/*******************************************************************************/
/
/ These are the examples of ABACUS program.
/
/*******************************************************************************/

# GPU Acceleration Examples

This directory contains examples for performing GPU-accelerated calculations in ABACUS. GPU acceleration can significantly speed up electronic structure calculations, particularly for large systems and computationally intensive tasks.

## What is GPU Acceleration?

GPU acceleration is the use of graphics processing units (GPUs) to perform computational tasks that were traditionally handled by central processing units (CPUs). GPUs are particularly well-suited for:

- Parallel computing tasks
- Matrix operations
- Electronic structure calculations
- Large-scale simulations

In ABACUS, GPU acceleration can provide significant speedups for:

- Plane wave basis set calculations
- LCAO basis set calculations
- Density matrix operations
- Self-consistent field (SCF) iterations

## GPU Acceleration in ABACUS

To enable GPU acceleration in ABACUS, set the following parameter in the INPUT file:

```
INPUT_PARAMETERS
gpu              1    # Enable GPU acceleration
```

### Key GPU Parameters:

- `gpu`: Set to 1 to enable GPU acceleration
- `gpu_device`: Specify which GPU device to use (default: 0)
- `gpu_stream`: Number of GPU streams to use (default: 4)
- `gpu_block_size`: Block size for GPU kernels (default: 256)

## Examples Included

### 1. 01_pw_Si16
- **System**: Silicon supercell with 16 atoms
- **Basis**: Plane wave (PW)
- **Purpose**: Demonstrates GPU-accelerated calculation for plane wave basis
- **Input Files**:
  - `INPUT`: Contains the GPU acceleration parameters
  - `KPT`: Defines the k-point sampling
  - `STRU`: Describes the silicon supercell structure

### 2. 02_lcao_Si16
- **System**: Silicon supercell with 16 atoms
- **Basis**: LCAO (Linear Combination of Atomic Orbitals)
- **Purpose**: Demonstrates GPU-accelerated calculation for LCAO basis
- **Input Files**:
  - `INPUT`: Contains the GPU acceleration parameters
  - `KPT`: Defines the k-point sampling
  - `STRU`: Describes the silicon supercell structure

## How to Run

1. Navigate to the example directory:
   ```bash
   cd /abacus/examples/34_gpu/01_pw_Si16
   ```

2. Run ABACUS:
   ```bash
   abacus
   ```

3. Check the output files in the `OUT.ABACUS` directory.

## Output Files

After running the calculation, you will find the following key output files in the `OUT.ABACUS` directory:

- `running_scf.log`: Log file for the SCF calculation, includes details about GPU setup

## Notes

- GPU acceleration requires a compatible NVIDIA GPU and CUDA installation
- The performance gain from GPU acceleration depends on the system size and the specific calculation type
- For small systems, the overhead of data transfer between CPU and GPU may outweigh the performance benefits
- For large systems, GPU acceleration can provide significant speedups (often 2-10x faster than CPU-only calculations)
- The examples provided are for demonstration purposes; for production calculations, you should carefully test the performance with different GPU parameters

## Applications

GPU-accelerated calculations in ABACUS can be used to study:

- Large supercells and nanostructures
- Complex materials with many atoms
- Molecular dynamics simulations
- High-throughput screening
- Systems requiring dense k-point sampling
- Time-dependent DFT calculations

## Performance Tips

To maximize GPU performance in ABACUS:

- Use larger systems to amortize the CPU-GPU data transfer overhead
- Choose appropriate GPU parameters based on your hardware
- For multi-GPU systems, consider using MPI to distribute the workload
- Monitor GPU memory usage to avoid out-of-memory errors
- Ensure that your CUDA drivers and ABACUS are properly configured for your GPU hardware