default:
				return intelReg[a]
			}
		}
	}
	return strings.ToLower(arg.String())
}

var intelOp = map[Op]string{
	JAE:       "jnb",
	JA:        "jnbe",
	JGE:       "jnl",
	JNE:       "jnz",
	JG:        "jnle",
	JE:        "jz",
	SETAE:     "setnb",
	SETA:      "setnbe",
	SETGE:     "setnl",
	SETNE:     "setnz",
	SETG:      "setnle",
	SETE:      "setz",

	instructions["JNA"] = x86.AJLS  /* alternate */
	instructions["JNAE"] = x86.AJCS /* alternate */
	instructions["JNB"] = x86.AJCC  /* alternate */
	instructions["JNBE"] = x86.AJHI /* alternate */
	instructions["JNC"] = x86.AJCC  /* alternate */
	instructions["JNE"] = x86.AJNE  /* not equal (ZF = 0) */
	instructions["JNG"] = x86.AJLE  /* alternate */
	instructions["JNGE"] = x86.AJLT /* alternate */

	TESTQ	BX, BX
	JEQ	move_0
	CMPQ	BX, $2
	JBE	move_1or2
	CMPQ	BX, $4
	JB	move_3
	JBE	move_4
	CMPQ	BX, $8
	JB	move_5through7
	JE	move_8
	CMPQ	BX, $16
	JBE	move_9through16
	CMPQ	BX, $32
	JBE	move_17through32
	CMPQ	BX, $64
	JBE	move_33through64
	CMPQ	BX, $128
	JBE	move_65through128
	CMPQ	BX, $256
	JBE	move_129through256

	TESTB	$1, runtime·useAVXmemmove(SB)
	JNZ	avxUnaligned

has_cpuid:
	MOVL	$0, AX
	CPUID
	MOVL	AX, SI
	CMPL	AX, $0
	JE	nocpuinfo

	CMPL	BX, $0x756E6547  // "Genu"
	JNE	notintel
	CMPL	DX, $0x49656E69  // "ineI"
	JNE	notintel
	CMPL	CX, $0x6C65746E  // "ntel"
	JNE	notintel
	MOVB	$1, runtime·isIntel(SB)
notintel:

	// Load EAX=1 cpuid flags
	MOVL	$1, AX
	CPUID
	MOVL	CX, DI // Move to global variable clobbers CX when generating PIC

	MOVL	(0*4)(BX), AX		// fn
	CALL	AX

	MOVL	AX, (4*4)(BX)		// r1
	MOVL	DX, (5*4)(BX)		// r2

	// Standard libc functions return -1 on error and set errno.
	CMPL	AX, $-1
	JNE	ok
	CMPL	DX, $-1
	JNE	ok

	// Get error code from libc.
	CALL	libc_errno(SB)
	MOVL	(AX), AX
	MOVW	AX, (6*4)(BX)		// err

ok:
	MOVL	$0, AX			// no error (it's ignored anyway)
	MOVL	BP, SP
	POPL	BP

	MOVQ	DI, BX
	CALL	libc_mach_absolute_time(SB)
	MOVQ	AX, 0(BX)
	MOVL	timebase<>+machTimebaseInfo_numer(SB), SI
	MOVL	timebase<>+machTimebaseInfo_denom(SB), DI // atomic read
	TESTL	DI, DI
	JNE	initialized

	SUBQ	$(machTimebaseInfo__size+15)/16*16, SP
	MOVQ	SP, DI
	CALL	libc_mach_timebase_info(SB)
	MOVL	machTimebaseInfo_numer(SP), SI
	MOVL	machTimebaseInfo_denom(SP), DI

	MOVQ	8(BX), SI		// arg 2 len
	MOVL	16(BX), DX		// arg 3 prot
	MOVL	20(BX), CX		// arg 4 flags
	MOVL	24(BX), R8		// arg 5 fid
	MOVL	28(BX), R9		// arg 6 offset
	CALL	libc_mmap(SB)
	XORL	DX, DX
	CMPQ	AX, $-1
	JNE	ok
	CALL	libc_errno(SB)
	MOVLQSX	(AX), DX		// errno
	XORQ	AX, AX
ok:
	MOVQ	AX, 32(BX)
	MOVQ	DX, 40(BX)
	RET

TEXT runtime·munmap_trampoline(SB),NOSPLIT,$0
	MOVQ	8(DI), SI		// arg 2 len

		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

		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