#moon-child and #rygorous discussing #GPU #floating-point and #FMA and dot products and round-to-negative vs. round-toward-zero etc.
on 02023-10-07how to get theoretical max #performance of 4 flops per cycle with #SSE intrinsics: manual loop unrolling and careful interleaving of multiplies and adds, at least for pre-#FMA processors.
on 02015-12-05