p.To.Type = obj.TYPE_REG
		p.To.Reg = regEntryTmp0
		if ctxt.Arch.Family == sys.I386 {
			p.As = ATESTL
		}

		// JNE checkargp (checkargp to be resolved later)
		jne := obj.Appendp(p, newprog)
		jne.As = AJNE
		jne.To.Type = obj.TYPE_BRANCH

		// end:
		//  NOP
		end := obj.Appendp(jne, newprog)
		end.As = obj.ANOP

		// Fast forward to end of function.
		var last *obj.Prog

 */
	CMPL	SI, DI
	JLS	back

/*
 * forward copy loop
 */
forward:
	// If REP MOVSB isn't fast, don't use it
	CMPB	internal∕cpu·X86+const_offsetX86HasERMS(SB), $1 // enhanced REP MOVSB/STOSB
	JNE	fwdBy4

	// Check alignment
	MOVL	SI, AX
	ORL	DI, AX
	TESTL	$3, AX
	JEQ	fwdBy4

	// Do 1 byte at a time
	MOVL	BX, CX
	REP;	MOVSB
	RET

fwdBy4:
	// Do 4 bytes at a time
	MOVL	BX, CX

	// Directly return DX, we don't need to accumulate
	// since we have <16 bytes.
	POPCNTL DX, DX
	MOVQ	DX, (R8)
	RET

avx2:
#ifndef hasAVX2
	CMPB   internal∕cpu·X86+const_offsetX86HasAVX2(SB), $1
	JNE sse
#endif
	MOVD AX, X0
	LEAQ -64(SI)(BX*1), R11
	LEAQ (SI)(BX*1), R13
	VPBROADCASTB  X0, Y1
	PCALIGN $32
avx2_loop:
	VMOVDQU (DI), Y2
	VMOVDQU 32(DI), Y4
	VPCMPEQB Y1, Y2, Y3

	JB loop33to63
fail_avx2:
	VZEROUPPER
fail:
	MOVQ $-1, (R11)
	RET
success_avx2:
	VZEROUPPER
	JMP success
sse42:
#ifndef hasSSE42
	CMPB internal∕cpu·X86+const_offsetX86HasSSE42(SB), $1
	JNE no_sse42
#endif
	CMPQ AX, $12
	// PCMPESTRI is slower than normal compare,
	// so using it makes sense only if we advance 4+ bytes per compare
	// This value was determined experimentally and is the ~same

TEXT runtime·settls(SB),NOSPLIT,$0
	RET

// void sigtramp(void *ureg, int8 *note)
TEXT runtime·sigtramp(SB),NOSPLIT|NOFRAME,$0
	get_tls(AX)

	// check that g exists
	MOVQ	g(AX), BX
	CMPQ	BX, $0
	JNE	3(PC)
	CALL	runtime·badsignal2(SB) // will exit
	RET

	// save args
	MOVQ	ureg+0(FP), CX
	MOVQ	note+8(FP), DX

	// change stack
	MOVQ	g_m(BX), BX
	MOVQ	m_gsignal(BX), R10

	MOVL	DX, 0(CX)(FS)
	RET

TEXT runtime·nanotime1(SB),NOSPLIT,$0-8
loop:
	MOVL	(_INTERRUPT_TIME+time_hi1), AX
	MOVL	(_INTERRUPT_TIME+time_lo), CX
	MOVL	(_INTERRUPT_TIME+time_hi2), DI
	CMPL	AX, DI
	JNE	loop

	// wintime = DI:CX, multiply by 100
	MOVL	$100, AX
	MULL	CX
	IMULL	$100, DI
	ADDL	DI, DX
	// wintime*100 = DX:AX
	MOVL	AX, ret_lo+0(FP)
	MOVL	DX, ret_hi+4(FP)
	RET

		PSHUFD $78, T0, T0
		PXOR T0, B2
		PXOR T1, B2

		MOVOU B2, (16*12)(dst)
		PSHUFD $78, B2, B3
		PXOR B2, B3
		MOVOU B3, (16*13)(dst)

		DECQ AX
		LEAQ (-16*2)(dst), dst
	JNE initLoop

	RET
#undef NR
#undef KS
#undef dst

// func gcmAesData(productTable *[256]byte, data []byte, T *[16]byte)
TEXT ·gcmAesData(SB),NOSPLIT,$0
#define pTbl DI
#define aut SI

	MOVL	m_vdsoSP(SI), DX
	MOVL	CX, 0(SP)
	MOVL	DX, 4(SP)

	LEAL	sec+0(FP), DX
	MOVL	-4(DX), CX
	MOVL	CX, m_vdsoPC(SI)
	MOVL	DX, m_vdsoSP(SI)

	CMPL	AX, m_curg(SI)	// Only switch if on curg.
	JNE	noswitch

	MOVL	m_g0(SI), DX
	MOVL	(g_sched+gobuf_sp)(DX), SP	// Set SP to g0 stack

noswitch:
	SUBL	$16, SP		// Space for results
	ANDL	$~15, SP	// Align for C code

cat >umod-amd64.s <<'EOF'
# tu_int __umodti3(tu_int x, tu_int y)
# x is rsi:rdi, y is rcx:rdx, return result is rdx:rax.
.globl __umodti3
__umodti3:
	# specialized to u128 % u64, so verify that
	test %rcx,%rcx
	jne 1f

	# save divisor
	movq %rdx, %r8

	# reduce top 64 bits mod divisor
	movq %rsi, %rax
	xorl %edx, %edx
	divq %r8

	# reduce full 128-bit mod divisor

	VMOVDQU Y6, Y4
	VMOVDQU Y7, Y5

	SUBQ $1, frame_SRND(SP)
	JNE  loop2

	addm(8*0(SI),AX)
	addm(8*1(SI),BX)
	addm(8*2(SI),CX)
	addm(8*3(SI),R8)
	addm(8*4(SI),DX)
	addm(8*5(SI),R9)
	addm(8*6(SI),R10)
	addm(8*7(SI),R11)

	MOVQ frame_INP(SP), DI
	ADDQ $128, DI
	CMPQ DI, frame_INPEND(SP)
	JNE  loop0

done_hash:
	VZEROUPPER