Home | History | Annotate | Download | only in kernels
      1 /* Copyright 2015 The TensorFlow Authors. All Rights Reserved.
      2 
      3 Licensed under the Apache License, Version 2.0 (the "License");
      4 you may not use this file except in compliance with the License.
      5 You may obtain a copy of the License at
      6 
      7     http://www.apache.org/licenses/LICENSE-2.0
      8 
      9 Unless required by applicable law or agreed to in writing, software
     10 distributed under the License is distributed on an "AS IS" BASIS,
     11 WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
     12 See the License for the specific language governing permissions and
     13 limitations under the License.
     14 ==============================================================================*/
     15 
     16 // See docs in ../ops/parse_ops.cc.
     17 
     18 #include <algorithm>
     19 #include "tensorflow/core/framework/op_kernel.h"
     20 #include "tensorflow/core/framework/tensor.h"
     21 #include "tensorflow/core/framework/tensor_shape.h"
     22 #include "tensorflow/core/framework/types.h"
     23 #include "tensorflow/core/lib/core/errors.h"
     24 #include "tensorflow/core/lib/io/zlib_compression_options.h"
     25 #include "tensorflow/core/lib/io/zlib_inputstream.h"
     26 
     27 namespace tensorflow {
     28 namespace {
     29 // Wrap memory buffer into InputStreamInterface
     30 class MemoryInputStream : public io::InputStreamInterface {
     31  public:
     32   explicit MemoryInputStream(const char* buffer, size_t length)
     33       : buf_(buffer), len_(length), pos_(0) {}
     34 
     35   ~MemoryInputStream() override {}
     36 
     37   Status ReadNBytes(int64 bytes_to_read, string* result) override {
     38     result->clear();
     39     if (bytes_to_read < 0) {
     40       return errors::InvalidArgument("Can't read a negative number of bytes: ",
     41                                      bytes_to_read);
     42     }
     43     int64 bytes = bytes_to_read;
     44     Status s = Status::OK();
     45     if (pos_ + bytes_to_read > len_) {
     46       bytes = len_ - pos_;
     47       s = errors::OutOfRange("reached end of file");
     48     }
     49     if (bytes > 0) {
     50       result->resize(bytes);
     51       memcpy(&(*result)[0], &buf_[pos_], bytes);
     52       pos_ += bytes;
     53     }
     54     return s;
     55   }
     56 
     57   int64 Tell() const override { return pos_; }
     58 
     59   Status Reset() override {
     60     pos_ = 0;
     61     return Status::OK();
     62   }
     63 
     64  private:
     65   const char* buf_;  // Not owned.
     66   int64 len_;
     67   int64 pos_ = 0;  // Tracks where we are in the file.
     68 };
     69 }  // namespace
     70 
     71 class DecodeCompressedOp : public OpKernel {
     72  public:
     73   explicit DecodeCompressedOp(OpKernelConstruction* context)
     74       : OpKernel(context) {
     75     OP_REQUIRES_OK(context,
     76                    context->GetAttr("compression_type", &compression_type_));
     77     OP_REQUIRES(context,
     78                 (compression_type_.empty() || compression_type_ == "ZLIB" ||
     79                  compression_type_ == "GZIP"),
     80                 errors::InvalidArgument(
     81                     "Only ZLIB, GZIP or NONE are supported compressions"));
     82   }
     83 
     84   void Compute(OpKernelContext* context) override {
     85     const Tensor* bytes_tensor;
     86     OP_REQUIRES_OK(context, context->input("bytes", &bytes_tensor));
     87     const auto& bytes_flat = bytes_tensor->flat<string>();
     88 
     89     Tensor* output_tensor = nullptr;
     90     OP_REQUIRES_OK(context,
     91                    context->allocate_output("output", bytes_tensor->shape(),
     92                                             &output_tensor));
     93     auto output_flat = output_tensor->flat<string>();
     94     if (compression_type_.empty()) {
     95       for (int64 i = 0; i < bytes_flat.size(); i++) {
     96         output_flat(i) = bytes_flat(i);
     97       }
     98     } else {
     99       const io::ZlibCompressionOptions zlib_options =
    100           compression_type_ == "ZLIB" ? io::ZlibCompressionOptions::DEFAULT()
    101                                       : io::ZlibCompressionOptions::GZIP();
    102       for (int64 i = 0; i < bytes_flat.size(); i++) {
    103         std::unique_ptr<MemoryInputStream> input_stream(
    104             new MemoryInputStream(bytes_flat(i).data(), bytes_flat(i).size()));
    105         std::unique_ptr<io::ZlibInputStream> zlib_stream(
    106             new io::ZlibInputStream(
    107                 input_stream.get(), static_cast<size_t>(kBufferSize),
    108                 static_cast<size_t>(kBufferSize), zlib_options));
    109         string output_string;
    110         Status s = zlib_stream->ReadNBytes(INT_MAX, &output_string);
    111         OP_REQUIRES(context, (s.ok() || errors::IsOutOfRange(s)), s);
    112         output_flat(i) = output_string;
    113       }
    114     }
    115   }
    116 
    117  private:
    118   enum { kBufferSize = 256 << 10 /* 256 kB */ };
    119   string compression_type_;
    120 };
    121 
    122 REGISTER_KERNEL_BUILDER(Name("DecodeCompressed").Device(DEVICE_CPU),
    123                         DecodeCompressedOp)
    124 
    125 }  // namespace tensorflow
    126