shift count is negative in int_to_fp_impl for arm64-apple-ios11.0-simulator

[itf]

After this commit: 6dfea56,

I see the error:

/int_to_fp_impl.inc:60:29: error: shift count is negative [-Werror,-Wshift-count-negative]
   60 |   const int dstExpBias = (1 << (dstExpBits - 1)) - 1;
      |                             ^  ~~~~~~~~~~~~~~~~
1 error generated.

to reproduce:

clang -nostdlibinc compiler-rt/lib/builtins/floatuntitf.c -fsyntax-only -target arm64-apple-ios11.0-simulator

the error seems to be caused by wrongly assuming the size of a long double

[asb]

Thanks for the report! I think we need to move up #if defined(CRT_HAS_TF_MODE) before the int_to_fp_impl.inc include and defines. That target it seems doesn't support __float128 and long double isn't a 128-bit float, so I think those helpers are never compiled.

It looks like int_to_fp.h also shouldn't be making the assumption that long double is a 128-bit float and have an ifdef similar to the one in fp_trunc.h. But I think that's separate to this bug.

[asb]

I believe this was fixed by 05a4212 (the original reported bug no longer reproduces). Please reopen if you still have issues.