[GitHub] [incubator-mxnet] anirudh2290 commented on a change in pull request #13401: [MXNET-1227] Adding CornerPooling operator

[GitBox <gi...@apache.org>]

anirudh2290 commented on a change in pull request #13401: [MXNET-1227] Adding CornerPooling operator
URL: https://github.com/apache/incubator-mxnet/pull/13401#discussion_r262304587
 
 

 ##########
 File path: src/operator/contrib/corner_pooling.cu
 ##########
 @@ -0,0 +1,293 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you 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.
+ */
+/*!
+ * Copyright (c) 2018 by Contributors
+ * \file corner_pooling.cu
+ * \brief corner pooling operator
+ * \author Jiajie Tang
+*/
+
+#include "./corner_pooling-inl.h"
+
+#include <vector>
+
+#include "mshadow/cuda/tensor_gpu-inl.cuh"
+#include "mshadow/tensor.h"
+
+
+namespace mxnet {
+namespace op {
+
+template<typename DType>
+__global__ void CornerPoolingForwardTBKernel(const int count,
+        const DType *in_data, const int batch,
+        const int channel,  const int height,
+        const int width, DType *out_data,
+        int h_step, int h_start, int h_end) {
+  for (int index = (blockIdx.x + blockIdx.y*gridDim.x)*blockDim.x + threadIdx.x;
+       index < count;
+       index += blockDim.x * gridDim.x * gridDim.y) {
+    using mshadow::red::limits::MinValue;
+    const int b = index / (channel * width);
+    const int c = index % (channel * width) / width;
+    const int w = index % (channel * width) % width;
+    DType max_val = MinValue<DType>();
+    in_data += (b*channel + c)*height*width + w;
+    out_data += (b*channel + c)*height*width + w;
+
+    for (int h = h_start; h != h_end; h += h_step) {
+      const int index = h * width;
+      max_val = max_val > in_data[index] ? max_val : in_data[index];
+      out_data[index] = max_val;
+    }
+  }
+}
+
+
+template<typename DType>
+__global__ void CornerPoolingBackwardTBKernel(const int count,
+        const DType *out_data, const int batch,
+        const int channel,  const int height,
+        const int width, const DType *out_grad, DType *in_grad,
+        int h_step, int h_start, int h_end) {
+  for (int index = (blockIdx.x + blockIdx.y*gridDim.x)*blockDim.x + threadIdx.x;
+       index < count;
+       index += blockDim.x * gridDim.x * gridDim.y) {
+    const int b = index / (channel * width);
+    const int c = index % (channel * width) / width;
+    const int w = index % (channel * width) % width;
+
+    out_data += (b*channel + c)*height*width + w;
+    out_grad += (b*channel + c)*height*width + w;
+    in_grad += (b*channel + c)*height*width + w;
+
+    int max_h_idx = h_start;
+    for (int h = h_start; h != h_end; h += h_step) {
+      const int index = h * width;
+      if (out_data[index] != out_data[max_h_idx]) {
+        max_h_idx = index;
+      }
+      in_grad[max_h_idx] += out_grad[index];
+    }
+  }
+}
+
+
+template<typename DType>
+__global__ void CornerPoolingForwardLRKernel(const int count,
+        const DType *in_data, const int batch,
+        const int channel, const int height,
+        const int width, DType *out_data,
+        int w_step, int w_start, int w_end) {
+  for (int index = (blockIdx.x + blockIdx.y*gridDim.x)*blockDim.x + threadIdx.x;
+       index < count;
+       index += blockDim.x * gridDim.x * gridDim.y) {
+    using mshadow::red::limits::MinValue;
+    const int b = index / (channel * height);
+    const int c = index % (channel * height) / height;
+    const int h = index % (channel * height) % height;
+    DType max_val = MinValue<DType>();
+    in_data += ((b*channel + c)*height + h)*width;
+    out_data += ((b*channel + c)*height + h)*width;
+
+    for (int w = w_start; w != w_end; w += w_step) {
+      const int index = w;
+      max_val = max_val > in_data[index] ? max_val : in_data[index];
+      out_data[index] = max_val;
+    }
+  }
+}
+
+
+template<typename DType>
+__global__ void CornerPoolingBackwardLRKernel(const int count,
+        const DType *out_data, const int batch,
+        const int channel, const int height,
+        const int width, const DType *out_grad, DType *in_grad,
+        int w_step, int w_start, int w_end) {
+  for (int index = (blockIdx.x + blockIdx.y*gridDim.x)*blockDim.x + threadIdx.x;
+       index < count;
+       index += blockDim.x * gridDim.x * gridDim.y) {
+    const int b = index / (channel * height);
+    const int c = index % (channel * height) / height;
+    const int h = index % (channel * height) % height;
+
+    out_data += ((b*channel + c)*height + h)*width;
+    out_grad += ((b*channel + c)*height + h)*width;
+    in_grad += ((b*channel + c)*height + h)*width;
+
+    int max_w_idx = w_start;
+    for (int w = w_start; w != w_end; w += w_step) {
+      const int index = w;
+      if (out_data[index] != out_data[max_w_idx]) {
+        max_w_idx = index;
+      }
+      in_grad[max_w_idx] += out_grad[index];
+    }
+  }
+}
+
+
+
+template<typename DType>
+inline void corner_pool(mshadow::Stream<gpu> *s, const DType *in_data,
+        const TShape &ishape, const int corner_pooling_type,
+        OpReqType req_type, DType *out_data) {
+  using mshadow::red::limits::MinValue;
+  CHECK_EQ(req_type, kWriteTo)
+      << "Only support req=kWriteTo in pooling operations";
+  int height = ishape[2], width = ishape[3];
+  if (corner_pooling_type == 0 || corner_pooling_type == 1) {
+    // top or bottom
+    int h_end = 0, h_start = 0, h_step = 0;
+    if (corner_pooling_type == 0) {
+      h_step = -1;
+      h_start = height - 1;
+      h_end = -1;
+    } else {
+      h_step = +1;
+      h_start = 0;
+      h_end = height;
+    }
+    const int count = ishape[0] * ishape[1] * width;
+    const int gridSize = (count + mshadow::cuda::kMaxThreadsPerBlock - 1) /
+                         mshadow::cuda::kMaxThreadsPerBlock;
+    dim3 dimGrid(mshadow::cuda::kMaxGridDim,
+                 (gridSize + mshadow::cuda::kMaxGridDim - 1) /
+                  mshadow::cuda::kMaxGridDim);
+    dim3 dimBlock(mshadow::cuda::kMaxThreadsPerBlock);
+    mshadow::cuda::CheckLaunchParam(dimGrid,
+            dimBlock, "Corner Pooling Forward");
+    cudaStream_t stream = mshadow::Stream<gpu>::GetStream(s);
+    CornerPoolingForwardTBKernel<DType><<<  dimGrid, dimBlock, 0, stream>>>
 
 Review comment:
   can we add launch bound guards similar to https://github.com/apache/incubator-mxnet/pull/13188

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services