MOVOU	48(DI), X1
	PCMPEQB X0, X1
	PMOVMSKB X1, AX
	XORQ	$0xffff, AX
	JNE	diff64

	ADDQ	$64, SI
	ADDQ	$64, DI
	SUBQ	$64, R8
	CMPQ	R8, $64
	JBE	loop
	JMP	big_loop
#endif

	// Compare 64-bytes per loop iteration.
	// Loop is unrolled and uses AVX2.
big_loop_avx2:
	VMOVDQU	(SI), Y2
	VMOVDQU	(DI), Y3
	VMOVDQU	32(SI), Y4
	VMOVDQU	32(DI), Y5
	VPCMPEQB Y2, Y3, Y0
	VPMOVMSKB Y0, AX

TEXT runtime·memequal(SB),NOSPLIT,$0-13
	MOVW	a+0(FP), R1
	MOVW	b+4(FP), R2
	BEQ	R1, R2, eq
	MOVW	size+8(FP), R3
	ADDU	R1, R3, R4
loop:
	BNE	R1, R4, test
	MOVW	$1, R1
	MOVB	R1, ret+12(FP)
	RET
test:
	MOVBU	(R1), R6
	ADDU	$1, R1
	MOVBU	(R2), R7
	ADDU	$1, R2
	BEQ	R6, R7, loop

	MOVB	R0, ret+12(FP)
	RET
eq:
	MOVW	$1, R1
	MOVB	R1, ret+12(FP)
	RET

// memequal_varlen(a, b unsafe.Pointer) bool

	CMP   R5, $0
	MOVD  R5, CTR         // Set up loop counter
	BEQ   tail            // If R5 = 0, we can't use the loop

	// Process 4 elements per iteration. Unrolling this loop
	// means a performance trade-off: we will lose performance
	// for small values of z_len (0.90x in the worst case), but
	// gain significant performance as z_len increases (up to
	// 1.45x).

	PCALIGN $16
loop:
	MOVD  8(R8), R11      // R11 = x[i]

	ADDV	$(1024*8), R3

	// Ask signaller to setgid
	MOVW	$1, R12
	DBAR
	MOVW	R12, ·Baton(SB)
	DBAR

	// Wait for setgid completion
loop:
	DBAR
	MOVW	·Baton(SB), R12
	OR	R13, R13, R13	// hint that we're in a spin loop
	BNE	R12, loop
	DBAR

	// Restore stack
	ADDV	$(-1024*8), R3

	var n int
loop: // we want to have 3 preds here, the function entry and both gotos
	if v := try(); v == 42 || v == 1337 { // the two || are to trick findIndVar
		if n < 30 { // this aims to be the matched block
			if shouldInc() {
				n++
				goto loop
			}
			n = N(n) // try to prevent some block joining
			goto loop
		}
	}

  "tf_saved_model.session_initializer"() {initializers = [@NoOp]} : () -> ()
  func.func @NoOp() attributes {tf_saved_model.exported_names = ["__tf_saved_model_session_initializer_NoOp"]} {
    func.return
  }
// CHECK: func @NoOp() attributes {tf_saved_model.exported_names = ["__tf_saved_model_session_initializer_NoOp"]}

	ADDV	$(1024*8), R29

	// Ask signaller to setgid
	MOVW	$1, R1
	SYNC
	MOVW	R1, ·Baton(SB)
	SYNC

	// Wait for setgid completion
loop:
	SYNC
	MOVW	·Baton(SB), R1
	OR	R2, R2, R2	// hint that we're in a spin loop
	BNE	R1, loop
	SYNC

	// Restore stack
	ADDV	$(-1024*8), R29

  // The control dependency is for constants in the cond graph, and other ops
  // that do not depend on the loop variables. This ensures that these ops are
  // in the while loop frame (since they will indirectly depend on an Enter node
  // defining the frame) and that they are executed once per loop iteration.
  //
  // TODO(skyewm): the control dep will be added to all nodes in the cond graph.

	// divided by the type of error:
	// leak, repair, full, outOfRange, duplicate, invalid, unknown
	clusterIPRepairIPErrors = metrics.NewCounterVec(
		&metrics.CounterOpts{
			Namespace:      namespace,
			Subsystem:      subsystem,
			Name:           "ip_errors_total",

	ADD	$(1024*8), X2

	// Ask signaller to setgid
	MOV	$1, X5
	FENCE
	MOVW	X5, ·Baton(SB)
	FENCE

	// Wait for setgid completion
loop:
	FENCE
	MOVW	·Baton(SB), X5
	OR	X6, X6, X6	// hint that we're in a spin loop
	BNE	ZERO, X5, loop
	FENCE

	// Restore stack
	ADD	$(-1024*8), X2