Re: Nop insertion

As our moderator says, it's normally done at assembly time, but that's
because the number of stall cycles is usually small (1-3 instructions)
and localized. Large stalls that are non-localized are not usually
handled by inserting fixed numbers of noops, especially not ones
crossing basic block (or whatever the coding unit is) boundary. I
have heard large precise timing models called "crystal clock"
architectures, refering to the fragiility of a clock made out of
glass. They are not robust, even with automated tool support.


At Intel, we have a crypto multiplier unit we put on certain chips and
it has a vary long latency, but we don't try to time it exactly even
though for any given multiplication there is a precise number of
cycles the multiplication will take. Doing so would only work for
very simple fixed architectures. The more out-of-order one allows,
the more variable the interactions can make the timing.


Thus, semaphores (and other locks) are good things. Use judiciously,
but use them.


Hope this helps,
-Chris


******************************************************************************
Chris Clark email: christopher.f.clark@compiler-resources.com
Compiler Resources, Inc. Web Site: http://world.std.com/~compres
23 Bailey Rd voice: (508) 435-5016
Berlin, MA 01503 USA twitter: @intel_chris

Return to the comp.compilers page.
Search the comp.compilers archives again.