Community
Participate
Working Groups
Build Identifier: 20110615-0604 I get an error `Function __builtin_ia32_vec_ext_v8hi could not be resolved` in a line using the SSE intrinsic function `_mm_extract_epi16(a,b)`. The `_mm_extract_epi16(a,b)` is defined in the SSE2 header `emmintrin.h`, which is in the path, gets parsed and for which the relevant macros are defined ('__SSE__' and '__SSE2__' needs to defined, and it is). The function is expanded with a macro to the gcc builtin function '__builtin_ia32_vec_ext_v8hi', which is defined nowhere (AFAIK), since it is built in. Maybe all functions starting with "__builtin" when using gcc should be treated as "defined"? This happens using gcc-4.5.3 Reproducible: Always
There is GCCBuiltinSymbolProvider that defines "other" built-ins from GCC, like __builtin_abs, etc. Maybe it would possible to have a GCCX86BuiltinSymbolProvider? I'm not sure how the types would map though.
(In reply to comment #1) > There is GCCBuiltinSymbolProvider that defines "other" built-ins from GCC, like > __builtin_abs, etc. Maybe it would possible to have a > GCCX86BuiltinSymbolProvider? I'm not sure how the types would map though. I would not mind adding target-specific builtins to GCCBuiltinSymbolProvider, they won't hurt on other platforms. However, there are lots of them and the gcc-manual does not even list all of them. The data type is 'short __attribute__ ((vector_size (16)))', which will be treated as a short by our parser. -> For the purpose of parsing it'd be sufficient to add int __builtin_ia32_vec_ext_v8hi(short, int) to GCCBuiltinSymbolProvider.
As of April 2013, CDT no longer gives an error for uses of symbols that begin with '__builtin_'. [1] They are still not syntax-colored as function calls; this could be accomplished by adding them to GCCBuiltinSymbolProvider as described in comment 2. However, since the undocumented ones are not meant to be used directly [2], but rather through their wrapper macros, and since there are so many of them, I don't think there's a point in doing that. I think this bug can be closed. [1] http://git.eclipse.org/c/cdt/org.eclipse.cdt.git/commit/?id=fa881229 [2] https://www.mail-archive.com/gcc@gcc.gnu.org/msg03310.html
Closing per comment 3.
