From edbccdfcaf46f11e631afe98796f3e6e170da5d0 Mon Sep 17 00:00:00 2001
From: Joseph Redmon <pjreddie@gmail.com>
Date: Sun, 26 Oct 2014 05:04:34 +0000
Subject: [PATCH] Maybe something changed?
---
src/im2col.c | 72 +++++++++++++++++++++--------------
1 files changed, 43 insertions(+), 29 deletions(-)
diff --git a/src/im2col.c b/src/im2col.c
index 899f73a..b743e34 100644
--- a/src/im2col.c
+++ b/src/im2col.c
@@ -1,32 +1,47 @@
#include "mini_blas.h"
+#include <stdio.h>
+inline float im2col_get_pixel(float *im, int height, int width, int channels,
+ int b, int row, int col, int channel, int pad)
+{
+ row -= pad;
+ col -= pad;
+
+ if (row < 0 || col < 0 ||
+ row >= height || col >= width) return 0;
+ return im[col + width*(row + height*(channel+b*channels))];
+}
//From Berkeley Vision's Caffe!
//https://github.com/BVLC/caffe/blob/master/LICENSE
-void im2col_cpu(float* data_im,
- const int batch, const int channels, const int height, const int width,
- const int ksize, const int stride, float* data_col)
+void im2col_cpu(float* data_im, int batch,
+ int channels, int height, int width,
+ int ksize, int stride, int pad, float* data_col)
{
int c,h,w,b;
int height_col = (height - ksize) / stride + 1;
int width_col = (width - ksize) / stride + 1;
+ if (pad){
+ height_col = 1 + (height-1) / stride;
+ width_col = 1 + (width-1) / stride;
+ pad = ksize/2;
+ }
int channels_col = channels * ksize * ksize;
- int im_size = height*width*channels;
int col_size = height_col*width_col*channels_col;
- for(b = 0; b < batch; ++b){
- for ( c = 0; c < channels_col; ++c) {
+ for (b = 0; b < batch; ++b) {
+ for (c = 0; c < channels_col; ++c) {
int w_offset = c % ksize;
int h_offset = (c / ksize) % ksize;
int c_im = c / ksize / ksize;
- for ( h = 0; h < height_col; ++h) {
- for ( w = 0; w < width_col; ++w) {
- data_col[(c * height_col + h) * width_col + w] =
- data_im[(c_im * height + h * stride + h_offset) * width
- + w * stride + w_offset];
+ for (h = 0; h < height_col; ++h) {
+ for (w = 0; w < width_col; ++w) {
+ int im_row = h_offset + h * stride;
+ int im_col = w_offset + w * stride;
+ int col_index = (c * height_col + h) * width_col + w + b*col_size;
+ data_col[col_index] = im2col_get_pixel(data_im, height, width, channels,
+ b, im_row, im_col, c_im, pad);
}
}
}
- data_im += im_size;
- data_col+= col_size;
}
}
@@ -48,9 +63,9 @@
}
-void im2col_ongpu(cl_mem data_im, const int batch,
- const int channels, const int height, const int width,
- const int ksize, const int stride, cl_mem data_col)
+void im2col_ongpu(cl_mem data_im, int batch,
+ int channels, int height, int width,
+ int ksize, int stride, int pad, cl_mem data_col)
{
cl_setup();
cl_kernel im2col_kernel = get_im2col_kernel();
@@ -64,29 +79,28 @@
cl.error = clSetKernelArg(im2col_kernel, i++, sizeof(width), (void*) &width);
cl.error = clSetKernelArg(im2col_kernel, i++, sizeof(ksize), (void*) &ksize);
cl.error = clSetKernelArg(im2col_kernel, i++, sizeof(stride), (void*) &stride);
+ cl.error = clSetKernelArg(im2col_kernel, i++, sizeof(pad), (void*) &pad);
cl.error = clSetKernelArg(im2col_kernel, i++, sizeof(data_col), (void*) &data_col);
check_error(cl);
int height_col = (height - ksize) / stride + 1;
int width_col = (width - ksize) / stride + 1;
int channels_col = channels * ksize * ksize;
+ if (pad){
+ height_col = 1 + (height-1) / stride;
+ width_col = 1 + (width-1) / stride;
+ }
- size_t global_size[2];
- size_t local_size[2];
- global_size[0] = batch;
- global_size[1] = channels_col;
- local_size[0] = height_col;
- local_size[1] = width_col;
+ size_t global_size = batch*channels_col*height_col*width_col;
- clEnqueueNDRangeKernel(queue, im2col_kernel, 2, 0,
- global_size, local_size, 0, 0, 0);
+ clEnqueueNDRangeKernel(queue, im2col_kernel, 1, 0,
+ &global_size, 0, 0, 0, 0);
check_error(cl);
}
-void im2col_gpu(float *data_im,
- const int batch, const int channels, const int height, const int width,
- const int ksize, const int stride,
- float *data_col)
+void im2col_gpu(float *data_im, int batch,
+ int channels, int height, int width,
+ int ksize, int stride, int pad, float *data_col)
{
cl_setup();
cl_context context = cl.context;
@@ -109,7 +123,7 @@
check_error(cl);
im2col_ongpu(im_gpu, batch, channels, height, width,
- ksize, stride, col_gpu);
+ ksize, stride, pad, col_gpu);
clEnqueueReadBuffer(queue, col_gpu, CL_TRUE, 0, size, data_col, 0, 0, 0);
check_error(cl);
--
Gitblit v1.10.0