Re: [PATCH RESEND 2/2] x86/locking: Use asm_inline for {,try_}cmpxchg{64,128} emulations

[Dave Hansen <dave.hansen@intel.com>]

On 2/13/25 11:14, Uros Bizjak wrote:
> According to [1], the usage of asm pseudo directives in the asm template
> can confuse the compiler to wrongly estimate the size of the generated
> code. ALTERNATIVE macro expands to several asm pseudo directives, so
> its usage in {,try_}cmpxchg{64,128} causes instruction length estimate
> to fail by an order of magnitude (the compiler estimates the length of
> an asm to be more than 20 instructions). 

Just curious, but how did you come up with the "20 instructions" number?

> This wrong estimate further causes unoptimal inlining decisions for
> functions that use these locking primitives.
> 
> Use asm_inline instead of just asm. For inlining purposes, the size of
> the asm is then taken as the minimum size, ignoring how many instructions
> compiler thinks it is.

So, the compiler is trying to decide whether to inline a function or
not. The bigger it is, the less likely, it is to be inlined. Since it is
over-estimating the size of {,try_}cmpxchg{64,128}, it will avoid
inlining it when it _should_ be inlining it.

Is that it?

Is any of this measurable? Is there any objective data to support that
this change is a good one?

It's quite possible that someone did the "asm" on purpose because
over-estimating the size was a good thing.