On Tue, Apr 11, 2023 at 1:15 PM robert bristow-johnson < r...@audioimagination.com> wrote:
> It really *isn't* that much logic to fix denorms going in and out of the > ALU. Isn't it? A normal float multiply requires one integer addition for the exponents and one (normalized) fixed-point multiplication of the mantissas. The denormal case adds a rescaling step (exponent add + mantissa shift), potentially on both input and output. Worse, that extra rescaling is conditional on the input/output values, so the CPU pipeline optimization gets hosed. Ethan D