diff options
author | Mariusz Moczala <mariusz.moczala@intel.com> | 2016-01-20 09:28:35 +0100 |
---|---|---|
committer | Mariusz Moczala <mariusz.moczala@intel.com> | 2016-01-20 09:28:35 +0100 |
commit | 581c1cc3fd6c04640c4b89e5ed003a40cd67e855 (patch) | |
tree | 63e74f8a09201e668c3da62403abf843ff9eac98 /src/caffe/util/im2col.cpp | |
parent | cff6f3d997616fa1201923fbfde77d24d0d395ad (diff) | |
download | caffeonacl-581c1cc3fd6c04640c4b89e5ed003a40cd67e855.tar.gz caffeonacl-581c1cc3fd6c04640c4b89e5ed003a40cd67e855.tar.bz2 caffeonacl-581c1cc3fd6c04640c4b89e5ed003a40cd67e855.zip |
Performance related update of im2col() and col2im() functions
Diffstat (limited to 'src/caffe/util/im2col.cpp')
-rw-r--r-- | src/caffe/util/im2col.cpp | 93 |
1 files changed, 61 insertions, 32 deletions
diff --git a/src/caffe/util/im2col.cpp b/src/caffe/util/im2col.cpp index 6e5ea875..114a86cb 100644 --- a/src/caffe/util/im2col.cpp +++ b/src/caffe/util/im2col.cpp @@ -5,6 +5,16 @@ namespace caffe { +// Function uses casting from int to unsigned to compare if value of +// parameter a is greater or equal to zero and lower than value of +// parameter b. The b parameter is of type signed and is always positive, +// therefore its value is always lower than 0x800... where casting +// negative value of a parameter converts it to value higher than 0x800... +// The casting allows to use one condition instead of two. +inline bool is_a_ge_zero_and_a_lt_b(int a, int b) { + return static_cast<unsigned>(a) < static_cast<unsigned>(b); +} + template <typename Dtype> void im2col_cpu(const Dtype* data_im, const int channels, const int height, const int width, const int kernel_h, const int kernel_w, @@ -12,22 +22,33 @@ void im2col_cpu(const Dtype* data_im, const int channels, const int stride_h, const int stride_w, const int dilation_h, const int dilation_w, Dtype* data_col) { - const int height_col = (height + 2 * pad_h - - (dilation_h * (kernel_h - 1) + 1)) / stride_h + 1; - const int width_col = (width + 2 * pad_w - - (dilation_w * (kernel_w - 1) + 1)) / stride_w + 1; - const int channels_col = channels * kernel_h * kernel_w; - for (int c_col = 0; c_col < channels_col; ++c_col) { - int w_offset = c_col % kernel_w; - int h_offset = (c_col / kernel_w) % kernel_h; - int c_im = c_col / kernel_h / kernel_w; - for (int h_col = 0; h_col < height_col; ++h_col) { - for (int w_col = 0; w_col < width_col; ++w_col) { - int h_im = h_col * stride_h - pad_h + h_offset * dilation_h; - int w_im = w_col * stride_w - pad_w + w_offset * dilation_w; - data_col[(c_col * height_col + h_col) * width_col + w_col] = - (h_im >= 0 && w_im >= 0 && h_im < height && w_im < width) ? - data_im[(c_im * height + h_im) * width + w_im] : 0; + const int output_h = (height + 2 * pad_h - + (dilation_h * (kernel_h - 1) + 1)) / stride_h + 1; + const int output_w = (width + 2 * pad_w - + (dilation_w * (kernel_w - 1) + 1)) / stride_w + 1; + const int channel_size = height * width; + for (int channel = channels; channel--; data_im += channel_size) { + for (int kernel_row = 0; kernel_row < kernel_h; kernel_row++) { + for (int kernel_col = 0; kernel_col < kernel_w; kernel_col++) { + int input_row = -pad_h + kernel_row * dilation_h; + for (int output_rows = output_h; output_rows; output_rows--) { + if (!is_a_ge_zero_and_a_lt_b(input_row, height)) { + for (int output_cols = output_w; output_cols; output_cols--) { + *(data_col++) = 0; + } + } else { + int input_col = -pad_w + kernel_col * dilation_w; + for (int output_col = output_w; output_col; output_col--) { + if (is_a_ge_zero_and_a_lt_b(input_col, width)) { + *(data_col++) = data_im[input_row * width + input_col]; + } else { + *(data_col++) = 0; + } + input_col += stride_w; + } + } + input_row += stride_h; + } } } } @@ -146,22 +167,30 @@ void col2im_cpu(const Dtype* data_col, const int channels, const int dilation_h, const int dilation_w, Dtype* data_im) { caffe_set(height * width * channels, Dtype(0), data_im); - const int height_col = (height + 2 * pad_h - - (dilation_h * (kernel_h - 1) + 1)) / stride_h + 1; - const int width_col = (width + 2 * pad_w - - (dilation_w * (kernel_w - 1) + 1)) / stride_w + 1; - const int channels_col = channels * kernel_h * kernel_w; - for (int c_col = 0; c_col < channels_col; ++c_col) { - int w_offset = c_col % kernel_w; - int h_offset = (c_col / kernel_w) % kernel_h; - int c_im = c_col / kernel_h / kernel_w; - for (int h_col = 0; h_col < height_col; ++h_col) { - for (int w_col = 0; w_col < width_col; ++w_col) { - int h_im = h_col * stride_h - pad_h + h_offset * dilation_h; - int w_im = w_col * stride_w - pad_w + w_offset * dilation_w; - if (h_im >= 0 && h_im < height && w_im >= 0 && w_im < width) - data_im[(c_im * height + h_im) * width + w_im] += - data_col[(c_col * height_col + h_col) * width_col + w_col]; + const int output_h = (height + 2 * pad_h - + (dilation_h * (kernel_h - 1) + 1)) / stride_h + 1; + const int output_w = (width + 2 * pad_w - + (dilation_w * (kernel_w - 1) + 1)) / stride_w + 1; + const int channel_size = height * width; + for (int channel = channels; channel--; data_im += channel_size) { + for (int kernel_row = 0; kernel_row < kernel_h; kernel_row++) { + for (int kernel_col = 0; kernel_col < kernel_w; kernel_col++) { + int input_row = -pad_h + kernel_row * dilation_h; + for (int output_rows = output_h; output_rows; output_rows--) { + if (!is_a_ge_zero_and_a_lt_b(input_row, height)) { + data_col += output_w; + } else { + int input_col = -pad_w + kernel_col * dilation_w; + for (int output_col = output_w; output_col; output_col--) { + if (is_a_ge_zero_and_a_lt_b(input_col, width)) { + data_im[input_row * width + input_col] += *data_col; + } + data_col++; + input_col += stride_w; + } + } + input_row += stride_h; + } } } } |