This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: Global code hoisting patch
- To: gcc-patches at gcc dot gnu dot org
- Subject: Re: Global code hoisting patch
- From: Dan Nicolaescu <dann at ics dot uci dot edu>
- Date: Mon, 04 Oct 1999 14:36:25 -0700
- Cc: Zack Weinberg <zack at bitmover dot com>, Jeffrey A Law <law at cygnus dot com>
- References: <199909211057.aa21590@gremlin-relay.ics.uci.edu> <199909211803.LAA29123@zack.bitmover.com>
Hi!
I have done some benchmarking for this using SPEC95. The machine I
used is a Sun Ultra5 with a 333 MHz UltraSPARC-IIi with 256Mb RAM
running Solaris-2.7 . The benchmarks and the compiler are on the local
disk.
The benchmarks were run 3 times, the input set used is the reference
one.
I used the 1999-09-27 snapshot.
The 126.gcc benchmark did not compile, so I did not use it.
The orig-O[2s] columns are the results when using the gcc snapshot and
compiling with -mcpu=v8 -0[2s]
The patched-O[2s] are the results when using the snapshot with Zack's
patch applied (see http://egcs.cygnus.com/ml/gcc-patches/1999-09/msg00956.html)
The fpppp orig-Os result looks weird, it's way to fast compared to the
other fpppp runs, I run it a couple of time, and it is
reproducible. I didn't try to see what is the cause for this.
It looks like Zack's patch helps most of the time..
Run times in seconds
Benchmarks orig-O2 orig-Os patched-O2 patched-Os
go 471 463 396 408
m88ksim 216 286 207 287
gcc -- -- -- --
compress 269 275 163 189
li 253 257 250 253
ijpeg 233 274 226 227
perl 192 198 183 202
vortex 300 314 310 306
tomcatv 391 465 362 471
swim 453 475 451 475
su2cor 292 316 285 331
hydro2d 433 509 433 500
mgrid 482 493 480 550
applu 372 503 372 486
turb3d 468 590 450 563
apsi 191 221 178 210
fpppp 1816 978 1683 1917
wave5 401 295 259 317
--dan