ASM CUDA Kernels

A pure, low-level collection of modular NVIDIA CUDA / PTX guest kernels and assembly routines, completely decoupled from host code and structured for autonomous compilation.

Repository Structure

The repository is organized by functional categories, where each kernel resides in its own self-contained leaf directory equipped with a local build target pointing to the shared template:

asm-cuda/
├── basics/                 # Fundamental GPU verification and test kernels
│   ├── gpu_clock/          # GPU cycle and clock measurement
│   ├── gpu_hello/          # Basic execution test
│   ├── gpu_pointertest/    # Pointer arithmetic and memory validation
│   ├── gpu_thread_write/   # Concurrent thread writing tests
│   └── gpu_write/          # Basic global memory write operations
├── docs/                   # Documentation and reference materials
├── finance/                # Financial and stochastic simulation kernels
│   └── gbm_monte_carlo/    # Geometric Brownian Motion (Monte Carlo path simulation)
├── include/
│   └── shared_kernel.mk    # Centralized, autonomous build template for all kernels
├── math/                   # Mathematical and vector operations
│   ├── complex_quadratic_solver/ # Direct analytic solver for complex quadratic coefficients
│   └── gpu_vector_add/     # Parallel vector addition
└── LICENSE

Build Instructions

Each kernel is completely autonomous. To build a specific kernel, navigate into its directory and use make:

# Navigate to a specific kernel directory
cd basics/gpu_clock

# Build in DEBUG mode (includes line info and SASS disassembly dump)
make debug

# Build in RELEASE mode (optimized with -O3)
make release

# Clean build artifacts
make clean

Requirements

• NVIDIA CUDA Toolkit (ptxas, nvdisasm)
• GNU Make