VP4DPWSSD 15(DX)(BX*8), [Z2-Z5], K4, Z17           // 62e26f4c528cda0f000000
	VP4DPWSSD 7(SI)(DI*1), [Z12-Z15], K4, Z17          // 62e21f4c528c3e07000000
	VP4DPWSSD 15(DX)(BX*8), [Z12-Z15], K4, Z17         // 62e21f4c528cda0f000000
	VP4DPWSSD 7(SI)(DI*1), [Z22-Z25], K4, Z17          // 62e24f44528c3e07000000
	VP4DPWSSD 15(DX)(BX*8), [Z22-Z25], K4, Z17         // 62e24f44528cda0f000000

	// so this function is reentrant.
	MOVQ	m_vdsoPC(BX), CX
	MOVQ	m_vdsoSP(BX), DX
	MOVQ	CX, 0(SP)
	MOVQ	DX, 8(SP)

	LEAQ	sec+0(FP), DX
	MOVQ	-8(DX), CX	// Sets CX to function return address.
	MOVQ	CX, m_vdsoPC(BX)
	MOVQ	DX, m_vdsoSP(BX)

	CMPQ	R14, m_curg(BX)	// Only switch if on curg.
	JNE	noswitch

	MOVQ	m_g0(BX), DX
	MOVQ	(g_sched+gobuf_sp)(DX), SP	// Set SP to g0 stack

noswitch:

	// frame is specially handled in the runtime. See the comment above LOCAL_RETVALID.
	LEAQ	LOCAL_REGARGS(SP), R12
	CALL	runtime·spillArgs(SB)
	MOVQ	DX, 24(SP) // outside of moveMakeFuncArgPtrs's arg area
	MOVQ	DX, 0(SP)
	MOVQ	R12, 8(SP)
	CALL	·moveMakeFuncArgPtrs(SB)
	MOVQ	24(SP), DX
	MOVQ	DX, 0(SP)
	LEAQ	argframe+0(FP), CX
	MOVQ	CX, 8(SP)
	MOVB	$0, LOCAL_RETVALID(SP)
	LEAQ	LOCAL_RETVALID(SP), AX
	MOVQ	AX, 16(SP)

	// CX = a_cap  (unused)
	// DI = b_base (want in DI)
	// SI = b_len  (want in DX)
	// R8 = b_cap  (unused)
	MOVQ	SI, DX
	MOVQ	AX, SI
	JMP	cmpbody<>(SB)

TEXT runtime·cmpstring<ABIInternal>(SB),NOSPLIT,$0-40
	// AX = a_base (want in SI)
	// BX = a_len  (want in BX)
	// CX = b_base (want in DI)
	// DI = b_len  (want in DX)
	MOVQ	AX, SI
	MOVQ	DI, DX
	MOVQ	CX, DI
	JMP	cmpbody<>(SB)

// input:
//   SI = a

	CALL _expand_key_128<>(SB)
Lexp_dec:
	// dec
	SUBQ $16, BX
	MOVUPS (BX), X1
	MOVUPS X1, (DX)
	DECQ CX
Lexp_dec_loop:
	MOVUPS -16(BX), X1
	AESIMC X1, X0
	MOVUPS X0, 16(DX)
	SUBQ $16, BX
	ADDQ $16, DX
	DECQ CX
	JNZ Lexp_dec_loop
	MOVUPS -16(BX), X0
	MOVUPS X0, 16(DX)
	RET

TEXT _expand_key_128<>(SB),NOSPLIT,$0
	PSHUFD $0xff, X1, X1
	SHUFPS $0x10, X0, X4
	PXOR X4, X0

	get_tls(BX)
	MOVQ	DX, g(BX)
	MOVQ	DI, g_m(DX)

	// Layout new m scheduler stack on os stack.
	MOVQ	SP, AX
	MOVQ	AX, (g_stack+stack_hi)(DX)
	SUBQ	$(0x100000), AX		// stack size
	MOVQ	AX, (g_stack+stack_lo)(DX)
	ADDQ	$const_stackGuard, AX
	MOVQ	AX, g_stackguard0(DX)
	MOVQ	AX, g_stackguard1(DX)

	// Someday the convention will be D is always cleared.
	CLD

	VPCMPEQB	Y2, Y3, Y5
	VPAND	Y4, Y5, Y6
	VPMOVMSKB Y6, DX
	ADDQ	$64, SI
	ADDQ	$64, DI
	SUBQ	$64, BX
	CMPL	DX, $0xffffffff
	JEQ	hugeloop_avx2
	VZEROUPPER
	XORQ	AX, AX	// return 0
	RET

bigloop_avx2:
	VZEROUPPER

	// 8 bytes at a time using 64-bit register
	PCALIGN $16
bigloop:
	CMPQ	BX, $8
	JBE	leftover
	MOVQ	(SI), CX
	MOVQ	(DI), DX
	ADDQ	$8, SI
	ADDQ	$8, DI
	SUBQ	$8, BX

	CMPQ    AX, R8
	JEQ     isPosInf
	MOVQ    y+8(FP), R9
	CMPQ    AX, R9
	JEQ     isPosInf
	// NaN special cases
	MOVQ    $~(1<<63), DX // bit mask
	MOVQ    $PosInf, AX
	MOVQ    R8, BX
	ANDQ    DX, BX // x = |x|
	CMPQ    AX, BX
	JLT     isMaxNaN
	MOVQ    R9, CX
	ANDQ    DX, CX // y = |y|
	CMPQ    AX, CX
	JLT     isMaxNaN
	// ±0 special cases
	ORQ     CX, BX
	JEQ     isMaxZero

	MOVQ    R8, X0

	MOVL	lo64+0(FP), CX
	MOVL	a_lo+4(FP), AX
	MULL	b+12(FP)
	MOVL	AX, 0(CX)
	MOVL	DX, BX
	MOVL	a_hi+8(FP), AX
	MULL	b+12(FP)
	ADDL	AX, BX
	ADCL	$0, DX
	MOVL	BX, 4(CX)
	MOVL	DX, AX
	MOVL	AX, hi32+16(FP)
	RET

TEXT runtime·_div64by32(SB), NOSPLIT, $0
	MOVL	r+12(FP), CX
	MOVL	a_lo+0(FP), AX
	MOVL	a_hi+4(FP), DX
	DIVL	b+8(FP)
	MOVL	DX, 0(CX)
	MOVL	AX, q+16(FP)

	VPCLMULQDQ $97, 15(DX)(BX*8), Z26, Z24             // 62632d404484da0f00000061 or 6263ad404484da0f00000061
	VPCLMULQDQ $97, Z9, Z0, Z12                        // 62537d4844e161 or 6253fd4844e161
	VPCLMULQDQ $97, Z3, Z0, Z12                        // 62737d4844e361 or 6273fd4844e361
	VPCLMULQDQ $97, 7(SI)(DI*1), Z0, Z12               // 62737d4844a43e0700000061 or 6273fd4844a43e0700000061