Attila Fülöp

D'oh, I was a zombie yesterday, no sleep no brain. > No it really is gone: Looks like a compiler bug. > add 0x10, \DATA_OFFSET That adds the *contents* at...

Great! I'll add the workaround.

So now that's a real problem, not easily fixed. Are both, macOS and Windows clangs at the same version?

I'd bet if you comment out the `breg` macro use, it will trip over the `xmmi` macro. And the `xmmi` one is crucial, without it one would need to expand...

Well, as far as I understand it, just replace `breg` with `.Lbreg` in the macro `.macro bytereg reg` in `isalc_reg_sizes.S` (line 88 ...) and do the same in `isalc_gcm_sse.S.` (lines...

I don't think that tacking a `.L` in front of the symbol would impair readability more than #ifdef'ing it. Rather it would make more clear what we are doing here....

> We could take align out of ENTRY_NP() and keep it separate `ENTRY_NP()` is used in other places as well IIRC, my suggestion was to do something like ```c #define...

Well, actually `.align 6` is llvm syntax and would translate to `.align 64` in GAS. The good new is that llvm also has a `.p2align` directive, so the Mac port...

@lundman When you do macOS you could try `.balign` first. It has the least impact (same syntax as `.align` on Linux) and should work fine wit one argument on macOS....

Unfortunately there's more information needed regarding the XSAVE build failure. What toolchain at what version are you using? What is the CPU model? What distro are you building on? Could...