mirror of https://github.com/xianyi/OpenBLAS.git
471 lines
12 KiB
ArmAsm
471 lines
12 KiB
ArmAsm
/*******************************************************************************
|
|
Copyright (c) 2024, 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.
|
|
*******************************************************************************/
|
|
|
|
#define ASSEMBLER
|
|
|
|
#include "common.h"
|
|
#include "loongarch64_asm.S"
|
|
|
|
/* Param */
|
|
#define M $r4
|
|
#define N $r5
|
|
#define A $r6
|
|
#define LDA $r7
|
|
#define X $r8
|
|
#define INCX $r9
|
|
#define Y $r10
|
|
#define INCY $r11
|
|
#define BUFFER $r16
|
|
#define ALPHA $f0
|
|
|
|
#define JY $r18
|
|
#define JX $r31
|
|
#define T0 $r19
|
|
#define T1 $r20
|
|
#define AO3 $r12
|
|
#define AO4 $r13
|
|
#define I $r14
|
|
#define J $r15
|
|
#define AO1 $r23
|
|
#define AO2 $r24
|
|
#define IX $r25
|
|
#define IY $r26
|
|
#define II $r27
|
|
#define T2 $r28
|
|
#define T3 $r29
|
|
#define T4 $r30
|
|
#define T5 $r17
|
|
#define T6 $r16
|
|
#define T7 $r12
|
|
|
|
/* LSX vectors */
|
|
#define U0 $vr31
|
|
#define U1 $vr1
|
|
#define U2 $vr2
|
|
#define U3 $vr3
|
|
#define U4 $vr4
|
|
#define U5 $vr5
|
|
#define U6 $vr6
|
|
#define U7 $vr7
|
|
#define U8 $vr8
|
|
#define U9 $vr9
|
|
#define U10 $vr10
|
|
#define U11 $vr11
|
|
#define U12 $vr12
|
|
#define U13 $vr13
|
|
#define U14 $vr14
|
|
#define U15 $vr15
|
|
#define U16 $vr16
|
|
#define VALPHA $vr17
|
|
|
|
#define a2 $f2
|
|
#define a3 $f3
|
|
#define a4 $f4
|
|
#define a5 $f5
|
|
#define a6 $f6
|
|
#define a7 $f7
|
|
#define a8 $f8
|
|
#define a9 $f9
|
|
|
|
.macro LOAD_Y_8
|
|
beqz T5, .L01_Y_0
|
|
add.d T2, IY, INCY
|
|
fldx.d $f4, Y, T2
|
|
add.d T2, T2, INCY
|
|
fldx.d $f5, Y, T2
|
|
add.d T2, T2, INCY
|
|
fldx.d $f6, Y, T2
|
|
add.d T2, T2, INCY
|
|
fldx.d $f7, Y, T2
|
|
|
|
add.d T2, T2, INCY
|
|
fldx.d $f8, Y, T2
|
|
add.d T2, T2, INCY
|
|
fldx.d $f9, Y, T2
|
|
add.d T2, T2, INCY
|
|
fldx.d $f10, Y, T2
|
|
add.d T2, T2, INCY
|
|
fldx.d $f11, Y, T2
|
|
|
|
vextrins.d U4, U5, 0x10
|
|
vextrins.d U6, U7, 0x10
|
|
vextrins.d U8, U9, 0x10
|
|
vextrins.d U10, U11, 0x10
|
|
b .L01_Y_1
|
|
.L01_Y_0:
|
|
add.d T7, IY, INCY
|
|
vldx U4, Y, T7
|
|
alsl.d T2, INCY, T7, 1
|
|
vldx U6, Y, T2
|
|
alsl.d T3, INCY, T2, 1
|
|
vldx U8, Y, T3
|
|
alsl.d T4, INCY, T3, 1
|
|
vldx U10, Y, T4
|
|
.L01_Y_1:
|
|
.endm
|
|
|
|
.macro LOAD_X_8
|
|
beqz T6, .L01_X_0
|
|
add.d T2, IX, INCX
|
|
fldx.d $f4, X, T2
|
|
add.d T2, T2, INCX
|
|
fldx.d $f5, X, T2
|
|
add.d T2, T2, INCX
|
|
fldx.d $f6, X, T2
|
|
add.d T2, T2, INCX
|
|
fldx.d $f7, X, T2
|
|
|
|
add.d T2, T2, INCX
|
|
fldx.d $f8, X, T2
|
|
add.d T2, T2, INCX
|
|
fldx.d $f9, X, T2
|
|
add.d T2, T2, INCX
|
|
fldx.d $f10, X, T2
|
|
add.d T2, T2, INCX
|
|
fldx.d $f11, X, T2
|
|
|
|
vextrins.d U4, U5, 0x10
|
|
vextrins.d U6, U7, 0x10
|
|
vextrins.d U8, U9, 0x10
|
|
vextrins.d U10, U11, 0x10
|
|
b .L01_X_1
|
|
.L01_X_0:
|
|
add.d T7, IX, INCX
|
|
vldx U4, X, T7
|
|
alsl.d T2, INCX, T7, 1
|
|
vldx U6, X, T2
|
|
alsl.d T3, INCX, T2, 1
|
|
vldx U8, X, T3
|
|
alsl.d T4, INCX, T3, 1
|
|
vldx U10, X, T4
|
|
.L01_X_1:
|
|
.endm
|
|
|
|
.macro STORE_Y_8
|
|
beqz T5, .L01_Y_2
|
|
vextrins.d U5, U4, 0x01
|
|
vextrins.d U7, U6, 0x01
|
|
vextrins.d U9, U8, 0x01
|
|
vextrins.d U11, U10, 0x01
|
|
|
|
add.d T2, IY, INCY
|
|
fstx.d $f4, Y, T2
|
|
add.d T2, T2, INCY
|
|
fstx.d $f5, Y, T2
|
|
add.d T2, T2, INCY
|
|
fstx.d $f6, Y, T2
|
|
add.d T2, T2, INCY
|
|
fstx.d $f7, Y, T2
|
|
|
|
add.d T2, T2, INCY
|
|
fstx.d $f8, Y, T2
|
|
add.d T2, T2, INCY
|
|
fstx.d $f9, Y, T2
|
|
add.d T2, T2, INCY
|
|
fstx.d $f10, Y, T2
|
|
add.d T2, T2, INCY
|
|
fstx.d $f11, Y, T2
|
|
b .L01_Y_3
|
|
.L01_Y_2:
|
|
vstx U4, Y, T7
|
|
vstx U6, Y, T2
|
|
vstx U8, Y, T3
|
|
vstx U10, Y, T4
|
|
.L01_Y_3:
|
|
.endm
|
|
|
|
PROLOGUE
|
|
|
|
addi.d $sp, $sp, -88
|
|
|
|
SDARG $r23, $sp, 0
|
|
SDARG $r24, $sp, 8
|
|
SDARG $r25, $sp, 16
|
|
SDARG $r26, $sp, 32
|
|
SDARG $r27, $sp, 40
|
|
SDARG $r28, $sp, 48
|
|
SDARG $r29, $sp, 56
|
|
SDARG $r30, $sp, 64
|
|
SDARG $r31, $sp, 72
|
|
ST ALPHA, $sp, 80
|
|
|
|
vldrepl.d VALPHA, $sp, 80
|
|
|
|
addi.d T5, INCY, -1
|
|
addi.d T6, INCX, -1
|
|
slli.d LDA, LDA, BASE_SHIFT
|
|
slli.d INCX, INCX, BASE_SHIFT
|
|
slli.d INCY, INCY, BASE_SHIFT
|
|
|
|
bge $r0, M, .L999
|
|
bge $r0, N, .L999
|
|
|
|
move J, $r0
|
|
move JY, $r0
|
|
move JX, $r0
|
|
move AO1, A
|
|
|
|
beq J, N, .L999
|
|
|
|
.L01:
|
|
vxor.v U2, U2, U2
|
|
fldx.d a6, X, JX
|
|
fmul.d a3, ALPHA, a6 //temp1
|
|
vshuf4i.d U3, U3, 0x00
|
|
|
|
mul.d T0, J, LDA
|
|
slli.d T1, J, BASE_SHIFT
|
|
add.d T0, T0, T1
|
|
fldx.d a6, AO1, T0
|
|
fldx.d a4, Y, JY
|
|
fmadd.d a4, a3, a6, a4
|
|
fstx.d a4, Y, JY
|
|
|
|
move IY, JY
|
|
move IX, JX
|
|
addi.d II, J, 1
|
|
move I, II
|
|
slli.d II, II, BASE_SHIFT
|
|
|
|
sub.d T0, M, J
|
|
addi.d T0, T0, -1
|
|
srai.d T0, T0, 3
|
|
add.d T0, T0, J
|
|
addi.d T0, T0, 1
|
|
beq I, T0, .L03
|
|
bge I, T0, .L03
|
|
|
|
mul.d T1, J, LDA
|
|
add.d T1, T1, II
|
|
|
|
.L02: /* /8 */
|
|
vldx U1, AO1, T1
|
|
addi.d T1, T1, 16
|
|
vldx U14, AO1, T1
|
|
addi.d T1, T1, 16
|
|
vldx U15, AO1, T1
|
|
addi.d T1, T1, 16
|
|
vldx U16, AO1, T1
|
|
addi.d T1, T1, 16
|
|
|
|
LOAD_Y_8
|
|
|
|
vfmadd.d U4, U3, U1, U4
|
|
vfmadd.d U6, U3, U14, U6
|
|
vfmadd.d U8, U3, U15, U8
|
|
vfmadd.d U10, U3, U16, U10
|
|
|
|
STORE_Y_8
|
|
|
|
alsl.d IY, INCY, IY, 3
|
|
|
|
LOAD_X_8
|
|
|
|
vfmadd.d U2, U1, U4, U2
|
|
vfmadd.d U2, U14, U6, U2
|
|
vfmadd.d U2, U15, U8, U2
|
|
vfmadd.d U2, U16, U10, U2
|
|
|
|
alsl.d IX, INCX, IX, 3
|
|
|
|
addi.d II, II, 64
|
|
addi.d I, I, 1
|
|
blt I, T0, .L02
|
|
|
|
// Acc U2
|
|
GACC vf, d, U4, U2
|
|
vilvl.d U2, U4, U4
|
|
|
|
.L03: /* &4 */
|
|
sub.d T0, M, J
|
|
addi.d T0, T0, -1
|
|
andi T0, T0, 4
|
|
beq $r0, T0, .L04
|
|
|
|
mul.d T1, J, LDA
|
|
add.d T1, T1, II
|
|
addi.d T2, T1, 16
|
|
|
|
vldx U1, AO1, T1
|
|
vldx U14, AO1, T2
|
|
|
|
add.d T1, IY, INCY
|
|
add.d T2, T1, INCY
|
|
add.d T3, T2, INCY
|
|
add.d T4, T3, INCY
|
|
|
|
fldx.d $f4, Y, T1
|
|
fldx.d $f5, Y, T2
|
|
fldx.d $f6, Y, T3
|
|
fldx.d $f7, Y, T4
|
|
|
|
vextrins.d U4, U5, 0x10
|
|
vextrins.d U6, U7, 0x10
|
|
|
|
vfmadd.d U4, U3, U1, U4
|
|
vfmadd.d U6, U3, U14, U6
|
|
|
|
vextrins.d U5, U4, 0x01
|
|
vextrins.d U7, U6, 0x01
|
|
|
|
fstx.d $f4, Y, T1
|
|
fstx.d $f5, Y, T2
|
|
fstx.d $f6, Y, T3
|
|
fstx.d $f7, Y, T4
|
|
|
|
slli.d T1, INCY, 2
|
|
add.d IY, IY, T1
|
|
|
|
add.d T1, IX, INCX
|
|
add.d T2, T1, INCX
|
|
add.d T3, T2, INCX
|
|
add.d T4, T3, INCX
|
|
|
|
fldx.d $f4, X, T1
|
|
fldx.d $f5, X, T2
|
|
fldx.d $f6, X, T3
|
|
fldx.d $f7, X, T4
|
|
|
|
vextrins.d U4, U5, 0x10
|
|
vextrins.d U6, U7, 0x10
|
|
|
|
vand.v $vr12, $vr2, $vr2
|
|
|
|
vfmadd.d U2, U1, U4, U2
|
|
vfsub.d U2, U2, $vr12
|
|
vfmadd.d U2, U14, U6, U2
|
|
|
|
vextrins.d U4, U2, 0x01
|
|
|
|
fadd.d $f2, $f2, $f4
|
|
fadd.d $f2, $f2, $f12
|
|
|
|
vextrins.d U2, U2, 0x10
|
|
|
|
slli.d T2, INCX, 2
|
|
add.d IX, IX, T2
|
|
|
|
addi.d II, II, 32
|
|
|
|
.L04: /* &2 */
|
|
sub.d T0, M, J
|
|
addi.d T0, T0, -1
|
|
andi T0, T0, 2
|
|
beq $r0, T0, .L05
|
|
|
|
mul.d T1, J, LDA
|
|
add.d T1, T1, II
|
|
|
|
vldx U1, AO1, T1
|
|
|
|
add.d T1, IY, INCY
|
|
add.d T2, T1, INCY
|
|
|
|
fldx.d $f6, Y, T1
|
|
fldx.d $f7, Y, T2
|
|
|
|
vextrins.d U6, U7, 0x10
|
|
vfmadd.d U6, U3, U1, U6
|
|
vextrins.d U7, U6, 0x01
|
|
|
|
fstx.d $f6, Y, T1
|
|
fstx.d $f7, Y, T2
|
|
|
|
slli.d T1, INCY, 1
|
|
add.d IY, IY, T1
|
|
|
|
add.d T1, IX, INCX
|
|
add.d T2, T1, INCX
|
|
|
|
fldx.d $f6, X, T1
|
|
fldx.d $f7, X, T2
|
|
|
|
vextrins.d U6, U7, 0x10
|
|
vand.v U12, U2, U2
|
|
|
|
vfmadd.d U2, U1, U6, U2
|
|
vfsub.d U2, U2, U12
|
|
|
|
vextrins.d U4, U2, 0x01
|
|
fadd.d $f2, $f2, $f4
|
|
fadd.d $f2, $f2, $f12
|
|
|
|
vextrins.d U2, U2, 0x10
|
|
|
|
slli.d T2, INCX, 1
|
|
add.d IX, IX, T2
|
|
|
|
addi.d II, II, 16
|
|
|
|
.L05: /* &1 */
|
|
sub.d T0, M, J
|
|
addi.d T0, T0, -1
|
|
andi T0, T0, 1
|
|
beq $r0, T0, .L06
|
|
|
|
mul.d T1, J, LDA
|
|
add.d T1, T1, II
|
|
|
|
fldx.d $f4, AO1, T1
|
|
add.d IY, IY, INCY
|
|
fldx.d $f6, Y, IY
|
|
fmadd.d $f6, $f3, $f4, $f6
|
|
fstx.d $f6, Y, IY
|
|
|
|
add.d IX, IX, INCX
|
|
fldx.d $f6, X, IX
|
|
fmadd.d $f2, $f4, $f6, $f2
|
|
|
|
addi.d II, II, 8
|
|
|
|
.L06:
|
|
fldx.d $f6, Y, JY
|
|
fmadd.d $f6, ALPHA, $f2, $f6
|
|
fstx.d $f6, Y, JY
|
|
|
|
add.d JX, JX, INCX
|
|
add.d JY, JY, INCY
|
|
|
|
addi.d J, J, 1
|
|
blt J, N, .L01
|
|
|
|
.L999:
|
|
LDARG $r23, $sp, 0
|
|
LDARG $r24, $sp, 8
|
|
LDARG $r25, $sp, 16
|
|
LDARG $r26, $sp, 32
|
|
LDARG $r27, $sp, 40
|
|
LDARG $r28, $sp, 48
|
|
LDARG $r29, $sp, 56
|
|
LDARG $r30, $sp, 64
|
|
LDARG $r31, $sp, 72
|
|
|
|
addi.d $sp, $sp, 88
|
|
jirl $r0, $r1, 0x0
|
|
|
|
EPILOGUE
|