From 0f7f2899b65343e56b0a1188f703d459d824d398 Mon Sep 17 00:00:00 2001
From: Joseph Redmon <pjreddie@gmail.com>
Date: Mon, 16 Nov 2015 03:51:26 +0000
Subject: [PATCH] Fix for cuda 7.5
---
src/softmax_layer_kernels.cu | 8 +++++---
1 files changed, 5 insertions(+), 3 deletions(-)
diff --git a/src/softmax_layer_kernels.cu b/src/softmax_layer_kernels.cu
index 0529f75..6637180 100644
--- a/src/softmax_layer_kernels.cu
+++ b/src/softmax_layer_kernels.cu
@@ -1,11 +1,13 @@
+#include "cuda_runtime.h"
+#include "curand.h"
+#include "cublas_v2.h"
+
extern "C" {
#include "softmax_layer.h"
#include "cuda.h"
#include "blas.h"
}
-#define BLOCK 256
-
__global__ void forward_softmax_layer_kernel(int n, int batch, float *input, float *output)
{
int b = (blockIdx.x + blockIdx.y*gridDim.x) * blockDim.x + threadIdx.x;
@@ -42,7 +44,7 @@
extern "C" void backward_softmax_layer_gpu(const softmax_layer layer, network_state state)
{
- copy_ongpu(layer.batch*layer.inputs, layer.delta_gpu, 1, state.delta, 1);
+ axpy_ongpu(layer.batch*layer.inputs, 1, layer.delta_gpu, 1, state.delta, 1);
}
/* This is if you want softmax w/o log-loss classification. You probably don't.
--
Gitblit v1.10.0