I have a FastQ file with 3.2 billion reads that I'm currently mapping with BWA aln (not mem, but I guess it could have the same behaviour here).

This is the output log on the commandline:
[bwa_aln_core] 2147221504 sequences have been processed. [bwa_aln_core] convert to sequence coordinate... 6.19 sec [bwa_aln_core] refine gapped alignments... 0.54 sec [bwa_aln_core] print alignments... 0.12 sec [bwa_aln_core] -2147483648 sequences have been processed. [bwa_aln_core] convert to sequence coordinate... 6.26 sec [bwa_aln_core] refine gapped alignments... 0.56 sec [bwa_aln_core] print alignments... 0.16 sec [bwa_aln_core] -2147221504 sequences have been processed. [bwa_aln_core] convert to sequence coordinate... 6.22 sec [bwa_aln_core] refine gapped alignments... 0.57 sec [bwa_aln_core] print alignments... 0.11 sec

As you can see, it counts the sequences correctly until C's MAX_INT and then starts using negative numbers at some point. Its more or less an optical thing, but still it would be more informative to have this fixed/enhanced.