FWIW, general expectations for PowerPC are that fused multiply-add is available by default. For example, the default for GCC on powerpc64-unknown-linux-gnu is to generate fmadd at -O1 and above. I may be missing the point, but I don't see a need for the default to be dependent on the vendor in this case.
My two cents, Bill -- Bill Schmidt, Ph.D. IBM Advance Toolchain for PowerLinux IBM Linux Technology Center [email protected] [email protected] On Tue, 2012-09-18 at 16:53 -0500, Hal Finkel wrote: > Lang, et al., > > Do you (or anyone else) have an opinion on this? Do we want to > introduce vendor-specific defaults for this? Should we wait until > contraction is fully implemented in clang and let them test the > differential performance (and what's the status on that)? > > Thanks again, > Hal > > On Fri, 31 Aug 2012 15:34:26 +0200 > Tobias von Koch <[email protected]> wrote: > > > New patch with test case attached. > > > > Thanks again, > > Tobias > > > > On 28/08/2012 20:00, Hal Finkel wrote: > > > Tobias, > > > > > > This also requires a test case. > > > > > > -Hal > > > > > > On Tue, 28 Aug 2012 18:54:18 +0200 > > > Tobias von Koch <[email protected]> wrote: > > > > > >> Dear all, > > >> > > >> This is part of a sequence of three (trivial) patches to allow > > >> Clang to be used as a drop-in replacement for GCC with the > > >> Freescale SDK. Support for Freescale processors has just been > > >> added to the PowerPC backend. > > >> > > >> Please review and commit if this is OK - I don't have commit > > >> access. > > >> > > >> Thanks! > > >> Tobias > > >> > > >> Description of patch: > > >> > > >> If vendor==Freescale, set fp-contract=fast unless specified > > >> otherwise. This is the default in GCC on Freescale targets. > > >> > > > > > > > > > > > > > > _______________________________________________ cfe-commits mailing list [email protected] http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
