Stride Calculation innacurate

https://github.com/siboehm/SGEMM_CUDA/blob/60cba6f9b20a198116c76f18de8047f44df8c8b8/src/kernels/5_kernel_2D_blocktiling.cuh#L50C1-L51C1 Stride calculation shouldn't be: const uint strideB = numThreadsBlocktile / BN; with a growing number of threads, the stride should decrease as each thread has less to calculate; the formula is backwards. it could be instead: BN*BK / numThreadsBlocktile