|
tket
|
Pass the raw gate unitary matrix elements into this class one by one, but let this class perform any extra optimisation. More...
#include <GateNodesBuffer.hpp>
Public Member Functions | |
| GateNodesBuffer (Eigen::MatrixXcd &matrix, double abs_epsilon) | |
| The full (2^n)*(2^n) unitaries of the nodes will be calculated, and left-multiply the matrix. | |
| ~GateNodesBuffer () | |
| void | push (const GateNode &node) |
| Process and store the next node, possibly with optimisation, e.g. | |
| void | add_global_phase (double) |
| void | flush () |
| The buffer might be cleared and optimised regularly, but the caller must call this at the end to ensure that any leftover nodes are also processed; updates the original matrix passed into the constructor. | |
Pass the raw gate unitary matrix elements into this class one by one, but let this class perform any extra optimisation.
(So, we leave open future options to optimise more aggressively when simulating).
E.g., if you had consecutive 1-qubit gates acting on the same qubit, it might be quicker to multiply the small 2x2 matrices together before expanding to a big (2^n)*(2^n) matrix.
If you had, e.g. a 2-qubit gate acting on {0,1} followed by another acting on {1,0}, it might be quicker to multiply the 4x4 matrices together, with an additional 4x4 lifted permutation generated by the qubit permutation {0,1} -> {1,0}.
If you had successive gates acting on qubits {0}, {1,2}, {0}, {3,4}, {2,1}, the gates on {0} can be combined into one gate, and also the gates on {1,2} and {2,1}, because the sets {0}, {1,2}, {3,4} are disjoint. Thus, we might store data for multiple gates.
If you had gates acting on {0,1,2} and {2,0}, or {0,1,2} and {0,2,3}, you might merge them, i.e. if two qubit index sets have a small symmetric difference than you can combine the small unitaries together before lifting and copying the entries.
Of course, this would all be simulating the exact same gates, just in a computationally more efficient way; allowing the gates themselves to be changed could give yet more speedup possibilities.
Definition at line 52 of file GateNodesBuffer.hpp.
| tket::tket_sim::internal::GateNodesBuffer::GateNodesBuffer | ( | Eigen::MatrixXcd & | matrix, |
| double | abs_epsilon | ||
| ) |
The full (2^n)*(2^n) unitaries of the nodes will be calculated, and left-multiply the matrix.
The matrix must remain valid throughout the lifetime of this object, and the caller should not alter the matrix in any other way.
| matrix | The matrix representing the full unitary of the circuit arising the gates added in sequence, in ILO-BE convention. |
| abs_epsilon | Used to convert almost-zero entries to zero entries: any z with std::abs(z) <= abs_epsilon is treated as zero. |
Definition at line 62 of file GateNodesBuffer.cpp.
| tket::tket_sim::internal::GateNodesBuffer::~GateNodesBuffer | ( | ) |
Definition at line 65 of file GateNodesBuffer.cpp.
| void tket::tket_sim::internal::GateNodesBuffer::add_global_phase | ( | double | ph | ) |
Definition at line 69 of file GateNodesBuffer.cpp.
| void tket::tket_sim::internal::GateNodesBuffer::flush | ( | ) |
The buffer might be cleared and optimised regularly, but the caller must call this at the end to ensure that any leftover nodes are also processed; updates the original matrix passed into the constructor.
Definition at line 73 of file GateNodesBuffer.cpp.
| void tket::tket_sim::internal::GateNodesBuffer::push | ( | const GateNode & | node | ) |
Process and store the next node, possibly with optimisation, e.g.
combining gates etc. Note that this class might not update the original matrix immediately.
Definition at line 67 of file GateNodesBuffer.cpp.
1.9.8