From 9d418102f4a44b2e2c437dec945189a646cbf3a4 Mon Sep 17 00:00:00 2001
From: Joseph Redmon <pjreddie@gmail.com>
Date: Sat, 21 Mar 2015 21:17:39 +0000
Subject: [PATCH] using caffe's im2col, it's so much better\!
---
src/convolutional_kernels.cu | 34 +++++++++++++++++-----------------
1 files changed, 17 insertions(+), 17 deletions(-)
diff --git a/src/convolutional_kernels.cu b/src/convolutional_kernels.cu
index 864d7fa..18a3b7d 100644
--- a/src/convolutional_kernels.cu
+++ b/src/convolutional_kernels.cu
@@ -56,7 +56,7 @@
extern "C" void forward_convolutional_layer_gpu(convolutional_layer layer, network_state state)
{
-clock_t time = clock();
+//clock_t time = clock();
int i;
int m = layer.n;
int k = layer.size*layer.size*layer.c;
@@ -64,31 +64,31 @@
convolutional_out_width(layer);
bias_output_gpu(layer.output_gpu, layer.biases_gpu, layer.batch, layer.n, n);
-cudaDeviceSynchronize();
-printf("bias %f\n", sec(clock() - time));
-time = clock();
+//cudaDeviceSynchronize();
+//printf("bias %f\n", sec(clock() - time));
+//time = clock();
-float imt=0;
-float gemt = 0;
+//float imt=0;
+//float gemt = 0;
for(i = 0; i < layer.batch; ++i){
-time = clock();
+//time = clock();
im2col_ongpu(state.input + i*layer.c*layer.h*layer.w, layer.c, layer.h, layer.w, layer.size, layer.stride, layer.pad, layer.col_image_gpu);
-cudaDeviceSynchronize();
-imt += sec(clock()-time);
-time = clock();
+//cudaDeviceSynchronize();
+//imt += sec(clock()-time);
+//time = clock();
float * a = layer.filters_gpu;
float * b = layer.col_image_gpu;
float * c = layer.output_gpu;
gemm_ongpu(0,0,m,n,k,1.,a,k,b,n,1.,c+i*m*n,n);
-cudaDeviceSynchronize();
-gemt += sec(clock()-time);
-time = clock();
+//cudaDeviceSynchronize();
+//gemt += sec(clock()-time);
+//time = clock();
}
activate_array_ongpu(layer.output_gpu, m*n*layer.batch, layer.activation);
-cudaDeviceSynchronize();
-printf("activate %f\n", sec(clock() - time));
-printf("im2col %f\n", imt);
-printf("gemm %f\n", gemt);
+//cudaDeviceSynchronize();
+//printf("activate %f\n", sec(clock() - time));
+//printf("im2col %f\n", imt);
+//printf("gemm %f\n", gemt);
}
extern "C" void backward_convolutional_layer_gpu(convolutional_layer layer, network_state state)
--
Gitblit v1.10.0