StackScanWork   int64
	GlobalsScanWork int64
}

func (c *GCController) Revise(d GCControllerReviseDelta) {
	c.heapLive.Add(d.HeapLive)
	c.heapScan.Add(d.HeapScan)
	c.heapScanWork.Add(d.HeapScanWork)
	c.stackScanWork.Add(d.StackScanWork)
	c.globalsScanWork.Add(d.GlobalsScanWork)
	c.revise()
}

func (c *GCController) EndCycle(bytesMarked uint64, assistTime, elapsed int64, gomaxprocs int) {

	/*509*/ uint16(xSetOp), uint16(ADD),
	/*511*/ uint16(xReadSlashR),
	/*512*/ uint16(xArgRM32),
	/*513*/ uint16(xArgR32),
	/*514*/ uint16(xMatch),
	/*515*/ uint16(xCondDataSize), 503, 509, 519,
	/*519*/ uint16(xSetOp), uint16(ADD),
	/*521*/ uint16(xReadSlashR),
	/*522*/ uint16(xArgRM64),
	/*523*/ uint16(xArgR64),
	/*524*/ uint16(xMatch),
	/*525*/ uint16(xSetOp), uint16(ADD),
	/*527*/ uint16(xReadSlashR),

}

var opToSSA = map[opAndType]ssa.Op{
	{ir.OADD, types.TINT8}:    ssa.OpAdd8,
	{ir.OADD, types.TUINT8}:   ssa.OpAdd8,
	{ir.OADD, types.TINT16}:   ssa.OpAdd16,
	{ir.OADD, types.TUINT16}:  ssa.OpAdd16,
	{ir.OADD, types.TINT32}:   ssa.OpAdd32,
	{ir.OADD, types.TUINT32}:  ssa.OpAdd32,
	{ir.OADD, types.TINT64}:   ssa.OpAdd64,
	{ir.OADD, types.TUINT64}:  ssa.OpAdd64,
	{ir.OADD, types.TFLOAT32}: ssa.OpAdd32F,

	if lockIndex >= 0 && inst.Prefix[lockIndex]&PrefixImplicit == 0 {
		switch inst.Op {
		// TODO(rsc): Perhaps this instruction class should be derived from the CSV.
		case ADD, ADC, AND, BTC, BTR, BTS, CMPXCHG, CMPXCHG8B, CMPXCHG16B, DEC, INC, NEG, NOT, OR, SBB, SUB, XOR, XADD, XCHG:
			if isMem(inst.Args[0]) {
				hasLock = true
				break
			}
			fallthrough
		default:
			inst.Prefix[lockIndex] |= PrefixInvalid
		}

# NONE:           %[[add:.*]], %[[add_control:.*]] = tf_executor.island wraps "tf.Add"(%[[ARG_0]], %[[ARG_1]])
# NONE:           fetch %[[add]]

# UNKNOWN-LABEL: func @main
# UNKNOWN-SAME:  (%[[ARG_0:[a-z0-9]+]]: tensor<*xi32>, %[[ARG_1:[a-z0-9]+]]: tensor<*xi32>) -> tensor<*xi32>
# UNKNOWN-SAME:  control_outputs = ""
# UNKNOWN-SAME:  inputs = "input0,input1"
# UNKNOWN-SAME:  outputs = "Add"

// CHECK-NEXT:    %[[ARG0]] = f32[] parameter(0)
// CHECK-NEXT:    %[[ARG1]] = f32[] parameter(1)
// CHECK-NEXT:    [[ADD:%.*]] = f32[] add(f32[] %[[ARG0]], f32[] %[[ARG1]])
// CHECK-NEXT:    ROOT %tuple.{{[0-9]+}} = (f32[]) tuple(f32[] [[ADD]])
// CHECK-NEXT:  }

// CHECK:       // InputMapping {0, 1}
// CHECK-NEXT:  // XlaInputShape f32[]
// CHECK-NEXT:  // XlaInputShape f32[]

# RUN: tf_tfl_translate -tf-input-arrays=input0,input1 -tf-input-shapes=4:4 -tf-input-data-types=DT_INT32,DT_INT32 -tf-output-arrays=Add %s -o - | flatbuffer_to_string - | FileCheck %s

# Add two tensor<4xi32> inputs and return the result

node {
  name: "Add"
  op: "Add"
  input: "input0"
  input: "input1"
  attr {
    key: "T"
    value {
      type: DT_INT32
    }
  }
}
node {
  name: "input0"
  op: "Placeholder"
  attr {

	VADD	V18.S4, V4.S4, V16.S4
	HASHUPDATEPARITY

	VADD	V18.S4, V5.S4, V16.S4
	HASHUPDATEPARITY

	VADD	V18.S4, V6.S4, V16.S4
	HASHUPDATEPARITY

	VADD	V18.S4, V7.S4, V16.S4
	HASHUPDATEPARITY

	SUB	$64, R3, R3                                  // message length - 64bytes, then compare with 64bytes
	VADD	V2.S4, V0.S4, V0.S4
	VADD	V1.S4, V20.S4, V20.S4
	CBNZ	R3, blockloop

sha1ret:

	WORD	$0x4D40E8DD

	VADD	V20.S4, V4.S4, V4.S4
	VADD	V21.S4, V5.S4, V5.S4
	VADD	V22.S4, V6.S4, V6.S4
	VADD	V23.S4, V7.S4, V7.S4
	VADD	V24.S4, V8.S4, V8.S4
	VADD	V25.S4, V9.S4, V9.S4
	VADD	V26.S4, V10.S4, V10.S4
	VADD	V27.S4, V11.S4, V11.S4
	VADD	V28.S4, V12.S4, V12.S4
	VADD	V29.S4, V13.S4, V13.S4
	VADD	V30.S4, V14.S4, V14.S4
	VADD	V31.S4, V15.S4, V15.S4

	VZIP1	V1.S4, V0.S4, V16.S4

	SHA512ROUND_LAST(V2, V3, V1, V4, V0, V25, V17)
	SHA512ROUND_LAST(V4, V2, V0, V1, V3, V26, V18)
	SHA512ROUND_LAST(V1, V4, V3, V0, V2, V27, V19)

	// add result to digest
	VADD	V0.D2, V8.D2, V8.D2
	VADD	V1.D2, V9.D2, V9.D2
	VADD	V2.D2, V10.D2, V10.D2
	VADD	V3.D2, V11.D2, V11.D2
	SUB	$128, R2
	CBNZ	R2, loop

	VST1	[V8.D2, V9.D2, V10.D2, V11.D2], (R0)