Quadrants

Quadrants is a high-performance parallel programming framework for GPU and CPU computing. Write Python-like code that compiles to optimized GPU kernels for CUDA, Metal, and Vulkan backends.

import quadrants as qd

qd.init(arch=qd.gpu)

@qd.kernel
def hello(a: qd.types.NDArray[qd.i32, 1]) -> None:
    for i in range(10):
        a[i] = i * 2

a = qd.ndarray(qd.i32, (10,))
hello(a)

Features

• Simple: annotate Python functions with @qd.kernel to run on GPU

• Fast: automatic parallelization of top-level for loops across GPU threads

• Portable: supports CUDA, Metal, AMD, and Vulkan backends

• Flexible: ndarrays, fields, structs, atomics, shared memory

Quadrants

• User guide
  • Getting started
  • Supported systems
  • Tensor types
  • Scalar tensors
  • Matrix and Vector
  • Per-thread matrix and vector operations
  • Per-thread linear algebra
  • Tensors
  • Compound types
  • BufferView
  • qd.static
  • Sub-functions
  • Parallelization
  • Numpy and Torch interop
  • Shared Metal command queue (PyTorch MPS)
  • Automatic differentiation
  • Atomics
  • Block primitives
  • Grid primitives
  • Math
  • Subgroup primitives
  • Tile16x16: register-resident 16x16 tiles
  • Algorithms
  • Fastcache
  • Graph
  • Streams
  • Performance Dispatch
  • qd.init options
  • Kernel code coverage
  • Python features in kernel scope
  • Python backend
  • Debug mode
  • Quirks
  • Troubleshooting
  • Contributing to quadrants
  • Building the CUDA graph conditional fatbin
• API Reference
  • quadrants