Use less FBOs for chained post-processing shaders #12921
Conversation
|
Cool. I'll help out with the Vulkan issue later, but maybe not tonight. |
|
With Vulkan pause menu is totally black as well for me now so there is some bigger issues I think behind. I could create a in game dump with renderdoc on my machine (the first shader is grayscale, the second should use it for the left part of the screen and the original for the right one). There is some code that look weird to me tho', I will comment on the code it self about that (might be irrelevant). |
| VkBuffer buffer_; | ||
|
|
||
| bool operator < (const DescriptorSetKey &other) const { | ||
| if (imageView_ < other.imageView_) return true; else if (imageView_ > other.imageView_) return false; | ||
| if (sampler_ < other.sampler_) return true; else if (sampler_ > other.sampler_) return false; | ||
| for (int i = 0; i < MAX_BOUND_TEXTURES; ++i) { |
There was a problem hiding this comment.
This loop seem weird to me, it have return in it so unless they are the same it will end asap, is this fine is this order? I'm not sure what is this used for tho'.
There was a problem hiding this comment.
Right, this is just a compare to say - is this key the same? It's so it can build the map. This should be fine.
-[Unknown]
Before, we created things at the wrong size, then recreated at the right. Now we just start at the correct size once.
If you apply 4 in a row at the same resolution, we only need two actual FBOs. This will use less VRAM.
So that we can initialize and bind samplers.
Not working properly though...
unknownbrackets
force-pushed
the
postshader
branch
from
9bfd2ab
to
804b673
Compare
…ationCommon::CopyToOutput. Validation caught an issue where old stuff lingered in sampler 1 and texture 1. Bug probably introduced in #12921, but could also be others.
…ationCommon::CopyToOutput. Validation caught an issue where old stuff lingered in sampler 1 and texture 1. Bug probably introduced in #12921, but could also be others.
A few changes (for v1.11.x):
This adds sampler1 to read the original source, which isn't tested thoroughly but seems to work in GL/D3D9/D3D11.
In Vulkan, I tried fixing it up, but if I assign the second sampler in the descriptor set update - the shader somehow doesn't apply (but the image still presents?) With RenderDoc attached it simply crashes, and there's no validation layer feedback. Arg.See anything?-[Unknown]