Interesting how companies keep reinventing the same idea. I came across a similar tool at IBM 10 years ago name fdprpro which did exactly the same: it was a post-link optimization tool, which used feedback information from an instrumented run (similar to PGO) and rebuild the binary with certain optimizations. The advantage was it did not required the source code, as BOLT, and it obtained the profile information directly from hardware counters.
I heard the DB2 team did use it and got some nice improvements in some workloads, in the order of 5% to 10%. On 19/06/2018 16:10, Prathamesh Kulkarni wrote: > Please ignore, sent to the wrong list. > Sorry for the noise. > > Regards, > Prathamesh > > On 20 June 2018 at 00:34, Prathamesh Kulkarni > <prathamesh.kulka...@linaro.org> wrote: >> Just came across Facebook's announcement of open-sourcing BOLT, a tool >> for optimising instruction cache and TLB misses. Looks interesting. >> They claim to get 2-15% perf improvement in their services with BOLT >> and have a port for aarch64: >> https://code.facebook.com/posts/605721433136474/accelerate-large-scale-applications-with-bolt/ >> >> Regards, >> Prathamesh > _______________________________________________ > linaro-toolchain mailing list > linaro-toolchain@lists.linaro.org > https://lists.linaro.org/mailman/listinfo/linaro-toolchain > _______________________________________________ linaro-toolchain mailing list linaro-toolchain@lists.linaro.org https://lists.linaro.org/mailman/listinfo/linaro-toolchain