RE: [GD-Windows] forcing value to bool 'true' or 'false' (performance warning)

[Brian S. <bs...@mi...>]

The compiler generated version doesn't cause a branch (this is VC 7):

752:          int a =3D 1000;
0003A370   mov         dword ptr [a],3E8h
753:          bool b =3D a;
0003A377   cmp         dword ptr [a],0
0003A37B   setne       al
0003A37E   mov         byte ptr [b],al

But it is using a register so that doesn't do the optimizer any favors.

--brian

> -----Original Message-----
> From: Jon Watte [mailto:hp...@mi...]=20
> Sent: Monday, April 22, 2002 6:16 PM
> To: Rich; gam...@li...
> Subject: RE: [GD-Windows] forcing value to bool 'true' or=20
> 'false' (performance warning)=20
>=20
>=20
>=20
> It's more than that. The problem is that any bytes in the int must=20
> count towards a "true" value. The re-write you propose will generate=20
> a branch, which will not predict well, which is a REALLY bad thing.
>=20
> Perhaps MSVC7 generates CMOV for this; I've never gotten MSVC 6 to=20
> do it unfortunately :-( Another branchless alternative (assuming=20
> they're OK with any non-0 value for "true" inside a bool, which may=20
> not be true) is to do some shifts and ors.
>=20
>   int i; bool b;
>=20
>   unsigned short temp16 =3D (i>>16) | i;
>   *(unsigned char *)&b =3D (temp16>>8) | temp16; // horrible!
>=20
> Cheers,
>=20
> 				/ h+
>=20
> > Well, the compiler will generate code to convert an int (4=20
> bytes) to a
> > bool (usually 1 byte), whether or not it matters for performance is
> > dubious.
> >=20
> > You can eliminate the warning by doing 'bool p =3D i ? true : =
false;'
>=20
>=20
>=20
> _______________________________________________
> Gamedevlists-windows mailing list
> Gam...@li...
> https://lists.sourceforge.net/lists/listinfo/gamedevlists-windows
> Archives:
> http://sourceforge.net/mailarchive/forum.php?forum_id=3D555
>=20