Skip to main content

Module cpp_extension

Module cpp_extension

Expand description

C++ Extension utilities for ToRSh

This module provides utilities for building C++ extensions that integrate with the ToRSh framework, similar to PyTorch’s cpp_extension module.

Structs§

BuildResult: Build result containing the path to the compiled extension
CppExtensionConfig: Configuration for building a C++ extension
CrossPlatformConfig: Cross-platform build configuration
CudaDeviceInfo: CUDA device information
CudaJitInfo: CUDA JIT compilation information
CudaKernelCompilationOptions: Advanced CUDA kernel compilation options
CudaLaunchConfig: CUDA kernel launch configuration
CustomOpDefinition: Custom operation definition
JitBuildInfo: JIT compilation build information
JitCompilationConfig: JIT compilation configuration
LinuxConfig: Linux-specific build configuration
MacOsConfig: macOS-specific build configuration
OpSchema: Operation schema for validation and optimization
RuntimeCudaKernel: Runtime CUDA kernel management
TensorType: Tensor type information
WindowsConfig: Windows-specific build configuration

Enums§

CompilerPreference: Compiler preference on Linux
CustomOpType: Custom operation type
MemoryRequirement: Memory requirement estimation
TargetPlatform: Target platform specification

Functions§

auto_tune_cuda_kernel: Auto-tune CUDA kernel launch parameters
build_cpp_extension: Build a C++ extension
compile_cuda_kernel_runtime: Compile CUDA kernel at runtime
cuda_is_available: Check if CUDA is available for building extensions
generate_extension_template: Generate a simple C++ extension template
get_cuda_arch_list: Get CUDA architectures available on the system
launch_cuda_kernel: Launch a runtime-compiled CUDA kernel
load_cpp_extension: Load a C++ extension from a shared library