From a284a7da8d1facbf984a22302665a2a50295a687 Mon Sep 17 00:00:00 2001
From: AlexeyAB <alexeyab84@gmail.com>
Date: Wed, 08 Aug 2018 16:08:58 +0000
Subject: [PATCH] Try to use avx_hs() - slow and requires alignment 4096 bits < (l.size*l.size*l.c) May be faster only from 8192 bits and more.
---
src/gru_layer.c | 8 ++++++++
1 files changed, 8 insertions(+), 0 deletions(-)
diff --git a/src/gru_layer.c b/src/gru_layer.c
index 4c720ce..b78e868 100644
--- a/src/gru_layer.c
+++ b/src/gru_layer.c
@@ -85,7 +85,15 @@
l.z_cpu = calloc(outputs*batch, sizeof(float));
l.h_cpu = calloc(outputs*batch, sizeof(float));
+ l.forward = forward_gru_layer;
+ l.backward = backward_gru_layer;
+ l.update = update_gru_layer;
+
#ifdef GPU
+ l.forward_gpu = forward_gru_layer_gpu;
+ l.backward_gpu = backward_gru_layer_gpu;
+ l.update_gpu = update_gru_layer_gpu;
+
l.forgot_state_gpu = cuda_make_array(l.output, batch*outputs);
l.forgot_delta_gpu = cuda_make_array(l.output, batch*outputs);
l.prev_state_gpu = cuda_make_array(l.output, batch*outputs);
--
Gitblit v1.10.0