1D Blocktiling Kernel has fundamental issues on data loading

Kernel 4: 1D Blocktiling for Calculating Multiple Results per Thread Each thread load only 1 element, while each thread compute TM results of C. How this will match?