Pull requests · pytorch/helion

[cutedsl] Consolidate fragment configuration and ownership checks

#3848yushangdi opened

Oct 8, 2026

Contributor

[cutedsl] Schedule pure producer prefixes on one warp with exact publications

#3847yushangdi opened

Oct 8, 2026

Contributor

[cutedsl] Transport immutable typed and derived snapshots through uniform frames

#3846yushangdi opened

Oct 8, 2026

Contributor

[cutedsl] Support write-only region outputs and bounded singleton views

#3845yushangdi opened

Oct 8, 2026

Contributor

[cutedsl] Compose uniform regions with current-call discovery and indexed reads

#3844yushangdi opened

Oct 8, 2026

Contributor

[cutedsl] Support fresh local histogram epochs in uniform branches and loops

#3843yushangdi opened

Oct 8, 2026

Contributor

[cutedsl] Export completed scans and preserve typed immutable loop snapshots

#3842yushangdi opened

Oct 8, 2026

Contributor

[cutedsl] Preserve uniform producer regions and resident loop storage epochs

#3841yushangdi opened

Oct 8, 2026

Contributor

[cutedsl] Lower bounded indexed gathers and hierarchical scan consumers

#3840yushangdi opened

Oct 8, 2026

Contributor

[cutedsl] Keep pure register producers across published scalar consumers

#3839yushangdi opened

Oct 8, 2026

Contributor

[cutedsl] Map packet loads and reductions to explicit CTA owners

#3838yushangdi opened

Oct 8, 2026

Contributor

[cutedsl] Fuse atomic consumers and hoist exact integer reductions

#3837yushangdi opened

Oct 8, 2026

Contributor

[cutedsl] Reuse published scalars and elide proved zero atomics

#3836yushangdi opened

Oct 8, 2026

Contributor

[cutedsl] Snapshot immutable fragment values in owner registers

#3835yushangdi opened

Oct 8, 2026

Contributor

[cutedsl] Keep local atomic tickets and consumers in owner registers

#3834yushangdi opened

Oct 8, 2026

Contributor

[cutedsl] Reuse fragment producers with warp scans and atomic aggregation

#3833yushangdi opened

Oct 8, 2026

Contributor

[cutedsl] Publish atomic results through ordered terminal consumers

#3832yushangdi opened

Oct 8, 2026

Contributor

[cutedsl] Support CTA-owned local atomics and readonly register loads

#3831yushangdi opened

Oct 8, 2026

Contributor

[cutedsl] Plan captured fragment storage

#3830yushangdi opened

Oct 8, 2026

Contributor

[cutedsl] Lower typed tiled fragment expressions and reductions

#3829yushangdi opened

Oct 8, 2026

Contributor

[cutedsl] Preserve ordered fragment phases

#3828yushangdi opened

Oct 8, 2026

Contributor

[cutedsl] Preserve bound runtime tensor disjointness proofs

#3827yushangdi opened

Oct 8, 2026

Contributor

[cutedsl] Support computed fragments and restore verification contracts

#3826yushangdi opened

Oct 8, 2026

Contributor

Codex/attention rebased final 20261007

#3825yushangdi opened

Oct 8, 2026

Contributor

[CuTe] Add GB300 pretuned sampling and variable-length selection kernels

#3824yushangdi opened

Oct 8, 2026

Contributor

Navigation Menu

Repository navigation

All pull requests

[cutedsl] Consolidate fragment configuration and ownership checks

[cutedsl] Schedule pure producer prefixes on one warp with exact publications

[cutedsl] Transport immutable typed and derived snapshots through uniform frames

[cutedsl] Support write-only region outputs and bounded singleton views

[cutedsl] Compose uniform regions with current-call discovery and indexed reads

[cutedsl] Support fresh local histogram epochs in uniform branches and loops

[cutedsl] Export completed scans and preserve typed immutable loop snapshots

[cutedsl] Preserve uniform producer regions and resident loop storage epochs

[cutedsl] Lower bounded indexed gathers and hierarchical scan consumers

[cutedsl] Keep pure register producers across published scalar consumers

[cutedsl] Map packet loads and reductions to explicit CTA owners

[cutedsl] Fuse atomic consumers and hoist exact integer reductions

[cutedsl] Reuse published scalars and elide proved zero atomics

[cutedsl] Snapshot immutable fragment values in owner registers

[cutedsl] Keep local atomic tickets and consumers in owner registers

[cutedsl] Reuse fragment producers with warp scans and atomic aggregation

[cutedsl] Publish atomic results through ordered terminal consumers

[cutedsl] Support CTA-owned local atomics and readonly register loads

[cutedsl] Plan captured fragment storage

[cutedsl] Lower typed tiled fragment expressions and reductions

[cutedsl] Preserve ordered fragment phases

[cutedsl] Preserve bound runtime tensor disjointness proofs

[cutedsl] Support computed fragments and restore verification contracts

Codex/attention rebased final 20261007

[CuTe] Add GB300 pretuned sampling and variable-length selection kernels