more careful float -> int range checks

After upgrading my emscripten toolchain I got a warning that
INT_MAX can't be represented with a float, so the comparison

    if (fbits > INT_MAX)

may not be doing the comparison we want.  In fact, it was not...
we want to exclude any fbits value that's equal to (float)INT_MAX,
since that's actually greater than INT_MAX.

So add explicit casts here to let that new toolchain know we know
what we're doing, and count INT_MAX among the infinite.  INT_MIN
is a power of two so it's exactly representable as a float and that
comparison works as we were intending.

We do seem to hit these cases outside fuzzers, but I think only
when trying to fit curves that we'll fail to fit anyway.

Change-Id: I427d7c8e1d5a6a13328859646dd8d178f0e8f647
Commit-Queue: Brian Osman <>
Auto-Submit: Mike Klein <>
Reviewed-by: Brian Osman <>
1 file changed