…mplementation yet and the rest may need tweaks.

Signed-off-by: cuneyt.ozdas <cuneyt.ozdas@autodesk.com>

- 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 <cuneyt.ozdas@autodesk.com>

- 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 <cuneyt.ozdas@autodesk.com>

Signed-off-by: cuneyt.ozdas <cuneyt.ozdas@autodesk.com>

…unction

- minor fixes

Signed-off-by: cuneyt.ozdas <cuneyt.ozdas@autodesk.com>

…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 <cuneyt.ozdas@autodesk.com>

Signed-off-by: cuneyt.ozdas <cuneyt.ozdas@autodesk.com>

…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 <cuneyt.ozdas@autodesk.com>

…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 <cuneyt.ozdas@autodesk.com>

Signed-off-by: cuneyt.ozdas <cuneyt.ozdas@autodesk.com>

…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 <cuneyt.ozdas@autodesk.com>

…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 <cuneyt.ozdas@autodesk.com>

Signed-off-by: cuneyt.ozdas <cuneyt.ozdas@autodesk.com>

Signed-off-by: cuneyt.ozdas <cuneyt.ozdas@autodesk.com>

…ate more.

Signed-off-by: cuneyt.ozdas <cuneyt.ozdas@autodesk.com>

…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 <cuneyt.ozdas@autodesk.com>