x86: Emulation of FDIV bug and incorrect result in 32 bit multiply operation

[brycecherry75]

Some may know that certain Intel 80386 CPUs have a marginal circuit which can result in an incorrect result from a 32 bit multiply operation along with certain software having such a check (this issue does not affect such CPUs with a double sigma marking on the CPU package).

Also, some may know that certain early steppings (B1 and B3 for 75-100 MHz and earlier than D for 60/66 MHz) of the Intel Pentium CPU have the FDIV bug (concerning divide, remaindering, tangent and arctangent floating point instructions) which can result in significant precision error along with certain software being dependent on this bug.

Both of these problems are worth emulating and/or trapping with an option switch (the FDIV bug has been emulated in DOSBox).

[angelosa]

You mean this stuff, but I'm curious to know what depends on any of this.

[cuavas]

Yeah, I know of software (e.g. Excel) that detects and applies a workaround for the FDIV bug, but I don’t know of any software that depends on its presence for correct operation.