From 0f645836f193e75c4c3b718369e6fab15b5d19c5 Mon Sep 17 00:00:00 2001
From: Joseph Redmon <pjreddie@gmail.com>
Date: Wed, 11 Feb 2015 03:41:03 +0000
Subject: [PATCH] Detection is back, baby\!
---
src/softmax_layer.c | 75 ++-----------------------------------
1 files changed, 4 insertions(+), 71 deletions(-)
diff --git a/src/softmax_layer.c b/src/softmax_layer.c
index ffc028f..aa5ab06 100644
--- a/src/softmax_layer.c
+++ b/src/softmax_layer.c
@@ -1,5 +1,6 @@
#include "softmax_layer.h"
-#include "mini_blas.h"
+#include "blas.h"
+#include "cuda.h"
#include <float.h>
#include <math.h>
#include <stdlib.h>
@@ -15,8 +16,8 @@
layer->delta = calloc(inputs*batch, sizeof(float));
layer->jacobian = calloc(inputs*inputs*batch, sizeof(float));
#ifdef GPU
- layer->output_cl = cl_make_array(layer->output, inputs*batch);
- layer->delta_cl = cl_make_array(layer->delta, inputs*batch);
+ layer->output_gpu = cuda_make_array(layer->output, inputs*batch);
+ layer->delta_gpu = cuda_make_array(layer->delta, inputs*batch);
#endif
return layer;
}
@@ -49,71 +50,3 @@
}
}
-#ifdef GPU
-
-void pull_softmax_layer_output(const softmax_layer layer)
-{
- cl_read_array(layer.output_cl, layer.output, layer.inputs*layer.batch);
-}
-
-cl_kernel get_softmax_forward_kernel()
-{
- static int init = 0;
- static cl_kernel kernel;
- if(!init){
- kernel = get_kernel("src/softmax_layer.cl", "forward", 0);
- init = 1;
- }
- return kernel;
-}
-
-void forward_softmax_layer_gpu(const softmax_layer layer, cl_mem input)
-{
- cl_kernel kernel = get_softmax_forward_kernel();
- cl_command_queue queue = cl.queue;
-
- cl_uint i = 0;
- cl.error = clSetKernelArg(kernel, i++, sizeof(layer.inputs), (void*) &layer.inputs);
- cl.error = clSetKernelArg(kernel, i++, sizeof(input), (void*) &input);
- cl.error = clSetKernelArg(kernel, i++, sizeof(layer.output_cl), (void*) &layer.output_cl);
- check_error(cl);
-
- const size_t global_size[] = {layer.batch};
-
- cl.error = clEnqueueNDRangeKernel(queue, kernel, 1, 0, global_size, 0, 0, 0, 0);
- check_error(cl);
-
- /*
- cl_read_array(layer.output_cl, layer.output, layer.inputs*layer.batch);
- int z;
- for(z = 0; z < layer.inputs*layer.batch; ++z) printf("%f,",layer.output[z]);
- */
-}
-
-void backward_softmax_layer_gpu(const softmax_layer layer, cl_mem delta)
-{
- copy_ongpu(layer.batch*layer.inputs, layer.delta_cl, 1, delta, 1);
-}
-#endif
-
-/* This is if you want softmax w/o log-loss classification. You probably don't.
- int i,j,b;
- for(b = 0; b < layer.batch; ++b){
- for(i = 0; i < layer.inputs; ++i){
- for(j = 0; j < layer.inputs; ++j){
- int d = (i==j);
- layer.jacobian[b*layer.inputs*layer.inputs + i*layer.inputs + j] =
- layer.output[b*layer.inputs + i] * (d - layer.output[b*layer.inputs + j]);
- }
- }
- }
- for(b = 0; b < layer.batch; ++b){
- int M = layer.inputs;
- int N = 1;
- int K = layer.inputs;
- float *A = layer.jacobian + b*layer.inputs*layer.inputs;
- float *B = layer.delta + b*layer.inputs;
- float *C = delta + b*layer.inputs;
- gemm(0,0,M,N,K,1,A,K,B,N,0,C,N);
- }
- */
--
Gitblit v1.10.0