175 lines
3.2 KiB
ArmAsm
175 lines
3.2 KiB
ArmAsm
#define ASSEMBLER
|
|
|
|
#include "common.h"
|
|
|
|
#define N $r4
|
|
#define X $r5
|
|
#define INCX $r6
|
|
#define I $r12
|
|
#define J $r13
|
|
#define t1 $r14
|
|
#define t2 $r18
|
|
#define t3 $r15
|
|
#define t4 $r17
|
|
#define TEMP $r16
|
|
#define m0 $vr8
|
|
#define x1 $vr9
|
|
#define x2 $vr10
|
|
#define x3 $vr11
|
|
#define x4 $vr12
|
|
#define VX0 $vr20
|
|
#define VX1 $vr21
|
|
#define VM0 $vr22
|
|
#define VM1 $vr23
|
|
|
|
PROLOGUE
|
|
|
|
bge $r0, N, .L999
|
|
bge $r0, INCX, .L999
|
|
li.d TEMP, 1
|
|
slli.d TEMP, TEMP, BASE_SHIFT
|
|
slli.d INCX, INCX, BASE_SHIFT
|
|
bne INCX, TEMP, .L20
|
|
vld VM0, X, 0
|
|
srai.d I, N, 3
|
|
bge $r0, I, .L12
|
|
.align 3
|
|
|
|
.L10:
|
|
vld VX0, X, 0 * SIZE
|
|
vld VX1, X, 4 * SIZE
|
|
addi.d I, I, -1
|
|
vfmin.s VM1, VX0, VX1
|
|
addi.d X, X, 8 * SIZE
|
|
vfmin.s VM0, VM0, VM1
|
|
blt $r0, I, .L10
|
|
.align 3
|
|
|
|
.L11:
|
|
vreplvei.w x1, VM0, 0
|
|
vreplvei.w x2, VM0, 1
|
|
vreplvei.w x3, VM0, 2
|
|
vreplvei.w x4, VM0, 3
|
|
vfmin.s VM1, x1, x2
|
|
vfmin.s VM0, x3, x4
|
|
vfmin.s VM0, VM0, VM1
|
|
.align 3
|
|
|
|
.L12: //INCX==1 and N<8
|
|
andi I, N, 7
|
|
li.d J, 4
|
|
bge J, I, .L13 // 4<N<8
|
|
vld VX0, X, 0
|
|
slli.d J, J, 1 // 8
|
|
sub.d I, J, I
|
|
slli.d I, I, BASE_SHIFT
|
|
vldx VX1, X, I
|
|
vfmin.s m0, VX0, VX1 //patial repeat read
|
|
vreplvei.w x1, m0, 0
|
|
vreplvei.w x2, m0, 1
|
|
vreplvei.w x3, m0, 2
|
|
vreplvei.w x4, m0, 3
|
|
vfmin.s m0, x1, x2
|
|
vfmin.s VM1, x3, x4
|
|
vfmin.s m0, m0, VM1
|
|
vfmin.s VM0, m0, VM0
|
|
fmov.s $f0, $f22
|
|
jirl $r0, $r1, 0x0
|
|
.align 3
|
|
|
|
.L13: //INCX==1 and 0<=N<=4
|
|
bge $r0, I, .L15
|
|
.align 3
|
|
|
|
.L14:
|
|
vld x1, X, 0
|
|
addi.d I, I, -1
|
|
vfmin.s VM0, VM0, x1
|
|
addi.d X, X, SIZE
|
|
blt $r0, I, .L14
|
|
.align 3
|
|
|
|
.L15:
|
|
fmov.s $f0, $f22
|
|
jirl $r0, $r1, 0x0
|
|
.align 3
|
|
|
|
.L20: // INCX!=1
|
|
move TEMP, X
|
|
ld.w t1, TEMP, 0 * SIZE
|
|
add.d TEMP, TEMP, INCX
|
|
vinsgr2vr.w VM0, t1, 0
|
|
srai.d I, N, 3
|
|
bge $r0, I, .L23
|
|
ld.w t2, TEMP, 0 * SIZE
|
|
add.d TEMP, TEMP, INCX
|
|
ld.w t3, TEMP, 0 * SIZE
|
|
add.d TEMP, TEMP, INCX
|
|
ld.w t4, TEMP, 0 * SIZE
|
|
add.d TEMP, TEMP, INCX
|
|
vinsgr2vr.w VM0, t2, 1
|
|
vinsgr2vr.w VM0, t3, 2
|
|
vinsgr2vr.w VM0, t4, 3
|
|
.align 3
|
|
|
|
.L21:
|
|
ld.w t1, X, 0 * SIZE
|
|
add.d X, X, INCX
|
|
ld.w t2, X, 0 * SIZE
|
|
add.d X, X, INCX
|
|
ld.w t3, X, 0 * SIZE
|
|
add.d X, X, INCX
|
|
ld.w t4, X, 0 * SIZE
|
|
add.d X, X, INCX
|
|
vinsgr2vr.w VX0, t1, 0
|
|
vinsgr2vr.w VX0, t2, 1
|
|
vinsgr2vr.w VX0, t3, 2
|
|
vinsgr2vr.w VX0, t4, 3
|
|
ld.w t1, X, 0 * SIZE
|
|
add.d X, X, INCX
|
|
ld.w t2, X, 0 * SIZE
|
|
add.d X, X, INCX
|
|
ld.w t3, X, 0 * SIZE
|
|
add.d X, X, INCX
|
|
ld.w t4, X, 0 * SIZE
|
|
add.d X, X, INCX
|
|
vinsgr2vr.w VX1, t1, 0
|
|
vinsgr2vr.w VX1, t2, 1
|
|
vinsgr2vr.w VX1, t3, 2
|
|
vinsgr2vr.w VX1, t4, 3
|
|
vfmin.s VM1, VX0, VX1
|
|
addi.d I, I, -1
|
|
vfmin.s VM0, VM0, VM1
|
|
blt $r0, I, .L21
|
|
.align 3
|
|
|
|
.L22:
|
|
vreplvei.w x1, VM0, 0
|
|
vreplvei.w x2, VM0, 1
|
|
vreplvei.w x3, VM0, 2
|
|
vreplvei.w x4, VM0, 3
|
|
vfmin.s VM1, x1, x2
|
|
vfmin.s VM0, x3, x4
|
|
vfmin.s VM0, VM0, VM1
|
|
.align 3
|
|
|
|
.L23: //INCX!=1 and N<8
|
|
andi I, N, 7
|
|
bge $r0, I, .L999
|
|
.align 3
|
|
|
|
.L24:
|
|
vld x1, X, 0
|
|
vfmin.s VM0, VM0, x1
|
|
addi.d I, I, -1
|
|
add.d X, X, INCX
|
|
blt $r0, I, .L24
|
|
.align 3
|
|
|
|
.L999:
|
|
fmov.s $f0, $f22
|
|
jirl $r0, $r1, 0x0
|
|
.align 3
|
|
|
|
EPILOGUE
|