Currently the "Float" and "Double" DllCall return types don't work, as they use the
xmm0 register and we only (implicitly) retrieve
rax. I spent hours trying to figure out how to retrieve the value of xmm0 using C++ intrinsics or let the assembly function determine which to return, and finally stumbled upon this article:
C ABI Hacks. It presents some excellent ideas, one of which I was able to adapt to fix floating-point return values:
Code:
; in x64call.asm
read_xmm0_float proc
option prologue:none, epilogue:none
ret
read_xmm0_float endp
// in script2.cpp
extern "C" float read_xmm0_float();
...
case DLL_ARG_FLOAT:
aResultToken.symbol = SYM_FLOAT;
aResultToken.value_double = Exp32or64(return_value.Float, read_xmm0_float());
break;
It relies on there being no use of the xmm registers (presumably no floating-point operations) between calling the function and retrieving the register. Similar code is needed for "Double" return values. The article used C functions (no assembly required
), but that meant putting the functions in a separate compilation unit and turning off link-time optimization.
I suppose the technique is comparable to this one I've used many times in AutoHotkey32:
Code:
str := DllCall("MulDiv", "int", ptr, "int", 1, "int", 1, "str")
Login
Register
FAQ
Search
