* presenter: render the game inside a ImGui window
* presenter: render the previous frame to keep the render rendering
* swapchain: fix swapchain image view format not being converted to unorm
* devtools: fix frame graph timing
* shader_recompiler: Account for instruction array flag in image type.
* shader_recompiler: Check da flag for all mimg instructions.
* shader_recompiler: Convert cube images into 2D arrays.
* shader_recompiler: Move image resource functions into sharp type.
* shader_recompiler: Use native AMD cube instructions when possible.
* specialization: Fix buffer storage mistake.
* video_core: Implement conversion for uncommon/unsupported number formats.
* shader_recompiler: Reinterpret image sample output as well.
* liverpool_to_vk: Remove mappings for remapped number formats.
These were poorly supported by drivers anyway.
* resource_tracking_pass: Fix image write swizzle mistake.
* amdgpu: Add missing specialization and move format mapping data to types
* reinterpret: Fix U/SToF input type.
* texture_cache: Stricter barriers on image upload
* buffer_cache: Stricter barrier for vkCmdUpdateBuffer
* vk_rasterizer: Barrier also normal buffers and make it apply to all stages
* texture_cache: Minor barrier cleanup
* Batch image and buffer barriers in a single command
* clang format
* Speed up LiverpoolToVK::SurfaceFormat
In Bloodborne this shows up as the function with the very highest cumulative "exclusive time". This is true both in scenes that perform poorly, and scenes that perform well.
I took (approximately) 10s samples using an 8khz sampling profiler.
In the Nightmare Grand Cathedral (looking towards the stairs, at the rest of the level):
- Reduced total time from 757.34ms to 82.61ms (out of ~10000ms).
- Reduced average frame times by 2ms (though according to the graph, the gap may be as big as 9ms every N frames).
In the Hunter's Dream (in the spawn position):
- Reduced the total time from 486.50ms to 53.83ms (out of ~10000ms).
- Average frame times appear to be roughly the same.
These are profiles of the change vs the version currently in the main branch. These improvements also improve things in the `threading` branch. They might improve them even more in that branch, but I didn't bother keeping track of my measurements as well in that branch. I believe this change will still be useful even when that branch is stabilized and merged.
It could be there are other bottlenecks in rendering on this branch that are preventing this code from being the critical path in places like the Hunter's Dream, where performance isn't currently as constrained. That might explain why the reduction in call times isn't resulting in a higher frame rate.
* Implement SurfaceFormat with derived lookup table instead of switch
* Clang format fixes
* shader_recompiler: Add swizzle support for unsupported formats.
* renderer_vulkan: Rework MRT swizzles and add unsupported format swizzle support.
* shader_recompiler: Clean up swizzle handling and handle ImageRead storage swizzle.
* shader_recompiler: Fix type errors
* liverpool_to_vk: Remove redundant clear color swizzles.
* shader_recompiler: Reduce CompositeConstruct to constants where possible.
* shader_recompiler: Fix ImageRead/Write and StoreBufferFormatF32 types.
* amdgpu: Add a few more unsupported format remaps.
* devtools: fix popen in non-windows environment
* devtools: fix frame crash assertion when hidden
* devtools: add search to shader list
* devtools: add copy name to shader list
* devtools: frame dump: search by shader name
* Clear RT if FCE was invoked before any draws
Co-authored-by: psucien <bad_cast@protonmail.com>
* address review comments
---------
Co-authored-by: psucien <bad_cast@protonmail.com>
* ir: Add heuristic based LDS barrier pass
* Attempts to insert barriers after zero-depth divergant conditional blocks in shaders that use shared memory
* lds_barriers: Limit to nvidia
* Intel has historically had problems with cs barriers, will debug other time
