VideoCommon: Update FastDepth's comment
This commit is contained in:
parent
6798a4763e
commit
3fbe1b1ccd
|
@ -609,11 +609,11 @@ static inline void GeneratePixelShader(T& out, DSTALPHA_MODE dstAlphaMode, API_T
|
|||
if (Pretest == AlphaTest::UNDETERMINED || (Pretest == AlphaTest::FAIL && bpmem.UseLateDepthTest()))
|
||||
WriteAlphaTest<T>(out, uid_data, ApiType, dstAlphaMode, per_pixel_depth);
|
||||
|
||||
|
||||
// TODO: Make more sense out of this comment
|
||||
// D3D9 doesn't support readback of depth in pixel shader, so we always have to calculate it again.
|
||||
// This shouldn't be a performance issue as the written depth is usually still from perspective division
|
||||
// but this isn't true for z-textures, so there will be depth issues between enabled and disabled z-textures fragments
|
||||
// FastDepth means to trust the depth generated in perspective division.
|
||||
// It should be correct, but it seems not to be as accurate as required. TODO: Find out why
|
||||
// So for disabled FastDepth, just calucate the depth value again.
|
||||
// This performance lack of this division doesn't matter, but it's a major performance issue
|
||||
// because of its forced off earlyZ.
|
||||
if (g_ActiveConfig.bFastDepthCalc)
|
||||
out.Write("float zCoord = rawpos.z;\n");
|
||||
else
|
||||
|
|
Loading…
Reference in New Issue