Lut-free built-ins #18
Open
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
cozdas
commented
Jul 20, 2024
cozdas
commented
- WIP Fixed-function for PQ<->Linear function. Doesn't have the GPU implementation yet and the rest may need tweaks.
doug-walker
reviewed
Jul 20, 2024
There was a problem hiding this comment.
I need to review more thoroughly, but you're on the right track.
There's a lot more to adding a new FixedFunction though. CTF support and Python support are required. Version checks must be updated to not allow the additions in config versions lower than 2.4. Unit tests must be added to test all of that.
Some of the other PRs that might be a useful reference are AcademySoftwareFoundation#1404, AcademySoftwareFoundation#1814, and AcademySoftwareFoundation#896.
cozdas
force-pushed
the
cozdas/main/Lut-Free-Built-Ins
branch
from
33cdc83 to
11d7e17
Compare
cozdas
commented
Aug 1, 2024
cozdas
commented
Aug 1, 2024
doug-walker
requested changes
Aug 12, 2024
There was a problem hiding this comment.
Please update tests/cpu/Config_tests.cpp, fixed_function_serialization, around line 4795 to validate Yaml serialization. Then add a check (there are other examples in that test) that an exception will be thrown on validation if the version is less than 2.4.
Please update tests/cpu/fileformats/FileFormatCTF_tests.cpp, ff_load_save_ctf, around line 3841 to validate CTF serialization.
I'm realizing that we have not been as rigorous on CTF versioning as we have on config versioning. Given that we are adding a lot for 2.4 we should bump the CTF version. This could be done in a follow-up.
…mplementation yet and the rest may need tweaks. Signed-off-by: cuneyt.ozdas <[email protected]>
- Python: Added docs - Python: added the builtin function style - Simplified the CPU code, it's still a prototype anyhow. Will optimize later. - Style->OpData now takes the direction into account. - Preliminary GPU implementation. This still clamps negative values. - Added GPU tests. Signed-off-by: cuneyt.ozdas <[email protected]>
- Added symmetry to the PQ curve both in CPU and GPU implementations - Added version check for the FIXED_FUNCTION_PQ_TO_LINEAR transform. - Adjusting the GPU test threshold as GPU and CPU (32f) are showing mismatch more than the distance each have to the ground truth. - Adding fixed function OP tests - Temporary change to the PQ curve CPU code for quickly switching between single and double precision floating point for error analysis. Signed-off-by: cuneyt.ozdas <[email protected]>
Signed-off-by: cuneyt.ozdas <[email protected]>
cozdas
force-pushed
the
cozdas/main/Lut-Free-Built-Ins
branch
from
17f0783 to
a2e5459
Compare
cozdas
commented
Aug 13, 2024
…unction - minor fixes Signed-off-by: cuneyt.ozdas <[email protected]>
…unction - implemented SSE versions of the PQ curve in both directions with and without fastPower - restored the old LUT-based PQ curve implementation that's used in the ST2084 Display builtin transform - added custom range options to gpu unit tests - changed the PQ curve test ranges to [-0.1, 1.1] in forward and [-0.1, 1001.1] in reverse directions Signed-off-by: cuneyt.ozdas <[email protected]>
Signed-off-by: cuneyt.ozdas <[email protected]>
…anoseconds, this is optional, so the existing clients don't need to change. - HLG curve in CIE_XYZ_D65_to_REC2100_HLG_1000nit_Functor is pulled into its own function to be reused later on. - HLG curve is no longer clipping at 0, it mirrors instead. clipping at 1 is currently untouched, may need to remove that too. - fine-tuned the unit test thresholds for PQ curve so that they pass in all build/use situations. Signed-off-by: cuneyt.ozdas <[email protected]>
…gcc and clang will keep on failing as I realized that those compilers don't have the _mm_pow_ps() SVML library implementation as msvc does. Will need to change the logic a bit. Signed-off-by: cuneyt.ozdas <[email protected]>
Signed-off-by: cuneyt.ozdas <[email protected]>
…her use the fast-SSE or scalar implementation, while Windows will be able to use the precise-SSE implementation too if needed. Fingers crossed. Signed-off-by: cuneyt.ozdas <[email protected]>
doug-walker
requested changes
Aug 24, 2024
…t which should be available since GL version 3.3 according to Khronos. Joy of cross-platform development :) Therefore removing the GPU render timing code from this PR. This is probably why GPU performance was not implemented in ocioperf already.
…dea to try to make all the compiler warning levels same/similar, some platforms being more picky about the same code than the others does not make much sense from the project's point of view, just frustration. Signed-off-by: cuneyt.ozdas <[email protected]>
Signed-off-by: cuneyt.ozdas <[email protected]>
Signed-off-by: cuneyt.ozdas <[email protected]>
…ate more. Signed-off-by: cuneyt.ozdas <[email protected]>
…er code review discussion - BT_2100 HGL clamp at 1 is removed. - Fixed function tests can now ask for fast math on or off explicitly. - Fixed function PQ curve tests are now performed both with and without fast math. With fast math on, the thresholds are increased to accommodate the accumulated error during roundtrip. Signed-off-by: cuneyt.ozdas <[email protected]>
doug-walker
approved these changes
Aug 27, 2024
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.