Return Styles: Pseud0ch, Terminal, Valhalla, NES, Geocities, Blue Moon.

Pages: 1-

RISC-V is bloated crapshit

Name: Anonymous 2018-05-23 18:12

https://content.riscv.org/wp-content/uploads/2017/05/riscv-spec-v2.2.pdf

It has variable instruction (wtf) and on the base architecture only it supports 11 different types of instruction encoding. I would not be surprised if it used half of the instruction length just selecting what the encoding is.

Why not just make mips cpus instead?

Name: Anonymous 2018-05-23 18:23

MIPS? Loongson shill detected.

Name: Anonymous 2018-05-24 16:59

What always triggered me about MIPS was the wasted bits for the unused shift amount field, which they could have easily aliased to the third register field (and introducing a new shift-arith-only instruction type), or introduce fused shift instructions like in ARM.
Well, they fixed that now.
By replacing it with a third (!) opcode field. 17 (7+3+7) of 32 bits needed to differentiate between an ADD and a MUL.

Then you think: well, okay, what should they have done with those 5 ``useless'' bits instead?

They should have made MIPS a variable length encoding of 16 bit wide instruction words from the very start instead of a 32 bit fixed width encoding.

Name: Anonymous 2018-05-24 18:01

ARM is the future. x86 is a sinking ship. RISC/SPARC/MIPS are useless.

Name: Anonymous 2018-05-24 22:18

>>2
Still waiting, Lemote.

Name: Anonymous 2018-05-25 0:04

>>5
oh geez I remember this faggot

Name: Anonymous 2018-06-01 5:27

>>3
But muh instruction decode complexity! John Hennessy told me this is bad; why would he lie?

Name: Anonymous 2018-06-01 7:28

>>7
I'm no HW designer so this *might* be a thing.
However, RISC-V seems to have variable instruction size, anyway (don't quote me on that).

Also, they tested (simulated?) a 16b encoding, that could work together with standard 32b instructions (when aligned)1.
Outcome: it's likely to be more energy efficient and faster2.

1980s HW design: no idea.
1995+ HW design: cache usage > a lot of other things.

1https://en.wikipedia.org/wiki/RISC-V#The_compressed_set
2https://www2.eecs.berkeley.edu/Pubs/TechRpts/2011/EECS-2011-63.html
On a system with a 16 KB direct-mapped instruction cache and 32 KB 2-way data cache, 9% of all main memory accesses (instruction and data) are eliminated, performance is 4% better, and 25% fewer instruction bits need be fetched.

Don't change these.
Name: Email:
Entire Thread Thread List