split_op_plugin.cu 11.0 KB
Newer Older
N
nhzlx 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14
// Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
//
// 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.

H
hjchen2 已提交
15 16
#include <cuda_fp16.h>
#include <algorithm>
N
nhzlx 已提交
17 18 19 20 21
#include "paddle/fluid/inference/tensorrt/plugin/split_op_plugin.h"

namespace paddle {
namespace inference {
namespace tensorrt {
22
namespace plugin {
N
nhzlx 已提交
23

H
hjchen2 已提交
24
template <typename T>
25 26 27 28 29 30 31 32 33 34
__device__ int upper_bound(T const* vals, int n, T const& key) {
  int i = 0;
  while (n > 0) {
    int m = n / 2;
    int j = i + m;
    if (!(key < vals[j])) {
      i = j + 1;
      n -= m + 1;
    } else {
      n = m;
H
hjchen2 已提交
35 36
    }
  }
37
  return i;
H
hjchen2 已提交
38 39
}

40 41
nvinfer1::Dims SplitPlugin::getOutputDimensions(
    int index, const nvinfer1::Dims* input_dims, int num_inputs) {
42 43 44 45 46 47 48 49 50 51 52
  PADDLE_ENFORCE_EQ(num_inputs, 1,
                    platform::errors::InvalidArgument(
                        "Invalid number of inputs of split TRT plugin. "
                        "Expected 1, received %d.",
                        num_inputs));
  PADDLE_ENFORCE_LT(
      index, this->getNbOutputs(),
      platform::errors::InvalidArgument(
          "Index of output should be less than the total number of outputs in "
          "split TensorRT plugin. Received index = %d >= total outputs = %d",
          index, this->getNbOutputs()));
53 54

  nvinfer1::Dims output_dims = input_dims[0];
55
  output_dims.d[axis_] = output_length_.at(index);
N
nhzlx 已提交
56 57 58
  return output_dims;
}

59 60 61 62 63 64 65 66 67 68
void SplitPlugin::shareData(const SplitPlugin* another) {
  outer_rows_ = another->outer_rows_;
  inner_cols_ = another->inner_cols_;
  same_shape_ = another->same_shape_;
  axis_shape_ = another->axis_shape_;
  d_segment_offsets_ = another->d_segment_offsets_;
  segment_offsets_ = another->segment_offsets_;
  d_output_ptrs_.resize(another->d_output_ptrs_.size(), nullptr);
}

N
nhzlx 已提交
69
int SplitPlugin::initialize() {
70 71 72 73 74
  PADDLE_ENFORCE_LE(axis_, nvinfer1::Dims::MAX_DIMS,
                    platform::errors::InvalidArgument(
                        "Axis dimension exceeds max dimension in TensorRT. "
                        "Received axis = %d > MAX_DIMS = %d",
                        axis_, nvinfer1::Dims::MAX_DIMS));
H
hjchen2 已提交
75 76 77 78 79 80 81 82 83 84 85
  // notice input dims is [C, H, W]
  nvinfer1::Dims dims = this->getInputDims(0);
  outer_rows_ = 1;
  inner_cols_ = 1;
  for (int i = 0; i < axis_; ++i) {
    outer_rows_ *= dims.d[i];
  }
  for (int i = axis_ + 1; i < dims.nbDims; ++i) {
    inner_cols_ *= dims.d[i];
  }
  same_shape_ = true;
N
nhzlx 已提交
86 87
  std::vector<int> segment_offsets(1, 0);
  for (int i = 0; i < this->getNbOutputs(); ++i) {
H
hjchen2 已提交
88 89 90
    if (output_length_[i] != output_length_[0]) {
      same_shape_ = false;
    }
91
    segment_offsets.push_back(segment_offsets.back() + output_length_[i]);
N
nhzlx 已提交
92
  }
93
  axis_shape_ = dims.d[axis_];
H
hjchen2 已提交
94 95 96 97 98 99
  d_segment_offsets_ = segment_offsets;
  segment_offsets_ = std::move(segment_offsets);
  d_output_ptrs_.resize(this->getNbOutputs(), nullptr);
  return 0;
}

100 101 102
// nothing to release according to initialize
void SplitPlugin::terminate() {}

103 104
// The following part of the code refers to onnx-tensorrt
// https://github.com/onnx/onnx-tensorrt/blob/master/Split.cu
H
hjchen2 已提交
105
template <typename T>
106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123
__global__ void split_kernel(int nsegment,
                             int const* __restrict__ segment_offsets,
                             T const* __restrict__ idata, T* const* odatas,
                             int inner_cols, int axis_shape, int outer_rows) {
  int x0 = threadIdx.x + blockIdx.x * blockDim.x;
  int src_y0 = threadIdx.y + blockIdx.y * blockDim.y;
  int z0 = threadIdx.z + blockIdx.z * blockDim.z;
  for (int z = z0; z < outer_rows; z += blockDim.z * gridDim.z) {
    for (int src_y = src_y0; src_y < axis_shape;
         src_y += blockDim.y * gridDim.y) {
      for (int x = x0; x < inner_cols; x += blockDim.x * gridDim.x) {
        int segment = upper_bound(segment_offsets, nsegment, src_y) - 1;
        int dst_y = src_y - segment_offsets[segment];
        int dst_ny = segment_offsets[segment + 1] - segment_offsets[segment];
        odatas[segment][x + inner_cols * (dst_y + dst_ny * z)] =
            idata[x + inner_cols * (src_y + axis_shape * z)];
      }
    }
N
nhzlx 已提交
124 125 126 127
  }
}

int SplitPlugin::enqueue(int batchSize, const void* const* inputs,
128
#if IS_TRT_VERSION_LT(8000)
N
nhzlx 已提交
129
                         void** outputs, void* workspace, cudaStream_t stream) {
130 131 132 133
#else
                         void* const* outputs, void* workspace,
                         cudaStream_t stream) {
#endif
134 135
  const int* d_segment_offsets_ptr =
      thrust::raw_pointer_cast(&d_segment_offsets_[0]);
H
hjchen2 已提交
136
  float const* input_ptr = reinterpret_cast<float const*>(inputs[0]);
137 138
  float* const* h_odatas = reinterpret_cast<float* const*>(outputs);
  float** output_ptrs = thrust::raw_pointer_cast(&d_output_ptrs_[0]);
139 140 141
  PADDLE_ENFORCE_CUDA_SUCCESS(cudaMemcpyAsync(
      output_ptrs, h_odatas, d_output_ptrs_.size() * sizeof(float*),
      cudaMemcpyHostToDevice, stream));
142 143 144 145 146 147 148 149 150 151 152

  int outer_rows = outer_rows_ * batchSize;

  dim3 block(32, 16);
  dim3 grid(std::min((inner_cols_ - 1) / block.x + 1, 65535u),
            std::min((axis_shape_ - 1) / block.y + 1, 65535u),
            std::min((outer_rows_ - 1) / block.z + 1, 65535u));

  split_kernel<<<grid, block, 0, stream>>>(
      d_segment_offsets_.size(), d_segment_offsets_ptr, input_ptr, output_ptrs,
      inner_cols_, axis_shape_, outer_rows);
N
nhzlx 已提交
153 154 155
  return cudaGetLastError() != cudaSuccess;
}

156 157 158 159
// Dynamic Plugin below.
#if IS_TRT_VERSION_GE(6000)
int SplitPluginDynamic::initialize() { return 0; }

160 161 162 163
size_t SplitPluginDynamic::getSerializationSize() const {
  return SerializedSize(axis_) + SerializedSize(output_length_) +
         SerializedSize(with_fp16_);
}
164

165 166 167 168 169
void SplitPluginDynamic::serialize(void* buffer) const {
  SerializeValue(&buffer, axis_);
  SerializeValue(&buffer, output_length_);
  SerializeValue(&buffer, with_fp16_);
}
170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193

nvinfer1::DimsExprs SplitPluginDynamic::getOutputDimensions(
    int output_index, const nvinfer1::DimsExprs* inputs, int nb_inputs,
    nvinfer1::IExprBuilder& expr_builder) {
  PADDLE_ENFORCE_EQ(nb_inputs, 1,
                    platform::errors::InvalidArgument(
                        "The Split plugin should be only one input."));
  PADDLE_ENFORCE_LT(output_index, output_length_.size(),
                    platform::errors::InvalidArgument(
                        "When GetOutputDimensions, the index(%d) should not "
                        "greater the num(%d) of the outpus.",
                        output_index, output_length_.size()));

  nvinfer1::DimsExprs output_dims = inputs[0];
  output_dims.d[axis_] = expr_builder.constant(output_length_.at(output_index));

  return output_dims;
}

bool SplitPluginDynamic::supportsFormatCombination(
    int pos, const nvinfer1::PluginTensorDesc* in_out, int nb_inputs,
    int nb_outputs) {
  PADDLE_ENFORCE_NOT_NULL(
      in_out, platform::errors::InvalidArgument(
194
                  "The input of split plugin should not be nullptr."));
195 196 197 198 199 200 201 202 203 204

  PADDLE_ENFORCE_LT(
      pos, nb_inputs + nb_outputs,
      platform::errors::InvalidArgument("The pos(%d) should be less than the "
                                        "num(%d) of the input and the output.",
                                        pos, nb_inputs + nb_outputs));
  (in_out && pos < (nb_inputs + nb_outputs));

  const nvinfer1::PluginTensorDesc& in = in_out[pos];
  if (pos == 0) {
205 206 207 208 209 210 211 212
    if (with_fp16_) {
      return (in.type == nvinfer1::DataType::kFLOAT ||
              in.type == nvinfer1::DataType::kHALF) &&
             (in.format == nvinfer1::TensorFormat::kLINEAR);
    } else {
      return (in.type == nvinfer1::DataType::kFLOAT) &&
             (in.format == nvinfer1::TensorFormat::kLINEAR);
    }
213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255
  }
  const nvinfer1::PluginTensorDesc& prev = in_out[pos - 1];
  // output
  return in.type == prev.type && in.format == prev.format;
}

nvinfer1::DataType SplitPluginDynamic::getOutputDataType(
    int index, const nvinfer1::DataType* input_types, int nb_inputs) const {
  return input_types[0];
}

int SplitPluginDynamic::enqueue(const nvinfer1::PluginTensorDesc* input_desc,
                                const nvinfer1::PluginTensorDesc* output_desc,
                                const void* const* inputs, void* const* outputs,
                                void* workspace, cudaStream_t stream) {
  auto input_dims = input_desc[0].dims;
  int outer_rows = 1;
  int inner_cols = 1;
  // with batch
  for (int i = 0; i < axis_; i++) {
    outer_rows *= input_dims.d[i];
  }

  for (int i = axis_ + 1; i < input_dims.nbDims; i++) {
    inner_cols *= input_dims.d[i];
  }

  std::vector<int> segment_offsets(1, 0);
  for (int i = 0; i < this->getNbOutputs(); i++) {
    segment_offsets.push_back(segment_offsets.back() + output_length_[i]);
  }
  int axis_shape = input_dims.d[axis_];
  thrust::device_vector<int> d_segment_offsets = segment_offsets;
  const int* d_segment_offsets_ptr =
      thrust::raw_pointer_cast(&d_segment_offsets[0]);

  dim3 block(32, 16);
  dim3 grid(std::min((inner_cols - 1) / block.x + 1, 65535u),
            std::min((axis_shape - 1) / block.y + 1, 65535u),
            std::min((outer_rows - 1) / block.z + 1, 65535u));

  auto input_type = input_desc[0].type;
  if (input_type == nvinfer1::DataType::kFLOAT) {
256
    VLOG(1) << "TRT Plugin DataType selected. Split-->fp32";
257 258 259 260 261 262 263
    thrust::device_vector<float*> d_output_ptrs;
    d_output_ptrs.resize(this->getNbOutputs(), nullptr);

    const float* input_ptr = static_cast<const float*>(inputs[0]);
    float* const* h_odatas = reinterpret_cast<float* const*>(outputs);
    float** output_ptrs = thrust::raw_pointer_cast(&d_output_ptrs[0]);

264 265 266
    PADDLE_ENFORCE_CUDA_SUCCESS(cudaMemcpyAsync(
        output_ptrs, h_odatas, d_output_ptrs.size() * sizeof(float*),
        cudaMemcpyHostToDevice, stream));
267 268 269 270 271

    split_kernel<<<grid, block, 0, stream>>>(
        d_segment_offsets.size(), d_segment_offsets_ptr, input_ptr, output_ptrs,
        inner_cols, axis_shape, outer_rows);
  } else if (input_type == nvinfer1::DataType::kHALF) {
272
    VLOG(1) << "TRT Plugin DataType selected. Split-->fp16";
273 274 275 276 277 278 279
    thrust::device_vector<half*> d_output_ptrs;
    d_output_ptrs.resize(this->getNbOutputs(), nullptr);

    const half* input_ptr = static_cast<const half*>(inputs[0]);
    half* const* h_odatas = reinterpret_cast<half* const*>(outputs);
    half** output_ptrs = thrust::raw_pointer_cast(&d_output_ptrs[0]);

280 281 282
    PADDLE_ENFORCE_CUDA_SUCCESS(cudaMemcpyAsync(
        output_ptrs, h_odatas, d_output_ptrs.size() * sizeof(half*),
        cudaMemcpyHostToDevice, stream));
283 284 285 286 287 288 289 290 291

    split_kernel<<<grid, block, 0, stream>>>(
        d_segment_offsets.size(), d_segment_offsets_ptr, input_ptr, output_ptrs,
        inner_cols, axis_shape, outer_rows);
  }
  return cudaGetLastError() != cudaSuccess;
}
#endif

292 293 294 295
}  // namespace plugin
}  // namespace tensorrt
}  // namespace inference
}  // namespace paddle