Re: copy/copy_backward/fill/fill_n/equal rework

[François Dumont]

Ping ?

On 9/9/19 8:34 PM, François Dumont wrote:

Hi

    This patch improves stl_algobase.h copy/copy_backward/fill/fill_n/equal implementations. The improvements are:

- activation of algo specialization for __gnu_debug::_Safe_iterator (w/o _GLIBCXX_DEBUG mode)

- activation of algo specialization for _Deque_iterator even if mixed with another kind of iterator.

- activation of algo specializations __copy_move_a2 for something else than pointers. For example this code:

std::vector<char> v { 'a', 'b', .... };

ostreambuf_iterator out(std::cout);

std::copy(v.begin(), v.end(), out);

is not calling the specialization __copy_move_a2(const char*, const char*, ostreambuf_iterator<>);

It also fix a _GLIBCXX_DEBUG issue where the __niter_base specialization was wrongly removing the _Safe_iterator<> layer. The testsuite/25_algorithms/copy/debug/1_neg.cc test case was failing on a debug assertion because _after_ the copy we were trying to increment the vector iterator after past-the-end. Of course the problem is the _after_, Debug mode should detect this _before_ it takes place which it does now.

Note that std::fill_n is now making use of std::fill for some optimizations dealing with random access iterators.

Performances are very good:

Before:

copy_backward_deque_iterators.cc    deque 2 deque 1084r 1084u 0s         0mem    0pf copy_backward_deque_iterators.cc    deque 2 vector 3373r 3372u 0s         0mem    0pf copy_backward_deque_iterators.cc    vector 2 deque 3316r 3316u 0s         0mem    0pf copy_backward_deque_iterators.cc    int deque 2 char vector 3610r 3609u    0s         0mem    0pf copy_backward_deque_iterators.cc    char vector 2 int deque 3552r 3552u    0s         0mem    0pf copy_backward_deque_iterators.cc    deque 2 list 10528r 10528u 0s         0mem    0pf copy_backward_deque_iterators.cc    list 2 deque 2161r 2162u 0s         0mem    0pf copy_deque_iterators.cc      deque 2 deque                 752r 751u    0s         0mem    0pf copy_deque_iterators.cc      deque 2 vector               3300r 3299u    0s         0mem    0pf copy_deque_iterators.cc      vector 2 deque               3144r 3140u    0s         0mem    0pf copy_deque_iterators.cc      int deque 2 char vector      3340r 3338u    1s         0mem    0pf copy_deque_iterators.cc      char vector 2 int deque      3132r 3132u    0s         0mem    0pf copy_deque_iterators.cc      deque 2 list                 10013r 10012u    0s         0mem    0pf copy_deque_iterators.cc      list 2 deque                 2274r 2275u    0s         0mem    0pf equal_deque_iterators.cc     deque vs deque               8676r 8675u    0s         0mem    0pf equal_deque_iterators.cc     deque vs vector              5870r 5870u    0s         0mem    0pf equal_deque_iterators.cc     vector vs deque              3163r 3163u    0s         0mem    0pf equal_deque_iterators.cc     int deque vs char vector     5845r 5845u    0s         0mem    0pf equal_deque_iterators.cc     char vector vs int deque     3307r 3307u    0s         0mem    0pf

After:

copy_backward_deque_iterators.cc    deque 2 deque  697r  697u 0s         0mem    0pf copy_backward_deque_iterators.cc    deque 2 vector  219r  218u 0s         0mem    0pf copy_backward_deque_iterators.cc    vector 2 deque  453r  453u 0s         0mem    0pf copy_backward_deque_iterators.cc    int deque 2 char vector 1914r 1915u    0s         0mem    0pf copy_backward_deque_iterators.cc    char vector 2 int deque 2112r 2111u    0s         0mem    0pf copy_backward_deque_iterators.cc    deque 2 list 7770r 7771u 0s         0mem    0pf copy_backward_deque_iterators.cc    list 2 deque 2194r 2193u 0s         0mem    0pf copy_deque_iterators.cc      deque 2 deque                 505r 504u    0s         0mem    0pf copy_deque_iterators.cc      deque 2 vector                221r 221u    0s         0mem    0pf copy_deque_iterators.cc      vector 2 deque                398r 397u    0s         0mem    0pf copy_deque_iterators.cc      int deque 2 char vector      1770r 1767u    0s         0mem    0pf copy_deque_iterators.cc      char vector 2 int deque      1995r 1993u    0s         0mem    0pf copy_deque_iterators.cc      deque 2 list                 7650r 7641u    2s         0mem    0pf copy_deque_iterators.cc      list 2 deque                 2270r 2270u    0s         0mem    0pf equal_deque_iterators.cc     deque vs deque                769r 768u    0s         0mem    0pf equal_deque_iterators.cc     deque vs vector               231r 230u    0s         0mem    0pf equal_deque_iterators.cc     vector vs deque               397r 397u    0s         0mem    0pf equal_deque_iterators.cc     int deque vs char vector     1541r 1541u    0s         0mem    0pf equal_deque_iterators.cc     char vector vs int deque     1623r 1623u    0s         0mem    0pf

In Debug Mode it is of course even better. I haven't had the patience to run the benches before the patch, it just takes hours to run. So here is just the After part:

copy_backward_deque_iterators.cc    deque 2 deque 1128r 1128u 0s         0mem    0pf copy_backward_deque_iterators.cc    deque 2 vector  616r  616u 0s         0mem    0pf copy_backward_deque_iterators.cc    vector 2 deque  856r  855u 0s         0mem    0pf copy_backward_deque_iterators.cc    int deque 2 char vector 2277r 2277u    0s         0mem    0pf copy_backward_deque_iterators.cc    char vector 2 int deque 2518r 2519u    0s         0mem    0pf copy_backward_deque_iterators.cc    deque 2 list 8029r 8028u 0s         0mem    0pf copy_backward_deque_iterators.cc    list 2 deque 10418r 10416u 0s         0mem    0pf copy_deque_iterators.cc      deque 2 deque                 931r 930u    0s         0mem    0pf copy_deque_iterators.cc      deque 2 vector                613r 613u    0s         0mem    0pf copy_deque_iterators.cc      vector 2 deque                794r 795u    0s         0mem    0pf copy_deque_iterators.cc      int deque 2 char vector      2192r 2192u    0s         0mem    0pf copy_deque_iterators.cc      char vector 2 int deque      2365r 2364u    0s         0mem    0pf copy_deque_iterators.cc      deque 2 list                 8009r 8010u    0s         0mem    0pf copy_deque_iterators.cc      list 2 deque                 10979r 10970u    1s         0mem    0pf equal_deque_iterators.cc     deque vs deque               1034r 1032u    0s         0mem    0pf equal_deque_iterators.cc     deque vs vector               481r 482u    0s         0mem    0pf equal_deque_iterators.cc     vector vs deque               646r 646u    0s         0mem    0pf equal_deque_iterators.cc     int deque vs char vector     1802r 1802u    0s         0mem    0pf equal_deque_iterators.cc     char vector vs int deque     1867r 1865u    0s         0mem    0pf

Note that copy/copy_backward 'list 2 deque' is still much slower because for the moment we can't remove the debug layer. I plan to do a refinement to also cover this use case.

In Debug implementations I have duplicated the Debug checks because those algo specialization will also be used when users are directly using <debug/vector> for instance without defining _GLIBCXX_DEBUG.

All algos tests passed except the constexpr ones in Debug mode, I've put this on my todo list.

Ok to commit once I completed all the other tests ?

François