Conversation
wren6991@types.pl

Luke Wren

feel like binutils could exploit the little-known algebraic property of "commutativity" to make this smaller

2
0
0
wren6991@types.pl

Luke Wren

Reply to @wren6991@types.pl

It already does the equivalent transformation for `add` so this is indeed a binutils issue not a GCC issue. I guess it was just missed when they added the Zcb compressions

1
0
0
palmer

Palmer Dabbelt

Reply to @wren6991@types.pl
@wren6991 ya, probably. I don't think anyone's really looking at the C sub-extensions all that closely -- or really any of the non-server stuff in general, that's where the vast majority of the upstream effort ended up going over the past few years. The embedded platforms tend to be a mess of vendor-specific extensions, it's just too chaotic for us to do a good job.

So if you file a GCC missed-optimization bug this will probably get fixed, but I'd expect there's a ton more of this kind of stuff floating around.

FWIW I think we should move away from the assembler tricks, but that's kind of a different discussion...
1
0
0
wren6991@types.pl

Luke Wren

Reply to @palmer

@palmer Turns out to be a really easy fix. I haven't done the GNU copyright paperwork, but do you think this patch would be accepted as a tiny-change?

Also I agree, the asm compression stuff is pretty leaky, and also poorly specified. Like if I have a c.j to a different section this seems to actually unconditionally emit a 32-bit opcode with an R_RISCV_JAL

2
0
0
palmer

Palmer Dabbelt

Reply to @wren6991@types.pl
@wren6991 most things don't need to be trivial anymore, they can just be DCO like normal projects.
0
0
1
palmer

Palmer Dabbelt

Reply to @wren6991@types.pl
@wren6991 and on the jump thing: you're probably right, there's historically been a ton of issues around C-type instructions where we accept stuff we shouldn't and just do something arbitrary -- that's true in binutils, but if you start looking closely you'll see it in hardware too.

Again something to fix, it's probably another trivial bug somewhere. It's really just a lack of review/testing, again there's just too many sub-targets to try and get everything up to a good quality bar.
0
0
0
wren6991@types.pl

Luke Wren

Reply to @wren6991@types.pl

Today on "why the assembler do that": using an uncompressed alias for load-immediate 0 into an x8-x15 register. There are a ton of these in MicroPython firmware. Weirdly they all seem to be in the soft float lib.

I assume they're going in as `addi` instead of `li`, and binutils doesn't have a compressed alias for that? Compressing these would save 254 bytes in a 376 kB binary, which isn't much but it's free real estate

1
0
0
wren6991@types.pl

Luke Wren

Reply to @wren6991@types.pl

So yeah, culprit: missing aliases for `mv rd, zero` and `move rd, zero` -> `c.li rd, 0`.

I don't love the way this is all hacked in with instruction aliases, but IMO if objdump spits it back out as `li rd, 0`, and that instruction is compressible, it should be compressed by the assembler too.

0
0
0
About social.kernel.org

Terms of service

  1. Please do not use this service in violation of the Linux Kernel Code of Conduct. Doing so will result in your account suspension with the referral of the matter to the CoC committee.

  2. "Repeating"/"boosting" someone else's status on this platform will be treated as endorsement and will fall under rule #1.

  3. You are encouraged to use this platform to promote your work on the Linux Kernel, but there is no restriction on permitted topics (with the exception of anything covered by #1 above).

  4. There is no requirement to post in English, but it should be considered the primary language of communication on this platform.

Privacy notice

The admins of this service have access to all posted statuses. They aren't looking, but if it's something they shouldn't know about, then you should not post it on this platform.

Please see the Linux Foundation Privacy Policy, which applies to this platform as well.

Getting your own account

If you would like an account on this instance, please check that the following applies to you:

  • You are listed in MAINTAINERS or CREDITS

  • OR: You have a kernel.org account or email address

  • OR: You have a long and established history of involvement with the Linux Kernel

If the above is true and you agree with the Terms of Service and Privacy Notice listed above, please use these instructions to request an account: