#include "textflag.h"

TEXT ·Index(SB),NOSPLIT,$0-56
	MOVQ a_base+0(FP), DI
	MOVQ a_len+8(FP), DX
	MOVQ b_base+24(FP), R8
	MOVQ b_len+32(FP), AX
	MOVQ DI, R10
	LEAQ ret+48(FP), R11
	JMP  indexbody<>(SB)

TEXT ·IndexString(SB),NOSPLIT,$0-40
	MOVQ a_base+0(FP), DI
	MOVQ a_len+8(FP), DX
	MOVQ b_base+16(FP), R8
	MOVQ b_len+24(FP), AX
	MOVQ DI, R10
	LEAQ ret+32(FP), R11
	JMP  indexbody<>(SB)

#include "go_asm.h"
#include "textflag.h"

TEXT ·Count(SB),NOSPLIT,$0-40
	MOVD	b_base+0(FP), R0
	MOVD	b_len+8(FP), R2
	MOVBU	c+24(FP), R1
	MOVD	$ret+32(FP), R8
	B	countbytebody<>(SB)

TEXT ·CountString(SB),NOSPLIT,$0-32
	MOVD	s_base+0(FP), R0
	MOVD	s_len+8(FP), R2
	MOVBU	c+16(FP), R1
	MOVD	$ret+24(FP), R8
	B	countbytebody<>(SB)

// input:
//   R0: data
//   R2: data len

	I32WrapI64
	I32Load8U c+24(FP)
	I64Load b_len+8(FP)
	I32WrapI64
	Call memchr<>(SB)
	I64ExtendI32S
	Set R0

	Get SP
	I64Const $-1
	Get R0
	I64Load b_base+0(FP)
	I64Sub
	Get R0
	I64Eqz $0
	Select
	I64Store ret+32(FP)

	RET

TEXT ·IndexByteString(SB), NOSPLIT, $0-32
	Get SP
	I64Load s_base+0(FP)
	I32WrapI64
	I32Load8U c+16(FP)
	I64Load s_len+8(FP)
	I32WrapI64
	Call memchr<>(SB)

//go:build !plan9

#include "go_asm.h"
#include "textflag.h"

TEXT	·IndexByte(SB), NOSPLIT, $0-40
	MOVQ b_base+0(FP), SI
	MOVQ b_len+8(FP), BX
	MOVB c+24(FP), AL
	LEAQ ret+32(FP), R8
	JMP  indexbytebody<>(SB)

TEXT	·IndexByteString(SB), NOSPLIT, $0-32
	MOVQ s_base+0(FP), SI
	MOVQ s_len+8(FP), BX
	MOVB c+16(FP), AL
	LEAQ ret+24(FP), R8
	JMP  indexbytebody<>(SB)

// input:
//   SI: data
//   BX: data len

// license that can be found in the LICENSE file.

#include "textflag.h"

TEXT ·IndexByte(SB),NOSPLIT,$0-40
	MOVD	b_base+0(FP), R0
	MOVD	b_len+8(FP), R2
	MOVBU	c+24(FP), R1
	MOVD	$ret+32(FP), R8
	B	indexbytebody<>(SB)

TEXT ·IndexByteString(SB),NOSPLIT,$0-32
	MOVD	s_base+0(FP), R0
	MOVD	s_len+8(FP), R2
	MOVBU	c+16(FP), R1
	MOVD	$ret+24(FP), R8
	B	indexbytebody<>(SB)

// input:
//   R0: data
//   R1: byte to search

	JEQ	2(PC)
	JMP	·countGeneric(SB)
#endif
	MOVQ	b_base+0(FP), SI
	MOVQ	b_len+8(FP), BX
	MOVB	c+24(FP), AL
	LEAQ	ret+32(FP), R8
	JMP	countbody<>(SB)

TEXT ·CountString(SB),NOSPLIT,$0-32
#ifndef hasPOPCNT
	CMPB	internal∕cpu·X86+const_offsetX86HasPOPCNT(SB), $1
	JEQ	2(PC)
	JMP	·countGenericString(SB)
#endif
	MOVQ	s_base+0(FP), SI
	MOVQ	s_len+8(FP), BX
	MOVB	c+16(FP), AL
	LEAQ	ret+24(FP), R8
	JMP	countbody<>(SB)

	MVC	$16, 0(R1), 0(R8) // update iv
	RET

// func xorBytes(dst, a, b []byte) int
TEXT ·xorBytes(SB),NOSPLIT,$0-80
	MOVD	dst_base+0(FP), R1
	MOVD	a_base+24(FP), R2
	MOVD	b_base+48(FP), R3
	MOVD	a_len+32(FP), R4
	MOVD	b_len+56(FP), R5
	CMPBLE	R4, R5, skip
	MOVD	R5, R4
skip:
	MOVD	R4, ret+72(FP)
	MOVD	$0, R5
	CMPBLT	R4, $8, tail
loop:
	MOVD	0(R2)(R5*1), R7
	MOVD	0(R3)(R5*1), R8
	XOR	R7, R8

	{"_seek<>(SB)", "_seek<>(SB)"},
	{"a2+16(FP)", "a2+16(FP)"},
	{"addr2+24(FP)", "addr2+24(FP)"},
	{"asmcgocall<>(SB)", "asmcgocall<>(SB)"},
	{"b+24(FP)", "b+24(FP)"},
	{"b_len+32(FP)", "b_len+32(FP)"},
	{"racecall<>(SB)", "racecall<>(SB)"},
	{"rcv_name+20(FP)", "rcv_name+20(FP)"},
	{"retoffset+28(FP)", "retoffset+28(FP)"},
	{"runtime·_GetStdHandle(SB)", "runtime._GetStdHandle(SB)"},

	MOVD y+48(FP), R4	// R4 = y = c
	MOVD z_len+8(FP), R11	// R11 = z_len

	CMP   R11, $0		// If z_len is zero, return
	BEQ   done

	// We will process the first iteration out of the loop so we capture
	// the value of c. In the subsequent iterations, we will rely on the
	// value of CA set here.
	MOVD  0(R8), R20	// R20 = x[i]
	ADD   $-1, R11		// R11 = z_len - 1
	ADDC  R20, R4, R6	// R6 = x[i] + c

	// Repeat until all 16-byte chunks are done.
	BRCTG R_ITER, vxloop

	// Skip to end if there are no trailing bytes.
	CIJ $EQ, R_LEN, $-1, vxret

	// Load 1-15 bytes and corresponding mask.
	// Note: only the low 32-bits of R_LEN are used for the index.
	VLL R_LEN, (R_PTR), V_VAL
	VLL R_LEN, (R_MPTR), V_MASK

	// Compare each byte in input chunk against byte to be counted.