Optimized multi resolution LBM #44
Merged
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
This reverts commit ead3a64.
|
From Windows |
|
From Linux (cuda:12.2.0.ubuntu.22.04) |
massimim
approved these changes
Nov 20, 2023
There was a problem hiding this comment.
Looks great!
I added only one comment on the use of the compile time for-loop for LBM.
This is an example of how it can be used:
Neon/benchmarks/lbm/src/DeviceD3QXX.h
Line 30 in 1d48965
Comment on lines
+142
to
+144
| if (offset.x == 0 && offset.y == 0 && offset.z == 0) { | ||
| return idx; | ||
| } |
There was a problem hiding this comment.
A case where the offset is zero should not be a common case that requires a special treatment. The rest of the method should cover it already.
| @@ -20,34 +20,30 @@ inline Neon::set::Container explosion(Neon::domain::mGrid& grid, | |||
| return [=] NEON_CUDA_HOST_DEVICE(const typename Neon::domain::mGrid::Idx& cell) mutable { | |||
| //If this cell has children i.e., it is been refined, then we should not work on it | |||
| //because this cell is only there to allow query and not to operate on | |||
| if (!pin.hasChildren(cell)) { | |||
| if (!pin.hasChildren(cell) && pin.hasParent(cell)) { | |||
| for (int8_t q = 0; q < Q; ++q) { | |||
There was a problem hiding this comment.
This could be automatically unrolled via templates.
Comment on lines
26
to
28
| if (dir.x == 0 && dir.y == 0 && dir.z == 0) { | ||
| continue; | ||
| } |
There was a problem hiding this comment.
The check can be removed by looping only on the direction that are not the center.
For example STLBM leverages the structure of the stencil to do that, however more easily if the previous loop is based on a template for, we can transform the if statement into a constexpr if.
| //if the neighbor cell has children, then this 'cell' is interfacing with L-1 (fine) along q direction | ||
| //we want to only work on cells that interface with L+1 (coarse) cell along q | ||
| if (!pin.hasChildren(cell, dir)) { |
There was a problem hiding this comment.
We could have a version of the hasChildren where the dir parameter is a template.
| //since we are on the finest level, we only need to do streaming and explosion (no coalescence) | ||
| //streaming is done as push | ||
|
|
||
| const Neon::int8_3d dir = getDir(q); |
There was a problem hiding this comment.
This could be passed as a template parameter.
| if (nghType == CellType::bulk) { | ||
| out(nghCell, q) = cellVal; | ||
| } else { | ||
| const int8_t opposte_q = latticeOppositeID[q]; |
There was a problem hiding this comment.
With q be a runtime parameter, latticeOppositeID is loaded to memory.
With a template parameter q would be resolved at compile time.
However because of the type this is the place where using a compile time q would not impact performance.
| const Neon::int8_3d dir = getDir(q); | ||
|
|
||
| //if the neighbor cell has children, then this 'cell' is interfacing with L-1 (fine) along q direction | ||
| const auto nghCell = out.helpGetNghIdx(cell, dir); |
There was a problem hiding this comment.
bGrid has a helpGetNghIdx version where dir is a template parameter. That version has quite less number of runtime checks.
Comment on lines
+276
to
+306
| auto fdecompose_shear = [&](const int q) -> T { | ||
| const T Nxz = Pi[0] - Pi[5]; | ||
| const T Nyz = Pi[3] - Pi[5]; | ||
| if (q == 9) { | ||
| return (2.0 * Nxz - Nyz) / 6.0; | ||
| } else if (q == 18) { | ||
| return (2.0 * Nxz - Nyz) / 6.0; | ||
| } else if (q == 3) { | ||
| return (-Nxz + 2.0 * Nyz) / 6.0; | ||
| } else if (q == 6) { | ||
| return (-Nxz + 2.0 * Nyz) / 6.0; | ||
| } else if (q == 1) { | ||
| return (-Nxz - Nyz) / 6.0; | ||
| } else if (q == 2) { | ||
| return (-Nxz - Nyz) / 6.0; | ||
| } else if (q == 12 || q == 24) { | ||
| return Pi[1] / 4.0; | ||
| } else if (q == 21 || q == 15) { | ||
| return -Pi[1] / 4.0; | ||
| } else if (q == 10 || q == 20) { | ||
| return Pi[2] / 4.0; | ||
| } else if (q == 19 || q == 11) { | ||
| return -Pi[2] / 4.0; | ||
| } else if (q == 8 || q == 4) { | ||
| return Pi[4] / 4.0; | ||
| } else if (q == 7 || q == 5) { | ||
| return -Pi[4] / 4.0; | ||
| } else { | ||
| return T(0); | ||
| } | ||
| }; |
There was a problem hiding this comment.
Using a compile time q parameter, this could be transform into a constexpr function and saving quite a lot of checks.
| @@ -22,7 +22,8 @@ inline Neon::set::Container coalescence(Neon::domain::mGrid& g | |||
| return [=] NEON_CUDA_HOST_DEVICE(const typename Neon::domain::mGrid::Idx& cell) mutable { | |||
| //If this cell has children i.e., it is been refined, than we should not work on it | |||
| //because this cell is only there to allow query and not to operate on | |||
| const int refFactor = pout.getRefFactor(level); | |||
| //const int refFactor = pout.getRefFactor(level); | |||
| constexpr T repRefFactor = 0.5; | |||
| if (!pin.hasChildren(cell)) { | |||
|
|
|||
| for (int q = 0; q < Q; ++q) { | |||
There was a problem hiding this comment.
- Consider template for loop.
apps/lbmMultiRes/sphere3.obj
Outdated
There was a problem hiding this comment.
For the sphere can we use an analytic sdf instead to replace the obj file?
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
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.
mGirdmGrid