home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!olivea!decwrl!concert!sas!mozart.unx.sas.com!walker
- From: walker@twix.unx.sas.com (Doug Walker)
- Newsgroups: comp.sys.amiga.programmer
- Subject: Re: SASC 6.0 (food for thought)
- Message-ID: <BtpHpo.9KB@unx.sas.com>
- Date: 28 Aug 92 18:22:35 GMT
- References: <25440014@hpsad.sad.hp.com> <BtHqsn.Cyv@unx.sas.com> <SIE.92Aug28144523@introitus.fulcrum.bt.co.uk>
- Sender: news@unx.sas.com (Noter of Newsworthy Events)
- Organization: SAS Institute Inc.
- Lines: 66
- Originator: walker@twix.unx.sas.com
- Nntp-Posting-Host: twix.unx.sas.com
-
-
- In article <SIE.92Aug28144523@introitus.fulcrum.bt.co.uk>, sie@fulcrum.bt.co.uk (Simon Raybould) writes:
- |> In article <BtLDsK.z9@unx.sas.com> walker@twix.unx.sas.com (Doug Walker) writes:
- |> In article <1992Aug25.221919.12437@ccu.umanitoba.ca>, kaarts@ccu.umanitoba.ca (Kenneth John Aarts) writes:
- |> |> In general, what is the average speedup for a generic program by generating
- |> |> 030 specific code?
- |>
- |> I got curious and tried the Dhrystone 2.0 benchmark on my A3000.
- |> I got the following results:
- |>
- |> V5.10, 68000: 6362.7 Dhrystones/sec
- |>
- |> V5.10, 68030: 7287.4 "
- |>
- |> V6.0, 68000: 8039.4 "
- |>
- |> V6.0, 68030: 8708.3 "
- |>
- |> This was with full optimization and registerized parameters
- |> in both versions.
- |>
- |>
- |> What about the 68040 ?
- |> There are quite a lot of amigas around with 68040's these days.
-
- |> Simon J Raybould (sie@fulcrum.bt.co.uk) // {o.o}
-
- OK, one more try... I found that I wasn't really cranking up
- the optimization as far as it would go, so I did it again.
- This time I set the inlining function up as high as it would
- go, which increased the code size significantly but bought
- some additional speed.
-
- The V5.10 numbers are the same, the V6 numbers are new. All numbers
- were generated on an Amiga 3000/25.
-
- V5.10, 68000: 6362.7 Dhrystones/sec
- V5.10, 68030: 7287.4 "
-
- V6.0, 68000 8321.8 "
- V6.0, 68030 9202.5 "
-
- Then I did the same set of tests on an Amiga 3000 with a 68040
- board installed:
-
- V5.10, 68000: 20761.2 Dhrystones/sec
- V5.10, 68040: 24477.2 "
-
- V6.0, 68000: 25603.8 "
- V6.0, 68040: 28301.9 "
-
- Note that since the Dhrystone benchmark doesn't use floating
- point, there would be no difference between code compiled for
- the 68030 or 68040. The only difference between SAS/C code
- generated for the '030 and the '040 is that floating point
- instructions which must be emulated on an '040 are not
- generated.
-
- --
- *****
- =*|_o_o|\\=====Doug Walker, Software Distiller====== BBS: (919)460-7430 =
- *|. o.| || 1200/2400/9600 Dual
- | o |// For all you do, this bug's for you!
- ======
- usenet: walker@unx.sas.com bix: djwalker
- Any opinions expressed are mine, not those of SAS Institute, Inc.
-