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

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

		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

	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

	CMOVQCS y_ptr, acc2
	CMOVQCS t1, acc3

	MOVQ acc0, (8*0)(res_ptr)
	MOVQ acc1, (8*1)(res_ptr)
	MOVQ acc2, (8*2)(res_ptr)
	MOVQ acc3, (8*3)(res_ptr)
	MOVQ res_ptr, x_ptr
	DECQ BX
	JNE  sqrLoop

	RET
/* ---------------------------------------*/
// func p256Mul(res, in1, in2 *p256Element)
TEXT ·p256Mul(SB),NOSPLIT,$0
	MOVQ res+0(FP), res_ptr
	MOVQ in1+8(FP), x_ptr
	MOVQ in2+16(FP), y_ptr

	"INW",
	"IRETL",
	"IRETQ",
	"IRETW",
	"JCC",
	"JCS",
	"JCXZL",
	"JCXZQ",
	"JCXZW",
	"JEQ",
	"JGE",
	"JGT",
	"JHI",
	"JLE",
	"JLS",
	"JLT",
	"JMI",
	"JNE",
	"JOC",
	"JOS",
	"JPC",
	"JPL",
	"JPS",
	"KADDB",
	"KADDD",
	"KADDQ",
	"KADDW",
	"KANDB",
	"KANDD",
	"KANDNB",
	"KANDND",
	"KANDNQ",
	"KANDNW",
	"KANDQ",