Merge pull request #233 from wernsaar/develop
added dgemv_n and some faster gemm_copy routines to BULLDOZER.
This commit is contained in:
Zhang Xianyi
commit
646e168d26
|
|
@ -1,20 +1,22 @@
|
|||
ZGEMVNKERNEL = zgemv_n_dup.S
|
||||
ZGEMVTKERNEL = zgemv_t_dup.S
|
||||
|
||||
DGEMVNKERNEL = dgemv_n_bulldozer.S
|
||||
|
||||
SGEMMKERNEL = sgemm_kernel_16x2_bulldozer.S
|
||||
SGEMMINCOPY = ../generic/gemm_ncopy_16.c
|
||||
SGEMMITCOPY = ../generic/gemm_tcopy_16.c
|
||||
SGEMMONCOPY = ../generic/gemm_ncopy_2.c
|
||||
SGEMMOTCOPY = ../generic/gemm_tcopy_2.c
|
||||
SGEMMONCOPY = gemm_ncopy_2_bulldozer.S
|
||||
SGEMMOTCOPY = gemm_tcopy_2_bulldozer.S
|
||||
SGEMMINCOPYOBJ = sgemm_incopy$(TSUFFIX).$(SUFFIX)
|
||||
SGEMMITCOPYOBJ = sgemm_itcopy$(TSUFFIX).$(SUFFIX)
|
||||
SGEMMONCOPYOBJ = sgemm_oncopy$(TSUFFIX).$(SUFFIX)
|
||||
SGEMMOTCOPYOBJ = sgemm_otcopy$(TSUFFIX).$(SUFFIX)
|
||||
DGEMMKERNEL = dgemm_kernel_8x2_bulldozer.S
|
||||
DGEMMINCOPY = ../generic/gemm_ncopy_8.c
|
||||
DGEMMITCOPY = ../generic/gemm_tcopy_8.c
|
||||
DGEMMONCOPY = ../generic/gemm_ncopy_2.c
|
||||
DGEMMOTCOPY = ../generic/gemm_tcopy_2.c
|
||||
DGEMMINCOPY = dgemm_ncopy_8_bulldozer.S
|
||||
DGEMMITCOPY = dgemm_tcopy_8_bulldozer.S
|
||||
DGEMMONCOPY = gemm_ncopy_2_bulldozer.S
|
||||
DGEMMOTCOPY = gemm_tcopy_2_bulldozer.S
|
||||
DGEMMINCOPYOBJ = dgemm_incopy$(TSUFFIX).$(SUFFIX)
|
||||
DGEMMITCOPYOBJ = dgemm_itcopy$(TSUFFIX).$(SUFFIX)
|
||||
DGEMMONCOPYOBJ = dgemm_oncopy$(TSUFFIX).$(SUFFIX)
|
||||
|
|
|
|||
File diff suppressed because it is too large
Load Diff
|
|
@ -0,0 +1,667 @@
|
|||
/*********************************************************************/
|
||||
/* 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 VMOVUPS_A1(OFF, ADDR, REGS) vmovups OFF(ADDR), REGS
|
||||
#define VMOVUPS_A2(OFF, ADDR, BASE, SCALE, REGS) vmovups OFF(ADDR, BASE, SCALE), REGS
|
||||
|
||||
#define A_PRE 256
|
||||
|
||||
#ifndef WINDOWS_ABI
|
||||
|
||||
#define N ARG1 /* rsi */
|
||||
#define M ARG2 /* rdi */
|
||||
#define A ARG3 /* rdx */
|
||||
#define LDA ARG4 /* rcx */
|
||||
#define B ARG5 /* r8 */
|
||||
|
||||
#define AO1 %r9
|
||||
#define AO2 %r10
|
||||
#define LDA3 %r11
|
||||
#define M8 %r12
|
||||
|
||||
#else
|
||||
|
||||
#define N ARG1 /* rdx */
|
||||
#define M ARG2 /* rcx */
|
||||
#define A ARG3 /* r8 */
|
||||
#define LDA ARG4 /* r9 */
|
||||
#define OLD_B 40 + 56(%rsp)
|
||||
|
||||
#define B %r12
|
||||
|
||||
#define AO1 %rsi
|
||||
#define AO2 %rdi
|
||||
#define LDA3 %r10
|
||||
#define M8 %r11
|
||||
#endif
|
||||
|
||||
#define I %rax
|
||||
|
||||
#define B0 %rbp
|
||||
#define B1 %r13
|
||||
#define B2 %r14
|
||||
#define B3 %r15
|
||||
|
||||
PROLOGUE
|
||||
PROFCODE
|
||||
|
||||
#ifdef WINDOWS_ABI
|
||||
pushq %rdi
|
||||
pushq %rsi
|
||||
#endif
|
||||
|
||||
pushq %r15
|
||||
pushq %r14
|
||||
pushq %r13
|
||||
pushq %r12
|
||||
pushq %rbp
|
||||
|
||||
#ifdef WINDOWS_ABI
|
||||
movq OLD_B, B
|
||||
#endif
|
||||
|
||||
subq $-16 * SIZE, B
|
||||
|
||||
movq M, B1
|
||||
movq M, B2
|
||||
movq M, B3
|
||||
|
||||
andq $-8, B1
|
||||
andq $-4, B2
|
||||
andq $-2, B3
|
||||
|
||||
imulq N, B1
|
||||
imulq N, B2
|
||||
imulq N, B3
|
||||
|
||||
leaq (B, B1, SIZE), B1
|
||||
leaq (B, B2, SIZE), B2
|
||||
leaq (B, B3, SIZE), B3
|
||||
|
||||
leaq (,LDA, SIZE), LDA
|
||||
leaq (LDA, LDA, 2), LDA3
|
||||
|
||||
leaq (, N, SIZE), M8
|
||||
|
||||
cmpq $8, N
|
||||
jl .L20
|
||||
ALIGN_4
|
||||
|
||||
.L11:
|
||||
subq $8, N
|
||||
|
||||
movq A, AO1
|
||||
leaq (A, LDA, 4), AO2
|
||||
leaq (A, LDA, 8), A
|
||||
|
||||
movq B, B0
|
||||
addq $64 * SIZE, B
|
||||
|
||||
movq M, I
|
||||
sarq $3, I
|
||||
jle .L14
|
||||
ALIGN_4
|
||||
|
||||
.L13:
|
||||
|
||||
prefetchnta A_PRE(AO1)
|
||||
VMOVUPS_A1(0 * SIZE, AO1, %xmm0)
|
||||
VMOVUPS_A1(2 * SIZE, AO1, %xmm1)
|
||||
VMOVUPS_A1(4 * SIZE, AO1, %xmm2)
|
||||
VMOVUPS_A1(6 * SIZE, AO1, %xmm3)
|
||||
|
||||
vmovups %xmm0, -16 * SIZE(B0)
|
||||
vmovups %xmm1, -14 * SIZE(B0)
|
||||
vmovups %xmm2, -12 * SIZE(B0)
|
||||
vmovups %xmm3, -10 * SIZE(B0)
|
||||
|
||||
|
||||
prefetchnta A_PRE(AO1, LDA, 1)
|
||||
VMOVUPS_A2(0 * SIZE, AO1, LDA, 1, %xmm0)
|
||||
VMOVUPS_A2(2 * SIZE, AO1, LDA, 1, %xmm1)
|
||||
VMOVUPS_A2(4 * SIZE, AO1, LDA, 1, %xmm2)
|
||||
VMOVUPS_A2(6 * SIZE, AO1, LDA, 1, %xmm3)
|
||||
|
||||
vmovups %xmm0, -8 * SIZE(B0)
|
||||
vmovups %xmm1, -6 * SIZE(B0)
|
||||
vmovups %xmm2, -4 * SIZE(B0)
|
||||
vmovups %xmm3, -2 * SIZE(B0)
|
||||
|
||||
|
||||
prefetchnta A_PRE(AO1, LDA, 2)
|
||||
VMOVUPS_A2(0 * SIZE, AO1, LDA, 2, %xmm0)
|
||||
VMOVUPS_A2(2 * SIZE, AO1, LDA, 2, %xmm1)
|
||||
VMOVUPS_A2(4 * SIZE, AO1, LDA, 2, %xmm2)
|
||||
VMOVUPS_A2(6 * SIZE, AO1, LDA, 2, %xmm3)
|
||||
|
||||
|
||||
vmovups %xmm0, 0 * SIZE(B0)
|
||||
vmovups %xmm1, 2 * SIZE(B0)
|
||||
vmovups %xmm2, 4 * SIZE(B0)
|
||||
vmovups %xmm3, 6 * SIZE(B0)
|
||||
|
||||
|
||||
prefetchnta A_PRE(AO1, LDA3, 1)
|
||||
VMOVUPS_A2(0 * SIZE, AO1, LDA3, 1, %xmm0)
|
||||
VMOVUPS_A2(2 * SIZE, AO1, LDA3, 1, %xmm1)
|
||||
VMOVUPS_A2(4 * SIZE, AO1, LDA3, 1, %xmm2)
|
||||
VMOVUPS_A2(6 * SIZE, AO1, LDA3, 1, %xmm3)
|
||||
|
||||
vmovups %xmm0, 8 * SIZE(B0)
|
||||
vmovups %xmm1, 10 * SIZE(B0)
|
||||
vmovups %xmm2, 12 * SIZE(B0)
|
||||
vmovups %xmm3, 14 * SIZE(B0)
|
||||
|
||||
prefetchnta A_PRE(AO2)
|
||||
VMOVUPS_A1(0 * SIZE, AO2, %xmm0)
|
||||
VMOVUPS_A1(2 * SIZE, AO2, %xmm1)
|
||||
VMOVUPS_A1(4 * SIZE, AO2, %xmm2)
|
||||
VMOVUPS_A1(6 * SIZE, AO2, %xmm3)
|
||||
|
||||
vmovups %xmm0, 16 * SIZE(B0)
|
||||
vmovups %xmm1, 18 * SIZE(B0)
|
||||
vmovups %xmm2, 20 * SIZE(B0)
|
||||
vmovups %xmm3, 22 * SIZE(B0)
|
||||
|
||||
prefetchnta A_PRE(AO2, LDA, 1)
|
||||
VMOVUPS_A2(0 * SIZE, AO2, LDA, 1, %xmm0)
|
||||
VMOVUPS_A2(2 * SIZE, AO2, LDA, 1, %xmm1)
|
||||
VMOVUPS_A2(4 * SIZE, AO2, LDA, 1, %xmm2)
|
||||
VMOVUPS_A2(6 * SIZE, AO2, LDA, 1, %xmm3)
|
||||
|
||||
vmovups %xmm0, 24 * SIZE(B0)
|
||||
vmovups %xmm1, 26 * SIZE(B0)
|
||||
vmovups %xmm2, 28 * SIZE(B0)
|
||||
vmovups %xmm3, 30 * SIZE(B0)
|
||||
|
||||
prefetchnta A_PRE(AO2, LDA, 2)
|
||||
VMOVUPS_A2(0 * SIZE, AO2, LDA, 2, %xmm0)
|
||||
VMOVUPS_A2(2 * SIZE, AO2, LDA, 2, %xmm1)
|
||||
VMOVUPS_A2(4 * SIZE, AO2, LDA, 2, %xmm2)
|
||||
VMOVUPS_A2(6 * SIZE, AO2, LDA, 2, %xmm3)
|
||||
|
||||
vmovups %xmm0, 32 * SIZE(B0)
|
||||
vmovups %xmm1, 34 * SIZE(B0)
|
||||
vmovups %xmm2, 36 * SIZE(B0)
|
||||
vmovups %xmm3, 38 * SIZE(B0)
|
||||
|
||||
prefetchnta A_PRE(AO2, LDA3, 1)
|
||||
VMOVUPS_A2(0 * SIZE, AO2, LDA3, 1, %xmm0)
|
||||
VMOVUPS_A2(2 * SIZE, AO2, LDA3, 1, %xmm1)
|
||||
VMOVUPS_A2(4 * SIZE, AO2, LDA3, 1, %xmm2)
|
||||
VMOVUPS_A2(6 * SIZE, AO2, LDA3, 1, %xmm3)
|
||||
|
||||
vmovups %xmm0, 40 * SIZE(B0)
|
||||
vmovups %xmm1, 42 * SIZE(B0)
|
||||
vmovups %xmm2, 44 * SIZE(B0)
|
||||
vmovups %xmm3, 46 * SIZE(B0)
|
||||
|
||||
addq $8 * SIZE, AO1
|
||||
addq $8 * SIZE, AO2
|
||||
leaq (B0, M8, 8), B0
|
||||
|
||||
decq I
|
||||
jg .L13
|
||||
ALIGN_4
|
||||
|
||||
.L14:
|
||||
testq $4, M
|
||||
jle .L16
|
||||
|
||||
VMOVUPS_A1(0 * SIZE, AO1, %xmm0)
|
||||
VMOVUPS_A1(2 * SIZE, AO1, %xmm1)
|
||||
VMOVUPS_A2(0 * SIZE, AO1, LDA, 1, %xmm2)
|
||||
VMOVUPS_A2(2 * SIZE, AO1, LDA, 1, %xmm3)
|
||||
|
||||
vmovups %xmm0, -16 * SIZE(B1)
|
||||
vmovups %xmm1, -14 * SIZE(B1)
|
||||
vmovups %xmm2, -12 * SIZE(B1)
|
||||
vmovups %xmm3, -10 * SIZE(B1)
|
||||
|
||||
VMOVUPS_A2(0 * SIZE, AO1, LDA, 2, %xmm0)
|
||||
VMOVUPS_A2(2 * SIZE, AO1, LDA, 2, %xmm1)
|
||||
VMOVUPS_A2(0 * SIZE, AO1, LDA3, 1, %xmm2)
|
||||
VMOVUPS_A2(2 * SIZE, AO1, LDA3, 1, %xmm3)
|
||||
|
||||
vmovups %xmm0, -8 * SIZE(B1)
|
||||
vmovups %xmm1, -6 * SIZE(B1)
|
||||
vmovups %xmm2, -4 * SIZE(B1)
|
||||
vmovups %xmm3, -2 * SIZE(B1)
|
||||
|
||||
VMOVUPS_A1(0 * SIZE, AO2, %xmm0)
|
||||
VMOVUPS_A1(2 * SIZE, AO2, %xmm1)
|
||||
VMOVUPS_A2(0 * SIZE, AO2, LDA, 1, %xmm2)
|
||||
VMOVUPS_A2(2 * SIZE, AO2, LDA, 1, %xmm3)
|
||||
|
||||
vmovups %xmm0, 0 * SIZE(B1)
|
||||
vmovups %xmm1, 2 * SIZE(B1)
|
||||
vmovups %xmm2, 4 * SIZE(B1)
|
||||
vmovups %xmm3, 6 * SIZE(B1)
|
||||
|
||||
VMOVUPS_A2(0 * SIZE, AO2, LDA, 2, %xmm0)
|
||||
VMOVUPS_A2(2 * SIZE, AO2, LDA, 2, %xmm1)
|
||||
VMOVUPS_A2(0 * SIZE, AO2, LDA3, 1, %xmm2)
|
||||
VMOVUPS_A2(2 * SIZE, AO2, LDA3, 1, %xmm3)
|
||||
|
||||
vmovups %xmm0, 8 * SIZE(B1)
|
||||
vmovups %xmm1, 10 * SIZE(B1)
|
||||
vmovups %xmm2, 12 * SIZE(B1)
|
||||
vmovups %xmm3, 14 * SIZE(B1)
|
||||
|
||||
addq $4 * SIZE, AO1
|
||||
addq $4 * SIZE, AO2
|
||||
subq $-32 * SIZE, B1
|
||||
ALIGN_4
|
||||
|
||||
.L16:
|
||||
testq $2, M
|
||||
jle .L18
|
||||
|
||||
VMOVUPS_A1(0 * SIZE, AO1, %xmm0)
|
||||
VMOVUPS_A2(0 * SIZE, AO1, LDA, 1, %xmm1)
|
||||
VMOVUPS_A2(0 * SIZE, AO1, LDA, 2, %xmm2)
|
||||
VMOVUPS_A2(0 * SIZE, AO1, LDA3, 1, %xmm3)
|
||||
|
||||
vmovups %xmm0, -16 * SIZE(B2)
|
||||
vmovups %xmm1, -14 * SIZE(B2)
|
||||
vmovups %xmm2, -12 * SIZE(B2)
|
||||
vmovups %xmm3, -10 * SIZE(B2)
|
||||
|
||||
VMOVUPS_A1(0 * SIZE, AO2, %xmm0)
|
||||
VMOVUPS_A2(0 * SIZE, AO2, LDA, 1, %xmm1)
|
||||
VMOVUPS_A2(0 * SIZE, AO2, LDA, 2, %xmm2)
|
||||
VMOVUPS_A2(0 * SIZE, AO2, LDA3, 1, %xmm3)
|
||||
|
||||
vmovups %xmm0, -8 * SIZE(B2)
|
||||
vmovups %xmm1, -6 * SIZE(B2)
|
||||
vmovups %xmm2, -4 * SIZE(B2)
|
||||
vmovups %xmm3, -2 * SIZE(B2)
|
||||
|
||||
addq $2 * SIZE, AO1
|
||||
addq $2 * SIZE, AO2
|
||||
subq $-16 * SIZE, B2
|
||||
ALIGN_4
|
||||
|
||||
.L18:
|
||||
testq $1, M
|
||||
jle .L19
|
||||
|
||||
vmovsd 0 * SIZE(AO1), %xmm0
|
||||
vmovsd 0 * SIZE(AO1, LDA), %xmm1
|
||||
vmovsd 0 * SIZE(AO1, LDA, 2), %xmm2
|
||||
vmovsd 0 * SIZE(AO1, LDA3), %xmm3
|
||||
|
||||
vunpcklpd %xmm1, %xmm0 , %xmm0
|
||||
vunpcklpd %xmm3, %xmm2 , %xmm2
|
||||
|
||||
vmovups %xmm0, -16 * SIZE(B3)
|
||||
vmovups %xmm2, -14 * SIZE(B3)
|
||||
|
||||
vmovsd 0 * SIZE(AO2), %xmm0
|
||||
vmovsd 0 * SIZE(AO2, LDA), %xmm1
|
||||
vmovsd 0 * SIZE(AO2, LDA, 2), %xmm2
|
||||
vmovsd 0 * SIZE(AO2, LDA3), %xmm3
|
||||
|
||||
vunpcklpd %xmm1, %xmm0 , %xmm0
|
||||
vunpcklpd %xmm3, %xmm2 , %xmm2
|
||||
|
||||
vmovups %xmm0, -12 * SIZE(B3)
|
||||
vmovups %xmm2, -10 * SIZE(B3)
|
||||
|
||||
subq $-8 * SIZE, B3
|
||||
ALIGN_4
|
||||
|
||||
.L19:
|
||||
cmpq $8, N
|
||||
jge .L11
|
||||
ALIGN_4
|
||||
|
||||
.L20:
|
||||
cmpq $4, N
|
||||
jl .L30
|
||||
|
||||
subq $4, N
|
||||
|
||||
movq A, AO1
|
||||
leaq (A, LDA, 2), AO2
|
||||
leaq (A, LDA, 4), A
|
||||
|
||||
movq B, B0
|
||||
addq $32 * SIZE, B
|
||||
|
||||
movq M, I
|
||||
sarq $3, I
|
||||
jle .L24
|
||||
ALIGN_4
|
||||
|
||||
.L23:
|
||||
|
||||
VMOVUPS_A1(0 * SIZE, AO1, %xmm0)
|
||||
VMOVUPS_A1(2 * SIZE, AO1, %xmm1)
|
||||
VMOVUPS_A1(4 * SIZE, AO1, %xmm2)
|
||||
VMOVUPS_A1(6 * SIZE, AO1, %xmm3)
|
||||
|
||||
vmovups %xmm0, -16 * SIZE(B0)
|
||||
vmovups %xmm1, -14 * SIZE(B0)
|
||||
vmovups %xmm2, -12 * SIZE(B0)
|
||||
vmovups %xmm3, -10 * SIZE(B0)
|
||||
|
||||
|
||||
VMOVUPS_A2(0 * SIZE, AO1, LDA, 1, %xmm0)
|
||||
VMOVUPS_A2(2 * SIZE, AO1, LDA, 1, %xmm1)
|
||||
VMOVUPS_A2(4 * SIZE, AO1, LDA, 1, %xmm2)
|
||||
VMOVUPS_A2(6 * SIZE, AO1, LDA, 1, %xmm3)
|
||||
|
||||
vmovups %xmm0, -8 * SIZE(B0)
|
||||
vmovups %xmm1, -6 * SIZE(B0)
|
||||
vmovups %xmm2, -4 * SIZE(B0)
|
||||
vmovups %xmm3, -2 * SIZE(B0)
|
||||
|
||||
VMOVUPS_A1(0 * SIZE, AO2, %xmm0)
|
||||
VMOVUPS_A1(2 * SIZE, AO2, %xmm1)
|
||||
VMOVUPS_A1(4 * SIZE, AO2, %xmm2)
|
||||
VMOVUPS_A1(6 * SIZE, AO2, %xmm3)
|
||||
|
||||
vmovups %xmm0, 0 * SIZE(B0)
|
||||
vmovups %xmm1, 2 * SIZE(B0)
|
||||
vmovups %xmm2, 4 * SIZE(B0)
|
||||
vmovups %xmm3, 6 * SIZE(B0)
|
||||
|
||||
VMOVUPS_A2(0 * SIZE, AO2, LDA, 1, %xmm0)
|
||||
VMOVUPS_A2(2 * SIZE, AO2, LDA, 1, %xmm1)
|
||||
VMOVUPS_A2(4 * SIZE, AO2, LDA, 1, %xmm2)
|
||||
VMOVUPS_A2(6 * SIZE, AO2, LDA, 1, %xmm3)
|
||||
|
||||
vmovups %xmm0, 8 * SIZE(B0)
|
||||
vmovups %xmm1, 10 * SIZE(B0)
|
||||
vmovups %xmm2, 12 * SIZE(B0)
|
||||
vmovups %xmm3, 14 * SIZE(B0)
|
||||
|
||||
addq $8 * SIZE, AO1
|
||||
addq $8 * SIZE, AO2
|
||||
leaq (B0, M8, 8), B0
|
||||
|
||||
decq I
|
||||
jg .L23
|
||||
ALIGN_4
|
||||
|
||||
.L24:
|
||||
testq $4, M
|
||||
jle .L26
|
||||
|
||||
VMOVUPS_A1(0 * SIZE, AO1, %xmm0)
|
||||
VMOVUPS_A1(2 * SIZE, AO1, %xmm1)
|
||||
VMOVUPS_A2(0 * SIZE, AO1, LDA, 1, %xmm2)
|
||||
VMOVUPS_A2(2 * SIZE, AO1, LDA, 1, %xmm3)
|
||||
|
||||
vmovups %xmm0, -16 * SIZE(B1)
|
||||
vmovups %xmm1, -14 * SIZE(B1)
|
||||
vmovups %xmm2, -12 * SIZE(B1)
|
||||
vmovups %xmm3, -10 * SIZE(B1)
|
||||
|
||||
VMOVUPS_A1(0 * SIZE, AO2, %xmm0)
|
||||
VMOVUPS_A1(2 * SIZE, AO2, %xmm1)
|
||||
VMOVUPS_A2(0 * SIZE, AO2, LDA, 1, %xmm2)
|
||||
VMOVUPS_A2(2 * SIZE, AO2, LDA, 1, %xmm3)
|
||||
|
||||
vmovups %xmm0, -8 * SIZE(B1)
|
||||
vmovups %xmm1, -6 * SIZE(B1)
|
||||
vmovups %xmm2, -4 * SIZE(B1)
|
||||
vmovups %xmm3, -2 * SIZE(B1)
|
||||
|
||||
addq $4 * SIZE, AO1
|
||||
addq $4 * SIZE, AO2
|
||||
subq $-16 * SIZE, B1
|
||||
ALIGN_4
|
||||
|
||||
.L26:
|
||||
testq $2, M
|
||||
jle .L28
|
||||
|
||||
VMOVUPS_A1(0 * SIZE, AO1, %xmm0)
|
||||
VMOVUPS_A2(0 * SIZE, AO1, LDA, 1, %xmm1)
|
||||
VMOVUPS_A1(0 * SIZE, AO2, %xmm2)
|
||||
VMOVUPS_A2(0 * SIZE, AO2, LDA, 1, %xmm3)
|
||||
|
||||
vmovups %xmm0, -16 * SIZE(B2)
|
||||
vmovups %xmm1, -14 * SIZE(B2)
|
||||
vmovups %xmm2, -12 * SIZE(B2)
|
||||
vmovups %xmm3, -10 * SIZE(B2)
|
||||
|
||||
addq $2 * SIZE, AO1
|
||||
addq $2 * SIZE, AO2
|
||||
subq $-8 * SIZE, B2
|
||||
ALIGN_4
|
||||
|
||||
.L28:
|
||||
testq $1, M
|
||||
jle .L30
|
||||
|
||||
vmovsd 0 * SIZE(AO1), %xmm0
|
||||
vmovsd 0 * SIZE(AO1, LDA), %xmm1
|
||||
vmovsd 0 * SIZE(AO2), %xmm2
|
||||
vmovsd 0 * SIZE(AO2, LDA), %xmm3
|
||||
|
||||
vunpcklpd %xmm1, %xmm0, %xmm0
|
||||
vunpcklpd %xmm3, %xmm2, %xmm2
|
||||
|
||||
vmovups %xmm0, -16 * SIZE(B3)
|
||||
vmovups %xmm2, -14 * SIZE(B3)
|
||||
subq $-4 * SIZE, B3
|
||||
ALIGN_4
|
||||
|
||||
.L30:
|
||||
cmpq $2, N
|
||||
jl .L40
|
||||
|
||||
subq $2, N
|
||||
|
||||
movq A, AO1
|
||||
leaq (A, LDA), AO2
|
||||
leaq (A, LDA, 2), A
|
||||
|
||||
movq B, B0
|
||||
addq $16 * SIZE, B
|
||||
|
||||
movq M, I
|
||||
sarq $3, I
|
||||
jle .L34
|
||||
ALIGN_4
|
||||
|
||||
.L33:
|
||||
|
||||
VMOVUPS_A1(0 * SIZE, AO1, %xmm0)
|
||||
VMOVUPS_A1(2 * SIZE, AO1, %xmm1)
|
||||
VMOVUPS_A1(4 * SIZE, AO1, %xmm2)
|
||||
VMOVUPS_A1(6 * SIZE, AO1, %xmm3)
|
||||
|
||||
vmovups %xmm0, -16 * SIZE(B0)
|
||||
vmovups %xmm1, -14 * SIZE(B0)
|
||||
vmovups %xmm2, -12 * SIZE(B0)
|
||||
vmovups %xmm3, -10 * SIZE(B0)
|
||||
|
||||
VMOVUPS_A1(0 * SIZE, AO2, %xmm0)
|
||||
VMOVUPS_A1(2 * SIZE, AO2, %xmm1)
|
||||
VMOVUPS_A1(4 * SIZE, AO2, %xmm2)
|
||||
VMOVUPS_A1(6 * SIZE, AO2, %xmm3)
|
||||
|
||||
vmovups %xmm0, -8 * SIZE(B0)
|
||||
vmovups %xmm1, -6 * SIZE(B0)
|
||||
vmovups %xmm2, -4 * SIZE(B0)
|
||||
vmovups %xmm3, -2 * SIZE(B0)
|
||||
|
||||
addq $8 * SIZE, AO1
|
||||
addq $8 * SIZE, AO2
|
||||
leaq (B0, M8, 8), B0
|
||||
|
||||
decq I
|
||||
jg .L33
|
||||
ALIGN_4
|
||||
|
||||
.L34:
|
||||
testq $4, M
|
||||
jle .L36
|
||||
|
||||
VMOVUPS_A1(0 * SIZE, AO1, %xmm0)
|
||||
VMOVUPS_A1(2 * SIZE, AO1, %xmm1)
|
||||
VMOVUPS_A1(0 * SIZE, AO2, %xmm2)
|
||||
VMOVUPS_A1(2 * SIZE, AO2, %xmm3)
|
||||
|
||||
vmovups %xmm0, -16 * SIZE(B1)
|
||||
vmovups %xmm1, -14 * SIZE(B1)
|
||||
vmovups %xmm2, -12 * SIZE(B1)
|
||||
vmovups %xmm3, -10 * SIZE(B1)
|
||||
|
||||
addq $4 * SIZE, AO1
|
||||
addq $4 * SIZE, AO2
|
||||
subq $-8 * SIZE, B1
|
||||
ALIGN_4
|
||||
|
||||
.L36:
|
||||
testq $2, M
|
||||
jle .L38
|
||||
|
||||
VMOVUPS_A1(0 * SIZE, AO1, %xmm0)
|
||||
VMOVUPS_A1(0 * SIZE, AO2, %xmm1)
|
||||
|
||||
vmovups %xmm0, -16 * SIZE(B2)
|
||||
vmovups %xmm1, -14 * SIZE(B2)
|
||||
|
||||
addq $2 * SIZE, AO1
|
||||
addq $2 * SIZE, AO2
|
||||
subq $-4 * SIZE, B2
|
||||
ALIGN_4
|
||||
|
||||
.L38:
|
||||
testq $1, M
|
||||
jle .L40
|
||||
|
||||
vmovsd 0 * SIZE(AO1), %xmm0
|
||||
vmovsd 0 * SIZE(AO2), %xmm1
|
||||
|
||||
vunpcklpd %xmm1, %xmm0, %xmm0
|
||||
|
||||
vmovups %xmm0, -16 * SIZE(B3)
|
||||
subq $-2 * SIZE, B3
|
||||
ALIGN_4
|
||||
|
||||
.L40:
|
||||
cmpq $1, N
|
||||
jl .L999
|
||||
|
||||
movq A, AO1
|
||||
|
||||
movq B, B0
|
||||
|
||||
movq M, I
|
||||
sarq $3, I
|
||||
jle .L44
|
||||
ALIGN_4
|
||||
|
||||
.L43:
|
||||
|
||||
VMOVUPS_A1(0 * SIZE, AO1, %xmm0)
|
||||
VMOVUPS_A1(2 * SIZE, AO1, %xmm1)
|
||||
VMOVUPS_A1(4 * SIZE, AO1, %xmm2)
|
||||
VMOVUPS_A1(6 * SIZE, AO1, %xmm3)
|
||||
|
||||
vmovups %xmm0, -16 * SIZE(B0)
|
||||
vmovups %xmm1, -14 * SIZE(B0)
|
||||
vmovups %xmm2, -12 * SIZE(B0)
|
||||
vmovups %xmm3, -10 * SIZE(B0)
|
||||
|
||||
addq $8 * SIZE, AO1
|
||||
leaq (B0, M8, 8), B0
|
||||
|
||||
decq I
|
||||
jg .L43
|
||||
ALIGN_4
|
||||
|
||||
.L44:
|
||||
testq $4, M
|
||||
jle .L45
|
||||
|
||||
VMOVUPS_A1(0 * SIZE, AO1, %xmm0)
|
||||
VMOVUPS_A1(2 * SIZE, AO1, %xmm1)
|
||||
|
||||
vmovups %xmm0, -16 * SIZE(B1)
|
||||
vmovups %xmm1, -14 * SIZE(B1)
|
||||
|
||||
addq $4 * SIZE, AO1
|
||||
subq $-4 * SIZE, B1
|
||||
ALIGN_4
|
||||
|
||||
.L45:
|
||||
testq $2, M
|
||||
jle .L46
|
||||
|
||||
VMOVUPS_A1(0 * SIZE, AO1, %xmm0)
|
||||
|
||||
vmovups %xmm0, -16 * SIZE(B2)
|
||||
|
||||
addq $2 * SIZE, AO1
|
||||
subq $-2 * SIZE, B2
|
||||
ALIGN_4
|
||||
|
||||
.L46:
|
||||
testq $1, M
|
||||
jle .L999
|
||||
|
||||
vmovsd 0 * SIZE(AO1), %xmm0
|
||||
|
||||
vmovsd %xmm0, -16 * SIZE(B3)
|
||||
jmp .L999
|
||||
ALIGN_4
|
||||
|
||||
.L999:
|
||||
popq %rbp
|
||||
popq %r12
|
||||
popq %r13
|
||||
popq %r14
|
||||
popq %r15
|
||||
|
||||
#ifdef WINDOWS_ABI
|
||||
popq %rsi
|
||||
popq %rdi
|
||||
#endif
|
||||
ret
|
||||
|
||||
EPILOGUE
|
||||
File diff suppressed because it is too large
Load Diff
|
|
@ -0,0 +1,360 @@
|
|||
/*********************************************************************/
|
||||
/* 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"
|
||||
|
||||
|
||||
#ifndef WINDOWS_ABI
|
||||
|
||||
#define M ARG1 /* rdi */
|
||||
#define N ARG2 /* rsi */
|
||||
#define A ARG3 /* rdx */
|
||||
#define LDA ARG4 /* rcx */
|
||||
#define B ARG5 /* r8 */
|
||||
|
||||
#define I %r9
|
||||
|
||||
#else
|
||||
|
||||
#define STACKSIZE 256
|
||||
|
||||
#define M ARG1 /* rcx */
|
||||
#define N ARG2 /* rdx */
|
||||
#define A ARG3 /* r8 */
|
||||
#define LDA ARG4 /* r9 */
|
||||
#define OLD_B 40 + 32 + STACKSIZE(%rsp)
|
||||
|
||||
#define B %r14
|
||||
#define I %r15
|
||||
|
||||
#endif
|
||||
|
||||
#define J %r10
|
||||
#define AO1 %r11
|
||||
#define AO2 %r12
|
||||
#define AO3 %r13
|
||||
#define AO4 %rax
|
||||
|
||||
PROLOGUE
|
||||
PROFCODE
|
||||
|
||||
#ifdef WINDOWS_ABI
|
||||
pushq %r15
|
||||
pushq %r14
|
||||
#endif
|
||||
pushq %r13
|
||||
pushq %r12
|
||||
|
||||
#ifdef WINDOWS_ABI
|
||||
subq $STACKSIZE, %rsp
|
||||
|
||||
vmovups %xmm6, 0(%rsp)
|
||||
vmovups %xmm7, 16(%rsp)
|
||||
vmovups %xmm8, 32(%rsp)
|
||||
vmovups %xmm9, 48(%rsp)
|
||||
vmovups %xmm10, 64(%rsp)
|
||||
vmovups %xmm11, 80(%rsp)
|
||||
vmovups %xmm12, 96(%rsp)
|
||||
vmovups %xmm13, 112(%rsp)
|
||||
vmovups %xmm14, 128(%rsp)
|
||||
vmovups %xmm15, 144(%rsp)
|
||||
|
||||
movq OLD_B, B
|
||||
#endif
|
||||
|
||||
leaq (,LDA, SIZE), LDA # Scaling
|
||||
|
||||
movq N, J
|
||||
sarq $1, J
|
||||
jle .L20
|
||||
ALIGN_4
|
||||
|
||||
.L01:
|
||||
movq A, AO1
|
||||
leaq (A, LDA), AO2
|
||||
leaq (A, LDA, 2), A
|
||||
|
||||
movq M, I
|
||||
sarq $3, I
|
||||
jle .L08
|
||||
ALIGN_4
|
||||
|
||||
.L03:
|
||||
|
||||
#ifndef DOUBLE
|
||||
vmovss 0 * SIZE(AO1), %xmm0
|
||||
vmovss 0 * SIZE(AO2), %xmm1
|
||||
vmovss 1 * SIZE(AO1), %xmm2
|
||||
vmovss 1 * SIZE(AO2), %xmm3
|
||||
vmovss 2 * SIZE(AO1), %xmm4
|
||||
vmovss 2 * SIZE(AO2), %xmm5
|
||||
vmovss 3 * SIZE(AO1), %xmm6
|
||||
vmovss 3 * SIZE(AO2), %xmm7
|
||||
|
||||
vmovss 4 * SIZE(AO1), %xmm8
|
||||
vmovss 4 * SIZE(AO2), %xmm9
|
||||
vmovss 5 * SIZE(AO1), %xmm10
|
||||
vmovss 5 * SIZE(AO2), %xmm11
|
||||
vmovss 6 * SIZE(AO1), %xmm12
|
||||
vmovss 6 * SIZE(AO2), %xmm13
|
||||
vmovss 7 * SIZE(AO1), %xmm14
|
||||
vmovss 7 * SIZE(AO2), %xmm15
|
||||
|
||||
vmovss %xmm0, 0 * SIZE(B)
|
||||
vmovss %xmm1, 1 * SIZE(B)
|
||||
vmovss %xmm2, 2 * SIZE(B)
|
||||
vmovss %xmm3, 3 * SIZE(B)
|
||||
vmovss %xmm4, 4 * SIZE(B)
|
||||
vmovss %xmm5, 5 * SIZE(B)
|
||||
vmovss %xmm6, 6 * SIZE(B)
|
||||
vmovss %xmm7, 7 * SIZE(B)
|
||||
|
||||
vmovss %xmm8, 8 * SIZE(B)
|
||||
vmovss %xmm9, 9 * SIZE(B)
|
||||
vmovss %xmm10, 10 * SIZE(B)
|
||||
vmovss %xmm11, 11 * SIZE(B)
|
||||
vmovss %xmm12, 12 * SIZE(B)
|
||||
vmovss %xmm13, 13 * SIZE(B)
|
||||
vmovss %xmm14, 14 * SIZE(B)
|
||||
vmovss %xmm15, 15 * SIZE(B)
|
||||
|
||||
#else
|
||||
prefetchw 256(B)
|
||||
|
||||
prefetchnta 256(AO1)
|
||||
vmovsd 0 * SIZE(AO1), %xmm0
|
||||
vmovsd 1 * SIZE(AO1), %xmm1
|
||||
vmovsd 2 * SIZE(AO1), %xmm2
|
||||
vmovsd 3 * SIZE(AO1), %xmm3
|
||||
vmovsd 4 * SIZE(AO1), %xmm4
|
||||
vmovsd 5 * SIZE(AO1), %xmm5
|
||||
vmovsd 6 * SIZE(AO1), %xmm6
|
||||
vmovsd 7 * SIZE(AO1), %xmm7
|
||||
|
||||
prefetchnta 256(AO2)
|
||||
vmovhpd 0 * SIZE(AO2), %xmm0 , %xmm0
|
||||
vmovhpd 1 * SIZE(AO2), %xmm1 , %xmm1
|
||||
vmovhpd 2 * SIZE(AO2), %xmm2 , %xmm2
|
||||
vmovhpd 3 * SIZE(AO2), %xmm3 , %xmm3
|
||||
vmovhpd 4 * SIZE(AO2), %xmm4 , %xmm4
|
||||
vmovhpd 5 * SIZE(AO2), %xmm5 , %xmm5
|
||||
vmovhpd 6 * SIZE(AO2), %xmm6 , %xmm6
|
||||
vmovhpd 7 * SIZE(AO2), %xmm7 , %xmm7
|
||||
|
||||
|
||||
prefetchw 256+64(B)
|
||||
vmovups %xmm0, 0 * SIZE(B)
|
||||
vmovups %xmm1, 2 * SIZE(B)
|
||||
vmovups %xmm2, 4 * SIZE(B)
|
||||
vmovups %xmm3, 6 * SIZE(B)
|
||||
vmovups %xmm4, 8 * SIZE(B)
|
||||
vmovups %xmm5, 10 * SIZE(B)
|
||||
vmovups %xmm6, 12 * SIZE(B)
|
||||
vmovups %xmm7, 14 * SIZE(B)
|
||||
|
||||
#endif
|
||||
|
||||
addq $8 * SIZE, AO1
|
||||
addq $8 * SIZE, AO2
|
||||
subq $-16 * SIZE, B
|
||||
decq I
|
||||
jg .L03
|
||||
ALIGN_4
|
||||
|
||||
|
||||
.L08:
|
||||
testq $4 , M
|
||||
je .L14
|
||||
|
||||
ALIGN_4
|
||||
|
||||
|
||||
.L13:
|
||||
#ifndef DOUBLE
|
||||
vmovss 0 * SIZE(AO1), %xmm0
|
||||
vmovss 0 * SIZE(AO2), %xmm1
|
||||
vmovss 1 * SIZE(AO1), %xmm2
|
||||
vmovss 1 * SIZE(AO2), %xmm3
|
||||
vmovss 2 * SIZE(AO1), %xmm4
|
||||
vmovss 2 * SIZE(AO2), %xmm5
|
||||
vmovss 3 * SIZE(AO1), %xmm6
|
||||
vmovss 3 * SIZE(AO2), %xmm7
|
||||
|
||||
vmovss %xmm0, 0 * SIZE(B)
|
||||
vmovss %xmm1, 1 * SIZE(B)
|
||||
vmovss %xmm2, 2 * SIZE(B)
|
||||
vmovss %xmm3, 3 * SIZE(B)
|
||||
vmovss %xmm4, 4 * SIZE(B)
|
||||
vmovss %xmm5, 5 * SIZE(B)
|
||||
vmovss %xmm6, 6 * SIZE(B)
|
||||
vmovss %xmm7, 7 * SIZE(B)
|
||||
#else
|
||||
|
||||
vmovsd 0 * SIZE(AO1), %xmm0
|
||||
vmovsd 1 * SIZE(AO1), %xmm1
|
||||
vmovsd 2 * SIZE(AO1), %xmm2
|
||||
vmovsd 3 * SIZE(AO1), %xmm3
|
||||
|
||||
vmovhpd 0 * SIZE(AO2), %xmm0 , %xmm0
|
||||
vmovhpd 1 * SIZE(AO2), %xmm1 , %xmm1
|
||||
vmovhpd 2 * SIZE(AO2), %xmm2 , %xmm2
|
||||
vmovhpd 3 * SIZE(AO2), %xmm3 , %xmm3
|
||||
|
||||
|
||||
vmovups %xmm0, 0 * SIZE(B)
|
||||
vmovups %xmm1, 2 * SIZE(B)
|
||||
vmovups %xmm2, 4 * SIZE(B)
|
||||
vmovups %xmm3, 6 * SIZE(B)
|
||||
#endif
|
||||
|
||||
addq $4 * SIZE, AO1
|
||||
addq $4 * SIZE, AO2
|
||||
subq $-8 * SIZE, B
|
||||
ALIGN_4
|
||||
|
||||
.L14:
|
||||
movq M, I
|
||||
andq $3, I
|
||||
jle .L16
|
||||
ALIGN_4
|
||||
|
||||
.L15:
|
||||
#ifndef DOUBLE
|
||||
vmovss 0 * SIZE(AO1), %xmm0
|
||||
vmovss 0 * SIZE(AO2), %xmm1
|
||||
|
||||
vmovss %xmm0, 0 * SIZE(B)
|
||||
vmovss %xmm1, 1 * SIZE(B)
|
||||
#else
|
||||
vmovsd 0 * SIZE(AO1), %xmm0
|
||||
vmovhpd 0 * SIZE(AO2), %xmm0 , %xmm0
|
||||
|
||||
vmovups %xmm0, 0 * SIZE(B)
|
||||
#endif
|
||||
|
||||
addq $SIZE, AO1
|
||||
addq $SIZE, AO2
|
||||
addq $2 * SIZE, B
|
||||
decq I
|
||||
jg .L15
|
||||
ALIGN_4
|
||||
|
||||
.L16:
|
||||
decq J
|
||||
jg .L01
|
||||
ALIGN_4
|
||||
|
||||
.L20:
|
||||
testq $1, N
|
||||
jle .L999
|
||||
|
||||
movq A, AO1
|
||||
|
||||
movq M, I
|
||||
sarq $2, I
|
||||
jle .L34
|
||||
ALIGN_4
|
||||
|
||||
.L33:
|
||||
#ifndef DOUBLE
|
||||
vmovups 0 * SIZE(AO1), %xmm0
|
||||
|
||||
vmovups %xmm0, 0 * SIZE(B)
|
||||
#else
|
||||
vmovups 0 * SIZE(AO1), %xmm0
|
||||
vmovups 2 * SIZE(AO1), %xmm1
|
||||
|
||||
vmovups %xmm0, 0 * SIZE(B)
|
||||
vmovups %xmm1, 2 * SIZE(B)
|
||||
#endif
|
||||
|
||||
addq $4 * SIZE, AO1
|
||||
subq $-4 * SIZE, B
|
||||
decq I
|
||||
jg .L33
|
||||
ALIGN_4
|
||||
|
||||
.L34:
|
||||
movq M, I
|
||||
andq $3, I
|
||||
jle .L999
|
||||
ALIGN_4
|
||||
|
||||
.L35:
|
||||
#ifndef DOUBLE
|
||||
vmovss 0 * SIZE(AO1), %xmm0
|
||||
vmovss %xmm0, 0 * SIZE(B)
|
||||
#else
|
||||
vmovsd 0 * SIZE(AO1), %xmm0
|
||||
vmovsd %xmm0, 0 * SIZE(B)
|
||||
#endif
|
||||
|
||||
addq $SIZE, AO1
|
||||
addq $1 * SIZE, B
|
||||
decq I
|
||||
jg .L35
|
||||
ALIGN_4
|
||||
|
||||
|
||||
.L999:
|
||||
#ifdef WINDOWS_ABI
|
||||
vmovups 0(%rsp), %xmm6
|
||||
vmovups 16(%rsp), %xmm7
|
||||
vmovups 32(%rsp), %xmm8
|
||||
vmovups 48(%rsp), %xmm9
|
||||
vmovups 64(%rsp), %xmm10
|
||||
vmovups 80(%rsp), %xmm11
|
||||
vmovups 96(%rsp), %xmm12
|
||||
vmovups 112(%rsp), %xmm13
|
||||
vmovups 128(%rsp), %xmm14
|
||||
vmovups 144(%rsp), %xmm15
|
||||
|
||||
addq $STACKSIZE, %rsp
|
||||
#endif
|
||||
|
||||
popq %r12
|
||||
popq %r13
|
||||
|
||||
#ifdef WINDOWS_ABI
|
||||
popq %r14
|
||||
popq %r15
|
||||
#endif
|
||||
ret
|
||||
|
||||
EPILOGUE
|
||||
|
|
@ -0,0 +1,374 @@
|
|||
/*********************************************************************/
|
||||
/* 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"
|
||||
|
||||
#ifndef WINDOWS_ABI
|
||||
|
||||
#define M ARG1 /* rdi */
|
||||
#define N ARG2 /* rsi */
|
||||
#define A ARG3 /* rdx */
|
||||
#define LDA ARG4 /* rcx */
|
||||
#define B ARG5 /* r8 */
|
||||
|
||||
#define I %r10
|
||||
#define J %rbp
|
||||
|
||||
#define AO1 %r9
|
||||
#define AO2 %r15
|
||||
#define AO3 %r11
|
||||
#define AO4 %r14
|
||||
#define BO1 %r13
|
||||
#define M8 %rbx
|
||||
#define BO %rax
|
||||
|
||||
#else
|
||||
|
||||
#define STACKSIZE 256
|
||||
|
||||
#define M ARG1 /* rcx */
|
||||
#define N ARG2 /* rdx */
|
||||
#define A ARG3 /* r8 */
|
||||
#define LDA ARG4 /* r9 */
|
||||
#define OLD_B 40 + 64 + STACKSIZE(%rsp)
|
||||
|
||||
#define B %rdi
|
||||
|
||||
#define I %r10
|
||||
#define J %r11
|
||||
|
||||
#define AO1 %r12
|
||||
#define AO2 %r13
|
||||
#define AO3 %r14
|
||||
#define AO4 %r15
|
||||
|
||||
#define BO1 %rsi
|
||||
#define M8 %rbp
|
||||
#define BO %rax
|
||||
|
||||
#endif
|
||||
|
||||
PROLOGUE
|
||||
PROFCODE
|
||||
|
||||
#ifdef WINDOWS_ABI
|
||||
pushq %rdi
|
||||
pushq %rsi
|
||||
#endif
|
||||
pushq %r15
|
||||
pushq %r14
|
||||
pushq %r13
|
||||
pushq %r12
|
||||
pushq %rbp
|
||||
pushq %rbx
|
||||
|
||||
#ifdef WINDOWS_ABI
|
||||
subq $STACKSIZE, %rsp
|
||||
|
||||
vmovups %xmm6, 0(%rsp)
|
||||
vmovups %xmm7, 16(%rsp)
|
||||
vmovups %xmm8, 32(%rsp)
|
||||
vmovups %xmm9, 48(%rsp)
|
||||
vmovups %xmm10, 64(%rsp)
|
||||
vmovups %xmm11, 80(%rsp)
|
||||
vmovups %xmm12, 96(%rsp)
|
||||
vmovups %xmm13, 112(%rsp)
|
||||
vmovups %xmm14, 128(%rsp)
|
||||
vmovups %xmm15, 144(%rsp)
|
||||
|
||||
movq OLD_B, B
|
||||
#endif
|
||||
|
||||
movq N, %rax
|
||||
andq $-2, %rax
|
||||
imulq M, %rax
|
||||
|
||||
leaq (B, %rax, SIZE), BO1
|
||||
|
||||
leaq (, LDA, SIZE), LDA
|
||||
leaq (, M, SIZE), M8
|
||||
|
||||
movq M, J
|
||||
sarq $1, J
|
||||
jle .L20
|
||||
ALIGN_4
|
||||
|
||||
.L01:
|
||||
movq A, AO1
|
||||
leaq (A, LDA ), AO2
|
||||
leaq (A, LDA, 2), A
|
||||
|
||||
movq B, BO
|
||||
addq $4 * SIZE, B
|
||||
|
||||
movq N, I
|
||||
sarq $3, I
|
||||
jle .L10
|
||||
ALIGN_4
|
||||
|
||||
|
||||
.L08:
|
||||
#ifndef DOUBLE
|
||||
|
||||
vmovsd 0 * SIZE(AO1), %xmm0
|
||||
vmovsd 2 * SIZE(AO1), %xmm2
|
||||
vmovsd 4 * SIZE(AO1), %xmm4
|
||||
vmovsd 6 * SIZE(AO1), %xmm6
|
||||
vmovsd 0 * SIZE(AO2), %xmm1
|
||||
vmovsd 2 * SIZE(AO2), %xmm3
|
||||
vmovsd 4 * SIZE(AO2), %xmm5
|
||||
vmovsd 6 * SIZE(AO2), %xmm7
|
||||
|
||||
vmovsd %xmm0, 0 * SIZE(BO)
|
||||
vmovsd %xmm1, 2 * SIZE(BO)
|
||||
leaq (BO, M8, 2), BO
|
||||
|
||||
vmovsd %xmm2, 0 * SIZE(BO)
|
||||
vmovsd %xmm3, 2 * SIZE(BO)
|
||||
leaq (BO, M8, 2), BO
|
||||
|
||||
vmovsd %xmm4, 0 * SIZE(BO)
|
||||
vmovsd %xmm5, 2 * SIZE(BO)
|
||||
leaq (BO, M8, 2), BO
|
||||
|
||||
vmovsd %xmm6, 0 * SIZE(BO)
|
||||
vmovsd %xmm7, 2 * SIZE(BO)
|
||||
leaq (BO, M8, 2), BO
|
||||
|
||||
|
||||
#else
|
||||
|
||||
prefetchnta 256(AO1)
|
||||
prefetchnta 256(AO2)
|
||||
vmovups 0 * SIZE(AO1), %xmm0
|
||||
vmovups 2 * SIZE(AO1), %xmm2
|
||||
vmovups 4 * SIZE(AO1), %xmm4
|
||||
vmovups 6 * SIZE(AO1), %xmm6
|
||||
vmovups 0 * SIZE(AO2), %xmm1
|
||||
vmovups 2 * SIZE(AO2), %xmm3
|
||||
vmovups 4 * SIZE(AO2), %xmm5
|
||||
vmovups 6 * SIZE(AO2), %xmm7
|
||||
|
||||
vmovups %xmm0, 0 * SIZE(BO)
|
||||
vmovups %xmm1, 2 * SIZE(BO)
|
||||
leaq (BO, M8, 2), BO
|
||||
|
||||
vmovups %xmm2, 0 * SIZE(BO)
|
||||
vmovups %xmm3, 2 * SIZE(BO)
|
||||
leaq (BO, M8, 2), BO
|
||||
|
||||
vmovups %xmm4, 0 * SIZE(BO)
|
||||
vmovups %xmm5, 2 * SIZE(BO)
|
||||
leaq (BO, M8, 2), BO
|
||||
|
||||
vmovups %xmm6, 0 * SIZE(BO)
|
||||
vmovups %xmm7, 2 * SIZE(BO)
|
||||
leaq (BO, M8, 2), BO
|
||||
|
||||
#endif
|
||||
|
||||
addq $8 * SIZE, AO1
|
||||
addq $8 * SIZE, AO2
|
||||
decq I
|
||||
jg .L08
|
||||
ALIGN_4
|
||||
|
||||
|
||||
|
||||
.L10:
|
||||
testq $4, N
|
||||
jle .L12
|
||||
#ifndef DOUBLE
|
||||
|
||||
vmovsd 0 * SIZE(AO1), %xmm0
|
||||
vmovsd 2 * SIZE(AO1), %xmm2
|
||||
vmovsd 0 * SIZE(AO2), %xmm1
|
||||
vmovsd 2 * SIZE(AO2), %xmm3
|
||||
|
||||
vmovsd %xmm0, 0 * SIZE(BO)
|
||||
vmovsd %xmm1, 2 * SIZE(BO)
|
||||
leaq (BO, M8, 2), BO
|
||||
|
||||
vmovsd %xmm2, 0 * SIZE(BO)
|
||||
vmovsd %xmm3, 2 * SIZE(BO)
|
||||
leaq (BO, M8, 2), BO
|
||||
|
||||
|
||||
#else
|
||||
|
||||
vmovups 0 * SIZE(AO1), %xmm0
|
||||
vmovups 2 * SIZE(AO1), %xmm2
|
||||
vmovups 0 * SIZE(AO2), %xmm1
|
||||
vmovups 2 * SIZE(AO2), %xmm3
|
||||
|
||||
vmovups %xmm0, 0 * SIZE(BO)
|
||||
vmovups %xmm1, 2 * SIZE(BO)
|
||||
leaq (BO, M8, 2), BO
|
||||
|
||||
vmovups %xmm2, 0 * SIZE(BO)
|
||||
vmovups %xmm3, 2 * SIZE(BO)
|
||||
leaq (BO, M8, 2), BO
|
||||
|
||||
#endif
|
||||
|
||||
addq $4 * SIZE, AO1
|
||||
addq $4 * SIZE, AO2
|
||||
ALIGN_4
|
||||
|
||||
|
||||
.L12:
|
||||
testq $2, N
|
||||
jle .L14
|
||||
#ifndef DOUBLE
|
||||
vmovsd 0 * SIZE(AO1), %xmm0
|
||||
vmovsd 0 * SIZE(AO2), %xmm1
|
||||
|
||||
vmovsd %xmm0, 0 * SIZE(BO)
|
||||
vmovsd %xmm1, 2 * SIZE(BO)
|
||||
#else
|
||||
vmovups 0 * SIZE(AO1), %xmm0
|
||||
vmovups 0 * SIZE(AO2), %xmm1
|
||||
|
||||
vmovups %xmm0, 0 * SIZE(BO)
|
||||
vmovups %xmm1, 2 * SIZE(BO)
|
||||
#endif
|
||||
|
||||
leaq (BO, M8, 2), BO
|
||||
addq $2 * SIZE, AO1
|
||||
addq $2 * SIZE, AO2
|
||||
ALIGN_4
|
||||
|
||||
.L14:
|
||||
testq $1, N
|
||||
jle .L19
|
||||
|
||||
#ifndef DOUBLE
|
||||
vmovss 0 * SIZE(AO1), %xmm0
|
||||
vmovss 0 * SIZE(AO2), %xmm1
|
||||
|
||||
vmovss %xmm0, 0 * SIZE(BO1)
|
||||
vmovss %xmm1, 1 * SIZE(BO1)
|
||||
#else
|
||||
vmovsd 0 * SIZE(AO1), %xmm0
|
||||
vmovhpd 0 * SIZE(AO2), %xmm0 , %xmm0
|
||||
|
||||
vmovups %xmm0, 0 * SIZE(BO1)
|
||||
#endif
|
||||
|
||||
addq $2 * SIZE, BO1
|
||||
ALIGN_4
|
||||
|
||||
.L19:
|
||||
decq J
|
||||
jg .L01
|
||||
ALIGN_4
|
||||
|
||||
.L20:
|
||||
testq $1, M
|
||||
jle .L999
|
||||
ALIGN_4
|
||||
|
||||
.L31:
|
||||
movq A, AO1
|
||||
movq B, BO
|
||||
|
||||
movq N, I
|
||||
sarq $1, I
|
||||
jle .L33
|
||||
ALIGN_4
|
||||
|
||||
.L32:
|
||||
#ifndef DOUBLE
|
||||
vmovsd 0 * SIZE(AO1), %xmm0
|
||||
vmovsd %xmm0, 0 * SIZE(BO)
|
||||
#else
|
||||
vmovups 0 * SIZE(AO1), %xmm0
|
||||
vmovups %xmm0, 0 * SIZE(BO)
|
||||
#endif
|
||||
|
||||
addq $2 * SIZE, AO1
|
||||
leaq (BO, M8, 2), BO
|
||||
decq I
|
||||
jg .L32
|
||||
ALIGN_4
|
||||
|
||||
.L33:
|
||||
testq $1, N
|
||||
jle .L999
|
||||
|
||||
#ifndef DOUBLE
|
||||
vmovss 0 * SIZE(AO1), %xmm0
|
||||
vmovss %xmm0, 0 * SIZE(BO1)
|
||||
#else
|
||||
vmovsd 0 * SIZE(AO1), %xmm0
|
||||
vmovsd %xmm0, 0 * SIZE(BO1)
|
||||
#endif
|
||||
addq $1 * SIZE, BO1
|
||||
ALIGN_4
|
||||
|
||||
.L999:
|
||||
#ifdef WINDOWS_ABI
|
||||
vmovups 0(%rsp), %xmm6
|
||||
vmovups 16(%rsp), %xmm7
|
||||
vmovups 32(%rsp), %xmm8
|
||||
vmovups 48(%rsp), %xmm9
|
||||
vmovups 64(%rsp), %xmm10
|
||||
vmovups 80(%rsp), %xmm11
|
||||
vmovups 96(%rsp), %xmm12
|
||||
vmovups 112(%rsp), %xmm13
|
||||
vmovups 128(%rsp), %xmm14
|
||||
vmovups 144(%rsp), %xmm15
|
||||
|
||||
addq $STACKSIZE, %rsp
|
||||
#endif
|
||||
|
||||
popq %rbx
|
||||
popq %rbp
|
||||
popq %r12
|
||||
popq %r13
|
||||
popq %r14
|
||||
popq %r15
|
||||
#ifdef WINDOWS_ABI
|
||||
popq %rsi
|
||||
popq %rdi
|
||||
#endif
|
||||
|
||||
ret
|
||||
|
||||
EPILOGUE
|
||||
Loading…
Reference in New Issue