cmd/compile: make UNDEF into a pseudo-instruction #16291
Labels
Milestone
Comments
|
I think treating UNDEF as NOP (no generated code) would be fine. I'm not worried about people depending on it generating real code; assembly isn't under the go1 guarantee. I remember there being a case where we need a nop that generated real code - maybe before deferreturn so we get correct line numbers? Just make sure we don't break that. |
|
The nop that generates real code is inserted via ginsnop (per arch). On x86, it uses XCHGL AX, AX. |
quentinmit
added
the
NeedsFix
|
Missed Go 1.8. Doesn't really seem worth it at all, at least not without more motivation. You'll probably spend more time debugging it than all the CPU saved ever. |
josharian
added a commit
to josharian/go
that referenced
this issue
May 23, 2017
DO NOT REVIEW [Breaks profiling. It appears the undef is also needed to provide a pc for tracebacks?] It was used by liveness analysis. Now that liveness works directly on SSA, it is unnecessary. Fixes golang#16291 name old object-bytes new object-bytes delta Template 386k ± 0% 386k ± 0% -0.07% (p=0.008 n=5+5) Unicode 202k ± 0% 202k ± 0% -0.02% (p=0.008 n=5+5) GoTypes 1.16M ± 0% 1.16M ± 0% -0.07% (p=0.008 n=5+5) Compiler 3.91M ± 0% 3.91M ± 0% -0.06% (p=0.008 n=5+5) SSA 7.91M ± 0% 7.86M ± 0% -0.53% (p=0.008 n=5+5) Flate 227k ± 0% 227k ± 0% -0.19% (p=0.008 n=5+5) GoParser 283k ± 0% 283k ± 0% -0.03% (p=0.008 n=5+5) Reflect 951k ± 0% 950k ± 0% -0.09% (p=0.008 n=5+5) Tar 187k ± 0% 187k ± 0% -0.05% (p=0.008 n=5+5) XML 406k ± 0% 406k ± 0% -0.04% (p=0.008 n=5+5) [Geo mean] 648k 647k -0.11% name old text-bytes new text-bytes delta HelloSize 684k ± 0% 680k ± 0% -0.60% (p=0.008 n=5+5) Change-Id: I7e026d06a33265481f73754bd3867e80b1d129ec
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
The compiler inserts
obj.AUNDEFinstructions into exit blocks without a return instruction, so that liveness analysis understands the control flow. However,obj.UNDEFis assembled into a two byte UD2 instruction (0f 0b). Something similar occurs on arm, and presumably on other architectures. Since that instruction is unreachable, we may as well omit it entirely, much like we do withobj.NOP. (Savings are small, approx 0.15% binary size, but might help a bit in small, hot, cache-sensitive functions.)One easy fix is to change the assemblers to treat
obj.AUNDEFthe same asobj.NOP. However, this will break any hand-written assembly that uses UNDEF and wants it to mean UD2.Another fix is to add
obj.AUNDEFNOP, which plive.go would treat as the same asobj.AUNDEF, but which would be converted toobj.NOPat assembly time.Opinions? I'd rather do the former.
@randall77
The text was updated successfully, but these errors were encountered: