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.

	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)

	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

	// remaining 0-4 bytes

	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)

	//	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)

	SBBL $(delta >> 32), DX

	// nano/100 = DX:AX
	// split into two decimal halves by div 1e9.
	// (decimal point is two spots over from correct place,
	// but we avoid overflow in the high word.)
	MOVL	$1000000000, CX
	DIVL	CX
	MOVL	AX, DI
	MOVL	DX, SI

	// DI = nano/100/1e9 = nano/1e11 = sec/100, DX = SI = nano/100%1e9
	// split DX into seconds and nanoseconds by div 1e7 magic multiply.
	MOVL	DX, AX
	MOVL	$1801439851, CX

#include "funcdata.h"

//
// System call support for AMD64, Darwin
//

// Trap # in AX, args in DI SI DX, return in AX DX

// func Syscall(trap, a1, a2, a3 uintptr) (r1, r2 uintptr, err Errno);
TEXT	·Syscall(SB),NOSPLIT,$0-56
	CALL	runtime·entersyscall<ABIInternal>(SB)
	MOVQ	a1+8(FP), DI
	MOVQ	a2+16(FP), SI
	MOVQ	a3+24(FP), DX
	MOVQ	trap+0(FP), AX	// syscall entry
	ADDQ	$0x2000000, AX
	SYSCALL
	JCC	ok
	MOVQ	$-1, r1+32(FP)