MOVQ	AX, DI	// DI = ptr
	XORQ	AX, AX

	// MOVOU seems always faster than REP STOSQ when Enhanced REP STOSQ is not available.
tail:
	// BSR+branch table make almost all memmove/memclr benchmarks worse. Not worth doing.
	TESTQ	BX, BX
	JEQ	_0
	CMPQ	BX, $2
	JBE	_1or2
	CMPQ	BX, $4
	JBE	_3or4
	CMPQ	BX, $8
	JB	_5through7
	JE	_8
	CMPQ	BX, $16
	JBE	_9through16
	CMPQ	BX, $32
	JBE	_17through32
	CMPQ	BX, $64

	// 386:"TESTL\t[$]63",-"DIVL",-"SHRL"
	// amd64:"TESTQ\t[$]63",-"DIVQ",-"SHRQ"
	// arm:"AND\t[$]63",-".*udiv",-"SRA"
	// arm64:"TST\t[$]63",-"UDIV",-"ASR",-"AND"
	// ppc64x:"ANDCC",-"RLDICL",-"SRAD",-"CMP"
	a := n1%64 == 0 // signed divisible

	// 386:"TESTL\t[$]63",-"DIVL",-"SHRL"
	// amd64:"TESTQ\t[$]63",-"DIVQ",-"SHRQ"
	// arm:"AND\t[$]63",-".*udiv",-"SRA"

	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
	MOVQ	0(DI), DI		// arg 1 addr
	CALL	libc_munmap(SB)
	TESTQ	AX, AX
	JEQ	2(PC)
	MOVL	$0xf1, 0xf1  // crash
	RET

TEXT runtime·madvise_trampoline(SB), NOSPLIT, $0
	MOVQ	8(DI), SI	// arg 2 len
	MOVL	16(DI), DX	// arg 3 advice
	MOVQ	0(DI), DI	// arg 1 addr

	MOVQ	SI, _rt0_amd64_lib_argv<>(SB)

	// Synchronous initialization.
	CALL	runtime·libpreinit(SB)

	// Create a new thread to finish Go runtime initialization.
	MOVQ	_cgo_sys_thread_create(SB), AX
	TESTQ	AX, AX
	JZ	nocgo

	// We're calling back to C.
	// Align stack per ELF ABI requirements.
	MOVQ	SP, BX  // Callee-save in C ABI
	ANDQ	$~15, SP
	MOVQ	$_rt0_amd64_lib_go(SB), DI
	MOVQ	$0, SI
	CALL	AX

	"SUBPD",
	"SUBPS",
	"SUBQ",
	"SUBSD",
	"SUBSS",
	"SUBW",
	"SWAPGS",
	"SYSCALL",
	"SYSENTER",
	"SYSENTER64",
	"SYSEXIT",
	"SYSEXIT64",
	"SYSRET",
	"TESTB",
	"TESTL",
	"TESTQ",
	"TESTW",
	"TPAUSE",
	"TZCNTL",
	"TZCNTQ",
	"TZCNTW",
	"UCOMISD",
	"UCOMISS",
	"UD1",
	"UD2",
	"UMWAIT",
	"UNPCKHPD",
	"UNPCKHPS",
	"UNPCKLPD",
	"UNPCKLPS",
	"UMONITOR",

			v0.Aux = symToAux(sym)
			v0.AddArg2(ptr, mem)
			return true
		}
		break
	}
	// match: (TESTQ a:(ANDQload [off] {sym} x ptr mem) a)
	// cond: a.Uses == 2 && a.Block == v.Block && clobber(a)
	// result: (TESTQ (MOVQload <a.Type> [off] {sym} ptr mem) x)
	for {
		for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
			a := v_0
			if a.Op != OpAMD64ANDQload {

	LEAQ  256(inp), inp
	LEAQ  256(oup), oup
	SUBQ  $256, inl
	JMP   openSSEMainLoop

openSSEMainLoopDone:
	// Handle the various tail sizes efficiently
	TESTQ inl, inl
	JE    openSSEFinalize
	CMPQ  inl, $64
	JBE   openSSETail64
	CMPQ  inl, $128
	JBE   openSSETail128
	CMPQ  inl, $192
	JBE   openSSETail192
	JMP   openSSETail256

openSSEFinalize:

	// hand can be used only when the memory regions don't overlap or the copy
	// direction is forward.
	//
	// BSR+branch table make almost all memmove/memclr benchmarks worse. Not worth doing.
	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

	MOVQ (8*2)(res_ptr), y_ptr
	MOVQ (8*3)(res_ptr), t1
	// Speculatively subtract
	SUBQ acc5, acc0
	SBBQ x_ptr, acc1
	SBBQ y_ptr, acc2
	SBBQ t1, acc3
	// If condition is 0, keep original value
	TESTQ t0, t0
	CMOVQEQ acc5, acc0
	CMOVQEQ x_ptr, acc1
	CMOVQEQ y_ptr, acc2
	CMOVQEQ t1, acc3
	// Store result
	MOVQ acc0, (8*0)(res_ptr)
	MOVQ acc1, (8*1)(res_ptr)
	MOVQ acc2, (8*2)(res_ptr)

	}

	if cursym.Func().Text.From.Sym.Wrapper() {
		// if g._panic != nil && g._panic.argp == FP {
		//   g._panic.argp = bottom-of-frame
		// }
		//
		//	MOVQ g_panic(g), regEntryTmp0
		//	TESTQ regEntryTmp0, regEntryTmp0
		//	JNE checkargp
		// end:
		//	NOP
		//  ... rest of function ...
		// checkargp:
		//	LEAQ (autoffset+8)(SP), regEntryTmp1
		//	CMPQ panic_argp(regEntryTmp0), regEntryTmp1