ELPH WF CACHE PREFILL: Difference between revisions
No edit summary |
No edit summary |
||
| (6 intermediate revisions by 2 users not shown) | |||
| Line 1: | Line 1: | ||
{{DISPLAYTITLE:ELPH_WF_CACHE_PREFILL}} | {{DISPLAYTITLE:ELPH_WF_CACHE_PREFILL}} | ||
{{TAGDEF|ELPH_WF_CACHE_PREFILL| | {{TAGDEF|ELPH_WF_CACHE_PREFILL|logical|.TRUE.}} | ||
Description: | Description: Pre-fills the wavefunction cache before the main electron-phonon loop begins. | ||
{{Available|6.5.0}} | {{Available|6.5.0}} | ||
---- | ---- | ||
When | When computing electron-phonon matrix elements, VASP caches wavefunctions fetched from remote MPI ranks (see {{TAG|ELPH_WF_CACHE_MB}}). With <code>ELPH_WF_CACHE_PREFILL = .TRUE.</code> (the default), all required wavefunctions are gathered into the cache in a single communication phase before the main loop starts. This means the loop itself runs with little or no inter-rank MPI traffic. | ||
This | |||
Setting <code>ELPH_WF_CACHE_PREFILL = .FALSE.</code> disables pre-filling; wavefunctions are then fetched on demand during the loop. This reduces the upfront communication cost but increases total MPI traffic and is generally slower. | |||
==Related tags and articles== | ==Related tags and articles== | ||
* {{TAG|ELPH_WF_CACHE_MB}} | |||
* {{TAG|ELPH_WF_REDISTRIBUTE}} | * {{TAG|ELPH_WF_REDISTRIBUTE}} | ||
* {{TAG| | * {{TAG|ELPH_WF_COMM_OPT}} | ||
[[Category:INCAR tag]][[Category:Electron-phonon_interactions]] | |||
Latest revision as of 13:35, 18 June 2026
ELPH_WF_CACHE_PREFILL = logical
Default: ELPH_WF_CACHE_PREFILL = .TRUE.
Description: Pre-fills the wavefunction cache before the main electron-phonon loop begins.
| Mind: Available as of VASP 6.5.0 |
When computing electron-phonon matrix elements, VASP caches wavefunctions fetched from remote MPI ranks (see ELPH_WF_CACHE_MB). With ELPH_WF_CACHE_PREFILL = .TRUE. (the default), all required wavefunctions are gathered into the cache in a single communication phase before the main loop starts. This means the loop itself runs with little or no inter-rank MPI traffic.
Setting ELPH_WF_CACHE_PREFILL = .FALSE. disables pre-filling; wavefunctions are then fetched on demand during the loop. This reduces the upfront communication cost but increases total MPI traffic and is generally slower.