}

// Given Value input_size, and known numbers filter_sz, dilation_rate, stride,
// calculate padding_low and padding_high for SAME padding.
void GetSamePaddingValues(OpBuilder &builder, Location loc, Value input_size,
                          int64_t filter_sz, int64_t dilation_rate,
                          int64_t stride, Value &padding_low,
                          Value &padding_high) {

                          XlaCompileOnDemandOp);                               \
  REGISTER_KERNEL_BUILDER(Name("XlaPad")                                       \
                              .HostMemory("padding_low")                       \
                              .HostMemory("padding_high")                      \
                              .HostMemory("padding_interior")                  \

      CompileTimeConstantOperand<TF::XlaDynamicSliceOp, 2>,  // $slice_indices
      CompileTimeConstantOperand<TF::XlaGatherOp, 2>,        // $slice_sizes
      // $padding_low, $padding_high, $padding_interior
      CompileTimeConstantOperand<TF::XlaPadOp, 2, 3, 4>,
      // $window_dimensions, $window_strides, $base_dilations,
      // $window_dilations, $padding

            loc, padding_needed.getType(), cond, padding_needed, zero);
        *padding_low =
            rewriter.create<arith::DivUIOp>(loc, padding_needed, two);
        *padding_high =
            rewriter.create<arith::SubIOp>(loc, padding_needed, *padding_low);
        break;
      }
    }
    return true;
  }

  LogicalResult matchAndRewriteDynamicConv(OpT op,

        operand_type.cloneWith(shape, rewriter.getI32Type());

    ArrayRef<int64_t> padding_low = op.getEdgePaddingLow();
    ArrayRef<int64_t> padding_high = op.getEdgePaddingHigh();
    SmallVector<int32_t> padding_value;
    for (int i = 0; i < rank; ++i) {
      padding_value.push_back(CastI64ToI32(padding_low[i]).value());
      padding_value.push_back(CastI64ToI32(padding_high[i]).value());
    }

    } else {
      const int64_t dilated_base = xla::window_util::DilatedBound(
          base_shape.getDimSize(i), dim.base_dilation());
      const int64_t padded_dilated_base =
          dim.padding_low() + dilated_base + dim.padding_high();
      const int64_t dilated_window =
          xla::window_util::DilatedBound(dim.size(), dim.window_dilation());

      output_dimensions[i] = xla::window_util::StridedBound(

    Arg<TF_I32OrI64Tensor, [{the padding to apply at the start of each input dimensions. Must
be a compile-time constant 1D tensor of length equal to rank of input.}]>:$padding_low,
    Arg<TF_I32OrI64Tensor, [{the padding to apply at the end of each input dimension. Must
be a compile-time constant 1D tensor of length equal to rank of input.}]>:$padding_high,

3746349612314,"touches":4,"min_t":1228851457,"max_t":1238022568,"mean_t":1233139606},{"name":"border_style-expected.png","kids":[],"cl_weight":0.006942015827872924,"touches":4,"min_t":1228851457,"max_t":1238022568,"mean_t":1233138547},{"name":"padding_top-expected.checksum","kids":[],"cl_weight":0.006942015827872924,"touches":4,"min_t":1228851457,"max_t":1238022568,"mean_t":1233138547},{"name":"margin_inline-expected.checksum","kids":[],"cl_weight":0.006942015827872924,"touches":4,"min_t":122885...