[cfe-dev] [Patch] __builtin_expect uses long, not int

[Douglas Gregor]

On Oct 31, 2009, at 4:19 AM, Ed Schouten wrote:

> Hi folks,
>
> Robert Watson discovered an issue with __builtin_expect() while  
> porting
> libdispatch to FreeBSD/amd64. According to the GCC docs,
> __builtin_expect looks like this:
>
> | long __builtin_expect (long exp, long c);
>
> But currently we use this line in Builtins.def:
>
> | BUILTIN(__builtin_expect, "iii"   , "nc")
>
> This means the arguments and the return value are of type int. This
> breaks libdispatch, because it tries to pass pointers through this
> function.

Ow, good catch.

> Patch:
>
> Index: include/clang/Basic/Builtins.def
> ===================================================================
> --- include/clang/Basic/Builtins.def	(revision 85650)
> +++ include/clang/Basic/Builtins.def	(working copy)
> @@ -329,7 +329,7 @@
> BUILTIN(__builtin___vfprintf_chk, "iP*icC*a", "FP:2:")
> BUILTIN(__builtin___vprintf_chk, "iicC*a", "FP:1:")
>
> -BUILTIN(__builtin_expect, "iii"   , "nc")
> +BUILTIN(__builtin_expect, "LiLiLi", "nc")
> BUILTIN(__builtin_prefetch, "vvC*.", "nc")
> BUILTIN(__builtin_abort, "v", "Fnr")
> BUILTIN(__builtin_trap, "v", "nr")

Thank you, committed here:

	http://lists.cs.uiuc.edu/pipermail/cfe-commits/Week-of-Mon-20091026/022975.html

> I've been thinking about adding some tests to the test suite. Maybe we
> could test something along these lines:
>
> | ULONG_MAX == __builtin_expect(ULONG_MAX, 1)

Since we avoid executable tests in the test-suite, we'd want to look  
at the generated IR to make sure it preserves the value of LONG_MAX.  
However, I settled for something a little simpler:

extern long int bi0;
extern __typeof__(__builtin_expect(0, 0)) bi0;

	- Doug

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20091031/989b5fde/attachment.html>