This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH/22885] Detect altered malloc chunks
- From: DJ Delorie <dj at redhat dot com>
- To: "Carlos O'Donell" <carlos at redhat dot com>
- Cc: libc-alpha at sourceware dot org
- Date: Tue, 13 Nov 2018 18:25:35 -0500
- Subject: Re: [PATCH/22885] Detect altered malloc chunks
"Carlos O'Donell" <carlos@redhat.com> writes:
> Are you available to run this patch through a performance test with
> the simulator and workloads? All the changes are fairly light weight
> and touching data we likely already have hot in the cache, and they
> are mostly mask/test operations. I don't expect them to have a visible
> impact.
I combined Mike's patch with Florian's two (no strict aliasing,
accessors - the fastbin one was already committed), Adam's, and mine,
and benchmarked them all together. I ran each workload through the
simulator 16 times and averaged the results, for each of four workloads,
against both pristine and patched sources. Results below. Deviations
are within 1% but no specific trend.
dj@envy pts/27 ~/tools
$ go oocalc okular-1 fluentd 389ds
---------- oocalc ----------
PRISTINE:
Total call time: 1,073,761,546 cycles
PATCHED:
Total call time: 1,079,895,621 cycles
---------- okular-1 ----------
PRISTINE:
Total call time: 3,271,045,477 cycles
PATCHED:
Total call time: 3,258,433,903 cycles
---------- fluentd ----------
PRISTINE:
Total call time: 6,317,728,016 cycles
PATCHED:
Total call time: 6,266,545,907 cycles
---------- 389ds ----------
PRISTINE:
Total call time: 7,613,754,098 cycles
PATCHED:
Total call time: 7,564,173,273 cycles