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/col2im.c | 92 +++++++++++++++++++++++++++++++++++++++-------
1 files changed, 78 insertions(+), 14 deletions(-)
diff --git a/src/col2im.c b/src/col2im.c
index 0520567..65db22a 100644
--- a/src/col2im.c
+++ b/src/col2im.c
@@ -1,19 +1,21 @@
-inline void col2im_set_pixel(float *im, int height, int width, int channels,
- int row, int col, int channel, int pad, float val)
+#include <stdio.h>
+#include <math.h>
+inline void col2im_add_pixel(float *im, int height, int width, int channels,
+ int b, int row, int col, int channel, int pad, float val)
{
row -= pad;
col -= pad;
if (row < 0 || col < 0 ||
row >= height || col >= width) return;
- im[col + width*(row + channel*height)] = val;
+ im[col + width*(row + height*(channel+b*channels))] += val;
}
//This one might be too, can't remember.
-void col2im_cpu(float* data_col,
- const int batch, const int channels, const int height, const int width,
- const int ksize, const int stride, int pad, float* data_im)
+void col2im_cpu(float* data_col, int batch,
+ int channels, int height, int width,
+ int ksize, int stride, int pad, float* data_im)
{
- int c,h,w,b;
+ int b,c,h,w;
int height_col = (height - ksize) / stride + 1;
int width_col = (width - ksize) / stride + 1;
if (pad){
@@ -22,9 +24,8 @@
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(b = 0; b < batch; ++b){
for (c = 0; c < channels_col; ++c) {
int w_offset = c % ksize;
int h_offset = (c / ksize) % ksize;
@@ -33,15 +34,78 @@
for (w = 0; w < width_col; ++w) {
int im_row = h_offset + h * stride;
int im_col = w_offset + w * stride;
- double val = data_col[(c * height_col + h) * width_col + w];
- col2im_set_pixel(data_im, height, width, channels,
- im_row, im_col, c_im, pad, val);
+ int col_index = (c * height_col + h) * width_col + w + b*col_size;
+ double val = data_col[col_index];
+ col2im_add_pixel(data_im, height, width, channels,
+ b, im_row, im_col, c_im, pad, val);
}
}
}
- data_im += im_size;
- data_col+= col_size;
}
}
+#ifdef GPU
+
+#include "opencl.h"
+
+cl_kernel get_col2im_kernel()
+{
+ static int init = 0;
+ static cl_kernel im2col_kernel;
+ if(!init){
+ im2col_kernel = get_kernel("src/col2im.cl", "col2im", 0);
+ init = 1;
+ }
+ return im2col_kernel;
+}
+
+void col2im_ongpu(cl_mem data_col, int batch,
+ int channels, int height, int width,
+ int ksize, int stride, int pad, cl_mem data_im)
+{
+ cl_setup();
+ cl_kernel kernel = get_col2im_kernel();
+ cl_command_queue queue = cl.queue;
+
+ cl_uint i = 0;
+ cl.error = clSetKernelArg(kernel, i++, sizeof(data_col), (void*) &data_col);
+ cl.error = clSetKernelArg(kernel, i++, sizeof(batch), (void*) &batch);
+ cl.error = clSetKernelArg(kernel, i++, sizeof(channels), (void*) &channels);
+ cl.error = clSetKernelArg(kernel, i++, sizeof(height), (void*) &height);
+ cl.error = clSetKernelArg(kernel, i++, sizeof(width), (void*) &width);
+ cl.error = clSetKernelArg(kernel, i++, sizeof(ksize), (void*) &ksize);
+ cl.error = clSetKernelArg(kernel, i++, sizeof(stride), (void*) &stride);
+ cl.error = clSetKernelArg(kernel, i++, sizeof(pad), (void*) &pad);
+ cl.error = clSetKernelArg(kernel, i++, sizeof(data_im), (void*) &data_im);
+ check_error(cl);
+
+ size_t global_size = channels*height*width*batch;
+
+ clEnqueueNDRangeKernel(queue, kernel, 1, 0,
+ &global_size, 0, 0, 0, 0);
+ check_error(cl);
+}
+
+void col2im_gpu(float *data_col, int batch,
+ int channels, int height, int width,
+ int ksize, int stride, int pad, float *data_im)
+{
+ int height_col = (height - ksize) / stride + 1;
+ int width_col = (width - ksize) / stride + 1;
+ int channels_col = channels * ksize * ksize;
+
+ size_t size = height_col*width_col*channels_col*batch;
+ cl_mem col_gpu = cl_make_array(data_col, size);
+ size = channels*height*width*batch;
+ cl_mem im_gpu = cl_make_array(data_im, size);
+
+ col2im_ongpu(col_gpu, batch, channels, height, width,
+ ksize, stride, pad, im_gpu);
+
+ cl_read_array(im_gpu, data_im, size);
+ clReleaseMemObject(col_gpu);
+ clReleaseMemObject(im_gpu);
+}
+
+#endif
--
Gitblit v1.10.0