MOVOU	(SI), X0
	MOVOU	(DI), X1
	PCMPEQB X0, X1
	PMOVMSKB X1, AX
	XORQ	$0xffff, AX
	JNE	diff16

	MOVOU	16(SI), X0
	MOVOU	16(DI), X1
	PCMPEQB X0, X1
	PMOVMSKB X1, AX
	XORQ	$0xffff, AX
	JNE	diff32

	MOVOU	32(SI), X0
	MOVOU	32(DI), X1
	PCMPEQB X0, X1
	PMOVMSKB X1, AX
	XORQ	$0xffff, AX
	JNE	diff48

	MOVOU	48(SI), X0
	MOVOU	48(DI), X1
	PCMPEQB X0, X1
	PMOVMSKB X1, AX

	MOVL	trap+0(FP), AX	// syscall entry
	// copy args down
	LEAL	a1+4(FP), SI
	LEAL	sysargs-144(SP), DI
	CLD
	MOVSL
	MOVSL
	MOVSL
	INT	$64
	MOVL	AX, r1+16(FP)
	MOVL	$0, r2+20(FP)
	CMPL	AX, $-1
	JNE	ok3

	LEAL	errbuf-128(SP), AX
	MOVL	AX, sysargs-144(SP)
	MOVL	$128, sysargs1-140(SP)
	MOVL	$SYS_ERRSTR, AX
	INT	$64
	CALL	runtime·exitsyscall(SB)
	MOVL	sysargs-144(SP), AX
	MOVL	AX, errbuf-148(SP)

	MOVL	(0*4)(BX), AX		// fn
	CALL	AX

	MOVL	AX, (4*4)(BX)		// r1
	MOVL	DX, (5*4)(BX)		// r2

	// Standard libc functions return -1 on error and set errno.
	CMPL	AX, $-1
	JNE	ok
	CMPL	DX, $-1
	JNE	ok

	// Get error code from libc.
	CALL	libc_errno(SB)
	MOVL	(AX), AX
	MOVW	AX, (6*4)(BX)		// err

ok:
	MOVL	$0, AX			// no error (it's ignored anyway)
	MOVL	BP, SP
	POPL	BP

func ptrEqual() {
	// amd64:-"JEQ",-"JNE"
	// ppc64x:-"BEQ",-"BNE"
	// s390x:-"BEQ",-"BNE"
	copy(x[:], x[:])
}

func ptrOneOffset() {
	// amd64:-"JEQ",-"JNE"
	// ppc64x:-"BEQ",-"BNE"
	// s390x:-"BEQ",-"BNE"
	copy(x[1:], x[:])
}

func ptrBothOffset() {
	// amd64:-"JEQ",-"JNE"
	// ppc64x:-"BEQ",-"BNE"
	// s390x:-"BEQ",-"BNE"
	copy(x[1:], x[2:])
}

	MOVQ	DI, BX
	CALL	libc_mach_absolute_time(SB)
	MOVQ	AX, 0(BX)
	MOVL	timebase<>+machTimebaseInfo_numer(SB), SI
	MOVL	timebase<>+machTimebaseInfo_denom(SB), DI // atomic read
	TESTL	DI, DI
	JNE	initialized

	SUBQ	$(machTimebaseInfo__size+15)/16*16, SP
	MOVQ	SP, DI
	CALL	libc_mach_timebase_info(SB)
	MOVL	machTimebaseInfo_numer(SP), SI
	MOVL	machTimebaseInfo_denom(SP), DI

aligned:
	MOVQ $0, AX // position in slices

	PCALIGN $16
loop16b:
	MOVOU (SI)(AX*1), X0   // XOR 16byte forwards.
	MOVOU (CX)(AX*1), X1
	PXOR  X1, X0
	MOVOU X0, (BX)(AX*1)
	ADDQ  $16, AX
	CMPQ  DX, AX
	JNE   loop16b
	RET

	PCALIGN $16
loop_1b:
	SUBQ  $1, DX           // XOR 1byte backwards.
	MOVB  (SI)(DX*1), DI
	MOVB  (CX)(DX*1), AX
	XORB  AX, DI
	MOVB  DI, (BX)(DX*1)

	// will clobber the test pattern created by the caller
	ADDL	$(1024 * 8), SP

	// Ask signaller to setgid
	MOVL	$1, (BX)

	// Wait for setgid completion
loop:
	PAUSE
	MOVL	(BX), AX
	CMPL	AX, $0
	JNE	loop

	// Restore stack
	SUBL	$(1024 * 8), SP

	MOVQ	trap+0(FP), BP	// syscall entry
	// copy args down
	LEAQ	a1+8(FP), SI
	LEAQ	sysargs-160(SP), DI
	CLD
	MOVSQ
	MOVSQ
	MOVSQ
	SYSCALL
	MOVQ	AX, r1+32(FP)
	MOVQ	$0, r2+40(FP)
	CMPL	AX, $-1
	JNE	ok3

	LEAQ	errbuf-128(SP), AX
	MOVQ	AX, sysargs-160(SP)
	MOVQ	$128, sysargs1-152(SP)
	MOVQ	$SYS_ERRSTR, BP
	SYSCALL
	CALL	runtime·exitsyscall(SB) // call via ABI wrapper, ensuring ABIInternal fixed registers are set

	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

	// w = DX:AX

	MOVQ	8(BX), SI		// arg 2 len
	MOVL	16(BX), DX		// arg 3 prot
	MOVL	20(BX), CX		// arg 4 flags
	MOVL	24(BX), R8		// arg 5 fid
	MOVL	28(BX), R9		// arg 6 offset
	CALL	libc_mmap(SB)
	XORL	DX, DX
	CMPQ	AX, $-1
	JNE	ok
	CALL	libc_errno(SB)
	MOVLQSX	(AX), DX		// errno
	XORQ	AX, AX
ok:
	MOVQ	AX, 32(BX)
	MOVQ	DX, 40(BX)
	RET

TEXT runtime·munmap_trampoline(SB),NOSPLIT,$0
	MOVQ	8(DI), SI		// arg 2 len