ROUND4(BP, AX, BX, CX, DX, 76)
	ROUND4(DX, BP, AX, BX, CX, 77)
	ROUND4(CX, DX, BP, AX, BX, 78)
	ROUND4(BX, CX, DX, BP, AX, 79)

	ADDL	64(SP), AX
	ADDL	68(SP), BX
	ADDL	72(SP), CX
	ADDL	76(SP), DX
	ADDL	80(SP), BP

	MOVL	88(SP), SI
	ADDL	$64, SI
	CMPL	SI, 84(SP)
	JB	loop

end:
	MOVL	dig+0(FP), DI
	MOVL	AX, (0*4)(DI)
	MOVL	BX, (1*4)(DI)
	MOVL	CX, (2*4)(DI)

	// Get popRet.
	MOVL	(12+callbackArgs_retPop)(SP), CX	// Can't use a callee-save register
	ADDL	$(12+callbackArgs__size), SP

	// restore registers as required for windows callback
	POPL	BX
	POPL	BP
	POPL	SI
	POPL	DI

	// remove callback parameters before return (as per Windows spec)
	POPL	DX
	ADDL	CX, SP
	PUSHL	DX

	CLD

	RET

// void tstart(M *newm);

	MOVL y+8(FP), BP
	LEAL (DI)(BX*4), DI
	LEAL (SI)(BX*4), SI
	NEGL BX			// i = -n
	MOVL $0, CX		// c = 0
	JMP E6

L6:	MOVL (SI)(BX*4), AX
	MULL BP
	ADDL CX, AX
	ADCL $0, DX
	ADDL AX, (DI)(BX*4)
	ADCL $0, DX
	MOVL DX, CX
	ADDL $1, BX		// i++

E6:	CMPL BX, $0		// i < 0
	JL L6

	MOVL CX, c+12(FP)

largeloop:
	CMPL	BP, $16
	JB	mediumloop
	MOVOU	(SI), X0
	MOVOU	(DI), X1
	PCMPEQB X0, X1
	PMOVMSKB X1, BX
	XORL	$0xffff, BX	// convert EQ to NE
	JNE	diff16	// branch if at least one byte is not equal
	ADDL	$16, SI
	ADDL	$16, DI
	SUBL	$16, BP
	JMP	largeloop

diff16:
	BSFL	BX, BX	// index of first byte that differs
	XORL	DX, DX
	MOVB	(SI)(BX*1), CX
	CMPB	CX, (DI)(BX*1)
	SETHI	DX

	PCMPEQB	X7, X6
	PAND	X2, X0
	PAND	X6, X4
	PAND	X4, X0
	PMOVMSKB X0, DX
	ADDL	$64, SI
	ADDL	$64, DI
	SUBL	$64, BX
	CMPL	DX, $0xffff
	JEQ	hugeloop
	MOVB	$0, (AX)
	RET

	// 4 bytes at a time using 32-bit register
bigloop:
	CMPL	BX, $4
	JBE	leftover
	MOVL	(SI), CX
	MOVL	(DI), DX
	ADDL	$4, SI
	ADDL	$4, DI
	SUBL	$4, BX
	CMPL	CX, DX
	JEQ	bigloop
	MOVB	$0, (AX)
	RET

 * check overlap
 */
back:
	MOVL	SI, CX
	ADDL	BX, CX
	CMPL	CX, DI
	JLS	forward
/*
 * whole thing backwards has
 * adjusted addresses
 */

	ADDL	BX, DI
	ADDL	BX, SI
	STD

/*
 * copy
 */
	MOVL	BX, CX
	SHRL	$2, CX
	ANDL	$3, BX

	SUBL	$4, DI
	SUBL	$4, SI
	REP;	MOVSL

	CLD
	ADDL	$4, DI
	ADDL	$4, SI
	SUBL	BX, DI
	SUBL	BX, SI
	JMP	tail

	ROUND4(AX, BX, CX, DX, BP, 75)
	ROUND4(BP, AX, BX, CX, DX, 76)
	ROUND4(DX, BP, AX, BX, CX, 77)
	ROUND4(CX, DX, BP, AX, BX, 78)
	ROUND4(BX, CX, DX, BP, AX, 79)

	ADDL	R11, AX
	ADDL	R12, BX
	ADDL	R13, CX
	ADDL	R14, DX
	ADDL	R15, BP

	ADDQ	$64, SI
	CMPQ	SI, DI
	JB	loop

end:
	MOVQ	dig+0(FP), DI
	MOVL	AX, (0*4)(DI)
	MOVL	BX, (1*4)(DI)
	MOVL	CX, (2*4)(DI)
	MOVL	DX, (3*4)(DI)

	// These tests check that property for backwards-compatibility.
	MOVL 2147483648(AX), AX  // 8b8000000080
	MOVL -2147483648(AX), AX // 8b8000000080
	ADDL 2147483648(AX), AX  // 038000000080
	ADDL -2147483648(AX), AX // 038000000080
	// Make sure MOV CR/DR continues to work after changing its movtabs.
	MOVL CR0, AX // 0f20c0
	MOVL CR0, DX // 0f20c2
	MOVL CR4, DI // 0f20e7
	MOVL AX, CR0 // 0f22c0

	MOVL	(_INTERRUPT_TIME+time_lo), CX
	MOVL	(_INTERRUPT_TIME+time_hi2), DI
	CMPL	AX, DI
	JNE	loop

	// w = DI:CX
	// multiply by 100
	MOVL	$100, AX
	MULL	CX
	IMULL	$100, DI
	ADDL	DI, DX
	// w*100 = DX:AX
	MOVL	AX, mono+12(FP)
	MOVL	DX, mono+16(FP)

wall:
	MOVL	(_SYSTEM_TIME+time_hi1), CX
	MOVL	(_SYSTEM_TIME+time_lo), AX
	MOVL	(_SYSTEM_TIME+time_hi2), DX
	CMPL	CX, DX
	JNE	wall

    .text
    .globl  _sum
_sum:
    movl    8(%esp), %eax
    addl    4(%esp), %eax