video_core: Various small improvements and bug fixes

[raphaelthegreat]

• Adds another case of missing shared memory barriers which fixes more LDS artifacts
• Improves texture cache overlap resolution to allow copying individual slices in a multi-layer mip-mapped image (credit to poly for testing and brainstorming)
• Fixes a crash when CE task finishes execution earlier than GFX task.
• Fixes crashes when emitting Quad/Rect tess shaders with default attributes

[Missake212]

This breaks Bloodborne

[squidbus]

Also breaks CUSA28193 loading into open world areas with this assert:

shadPS4/src/video_core/texture_cache/image_info.cpp

Line 253 in 207facd

ASSERT(mip != 0);

[raphaelthegreat]

Also breaks CUSA28193 loading into open world areas with this assert:

shadPS4/src/video_core/texture_cache/image_info.cpp

Line 253 in 207facd

ASSERT(mip != 0);

This assert should probably be removed, as the function should be allowed to return mip 0. I can imagine this happening in cases where individual slice of first mip is accessed, so it takes right overlap path

[squidbus]

This assert should probably be removed, as the function should be allowed to return mip 0. I can imagine this happening in cases where individual slice of first mip is accessed, so it takes right overlap path

Yes seems fine when removed.

[squidbus]

This breaks Bloodborne

For this, I can see the same sort of effect as well in CUSA16404 on characters.

[raphaelthegreat]

This probably has something to do with the new overlap handling commit

[raphaelthegreat]

@Missake212 The bug should be fixed now

[squidbus]

No longer seeing any of reported issues so far.