Optimize Topk when height is large. (#13710)

paddle/fluid/operators/top_k_op.cu

@@ -256,15 +256,20 @@ __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) {
+    output += i * output_stride;
+    indices += i * k;
 
     Pair<T> topk[MaxLength];
     int beam = MaxLength;
@@ -276,16 +281,40 @@ __global__ void KeMatrixTopK(T* output, int output_stride, int64_t* indices,
       topk[k].set(-INFINITY, -1);
     }
     while (k) {
-    ThreadGetTopK<T, MaxLength, BlockSize>(topk, &beam, k,
-                                           src + blockIdx.x * lds, &firststep,
-                                           &is_empty, &max, dim, tid);
+      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);
     }
+  }
 }
 
+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:
@@ -310,18 +339,26 @@ class TopkOpCUDAKernel : public framework::OpKernel<T> {
     // 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);
+    const int kMaxHeight = 2048;
+    int gridx = input_height < kMaxHeight ? input_height : kMaxHeight;
+    auto& dev_ctx = ctx.cuda_device_context();
 
-    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));
+    switch (GetDesiredBlockDim(input_width)) {
+      FIXED_BLOCK_DIM(
+          KeMatrixTopK<T, 5,
+                       kBlockDim><<<gridx, kBlockDim, 0, dev_ctx.stream()>>>(
+              output_data, output->dims()[1], 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