I'm sure you know, but for others reading: even on the same architecture, there is more to floating point determinism than just running the same "x = a + b" code on each system. There's also the state of the FPU (eg: rounding modes) that can affect results.
On older versions of DirectX (maybe even in some modern Windows APIs?) there were cases where it would internally change the FPU mode, causing chaos for callers trying to use floats deterministically[1].
Comments
I'm sure you know, but for others reading: even on the same architecture, there is more to floating point determinism than just running the same "x = a + b" code on each system. There's also the state of the FPU (eg: rounding modes) that can affect results.
On older versions of DirectX (maybe even in some modern Windows APIs?) there were cases where it would internally change the FPU mode, causing chaos for callers trying to use floats deterministically[1].
[1] https://gafferongames.com/post/floating_point_determinism/ (see the Elijah quote, especially)