Add vulkan+rtx renderer #9
Draft
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Improvement geometry smoothing (normals)
Update patches
- Do not replace inherited base color texture - Do not acquire default base color texture Fixes #638
Detect changes in alternate_anims sequences. Fixes #640
Fixes a typo that rewrote roughness value with garbage. Also adds a few more debug channel displays for lighting phases. And prints out available debug displays. Fixes #641
This makes it easily switchable at any point in time. Still not sure how to properly manage log verbosity cvars: - cvars are loaded after initialization and map load, so we can't really depend on saved cvar values. - reloading cvars each frame cancels `-vkverboselogs` arg that is supposed to work around the above limitation
Adds `_xvk_tex_rotate` field for surface patches. Angle is specified in degrees. Note that it rotates the texture around origin, which might be very far away. So offset patching might be needed to re-align.
We were using a negative value for `SURF_SKY` surfaces, which hit an assert on uploading kusochki.
Pass current entity to compute wave height. Still doesn't switch to negative height when underwater for some reason. And still displays water sides. Add some notes regarding water geometry generation and drawing.
Things done during stream E326 - [x] list supported arguments for `rt_debug_display_only` - [x] make `vk_debug_log` a command - [x] rotate your owl, fix #625 - [x] fix skybox surface assert - [x] half-fix water levels - [x] add alternate anims to infotool
Minor changes (replacement by _xvk_tex_rotate)
Update patches
Add a section for this specific project
Many static worldmodel surfaces still contain alternate texture chains. That was making these surfaces dynamic, even though there's no way to trigger these alternate anims. Make worldmodel ignore alternate_anims when looking for animated surfaces. Fixes #644
Not sure that we need it.
Worldmodel draws all water surfaces regardless of their orientation/PLANE_Z. All other entities only draw sides when EF_WATERSIDES effect bit is set.
Stuff done during stream E327 - [x] Fix alternate worldmodel anims #644 - [x] Hide water sides appropriately
Adds a few things: - verbose logs about surface flags/type/orientation - same about surface subdivided polys - tries culling back water surfaces (doesn't work that well) - tries culling in rt shaders This still unfinished
Cull everything except: - opaque geometry: to avoid shadow and reflection holes - bleded geometry: particles and such should be visible from any angle Alpha-tested geometries result in visual glitches when not culled (double-sided ladders, double shadows, etc). These glitches are worse than slightly misaligned shadows. Translucent geometries are still matter of research.
Only leave water surfaces directed towards "air". This is consistent with: - non-`kRenderNormal` surfaces, for which there's only one surface present - opt2 for translucent surfaces, where we aim to detect medium boundary direction based on normal-vs-ray-direction Also add notes about translucent surfaces in general, more info about water rendering under ref_gl, etc. Fixes #264
E328-E330: Improve backface culling - [x] Turn on backface culling in RT shaders for everything. - [x] Force-disable culling for everything except alpha blended geometries. - [x] Only leave water surfaces that are directed towards "air", fixes #264
Makes acid water surfaces emissive again. However, they are not assinged proper emissive color values for direct ray hits yet, so they do emit light, but look dar still. Ref: #56
Now indirect diffuse channel gets blurred with 2 Á-Trous passes. It mostly follows the paper, except for Σd_i, which gives very bad viusal artifacts.
Under `-vkvalidate` enables `debugPrintfEXT()` in shaders, allowing extra validation and error reporting from shaders. Severely degrades performance, so should be enabled only for debugging.
Hand-made organic bespoke GMO-free BRDFs - [x] Fix black metals in #666 - [x] Better diffuse-vs-specular channels tracking - [x] specular reflections - [x] Fix new `test_material` glitches - [x] Fix #461 - [x] Fix #151 - [x] Fix #266 - [x] Filter out zero-area degenerate triangles (NOTE: might affect normal smoothing) - [x] Implement #126 - [x] sun solid angle parametrization - [x] Make new brdfs sampling functions - [ ] distant light circular glitches - [x] not enough float precision, addressed to a degree with slight modifications to equations - [x] Address glow over weapons in RT, see #442 - [x] bounces - [x] diffuse - [x] why is it so dark? - [x] specular - [x] Peformance differences. Why is there no CPU-GPU parallelism anymore? - [x] `VK_VALIDATION_FEATURE_ENABLE_DEBUG_PRINTF_EXT` forces cpu-gpu sync. 💩 - [x] Make a special flag for it - [x] Update rendertests - [x] add more channels - [x] add white furnace display test
Once upon a time, there were some BLASes. There were the BLASes for dynamic geometry, that were to be used every frame to draw very dynamic things, like sprites and beams. They were initialized at the very begining of the renderer's lifetime, and were expected to live happily for the entire process duration. However, an evil NewMap appeared. It didn't care abount anything or anyone, so when it came, it just cleared all the allocators, and allowed other BLASes to be allocated into the same space as dynamic BLASes were already given. This made BLASes live on top of each other, use each others toys and make them fight. They weren't happy. So we just kill them and creat them again from scratch, when the evil NewMap comes. The end. Fixes #729
vk: rt: add workaround for holes in geometry Once upon a time, there were some BLASes. There were the BLASes for dynamic geometry, that were to be used every frame to draw very dynamic things, like sprites and beams. They were initialized at the very begining of the renderer's lifetime, and were expected to live happily for the entire process duration. However, an evil NewMap appeared. It didn't care abount anything or anyone, so when it came, it just cleared all the allocators, and allowed other BLASes to be allocated into the same space as dynamic BLASes were already given. This made BLASes live on top of each other, use each others toys and make them fight. They weren't happy. So we just kill them and creat them again from scratch, when the evil NewMap comes. The end. Fixes #729
It is not wired up to render anything yet, but we just made sure that it builds and gets destroyed. Also, fixup dynamic BLAS double-free.
It does it in a super dirty way, trad renderer needs a massive refactoring. It also displays patched skybox, however, we need it to display the oridingal one.
Also fix: - a bunch of texture/image destruction issues - `_xvk_remove_all_sky_surfaces` getting stuck
E373: Draw skybox in traditional renderer Fixes #732
- `mem->priv_.devmem` -> `slot_index`; - Move static `VK_DevMemUsageTypeString()` out of `devmem.h` header.
profiler: track memory usage (#502) * [x] vulkan device memory: * [x] allocated * [x] used as storage for buffers/images/etc * [x] unusable alignment holes * [x] vulkan buffers * [x] allocated * [x] used as storage * [x] unusable alignment holes * [x] images memory total * [ ] cpu memory allocated in ref/vk (copied & pasted from Issue #502)
Fixes sync hazard validation error for depth buffer.
`rt_force_disable` is perfectly fine, it just needs to be written into either `vk.cfg` or `video.cfg` to work. Other configs are read *after* `R_VkInit()` function get called.
This is a very dirty way of silencing errors, trading them for a waterfall of warnings. But it will allow making progress for now.
Trad rendering still works with this slightly more tight sync model. It is suboptimal as it doesn't really know the previous op on the buffer, so it has to do the ALL_COMMANDS stages, which makes validation a bit sore.
Not sure what's happening with the rest, but first couple works.
This makes RT rendering work w/o sync hazards. But it will still fail with one if menu is called. It's as much as it can go w/o proper state tracking.
Synchronization stuff started on stream E374 Addresses a bunch of validation synchronization hazards. Not all of them. And the remaining ones require more involved state tracking, a big effort. Also contains a few minor tweaks: - stop using SDL_GL functions, silence errors - increase acceleration structures buffer size -- with recent mesa update BLASes got bigger somewhat (?)
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This is here for now only for tracking mergeability + CI purposes