Re:David Patterson Says It’s Time for New Computer Architect...
by m_mueller on Tuesday September 18, 2018 @12:59AM
(#18012576)
On GPUs you can program the L1 cache at least. I’ve never had much success with it though, as in more recent architectures I found the default L1 and the schedulers are doing a good job at keeping occupancy high and by wiring it you either don’t win anything or it often even gets slower - and there’s no way to know in advance since scheduler and cache architecture are largely a trade secret. In general, GPUs sre optimized to use their high memory bandwidth through an extreme amount of hyperthreading, and the number of possible threads goes down with the ressources your kernel occupies.
Some embedded systems had scratchpads and it is possible to 'wire down cache' on some architectures to similar effect. But it would be tricky in general at the language level.
I have a project half on the boil that's aiming to produce a language aimed at these kind of usages. I'm figuring that the supposed insult hurled at C ("structured assembly") isn't really an insult at all but that someone should actually do that. At the moment C, with its giant raft of Undefined Behavior, isn't really that language any more.