From a0ada4fc9e0075d8ba57f128a111fd5e214266c8 Mon Sep 17 00:00:00 2001
From: Jeff Donahue <jeff.donahue@gmail.com>
Date: Sat, 1 Nov 2014 21:22:11 -0700
Subject: [PATCH] Unroll kernels in SoftmaxLayer...from terrible performance to
 mediocre performance.

---
 src/caffe/layers/softmax_layer.cu | 48 +++++++++++++++++++--------------------
 1 file changed, 24 insertions(+), 24 deletions(-)

diff --git a/src/caffe/layers/softmax_layer.cu b/src/caffe/layers/softmax_layer.cu
index 292ad2b..6b8871a 100644
--- a/src/caffe/layers/softmax_layer.cu
+++ b/src/caffe/layers/softmax_layer.cu
@@ -25,14 +25,13 @@ __global__ void kernel_channel_max(const int num, const int channels,
 }
 
 template <typename Dtype>
-__global__ void kernel_channel_subtract(const int num, const int channels,
-    const int spatial_dim, Dtype* data, const Dtype* channel_max) {
-  CUDA_KERNEL_LOOP(index, num * spatial_dim) {
-    int n = index / spatial_dim;
+__global__ void kernel_channel_subtract(const int count,
+    const int num, const int channels,
+    const int spatial_dim, const Dtype* channel_max, Dtype* data) {
+  CUDA_KERNEL_LOOP(index, count) {
+    int n = index / channels / spatial_dim;
     int s = index % spatial_dim;
-    for (int c = 0; c < channels; ++c) {
-      data[(n * channels + c) * spatial_dim + s] -= channel_max[index];
-    }
+    data[index] -= channel_max[n * spatial_dim + s];
   }
 }
 
@@ -58,14 +57,13 @@ __global__ void kernel_channel_sum(const int num, const int channels,
 }
 
 template <typename Dtype>
-__global__ void kernel_channel_div(const int num, const int channels,
-    const int spatial_dim, Dtype* data, const Dtype* channel_sum) {
-  CUDA_KERNEL_LOOP(index, num * spatial_dim) {
-    int n = index / spatial_dim;
+__global__ void kernel_channel_div(const int count,
+    const int num, const int channels,
+    const int spatial_dim, const Dtype* channel_sum, Dtype* data) {
+  CUDA_KERNEL_LOOP(index, count) {
+    int n = index / channels / spatial_dim;
     int s = index % spatial_dim;
-    for (int c = 0; c < channels; ++c) {
-      data[(n * channels + c) * spatial_dim + s] /= channel_sum[index];
-    }
+    data[index] /= channel_sum[n * spatial_dim + s];
   }
 }
 
@@ -91,10 +89,11 @@ void SoftmaxLayer<Dtype>::Forward_gpu(const vector<Blob<Dtype>*>& bottom,
   const Dtype* bottom_data = bottom[0]->gpu_data();
   Dtype* top_data = top[0]->mutable_gpu_data();
   Dtype* scale_data = scale_.mutable_gpu_data();
+  int count = bottom[0]->count();
   int num = bottom[0]->num();
   int channels = bottom[0]->channels();
   int spatial_dim = bottom[0]->height() * bottom[0]->width();
-  caffe_copy(bottom[0]->count(), bottom_data, top_data);
+  caffe_copy(count, bottom_data, top_data);
   // We need to subtract the max to avoid numerical issues, compute the exp,
   // and then normalize.
   // compute max
@@ -104,9 +103,9 @@ void SoftmaxLayer<Dtype>::Forward_gpu(const vector<Blob<Dtype>*>& bottom,
       scale_data);
   // subtract
   // NOLINT_NEXT_LINE(whitespace/operators)
-  kernel_channel_subtract<Dtype><<<CAFFE_GET_BLOCKS(num * spatial_dim),
-      CAFFE_CUDA_NUM_THREADS>>>(num, channels, spatial_dim, top_data,
-      scale_data);
+  kernel_channel_subtract<Dtype><<<CAFFE_GET_BLOCKS(count),
+      CAFFE_CUDA_NUM_THREADS>>>(count, num, channels, spatial_dim,
+      scale_data, top_data);
   // exponentiate
   // NOLINT_NEXT_LINE(whitespace/operators)
   kernel_exp<Dtype><<<CAFFE_GET_BLOCKS(num * channels * spatial_dim),
@@ -119,9 +118,9 @@ void SoftmaxLayer<Dtype>::Forward_gpu(const vector<Blob<Dtype>*>& bottom,
       scale_data);
   // divide
   // NOLINT_NEXT_LINE(whitespace/operators)
-  kernel_channel_div<Dtype><<<CAFFE_GET_BLOCKS(num * spatial_dim),
-      CAFFE_CUDA_NUM_THREADS>>>(num, channels, spatial_dim, top_data,
-      scale_data);
+  kernel_channel_div<Dtype><<<CAFFE_GET_BLOCKS(count),
+      CAFFE_CUDA_NUM_THREADS>>>(count, num, channels, spatial_dim,
+      scale_data, top_data);
 }
 
 template <typename Dtype>
@@ -131,6 +130,7 @@ void SoftmaxLayer<Dtype>::Backward_gpu(const vector<Blob<Dtype>*>& top,
   const Dtype* top_data = top[0]->gpu_data();
   Dtype* bottom_diff = bottom[0]->mutable_gpu_diff();
   Dtype* scale_data = scale_.mutable_gpu_data();
+  int count = top[0]->count();
   int num = top[0]->num();
   int channels = top[0]->channels();
   int spatial_dim = top[0]->height() * top[0]->width();
@@ -141,9 +141,9 @@ void SoftmaxLayer<Dtype>::Backward_gpu(const vector<Blob<Dtype>*>& top,
       CAFFE_CUDA_NUM_THREADS>>>(num, channels, spatial_dim, top_diff, top_data,
       scale_data);
   // NOLINT_NEXT_LINE(whitespace/operators)
-  kernel_channel_subtract<Dtype><<<CAFFE_GET_BLOCKS(num * spatial_dim),
-      CAFFE_CUDA_NUM_THREADS>>>(num, channels, spatial_dim, bottom_diff,
-      scale_data);
+  kernel_channel_subtract<Dtype><<<CAFFE_GET_BLOCKS(count),
+      CAFFE_CUDA_NUM_THREADS>>>(count, num, channels, spatial_dim,
+      scale_data, bottom_diff);
   // elementwise multiplication
   caffe_gpu_mul<Dtype>(top[0]->count(), bottom_diff, top_data, bottom_diff);
 }
-- 
2.7.4