Enable userdata derivatives for interpolated params on GPU #1657
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
lgritz
requested changes
Mar 20, 2023
src/liboslexec/llvm_instance.cpp
Outdated
| * 3 // make room for derivs by default | ||
| : type.numelements(); | ||
| type.arraylen = n; | ||
| type.arraylen = type.numelements() * 3; // make room for derivs by default |
There was a problem hiding this comment.
Now that I'm looking at this, I think that we should only multiply by 3 if type.basetype == TypeDesc::FLOAT. Because anything based on int or string can't possibly have derivatives, so perhaps you can reclaim a little space that way (if indeed anything is ever declared userdata with those types, but it is possible).
Also, I'm seeing a CI complaint about clang-format on this line.
There was a problem hiding this comment.
Sounds good, I pushed an update addressing both points.
Signed-off-by: Chris Hellmuth <chellmuth@gmail.com>
chellmuth
force-pushed
the
gpu-userdata-derivs
branch
from
e789908
to
466b5e7
Compare
lgritz
pushed a commit
that referenced
this pull request
Apr 14, 2023
When generating calls to osl_bind_interpolated_param in llvm_instance.cpp, the userdata_has_derivs field is set to true if the parameter's derivatives are needed, but the code does not take into account that on gpu there is no space allocated for these derivatives. This leads to an incorrect derivatives pointer getting passed into the renderer, which the renderer will use to write into unrelated groupdata fields. For our renders, getting the correct image was worth the 1-2% inflation of groupdata struct size that we observed by simply re-enabling gpu userdata derivatives. If desired, adding an option to turn them on/off is straightforward, but I didn't want to preemptively bloat the code if it's not necessary. Signed-off-by: Chris Hellmuth <chellmuth@gmail.com>
5 tasks
lgritz
pushed a commit
that referenced
this pull request
May 20, 2023
…1685) Following #1657, it turns out we forgot to zero initialize the userdata derivatives for the optix path. This leads to some rendering glitches caused by the memory garbage left in those derivatives. Signed-off-by: Pascal Lecocq <pascal.lecocq@gmail.com> Co-authored-by: Pascal Lecocq <plecocq@imageworks.com>
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
Enable userdata derivatives for interpolated parameters on the GPU.
When generating calls to
osl_bind_interpolated_paramin llvm_instance.cpp, theuserdata_has_derivsfield is set to true if the parameter's derivatives are needed, but the code does not take into account that on gpu there is no space allocated for these derivatives.This leads to an incorrect derivatives pointer getting passed into the renderer, which the renderer will use to write into unrelated groupdata fields.
For our renders, getting the correct image was worth the 1-2% inflation of groupdata struct size that we observed by simply re-enabling gpu userdata derivatives. If desired, adding an option to turn them on/off is straightforward, but I didn't want to preemptively bloat the code if it's not necessary.
Tests
Checklist: