Create QsharpCore targeting package - #352
Conversation
Sorry, something went wrong.
This change splits all the intrinsics and decompositions defined in QsharpCore out into their own files to allow them to be individually combined into a target definition package. Q# should include decompositions to alternate quantum gate sets #249
|
This PR is just to make viewing the incremental changes easier. We can have a larger review when this eventually makes it from my swernli/decomp-working branch to the feature/decomp branch (after resolving the current pending merge of master there). |
Sorry, something went wrong.
| @@ -0,0 +1,39 @@ | |||
| <?xml version="1.0" encoding="utf-8"?> | |||
There was a problem hiding this comment.
Alan Geller (@alan-geller) FYI, this is what target definition "package" will look like. This allows any project to pull in the API surface defined for this target by just importing this props file in the project. Let me know if you have any questions or concerns regarding this approach!
Sorry, something went wrong.
There was a problem hiding this comment.
I like it!
I can see how for simulators like Toffoli or CHP (or the full state simulator, of course), and for the hardware targets as well, it's very clean to get exactly the operations you want. It also avoids the need to repeat the decompositions for each simulator or hardware target they apply to.
I assume that if I have a simulator (or hardware target, for that matter) that implements some additional operation as an intrinsic, I would just add an additional file to the Intrinsic directory and include that rather than the Decompositions version? So, for instance, if I had an intrinsic SWAP, I'd create a SWAP.qs in the Intrinsic directory and include that in my project instead of the one from the Decompositions folder?
I guess I could also create a TargetDefinition.qs file in my project where I include any "overrides" like that. Which do you think will be better?
Sorry, something went wrong.
There was a problem hiding this comment.
Regarding how to extend this, I was imagining the first option you describe: create a new SWAP.qs in the Intrinsic folder and link against that one instead of the decomposition in your props file. Right now this is all very clean and easy, where the names of the files match the names of the callable within, but that won't last forever. For example, if two targets have different intrinsics, they might need different implementations of the same decomposed callable, so you could end up with something like SWAPfromCNOT.qs vs SWAPfromControlledR.qs where the name is meant to be succinctly but uniquely descriptive of what's inside even though both contain an implementation of the same callable, namely Microsoft.Quantum.Intrinsic.SWAP.
I'll be getting into this very shortly with my next working branch PR and we can continue to discuss the approach there.
Sorry, something went wrong.
| Adjoint _flipToBasis(stateId, qubits); | ||
| AssertMeasurementProbability(observable, qubits, Zero, expectedZeroProbability, $"", Accuracy()); | ||
| AssertMeasurementProbability(observable, qubits, One, expectedOneProbability, $"", Accuracy()); | ||
| Adjoint flipToBasis(stateId, qubits); |
There was a problem hiding this comment.
Could you do this with a within/apply construct? within flipToBasis apply the two assserts?
Sorry, something went wrong.
There was a problem hiding this comment.
Very likely! I didn't look too hard at the implementations from the perspective of refactoring to use the latest syntax, but I think that's something we should definitely do at somepoint.
Sorry, something went wrong.
| @@ -0,0 +1,39 @@ | |||
| <?xml version="1.0" encoding="utf-8"?> | |||
There was a problem hiding this comment.
I like it!
I can see how for simulators like Toffoli or CHP (or the full state simulator, of course), and for the hardware targets as well, it's very clean to get exactly the operations you want. It also avoids the need to repeat the decompositions for each simulator or hardware target they apply to.
I assume that if I have a simulator (or hardware target, for that matter) that implements some additional operation as an intrinsic, I would just add an additional file to the Intrinsic directory and include that rather than the Decompositions version? So, for instance, if I had an intrinsic SWAP, I'd create a SWAP.qs in the Intrinsic directory and include that in my project instead of the one from the Decompositions folder?
I guess I could also create a TargetDefinition.qs file in my project where I include any "overrides" like that. Which do you think will be better?
Sorry, something went wrong.
This change splits all the intrinsics and decompositions defined in QsharpCore out into their own files to allow them to be individually combined into a target definition package. Q# should include decompositions to alternate quantum gate sets #249
This change splits all the intrinsics and decompositions defined in QsharpCore out into their own files to allow them to be individually combined into a target definition package.
Q# should include decompositions to alternate quantum gate sets #249