This is the mail archive of the
gcc-help@gcc.gnu.org
mailing list for the GCC project.
Re: Inline assembly without inputs considered const/pure?
- From: Andrew Haley <aph at redhat dot com>
- To: Matthias Pfaller <leo at marco dot de>, gcc-help at gcc dot gnu dot org
- Date: Thu, 25 Feb 2016 17:49:41 +0000
- Subject: Re: Inline assembly without inputs considered const/pure?
- Authentication-results: sourceware.org; auth=none
- References: <56CC165A dot 6080704 at marco dot de> <56CC74F3 dot 40304 at marco dot de> <20160223201135 dot GB31896 at unpythonic dot net> <56CD4D70 dot 2060307 at marco dot de> <56CF1F81 dot 3000409 at marco dot de> <56CF2063 dot 9080405 at redhat dot com> <56CF25BA dot 3060501 at marco dot de> <56CF2A4D dot 8050000 at redhat dot com> <56CF36C9 dot 1060309 at marco dot de>
On 02/25/2016 05:15 PM, Matthias Pfaller wrote:
> So that's not a solution I would like to use. I could use "i" (&poo)
> (which would yeld an #poo to get discarded), but I does that count
> as a memory reference?
It's tough to persuade GCC that data magically appears from nowhere,
but it's not so volatile that you have to read it (as in a
memory-mapped peripheral, where register must be read because they
have side effects). Anything I tell you know beyond the most
straightforward interpretation might break in the future, or indeed
the past. I can tell you that GCC 6 on AArch64 doesn't need the
volatile memory input. Maybe some other version will.
Andrew.