Test QPP state allocation #1143
Closed
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
* Add a reference to the Quick Start Guide. (NVIDIA#1109) Co-authored-by: Bettina Heim <heimb@outlook.com> * UCCSD Bug Fixes NVIDIA#952 (NVIDIA#1099) * Fix uccsd issue NVIDIA#952 --------- Signed-off-by: Alex McCaskey <amccaskey@nvidia.com> * REST Server-Client platform (NVIDIA#1012) * [WIP] prototyping a server-client platform * minor code edits * [WIP] prototype remote execution * [WIP] Work on remote server * [WIP] REST remote server-client implementation - Server [sim]: ability to run MPI; set simulator dynamically, start on a specific port - Client: make is multi-qpu. - Target config: pass urls and remote simulator. * some code clean-up * more code tidy-up * Better code structure * EOF missing * unused var * properly handle observe * Make passes configurable * code format * Fix CI * Clang build fixes * Fix issue when switching simulators * Don't build server app for wheels * Fix build after merge * [WIP] work on library mode support * More work for library mode support - Handle kernel name lookup with manged name. - Fix issues with curl for large POST requests. Use bitcode to save bandwidth. * [WIP] More experimenation to support library mode * Prevent local execution when using remote platform in library mode * Rejit the library mode support - Workaround a bug in 'symbol responsibility' assertion when using mlir::Execution engine for pure llvm bitcode. Use LLJIT directly. - Basic support for argument passing. - Don't use the entry/exit marker functions. Rather add a redirection similar to std::invoke for clarity. * code format * Fixed cmake script * Add auto launch feature * Another code format * Fix clang build * Add JSON schema version check * Add wait to auto-launch * Add some basic tests * Add observe/observe_async tests Fixed an oversight of not passing the full sample_result back from the server Fix const correctness in MeasureCounts to enable Json conversions. * add kernel_builder tests for remote REST platform * Add Python tests * Fix for build tree vs. install tree * Add tests and fixes for free function kernels * Add tests for get_state/get_state_async with remote platform * Attempt to fix CI test failures on ARM * some cleanup and docs * some more code docs * Fixed Doxygen parser bugs * Add docs * Fixed spelling * Update docs/sphinx/using/cudaq/platform.rst Co-authored-by: Pradnya Khalate <148914294+khalatepradnya@users.noreply.github.com> * Address code review comments: code readability * Address code review comments: update code comments * Fix a typo in doc * Add OpenAPI spec for the Json pauloads * Bypass license header check for the OpenAPI spec yml and conform the payload data to the spec * BypaAddress code review: escape code in docstring * Some docs changes * Update docs/sphinx/using/cudaq/platform.rst Co-authored-by: Eric Schweitz <eschweitz@nvidia.com> * Update docs/sphinx/using/cudaq/platform.rst Co-authored-by: Eric Schweitz <eschweitz@nvidia.com> * Update docs/sphinx/using/cudaq/platform.rst Co-authored-by: Eric Schweitz <eschweitz@nvidia.com> * Address code review - Add thread_id getter to base QuantumExecutionQueue - Rename RemoteSimulatorQPU.cpp - Remove outdated comment about .qplt file. * Code review: simplify if/else return value and code format * Add comments to OpenAPI REST spec YAML * Remove dead code: unused class * Core review: REST server to report errors back as HTTP status 500 * Minor tweak for kernel_builder Use the canonical pattern of calling jitCode(). * Walkthrough followup: First pass of code refactor Main goal is to add abstraction to make the code extensible: - Adding interfaces to abstract server and client. - Using LLVM registry to inject the REST-specific impl. Next steps: high-level integration: unifying with the mqpu platform and moving the rest server to tools/ * Walkthrough follow-up #2: Code restructure (1) Merging remote platform (client-side) into the existing `mqpu` platform: - Abstract the QPU (GPU-emulated or remote) via plugin. - Isolate CUDA-dependency to build `mqpu` platform with/without CUDA. (2) Move the server app to tools/, add proper CLI args parsing, rename it to `cudaq-qpud`. * Fix CMake fallout from my refactor Find CUDA was not properly relocated to the GPU-only subdirectory (now mqpu doesn't require CUDA). Fix a typo in target name. * Fix another CMake refactor fallout: missing fmt linking, didn't catch locally * Address code review: make sure that no custom host compiler was used * Code review: Enhance cudaq-qpud autolaunch helper Better error handling and reporting. Add a polling loop to guarantee connection rather than wait for fixed amount of time. * Change target config key to be compatible with Python * Fixes some minor issues found - Handle the case whereby the kernel is wrapped in `cudaq::gradient` before passing to `cudaq::observe`. - Fix a fallout from code refactor: set_seed on the server side needs to be after the simulator lib loading since it'll try to access the simulator. * Missing a field in the ExecutionContext Json serialization * Update JSON schema Nest response execution context as a sub-field with a proper key so that it's in the same path as the request. Update the openAPI yaml accordingly. * Minor tweaks to get the examples working for the remote platform Don't pass kernels as arguments to entry-point invocation. Make sure that a simple compilation with `--target remote-sim` just works. * Fix spell check * Fixed an issue for lambda kernels in library mode and add a test to cover that * Minor updates to the docs * Debug container validation CI * Revert "Debug container validation CI" This reverts commit 0ddf6b7. * Fix CI: the app needs to be able to work without MPI support * Code review: remove SYNTAX_CHECK in tests * Update docs/sphinx/using/cudaq/platform.rst Co-authored-by: Ben Howe <141149032+bmhowe23@users.noreply.github.com> * Update docs/sphinx/using/cudaq/platform.rst Co-authored-by: Ben Howe <141149032+bmhowe23@users.noreply.github.com> * Update docs/sphinx/using/cudaq/platform.rst Co-authored-by: Ben Howe <141149032+bmhowe23@users.noreply.github.com> * Update runtime/common/KernelWrapper.h Co-authored-by: Ben Howe <141149032+bmhowe23@users.noreply.github.com> * Update runtime/common/KernelWrapper.h Co-authored-by: Ben Howe <141149032+bmhowe23@users.noreply.github.com> * Code review: Update docs * Code review: Add code comment * Code review: enum -> enum class * Update runtime/cudaq/cudaq.cpp Co-authored-by: Ben Howe <141149032+bmhowe23@users.noreply.github.com> * Code review: remore a stray comment * Code review: code format change * Code review: fix log messages * Code review: use make_unique * Code review: make args parsing more robust * Update runtime/cudaq/platform/mqpu/MultiQPUPlatform.cpp Co-authored-by: Ben Howe <141149032+bmhowe23@users.noreply.github.com> * Code review: use C++20 starts_with * Code review: thread_safety for static default context and code comment * Adding code comments about a skipped test and rest server request handling * Fixed white spaces * Fixed RST underline too short error in docs build * Reduce CI time * Attempt to fix race conditions when multiple apps with auto-launch starting concurrently * Debug container validation logs * Revert changes to examples * Debug CI container validation * Exclude tests with callable args * Handle base64 encoded option * Better handling no-MPI case: don't repeatedly query MPI plugin registry. Only need to check once * Add error logging for kernel callable as args and tests * Revert debug codes * Code review: simplify the code * Add new deps to the NOTICE file * Fixes after merge - Adjust CMake for installer build and add missing headers for OSX - Fixed installer validation for remote-sim The validate_container.sh is shared b/w docker and installer validation. The installer validation requires an additional step install_sshclient.sh to run examples that use cudaq::mpi API (remote-sim). Add a separate remote sim test for the installer - Update docs * Update docs/sphinx/using/cudaq/platform.rst Co-authored-by: Bettina Heim <heimb@outlook.com> * Update docs/sphinx/using/cudaq/platform.rst Co-authored-by: Bettina Heim <heimb@outlook.com> * Update docs/sphinx/using/cudaq/platform.rst Co-authored-by: Bettina Heim <heimb@outlook.com> * Update docs/sphinx/using/cudaq/platform.rst Co-authored-by: Bettina Heim <heimb@outlook.com> * Update platform name and docs - Rename remote-sim -> remote-mqpu. - Update docs: + Adding back some intro about auto-launch. + Add examples. * Fix Cmake for added docs snippet tests * Fix a ref in docs * Fix lit test RUN lines after merge * Revert "Fix lit test RUN lines after merge" This reverts commit 047baa8. * Try another fix * Add reference to Github issue for disabled tests * Update docs/sphinx/using/cudaq/platform.rst Co-authored-by: Bettina Heim <heimb@outlook.com> * Update docs according to review comments * Update platform.rst * Update sample_async_remote.py * Guard the example test that requires cutensornet --------- Co-authored-by: Pradnya Khalate <148914294+khalatepradnya@users.noreply.github.com> Co-authored-by: Eric Schweitz <eschweitz@nvidia.com> Co-authored-by: Ben Howe <141149032+bmhowe23@users.noreply.github.com> Co-authored-by: Bettina Heim <heimb@outlook.com> * MPI plugin docs (NVIDIA#1095) * Fix deployment validation post merge (NVIDIA#1116) * Including uninstall script when installed via installer (NVIDIA#1112) * Fixing wheelname in python_wheels.yml * Fix an oversight in MultiGPU refactor (NVIDIA#1119) --------- Signed-off-by: Alex McCaskey <amccaskey@nvidia.com> Co-authored-by: Bettina Heim <heimb@outlook.com> Co-authored-by: Alex McCaskey <amccaskey@nvidia.com> Co-authored-by: Thien Nguyen <58006629+1tnguyen@users.noreply.github.com> Co-authored-by: Pradnya Khalate <148914294+khalatepradnya@users.noreply.github.com> Co-authored-by: Ben Howe <141149032+bmhowe23@users.noreply.github.com>
…ion (NVIDIA#1115) Initial library mode implementation --------- Signed-off-by: Alex McCaskey <amccaskey@nvidia.com> Co-authored-by: A.M. Santana <39563805+anthony-santana@users.noreply.github.com>
…st (NVIDIA#1134) to the bridge.
…VIDIA#1141) * Add support for the new qvector constructor with an initialization list to the bridge. * Update add-dealloc to account for InitializeStateOp being present. Add test.
Signed-off-by: A.M. Santana <anthonys@nvidia.com>
anthony-santana
commented
Jan 26, 2024
Signed-off-by: A.M. Santana <anthonys@nvidia.com>
schweitzpgi
reviewed
Jan 31, 2024
| qpp::ket initState = | ||
| qpp::ket::Map(const_cast<std::complex<double> *>(stateData), count); | ||
| qpp::ket initState = qpp::ket::Map( | ||
| const_cast<std::complex<double> *>(stateData), (1UL << count)); |
There was a problem hiding this comment.
Might want to note in the doxygen comment that count is the power-of-2 exponent.
schweitzpgi
force-pushed
the
features/qalloc
branch
2 times, most recently
from
dec7672
to
c6239f6
Compare
…ion (NVIDIA#1115) Initial library mode implementation --------- Signed-off-by: Alex McCaskey <amccaskey@nvidia.com> Co-authored-by: A.M. Santana <39563805+anthony-santana@users.noreply.github.com>
…st (NVIDIA#1134) to the bridge.
…VIDIA#1141) * Add support for the new qvector constructor with an initialization list to the bridge. * Update add-dealloc to account for InitializeStateOp being present. Add test.
* Fixes for the lower to QIR pass.
- Splits subgraph matching from the conversion pass. The conversion
rewriter doesn't really let one do both at the same time given the
way it traversing the IR.
- Add a new codegen dialect. Only used in codegen. Part of the
previous bullet. The dialect will be the logical point to add any
future combination ops.
- Removes the extra canonicalizer pass
* clang-format
These changes adjust the placement of quake.dealloc ops to account for the presence of state initializers in local scopes.
* [Core] Add complex types and values. Fix traversals of casts. Cast expressions sometimes didn't actually traverse the information required. Update the test. Get the test case working end-to-end. * clang-format * Update test for run variations. * Reformat the comments.
schweitzpgi
force-pushed
the
features/qalloc
branch
from
c29e92a
to
2f94cd8
Compare
* Add qudit initializer to bridge. * Add a hack to work around visiting allocators, etc. Need to eventually handle "NOP" types and arguments in a better way than skipping over them when it is convenient.
…er (NVIDIA#1474) at runtime According to MLIR: "The process of loading a Dialect in the context is not thread-safe, which forces all involved Dialects to be loaded before the multi-threaded pass manager starts the execution." This PR makes sure that we load all the dialects when MLIR is needed at runtime, hence may be used in a multi-threading environment (async. execution). In particular, add a method to register the internal CodeGenDialect, which is only used by passes (not the frontend). Make sure that we load all Dialects in C++ (CUDAQ MLIRRuntime) and Python (builder/decorator). Tested by: running python/tests/remote/test_remote_platform.py with these changes on the features/qalloc branch.
Signed-off-by: A.M. Santana <anthonys@nvidia.com>
Signed-off-by: A.M. Santana <anthonys@nvidia.com>
…ion (NVIDIA#1115) Initial library mode implementation --------- Signed-off-by: Alex McCaskey <amccaskey@nvidia.com> Co-authored-by: A.M. Santana <39563805+anthony-santana@users.noreply.github.com>
…st (NVIDIA#1134) to the bridge.
…VIDIA#1141) * Add support for the new qvector constructor with an initialization list to the bridge. * Update add-dealloc to account for InitializeStateOp being present. Add test.
* Fixes for the lower to QIR pass.
- Splits subgraph matching from the conversion pass. The conversion
rewriter doesn't really let one do both at the same time given the
way it traversing the IR.
- Add a new codegen dialect. Only used in codegen. Part of the
previous bullet. The dialect will be the logical point to add any
future combination ops.
- Removes the extra canonicalizer pass
* clang-format
These changes adjust the placement of quake.dealloc ops to account for the presence of state initializers in local scopes.
* [Core] Add complex types and values. Fix traversals of casts. Cast expressions sometimes didn't actually traverse the information required. Update the test. Get the test case working end-to-end. * clang-format * Update test for run variations. * Reformat the comments.
* Add qudit initializer to bridge. * Add a hack to work around visiting allocators, etc. Need to eventually handle "NOP" types and arguments in a better way than skipping over them when it is convenient.
…er (NVIDIA#1474) at runtime According to MLIR: "The process of loading a Dialect in the context is not thread-safe, which forces all involved Dialects to be loaded before the multi-threaded pass manager starts the execution." This PR makes sure that we load all the dialects when MLIR is needed at runtime, hence may be used in a multi-threading environment (async. execution). In particular, add a method to register the internal CodeGenDialect, which is only used by passes (not the frontend). Make sure that we load all Dialects in C++ (CUDAQ MLIRRuntime) and Python (builder/decorator). Tested by: running python/tests/remote/test_remote_platform.py with these changes on the features/qalloc branch.
* Compute the correct number of qubits. The initializer length must be 2**q where q is the number of qubits. * Make the log calls consistent. Fix up the QIR checks.
schweitzpgi
force-pushed
the
features/qalloc
branch
from
e4ceb2a
to
82802de
Compare
…er (NVIDIA#1490) Add power-of-two tests.
* [featues/qalloc] Use cc dialect. Port the intrinsics to cc dialect, which should have everything required. Add external/private linkage control. * Review comment: replace the inline MLIR string with direct builder calls. * Replace expanded complex types. * Remove more LLVMIR code.
…eVecSimulator kron prod (NVIDIA#1482) * Update builder to support vector state input. Implement Cusv kronprod Signed-off-by: Alex McCaskey <amccaskey@nvidia.com>
Signed-off-by: A.M. Santana <anthonys@nvidia.com>
Signed-off-by: A.M. Santana <anthonys@nvidia.com>
Signed-off-by: A.M. Santana <anthonys@nvidia.com>
Signed-off-by: A.M. Santana <anthonys@nvidia.com>
anthony-santana
requested review from
bettinaheim,
amccaskey and
bmhowe23
as code owners
anthony-santana
changed the base branch from
features/qalloc
to
experimental/stateHandling
Signed-off-by: A.M. Santana <anthonys@nvidia.com>
schweitzpgi
reviewed
Apr 26, 2024
CMakeLists.txt
Outdated
| @@ -129,6 +129,7 @@ option(CUDAQ_TEST_MOCK_SERVERS "Enable Remote QPU Tests via Mock Servers." OFF) | |||
| option(CUDAQ_DISABLE_RUNTIME "Build without the CUDA Quantum runtime, just the compiler toolchain." OFF) | |||
| option(CUDAQ_SKIP_MPI "Do not build with MPI, even when it is available." OFF) | |||
| option(CUDAQ_DISABLE_CPP_FRONTEND "Build without the CUDA Quantum C++ Clang-based Frontend." OFF) | |||
| option(CUDAQ_ENABLE_CC "Enable CUDA Quantum code coverage generation." OFF) | |||
There was a problem hiding this comment.
Can we use something other than CC for code coverage?
(CC is already overloaded with it being an environment variable, a dialect, and a C compiler.)
schweitzpgi
reviewed
Apr 26, 2024
Signed-off-by: A.M. Santana <anthonys@nvidia.com>
Signed-off-by: A.M. Santana <anthonys@nvidia.com>
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
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.
Description