refactor(pre-dequant): extract Phase 0 + 0b (NVFP4 loader) to its own TU#305
Closed
kekzl wants to merge 5 commits into
Closed
refactor(pre-dequant): extract Phase 0 + 0b (NVFP4 loader) to its own TU#305kekzl wants to merge 5 commits into
kekzl wants to merge 5 commits into
Conversation
Anonymous-namespace and file-scope static helpers (infer_tier_from_wcache, borrow_payload_from_wcache, nvfp4_beneficial, deduct_budget, create_fused_weight_pair, for_each_dense_weight) move to a new header src/exec/pre_dequant_internal.h so subsequent per-phase extractions can share them across translation units. No behavior change. executor_pre_dequant.cu still contains every phase function -- those move to their own TUs in Tasks 2-7. Phase 3 of docs/superpowers/specs/2026-05-20-architecture-refactor-roadmap-design.md Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Moves GraphExecutor::pre_dequant_phase1_fp16_cache_ from src/exec/executor_pre_dequant.cu to a new TU src/exec/pre_dequant_phase1_fp16_cache.cu. The declaration in src/exec/executor.h is unchanged. The function body is byte-identical (verified by reading both before and after). Phase 3 of docs/superpowers/specs/2026-05-20-architecture-refactor-roadmap-design.md Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Moves GraphExecutor::pre_dequant_phase2_fp8_cache_ from src/exec/executor_pre_dequant.cu to a new TU src/exec/pre_dequant_phase2_fp8_cache.cu. Declaration in src/exec/executor.h unchanged. Function body byte-identical. Phase 3 of docs/superpowers/specs/2026-05-20-architecture-refactor-roadmap-design.md Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Moves GraphExecutor::pre_dequant_phase4_tensor_registry_ (~378 LOC) from src/exec/executor_pre_dequant.cu to a new TU src/exec/pre_dequant_phase4_tensor_registry.cu. Declaration in src/exec/executor.h unchanged. Function body byte-identical. Phase 3 of docs/superpowers/specs/2026-05-20-architecture-refactor-roadmap-design.md Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Moves GraphExecutor::pre_dequant_phase0_promote_nvfp4_sidecars_ (~230 LOC) and GraphExecutor::pre_dequant_phase0b_register_cutlass_nvfp4_ (~93 LOC) to a combined file src/exec/pre_dequant_phase0_nvfp4_loader.cu. Both phases are NVFP4 loader-side concerns that run consecutively; colocating them keeps related logic in one place. Declarations unchanged, bodies byte-identical. Phase 3 of docs/superpowers/specs/2026-05-20-architecture-refactor-roadmap-design.md Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2 tasks
4 tasks
Owner
Author
|
Superseded by #330 — rebased onto current main since this branch couldn't be force-pushed via the harness. |
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
Move `GraphExecutor::pre_dequant_phase0_promote_nvfp4_sidecars_` + `GraphExecutor::pre_dequant_phase0b_register_cutlass_nvfp4_` to a combined TU `src/exec/pre_dequant_phase0_nvfp4_loader.cu`. Both phases are NVFP4 loader-side concerns running consecutively — colocated by design.
Stacking
Stack chain: #301 → #302 → #303 → #304 → this. Rebase as parents merge.
LOC
`executor_pre_dequant.cu` 1891 → 1567 (-324). New file 350 LOC.
Test plan
Phase 3 Task 5 of `docs/superpowers/specs/2026-05-20-architecture-refactor-roadmap-design.md`.
🤖 Generated with Claude Code