Bug 234333 – [libpas] Implement atomics in inline assembly if compiler is using ARM64

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

[Yusuke Suzuki]

[libpas] Implement atomics in inline assembly if compiler is using ARM64

[Yusuke Suzuki]

Created attachment 447194 [details]
Patch

[Yusuke Suzuki]

Created attachment 447197 [details]
Patch

[Yusuke Suzuki]

Created attachment 447222 [details]
Patch

[Filip Pizlo]

Comment on attachment 447222 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=447222&action=review

I'm not sure how much I care, but it would be nice if store_atomic was a pas_utils thing.

> Source/bmalloc/libpas/src/libpas/pas_lock.h:116
>  #if PAS_COMPILER(CLANG)
> +#if PAS_COMPILER(CLANG_ARM64_ATOMICS_DEPENDENCY)
> +    asm volatile (
> +        "stlrb wzr, [%x[ptr]]\t\n"
> +        /* outputs */  :
> +        /* inputs  */  : [ptr]"r"(&lock->lock)
> +        /* clobbers */ : "memory"
> +    );
> +#else
>      __c11_atomic_store((_Atomic bool*)&lock->lock, false, __ATOMIC_SEQ_CST);
> +#endif
>  #else
>      __atomic_store_n((bool*)&lock->lock, false, __ATOMIC_SEQ_CST);
>  #endif

I think it would be way better if this atomic store thing was factored out into pas_utils.h.

[Yusuke Suzuki]

Comment on attachment 447222 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=447222&action=review

>> Source/bmalloc/libpas/src/libpas/pas_lock.h:116
>>  #endif
> 
> I think it would be way better if this atomic store thing was factored out into pas_utils.h.

OK! Extracted atomic part.

[Yusuke Suzuki]

Committed r287092 (245284@trunk): <https://commits.webkit.org/245284@trunk>

[Radar WebKit Bug Importer]

<rdar://problem/86535186>