009d73fdf6
renderer_vulkan: Allocate descriptor sets during reinterpretation
2022-12-25 22:19:25 +02:00
42af22f8fd
renderer_vulkan: Enable logic ops and fix swapchain resizing
2022-12-25 22:19:25 +02:00
b85d15b035
renderer_vulkan: Clear stencil with renderpass
...
* Fixes outline retension in pokemon games
2022-12-25 22:19:25 +02:00
130e376c0c
renderer_vulkan: Fix pipeline cache crashes
2022-12-25 22:19:25 +02:00
f884986257
renderer_vulkan: Optimize tiled format convertion + fix vertex buffer alignment
...
* Integrate format convertion to the morton copy function, removing the need for an intermediate copy and convertion pass. This should be beneficial for performance especially since most games use tiled textures
* Also bump vertex buffer size to avoid crashes with hardware shaders and provide correct offset on normal draws which fixes glitches in pokemon Y
* Reduce the local group size to 8 in the D24S8 compute shader which fixes graphical issues in the afformentioned pokemon games at native resolution
* Set LOD to 0 instead of 0.25 to fix another glitch in pokemon y
2022-12-25 22:19:25 +02:00
e23dc3efb1
renderer_opengl: Fix broken texture copy
...
* Resolves graphical bugs in Professor Layton vs Ace Attorney when using OpenGL
2022-12-25 22:19:25 +02:00
dd5e95d7c6
renderer_vulkan: Pipeline cache fixes
...
* Delete cache file if found invalid
* Name it after the vendor/device ids so each physical devices gets a separate cache
2022-12-25 22:19:25 +02:00
b72289eadd
video_core: Fix renderpass cache bug and introduce RGBA -> BGR converter
2022-12-25 22:19:25 +02:00
7a5d4f03da
renderer_opengl: Specify precision in compute shader and add RGB5A1 converter
...
* Fixes OpenGLES crash
2022-12-25 22:19:25 +02:00
18fa277c71
renderer_vulkan: Complete hardware shader support
...
* With these changes all commercial games I tested work fine and get a massive performance boost
2022-12-25 22:19:25 +02:00
73cc764091
renderer_vulkan: Begin hardware shader support
...
* Still experimental and works only with homebrew
2022-12-25 22:19:25 +02:00
9dea514d45
citra: Fix build issues with MinGW and MSVC
2022-12-25 22:19:25 +02:00
0bfaa035b9
renderer_vulkan: Fix warnings and cleanup
2022-12-25 22:19:06 +02:00
85df778785
code: Run clang-format
2022-12-25 22:19:06 +02:00
e54a92c252
code: Address build issues
2022-12-25 22:18:04 +02:00
9145d4cec8
video_core: Re-implement format reinterpretation
...
* Same as before but D24S8 to RGBA8 is switched to a compute shader which should provide better throughput and is much simpler to implement in Vulkan
2022-12-25 22:18:04 +02:00
c611592db6
citra_qt: Add physical device selection dialog
2022-12-25 22:18:04 +02:00
6aba809da8
renderer_opengl: Unbind unused framebuffer targets
...
* Fixes graphical glitches in many games for some reason
2022-12-25 22:12:42 +02:00
f0449d79fd
renderer_opengl: Emulate texture copy with blit for now
2022-12-25 22:12:42 +02:00
33481ada7f
renderer_opengl: Address buffer overflow
2022-12-25 22:12:42 +02:00
67195974e7
video_core: Small code improvements
2022-12-25 22:12:42 +02:00
70c2376fd0
renderer_vulkan: Don't sample from mipmaps when using texture cubes
...
* Mipmaps for texture cubes are unimplemented in the rasterizer cache, so sampling from mipmaps will return nothing
2022-12-25 22:12:42 +02:00
177c7de4f9
input_common: Small fix
2022-12-25 22:12:42 +02:00
eea914ba84
citra_qt: Improve graphics API intergration
...
* Add renderer debug option which toggles debug output in OpenGL/validation layers in Vulkan
* Fix many warnings and replace deprecated Qt functionailty with newer alternatives
2022-12-25 22:12:41 +02:00
62e88fbeb3
rasterizer_cache: Code cleanup
...
* Merge utils and types to a single header
2022-12-25 22:07:46 +02:00
5ce27d8341
texture_decode: Prefer std::memcpy where possible
2022-12-25 22:07:46 +02:00
eaf62eb635
renderer_vulkan: Rework format handling
...
* This is a pretty large commit that aims to solve some issues with the current format system
* The instance now builds at application initialization an array of format traits for each pixel format
that includes information such as blit/attachment/storage support and fallback formats
* The runtime doesn't ask the instance for formats but receives these traits and can dedice on its own what to build
For now we do the same as before, we require both blit and attachment support
* Morton swizzling also sees many bug fixes. The previous code was very hacky and didn't work for partial
texture updates. It was also inconsistent, as it would take a tiled_buffer and write to the middle of linear
* Now the functions have been greatly simplified and adjusted to work better with std::span. This fixes out of bounds
errors and texture glitches (like the display in Mario Kart 7)
2022-12-25 22:07:46 +02:00
9675811bbe
renderer_vulkan: Add experimental Vulkan renderer
2022-12-25 22:07:46 +02:00
945faf8e92
externals: Add vulkan headers and vma
2022-12-25 22:02:59 +02:00
9403049671
rasterizer_cache: Refactor texture cube interface
...
* Reuse our Surface class instead of having a separate one, to avoid reimplementing stuff in the backend
2022-12-25 22:02:59 +02:00
40159d9779
gl_texture_runtime: Clean up texture upload/download code
...
* Improve readability and code clarity
2022-12-25 22:02:59 +02:00
f63653a5b9
rasterizer_cache: Use Common::Rectangle everywhere
...
* Make a nice alias for it and use it instead of having Rect2D/Region2D. Makes the new design less intrusive to the current cache
2022-12-25 22:02:59 +02:00
c71dbb5d19
rasterizer_cache: Make into template
...
* This is the final step, now RasterizerCache is compltely decoupled from OpenGL (technically not yet, but that's talking details). For now texture filtering and some GLES paths have been disabled and will be reimplemented in the following commits
2022-12-25 22:02:59 +02:00
0f4df2c012
rasterizer_cache: Use PBO staging buffer cache for texture uploads/downloads
2022-12-25 22:02:11 +02:00
c6fc4f5a87
rasterizer_cache: Reorder methods
2022-12-25 22:02:11 +02:00
916afa194d
rasterizer_cache: Remove remnants of cached_pages
2022-12-25 22:02:11 +02:00
6f2cd11a85
rasterizer_cache: Fix texture cube blitting
...
* The target was GL_TEXTURE_2D instead of GL_TEXTURE_CUBE_MAP_*
2022-12-25 22:02:11 +02:00
14652d52bc
morton_swizzle: Implement texture formats in UNSWIZZLE_TABLE
...
* I can now remove that loop that has been messing with my OCD
2022-12-25 22:02:11 +02:00
a57ee7cdf2
morton_swizzle: Use tiled_buffer instead of reading data from g_memory
...
* It's much safer and removes hardcoded global state usage
2022-12-25 22:02:11 +02:00
dbd3e6c29b
rasterizer_accelerated: Zero intialize cached_pages
...
* Resolves random crashes because count takes random values
2022-12-25 22:02:11 +02:00
665cbca17c
texture_runtime: Add staging buffer lock mechanism
2022-12-25 22:02:11 +02:00
efb9e9f40f
cached_surface: Remove custom texture logic
...
* Makes things more complicated and is in the way. It's probably already
broken by recent changes, so I'll need to reimplement it anyway
2022-12-25 22:02:11 +02:00
8d35118f63
renderer_opengl: Add driver class to report info/bugs
2022-12-25 22:02:11 +02:00
553c85456e
rasterizer_cache: Add staging buffer cache for uploads/downloads
...
* In addition bump context version to 4.4 to enforce ARB_buffer_storage and use EXT_buffer_storage for GLES which is support on many mobile devices
2022-12-25 22:02:11 +02:00
68ca206d53
rasterizer_cache: Improve TextureRuntime API
...
* This makes every operation more explicit and mimics more the Vulkan API
2022-12-25 22:02:11 +02:00
e30e977140
renderer_opengl: Encapsulate sync objects in OGLSync
2022-12-25 22:02:10 +02:00
f13738d252
morton_swizzle: Optimize and use std::span
2022-12-25 22:02:10 +02:00
04b927ab7f
morton_swizzle: Avoid buffer underflow
...
* Check the y coordinate before decrementing linear_buffer
2022-12-25 22:02:10 +02:00
993d172de9
morton_swizzle: Move out of bounds texture check out of the decode loop
...
* Running relative expensive checks like this on a hot path causes small but measurable performance loss. Tested SMD wit this and it doesn't crash
2022-12-25 22:02:10 +02:00
695447611e
rasterizer_cache: Use SurfaceType instead of Aspect
...
* It was doing pointless enum conversions when both enums described the same thing
2022-12-25 22:02:10 +02:00
