//go:nosplit
func asanread(addr unsafe.Pointer, sz uintptr) {
	sp := getcallersp()
	pc := getcallerpc()
	doasanread(addr, sz, sp, pc)
}

//go:linkname asanwrite
//go:nosplit
func asanwrite(addr unsafe.Pointer, sz uintptr) {
	sp := getcallersp()
	pc := getcallerpc()
	doasanwrite(addr, sz, sp, pc)
}

//go:noescape
func doasanread(addr unsafe.Pointer, sz, sp, pc uintptr)

//go:noescape

	// The values are in registers.
	MOVQ	CX, (16+0)(SP)
	MOVQ	DX, (16+8)(SP)
	MOVQ	R8, (16+16)(SP)
	MOVQ	R9, (16+24)(SP)
	// R8 = address of args vector
	LEAQ	(16+0)(SP), R8

	// remove return address from stack, we are not returning to callbackasm, but to its caller.
	MOVQ	0(SP), AX
	ADDQ	$8, SP

	// determine index into runtime·cbs table
	MOVQ	$runtime·callbackasm(SB), DX

	// Align stack to call C function.
	// We moved SP to BP above, but BP was clobbered by the libpreinit call.
	MOVL	SP, BP
	ANDL	$~15, SP

	MOVL	$_rt0_386_lib_go(SB), BX
	MOVL	BX, 0(SP)
	MOVL	$0, 4(SP)

	CALL	AX

	MOVL	BP, SP

	JMP	restore

nocgo:
	MOVL	$0x800000, 0(SP)                    // stacksize = 8192KB
	MOVL	$_rt0_386_lib_go(SB), AX
	MOVL	AX, 4(SP)                           // fn

	MOVQ	R15, r15-(14*8+8)(SP)
	MOVQ	R14, r14-(13*8+8)(SP)
	MOVQ	R13, r13-(12*8+8)(SP)
	MOVQ	R12, r12-(11*8+8)(SP)
	MOVQ	R11, r11-(10*8+8)(SP)
	MOVQ	R10, r10-(9*8+8)(SP)
	MOVQ	R9, r9-(8*8+8)(SP)
	MOVQ	R8, r8-(7*8+8)(SP)
	MOVQ	DI, di-(6*8+8)(SP)
	MOVQ	SI, si-(5*8+8)(SP)
	MOVQ	BP, bp-(4*8+8)(SP)
	MOVQ	BX, bx-(3*8+8)(SP)
	MOVQ	DX, dx-(2*8+8)(SP)

	MOV	$8(SP), (X5)			// ERROR "address load must target register"
	MOVB	$8(SP), X5			// ERROR "unsupported address load"
	MOVH	$8(SP), X5			// ERROR "unsupported address load"
	MOVW	$8(SP), X5			// ERROR "unsupported address load"
	MOVF	$8(SP), X5			// ERROR "unsupported address load"
	MOV	$1234, 0(SP)			// ERROR "constant load must target register"
	MOV	$1234, 8(SP)			// ERROR "constant load must target register"

	increment(5)
	MOVOU T0, (8*16 + 6*16)(SP)
	increment(6)
	MOVOU T0, (8*16 + 7*16)(SP)
	increment(7)

	MOVOU (8*16 + 0*16)(SP), B0
	MOVOU (8*16 + 1*16)(SP), B1
	MOVOU (8*16 + 2*16)(SP), B2
	MOVOU (8*16 + 3*16)(SP), B3
	MOVOU (8*16 + 4*16)(SP), B4
	MOVOU (8*16 + 5*16)(SP), B5
	MOVOU (8*16 + 6*16)(SP), B6
	MOVOU (8*16 + 7*16)(SP), B7

	aesRound(1)
	increment(0)
	aesRound(2)
	increment(1)
	aesRound(3)

//
// The stack holds the intermediate word array - 16 uint32s - at 0(SP) up to 64(SP).
// The saved a, b, c, d, e (R11 through R15 on amd64) are at 64(SP) up to 84(SP).
// The saved limit pointer (DI on amd64) is at 84(SP).
// The saved data pointer (SI on amd64) is at 88(SP).

#define LOAD(index, e) \
	MOVL	88(SP), SI; \
	MOVL	(index*4)(SI), DI; \
	BSWAPL	DI; \
	MOVL	DI, (index*4)(SP); \
	ADDL	DI, e

#define SHUFFLE(index, e) \

	return dst == src &&
		r.Overlaps(r.Add(sp.Sub(r.Min))) &&
		(sp.Y < r.Min.Y || (sp.Y == r.Min.Y && sp.X < r.Min.X))
}

// Draw calls [DrawMask] with a nil mask.
func Draw(dst Image, r image.Rectangle, src image.Image, sp image.Point, op Op) {
	DrawMask(dst, r, src, sp, nil, image.Point{}, op)
}

			matchFile := (loc.Mapping != nil && sp.sym.MatchString(loc.Mapping.File))
			for j, line := range loc.Line {
				if (j == 0 && matchFile) || matches(line) {
					markInterest(addr, loc, j)
				}
			}
		}
	}

	sp.expandAddresses(rpt, addrs, flat)
	sp.initSamples(flat, cum)
	return sp
}

func (sp *sourcePrinter) close() {
	for _, objFile := range sp.objects {
		if objFile != nil {

	SHLL	$6,		DX

	LEAL	(SI)(DX*1),	DI
	MOVL	(0*4)(BP),	AX
	MOVL	(1*4)(BP),	BX
	MOVL	(2*4)(BP),	CX
	MOVL	(3*4)(BP),	DX

	CMPL	SI,		DI
	JEQ	end

	MOVL	DI,		16(SP)

loop:
	MOVL	AX,		0(SP)
	MOVL	BX,		4(SP)
	MOVL	CX,		8(SP)
	MOVL	DX,		12(SP)

	MOVL	(0*4)(SI),	DI
	MOVL	DX,		BP

	ROUND1(AX,BX,CX,DX, 1,0xd76aa478, 7);
	ROUND1(DX,AX,BX,CX, 2,0xe8c7b756,12);
	ROUND1(CX,DX,AX,BX, 3,0x242070db,17);