ROUND1(CX,DX,AX,BX, 3,0x242070db,17);
	ROUND1(BX,CX,DX,AX, 4,0xc1bdceee,22);
	ROUND1(AX,BX,CX,DX, 5,0xf57c0faf, 7);
	ROUND1(DX,AX,BX,CX, 6,0x4787c62a,12);
	ROUND1(CX,DX,AX,BX, 7,0xa8304613,17);
	ROUND1(BX,CX,DX,AX, 8,0xfd469501,22);
	ROUND1(AX,BX,CX,DX, 9,0x698098d8, 7);
	ROUND1(DX,AX,BX,CX,10,0x8b44f7af,12);
	ROUND1(CX,DX,AX,BX,11,0xffff5bb1,17);
	ROUND1(BX,CX,DX,AX,12,0x895cd7be,22);

	MOVL	c, R9; \
	ADDL	b, a

	ROUND1(AX,BX,CX,DX, 1,0xd76aa478, 7);
	ROUND1(DX,AX,BX,CX, 2,0xe8c7b756,12);
	ROUND1(CX,DX,AX,BX, 3,0x242070db,17);
	ROUND1(BX,CX,DX,AX, 4,0xc1bdceee,22);
	ROUND1(AX,BX,CX,DX, 5,0xf57c0faf, 7);
	ROUND1(DX,AX,BX,CX, 6,0x4787c62a,12);
	ROUND1(CX,DX,AX,BX, 7,0xa8304613,17);
	ROUND1(BX,CX,DX,AX, 8,0xfd469501,22);
	ROUND1(AX,BX,CX,DX, 9,0x698098d8, 7);
	ROUND1(DX,AX,BX,CX,10,0x8b44f7af,12);

	ADDQ AX, acc3
	ADCQ DX, acc4
	ADCQ $0, acc5
	// x * y[1]
	MOVQ (8*1)(y_ptr), t0

	MOVQ (8*0)(x_ptr), AX
	MULQ t0
	ADDQ AX, acc1
	ADCQ $0, DX
	MOVQ DX, t1

	MOVQ (8*1)(x_ptr), AX
	MULQ t0
	ADDQ t1, acc2
	ADCQ $0, DX
	ADDQ AX, acc2
	ADCQ $0, DX
	MOVQ DX, t1

	MOVQ (8*2)(x_ptr), AX
	MULQ t0
	ADDQ t1, acc3
	ADCQ $0, DX
	ADDQ AX, acc3
	ADCQ $0, DX
	MOVQ DX, t1

	ROUND1(BP, AX, BX, CX, DX, 1)
	ROUND1(DX, BP, AX, BX, CX, 2)
	ROUND1(CX, DX, BP, AX, BX, 3)
	ROUND1(BX, CX, DX, BP, AX, 4)
	ROUND1(AX, BX, CX, DX, BP, 5)
	ROUND1(BP, AX, BX, CX, DX, 6)
	ROUND1(DX, BP, AX, BX, CX, 7)
	ROUND1(CX, DX, BP, AX, BX, 8)
	ROUND1(BX, CX, DX, BP, AX, 9)
	ROUND1(AX, BX, CX, DX, BP, 10)
	ROUND1(BP, AX, BX, CX, DX, 11)
	ROUND1(DX, BP, AX, BX, CX, 12)
	ROUND1(CX, DX, BP, AX, BX, 13)

	ADCQ   DX, SI

	// r0 += 19×a4×b1
	MOVQ   32(CX), AX
	IMUL3Q $0x13, AX, AX
	MULQ   8(BX)
	ADDQ   AX, DI
	ADCQ   DX, SI

	// r1 = a0×b1
	MOVQ (CX), AX
	MULQ 8(BX)
	MOVQ AX, R9
	MOVQ DX, R8

	// r1 += a1×b0
	MOVQ 8(CX), AX
	MULQ (BX)
	ADDQ AX, R9
	ADCQ DX, R8

	// r1 += 19×a2×b4
	MOVQ   16(CX), AX
	IMUL3Q $0x13, AX, AX
	MULQ   32(BX)
	ADDQ   AX, R9
	ADCQ   DX, R8

	ROUND2(AX, BX, CX, DX, BP, 30)
	ROUND2(BP, AX, BX, CX, DX, 31)
	ROUND2(DX, BP, AX, BX, CX, 32)
	ROUND2(CX, DX, BP, AX, BX, 33)
	ROUND2(BX, CX, DX, BP, AX, 34)
	ROUND2(AX, BX, CX, DX, BP, 35)
	ROUND2(BP, AX, BX, CX, DX, 36)
	ROUND2(DX, BP, AX, BX, CX, 37)
	ROUND2(CX, DX, BP, AX, BX, 38)
	ROUND2(BX, CX, DX, BP, AX, 39)

	ROUND3(AX, BX, CX, DX, BP, 40)
	ROUND3(BP, AX, BX, CX, DX, 41)

	MOVQ	m_g0(CX), DX		// g

	// Layout new m scheduler stack on os stack.
	MOVQ	SP, AX
	MOVQ	AX, (g_stack+stack_hi)(DX)
	SUBQ	$(64*1024), AX		// initial stack size (adjusted later)
	MOVQ	AX, (g_stack+stack_lo)(DX)
	ADDQ	$const_stackGuard, AX
	MOVQ	AX, g_stackguard0(DX)
	MOVQ	AX, g_stackguard1(DX)

	// Set up tls.
	LEAQ	m_tls(CX), DI
	MOVQ	CX, g_m(DX)
	MOVQ	DX, g(DI)

	MOVL	val+4(FP), CX
casloop:
	MOVL 	CX, DX
	MOVL	(BX), AX
	ANDL	AX, DX
	LOCK
	CMPXCHGL	DX, (BX)
	JNZ casloop
	MOVL 	AX, ret+8(FP)
	RET

// func Or32(addr *uint32, v uint32) old uint32
TEXT ·Or32(SB), NOSPLIT, $0-12
	MOVL	ptr+0(FP), BX
	MOVL	val+4(FP), CX
casloop:
	MOVL 	CX, DX
	MOVL	(BX), AX
	ORL	AX, DX
	LOCK
	CMPXCHGL	DX, (BX)
	JNZ casloop
	MOVL 	AX, ret+8(FP)
	RET

	MOVL	g(CX), BX	// BX = g
	MOVL	g_m(BX), DX	// DX = curm

	// set g to gcrash
	LEAL	runtime·gcrash(SB), BX // g = &gcrash
	MOVL	DX, g_m(BX)            // g.m = curm
	MOVL	BX, m_g0(DX)           // curm.g0 = g
	get_tls(CX)
	MOVL	BX, g(CX)

	// switch to crashstack
	MOVL	(g_stack+stack_hi)(BX), DX
	SUBL	$(4*8), DX
	MOVL	DX, SP

	// call target function
	MOVL	AX, DX
	MOVL	0(AX), AX
	CALL	AX

	MOVL	((index-2)*4)(BP), AX; \
	MOVL	AX, CX; \
	RORL	$17, AX; \
	MOVL	CX, DX; \
	RORL	$19, CX; \
	SHRL	$10, DX; \
	MOVL	((index-15)*4)(BP), BX; \
	XORL	CX, AX; \
	MOVL	BX, CX; \
	XORL	DX, AX; \
	RORL	$7, BX; \
	MOVL	CX, DX; \
	SHRL	$3, DX; \
	RORL	$18, CX; \
	ADDL	((index-7)*4)(BP), AX; \
	XORL	CX, BX; \
	XORL	DX, BX; \
	ADDL	((index-16)*4)(BP), BX; \
	ADDL	BX, AX; \
	MOVL	AX, ((index)*4)(BP)