// CHECK: %[[MUL:.*]] = tfl.mul %arg0, %[[BROADCAST_TO]] {fused_activation_function = "NONE"} : tensor<8x7x6x5x4x3x2x1xf32>
  // CHECK: return %[[MUL]] : tensor<8x7x6x5x4x3x2x1xf32>
}

func.func @mul_with_high_dims_dynamic_shape_both_sides(%arg0: tensor<8x7x6x5x?x3x2x1xf32>, %arg1: tensor<?x3x2x1xf32>) -> tensor<8x7x6x5x?x3x2x1xf32> {

  // CHECK: tfl.mul %arg0, %arg1 {fused_activation_function = "RELU6"}
  %0 = tfl.mul %arg0, %arg1 {fused_activation_function = "RELU6"} : tensor<? x i16>
  func.return %0#0 : tensor<? x i16>
}

// CHECK-LABEL: testMulComplex
func.func @testMulComplex(tensor<? x complex<f32>>, tensor<? x complex<f32>>) -> tensor<? x complex<f32>> {
^bb0(%arg0: tensor<? x complex<f32>>, %arg1: tensor<? x complex<f32>>):

// CHECK{LITERAL}: %cst_1 = arith.constant dense<[[[[1.000000e+00], [2.000000e+00]], [[2.000000e+00], [4.000000e+00]]]]> : tensor<1x2x2x1xf32>
// CHECK: %3 = tfl.mul %2, %cst_1 {fused_activation_function = "NONE"} : tensor<1x2x2x1xf32>
// CHECK: %cst_2 = arith.constant dense<[0, 3, 1, 2]> : tensor<4xi32>

(Div16u x y) => (DIVWU (ZeroExt16to32 x) (ZeroExt16to32 y))
(Div8 x y) => (DIVW  (SignExt8to32 x) (SignExt8to32 y))
(Div8u x y) => (DIVWU (ZeroExt8to32 x) (ZeroExt8to32 y))

(Hmul(64|64u|32|32u) ...) => (MULH(D|DU|W|WU) ...)

(Mul(32|64)F ...) => ((FMULS|FMUL) ...)

(Div(32|64)F ...) => ((FDIVS|FDIV) ...)

// Lowering float <=> int
(Cvt32to(32|64)F x) => ((FCFIDS|FCFID) (MTVSRD (SignExt32to64 x)))

	{ir.OMUL, types.TINT8}:    ssa.OpMul8,
	{ir.OMUL, types.TUINT8}:   ssa.OpMul8,
	{ir.OMUL, types.TINT16}:   ssa.OpMul16,
	{ir.OMUL, types.TUINT16}:  ssa.OpMul16,
	{ir.OMUL, types.TINT32}:   ssa.OpMul32,
	{ir.OMUL, types.TUINT32}:  ssa.OpMul32,
	{ir.OMUL, types.TINT64}:   ssa.OpMul64,
	{ir.OMUL, types.TUINT64}:  ssa.OpMul64,
	{ir.OMUL, types.TFLOAT32}: ssa.OpMul32F,

		default:
			return ast.NewIdent(t.Name())
		}
	case *types.Pointer:
		x := TypeExpr(f, pkg, t.Elem())
		if x == nil {
			return nil
		}
		return &ast.UnaryExpr{
			Op: token.MUL,
			X:  x,
		}
	case *types.Array:
		elt := TypeExpr(f, pkg, t.Elem())
		if elt == nil {
			return nil
		}
		return &ast.ArrayType{
			Len: &ast.BasicLit{
				Kind:  token.INT,

	}

	// ±0 - y
	// x - ±Inf
	return z.Neg(y)
}

// Mul sets z to the rounded product x*y and returns z.
// Precision, rounding, and accuracy reporting are as for [Float.Add].
// Mul panics with [ErrNaN] if one operand is zero and the other
// operand an infinity. The value of z is undefined in that case.
func (z *Float) Mul(x, y *Float) *Float {
	if debugFloat {
		x.validate()
		y.validate()
	}

class OperandsSameElementTypeConstraintBase<string op> :
  PredOpTrait<op # " operands have same element type",
    OperandsSameElementTypeConstraintBasePred>;

// This is a constraint for most of the binary ops, e.g., add, mul, div, etc.
// Binary ops lhs & rhs should have the same value type, and is capable to
// compare quantization types as well.
def BinaryOpSameElementTypeConstraint :
  OperandsSameElementTypeConstraintBase<"binary op">;

    *   Unary GPU kernels: Abs, Atanh, Acos, Acosh, Asin, Asinh, Atan, Cos, Cosh, Sin, Sinh, Tan, Tanh.
    *   Binary GPU kernels: AddV2, Sub, Div, DivNoNan, Mul, MulNoNan, FloorDiv, Equal, NotEqual, Greater, GreaterEqual, LessEqual, Less.

* `tf.lite`
    * Add experimental supports conversion of models that may be larger than 2GB before buffer deduplication

		v.AuxInt = float64ToAuxInt(math.Abs(x))
		return true
	}
	return false
}
func rewriteValuePPC64_OpPPC64FADD(v *Value) bool {
	v_1 := v.Args[1]
	v_0 := v.Args[0]
	// match: (FADD (FMUL x y) z)
	// cond: x.Block.Func.useFMA(v)
	// result: (FMADD x y z)
	for {
		for _i0 := 0; _i0 <= 1; _i0, v_0, v_1 = _i0+1, v_1, v_0 {
			if v_0.Op != OpPPC64FMUL {
				continue
			}