Extend the small inverse pow5 offset table

[keith-packard]

The maximum value for the invPow5 function is 342, which means the offset table needs 342 * 2 = 684 bits or > 21 32-bit values. Add the missing values by computing them based off the full table values.

[ulfjack]

It would be good to add tests for this case; are there numbers that don't work correctly due to the missing entries?

[keith-packard]

There are cases which access beyond the bounds of the array; that's what I discovered when I enabled array bounds checking and ran my test suite. I'm not sure I can easily re-create that situation though, it's been a long time.

[ulfjack]

Ok, got it. The asserts are correctly failing the tests, and - for parsing - we need to support at least up to 324 (is the 342 a typo?). I filed #255 for this. Note that the table is generated from Java code. I'll update the generator and then update the file.

[keith-packard]

Ok, got it. The asserts are correctly failing the tests, and - for parsing - we need to support at least up to 324 (is the 342 a typo?). I filed #255 for this. Note that the table is generated from Java code. I'll update the generator and then update the file.

Awesome. Thanks for digging into this; I'll update my C library with the new values.

[keith-packard]

and, of course, if you ever decide to generate tables and code for 80- bit (intel/m68k) and 128-bit (ieee binary 128) forms, that'd be super exciting. I still have janky code for those two types and just haven't found the time to figure out how to fix that.