[Libre-soc-dev] WIP demo of deficiency of 6600-derived architecture compared to register renaming

Hendrik Boom hendrik at topoi.pooq.com
Wed Oct 28 21:41:32 GMT 2020

On Wed, Oct 28, 2020 at 03:04:17PM +0000, Luke Kenneth Casson Leighton wrote:
> btw just to say, i hear you about the compilers not wanting to do
> reg-rename, bit of a pain about that.

There definitely have been compilers in the past that modulated their 
register allocation based on detailed understansing of the number and 
timing of the various functinoal units in the hardware.

I think that went out of fashion when
  (a) the CPU's got so comples that the timing was well-night 
unpredictable.  In effect, the CPU's were taking the register allocation 
away from the compilers,
  (b) there started to be many CPU's with the same instruction set but 
many different internal architectures and timing.  Even with a single 
CPU, all that would be needed is a minor microcode update for a 
compiler's carefullt crafted machine-dependent optimisation to be thrown 
out of whack.

-- hendrik

More information about the Libre-soc-dev mailing list