Re: [musl] roundf() (and round(), and ...)

[Markus Wichmann <nullplan@gmx.net>]

Am Wed, Jun 26, 2024 at 01:49:09AM +0000 schrieb Thorsten Glaser:
> Stories of compilers doing that abound… IIRC it was the compiler
> recognising the implementation pattern of memcpy and replacing
> that by a call to memcpy… in some libc’s that was not careful to
> use -ffreestanding memcpy implementation.

memcpy() is special, in that GCC reserves the right to create calls to
it (and memmove(), memcmp(), and memset()) even in freestanding mode. It
doesn't do that for any other builtin. Your only options are to write
those functions in assembler, or write them in C and constantly curate
the options needed to compile the functions so that no recursive calls
happen.

> 15:41⎜<Lo-lan-do:#fusionforge> Somebody write a testsuite for helloworld :-)
$ ./hw > /dev/full && echo FAIL || echo PASS

And you wouldn't believe how many implementations fail that.

Ciao,
Markus