home *** CD-ROM | disk | FTP | other *** search
- Path: lou.teclink.net!usenet
- From: rad@teclink.net (rad)
- Newsgroups: comp.sys.amiga.programmer
- Subject: Re: Processors
- Date: 20 Mar 1996 03:20:25 GMT
- Organization: TECLink Internet Services: info@TECLink.Net
- Message-ID: <4024.6651T1376T3@teclink.net>
- References: <1880.6651T550T1538@darwin.topend.com.au>
- NNTP-Posting-Host: tc2_43.teclink.net
- X-Newsreader: THOR 2.2 (Amiga;TCP/IP) *UNREGISTERED*
-
- travis <envision@darwin.topend.com.au> writes:
- >I looking for non emotional comments :-) about the differences between
- >Motorola processors and Intel equivs.
- >
- >Recently, In the Imagine mailing list, we have been discussing the the very
- >obvious performance difference between Amiga Imagine and PC Imagine.
- >Comparing similar clock speeds, the PC version of Imagine does run much
- >faster than the Amiga version.
- ...
- >Could it be that the difference in performance is due to the fact that when
- >the 040 was released, some of the original FPU functions must now be
- >emulated?
-
- First of all the emulated function are VERY rarely called in Imagine. Even if
- emulation took 10-20 times longer it would be negligible. Second the 68040
- added special features to make emulation fast and easy. (All operands are
- fetched for you, you don't need to worry about where the result goes all you
- need do is compute the result.) The Motorola emulation package (which Amiga
- uses) runs transcendental functions at the speed equal or better than a
- 33MHz 68882.
-
- >Have the Intel processors maintained the same FPU instructions right through
- >the various CPU versions? This might then explain the sometimes large
- >differences we are seeing.
-
- Yes, intel had nearly all there original 8087 instructions through out the
- family; however, Intel started with only 6 transcendentals (compared to
- Motorola's 20). Keeping those could hardly be considered a feat. Further,
- even in Imagine these functions are hardly ever used. This has nothing to do
- with the speed differences.
-
- I think if you really look into it you will find that their are 3 major
- factors that cause the difference (none of which have anything to do with the
- CPU itself).
-
- 1. The compilers used to develop the two different versions are drastically
- different. The one used for the pc port was probably developed with 10
- times the man power and budget as the one for used for the Amiga port. It
- is safe to assume that pc compiler has gone to great lengths to squeeze
- that last bit of performance out of the intel CPU. Something the Amiga
- compiler could only put a modest effort toward.
-
- 2. The writers of Imagine probably put more effort into optimizing the pc port
- for speed (possible at the expense of the Amiga port if they insisted on
- the same source code) since the pc market is assumed to give a higher
- return and is more competitive market and since they probably consider
- the Amiga a dying market.
-
- 3. Typical pc's often have additional features such as level-2 caches, Hard
- disk hardware caches not seen on typical Amigas. Further, it is likely
- that much of the Hardware on the benchmark systems was not equavilent RAM
- size & speed, Hard drive access time & transfer rate, just to name a
- couple.
-
- >I know that when Imagine textures are compiled for 040 only use they are some
- >20- 30% faster than the equivalent 68020+FPU code at the same clock rate.
- >
- >Considering the differnce between 68020+FPU and the 040's FPU should a
- >program like Imagine be compiled for 040 (or 060) only use? I guess one
- >would have to know which FPU instructions Imagine was using though.
-
- The performance difference you see in the example above is due to pipelining
- and FPU/Integer parallelization differences between the 68020/881 & the 68040.
- The emulated functions would probably only account for a percentage point or
- so. Even optimizing for the 040 while still using transcendental emulation
- alone would probably be of significant benefit to Imagine.
-
- >Well I don't want to start a CPU war here, I'm just looking for sensible
- >comment.
-
- Likewise.
-
- ------------------------------------------------------------------------------
- - Richard Deken EMail: (personal) rad@teclink.net -
- - VLSI Design Engineer (AuE) rad@aue.com -
- - Advanced Microelectronics PGP public key available -
- ------------------------------------------------------------------------------
-
-