Adjusts velocity boundary conditions for accelerated flows #108

marinlauber · 2024-01-29T14:23:06Z

I have adjusted the velocity boundary conditions to allow for either a Tuple or a Function(i,t). This is required if we have an accelerated flow without periodic BC in the forced direction. The resulting body force is computed with ForwardDiff on the body velocity function.

Either boundary condition type is dispatched to a BCTuple function that generates a Tuple that is then passed to the BC!() function.

I added an example, examples/TwoD_SlowStartCircle.jl, that demonstrates how to use the time-varying velocity boundary conditions.

I have added some tests (not run on the GPU yet) for the new functions.

codecov · 2024-01-29T15:33:55Z

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (5dea927) 96.44% compared to head (c90de26) 96.55%.
Report is 15 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #108      +/-   ##
==========================================
+ Coverage   96.44%   96.55%   +0.11%     
==========================================
  Files          12       12              
  Lines         450      465      +15     
==========================================
+ Hits          434      449      +15     
  Misses         16       16

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

b-fg · 2024-01-30T09:52:04Z

This requires storing the velocity BC as an Array instead of a Tuple. I am not sure if this needs to be stored on the GPU or not (I will test this).

That's indeed required for the GPU. We cannot have a small CuArray and access it by index as with an Array, so this is why tuples are used instead.

src/Flow.jl

b-fg · 2024-01-30T16:32:51Z

Have a look here: https://juliaarrays.github.io/StaticArrays.jl/stable/pages/api/#Mutable-arrays:-MVector,-MMatrix-and-MArray

I think MVector structures is what you are looking for.

weymouth · 2024-01-30T20:12:27Z

Shouldn't we just use functions instead of arrays? @TzuYaoHuang's g function works fine on GPUs. Why do we need an array or Tuple at all?

marinlauber · 2024-02-01T08:55:59Z

Shouldn't we just use functions instead of arrays? @TzuYaoHuang's g function works fine on GPUs. Why do we need an array or Tuple at all?

Yes, we can, but this will mean that the boundary condition function will now call these functions every time.

b-fg · 2024-02-01T09:26:59Z

Maybe we can dispatch to the tuple for static conditions and dispatch to the functions for dynamic ones.

…nd updated documentation in src/Waterlily

src/Flow.jl

test/runtests.jl

b-fg · 2024-02-06T13:29:01Z

I think this is ready to merge? @weymouth @marinlauber

marinlauber · 2024-02-07T12:46:10Z

This is ready!

The tests are passing on CPU/GPU

Test Summary: | Pass  Total   Time
util.jl       |   34     34  20.8s
Test Summary: | Pass  Total   Time
Poisson.jl    |   14     14  18.8s
Test Summary:        | Pass  Total  Time
MultiLevelPoisson.jl |   13     13  6.2s
Test Summary: | Pass  Total  Time
Flow.jl       |   18     18  9.4s
Test Summary: | Pass  Total  Time
Body.jl       |    3      3  0.1s
Test Summary: | Pass  Total  Time
AutoBody.jl   |   11     11  5.3s
Test Summary:        | Pass  Total  Time
Flow.jl periodic TGV |    2      2  5.1s
Test Summary:                      | Pass  Total   Time
Flow.jl with increasing body force |    2      2  10.0s
Test Summary: | Pass  Total  Time
Metrics.jl    |   17     17  6.8s
Test Summary: | Pass  Total  Time
WaterLily.jl  |   34     34  6.4s
Test Summary: | Pass  Total   Time
VTKExt.jl     |   28     28  14.2s
     Testing WaterLily tests passed

And the benchmark results compared to master (d0ec1d2).

--------------------------------------
Running benchmark tests for:
 - Julia:        1.10.0
 - Backends:     Array CuArray
 - CPU threads:  1 6
 - Cases:        tgv jelly
 - Size:         6,7 5,6
 - Sim. steps:   100 100
 - Data type:    Float32 Float32
--------------------------------------

Julia 1.10.0 benchmaks
Updating environment to Julia 1.10.0
   Resolving package versions...
  No Changes to `~/Workspace/WaterLily.jl/benchmark/Project.toml`
  No Changes to `~/Workspace/WaterLily.jl/benchmark/Manifest.toml`
    Updating registry at `/scratch/marinlauber/.julia/registries/General.toml`
  No Changes to `~/Workspace/WaterLily.jl/benchmark/Project.toml`
  No Changes to `~/Workspace/WaterLily.jl/benchmark/Manifest.toml`
Running: julia --project --startup-file=no -t 1 benchmark.jl --backend=Array --cases=["tgv","jelly"] --log2p=[(6,7),(5,6)] --max_steps=[100,100] --ftype=[Float32,Float32]
Benchmarking: tgv
(1/2) benchmarking "6"...
  (1/1) benchmarking "sim_step!"...
  done (took 9.153122932 seconds)
done (took 9.664080328 seconds)
(2/2) benchmarking "7"...
  (1/1) benchmarking "sim_step!"...
  done (took 55.775909505 seconds)
done (took 56.219365421 seconds)
Benchmarking: jelly
(1/2) benchmarking "5"...
  (1/1) benchmarking "sim_step!"...
  done (took 10.31607692 seconds)
done (took 10.762646285 seconds)
(2/2) benchmarking "6"...
  (1/1) benchmarking "sim_step!"...
  done (took 48.866873048 seconds)
done (took 49.323129197 seconds)
Running: julia --project --startup-file=no -t 6 benchmark.jl --backend=Array --cases=["tgv","jelly"] --log2p=[(6,7),(5,6)] --max_steps=[100,100] --ftype=[Float32,Float32]
Benchmarking: tgv
(1/2) benchmarking "6"...
  (1/1) benchmarking "sim_step!"...
  done (took 3.224376419 seconds)
done (took 3.562162138 seconds)
(2/2) benchmarking "7"...
  (1/1) benchmarking "sim_step!"...
  done (took 15.920800915 seconds)
done (took 16.161406035 seconds)
Benchmarking: jelly
(1/2) benchmarking "5"...
  (1/1) benchmarking "sim_step!"...
  done (took 3.995909223 seconds)
done (took 4.262310484 seconds)
(2/2) benchmarking "6"...
  (1/1) benchmarking "sim_step!"...
  done (took 18.140882184 seconds)
done (took 18.3939807 seconds)
Running: julia --project --startup-file=no benchmark.jl --backend=CuArray --cases=["tgv","jelly"] --log2p=[(6,7),(5,6)] --max_steps=[100,100] --ftype=[Float32,Float32]
Benchmarking: tgv
(1/2) benchmarking "6"...
  (1/1) benchmarking "sim_step!"...
  done (took 1.097216626 seconds)
done (took 1.657903387 seconds)
(2/2) benchmarking "7"...
  (1/1) benchmarking "sim_step!"...
  done (took 1.639622323 seconds)
done (took 2.123524882 seconds)
Benchmarking: jelly
(1/2) benchmarking "5"...
  (1/1) benchmarking "sim_step!"...
  done (took 1.510458438 seconds)
done (took 2.007634304 seconds)
(2/2) benchmarking "6"...
  (1/1) benchmarking "sim_step!"...
  done (took 2.660584546 seconds)
done (took 3.166380567 seconds)

Benchmark environment: tgv sim_step! (max_steps=100)
▶ log2p = 6
┌─────────┬───────────┬────────┬───────────┬─────────────┬────────┬──────────┬──────────┐
│ Backend │ WaterLily │ Julia  │ Precision │ Allocations │ GC [%] │ Time [s] │ Speed-up │
├─────────┼───────────┼────────┼───────────┼─────────────┼────────┼──────────┼──────────┤
│   CPUx1 │   afa03a3 │ 1.10.0 │   Float32 │     1271464 │   0.09 │     8.67 │     1.00 │
│   CPUx1 │   d0ec1d2 │ 1.10.0 │   Float32 │     1357166 │   0.23 │     8.95 │     0.97 │
│   CPUx6 │   afa03a3 │ 1.10.0 │   Float32 │     4677363 │   1.00 │     2.92 │     2.97 │
│   CPUx6 │   d0ec1d2 │ 1.10.0 │   Float32 │     4999132 │   0.94 │     3.13 │     2.77 │
│     GPU │   afa03a3 │ 1.10.0 │   Float32 │     2304525 │   1.56 │     0.57 │    15.17 │
│     GPU │   d0ec1d2 │ 1.10.0 │   Float32 │     2467970 │   1.58 │     0.60 │    14.57 │
└─────────┴───────────┴────────┴───────────┴─────────────┴────────┴──────────┴──────────┘
▶ log2p = 7
┌─────────┬───────────┬────────┬───────────┬─────────────┬────────┬──────────┬──────────┐
│ Backend │ WaterLily │ Julia  │ Precision │ Allocations │ GC [%] │ Time [s] │ Speed-up │
├─────────┼───────────┼────────┼───────────┼─────────────┼────────┼──────────┼──────────┤
│   CPUx1 │   afa03a3 │ 1.10.0 │   Float32 │     1145360 │   0.09 │    55.34 │     1.00 │
│   CPUx1 │   d0ec1d2 │ 1.10.0 │   Float32 │     1329041 │   0.14 │    57.37 │     0.96 │
│   CPUx6 │   afa03a3 │ 1.10.0 │   Float32 │     3994709 │   0.40 │    15.65 │     3.54 │
│   CPUx6 │   d0ec1d2 │ 1.10.0 │   Float32 │     4683048 │   0.67 │    16.54 │     3.35 │
│     GPU │   afa03a3 │ 1.10.0 │   Float32 │     1918194 │   0.38 │     1.16 │    47.86 │
│     GPU │   d0ec1d2 │ 1.10.0 │   Float32 │     2263519 │   0.68 │     1.31 │    42.10 │
└─────────┴───────────┴────────┴───────────┴─────────────┴────────┴──────────┴──────────┘

Benchmark environment: jelly sim_step! (max_steps=100)
▶ log2p = 5
┌─────────┬───────────┬────────┬───────────┬─────────────┬────────┬──────────┬──────────┐
│ Backend │ WaterLily │ Julia  │ Precision │ Allocations │ GC [%] │ Time [s] │ Speed-up │
├─────────┼───────────┼────────┼───────────┼─────────────┼────────┼──────────┼──────────┤
│   CPUx1 │   afa03a3 │ 1.10.0 │   Float32 │     2263424 │   0.15 │     9.86 │     1.00 │
│   CPUx1 │   d0ec1d2 │ 1.10.0 │   Float32 │     2701319 │   0.17 │    10.94 │     0.90 │
│   CPUx6 │   afa03a3 │ 1.10.0 │   Float32 │     8318725 │   1.44 │     3.72 │     2.65 │
│   CPUx6 │   d0ec1d2 │ 1.10.0 │   Float32 │     9932893 │   1.35 │     4.29 │     2.30 │
│     GPU │   afa03a3 │ 1.10.0 │   Float32 │     4202842 │   1.23 │     1.00 │     9.87 │
│     GPU │   d0ec1d2 │ 1.10.0 │   Float32 │     5081193 │   1.40 │     1.18 │     8.34 │
└─────────┴───────────┴────────┴───────────┴─────────────┴────────┴──────────┴──────────┘
▶ log2p = 6
┌─────────┬───────────┬────────┬───────────┬─────────────┬────────┬──────────┬──────────┐
│ Backend │ WaterLily │ Julia  │ Precision │ Allocations │ GC [%] │ Time [s] │ Speed-up │
├─────────┼───────────┼────────┼───────────┼─────────────┼────────┼──────────┼──────────┤
│   CPUx1 │   afa03a3 │ 1.10.0 │   Float32 │     3192831 │   0.19 │    48.40 │     1.00 │
│   CPUx1 │   d0ec1d2 │ 1.10.0 │   Float32 │     4208670 │   0.22 │    53.38 │     0.91 │
│   CPUx6 │   afa03a3 │ 1.10.0 │   Float32 │    11729531 │   0.84 │    17.86 │     2.71 │
│   CPUx6 │   d0ec1d2 │ 1.10.0 │   Float32 │    15448298 │   0.88 │    20.85 │     2.32 │
│     GPU │   afa03a3 │ 1.10.0 │   Float32 │     5958203 │   0.89 │     2.15 │    22.48 │
│     GPU │   d0ec1d2 │ 1.10.0 │   Float32 │     7945566 │   1.02 │     2.69 │    17.99 │
└─────────┴───────────┴────────┴───────────┴─────────────┴────────┴──────────┴──────────┘

src/Flow.jl

test/runtests.jl

examples/TwoD_SlowStartCircle.jl

marinlauber · 2024-02-14T14:18:46Z

I think this is ready to merge, or do you want me to change other things? @b-fg @weymouth

b-fg · 2024-02-16T15:34:18Z

Looks fine to me.

Update to 1.0.4

marinlauber added 2 commits January 29, 2024 14:41

adjusts BCs for forced flows

167e99c

fixed velocity integral, updated tests and added Slow Start Circle case

85ed332

moving reference frame via velocity function

7d0c016

This comment was marked as outdated.

Sign in to view

updated examples

a9ff87c

marinlauber commented Jan 30, 2024

View reviewed changes

src/Flow.jl Outdated Show resolved Hide resolved

marinlauber added 2 commits February 2, 2024 13:03

combined Tuple and Function into flow.U

93a5818

removed unused Setfield from src/Flow.jl, revert boxflow in runtest a…

24de213

…nd updated documentation in src/Waterlily

b-fg reviewed Feb 2, 2024

View reviewed changes

src/Flow.jl Show resolved Hide resolved

b-fg reviewed Feb 2, 2024

View reviewed changes

src/Flow.jl Outdated Show resolved Hide resolved

Fixed U in accelerate!() and updated tests

95c29a0

b-fg reviewed Feb 2, 2024

View reviewed changes

test/runtests.jl Outdated Show resolved Hide resolved

removed useless ; from test/runtest.jl

afa03a3

fix allowscalar error in test/runtests.jl

415bedd