From bfffadc75502cadb5d05909435a2167db5204325 Mon Sep 17 00:00:00 2001
From: Joseph Redmon <pjreddie@gmail.com>
Date: Wed, 04 Feb 2015 20:41:20 +0000
Subject: [PATCH] Stable place to commit
---
src/convolutional_kernels.cu | 23 ++++++++++++++---------
1 files changed, 14 insertions(+), 9 deletions(-)
diff --git a/src/convolutional_kernels.cu b/src/convolutional_kernels.cu
index 6461aff..8645fbf 100644
--- a/src/convolutional_kernels.cu
+++ b/src/convolutional_kernels.cu
@@ -28,11 +28,11 @@
check_error(cudaPeekAtLastError());
}
-__global__ void learn_bias(int batch, int n, int size, float *delta, float *bias_updates)
+__global__ void learn_bias(int batch, int n, int size, float *delta, float *bias_updates, float scale)
{
__shared__ float part[BLOCK];
int i,b;
- int filter = (blockIdx.x + blockIdx.y*gridDim.x);
+ int filter = blockIdx.x;
int p = threadIdx.x;
float sum = 0;
for(b = 0; b < batch; ++b){
@@ -44,16 +44,16 @@
part[p] = sum;
__syncthreads();
if(p == 0){
- for(i = 0; i < BLOCK; ++i) bias_updates[filter] += part[i];
+ for(i = 0; i < BLOCK; ++i) bias_updates[filter] += scale * part[i];
}
}
extern "C" void learn_bias_convolutional_layer_ongpu(convolutional_layer layer)
{
int size = convolutional_out_height(layer)*convolutional_out_width(layer);
+ float alpha = 1./layer.batch;
-
- learn_bias<<<cuda_gridsize(layer.n), BLOCK>>>(layer.batch, layer.n, size, layer.delta_gpu, layer.bias_updates_gpu);
+ learn_bias<<<layer.n, BLOCK>>>(layer.batch, layer.n, size, layer.delta_gpu, layer.bias_updates_gpu, alpha);
check_error(cudaPeekAtLastError());
}
@@ -96,13 +96,11 @@
gemm_ongpu(0,0,m,n,k,1.,a,k,b,n,1.,c+i*m*n,n);
}
activate_array_ongpu(layer.output_gpu, m*n*layer.batch, layer.activation);
- cuda_pull_array(layer.output_gpu, layer.output, m*n*layer.batch);
- //for(i = 0; i < m*n*layer.batch; ++i) printf("%f, ", layer.output[i]);
- //printf("\n");
}
extern "C" void backward_convolutional_layer_gpu(convolutional_layer layer, float *in, float *delta_gpu)
{
+ float alpha = 1./layer.batch;
int i;
int m = layer.n;
int n = layer.size*layer.size*layer.c;
@@ -119,7 +117,7 @@
float * c = layer.filter_updates_gpu;
im2col_ongpu(in, i*layer.c*layer.h*layer.w, layer.c, layer.h, layer.w, layer.size, layer.stride, layer.pad, layer.col_image_gpu);
- gemm_ongpu(0,1,m,n,k,1,a + i*m*k,k,b,k,1,c,n);
+ gemm_ongpu(0,1,m,n,k,alpha,a + i*m*k,k,b,k,1,c,n);
if(delta_gpu){
@@ -153,6 +151,13 @@
extern "C" void update_convolutional_layer_gpu(convolutional_layer layer)
{
int size = layer.size*layer.size*layer.c*layer.n;
+
+/*
+ cuda_pull_array(layer.filter_updates_gpu, layer.filter_updates, size);
+ cuda_pull_array(layer.filters_gpu, layer.filters, size);
+ printf("Filter: %f updates: %f\n", mag_array(layer.filters, size), layer.learning_rate*mag_array(layer.filter_updates, size));
+ */
+
axpy_ongpu(layer.n, layer.learning_rate, layer.bias_updates_gpu, 1, layer.biases_gpu, 1);
scal_ongpu(layer.n,layer.momentum, layer.bias_updates_gpu, 1);
--
Gitblit v1.10.0