warpctc_op.h 15.2 KB
Newer Older
1
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved.
Y
Yiqun Liu 已提交
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License. */

#pragma once

17
#include <vector>
Y
Yi Wang 已提交
18 19 20 21 22
#include "paddle/fluid/framework/op_registry.h"
#include "paddle/fluid/operators/math/math_function.h"
#include "paddle/fluid/operators/math/sequence_padding.h"
#include "paddle/fluid/operators/math/sequence_scale.h"
#include "paddle/fluid/platform/dynload/warpctc.h"
Y
Yiqun Liu 已提交
23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67

namespace paddle {
namespace operators {

using Tensor = framework::Tensor;
using LoDTensor = framework::LoDTensor;

template <typename DeviceContext>
class WarpCTCFunctor {
 public:
  /*
   * \brief Compute the connectionist temporal classification loss,
   *        and optionally compute the gradient with respect to the inputs.
   *
   * If gradient is nullptr, it only computes the ctc loss,
   * or computes both ctc loss and gradient.
   *
   * \param ctx               execution context of this functor
   * \param input             batch matrix of input probabilities, in
   *                          max_sequence_length x num_sequences x
   *                          sequence_width, (row-major) format
   * \param gradient          batch matrix of gradient, with the same shape as
   *                          input.
   * \param cpu_labels        labels always in CPU memory.
   * \param cpu_label_lengths length of all labels in CPU memory.
   * \param cpu_input_lengths length of all sequences in CPU memory.
   * \param sequence_width    number of possible output symbols.
   * \param num_sequences     number of sequence.
   * \param blank             blank label used in ctc loss function.
   * \param cpu_losss         cost of each sequence in CPU memory.
   */
  void operator()(const framework::ExecutionContext& ctx, const float* input,
                  float* gradient, const int* cpu_labels,
                  const int* cpu_label_lengths, const int* cpu_input_lengths,
                  const size_t sequence_width, const size_t num_sequences,
                  const size_t blank, float* cpu_loss) {
    // Init warp-ctc options
    init(ctx, blank);

    // Compute the required workspace size.
    // There is no memory allocated operations within warp-ctc.
    size_t workspace_bytes = 0;
    ctcStatus_t status = platform::dynload::get_workspace_size(
        cpu_label_lengths, cpu_input_lengths, static_cast<int>(sequence_width),
        static_cast<int>(num_sequences), options_, &workspace_bytes);
68 69 70 71 72 73 74 75 76 77 78 79

    PADDLE_ENFORCE_EQ(
        CTC_STATUS_SUCCESS, status,
        platform::errors::PreconditionNotMet(
            "warp-ctc [version %d] Error in get_workspace_size: %s",
            warpctc_version_, platform::dynload::ctcGetStatusString(status)));
    PADDLE_ENFORCE_GT(
        workspace_bytes, 0UL,
        platform::errors::InvalidArgument(
            "Bytes of workspace got by warp-ctc function, "
            "get_workspace_size() should be larger than 0, but received %d",
            workspace_bytes));
Y
Yiqun Liu 已提交
80

L
Li Fuchen 已提交
81
    auto& dev_ctx = ctx.template device_context<DeviceContext>();
Y
Yiqun Liu 已提交
82
    size_t workspace_elements = workspace_bytes / sizeof(float) + 1UL;
L
Li Fuchen 已提交
83
    Tensor workspace = ctx.AllocateTmpTensor<float, DeviceContext>(
Y
Yiqun Liu 已提交
84
        framework::make_ddim({static_cast<int64_t>(workspace_elements)}),
L
Li Fuchen 已提交
85 86
        dev_ctx);
    float* workspace_data = workspace.data<float>();
Y
Yiqun Liu 已提交
87 88 89 90 91 92 93 94 95
    math::SetConstant<DeviceContext, float>()(
        ctx.template device_context<DeviceContext>(), &workspace,
        static_cast<float>(0));

    // compute loss and gradient
    status = platform::dynload::compute_ctc_loss(
        input, gradient, cpu_labels, cpu_label_lengths, cpu_input_lengths,
        static_cast<int>(sequence_width), static_cast<int>(num_sequences),
        cpu_loss, workspace_data, options_);
96 97 98 99 100 101

    PADDLE_ENFORCE_EQ(
        CTC_STATUS_SUCCESS, status,
        platform::errors::PreconditionNotMet(
            "warp-ctc [version %d] Error in get_workspace_size: %s",
            warpctc_version_, platform::dynload::ctcGetStatusString(status)));
Y
Yiqun Liu 已提交
102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138
  }

 protected:
  void init(const framework::ExecutionContext& ctx, const size_t blank) {
    warpctc_version_ = platform::dynload::get_warpctc_version();

    if (platform::is_gpu_place(ctx.GetPlace())) {
#ifdef PADDLE_WITH_CUDA
      options_.loc = CTC_GPU;
      options_.stream = reinterpret_cast<const platform::CUDADeviceContext&>(
                            ctx.device_context())
                            .stream();
#else
      PADDLE_THROW("[warpctc init] GPU is not enabled.");
#endif
    } else {
      options_.loc = CTC_CPU;
      options_.num_threads = 1;
    }

    options_.blank_label = blank;
  }

 private:
  int warpctc_version_;
  ctcOptions options_;
};

template <typename DeviceContext, typename T>
class WarpCTCKernel : public framework::OpKernel<T> {
 public:
  void Compute(const framework::ExecutionContext& ctx) const override {
    auto* logits = ctx.Input<LoDTensor>("Logits");
    auto* label = ctx.Input<LoDTensor>("Label");
    auto* warpctc_grad = ctx.Output<Tensor>("WarpCTCGrad");
    auto* loss = ctx.Output<Tensor>("Loss");

139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158
    size_t num_sequences, sequence_width, max_sequence_length;
    framework::Vector<size_t> logits_lod;
    framework::Vector<size_t> label_lod;

    if (ctx.HasInput("LogitsLength") && ctx.HasInput("LabelLength")) {
      num_sequences = logits->dims()[1];
      sequence_width = logits->dims()[2];
      max_sequence_length = logits->dims()[0];

      auto* logits_length = ctx.Input<framework::Tensor>("LogitsLength");
      auto* labels_length = ctx.Input<framework::Tensor>("LabelLength");
      framework::Tensor logits_length_cpu;
      framework::Tensor labels_length_cpu;
      framework::TensorCopy(*logits_length, platform::CPUPlace(),
                            &logits_length_cpu);
      framework::TensorCopy(*labels_length, platform::CPUPlace(),
                            &labels_length_cpu);

      logits_lod.push_back(0);
      label_lod.push_back(0);
159
      for (size_t i = 0; i < num_sequences; i++) {
160 161 162 163 164 165
        logits_lod.push_back(logits_lod[i] +
                             logits_length_cpu.data<int64_t>()[i]);
        label_lod.push_back(label_lod[i] +
                            labels_length_cpu.data<int64_t>()[i]);
      }
    } else {
166 167 168 169 170 171 172 173 174
      PADDLE_ENFORCE_GT(logits->NumLevels(), 0UL,
                        platform::errors::InvalidArgument(
                            "Input(Logits) Tensor of WarpCTC "
                            "does not contain LoD information."));
      PADDLE_ENFORCE_GT(label->NumLevels(), 0UL,
                        platform::errors::InvalidArgument(
                            "Input(Label) Tensor of WarpCTC "
                            "does not contain LoD information."));

175 176
      logits_lod = framework::ToAbsOffset(logits->lod())[0];
      auto logits_dims = logits->dims();
177

178 179
      PADDLE_ENFORCE_EQ(
          logits_dims[0], static_cast<int64_t>(logits_lod.back()),
180 181 182 183
          platform::errors::InvalidArgument(
              "The first dimension of Input(Logits) should be equal to "
              "the sum of all sequences' lengths = %d., but received %d. ",
              static_cast<int64_t>(logits_lod.back()), logits_dims[0]));
184 185 186

      label_lod = framework::ToAbsOffset(label->lod())[0];
      auto label_dims = label->dims();
187 188 189 190 191
      PADDLE_ENFORCE_EQ(label_dims[1], 1,
                        platform::errors::InvalidArgument(
                            "The last dimension of Input(Label) should be 1, "
                            "but received %d",
                            label_dims[1]));
192 193

      num_sequences = logits_lod.size() - 1;
194 195 196 197 198 199
      PADDLE_ENFORCE_EQ(
          num_sequences, label_lod.size() - 1,
          platform::errors::InvalidArgument(
              "The number of sequences of Input(Logits) should be "
              "equal to that of Input(Label) = %d, but received %d",
              label_lod.size() - 1, num_sequences));
200 201 202 203 204

      sequence_width = logits->numel() / logits_dims[0];
      max_sequence_length = math::MaximumSequenceLength(logits_lod);
    }

Y
Yiqun Liu 已提交
205 206 207 208
    auto loss_dims =
        framework::make_ddim({static_cast<int64_t>(num_sequences), 1});

    // warpctc needs sequences data stored in transposed padding format
F
fengjiayi 已提交
209
    LoDTensor warpctc_logits;
Y
Yiqun Liu 已提交
210 211 212 213
    auto warpctc_logits_dims =
        framework::make_ddim({static_cast<int64_t>(max_sequence_length),
                              static_cast<int64_t>(num_sequences),
                              static_cast<int64_t>(sequence_width)});
L
Li Fuchen 已提交
214 215 216 217
    auto& dev_ctx = ctx.template device_context<DeviceContext>();
    Tensor warpctc_logits_tmp =
        ctx.AllocateTmpTensor<T, DeviceContext>(warpctc_logits_dims, dev_ctx);
    warpctc_logits.ShareDataWith(warpctc_logits_tmp);
218 219
    if (ctx.HasInput("LogitsLength")) {
      TensorCopySync(*logits, ctx.GetPlace(), &warpctc_logits);
F
fengjiayi 已提交
220
    } else {
221 222 223 224 225 226 227 228 229 230 231 232 233 234 235
      LoDTensor cpu_pad_value;
      T* pad_value_data =
          cpu_pad_value.mutable_data<T>({1}, platform::CPUPlace());
      *pad_value_data = static_cast<T>(0);
      LoDTensor pad_value;
      if (platform::is_cpu_place(ctx.GetPlace())) {
        pad_value = cpu_pad_value;
      } else {
        TensorCopySync(cpu_pad_value, ctx.GetPlace(), &pad_value);
      }

      math::PaddingLoDTensorFunctor<DeviceContext, T>()(
          ctx.template device_context<DeviceContext>(), *logits,
          &warpctc_logits, pad_value, -1, 0, false /* norm_by_times */,
          math::kLengthBatchWidth);
F
fengjiayi 已提交
236
    }
Y
Yiqun Liu 已提交
237 238 239 240 241 242
    const T* warpctc_logits_data = warpctc_logits.data<T>();

    std::vector<int> warpctc_label_lengths(num_sequences);
    std::vector<int> warpctc_logits_lengths(num_sequences);

    for (size_t i = 0; i < num_sequences; ++i) {
243 244
      warpctc_label_lengths[i] = label_lod[i + 1] - label_lod[i];
      warpctc_logits_lengths[i] = logits_lod[i + 1] - logits_lod[i];
Y
Yiqun Liu 已提交
245 246 247 248 249 250 251
    }

    // warpctc computes loss and gradient in one call, gradient data also stored
    // in batch format
    T* warpctc_grad_data =
        warpctc_grad->mutable_data<T>(warpctc_logits.dims(), ctx.GetPlace());

252 253 254 255
    math::SetConstant<DeviceContext, T>()(
        ctx.template device_context<DeviceContext>(), warpctc_grad,
        static_cast<T>(0));

Y
Yiqun Liu 已提交
256
    // warpctc accesses labels in CPU memory
W
whs 已提交
257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285
    LoDTensor warpctc_label;
    if (ctx.HasInput("LogitsLength")) {
      warpctc_label.mutable_data<int>(
          {static_cast<int64_t>(math::TotalSequenceLength(label_lod)), 1},
          platform::CPUPlace());
      std::vector<framework::Vector<size_t>> lod;
      lod.push_back(label_lod);
      warpctc_label.set_lod(lod);

      if (platform::is_cpu_place(ctx.GetPlace())) {
        math::UnpaddingLoDTensorFunctor<DeviceContext, int>()(
            ctx.template device_context<DeviceContext>(), *label,
            &warpctc_label, label->dims()[1] /*pad_seq_len*/, 0 /*lod_level*/,
            false /*norm_by_times*/, math::kBatchLengthWidth);
      } else {
        LoDTensor gpu_label;
        gpu_label.mutable_data<int>(
            {static_cast<int64_t>(math::TotalSequenceLength(label_lod)), 1},
            ctx.GetPlace());
        gpu_label.set_lod(lod);
        math::UnpaddingLoDTensorFunctor<DeviceContext, int>()(
            ctx.template device_context<DeviceContext>(), *label, &gpu_label,
            label->dims()[1] /*pad_seq_len*/, 0 /*lod_level*/,
            false /*norm_by_times*/, math::kBatchLengthWidth);
        TensorCopySync(gpu_label, platform::CPUPlace(), &warpctc_label);
      }
    } else {
      TensorCopySync(*label, platform::CPUPlace(), &warpctc_label);
    }
286

Y
Yiqun Liu 已提交
287 288 289 290 291 292 293 294 295 296 297 298 299 300
    const int* warpctc_label_data = warpctc_label.data<int>();
    // warpctc stores loss in CPU memory
    Tensor warpctc_loss;
    T* warpctc_loss_data =
        warpctc_loss.mutable_data<T>(loss_dims, platform::CPUPlace());

    const size_t blank = static_cast<size_t>(ctx.Attr<int>("blank"));

    WarpCTCFunctor<DeviceContext>()(
        ctx, warpctc_logits_data, warpctc_grad_data, warpctc_label_data,
        warpctc_label_lengths.data(), warpctc_logits_lengths.data(),
        sequence_width, num_sequences, blank, warpctc_loss_data);

    // Copy the loss back
Y
Yi Wang 已提交
301
    TensorCopy(warpctc_loss, ctx.GetPlace(), ctx.device_context(), loss);
Y
Yiqun Liu 已提交
302 303 304 305 306 307 308
  }
};

template <typename DeviceContext, typename T>
class WarpCTCGradKernel : public framework::OpKernel<T> {
 public:
  void Compute(const framework::ExecutionContext& ctx) const override {
F
fengjiayi 已提交
309
    auto* warpctc_grad = ctx.Input<LoDTensor>("WarpCTCGrad");
Y
Yiqun Liu 已提交
310
    auto* logits_grad = ctx.Output<LoDTensor>(framework::GradVarName("Logits"));
W
wanghaoshuang 已提交
311 312 313
    const Tensor* loss_grad = ctx.Input<Tensor>(framework::GradVarName("Loss"));

    logits_grad->mutable_data<T>(ctx.GetPlace());
Y
Yiqun Liu 已提交
314
    bool norm_by_times = ctx.Attr<bool>("norm_by_times");
315 316 317 318 319 320

    if (ctx.HasInput("LogitsLength")) {
      size_t max_seq_length = warpctc_grad->dims()[0];
      size_t num_sequences = warpctc_grad->dims()[1];
      size_t seq_width = warpctc_grad->dims()[2];

321 322 323 324 325
      auto* logits_length = ctx.Input<framework::Tensor>("LogitsLength");
      framework::Tensor logits_length_cpu;
      framework::TensorCopy(*logits_length, platform::CPUPlace(),
                            &logits_length_cpu);

326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347
      LoDTensor logits_grad_with_lod;
      auto logits_grad_dims =
          framework::make_ddim({static_cast<int64_t>(max_seq_length),
                                static_cast<int64_t>(num_sequences),
                                static_cast<int64_t>(seq_width)});
      T* logits_grad_cpu_data = logits_grad_with_lod.mutable_data<T>(
          logits_grad_dims, platform::CPUPlace());

      TensorCopySync(*warpctc_grad, platform::CPUPlace(),
                     &logits_grad_with_lod);

      Tensor loss_grad_cpu;
      loss_grad_cpu.mutable_data<T>(loss_grad->dims(), platform::CPUPlace());
      TensorCopySync(*loss_grad, platform::CPUPlace(), &loss_grad_cpu);

      LoDTensor scaled_logits;
      T* scaled_logits_data =
          scaled_logits.mutable_data<T>(logits_grad_dims, platform::CPUPlace());

      const T* loss_grad_data = loss_grad_cpu.data<T>();
      for (size_t i = 0; i < max_seq_length; ++i) {
        for (size_t j = 0; j < num_sequences; ++j) {
348 349 350 351
          T scale = 1.0;
          if (norm_by_times) {
            scale = 1.0 / static_cast<T>(logits_length_cpu.data<int64_t>()[j]);
          }
352 353 354
          for (size_t k = 0; k < seq_width; ++k) {
            size_t idx = i * (num_sequences * seq_width) + j * seq_width + k;
            scaled_logits_data[idx] =
355
                logits_grad_cpu_data[idx] * loss_grad_data[j] * scale;
356 357 358 359 360 361 362 363 364 365 366 367 368 369 370
          }
        }
      }

      TensorCopySync(scaled_logits, ctx.GetPlace(), logits_grad);
    } else {
      math::UnpaddingLoDTensorFunctor<DeviceContext, T>()(
          ctx.template device_context<DeviceContext>(), *warpctc_grad,
          logits_grad, -1, 0, norm_by_times, math::kLengthBatchWidth);

      const T* loss_grad_data = loss_grad->data<T>();
      math::ScaleLoDTensorFunctor<DeviceContext, T>()(
          ctx.template device_context<DeviceContext>(), loss_grad_data,
          logits_grad);
    }
Y
Yiqun Liu 已提交
371 372 373 374 375
  }
};

}  // namespace operators
}  // namespace paddle