[d3d8] Various D3D8 Wine device test fixes and other nits #4732
Conversation
WinterSnowfall
force-pushed
the
d3d8-testsfix
branch
5 times, most recently
from
96927bf to
785c3ae
Compare
WinterSnowfall
changed the title
WinterSnowfall
force-pushed
the
d3d8-testsfix
branch
8 times, most recently
from
b27d899 to
9979144
Compare
WinterSnowfall
changed the title
WinterSnowfall
force-pushed
the
d3d8-testsfix
branch
6 times, most recently
from
b06e20f to
0a07fab
Compare
|
Seems fine, didn't run into any regressions with ~50 d3d8/9 games. |
There was a problem hiding this comment.
Couple comments / questions.
I think for the texture validations specifically, depending on how well understood these are and how important fixing them is, we might want to hold off until after 2.6, but I'd also like a second pair of eyes here (@K0bin).
| // CreateImageSurface and D3D8 cube textures outside of D3DPOOL_DEFAULT | ||
| if ((m_mipLevel == 0 && isBlockAlignedFormat && desc.Pool == D3DPOOL_DEFAULT) |
There was a problem hiding this comment.
does this change anything for BC4/5 formats that the d3d9 runtime doesn't know about?
There was a problem hiding this comment.
Yes, ATI1/2 will get validated as well. The failing Wine test checked all DXT formats, ATI1/2, YUV2 and UYVY.
| @@ -119,9 +119,9 @@ namespace dxvk { | |||
| || static_cast<LONG>(pBox->Bottom) - static_cast<LONG>(pBox->Top) <= 0 | |||
| || static_cast<LONG>(pBox->Back) - static_cast<LONG>(pBox->Front) <= 0 | |||
| // Exceeding surface dimensions | |||
| || pBox->Right > desc.Width | |||
| || pBox->Bottom > desc.Height | |||
| || pBox->Back > desc.Depth) | |||
There was a problem hiding this comment.
Can 3D be block-compressed in D3D9?
There was a problem hiding this comment.
Not entirely sure if there's a validation against it, but I don't know of any game that does that.
| @@ -161,8 +161,8 @@ namespace dxvk { | |||
| || pRect->right - pRect->left <= 0 | |||
| || pRect->bottom - pRect->top <= 0 | |||
| // Exceeding surface dimensions | |||
| || static_cast<UINT>(pRect->right) > desc.Width | |||
| || static_cast<UINT>(pRect->bottom) > desc.Height) | |||
There was a problem hiding this comment.
similar question here, does this affect any potential edge case with compressed formats? I know they are somewhat restricted in D3D9 in general, but is it e.g. possible to have a 12x8 BC1 texture with 2 mips where mip 1 would be 6x4 (so at least one block in each dimension, but not block-aligned), and if so, is this tightly validated or would the runtime let you lock an 8x4 region?
There was a problem hiding this comment.
Wine tests seem to suggest this is tightly validated and mips > 0 don't have to be block aligned. And the tests in question are particularly targeted at compressed formats.
WinterSnowfall
force-pushed
the
d3d8-testsfix
branch
from
0a07fab to
61237b7
Compare
WinterSnowfall
force-pushed
the
d3d8-testsfix
branch
from
5ab1903 to
25d9a51
Compare
WinterSnowfall
force-pushed
the
d3d8-testsfix
branch
2 times, most recently
from
2e35cdc to
61237b7
Compare
WinterSnowfall
force-pushed
the
d3d8-testsfix
branch
from
61237b7 to
bbb5aa3
Compare
This adds validation capabilities to our d3d8->d3d9 shader translation logic and fixes a FVF normals component Wine test specific to d3d8. The test in question seems to suggest the validation is needed by 3DMark2001, but at least 3DMark2001 SE does not appear to hit it, so perhaps it was ultimately fixed? Regardless, the validation is in line with native behavior.
While I was at it, I also changed the shader creation logic a bit to not needlessly store partially created vertex shader objects if something should fail along the way.
Might look into a few more of the remaining tests, after which I'll test a bit and undraft.