Closed mscdex closed 2 years ago
Hm, as mentioned in #96, I think there might be an easier fix, which is to have clang
always load the encoding table with the vld1q_u8_x4
intrinsic. As far as I know, that intrinsic gets compiled correctly by clang
, and will allocate four sequential registers. (GCC does not compile the intrinsic correctly and chooses four random registers.)
This may not be the best fix, but it does allow compiling of debug builds with clang.
Also, as far as I can tell the NEON32 inline asm is not affected.
Fixes https://github.com/aklomp/base64/issues/96