Russ Cox <******@****.***> 1638378945 -0500

	ANDL    $0x7FF, BX
	SUBL    $0x3FE, BX
	XORPS   X1, X1 // break dependency for CVTSL2SD
	CVTSL2SD BX, X1 // x1= k, x2= f1
	// if f1 < math.Sqrt2/2 { k -= 1; f1 *= 2 }
	MOVSD   $HSqrt2, X0 // x0= 0.7071, x1= k, x2= f1
	CMPSD   X2, X0, 5 // cmpnlt; x0= 0 or ^0, x1= k, x2 = f1
	MOVSD   $1.0, X3 // x0= 0 or ^0, x1= k, x2 = f1, x3= 1
	ANDPD   X0, X3 // x0= 0 or ^0, x1= k, x2 = f1, x3= 0 or 1
	SUBSD   X3, X1 // x0= 0 or ^0, x1= k, x2 = f1, x3= 0 or 1

		// Test AVX512 suffixes.
		{"VADDPD.A X0, X1, X2", `unknown suffix "A"`},
		{"VADDPD.A.A X0, X1, X2", `unknown suffix "A"; duplicate suffix "A"`},
		{"VADDPD.A.A.A X0, X1, X2", `unknown suffix "A"; duplicate suffix "A"`},
		{"VADDPD.A.B X0, X1, X2", `unknown suffix "A"; unknown suffix "B"`},
		{"VADDPD.Z.A X0, X1, X2", `Z suffix should be the last; unknown suffix "A"`},
		{"VADDPD.Z.Z X0, X1, X2", `Z suffix should be the last; duplicate suffix "Z"`},

	LEAQ	-16(SI)(BX*1), AX	// AX = address of last 16 bytes
	JMP	sseloopentry

	PCALIGN $16
sseloop:
	// Move the next 16-byte chunk of the data into X1.
	MOVOU	(DI), X1
	// Compare bytes in X0 to X1.
	PCMPEQB	X0, X1
	// Take the top bit of each byte in X1 and put the result in DX.
	PMOVMSKB X1, DX
	// Find first set bit, if any.
	BSFL	DX, DX
	JNZ	ssesuccess
	// Advance to next block.
	ADDQ	$16, DI
sseloopentry:
	CMPQ	DI, AX

	AESENC X1, X0
	MOVUPS 48(AX), X1
	AESENC X1, X0
	MOVUPS 64(AX), X1
	AESENC X1, X0
	MOVUPS 80(AX), X1
	AESENC X1, X0
	MOVUPS 96(AX), X1
	AESENC X1, X0
	MOVUPS 112(AX), X1
	AESENC X1, X0
	MOVUPS 128(AX), X1
	AESENC X1, X0
	MOVUPS 144(AX), X1
	AESENCLAST X1, X0
	MOVUPS X0, 0(DX)
	RET

// func decryptBlockAsm(nr int, xk *uint32, dst, src *byte)
TEXT ·decryptBlockAsm(SB),NOSPLIT,$0
	MOVQ nr+0(FP), CX

	LEAQ	-16(SI)(BX*1), AX	// AX = address of last 16 bytes
	JMP	sseloopentry

	PCALIGN $16
sseloop:
	// Move the next 16-byte chunk of the data into X1.
	MOVOU	(DI), X1
	// Compare bytes in X0 to X1.
	PCMPEQB	X0, X1
	// Take the top bit of each byte in X1 and put the result in DX.
	PMOVMSKB X1, DX
	// Count number of matching bytes
	POPCNTL DX, DX
	// Accumulate into R12
	ADDQ DX, R12
	// Advance to next block.
	ADDQ	$16, DI

		AB
		C
	}
	ABC3 interface {
		A
		BC
	}
)

var (
	x1 ABC1
	x2 ABC2
	x3 ABC3
)

func _() {
	// all types have the same method set
	x1 = x2
	x2 = x1

	x1 = x3
	x3 = x1

	x2 = x3
	x3 = x2

	// all methods return the same type again
	x1 = x1.a()
	x1 = x1.b()
	x1 = x1.c()

	x2 = x2.a()
	x2 = x2.b()
	x2 = x2.c()

	x3 = x3.a()
	x3 = x3.b()

	//   = x1*B + x0 - d*((x1*m+x1*B+x0)/B-alpha)                                   0 <= alpha < 1
	//   = x1*B + x0 - x1*d/B*m                         - x1*d - x0*d/B + d*alpha
	//   = x1*B + x0 - x1*d/B*⎣(B^2-1)/d-B⎦             - x1*d - x0*d/B + d*alpha
	//   = x1*B + x0 - x1*d/B*⎣(B^2-1)/d-B⎦             - x1*d - x0*d/B + d*alpha
	//   = x1*B + x0 - x1*d/B*((B^2-1)/d-B-beta)        - x1*d - x0*d/B + d*alpha   0 <= beta < 1

	// Usage of rounding with invalid operands.
	VADDPD.RU_SAE X3, X2, K1, X1     // ERROR "unsupported rounding"
	VADDPD.RD_SAE X3, X2, K1, X1     // ERROR "unsupported rounding"
	VADDPD.RZ_SAE X3, X2, K1, X1     // ERROR "unsupported rounding"
	VADDPD.RN_SAE X3, X2, K1, X1     // ERROR "unsupported rounding"
	VADDPD.RU_SAE Y3, Y2, K1, Y1     // ERROR "unsupported rounding"

	VL   80(P1ptr), X1        // Z1H
	VPDI $0x4, X1, X1, X1
	VL   64(P1ptr), X0        // Z1L
	VPDI $0x4, X0, X0, X0
	VLR  X0, Y0
	VLR  X1, Y1
	CALL p256SqrInternal<>(SB)

	// SUB(X<X1-T)            // T2 = X1-T1
	VL   16(P1ptr), X1H
	VPDI $0x4, X1H, X1H, X1H
	VL   0(P1ptr), X1L
	VPDI $0x4, X1L, X1L, X1L
	p256SubInternal(X1,X0,X1H,X1L,T1,T0)