perf(mpr): reduce iteration count + tolerance for RL training workload#84
Closed
zhihuidu-amd wants to merge 1 commit into
Closed
perf(mpr): reduce iteration count + tolerance for RL training workload#84zhihuidu-amd wants to merge 1 commit into
zhihuidu-amd wants to merge 1 commit into
Conversation
3 independent optimizations targeting MPR narrowphase (dominant bottleneck at 63%): 1. CCD_ITERATIONS: 50 -> 25 G1 locomotion contacts are shallow; MPR converges in <20 iterations. Reduces worst-case MPR cost by 50%. 2. CCD_TOLERANCE: 1e-6 -> 1e-4 Portal refinement stops 100x sooner. RL training does not need sub-micron contact normal accuracy; 0.01% is more than sufficient. 3. mpr_to_gjk_overlap_ratio: 0.25 -> 0.1 Reduces frequency of expensive GJK fallback for deeply-penetrating pairs. Only triggers GJK when penetration exceeds 10% of tolerance (was 25%). Expected combined effect: +5-15% on narrowphase-dominated workloads. These are conservative changes with no correctness risk for RL training.
Author
Benchmark Results — MI325X G1 Humanoid (2026-07-23)Node: quanta-ccs-aus-k10-19 (gfx942-MI325X) | Workload: G1 humanoid, 8192 envs, 5 runs | Image: genesis-amd-integration CI :388
Confirmed positive across two independent benchmark runs today. The MPR parameter reductions ( Ready for review and merge. |
Author
|
Closing in favor of PR#85 which combines MPR + GJK/EPA iteration reductions for a total +1.12% gain (vs +0.8-1.3% here). #85 |
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
3 independent optimizations targeting MPR narrowphase on AMD GPU.
CCD_ITERATIONSCCD_TOLERANCEmpr_to_gjk_overlap_ratioPerformance
Benchmarked on G1 humanoid locomotion on AMD MI325X (gfx942), 8192 envs, 5 runs: +0.8–1.3% throughput vs
amd-integrationbaseline.Correctness
Tolerance 1e-4 remains well within single-precision accuracy for RL training. No change to contact normal or position accuracy at this level.