This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH] Some ia64 speedups


On 11/2/07, Jakub Jelinek <jakub@redhat.com> wrote:
> Hi!
>
> On PR rtl-optimization/33922 testcase where PRE completely misoptimizes that testcase
> a huge amount of time is spent in safe_group_barrier_needed and
> related functions.
>
> The attached patch should just speed up some low hanging fruit, not by doing
> any algorithmic changes, but make the huge array smaller, avoid computing
> stuff that is never used and avoid huge memcpy calls (1604 bytes twice)
> in each safe_group_barrier_needed.
>
> Here are some timings with yesterday's svn.  I have timed user time for
> bootstrap of --enable-checking=release build several times (for checking
> there is an improvement as well) and below is also timing for the pr33922.c
> testcase which was speeded up by 42%.
>
> unpatched
> ../configure --enable-languages=c,c++,fortran,java,objc,obj-c++ --enable-checking=release; time make -j32 > /dev/null 2>&1
> user    256m13.004s
> user    256m27.199s
> user    255m46.884s
> user    256m8.872s
> time ./cc1 -quiet -O3 pr33922.c
> user    0m23.064s
> patched
> ../configure --enable-languages=c,c++,fortran,java,objc,obj-c++ --enable-checking=release; time make -j32 > /dev/null 2>&1
> user    254m48.218s
> user    254m33.087s
> user    254m17.814s
> user    254m11.025s
> user    254m16.007s
> time ./cc1 -quiet -O3 pr33922.c
> user    0m13.296s
>
> Bootstrapped/regtested on ia64-linux, ok for trunk?

Please make rws_insn_set and rws_insn_test static inline functions
instead of macros.  Ok with this change if you also tested this with
checking disabled.

Thanks,
Richard.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]