Development

This guide is for contributors extending the solver: build system, adding new kernels or solvers, and running the test suite.

Build System

Dual build approach for flexibility: - Makefile: Primary build for CUDA/MPI binaries - CMake: Testing framework with Google Test

# Release build (default)
make

# Debug build with GPU debugging (-g -G)
make BUILD_TYPE=debug

# Build specific targets
make cg_solver_mgpu_stencil
make generate_matrix

# Run tests
cd tests && mkdir build && cd build
cmake .. && make && ./test_runner

Adding Features

1. New SpMV kernel: Implement in src/spmv/, register in get_operator()
2. New solver: Add to src/solvers/, create entry point in src/main/
3. Performance metrics: Extend benchmark_stats_mgpu_partitioned.cu

Testing

# All tests
./test_runner

# Specific test suite
./test_runner --gtest_filter="PartitionedSolver*"