Skip to content

Releases: unitaryfund/qrack

Stability improvements and API refactor

01 May 04:23
f461f7d
Compare
Choose a tag to compare
  1. Cross entropy benchmarks have helped us debug QUnit to its most stable, accurate, and performant point to date.
  2. Builds have been tested on Linux, Windows, Apple, and with VC4CL on the Raspberry Pi 3.
  3. QFusion was determined to be obsolete, including for performance considerations, and it was removed.
  4. The "Hash" method has been added, which effectively lets the user specify any one-to-one function (without phase effects) for unitary qubit state transformation via a classical hash table.
  5. CMake options now include arbitrary (power-of-two) maximum qubit capacity in a single QUnit, with the Boost libraries for big integers. Of course, allocation is still limited by system resources, and the maximum OpenCL "shard" indexing is 64 qubits or less, (which is safely outside of possible resource limits for "Schrödinger method" simulation).
  6. Documentation has been brought into line with the features and benchmarks of the new release.

Qrack stack integration

25 Jan 00:20
696b1f0
Compare
Choose a tag to compare

At this point, Qrack has been integrated with and extensively tested against a wide, open source stack of third-party quantum computing libraries, as a first-order and a second-order dependency. This is also the state of development at which we plan submit an academic report on benchmarks, for peer review, thanks to the generous support of the Unitary Fund micro-grant program. In overview, major points of improvement include:

  • Debugging all existing work to pass extensive batteries of open source, third party unit tests, (particularly see the recent continuous integration tests and Travis definition file)
  • QUnit "phase gate buffering" optimizations, which cache, commute, and "fuse" 2-qubit controlled gates to further reduce required "representational entanglement" (re: "Schmidt decomposition")
  • Replacing float rounding neighborhoods with exact float value checks wherever applicable, and without loss of performance (which, upon scrutinization, turns out to more-or-less entirely generally)
  • Full support as a Qiskit plugin, in addition to existing support for ProjectQ, as well as tested support as a second-order dependency through those projects

Stable Release (for DOI reference)

16 Aug 03:01
18d7688
Compare
Choose a tag to compare

Qrack has been stable and "quiescent," for about the past month. The vm6502q.v3.3 "minor release" is primarily a fixed point of reference to create an official DOI. Documentation is also tagged.

Windows 10 and Mac support, and ApproxCompare() fix

17 Feb 04:22
a784caa
Compare
Choose a tag to compare

This build debugs support on Mac Mojave, adds support for building with Visual Studio on Windows 10, and fixes a longstanding bug in the ApproxCompare() method, which had been leading to unit test failures. The changes in code itself are minor, but since they add wide platform support, and stable general platform support was the aim of v3.1, we're tagging this version before any new functionality is added.

General hardware support and stack-readiness

09 Feb 01:20
074a808
Compare
Choose a tag to compare

This release improves stability across all tested hardware platforms and debugs support for optional integration with the vm6502q fork of ProjectQ.

What's new in Qrack v3.1:

  • Supports building on the Raspberry Pi 3 with VC4CL
  • Optional integration with vm6502q/ProjectQ fork has been completed
  • "Uniformly controlled" gates added to the API
  • Bug fix in TimeEvolve() method
  • General stability improvements across many different platforms

Known Issues:

ApproxCompare() has a sensitive threshold for state comparison, which causes unit tests based on it to fail sporadically. TryDecompose() and TrySeparate() also use ApproxCompare(). A low threshold means more conservative behavior. TryDecompose() and TrySeparate() are less likely to give false positives for true separability, but they are more likely to give false negatives.

API and performance improvements, gate fusion, and ProjectQ compatibility

26 Dec 17:51
fadccd5
Compare
Choose a tag to compare

This release changes the API to allow arbitrary numbers of control bits, adds a gate fusion layer called Qrack::QFusion, and makes major improvements to speed and memory usage.

OpenCL single gates now dispatch entirely asynchronously, returning to the main thread after adding gates to the OpenCL queue, so that work can continue on the CPU without blocking.

If possible, by default, Qrack will attempt to allocate its state vectors entirely on OpenCL device memory, which can recover up to about 1 or 2 GB of general heap for a small enough state vector, for many modern accelerators.

General performance improvements, including fewer redundant normalizations, have resulted in a large factor of improvement in overall speed, without sacrificing accuracy.

Bugs were fixed in QUnit, for edge cases for logic pertaining to explicit separability of qubits.

New methods have been added to help support quantum mechanics simulation, including the "TimeEvolve" method and a matrix exponentiation gate. The API has been generally updated to help support integration as a back end for ProjectQ. (https://github.com/ProjectQ-Framework/ProjectQ)

v2.0 - OpenCL Performance Improvements

16 Aug 02:48
Compare
Choose a tag to compare

What’s New in Qrack v2.0

  • Greatly improved all-around OpenCL performance (See https://qrack.readthedocs.io/en/two_point_zero/performance.html)
  • "Full OpenCL coverage” - QEngineOCL no longer inherits from QEngineCPU at all, and QEngineOCL state vector manipulations are virtually entirely done with OpenCL kernels and the OpenCL API.
  • Operator exponentiation methods have been added to the public API, (“Exp,” “ExpX,” “ExpY,” etc..) as well as a single bit gate method with a 2x2 complex matrix specified arbitrarily by the user
  • Experimental multi-processor engine, QEngineOCLMulti
  • Better explicit qubit separation in QUnit (less RAM, often greater speed, depending on use case)
  • Tested and debugged for single and multi-processor compatibility with the Intel HD. Issues diagnosed and fixed for the HD include OpenCL compilation for single-accuracy-float-only devices, as well as logical compatibility of kernel calls with an arbitrary number of processing elements on a device, as opposed to an exact power of 2 processing elements.
  • General minor bug fixes, (including small memory leaks, bad OpenCL group sizes, and others)

v1.7 - Public Release

25 May 17:55
f209e46
Compare
Choose a tag to compare

This release of the Qrack Quantum Simulator and the associated VM6502Q toolchain is suitable for public use and large scale deployment in production.