This is the mail archive of the
gcc-bugs@gcc.gnu.org
mailing list for the GCC project.
[Bug target/71300] Vector ABI bug for some AVX vectorized variants
- From: "jakub at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>
- To: gcc-bugs at gcc dot gnu dot org
- Date: Thu, 26 May 2016 18:35:32 +0000
- Subject: [Bug target/71300] Vector ABI bug for some AVX vectorized variants
- Auto-submitted: auto-generated
- References: <bug-71300-4 at http dot gcc dot gnu dot org/bugzilla/>
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71300
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |RESOLVED
CC| |jakub at gcc dot gnu.org
Resolution|--- |INVALID
--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
???
At least reading to 0.9.8 version of the ABI pdf, I see for YMM1 class:
sizeof (bytes) VLEN=2 VLEN=4 VLEN=8 VLEN=16
...
any-type * 4/8 1*MI128/ 1*MI128/ 2*MI128/ 4*MI128/
any-type (*)() 1*MI128 2*MI128 4*MI128 8*MI128
...
double 8 1*MD128 1*MD256 2*MD256 4*MD256
The characteristic type is double in this function, therefore for YMM1 class
(letter c) this function is VLEN=4, which means the first argument is passed
in __m256d (holding 4 doubles) and the second argument is passed in a pair of
__m128i (each holding 2 pointers).
This is because AVX generally doesn't support 256-bit integer vector
operations.
If you look at YMM2 class instead, pointers are passed in __m256i instead (for
VLEN=4 and above).