i'm interested to know whether it is possible that there is a difference in the performance between the QR decomp. and a Cholesky decomp. I know that both algorithms are of cubic
complexity (only differing in the constant factor) but my profiling results show strange behaviour. I have an 2 algorithms avoiding these decompositions and the interesting thing is that although both algorithms are structurally identical, the Qr-avoiding algorithm profits very much more of avoiding the QR decompositions in contrast to the Cholesky-avoiding one.
Therefore, i would like to know whether it is somehow possible that the performance of the QR decomposition implementation is somehow not that much optimized as the cholesky decomposition. Could this be possible?
Are there some performance measurements of both decompositions or profiling results i could have a look on?
Thanks for any advice/comment and help.