[Bug tree-optimization/88531] Index data types when targeting AVX-512 vectorization with gather/scatter
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88531 H.J. Lu changed: What|Removed |Added Target Milestone|--- |12.0 Resolution|--- |FIXED Status|NEW |RESOLVED --- Comment #15 from H.J. Lu --- Fixed for GCC 12.
[Bug tree-optimization/88531] Index data types when targeting AVX-512 vectorization with gather/scatter
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88531 --- Comment #14 from CVS Commits --- The master branch has been updated by H.J. Lu : https://gcc.gnu.org/g:6866f4819ad8e6e62fef2177520f9fb217dfa353 commit r12-2795-g6866f4819ad8e6e62fef2177520f9fb217dfa353 Author: H.J. Lu Date: Sat Aug 7 07:29:04 2021 -0700 Add tests for PR tree-optimization/88531 PR tree-optimization/88531 * gcc.target/i386/pr88531-1a.c: New test. * gcc.target/i386/pr88531-1b.c: Likewise. * gcc.target/i386/pr88531-1c.c: Likewise. * gcc.target/i386/pr88531-2a.c: Likewise. * gcc.target/i386/pr88531-2b.c: Likewise. * gcc.target/i386/pr88531-2c.c: Likewise.
[Bug tree-optimization/88531] Index data types when targeting AVX-512 vectorization with gather/scatter
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88531
--- Comment #13 from H.J. Lu ---
Here is the equivalent C code:
---
#include
#define loop_t uint32_t
#define idx_t uint32_t
void loop(double * const __restrict__ dst,
double const * const __restrict__ src,
idx_t const * const __restrict__ idx,
loop_t const begin,
loop_t const end)
{
for (loop_t i = begin; i < end; ++i)
dst[i] = 42.0 * src[idx[i]];
}
---
[Bug tree-optimization/88531] Index data types when targeting AVX-512 vectorization with gather/scatter
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88531 --- Comment #12 from H.J. Lu --- For some reason, -march=x86-64 -mx32 and -march=x86-64 -m32 -mfpmath=sse won't vectorize the loop.
[Bug tree-optimization/88531] Index data types when targeting AVX-512 vectorization with gather/scatter
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88531
--- Comment #11 from Richard Biener ---
OK, that probably was an unintended side-effect of now doing
/* Include the conversion if it is widening and we're using
the IFN path or the target can handle the converted from
offset or the current size is not already the same as the
data vector element size. */
if ((TYPE_PRECISION (TREE_TYPE (op0))
< TYPE_PRECISION (TREE_TYPE (off)))
&& (use_ifn_p
|| (DR_IS_READ (dr)
? (targetm.vectorize.builtin_gather
&& targetm.vectorize.builtin_gather (vectype,
TREE_TYPE (op0),
scale))
: (targetm.vectorize.builtin_scatter
&& targetm.vectorize.builtin_scatter (vectype,
TREE_TYPE (op0),
scale)))
|| !operand_equal_p (TYPE_SIZE (TREE_TYPE (off)),
TYPE_SIZE (TREE_TYPE (vectype)), 0)))
{
off = op0;
offtype = TREE_TYPE (off);
STRIP_NOPS (off);
continue;
}
that is we no longer try to consume the conversion because with the conversion
source the gather is not supported and the offset is also already of the
size of the data.
We should probably add this testcase to make sure any other heuristic
improvements in the above code doesn't break it again.
[Bug tree-optimization/88531] Index data types when targeting AVX-512 vectorization with gather/scatter
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88531 --- Comment #10 from H.J. Lu --- It is fixed by r12-2733.
[Bug tree-optimization/88531] Index data types when targeting AVX-512 vectorization with gather/scatter
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88531 --- Comment #9 from Hongtao.liu --- I notice this testcase can be vectorized w/ gcc version 12.0.0 20210805 (experimental) (GCC) Guess it's related to Richi's https://gcc.gnu.org/pipermail/gcc-patches/2021-August/576527.html
[Bug tree-optimization/88531] Index data types when targeting AVX-512 vectorization with gather/scatter
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88531 Andrew Pinski changed: What|Removed |Added Keywords||missed-optimization Blocks||53947 Severity|normal |enhancement Referenced Bugs: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=53947 [Bug 53947] [meta-bug] vectorizer missed-optimizations
[Bug tree-optimization/88531] Index data types when targeting AVX-512 vectorization with gather/scatter
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88531 --- Comment #8 from Florian Schornbaum --- They are definitely a good source to ask. We'll try to get in contact with them and see if we can get help/insight. Thanks for all your input so far!
[Bug tree-optimization/88531] Index data types when targeting AVX-512 vectorization with gather/scatter
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88531 --- Comment #7 from Richard Biener --- (In reply to Florian Schornbaum from comment #6) > Thanks Jakub. That's good information to have. > > We would certainly be willing to help since this is something that we would > really like GCC to be able to handle. > > Does it make sense for us, as developers that have never been involved in > GCC development, to have a look if you give as some pointers on where to > look? Since Mentor and thus CodeSourcery is now part of the Siemens family you could even get support for this from various GCC maintainers in-house ;)
[Bug tree-optimization/88531] Index data types when targeting AVX-512 vectorization with gather/scatter
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88531 --- Comment #6 from Florian Schornbaum --- Thanks Jakub. That's good information to have. We would certainly be willing to help since this is something that we would really like GCC to be able to handle. Does it make sense for us, as developers that have never been involved in GCC development, to have a look if you give as some pointers on where to look?
[Bug tree-optimization/88531] Index data types when targeting AVX-512 vectorization with gather/scatter
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88531 --- Comment #5 from Jakub Jelinek --- Certainly not in the immediate future, GCC 9 is now in stage4 and will remains in regression bugfixing mode until usually mid April. Even after that, I'll be busy with OpenMP 5, but somebody else could implement it too.
[Bug tree-optimization/88531] Index data types when targeting AVX-512 vectorization with gather/scatter
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88531 --- Comment #4 from Florian Schornbaum --- Hi Jakub, Richard, I hope you both had a good start into 2019. I'm still wondering if there are any plans to make arbitrary index data types work with gather/scatter? If there are no such plans at the moment, we will work around this issue on our side.
[Bug tree-optimization/88531] Index data types when targeting AVX-512 vectorization with gather/scatter
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88531 --- Comment #3 from Florian Schornbaum --- Thank you for your very quick replies! I'm aware of 88464 (I think this is the recent work you are referring to?), but this had no effect on the index data type issue I was describing. Even if the gathers/scatters do widening themselves, my code example is not vectorized when using int16_t. Probably because no gather/scatter is created by GCC in the first place? As for uint32_t with -m64 (= unsgined int on x86-64, and sadly the problem that we are facing): I'm aware that manually transforming the index array from uint32_t to int64_t is a solution, but one that comes at a cost for us. Looking at clang, they use "vpmovzxdq" when loading the data. Which is the only difference to the int64_t/uint64_t version, which uses a different load. Are there any plans for GCC to make these "unfitting" index data types work with AVX-512 gathers/scatters?
[Bug tree-optimization/88531] Index data types when targeting AVX-512 vectorization with gather/scatter
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88531
Jakub Jelinek changed:
What|Removed |Added
CC||hjl.tools at gmail dot com
--- Comment #2 from Jakub Jelinek ---
I think I haven't changed anything on this. AVX2 and AVX512{F,VL} gathers and
scatters only do signed widening themselves. Thus, if the index is uint32_t
and -m64, the only hope would be to perform vectorization conversion of it to
uint64_t/int64_t vector first and then use gathers or scatters with V*DImode
indexes.
[Bug tree-optimization/88531] Index data types when targeting AVX-512 vectorization with gather/scatter
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88531 Richard Biener changed: What|Removed |Added Target||x86_64-*-*, i?86-*-* Status|UNCONFIRMED |NEW Last reconfirmed||2018-12-17 CC||jakub at gcc dot gnu.org Component|c++ |tree-optimization Ever confirmed|0 |1 --- Comment #1 from Richard Biener --- There were recent updates to scatter/gather support for AVX512 and GCC 9. I'm not sure that generic widening is done though.
