[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Bug-gnubg] Vectorization of neural nets just commited into CVS
From: |
Jim Segrave |
Subject: |
Re: [Bug-gnubg] Vectorization of neural nets just commited into CVS |
Date: |
Wed, 4 May 2005 19:16:04 +0200 |
User-agent: |
Mutt/1.4.2.1i |
> That's strange as in NeuralNetCreate() arHiddenWeight is using
> sse_malloc() which allocs this to a 16 byte boundary (using _mm_malloc).
>
> Try breaking in NeuralNetCreate() and checking the value arHiddenWeight
> gets set to.
>
> It may be easier to debug if you omit the -O3 optimisation (if it still
> works/goes wrong).
I've fixed it by avoiding calling NeraulNetCreateDirect if using sse,
as having mmap around for other things may be a good thing and for non
sse architectures, we can leave mmaping the weights
I also added a CVS tag to sse.h for completeness' sake
--
Jim Segrave address@hidden