Uh oh!
There was an error while loading. Please reload this page.
pushpull: device dispatch for pull/push/count/grad_backward - #46
Merged
Conversation
Thin hub wrappers, identical in shape to the forward four: check that every operand is on the same device, then forward to FF_CUDA:: or FF_CPU::. Bumps both submodule pins to the commits that add the ops. Together with the layers below, this is what lets `fastfields.torch` differentiate `pull`/`push` through the sampling `grid` (fastfields-torch#16). MIGRATION.md: pushpull's row now reads pull/push/count/grad *and* their four adjoints (only `hess` remains impl-only), and records the four latent jitfields bugs the new finite-difference oracle turned up plus the build-cost reasoning behind the separate `pushpull_backward` module. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016AjQcY78NgbagPSbPJRr6Z
This was referenced Jul 31, 2026
Merged
… conflicts Both main and this branch independently re-pinned cpu (-> cpu-lib#49's merge, 6aaa159) and cuda (-> cuda-lib#32's merge, 3be8587) since their common ancestor. Both merged SHAs are descendants of what main had (cpu-lib#48's and cuda-lib#31's merges respectively), so they are the correct resolution. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016AjQcY78NgbagPSbPJRr6Z
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 freeto join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Part of the stack for fastfields-torch#16 (backpropagate through the pushpull sampling
grid).What
Thin hub wrappers, identical in shape to the forward four:
require_same_deviceacross every operand, then forward toFF_CUDA::orFF_CPU::. Plus the two submodule pin bumps.Together with the layers below, this is what lets
fastfields.torchdifferentiatepull/pushthrough the samplinggrid.MIGRATION.md: pushpull's row now reads pull/push/count/grad and their four adjoints (onlyhessremains impl-only), and records the four latent jitfields bugs the new finite-difference oracle turned up, plus the build-cost reasoning behind the separatepushpull_backwardmodule.Verification
libfastfields.solinks and exports all four:Downstream, the full Python stack was rebuilt against this and passes:
fastfields-dlpack36 tests,fastfields-torch104 tests (including 26 newgradchecks that exercise these entry points end-to-end).Workstream: claude-jitfields-to-fastfields
🤖 Generated with Claude Code
Generated by Claude Code