qat.experimental.conversion.pulse_to_q1.rewrite_patterns module

Rewrite patterns for the Pulse-to-Q1 phase legalisation and lowering stages.

class LowerArithIntegerConstantToMoveOp

Bases: ModulePass, RewritePattern

Lower arith.constant with an integer or index type to q1.ir.move.

An arith.constant whose result type is IntegerType or IndexType is replaced by a q1.ir.move instruction that loads the same immediate value into a virtual register. Constants whose result type is not an integer or index (e.g. floating-point) are left unchanged.

The immediate must fit in the 32-bit signed-or-unsigned range accepted by q1.ir.move (SU32Imm: [-2**31, 2**32 - 1]). Values outside that range raise PassFailedException.

apply(ctx, op)

Return type:: None

match_and_rewrite(op: Operation, rewriter: PatternRewriter) → None

Match an operation, and optionally perform a rewrite using the rewriter.

Return type:: None

name: ClassVar[str] = 'lower-arith-integer-constant-to-move'

class RewritePhaseSetOp(target_data, rewrite_callable)

Bases: RewritePattern

Match pulse.phase_set and delegate stage policy to rewrite_callable.

The same matcher is used in legalisation and lowering. The callable decides whether the rewrite remains in Pulse or emits Q1 instructions.

match_and_rewrite(op: Operation, rewriter: PatternRewriter) → None

Match an operation, and optionally perform a rewrite using the rewriter.

Return type:: None

class RewritePhaseShiftOp(target_data, rewrite_callable)

Bases: RewritePattern

Match pulse.phase_shift and delegate stage policy to rewrite_callable.

Mirrors RewritePhaseSetOp in structure.

match_and_rewrite(op: Operation, rewriter: PatternRewriter) → None

Match an operation, and optionally perform a rewrite using the rewriter.

Return type:: None

class RewritePreQ1AcquireOp(target_data)

Bases: RewritePattern

Lower PreQ1AcquireOp to a Q1 acquire instruction.

AcquireImmRsImmOp when no integration weights are present.
AcquireWeightedImmRsRsRsImmOp when a WeightsAttr is present.

The acquisition, and any integration weights, are registered on the enclosing SequenceOp and assigned hardware indices in encounter order. Q1 requires the bin and weight indices to be supplied in registers. Bin indices retain their upstream value, while static weight indices are materialised directly as Q1 moves to unallocated integer registers. A later register-allocation pass assigns concrete Q1 GPRs.

match_and_rewrite(op: Operation, rewriter: PatternRewriter) → None

Match an operation, and optionally perform a rewrite using the rewriter.

Return type:: None

class RewritePulseOp(target_data)

Bases: RewritePattern

Lowers a pulse.pulse op to a Q1 play instruction.

The pattern resolves the pulse’s sampled waveform, registers its real (I) and imaginary (Q) sample vectors in the enclosing sequence’s waveform table, and replaces the pulse.pulse op with a PlayImmImmImmOp that references the two table indices and the waveform duration. The now-unused waveform ConstantOp[SampledWaveformAttr] is erased once no other op consumes it.

Distinct sampled waveforms are assumed to have been folded and de-duplicated upstream, so identical waveforms share a single ConstantOp. Each such op is therefore registered in the waveform table only once per sequence. Subsequent pulses referencing the same op reuse the cached table indices instead of re-comparing sample arrays.

Initialise the pattern.

Parameters:: target_data¶ (QbloxTargetData) – Qblox target data used during lowering.

match_and_rewrite(op: Operation, rewriter: PatternRewriter) → None

Match an operation, and optionally perform a rewrite using the rewriter.

Return type:: None

class RewriteStartContinuousWaveformOp(target_data)

Bases: RewritePattern

Lowers a pulse.start_continuous_waveform op to a Q1 AWG offset.

A continuous waveform is emitted on Q1 hardware by latching a constant AWG offset on both output paths. This pattern replaces the pulse op with a SetAwgOffsImmImmOp carrying the I and Q offsets scaled to the DAC range. Assumes Square waves have been legalised to StartContinuousWaveformOp, Delay, and StopContinuousWaveformOp.

Initialise the pattern.

Parameters:: target_data¶ (QbloxTargetData) – Qblox target data used during lowering.

match_and_rewrite(op: Operation, rewriter: PatternRewriter) → None

Match an operation, and optionally perform a rewrite using the rewriter.

Return type:: None

class RewriteStopContinuousWaveformOp(target_data)

Bases: RewritePattern

Lowers a pulse.stop_continuous_waveform op to a Q1 AWG offset reset.

Stopping a continuous waveform corresponds to clearing the latched AWG offset on both output paths, emitted as a SetAwgOffsImmImmOp with zero offsets. Assumes Square waves have been legalised to StartContinuousWaveformOp, Delay, and StopContinuousWaveformOp.

Initialise the pattern.

Parameters:: target_data¶ (QbloxTargetData) – Qblox target data used during lowering.

match_and_rewrite(op: Operation, rewriter: PatternRewriter) → None

Match an operation, and optionally perform a rewrite using the rewriter.

Return type:: None

class RewriteWaitOp(target_data)

Bases: RewritePattern

Lower pulse.wait to one or more q1.wait instructions.

The requested duration is converted from seconds to nanoseconds and aligned up to the sequencer grid time. Durations that exceed the maximum wait immediate are split into a chain of q1.wait instructions whose durations sum to the requested value. The frame carried by pulse.wait is forwarded to downstream operations.

Register-driven durations that do not fold to a compile-time constant are left untouched. Those are handled elsewhere and are out of scope for this lowering.

match_and_rewrite(op: Operation, rewriter: PatternRewriter) → None

Match an operation, and optionally perform a rewrite using the rewriter.

Return type:: None

create_legalisation_patterns()

Create the rewrite set used by the legalisation stage.

Canonicalises pulse.phase_set and pulse.phase_shift operands inside the Pulse dialect. New legalisation behaviour can be added here without touching the lowering factory.

Return type:: tuple[RewritePattern, ...]
Returns:: Ordered pattern tuple for the legalisation pass.

create_pulse_to_q1_lowering_patterns(target_data=QbloxTargetData(max_acquisitions=10000, default_shots=1000, QUBIT_DATA=QubitDescription(sample_time=1e-09, samples_per_clock_cycle=1, instruction_memory_size=50000, waveform_memory_size=1500, pulse_duration_min=6.4e-08, pulse_duration_max=0.001, pulse_channel_lo_freq_min=1000000, pulse_channel_lo_freq_max=10000000000, pulse_channel_if_freq_min=0, pulse_channel_if_freq_max=10000000000, passive_reset_time=0.001), RESONATOR_DATA=ResonatorDescription(sample_time=1e-09, samples_per_clock_cycle=1, instruction_memory_size=50000, waveform_memory_size=1500, pulse_duration_min=6.4e-08, pulse_duration_max=0.001, pulse_channel_lo_freq_min=1000000, pulse_channel_lo_freq_max=10000000000, pulse_channel_if_freq_min=0, pulse_channel_if_freq_max=10000000000), driver_version=SemanticVersion(major=1, minor=3, patch=1, prerelease=None, build=None), fw_version=SemanticVersion(major=0, minor=13, patch=0, prerelease=None, build=None), Q1ASM_DATA=Q1asmDescription(min_gain=-32768, max_gain=32767, min_offset=-32768, max_offset=32767, max_wait_time=65532, register_size=4294967295, loop_unroll_threshold=4), CONTROL_SEQUENCER_DATA=ControlSequencerDescription(grid_time=4, nco_min_freq=-500000000.0, nco_max_freq=500000000.0, nco_max_phase_steps=1000000000, nco_phase_steps_per_deg=2777777.777777778, nco_freq_steps_per_hz=4, nco_freq_limit_steps=2000000000.0, number_of_registers=64, min_acq_integration_length=4, max_acq_integration_length=16777212, min_acq_threshold=-16777212, max_acq_threshold=16777212, max_sample_size_waveforms=16384, max_num_instructions=16384, sample_rate=1000000000.0), READOUT_SEQUENCER_DATA=ReadoutSequencerDescription(grid_time=4, nco_min_freq=-500000000.0, nco_max_freq=500000000.0, nco_max_phase_steps=1000000000, nco_phase_steps_per_deg=2777777.777777778, nco_freq_steps_per_hz=4, nco_freq_limit_steps=2000000000.0, number_of_registers=64, min_acq_integration_length=4, max_acq_integration_length=16777212, min_acq_threshold=-16777212, max_acq_threshold=16777212, max_sample_size_waveforms=16384, max_num_instructions=12288, sample_rate=1000000000.0), QCM_DATA=QcmDescription(number_of_sequencers=6, min_qcm_offset_v=-2.5, max_qcm_offset_v=2.5, output_connections={0: [0, 1, 2, 3, 4, 5], 1: [0, 1, 2, 3, 4, 5], 2: [0, 1, 2, 3, 4, 5], 3: [0, 1, 2, 3, 4, 5]}), QCM_RF_DATA=QcmRfDescription(number_of_sequencers=6, min_qcm_rf_offset_mv=-84, max_qcm_rf_offset_mv=73, min_out_att_db=0, max_out_att_db=60, output_connections={0: [0, 1, 2, 3, 4, 5], 1: [0, 1, 2, 3, 4, 5]}), QRM_DATA=QrmDescription(number_of_sequencers=6, min_sample_size_scope_acquisitions=4, max_sample_size_scope_acquisitions=16384, max_binned_acquisitions=3000000, min_qrm_offset_v=-0.09, max_qrm_offset_v=0.09, output_connections={0: [0, 1, 2, 3, 4, 5], 1: [0, 1, 2, 3, 4, 5]}, input_connections={0: [0, 1, 2, 3, 4, 5], 1: [0, 1, 2, 3, 4, 5]}), QRM_RF_DATA=QrmRfDescription(number_of_sequencers=6, min_sample_size_scope_acquisitions=4, max_sample_size_scope_acquisitions=16384, max_binned_acquisitions=3000000, min_qrm_rf_offset_v=-0.09, max_qrm_rf_offset_v=0.09, min_out_att_db=0, max_out_att_db=60, min_in_att_db=0, max_in_att_db=30, output_connections={0: [0, 1, 2, 3, 4, 5]}, input_connections={0: [0, 1, 2, 3, 4, 5]}), QRC_DATA=QrcDescription(number_of_sequencers=12, min_sample_size_scope_acquisitions=4, max_sample_size_scope_acquisitions=16384, max_binned_acquisitions=7000000, number_of_readout_sequencers=8, number_of_control_sequencers=4, min_out_att_db=0.0, max_out_att_db=31.5, min_in_att_db=0.0, max_in_att_db=31.5, output_connections={0: [0, 1, 2, 3, 4, 5, 6, 7], 1: [0, 1, 2, 3, 4, 5, 6, 7], 2: [0, 4, 8, 9, 10, 11], 3: [1, 5, 8, 9, 10, 11], 4: [2, 6, 8, 9, 10, 11], 5: [3, 7, 8, 9, 10, 11]}, input_connections={0: [0, 1, 2, 3, 4, 5, 6, 7], 1: [0, 1, 2, 3, 4, 5, 6, 7]})))

Create the rewrite set used by the lowering stage.

Phase entries are configured with PhaseLowering. Wait, finite pulse, continuous-waveform, and acquisition patterns lower their Pulse or pre-Q1 operations to Q1 instructions. Pulse preprocessing owns synchronization, while hardware configuration resolution owns initial NCO frequency.

Parameters:: target_data¶ (QbloxTargetData) – Qblox limits used during lowering.
Return type:: tuple[RewritePattern, ...]
Returns:: Ordered pattern tuple for the lowering pass.