Use subpackage Quasar for parsing, handle batch-running locally

[kshyatt-aws]

Issue #, if available: N/A

Description of changes:

• Use new subpackage Quasar.jl for parsing, based on Automa.jl. This allows precompilation of the parsing logic.
• Move translation of Python to Julia types to the Python wrapper package amazon-braket-simulator-v2-python
• Include batch processing logic in simulator code

Testing done:

• Unit tests passed
• New precompilation and parsing logic drops TTFX to 0.68

Merge Checklist

Put an x in the boxes that apply. You can also fill these out after creating the PR. If you're unsure about any of them, don't hesitate to ask. We're here to help! This is simply a reminder of what we are going to look for before merging your pull request.

General

• I have read the CONTRIBUTING doc
• I used the commit message format described in CONTRIBUTING
• I have updated any necessary documentation, including READMEs and API docs (if appropriate)

Tests

• I have added tests that prove my fix is effective or that my feature works (if appropriate)
• I have checked that my tests are not configured for a specific region or account (if appropriate)

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.

[speller26]

Just some small questions, still need to go through the big files (BraketSimulator.jl and Quasar.jl)

[speller26]

Is it worth mentioning that one control is equivalent to the phase gate and two are equivalent to cphase, just like the MultiRZ docstring?

[kshyatt-aws]

Not sure, it might be confusing? I can add though

[speller26]

Why add the arbitrary matrix param type? Same for other two.

[kshyatt-aws]

For Unitary, which doesn't have a SMatrix as its matrix representation (because you odn't know the size beforehand)

[speller26]

Makes sense; does this enable Unitary support that we didn't have before?

[kshyatt-aws]

It should allow arbitrary controlled unitaries on one qubit, with logic we can extend to controlled unitaries of any qubit count

[speller26]

Are we sure this is exhaustive? Just took a quick skim, but I think there are some missing characters👀

[kshyatt-aws]

Look at the gen_regex.jl script to see how this was generated

[speller26]

I didn't actually read this at all, I just have a chuckle every time I run across enormous auto-generated files as I'm reviewing code especially if they're non-human-readable

[kshyatt-aws]

We're actually not using this right now because it causes LLVM to choke to death. I have a to-do to improve this regex string which should improve TTFX substantially :)

[speller26]

Why move into this block?

[kshyatt-aws]

Look at the PrecompileTools.jl documentation for an explanation.

[speller26]

Has this been moved here from simulator-v2 as part of the transition to Julia OQ parsing?

[kshyatt-aws]

What? I don't understand this question. This is a purely Julia block for precompilation, why would it be in the Python wrapper?

[speller26]

Oh, i was just asking because this is newly introduced, but in the old world, I remember the circuits for pre compilation lived in the Python wrapper

[kshyatt-aws]

Yeah we moved all that here so you just precompile at first import, not every import

[speller26]

Is this case for precompiling classical control flow? If so, does it cover everything on that front?

[kshyatt-aws]

It's for precompiling a bunch of instructions, including subroutines, if-else, etc. We can add more workloads if necessary.

[speller26]

Just for clarification: is it necessary to go OQ -> Circuit -> Program, or can the middle step be cut out? Does the second step (Circuit -> Program) require another iteration over the circuit?

[kshyatt-aws]

We could cut out the middle step but I used it here to mirror what the existing sim does. The last step doesn't require another iteration over the circuit.

[speller26]

Is the shots > 0 check still needed? Looks like it's already done above

[kshyatt-aws]

We still don't want to add in spurious basis rotation instructions a weirdo user might add even if they intend to run shots=0

[speller26]

Why default to 32? Will this cause problems for CPUs with fewer than 32 logical cores?

[kshyatt-aws]

The reason is to avoid spawning too many short lived tasks if you have, for example, a batch of 10,000 3-qubit circuits. It's a relatively arbitrary pick that keeps Julia's scheduler from being overwhelmed if you have more than 32 cores. For machines with smaller core counts, the maximum will be Threads.nthreads() which should be equal to the CPU count

[speller26]

nit: Might be easier to read if the visitor was placed in a separate file from the parser

[speller26]

What's this TODO for?

[kshyatt-aws]

Casting to non-bool values explicitly. Currently none of our tests or examples are doing this.

[speller26]

As well as separating out the Braket-specific implementation

[speller26]

Definitely not to be done in this PR, but is it possible to give names to the args for each expression type, instead of relying on indices? That can make reading much easier.

[kshyatt-aws]

It's kind of hard because the args are a Vector{Any} by design, unless you mean instantiating some variable to be = expr.args[2]?

[speller26]

Exactly that, but only if it's done within each Expr subtype