cudnn version. staged.

paddle/fluid/operators/top_k_op.cc

@@ -50,7 +50,7 @@ class TopkOpMaker : public framework::OpProtoAndCheckerMaker {
  public:
   void Make() override {
     AddInput("X", "(Tensor) The input of Topk op");
-    AddOutput("Out", "(Tensor) The output tensor of Topk op").Reuse("X");
+    AddOutput("Out", "(Tensor) The output tensor of Topk op");
     AddOutput("Indices", "(Tensor) The indices of Topk elements of input");
     AddComment(R"DOC(
 Top K operator

paddle/fluid/operators/top_k_op.cu

@@ -256,36 +256,65 @@ __device__ __forceinline__ void BlockReduce(Pair<T>* sh_topk, int* maxid,
  * 3. go to the second setp, until one thread's topk value is null;
  * 4. go to the first setp, until get the topk value.
  */
+
 template <typename T, int MaxLength, int BlockSize>
 __global__ void KeMatrixTopK(T* output, int output_stride, int64_t* indices,
-                             const T* src, int lds, int dim, int k) {
+                             const T* src, int lds, int dim, int k,
+                             int grid_dim, int num) {
   __shared__ Pair<T> sh_topk[BlockSize];
-  __shared__ int maxid[BlockSize / 2];
   const int tid = threadIdx.x;
   const int warp = threadIdx.x / 32;
-  output += blockIdx.x * output_stride;
-  indices += blockIdx.x * k;
 
+  const int bid = blockIdx.x;
+  for (int i = bid; i < num; i += grid_dim) {
+    int top_num = k;
+    __shared__ int maxid[BlockSize / 2];
+    T* out = output + i * output_stride;
+    int64_t* inds = indices + i * k;
     Pair<T> topk[MaxLength];
     int beam = MaxLength;
     Pair<T> max;
     bool is_empty = false;
     bool firststep = true;
 
-  for (int k = 0; k < MaxLength; k++) {
-    topk[k].set(-INFINITY, -1);
+    for (int j = 0; j < MaxLength; j++) {
+      topk[j].set(-INFINITY, -1);
     }
-  while (k) {
-    ThreadGetTopK<T, MaxLength, BlockSize>(topk, &beam, k,
-                                           src + blockIdx.x * lds, &firststep,
-                                           &is_empty, &max, dim, tid);
+    while (top_num) {
+      ThreadGetTopK<T, MaxLength, BlockSize>(
+          topk, &beam, k, src + i * lds, &firststep, &is_empty, &max, dim, tid);
 
       sh_topk[tid] = topk[0];
-    BlockReduce<T, MaxLength, BlockSize>(sh_topk, maxid, topk, &output,
-                                         &indices, &beam, &k, tid, warp);
+      BlockReduce<T, MaxLength, BlockSize>(sh_topk, maxid, topk, &out, &inds,
+                                           &beam, &top_num, tid, warp);
+    }
   }
 }
 
+inline static int GetDesiredBlockDim(int dim) {
+  if (dim > 128) {
+    return 256;
+  } else if (dim > 64) {
+    return 128;
+  } else if (dim > 32) {
+    return 64;
+  } else {
+    return 32;
+  }
+}
+
+#define FIXED_BLOCK_DIM_BASE(dim, ...) \
+  case (dim): {                        \
+    constexpr auto kBlockDim = (dim);  \
+    __VA_ARGS__;                       \
+  } break
+
+#define FIXED_BLOCK_DIM(...)                \
+  FIXED_BLOCK_DIM_BASE(256, ##__VA_ARGS__); \
+  FIXED_BLOCK_DIM_BASE(128, ##__VA_ARGS__); \
+  FIXED_BLOCK_DIM_BASE(64, ##__VA_ARGS__);  \
+  FIXED_BLOCK_DIM_BASE(32, ##__VA_ARGS__)
+
 template <typename T>
 class TopkOpCUDAKernel : public framework::OpKernel<T> {
  public:
@@ -298,30 +327,38 @@ class TopkOpCUDAKernel : public framework::OpKernel<T> {
     size_t k = static_cast<int>(ctx.Attr<int>("k"));
 
     const T* input_data = input->data<T>();
-
     T* output_data = output->mutable_data<T>(ctx.GetPlace());
     // FIXME(typhoonzero): data is always converted to type T?
     int64_t* indices_data = indices->mutable_data<int64_t>(ctx.GetPlace());
 
-    size_t input_height = input->dims()[0];
-    size_t input_width = input->dims()[1];
+    framework::DDim inputdims = input->dims();
+    const size_t input_height = framework::product(
+        framework::slice_ddim(inputdims, 0, inputdims.size() - 1));
+    const size_t input_width = inputdims[inputdims.size() - 1];
+
     if (k > input_width) k = input_width;
 
     // NOTE: pass lds and dim same to input width.
     // NOTE: old matrix implementation of stride is different to eigen.
     // TODO(typhoonzero): refine this kernel.
-    dim3 threads(256, 1);
-    dim3 grid(input_height, 1);
-
-    KeMatrixTopK<T, 5, 256><<<
-        grid, threads, 0, reinterpret_cast<const platform::CUDADeviceContext&>(
-                              ctx.device_context())
-                              .stream()>>>(
-        output_data, output->dims()[1], indices_data, input_data, input_width,
-        input_width, static_cast<int>(k));
+    const int kMaxHeight = 2048;
+    int gridx = input_height < kMaxHeight ? input_height : kMaxHeight;
+    auto& dev_ctx = ctx.cuda_device_context();
+    switch (GetDesiredBlockDim(input_width)) {
+      FIXED_BLOCK_DIM(
+          KeMatrixTopK<T, 5,
+                       kBlockDim><<<gridx, kBlockDim, 0, dev_ctx.stream()>>>(
+              output_data, k, indices_data, input_data, input_width,
+              input_width, static_cast<int>(k), gridx, input_height));
+      default:
+        PADDLE_THROW("Error");
+    }
   }
 };
 
+#undef FIXED_BLOCK_DIM_BASE
+#undef FIXED_BLOCK_DIM
+
 }  // namespace operators
 }  // namespace paddle
 

paddle/fluid/operators/top_k_op.h

@@ -34,7 +34,6 @@ class TopkKernel : public framework::OpKernel<T> {
  public:
   void Compute(const framework::ExecutionContext& ctx) const override {
     // Get the top k elements of each row of input tensor
-    // FIXME: only deal with matrix(2d tensor).
     auto* input = ctx.Input<Tensor>("X");
     auto* output = ctx.Output<Tensor>("Out");
     auto* indices = ctx.Output<Tensor>("Indices");
@@ -44,8 +43,6 @@ class TopkKernel : public framework::OpKernel<T> {
     T* output_data = output->mutable_data<T>(ctx.GetPlace());
     int64_t* indices_data = indices->mutable_data<int64_t>(ctx.GetPlace());
 
-    auto eg_input = EigenMatrix<T>::From(*input);
-
     // reshape input to a flattern matrix(like flat_inner_dims)
     framework::DDim inputdims = input->dims();
     const size_t row = framework::product(
@@ -53,7 +50,7 @@ class TopkKernel : public framework::OpKernel<T> {
     const size_t col = inputdims[inputdims.size() - 1];
     Eigen::DSizes<int, 2> flat2dims(row, col);
     // NOTE: eigen shape doesn't affect paddle tensor.
-    eg_input.reshape(flat2dims);
+    auto eg_input = EigenMatrix<T>::Reshape(*input, inputdims.size() - 1);
 
 #ifdef PADDLE_WITH_MKLML
 #pragma omp parallel for