Worked on enhancing the TritonGPU WMMA layout path within the facebookexperimental/triton repository, focusing on compiler development and GPU programming using C++. Addressed a critical bug by refining the logic for disabling swizzling on the B operand, ensuring it only applies when the k dimension is not contiguous. Introduced a heuristic approach to more accurately determine layout parameters such as vectorSize, perPhase, and maxPhase, which improved the stability and correctness of WMMA layout calculations. The work demonstrated a strong grasp of low-level optimization and cross-hardware considerations, contributing to more reliable and portable performance in the TritonGPU dialect.