TEXT ·p256Sqr(SB),NOSPLIT,$0
	MOVD	res+0(FP), res_ptr
	MOVD	in+8(FP), a_ptr
	MOVD	n+16(FP), b_ptr

	MOVD	p256const0<>(SB), const0
	MOVD	p256const1<>(SB), const1

	LDP	0*16(a_ptr), (x0, x1)
	LDP	1*16(a_ptr), (x2, x3)

sqrLoop:
	SUB	$1, b_ptr
	CALL	p256SqrInternal<>(SB)
	MOVD	y0, x0
	MOVD	y1, x1
	MOVD	y2, x2
	MOVD	y3, x3
	CBNZ	b_ptr, sqrLoop

// Use a local trampoline, to avoid taking the address of a dynamically exported
// function.
TEXT ·set_crosscall2(SB),NOSPLIT,$0-0
	MOVD	_crosscall2_ptr(SB), R1
	MOVD	$crosscall2_trampoline<>(SB), R2
	MOVD	R2, (R1)
	RET

TEXT crosscall2_trampoline<>(SB),NOSPLIT,$0-0
	JMP	crosscall2(SB)

// Called by C code generated by cmd/cgo.

	RET
#else
	CMP	R5, $8
	BLT	check0_7
	// Load sX[0:7] and compare.
	MOVD	(R8), R6
	MOVD	(R4), R7
	CMP	R6, R7
	ISEL	CR0EQ, R11, R0, R5
	// Load sX[len(sX)-8:len(sX)] and compare.
	MOVD	-8(R9), R6
	MOVD	-8(R10), R7
	CMP	R6, R7
	ISEL	CR0EQ, R5, R0, R3
	RET

check0_7:
	CMP	R5,$0
	MOVD	$1, R3
	BEQLR		// return if len == 0

	VN	V10, V11, V13
	VN	V12, V13, V14
	VCEQGS	V14, V15, V16
	BEQ	found
	MOVD	$1(R7), R7
	CMPBLE  R7, R2, index49to64loop
notfound:
	MOVD	$-1, (R5)
	RET

index65plus:
	// not implemented
	MOVD	$0, (R0)
	RET

foundV17: // index is in doubleword V17[0]
	VLGVG	$0, V17, R8
	ADD	R8, R7
found:
	SUB	R1, R7
	MOVD	R7, (R5)

TEXT ·IndexByte(SB),NOSPLIT|NOFRAME,$0-40
	MOVD	b_base+0(FP), R3// b_base => R3
	MOVD	b_len+8(FP), R4 // b_len => R4
	MOVBZ	c+24(FP), R5    // c => R5
	MOVD	$ret+32(FP), R2 // &ret => R9
	BR	indexbytebody<>(SB)

TEXT ·IndexByteString(SB),NOSPLIT|NOFRAME,$0-32
	MOVD	s_base+0(FP), R3// s_base => R3
	MOVD	s_len+8(FP), R4 // s_len => R4
	MOVBZ	c+16(FP), R5    // c => R5
	MOVD	$ret+24(FP), R2 // &ret => R9
	BR	indexbytebody<>(SB)

	BEQ	equal
notequal:
	MOVB	$0, 0(R7)
	RET
equal:
	MOVB	$1, 0(R7)
	RET
tiny:
	MOVD	$0, R2
	CMPBLT	R6, $16, lt16
	MOVD	0(R3), R8
	MOVD	0(R5), R9
	CMPBNE	R8, R9, notequal
	MOVD	8(R3), R8
	MOVD	8(R5), R9
	CMPBNE	R8, R9, notequal
	LA	16(R2), R2
	SUB	$16, R6
lt16:
	CMPBLT	R6, $8, lt8
	MOVD	0(R3)(R2*1), R8
	MOVD	0(R5)(R2*1), R9
	CMPBNE	R8, R9, notequal
	LA	8(R2), R2
	SUB	$8, R6
lt8:

#include "textflag.h"

// func kimd(function code, chain *[200]byte, src []byte)
TEXT ·kimd(SB), NOFRAME|NOSPLIT, $0-40
	MOVD function+0(FP), R0
	MOVD chain+8(FP), R1
	LMG  src+16(FP), R2, R3 // R2=base, R3=len

continue:
	WORD $0xB93E0002 // KIMD --, R2
	BVS  continue    // continue if interrupted
	MOVD $0, R0      // reset R0 for pre-go1.8 compilers
	RET

// func klmd(function code, chain *[200]byte, dst, src []byte)

TEXT ·archMax(SB),NOSPLIT,$0
	MOVD	x+0(FP), F0
	MOVD	y+8(FP), F1
	FCLASSD	F0, X5
	FCLASSD	F1, X6

	// +Inf special cases
	MOV	$PosInf, X7
	BEQ	X7, X5, isMaxX
	BEQ	X7, X6, isMaxY

	// NaN special cases
	MOV	$NaN, X7
	BEQ	X7, X5, isMaxX
	BEQ	X7, X6, isMaxY

	// normal case
	FMAXD	F0, F1, F0
	MOVD	F0, ret+16(FP)
	RET

isMaxX: // return x
	MOVD	F0, ret+16(FP)
	RET

// No arg size here; runtime pulls arg map out of the func value.
TEXT ·makeFuncStub(SB),(NOSPLIT|WRAPPER),$40
	NO_LOCAL_POINTERS

	MOVD CTXT, 0(SP)

	Get SP
	Get SP
	I64ExtendI32U
	I64Const $argframe+0(FP)
	I64Add
	I64Store $8

	MOVB $0, 32(SP)
	MOVD $32(SP), 16(SP)
	MOVD $0, 24(SP)

	CALL ·callReflect(SB)
	RET

// methodValueCall is the code half of the function returned by makeMethodValue.

#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