Run boolean and bitwise kernels for longer to reduce benchmark noise#8872
Run boolean and bitwise kernels for longer to reduce benchmark noise#8872alamb wants to merge 2 commits intoapache:mainfrom
Conversation
d6c560f to
53e05b2
Compare
53e05b2 to
dcd00cf
Compare
| fn bitwise_array_benchmark(c: &mut Criterion) { | ||
| let size = 64 * 1024_usize; |
There was a problem hiding this comment.
We often run these kernels on 8k elements, not 64k elements, so adjust the benchmark to reflect that
|
🤖 |
|
🤖 |
|
run benchmark bitwise_kernel |
|
🤖: Benchmark completed Details
|
|
🤖 |
|
🤖: Benchmark completed Details
|
|
Roughly speaking this shows what I expect which is the benchmarks taking longer |
|
I am not convinced this is the right approach. I think a better approach would be to run the kernels on multiple different sizes / allocations. I suspect alignment is causing benchmark artifacts |
Note to self:
Which issue does this PR close?
We generally require a GitHub issue to be filed for all bug fixes and enhancements and this helps us generate change logs for our releases. You can link an issue to this PR using the GitHub syntax.
Rationale for this change
The boolean and bitwise kernels are currently very noisy and sometimes get nonsensical results (e.g. #8854 (comment)) I think this is because they are so fast. For example, Here is an output from a recent run (note the benchmark each takes only nanoseconds
What changes are included in this PR?
Are these changes tested?
I will benchmark then
Are there any user-facing changes?
No these are benchmarks