Thank you for the PR! It looks great.

Before merging it to master, I would like to request some modifications.

• ilog10ull
  • please rename the function; all functions should be prefixed with _qrintf_
  • please add asserts that check the size of long long is 64 bits
  • use unsigned long long instead of uint64_t
• _qrintf_int_core
  • please accept a pointer next to the last char instead of a pointer to the last char, and use *--p instead of *p--
• the PR does not perform well on 32-bit architecture, please create long versions of ilog10ull, _qrintf_int_core and use them when stringifying the smaller types
  • IMO it can be implemented like https://gist.github.com/kazuho/83cde8ab0579b463f7bc

@kazuho
Thanks for the feedback. I updated the branch with the suggested changes:

• replace ilog10ull(uint64_t) to _qrintf_ilog10ull(unsigned long long) (6d84553)
• fix _qrintf_int_core. This api now accept a pointer next to the last char and use *--p (7148ea6)

And I fixed a bug which cause tests failure on 32bit architecture. (b7cb770)
If we still need to add long versions of _qrintf_ilog10ull, _qrintf_int_core, I will continue to work to implement these apis.

Thank you for the changes.

If we still need to add long versions of _qrintf_ilog10ull, _qrintf_int_core, I will continue to work to implement these apis.

It would be great if you could implement them. The reason I ask for them is because I do see performance degradation compared to master on 32-bit architecture. Your proposed optimization should work fine for both 32-bit and 64-bit platforms, and I am hoping that it would be implemented as such.

$ git checkout master
Switched to branch 'master'
Your branch is up-to-date with 'origin/master'.
$ bin/qrintf-gcc -m32 -O3 examples/ipv4addr.c && time ./a.out 1234567890
result: 73.150.2.210

real    0m0.352s
user    0m0.348s
sys 0m0.001s
$ git checkout -
Switched to branch 'pr/9'
Your branch is up-to-date with 'origin/pr/9'.
$ bin/qrintf-gcc -m32 -O3 examples/ipv4addr.c && time ./a.out 1234567890
result: 73.150.2.210

real    0m0.509s
user    0m0.504s
sys 0m0.003s

@kazuho
Thank you for your comment.
Performance degradations caused by this PR is not acceptable for me. So I'll try to fix it.

I've revisited this and I think the performance degradation was removed.
@kazuho, Could you check it?

$ git co master
Switched to branch 'master'
Your branch is up-to-date with 'origin/master'.

$./bin/qrintf-gcc  -O3 examples/ipv4addr.c && time ./a.out 1234567890
result: 73.150.2.210
./a.out 1234567890  0.22s user 0.00s system 99% cpu 0.219 total

$ ./bin/qrintf-gcc -m32 -O3 examples/ipv4addr.c && time ./a.out 1234567890
result: 73.150.2.210
./a.out 1234567890  0.27s user 0.00s system 99% cpu 0.268 total

$ git co optimize_itoa
Switched to branch 'optimize_itoa'
Your branch is up-to-date with 'imasahiro/optimize_itoa'.

$./bin/qrintf-gcc  -O3 examples/ipv4addr.c && time ./a.out 1234567890
result: 73.150.2.210
./a.out 1234567890  0.12s user 0.00s system 99% cpu 0.122 total

$ ./bin/qrintf-gcc -m32 -O3 examples/ipv4addr.c && time ./a.out 1234567890
result: 73.150.2.210
./a.out 1234567890  0.15s user 0.00s system 99% cpu 0.154 total

Thank you for the hard work! Merged to master.