`Multistep(carrier_frame="scalar")` is refused on device builds, and costs more than it should on cheap sources
I maintainer di solito rispondono entro 1 giorno
Nessuno ha ancora preso questa issue.
Valutazione
- Difficoltà
- 5/5
- Tempo stimato
- Più di una settimana
- Idoneità per principianti
- 12/100
- Tipo di issue
- Funzionalità
- Chiarezza
- Abbastanza chiara
- Stato di attività
- Attiva
- Stack tecnologico
- python
- Ambito
- backend, performance
Direzione di ricerca
Start with native/src/solprop_native/backend.py around the device-build refusal of carrier_frame="scalar" and tests/backends/test_native_march_admission.py where the test is skipped on devices. The GPU parity tests are in tests/backends/test_carrier_frame.py, and the first step is to read SEP 39 against the current march before running anything. Done means both vendors either pass the budgets and the refusal is lifted, or the recorded reason is kept, plus a before/after Kerr-only ratio on CPU.
Scritto dal modello di indicizzazione a partire dal testo della issue.
Descrizione
What happens
Multistep(carrier_frame="scalar") (SEP 35) removes the pulse's mean phase velocity from the Adams march's history. Two gaps remain.
Device builds refuse it by name. backend.py passes no delivered policy on a device, the table says so, and the admission test skips there. The plan was a short GPU segment on one NVIDIA and one AMD device, running the compiled parity tests and the ten-step timing. No such run is recorded. Since then SEP 39 moved the chord's upload, shift and rotations into the native engine, so that plan has to be read against the current march. Read, not run.
It costs more than it should on a cheap source, one that is mostly Kerr. These are SEP 35's numbers, on native and on CPU:
- Ten-step timing: 1.06 per step on a Kerr-only source, 1.01 with Raman, the source of the roadmap's "4 to 7 %".
- At matched worst-plane error, which is larger: 1.20 to 1.33 times
"off"on helium case A with Kerr alone, and 2.06 to 2.60 times the best other method, IP-DOPRI5, onhe_rdw_averaged. - Forming
exp(x - sigma)asexp(x) exp(-sigma)took native's per-step ratio from 1.076 to 1.057.
Why it matters
A user on a GPU cannot use a method SEP 35 found fastest on ctc_reduced at matched error. On a cheap source the cost counts against the policy's 1.1 times criterion wherever it saves few calls.
To do
- Run the parity tests that use
carrier_increment_budgetand the ten-step timing on one NVIDIA and one AMD data-centre GPU. It can share #353's rental. - If they pass, lift the refusal on device builds, unskip the admission test, and update the table and the guide. If they do not, record why and keep the refusal.
- On CPU, fuse the rotation into the defect reducers' divided-difference rows, the one lever SEP 35 names. Unchecked whether it would close the gap.
What must not change: carrier_frame="off" and the default, on any build. #97 holds other carrier-frame follow-ups, and #77 the frame's choice beside third-harmonic generation.
Done when
Device builds run "scalar" within the same budgets on both vendors, or refuse it with the recorded reason. On native CPU the ten-step Kerr-only ratio is measured before and after the fusion on one idle machine, and the fusion is kept only if it lowers the ratio with the same steps.
Checked against commit 8a9f764. Takes over the roadmap's entry "The Adams carrier frame on devices and on cheap sources".
- Lingua principale
- Python
- Stelle
- 3
- Fork
- 3
- Merge medio
- 6h 28m
- PR unite (30g)
- 126
Preparare l'ambiente
- Nessun Dockerfile né file Docker Compose
- Nessun modello di pull request
- Leggi la guida per i contributori
Come iniziare
- Leggi tutta la issue e poi la guida ai contributi del progetto.
- Commenta sulla issue per dire che te ne occupi tu — evita che due persone facciano lo stesso lavoro.
- Fai un fork del repository e lavora su un branch.
- Apri una pull request che faccia riferimento al numero della issue.
Altre issue di LupoLab/solprop
-
A recipe missing a required keyword raises a bare TypeError from build, modes and preview_inputApertaarea: gui bug priority: medium
Difficoltà 2/5 1-3 ore Idoneità per principianti 72/100
I maintainer di solito rispondono entro 1 giorno
-
The TDSE steady-current certificate can refuse a resolved taper on a band far above its frequenciesApertaarea: tdse enhancement priority: low
Difficoltà 2/5 1-3 ore Idoneità per principianti 60/100
I maintainer di solito rispondono entro 1 giorno
-
The band-edge term that floors comparisons between windows is recorded only in an archived SEPApertaarea: boundaries documentation priority: low
Difficoltà 2/5 Mezza giornata Idoneità per principianti 62/100
I maintainer di solito rispondono entro 1 giorno
-
area: backends bug priority: low
Difficoltà 2/5 Mezza giornata Idoneità per principianti 66/100
I maintainer di solito rispondono entro 1 giorno
-
area: interfaces bug
Difficoltà 4/5 3-5 giorni Idoneità per principianti 55/100
I maintainer di solito rispondono entro 1 giorno
Tutte le issue di LupoLab/solprop
Issue simili
-
Difficoltà 2/5 1-3 ore Idoneità per principianti 76/100
autowarefoundation/autoware_carla_scenario#1485 ·
I maintainer di solito rispondono entro 1 giorno
-
needs-area-label
Difficoltà 2/5 1-3 ore Idoneità per principianti 66/100
I maintainer di solito rispondono entro 2 giorni
-
enhancement good first issue
Difficoltà 2/5 1-3 ore Idoneità per principianti 84/100
Official-Space-AI/OpenPhysGraspGen#114 ·
I maintainer di solito rispondono entro 2 giorni
-
changelog investigate
Difficoltà 2/5 1-3 ore Idoneità per principianti 62/100
ramnes/notion-sdk-py#441 ·
-
copy_tree splits same-file errors into charactersForse già presa @Gambit-Checkmate l’ha presa oggi. Aperta
Difficoltà 2/5 1-3 ore Idoneità per principianti 68/100
I maintainer di solito rispondono entro 14 giorni