Partial implementation
This commit is contained in:
parent
3edd18c4fd
commit
35b8af8b2d
|
|
@ -517,6 +517,59 @@ pub fn RHS_trad(
|
|||
SAT_characteristics(op, k, y, metrics, boundaries);
|
||||
}
|
||||
|
||||
#[allow(non_snake_case)]
|
||||
pub fn RHS_no_SAT(
|
||||
op: &dyn SbpOperator2d,
|
||||
k: &mut Diff,
|
||||
y: &Field,
|
||||
metrics: &Metrics,
|
||||
tmp: &mut (Field, Field, Field, Field, Field, Field),
|
||||
) {
|
||||
let ehat = &mut tmp.0;
|
||||
let fhat = &mut tmp.1;
|
||||
fluxes((ehat, fhat), y, metrics, &mut tmp.2);
|
||||
let dE = &mut tmp.2;
|
||||
let dF = &mut tmp.3;
|
||||
|
||||
op.diffxi(ehat.rho(), dE.rho_mut());
|
||||
op.diffxi(ehat.rhou(), dE.rhou_mut());
|
||||
op.diffxi(ehat.rhov(), dE.rhov_mut());
|
||||
op.diffxi(ehat.e(), dE.e_mut());
|
||||
|
||||
op.diffeta(fhat.rho(), dF.rho_mut());
|
||||
op.diffeta(fhat.rhou(), dF.rhou_mut());
|
||||
op.diffeta(fhat.rhov(), dF.rhov_mut());
|
||||
op.diffeta(fhat.e(), dF.e_mut());
|
||||
|
||||
if let Some(diss_op) = op.upwind() {
|
||||
let ad_xi = &mut tmp.4;
|
||||
let ad_eta = &mut tmp.5;
|
||||
upwind_dissipation(
|
||||
&*diss_op,
|
||||
(ad_xi, ad_eta),
|
||||
y,
|
||||
metrics,
|
||||
(&mut tmp.0, &mut tmp.1),
|
||||
);
|
||||
|
||||
azip!((out in &mut k.0,
|
||||
eflux in &dE.0,
|
||||
fflux in &dF.0,
|
||||
ad_xi in &ad_xi.0,
|
||||
ad_eta in &ad_eta.0,
|
||||
detj in &metrics.detj().broadcast((4, y.ny(), y.nx())).unwrap()) {
|
||||
*out = (-eflux - fflux + ad_xi + ad_eta)/detj
|
||||
});
|
||||
} else {
|
||||
azip!((out in &mut k.0,
|
||||
eflux in &dE.0,
|
||||
fflux in &dF.0,
|
||||
detj in &metrics.detj().broadcast((4, y.ny(), y.nx())).unwrap()) {
|
||||
*out = (-eflux - fflux )/detj
|
||||
});
|
||||
}
|
||||
}
|
||||
|
||||
#[allow(non_snake_case)]
|
||||
pub fn RHS_upwind(
|
||||
op: &dyn SbpOperator2d,
|
||||
|
|
|
|||
|
|
@ -18,3 +18,4 @@ json5 = "0.3.0"
|
|||
indexmap = { version = "1.5.2", features = ["serde-1"] }
|
||||
argh = "0.1.4"
|
||||
evalexpr = "6.3.0"
|
||||
crossbeam-channel = "0.5.0"
|
||||
|
|
|
|||
|
|
@ -1,8 +1,6 @@
|
|||
use argh::FromArgs;
|
||||
use rayon::prelude::*;
|
||||
|
||||
use euler::eval::Evaluator;
|
||||
use sbp::operators::SbpOperator2d;
|
||||
use sbp::*;
|
||||
|
||||
mod file;
|
||||
|
|
@ -10,207 +8,8 @@ mod input;
|
|||
mod parsing;
|
||||
use file::*;
|
||||
mod eval;
|
||||
|
||||
struct System {
|
||||
fnow: Vec<euler::Field>,
|
||||
fnext: Vec<euler::Field>,
|
||||
wb: Vec<euler::WorkBuffers>,
|
||||
k: [Vec<euler::Diff>; 4],
|
||||
grids: Vec<grid::Grid>,
|
||||
metrics: Vec<grid::Metrics>,
|
||||
bt: Vec<euler::BoundaryCharacteristics>,
|
||||
eb: Vec<euler::BoundaryStorage>,
|
||||
time: Float,
|
||||
operators: Vec<Box<dyn SbpOperator2d>>,
|
||||
}
|
||||
|
||||
use std::sync::atomic::{AtomicBool, Ordering};
|
||||
pub(crate) static MULTITHREAD: AtomicBool = AtomicBool::new(false);
|
||||
|
||||
impl integrate::Integrable for System {
|
||||
type State = Vec<euler::Field>;
|
||||
type Diff = Vec<euler::Diff>;
|
||||
|
||||
fn scaled_add(s: &mut Self::State, o: &Self::Diff, scale: Float) {
|
||||
if MULTITHREAD.load(Ordering::Acquire) {
|
||||
s.par_iter_mut()
|
||||
.zip(o.par_iter())
|
||||
.for_each(|(s, o)| euler::Field::scaled_add(s, o, scale))
|
||||
} else {
|
||||
s.iter_mut()
|
||||
.zip(o.iter())
|
||||
.for_each(|(s, o)| euler::Field::scaled_add(s, o, scale))
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
impl System {
|
||||
fn new(
|
||||
grids: Vec<grid::Grid>,
|
||||
bt: Vec<euler::BoundaryCharacteristics>,
|
||||
operators: Vec<Box<dyn SbpOperator2d>>,
|
||||
) -> Self {
|
||||
let fnow = grids
|
||||
.iter()
|
||||
.map(|g| euler::Field::new(g.ny(), g.nx()))
|
||||
.collect::<Vec<_>>();
|
||||
let fnext = fnow.clone();
|
||||
let wb = grids
|
||||
.iter()
|
||||
.map(|g| euler::WorkBuffers::new(g.ny(), g.nx()))
|
||||
.collect();
|
||||
let k = grids
|
||||
.iter()
|
||||
.map(|g| euler::Diff::zeros((g.ny(), g.nx())))
|
||||
.collect::<Vec<_>>();
|
||||
let k = [k.clone(), k.clone(), k.clone(), k];
|
||||
let metrics = grids
|
||||
.iter()
|
||||
.zip(&operators)
|
||||
.map(|(g, op)| g.metrics(&**op).unwrap())
|
||||
.collect::<Vec<_>>();
|
||||
|
||||
let eb = bt
|
||||
.iter()
|
||||
.zip(&grids)
|
||||
.map(|(bt, grid)| euler::BoundaryStorage::new(bt, grid))
|
||||
.collect();
|
||||
|
||||
Self {
|
||||
fnow,
|
||||
fnext,
|
||||
k,
|
||||
wb,
|
||||
grids,
|
||||
metrics,
|
||||
bt,
|
||||
eb,
|
||||
time: 0.0,
|
||||
operators,
|
||||
}
|
||||
}
|
||||
|
||||
fn vortex(&mut self, t: Float, vortex_params: &euler::VortexParameters) {
|
||||
for (f, g) in self.fnow.iter_mut().zip(&self.grids) {
|
||||
f.vortex(g.x(), g.y(), t, &vortex_params);
|
||||
}
|
||||
}
|
||||
|
||||
fn advance(&mut self, dt: Float) {
|
||||
let metrics = &self.metrics;
|
||||
let grids = &self.grids;
|
||||
let bt = &self.bt;
|
||||
let wb = &mut self.wb;
|
||||
let eb = &mut self.eb;
|
||||
let operators = &self.operators;
|
||||
|
||||
let rhs = move |fut: &mut Vec<euler::Diff>, prev: &Vec<euler::Field>, time: Float| {
|
||||
let prev_all = &prev;
|
||||
if MULTITHREAD.load(Ordering::Acquire) {
|
||||
rayon::scope(|s| {
|
||||
for (((((((fut, prev), wb), grid), metrics), op), bt), eb) in fut
|
||||
.iter_mut()
|
||||
.zip(prev.iter())
|
||||
.zip(wb.iter_mut())
|
||||
.zip(grids)
|
||||
.zip(metrics.iter())
|
||||
.zip(operators.iter())
|
||||
.zip(bt.iter())
|
||||
.zip(eb.iter_mut())
|
||||
{
|
||||
s.spawn(move |_| {
|
||||
let bc = euler::boundary_extracts(prev_all, bt, prev, grid, eb, time);
|
||||
if op.upwind().is_some() {
|
||||
euler::RHS_upwind(&**op, fut, prev, metrics, &bc, &mut wb.0);
|
||||
} else {
|
||||
euler::RHS_trad(&**op, fut, prev, metrics, &bc, &mut wb.0);
|
||||
}
|
||||
})
|
||||
}
|
||||
});
|
||||
} else {
|
||||
for (((((((fut, prev), wb), grid), metrics), op), bt), eb) in fut
|
||||
.iter_mut()
|
||||
.zip(prev.iter())
|
||||
.zip(wb.iter_mut())
|
||||
.zip(grids)
|
||||
.zip(metrics.iter())
|
||||
.zip(operators.iter())
|
||||
.zip(bt.iter())
|
||||
.zip(eb.iter_mut())
|
||||
{
|
||||
let bc = euler::boundary_extracts(prev_all, bt, prev, grid, eb, time);
|
||||
if op.upwind().is_some() {
|
||||
euler::RHS_upwind(&**op, fut, prev, metrics, &bc, &mut wb.0);
|
||||
} else {
|
||||
euler::RHS_trad(&**op, fut, prev, metrics, &bc, &mut wb.0);
|
||||
}
|
||||
}
|
||||
}
|
||||
};
|
||||
|
||||
integrate::integrate::<integrate::Rk4, System, _>(
|
||||
rhs,
|
||||
&self.fnow,
|
||||
&mut self.fnext,
|
||||
&mut self.time,
|
||||
dt,
|
||||
&mut self.k,
|
||||
);
|
||||
|
||||
std::mem::swap(&mut self.fnow, &mut self.fnext);
|
||||
}
|
||||
|
||||
/// Suggested maximum dt for this problem
|
||||
fn max_dt(&self) -> Float {
|
||||
let is_h2 = self
|
||||
.operators
|
||||
.iter()
|
||||
.any(|op| op.is_h2xi() || op.is_h2eta());
|
||||
let c_max = if is_h2 { 0.5 } else { 1.0 };
|
||||
let mut max_dt: Float = Float::INFINITY;
|
||||
|
||||
for (field, metrics) in self.fnow.iter().zip(self.metrics.iter()) {
|
||||
let nx = field.nx();
|
||||
let ny = field.ny();
|
||||
|
||||
let rho = field.rho();
|
||||
let rhou = field.rhou();
|
||||
let rhov = field.rhov();
|
||||
|
||||
let mut max_u: Float = 0.0;
|
||||
let mut max_v: Float = 0.0;
|
||||
|
||||
for ((((((rho, rhou), rhov), detj_dxi_dx), detj_dxi_dy), detj_deta_dx), detj_deta_dy) in
|
||||
rho.iter()
|
||||
.zip(rhou.iter())
|
||||
.zip(rhov.iter())
|
||||
.zip(metrics.detj_dxi_dx())
|
||||
.zip(metrics.detj_dxi_dy())
|
||||
.zip(metrics.detj_deta_dx())
|
||||
.zip(metrics.detj_deta_dy())
|
||||
{
|
||||
let u = rhou / rho;
|
||||
let v = rhov / rho;
|
||||
|
||||
let uhat: Float = detj_dxi_dx * u + detj_dxi_dy * v;
|
||||
let vhat: Float = detj_deta_dx * u + detj_deta_dy * v;
|
||||
|
||||
max_u = max_u.max(uhat.abs());
|
||||
max_v = max_v.max(vhat.abs());
|
||||
}
|
||||
|
||||
let dx = 1.0 / nx as Float;
|
||||
let dy = 1.0 / ny as Float;
|
||||
|
||||
let c_dt = Float::max(max_u / dx, max_v / dy);
|
||||
|
||||
max_dt = Float::min(max_dt, c_max / c_dt);
|
||||
}
|
||||
|
||||
max_dt
|
||||
}
|
||||
}
|
||||
mod system;
|
||||
use system::*;
|
||||
|
||||
#[derive(Debug, FromArgs)]
|
||||
/// Options for configuring and running the solver
|
||||
|
|
@ -312,13 +111,7 @@ fn main() {
|
|||
|
||||
{
|
||||
let nthreads = opt.jobs.unwrap_or(1);
|
||||
if nthreads > 1 {
|
||||
MULTITHREAD.store(true, Ordering::Release);
|
||||
rayon::ThreadPoolBuilder::new()
|
||||
.num_threads(nthreads)
|
||||
.build_global()
|
||||
.unwrap();
|
||||
}
|
||||
todo!("nthreads");
|
||||
}
|
||||
|
||||
let should_output = |itime| {
|
||||
|
|
|
|||
|
|
@ -0,0 +1,579 @@
|
|||
use crate::utils::Direction;
|
||||
use crossbeam_channel::{Receiver, Select, Sender};
|
||||
use euler::{
|
||||
eval::{self, Evaluator},
|
||||
Diff, Field, VortexParameters, WorkBuffers,
|
||||
};
|
||||
use ndarray::Array2;
|
||||
use sbp::grid::{Grid, Metrics};
|
||||
use sbp::operators::{InterpolationOperator, SbpOperator2d};
|
||||
use sbp::*;
|
||||
use std::sync::{Arc, Barrier};
|
||||
|
||||
pub struct System {
|
||||
pub fnow: Vec<euler::Field>,
|
||||
pub fnext: Vec<euler::Field>,
|
||||
pub wb: Vec<euler::WorkBuffers>,
|
||||
pub k: [Vec<euler::Diff>; 4],
|
||||
pub grids: Vec<grid::Grid>,
|
||||
pub metrics: Vec<grid::Metrics>,
|
||||
pub bt: Vec<euler::BoundaryCharacteristics>,
|
||||
pub eb: Vec<euler::BoundaryStorage>,
|
||||
pub time: Float,
|
||||
pub operators: Vec<Box<dyn SbpOperator2d>>,
|
||||
}
|
||||
|
||||
impl integrate::Integrable for System {
|
||||
type State = Vec<euler::Field>;
|
||||
type Diff = Vec<euler::Diff>;
|
||||
|
||||
fn scaled_add(s: &mut Self::State, o: &Self::Diff, scale: Float) {
|
||||
s.iter_mut()
|
||||
.zip(o.iter())
|
||||
.for_each(|(s, o)| euler::Field::scaled_add(s, o, scale))
|
||||
}
|
||||
}
|
||||
|
||||
impl System {
|
||||
pub fn new(
|
||||
grids: Vec<grid::Grid>,
|
||||
bt: Vec<euler::BoundaryCharacteristics>,
|
||||
operators: Vec<Box<dyn SbpOperator2d>>,
|
||||
) -> Self {
|
||||
let fnow = grids
|
||||
.iter()
|
||||
.map(|g| euler::Field::new(g.ny(), g.nx()))
|
||||
.collect::<Vec<_>>();
|
||||
let fnext = fnow.clone();
|
||||
let wb = grids
|
||||
.iter()
|
||||
.map(|g| euler::WorkBuffers::new(g.ny(), g.nx()))
|
||||
.collect();
|
||||
let k = grids
|
||||
.iter()
|
||||
.map(|g| euler::Diff::zeros((g.ny(), g.nx())))
|
||||
.collect::<Vec<_>>();
|
||||
let k = [k.clone(), k.clone(), k.clone(), k];
|
||||
let metrics = grids
|
||||
.iter()
|
||||
.zip(&operators)
|
||||
.map(|(g, op)| g.metrics(&**op).unwrap())
|
||||
.collect::<Vec<_>>();
|
||||
|
||||
let eb = bt
|
||||
.iter()
|
||||
.zip(&grids)
|
||||
.map(|(bt, grid)| euler::BoundaryStorage::new(bt, grid))
|
||||
.collect();
|
||||
|
||||
Self {
|
||||
fnow,
|
||||
fnext,
|
||||
k,
|
||||
wb,
|
||||
grids,
|
||||
metrics,
|
||||
bt,
|
||||
eb,
|
||||
time: 0.0,
|
||||
operators,
|
||||
}
|
||||
}
|
||||
|
||||
pub fn vortex(&mut self, t: Float, vortex_params: &euler::VortexParameters) {
|
||||
for (f, g) in self.fnow.iter_mut().zip(&self.grids) {
|
||||
f.vortex(g.x(), g.y(), t, &vortex_params);
|
||||
}
|
||||
}
|
||||
|
||||
pub fn advance(&mut self, dt: Float) {
|
||||
let metrics = &self.metrics;
|
||||
let grids = &self.grids;
|
||||
let bt = &self.bt;
|
||||
let wb = &mut self.wb;
|
||||
let eb = &mut self.eb;
|
||||
let operators = &self.operators;
|
||||
|
||||
let rhs = move |fut: &mut Vec<euler::Diff>, prev: &Vec<euler::Field>, time: Float| {
|
||||
let prev_all = &prev;
|
||||
for (((((((fut, prev), wb), grid), metrics), op), bt), eb) in fut
|
||||
.iter_mut()
|
||||
.zip(prev.iter())
|
||||
.zip(wb.iter_mut())
|
||||
.zip(grids)
|
||||
.zip(metrics.iter())
|
||||
.zip(operators.iter())
|
||||
.zip(bt.iter())
|
||||
.zip(eb.iter_mut())
|
||||
{
|
||||
let bc = euler::boundary_extracts(prev_all, bt, prev, grid, eb, time);
|
||||
if op.upwind().is_some() {
|
||||
euler::RHS_upwind(&**op, fut, prev, metrics, &bc, &mut wb.0);
|
||||
} else {
|
||||
euler::RHS_trad(&**op, fut, prev, metrics, &bc, &mut wb.0);
|
||||
}
|
||||
}
|
||||
};
|
||||
|
||||
integrate::integrate::<integrate::Rk4, System, _>(
|
||||
rhs,
|
||||
&self.fnow,
|
||||
&mut self.fnext,
|
||||
&mut self.time,
|
||||
dt,
|
||||
&mut self.k,
|
||||
);
|
||||
|
||||
std::mem::swap(&mut self.fnow, &mut self.fnext);
|
||||
}
|
||||
|
||||
/// Suggested maximum dt for this problem
|
||||
pub fn max_dt(&self) -> Float {
|
||||
let is_h2 = self
|
||||
.operators
|
||||
.iter()
|
||||
.any(|op| op.is_h2xi() || op.is_h2eta());
|
||||
let c_max = if is_h2 { 0.5 } else { 1.0 };
|
||||
let mut max_dt: Float = Float::INFINITY;
|
||||
|
||||
for (field, metrics) in self.fnow.iter().zip(self.metrics.iter()) {
|
||||
let nx = field.nx();
|
||||
let ny = field.ny();
|
||||
|
||||
let rho = field.rho();
|
||||
let rhou = field.rhou();
|
||||
let rhov = field.rhov();
|
||||
|
||||
let mut max_u: Float = 0.0;
|
||||
let mut max_v: Float = 0.0;
|
||||
|
||||
for ((((((rho, rhou), rhov), detj_dxi_dx), detj_dxi_dy), detj_deta_dx), detj_deta_dy) in
|
||||
rho.iter()
|
||||
.zip(rhou.iter())
|
||||
.zip(rhov.iter())
|
||||
.zip(metrics.detj_dxi_dx())
|
||||
.zip(metrics.detj_dxi_dy())
|
||||
.zip(metrics.detj_deta_dx())
|
||||
.zip(metrics.detj_deta_dy())
|
||||
{
|
||||
let u = rhou / rho;
|
||||
let v = rhov / rho;
|
||||
|
||||
let uhat: Float = detj_dxi_dx * u + detj_dxi_dy * v;
|
||||
let vhat: Float = detj_deta_dx * u + detj_deta_dy * v;
|
||||
|
||||
max_u = max_u.max(uhat.abs());
|
||||
max_v = max_v.max(vhat.abs());
|
||||
}
|
||||
|
||||
let dx = 1.0 / nx as Float;
|
||||
let dy = 1.0 / ny as Float;
|
||||
|
||||
let c_dt = Float::max(max_u / dx, max_v / dy);
|
||||
|
||||
max_dt = Float::min(max_dt, c_max / c_dt);
|
||||
}
|
||||
|
||||
max_dt
|
||||
}
|
||||
|
||||
/// Spreads the computation over n threads, in a thread per grid way.
|
||||
/// This system can only be called once for ntime calls.
|
||||
pub fn distribute(self, ntime: usize) -> DistributedSystem {
|
||||
let nthreads = self.grids.len();
|
||||
let time = 0.0;
|
||||
// alt: crossbeam::WaitGroup
|
||||
let b = Arc::new(Barrier::new(nthreads + 1));
|
||||
let dt = self.max_dt();
|
||||
|
||||
// Build up the boundary conditions
|
||||
// Assume all boundaries are push/pull through channels
|
||||
let channels = (0..nthreads)
|
||||
.map(|_| {
|
||||
use crossbeam_channel::unbounded;
|
||||
Direction {
|
||||
north: unbounded(),
|
||||
south: unbounded(),
|
||||
west: unbounded(),
|
||||
east: unbounded(),
|
||||
}
|
||||
})
|
||||
.collect::<Vec<_>>();
|
||||
|
||||
// TODO: Iterate through all grids and see if they need ourself to push
|
||||
let mut requested_channels = (0..nthreads)
|
||||
.map(|_| Direction::default())
|
||||
.collect::<Vec<Direction<bool>>>();
|
||||
|
||||
let mut tids = Vec::new();
|
||||
for (((((((current, fut), grid), metrics), sbp), wb), bt), req_channel) in self
|
||||
.fnow
|
||||
.into_iter()
|
||||
.zip(self.fnext.into_iter())
|
||||
.zip(self.grids.into_iter())
|
||||
.zip(self.metrics.into_iter())
|
||||
.zip(self.operators.into_iter())
|
||||
.zip(self.wb.into_iter())
|
||||
.zip(self.bt)
|
||||
.zip(requested_channels)
|
||||
{
|
||||
let builder = std::thread::Builder::new().name(format!("eulersolver: {}", "smth"));
|
||||
let barrier = b.clone();
|
||||
|
||||
let Direction {
|
||||
north: bt_north,
|
||||
south: bt_south,
|
||||
west: bt_west,
|
||||
east: bt_east,
|
||||
} = bt;
|
||||
let boundary_conditions = Direction {
|
||||
north: match bt_north {
|
||||
euler::BoundaryCharacteristic::This => DistributedBoundaryConditions::This,
|
||||
euler::BoundaryCharacteristic::Grid(i) => {
|
||||
*requested_channels[i].south_mut() = true;
|
||||
DistributedBoundaryConditions::Channel(channels[i].south().1.clone())
|
||||
}
|
||||
euler::BoundaryCharacteristic::Interpolate(i, int_op) => {
|
||||
*requested_channels[i].south_mut() = true;
|
||||
DistributedBoundaryConditions::Interpolate(
|
||||
channels[i].south().1.clone(),
|
||||
int_op,
|
||||
)
|
||||
}
|
||||
euler::BoundaryCharacteristic::MultiGrid(_) => unimplemented!(),
|
||||
euler::BoundaryCharacteristic::Vortex(vp) => {
|
||||
DistributedBoundaryConditions::Vortex(vp)
|
||||
}
|
||||
euler::BoundaryCharacteristic::Eval(eval) => {
|
||||
DistributedBoundaryConditions::Eval(eval)
|
||||
}
|
||||
},
|
||||
south: match bt_south {
|
||||
euler::BoundaryCharacteristic::This => DistributedBoundaryConditions::This,
|
||||
euler::BoundaryCharacteristic::Grid(i) => {
|
||||
*requested_channels[i].north_mut() = true;
|
||||
DistributedBoundaryConditions::Channel(channels[i].north().1.clone())
|
||||
}
|
||||
euler::BoundaryCharacteristic::Interpolate(i, int_op) => {
|
||||
*requested_channels[i].north_mut() = true;
|
||||
DistributedBoundaryConditions::Interpolate(
|
||||
channels[i].north().1.clone(),
|
||||
int_op,
|
||||
)
|
||||
}
|
||||
euler::BoundaryCharacteristic::MultiGrid(_) => unimplemented!(),
|
||||
euler::BoundaryCharacteristic::Vortex(vp) => {
|
||||
DistributedBoundaryConditions::Vortex(vp)
|
||||
}
|
||||
euler::BoundaryCharacteristic::Eval(eval) => {
|
||||
DistributedBoundaryConditions::Eval(eval)
|
||||
}
|
||||
},
|
||||
east: match bt_east {
|
||||
euler::BoundaryCharacteristic::This => DistributedBoundaryConditions::This,
|
||||
euler::BoundaryCharacteristic::Grid(i) => {
|
||||
*requested_channels[i].west_mut() = true;
|
||||
DistributedBoundaryConditions::Channel(channels[i].west().1.clone())
|
||||
}
|
||||
euler::BoundaryCharacteristic::Interpolate(i, int_op) => {
|
||||
*requested_channels[i].west_mut() = true;
|
||||
DistributedBoundaryConditions::Interpolate(
|
||||
channels[i].west().1.clone(),
|
||||
int_op,
|
||||
)
|
||||
}
|
||||
euler::BoundaryCharacteristic::MultiGrid(_) => unimplemented!(),
|
||||
euler::BoundaryCharacteristic::Vortex(vp) => {
|
||||
DistributedBoundaryConditions::Vortex(vp)
|
||||
}
|
||||
euler::BoundaryCharacteristic::Eval(eval) => {
|
||||
DistributedBoundaryConditions::Eval(eval)
|
||||
}
|
||||
},
|
||||
west: match bt_west {
|
||||
euler::BoundaryCharacteristic::This => DistributedBoundaryConditions::This,
|
||||
euler::BoundaryCharacteristic::Grid(i) => {
|
||||
*requested_channels[i].east_mut() = true;
|
||||
DistributedBoundaryConditions::Channel(channels[i].east().1.clone())
|
||||
}
|
||||
euler::BoundaryCharacteristic::Interpolate(i, int_op) => {
|
||||
*requested_channels[i].east_mut() = true;
|
||||
DistributedBoundaryConditions::Interpolate(
|
||||
channels[i].east().1.clone(),
|
||||
int_op,
|
||||
)
|
||||
}
|
||||
euler::BoundaryCharacteristic::MultiGrid(_) => unimplemented!(),
|
||||
euler::BoundaryCharacteristic::Vortex(vp) => {
|
||||
DistributedBoundaryConditions::Vortex(vp)
|
||||
}
|
||||
euler::BoundaryCharacteristic::Eval(eval) => {
|
||||
DistributedBoundaryConditions::Eval(eval)
|
||||
}
|
||||
},
|
||||
};
|
||||
|
||||
tids.push(
|
||||
builder
|
||||
.spawn(move || {
|
||||
let mut sys = DistributedSystemPart {
|
||||
barrier,
|
||||
ntime,
|
||||
dt,
|
||||
current,
|
||||
fut,
|
||||
k: [todo!(); 4],
|
||||
boundary_conditions,
|
||||
grid: (grid, metrics),
|
||||
output: (),
|
||||
push: todo!(),
|
||||
sbp,
|
||||
t: time,
|
||||
wb,
|
||||
};
|
||||
sys.advance();
|
||||
})
|
||||
.unwrap(),
|
||||
);
|
||||
}
|
||||
// Set up communicators
|
||||
// Spawn a new communicator
|
||||
|
||||
DistributedSystem {
|
||||
ntime,
|
||||
start: b,
|
||||
sys: tids,
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// single-threaded items: clone_from
|
||||
// sync points: scaled_add, rhs
|
||||
//
|
||||
// Could instead make every thread (of a threadpool?) carry one grid themselves,
|
||||
// and instead only wait on obtaining bc, which would make synchronisation
|
||||
// be somewhat minimal
|
||||
//
|
||||
// Difficulties/implementation notes:
|
||||
// * How to get output from each thread? Push to the multi-producer, single consumer queue
|
||||
// with capacity of 2*N (which ensures internal synchronisation stays active), or many
|
||||
// channels which are select!'ed. Choose instead a (name, itime, data) tuple, which is
|
||||
// communicated. Also initialise the output file all at the start, since we know the
|
||||
// number of time steps, and which timesteps will be used (should_output).
|
||||
// * How to balance the number of threads/jobs?
|
||||
// Each grid will be pushed to a thread, and a thread pool created for use by all threads
|
||||
// combined. Set affinity to limit the number of cores available (could have side effects
|
||||
// on heat etc, although this would also be the case for multi-threaded code). A
|
||||
// thread-per-core architecture + the output thread if we have enough cores, else let os
|
||||
// control the scheduling for us.
|
||||
// Use cgroups to artificially limit the number of cores.
|
||||
// Use taskset to artificially limit the number of cores.
|
||||
// * Mechanism to wait for bc, is select available? Use a channel with 0-4 capacity, each
|
||||
// other grid additonally pushes their direction to always use the available
|
||||
// Use select or similar from crossbeam_channel
|
||||
// * Mechanism to push bc: try_push and then move to next bc if not available. Spawn a
|
||||
// thread into a threadpool to handle this.
|
||||
// * Each grid is spawned onto an async threadpool with M threads allocated to itself, this should
|
||||
// allow async waiting on the channel (or in the future from process/internet/bc processor)
|
||||
// * BC: A local thread can have a handle
|
||||
pub struct DistributedSystem {
|
||||
/// Simple messaging system to be replaced by a more sophisticated system (e.g. run 5 steps,
|
||||
/// collect, initialise, return something to main)
|
||||
/// This simply waits until all threads are ready, then starts the computation on all threads
|
||||
start: Arc<Barrier>,
|
||||
ntime: usize,
|
||||
/// These should be joined to mark the end of the computation
|
||||
sys: Vec<std::thread::JoinHandle<()>>,
|
||||
}
|
||||
|
||||
impl DistributedSystem {
|
||||
fn run(self) {
|
||||
// This should start as we have n thread, but barrier blocks on n+1
|
||||
self.start.wait();
|
||||
self.sys.into_iter().for_each(|tid| tid.join().unwrap());
|
||||
}
|
||||
}
|
||||
|
||||
// #[derive(Debug)]
|
||||
pub enum DistributedBoundaryConditions {
|
||||
This,
|
||||
|
||||
Vortex(VortexParameters),
|
||||
Eval(std::sync::Arc<dyn eval::Evaluator<ndarray::Ix1>>),
|
||||
|
||||
Interpolate(Receiver<Array2<Float>>, Box<dyn InterpolationOperator>),
|
||||
Channel(Receiver<Array2<Float>>),
|
||||
}
|
||||
|
||||
impl DistributedBoundaryConditions {
|
||||
fn channel(&self) -> Option<&Receiver<Array2<Float>>> {
|
||||
match self {
|
||||
Self::Interpolate(r, _) | Self::Channel(r) => Some(r),
|
||||
_ => None,
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#[derive(Debug, Clone)]
|
||||
enum PushCommunicator {
|
||||
Channel(Sender<Array2<Float>>),
|
||||
None,
|
||||
}
|
||||
|
||||
impl Default for PushCommunicator {
|
||||
fn default() -> Self {
|
||||
Self::None
|
||||
}
|
||||
}
|
||||
|
||||
struct DistributedSystemPart {
|
||||
grid: (Grid, Metrics),
|
||||
sbp: Box<dyn SbpOperator2d + 'static>,
|
||||
|
||||
boundary_conditions: Direction<DistributedBoundaryConditions>,
|
||||
/// Subscribers to the boundaries of self
|
||||
push: Direction<PushCommunicator>,
|
||||
|
||||
current: Field,
|
||||
fut: Field,
|
||||
k: [Diff; 4],
|
||||
wb: WorkBuffers,
|
||||
barrier: Arc<Barrier>,
|
||||
output: (), // hdf5::Dataset eventually,
|
||||
t: Float,
|
||||
dt: Float,
|
||||
ntime: usize,
|
||||
}
|
||||
|
||||
impl DistributedSystemPart {
|
||||
fn advance(&mut self) {
|
||||
self.barrier.wait();
|
||||
for i in 0..self.ntime {
|
||||
let metrics = &self.grid.1;
|
||||
let wb = &mut self.wb.0;
|
||||
let sbp = &self.sbp;
|
||||
let push = &self.push;
|
||||
let boundary_conditions = &self.boundary_conditions;
|
||||
let grid = &self.grid.0;
|
||||
|
||||
let mut rhs = |k: &mut euler::Diff, y: &euler::Field, time: Float| {
|
||||
// Send off the boundaries optimistically, in case some grid is ready
|
||||
match &push.north {
|
||||
PushCommunicator::None => (),
|
||||
PushCommunicator::Channel(s) => s.send(y.north().to_owned()).unwrap(),
|
||||
}
|
||||
match &push.south {
|
||||
PushCommunicator::None => (),
|
||||
PushCommunicator::Channel(s) => s.send(y.south().to_owned()).unwrap(),
|
||||
}
|
||||
match &push.east {
|
||||
PushCommunicator::None => (),
|
||||
PushCommunicator::Channel(s) => s.send(y.east().to_owned()).unwrap(),
|
||||
}
|
||||
match &push.west {
|
||||
PushCommunicator::None => (),
|
||||
PushCommunicator::Channel(s) => s.send(y.west().to_owned()).unwrap(),
|
||||
}
|
||||
|
||||
use std::ops::Deref;
|
||||
// This computation does not depend on the boundaries
|
||||
euler::RHS_no_SAT(sbp.deref(), k, y, metrics, wb);
|
||||
|
||||
fn north_sat() {
|
||||
todo!()
|
||||
}
|
||||
|
||||
// Get boundaries, but be careful and maximise the amount of work which can be
|
||||
// performed before we have all of them, whilst ensuring threads can sleep for as
|
||||
// long as possible
|
||||
let mut select = Select::new();
|
||||
let mut selectable = 0;
|
||||
let recv_north = match boundary_conditions.north() {
|
||||
DistributedBoundaryConditions::Channel(r)
|
||||
| DistributedBoundaryConditions::Interpolate(r, _) => Some(r),
|
||||
DistributedBoundaryConditions::This => {
|
||||
todo!()
|
||||
}
|
||||
DistributedBoundaryConditions::Vortex(vp) => {
|
||||
let mut data = y.north().to_owned();
|
||||
let mut fiter = data.outer_iter_mut();
|
||||
let (rho, rhou, rhov, e) = (
|
||||
fiter.next().unwrap(),
|
||||
fiter.next().unwrap(),
|
||||
fiter.next().unwrap(),
|
||||
fiter.next().unwrap(),
|
||||
);
|
||||
let (x, y) = grid.north();
|
||||
vp.evaluate(time, x, y, rho, rhou, rhov, e);
|
||||
|
||||
north_sat();
|
||||
None
|
||||
}
|
||||
DistributedBoundaryConditions::Eval(eval) => {
|
||||
todo!()
|
||||
}
|
||||
_ => None,
|
||||
};
|
||||
let recv_south = if let Some(r) = boundary_conditions.south().channel() {
|
||||
selectable += 1;
|
||||
Some(select.recv(r))
|
||||
} else {
|
||||
// Do SAT boundary from other BC
|
||||
None
|
||||
};
|
||||
let recv_west = if let Some(r) = boundary_conditions.west().channel() {
|
||||
selectable += 1;
|
||||
Some(select.recv(r))
|
||||
} else {
|
||||
// Do SAT boundary from other BC
|
||||
None
|
||||
};
|
||||
let recv_east = if let Some(r) = boundary_conditions.east().channel() {
|
||||
selectable += 1;
|
||||
Some(select.recv(r))
|
||||
} else {
|
||||
// Do SAT boundary from other BC
|
||||
None
|
||||
};
|
||||
|
||||
// Get an item off each channel, waiting minimally before processing that boundary.
|
||||
// The waiting ensures other grids can be processed by the core in case of
|
||||
// oversubscription (in case of a more grids than core scenario)
|
||||
// This minimises the amount of time waiting on boundary conditions
|
||||
while selectable != 0 {
|
||||
let s = select.select();
|
||||
let sindex = s.index();
|
||||
match Some(sindex) {
|
||||
recv_north => {
|
||||
let r = s.recv(boundary_conditions.north().channel().unwrap());
|
||||
// process into boundary SAT here
|
||||
}
|
||||
recv_south => {
|
||||
let r = s.recv(boundary_conditions.south().channel().unwrap());
|
||||
// process into boundary SAT here
|
||||
}
|
||||
recv_west => {
|
||||
let r = s.recv(boundary_conditions.west().channel().unwrap());
|
||||
// process into boundary SAT here
|
||||
}
|
||||
recv_east => {
|
||||
let r = s.recv(boundary_conditions.east().channel().unwrap());
|
||||
// process into boundary SAT here
|
||||
}
|
||||
_ => unreachable!(),
|
||||
}
|
||||
select.remove(sindex);
|
||||
selectable -= 1;
|
||||
}
|
||||
};
|
||||
integrate::integrate::<integrate::Rk4, Field, _>(
|
||||
rhs,
|
||||
&self.current,
|
||||
&mut self.fut,
|
||||
&mut self.t,
|
||||
self.dt,
|
||||
&mut self.k,
|
||||
)
|
||||
}
|
||||
}
|
||||
}
|
||||
Loading…
Reference in New Issue