Always thought it was strange that fptan also pushes 1 to the register stack. It now makes sense it’s so existing code for the 8087 which was expected to do y/x to get the actual tangent could keep working by doing y/1 on newer processors.

Author here for your 8087 questions...

Love these deep dives. It’s fascinating to see how real, low-level engineering we all take for granted today unfolded. It’s incredible that the rough equivalent of an entire 160lb digital computer was built and baked into silicon.

Also incredible reverse engineering. In a time when seemingly all appreciation for expertise is gone it’s so refreshing to see.

The combination of numerical methods and low-level hardware is fascinating. Those 20th century engineers were both smart and creative.