[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Discuss-gnuradio] [VOLK] Test-case behaviour
From: |
Martin Braun |
Subject: |
Re: [Discuss-gnuradio] [VOLK] Test-case behaviour |
Date: |
Thu, 12 Nov 2015 10:48:36 -0800 |
User-agent: |
Mozilla/5.0 (X11; Linux x86_64; rv:38.0) Gecko/20100101 Thunderbird/38.3.0 |
On 11.11.2015 19:29, Tom Rondeau wrote:
> Ah, interesting. This is something we should take up in the VOLK working
> group. We don't claim or not claim that you can use the same memory for
> the input and output buffers. Some kernels are known to work with these
> inplace operations. Looks like this one doesn't. Perhaps just testing
> and knowing which kernels do and do not support this and protect those
> that do not as const (which can be easily gotten around) and saying as
> much in the kernel's documentation.
This came up at one of the recent dev calls, and we agreed that it
*should* work, and that we'll put it into the docs. If we can
document/const this on a per-kernel basis, that's even better. However,
we need to make sure that adding new kernels won't change this
behaviour; i.e. say we have a function that we believe is safe for
in-place ops, and then we add a NEON kernel, and it stops working, that
would be confusing.
Cheers,
Martin