// Package loopclosure defines an Analyzer that checks for references to
// enclosing loop variables from within nested functions.
//
// # Analyzer loopclosure
//
// loopclosure: check references to loop variables from within nested functions
//
// This analyzer reports places where a function literal references the
// iteration variable of an enclosing loop, and the loop calls the function
// in such a way (e.g. with go or defer) that it may outlive the loop

it does not call the loop body again after it has returned false or
after the entire loop has exited (it might retain a copy of the body
function, or pass it to another goroutine) -- each generated loop has
its own #stateK variable that is used to check for permitted call
patterns to the yield function for a loop body.

The state values are:

abi.RF_DONE = 0      // body of loop has exited in a non-panic way

        //   ApplyWeightUpdates_1 -> Computation_B
        //   ApplyWeightUpdates_2 -> Computation_C
        //   Computation_A -> NoOp
        //   Computation_B -> NoOp
        //   Computation_C -> NoOp
        //   "iteration++" -> NoOp
        // }
        //
        // In the graph above we can't cluster iteration++ with any of the

		for i := 0; i < x; i++ { // ERROR "Branch prediction rule stay in loop"
			if x == 4 { // ERROR "Branch prediction rule stay in loop"
				return a
			}
			for j := 0; j < y; j++ { // ERROR "Branch prediction rule stay in loop"
				for k := 0; k < z; k++ { // ERROR "Branch prediction rule stay in loop"
					a -= j * i
				}
				a += j
			}
		}
	}
	return a
}

// Checks that 2 `NoOp`s having control dependencies to each of the initializer
// functions are created.
// CHECK-DAG: %[[CTL_2:.*]] = tf_executor.island(%[[CTL_0]]) wraps "tf.NoOp"()
// CHECK-DAG: %[[CTL_3:.*]] = tf_executor.island(%[[CTL_1]]) wraps "tf.NoOp"()

// CHECK: tf_executor.fetch
// CHECK-SAME: !tf_executor.control, !tf_executor.control

	ADD	X10, X11	// end

	PCALIGN	$16
loop:
	BEQ	X10, X11, done
	MOVBU	(X10), X15
	ADD	$1, X10
	BNE	X12, X15, loop
	ADD	$1, X14
	JMP	loop

done:
	MOV	X14, X10
	RET

TEXT ·CountString<ABIInternal>(SB),NOSPLIT,$0-32
	// X10 = s_base
	// X11 = s_len
	// X12 = byte to count
	AND	$0xff, X12
	MOV	ZERO, X14	// count
	ADD	X10, X11	// end

	PCALIGN	$16
loop:
	BEQ	X10, X11, done
	MOVBU	(X10), X15

				}
				i++
			}
			if i != 4 {
				t.Fatalf("should loop four times")
			}
		}},
		{"int8", ValueOf(int8(4)), func(t *testing.T, s iter.Seq[Value]) {
			i := int8(0)
			for v := range s {
				if v.Interface().(int8) != i {
					t.Fatalf("got %d, want %d", v.Int(), i)
				}
				i++
			}
			if i != 4 {
				t.Fatalf("should loop four times")
			}
		}},

// Utility functions for constructing while loop gradients

namespace tensorflow {

// Adds the gradient computation for the while loop associated with
// `while_ctx`. `grad_inputs` are the partial derivatives w.r.t. the loop
// outputs, i.e. the exit nodes.  The partial derivatives w.r.t. the loop
// inputs, i.e. the input loop vars, are returned in `grad_outputs`.

	// X11 = b_len
	// X12 = b_cap (unused)
	// X13 = byte to find
	AND	$0xff, X13
	MOV	X10, X12		// store base for later
	ADD	X10, X11		// end
	SUB	$1, X10

loop:
	ADD	$1, X10
	BEQ	X10, X11, notfound
	MOVBU	(X10), X14
	BNE	X13, X14, loop

	SUB	X12, X10		// remove base
	RET

notfound:
	MOV	$-1, X10
	RET

TEXT ·IndexByteString<ABIInternal>(SB),NOSPLIT,$0-32
	// X10 = b_base
	// X11 = b_len

	ADCS	$0, R4
	MOVD.P	R4, 8(R3)
	SUB	$1, R0
two:				// do it twice
	TBZ	$1, R0, loop
	LDP.P	16(R1), (R4, R5)
	ADCS	$0, R4, R8	// c, z[i] = x[i] + c
	ADCS	$0, R5, R9
	STP.P	(R8, R9), 16(R3)
	SUB	$2, R0
loop:				// do four times per round
	vwOneIter(ADCS, R0, len1)
	B	loop
len1:
	CSET	HS, R2		// extract carry flag
len0:
	MOVD	R2, c+56(FP)
done:
	RET
large: