8.6.4. Best Practices - 5.2 English - 57404

1. Use Shared Library for Development: Faster iteration during development

2. Test with Static Library Before Deployment: Verify performance matches shared library

3. Set Threading Explicitly: Don’t rely on defaults

   dlp_thread_set_num_threads(8);
   

4. Reuse Reordered Matrices: For repeated GEMM with the same weights

5. Check CPU Features at Runtime: AOCL-DLP automatically selects best kernel for your CPU

6. Use Post-Operations for Fused Kernels: Leverage fused operations for better performance

For comprehensive integration documentation including troubleshooting and advanced topics, see the Integration Guide.