The Power of the NORC 3764
(John Savard) wrote, in part:
I was too hasty.
All of them. The couple of EE degree programs were a joke, too. (This was in the late...
SR-3108 9.1, the buttembly language manual for the T90 and its predecessors, does give opcodes in an appendix.
Reviewing the table of registers provided, though, I find that while the number of values in each of the eight vector registers was increased from 64 to 128 in the T90, those machines did *not* have a larger set of vector registers with 64 elements or anything like that.
Nor, apparently, did the Cray 2.
They did have something called "local memory", which is a lot like the main memory on the CDC 6600.
In any event, in designing my illustrative computer architecture, I decided to illustrate how vector computers worked - but my attempt to illustrate Cray-like operation failed to an extent.
The Cray machines were pipelined, not parallel. I started out naively buttuming vector registers with 64 elements implied 64 ALUs, and *then* came up with elaborate scatter-gather circuitry to allow pumping 64 arguments of varying length to them, so as to simulate pipelined vector behavior (unlike the raw parallel behavior of MMX and suchlike, which the *short* vector unit illustrates). Intimately tied in with that is the circuitry required to use partially-filled cache lines... to let the computer divide conventional 256-bit wide memory... into things like 36-bit words.
Of course. But our biz' roots stem from that kind of small thinking. It was the time when production line BSs became normal. The...
To equal what I *thought* Cray machines after the Cray 1 were doing, I added a bank of sixty-four vector registers to the architecture... and now I find it based on a misreading.
John Savard Usenet Zone Free Binaries Usenet Server More than 140,000 groups Unlimited download
Alt Folklore Computers from Newsgroups