Commit d6c2a7c3 authored by Mads R. B. Kristensen's avatar Mads R. B. Kristensen

Updated the chapel/heat_equation

parent 1d8e3963
......@@ -6,38 +6,56 @@ In this example, we solve the heat equation. The idea is to apply a 5-point sten
Sequential
----------
`sequential.chpl <src/sequential.chpl>` is a sequential implementation of the heat equation written in Chapel. The stencil computation is the most time consuming part of the code and look like::
[sequential.chpl](src/sequential.chpl) is a sequential implementation of the heat equation written in Chapel. The stencil computation is the most time consuming part of the code and look like:
```
for (i,j) in Interior do//Iterate over all non-border cells
{
//Assign each cell in 'T' the mean of its neighboring cells in 'A'
T[i,j] = (A[i,j] + A[i-1,j] + A[i+1,j] + A[i,j-1] + A[i,j+1]) / 5;
}
```
Basically, each *interior* element in ``T`` gets the mean of the corresponding element in ``A`` as well as the neighboring elements. Since ``for`` is a sequential language construct in Chapel, a single CPU-core will execute this code.
Basically, each *interior* element in `T` gets the mean of the corresponding element in `A` as well as the neighboring elements. Since `for` is a sequential language construct in Chapel, a single CPU-core will execute this code.
Now, let's run it:
```
./bin/heat_equation -nl 1 --size=5000*10
Heat Equation (sequential) - n: 5000, iterations: 10, elapsed-time: 381.5 seconds
```
Multi-core
----------
In order to improve the performance, we can tell Chapel to use threads to execute the stencil operations in parallel (`single_machine.chpl <src/single_machine.chpl>`). We do that by replacing ``for`` with ``forall``, which tells Chapel to execute each iteration in ``Interior`` parallel.
It is our responsibility to make sure that each iteration in the ``forall`` loop is independent in order not to introduce race conditions.
In order to improve the performance, we can tell Chapel to use threads to execute the stencil operations in parallel ([single_machine.chpl](src/single_machine.chpl)). We do that by replacing `for` with `forall`, which tells Chapel to execute each iteration in `Interior` parallel.
It is our responsibility to make sure that each iteration in the `forall` loop is independent in order not to introduce race conditions.
Clearly in this case iteration is clearly independent since we do not read ``T``::
Clearly in this case iteration is clearly independent since we do not read `T`:
```
forall (i,j) in Interior do//Iterate over all non-border cells
{
//Assign each cell in 'T' the mean of its neighboring cells in 'A'
T[i,j] = (A[i,j] + A[i-1,j] + A[i+1,j] + A[i,j-1] + A[i,j+1]) / 5;
}
```
Now, let's run it (note that `CHPL_RT_NUM_THREADS_PER_LOCALE` tells Chapel the number of threads to use)::
```
export CHPL_RT_NUM_THREADS_PER_LOCALE=16
./bin/heat_equation -nl 1 --size=5000*10
Heat Equation (single machine) - n: 5000, iterations: 10, elapsed-time: 25.7052 seconds
```
Multiple Machines
-----------------
In order to improve the performance even further, we can tell Chapel to execute the stencil operation in parallel on multiple machines (`multiple_machines.chpl <src/multiple_machines.chpl>`).
We still use the ``forall`` loop construct, be we have to tell Chapel how to distributes ``A`` and ``T`` between the multiple machines. For that, we use the ``dmapped`` language construct when defining the ``Grid`` and ``Interior`` domain::
We still use the `forall` loop construct, be we have to tell Chapel how to distributes `A` and `T` between the multiple machines. For that, we use the `dmapped` language construct when defining the `Grid` and `Interior` domain:
```
//A n+2 by n+2 domain.
const Grid = {0..n+1, 0..n+1} dmapped Block({1..n, 1..n});
......@@ -45,7 +63,23 @@ We still use the ``forall`` loop construct, be we have to tell Chapel how to dis
const Interior = {1..n, 1..n} dmapped Block({1..n, 1..n});
var A, T : [Grid] real;//Zero initialized as default
```
We tell Chapel to use the same *block* distribution of the `Grid` and `Interior` domain such that each index in `Grid` has the same location as the corresponding index in `Interior`. Because they use the same distribution, no communication is needed when accessing the same index. For example, the operations `A[2,4] + T[2,4]` can be done locally on the machine that *owns* index `[2,4]`. However, it also means that a operations such as `A[2,4] + T[3,4]` will generally require communication.
Now, let's run it (note that `-nl 8` tells Chapel to use eight locations):
```
export CHPL_RT_NUM_THREADS_PER_LOCALE=16
./bin/heat_equation -nl 8 --size=5000*10
Heat Equation (multiple machines) - n: 5000, iterations: 10, elapsed-time: 5.13 seconds
```
It is very importation that all arrays in the calculation uses similar `dmapped` layouts. For example, if we do not use `dmapped` when defines `Interior` we get horrible performance:
We tell Chapel to use the same *block* distribution of the ``Grid`` and ``Interior`` domain such that each index in ``Grid`` has the same location as the corresponding index in ``Interior``. Because they use the same distribution, no communication is needed when accessing the same index. For example, the operations ``A[2,4] + T[2,4]`` can be done locally on the machine that *owns* index ``[2,4]``. However, it also means that a operations such as ``A[2,4] + T[3,4]`` will generally require communication.
```
export CHPL_RT_NUM_THREADS_PER_LOCALE=16
./bin/heat_equation -nl 8 --size=5000*10
Heat Equation (multiple machines) - n: 5000, iterations: 10, elapsed-time: 1823.23 seconds
```
In relation to HPC, it is very importation use ``dmapped`` such that you minimize the communication requirements of your application.
......@@ -2,13 +2,17 @@
//The first integer is the domain size squired and the second integer is
//the number of iterations.
config const size = "100*10";//Default, 100 by 100 domain and 10 iterations
config const epsilon = 1.0e-10;//Stop condition in amount of change
//Stop condition in amount of change (ignored when 'iterations' are non-zero).
config const epsilon = 1.0e-10;
//Parse the --size argument into 'n' and 'iterations'
use Regexp;
const arg = size.matches(compile("(\\d+)*(\\d+)"));
const n = size.substring(arg[1][1]) : int;
const iterations = size.substring(arg[2][1]) : int;
const arg_n = arg[1][1];
const arg_i = arg[2][1];
const n = size[arg_n.offset+1..arg_n.offset+arg_n.length] : int;
const iterations = size[arg_i.offset+1..arg_i.offset+arg_i.length]: int;
//Initiate a Timer object
use Time;
......@@ -49,9 +53,21 @@ do{
//Copy back the non-border cells
A[Interior] = T[Interior];
//When 'delta' is smaller than 'epsilon' the calculation has converged
iter_count += 1;
} while (delta > epsilon && iter_count >= iterations);
//if 'iterations' is non-zero we stop after a fixed number of iterations
//otherwise we stop when the calculation has converged, i.e. 'delta' is smaller than 'epsilon'.
var stop = false;
if(iterations > 0)
{
if iter_count >= iterations then
stop = true;
}
else
{
if delta < epsilon then
stop = true;
}
} while (!stop);
timer.stop();
writeln("Heat Equation (multiple machines) - n: ",n,
......
......@@ -2,13 +2,17 @@
//The first integer is the domain size squired and the second integer is
//the number of iterations.
config const size = "100*10";//Default, 100 by 100 domain and 10 iterations
config const epsilon = 1.0e-10;//Stop condition in amount of change
//Stop condition in amount of change (ignored when 'iterations' are non-zero).
config const epsilon = 1.0e-10;
//Parse the --size argument into 'n' and 'iterations'
use Regexp;
const arg = size.matches(compile("(\\d+)*(\\d+)"));
const n = size.substring(arg[1][1]) : int;
const iterations = size.substring(arg[2][1]) : int;
const arg_n = arg[1][1];
const arg_i = arg[2][1];
const n = size[arg_n.offset+1..arg_n.offset+arg_n.length] : int;
const iterations = size[arg_i.offset+1..arg_i.offset+arg_i.length]: int;
//Initiate a Timer object
use Time;
......@@ -46,9 +50,21 @@ do{
//Copy back the non-border cells
A[Interior] = T[Interior];
//When 'delta' is smaller than 'epsilon' the calculation has converged
iter_count += 1;
} while (delta > epsilon && iter_count >= iterations);
//if 'iterations' is non-zero we stop after a fixed number of iterations
//otherwise we stop when the calculation has converged, i.e. 'delta' is smaller than 'epsilon'.
var stop = false;
if(iterations > 0)
{
if iter_count >= iterations then
stop = true;
}
else
{
if delta < epsilon then
stop = true;
}
} while (!stop);
timer.stop();
writeln("Heat Equation (sequential) - n: ",n,
......
......@@ -2,13 +2,17 @@
//The first integer is the domain size squired and the second integer is
//the number of iterations.
config const size = "100*10";//Default, 100 by 100 domain and 10 iterations
config const epsilon = 1.0e-10;//Stop condition in amount of change
//Stop condition in amount of change (ignored when 'iterations' are non-zero).
config const epsilon = 1.0e-10;
//Parse the --size argument into 'n' and 'iterations'
use Regexp;
const arg = size.matches(compile("(\\d+)*(\\d+)"));
const n = size.substring(arg[1][1]) : int;
const iterations = size.substring(arg[2][1]) : int;
const arg_n = arg[1][1];
const arg_i = arg[2][1];
const n = size[arg_n.offset+1..arg_n.offset+arg_n.length] : int;
const iterations = size[arg_i.offset+1..arg_i.offset+arg_i.length]: int;
//Initiate a Timer object
use Time;
......@@ -46,9 +50,21 @@ do{
//Copy back the non-border cells
A[Interior] = T[Interior];
//When 'delta' is smaller than 'epsilon' the calculation has converged
iter_count += 1;
} while (delta > epsilon && iter_count >= iterations);
//if 'iterations' is non-zero we stop after a fixed number of iterations
//otherwise we stop when the calculation has converged, i.e. 'delta' is smaller than 'epsilon'.
var stop = false;
if(iterations > 0)
{
if iter_count >= iterations then
stop = true;
}
else
{
if delta < epsilon then
stop = true;
}
} while (!stop);
timer.stop();
writeln("Heat Equation (single machine) - n: ",n,
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment