ahatanak added a comment.
The test is failing because `float16x4_t` is defined as a vector of `__fp16` in
arm_neon.h and Sema promotes vectors of `__fp16` to vectors of float since
`__fp16` is a storage-only type. The test used to pass because of a bug in
function `isVector`, which was fixed in the commit that was reverted.
typedef __fp16 float16_t;
typedef __attribute__((neon_vector_type(4))) float16_t float16x4_t;
For example, in `vneg_f16`, `__p0` is promoted to a vector of float before
it's negated.
__ai float16x4_t vneg_f16(float16x4_t __p0) {
float16x4_t __ret;
__ret = -__p0;
return __ret;
}
Is it correct to use `__fp16` for the element type of `float16x4_t`?
BRANCHES
master
Users:
ahatanak (Author)
https://reviews.llvm.org/rG825235c140e7
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
