//     Maj(x, y, z) = (x AND y) XOR (x AND z) XOR (y AND z)
#define SHA512T2(a, b, c) \
	ROR	$28, a, X6; \
	ROR	$34, a, X7; \
	XOR	X7, X6; \
	ROR	$39, a, X8; \
	XOR	X8, X6; \
	AND	a, b, X7; \
	AND	a, c, X8; \
	XOR	X8, X7; \
	AND	b, c, X9; \
	XOR	X9, X7; \
	ADD	X7, X6

// Calculate T1 and T2, then e = d + T1 and a = T1 + T2.
// The values for e and a are stored in d and h, ready for rotation.

	AND	$7, X10, X7
	AND	$7, X12, X8
	BNE	X7, X8, check8_unaligned
	BEQZ	X7, compare32

	// Check one byte at a time until we reach 8 byte alignment.
	SUB	X7, X0, X7
	ADD	$8, X7, X7
	SUB	X7, X5, X5
align:
	SUB	$1, X7
	MOVBU	0(X10), X8
	MOVBU	0(X12), X9
	BNE	X8, X9, cmp
	ADD	$1, X10
	ADD	$1, X12
	BNEZ	X7, align

check32:
	// X6 contains $32

TEXT addMulVVWx(SB),NOFRAME|NOSPLIT,$0
	MOV	z+0(FP), X5
	MOV	x+8(FP), X7
	MOV	y+16(FP), X6
	MOV	$0, X29

	BEQZ	X30, done
loop:
	MOV	0*8(X5), X10	// z[0]
	MOV	1*8(X5), X13	// z[1]
	MOV	2*8(X5), X16	// z[2]
	MOV	3*8(X5), X19	// z[3]

	MOV	0*8(X7), X8	// x[0]
	MOV	1*8(X7), X11	// x[1]
	MOV	2*8(X7), X14	// x[2]
	MOV	3*8(X7), X17	// x[3]

	MULHU	X8, X6, X9	// z_hi[0] = x[0] * y

	VREPF $1, KEY0, X5
	VREPF $2, KEY0, X6
	VREPF $3, KEY0, X7
	VREPF $0, KEY1, X8
	VREPF $1, KEY1, X9
	VREPF $2, KEY1, X10
	VREPF $3, KEY1, X11
	VLR   CTR, X12
	VREPF $1, NONCE, X13
	VREPF $2, NONCE, X14
	VREPF $3, NONCE, X15

	MOVD $(NUM_ROUNDS/2), R1

loop:
	ROUND4(X0, X4, X12,  X8, X1, X5, X13,  X9, X2, X6, X14, X10, X3, X7, X15, X11)

	QR(X1, X5, X9, X13, X15)
	MOVOU (15*16)(BX), X15 // reload X15; temp now X4
	QR(X2, X6, X10, X14, X4)
	QR(X3, X7, X11, X15, X4)

	QR(X0, X5, X10, X15, X4)
	MOVOU X15, (15*16)(BX) // save X15
	QR(X1, X6, X11, X12, X4)
	MOVOU (4*16)(BX), X4  // reload X4; temp now X15
	QR(X2, X7, X8, X13, X15)
	QR(X3, X4, X9, X14, X15)

	DECL DX
	JNZ loop

	// Store interlaced blocks back to output buffer,

		x2 := byteorder.LeUint32(q[4*0x2:])
		x3 := byteorder.LeUint32(q[4*0x3:])
		x4 := byteorder.LeUint32(q[4*0x4:])
		x5 := byteorder.LeUint32(q[4*0x5:])
		x6 := byteorder.LeUint32(q[4*0x6:])
		x7 := byteorder.LeUint32(q[4*0x7:])
		x8 := byteorder.LeUint32(q[4*0x8:])
		x9 := byteorder.LeUint32(q[4*0x9:])
		xa := byteorder.LeUint32(q[4*0xa:])
		xb := byteorder.LeUint32(q[4*0xb:])
		xc := byteorder.LeUint32(q[4*0xc:])

// Use a local trampoline, to avoid taking the address of a dynamically exported
// function.
TEXT ·set_crosscall2(SB),NOSPLIT,$0-0
	MOV	_crosscall2_ptr(SB), X7
	MOV	$crosscall2_trampoline<>(SB), X8
	MOV	X8, (X7)
	RET

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

// Called by C code generated by cmd/cgo.
// func crosscall2(fn, a unsafe.Pointer, n int32, ctxt uintptr)

// Code generated by mkpreempt.go; DO NOT EDIT.

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

TEXT ·asyncPreempt(SB),NOSPLIT|NOFRAME,$0-0
	MOV X1, -464(X2)
	SUB $464, X2
	MOV X5, 8(X2)
	MOV X6, 16(X2)
	MOV X7, 24(X2)
	MOV X8, 32(X2)
	MOV X9, 40(X2)
	MOV X10, 48(X2)
	MOV X11, 56(X2)
	MOV X12, 64(X2)
	MOV X13, 72(X2)
	MOV X14, 80(X2)
	MOV X15, 88(X2)
	MOV X16, 96(X2)
	MOV X17, 104(X2)
	MOV X18, 112(X2)

hugeloop:
	CMPQ	BX, $64
	JB	bigloop
	MOVOU	(SI), X0
	MOVOU	(DI), X1
	MOVOU	16(SI), X2
	MOVOU	16(DI), X3
	MOVOU	32(SI), X4
	MOVOU	32(DI), X5
	MOVOU	48(SI), X6
	MOVOU	48(DI), X7
	PCMPEQB	X1, X0
	PCMPEQB	X3, X2
	PCMPEQB	X5, X4
	PCMPEQB	X7, X6
	PAND	X2, X0
	PAND	X6, X4
	PAND	X4, X0
	PMOVMSKB X0, DX
	ADDQ	$64, SI
	ADDQ	$64, DI
	SUBQ	$64, BX
	CMPL	DX, $0xffff
	JEQ	hugeloop
	XORQ	AX, AX	// return 0
	RET

type Numeric4 interface{t4 /* ERROR "not a type" */ }
func t4[T Numeric4](s[]T){if( /* ERROR "non-boolean" */ 0){*s /* ERROR "cannot indirect" */ [0]}}

// crash 7
type foo7 interface { bar() }
type x7[A any] struct{ foo7 }
func main7() { var _ foo7 = x7[int]{} }

// crash 8
type foo8[A any] interface { ~A /* ERROR "cannot be a type parameter" */ }
func bar8[A foo8[A]](a A) {}

// crash 9