Support static parameters for GlobalPhase and multiRZ gates. #1396
+92
−28
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
mehrdad2m
commented
Dec 20, 2024
frontend/catalyst/jax_tracer.py
Outdated
| @@ -201,7 +200,7 @@ def _eval_jaxpr(*args, **kwargs): | |||
| # Take care when adding primitives to this set in order to avoid introducing a quadratic number of | |||
| # edges to the jaxpr equation graph in ``sort_eqns()``. Each equation with a primitive in this set | |||
| # is constrained to occur before all subsequent equations in the quantum operations trace. | |||
| FORCED_ORDER_PRIMITIVES = {qdevice_p, gphase_p} | |||
| FORCED_ORDER_PRIMITIVES = {qdevice_p} | |||
There was a problem hiding this comment.
TODO: This needs to be handled in another way.
There was a problem hiding this comment.
Oh I wonder if this is the reason the primitive exists. Without qubits, it could be reordered before the device is instantiated, which would result in a crash I think
erick-xanadu
requested changes
Jan 2, 2025
There was a problem hiding this comment.
Thanks @mehrdad2m! I do see why this would be nice, but given the current constraints with the topological sorting, I believe the best, temporary, course of action is to leave the gphase_p primitive as is. Whenever we have time we should determine why the instructions get reordered. (I believe this has to do with how we trace.) However, depending on some future changes, this investigation may no longer be needed.
Regarding MultiRZ, I think it is ok to merge in a single primitive but it is not incredibly important. I'll leave it up to you if you want to merge it or close this PR without merging.
|
Agreed, let's just keep the |
…states, and use it to disentangle CNOTs and SWAPs (#1154) **Context:** In circuits that mostly involve the six Pauli eigenstates, the effect of gates can be described by a simple six-state finite state machine. This allows us to deduce their midcircuit state at compile time. The knowledge of the snapshot midcircuit states can be useful when: - Deducing CNOT/Toffoli/SWAP/... effect and decouple them. We decouple CNOT and SWAP in this PR. - Replace complicated circuits whose results are |01+-LR> with standard preparation routines - Potentially more **Description of the Change:** Added an analysis for propagating simple states (and tests). Also added CNOT and SWAP decomposition pass `disentangle-CNOT`, `disentangle-SWAP` (and tests). Credit to [Ritu Thombre](https://github.com/ritu-thombre99) for the work on the SWAP gates! **Benefits:** More compile time circuit information for potential compile time transformation/optimization. **Possible Drawbacks:** This algorithm is from an external paper. We need to properly cite it. https://arxiv.org/abs/2012.07711 (fig.5) **Related GitHub Issues:** closes #1268 [sc-74649] [sc-74650] --------- Co-authored-by: Ritu Thombre <[email protected]> Co-authored-by: Ritu Thombre <[email protected]>
**Context:** Forgotten changelog entry for PR #1387 **Description of the Change:** **Benefits:** **Possible Drawbacks:** **Related GitHub Issues:**
**Context:** Braket tests were removed temporarily because the dev version of PennyLane needed for Catalyst tests was incompatible with the Braket plugin in the interim between removing legacy operator arithmetic from PennyLane and releasing a version of the Braket plugin with no references to legacy opmath. Braket has released, so we expect things to work again now. **Description of the Change:** This PR reverses the changes disabling Braket tests in #1308 and #1321. [sc-78774]
**Context:** The [wheels are failing](https://github.com/PennyLaneAI/catalyst/actions/runs/12563739994/job/35026196938) after we put the Braket tests stuff back Looking at the wheels, it's ending up installing version 1.28 (we set the minimum to 1.27). We need a more recent version because `QubitDevice` isn't accessible at top level and legacy opmath has been removed. **Description of the Change:** Install with `>=0.31.0` instead. We also remove the `boto3` pin, because it was creating version conflicts and seems to no longer be needed.
**Context:** A number of OQD packages, modules and data files had been added recently (e.g. #1348 and #1355). However, as support for OQD devices is still under development, these files should not be included in the wheels. **Description of the Change:** Excludes _all_ OQD packages, modules and data files (such as toml config files) from the wheels. Furthermore, the OQD device entry in the [Supported devices documentation](https://docs.pennylane.ai/projects/catalyst/en/stable/dev/devices.html) has also been removed.
…ent from the wheels (#1406) **Context:** Switching to a different branch after setting up the dev wheel environment would break the latter. **Description of the Change:** Accept a branch to switch to as an optional argument, and switch to it before hard-linking the frontend source code files. **Benefits:** Start working directly on an existing branch. **Possible Drawbacks:** Switching to another branch using git would keep breaking the installation.
**Context:** Changelog --------- Co-authored-by: Joey Carter <[email protected]>
**Context:** In preparation for the Catalyst 0.10.0 release candidate (RC), the GitHub action must point to the corresponding RC branch, `v0.10.0-rc`. **Description of the Change:** Update the Catalyst RC branch `v0.9.0-rc` -> `v0.10.0-rc`, as well as the corresponding RC branches for Lightning and PennyLane from `v0.39.0` to `v0.40.0`.
**Context:** Preparing for the Catalyst v0.10.0 release candidate (RC). **Description of the Change:** Update versions in release notes and _version.py.
**Context:** Preparing the main branch for Catalyst v0.11.0-dev (development release). **Description of the Change:** Bump version to "0.11.0-dev" and add new changelog. --------- Co-authored-by: erick-xanadu <[email protected]>
**Context:** The new development release is v0.11.0, not v0.10.0. :confounded: **Description of the Change:** Fix typo in title of changelog for development release.
**Context:**  **Description of the Change:**  **Benefits:** No floating `}` and well formatted contents. **Possible Drawbacks:** Do we also want to show the libraries we use? **Related GitHub Issues:**
**Context:** Fix some typos and phrasing in the docs
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Context:
PR #1387, adds support for static parameters in customOp. However support for GlobalPhase and MultiRZ gates are missing.
Description of the Change:
This PR adds support for the missing two gates. We unified the way
GlobalPhaseandMultiRZare treated in the frontend. right now, GlobalPase has its own primitive (qphase_p) howeverMultiRZis binded through theqinst_pprimitive.qphase_pis now removed from frontend and binded throughqints_psimilar to MultiRZ.If a
MultiRZorGlobalPhasegate has static parameters parameters, astaticCustomOpis created which will be lowered to the respective gate in static_custom_lowering pass.Benefits:
removed inconsistency in the definition of odd custom gates (MultiRZ and GlobalPhase)
static circuit IR now supports more parameterized gates.
Possible Drawbacks:
Related GitHub Issues: