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/network.c | 2 ++
1 files changed, 2 insertions(+), 0 deletions(-)
diff --git a/src/network.c b/src/network.c
index 63b76a8..a62f6d0 100644
--- a/src/network.c
+++ b/src/network.c
@@ -862,6 +862,8 @@
if (l->xnor) {
//printf("\n %d \n", j);
size_t ldb_align = 256; // 256bit for AVX2
+ if (l->size*l->size*l->c > 4096) ldb_align = 4096;
+
binary_transpose_align_weights(l, ldb_align);
}
}
--
Gitblit v1.10.0