//	func f(x uint64) (uint64, uint64) { return x/1000000000, x%1000000000 }
	// adapted to reduce duplication
	MOVQ	AX, CX
	MOVQ	$1360296554856532783, AX
	MULQ	CX
	ADDQ	CX, DX
	RCRQ	$1, DX
	SHRQ	$29, DX
	MOVQ	DX, sec+0(FP)
	IMULQ	$1000000000, DX
	SUBQ	DX, CX
	MOVL	CX, nsec+8(FP)
	RET

TEXT runtime·notify(SB),NOSPLIT,$0
	MOVQ	$28, BP
	SYSCALL
	MOVL	AX, ret+8(FP)
	RET

	MULQ R9
	ADDQ CX, AX
	ADCQ $0, DX
	MOVQ AX, (0*8)(R10)(BX*8)
	MOVQ DX, CX
	MOVQ (1*8)(R8)(BX*8), AX
	MULQ R9
	ADDQ CX, AX
	ADCQ $0, DX
	MOVQ AX, (1*8)(R10)(BX*8)
	MOVQ DX, CX
	MOVQ (2*8)(R8)(BX*8), AX
	MULQ R9
	ADDQ CX, AX
	ADCQ $0, DX
	MOVQ AX, (2*8)(R10)(BX*8)
	MOVQ DX, CX
	MOVQ (3*8)(R8)(BX*8), AX
	MULQ R9
	ADDQ CX, AX
	ADCQ $0, DX
	MOVQ AX, (3*8)(R10)(BX*8)
	MOVQ DX, CX

	PMOVMSKB X1, DX	// Move result bits to integer register.
	MOVL	BX, CX
	SHLL	CX, DX
	SHRL	$16, DX	// Shift desired bits down to bottom of register.
	BSFL	DX, DX	// Find first set bit.
	JZ	failure	// No set bit, failure.
	MOVQ	DX, (R8)
	RET

avx2:
#ifndef hasAVX2
	CMPB   internal∕cpu·X86+const_offsetX86HasAVX2(SB), $1
	JNE sse
#endif
	MOVD AX, X0
	LEAQ -32(SI)(BX*1), R11

	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

	MOVOU	(SI), X1
	// Compare target byte with each byte in data.
	PCMPEQB	X0, X1
	// Move result bits to integer register.
	PMOVMSKB X1, DX
	// Apply mask
	ANDQ R10, DX
	POPCNTL DX, DX
	// Directly return DX, we don't need to accumulate
	// since we have <16 bytes.
	MOVQ	DX, (R8)
	RET
endzero:
	MOVQ $0, (R8)
	RET

endofpage:
	// We must ignore low bytes as they aren't part of our slice.

	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)

	VADDPD 2032(DX), X29, X0        // 62f1950058427f
	VADDPD 2032(DX), X1, X29        // 6261f508586a7f
	VADDPD 2032(DX), X29, X28       // 6261950058627f
	VADDPD 2032(DX)(AX*2), X29, X0  // 62f195005844427f
	VADDPD 2032(DX)(AX*2), X1, X29  // 6261f508586c427f
	VADDPD 2032(DX)(AX*2), X29, X28 // 626195005864427f
	VADDPD 4064(DX), Y0, Y29        // 6261fd28586a7f
	VADDPD 4064(DX), Y29, Y1        // 62f19520584a7f

	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	((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)