This is the mail archive of the libc-alpha@sourceware.org mailing list for the glibc project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH] Don't write beyond destination in __mempcpy_avx512_no_vzeroupper (bug 23196)

On 05/22/2018 03:21 PM, Andreas Schwab wrote:

On Mai 22 2018, Florian Weimer <fweimer@redhat.com> wrote:


It has to do with copy sizes and relation to the cache size.  I had to
adjust your reproducer to make it fail on KNL, basically increasing the
copy size by a factor of 10.


What is the architectural maximum cache size?
My test machine had an L2 cache size of 1 MiB. I think the cut-over happens at half of that (at least for our backported version), which is why 

#define N 976999

in your reproducer word for me (along with -O0 -fno-builtin).
Thanks,
Florian
Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]