How do I get float to int convertion faster than regular typecasting? I wrote a function with an inline assembler routine, that looks like this: uint32 fistp(float * f) { uint32 * tmp; __asm { mov eax, f fld [eax] fistp [ebx] mov tmp, eax } return *tmp; } but it turned out to be slower than int i = (float)f. It feels like this could be done faster, but I just don''t have enough knowledge of how the FPU works and how MSVC deals with inline asm, so any links on that topic would be great. thanks. //pl

If only I knew ASM. That would be cool for both of us.

Dennis

there is a slight problem with your function:

uint32 *tmp;

does not allocate any memory to store the unsigned integer. secondly fistp does not return unsigned integers.

try this:

int32 __fistp(float f)
{
int32 iresult;

__asm {
fld f
fistp iresult;
}
return (iresult);
}

this still won''t be faster than:

float f;
int i = (int)f;

because of the function call and stack pushes and pops. you can try __inline though which should speed up the process by eliminating the function call.

try your function as

int32 __declspec(naked) __fistp(float f)
{
__asm
{
fld eax;
fistp eax;
ret
}
}

i dont know if it works so, i cant boot vc++ currently i dont have the time, but i try it when i can

we wanna play, not watch the pictures