Make recent run-layer optimizations optional, and fix init ops-related false positive #1672
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
… false positive Signed-off-by: Chris Hellmuth <chellmuth@gmail.com>
lgritz
approved these changes
Apr 20, 2023
|
Is the new |
AlexMWells
reviewed
Apr 20, 2023
| @@ -1024,6 +1024,7 @@ ShadingSystemImpl::ShadingSystemImpl(RendererServices* renderer, | |||
| , m_opt_middleman(true) | |||
| , m_opt_texture_handle(true) | |||
| , m_opt_seed_bblock_aliases(true) | |||
| , m_opt_useparam(false) | |||
There was a problem hiding this comment.
Will then be enabled by default once confidence in its correctness is established?
There was a problem hiding this comment.
Right. Chris is reverting most of the original change, and also putting this option in that defaults off but will allow it to be turned on selectively (after it's re-introduced, hopefully fixed) while we're gaining confidence, then will eventually be switched to default on (with ability to turn it off if anything fishy happens).
chellmuth
pushed a commit
to chellmuth/OpenShadingLanguage
that referenced
this pull request
Sep 6, 2024
* Don't insert redundant run layer calls inside a basic block (AcademySoftwareFoundation#1665) * Apply recent run layer call optimization to batched execution (AcademySoftwareFoundation#1669) * Do-over: Make recent run-layer optimizations optional, and fix init ops-related false positive (AcademySoftwareFoundation#1672) * fix for ReParameterization corner case (AcademySoftwareFoundation#1670) * RendererServices API for letting get_texture_handle consider colorspace (AcademySoftwareFoundation#1641) * Set up ray types for testrender (AcademySoftwareFoundation#1648) * testrender: Don't use the cached background map on the first bounce (AcademySoftwareFoundation#1649) * Switch lockgeom to interpolated and interactive (AcademySoftwareFoundation#1662) * Add type information for needed attributes. (AcademySoftwareFoundation#1650) Note that this bump to 1.13.3.x is only for Arnold at this point, and not yet for the platforms, so to make CI go faster I have rigged .spdev.yaml to only build the sparnold versions at this time. Also updated ABI dump files. See merge request spi/dev/3rd-party/osl-feedstock!47
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.
Description
My initial implementation of the run layer check optimizations didn’t account for the fact that init ops for symbols have their llvm code generated before we build basic blocks from the OSL IR.
This caused the code to be wrong in two ways: 1) the set of known run layer calls was being reset between instances too late in the pipeline, so a useparam inside an init op would actually be checking the previously compiled layer’s set, and 2) even when reset earlier, it assumed valid basic block ids existed for the init ops when they did not.
So, if a layer had a useparam inside one of its init ops that generally matched the same code location as a useparam in the previously compiled layer, the run layer check could be incorrectly omitted.
I’ve updated the code so that the default optimization behavior is reverted to the way it was before #1665, and added a new option
opt_useparamthat can be turned on to enable a fixed, restricted version of the optimization that only applies to code in the main section.Tests
I'm working on a test that properly triggers the issue.
Checklist: