Add workarounds for Samsung Xclipse GPUs:
- Disable extendedDynamicState3ColorBlendEquation as it is broken in Samsung
drivers, similar to AMD drivers
- Add Samsung's proprietary driver to the validated driver list for clock
boosting
- Fix log message to indicate both AMD and Samsung drivers have broken
color blend equation support
Remove stray logical OR operator from validated_driver condition.
Refactors and improves the texture format conversion system in the Vulkan
renderer:
- Adds proper sRGB to linear conversion for depth formats
- Improves shader accuracy for ABGR8 SRGB to D24S8 conversion
- Adds gamma correction and proper depth range clamping
- Moves GetSupportedFormat implementation to header
- Cleans up format conversion switch statement
- Removes redundant format conversion paths
The changes improve accuracy when converting between color and depth
formats, particularly for sRGB sources. The shader improvements ensure
proper gamma correction and depth range handling.
Technical changes:
- Improves sRGB to linear conversion in fragment shader
- Adds proper depth value clamping
- Consolidates format conversion logic
- Removes duplicate GetSupportedFormat implementation
Implements hardware-accelerated MSAA resolve functionality in the Vulkan
texture cache instead of relying on compute shaders. This change:
- Adds proper MSAA to non-MSAA image copy support using VkResolveImage
- Creates temporary resolve images with appropriate memory allocation
- Handles format compatibility checks with proper fallback to compute
- Manages image layout transitions and memory barriers
- Preserves existing compute shader fallback for unsupported formats
The implementation follows Vulkan best practices for MSAA resolve
operations and should provide better performance for supported formats.
Adds special handling for memory allocation size on AMD and Qualcomm (Adreno)
drivers by aligning allocations to 4KB boundaries. This fixes potential memory
allocation issues on these drivers where unaligned allocations may fail or
cause undefined behavior.
Affected drivers:
- AMD Proprietary (AMDVLK)
- AMD Open Source (RADV)
- Qualcomm Proprietary (Adreno)
Adds several new shader-based format conversion pipelines to support additional
texture formats and operations:
- RGBA8 to BGRA8 conversion
- YUV420/RGB conversions
- BC7 to RGBA8 decompression
- ASTC HDR to RGBA16F decompression
- RGBA16F to RGBA8 conversion
- Temporal dithering
- Dynamic resolution scaling
Updates the texture cache runtime to handle these new conversion paths and adds
helper functions to check format compatibility for dithering and scaling
operations.
The changes include:
- New shader files and CMake entries
- Additional conversion pipeline setup in BlitImageHelper
- Extended format conversion logic in TextureCacheRuntime
- New format compatibility check helpers
Implements conversion from sRGB color formats to D24S8 depth-stencil format
in the Vulkan renderer. This change includes:
- New fragment shader convert_abgr8_srgb_to_d24s8.frag that handles proper
sRGB to linear conversion before depth calculation
- Added shader to CMake build system
- Extended BlitImageHelper with new conversion pipeline and methods
- Updated texture cache to handle sRGB to D24S8 format conversion paths
The conversion properly handles sRGB color space by first converting to
linear space before calculating luminance values for the depth component,
while preserving alpha channel data for the stencil component.
Adjusts VRAM allocation strategy to be more conservative while maintaining
performance:
- Increases reserve memory from 1/8th to 1/4th (max 2GB) for discrete GPUs
- Increases base memory limit from 6GB to 8GB
- Doubles resolution scaling memory from 1GB to 2GB per scale factor
- Reduces system memory reservation from 8GB to 4GB for integrated GPUs
- Increases maximum memory limit from 4GB to 6GB for integrated GPUs
These changes help prevent memory leaks while still providing adequate
VRAM for optimal performance.
Adds missing include for common/logging/log.h in gpu.h which was causing
compilation failures on Android. This header is needed for logging
functionality used in GPU-related operations.
The include was previously indirectly available through other headers,
but making it explicit improves code clarity and prevents potential
future compilation issues.
Reverts overly restrictive storage buffer validation and size calculation
that was causing rendering issues in The Legend of Zelda: Tears of the
Kingdom, particularly in underground/depth areas. The simplified approach:
- Uses GetMemoryLayoutSize() instead of manual page probing
- Removes unnecessary 4GB memory bounds validation
- Streamlines address translation and alignment handling
This fixes numerous reported cases of missing or corrupted rendering in
TOTK's underground areas where storage buffer operations are heavily used
for depth-related effects.
Relaxes validation checks in the NVDRV GPU service and improves error notifier
handling to prevent potential hangs. Key changes:
- Remove strict size validation in SetErrorNotifier
- Relax GPFIFO entry count validation to only check for non-zero values
- Add proper error notifier state tracking in GPU class
- Improve debug logging messages
The previous strict validation was causing issues with some games like ACNH.
These changes maintain necessary checks while being more permissive with
edge cases that don't impact functionality.
Technical changes:
- Store error notifier state in GPU class for future implementation
- Remove upper bound check on GPFIFO entries
- Simplify error notifier setup flow
This should resolve hanging issues while maintaining core functionality.
Implements several previously stubbed functions in the NVDRV service:
- Initialize proper transfer memory handling
- Add error notifier configuration
- Implement channel timeout and timeslice management
- Add object context allocation and tracking
- Add GPU interface stubs for new functionality
The changes improve the accuracy of GPU-related operations while maintaining
compatibility with the existing codebase. All functions now properly validate
parameters and handle endianness correctly using _le types.
Two main changes in this commit:
1. Replace NVIDIA-specific GL_NV_gpu_shader5 extension with the more widely
supported GL_EXT_shader_explicit_arithmetic_types_float16 in the scaleforce
shader. This improves compatibility across different GPU vendors.
2. Refactor ARM32 exception handling:
- Restructure exception cases for better readability
- Update exception handling to match current Dynarmic API
- Fix indentation in switch statement
- Remove AccessViolation case as it's no longer supported in current API
These changes improve shader compatibility and align the exception handling
with the current Dynarmic implementation.
Previously, the code would unconditionally add a constant buffer descriptor
at index 0 whenever storage buffers were present, which could cause conflicts
and crashes. This change:
- Adds validation to check if constant buffer 0 already exists
- Only adds the descriptor if it's not already present
- Prevents potential descriptor conflicts in shaders
This should resolve crashes in Vulkan games related to invalid descriptor
layouts and resource binding conflicts.
Add support for bindless texture constant buffers in the compute pipeline
creation process. When storage buffer descriptors are present, create a
constant buffer descriptor to handle bindless textures. This fixes the
"Failed to track bindless texture constant buffer" error.
Changes:
- Add constant buffer descriptor with index 0 and count 1 when storage
buffers are present
- Place descriptor creation before SPIR-V code generation to ensure proper
shader compilation
This resolves issues with bindless texture access in compute shaders.
Fixes the StorageBufferBinding function to properly handle memory validation
and size detection. Key changes include:
- Fix ReadBlock usage to properly handle void return values
- Implement safer memory validation using byte-level reads
- Improve size detection logic for storage buffers
- Fix NVN buffer size reading
- Add proper bounds checking for device memory addresses
- Add better error logging for invalid conditions
This addresses the "Failed to find storage buffer for cbuf index 0" errors
by implementing more robust memory validation and size detection. The changes
ensure proper handling of invalid memory addresses and prevent crashes from
accessing out-of-bounds memory.
- Modify DmaPusher to use safe memory reads when handling compute
operations at High GPU accuracy
- Prevent potential memory corruption issues that could lead to
invalid dispatch parameters
- Previously, unsafe reads could result in corrupted launch_description
data in KeplerCompute::ProcessLaunch, causing invalid vkCmdDispatch
calls
- By enforcing safe reads specifically for compute operations, we
maintain performance for other GPU tasks while ensuring compute
dispatch stability
This change requires >= High GPU accuracy level to take effect.
Updated to fmt 11 with the required source changes for it to work.
Also updated vcpkg for this, and as an added benefit it fixes the `Unable to find a valid Visual Studio instance` error, and the VS 2019 build tools are no longer required. Just make sure to delete the existing downloaded vcpkg tool and binaries in `externals/vcpkg` if you have compiled before, or else it will continue to use the old version and give the error.
Reviewed-on: http://vub63vv26q6v27xzv2dtcd25xumubshogm67yrpaz2rculqxs7jlfqad.onion/torzu-emu/torzu/pulls/68
Co-authored-by: lui <lui@vub63vv26q6v27xzv2dtcd25xumubshogm67yrpaz2rculqxs7jlfqad.onion>
Co-committed-by: lui <lui@vub63vv26q6v27xzv2dtcd25xumubshogm67yrpaz2rculqxs7jlfqad.onion>
If unsafe read is done there can sometimes be corrupt data in the
KeplerCompute::ProcessLaunch qmd structure.
Fixes GPU crashes in 'Princess Peach: Showtime!' when using vulkan
renderer. Requires using "Accuracy Level High" (crashes will still
happen if using "Normal").
Tested on Radeon 6750XT, Linux 6.11.2, Mesa 24.2.5 (RADV driver).
Unsafe read was introduced in 115792158d
"VideoCore: Implement DispatchIndirect"
How did I debug this:
- Used VK_LAYER_KHRONOS_validation which found invalid vkCmdDispatch
(along with a lot of other noise!)
- Instrumented all calls to vulkan Dispatch(), set breakpoint when
grid_dim_x > 1024 (an obviously invalid value). Found dispatch came
from RasterizerVulkan::DispatchCompute().
- Commented out DispatchCompute() entirely, game runs with no crashes
but some graphics effects are missing.
- Keep going one layer up, observe corrupted `launch_description` in
KeplerCompute::ProcessLaunch()
- Attempted safe ReadBlock (`which = VideoCommon::CacheType::All`)
instead of ReadBlockUnsafe in KeplerCompute::ProcessLaunch(), did not
help
- Go one layer up to DmaPusher. Switch to safe_process(). No more
corrupt `launch_description`.
Adds Area Sampling to the list of scaling options. Works well to achieve a high-quality, smooth super-sampling effect. Dolphin has had this for a while, and now Ryujinx has recently added it too, so I decided to port it.
Not sure if adding the extra uniform to the OpenGL WindowAdaptPass was a good idea or not, or if using the push constants under Vulkan was either, but I wasn't sure about the best way to get the window size for use in the shader, and other scaling methods still work fine. Implementation seems to work fine under both Vulkan and OpenGL, but might still need some minor tweaks to the shader. Should definitely do some testing before merging, I have tested on an Nvidia RTX 3080 under Windows.
Adapted from these two PRs:
https://github.com/Ryujinx/Ryujinx/pull/7304https://github.com/dolphin-emu/dolphin/pull/11999
Reviewed-on: http://vub63vv26q6v27xzv2dtcd25xumubshogm67yrpaz2rculqxs7jlfqad.onion/torzu-emu/torzu/pulls/57
Co-authored-by: lui <lui@vub63vv26q6v27xzv2dtcd25xumubshogm67yrpaz2rculqxs7jlfqad.onion>
Co-committed-by: lui <lui@vub63vv26q6v27xzv2dtcd25xumubshogm67yrpaz2rculqxs7jlfqad.onion>
Adds a new option "On Load" to the "Optimize SPIRV output" option that turns on optimizations during the loading of the shader cache from disk, but turns it off after that.
The previous checkbox states have been named "Never" for unchecked and "Always" for checked.
The idea is that once the shader cache has most of the shaders in a game cached they can be optimized during initial game startup (where a performance hit matters less) and the few shaders that get compiled during runtime are not optimized to reduce performance hits.
Most of the commit is adding the setting to the Android app, the main logic is in the `gl_shader_cache.cpp` and `vk_pipeline_cache.cpp` files.
Reviewed-on: http://vub63vv26q6v27xzv2dtcd25xumubshogm67yrpaz2rculqxs7jlfqad.onion/torzu-emu/torzu/pulls/13
Co-authored-by: echosys <echosys@noreply.localhost>
Co-committed-by: echosys <echosys@noreply.localhost>
When compiling with -DCMAKE_BUILD_TYPE=Debug, GCC would (correctly) fail to
compile intrinsics in stb and host1x due to lack of optimizations.
Sadly, the compilation error given is bogus and Clang completing the builds
without issues does raise some eyebrows.
Therefore, force optimizations for the offending files under GCC when
creating Debug builds.
Signed-off-by: voidanix <voidanix@keyedlimepie.org>
