distribute_fpn_proposals_op.cu 7.9 KB
Newer Older
J
jerrywgz 已提交
1
/* Copyright (c) 2019 PaddlePaddle Authors. All Rights Reserved.
J
jerrywgz 已提交
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17

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. */

#include <paddle/fluid/memory/allocation/allocator.h>
#include "cub/cub.cuh"
#include "paddle/fluid/memory/memcpy.h"
18
#include "paddle/fluid/operators/detection/bbox_util.h"
J
jerrywgz 已提交
19 20
#include "paddle/fluid/operators/detection/distribute_fpn_proposals_op.h"
#include "paddle/fluid/operators/gather.cu.h"
21
#include "paddle/fluid/operators/math/math_function.h"
J
jerrywgz 已提交
22 23 24 25 26 27 28 29 30
#include "paddle/fluid/platform/cuda_primitives.h"
#include "paddle/fluid/platform/for_range.h"

namespace paddle {
namespace operators {

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

31
static constexpr int kNumCUDAThreads = 64;
J
jerrywgz 已提交
32 33 34 35 36 37 38 39 40 41 42 43 44 45
static constexpr int kNumMaxinumNumBlocks = 4096;

#define CUDA_1D_KERNEL_LOOP(i, n)                              \
  for (int i = blockIdx.x * blockDim.x + threadIdx.x; i < (n); \
       i += blockDim.x * gridDim.x)

int const BBoxSize = 4;

static inline int NumBlocks(const int N) {
  return std::min((N + kNumCUDAThreads - 1) / kNumCUDAThreads,
                  kNumMaxinumNumBlocks);
}

template <class T>
46
__global__ void GPUDistFpnProposalsHelper(
J
jerrywgz 已提交
47 48 49 50 51 52 53
    const int nthreads, const T* rois, const int lod_size,
    const int refer_level, const int refer_scale, const int max_level,
    const int min_level, int* roi_batch_id_data, int* sub_lod_list,
    int* target_lvls) {
  CUDA_1D_KERNEL_LOOP(i, nthreads) {
    const T* offset_roi = rois + i * BBoxSize;
    int roi_batch_ind = roi_batch_id_data[i];
J
jerrywgz 已提交
54
    // get the target level of current rois
J
jerrywgz 已提交
55 56
    T roi_area = RoIArea(offset_roi, false);
    T roi_scale = sqrt(roi_area);
57 58
    int tgt_lvl = floor(
        log2(roi_scale / static_cast<T>(refer_scale) + (T)1e-6) + refer_level);
J
jerrywgz 已提交
59 60
    tgt_lvl = min(max_level, max(tgt_lvl, min_level));
    target_lvls[i] = tgt_lvl;
J
jerrywgz 已提交
61
    // compute number of rois in the same batch and same target level
62 63
    platform::CudaAtomicAdd(
        sub_lod_list + (tgt_lvl - min_level) * lod_size + roi_batch_ind, 1);
J
jerrywgz 已提交
64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82
  }
}

template <typename DeviceContext, typename T>
class GPUDistributeFpnProposalsOpKernel : public framework::OpKernel<T> {
 public:
  void Compute(const framework::ExecutionContext& ctx) const override {
    auto* fpn_rois = ctx.Input<paddle::framework::LoDTensor>("FpnRois");

    auto multi_fpn_rois = ctx.MultiOutput<LoDTensor>("MultiFpnRois");
    auto* restore_index = ctx.Output<Tensor>("RestoreIndex");

    const int min_level = ctx.Attr<int>("min_level");
    const int max_level = ctx.Attr<int>("max_level");
    const int refer_level = ctx.Attr<int>("refer_level");
    const int refer_scale = ctx.Attr<int>("refer_scale");
    int num_level = max_level - min_level + 1;

    // check that the fpn_rois is not empty
83 84 85 86
    PADDLE_ENFORCE_EQ(
        fpn_rois->lod().size(), 1UL,
        platform::errors::InvalidArgument("DistributeFpnProposalsOp needs LoD"
                                          "with one level"));
J
jerrywgz 已提交
87 88 89 90 91 92 93

    auto fpn_rois_lod = fpn_rois->lod().back();
    int lod_size = fpn_rois_lod.size() - 1;
    int roi_num = fpn_rois_lod[lod_size];

    auto& dev_ctx = ctx.template device_context<DeviceContext>();

J
jerrywgz 已提交
94
    // get batch id by lod in CPU
J
jerrywgz 已提交
95 96 97 98 99 100 101 102 103
    Tensor roi_batch_id_list;
    roi_batch_id_list.Resize({roi_num});
    int* roi_batch_id_data =
        roi_batch_id_list.mutable_data<int>(platform::CPUPlace());
    for (int n = 0; n < lod_size; ++n) {
      for (size_t i = fpn_rois_lod[n]; i < fpn_rois_lod[n + 1]; ++i) {
        roi_batch_id_data[i] = n;
      }
    }
J
jerrywgz 已提交
104
    // copy batch id list to GPU
J
jerrywgz 已提交
105 106 107 108 109 110 111
    Tensor roi_batch_id_list_gpu;
    framework::TensorCopySync(roi_batch_id_list, dev_ctx.GetPlace(),
                              &roi_batch_id_list_gpu);

    Tensor sub_lod_list;
    sub_lod_list.Resize({num_level, lod_size});
    int* sub_lod_list_data = sub_lod_list.mutable_data<int>(dev_ctx.GetPlace());
112 113 114
    math::SetConstant<platform::CUDADeviceContext, int> set_zero;
    set_zero(dev_ctx, &sub_lod_list, static_cast<int>(0));

J
jerrywgz 已提交
115 116 117 118
    Tensor target_lvls;
    target_lvls.Resize({roi_num});
    int* target_lvls_data = target_lvls.mutable_data<int>(dev_ctx.GetPlace());

119
    int dist_blocks = NumBlocks(roi_num);
J
jerrywgz 已提交
120
    int threads = kNumCUDAThreads;
J
jerrywgz 已提交
121
    // get target levels and sub_lod list
122
    GPUDistFpnProposalsHelper<T><<<dist_blocks, threads>>>(
J
jerrywgz 已提交
123 124 125
        roi_num, fpn_rois->data<T>(), lod_size, refer_level, refer_scale,
        max_level, min_level, roi_batch_id_list_gpu.data<int>(),
        sub_lod_list_data, target_lvls_data);
126
    dev_ctx.Wait();
127
    auto place = BOOST_GET_CONST(platform::CUDAPlace, dev_ctx.GetPlace());
J
jerrywgz 已提交
128 129 130 131 132 133 134 135 136 137 138 139 140

    Tensor index_in_t;
    int* idx_in = index_in_t.mutable_data<int>({roi_num}, dev_ctx.GetPlace());
    platform::ForRange<platform::CUDADeviceContext> for_range(dev_ctx, roi_num);
    for_range(RangeInitFunctor{0, 1, idx_in});

    Tensor keys_out_t;
    int* keys_out = keys_out_t.mutable_data<int>({roi_num}, dev_ctx.GetPlace());
    Tensor index_out_t;
    int* idx_out = index_out_t.mutable_data<int>({roi_num}, dev_ctx.GetPlace());

    // Determine temporary device storage requirements
    size_t temp_storage_bytes = 0;
141 142 143
    cub::DeviceRadixSort::SortPairs<int, int>(nullptr, temp_storage_bytes,
                                              target_lvls_data, keys_out,
                                              idx_in, idx_out, roi_num);
J
jerrywgz 已提交
144
    // Allocate temporary storage
145
    auto d_temp_storage = memory::Alloc(place, temp_storage_bytes);
J
jerrywgz 已提交
146 147

    // Run sorting operation
J
jerrywgz 已提交
148
    // sort target level to get corresponding index
149
    cub::DeviceRadixSort::SortPairs<int, int>(
J
jerrywgz 已提交
150 151 152 153 154
        d_temp_storage->ptr(), temp_storage_bytes, target_lvls_data, keys_out,
        idx_in, idx_out, roi_num);

    int* restore_idx_data =
        restore_index->mutable_data<int>({roi_num, 1}, dev_ctx.GetPlace());
J
jerrywgz 已提交
155
    // sort current index to get restore index
156
    cub::DeviceRadixSort::SortPairs<int, int>(
J
jerrywgz 已提交
157 158 159
        d_temp_storage->ptr(), temp_storage_bytes, idx_out, keys_out, idx_in,
        restore_idx_data, roi_num);

160
    int start = 0;
J
jerrywgz 已提交
161 162 163
    for (int i = 0; i < num_level; ++i) {
      Tensor sub_lod = sub_lod_list.Slice(i, i + 1);
      int* sub_lod_data = sub_lod.data<int>();
J
jerrywgz 已提交
164
      // transfer length-based lod to offset-based lod
165 166 167 168 169 170 171 172
      std::vector<size_t> offset(1, 0);
      std::vector<int> sub_lod_cpu(lod_size);
      memory::Copy(platform::CPUPlace(), sub_lod_cpu.data(), place,
                   sub_lod_data, sizeof(int) * lod_size, dev_ctx.stream());
      dev_ctx.Wait();
      for (int j = 0; j < lod_size; ++j) {
        offset.emplace_back(offset.back() + sub_lod_cpu[j]);
      }
J
jerrywgz 已提交
173

174 175 176 177 178 179 180 181 182 183 184 185 186
      int sub_rois_num = offset.back();

      int end = start + sub_rois_num;
      if (end > start) {
        Tensor sub_idx = index_out_t.Slice(start, end);
        start = end;
        multi_fpn_rois[i]->mutable_data<T>({sub_rois_num, kBoxDim},
                                           dev_ctx.GetPlace());
        GPUGather<T>(dev_ctx, *fpn_rois, sub_idx, multi_fpn_rois[i]);
      } else {
        multi_fpn_rois[i]->mutable_data<T>({sub_rois_num, kBoxDim},
                                           dev_ctx.GetPlace());
      }
J
jerrywgz 已提交
187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203
      framework::LoD lod;
      lod.emplace_back(offset);
      multi_fpn_rois[i]->set_lod(lod);
    }
  }
};

}  // namespace operators
}  // namespace paddle

namespace ops = paddle::operators;
REGISTER_OP_CUDA_KERNEL(
    distribute_fpn_proposals,
    ops::GPUDistributeFpnProposalsOpKernel<paddle::platform::CUDADeviceContext,
                                           float>,
    ops::GPUDistributeFpnProposalsOpKernel<paddle::platform::CUDADeviceContext,
                                           double>);