home *** CD-ROM | disk | FTP | other *** search
- Path: grafix.xs4all.nl!john.hendrikx
- Date: Sat, 06 Jan 96 15:45:44 GMT+1
- Newsgroups: comp.sys.amiga.programmer
- Distribution: world
- Subject: Re: >>>>>>> The Future: Amiga goes P
- MIME-Version: 1.0
- Content-Type: text/plain; charset=iso-8859-1
- Content-Transfer-Encoding: 8bit
- From: john.hendrikx@grafix.xs4all.nl (John Hendrikx)
- Message-ID: <john.hendrikx.4458@grafix.xs4all.nl>
- Organization: Grafix Attack BBS Holland
-
- In a message of 02 Jan 96 Th.huber wrote to All:
-
- >> Get a clue, AKIKO uses slow $dffxxx-type registers, so the first 2 moves
- >> are just as slow as ChipRAM moves.
-
- Th> Yep, $dffxxx registers are slow, that`s true. But I`m refering to
- Th> machines I know:
-
- Th> - A1200 with 68030/50Mhz with overdrive CDROM.
-
- Then why did you state in your original post that AKIKO is even faster than
- 040's? This is simply not true.
-
- Th> - CD32 expanded with keyboard and 68030/50Mhz.
-
- Th> (On sysinfo both machines show equal speed values.)
-
- Th> I tried many C2P-codes, but noone could beat the 3.1 OSfunktion on the
- Th> CD32.
-
- Give us some timings (with an exact description of what you're doing), maybe
- you haven't been using the really fast C2P routines out there.
-
- Th> In my opinion this results by the high number of cache misses you get
- Th> with C2P by CPU. Most values are read once, thus not cacheable.
-
- And I guess with AKIKO the values aren't read once? AKIKO will have the same
- amount of cache misses (if there are any, on 040 there probably is only 1 cache
- miss for every 16 bytes you read in most C2P routines)
-
- Th> Instructioncache is very effective with most of the C2Pcodes, but it
- Th> doesn`t give such a boost that you could reach akikos registers.
-
- You just don't get it; do you know what it means when people say that on 040
- the only thing limiting the speed of the C2P conversion is ChipRAM speed? This
- means that all the CPU does is wait for the ChipRAM bus all the time and during
- this waiting it can do the entire C2P process all while waiting for the ChipRAM
- bus. On 040's a loop like this:
-
- .Loop
- Move.l d0,(a0)+ ;a0 points to ChipRAM
- Dbra d1,.Loop
-
- is just as fast as:
-
- .Loop
- Move.l d0,(a0)+ ;a0 points to ChipRAM
- Move.l a1,a1
- Move.l a1,a1
- Move.l a1,a1
- Move.l a1,a1
- Move.l a1,a1
- Move.l a1,a1
- Move.l a1,a1
- Move.l a1,a1
- Move.l a1,a1
- Move.l a1,a1
- Move.l a1,a1
- Move.l a1,a1
- Move.l a1,a1
- Move.l a1,a1 ;probably even more.
- Dbra d1,.Loop
-
- The 'Move.l a1,a1' instructions can be exchanged for any instruction you like
- as long as it doesn't access memory. So you could do And, Or, Add, Lsr, Swap,
- whatever and still the routine runs just as fast, all because of the slow
- ChipRAM bus.
-
- GOOD C2P routines take advantage of this, AKIKO can never go 'faster' than this
- simply because ChipRAM is the limiting factor.
-
- Grtz John
-
- -----------------------------------------------------------------------
- John.Hendrikx@grafix.xs4all.nl TextDemo/FastView/Etc... development
- -----------------------------------------------------------------------
- -- Via Xenolink 1.985B1, XenolinkUUCP 1.1
-