add direct ccu alltoall - #71

Closed

ChinaLinZhen wants to merge 44 commits into

mainfrom

codex/ccu-alltoall

Commits on Jul 6, 2026

feat: add direct CCU runtime support
Kur0x
committed
217dbdd
Merge remote-tracking branch 'origin/main' into codex/direct-ccu-rebased
Kur0x
committed
8e63d9e

Commits on Jul 7, 2026

docs: design CCU API isolation
Kur0x
committed
c09eca0
docs: plan CCU API isolation
Kur0x
committed
4b1a7b4
feat: add collective backend options
Kur0x
committed
ac8bc42
feat: route collective backend selection
Kur0x
committed
3015448
fix: enforce forced collective backend dispatch
Kur0x
committed
f9f17c2
refactor: remove public direct CCU API
Kur0x
committed
411f565
refactor: introduce internal CCU backend
Kur0x
committed
f62c4f8
fix: preserve CCU backend internals
Kur0x
committed
775ee7c
fix: migrate CCU lower-layer source guards
Kur0x
committed
84857b0
refactor: split CCU backend internals
Kur0x
committed
130ee97
fix: update CCU split source guards
Kur0x
committed
b4de639
fix: guard CCU diagnostic internals
Kur0x
committed
fbb4468
feat: connect CCU collective backend dispatch
Kur0x
committed
59e84d0

Commits on Jul 8, 2026

Merge pull request #54 from LingquLab/codex/ccu-api-isolation-impl
ChinaLinZhen
authored
f0fe731
fix(ccu): apply P2P direct copy smoke defaults
linzhen 00654177
committed
7267da2
fix(ccu): adapt smoke probe to internal backend
linzhen 00654177
committed
8b5a414

Commits on Jul 10, 2026

fix(ccu): harden direct runtime lifecycle
linzhen 00654177
committed
ca88fd8
docs: add ccu signal wait design
linzhen 00654177
committed
8bc1207

Commits on Jul 13, 2026

feat(ccu): add direct signal wait backend
linzhen 00654177
committed
83f937c

Commits on Jul 14, 2026

fix(ccu): initialize signal wait source cke
linzhen 00654177
committed
81f2210
docs: add ccu alltoall mvp design
linzhen 00654177
committed
d192cbb

Commits on Jul 15, 2026

feat(ccu): add host-phased direct alltoall MVP
linzhen 00654177
committed
b95c63b

Commits on Jul 16, 2026

feat(ccu): harden direct alltoall validation
linzhen 00654177
committed
82cb4eb

Commits on Jul 17, 2026

ccu alltoall bidirection loop0 error loop1-63 success
linzhen 00654177
committed
2dc0f26

Commits on Jul 20, 2026

fix(ccu): complete direct all-to-all synchronization
GoodTimeLZ
committed
8fabf18
feat(ccu): support repeated direct all-to-all submissions
GoodTimeLZ
committed
b435f2a
docs(ccu): design four-rank all-to-all mesh
GoodTimeLZ
committed
bc50d9b
docs(ccu): retarget four-rank validation server
GoodTimeLZ
committed
c86e318
docs(ccu): plan four-rank all-to-all mesh
GoodTimeLZ
committed
c035759
feat(ccu): encode local memory transfers
GoodTimeLZ
committed
3e97f94
feat(ccu): build four-rank all-to-all mesh mission
GoodTimeLZ
committed
88104c1
feat(ccu): package four-rank all-to-all resources
GoodTimeLZ
committed
b65444a
feat(ccu): plan four-rank all-to-all endpoints
GoodTimeLZ
committed
043da13
test(ccu): exercise repeated four-rank all-to-all
GoodTimeLZ
committed
1acceec
test(ccu): run all-to-all smoke across four ranks
GoodTimeLZ
committed
5b41bbb

Commits on Jul 23, 2026

feat(ccu): enable 4-rank all-to-all 128KB
GoodTimeLZ
committed
b9d42d2

Commits on Jul 27, 2026

Merge remote-tracking branch 'origin/main' into direct-ccu-rebased
GoodTimeLZ
committed
abe4bd3
feat(ccu): generalize direct all-to-all mesh
GoodTimeLZ
committed
01d358d

Commits on Jul 28, 2026

docs: remove internal CCU notes from PR
GoodTimeLZ
committed
0464996
docs: remove CCU implementation plans from PR
GoodTimeLZ
committed
ce9951b
docs(ccu): update all-to-all runtime design
GoodTimeLZ
committed
0c385b4
test(ccu): isolate long-mission loop timing
GoodTimeLZ
committed
2c4edec