if err != nil {
			return nil, err
		}
		if allocateRemainingFrom(preferred.Intersection(aligned)) {
			return allocated, nil
		}
		// Then fallback to allocate from the aligned set if no preferred list
		// is returned (or not enough devices are returned in that list).
		if allocateRemainingFrom(aligned) {
			return allocated, nil
		}

                        if (it.getId().getGroup().startsWith("nebula")) {
                            it.belongsTo("aligned-group:nebula:\${it.getId().getVersion()}")
                        }
                        if (it.getId().getGroup().startsWith("proto")) {
                            it.belongsTo("aligned-group:proto:\${it.getId().getVersion()}")
                        }

// available in the pool.
func (bp *BytePoolCap) Get() (b []byte) {
	if bp == nil {
		return nil
	}
	select {
	case b = <-bp.c:
		// reuse existing buffer
	default:
		// create new aligned buffer
		if bp.wcap > 0 {
			b = reedsolomon.AllocAligned(1, bp.wcap)[0][:bp.w]
		} else {
			b = reedsolomon.AllocAligned(1, bp.w)[0]
		}
	}
	return
}

// Put returns the given Buffer to the BytePool.

Starting with Go 1.12, Go asm supports the PCALIGN directive, which indicates
that the next instruction should be aligned to the specified value. Currently
8 and 16 are the only supported values, and a maximum of 2 NOPs will be added
to align the code. That means in the case where the code is aligned to 4 but
PCALIGN $16 is at that location, the code will only be aligned to 8 to avoid
adding 3 NOPs.

	DMB	MB_ISHST

	STREXD	R2, (R1), R0	// stores R2 and R3
	CMP	$0, R0
	BNE	store64loop

	DMB	MB_ISH
	RET

// The following functions all panic if their address argument isn't
// 8-byte aligned. Since we're calling back into Go code to do this,
// we have to cooperate with stack unwinding. In the normal case, the
// functions tail-call into the appropriate implementation, which

// two's-complement way.
//
//go:nosplit
func (i *Int32) Add(delta int32) int32 {
	return Xaddint32(&i.value, delta)
}

// Int64 is an atomically accessed int64 value.
//
// 8-byte aligned on all platforms, unlike a regular int64.
//
// An Int64 must not be copied.
type Int64 struct {
	noCopy noCopy
	_      align64
	value  int64
}

// Load accesses and returns the value atomically.
//

	MOVV	ptr+0(FP), R4
	MOVBU	val+8(FP), R5
	// Align ptr down to 4 bytes so we can use 32-bit load/store.
	MOVV	$~3, R6
	AND	R4, R6
	// R7 = ((ptr & 3) * 8)
	AND	$3, R4, R7
	SLLV	$3, R7
	// Shift val for aligned ptr. R5 = val << R4
	SLLV	R7, R5

	DBAR
	LL	(R6), R7
	OR	R5, R7
	SC	R7, (R6)
	BEQ	R7, -4(PC)
	DBAR
	RET

// void	And8(byte volatile*, byte);
TEXT ·And8(SB), NOSPLIT, $0-9

// using the initial stack allocated by the OS.
// It calls back into standard C using the BL below.
// To do that, the stack pointer must be 8-byte-aligned.
TEXT runtime·_initcgo(SB),NOSPLIT|NOFRAME,$0
	MOVM.DB.W [R4, R14], (R13)	// push {r4, lr}

	// Ensure stack is 8-byte aligned before calling C code
	MOVW	R13, R4
	BIC	$0x7, R13

	// Allocate a TLS slot to hold g across calls to external code

func rsaAlignOf(salen int) int {
	salign := sizeofPtr
	if darwin64Bit {
		// Darwin kernels require 32-bit aligned access to
		// routing facilities.
		salign = 4
	} else if netbsd32Bit {
		// NetBSD 6 and beyond kernels require 64-bit aligned
		// access to routing facilities.
		salign = 8
	} else if runtime.GOOS == "freebsd" {
		// In the case of kern.supported_archs="amd64 i386",

	next *traceRegionAllocBlock
	off  atomic.Uintptr
}

const traceRegionAllocBlockData = 64<<10 - unsafe.Sizeof(traceRegionAllocBlockHeader{})

// alloc allocates n-byte block. The block is always aligned to 8 bytes, regardless of platform.
func (a *traceRegionAlloc) alloc(n uintptr) *notInHeap {
	n = alignUp(n, 8)
	if n > traceRegionAllocBlockData {
		throw("traceRegion: alloc too large")
	}
	if a.dropping.Load() {