home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!dtix!darwin.sura.net!mips!suneel
- From: suneel@mips.com (Suneel Jain)
- Newsgroups: comp.sys.sgi
- Subject: Re: R4000 compiler directive, is there one ???
- Message-ID: <l8vqdsINNirj@spim.mips.com>
- Date: 17 Aug 92 18:02:36 GMT
- References: <oisl0q0@zuni.esd.sgi.com> <1992Aug16.210505.18316@megatek.uucp>
- Organization: MIPS Computer Systems, Sunnyvale, California
- Lines: 63
- NNTP-Posting-Host: limca.mips.com
-
- In article <1992Aug16.210505.18316@megatek.uucp> rgs@megatek.uucp (Rusty Sanders) writes:
- >From article <oisl0q0@zuni.esd.sgi.com>, by olson@anchor.esd.sgi.com (Dave Olson):
- >> In <1992Aug14.180101.14529@megatek.uucp> rgs@megatek.uucp (Rusty Sanders) writes:
- >> | Net result; if you're using floating point on an R4000 you probably
- >> | should use -mips2 -r4000 to get best performance. Does MIPS/SGI
- >> | actually support (i.e. accept bug report calls) on the undocumented
- >> | -r4000 switch?
- >>
- >> I don't think so. If so, we would have used it in the specmark
- >> tests, more than likely. I don't know why it isn't supported;
- >> it may be for lack of testing, or introduction of bugs.
- >
- >It's my understanding that MIPS specifically did NOT want to use
- >code optimized specifically for the R4000 in the Specmark suite.
- >To wit: I believe the code wasn't even compiled -mips2 (which
- >is clearly supported), much less with -r4000.
- >
- >I have been lead to believe this is because MIPS wanted to
- >accurately represent processor performance with code compiled
- >to run on any MIPS machine, hence the -mips1 compile switch.
-
- I don't think the above is true. The SPEC numbers released by SGI for
- Crimson and R4000 Indigo used the -mips2 option.
-
- >
- >I don't believe I've ever seen Specmark numbers for an R4000
- >with the code compiled -mips2 -r4000, or even just -mips2
- >(although I would dearly love to).
- >
- >One would think it prudent marketing for MIPS/SGI to release
- >both sets of numbers. One for compatibility mode (useful for
- >software houses, which will compile -mips1), and one for raw
- >"fast as we can get it" mode, which is the most useful number
- >for the embedded market and end-users. Not that Spec is terribly
- >useful as a performance indicator in the embedded market, but
- >it would make my discussions with management SO much easier.
-
-
- The -r4000 option is fully supported in the 3.10 release of the MIPS
- compilers. The SPEC numbers released for the Magnum R4000PC used the
- -mips2 and the -r4000 options.
-
- Currently, the processors based on the MIPS architecture that are out in
- the market are R3000, R4000 and the R6000. Some time in the future, these
- will be joined by TFP, T5, VRX, ....
-
- There are differences in the instruction scheduling rules for each of
- these processors. Some of the differences are minor, but some are major,
- especially in the area of FP scheduling. Most software applications will
- be compiled only one way that allows it to run on all MIPS architecture
- boxes. That lowest common denominator option today is -mips1. Sometime
- in the future, the lowest common denominator option maybe "-mips2" or
- maybe even "-mips2 -r4000".
-
- At the same time, there are some programs that want the best scheduling
- for a particular target. To address such needs, we have added the -r4000
- and -r6000 options to the 3.10 compiler. These are completely orthogonal
- to the -mips[1,2,3] options, which define the instruction set to be used.
- --
- Suneel Jain
-
- EMAIL : suneel@mips.com
- USMAIL: Silicon Graphics, Inc., 928 Arques Avenue, Sunnyvale CA 94088-3650.
-