Use fast `measure(body)` in `measure!(flow)` #155

weymouth · 2024-07-29T12:41:50Z

We don't need normal or velocity measurements outside a very narrow band in the solver, so measure(fast) can be applied there as well, but the band is variable. Changed fast=false to fastd²=Inf and n,V are skipped when d^2>fastd².

Applied in measure!(flow) and tests passing, but I haven't benchmarked yet. Should be a bit faster for the Jelly and will really speed up ParametricBody simulations.

We don't need normal or velocity measurements outside a very narrow band in the solver, so `measure(fast)` can be applied there as well, but the band is variable. Changed `fast` to `fastd²=Inf` and n,V are skipped when `d^2>fastd²`.

b-fg · 2024-07-29T12:44:33Z

You could locally merge #152 into this PR (no need to commit it) and do the benchmark with that :)

weymouth · 2024-07-29T12:44:33Z

src/Body.jl

            for i ∈ 1:N
-                dᵢ,nᵢ,Vᵢ = measure(body,loc(i,I,T),t)
+                dᵢ,nᵢ,Vᵢ = measure(body,loc(i,I,T),t,fastd²=d²)


This line won't do too much, since the if statement above is already catching most of them. I tried to reduce to d=epsilon, but this broke the code.

weymouth · 2024-07-29T12:45:02Z

src/Body.jl

    @fastmath @inline function fill!(μ₀,μ₁,V,d,I)
-        d[I] = sdf(body,loc(0,I,T),t)
-        if abs(d[I])<2+ϵ
+        d[I] = sdf(body,loc(0,I,T),t,fastd²=d²)


This is the important line for ParametricBodies.

weymouth · 2024-07-29T12:47:18Z

You could locally merge #152 into this PR (no need to commit it) and do the benchmark with that :)

But I need to compare with master. How would that work? Also, I'm working from home so I don't have my GPU available today.

b-fg · 2024-07-29T12:53:54Z

Merge that branch into this one and run the benchmarks.
Checkout to that branch and run the benchmarks (I have just updated that branch with master, so you will be basically benchmarking master with this).
Compare benchmarks from 1 and 2.

But yes, ideally we would like to see GPU results too.

--

Or I could also merge that with master already, because I think it is finished now.

b-fg · 2024-07-29T13:02:08Z

In fact, you do need to use that PR because otherwise the changes in measure will not be actually benchmarked. I fixed this bug in that PR.

weymouth · 2024-07-29T20:19:18Z

CPU benchmarks on my machine show a slow down??!??

Benchmark environment: tgv sim_step! (max_steps=100)
▶ log2p = 6
┌─────────┬───────────┬────────┬───────────┬─────────────┬────────┬──────────┬──────────────────┬──────────┐
│ Backend │ WaterLily │ Julia  │ Precision │ Allocations │ GC [%] │ Time [s] │ Cost [ns/DOF/dt] │ Speed-up │
├─────────┼───────────┼────────┼───────────┼─────────────┼────────┼──────────┼──────────────────┼──────────┤
│  CPUx01 │   18690fa │ 1.10.0 │   Float32 │       78733 │   0.00 │    16.86 │           643.06 │     1.00 │
│  CPUx01 │   1dbc2d2 │ 1.10.0 │   Float32 │       78733 │   0.00 │    17.39 │           663.25 │     0.97 │
│  CPUx04 │   18690fa │ 1.10.0 │   Float32 │     2274514 │   0.40 │     9.42 │           359.26 │     1.79 │
│  CPUx04 │   1dbc2d2 │ 1.10.0 │   Float32 │     2274514 │   0.45 │     7.73 │           294.88 │     2.18 │
└─────────┴───────────┴────────┴───────────┴─────────────┴────────┴──────────┴──────────────────┴──────────┘
▶ log2p = 7
┌─────────┬───────────┬────────┬───────────┬─────────────┬────────┬──────────┬──────────────────┬──────────┐
│ Backend │ WaterLily │ Julia  │ Precision │ Allocations │ GC [%] │ Time [s] │ Cost [ns/DOF/dt] │ Speed-up │
├─────────┼───────────┼────────┼───────────┼─────────────┼────────┼──────────┼──────────────────┼──────────┤
│  CPUx01 │   18690fa │ 1.10.0 │   Float32 │       70606 │   0.00 │   117.36 │           559.61 │     1.00 │
│  CPUx01 │   1dbc2d2 │ 1.10.0 │   Float32 │       70606 │   0.00 │   106.49 │           507.78 │     1.10 │
│  CPUx04 │   18690fa │ 1.10.0 │   Float32 │     2021882 │   0.06 │    61.00 │           290.85 │     1.92 │
│  CPUx04 │   1dbc2d2 │ 1.10.0 │   Float32 │     2021882 │   0.07 │    51.39 │           245.06 │     2.28 │
└─────────┴───────────┴────────┴───────────┴─────────────┴────────┴──────────┴──────────────────┴──────────┘
Benchmark environment: jelly sim_step! (max_steps=100)
▶ log2p = 5
┌─────────┬───────────┬────────┬───────────┬─────────────┬────────┬──────────┬──────────────────┬──────────┐
│ Backend │ WaterLily │ Julia  │ Precision │ Allocations │ GC [%] │ Time [s] │ Cost [ns/DOF/dt] │ Speed-up │
├─────────┼───────────┼────────┼───────────┼─────────────┼────────┼──────────┼──────────────────┼──────────┤
│  CPUx01 │   18690fa │ 1.10.0 │   Float32 │      196747 │   0.00 │    15.02 │          1145.60 │     1.00 │
│  CPUx01 │   1dbc2d2 │ 1.10.0 │   Float32 │      196739 │   0.00 │    13.55 │          1034.03 │     1.11 │
│  CPUx04 │   18690fa │ 1.10.0 │   Float32 │     5549636 │   0.54 │    10.56 │           805.81 │     1.42 │
│  CPUx04 │   1dbc2d2 │ 1.10.0 │   Float32 │     5549382 │   0.70 │     8.65 │           659.80 │     1.74 │
└─────────┴───────────┴────────┴───────────┴─────────────┴────────┴──────────┴──────────────────┴──────────┘
▶ log2p = 6
┌─────────┬───────────┬────────┬───────────┬─────────────┬────────┬──────────┬──────────────────┬──────────┐
│ Backend │ WaterLily │ Julia  │ Precision │ Allocations │ GC [%] │ Time [s] │ Cost [ns/DOF/dt] │ Speed-up │
├─────────┼───────────┼────────┼───────────┼─────────────┼────────┼──────────┼──────────────────┼──────────┤
│  CPUx01 │   18690fa │ 1.10.0 │   Float32 │      230016 │   0.00 │   124.63 │          1188.58 │     1.00 │
│  CPUx01 │   1dbc2d2 │ 1.10.0 │   Float32 │      230016 │   0.00 │   111.15 │          1060.02 │     1.12 │
│  CPUx04 │   18690fa │ 1.10.0 │   Float32 │     6552597 │   0.08 │    74.73 │           712.64 │     1.67 │
│  CPUx04 │   1dbc2d2 │ 1.10.0 │   Float32 │     6552597 │   0.08 │    61.22 │           583.83 │     2.04 │
└─────────┴───────────┴────────┴───────────┴─────────────┴────────┴──────────┴──────────────────┴──────────┘

b-fg · 2024-07-30T08:43:39Z

Here are my benchmarks. 1dbc2d2 is "master" (the benchmark PR) and 8d348ef is the current PR. I do not see any major performance speedup or regression.

Benchmark environment: tgv sim_step! (max_steps=100)
▶ log2p = 6
┌────────────┬───────────┬────────┬───────────┬─────────────┬────────┬──────────┬──────────────────┬──────────┐
│  Backend   │ WaterLily │ Julia  │ Precision │ Allocations │ GC [%] │ Time [s] │ Cost [ns/DOF/dt] │ Speed-up │
├────────────┼───────────┼────────┼───────────┼─────────────┼────────┼──────────┼──────────────────┼──────────┤
│     CPUx01 │   1dbc2d2 │ 1.10.4 │   Float32 │       78733 │   0.00 │     9.13 │           348.45 │     1.00 │
│     CPUx01 │   8d348ef │ 1.10.4 │   Float32 │       78733 │   0.00 │     9.34 │           356.17 │     0.98 │
│     CPUx04 │   1dbc2d2 │ 1.10.4 │   Float32 │     2274514 │   0.00 │     3.11 │           118.73 │     2.93 │
│     CPUx04 │   8d348ef │ 1.10.4 │   Float32 │     2274514 │   0.00 │     3.20 │           121.88 │     2.86 │
│     CPUx08 │   1dbc2d2 │ 1.10.4 │   Float32 │     3555070 │   0.00 │     3.19 │           121.53 │     2.87 │
│     CPUx08 │   8d348ef │ 1.10.4 │   Float32 │     3555070 │   0.00 │     3.63 │           138.55 │     2.51 │
│ GPU-NVIDIA │   1dbc2d2 │ 1.10.4 │   Float32 │     2671216 │   0.00 │     0.65 │            24.97 │    13.96 │
│ GPU-NVIDIA │   8d348ef │ 1.10.4 │   Float32 │     2672028 │   0.00 │     0.66 │            25.16 │    13.85 │
└────────────┴───────────┴────────┴───────────┴─────────────┴────────┴──────────┴──────────────────┴──────────┘
▶ log2p = 7
┌────────────┬───────────┬────────┬───────────┬─────────────┬────────┬──────────┬──────────────────┬──────────┐
│  Backend   │ WaterLily │ Julia  │ Precision │ Allocations │ GC [%] │ Time [s] │ Cost [ns/DOF/dt] │ Speed-up │
├────────────┼───────────┼────────┼───────────┼─────────────┼────────┼──────────┼──────────────────┼──────────┤
│     CPUx01 │   1dbc2d2 │ 1.10.4 │   Float32 │       70606 │   0.00 │    73.55 │           350.70 │     1.00 │
│     CPUx01 │   8d348ef │ 1.10.4 │   Float32 │       70606 │   0.00 │    58.92 │           280.96 │     1.25 │
│     CPUx04 │   1dbc2d2 │ 1.10.4 │   Float32 │     2021882 │   0.00 │    18.47 │            88.09 │     3.98 │
│     CPUx04 │   8d348ef │ 1.10.4 │   Float32 │     2021882 │   0.00 │    19.25 │            91.81 │     3.82 │
│     CPUx08 │   1dbc2d2 │ 1.10.4 │   Float32 │     3159482 │   0.00 │    19.09 │            91.01 │     3.85 │
│     CPUx08 │   8d348ef │ 1.10.4 │   Float32 │     3159482 │   0.00 │    19.26 │            91.84 │     3.82 │
│ GPU-NVIDIA │   1dbc2d2 │ 1.10.4 │   Float32 │     2347006 │   0.00 │     3.12 │            14.89 │    23.56 │
│ GPU-NVIDIA │   8d348ef │ 1.10.4 │   Float32 │     2347806 │   0.00 │     3.15 │            15.02 │    23.35 │
└────────────┴───────────┴────────┴───────────┴─────────────┴────────┴──────────┴──────────────────┴──────────┘
Benchmark environment: jelly sim_step! (max_steps=100)
▶ log2p = 5
┌────────────┬───────────┬────────┬───────────┬─────────────┬────────┬──────────┬──────────────────┬──────────┐
│  Backend   │ WaterLily │ Julia  │ Precision │ Allocations │ GC [%] │ Time [s] │ Cost [ns/DOF/dt] │ Speed-up │
├────────────┼───────────┼────────┼───────────┼─────────────┼────────┼──────────┼──────────────────┼──────────┤
│     CPUx01 │   1dbc2d2 │ 1.10.4 │   Float32 │      196731 │   0.00 │     7.77 │           592.83 │     1.00 │
│     CPUx01 │   8d348ef │ 1.10.4 │   Float32 │      196756 │   0.00 │     7.92 │           604.10 │     0.98 │
│     CPUx04 │   1dbc2d2 │ 1.10.4 │   Float32 │     5549128 │   1.54 │     4.55 │           346.84 │     1.71 │
│     CPUx04 │   8d348ef │ 1.10.4 │   Float32 │     5549932 │   0.00 │     4.55 │           346.92 │     1.71 │
│     CPUx08 │   1dbc2d2 │ 1.10.4 │   Float32 │     8644684 │   2.59 │     5.01 │           382.28 │     1.55 │
│     CPUx08 │   8d348ef │ 1.10.4 │   Float32 │     8645956 │   1.04 │     5.06 │           386.02 │     1.54 │
│ GPU-NVIDIA │   1dbc2d2 │ 1.10.4 │   Float32 │     6539350 │   0.00 │     1.49 │           113.83 │     5.21 │
│ GPU-NVIDIA │   8d348ef │ 1.10.4 │   Float32 │     6536576 │   0.00 │     1.46 │           111.14 │     5.33 │
└────────────┴───────────┴────────┴───────────┴─────────────┴────────┴──────────┴──────────────────┴──────────┘
▶ log2p = 6
┌────────────┬───────────┬────────┬───────────┬─────────────┬────────┬──────────┬──────────────────┬──────────┐
│  Backend   │ WaterLily │ Julia  │ Precision │ Allocations │ GC [%] │ Time [s] │ Cost [ns/DOF/dt] │ Speed-up │
├────────────┼───────────┼────────┼───────────┼─────────────┼────────┼──────────┼──────────────────┼──────────┤
│     CPUx01 │   1dbc2d2 │ 1.10.4 │   Float32 │      230016 │   0.00 │    58.94 │           562.14 │     1.00 │
│     CPUx01 │   8d348ef │ 1.10.4 │   Float32 │      230016 │   0.00 │    59.67 │           569.03 │     0.99 │
│     CPUx04 │   1dbc2d2 │ 1.10.4 │   Float32 │     6552597 │   0.37 │    20.24 │           193.04 │     2.91 │
│     CPUx04 │   8d348ef │ 1.10.4 │   Float32 │     6552597 │   0.00 │    21.47 │           204.78 │     2.75 │
│     CPUx08 │   1dbc2d2 │ 1.10.4 │   Float32 │    10222545 │   0.76 │    21.64 │           206.34 │     2.72 │
│     CPUx08 │   8d348ef │ 1.10.4 │   Float32 │    10222545 │   0.66 │    20.88 │           199.14 │     2.82 │
│ GPU-NVIDIA │   1dbc2d2 │ 1.10.4 │   Float32 │     7787222 │   0.00 │     4.69 │            44.76 │    12.56 │
│ GPU-NVIDIA │   8d348ef │ 1.10.4 │   Float32 │     7787224 │   0.00 │     4.67 │            44.57 │    12.61 │
└────────────┴───────────┴────────┴───────────┴─────────────┴────────┴──────────┴──────────────────┴──────────┘

weymouth · 2024-07-30T17:11:59Z

Yeah, that's actually encouraging. It means the current measure function is already optimized for AutoBodies.

For NURBS ParametricBodies, I measured a 100% speed up with this PR, so it's certainly worth doing.

b-fg · 2024-07-31T11:58:19Z

So can we merge this?

Make fast a function and use in measure!(flow)

425844d

We don't need normal or velocity measurements outside a very narrow band in the solver, so `measure(fast)` can be applied there as well, but the band is variable. Changed `fast` to `fastd²=Inf` and n,V are skipped when `d^2>fastd²`.

weymouth commented Jul 29, 2024

View reviewed changes

b-fg changed the title ~~Use fast measure(body) in measure!(flow)~~ Use fast measure(body) in measure!(flow) Jul 29, 2024

weymouth merged commit cf2a247 into master Jul 31, 2024
42 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Use fast `measure(body)` in `measure!(flow)` #155

Use fast `measure(body)` in `measure!(flow)` #155

weymouth commented Jul 29, 2024

b-fg commented Jul 29, 2024 •

edited

Loading

weymouth Jul 29, 2024

weymouth Jul 29, 2024

weymouth commented Jul 29, 2024

b-fg commented Jul 29, 2024 •

edited

Loading

b-fg commented Jul 29, 2024

weymouth commented Jul 29, 2024

b-fg commented Jul 30, 2024

weymouth commented Jul 30, 2024

b-fg commented Jul 31, 2024

Use fast measure(body) in measure!(flow) #155

Use fast measure(body) in measure!(flow) #155

Conversation

weymouth commented Jul 29, 2024

b-fg commented Jul 29, 2024 • edited Loading

weymouth Jul 29, 2024

Choose a reason for hiding this comment

weymouth Jul 29, 2024

Choose a reason for hiding this comment

weymouth commented Jul 29, 2024

b-fg commented Jul 29, 2024 • edited Loading

b-fg commented Jul 29, 2024

weymouth commented Jul 29, 2024

b-fg commented Jul 30, 2024

weymouth commented Jul 30, 2024

b-fg commented Jul 31, 2024

Use fast `measure(body)` in `measure!(flow)` #155

Use fast `measure(body)` in `measure!(flow)` #155

b-fg commented Jul 29, 2024 •

edited

Loading

b-fg commented Jul 29, 2024 •

edited

Loading