CUDA BLAS

A CUDA implementation of Basic Linear Algebra Subprograms (BLAS).

Current Features

Level 1 BLAS Operations

• axpy: Compute y = αx + y
• scal: Compute x = αx
• asum: Compute the sum of absolute values

All operations currently support single-precision floating point (float) with plans to extend to other data types.

Requirements

• CUDA Toolkit
• C++ Compiler with C++11 support
• Make

Project Structure

cuda-blas/
├── include/          # Header files
├── src/             # Implementation files
├── examples/        # Example usage
└── tests/           # Test files

Building

# Build everything
make all

# Run examples
./bin/level1_examples

# Clean build files
make clean

Usage Example

#include "blas.cuh"

int main() {
  const int n = 5;
  const float alpha = 2.0f;
  float x[] = {1, 2, 3, 4, 5};
  float y[] = {1, 1, 1, 1, 1};

  // Compute y = αx + y
  axpy(n, alpha, x, 1, y, 1);
  
  return 0;
}

Planned Features

• Support for double precision
• Support for complex numbers
• Additional Level 1 BLAS operations
• Implementation of Level 2 and 3 BLAS operations

---

Feel free to reach out with questions or feedback!