IDA declare a function that passes return value on FPU stack [duplicate]

Question

This question already has an answer here:

• Which calling convention to use for EAX/EDX in IDA 3 answers

The usual way to return a value is by passing it in EAX. I know also that a 64 Bit value would be returned in EAX:EDX. However I have now a function that calcluates the localtime to miliseconds into a 64Bit double value, and the function returns it in the FPU stack, so the caller, upon return, pops it off from there.

IDA identified this function as void and when I try to declare it as double it says that it is a "bad declaraton". Even if IDA would accept it, this would mean that the value would have been passed back using EAX:EDX right?

According to Wikipedia calling conventions, float/doubles are returned using the x87 registers as a standard. So how do I tell IDA this, resp. why doesn't it accept double as the return value?

Update

I found another function which returns a double, and this one is correctly using the double as the return value, when I press Y to change the function. So whats wrong with this one?

void __userpurge GetTimeInMilliseconds(unsigned __int16 nHour<ax>, unsigned __int16 nMinute<dx>, unsigned __int16 nSeconds<cx>, unsigned __int32 nMilliseconds)

When I try to change the void to double, then it failes. I assume that I may have to specify the returnvalue manually, but how do I do this for a FPU register?

Something like this, also doesn't work, so probalby there is some different name for the FPU?

void __userpurge GetTimeInMilliseconds<st>(unsigned __int16 nHour<ax>, unsigned __int16 nMinute<dx>, unsigned __int16 nSeconds<cx>, unsigned __int32 nMilliseconds)