2, %xmm10
movaps -16 * SIZE(BO), %xmm2
addpd %xmm3, %xmm14
movaps %xmm6, %xmm3
mulpd %xmm0, %xmm6
mulpd %xmm1, %xmm3
addpd %xmm4, %xmm11
movaps -14 * SIZE(BO), %xmm4
addpd %xmm5, %xmm15
movaps %xmm7, %xmm5
mulpd %xmm0, %xmm7
mulpd %xmm1, %xmm5
addpd %xmm6, %xmm8
movaps -12 * SIZE(BO), %xmm6
addpd %xmm3, %xmm12
addq $4 * SIZE, AO
movaps %xmm2, %xmm3
mulpd %xmm0, %xmm2
mulpd %xmm1, %xmm3
addpd %xmm7, %xmm9
movaps -10 * SIZE(BO), %xmm7
addpd %xmm5, %xmm13
addq $8 * SIZE, BO
movaps %xmm4, %xmm5
mulpd %xmm0, %xmm4
movaps -16 * SIZE(AO), %xmm0
mulpd %xmm1, %xmm5
movaps -14 * SIZE(AO), %xmm1
subq $1, %rax
BRANCH
jg .L16
ALIGN_4
.L18:
movapd ALPHA, %xmm7
addpd %xmm2, %xmm10
addpd %xmm3, %xmm14
addpd %xmm4, %xmm11
addpd %xmm5, %xmm15
movsd 0 * SIZE(CO1), %xmm0
movhpd 1 * SIZE(CO1), %xmm0
movsd 2 * SIZE(CO1), %xmm1
movhpd 3 * SIZE(CO1), %xmm1
movsd 4 * SIZE(CO1), %xmm2
movhpd 5 * SIZE(CO1), %xmm2
movsd 6 * SIZE(CO1), %xmm3
movhpd 7 * SIZE(CO1), %xmm3
movddup %xmm8, %xmm4
unpckhpd %xmm8, %xmm8
movddup %xmm12, %xmm5
unpckhpd %xmm12, %xmm12
mulpd %xmm7, %xmm4
mulpd %xmm7, %xmm8
mulpd %xmm7, %xmm5
mulpd %xmm7, %xmm12
addpd %xmm4, %xmm0
addpd %xmm8, %xmm1
addpd %xmm5, %xmm2
addpd %xmm12, %xmm3
movsd %xmm0, 0 * SIZE(CO1)
movhpd %xmm0, 1 * SIZE(CO1)
movsd %xmm1, 2 * SIZE(CO1)
movhpd %xmm1, 3 * SIZE(CO1)
movsd %xmm2, 4 * SIZE(CO1)
movhpd %xmm2, 5 * SIZE(CO1)
movsd %xmm3, 6 * SIZE(CO1)
movhpd %xmm3, 7 * SIZE(CO1)
movsd 0 * SIZE(CO2), %xmm0
movhpd 1 * SIZE(CO2), %xmm0
movsd 2 * SIZE(CO2), %xmm1
movhpd 3 * SIZE(CO2), %xmm1
movsd 4 * SIZE(CO2), %xmm2
movhpd 5 * SIZE(CO2), %xmm2
movsd 6 * SIZE(CO2), %xmm3
movhpd 7 * SIZE(CO2), %xmm3
movddup %xmm9, %xmm4
unpckhpd %xmm9, %xmm9
movddup %xmm13, %xmm5
unpckhpd %xmm13, %xmm13
mulpd %xmm7, %xmm4
mulpd %xmm7, %xmm9
mulpd %xmm7, %xmm5
mulpd %xmm7, %xmm13
addpd %xmm4, %xmm0
addpd %xmm9, %xmm1
addpd %xmm5, %xmm2
addpd %xmm13, %xmm3
movsd %xmm0, 0 * SIZE(CO2)
movhpd %xmm0, 1 * SIZE(CO2)
movsd %xmm1, 2 * SIZE(CO2)
movhpd %xmm1, 3 * SIZE(CO2)
movsd %xmm2, 4 * SIZE(CO2)
movhpd %xmm2, 5 * SIZE(CO2)
movsd %xmm3, 6 * SIZE(CO2)
movhpd %xmm3, 7 * SIZE(CO2)
movsd 0 * SIZE(CO1, LDC, 2), %xmm0
movhpd 1 * SIZE(CO1, LDC, 2), %xmm0
movsd 2 * SIZE(CO1, LDC, 2), %xmm1
movhpd 3 * SIZE(CO1, LDC, 2), %xmm1
movsd 4 * SIZE(CO1, LDC, 2), %xmm2
movhpd 5 * SIZE(CO1, LDC, 2), %xmm2
movsd 6 * SIZE(CO1, LDC, 2), %xmm3
movhpd 7 * SIZE(CO1, LDC, 2), %xmm3
movddup %xmm10, %xmm4
unpckhpd %xmm10, %xmm10
movddup %xmm14, %xmm5
unpckhpd %xmm14, %xmm14
mulpd %xmm7, %xmm4
mulpd %xmm7, %xmm10
mulpd %xmm7, %xmm5
mulpd %xmm7, %xmm14
addpd %xmm4, %xmm0
addpd %xmm10, %xmm1
addpd %xmm5, %xmm2
addpd %xmm14, %xmm3
movsd %xmm0, 0 * SIZE(CO1, LDC, 2)
movhpd %xmm0, 1 * SIZE(CO1, LDC, 2)
movsd %xmm1, 2 * SIZE(CO1, LDC, 2)
movhpd %xmm1, 3 * SIZE(CO1, LDC, 2)
movsd %xmm2, 4 * SIZE(CO1, LDC, 2)
movhpd %xmm2, 5 * SIZE(CO1, LDC, 2)
movsd %xmm3, 6 * SIZE(CO1, LDC, 2)
movhpd %xmm3, 7 * SIZE(CO1, LDC, 2)
movsd 0 * SIZE(CO2, LDC, 2), %xmm0
movhpd 1 * SIZE(CO2, LDC, 2), %xmm0
movsd 2 * SIZE(CO2, LDC, 2), %xmm1
movhpd 3 * SIZE(CO2, LDC, 2), %xmm1
movsd 4 * SIZE(CO2, LDC, 2), %xmm2
movhpd 5 * SIZE(CO2, LDC, 2), %xmm2
movsd 6 * SIZE(CO2, LDC, 2), %xmm3
movhpd 7 * SIZE(CO2, LDC, 2), %xmm3
movddup %xmm11, %xmm4
unpckhpd %xmm11, %xmm11
movddup %xmm15, %xmm5
unpckhpd %xmm15, %xmm15
mulpd %xmm7, %xmm4
mulpd %xmm7, %xmm11
mulpd %xmm7, %xmm5
mulpd %xmm7, %xmm15
addpd %xmm4, %xmm0
addpd %xmm11, %xmm1
addpd %xmm5, %xmm2
addpd %xmm15, %xmm3
movsd %xmm0, 0 * SIZE(CO2, LDC, 2)
movhpd %xmm0, 1 * SIZE(CO2, LDC, 2)
movsd %xmm1, 2 * SIZE(CO2, LDC, 2)
movhpd %xmm1, 3 * SIZE(CO2, LDC, 2)
movsd %xmm2, 4 * SIZE(CO2, LDC, 2)
movhpd %xmm2, 5 * SIZE(CO2, LDC, 2)
movsd %xmm3, 6 * SIZE(CO2, LDC, 2)
movhpd %xmm3, 7 * SIZE(CO2, LDC, 2)
addq $8 * SIZE, CO1 # coffset += 4
addq $8 * SIZE, CO2 # coffset += 4
decq I # i --
BRANCH
jg .L11
ALIGN_4
.L20:
testq $2, M
BRANCH
jle .L30
#if !defined(TRMMKERNEL) || \
(defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \
(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
leaq 16 * SIZE + BUFFER, BO
#else
leaq 16 * SIZE + BUFFER, BO
movq KK, %rax
leaq (, %rax, SIZE), %rax
leaq (AO, %rax, 2), AO
leaq (BO, %rax, 8), BO
#endif
pxor %xmm8, %xmm8
movapd -16 * SIZE(AO), %xmm0
pxor %xmm9, %xmm9
pxor %xmm10, %xmm10
pxor %xmm11, %xmm11
movapd %xmm8, %xmm2
movapd %xmm9, %xmm3
movapd %xmm10, %xmm4
movapd %xmm11, %xmm5
#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
#else
addq $4, %rax
#endif
movq %rax, KKK
#endif
sarq $2, %rax
NOBRANCH
jle .L25
ALIGN_4
.L21:
PREFETCH (PREFETCHSIZE + 0) * SIZE(AO)
addpd %xmm2, %xmm8
movapd -16 * SIZE(BO), %xmm2
mulpd %xmm0, %xmm2
addpd %xmm3, %xmm9
movapd -14 * SIZE(BO), %xmm3
mulpd %xmm0, %xmm3
addpd %xmm4, %xmm10
movapd -12 * SIZE(BO), %xmm4
mulpd %xmm0, %xmm4
addpd %xmm5, %xmm11
movapd -10 * SIZE(BO), %xmm5
mulpd %xmm0, %xmm5
movapd -14 * SIZE(AO), %xmm0
addpd %xmm2, %xmm8
movapd -8 * SIZE(BO), %xmm2
mulpd %xmm0, %xmm2
addpd %xmm3, %xmm9
movapd -6 * SIZE(BO), %xmm3
mulpd %xmm0, %xmm3
addpd %xmm4, %xmm10
movapd -4 * SIZE(BO), %xmm4
mulpd %xmm0, %xmm4
addpd %xmm5, %xmm11
movapd -2 * SIZE(BO), %xmm5
mulpd %xmm0, %xmm5
movapd -12 * SIZE(AO), %xmm0
addpd %xmm2, %xmm8
movapd 0 * SIZE(BO), %xmm2
mulpd %xmm0, %xmm2
addpd %xmm3, %xmm9
movapd 2 * SIZE(BO), %xmm3
mulpd %xmm0, %xmm3
addpd %xmm4, %xmm10
movapd 4 * SIZE(BO), %xmm4
mulpd %xmm0, %xmm4
addpd %xmm5, %xmm11
movapd 6 * SIZE(BO), %xmm5
mulpd %xmm0, %xmm5
movapd -10 * SIZE(AO), %xmm0
addpd %xmm2, %xmm8
movapd 8 * SIZE(BO), %xmm2
mulpd %xmm0, %xmm2
addpd %xmm3, %xmm9
movapd 10 * SIZE(BO), %xmm3
mulpd %xmm0, %xmm3
addpd %xmm4, %xmm10
movapd 12 * SIZE(BO), %xmm4
mulpd %xmm0, %xmm4
addpd %xmm5, %xmm11
movapd 14 * SIZE(BO), %xmm5
mulpd %xmm0, %xmm5
movapd -8 * SIZE(AO), %xmm0
subq $ -8 * SIZE, AO
subq $-32 * SIZE, BO
subq $1, %rax
BRANCH
jg .L21
ALIGN_4
.L25:
movapd ALPHA, %xmm7
#ifndef TRMMKERNEL
movq K, %rax
#else
movq KKK, %rax
#endif
andq $3, %rax # if (k & 1)
BRANCH
je .L28
ALIGN_4
.L26:
addpd %xmm2, %xmm8
movapd -16 * SIZE(BO), %xmm2
mulpd %xmm0, %xmm2
addpd %xmm3, %xmm9
movapd -14 * SIZE(BO), %xmm3
mulpd %xmm0, %xmm3
addpd %xmm4, %xmm10
movapd -12 * SIZE(BO), %xmm4
mulpd %xmm0, %xmm4
addpd %xmm5, %xmm11
movapd -10 * SIZE(BO), %xmm5
mulpd %xmm0, %xmm5
movapd -14 * SIZE(AO), %xmm0
addq $2 * SIZE, AO
addq $8 * SIZE, BO
subq $1, %rax
BRANCH
jg .L26
ALIGN_4
.L28:
addpd %xmm2, %xmm8
addpd %xmm3, %xmm9
addpd %xmm4, %xmm10
addpd %xmm5, %xmm11
movsd 0 * SIZE(CO1), %xmm0
movhpd 1 * SIZE(CO1), %xmm0
movsd 2 * SIZE(CO1), %xmm1
movhpd 3 * SIZE(CO1), %xmm1
movddup %xmm8, %xmm4
unpckhpd %xmm8, %xmm8
mulpd %xmm7, %xmm4
mulpd %xmm7, %xmm8
addpd %xmm4, %xmm0
addpd %xmm8, %xmm1
movsd %xmm0, 0 * SIZE(CO1)
movhpd %xmm0, 1 * SIZE(CO1)
movsd %xmm1, 2 * SIZE(CO1)
movhpd %xmm1, 3 * SIZE(CO1)
movsd 0 * SIZE(CO2), %xmm0
movhpd 1 * SIZE(CO2), %xmm0
movsd 2 * SIZE(CO2), %xmm1
movhpd 3 * SIZE(CO2), %xmm1
movddup %xmm9, %xmm4
unpckhpd %xmm9, %xmm9
mulpd %xmm7, %xmm4
mulpd %xmm7, %xmm9
addpd %xmm4, %xmm0
addpd %xmm9, %xmm1
movsd %xmm0, 0 * SIZE(CO2)
movhpd %xmm0, 1 * SIZE(CO2)
movsd %xmm1, 2 * SIZE(CO2)
movhpd %xmm1, 3 * SIZE(CO2)
movsd 0 * SIZE(CO1, LDC, 2), %xmm0
movhpd 1 * SIZE(CO1, LDC, 2), %xmm0
movsd 2 * SIZE(CO1, LDC, 2), %xmm1
movhpd 3 * SIZE(CO1, LDC, 2), %xmm1
movddup %xmm10, %xmm4
unpckhpd %xmm10, %xmm10
mulpd %xmm7, %xmm4
mulpd %xmm7, %xmm10
addpd %xmm4, %xmm0
addpd %xmm10, %xmm1
movsd %xmm0, 0 * SIZE(CO1, LDC, 2)
movhpd %xmm0, 1 * SIZE(CO1, LDC, 2)
movsd %xmm1, 2 * SIZE(CO1, LDC, 2)
movhpd %xmm1, 3 * SIZE(CO1, LDC, 2)
movsd 0 * SIZE(CO2, LDC, 2), %xmm0
movhpd 1 * SIZE(CO2, LDC, 2), %xmm0
movsd 2 * SIZE(CO2, LDC, 2), %xmm1
movhpd 3 * SIZE(CO2, LDC, 2), %xmm1
movddup %xmm11, %xmm4
unpckhpd %xmm11, %xmm11
mulpd %xmm7, %xmm4
mulpd %xmm7, %xmm11
addpd %xmm4, %xmm0
addpd %xmm11, %xmm1
movsd %xmm0, 0 * SIZE(CO2, LDC, 2)
movhpd %xmm0, 1 * SIZE(CO2, LDC, 2)
movsd %xmm1, 2 * SIZE(CO2, LDC, 2)
movhpd %xmm1, 3 * SIZE(CO2, LDC, 2)
addq $4 * SIZE, CO1 # coffset += 4
addq $4 * SIZE, CO2 # coffset += 4
ALIGN_4
.L30:
testq $1, M
BRANCH
jle .L39
#if !defined(TRMMKERNEL) || \
(defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \
(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
leaq 16 * SIZE + BUFFER, BO
#else
leaq 16 * SIZE + BUFFER, BO
movq KK, %rax
leaq (, %rax, SIZE), %rax
leaq (AO, %rax, 1), AO
leaq (BO, %rax, 8), BO
#endif
pxor %xmm8, %xmm8
movsd -16 * SIZE(AO), %xmm0
pxor %xmm9, %xmm9
pxor %xmm10, %xmm10
pxor %xmm11, %xmm11
movapd %xmm8, %xmm2
movapd %xmm9, %xmm3
movapd %xmm10, %xmm4
movapd %xmm11, %xmm5
#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
#else
addq $4, %rax
#endif
movq %rax, KKK
#endif
sarq $2, %rax
NOBRANCH
jle .L35
ALIGN_4
.L31:
PREFETCH (PREFETCHSIZE + 0) * SIZE(AO)
addsd %xmm2, %xmm8
movsd -16 * SIZE(BO), %xmm2
mulsd %xmm0, %xmm2
addsd %xmm3, %xmm9
movsd -14 * SIZE(BO), %xmm3
mulsd %xmm0, %xmm3
addsd %xmm4, %xmm10
movsd -12 * SIZE(BO), %xmm4
mulsd %xmm0, %xmm4
addsd %xmm5, %xmm11
movsd -10 * SIZE(BO), %xmm5
mulsd %xmm0, %xmm5
movsd -15 * SIZE(AO), %xmm0
addsd %xmm2, %xmm8
movsd -8 * SIZE(BO), %xmm2
mulsd %xmm0, %xmm2
addsd %xmm3, %xmm9
movsd -6 * SIZE(BO), %xmm3
mulsd %xmm0, %xmm3
addsd %xmm4, %xmm10
movsd -4 * SIZE(BO), %xmm4
mulsd %xmm0, %xmm4
addsd %xmm5, %xmm11
movsd -2 * SIZE(BO), %xmm5
mulsd %xmm0, %xmm5
movsd -14 * SIZE(AO), %xmm0
addsd %xmm2, %xmm8
movsd 0 * SIZE(BO), %xmm2
mulsd %xmm0, %xmm2
addsd %xmm3, %xmm9
movsd 2 * SIZE(BO), %xmm3
mulsd %xmm0, %xmm3
addsd %xmm4, %xmm10
movsd 4 * SIZE(BO), %xmm4
mulsd %xmm0, %xmm4
addsd %xmm5, %xmm11
movsd 6 * SIZE(BO), %xmm5
mulsd %xmm0, %xmm5
movsd -13 * SIZE(AO), %xmm0
addsd %xmm2, %xmm8
movsd 8 * SIZE(BO), %xmm2
mulsd %xmm0, %xmm2
addsd %xmm3, %xmm9
movsd 10 * SIZE(BO), %xmm3
mulsd %xmm0, %xmm3
addsd %xmm4, %xmm10
movsd 12 * SIZE(BO), %xmm4
mulsd %xmm0, %xmm4
addsd %xmm5, %xmm11
movsd 14 * SIZE(BO), %xmm5
mulsd %xmm0, %xmm5
movsd -12 * SIZE(AO), %xmm0
subq $ -4 * SIZE, AO
subq $-32 * SIZE, BO
subq $1, %rax
BRANCH
jg .L31
ALIGN_4
.L35:
movapd ALPHA, %xmm7
#ifndef TRMMKERNEL
movq K, %rax
#else
movq KKK, %rax
#endif
andq $3, %rax # if (k & 1)
BRANCH
je .L38
ALIGN_4
.L36:
addsd %xmm2, %xmm8
movsd -16 * SIZE(BO), %xmm2
mulsd %xmm0, %xmm2
addsd %xmm3, %xmm9
movsd -14 * SIZE(BO), %xmm3
mulsd %xmm0, %xmm3
addsd %xmm4, %xmm10
movsd -12 * SIZE(BO), %xmm4
mulsd %xmm0, %xmm4
addsd %xmm5, %xmm11
movsd -10 * SIZE(BO), %xmm5
mulsd %xmm0, %xmm5
movsd -15 * SIZE(AO), %xmm0
addq $1 * SIZE, AO
addq $8 * SIZE, BO
subq $1, %rax
BRANCH
jg .L36
ALIGN_4
.L38:
addsd %xmm2, %xmm8
addsd %xmm3, %xmm9
addsd %xmm4, %xmm10
addsd %xmm5, %xmm11
movsd 0 * SIZE(CO1), %xmm0
movhpd 1 * SIZE(CO1), %xmm0
movddup %xmm8, %xmm4
mulpd %xmm7, %xmm4
addpd %xmm4, %xmm0
movsd %xmm0, 0 * SIZE(CO1)
movhpd %xmm0, 1 * SIZE(CO1)
movsd 0 * SIZE(CO2), %xmm0
movhpd 1 * SIZE(CO2), %xmm0
movddup %xmm9, %xmm4
mulpd %xmm7, %xmm4
addpd %xmm4, %xmm0
movsd %xmm0, 0 * SIZE(CO2)
movhpd %xmm0, 1 * SIZE(CO2)
movsd 0 * SIZE(CO1, LDC, 2), %xmm0
movhpd 1 * SIZE(CO1, LDC, 2), %xmm0
movddup %xmm10, %xmm4
mulpd %xmm7, %xmm4
addpd %xmm4, %xmm0
movsd %xmm0, 0 * SIZE(CO1, LDC, 2)
movhpd %xmm0, 1 * SIZE(CO1, LDC, 2)
movsd 0 * SIZE(CO2, LDC, 2), %xmm0
movhpd 1 * SIZE(CO2, LDC, 2), %xmm0
movddup %xmm11, %xmm4
mulpd %xmm7, %xmm4
addpd %xmm4, %xmm0
movsd %xmm0, 0 * SIZE(CO2, LDC, 2)
movhpd %xmm0, 1 * SIZE(CO2, LDC, 2)
ALIGN_4
.L39:
#if defined(TRMMKERNEL) && !defined(LEFT)
addl $4, KK
#endif
leaq (C, LDC, 4), C
subq $1, J
BRANCH
jg .L01
ALIGN_4
.L40:
testq $2, N
BRANCH
jle .L80
ALIGN_4
.L41:
/* Copying to Sub Buffer */
leaq BUFFER, BO
#if defined(TRMMKERNEL) && defined(LEFT)
movq OFFSET, %rax
movq %rax, KK
#endif
movq K, %rax
sarq $3, %rax
jle .L43
addq %rax, %rax
ALIGN_4
.L42:
movddup -16 * SIZE(B), %xmm8
movddup -15 * SIZE(B), %xmm9
movddup -14 * SIZE(B), %xmm10
movddup -13 * SIZE(B), %xmm11
movddup -12 * SIZE(B), %xmm12
movddup -11 * SIZE(B), %xmm13
movddup -10 * SIZE(B), %xmm14
movddup -9 * SIZE(B), %xmm15
movapd %xmm8, 0 * SIZE(BO)
movapd %xmm9, 2 * SIZE(BO)
movapd %xmm10, 4 * SIZE(BO)
movapd %xmm11, 6 * SIZE(BO)
movapd %xmm12, 8 * SIZE(BO)
movapd %xmm13, 10 * SIZE(BO)
movapd %xmm14, 12 * SIZE(BO)
movapd %xmm15, 14 * SIZE(BO)
addq $8 * SIZE, B
addq $16 * SIZE, BO
subq $1, %rax
jne .L42
ALIGN_4
.L43:
movq K, %rax
andq $7, %rax
BRANCH
jle .L45
ALIGN_4
.L44:
movddup -16 * SIZE(B), %xmm8
movddup -15 * SIZE(B), %xmm9
movapd %xmm8, 0 * SIZE(BO)
movapd %xmm9, 2 * SIZE(BO)
addq $2 * SIZE, B
addq $4 * SIZE, BO
subq $1, %rax
jne .L44
ALIGN_4
.L45:
movq C, CO1
leaq (C, LDC, 1), CO2
movq A, AO # aoffset = a
movq M, I
sarq $2, I # i = (m >> 2)
jle .L60
ALIGN_4
.L50:
#if !defined(TRMMKERNEL) || \
(defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \
(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
leaq 16 * SIZE + BUFFER, BO
#else
leaq 16 * SIZE + BUFFER, BO
movq KK, %rax
leaq (, %rax, SIZE), %rax
leaq (AO, %rax, 4), AO
leaq (BO, %rax, 4), BO
#endif
pxor %xmm8, %xmm8
pxor %xmm9, %xmm9
prefetcht0 3 * SIZE(CO1)
pxor %xmm12, %xmm12
prefetcht0 3 * SIZE(CO2)
pxor %xmm13, %xmm13
movapd -16 * SIZE(AO), %xmm0
movapd -14 * SIZE(AO), %xmm1
movapd %xmm8, %xmm2
movapd %xmm8, %xmm3
movapd %xmm8, %xmm4
movapd %xmm8, %xmm5
#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 $4, %rax
#else
addq $2, %rax
#endif
movq %rax, KKK
#endif
sarq $2, %rax
jle .L55
ALIGN_4
.L51:
PREFETCH (PREFETCHSIZE + 0) * SIZE(AO)
addpd %xmm2, %xmm8
movapd -16 * SIZE(BO), %xmm2
addpd %xmm3, %xmm12
movapd %xmm2, %xmm3
mulpd %xmm0, %xmm2
mulpd %xmm1, %xmm3
addpd %xmm4, %xmm9
movapd -14 * SIZE(BO), %xmm4
addpd %xmm5, %xmm13
movapd %xmm4, %xmm5
mulpd %xmm0, %xmm4
movapd -12 * SIZE(AO), %xmm0
mulpd %xmm1, %xmm5
movapd -10 * SIZE(AO), %xmm1
addpd %xmm2, %xmm8
movapd -12 * SIZE(BO), %xmm2
addpd %xmm3, %xmm12
movapd %xmm2, %xmm3
mulpd %xmm0, %xmm2
mulpd %xmm1, %xmm3
addpd %xmm4, %xmm9
movapd -10 * SIZE(BO), %xmm4
addpd %xmm5, %xmm13
movapd %xmm4, %xmm5
mulpd %xmm0, %xmm4
movapd -8 * SIZE(AO), %xmm0
mulpd %xmm1, %xmm5
movapd -6 * SIZE(AO), %xmm1
PREFETCH (PREFETCHSIZE + 8) * SIZE(AO)
addpd %xmm2, %xmm8
movapd -8 * SIZE(BO), %xmm2
addpd %xmm3, %xmm12
movapd %xmm2, %xmm3
mulpd %xmm0, %xmm2
mulpd %xmm1, %xmm3
addpd %xmm4, %xmm9
movapd -6 * SIZE(BO), %xmm4
addpd %xmm5, %xmm13
movapd %xmm4, %xmm5
mulpd %xmm0, %xmm4
movapd -4 * SIZE(AO), %xmm0
mulpd %xmm1, %xmm5
movapd -2 * SIZE(AO), %xmm1
addpd %xmm2, %xmm8
movapd -4 * SIZE(BO), %xmm2
addpd %xmm3, %xmm12
movapd %xmm2, %xmm3
mulpd %xmm0, %xmm2
mulpd %xmm1, %xmm3
addpd %xmm4, %xmm9
movapd -2 * SIZE(BO), %xmm4
addpd %xmm5, %xmm13
movapd %xmm4, %xmm5
mulpd %xmm0, %xmm4
movapd 0 * SIZE(AO), %xmm0
mulpd %xmm1, %xmm5
movapd 2 * SIZE(AO), %xmm1
subq $-16 * SIZE, AO
subq $-16 * SIZE, BO
subq $1, %rax
jg .L51
ALIGN_4
.L55:
movapd ALPHA, %xmm7
#ifndef TRMMKERNEL
movq K, %rax
#else
movq KKK, %rax
#endif
andq $3, %rax # if (k & 1)
je .L58
ALIGN_4
.L56:
addpd %xmm2, %xmm8
movapd -16 * SIZE(BO), %xmm2
addpd %xmm3, %xmm12
movapd %xmm2, %xmm3
mulpd %xmm0, %xmm2
mulpd %xmm1, %xmm3
addpd %xmm4, %xmm9
movapd -14 * SIZE(BO), %xmm4
addpd %xmm5, %xmm13
movapd %xmm4, %xmm5
mulpd %xmm0, %xmm4
movapd -12 * SIZE(AO), %xmm0
mulpd %xmm1, %xmm5
movapd -10 * SIZE(AO), %xmm1
addq $4 * SIZE, AO
addq $4 * SIZE, BO
subq $1, %rax
jg .L56
ALIGN_4
.L58:
addpd %xmm2, %xmm8
addpd %xmm3, %xmm12
addpd %xmm4, %xmm9
addpd %xmm5, %xmm13
movsd 0 * SIZE(CO1), %xmm0
movhpd 1 * SIZE(CO1), %xmm0
movsd 2 * SIZE(CO1), %xmm1
movhpd 3 * SIZE(CO1), %xmm1
movsd 4 * SIZE(CO1), %xmm2
movhpd 5 * SIZE(CO1), %xmm2
movsd 6 * SIZE(CO1), %xmm3
movhpd 7 * SIZE(CO1), %xmm3
movddup %xmm8, %xmm4
unpckhpd %xmm8, %xmm8
movddup %xmm12, %xmm5
unpckhpd %xmm12, %xmm12
mulpd %xmm7, %xmm4
mulpd %xmm7, %xmm8
mulpd %xmm7, %xmm5
mulpd %xmm7, %xmm12
addpd %xmm4, %xmm0
addpd %xmm8, %xmm1
addpd %xmm5, %xmm2
addpd %xmm12, %xmm3
movsd %xmm0, 0 * SIZE(CO1)
movhpd %xmm0, 1 * SIZE(CO1)
movsd %xmm1, 2 * SIZE(CO1)
movhpd %xmm1, 3 * SIZE(CO1)
movsd %xmm2, 4 * SIZE(CO1)
movhpd %xmm2, 5 * SIZE(CO1)
movsd %xmm3, 6 * SIZE(CO1)
movhpd %xmm3, 7 * SIZE(CO1)
movsd 0 * SIZE(CO2), %xmm0
movhpd 1 * SIZE(CO2), %xmm0
movsd 2 * SIZE(CO2), %xmm1
movhpd 3 * SIZE(CO2), %xmm1
movsd 4 * SIZE(CO2), %xmm2
movhpd 5 * SIZE(CO2), %xmm2
movsd 6 * SIZE(CO2), %xmm3
movhpd 7 * SIZE(CO2), %xmm3
movddup %xmm9, %xmm4
unpckhpd %xmm9, %xmm9
movddup %xmm13, %xmm5
unpckhpd %xmm13, %xmm13
mulpd %xmm7, %xmm4
mulpd %xmm7, %xmm9
mulpd %xmm7, %xmm5
mulpd %xmm7, %xmm13
addpd %xmm4, %xmm0
addpd %xmm9, %xmm1
addpd %xmm5, %xmm2
addpd %xmm13, %xmm3
movsd %xmm0, 0 * SIZE(CO2)
movhpd %xmm0, 1 * SIZE(CO2)
movsd %xmm1, 2 * SIZE(CO2)
movhpd %xmm1, 3 * SIZE(CO2)
movsd %xmm2, 4 * SIZE(CO2)
movhpd %xmm2, 5 * SIZE(CO2)
movsd %xmm3, 6 * SIZE(CO2)
movhpd %xmm3, 7 * SIZE(CO2)
addq $8 * SIZE, CO1
addq $8 * SIZE, CO2
subq $1, I
jg .L50
ALIGN_4
.L60:
testq $2, M
jle .L70
#if !defined(TRMMKERNEL) || \
(defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \
(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
leaq 16 * SIZE + BUFFER, BO
#else
leaq 16 * SIZE + BUFFER, BO
movq KK, %rax
leaq (, %rax, SIZE), %rax
leaq (AO, %rax, 2), AO
leaq (BO, %rax, 4), BO
#endif
pxor %xmm8, %xmm8
movapd -16 * SIZE(AO), %xmm0
pxor %xmm9, %xmm9
movapd -14 * SIZE(AO), %xmm1
pxor %xmm10, %xmm10
pxor %xmm11, %xmm11
movapd %xmm8, %xmm2
movapd %xmm8, %xmm3
movapd %xmm8, %xmm4
movapd %xmm8, %xmm5
#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
#else
addq $2, %rax
#endif
movq %rax, KKK
#endif
sarq $2, %rax
jle .L65
ALIGN_4
.L61:
PREFETCH (PREFETCHSIZE + 0) * SIZE(AO)
addpd %xmm2, %xmm8
movapd -16 * SIZE(BO), %xmm2
mulpd %xmm0, %xmm2
addpd %xmm3, %xmm9
movapd -14 * SIZE(BO), %xmm3
mulpd %xmm0, %xmm3
movapd -12 * SIZE(AO), %xmm0
addpd %xmm4, %xmm10
movapd -12 * SIZE(BO), %xmm4
mulpd %xmm1, %xmm4
addpd %xmm5, %xmm11
movapd -10 * SIZE(BO), %xmm5
mulpd %xmm1, %xmm5
movapd -10 * SIZE(AO), %xmm1
addpd %xmm2, %xmm8
movapd -8 * SIZE(BO), %xmm2
mulpd %xmm0, %xmm2
addpd %xmm3, %xmm9
movapd -6 * SIZE(BO), %xmm3
mulpd %xmm0, %xmm3
movapd -8 * SIZE(AO), %xmm0
addpd %xmm4, %xmm10
movapd -4 * SIZE(BO), %xmm4
mulpd %xmm1, %xmm4
addpd %xmm5, %xmm11
movapd -2 * SIZE(BO), %xmm5
mulpd %xmm1, %xmm5
movapd -6 * SIZE(AO), %xmm1
subq $ -8 * SIZE, AO
subq $-16 * SIZE, BO
subq $1, %rax
jg .L61
ALIGN_4
.L65:
movapd ALPHA, %xmm7
#ifndef TRMMKERNEL
movq K, %rax
#else
movq KKK, %rax
#endif
andq $3, %rax # if (k & 1)
je .L68
ALIGN_4
.L66:
addpd %xmm2, %xmm8
movapd -16 * SIZE(BO), %xmm2
mulpd %xmm0, %xmm2
addpd %xmm3, %xmm9
movapd -14 * SIZE(BO), %xmm3
mulpd %xmm0, %xmm3
movapd -14 * SIZE(AO), %xmm0
addq $2 * SIZE, AO
addq $4 * SIZE, BO
subq $1, %rax
jg .L66
ALIGN_4
.L68:
addpd %xmm2, %xmm8
addpd %xmm3, %xmm9
addpd %xmm4, %xmm10
addpd %xmm5, %xmm11
addpd %xmm10, %xmm8
addpd %xmm11, %xmm9
movsd 0 * SIZE(CO1), %xmm0
movhpd 1 * SIZE(CO1), %xmm0
movsd 2 * SIZE(CO1), %xmm1
movhpd 3 * SIZE(CO1), %xmm1
movddup %xmm8, %xmm4
unpckhpd %xmm8, %xmm8
mulpd %xmm7, %xmm4
mulpd %xmm7, %xmm8
addpd %xmm4, %xmm0
addpd %xmm8, %xmm1
movsd %xmm0, 0 * SIZE(CO1)
movhpd %xmm0, 1 * SIZE(CO1)
movsd %xmm1, 2 * SIZE(CO1)
movhpd %xmm1, 3 * SIZE(CO1)
movsd 0 * SIZE(CO2), %xmm0
movhpd 1 * SIZE(CO2), %xmm0
movsd 2 * SIZE(CO2), %xmm1
movhpd 3 * SIZE(CO2), %xmm1
movddup %xmm9, %xmm4
unpckhpd %xmm9, %xmm9
mulpd %xmm7, %xmm4
mulpd %xmm7, %xmm9
addpd %xmm4, %xmm0
addpd %xmm9, %xmm1
movsd %xmm0, 0 * SIZE(CO2)
movhpd %xmm0, 1 * SIZE(CO2)
movsd %xmm1, 2 * SIZE(CO2)
movhpd %xmm1, 3 * SIZE(CO2)
addq $4 * SIZE, CO1 # coffset += 4
addq $4 * SIZE, CO2 # coffset += 4
ALIGN_4
.L70:
testq $1, M
jle .L79
#if !defined(TRMMKERNEL) || \
(defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \
(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
leaq 16 * SIZE + BUFFER, BO
#else
leaq 16 * SIZE + BUFFER, BO
movq KK, %rax
leaq (, %rax, SIZE), %rax
leaq (AO, %rax, 1), AO
leaq (BO, %rax, 4), BO
#endif
movsd -16 * SIZE(AO), %xmm0
movsd -15 * SIZE(AO), %xmm1
pxor %xmm8, %xmm8
pxor %xmm9, %xmm9
pxor %xmm10, %xmm10
pxor %xmm11, %xmm11
movapd %xmm8, %xmm2
movapd %xmm8, %xmm3
movapd %xmm8, %xmm4
movapd %xmm8, %xmm5
#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
#else
addq $2, %rax
#endif
movq %rax, KKK
#endif
sarq $2, %rax
jle .L75
ALIGN_4
.L71:
PREFETCH (PREFETCHSIZE + 0) * SIZE(AO)
addsd %xmm2, %xmm8
movsd -16 * SIZE(BO), %xmm2
mulsd %xmm0, %xmm2
addsd %xmm3, %xmm9
movsd -14 * SIZE(BO), %xmm3
mulsd %xmm0, %xmm3
movsd -14 * SIZE(AO), %xmm0
addsd %xmm4, %xmm10
movsd -12 * SIZE(BO), %xmm4
mulsd %xmm1, %xmm4
addsd %xmm5, %xmm11
movsd -10 * SIZE(BO), %xmm5
mulsd %xmm1, %xmm5
movsd -13 * SIZE(AO), %xmm1
addsd %xmm2, %xmm8
movsd -8 * SIZE(BO), %xmm2
mulsd %xmm0, %xmm2
addsd %xmm3, %xmm9
movsd -6 * SIZE(BO), %xmm3
mulsd %xmm0, %xmm3
movsd -12 * SIZE(AO), %xmm0
addsd %xmm4, %xmm10
movsd -4 * SIZE(BO), %xmm4
mulsd %xmm1, %xmm4
addsd %xmm5, %xmm11
movsd -2 * SIZE(BO), %xmm5
mulsd %xmm1, %xmm5
movsd -11 * SIZE(AO), %xmm1
subq $ -4 * SIZE, AO
subq $-16 * SIZE, BO
subq $1, %rax
jg .L71
ALIGN_4
.L75:
movapd ALPHA, %xmm7
#ifndef TRMMKERNEL
movq K, %rax
#else
movq KKK, %rax
#endif
andq $3, %rax # if (k & 1)
je .L78
ALIGN_4
.L76:
addsd %xmm2, %xmm8
movsd -16 * SIZE(BO), %xmm2
mulsd %xmm0, %xmm2
addsd %xmm3, %xmm9
movsd -14 * SIZE(BO), %xmm3
mulsd %xmm0, %xmm3
movsd -15 * SIZE(AO), %xmm0
addq $1 * SIZE, AO
addq $4 * SIZE, BO
subq $1, %rax
jg .L76
ALIGN_4
.L78:
addsd %xmm2, %xmm8
addsd %xmm3, %xmm9
addsd %xmm4, %xmm10
addsd %xmm5, %xmm11
addsd %xmm10, %xmm8
addsd %xmm11, %xmm9
movsd 0 * SIZE(CO1), %xmm0
movhpd 1 * SIZE(CO1), %xmm0
movddup %xmm8, %xmm4
mulpd %xmm7, %xmm4
addpd %xmm4, %xmm0
movsd %xmm0, 0 * SIZE(CO1)
movhpd %xmm0, 1 * SIZE(CO1)
movsd 0 * SIZE(CO2), %xmm0
movhpd 1 * SIZE(CO2), %xmm0
movddup %xmm9, %xmm4
mulpd %xmm7, %xmm4
addpd %xmm4, %xmm0
movsd %xmm0, 0 * SIZE(CO2)
movhpd %xmm0, 1 * SIZE(CO2)
ALIGN_4
.L79:
#if defined(TRMMKERNEL) && !defined(LEFT)
addl $2, KK
#endif
leaq (C, LDC, 2), C
ALIGN_4
.L80:
testq $1, N
BRANCH
jle .L999
ALIGN_4
.L81:
/* Copying to Sub Buffer */
leaq BUFFER, BO
#if defined(TRMMKERNEL) && defined(LEFT)
movq OFFSET, %rax
movq %rax, KK
#endif
movq K, %rax
sarq $4, %rax
jle .L83
addq %rax, %rax
ALIGN_4
.L82:
movddup -16 * SIZE(B), %xmm8
movddup -15 * SIZE(B), %xmm9
movddup -14 * SIZE(B), %xmm10
movddup -13 * SIZE(B), %xmm11
movddup -12 * SIZE(B), %xmm12
movddup -11 * SIZE(B), %xmm13
movddup -10 * SIZE(B), %xmm14
movddup -9 * SIZE(B), %xmm15
movapd %xmm8, 0 * SIZE(BO)
movapd %xmm9, 2 * SIZE(BO)
movapd %xmm10, 4 * SIZE(BO)
movapd %xmm11, 6 * SIZE(BO)
movapd %xmm12, 8 * SIZE(BO)
movapd %xmm13, 10 * SIZE(BO)
movapd %xmm14, 12 * SIZE(BO)
movapd %xmm15, 14 * SIZE(BO)
addq $ 8 * SIZE, B
subq $-16 * SIZE, BO
subq $1, %rax
jne .L82
ALIGN_4
.L83:
movq K, %rax
andq $15, %rax
BRANCH
jle .L85
ALIGN_4
.L84:
movddup -16 * SIZE(B), %xmm8
movapd %xmm8, 0 * SIZE(BO)
addq $1 * SIZE, B
addq $2 * SIZE, BO
subq $1, %rax
jne .L84
ALIGN_4
.L85:
movq C, CO1
movq A, AO
movq M, I
sarq $2, I
jle .L100
ALIGN_4
.L90:
#if !defined(TRMMKERNEL) || \
(defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \
(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
leaq 16 * SIZE + BUFFER, BO
#else
leaq 16 * SIZE + BUFFER, BO
movq KK, %rax
leaq (, %rax, SIZE), %rax
leaq (AO, %rax, 4), AO
leaq (BO, %rax, 2), BO
#endif
pxor %xmm8, %xmm8
movapd -16 * SIZE(BO), %xmm4
pxor %xmm9, %xmm9
movapd -14 * SIZE(BO), %xmm5
pxor %xmm12, %xmm12
movapd -12 * SIZE(BO), %xmm6
pxor %xmm13, %xmm13
movapd -10 * SIZE(BO), %xmm7
movapd %xmm8, %xmm0
prefetcht0 3 * SIZE(CO1)
movapd %xmm8, %xmm1
movapd %xmm8, %xmm2
movapd %xmm8, %xmm3
#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 $4, %rax
#else
addq $1, %rax
#endif
movq %rax, KKK
#endif
sarq $2, %rax
jle .L95
ALIGN_4
.L91:
PREFETCH (PREFETCHSIZE + 0) * SIZE(AO)
addpd %xmm0, %xmm8
movapd -16 * SIZE(AO), %xmm0
mulpd %xmm4, %xmm0
addpd %xmm1, %xmm12
movapd -14 * SIZE(AO), %xmm1
mulpd %xmm4, %xmm1
movapd -8 * SIZE(BO), %xmm4
addpd %xmm2, %xmm9
movapd -12 * SIZE(AO), %xmm2
mulpd %xmm5, %xmm2
addpd %xmm3, %xmm13
movapd -10 * SIZE(AO), %xmm3
mulpd %xmm5, %xmm3
movapd -6 * SIZE(BO), %xmm5
PREFETCH (PREFETCHSIZE + 8) * SIZE(AO)
addpd %xmm0, %xmm8
movapd -8 * SIZE(AO), %xmm0
mulpd %xmm6, %xmm0
addpd %xmm1, %xmm12
movapd -6 * SIZE(AO), %xmm1
mulpd %xmm6, %xmm1
movapd -4 * SIZE(BO), %xmm6
addpd %xmm2, %xmm9
movapd -4 * SIZE(AO), %xmm2
mulpd %xmm7, %xmm2
addpd %xmm3, %xmm13
movapd -2 * SIZE(AO), %xmm3
mulpd %xmm7, %xmm3
movapd -2 * SIZE(BO), %xmm7
subq $-16 * SIZE, AO
subq $ -8 * SIZE, BO
subq $1, %rax
jg .L91
ALIGN_4
.L95:
movapd ALPHA, %xmm7
#ifndef TRMMKERNEL
movq K, %rax
#else
movq KKK, %rax
#endif
andq $3, %rax # if (k & 1)
je .L98
ALIGN_4
.L96:
addpd %xmm0, %xmm8
movapd -16 * SIZE(AO), %xmm0
mulpd %xmm4, %xmm0
addpd %xmm1, %xmm12
movapd -14 * SIZE(AO), %xmm1
mulpd %xmm4, %xmm1
movapd -14 * SIZE(BO), %xmm4
addq $4 * SIZE, AO
addq $2 * SIZE, BO
subq $1, %rax
jg .L96
ALIGN_4
.L98:
addpd %xmm0, %xmm8
addpd %xmm1, %xmm12
addpd %xmm2, %xmm9
addpd %xmm3, %xmm13
addpd %xmm9, %xmm8
addpd %xmm13, %xmm12
movsd 0 * SIZE(CO1), %xmm0
movhpd 1 * SIZE(CO1), %xmm0
movsd 2 * SIZE(CO1), %xmm1
movhpd 3 * SIZE(CO1), %xmm1
movsd 4 * SIZE(CO1), %xmm2
movhpd 5 * SIZE(CO1), %xmm2
movsd 6 * SIZE(CO1), %xmm3
movhpd 7 * SIZE(CO1), %xmm3
movddup %xmm8, %xmm4
unpckhpd %xmm8, %xmm8
movddup %xmm12, %xmm5
unpckhpd %xmm12, %xmm12
mulpd %xmm7, %xmm4
mulpd %xmm7, %xmm8
mulpd %xmm7, %xmm5
mulpd %xmm7, %xmm12
addpd %xmm4, %xmm0
addpd %xmm8, %xmm1
addpd %xmm5, %xmm2
addpd %xmm12, %xmm3
movsd %xmm0, 0 * SIZE(CO1)
movhpd %xmm0, 1 * SIZE(CO1)
movsd %xmm1, 2 * SIZE(CO1)
movhpd %xmm1, 3 * SIZE(CO1)
movsd %xmm2, 4 * SIZE(CO1)
movhpd %xmm2, 5 * SIZE(CO1)
movsd %xmm3, 6 * SIZE(CO1)
movhpd %xmm3, 7 * SIZE(CO1)
addq $8 * SIZE, CO1 # coffset += 4
subq $1, I
jg .L90
ALIGN_4
.L100:
testq $2, M
jle .L110
#if !defined(TRMMKERNEL) || \
(defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \
(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
leaq 16 * SIZE + BUFFER, BO
#else
leaq 16 * SIZE + BUFFER, BO
movq KK, %rax
leaq (, %rax, SIZE), %rax
leaq (AO, %rax, 2), AO
leaq (BO, %rax, 2), BO
#endif
pxor %xmm8, %xmm8
movapd -16 * SIZE(BO), %xmm4
pxor %xmm9, %xmm9
movapd -14 * SIZE(BO), %xmm5
pxor %xmm10, %xmm10
movapd -12 * SIZE(BO), %xmm6
pxor %xmm11, %xmm11
movapd -10 * SIZE(BO), %xmm7
movapd %xmm8, %xmm0
movapd %xmm8, %xmm1
movapd %xmm8, %xmm2
movapd %xmm8, %xmm3
#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
#else
addq $1, %rax
#endif
movq %rax, KKK
#endif
sarq $2, %rax
jle .L105
ALIGN_4
.L101:
PREFETCH (PREFETCHSIZE + 0) * SIZE(AO)
addpd %xmm0, %xmm8
movapd -16 * SIZE(AO), %xmm0
mulpd %xmm4, %xmm0
movapd -8 * SIZE(BO), %xmm4
addpd %xmm1, %xmm9
movapd -14 * SIZE(AO), %xmm1
mulpd %xmm5, %xmm1
movapd -6 * SIZE(BO), %xmm5
addpd %xmm2, %xmm10
movapd -12 * SIZE(AO), %xmm2
mulpd %xmm6, %xmm2
movapd -4 * SIZE(BO), %xmm6
addpd %xmm3, %xmm11
movapd -10 * SIZE(AO), %xmm3
mulpd %xmm7, %xmm3
movapd -2 * SIZE(BO), %xmm7
subq $-8 * SIZE, AO
subq $-8 * SIZE, BO
subq $1, %rax
jg .L101
ALIGN_4
.L105:
movapd ALPHA, %xmm7
#ifndef TRMMKERNEL
movq K, %rax
#else
movq KKK, %rax
#endif
andq $3, %rax # if (k & 1)
je .L108
ALIGN_4
.L106:
addpd %xmm0, %xmm8
movapd -16 * SIZE(AO), %xmm0
mulpd %xmm4, %xmm0
movapd -14 * SIZE(BO), %xmm4
addq $2 * SIZE, AO
addq $2 * SIZE, BO
subq $1, %rax
jg .L106
ALIGN_4
.L108:
addpd %xmm0, %xmm8
addpd %xmm1, %xmm9
addpd %xmm2, %xmm10
addpd %xmm3, %xmm11
addpd %xmm10, %xmm8
addpd %xmm11, %xmm9
addpd %xmm9, %xmm8
movsd 0 * SIZE(CO1), %xmm0
movhpd 1 * SIZE(CO1), %xmm0
movsd 2 * SIZE(CO1), %xmm1
movhpd 3 * SIZE(CO1), %xmm1
movddup %xmm8, %xmm4
unpckhpd %xmm8, %xmm8
mulpd %xmm7, %xmm4
mulpd %xmm7, %xmm8
addpd %xmm4, %xmm0
addpd %xmm8, %xmm1
movsd %xmm0, 0 * SIZE(CO1)
movhpd %xmm0, 1 * SIZE(CO1)
movsd %xmm1, 2 * SIZE(CO1)
movhpd %xmm1, 3 * SIZE(CO1)
addq $4 * SIZE, CO1
ALIGN_4
.L110:
testq $1, M
jle .L999
#if !defined(TRMMKERNEL) || \
(defined(TRMMKERNEL) && defined(LEFT) && defined(TRANSA)) || \
(defined(TRMMKERNEL) && !defined(LEFT) && !defined(TRANSA))
leaq 16 * SIZE + BUFFER, BO
#else
leaq 16 * SIZE + BUFFER, BO
movq KK, %rax
leaq (, %rax, SIZE), %rax
leaq (AO, %rax, 1), AO
leaq (BO, %rax, 2), BO
#endif
pxor %xmm8, %xmm8
movsd -16 * SIZE(BO), %xmm4
pxor %xmm9, %xmm9
movsd -14 * SIZE(BO), %xmm5
pxor %xmm10, %xmm10
movsd -12 * SIZE(BO), %xmm6
pxor %xmm11, %xmm11
movsd -10 * SIZE(BO), %xmm7
movapd %xmm8, %xmm0
movapd %xmm8, %xmm1
movapd %xmm8, %xmm2
movapd %xmm8, %xmm3
#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
#else
addq $1, %rax
#endif
movq %rax, KKK
#endif
sarq $2, %rax
jle .L115
ALIGN_4
.L111:
PREFETCH (PREFETCHSIZE + 0) * SIZE(AO)
addpd %xmm0, %xmm8
movsd -16 * SIZE(AO), %xmm0
mulpd %xmm4, %xmm0
movsd -8 * SIZE(BO), %xmm4
addpd %xmm1, %xmm9
movsd -15 * SIZE(AO), %xmm1
mulpd %xmm5, %xmm1
movsd -6 * SIZE(BO), %xmm5
addpd %xmm2, %xmm10
movsd -14 * SIZE(AO), %xmm2
mulpd %xmm6, %xmm2
movsd -4 * SIZE(BO), %xmm6
addpd %xmm3, %xmm11
movsd -13 * SIZE(AO), %xmm3
mulpd %xmm7, %xmm3
movsd -2 * SIZE(BO), %xmm7
subq $-4 * SIZE, AO
subq $-8 * SIZE, BO
subq $1, %rax
jg .L111
ALIGN_4
.L115:
movapd ALPHA, %xmm7
#ifndef TRMMKERNEL
movq K, %rax
#else
movq KKK, %rax
#endif
andq $3, %rax # if (k & 1)
je .L118
ALIGN_4
.L116:
addsd %xmm0, %xmm8
movsd -16 * SIZE(AO), %xmm0
mulsd %xmm4, %xmm0
movsd -14 * SIZE(BO), %xmm4
addq $1 * SIZE, AO
addq $2 * SIZE, BO
subq $1, %rax
jg .L116
ALIGN_4
.L118:
addsd %xmm0, %xmm8
addsd %xmm1, %xmm9
addsd %xmm2, %xmm10
addsd %xmm3, %xmm11
addsd %xmm10, %xmm8
addsd %xmm11, %xmm9
addsd %xmm9, %xmm8
movsd 0 * SIZE(CO1), %xmm0
movhpd 1 * SIZE(CO1), %xmm0
movddup %xmm8, %xmm4
mulpd %xmm7, %xmm4
addpd %xmm4, %xmm0
movsd %xmm0, 0 * SIZE(CO1)
movhpd %xmm0, 1 * SIZE(CO1)
ALIGN_4
.L999:
movq %r15, %rsp
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