Add support for other multi-threading APIs#1667
Add support for other multi-threading APIs#1667krzikalla wants to merge 21 commits intogoogle:mainfrom
Conversation
Fork update
Support the benchmarking of code, which relies on other multi-threading APIs, e.g. OpenMP.
|
This is a follow-up to PR #1027 |
|
I guess i will need to review this. It's unfortunate that so much time has passed |
a usual State object instead of copying it. This fixes a bug, which otherwise happens, if one thread has already finished the benchmark loop and merged its results to the parent state, while another one hasn't started yet.
Support the benchmarking of code, which relies on other multi-threading APIs, e.g. OpenMP.
a usual State object instead of copying it. This fixes a bug, which otherwise happens, if one thread has already finished the benchmark loop and merged its results to the parent state, while another one hasn't started yet.
… into threadingapi_2
Any chance to get a review on this soon? Do I have to do something before you start? The issues I still have in the CI runners don't seem to be related to my additions. |
LebedevRI
left a comment
LebedevRI
left a comment
There was a problem hiding this comment.
I guess i will need to review this. It's unfortunate that so much time has passed since the previous attempt, i basically do not recall what specific issues i had there.
Any chance to get a review on this soon?
Wouldn't that be nice :(
Do I have to do something before you start?
No.
| // Don't create threads. Let the user evaluate state.threads and/or use | ||
| // ThreadState. | ||
| Benchmark* ManualThreading() { |
There was a problem hiding this comment.
This presumably should assert that thread_counts_ is empty.
There was a problem hiding this comment.
It is actually fine to have a non-empty thread_counts_. In that case the benchmark is executed with each thread count and the user is responsible for creating the appropriate number of threads. The first two examples in the user guide use this combination. Later src/benchmark_runner.cc:140 checks, if the thread count and the number of ThreadState created match.
| @@ -1274,6 +1296,13 @@ class BENCHMARK_EXPORT Benchmark { | |||
| // Equivalent to ThreadRange(NumCPUs(), NumCPUs()) | |||
| Benchmark* ThreadPerCpu(); | |||
There was a problem hiding this comment.
These all should probably assert !manual_threading_.
There was a problem hiding this comment.
As said above, it is fine to combine these functions with manual threading.
| int threads_; // Number of concurrent threads to us | ||
| bool manual_threading_; | ||
| bool explicit_threading_; // true: Number of threads come from a Threads() | ||
| // call |
There was a problem hiding this comment.
I haven't fully looked at things, but preliminary, i don't really like the names chosen.
and "Improve thread sanitzer statements"
Support the benchmarking of code, which relies on other multi-threading APIs, e.g. OpenMP.
… into threadingapi_2
|
Took a second look at this, after taking an extended mental break. So there's really a two distinct features there:
@krzikalla if this is still relevant, i think the 1. makes sense, |
|
OK, I have to get back into this, but I am fine with separating the PR. |
3 participants
Support the benchmarking of code, which relies on other multi-threading APIs, e.g. OpenMP.