68 lines
1.1 KiB
ArmAsm
68 lines
1.1 KiB
ArmAsm
|
// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
|
||
|
|
// SPDX-License-Identifier: BSD-3-Clause
|
||
|
|
|
||
|
|
#include "textflag.h"
|
||
|
|
|
||
|
|
// func add(a, b int64) int64
|
||
|
|
TEXT ·add(SB), NOSPLIT, $0-24
|
||
|
|
MOVQ a+0(FP), AX
|
||
|
|
ADDQ b+8(FP), AX
|
||
|
|
MOVQ AX, ret+16(FP)
|
||
|
|
RET
|
||
|
|
|
||
|
|
// func sum(data []int64) int64
|
||
|
|
// Sums all elements of the slice.
|
||
|
|
TEXT ·sum(SB), NOSPLIT, $0-32
|
||
|
|
MOVQ data_base+0(FP), SI
|
||
|
|
MOVQ data_len+8(FP), CX
|
||
|
|
XORQ AX, AX
|
||
|
|
TESTQ CX, CX
|
||
|
|
JZ sum_done
|
||
|
|
|
||
|
|
sum_loop:
|
||
|
|
ADDQ (SI), AX
|
||
|
|
ADDQ $8, SI
|
||
|
|
DECQ CX
|
||
|
|
JNZ sum_loop
|
||
|
|
|
||
|
|
sum_done:
|
||
|
|
MOVQ AX, ret+24(FP)
|
||
|
|
RET
|
||
|
|
|
||
|
|
// func wideCopy(dst, src []byte)
|
||
|
|
// Non-overlapping copy of min(len(dst), len(src)) bytes using 32-byte moves.
|
||
|
|
TEXT ·wideCopy(SB), NOSPLIT, $0-48
|
||
|
|
MOVQ dst_base+0(FP), DI
|
||
|
|
MOVQ dst_len+8(FP), BX
|
||
|
|
MOVQ src_base+24(FP), SI
|
||
|
|
MOVQ src_len+32(FP), R8
|
||
|
|
CMPQ BX, R8
|
||
|
|
JLE wc_have_n
|
||
|
|
MOVQ R8, BX
|
||
|
|
|
||
|
|
wc_have_n:
|
||
|
|
CMPQ BX, $32
|
||
|
|
JB wc_small
|
||
|
|
|
||
|
|
VMOVDQU (SI), Y0
|
||
|
|
VMOVDQU Y0, (DI)
|
||
|
|
VMOVDQU -32(SI)(BX*1), Y0
|
||
|
|
VMOVDQU Y0, -32(DI)(BX*1)
|
||
|
|
VZEROUPPER
|
||
|
|
RET
|
||
|
|
|
||
|
|
wc_small:
|
||
|
|
TESTQ BX, BX
|
||
|
|
JZ wc_done
|
||
|
|
|
||
|
|
wc_byte:
|
||
|
|
MOVB (SI), R8B
|
||
|
|
MOVB R8B, (DI)
|
||
|
|
INCQ SI
|
||
|
|
INCQ DI
|
||
|
|
DECQ BX
|
||
|
|
JNZ wc_byte
|
||
|
|
|
||
|
|
wc_done:
|
||
|
|
RET
|