Benchmark compilation time for cancel_inverses and merge_rotations, catalyst-vs-PL #1207
Conversation
|
Hello. You may have forgotten to update the changelog!
|
|
drafting this --- no need to waste CI on this |
paul0403
changed the title
paul0403
changed the title
|
|
||
| auto stop = std::chrono::high_resolution_clock::now(); | ||
| auto duration = std::chrono::duration_cast<std::chrono::microseconds>(stop - start); | ||
| llvm::errs() << "merge rotation pass runtime: " << duration.count() << " microseconds\n"; |
There was a problem hiding this comment.
@paul0403 quick question, I saw I was tagged for review but I assume this branch is not going to be merged into main? 😅
There was a problem hiding this comment.
No need to review, this is just for visibility, or if others want to use the compile time timing/plotting script (which I will add)
|
There is no need for bespoke timing code, that's what the instrumentation is for :) |
| @@ -323,6 +332,7 @@ def run_writing_command(command: List[str], compile_options: Optional[CompileOpt | |||
|
|
|||
| DEFAULT_ASYNC_PIPELINES = [ | |||
| ENFORCE_RUNTIME_INVARIANTS_PASS, | |||
| PEEPHOLE_BENCHMARK_PASS, | |||
There was a problem hiding this comment.
irrelevant, as async is not used in the benchmark
paul0403
changed the title
peephole_benchmark_driver.py
Outdated
| def run_one_circuit(timings, core_PL_timings, num_of_iters): | ||
| do(f"python3 my_toy_circuit.py {num_of_iters}") | ||
|
|
||
| with open('my_toy_circuit.yml', 'r') as f: |
There was a problem hiding this comment.
Oh no, you can use a yaml package to parse it for you and access the elements like dictionaries, lists, etc :)
|
Might want to run |
|
Please forgive my abhorrent code quality in these scripts (which are only supposed to be used by me) : ))))) |
| print(loopsizes, walltimes, cputimes, programsizes, core_PL_times) | ||
|
|
||
|
|
||
| plt.figure(figsize=(15, 9)) |
There was a problem hiding this comment.
One thing we may want to consider when harmonizing our plot styles is to use a consistent figsize, since that will change the size of the text relative to the rest of the figure when pasting these plots into slides or a document, for example.
There was a problem hiding this comment.
Good point, I actually had this in mind (hence why I chose the same colors as your plots) but didn't think about the font!
Let's talk about it when the data for all three plots are locked down. It also depends on the specs of the final presentation, etc.
cc @josh146
Context:
Add a timer to cancel inverse and merge rotation passes for compile time measurements.
https://app.shortcut.com/xanaduai/epic/70803
Branch usage: