home *** CD-ROM | disk | FTP | other *** search
- Path: fc.hp.com!news
- From: koren@hpsrk.fc.hp.com (Steve Koren)
- Newsgroups: comp.sys.amiga.misc
- Subject: Re: DEC ALPHA card for AMIGA
- Date: 29 Jan 1996 12:13:07 -0700
- Organization: HP Fort Collins Site
- Sender: koren@hpsrk.fc.hp.com
- Message-ID: <oj6d982u858.fsf@hpsrk.fc.hp.com>
- References: <4dpvmp$2hc@ixnews8.ix.netcom.com> <4e8olr$h9a@news.voicenet.com>
- NNTP-Posting-Host: hpsrk.fc.hp.com
- In-reply-to: frotz@omni.voicenet.com's message of 25 Jan 1996 20:20:11 GMT
- X-Newsreader: Gnus v5.0.9
-
-
- frotz@omni.voicenet.com (Steve Bara) wrote:
-
- > libraries, since the original amiga math libraries were so slow. Yet
- > another case of hardware banging in "professional" amiga apps that keeps
-
- This so called "hardware banging" is used throughout the computer
- industry, on most platforms, and for good reason.
-
- Many modern CPUs (and note I am *not* including the 680x0 as modern CPUs
- here) do FP very quickly. In many cases FP math is faster than integer
- math by often a factor of 3 or 4. The time spent doing any sort of API
- call would be long compared to the time spent actually doing the
- calculation. Even if the API added only one clock cycle (which is not
- very likely) that would half the FP performance for some things. In
- practice, it would be worse than that.
-
- In other words, your reason for wanting the API is to take advantage of
- faster FP hardware, but by the very act of using the API, you destroy
- any advantage that faster hardware might have gotten you. Only if the
- API overhead is small compared to the calculation time does it make
- sense to use the API approach for FP calculations. But that hasn't been
- true for a long time. So we use inline FP instructions instead.
-
- - steve
-