This doesn’t make any sense, since the IDT WinChip is a single pipeline, in-order execution microarchitecture, so this new code shouldn’t be faster on this CPU. My guess is that SHLD/SHRD instructions are really slow on the WinChip 🤔

Holy bananas! I’ve found the CPU that gets the biggest FPS boost from the new code. I didn’t expect this at all! The IDT WinChip 200 is the winner, with nearly a 35% performance increase: from 110 FPS to 150 FPS in FDOOMVBR, and from 33 FPS to 49 FPS in FDM400R

Some new results comparing the Pentium 166 and the Pentium 166 MMX (thanks, Néstor!). I think this clearly shows that at higher frequencies, the original P54CS hits a barrier that was fixed in the P55C.

This new Pentium-optimized code runs faster on the slower variants of the Pentium. I’m not sure why, but on faster models the difference is smaller. In fact, on the Pentium 200 the new code is actually slower. Maybe I’m hitting some kind of architectural limitation?

Tested the AMD K5 at 100 MHz. As expected, it doesn’t like this code (640x400):

- Old code: 26.453 fps
- New code: 25.563 fps (-3.3%)