Whoops, sent this to just Stas the first time ... sorry, Stas! On Mon, Sep 14, 2009 at 11:26 AM, Stas Sergeev <stsp@xxxxxxxx> wrote: > Hello. > > Bart Oldeman wrote: >> The fpu save/restore operations aren't cheap, so perhaps one could >> only use them around the sound code instead of for every vm86 call. >> Though that could be messy. > > Indeed - so what was the problem with fninit/ldmxcsr then? > Looks like it would load only the control/status word, so > that might not be as expensive as the full reload? > Also, I think fxrstor doesn't wait for completion, so > the careful implementation may not be too expensive. > There are also some tricks possible, like, for example, set > FPU to something that will SIGFPE on any FP, and initialize > it properly in a signal handler, but I guess this would > be an overkill. Well, I believe that's essentially what Intel suggests OSes do on task-switch -- leave some flag set/cleared that will cause the next FP operation to give an exception so that it can switch the FP registers at that time. Whether or not this would be a good idea for DOSEMU depends on how often you do a vm86 call vs how often the FP registers are needed, and on how expensive it is to actually enter a state from which the next FP access will cause a signal compared to just doing the reload each time. -- To unsubscribe from this list: send the line "unsubscribe linux-msdos" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
