johnplatts wrote:
I have discovered that there is a bug in the VSX `vec_pack_to_short_fp32(a, b)`
intrinsic with GCC on big-endian POWER9 (but not little-endian POWER9), where
`vec_pack_to_short_fp32(a, b)` returns `{b[0], b[1], b[2], b[3], a[0], a[1],
a[2], a[3]}` instead of the expected `{a[0], a[1], a[2], a[3], b[0], b[1],
b[2], b[3]}`.
PVIPR expects the `vec_pack_to_short_fp32(a, b)` intrinsic to return the same
results on both big-endian POWER9 and little-endian POWER9.
https://github.com/llvm/llvm-project/pull/129923
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
