svadev AT lists.siebelschool.illinois.edu
Subject: Svadev mailing list
List archive
- From: Santosh Nagarakatte <santoshn AT cis.upenn.edu>
- To: John Criswell <criswell AT illinois.edu>, svadev AT cs.uiuc.edu
- Subject: [svadev] Quantitative performance slowdowns with SafeCode from llvm trunk
- Date: Thu, 15 Dec 2011 11:55:47 -0500
- List-archive: <http://lists.cs.uiuc.edu/pipermail/svadev>
- List-id: <svadev.cs.uiuc.edu>
Hi all,
I ran SafeCode from the llvm trunk with the set of SPEC 2k6, 2K, 95
benchmarks which we use to performance test SoftBound.
I have attached the graph of the performance slowdown. On an average,
Geometric mean of slowdown with SafeCode in a release build is 24.5X
I used clang -O3 -fmemsafety <....> to compile the benchmarks.
All the performance experiments were performed on a Intel Core 2
processor 2.66GhZ with 4GB ram. I used ref inputs for all the SPEC
benchmarks.
There is one bar for each benchmark in the figure.
Y axis is Slowdown. For example, a of value 10X means the program
runs 10 times slower compared to the baseline.
X axis is Spec benchmarks with abc.2k6, abc.2k and abc.95 means
benchmark abc is a Spec 2K6, Spec 2K, and Spec 95 benchmark
respectively.
Hope this is useful.
Thanks,
Santosh
--
Santosh G Nagarakatte,
PhD Student,
Computer and Information Science Department
University of Pennsylvania,
Philadelphia-19104
http://www.cis.upenn.edu/~santoshn
Attachment:
safecode_graph.pdf
Description: Adobe PDF document
- [svadev] Quantitative performance slowdowns with SafeCode from llvm trunk, Santosh Nagarakatte, 12/15/2011
Archive powered by MHonArc 2.6.16.