Kernels from zero
One CUDA kernel, a matrix transpose, rewritten one idea at a time: first correct, then coalesced, then tiled, then measured.
RoobLiSeries
Essays that come in parts, read in order, each with its track.
One CUDA kernel, a matrix transpose, rewritten one idea at a time: first correct, then coalesced, then tiled, then measured.
1 item under this folder.