CUDA-Q Logical

Getting started

  • Quick start
    • Install
    • Running the shipped examples
    • Step 1 — Estimate an existing CUDA-Q kernel
    • Step 2 — Author a portable P0 program
    • Step 3 — Choose a code and a gadget (P2)
    • Step 4 — Lower to physical events and schedule them (P3)
    • Step 5 — Emit Stim text from the command line
    • Where to go next
  • The logical programming stack
    • 1. Write the logical program
    • 2. Place the qubits on a logical machine
    • 3. Define a QEC realization
    • 4. Estimate a selected realization
    • 5. Emit a verified gadget as Stim
    • What each stage owns
    • Where to go next
  • Core concepts
    • 1. Four stages, one direction
    • 2. Facets, not extra stages
    • 3. Linear ownership
    • 4. Codes, profiles, encodings — three different things
    • 5. Objectives, gadgets, protocols — claims and proofs
    • 6. Evidence follows the program
    • 7. There is no registry — imports are the linker
    • 8. Global phase is not observable
    • Where the pieces live
    • Where to go next
  • CUDA-Q Logical by Example
    • CUDA-Q kernel examples
      • Logical Resource Estimate
      • Clifford+T Synthesis
      • CUDA-Q Logical Resource Estimate
      • Fermi–Hubbard Estimation
      • Carbon Code Target
      • Gidney–Ekerå RSA-2048
    • Standalone compiler examples
      • Portable P0 Bell Program
      • P1 Placement
      • Code and Gadget
      • T-State Distillation
      • Standalone Physical Schedule
      • Gidney–Ekerå Lookup Addition

Use cases

  • Defining codes
    • CSS codes: state the checks once
    • Distance is evidence, not an integer
    • The catalog and parameterized families
    • What you get for free
    • Where to go next
  • Gadgets and verification
    • What implements= takes
      • Author the objective in CUDA-Q Logical
      • Reuse a CUDA-Q kernel as the claim
      • What the signature and body add
    • Typed records at the boundary
    • Preparation and destructive measurement
    • Selection: retry and postselection belong to the protocol
    • Verification: claims are checked, not trusted
    • Design rules that keep the model crisp
    • Where to go next
  • Devices and placement
    • Machines, devices, architectures
      • Stop at the layer your study needs
    • Automatic placement: constraints, preferences, witnesses
    • Code-agnostic P1
    • Where the firewall pays off
    • Continue from here
  • Magic states and protocols
    • Phase convention
      • Authoring exact angles with cudaq.logical.algebra.pi
    • Typed resource kinds
    • 15-to-1: a concrete factory
  • Logical Clifford+T synthesis
    • Synthesize a program
    • Precision
    • Inspect and replay the result
    • Use the pipeline form
    • Pauli-based computation, still at P0
    • When not to synthesize
    • Continue from here
  • Resource estimation
    • Tier.LOGICAL — cost the algorithm before any QEC choice
    • Tier.STATIC — count one selected P2 realization
    • Tier.ANALYTICAL — apply an explicit physical model
    • Tier.SCHEDULE — cost the physical event graph
    • The rules that keep estimates honest
    • Direct spellings
    • Sweeps and reproducibility
    • Paper-specific projections live in the open
    • Estimating ordinary CUDA-Q kernels
  • Stim emission
    • Emit from the command line
    • Fail-closed boundaries

Reference

  • CUDA-Q Logical for Stim users
    • The translation table
    • Worked emission
    • What has no Stim equivalent
    • When to just use Stim
  • Capabilities and status
    • What the product is
    • Subsystem status
    • Present but not yet exercised
    • Documented boundaries (all fail closed)
  • Architecture
    • The dialect stack
    • The semantic spine and its facets
    • Definitions versus builds
    • The artifact model
    • Data ownership
    • The compile pipeline
    • The P2 patch graph
    • Verification layers
    • Typed inspection
    • Package map
    • Extensibility
    • Where to go next
  • Building against CUDA-Q
    • Option 1 (default): the cudaq-devel wheel
    • Option 2: a CUDA-Q installation built from source
    • Build options
    • Verifying the build
    • Troubleshooting

CUDA-Q

  • Main CUDA-Q documentation
CUDA-Q Logical
  • Search

© Copyright Copyright © 2026, NVIDIA Corporation & affiliates. All rights reserved..

Built with Sphinx using a theme provided by Read the Docs.