vs_2_0's sgn has two extra parameters - temporary registers for the
instruction. This patch fixes the crash of the d3d9:visual test on windows and
properly handles the vs_2_0 version of sgn.
GL_TEXTURE_BASE_LEVEL matches the basetexture::SetLOD functionality.
D3DSAMP_MAXMIPLEVEL essentially does the same as SetLOD. The test included in
this patch shows that the smallest mipmap level is used.
SCS is unfortunately a fragment program only instruction. If we have the NV
extensions we can use SIN and COS. Otherwise we have to approximate sine and
cosine with a taylor series. Luckily we're provided with the necessary
constants by the application.
The fog settings do not depend on wether the shader writes to oFog or not,
instead they depend on the FOGVERTEXMODE and FOGTABLEMODE settings, and if a
vertex shader is bound at all.
It works the same way as with the fixed function, and having a vertex shader
is the same as using pretransformed vertices, just that the fog coord comes
from the shader instead of the specular color:
FOGTABLEMODE != NONE: The Z coord is used, oFog is ignored
FOGTABLEMODE == NONE, with VS: oFog is used
FOGTABLEMODE == NONE, no VS, XYZ: Z is used
FOGTABLEMODE == NONE, no VS, XYZRHW: diffuse color is used
This allows us to drop the load time conversion and the clear
readback hack and replaces it with a color fixup in the fixed
function pipeline replacement.
Add a test that checks what happens if D3DRS_POINTSIZE, D3DRS_POINTSIZE_MIN
and POINTSIZE_MAX have conflicting values. D3DRS_POINTSIZE_MAX trumps
D3DRS_POINTSIZE_MIN, and both MIN and MAX clamp the D3DRS_POINTSIZE value if it is
outside of their range.
caps.MaxTextureWidth / caps.MaxTextureHeight can potentially be very
large, which could cause a driver to run out of memory. Arguably the
driver shouldn't report such large limits, but there's no reason for
us to create a texture larger than 1024x1024 either.
