Found 5 packages
Package Description
Production-ready NVIDIA CUDA GPU backend for DotCompute. Provides GPU acceleration (21-92x speedup) through CUDA with NVRTC compilation, P2P transfers, Ring Kernels with NCCL support, and unified memory. Requires CUDA 12.0+ and Compute Capability 5.0+ NVIDIA GPU. Benchmarked on RTX 2000 Ada (CC 8.9).
Attributes and enums for DotCompute kernel definition. This package provides the [Kernel], [RingKernel], and [RingKernelMessage] attributes used to define compute kernels.
Production-ready CPU compute backend for DotCompute. Provides SIMD vectorization (3.7x faster) using AVX2/AVX512/NEON instructions, multi-threaded kernel execution, and Ring Kernel simulation. Benchmarked: Vector Add (100K elements) 2.14ms → 0.58ms. Native AOT compatible with sub-10ms startup.
Production-ready OpenCL backend for DotCompute. Cross-platform GPU acceleration for NVIDIA, AMD, Intel, ARM Mali, and Qualcomm Adreno GPUs. Supports OpenCL 1.2+, Ring Kernels with atomic message queues, runtime kernel compilation, and multi-device workload distribution. Works with nvidia-opencl-icd, ROCm, intel-opencl-icd, and vendor drivers.