Fix #129420: precision issue in light tree distant light angle #129489

Weizhen Huang · 2024-10-28T22:13:06+01:00

Weizhen Huang commented

2024-10-28 22:13:06 +01:00

In volume segment, the minimal angle formed by the emitter bounding cone
axis and the vector pointing from the cluster centroid to any point on
the ray is computed via dot(bcone.axis, point_to_centroid), see Fig.8.
in paper.
For distant light this angle is 0, but due to numerical issues this is
not always true. Therefore explicitly assign -bcone.axis to
point_to_centroid in this case.

In volume segment, the minimal angle formed by the emitter bounding cone axis and the vector pointing from the cluster centroid to any point on the ray is computed via `dot(bcone.axis, point_to_centroid)`, see Fig.8. in paper. For distant light this angle is 0, but due to numerical issues this is not always true. Therefore explicitly assign `-bcone.axis` to `point_to_centroid` in this case.

Weizhen Huang added the

Module

Render & Cycles

label 2024-10-28 22:13:06 +01:00

Weizhen Huang added 1 commit 2024-10-28 22:13:16 +01:00

Fix #129420 : precision issue in light tree distant light angle

buildbot/vexp-code-patch-lint Build done.

Details

buildbot/vexp-code-patch-linux-x86_64 Build done.

Details

buildbot/vexp-code-patch-darwin-arm64 Build done.

Details

buildbot/vexp-code-patch-darwin-x86_64 Build done.

Details

buildbot/vexp-code-patch-windows-amd64 Build done.

Details

buildbot/vexp-code-patch-coordinator Build done.

Details

655ccd6ed4

In volume segment, the minimal angle formed by the bounding cone axis
and the vector pointing from the cluster centroid and any point on the
ray is computed via `dot(bcone.axis, point_to_centroid)`, see Fig.8. in
paper.
For distant light this angle is 0, but due to numerical issues this is
not always true. Therefore explicitly assign `-bcone.axis` to
`point_to_centroid` in this case.

Weizhen Huang commented

2024-10-28 22:16:38 +01:00

The logic here of conditionally assigning some variables but change them later is quite ugly. I tried to rewrite but it ended up with too many changes (!129483), which is probably unsuitable for 4.3 and 4.2. I think it's better to do that refactor in 4.4.

Weizhen Huang commented

2024-10-28 22:16:55 +01:00

@blender-bot build

Weizhen Huang requested review from Sergey Sharybin 2024-10-29 11:35:44 +01:00

Sergey Sharybin approved these changes 2024-10-29 11:47:26 +01:00

Sergey Sharybin left a comment

For the 4.3 this indeed seems to be the easiest and safest way forward.
A bit annoying to fetch the KernelLight and check the type again, but that is what hopefully can be solved by the !129483 :)

For the 4.3 this indeed seems to be the easiest and safest way forward. A bit annoying to fetch the KernelLight and check the type again, but that is what hopefully can be solved by the !129483 :)

Weizhen Huang merged commit 219e655119 into blender-v4.3-release

2024-10-29 18:01:11 +01:00

Weizhen Huang referenced this issue from a commit

2024-10-29 18:01:12 +01:00

Fix #129420: precision issue in light tree distant light angle

Weizhen Huang deleted branch fix-129420-4.3

2024-10-29 18:01:19 +01:00

casey-bianco-davis referenced this issue from a commit

2024-10-30 04:13:07 +01:00

Fix #129420: precision issue in light tree distant light angle

Philipp Oeser referenced this issue from a commit

2024-11-13 12:10:56 +01:00