| Age | Commit message (Collapse) | Author |
|---|
|
to unintended behavior)
|
|
|
|
* Fix calling setTextureAddressModeFast and setTextureFilteringOptionFast with invalid tex type during fast binds
* Restore mRT->screen to GL_RGBA16F to fix lighting banding
|
|
|
|
* Remove GLM sse flag from cmake that was moved to llpreprocessor.h
* Further reduce performance loss of HDR and Sharpening on bandwith-constrained gpu by combining gamma correction into tonemap/sharpening shader passes
* Update SSE2NEON to 1.8.0 to fix random render nans
* Fix occasional startup crash from LLCachedControl being declared in global scope
|
|
Mac Arm
|
|
|
|
|
|
|
|
a non-float format
|
|
|
|
Fix: Remove potentially redundant RenderAutoHideSurfaceAreaLimit sett…
|
|
Merge 2025.03 release into develop.
|
|
registration
This commit removes a seemingly duplicated `connectRefreshCachedSettingsSafe` call in `LLPipeline::init()` for the `RenderAutoHideSurfaceAreaLimit` setting.
A duplicated registration for this setting was identified during a review of `LLPipeline::init()`. Double registration can lead to unexpected behavior, including potential CPU overhead.
The duplication *may* have been introduced with commit 440c7b2 (Added CollectFontVertexBuffers feature), though this requires further confirmation.
Testing Performed:
After removing the duplicate registration, the `RenderAutoHideSurfaceAreaLimit` functionality was validated by ensuring the following behavior (consistent with the existing code):
* A value of 0 (zero) causes all objects to appear regardless of size.
* Values slightly above zero result in only small objects appearing, with all others hidden.
* Increasing the value causes objects of increasing size to appear, while smaller objects remain visible.
This change merits careful review to ensure it has no unintended side effects, and to confirm the accuracy of these observations from other developers.
|
|
Resolves the root cause of shadow rendering failures when changing RenderShadowResolutionScale immediately after modifying other graphics settings (e.g., SSAO, HDR).
Investigation revealed that LLPipeline::createGLBuffers, which is called during certain graphics setting changes that require full buffer recreation, contained lines that incorrectly set mRT->width and mRT->height to zero *after* the call to allocateScreenBuffer had already established the correct dimensions.
This created a state inconsistency. If RenderShadowResolutionScale was changed immediately following the graphics setting change, the subsequent call to LLPipeline::resizeShadowTexture (triggered via handleShadowsResized) would read these incorrect zero dimensions from mRT. This led to failed shadow buffer allocation (allocateShadowBuffer(0, 0)) and resulted in corrupted or missing shadows.
This commit removes the erroneous mRT->width = 0 and mRT->height = 0 lines from the end of createGLBuffers. This ensures that the render target dimensions remain valid after buffer recreation.
With this fix, resizeShadowTexture now correctly reads the valid screen dimensions immediately following a graphics setting change and successfully resizes the shadow buffers without delay or error. This eliminates the need for previous workarounds like guard conditions or forced shader recompiles.
Ref: #3719
|
|
dimensions after shader changes; **Replaces forced shader refresh with lightweight guard**
This commit introduces a guard in `LLPipeline::resizeShadowTexture()` to prevent shadow texture resizing when the shadow render target (mRT) has invalid (zero) dimensions. **This replaces a previous, less efficient approach of forcing a full shader recompile whenever `RenderShadowResolutionScale` was changed in-session.**
**Background and Problem:**
Previously, the code forced a full shader recompile whenever `RenderShadowResolutionScale` changed in-session (after toggling advanced graphics settings like SSAO or HDR). While this “sledgehammer” approach did fix broken shadow rendering, it unnecessarily thrashed the shader cache and reset many pipeline states.
**Solution:**
This commit removes the forced shader recompile in favor of a guard check in `LLPipeline::resizeShadowTexture()`. The guard ensures mRT (the shadow render target) has non-zero dimensions before resizing. If mRT is zero for that frame, the resize operation is skipped, and a warning is logged. Once mRT becomes valid (usually in the next frame), the shadow texture is resized successfully without requiring a full shader refresh.
**Detailed changes:**
- Reverted the binding of `RenderShadowResolutionScale` to `handleSetShaderChanged`.
- Restored the original `handleShadowsResized` listener for `RenderShadowResolutionScale` in `llviewercontrol.cpp`.
- Added guard checks in `LLPipeline::resizeShadowTexture()` to skip resizing when `mRT->width` or `mRT->height` is zero.
- Added logging statements to track how many frames are skipped.
**Benefits:**
- Prevents shader thrashing while still avoiding shadow corruption.
- Shadows now update correctly as soon as mRT dimensions are valid.
- Maintains a detailed record of frames skipped.
- **Lightweight and targeted interim solution, much less disruptive than a full shader recompile.**
Testing:
1. Reproduce the bug as described in the bug report (toggle SSAO, then change RenderShadowResolutionScale).
2. Verify that shadows are no longer broken after these steps.
3. Check the logs for the warning message indicating skipped frames when the bug is triggered.
4. Confirm that under normal operation (without shader changes causing mRT issues), shadow resizing works as expected without excessive warnings.
Documentation:
No user-facing documentation changes are needed for this interim fix. However, internal developer documentation should note this guard and the ongoing investigation into the root cause.
Further Development:
This guard is a temporary fix. The root cause of why mRT becomes invalid after shader changes needs to be investigated and resolved. See the bug report for detailed next steps for investigation.
|
|
|
|
Likely specific to headless client
|
|
|
|
|
|
Fix disabling renderdoc support
Improve ll_aligned_alloc functions on darwin for 32 and 64byte aligned by utilizing posix_memalign
|
|
* #3455 Add support for water exclusion surfaces
|
|
LLVector3/4 and fall back mul_mat4_vec3 to scalar implementation to attempt crash mitigation (#3339)
|
|
|
|
|
|
3.1/3.2/3.3 (#3184)
|
|
|
|
|
|
|
|
|
|
|
|
At the moment slider is in general settings, not per environment
|
|
* #3103 Add the ability to disable the emissive buffer for older GPUs with low memory bandwidth.
* #3135 Add a "vintage" mode for slower GPUs
* #2719 Fix for skies being overbrightened
* #2632 Do not apply tonemapping on legacy skies
|
|
* Partial solution to secondlife/viewer#2744 crash with better error handling.
Handles shader compile errors better, and should turn crash into an LL_ERRS assertion failure.
Strengthed more assertions and improved shader error line numbers
* Even more error handling to get a handle on crash secondlife/viewer#2744
* Improved GLSL correctness on Intel chips that lack OpenGL 4.6 support. secondlife/viewer#2744
* Removed non-working fallback code for gDeferredPostProgram
* Fixed incorrect llmax call
|
|
* OpenGL 3.3 compatibility pass. Fix for FBO driven downscaling corrupting textures.
* Increase maximum texture bias, immediately scale down when textures are loaded higher resolution than desired
* #2590 Fix for some frame stalls on Intel Macs
|
|
(#2760)
fix secondlife/viewer#2756
|
|
for testing purposes
|
|
tonemapping (#2659)
Co-authored-by: Rye Cogtail <rye@lindenlab.com>
|
|
|
|
|
|
Raise resolution of local baked texture preview from 512 to 2048
|
|
|
|
|
|
|
|
|
|
|
|
|
|
to control tonemap mix (#2464) (#2468)
|
|
Add SMAA buffer generation passes
Add quality levels for both FXAA and SMAA
Separate gamma correction and tonemapping for effects that require linear-but-tonemapped inputs
Move application of noise to final render pass to screen to avoid damaging other post process effects
|
|
|
