# MLIR:         %[[conv:.*]] = "tfl.conv_2d"(%[[ARG_0]], %[[weight]], %[[bias]]) <{dilation_h_factor = 1 : i32, dilation_w_factor = 1 : i32, fused_activation_function = "NONE", padding = "SAME", stride_h = 1 : i32, stride_w = 1 : i32}
# MLIR:         %[[reshape:.*]] = "tfl.reshape"(%[[conv]], %[[shape]]) : (tensor<1x1x1x186x!quant.uniform<i8:f32, 0.09363494573854933:22>>, tensor<3xi32>)

# MLIR:         %[[conv:.*]] = "tfl.conv_2d"(%[[ARG_0]], %[[weight]], %[[bias]]) <{dilation_h_factor = 1 : i32, dilation_w_factor = 1 : i32, fused_activation_function = "NONE", padding = "SAME", stride_h = 1 : i32, stride_w = 1 : i32}
# MLIR:         %[[reshape:.*]] = "tfl.reshape"(%[[conv]], %[[shape]]) : (tensor<1x1x1x186x!quant.uniform<i8:f32, 0.09363494573854933:22>>, tensor<3xi32>)

        "MLIR dump file name.">,
    Option<"merge_fusion_with_dequantize_",
        "merge-fusion-with-dequantize",
        "bool", /*default=*/"false",
        "Whether to merge quantized conv/dot_general fusion with subsequent dequantize.">,
  ];
  let dependentDialects = [
    "mlir::arith::ArithDialect",
    "mlir::stablehlo::StablehloDialect",
    "mlir::quant::QuantizationDialect",

    %1 = "tf.PartitionedCall"(%arg0, %cst, %cst_0) {_tfl_quant_trait = "fully_quantizable", config = "", config_proto = "", executor_type = "", f = @composite_conv2d_with_bias_and_relu6_fn_1} : (tensor<1x2x2x3xf32>, tensor<2x2x3x2xf32>, tensor<2xf32>) -> tensor<*xf32> loc(callsite("test@conv"("Conv2D_1") at "QuantizationUnit(\12\08Conv2D_1\1a\04conv)"))

  // The result of the conv must be the first input to BiasAdd to be fusable.
  %1 = "tf.BiasAdd"(%arg0, %0) {data_format = "NHWC"} : (tensor<*xf32>, tensor<*xf32>) -> tensor<*xf32>
  %2 = "tf.Elu"(%1) : (tensor<*xf32>) -> tensor<*xf32>

//	LMOVW ximm ',' rreg
//	{
//		outcode(int($1), &$2, 0, &$4);
//	}
	MOVW	$1, R1
	MOVW	$foo(SB), R1
	MOVV	$1, R1
	MOVV	$foo(SB), R1

//
// floating point operate
//
//	LFCONV freg ',' freg
//	{
//		outcode(int($1), &$2, 0, &$4);
//	}
	ABSD	F1, F2

//	LFADD freg ',' freg
//	{
//		outcode(int($1), &$2, 0, &$4);
//	}
	ADDD	F1, F2

# MLIR:         %[[conv:.*]] = "tfl.conv_2d"(%[[ARG_0]], %[[weight]], %[[bias]]) <{dilation_h_factor = 1 : i32, dilation_w_factor = 1 : i32, fused_activation_function = "NONE", padding = "SAME", stride_h = 1 : i32, stride_w = 1 : i32}
# MLIR:         %[[reshape:.*]] = "tfl.reshape"(%[[conv]], %[[shape]]) : (tensor<1x1x1x186x!quant.uniform<i8:f32, 0.09363494573854933:22>>, tensor<3xi32>)

// these tests pass with `-gcflags=-l`.
func float32bits(f float32) uint32 { // ERROR "can inline float32bits"
	return *(*uint32)(unsafe.Pointer(&f))
}

// Ensure OCONVNOP is zero cost.
func Conv(v uint64) uint64 { // ERROR "can inline Conv"
	return conv2(conv2(conv2(v))) // ERROR "inlining call to (conv1|conv2)"
}
func conv2(v uint64) uint64 { // ERROR "can inline conv2"

# MLIR:         %[[conv:.*]] = "tfl.conv_2d"(%[[ARG_0]], %[[weight]], %[[bias]]) <{dilation_h_factor = 1 : i32, dilation_w_factor = 1 : i32, fused_activation_function = "NONE", padding = "SAME", stride_h = 1 : i32, stride_w = 1 : i32}
# MLIR:         %[[reshape:.*]] = "tfl.reshape"(%[[conv]], %[[shape]]) : (tensor<1x1x1x186x!quant.uniform<i8:f32, 0.09363494573854933:22>>, tensor<3xi32>)

		// pick off before type-checking arguments
		arg, ok := needOneArg(n, "conversion to %v", l.Type())
		if !ok {
			n.SetType(nil)
			return n
		}

		n := ir.NewConvExpr(n.Pos(), ir.OCONV, nil, arg)
		n.SetType(l.Type())
		return tcConv(n)
	}

	RewriteNonNameCall(n)
	typecheckargs(n)
	t := l.Type()
	if t == nil {
		n.SetType(nil)
		return n
	}
	types.CheckSize(t)

	switch l.Op() {