Mapping colorop to pixelshader is necessary because misc_state_template
defines a pixel shader handler. We have to use the same representative
in all pipeline parts.
Wined3d_cs_exec_set_texture dirtifies the pixel shader state when the
fixup of a texture changes. Atifs no longer needs a pixel shader handler
on its own since vertex shader updates aren't delayed any more when the
pixel shader state is dirty.
The abilities of this hardware is too limited to support generic sign /
swizzle fixups. A generic handler would consume 4 of the 8 available
color instruction slots and 2 alpha instruction slots.
The bump mapping handler code has its own way of handling the color
fixups. It merges the fixup into the perturbation calculation without
requiring extra shader instructions. In theory this is possible for the
majority of d3d texture ops as well, but I don't think this is worth the
effort. I expect that this code will only be used for the ddraw signed
format test in practice.
Notice that I'm using floats instead of doubles in the new function,
mostly to be able to use struct wined3d_matrix and multiply_matrix().
At a rough estimate the precision should still be good enough.
Instead of a fixed array of wined3d_shader_signature_element structures.
Shader model 4 shaders can have different semantics in a single
register, e.g. v1.xy TEXCOORD0 and v1.zw TEXCOORD1, so having a single
wined3d_shader_signature_element structure per register isn't
necessarily sufficient.
It's mostly a copy of vertexdeclaration() from state.c, with a few
differences due to the new function being private to the GLSL backend
(e.g. d3d_info.vs_clipping is known to be TRUE) and the fragment fog
update part being split out.
We need GL_ARB_instanced_arrays to be able to take the fastest path and
that's already checked in draw_primitive. Note that
GL_ARB_instanced_arrays provides glDrawElementsInstanced if
GL_ARB_draw_instanced is not supported so we don't need to explicitly
check for the latter.
Alignment should be taken care by the surface pitch and
GL_UNPACK_ROW_LENGTH alone just fine, as long as the pitch is a
multiple of the pixel size. Client APIs respect that,
internally we need to do the same and the changes to
surface_load_texture and wined3d_volume_upload_data are there
for this reason (e.g. WINED3DFMT_R16G16_SNORM is emulated with
GL_RGB16 when NV_texture_shader isn't supported).
The idea is to make it clear that wined3d_texture_load is responsible
for checking the color key for changes. Once we can do color keying in
the shader we also have one place where we can disable the surface
conversion.
This fixes a regression with color keyed mipmaps that was introduced by
ee8a5b7dd1. surface_load for level n+1
would call texture_force_reload, thus removing level n from the GL
texture.
Stefan Dösinger