|
kusano |
2b45e8 |
/*********************************************************************/
|
|
kusano |
2b45e8 |
/* Copyright 2009, 2010 The University of Texas at Austin. */
|
|
kusano |
2b45e8 |
/* All rights reserved. */
|
|
kusano |
2b45e8 |
/* */
|
|
kusano |
2b45e8 |
/* Redistribution and use in source and binary forms, with or */
|
|
kusano |
2b45e8 |
/* without modification, are permitted provided that the following */
|
|
kusano |
2b45e8 |
/* conditions are met: */
|
|
kusano |
2b45e8 |
/* */
|
|
kusano |
2b45e8 |
/* 1. Redistributions of source code must retain the above */
|
|
kusano |
2b45e8 |
/* copyright notice, this list of conditions and the following */
|
|
kusano |
2b45e8 |
/* disclaimer. */
|
|
kusano |
2b45e8 |
/* */
|
|
kusano |
2b45e8 |
/* 2. Redistributions in binary form must reproduce the above */
|
|
kusano |
2b45e8 |
/* copyright notice, this list of conditions and the following */
|
|
kusano |
2b45e8 |
/* disclaimer in the documentation and/or other materials */
|
|
kusano |
2b45e8 |
/* provided with the distribution. */
|
|
kusano |
2b45e8 |
/* */
|
|
kusano |
2b45e8 |
/* THIS SOFTWARE IS PROVIDED BY THE UNIVERSITY OF TEXAS AT */
|
|
kusano |
2b45e8 |
/* AUSTIN ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, */
|
|
kusano |
2b45e8 |
/* INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF */
|
|
kusano |
2b45e8 |
/* MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE */
|
|
kusano |
2b45e8 |
/* DISCLAIMED. IN NO EVENT SHALL THE UNIVERSITY OF TEXAS AT */
|
|
kusano |
2b45e8 |
/* AUSTIN OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, */
|
|
kusano |
2b45e8 |
/* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES */
|
|
kusano |
2b45e8 |
/* (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE */
|
|
kusano |
2b45e8 |
/* GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR */
|
|
kusano |
2b45e8 |
/* BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF */
|
|
kusano |
2b45e8 |
/* LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT */
|
|
kusano |
2b45e8 |
/* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT */
|
|
kusano |
2b45e8 |
/* OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE */
|
|
kusano |
2b45e8 |
/* POSSIBILITY OF SUCH DAMAGE. */
|
|
kusano |
2b45e8 |
/* */
|
|
kusano |
2b45e8 |
/* The views and conclusions contained in the software and */
|
|
kusano |
2b45e8 |
/* documentation are those of the authors and should not be */
|
|
kusano |
2b45e8 |
/* interpreted as representing official policies, either expressed */
|
|
kusano |
2b45e8 |
/* or implied, of The University of Texas at Austin. */
|
|
kusano |
2b45e8 |
/*********************************************************************/
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
.set noat
|
|
kusano |
2b45e8 |
.set noreorder
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
#define ASSEMBLER
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
#include "common.h"
|
|
kusano |
2b45e8 |
#include "version.h"
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
.globl NAME
|
|
kusano |
2b45e8 |
.ent NAME
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
NAME:
|
|
kusano |
2b45e8 |
#ifdef PROFILE
|
|
kusano |
2b45e8 |
ldgp $gp, 0($27)
|
|
kusano |
2b45e8 |
lda $28, _mcount
|
|
kusano |
2b45e8 |
jsr $28, ($28), _mcount
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
#ifndef C_INTERFACE
|
|
kusano |
2b45e8 |
ldl $16, 0($16) # n
|
|
kusano |
2b45e8 |
mov $18, $20 # Store Address
|
|
kusano |
2b45e8 |
ldl $19, 0($19) # incx
|
|
kusano |
2b45e8 |
nop
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
LD $f1, 0($17) # alpha
|
|
kusano |
2b45e8 |
#else
|
|
kusano |
2b45e8 |
mov $18, $20 # Store Address
|
|
kusano |
2b45e8 |
fmov $f17, $f1 # alpha
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
#ifndef PROFILE
|
|
kusano |
2b45e8 |
.prologue 0
|
|
kusano |
2b45e8 |
#else
|
|
kusano |
2b45e8 |
.prologue 1
|
|
kusano |
2b45e8 |
#endif
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
sra $16, 1, $21 # 4-unrolling
|
|
kusano |
2b45e8 |
ble $16, $End
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
lda $23, -1($19)
|
|
kusano |
2b45e8 |
ble $19, $End
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
bgt $23, $INC_NOT_1
|
|
kusano |
2b45e8 |
.align 4
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
ble $21, $Sub
|
|
kusano |
2b45e8 |
lda $21, -1($21)
|
|
kusano |
2b45e8 |
LD $f10, 0*SIZE($18)
|
|
kusano |
2b45e8 |
LD $f11, 1*SIZE($18)
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
LD $f12, 2*SIZE($18)
|
|
kusano |
2b45e8 |
LD $f13, 3*SIZE($18)
|
|
kusano |
2b45e8 |
lda $18, 4*SIZE($18)
|
|
kusano |
2b45e8 |
ble $21, $MainRemain
|
|
kusano |
2b45e8 |
.align 4
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
$MainLoop:
|
|
kusano |
2b45e8 |
MUL $f10, $f1, $f20
|
|
kusano |
2b45e8 |
LD $f10, 0*SIZE($18)
|
|
kusano |
2b45e8 |
MUL $f11, $f1, $f21
|
|
kusano |
2b45e8 |
LD $f11, 1*SIZE($18)
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
MUL $f12, $f1, $f22
|
|
kusano |
2b45e8 |
LD $f12, 2*SIZE($18)
|
|
kusano |
2b45e8 |
MUL $f13, $f1, $f23
|
|
kusano |
2b45e8 |
LD $f13, 3*SIZE($18)
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
lda $18, 4*SIZE($18)
|
|
kusano |
2b45e8 |
lda $21, -1($21)
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
ST $f20, 0*SIZE($20)
|
|
kusano |
2b45e8 |
ST $f21, 1*SIZE($20)
|
|
kusano |
2b45e8 |
ST $f22, 2*SIZE($20)
|
|
kusano |
2b45e8 |
ST $f23, 3*SIZE($20)
|
|
kusano |
2b45e8 |
lda $20, 4*SIZE($20)
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
bgt $21, $MainLoop
|
|
kusano |
2b45e8 |
.align 4
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
$MainRemain:
|
|
kusano |
2b45e8 |
MUL $f10, $f1, $f20
|
|
kusano |
2b45e8 |
MUL $f11, $f1, $f21
|
|
kusano |
2b45e8 |
MUL $f12, $f1, $f22
|
|
kusano |
2b45e8 |
MUL $f13, $f1, $f23
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
ST $f20, 0*SIZE($20)
|
|
kusano |
2b45e8 |
ST $f21, 1*SIZE($20)
|
|
kusano |
2b45e8 |
ST $f22, 2*SIZE($20)
|
|
kusano |
2b45e8 |
ST $f23, 3*SIZE($20)
|
|
kusano |
2b45e8 |
lda $20, 4*SIZE($20)
|
|
kusano |
2b45e8 |
.align 4
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
$Sub:
|
|
kusano |
2b45e8 |
blbc $16, $End
|
|
kusano |
2b45e8 |
LD $f10, 0*SIZE($18)
|
|
kusano |
2b45e8 |
LD $f11, 1*SIZE($18)
|
|
kusano |
2b45e8 |
MUL $f10, $f1, $f20
|
|
kusano |
2b45e8 |
MUL $f11, $f1, $f21
|
|
kusano |
2b45e8 |
ST $f20, 0*SIZE($20)
|
|
kusano |
2b45e8 |
ST $f21, 1*SIZE($20)
|
|
kusano |
2b45e8 |
.align 4
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
$End:
|
|
kusano |
2b45e8 |
ret
|
|
kusano |
2b45e8 |
.align 4
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
$INC_NOT_1:
|
|
kusano |
2b45e8 |
addl $19, $19, $19
|
|
kusano |
2b45e8 |
ble $21, $INC_Sub
|
|
kusano |
2b45e8 |
lda $21, -1($21)
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
LD $f10, 0*SIZE($18)
|
|
kusano |
2b45e8 |
LD $f11, 1*SIZE($18)
|
|
kusano |
2b45e8 |
SXADDQ $19, $18, $18
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
LD $f12, 0*SIZE($18)
|
|
kusano |
2b45e8 |
LD $f13, 1*SIZE($18)
|
|
kusano |
2b45e8 |
SXADDQ $19, $18, $18
|
|
kusano |
2b45e8 |
ble $21, $INC_MainRemain
|
|
kusano |
2b45e8 |
.align 4
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
$INC_MainLoop:
|
|
kusano |
2b45e8 |
MUL $f10, $f1, $f20
|
|
kusano |
2b45e8 |
LD $f10, 0*SIZE($18)
|
|
kusano |
2b45e8 |
MUL $f11, $f1, $f21
|
|
kusano |
2b45e8 |
LD $f11, 1*SIZE($18)
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
SXADDQ $19, $18, $18
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
MUL $f12, $f1, $f22
|
|
kusano |
2b45e8 |
LD $f12, 0*SIZE($18)
|
|
kusano |
2b45e8 |
MUL $f13, $f1, $f23
|
|
kusano |
2b45e8 |
LD $f13, 1*SIZE($18)
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
SXADDQ $19, $18, $18
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
ST $f20, 0*SIZE($20)
|
|
kusano |
2b45e8 |
lda $21, -1($21)
|
|
kusano |
2b45e8 |
ST $f21, 1*SIZE($20)
|
|
kusano |
2b45e8 |
SXADDQ $19, $20, $20
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
ST $f22, 0*SIZE($20)
|
|
kusano |
2b45e8 |
ST $f23, 1*SIZE($20)
|
|
kusano |
2b45e8 |
SXADDQ $19, $20, $20
|
|
kusano |
2b45e8 |
unop
|
|
kusano |
2b45e8 |
bgt $21, $INC_MainLoop
|
|
kusano |
2b45e8 |
.align 4
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
$INC_MainRemain:
|
|
kusano |
2b45e8 |
MUL $f10, $f1, $f20
|
|
kusano |
2b45e8 |
MUL $f11, $f1, $f21
|
|
kusano |
2b45e8 |
MUL $f12, $f1, $f22
|
|
kusano |
2b45e8 |
MUL $f13, $f1, $f23
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
ST $f20, 0*SIZE($20)
|
|
kusano |
2b45e8 |
ST $f21, 1*SIZE($20)
|
|
kusano |
2b45e8 |
SXADDQ $19, $20, $20
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
ST $f22, 0*SIZE($20)
|
|
kusano |
2b45e8 |
ST $f23, 1*SIZE($20)
|
|
kusano |
2b45e8 |
SXADDQ $19, $20, $20
|
|
kusano |
2b45e8 |
.align 4
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
$INC_Sub:
|
|
kusano |
2b45e8 |
blbc $16, $INC_End
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
LD $f10, 0*SIZE($18)
|
|
kusano |
2b45e8 |
LD $f11, 1*SIZE($18)
|
|
kusano |
2b45e8 |
MUL $f10, $f1, $f20
|
|
kusano |
2b45e8 |
MUL $f11, $f1, $f21
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
ST $f20, 0*SIZE($20)
|
|
kusano |
2b45e8 |
ST $f21, 1*SIZE($20)
|
|
kusano |
2b45e8 |
.align 4
|
|
kusano |
2b45e8 |
|
|
kusano |
2b45e8 |
$INC_End:
|
|
kusano |
2b45e8 |
ret
|
|
kusano |
2b45e8 |
.end NAME
|
|
kusano |
2b45e8 |
.ident VERSION
|