Bug 210028 – ensureStillAliveHere can take the value in any location

• Status: RESOLVED
• Resolution: FIXED
• Component: New Bugs

[Keith Miller]

ensureStillAliveHere can take the value in any location

[Keith Miller]

Created attachment 395509 [details]
Patch

[Mark Lam]

Comment on attachment 395509 [details]
Patch

r=me.  Please also fix the variant in JSValue too.

[Keith Miller]

Created attachment 395511 [details]
Patch for landing

[Keith Miller]

(In reply to Mark Lam from comment #2)
> Comment on attachment 395509 [details]
> Patch
> 
> r=me.  Please also fix the variant in JSValue too.

Ah, I missed that one. Done.

[Mark Lam]

Actually, I think the “clobbers memory” declaration is needed.  Otherwise,the compiler can optimize the whole asm statement away.

https://gcc.gnu.org/onlinedocs/gcc/Extended-Asm.html
“Note, however, that if the code that follows the asm statement makes no use of any of the output operands, the GCC optimizers may discard the asm statement as unneeded (see Volatile).”

I think :memory takes care of this.

[Keith Miller]

(In reply to Mark Lam from comment #5)
> Actually, I think the “clobbers memory” declaration is needed. 
> Otherwise,the compiler can optimize the whole asm statement away.
> 
> https://gcc.gnu.org/onlinedocs/gcc/Extended-Asm.html
> “Note, however, that if the code that follows the asm statement makes no use
> of any of the output operands, the GCC optimizers may discard the asm
> statement as unneeded (see Volatile).”
> 
> I think :memory takes care of this.

That's what the volatile does.
https://gcc.gnu.org/onlinedocs/gcc/Extended-Asm.html#Volatile
GCC’s optimizers sometimes discard asm statements if they determine there is no need for the output variables. Also, the optimizers may move code out of loops if they believe that the code will always return the same result (i.e. none of its input values change between calls). Using the volatile qualifier disables these optimizations.

[Mark Lam]

Comment on attachment 395511 [details]
Patch for landing

Nevermind.  https://gcc.gnu.org/onlinedocs/gcc/Extended-Asm.html#Volatile

“ Using the volatile qualifier disables these optimizations. asm statements that have no output operands, including asm goto statements, are implicitly volatile.”

[EWS]

Patch 395511 does not build

[Keith Miller]

Created attachment 395516 [details]
Patch for landing

[EWS]

Committed r259554: <https://trac.webkit.org/changeset/259554>

All reviewed patches have been landed. Closing bug and clearing flags on attachment 395516 [details].

[Radar WebKit Bug Importer]

<rdar://problem/61318951>

[Yusuke Suzuki]

I think this is not correct. This needs to be a compiler-fence and "memory" is used to make it so.
e.g. WTF::compilerFence.

241 // Just a compiler fence. Has no effect on the hardware, but tells the compiler
242 // not to move things around this call. Should not affect the compiler's ability
243 // to do things like register allocation and code motion over pure operations.
244 inline void compilerFence()
245 {
246 #if OS(WINDOWS) && !COMPILER(GCC_COMPATIBLE)
247     _ReadWriteBarrier();
248 #else
249     asm volatile("" ::: "memory");
250 #endif
251 }

[Yusuke Suzuki]

Committed r259558: <https://trac.webkit.org/changeset/259558>