[Bug c/66683] Large macro with float multiplications fails

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66683

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
Note that most of the memory goes to building the expression in GCCs internal
representation.  The most compile-time spent in a single routine is parsing
the float constants (real_from_string, running into mpfr mpfr_strtofr
and real_from_mpfr).

Memory requirement could be reduced if sharing REAL_CSTs, but of course
adjusting the testcase to use different float values would make that even
worse.