There are prime counts up to 1e16 with a step of 1e9
- http://www.primefan.ru/stuff/primes/pi_1e16.zip (19 Mb)
The archive contains the data file with 2-byte differences between the successive values of Round[(pi(x)-li(x))*3/2], and also a small delphi program to generate a plain text file.
The multiplier of 3/2 allows:
1) to fit the 2-byte range [-32768..+32767] for the differences
2) to compute li(x) with an absolute error of up to 1/6, so the Ramanujan method with extended floating point arithmetics works well up to at least 1e18
[Non-text portions of this message have been removed]