🚀 go-pugleaf

In article <TMH.92Aug26230950@doppel.first.gmd.de>, tmh@doppel.first.gmd.de (Thomas Hoberg) writes:

| Why did it fail, though? While there might be better CPUs today, I
| thought it pretty good when it came out. Was there anything seriously
| wrong with it (except that it'd do virtual caching)? Was it just that
| the various vendors were too busy promoting their own designs or
| tuning their old architectures? Is it just bad luck that nobody
| decided to pick it up or is there something about the i860 that I
| missed?

  Please realize that I am a casual 860 user who talks to serious
users... but the problem seems to be that there is not (or maybe was
not) a good C compiler which would generally take advantage of the
performance of the chip. During another discussion I had several people
tell me that the 860 would run about 4x faster hand coded than with the
best compiler they could buy. These were people who were using it for
image processing and DSP and who /liked/ the chip, so I don't think it's
being badmouthed.

  To some extent this is true of the FPU in the x87 line as well, but
not nearly to that extent. I did hand code one inner loop (11
instructions) for speed a few years ago, and got about 40% by playing
with the FP stack like a pipe organ. I would not expect a compiler to do
the things I did, and the small gain indicates that while there is room
for improvement the compiler do pretty well.

  The inner loop ran 600 million times per frame, so it was definitely
worth it!

-- 
bill davidsen, GE Corp. R&D Center; Box 8; Schenectady NY 12345
    I admit that when I was in school I wrote COBOL. But I didn't compile.