for i := 0; i < p; i++ {
		i := i
		go func() {
			for j := 0; j < N; j++ {
				// Write a pointer using memmove.
				copy(collider[i:i+1], ptrs[i:i+1])
				// Write nil using memclr.
				// (This is a magic loop that gets lowered to memclr.)
				r := collider[i : i+1]
				for k := range r {
					r[k] = nil
				}
			}
			done <- struct{}{}
		}()
	}
	// Start readers.  They read pointers from slots

// asmcheck

// Copyright 2022 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.

// Test that optimized range memclr works with pointers to arrays.
// The clears get inlined, see https://github.com/golang/go/issues/56997

package codegen

type T struct {
	a *[10]int
	b [10]int
}

func (t *T) f() {
	// amd64:-".*runtime.memclrNoHeapPointers"

}

func SliceExtensionInt64(s []int, l64 int64) []int {
	// 386:`.*runtime\.makeslice`
	// 386:-`.*runtime\.memclr`
	return append(s, make([]int, l64)...)
}

// ------------------ //
//      Make+Copy     //
// ------------------ //

// Issue #26252 - avoid memclr for make+copy

func SliceMakeCopyLen(s []int) []int {
	// amd64:`.*runtime\.mallocgc`
	// amd64:`.*runtime\.memmove`

TEXT runtime·memclrNoHeapPointers(SB), NOSPLIT, $0-8
	MOVL	ptr+0(FP), DI
	MOVL	n+4(FP), BX
	XORL	AX, AX

	// MOVOU seems always faster than REP STOSL.
tail:
	// BSR+branch table make almost all memmove/memclr benchmarks worse. Not worth doing.
	TESTL	BX, BX
	JEQ	_0
	CMPL	BX, $2
	JBE	_1or2
	CMPL	BX, $4
	JB	_3
	JE	_4
	CMPL	BX, $8
	JBE	_5through8
	CMPL	BX, $16
	JBE	_9through16
#ifdef GO386_softfloat

	// AX = ptr
	// BX = n
	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

	// 128 because that is the maximum SSE register load (loading all data
	// into registers lets us ignore copy direction).
tail:
	// BSR+branch table make almost all memmove/memclr benchmarks worse. Not worth doing.
	TESTL	BX, BX
	JEQ	move_0
	CMPL	BX, $2
	JBE	move_1or2
	CMPL	BX, $4
	JB	move_3
	JE	move_4
	CMPL	BX, $8
	JBE	move_5through8
	CMPL	BX, $16
	JBE	move_9through16

}

// arrayClear constructs a call to runtime.memclr for fast zeroing of slices and arrays.
func arrayClear(wbPos src.XPos, a ir.Node, nrange *ir.RangeStmt) ir.Node {
	elemsize := typecheck.RangeExprType(a.Type()).Elem().Size()
	if elemsize <= 0 {
		return nil
	}

	// Convert to
	// if len(a) != 0 {
	// 	hp = &a[0]
	// 	hn = len(a)*sizeof(elem(a))
	// 	memclr{NoHeap,Has}Pointers(hp, hn)
	// 	i = len(a) - 1
	// }

//	      s = growslice(T, s.ptr, n, s.cap, l2, T)
//	    }
//	    // clear the new portion of the underlying array.
//	    hp := &s[len(s)-l2]
//	    hn := l2 * sizeof(T)
//	    memclr(hp, hn)
//	  }
//	}
//	s
//
//	if T has pointers, the final memclr can go inside the "then" branch, as
//	growslice will have done the clearing for us.

func extendSlice(n *ir.CallExpr, init *ir.Nodes) ir.Node {

	// registers before writing it back.  move_256through2048 on the other
	// 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

 * bytecode manipulation libraries tools like ASM.
 *
 * <p>The notion of "member" here is similar to, but broader than
 * {@link java.lang.reflect.Member}. The latter is essentially an abstraction over fields,
 * methods and constructors; this Member and its subtypes represent not only fields and
 * methods, but also classes, inner classes, annotations and their values, and more. This