MOVD	dec+24(FP), R11
	LDP	rotInvSRows<>(SB), (R0, R1)
	VMOV	R0, V3.D[0]
	VMOV	R1, V3.D[1]
	VEOR	V0.B16, V0.B16, V0.B16 // All zeroes
	MOVW	$1, R13
	TBZ	$1, R8, ks192
	TBNZ	$2, R8, ks256
	LDPW	(R9), (R4, R5)
	LDPW	8(R9), (R6, R7)
	STPW.P	(R4, R5), 8(R10)
	STPW.P	(R6, R7), 8(R10)
	MOVW	$0x1b, R14
ks128Loop:
		VMOV	R7, V2.S[0]
		WORD	$0x4E030042       // TBL V3.B16, [V2.B16], V2.B16

//go:build !purego

#include "textflag.h"

#define HASHUPDATECHOOSE \
	SHA1C	V16.S4, V1, V2 \
	SHA1H	V3, V1 \
	VMOV	V2.B16, V3.B16

#define HASHUPDATEPARITY \
	SHA1P	V16.S4, V1, V2 \
	SHA1H	V3, V1 \
	VMOV	V2.B16, V3.B16

#define HASHUPDATEMAJ \
	SHA1M	V16.S4, V1, V2 \
	SHA1H	V3, V1 \
	VMOV	V2.B16, V3.B16

// func sha1block(h []uint32, p []byte, k []uint32)
TEXT ·sha1block(SB),NOSPLIT,$0

	VLD1.P	16(R1), [V6.B16]                            // load 16bytes message
	VLD1.P	16(R1), [V7.B16]                            // load 16bytes message
	VMOV	V0.B16, V2.B16                              // backup: VO h(dcba)
	VMOV	V1.B16, V3.B16                              // backup: V1 h(hgfe)
	VMOV	V2.B16, V8.B16
	VREV32	V4.B16, V4.B16                              // prepare for using message in Byte format
	VREV32	V5.B16, V5.B16
	VREV32	V6.B16, V6.B16

	// long enough to prefetch
	PRFM	(R3), PLDL3KEEP
	// load digest
	VLD1	(R0), [V8.D2, V9.D2, V10.D2, V11.D2]
loop:
	// load digest in V0-V3 keeping original in V8-V11
	VMOV	V8.B16, V0.B16
	VMOV	V9.B16, V1.B16
	VMOV	V10.B16, V2.B16
	VMOV	V11.B16, V3.B16

	// load message data in V12-V19
	VLD1.P	64(R1), [V12.D2, V13.D2, V14.D2, V15.D2]
	VLD1.P	64(R1), [V16.D2, V17.D2, V18.D2, V19.D2]

	VCMEQ	V1.D2, V5.D2, V9.D2
	VCMEQ	V2.D2, V6.D2, V10.D2
	VCMEQ	V3.D2, V7.D2, V11.D2
	VAND	V8.B16, V9.B16, V8.B16
	VAND	V8.B16, V10.B16, V8.B16
	VAND	V8.B16, V11.B16, V8.B16
	CMP	R0, R6
	VMOV	V8.D[0], R4
	VMOV	V8.D[1], R5
	CBZ	R4, not_equal
	CBZ	R5, not_equal
	BNE	chunk64_loop
	AND	$0x3f, R2, R2
	CBZ	R2, equal
chunk16:
	// work with 16-byte chunks
	BIC	$0xf, R2, R3
	CBZ	R3, tail

// err | A2          | part of A0
TEXT ·Syscall6<ABIInternal>(SB),NOSPLIT,$0-80
	MOV	A0, A7
	MOV	A1, A0
	MOV	A2, A1
	MOV	A3, A2
	MOV	A4, A3
	MOV	A5, A4
	MOV	A6, A5
	ECALL
	MOV	$-4096, T0
	BLTU	T0, A0, err
	// r1 already in A0
	// r2 already in A1
	MOV	ZERO, A2 // errno
	RET
err:
	SUB	A0, ZERO, A2 // errno
	MOV	$-1, A0	     // r1
	MOV	ZERO, A1     // r2

	MOV	(0*8)(X20), X5
	MOV	(1*8)(X20), X6
	MOV	(2*8)(X20), X7
	MOV	(3*8)(X20), X8
	ADD	X5, X10		// H0 = a + H0
	ADD	X6, X11		// H1 = b + H1
	ADD	X7, X12		// H2 = c + H2
	ADD	X8, X13		// H3 = d + H3
	MOV	X10, (0*8)(X20)
	MOV	X11, (1*8)(X20)
	MOV	X12, (2*8)(X20)
	MOV	X13, (3*8)(X20)
	MOV	(4*8)(X20), X5
	MOV	(5*8)(X20), X6
	MOV	(6*8)(X20), X7
	MOV	(7*8)(X20), X8

	// 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 {
	// amd64:"MOV(B|L)\t[$]1",-"FCMP",-"MOV(B|L)\t[$]0"
	// s390x:"MOV(B|BZ|D)\t[$]1,",-"FCMPU",-"MOV(B|BZ|D)\t[$]0,"
	return float32(0.5) <= float32(int64(1)) && float32(1.5) >= float32(int32(-1<<31))