Worked on performance optimization for the GEMM Block Scale Kernel within the ROCm/composable_kernel repository, focusing on enhancing throughput for core GPU workloads. Developed and integrated a hot loop scheduler to improve the overlap between data loading and computation, directly targeting bottlenecks in the kernel’s execution flow. Leveraged C++ and GPU programming expertise to refine data loading mechanisms and applied clang-format for consistent code style across the kernel. Collaborated closely with another contributor to ensure robust implementation. This work established a foundation for measurable GEMM throughput improvements and streamlined future optimizations, demonstrating depth in performance engineering and collaborative development practices.