This is the mail archive of the
mailing list for the GCC project.
Re: [RFC, VECTOR ABI] Allow __attribute__((vector)) in GCC by default.
- From: Jeff Law <law at redhat dot com>
- To: Ramana Radhakrishnan <ramana dot gcc at googlemail dot com>, Joseph Myers <joseph at codesourcery dot com>
- Cc: Kirill Yukhin <kirill dot yukhin at gmail dot com>, Jakub Jelinek <jakub at redhat dot com>, "gcc at gcc dot gnu dot org" <gcc at gcc dot gnu dot org>, GNU C Library <libc-alpha at sourceware dot org>
- Date: Wed, 7 Oct 2015 12:17:41 -0600
- Subject: Re: [RFC, VECTOR ABI] Allow __attribute__((vector)) in GCC by default.
- Authentication-results: sourceware.org; auth=none
- References: <20151005130733 dot GB62312 at msticlxl57 dot ims dot intel dot com> <alpine dot DEB dot 2 dot 10 dot 1510051320120 dot 16355 at digraph dot polyomino dot org dot uk> <561551B0 dot 70507 at redhat dot com> <alpine dot DEB dot 2 dot 10 dot 1510071719530 dot 29388 at digraph dot polyomino dot org dot uk> <CAJA7tRZr-d5J7Fdgo6WV3gWGNuVj62UnxNgf=pB1VxhK3-gZRg at mail dot gmail dot com>
On 10/07/2015 11:34 AM, Ramana Radhakrishnan wrote:
ISTM it should be a nop in those cases and yes, documenting and testing
those cases seems wise. Anything else is just asking for trouble. So
Kirill, some simple tests for those cases seems advisable.
On Wed, Oct 7, 2015 at 6:22 PM, Joseph Myers <firstname.lastname@example.org> wrote:
On Wed, 7 Oct 2015, Jeff Law wrote:
I'm not sure why this attribute isn't documented, but clearly that should be
I assume the reasoning was: we document support for Cilk+ (and what's
included in Cilk+ is externally documented). When it becomes a generic
GNU C attribute, that reasoning no longer applies.
I do not understand enough of this feature but since this is becoming
a GNU C attribute, what happens with functions which are marked
__attribute__((vector)) but are compiled for platforms that do not
support vector clones or have options to compile for soft float - what
is the expected behaviour for the compiler in those situations ?
Presumably the __attribute__((vector)) has no effect ? Might also be
useful to document that.
ISTM that in an ideal world, for all targets with vector hardware that
we ought to work towards getting clone support and vector entry points
in the math library. Not sure where that'll land on everyone's priority