// already zeroed. Otherwise if needzero is true, objects are zeroed as
// they are allocated. There are various benefits to delaying zeroing
// this way:
//
//	1. Stack frame allocation can avoid zeroing altogether.
//
//	2. It exhibits better temporal locality, since the program is
//	   probably about to write to the memory.
//
//	3. We don't zero pages that never get reused.

			(MOVOstoreconst [makeValAndOff(0,16)] destptr
				(MOVOstoreconst [makeValAndOff(0,0)] destptr mem))))

// Medium zeroing uses a duff device.
(Zero [s] destptr mem)
	&& s > 64 && s <= 1024 && s%16 == 0 && !config.noDuffDevice =>
	(DUFFZERO [s] destptr mem)

// Large zeroing uses REP STOSQ.
(Zero [s] destptr mem)
	&& (s > 1024 || (config.noDuffDevice && s > 64 || !config.useSSE && s > 32))
	&& s%8 == 0 =>

	VPEXTRW $17, X20, (SP)(AX*2)      // 62e37d0815244411 or 62e3fd0815244411
	VPEXTRW $127, X20, (SP)(AX*2)     // 62e37d081524447f or 62e3fd081524447f
	// EVEX: embedded zeroing.
	VADDPD.Z X30, X1, K7, X0  // 6291f58f58c6
	VMAXPD.Z (AX), Z2, K1, Z1 // 62f1edc95f08
	// EVEX: embedded rounding.
	VADDPD.RU_SAE Z3, Z2, K1, Z1   // 62f1ed5958cb
	VADDPD.RD_SAE Z3, Z2, K1, Z1   // 62f1ed3958cb

			(MOVBstore [0] ptr (MOVWconst [0]) mem)))

// Medium zeroing uses a duff device
// 4 and 128 are magic constants, see runtime/mkduff.go
(Zero [s] {t} ptr mem)
	&& s%4 == 0 && s > 4 && s <= 512
	&& t.Alignment()%4 == 0 && !config.noDuffDevice =>
	(DUFFZERO [4 * (128 - s/4)] ptr (MOVWconst [0]) mem)

// Large zeroing uses a loop
(Zero [s] {t} ptr mem)

				clobbers: buildReg("R16 R17 R20 R30"),
			},
			faultOnNilArg0: true,
			unsafePoint:    true, // FP maintenance around DUFFZERO can be clobbered by interrupts
		},

		// large zeroing
		// arg0 = address of memory to zero (in R16 aka arm64.REGRT1, changed as side effect)
		// arg1 = address of the last 16-byte unit to zero
		// arg2 = mem
		// returns mem
		//	STP.P	(ZR,ZR), 16(R16)

		// and extended for more than just bytes (like nibbles
		// and uint16s) by using an appropriate constant.
		//
		// To summarize the technique, quoting from that page:
		// "[It] works by first zeroing the high bits of the [8]
		// bytes in the word. Subsequently, it adds a number that
		// will result in an overflow to the high bit of a byte if
		// any of the low bits were initially set. Next the high

		typ = *(**_type)(unsafe.Pointer(addr))
		addr += mallocHeaderSize
	} else {
		typ = span.largeType
		if typ == nil {
			// Allow a nil type here for delayed zeroing. See mallocgc.
			return typePointers{}
		}
	}
	gcdata := typ.GCData
	return typePointers{elem: addr, addr: addr, mask: readUintptr(gcdata), typ: typ}
}

		// value of the argument type. This avoids including
		// string.h for memset, and is also robust to C++
		// types with constructors. Both GCC and LLVM optimize
		// this into just zeroing _cgo_a.
		fmt.Fprintf(fgcc, "\ttypedef %s %v _cgo_argtype;\n", ctype, p.packedAttribute())
		fmt.Fprintf(fgcc, "\tstatic _cgo_argtype _cgo_zero;\n")
		fmt.Fprintf(fgcc, "\t_cgo_argtype _cgo_a = _cgo_zero;\n")

	B := uint8(0)
	for overLoadFactor(hint, B) {
		B++
	}
	h.B = B

	// allocate initial hash table
	// if B == 0, the buckets field is allocated lazily later (in mapassign)
	// If hint is large zeroing this memory could take a while.
	if h.B != 0 {
		var nextOverflow *bmap
		h.buckets, nextOverflow = makeBucketArray(t, h.B, nil)
		if nextOverflow != nil {
			h.extra = new(mapextra)

					&BaseEndpointInfo{ip: "10.0.1.5", port: 80, endpoint: "10.0.1.5:80", isLocal: true, ready: true, serving: true, terminating: false},
					&BaseEndpointInfo{ip: "10.0.2.1", port: 80, endpoint: "10.0.2.1:80", isLocal: false, ready: true, serving: true, terminating: false},
					&BaseEndpointInfo{ip: "10.0.2.2", port: 80, endpoint: "10.0.2.2:80", isLocal: true, ready: true, serving: true, terminating: false},