home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!decwrl!pa.dec.com!nntpd2.cxo.dec.com!nntpd.lkg.dec.com!sousa.tay.dec.com!human.enet.dec.com!supnik
- From: supnik@human.enet.dec.com (Bob Supnik)
- Newsgroups: comp.arch
- Subject: Re: DEC Alpha AXP System INTEGER Performance
- Summary: Compilers impacts
- Message-ID: <2226@sousa.tay.dec.com>
- Date: 15 Nov 92 18:33:26 GMT
- References: <1992Nov10.153629.27510@ryn.mro4.dec.com> <martin.721554717@bert> <1992Nov12.101520.2828@crl.dec.com> <1698@niktow.canisius.edu> <jdd.721687838@cdf.toronto.edu> <lgbhcmINNmbc@exodus.Eng.Sun.COM>
- Sender: newsa@sousa.tay.dec.com
- Organization: Digital Equipment Corporation
- Lines: 32
-
-
- In article <lgbhcmINNmbc@exodus.Eng.Sun.COM>, grover@brahmand.Eng.Sun.COM (Vinod Grover) writes...
- > [text deleted]
- >The third is to have clever compilers, that make the machine "appear"
- >fast. How do you tell the difference?. If you measured VAX-780 with
- >modern compilers today, I bet it would do more than 1 SPEC{fp,int}92
- >mark.
- > [text deleted]
-
- This experiment has been done, at least in part, by applying the KAP
- preprocessor and an up to date FORTRAN compiler to classic VAX's. For
- SPECmark89, the 11/780 is indeed a "1.3 SPECmark89" machine -- mostly
- because of the inflation of MATRIX300 seen in other systems. For SPEC92,
- there is less impact.
-
- The reason better compilers can do little with the 11/780 is that it is
- not, at base, a pipelined machine. The principal optimizations to make
- for the 780 would be:
-
- - Inlining, to minimize use of the high overhead procedure
- call and return instructions
- - Cache blocking, to maximize use of the 8KB cache
- - Global register allocation, to minimize register spills
- - Careful instruction selection, to use the fastest sequences
- among alternatives
-
- Recent VAX processors, which are heavily pipelined, benefit for more
- aggressive compiler techniques just like RISC processors.
-
- Bob Supnik >Supnik@human.enet.dec.com
- >All opinions expressed are those of a hardline microcoder
- >and do not reflect those of Digital Equipment Corporation
-