From 809f924db2823b9e1eaf3efb9370380edc1f76ed Mon Sep 17 00:00:00 2001
From: Joseph Redmon <pjreddie@gmail.com>
Date: Fri, 23 Jan 2015 00:38:24 +0000
Subject: [PATCH] CUDA so fast
---
src/col2im.c | 66 ---------------------------------
1 files changed, 0 insertions(+), 66 deletions(-)
diff --git a/src/col2im.c b/src/col2im.c
index f986071..0fb3db4 100644
--- a/src/col2im.c
+++ b/src/col2im.c
@@ -41,69 +41,3 @@
}
}
-
-#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 offset,
- int channels, int height, int width,
- int ksize, int stride, int pad, cl_mem data_im)
-{
- 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(offset), (void*) &offset);
- 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;
-
- cl.error = 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