Re:MMIX 2009: A RISC computer for the third millennium (2011)
by Koshkin on Friday June 09, 2017 @09:19AM
(#14521136)
How do RISC architecture designers deal with the problem of low code density when modern chip technology makes RAM access considerably slower than anything that happens inside the CPU? One answer could be to have a huge instruction cache, but I doubt that it is economically possible.
[ Parent
| on HN ]
Re:MMIX 2009: A RISC computer for the third millennium (2011)
by 0xcde4c3db on Friday June 09, 2017 @09:37AM
(#14521235)
As far as I know, the two major approaches are
1) Caches
2) Supporting a "compressed" variant of the instruction set (ARM Thumb, MIPS16, RISC-V Compressed). This can be put together by using the same field for target and destination registers, shrinking the set of named registers, making shifts and conditions into explicit instructions instead of fields (for ARM), and similar tweaks.
Re:MMIX 2009: A RISC computer for the third millennium (2011)
by deepnotderp on Friday June 09, 2017 @11:53AM
(#14522230)
A number of things. The biggest one is probably compressed instruction sets, which even x86 "CISC" processors use.
For example, RISC-V and Thumb use this in their processors and they're a big boon.
Also, I think that you've been lied to about the relative code density of RISC vs CISC. CISC is only about ~50% more dense than RISC in the best cases, unless you're talking about polynomial evaluation on VAX or something crazy like that.