MFC
Exascale flow solver
Loading...
Searching...
No Matches
m_phase_timing Module Reference

Per-phase wall-time budget of the AMR step, reported when rank_time_wrt is on. Brackets are disjoint (regrid sub-phases nest inside regrid) and drain the device first, so a phase's time includes the GPU work it launched. More...

Functions/Subroutines

impure subroutine, public s_phase_tic (id)
impure subroutine, public s_phase_toc (id)
impure subroutine, public s_phase_step_tic ()
impure subroutine, public s_phase_step_toc ()
impure subroutine, public s_phase_report ()
 Mean and max over ranks per phase, share of the step-loop wall, imbalance (max/mean) and calls; the residual is the wall minus the sum of the means.
impure subroutine, public s_mem_note (id, bytes)
 Add bytes to group id of the end-of-run AMR memory table.
impure subroutine, public s_mem_note_fields (id, f)
 s_mem_note for the fields of f that are allocated.

Variables

integer, parameter, public ph_halo = 1
 coarse cons halo exchange (once per stage)
integer, parameter, public ph_gather = 2
 fill waves: coarse-patch gather into every block
integer, parameter, public ph_gfill = 3
 ghost prolongation from the gathered patch
integer, parameter, public ph_seam = 4
 fine-fine seam halo
integer, parameter, public ph_rhs = 5
 fine-block RHS
integer, parameter, public ph_rk = 6
 fine-block RK update + IB
integer, parameter, public ph_reflux = 7
 level-1 reflux (faces wave + apply)
integer, parameter, public ph_regrid = 8
 regrid
integer, parameter, public ph_coarse = 9
 coarse (base-grid) solver work
integer, parameter, public ph_swap = 10
 per-batch grid-state swap and restore
integer, parameter, public ph_restr = 11
 post-stage restrict + reflux-to-parent fold
integer, parameter, public ph_lerp = 12
 subcycled coarse time interpolation for the level-1 fills
integer, parameter ph_ntop = 12
 the disjoint brackets above; the regrid sub-phases below nest in PH_REGRID
integer, parameter, public ph_rg_halo = 13
 coarse halo + host copy for tagging
integer, parameter, public ph_rg_tag = 14
 tagging
integer, parameter, public ph_rg_clus = 15
 clustering, shaping, nesting and box checks
integer, parameter, public ph_rg_own = 16
 stash and owner assignment
integer, parameter, public ph_rg_mig = 17
 fine-state migration
integer, parameter, public ph_rg_build = 18
 new-block build (fills, prolongation, overlap copy)
integer, parameter, public ph_rg_rec = 19
 slot reconcile, IB setup, seam check
integer, parameter ph_n = 19
character(len=8), dimension(ph_n), parameter ph_name = [character(len=8)::'halo','gather', 'gfill', 'seam', 'rhs', 'rk', 'reflux', 'regrid', 'coarse', 'swap', 'restr', 'lerp', ' rg.halo', ' rg.tag', ' rg.clus', ' rg.own', ' rg.mig', ' rg.build', ' rg.rec']
real(wp), dimension(ph_n) acc = 0._wp
real(wp) wall = 0._wp
 wall is the accumulated step-loop time the phases are budgeted against
integer(8), dimension(ph_n) ncall = 0
integer(8), dimension(ph_n) tic_c = 0
integer(8) wall_c = 0
integer, dimension(ph_n) depth = 0
 a re-entered bracket (shared routine) measures only its outermost pair
integer, parameter, public mem_store = 1
 AMR allocations by group, noted by each owner before it frees them (s_mem_note) and printed with the phase table.
integer, parameter, public mem_scratch = 2
 batched-advance scratch (prim, rhs) and the block bridge
integer, parameter, public mem_patch = 3
 coarse patch pool and its subcycling time cache
integer, parameter, public mem_reg = 4
 flux registers
integer, parameter, public mem_ib = 5
 fine IB ghost-point park
integer, parameter, public mem_ibhost = 6
 per-slot fine IB marker copies (host only)
integer, parameter mem_n = 6
character(len=8), dimension(mem_n+1), parameter mem_name = [character(len=8)::'store','scratch', 'patch', 'register', 'ib park', 'ib marks', 'total']
real(wp), dimension(mem_n) mem_mib = 0._wp

Detailed Description

Per-phase wall-time budget of the AMR step, reported when rank_time_wrt is on. Brackets are disjoint (regrid sub-phases nest inside regrid) and drain the device first, so a phase's time includes the GPU work it launched.

Function/Subroutine Documentation

◆ s_mem_note()

impure subroutine, public m_phase_timing::s_mem_note ( integer, intent(in) id,
integer(kind=8), intent(in) bytes )

Add bytes to group id of the end-of-run AMR memory table.

Definition at line 535 of file m_phase_timing.fpp.f90.

Here is the caller graph for this function:

◆ s_mem_note_fields()

impure subroutine, public m_phase_timing::s_mem_note_fields ( integer, intent(in) id,
type(scalar_field), dimension(:), intent(in) f )

s_mem_note for the fields of f that are allocated.

Definition at line 545 of file m_phase_timing.fpp.f90.

Here is the call graph for this function:

◆ s_phase_report()

impure subroutine, public m_phase_timing::s_phase_report

Mean and max over ranks per phase, share of the step-loop wall, imbalance (max/mean) and calls; the residual is the wall minus the sum of the means.

Definition at line 494 of file m_phase_timing.fpp.f90.

Here is the call graph for this function:

◆ s_phase_step_tic()

impure subroutine, public m_phase_timing::s_phase_step_tic

Definition at line 476 of file m_phase_timing.fpp.f90.

Here is the caller graph for this function:

◆ s_phase_step_toc()

impure subroutine, public m_phase_timing::s_phase_step_toc

Definition at line 482 of file m_phase_timing.fpp.f90.

Here is the caller graph for this function:

◆ s_phase_tic()

impure subroutine, public m_phase_timing::s_phase_tic ( integer, intent(in) id)

Definition at line 427 of file m_phase_timing.fpp.f90.

Here is the caller graph for this function:

◆ s_phase_toc()

impure subroutine, public m_phase_timing::s_phase_toc ( integer, intent(in) id)

Definition at line 450 of file m_phase_timing.fpp.f90.

Here is the caller graph for this function:

Variable Documentation

◆ acc

real(wp), dimension(ph_n) m_phase_timing::acc = 0._wp
private

Definition at line 409 of file m_phase_timing.fpp.f90.

◆ depth

integer, dimension(ph_n) m_phase_timing::depth = 0
private

a re-entered bracket (shared routine) measures only its outermost pair

Definition at line 411 of file m_phase_timing.fpp.f90.

◆ mem_ib

integer, parameter, public m_phase_timing::mem_ib = 5

fine IB ghost-point park

Definition at line 418 of file m_phase_timing.fpp.f90.

◆ mem_ibhost

integer, parameter, public m_phase_timing::mem_ibhost = 6

per-slot fine IB marker copies (host only)

Definition at line 419 of file m_phase_timing.fpp.f90.

◆ mem_mib

real(wp), dimension(mem_n) m_phase_timing::mem_mib = 0._wp
private

Definition at line 423 of file m_phase_timing.fpp.f90.

◆ mem_n

integer, parameter m_phase_timing::mem_n = 6
private

Definition at line 420 of file m_phase_timing.fpp.f90.

◆ mem_name

character(len=8), dimension(mem_n + 1), parameter m_phase_timing::mem_name = [character(len=8)::'store','scratch', 'patch', 'register', 'ib park', 'ib marks', 'total']
private

Definition at line 421 of file m_phase_timing.fpp.f90.

◆ mem_patch

integer, parameter, public m_phase_timing::mem_patch = 3

coarse patch pool and its subcycling time cache

Definition at line 416 of file m_phase_timing.fpp.f90.

◆ mem_reg

integer, parameter, public m_phase_timing::mem_reg = 4

flux registers

Definition at line 417 of file m_phase_timing.fpp.f90.

◆ mem_scratch

integer, parameter, public m_phase_timing::mem_scratch = 2

batched-advance scratch (prim, rhs) and the block bridge

Definition at line 415 of file m_phase_timing.fpp.f90.

◆ mem_store

integer, parameter, public m_phase_timing::mem_store = 1

AMR allocations by group, noted by each owner before it frees them (s_mem_note) and printed with the phase table.

flat block store (cons + RK stage copy)

Definition at line 414 of file m_phase_timing.fpp.f90.

◆ ncall

integer(8), dimension(ph_n) m_phase_timing::ncall = 0
private

Definition at line 410 of file m_phase_timing.fpp.f90.

◆ ph_coarse

integer, parameter, public m_phase_timing::ph_coarse = 9

coarse (base-grid) solver work

Definition at line 392 of file m_phase_timing.fpp.f90.

◆ ph_gather

integer, parameter, public m_phase_timing::ph_gather = 2

fill waves: coarse-patch gather into every block

Definition at line 385 of file m_phase_timing.fpp.f90.

◆ ph_gfill

integer, parameter, public m_phase_timing::ph_gfill = 3

ghost prolongation from the gathered patch

Definition at line 386 of file m_phase_timing.fpp.f90.

◆ ph_halo

integer, parameter, public m_phase_timing::ph_halo = 1

coarse cons halo exchange (once per stage)

Definition at line 384 of file m_phase_timing.fpp.f90.

◆ ph_lerp

integer, parameter, public m_phase_timing::ph_lerp = 12

subcycled coarse time interpolation for the level-1 fills

Definition at line 395 of file m_phase_timing.fpp.f90.

◆ ph_n

integer, parameter m_phase_timing::ph_n = 19
private

Definition at line 404 of file m_phase_timing.fpp.f90.

◆ ph_name

character(len=8), dimension(ph_n), parameter m_phase_timing::ph_name = [character(len=8)::'halo','gather', 'gfill', 'seam', 'rhs', 'rk', 'reflux', 'regrid', 'coarse', 'swap', 'restr', 'lerp', ' rg.halo', ' rg.tag', ' rg.clus', ' rg.own', ' rg.mig', ' rg.build', ' rg.rec']
private

Definition at line 405 of file m_phase_timing.fpp.f90.

◆ ph_ntop

integer, parameter m_phase_timing::ph_ntop = 12
private

the disjoint brackets above; the regrid sub-phases below nest in PH_REGRID

Definition at line 396 of file m_phase_timing.fpp.f90.

◆ ph_reflux

integer, parameter, public m_phase_timing::ph_reflux = 7

level-1 reflux (faces wave + apply)

Definition at line 390 of file m_phase_timing.fpp.f90.

◆ ph_regrid

integer, parameter, public m_phase_timing::ph_regrid = 8

regrid

Definition at line 391 of file m_phase_timing.fpp.f90.

◆ ph_restr

integer, parameter, public m_phase_timing::ph_restr = 11

post-stage restrict + reflux-to-parent fold

Definition at line 394 of file m_phase_timing.fpp.f90.

◆ ph_rg_build

integer, parameter, public m_phase_timing::ph_rg_build = 18

new-block build (fills, prolongation, overlap copy)

Definition at line 402 of file m_phase_timing.fpp.f90.

◆ ph_rg_clus

integer, parameter, public m_phase_timing::ph_rg_clus = 15

clustering, shaping, nesting and box checks

Definition at line 399 of file m_phase_timing.fpp.f90.

◆ ph_rg_halo

integer, parameter, public m_phase_timing::ph_rg_halo = 13

coarse halo + host copy for tagging

Definition at line 397 of file m_phase_timing.fpp.f90.

◆ ph_rg_mig

integer, parameter, public m_phase_timing::ph_rg_mig = 17

fine-state migration

Definition at line 401 of file m_phase_timing.fpp.f90.

◆ ph_rg_own

integer, parameter, public m_phase_timing::ph_rg_own = 16

stash and owner assignment

Definition at line 400 of file m_phase_timing.fpp.f90.

◆ ph_rg_rec

integer, parameter, public m_phase_timing::ph_rg_rec = 19

slot reconcile, IB setup, seam check

Definition at line 403 of file m_phase_timing.fpp.f90.

◆ ph_rg_tag

integer, parameter, public m_phase_timing::ph_rg_tag = 14

tagging

Definition at line 398 of file m_phase_timing.fpp.f90.

◆ ph_rhs

integer, parameter, public m_phase_timing::ph_rhs = 5

fine-block RHS

Definition at line 388 of file m_phase_timing.fpp.f90.

◆ ph_rk

integer, parameter, public m_phase_timing::ph_rk = 6

fine-block RK update + IB

Definition at line 389 of file m_phase_timing.fpp.f90.

◆ ph_seam

integer, parameter, public m_phase_timing::ph_seam = 4

fine-fine seam halo

Definition at line 387 of file m_phase_timing.fpp.f90.

◆ ph_swap

integer, parameter, public m_phase_timing::ph_swap = 10

per-batch grid-state swap and restore

Definition at line 393 of file m_phase_timing.fpp.f90.

◆ tic_c

integer(8), dimension(ph_n) m_phase_timing::tic_c = 0
private

Definition at line 410 of file m_phase_timing.fpp.f90.

◆ wall

real(wp) m_phase_timing::wall = 0._wp
private

wall is the accumulated step-loop time the phases are budgeted against

Definition at line 409 of file m_phase_timing.fpp.f90.

◆ wall_c

integer(8) m_phase_timing::wall_c = 0
private

Definition at line 410 of file m_phase_timing.fpp.f90.