Re: Superstitious learning in Computer Architecture

jsavard@xxxxxxxxx wrote:

(SNIP)

My own pet peeve is the popularity of RISC. A simple, clean, CISC
design would save having to have two instructions for many common
operations - but, instead, our only two choices are RISC or the
baroque, legacy-rich, x86 architecture. (Needless to say, I mourn the
loss of the 68K architecture.)

My complaint about RISC is a little different. RISC depends on the
compiler to generate optimal code. While that isn't so bad, it
makes the idea of an upward compatible architecture hard. Newer
processors will need different code sequences, which means recompiling.

Note that z/Architecture machines IBM produces today will still run
OS/360 code compiled 40 years ago! But CISC depends on the processor
to optimize the code sequences at run time, which the 360/91 could do,
but that also isn't optimal.

What I would like to see is an intermediate code for a line of RISC
processors that could easily be translated to the specific code sequences optimal for the specific sub-architecture. Not completely
general, but without the micro-optimizations that RISC needs.
Then, at either program load time or install time the specific
optimizations are done.

Note that even CISC doesn't avoid the problem, with specific
optimizations for different models in the x86 family.

-- glen

.

Relevant Pages

  • Re: Code density and performance?
    ... CISC was coined to describe "the rest", which covered an immense range of different ISAs. ... IBM S/360, X86, and VAX are *different*, and an analysis comparing X86 to RISC does nothing to invalidate earlier comparison of VAX to RISC. ... Previous generations of RISC were as variable as 1980s CISC. ... Well, in the *real* world of CPU architecture, there were a fairly modest number of people who actually did this for a living, and many of them knew each other, moved among companies, went to the same ...
    (comp.arch)
  • Re: after SIMD implementation, is it still a RISC?
    ... processor from being RISC, since the PowerPC, a RISC architecture, has ... The original PowerPC architecture did not have AltiVec. ... Hmm, the fused multiply-add instruction? ...
    (comp.arch)
  • Re: 32 or even 64 registers for x86-64?
    ... to PA-RISC, did HP call Spectrum a RISC architecture? ... While PA-RISC is often scoffed at by RISC purists for not being "reduced" ... machine and execute all the instructions you need in a pipelined "single ...
    (comp.arch)
  • Compiler/Architecture course on modern hardware
    ... density of architecture nerds - I have a question about how best to ... Most compiler/architecture courses teached at under graduate/graduate ... implement a simple 1 cycle RISC style arch ... Most students get a surprised when they encounter ...
    (comp.arch)
  • Re: IPS Reliability/Availability
    ... Our benchmarks have shown, with the basic 7 processor model, that it can sustain over 2Gbps of actual customer traffic, and that using this architecture we can actually scale the product with more processors to hit the 10G mark. ... The only two IPS companies I know of that use this technology are NFR and SourceFire so far. ... ASICs have been around a long time, and people have a kind of warm fuzzy from that older technology. ... This RISC architecture was great because it let us take the years of development we did on the x86 platform and cut it over to this RISC model very cleanly, with little development time or cost. ...
    (Focus-IDS)