datadeps: cache hierarchical schedules; fix similar; fix JuMPExt import - #722

Draft

riteshbhirud wants to merge 46 commits into

JuliaParallel:jps/riteshsc26from

riteshbhirud:fix-eft-runtime-tuple-fargs

Commits on Jul 17, 2026

datadeps: fix _eft_runtime_ns view() on Tuple fargs
riteshbhirud
committed
00e67b0
datadeps: cache hierarchical schedules; fix similar; fix JuMPExt import
riteshbhirud
committed
80df355

Commits on Jul 18, 2026

datadeps: add wall-clock time_limit_sec to IG, SA, OptimizingScheduler
riteshbhirud
committed
53b61ae
bench: fix TimedScheduler under hierarchical partitioning
riteshbhirud
committed
029e9c1

Commits on Jul 19, 2026

Commits on Jul 20, 2026

bench: CPU+GPU heterogeneous support and Sinnen-Sousa random DAG workload

riteshbhirud
and
claude
committed
d39b1b5
bench: tile ownership restricted to CPU procs; task-time HtoD path only

riteshbhirud
and
claude
committed
c6b7b74
bench: GPU-forced warmup pass to prime cost model before Greedy runs

riteshbhirud
and
claude
committed
b8afecb
datadeps: GC.@preserve from_raw/to_raw in RemainderAliasing move!

riteshbhirud
and
claude
committed
58a8941
datadeps+bench: instrumentation for cross-GPU unsafe_free! + single-GPU safety net

riteshbhirud
and
claude
committed
3a7bcb2
CUDAExt+ROCExt: synchronize device before unsafe_free! to fix cross-GPU move! race

riteshbhirud
and
claude
committed
9229585
datadeps: fix free-task syncdep gap in hierarchical partitioning

Bhirud R
and
claude
committed
de008ce
bench: bound MetricsTracker cache to prevent write-path degradation

Bhirud R
and
claude
committed
383c151
MetricsTracker/Sch: Optimize metric insertion
jpsamaroo
committed
845e0ef
Sch: Optimize steal allocations
jpsamaroo
committed
9a57097
Sch: Optimize processor queue kick
jpsamaroo
committed
8d198c2
Sch: Optimize do_task move
jpsamaroo
committed
b957755
datadeps: K-slot capacity-aware EFT for multi-stream processors

Bhirud R
and
claude
committed
23fb03d
Merge branch 'fix-eft-runtime-tuple-fargs' of https://github.com/riteshbhirud/Dagger.jl into fix-eft-runtime-tuple-fargs
Bhirud R
committed
e5cbcc3
bench: scope datadeps regions to all procs so GPUs reach the scheduler

Bhirud R
and
claude
committed
0601649
bench: add tiled LU workload

Bhirud R
and
claude
committed
7556885
datadeps: fix EFT signature mismatch that flattened the cost model

Bhirud R
and
claude
committed
4289c3b
datadeps: translate sig for GPU cost lookup so post-dispatch metrics reach the scheduler

Bhirud R
and
claude
committed
e453fdd
metrics: make cache bound runtime-settable via metrics_cache_max_tasks!

Bhirud R
and
claude
committed
e490572
datadeps: record FromSpace/ToSpace/MoveSize metrics at move! execution site

Bhirud R
and
claude
committed
ed80723
datadeps: median-of-per-move-rates aggregation in metrics_lookup_move_rate

Bhirud R
and
claude
committed
2f3461f

Commits on Jul 21, 2026

datadeps: stochastic reconstruction for IteratedGreedy (was deterministic == Greedy)

Bhirud R
and
claude
committed
6382a52
datadeps: faithful Ruiz-Stützle acceptance + destruction for IteratedGreedy

Bhirud R
and
claude
committed
4ab6d4e
datadeps: Orsila SA tuning — k=2 temperature range + ECP critical-path move

Bhirud R
and
claude
committed
6940851
datadeps: producer-finish term in greedy data-ready (critical-path awareness)

Bhirud R
and
claude
committed
389f222
bench: paper Table-1 grid harness + MILP-objective instrumentation

Bhirud R
and
claude
committed
a39797b
bench: Tuolumne MI300a one-shot benchmark (M1/M2/M4 + optgap + cache demo)

Bhirud R
and
claude
committed
673bff9

Commits on Jul 31, 2026

bench: Tuolumne partitioned run via REGIME/WORKLOAD filter env vars

riteshbhirud
and
claude
committed
fdcc4d7