This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH v3] Add random memcpy test
On Sun, 2017-03-26 at 18:12 +0530, Siddhesh Poyarekar wrote:
> On Sunday 26 March 2017 12:54 AM, Wilco Dijkstra wrote:
> >
> > Yes it seems I forgot to do a git add of the actual file... Sorry
> > about that.
> >
> > Since I'm on holiday with no access to my GLIBC tree, it's probably best
> > to revert the Makefile change for now and I'll redo it when I get back.
> Don't worry about it, I'll push the file.
>
> Siddhesh
I see the bench-memcpy-random.c file now but I think there is still
something wrong. I am using a glibc that has my aarch64 IFUNC version
of memcpy and memmove implemented in it but I do not get measurements
of the different implementations for the random memcpy test like I do
for the other two memcpy tests.
In bench-memcpy.out I have:
builtin_memcpy simple_memcpy __memcpy_thunderx __memcpy_generic
Length 1, alignment 0/ 0: 39.2188 19.375 22.03
12 23.5938
Length 1, alignment 0/ 0: 27.3438 16.25 22.187
5 22.1875
.
.
In bench-memcpy-large.out I have:
__memcpy_thunderx __memcpy_generi
c
Length 65543, alignment 0/ 0: 8937.5 17000
Length 65551, alignment 0/ 3: 27275.6 35756.9
.
.
In bench-memcpy-random.out I have:
memcpy
Memory size 4096: 105751
Memory size 8192: 103559
Memory size 16384: 112045
Memory size 32768: 118391
Memory size 65536: 158526
I.e. there are no seperate timings for the generic and thunderx
versions of memcpy like I have in bench-memcpy and bench-memcpy-large.
There is code in bench-memcpy-random.c to do the different
measurements, I see 'FOR_EACH_IMPL' but it doesn't seeem to be working.
Has anyone else seen this?
Steve Ellcey