Fix 32 bit rendering about open1560

[0x1F9F1]

Using -nomip helps to avoid a bug (unsure if caused by MM1 or dgVoodoo) where the pitch of a surface is half of what it should be (width * (bpp+7)/8)

[0x1F9F1]

agiD3DTexDef::BeginGfx calculates this->SurfaceSize = 2 * this->SurfaceDesc.dwWidth * this->SurfaceDesc.dwHeight though it is not clear if this field is used anywhere

[0x1F9F1]

agiDDPipeline::CopyBitmap expects a 16-bit frame buffer when agiCurState.State.PolyMode == 3

[0x1F9F1]

When some of the listed issues are hackily fixed, you are able to get in game, though there are still some graphical issues, mostly being broken reflections and traffic lights:

[0x1F9F1]

mmCullCity::UpdateSnowTextures supports 8-bit and 16-bit snow textures, though this is implemented as an if/else and as such defaults to treating the texture as 8-bit.

[0x1F9F1]

agiSurfaceDesc::Reload only supports annotating 8-bit and 16-bit textures, though this happens before textures are converted to 32-bit so it's not much of a problem

[0x1F9F1]

Using -depth -nomip and setting bAdditiveBlending=0 in video.cfg fixes the graphical issues:

[0x1F9F1]

Commit 910dad3 fixes -nomip being required with dgVoodoo by clearing the DDSD_PITCH flag before passing the DDSURFACEDESC2 to IDirectDraw4::CreateSurface. Because the DDSURFACEDESC2 was originally from a 16-bit image, and only the ddpfPixelFormat was overwritten, the pitch was incorrect. It seems native DDRAW doesn't trust this value, but dgVodooo does.

[0x1F9F1]

dxiScreenShot only supports 555 and 565 surfaces

[0x1F9F1]

Using bAdditiveBlending=1 and premultiplying the alpha in copyrow4444_to_8888 also (mostly) fixes the transparency issues.

[0x1F9F1]

Using premultipled alpha with copyrow4444_to_8888 causes certain textures to be darker/less visible. Most noticeably: Smoke, Rain and Snow (maybe all particles?).

[0x1F9F1]

With premultiplied alpha the reflections also seem to be over saturated (NOTE: This is only true for RV3, so it may just be a problem with those textures).

[0x1F9F1]

It looks like premultipled alpha should only be used with textures that have the "g" flag (in their tex sheet), though since that information isn't stored in the agiSurfaceDesc, it will likely require a rewrite of agiD3DTexDef::BeginGfx

[0x1F9F1]

467188a fixes texture conversion by passing agiTexParameters to agiSurfaceDesc::CopyFrom, allowing a check whether the surface needs to have the alpha premultiplied

[0x1F9F1]

All 32-bit texture issues should be fixed at this point.