Fix libstdc++/39629 (partial) + Fix operator precedence in _ShiftMin1

[Ed Smith-Rowland]

Paolo Carlini wrote:
> Hi,
>
> I committed the below. With it we are down to the following 4 fails in
> 32-bit mode:
>
> FAIL: 26_numerics/random/knuth_b.cc execution test
> FAIL:
> 26_numerics/random/piecewise_linear_distribution/operators/serialize.cc
> execution test
> FAIL: 26_numerics/random/ranlux48.cc execution test
> FAIL: 26_numerics/random/ranlux48_base.cc execution test
>
> Thanks,
> Paolo.
>
> ///////////////////
>   
Greetings,

Thanks for the fixes.  I'm not sure why the patch failed to apply.

I did both types of checks and the 64-bit was clean and the 32-bit had 
14 fails.  The patch fixed fails in both.

I wonder if the last fails might be test case fails having to do with 
comparison of literals.  I'll try to do some casting with uint_fast64_t 
if you haven't already.

Should the templates force usage of long long integers?  In the 
standard?  It seems bothersome if a 32-bit system has to supply such a 
seemingly strange type to use.

Ed