374 lines
8.6 KiB
ArmAsm
374 lines
8.6 KiB
ArmAsm
#define ASSEMBLER
|
|
|
|
#include "common.h"
|
|
|
|
#define N $r4
|
|
#define X $r5
|
|
#define INCX $r6
|
|
#define I $r12
|
|
#define t1 $r13
|
|
#define t2 $r15
|
|
#define t3 $r18
|
|
#define t4 $r16
|
|
#define i0 $r17
|
|
#define i1 $r14
|
|
#define TEMP $r19
|
|
#define x1 $xr9
|
|
#define x2 $xr10
|
|
#define x3 $xr11
|
|
#define x4 $xr12
|
|
#define VX0 $xr13
|
|
#define VX1 $xr14
|
|
#define VM0 $xr15
|
|
#define VM1 $xr16
|
|
#define VINC4 $xr17
|
|
#define VINC8 $xr18
|
|
#define VI0 $xr20
|
|
#define VI1 $xr21
|
|
#define VI2 $xr22
|
|
#define VI3 $xr8
|
|
#define VI4 $xr19
|
|
#define VT0 $xr23
|
|
|
|
PROLOGUE
|
|
li.d i0, 0
|
|
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
|
|
xvld VM0, X, 0
|
|
addi.w i0, i0, 1
|
|
srai.d I, N, 3
|
|
bge $r0, I, .L21
|
|
slli.w i0, i0, 3 //8
|
|
xvreplgr2vr.w VINC8, i0
|
|
addi.w i0, i0, -15
|
|
xvinsgr2vr.w VI1, i0, 0 //initialize the index value for vectorization
|
|
addi.w i0, i0, 1
|
|
xvinsgr2vr.w VI1, i0, 1
|
|
addi.w i0, i0, 1
|
|
xvinsgr2vr.w VI1, i0, 2
|
|
addi.w i0, i0, 1
|
|
xvinsgr2vr.w VI1, i0, 3
|
|
addi.w i0, i0, 1
|
|
xvinsgr2vr.w VI1, i0, 4
|
|
addi.w i0, i0, 1
|
|
xvinsgr2vr.w VI1, i0, 5
|
|
addi.w i0, i0, 1
|
|
xvinsgr2vr.w VI1, i0, 6
|
|
addi.w i0, i0, 1
|
|
xvinsgr2vr.w VI1, i0, 7
|
|
addi.w i0, i0, 1
|
|
xvinsgr2vr.w VI0, i0, 0 //1
|
|
addi.w i0, i0, 1
|
|
xvinsgr2vr.w VI0, i0, 1 //2
|
|
addi.w i0, i0, 1
|
|
xvinsgr2vr.w VI0, i0, 2 //3
|
|
addi.w i0, i0, 1
|
|
xvinsgr2vr.w VI0, i0, 3 //4
|
|
addi.w i0, i0, 1
|
|
xvinsgr2vr.w VI0, i0, 4 //5
|
|
addi.w i0, i0, 1
|
|
xvinsgr2vr.w VI0, i0, 5 //6
|
|
addi.w i0, i0, 1
|
|
xvinsgr2vr.w VI0, i0, 6 //7
|
|
addi.w i0, i0, 1
|
|
xvinsgr2vr.w VI0, i0, 7 //8
|
|
.align 3
|
|
|
|
.L10:
|
|
xvld VX0, X, 0 * SIZE
|
|
xvadd.w VI1, VI1, VINC8
|
|
xvfcmp.clt.s VT0, VX0, VM0
|
|
addi.d I, I, -1
|
|
xvbitsel.v VM0, VM0, VX0, VT0
|
|
xvbitsel.v VI0, VI0, VI1, VT0
|
|
addi.d X, X, 8 * SIZE
|
|
blt $r0, I, .L10
|
|
.align 3
|
|
|
|
.L15:
|
|
xvxor.v VX0, VX0, VX0
|
|
xvor.v VX0, VI0, VX0
|
|
xvxor.v VX1, VX1, VX1
|
|
xvor.v VX1, VM0, VX1
|
|
xvpickve.w VI1, VI0, 0
|
|
xvpickve.w VI2, VI0, 1
|
|
xvpickve.w VI3, VI0, 2
|
|
xvpickve.w VI4, VI0, 3
|
|
xvpickve.w x1, VM0, 0
|
|
xvpickve.w x2, VM0, 1
|
|
xvpickve.w x3, VM0, 2
|
|
xvpickve.w x4, VM0, 3
|
|
xvfcmp.clt.s VT0, x2, x1
|
|
xvbitsel.v VM1, x1, x2, VT0
|
|
xvbitsel.v VINC4, VI1, VI2, VT0
|
|
xvfcmp.clt.s VT0, x4, x3
|
|
xvbitsel.v VM0, x3, x4, VT0
|
|
xvbitsel.v VINC8, VI3, VI4, VT0
|
|
xvfcmp.clt.s VT0, VM1, VM0
|
|
xvbitsel.v VM0, VM0, VM1, VT0
|
|
xvbitsel.v VI0, VINC8, VINC4, VT0
|
|
li.d TEMP, 1 //处理尾数相等时取最小序号
|
|
movgr2fr.w $f17, TEMP
|
|
ffint.s.w $f17, $f17
|
|
xvfcmp.ceq.s VT0, x1, VM0
|
|
fcmp.ceq.s $fcc0, $f23, $f17
|
|
bceqz $fcc0, .L26
|
|
xvfcmp.clt.s VT0, VI1, VI0
|
|
xvbitsel.v VI0, VI0, VI1, VT0
|
|
b .L26
|
|
.align 3
|
|
|
|
.L20: // INCX!=1
|
|
move TEMP, X
|
|
addi.w i0, i0, 1
|
|
ld.w t1, TEMP, 0 * SIZE
|
|
add.d TEMP, TEMP, INCX
|
|
srai.d I, N, 3
|
|
bge $r0, I, .L21
|
|
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
|
|
xvinsgr2vr.w VM0, t1, 0
|
|
xvinsgr2vr.w VM0, t2, 1
|
|
xvinsgr2vr.w VM0, t3, 2
|
|
xvinsgr2vr.w VM0, t4, 3
|
|
ld.w t1, TEMP, 0 * SIZE
|
|
add.d TEMP, TEMP, INCX
|
|
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
|
|
xvinsgr2vr.w VM0, t1, 4
|
|
xvinsgr2vr.w VM0, t2, 5
|
|
xvinsgr2vr.w VM0, t3, 6
|
|
xvinsgr2vr.w VM0, t4, 7
|
|
slli.w i0, i0, 3 //8
|
|
xvreplgr2vr.w VINC8, i0
|
|
addi.w i0, i0, -15
|
|
xvinsgr2vr.w VI1, i0, 0 //initialize the index value for vectorization
|
|
addi.w i0, i0, 1
|
|
xvinsgr2vr.w VI1, i0, 1
|
|
addi.w i0, i0, 1
|
|
xvinsgr2vr.w VI1, i0, 2
|
|
addi.w i0, i0, 1
|
|
xvinsgr2vr.w VI1, i0, 3
|
|
addi.w i0, i0, 1
|
|
xvinsgr2vr.w VI1, i0, 4
|
|
addi.w i0, i0, 1
|
|
xvinsgr2vr.w VI1, i0, 5
|
|
addi.w i0, i0, 1
|
|
xvinsgr2vr.w VI1, i0, 6
|
|
addi.w i0, i0, 1
|
|
xvinsgr2vr.w VI1, i0, 7
|
|
addi.w i0, i0, 1
|
|
xvinsgr2vr.w VI0, i0, 0 //1
|
|
addi.w i0, i0, 1
|
|
xvinsgr2vr.w VI0, i0, 1 //2
|
|
addi.w i0, i0, 1
|
|
xvinsgr2vr.w VI0, i0, 2 //3
|
|
addi.w i0, i0, 1
|
|
xvinsgr2vr.w VI0, i0, 3 //4
|
|
addi.w i0, i0, 1
|
|
xvinsgr2vr.w VI0, i0, 4 //5
|
|
addi.w i0, i0, 1
|
|
xvinsgr2vr.w VI0, i0, 5 //6
|
|
addi.w i0, i0, 1
|
|
xvinsgr2vr.w VI0, i0, 6 //7
|
|
addi.w i0, i0, 1
|
|
xvinsgr2vr.w VI0, i0, 7 //8
|
|
.align 3
|
|
|
|
.L24:
|
|
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
|
|
xvinsgr2vr.w VX0, t1, 0
|
|
xvinsgr2vr.w VX0, t2, 1
|
|
xvinsgr2vr.w VX0, t3, 2
|
|
xvinsgr2vr.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
|
|
xvinsgr2vr.w VX0, t1, 4
|
|
xvinsgr2vr.w VX0, t2, 5
|
|
xvinsgr2vr.w VX0, t3, 6
|
|
xvinsgr2vr.w VX0, t4, 7
|
|
xvadd.w VI1, VI1, VINC8
|
|
xvfcmp.clt.s VT0, VX0, VM0
|
|
addi.d I, I, -1
|
|
xvbitsel.v VM0, VM0, VX0, VT0
|
|
xvbitsel.v VI0, VI0, VI1, VT0
|
|
blt $r0, I, .L24
|
|
.align 3
|
|
|
|
.L25:
|
|
xvxor.v VX0, VX0, VX0
|
|
xvor.v VX0, VI0, VX0
|
|
xvxor.v VX1, VX1, VX1
|
|
xvor.v VX1, VM0, VX1
|
|
xvpickve.w VI1, VI0, 0
|
|
xvpickve.w VI2, VI0, 1
|
|
xvpickve.w VI3, VI0, 2
|
|
xvpickve.w VI4, VI0, 3
|
|
xvpickve.w x1, VM0, 0
|
|
xvpickve.w x2, VM0, 1
|
|
xvpickve.w x3, VM0, 2
|
|
xvpickve.w x4, VM0, 3
|
|
xvfcmp.clt.s VT0, x2, x1
|
|
xvbitsel.v VM1, x1, x2, VT0
|
|
xvbitsel.v VINC4, VI1, VI2, VT0
|
|
xvfcmp.clt.s VT0, x4, x3
|
|
xvbitsel.v VM0, x3, x4, VT0
|
|
xvbitsel.v VINC8, VI3, VI4, VT0
|
|
xvfcmp.clt.s VT0, VM1, VM0
|
|
xvbitsel.v VM0, VM0, VM1, VT0
|
|
xvbitsel.v VI0, VINC8, VINC4, VT0
|
|
li.d TEMP, 1 //处理尾数相等时取最小序号
|
|
movgr2fr.w $f17, TEMP
|
|
ffint.s.w $f17, $f17
|
|
xvfcmp.ceq.s VT0, VM0, x1
|
|
fcmp.ceq.s $fcc0, $f23, $f17
|
|
bceqz $fcc0, .L26
|
|
xvfcmp.clt.s VT0, VI1, VI0
|
|
xvbitsel.v VI0, VI0, VI1, VT0
|
|
.align 3
|
|
|
|
.L26:
|
|
xvfcmp.ceq.s VT0, VM0, x2
|
|
fcmp.ceq.s $fcc0, $f23, $f17
|
|
bceqz $fcc0, .L27
|
|
xvfcmp.clt.s VT0, VI2, VI0
|
|
xvbitsel.v VI0, VI0, VI2, VT0
|
|
.align 3
|
|
|
|
.L27:
|
|
xvfcmp.ceq.s VT0, VM0, x3
|
|
fcmp.ceq.s $fcc0, $f23, $f17
|
|
bceqz $fcc0, .L28
|
|
xvfcmp.clt.s VT0, VI3, VI0
|
|
xvbitsel.v VI0, VI0, VI3, VT0
|
|
.align 3
|
|
|
|
.L28:
|
|
xvfcmp.ceq.s VT0, VM0, x4
|
|
fcmp.ceq.s $fcc0, $f23, $f17
|
|
bceqz $fcc0, .L29
|
|
xvfcmp.clt.s VT0, VI4, VI0
|
|
xvbitsel.v VI0, VI0, VI4, VT0
|
|
.align 3
|
|
|
|
.L29:
|
|
fmov.s $f16, $f20
|
|
.align 3
|
|
|
|
.L252:
|
|
xvxor.v VI0, VI0, VI0
|
|
xvor.v VI0, VI0, VX0
|
|
fmov.s $f13, $f15
|
|
xvxor.v VM0, VM0, VM0
|
|
xvor.v VM0, VM0, VX1
|
|
xvpickve.w VI1, VI0, 4
|
|
xvpickve.w VI2, VI0, 5
|
|
xvpickve.w VI3, VI0, 6
|
|
xvpickve.w VI4, VI0, 7
|
|
xvpickve.w x1, VM0, 4
|
|
xvpickve.w x2, VM0, 5
|
|
xvpickve.w x3, VM0, 6
|
|
xvpickve.w x4, VM0, 7
|
|
xvfcmp.clt.s VT0, x2, x1
|
|
xvbitsel.v x1, x1, x2, VT0
|
|
xvbitsel.v VINC4, VI1, VI2, VT0
|
|
xvfcmp.clt.s VT0, x4, x3
|
|
xvbitsel.v VM0, x3, x4, VT0
|
|
xvbitsel.v VINC8, VI3, VI4, VT0
|
|
xvfcmp.clt.s VT0, x1, VM0
|
|
xvbitsel.v VM0, VM0, x1, VT0
|
|
xvbitsel.v VI0, VINC8, VINC4, VT0
|
|
li.d TEMP, 1 //处理尾数相等时取最小序号
|
|
movgr2fr.w $f17, TEMP
|
|
ffint.s.w $f17, $f17
|
|
xvfcmp.ceq.s VT0, VM0, x1
|
|
fcmp.ceq.s $fcc0, $f23, $f17
|
|
bceqz $fcc0, .L262
|
|
xvfcmp.clt.s VT0, VI1, VI0
|
|
xvbitsel.v VI0, VI0, VI1, VT0
|
|
.align 3
|
|
|
|
.L262:
|
|
xvfcmp.ceq.s VT0, VM0, x2
|
|
fcmp.ceq.s $fcc0, $f23, $f17
|
|
bceqz $fcc0, .L272
|
|
xvfcmp.clt.s VT0, VI2, VI0
|
|
xvbitsel.v VI0, VI0, VI2, VT0
|
|
.align 3
|
|
|
|
.L272:
|
|
xvfcmp.ceq.s VT0, VM0, x3
|
|
fcmp.ceq.s $fcc0, $f23, $f17
|
|
bceqz $fcc0, .L282
|
|
xvfcmp.clt.s VT0, VI3, VI0
|
|
xvbitsel.v VI0, VI0, VI3, VT0
|
|
.align 3
|
|
|
|
.L282:
|
|
xvfcmp.ceq.s VT0, VM0, x4
|
|
fcmp.ceq.s $fcc0, $f23, $f17
|
|
bceqz $fcc0, .L292
|
|
xvfcmp.clt.s VT0, VI4, VI0
|
|
xvbitsel.v VI0, VI0, VI4, VT0
|
|
.align 3
|
|
|
|
.L292:
|
|
fcmp.clt.s $fcc0, $f13, $f15
|
|
fsel $f15, $f15, $f13, $fcc0
|
|
fsel $f20, $f20, $f16, $fcc0
|
|
movfr2gr.s i0, $f20
|
|
|
|
.L21: //N<8
|
|
andi I, N, 7
|
|
bge $r0, I, .L999
|
|
srai.d i1, N, 3
|
|
slli.d i1, i1, 3
|
|
addi.d i1, i1, 1 //current index
|
|
movgr2fr.d $f21, i1
|
|
movgr2fr.d $f20, i0
|
|
.align 3
|
|
|
|
.L22:
|
|
fld.d $f9, X, 0
|
|
addi.d I, I, -1
|
|
fcmp.clt.s $fcc0, $f9, $f15
|
|
fsel $f15, $f15, $f9, $fcc0
|
|
fsel $f20, $f20, $f21, $fcc0
|
|
addi.d i1, i1, 1
|
|
movgr2fr.d $f21, i1
|
|
add.d X, X, INCX
|
|
blt $r0, I, .L22
|
|
movfr2gr.s i0, $f20
|
|
.align 3
|
|
|
|
.L999:
|
|
move $r4, $r17
|
|
jirl $r0, $r1, 0x0
|
|
.align 3
|
|
|
|
EPILOGUE |