Fail when calculating complexity

[Skabunkel]

When calculating time complexity and the range includes 0 it fails upon evaluation.

void BM_Example(benchmark::State& state)
{
  while (state.KeepRunning())
  {
    int i = state.range(0);
  }
  state.SetComplexityN(state.range(0));
}

BENCHMARK(BM_JumpTable)->RangeMultiplier(2)->Range(0, 1)->Complexity(benchmark::o1);

Result

F:\cruft\benchmark\src\complexity.cc:270: benchmark::ComputeBigO: Check (run.complexity_n) > (0)' Failed. Did you forget to call SetComplexityN?

[dmah42]

I get:

../src/complexity.cc:270: ComputeBigO: Check `(run.complexity_n) > (0)' failed. Did you forget to call SetComplexityN?

when i try your snippit.

i believe it's because you're using SetComplexityN with an argument 0. See the documentation on SetComplexityN:

404   // If this routine is called with complexity_n > 0 and complexity
report is
405   // requested for the

406   // family benchmark, then current benchmark will be part of the
computation
407   // and complexity_n will

408   // represent the length of N.

I'm not convincing myself that setting it to 0 is wrong, but it is unexpected by the library, at least. Perhaps we shouldn't use '0' as a signal that SetComplexityN wasn't called.

Dominic Hamon | Google There are no bad ideas; only good ideas that go horribly wrong.

On Fri, Jun 30, 2017 at 9:46 AM, Niklas Augustsson <[email protected]

wrote:

When calculating code complexity and the range includes 0 it crashes upon evaluation.

void BM_JumpTable(benchmark::State& state) { while (state.KeepRunning()) { int i = state.range(0); } state.SetComplexityN(state.range(0)); }

BENCHMARK(BM_JumpTable)->RangeMultiplier(2)->Range(0, 1)->Complexity(benchmark::o1);

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/google/benchmark/issues/410, or mute the thread https://github.com/notifications/unsubscribe-auth/AAfIMoUVoNMp_MJfmKKRnexUrAxtinLnks5sJKfVgaJpZM4OKQUj .

[Skabunkel]

The primary reason it is 0 is because of a copy paste fail on my part, from the section Calculate asymptotic complexity (Big O).

I do however consider 0 to a possible value. A possible solution would be to check if n amount of them satisfy the condition complexity_n > 0 where n is a significant subset of the total amount of benchmarks.

i'm unsure if i can come with any better suggestions since i have little reference regarding the code flow of complexity_n.

[dmah42]

i think we should differentiate between SetComplexityN wasn't called and SetComplexityN was called with 0.