From cad4d1618fee74471d335314cb77070fee951a42 Mon Sep 17 00:00:00 2001
From: AlexeyAB <alexeyab84@gmail.com>
Date: Sun, 25 Feb 2018 13:29:44 +0000
Subject: [PATCH] Added support for Tensor Cores CC >= 7.0 (V100). For FP16/32 (mixed precision) define CUDNN_HALF should be used.
---
src/network_kernels.cu | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)
diff --git a/src/network_kernels.cu b/src/network_kernels.cu
index 6090bb0..503a1b8 100644
--- a/src/network_kernels.cu
+++ b/src/network_kernels.cu
@@ -117,7 +117,7 @@
int i;
for (i = 0; i < net.n; ++i) {
layer l = net.layers[i];
- cuda_convert_f32_to_f16(l.weights_gpu, l.c*l.n*l.size*l.size, (half *)l.weights_gpu16);
+ cuda_convert_f32_to_f16(l.weights_gpu, l.c*l.n*l.size*l.size, l.weights_gpu16);
}
#endif
forward_network_gpu(net, state);
--
Gitblit v1.10.0