input_tensor,
            self.filters,
            strides=strides,
            dilations=dilations,
            padding=padding,
            data_format='NHWC',
            name='sample/conv',
        )
        if bias_fn is not None:
          out = nn_ops.bias_add(out, self.bias)
        if has_batch_norm:
          # Fusing is supported for non-training case.

#include "tensorflow/compiler/mlir/tensorflow/ir/tf_ops.h"
#include "xla/xla_data.pb.h"

namespace mlir::quant {
namespace {

constexpr StringRef kTfQuantCreatedEinsum = "__tf_quant_created_einsum";

// Replaces mixed-type Conv and Matmul cast hacks with TF XLA ops.
// TODO(b/228403741): Support conversion for dynamic-shaped TF ops.
class ReplaceCastHacksWithTFXLAOpsPass
    : public PassWrapper<ReplaceCastHacksWithTFXLAOpsPass,

                                                Operation* target_op,
                                                int operand_index);

  // Adjusts bias scale that is derived from other scales (fc, conv ops) to
  // prevent overflow of quantized bias values. This also changes quantization
  // state of other inputs when needed.
  bool SetBiasParamsWithAdjustments(Operation* op, int bias_index,

			return true
		}
		// We may have skipped past one or more OCONVNOPs, so
		// use conv to ensure r is assignable to l (#13263).
		dst := ir.Node(l)
		if loff != 0 || !types.Identical(typ, l.Type()) {
			dst = ir.NewNameOffsetExpr(base.Pos, l, loff, typ)
		}
		s.append(ir.NewAssignStmt(base.Pos, dst, typecheck.Conv(r, typ)))
		return true

	case ir.ONIL:
		return true

	case ir.OLITERAL:

    Option<"is_signed_", "is-signed", "bool", "false",
               "Is the corresponding integer signed">,
  ];
}

def IdentifyDilatedConvPass : Pass<"tfl-identify-dilated-conv", "mlir::func::FuncOp"> {
  let summary = "Convert dense tensor to sparse format.";
  let constructor = "CreateIdentifyDilatedConvPass()";
  let dependentDialects = ["TFL::TensorFlowLiteDialect"];
}

	for _, svc := range services {
		hostname := kube.ServiceHostname(svc.Name, svc.Namespace, c.opts.DomainSuffix)
		c.Lock()
		conv, f := c.servicesMap[hostname]
		c.Unlock()
		if !f {
			return
		}
		shard := model.ShardKeyFromRegistry(c)
		endpoints := c.buildEndpointsForService(conv, true)
		if len(endpoints) > 0 {
			c.opts.XDSUpdater.EDSCacheUpdate(shard, string(hostname), svc.Namespace, endpoints)
		}

	{name: "recv", format: "package main; var x = «<-»x"},
	{name: "call", format: "package main; var x = «f(«1»)»", parseMultiplier: 2},    // Parser nodes: Ident, CallExpr
	{name: "conv", format: "package main; var x = «(*T)(«1»)»", parseMultiplier: 2}, // Parser nodes: ParenExpr, CallExpr
	{name: "label", format: "package main; func main() { «Label:» }"},

  // For now, restrict scale adjustment to ops with affine quantized weights,
  // and having weights and biases as constants. This currently only applies to
  // FC and Conv* ops. Restriction for the weight can be relaxed if there are
  // needs for adjusting scale of variable weights.
  auto affine_op = dyn_cast<AffineQuantizedOpInterface>(op);

			return false
		}
	}

	// DefaultLit is necessary for non-constants too: n might be 1.1<<k.
	n = DefaultLit(n, types.Types[types.TINT])
	*np = n

	return true
}

func Conv(n ir.Node, t *types.Type) ir.Node {
	if types.IdenticalStrict(n.Type(), t) {
		return n
	}
	n = ir.NewConvExpr(base.Pos, ir.OCONV, nil, n)
	n.SetType(t)
	n = Expr(n)
	return n
}