[PATCH 1/4] wined3d: Properly up-scale WINED3DFMT_R5G5_SNORM_L6_UNORM.
matteo.mystral at gmail.com
Tue Mar 3 05:18:23 CST 2015
2015-03-02 22:29 GMT+01:00 Stefan Dösinger <stefan at codeweavers.com>:
> Simply left-shifting is not enough in positive ranges. 0xff / 255 !=
> 0xff00 / 65535.
> Using the compiler to read 5 bit signed values seems a lot easier than
> operating with masks.
> dlls/wined3d/utils.c | 44 +++++++++++++++++++++++++-------------------
> 1 file changed, 25 insertions(+), 19 deletions(-)
> diff --git a/dlls/wined3d/utils.c b/dlls/wined3d/utils.c
> index 1ce23ac..67457d5 100644
> --- a/dlls/wined3d/utils.c
> +++ b/dlls/wined3d/utils.c
> @@ -290,6 +290,15 @@ static void convert_l4a4_unorm(const BYTE *src, BYTE *dst, UINT src_row_pitch, U
> +#include <pshpack1.h>
> +struct r5g5l6
> + signed r : 5;
> + signed g : 5;
> + unsigned l : 6;
> +#include <poppack.h>
AFAIK C doesn't give any guarantee on a specific allocation of
bitfields in memory. I'm sure this works in practice with GCC and
maybe also with the other compilers we care about, but if it doesn't
get too ugly I'd prefer explicit shifts and masks and be safe.
In any case, this should use signed / unsigned short at least.
More information about the wine-devel