JMP	·Store64(SB)

// void	·Or8(byte volatile*, byte);
TEXT ·Or8(SB), NOSPLIT, $0-9
	MOVQ	ptr+0(FP), AX
	MOVB	val+8(FP), BX
	LOCK
	ORB	BX, (AX)
	RET

// void	·And8(byte volatile*, byte);
TEXT ·And8(SB), NOSPLIT, $0-9
	MOVQ	ptr+0(FP), AX
	MOVB	val+8(FP), BX
	LOCK
	ANDB	BX, (AX)
	RET

// func Or(addr *uint32, v uint32)
TEXT ·Or(SB), NOSPLIT, $0-12
	MOVQ	ptr+0(FP), AX

loop:
	MOVD	0(R2)(R5*1), R7
	MOVD	0(R3)(R5*1), R8
	XOR	R7, R8
	MOVD	R8, 0(R1)(R5*1)
	LAY	8(R5), R5
	SUB	$8, R4
	CMPBGE	R4, $8, loop
tail:
	CMPBEQ	R4, $0, done
	MOVB	0(R2)(R5*1), R7
	MOVB	0(R3)(R5*1), R8
	XOR	R7, R8
	MOVB	R8, 0(R1)(R5*1)
	LAY	1(R5), R5
	SUB	$1, R4
	BR	tail
done:
	RET

// func cryptBlocksGCM(fn code, key, dst, src, buf []byte, cnt *[16]byte)
TEXT ·cryptBlocksGCM(SB),NOSPLIT,$0-112

	CMPQ R15, $16
	JAE  loop

bytes_between_0_and_15:
	TESTQ R15, R15
	JZ    done
	MOVQ  $1, BX
	XORQ  CX, CX
	XORQ  R13, R13
	ADDQ  R15, SI

flush_buffer:
	SHLQ $8, BX, CX
	SHLQ $8, BX
	MOVB -1(SI), R13
	XORQ R13, BX
	DECQ SI
	DECQ R15
	JNZ  flush_buffer

	ADDQ BX, R8
	ADCQ CX, R9
	ADCQ $0, R10
	MOVQ $16, R15
	JMP  multiply

done:
	MOVQ R8, 0(DI)
	MOVQ R9, 8(DI)

	AND	$7, R4, R8
	BEQ	R8, words
	MOVB	(R5), R7
	ADDV	$1, R5
	MOVB	R7, (R4)
	ADDV	$1, R4
	JMP	-6(PC)

words:
	// do 8 bytes at a time if there is room
	ADDV	$-7, R9, R6 // R6 is end pointer-7

	PCALIGN	$16
	SGTU	R6, R4, R8
	BEQ	R8, out
	MOVV	(R5), R7
	ADDV	$8, R5
	MOVV	R7, (R4)
	ADDV	$8, R4
	JMP	-6(PC)

out:
	BEQ	R4, R9, done
	MOVB	(R5), R7
	ADDV	$1, R5
	MOVB	R7, (R4)

	_LHBEX	(R0)(R6),R11
	_LHBEX	(R9)(R5),R12
	_LHBEX	(R9)(R6),R14
	RLDIMI	$32,R10,$0,R12
	RLDIMI	$32,R11,$0,R14
	CMPU	R12,R14
	BR	cmp0

	PCALIGN $16
cmp1:
	CMP	R9,$0
	BEQ	cmp0
	MOVBZ	(R5),R10
	MOVBZ	(R6),R11
	CMPU	R10,R11
cmp0:
	SETB_CR0(R6)
	ISEL	CR0EQ,R3,R6,R3
	RET

	return math.Float32frombits(u32+1) + 1
}

// Test that comparisons with constants converted to float
// are evaluated at compile-time

func constantCheck64() bool {
	// amd64:"(MOVB\t[$]0)|(XORL\t[A-Z][A-Z0-9]+, [A-Z][A-Z0-9]+)",-"FCMP",-"MOVB\t[$]1"
	// s390x:"MOV(B|BZ|D)\t[$]0,",-"FCMPU",-"MOV(B|BZ|D)\t[$]1,"
	return 0.5 == float64(uint32(1)) || 1.5 > float64(uint64(1<<63))
}

func constantCheck32() bool {

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

//go:build !purego

#include "textflag.h"

// func block(dig *digest, p []byte)
TEXT ·block(SB), NOSPLIT|NOFRAME, $0-32
	MOVBZ  ·useAsm(SB), R4
	LMG    dig+0(FP), R1, R3            // R2 = &p[0], R3 = len(p)
	MOVBZ  $1, R0                       // SHA-1 function code
	CMPBEQ R4, $0, generic

loop:
	KIMD R0, R2      // compute intermediate message digest (KIMD)

#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
//   R1: byte to find

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

//go:build !purego

#include "textflag.h"

// func block(dig *digest, p []byte)
TEXT ·block(SB), NOSPLIT|NOFRAME, $0-32
	MOVBZ  ·useAsm(SB), R4
	LMG    dig+0(FP), R1, R3            // R2 = &p[0], R3 = len(p)
	MOVBZ  $2, R0                       // SHA-256 function code
	CMPBEQ R4, $0, generic

loop:
	KIMD R0, R2      // compute intermediate message digest (KIMD)

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

//go:build !purego

#include "textflag.h"

// func block(dig *digest, p []byte)
TEXT ·block(SB), NOSPLIT|NOFRAME, $0-32
	MOVBZ  ·useAsm(SB), R4
	LMG    dig+0(FP), R1, R3            // R2 = &p[0], R3 = len(p)
	MOVBZ  $3, R0                       // SHA-512 function code
	CMPBEQ R4, $0, generic

loop:
	KIMD R0, R2      // compute intermediate message digest (KIMD)