bmahjour added a comment.
Thanks for this patch @ZarkoCA.
We have six of these functions in altivec: `vec_ctd, vec_ctf, vec_cts,
vec_ctsl, vec_ctu, vec_ctul`. They are all defined as macros in altivec.h and
not all of them map to the builtins checked in this patch (eg `vec_ctul`). I
realize this is an improvement over what we have, but just wondering if you've
considered diagnosing cases like `vec_ctul` or the following:
vector double a;
vector unsigned int res = vec_ctu(a, 32);
One solution I can think of is to introduce a noop pass-through builtin whose
only purpose would be to allow semantic checking in places like
`Sema::CheckPPCBuiltinFunctionCall`. For example say the builtin is called
`_builtin_pass_through_with_rcheck_1_32` then we can change `vec_ctul` in
altivec.h to the following:
#define vec_ctul(__a, __b)
\
_Generic((__a), vector float
\
: __extension__({
\
vector float __ret =
\
(vector float)(__a) *
\
(vector float)(vector unsigned)((0x7f +
(__builtin_pass_through_with_rcheck_1_32(__b))) << 23); \
__builtin_vsx_xvcvspuxds(
\
__builtin_vsx_xxsldwi(__ret, __ret, 1));
\
}),
\
and the sema check would look something like:
case PPC::BI__builtin_pass_through_with_rcheck_1_32:
return SemaBuiltinConstantArgRange(TheCall, 0, 0, 31);
Any thoughts @nemanjai ?
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D106409/new/
https://reviews.llvm.org/D106409
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
