Patch: hoist more loop invariants

[Andreas Jaeger]

Dale Johannesen <dalej@apple.com> writes:

> On Thursday, February 27, 2003, at 05:48  PM, Richard Henderson wrote:
>
>> On Wed, Feb 26, 2003 at 05:32:34PM -0800, Dale Johannesen wrote:
>>> This is about a 2% win on SPECint (ppc Darwin).
>>> Bootstrapped and tested on Darwin.
>>
>> Seems ok.  Do you have a notion of where it helped the most?
>
> It was fairly pervasive, with eon and twolf being the biggest
> individual winners.
> It did not make any Specmark worse.

I get other results on GCC CVS mainline with:
                                   HARDWARE
                                   --------
     Hardware Vendor: AMD
          Model Name: Athlon 1.133 Ghz Box
                 CPU: AMD Athlon(tm) Processor
             CPU MHz: 1133.380
                 FPU: Integrated
      CPU(s) enabled: 1
    CPU(s) orderable: 1
            Parallel: No
       Primary Cache: 64KBI + 64KBD on chip
     Secondary Cache: 256 KB
            L3 Cache: N/A
         Other Cache: N/A
              Memory: 512 MB
      Disk Subsystem: 1x IBM DTLA 307030 30GB EIDE Disk
      Other Hardware: Ethernet
                                   SOFTWARE
                                   --------
    Operating System: SuSE Linux 8.0 (i386)
            Compiler: GCC CVS
         File System: Linux/reiserfs
        System State: Multi-User


                                    NOTES
                                    -----
     
     Base flags: -O2 -march=athlon -malign-double
     Compiler: GCC CVS mainline
     Peak flags: -O2 -march=athlon -malign-double
     Compiler: base+Patch:  hoist more loop invariants

                                     Estimated                     Estimated
                   Base      Base      Base      Peak      Peak      Peak
   Benchmarks    Ref Time  Run Time   Ratio    Ref Time  Run Time   Ratio
   ------------  --------  --------  --------  --------  --------  --------
   164.gzip          1400       288       486*     1400       290       483*
   175.vpr           1400       449       312*     1400       449       311*
   176.gcc           1100       292       377*     1100       293       375*
   181.mcf           1800       815       221*     1800       815       221*
   186.crafty        1000       173       576*     1000       174       576*
   197.parser        1800       532       338*     1800       533       338*
   252.eon           1300       218       598*     1300       217       599*
   253.perlbmk       1800       346       520*     1800       344       524*
   254.gap           1100       274       401*     1100       272       404*
   255.vortex        1900       414       459*     1900       416       456*
   256.bzip2         1500       438       343*     1500       437       343*
   300.twolf         3000       877       342*     3000       873       344*
   Est. SPECint_base2000                  399
   Est. SPECint2000                                                     400

So, it's just one point which is not really meaningfull.  Or should I
have used different compiler options for this?

Andreas
-- 
 Andreas Jaeger
  SuSE Labs aj@suse.de
   private aj@arthur.inka.de
    http://www.suse.de/~aj