This enables most of the GPU compiler's optimizations while -ffast-math
isn't set at DPC++ level.
It brings an overall 1% speedup and currently doesn't change the unit
tests pass rate.
This enables three additional math optimizations:
-ffp-contract=fast (enables FMA generation)
-freciprocal-math (enables x/y -> x*(1/y))
-fassociative-math (enables e.g. a*b + c*b -> (a+c)*b)
These are used on Windows and HIP anyways, so our code can't expect exact IEEE
semantics in any case.
The only difference between the new set and -ffast-math is that we don't use
-ffinite-math-only since this causes issues with the BVH (see ce1f2e271d) and
breaks e.g. isnan.
This causes a ~1.5% speedup in my very quick test, but might be higher for some
more math-intensive cases.
Pull Request: https://projects.blender.org/blender/blender/pulls/128342
The only difference between Windows+Clang and the others is a prefix, so use
some CMake logic to just prepend that to all flags instead of duplicating them.
Pull Request: https://projects.blender.org/blender/blender/pulls/128342
Happens for renders from command line, when kernel specialization
thread is still working after the allocators on the Blender side
have been deinitialized.
Add an explicit deinitializaiton, which ensures all Cycles worker
and cache threads are finished before the allocators are deinitialized.
This should solve occasional crashes when running regression tests
for Metal or Metal-RT.
Pull Request: https://projects.blender.org/blender/blender/pulls/128239
This improve the API in multiple aspects:
* No need for an additional `lookup` call to get the current attribute. This
would internally iterate over all attributes again. This leads to O(n^2)
behavior. Note that there are still other reasons for O(n^2) behavior when
processing attributes (where n is the number of attributes).
* Remove the need to return a value from the iteration code to indicate that the
iteration should continue. This is now the default behavior. The iteration can
still be stopped by calling `iter.stop()`.
* Easier access to `is_builtin` property.
* Iterator callback only has a single parameter instead of two (of which one is
sometimes unused).
Pull Request: https://projects.blender.org/blender/blender/pulls/128128
The issue was that cycles did not take ownership of the entire `VolumeGridData`.
The `VolumeTreeAccessToken` that was used already only makes sure that the
grid is not unloaded when it was read from disk for example. It does not extend
the lifetime of the `VolumeGridData` as a whole.
This change switches Cycles to an opensource HIP-RT library which
implements hardware ray-tracing. This library is now used on
both Windows and Linux. While there should be no noticeable changes
on Windows, on Linux this adds support for hardware ray-tracing on
AMD GPUs.
The majority of the change is typical platform code to add new
library to the dependency builder, and a change in the way how
ahead-of-time (AoT) kernels are compiled. There are changes in
Cycles itself, but they are rather straightforward: some APIs
changed in the opensource version of the library.
There are a couple of extra files which are needed for this to
work: hiprt02003_6.1_amd.hipfb and oro_compiled_kernels.hipfb.
There are some assumptions in the HIP-RT library about how they
are available. Currently they follow the same rule as AoT
kernels for oneAPI:
- On Windows they are next to blender.exe
- On Linux they are in the lib/ folder
Performance comparison on Ubuntu 22.04.5:
```
GPU: AMD Radeon PRO W7800
Driver: amdgpu-install_6.1.60103-1_all.deb
main hip-rt
attic 0.1414s 0.0932s
barbershop_interior 0.1563s 0.1258s
bistro 0.2134s 0.1597s
bmw27 0.0119s 0.0099s
classroom 0.1006s 0.0803s
fishy_cat 0.0248s 0.0178s
junkshop 0.0916s 0.0713s
koro 0.0589s 0.0720s
monster 0.0435s 0.0385s
pabellon 0.0543s 0.0391s
sponza 0.0223s 0.0180s
spring 0.1026s 1.5145s
victor 0.1901s 0.1239s
wdas_cloud 0.1153s 0.1125s
```
Co-authored-by: Brecht Van Lommel <brecht@blender.org>
Co-authored-by: Ray Molenkamp <github@lazydodo.com>
Co-authored-by: Sergey Sharybin <sergey@blender.org>
Pull Request: https://projects.blender.org/blender/blender/pulls/121050
Avoid printing the word `size` twice when printing information about
the scheduled tile size to terminal when Blender is launched with
with `--debug-cycles --verbose 3` or higher.
Pull Request: https://projects.blender.org/blender/blender/pulls/127885
Since their introduction, generated UDIM images/tiles would not be
loaded in Cycles. In relative recent history 72ab6faf5d added the
ability to track which tiles were still marked as being "generated" and
unsaved. However, that check was never implemented into the Cycles code.
With this PR, these two additional scenarios should now work:
- Properly load pixels if all tiles are generated
- Properly load pixels if a UDIM image has been loaded from a file and
new generated tiles have been added but not saved yet etc.
Pull Request: https://projects.blender.org/blender/blender/pulls/127673
Temporarily disable point cloud rendering in HIPRT to fix a performance
regression triggered by increased register preasure until
a better solution can be developed.
Pull Request: https://projects.blender.org/blender/blender/pulls/127738
Deforming motion blurred point clouds do not render in Cycles
HIP-RT when BVH timesteps != 0 if Blender is launched with
debug memory.
The root cause is that the size of allocated memory for the
bounding boxes is reported to HIP-RT not the number of valid
bounding boxes.
Pull Request: https://projects.blender.org/blender/blender/pulls/127432
VDB files would fail to render in HIP-RT because NanoVDB wasn't
enabled when compiling HIP-RT kernels, resulting in NanoVDB textures
not being sampled and a blank result being returned instead.
The fix is to enable NanoVDB when compiling HIP-RT kernels.
Ref: #125086
Pull Request: https://projects.blender.org/blender/blender/pulls/127384
Fix the unnecessary recreation of the denoiser that occurs if
Cycles had fallen back to an alternative denoiser in a previous
interation. (E.g. Fallback from OptiX to OIDN)
This issue occured because Cycles didn't understand that when it
previously setup the denoising device, that it had fallen back to
something else. So it thinks the denoising settings have been changes
and tries to recreate the denoiser.
The solution is to first compute the settings change due to
the fallback, then check to see if it's different from the current
denoiser, then recreate the denoiser device if neccesary.
Pull Request: https://projects.blender.org/blender/blender/pulls/125453
The same random number was used for sampling color channel at each step,
which leads to bias. Fixed by rescaling the random number.
Another possibility would be to scramble `rng_offset` and use a new
random number each time, similar as in subsurface scattering, but
rescaling random number should be faster than computing a new one, and
is favorable here since the precision here is not very important
Pull Request: https://projects.blender.org/blender/blender/pulls/127454
IGC 1.0.17384, ocloc 24.31.30508, which:
- add support for Battlemage and Lunar Lake GPUs
- recover from recent performance regression on Linux
- allow to drop older work-around
(9d5164d472) and need for a patched
version on Windows
- ocloc now needs "dg2,mtl" naming for fat binaries.
opencl-clang patches don't get applied anymore by igc build scripts
when llvm is not a git repository, hence I could also drop we can drop
current patch disabling patching.
I've only slightly pushed min-driver-version updates after carefull
testing, instead of jumping to the same version as ocloc as we use to.
Pull Request: https://projects.blender.org/blender/blender/pulls/127251
This new version of the graphics compiler solves a performance
regression on Arc, adds support for Battlemage and Lunar Lake GPUs, and
allows to drop older patch to build fat binaries with broad
compatibility.
This latter change requires using -device dg2,mtl naming instead of
passing architecture ids.
Pull Request: https://projects.blender.org/blender/blender/pulls/127371
The device code was disabled for primitives with deformation blur
and the intersection function always returned false, hence no
rendered primitive.
Other than that, there were a few bugs on both device and host codes
(e.g., the order of current and previous times and the primitive name.)
Pull Request: https://projects.blender.org/blender/blender/pulls/127163
Previously, the `AttributeIDRef` wrapper was needed because it also had to
contain a pointer to an `AnonymousAttributeID`. However, since
b279a6d703 this is not necessary anymore.
Therefore we can use "raw" `StringRef` now which reduces the mental overhead
when working with attributes and also simplifies code.
Pull Request: https://projects.blender.org/blender/blender/pulls/127140
Fixes a few issues with point clouds with HIPRT.
1. Crashing when building the BLAS due to an incorrect sized array.
2. A typo leading to all point cloud intersections being skipped.
3. A typo leading to some motion blurred point clouds rendering
as if they were stationary, or not rendering at all.
Pointclouds, with deformable motion blur, with BVH time steps set to >0
still do not render. Curves seem to have the same issue.
Ref #125086
Pull Request: https://projects.blender.org/blender/blender/pulls/125834
`atan2(0, 0)` is undefined on many platforms. To ensure consistent
result across platforms, we return `0` in this case.
Note only the behavior of the shader node `Artan2` is changed here.
During shading, we might still produce `atan2(0, 0)` internally and
cause different results across platforms, but that usually happens with
single samples and is not obvious, plus checking this condition all the
time is costly. If later we find out it's indeed necessary to change all
the invocation of `atan2(0, 0)`, we could change the wrapper functions
in `metal/compat.h` and `mtl_shader_defines.msl`.
Pull Request: https://projects.blender.org/blender/blender/pulls/126951
The kernel zeroing memory since we've added host memory fallback didn't
expect large inputs, so with these scenes, it was running into
"Provided range is out of integer limits. Pass
`-fno-sycl-id-queries-fit-in-int' to disable range check" error.
This kernel was used instead of memset to avoid some issues with the
free_memory queries not always being updated.
As we can't reproduce these with recent drivers, we now use memset,
which fixes rendering with BVH2.
Use a bounding sphere instead of the corners of a bounding box to
compute the subtended angle of a light tree node.
Using the corners of the bounding box was an underestimate in some
scenes, causing some light tree nodes being incorrectly skipped.
Using the subtended angle of a bounding sphere is an overestimate, but
it covers the entire node and would not skip any valid contribution,
and no other reliable algorithm to compute the minimal enclosing angle
is known to us.
We expect some increase in noise due to overestimation, but this has
not been observed yet, in our benchmark scenes only a difference in
noise is visible.
Thanks to Weizhen for the suggestion to use the bounding sphere.
Pull Request: https://projects.blender.org/blender/blender/pulls/126625
Add Metallic BSDF Node to the shader editor.
This node can primarily be used to create more realistic looking
metallic materials than the existing Glossy BSDF node.
This commit does not add any new closures to Cycles, it simply exposes
existing closures that were previous hard to access on their own.
- Exposes the F82 fresnel type that is currently used by the
metallic component of the Principled BSDF. Results should match
between the Metallic BSDF and Principled BSDF when using the same
settings.
- Exposes the Physical Conductor fresnel type that was previously
limited to custom OSL scripts. The Conductor fresnel type accepts
IOR and Extinction coefficients to define the appearance of the
material based off real life measurements.
EEVEE only supports the F82 fresnel type with internal code to convert
the the physical conductor inputs in to a colour format for F82,
which can lead to noticeable rendering differences with
some configurations.
Pull Request: https://projects.blender.org/blender/blender/pulls/114958
Some of the device memory objects had their host_pointer overwritten
with another CPU-side buffer after allocation. This leads to a leak of
host memory allocated by the device_memory.
There are few remaining places where the host_pointer is assigned and
those seems to be fine because the memory was not yet allocated with
a alloc() call.
While the approach in this change is not very ideal, it is small and
potentially could be ported to the LTS tracks. More ideal solution
would be to utilize device_vector::give_data().
Pull Request: https://projects.blender.org/blender/blender/pulls/126788
when tracing shadow ray through a volume and no hit is registered, we
consider the whole ray segment inside the volume.
However, no hit registered could also happen when the volume is
invisible to shadow ray. We should explicitly check this case and skip
rendering the volume segment instead.
Pull Request: https://projects.blender.org/blender/blender/pulls/126139
