Antonin Portelli portelli
0 Followers · 0 Following
  • Joined on 2017-10-25
Repositories
6
 Projects Packages Public Activity Starred Repositories
1
portelli synced commits to refs/heads/develop at portelli/Grid from mirror 2026-08-28 21:39:35 +01:00
f367bc8bce Solving memory pressure in NRHS=12 solver
63c2cdb712 Debug ulimit as core files driving me crazy
9b4c388585 Update for latt parameter; claude's predeliction for using shell variables rather than "grid" style command line args is getting annoying
576cebd56a Job for debug
86da22fb78 Debug page fault
Compare 5 commits »
portelli synced commits to refs/heads/develop at portelli/Grid from mirror 2026-08-28 13:29:35 +01:00
27e32b1633 Stop mucking about with THREAD_MULTIPLE
portelli synced commits to refs/heads/develop at portelli/Grid from mirror 2026-08-28 05:19:35 +01:00
2264afda84 Allreduce benchmarking
792d2a8258 Masked/orthog dim CartesianRingAllReduce and CartesianRingAllGather
7abc19dc03 Allow subsets of dimensions for spatial only sums
Compare 3 commits »
portelli synced commits to refs/heads/develop at portelli/Grid from mirror 2026-08-27 21:09:36 +01:00
074bddd48b Probing environment effects on dense inversion time
6cea81bc2c More probing why test harness vs. slate was a little faster running Grid's dense inverse than normal environment
68b9fa86d7 Faster leaf inverse
d44964d923 Faster dense inverse, I hope
52998539bc Bigger leaf operations should spead up distributed inverse
Compare 5 commits »
portelli synced commits to refs/heads/develop at portelli/Grid from mirror 2026-08-27 12:59:36 +01:00
0979371c45 SUMMA telemetry: ring time histogram by message size
96071aa492 Dense allgather: stage/scatter x by global-site order, not rank-major block
Compare 2 commits »
portelli synced commits to refs/heads/develop at portelli/Grid from mirror 2026-08-27 04:49:36 +01:00
e5f1553232 Better reduction
06c6011229 Timestamp improvement
cbe97cc40d Smoother search improvement
dc1ae3185a Ring allgather too. Let's nail the Dense CoarseCoarse
8e4a304626 Less aggressive L5
Compare 5 commits »
portelli synced commits to refs/heads/develop at portelli/Grid from mirror 2026-08-26 20:39:36 +01:00
5f8f396e9b More jobs for testing Mgrid
841e59e8c0 GCR coefficient recording and playback
b55ca3d50d Consolidate polynomial smoothers + logging time breakdown in Dense Coarsest invers
Compare 3 commits »
portelli synced commits to refs/heads/develop at portelli/Grid from mirror 2026-08-26 12:29:36 +01:00
d7c9076228 Power method added
portelli synced commits to refs/heads/develop at portelli/Grid from mirror 2026-08-26 04:19:37 +01:00
7a292bc4c6 Slate compile fix
9213f6e533 Improvements for mulktigrid. + Schur setup improvements
5685203b7b Frontier jobs
22dda4a8ed Update
eb22145f49 Updated slate
Compare 5 commits »
portelli synced commits to refs/heads/develop at portelli/Grid from mirror 2026-08-25 20:09:36 +01:00
229f4808c7 Ring all reduce first cut
167753aa3f Adding ring all reduce for large vectors
dbbd5dba35 Potentially faster GCR linalg
Compare 3 commits »
portelli synced commits to refs/heads/develop at portelli/Grid from mirror 2026-08-25 03:49:35 +01:00
33757c3c94 FOrmat warn avoidance
7438e56f94 More controls
47416230b1 Verbose on finalize improvement
6a854ebd0d Print trace for teh GCR step
8c6bcddbab Coefficent of GCR printing for poly recording
Compare 5 commits »
portelli synced commits to refs/heads/develop at portelli/Grid from mirror 2026-08-24 19:39:35 +01:00
7792d7a1a7 Verbose prettier
8261dc39d9 More tracing
Compare 2 commits »
portelli synced commits to refs/heads/develop at portelli/Grid from mirror 2026-08-24 11:29:35 +01:00
9ed7049736 COMpile fixes
portelli synced commits to refs/heads/develop at portelli/Grid from mirror 2026-08-24 03:19:35 +01:00
5a70066e5b Claude build fixeds
6be8b478c3 Claude 2D block cyclic inverse work
34a1220035 Claude 2D block cyclic inverse
Compare 3 commits »
portelli synced commits to refs/heads/develop at portelli/Grid from mirror 2026-08-23 19:09:35 +01:00
c97cfe9f5b This fails
01858754ed Unhappy test looking for work aroudn to avoid allgather
c388dd98d6 More verbose phasing on finish
38344e035f Don't print Finalize until made collective
Compare 4 commits »
portelli synced commits to refs/heads/develop at portelli/Grid from mirror 2026-08-23 10:59:36 +01:00
14d2f2c06e Improvements
portelli synced commits to refs/heads/develop at portelli/Grid from mirror 2026-08-23 02:49:36 +01:00
06fa2fc160 use shm buffers
a9c22d787a Benchmark comms with the exact sequence in Stencil as it is now.
Compare 2 commits »
portelli synced commits to refs/heads/develop at portelli/Grid from mirror 2026-08-22 18:39:36 +01:00
1e287b6b30 Changes to dense inverse AND improved timing on comms
152e9ecf3b Bandwidth measures inside stencil
Compare 2 commits »
portelli synced commits to refs/heads/develop at portelli/Grid from mirror 2026-08-22 10:29:35 +01:00
2e0c19c915 Allgather test
portelli synced commits to refs/heads/develop at portelli/Grid from mirror 2026-08-21 18:09:36 +01:00
4d6931620b Gather based distributed dense inverse