Adding an Oceananigans benchmark by jlk9 · Pull Request #2362 · EnzymeAD/Reactant.jl

jlk9 · 2026-02-06T04:49:04Z

This adds an Oceananigans model run to the Reactant benchmark.

github-actions

Remaining comments which cannot be posted as a review comment to avoid GitHub Rate Limit

JuliaFormatter

[JuliaFormatter] _{reported by reviewdog 🐶}

Reactant.jl/benchmark/oceananigans/abernathey_channel.jl

Line 282 in 4a7db89

@trace mincut = true checkpointing = true track_numbers = false for i = 1:100

[JuliaFormatter] _{reported by reviewdog 🐶}

Reactant.jl/benchmark/oceananigans/abernathey_channel.jl

Line 288 in 4a7db89

    
           function run_reentrant_channel_model!(model, Tᵢ, Sᵢ, u_wind_stress, v_wind_stress, temp_flux)

[JuliaFormatter] _{reported by reviewdog 🐶}

Reactant.jl/benchmark/oceananigans/abernathey_channel.jl

Lines 306 to 308 in 4a7db89

    
           function estimate_tracer_error(model, initial_temperature, initial_salinity, u_wind_stress, v_wind_stress, temp_flux, Δz, mld) 
        
               run_reentrant_channel_model!(model, initial_temperature, initial_salinity, u_wind_stress, v_wind_stress, temp_flux)

[JuliaFormatter] _{reported by reviewdog 🐶}

Reactant.jl/benchmark/oceananigans/abernathey_channel.jl

Line 312 in 4a7db89

    
           zonal_transport = (model.velocities.u[x_midpoint,1:Ny,1:Nz] .* model.grid.Δyᵃᶜᵃ) .* Δz

[JuliaFormatter] _{reported by reviewdog 🐶}

Reactant.jl/benchmark/oceananigans/abernathey_channel.jl

Lines 317 to 329 in 4a7db89

    
           function differentiate_tracer_error(model, Tᵢ, Sᵢ, u_wind_stress, v_wind_stress, temp_flux, Δz, mld, 
        
                                              dmodel, dTᵢ, dSᵢ, du_wind_stress, dv_wind_stress, dtemp_flux, dΔz, dmld) 
        
               dedν = autodiff(set_strong_zero(Enzyme.ReverseWithPrimal), 
        
                               estimate_tracer_error, Active, 
        
                               Duplicated(model, dmodel), 
        
                               Duplicated(Tᵢ, dTᵢ), 
        
                               Duplicated(Sᵢ, dSᵢ), 
        
                               Duplicated(u_wind_stress, du_wind_stress), 
        
                               Duplicated(v_wind_stress, dv_wind_stress), 
        
                               Duplicated(temp_flux, dtemp_flux), 
        
                               Duplicated(Δz, dΔz), 
        
                               Duplicated(mld, dmld))

[JuliaFormatter] _{reported by reviewdog 🐶}

Reactant.jl/benchmark/oceananigans/abernathey_channel.jl

Line 342 in 4a7db89

Δt₀ = 2.5minutes

[JuliaFormatter] _{reported by reviewdog 🐶}

Reactant.jl/benchmark/oceananigans/abernathey_channel.jl

Lines 345 to 347 in 4a7db89

    
           grid          = make_grid(architecture, Nx, Ny, Nz, z_faces) 
        
           model         = build_model(grid, Δt₀, parameters) 
        
           T_flux        = T_flux_init(model.grid, parameters)

[JuliaFormatter] _{reported by reviewdog 🐶}

Reactant.jl/benchmark/oceananigans/abernathey_channel.jl

Lines 350 to 352 in 4a7db89

    
           Tᵢ, Sᵢ        = temperature_salinity_init(model.grid, parameters) 
        
           mld           = Field{Center, Center, Nothing}(model.grid) # Not used for now 
        
           Δz            = Reactant.ConcreteRArray(Δz)

[JuliaFormatter] _{reported by reviewdog 🐶}

Reactant.jl/benchmark/oceananigans/abernathey_channel.jl

Lines 356 to 363 in 4a7db89

    
           dmodel         = Enzyme.make_zero(model) 
        
           dTᵢ            = Field{Center, Center, Center}(model.grid) 
        
           dSᵢ            = Field{Center, Center, Center}(model.grid) 
        
           du_wind_stress = Field{Face, Center, Nothing}(model.grid) 
        
           dv_wind_stress = Field{Center, Face, Nothing}(model.grid) 
        
           dT_flux        = Field{Center, Center, Nothing}(model.grid) 
        
           dmld           = Field{Center, Center, Nothing}(model.grid) 
        
           dΔz            = Enzyme.make_zero(Δz)

[JuliaFormatter] _{reported by reviewdog 🐶}

Reactant.jl/benchmark/oceananigans/abernathey_channel.jl

Line 369 in 4a7db89

    
           rspinup_reentrant_channel_model! = @compile raise_first=true raise=true sync=true  spinup_reentrant_channel_model!(model, Tᵢ, Sᵢ, u_wind_stress, v_wind_stress, T_flux)

[JuliaFormatter] _{reported by reviewdog 🐶}

Reactant.jl/benchmark/oceananigans/abernathey_channel.jl

Lines 371 to 372 in 4a7db89

    
           rdifferentiate_tracer_error = @compile raise_first=true raise=true sync=true  differentiate_tracer_error(model, Tᵢ, Sᵢ, u_wind_stress, v_wind_stress, T_flux, Δz, mld, 
        
                                                                                                                   dmodel, dTᵢ, dSᵢ, du_wind_stress, dv_wind_stress, dT_flux, dΔz, dmld)

[JuliaFormatter] _{reported by reviewdog 🐶}

Reactant.jl/benchmark/oceananigans/abernathey_channel.jl

Line 418 in 4a7db89

    
           dedν = rdifferentiate_tracer_error(model, Tᵢ, Sᵢ, u_wind_stress, v_wind_stress, T_flux, Δz, mld, dmodel, dTᵢ, dSᵢ, du_wind_stress, dv_wind_stress, dT_flux, dΔz, dmld)

[JuliaFormatter] _{reported by reviewdog 🐶}

Reactant.jl/benchmark/oceananigans/abernathey_channel.jl

Line 449 in 4a7db89

=#

benchmark/oceananigans/abernathey_channel.jl

jlk9 · 2026-02-06T05:34:11Z

This adds a sample Oceananigans script to the benchmark directory for CI runs. I'm not sure how to update the 'benchmark.yml' file so that this script is run. I know the benchmark directory is listed here:

https://github.com/EnzymeAD/Reactant.jl/blob/9227d57f248250440ef77a25d954d9a3b629e1a2/.github/workflows/benchmark.yml#L16C1-L23C1

but none of the subdirectories are.

Also, I attached the script's dependencies in a separate Project.toml file in the new benchmark/oceananigans subdirectory. Maybe those dependencies should just go into the toml in the benchmark folder itself - let me know if that's better.

@avik-pal @wsmoses

benchmark/oceananigans/abernathey_channel.jl

github-actions

Reactant.jl Benchmarks

Details

Benchmark suite	Current: `9227d57`	Previous: `58e2ca8`	Ratio
`DGCNN [3, 128, 256]/reverse/CPU/DefaultAfterEnzyme`	`0.398922009` s	`0.398070836` s	`1.00`
`jacobi_2d [512, 512, 1024]/primal/CPU/Default_manual_vectorized`	`0.602496558` s	`0.617874608` s	`0.98`
`NewtonSchulz [4096 x 4096]/primal/CPU/Default`	`2.324246022` s	`2.316913006` s	`1.00`
`gesummv [4096]/primal/CPU/Default_manual_vectorized`	`0.064501904` s	`0.063009809` s	`1.02`
`bicg [2048, 4096]/primal/CPU/Default_manual_vectorized`	`0.000840299` s	`0.000743054` s	`1.13`
`NewtonSchulz [1024 x 1024]/primal/CPU/StructuredTensors`	`0.193011207` s	`0.189669149` s	`1.02`
`atax [2048]/primal/CPU/Default_manual_vectorized`	`0.000504998` s	`0.000425077` s	`1.19`
`covariance [2048, 2048]/primal/CPU/Default`	`0.025072572` s	`0.02273972` s	`1.10`
`NewtonSchulz [256 x 256]/primal/CPU/Default`	`0.00745019` s	`0.008141407` s	`0.92`
`NewtonSchulz [1024 x 1024]/primal/CPU/Default`	`0.05022749` s	`0.050128252` s	`1.00`
`bloch_rf [128 spins]/reverse/CPU/Default`	`0.006759542` s	`0.006765288` s	`1.00`
`syrk [2048]/primal/CPU/Julia`	`36.765276186` s	`34.885234172000004` s	`1.05`
`2mm [2048]/primal/CPU/Default_manual_vectorized`	`0.021594917` s	`0.023451378` s	`0.92`
`doitgen [256, 1024, 512]/primal/CPU/Default_manual_vectorized`	`0.074364702` s	`0.076964094` s	`0.97`
`DeepONet ([64, 1024], [1, 128])/reverse/CPU/DefaultAll`	`0.0045181` s	`0.004539621` s	`1.00`
`NewtonSchulz [4096 x 4096]/primal/CPU/StructuredTensors`	`7.32011585` s	`7.309533224` s	`1.00`
`2mm [2048]/primal/CPU/Default`	`0.018274122` s	`0.017374886` s	`1.05`
`doitgen [256, 1024, 512]/primal/CPU/Default`	`0.111491519` s	`0.111187316` s	`1.00`
`bloch_rf [8192 spins]/reverse/CPU/Default`	`0.063466459` s	`0.064630334` s	`0.98`
`bicg [2048, 4096]/primal/CPU/Julia`	`0.055524708000000006` s	`0.053497277` s	`1.04`
`NewtonSchulz [256 x 256]/primal/CPU/StructuredTensors`	`0.009151546` s	`0.010075531` s	`0.91`
`NewtonSchulz [1024 x 1024]/primal/CPU/StructuredTensors (Only Detection)`	`0.150897501` s	`0.151394765` s	`1.00`
`DGCNN [3, 128, 256]/primal/CPU/Default`	`0.084876016` s	`0.084878564` s	`1.00`
`jacobi_1d [2048, 1024]/primal/CPU/Default_manual_vectorized`	`0.007331776` s	`0.006895479` s	`1.06`
`covariance [2048, 2048]/primal/CPU/Default_manual_vectorized`	`0.025170141` s	`0.023050009` s	`1.09`
`atax [2048]/primal/CPU/Julia`	`0.027179259` s	`0.026390141000000002` s	`1.03`
`syrk [2048]/primal/CPU/Default_manual_vectorized`	`0.009459075` s	`0.009623455` s	`0.98`
`correlation [2048, 2048]/primal/CPU/Default`	`0.040496038` s	`0.036684941` s	`1.10`
`correlation [2048, 2048]/primal/CPU/Julia`	`22.956283347000003` s	`22.920312024` s	`1.00`
`DeepONet ([64, 1024], [1, 128])/primal/CPU/Default`	`0.001641677` s	`0.001395163` s	`1.18`
`atax [2048]/primal/CPU/Default`	`0.000503268` s	`0.000499805` s	`1.01`
`DGCNN [3, 128, 256]/reverse/CPU/DefaultBeforeEnzyme`	`0.576965901` s	`0.578087553` s	`1.00`
`NewtonSchulz [4096 x 4096]/primal/CPU/StructuredTensors (Only Detection)`	`5.510028065` s	`5.472108618` s	`1.01`
`correlation [2048, 2048]/primal/CPU/Default_manual_vectorized`	`0.03578499` s	`0.034047847` s	`1.05`
`3mm [256, 1024, 2048, 4096]/primal/CPU/Default`	`0.006792493` s	`0.006639493` s	`1.02`
`bloch_rf [8192 spins]/reverse/CPU/Default_NoBatching`	`0.160148244` s	`0.15287961` s	`1.05`
`heat_3d [128, 128, 128, 256]/primal/CPU/Default`	`0.6286878` s	`0.696132764` s	`0.90`
`bloch_rf [16384 spins]/reverse/CPU/Default`	`0.124762969` s	`0.121696235` s	`1.03`
`bloch_rf [16384 spins]/reverse/CPU/Default_Checkpointing`	`0.226232159` s	`0.228674484` s	`0.99`
`heat_3d [128, 128, 128, 256]/primal/CPU/Default_manual_vectorized`	`0.635809699` s	`0.64505205` s	`0.99`
`DGCNN [3, 128, 256]/reverse/CPU/DisableTransposeReshapeAfterEnzyme`	`0.415978495` s	`0.416717894` s	`1.00`
`gemmver [2048]/primal/CPU/Default_manual_vectorized`	`0.01021365` s	`0.011180861` s	`0.91`
`FNO [64, 64, 1, 4]/reverse/CPU/DefaultBeforeEnzyme`	`0.159975727` s	`0.156646088` s	`1.02`
`2mm [2048]/primal/CPU/Julia`	`56.765553326` s	`57.74098089100001` s	`0.98`
`bloch_rf [128 spins]/reverse/CPU/Default_NoBatching`	`0.011326447` s	`0.010269736` s	`1.10`
`mvt [4096]/primal/CPU/Default_manual_vectorized`	`0.007802637` s	`0.007670336` s	`1.02`
`DGCNN [3, 128, 256]/reverse/CPU/DisableTransposeReshapeBeforeEnzyme`	`0.424982611` s	`0.432458673` s	`0.98`
`gemmver [2048]/primal/CPU/Default`	`0.003459995` s	`0.003356295` s	`1.03`
`gemmver [2048]/primal/CPU/Julia`	`0.036043339` s	`0.034191986` s	`1.05`
`DeepONet ([64, 1024], [1, 128])/reverse/CPU/DefaultBeforeEnzyme`	`0.004545098` s	`0.004537273` s	`1.00`
`bloch_rf [128 spins]/reverse/CPU/Default_NoBatching_Checkpointing`	`0.028099068` s	`0.029549591` s	`0.95`
`bloch_rf [16384 spins]/reverse/CPU/Default_NoBatching`	`0.310841634` s	`0.311803138` s	`1.00`
`gesummv [4096]/primal/CPU/Julia`	`0.376272724` s	`0.37708815100000004` s	`1.00`
`jacobi_1d [2048, 1024]/primal/CPU/Default`	`0.006900504` s	`0.006919381` s	`1.00`
`jacobi_2d [512, 512, 1024]/primal/CPU/Julia`	`1.6186772580000002` s	`1.743826184` s	`0.93`
`NewtonSchulz [1024 x 1024]/primal/CPU/Julia`	`0.090992924` s	`0.096990111` s	`0.94`
`bloch_rf [1024 spins]/reverse/CPU/Default_Checkpointing`	`0.02449167` s	`0.024760363` s	`0.99`
`FNO [64, 64, 1, 4]/reverse/CPU/DefaultAfterEnzyme`	`0.170899858` s	`0.168079487` s	`1.02`
`DGCNN [3, 128, 256]/reverse/CPU/DisableTransposeReshapeAll`	`0.429761764` s	`0.432814088` s	`0.99`
`bloch_rf [128 spins]/reverse/CPU/Julia`	`0.005871141000000001` s	`0.006579594` s	`0.89`
`fdtd_2d [1024, 2048, 256]/primal/CPU/Julia`	`28.566648037` s	`26.824709781000003` s	`1.06`
`bloch_rf [128 spins]/reverse/CPU/Default_Checkpointing`	`0.01823615` s	`0.017465801` s	`1.04`
`DGCNN [3, 128, 256]/reverse/CPU/NoOpt`	`0.431072964` s	`0.428403831` s	`1.01`
`FNO [64, 64, 1, 4]/reverse/CPU/NoOpt`	`0.155458936` s	`0.152382598` s	`1.02`
`gemm [2048, 4096]/primal/CPU/Default`	`0.01723814` s	`0.017096556` s	`1.01`
`jacobi_1d [2048, 1024]/primal/CPU/Julia`	`0.0005626380000000001` s	`0.000568096` s	`0.99`
`mvt [4096]/primal/CPU/Julia`	`0.15012155600000002` s	`0.200703134` s	`0.75`
`bloch_rf [1024 spins]/reverse/CPU/Default`	`0.011250229` s	`0.010971558` s	`1.03`
`bicg [2048, 4096]/primal/CPU/Default`	`0.000836915` s	`0.000698753` s	`1.20`
`doitgen [256, 1024, 512]/primal/CPU/Julia`	`267.821386008` s	`376.430725777` s	`0.71`
`gesummv [4096]/primal/CPU/Default`	`0.002210536` s	`0.001836795` s	`1.20`
`FNO [64, 64, 1, 4]/reverse/CPU/DefaultAll`	`0.169654797` s	`0.168534654` s	`1.01`
`jacobi_2d [512, 512, 1024]/primal/CPU/Default`	`0.625673918` s	`0.6296972` s	`0.99`
`bloch_rf [16384 spins]/reverse/CPU/Julia`	`0.546503321` s	`0.5568492260000001` s	`0.98`
`DeepONet ([64, 1024], [1, 128])/primal/CPU/NoOpt`	`0.001420488` s	`0.00144888` s	`0.98`
`gemm [2048, 4096]/primal/CPU/Julia`	`300.773966141` s	`285.31838474200003` s	`1.05`
`bloch_rf [1024 spins]/reverse/CPU/Julia`	`0.038267074000000005` s	`0.038497590000000005` s	`0.99`
`fdtd_2d [1024, 2048, 256]/primal/CPU/Default`	`0.534963059` s	`0.548909331` s	`0.97`
`DGCNN [3, 128, 256]/primal/CPU/NoOpt`	`0.098497035` s	`0.096644299` s	`1.02`
`fdtd_2d [1024, 2048, 256]/primal/CPU/Default_manual_vectorized`	`0.844111791` s	`0.865483869` s	`0.98`
`covariance [2048, 2048]/primal/CPU/Julia`	`22.94589364` s	`22.946513432` s	`1.00`
`syr2k [2048]/primal/CPU/Default_manual_vectorized`	`0.019910847` s	`0.019253731` s	`1.03`
`DeepONet ([64, 1024], [1, 128])/reverse/CPU/NoOpt`	`0.004097683` s	`0.004432937` s	`0.92`
`DGCNN [3, 128, 256]/reverse/CPU/DefaultAll`	`0.371169488` s	`0.369851723` s	`1.00`
`bloch_rf [8192 spins]/reverse/CPU/Julia`	`0.275172095` s	`0.27385626500000004` s	`1.00`
`bloch_rf [8192 spins]/reverse/CPU/Default_Checkpointing`	`0.118576998` s	`0.119207904` s	`0.99`
`3mm [256, 1024, 2048, 4096]/primal/CPU/Julia`	`15.775128842` s	`15.034657282000001` s	`1.05`
`NewtonSchulz [256 x 256]/primal/CPU/Julia`	`0.0036593940000000003` s	`0.0033585240000000003` s	`1.09`
`NewtonSchulz [256 x 256]/primal/CPU/StructuredTensors (Only Detection)`	`0.008095428` s	`0.009733051` s	`0.83`
`bloch_rf [16384 spins]/reverse/CPU/Default_NoBatching_Checkpointing`	`0.490487018` s	`0.477110539` s	`1.03`
`syr2k [2048]/primal/CPU/Julia`	`39.021939283` s	`39.007653261` s	`1.00`
`heat_3d [128, 128, 128, 256]/primal/CPU/Julia`	`7.8999029830000005` s	`11.672141676` s	`0.68`
`mvt [4096]/primal/CPU/Default`	`0.007581925` s	`0.007498994` s	`1.01`
`bloch_rf [8192 spins]/reverse/CPU/Default_NoBatching_Checkpointing`	`0.236782268` s	`0.235839559` s	`1.00`
`NewtonSchulz [4096 x 4096]/primal/CPU/Julia`	`4.686436541` s	`4.64022502` s	`1.01`
`FNO [64, 64, 1, 4]/primal/CPU/NoOpt`	`0.071988439` s	`0.071582932` s	`1.01`
`DGCNN [3, 128, 256]/primal/CPU/DisableTransposeReshape`	`0.09569326` s	`0.095659247` s	`1.00`
`DeepONet ([64, 1024], [1, 128])/reverse/CPU/DefaultAfterEnzyme`	`0.004559171` s	`0.004409539` s	`1.03`
`FNO [64, 64, 1, 4]/primal/CPU/Default`	`0.070489744` s	`0.070216447` s	`1.00`
`bloch_rf [1024 spins]/reverse/CPU/Default_NoBatching`	`0.025769687` s	`0.023443795` s	`1.10`
`bloch_rf [1024 spins]/reverse/CPU/Default_NoBatching_Checkpointing`	`0.053763732` s	`0.056046882` s	`0.96`
`gemm [2048, 4096]/primal/CPU/Default_manual_vectorized`	`0.019783398` s	`0.020302257` s	`0.97`
`syrk [2048]/primal/CPU/Default`	`0.009323441` s	`0.009327635` s	`1.00`
`3mm [256, 1024, 2048, 4096]/primal/CPU/Default_manual_vectorized`	`0.006767557` s	`0.006828708` s	`0.99`
`syr2k [2048]/primal/CPU/Default`	`0.018834526` s	`0.019498516` s	`0.97`
`bloch_rf [16384 spins]/reverse/CUDA/Default_NoBatching_Checkpointing`	`0.264304473` s	`0.261457892` s	`1.01`
`NewtonSchulz [256 x 256]/primal/CUDA/Default`	`0.000574662` s	`0.000554159` s	`1.04`
`bloch_rf [1024 spins]/reverse/CUDA/Default_NoBatching`	`0.073729375` s	`0.073598594` s	`1.00`
`ViT tiny [256, 256, 3, 4]/reverse/CUDA/DisableScatterGatherPadAll`	`0.010376032` s	`0.010298557` s	`1.01`
`gemmver [2048]/primal/CUDA/Default_manual_vectorized`	`0.000060768` s	`0.000058257` s	`1.04`
`ViT tiny [256, 256, 3, 4]/reverse/CUDA/DisableTransposeReshapeAll`	`0.01054224` s	`0.0104369` s	`1.01`
`syr2k [2048]/primal/CUDA/Default`	`0.000499565` s	`0.000499709` s	`1.00`
`FNO [64, 64, 1, 4]/reverse/CUDA/DefaultAll`	`0.003151842` s	`0.003121849` s	`1.01`
`atax [2048]/primal/CUDA/Default_manual_vectorized`	`0.000025875` s	`0.000025575` s	`1.01`
`gesummv [4096]/primal/CUDA/Default_manual_vectorized`	`0.000206178` s	`0.000206803` s	`1.00`
`fdtd_2d [1024, 2048, 256]/primal/CUDA/Default_manual_vectorized`	`0.028675672` s	`0.027458502` s	`1.04`
`bloch_rf [8192 spins]/reverse/CUDA/Default_NoBatching_Checkpointing`	`0.261291461` s	`0.256914282` s	`1.02`
`VGG11 bn=true [224, 224, 3, 4]/primal/CUDA/Default`	`0.001917002` s	`0.001904037` s	`1.01`
`gemm [2048, 4096]/primal/CUDA/Default`	`0.000446583` s	`0.000445421` s	`1.00`
`bloch_rf [16384 spins]/reverse/CUDA/Default`	`0.034818093` s	`0.034181084` s	`1.02`
`bloch_rf [8192 spins]/reverse/CUDA/Default_Checkpointing`	`0.147638926` s	`0.146332433` s	`1.01`
`ViT tiny [256, 256, 3, 4]/primal/CUDA/NoOpt`	`0.003158564` s	`0.00314326` s	`1.00`
`3mm [256, 1024, 2048, 4096]/primal/CUDA/Default_manual_vectorized`	`0.000159101` s	`0.000159924` s	`0.99`
`jacobi_2d [512, 512, 1024]/primal/CUDA/Default_manual_vectorized`	`0.02091506` s	`0.020097093` s	`1.04`
`NewtonSchulz [1024 x 1024]/primal/CUDA/StructuredTensors`	`0.00701279` s	`0.00686823` s	`1.02`
`ViT tiny [256, 256, 3, 4]/primal/CUDA/Default`	`0.002877501` s	`0.002651615` s	`1.09`
`DGCNN [3, 128, 256]/primal/CUDA/DisableTransposeReshape`	`0.001247968` s	`0.001242163` s	`1.00`
`ViT tiny [256, 256, 3, 4]/reverse/CUDA/DisableScatterGatherAll`	`0.010460293` s	`0.01062204` s	`0.98`
`DGCNN [3, 128, 256]/reverse/CUDA/DisableTransposeReshapeAfterEnzyme`	`0.003393965` s	`0.003389311` s	`1.00`
`bloch_rf [16384 spins]/reverse/CUDA/Default_Checkpointing`	`0.145979048` s	`0.146557366` s	`1.00`
`DeepONet ([64, 1024], [1, 128])/reverse/CUDA/DefaultAll`	`0.000527248` s	`0.000575929` s	`0.92`
`bloch_rf [128 spins]/reverse/CUDA/Default`	`0.037041434` s	`0.034459056` s	`1.07`
`gemm [2048, 4096]/primal/CUDA/Default_manual_vectorized`	`0.00044178` s	`0.000442725` s	`1.00`
`DeepONet ([64, 1024], [1, 128])/primal/CUDA/Default`	`0.000162614` s	`0.000213511` s	`0.76`
`DGCNN [3, 128, 256]/reverse/CUDA/DisableTransposeReshapeAll`	`0.003345437` s	`0.003340038` s	`1.00`
`VGG11 bn=true [224, 224, 3, 4]/reverse/CUDA/DefaultAfterEnzyme`	`0.007075102` s	`0.007045761` s	`1.00`
`NewtonSchulz [4096 x 4096]/primal/CUDA/StructuredTensors (Only Detection)`	`0.109337738` s	`0.109336512` s	`1.00`
`covariance [2048, 2048]/primal/CUDA/Default_manual_vectorized`	`0.000262827` s	`0.000263588` s	`1.00`
`NewtonSchulz [4096 x 4096]/primal/CUDA/StructuredTensors`	`0.109759335` s	`0.110324879` s	`0.99`
`NewtonSchulz [4096 x 4096]/primal/CUDA/Default`	`0.043339614` s	`0.043750541` s	`0.99`
`correlation [2048, 2048]/primal/CUDA/Default`	`0.000295359` s	`0.000296845` s	`0.99`
`3mm [256, 1024, 2048, 4096]/primal/CUDA/Default`	`0.000158985` s	`0.000161757` s	`0.98`
`doitgen [256, 1024, 512]/primal/CUDA/Default_manual_vectorized`	`0.001468615` s	`0.001470557` s	`1.00`
`jacobi_1d [2048, 1024]/primal/CUDA/Default_manual_vectorized`	`0.024257626` s	`0.024096308` s	`1.01`
`VGG11 bn=true [224, 224, 3, 4]/reverse/CUDA/DefaultBeforeEnzyme`	`0.007131914` s	`0.007119492` s	`1.00`
`DGCNN [3, 128, 256]/primal/CUDA/NoOpt`	`0.001266264` s	`0.001263411` s	`1.00`
`covariance [2048, 2048]/primal/CUDA/Default`	`0.000283074` s	`0.000286435` s	`0.99`
`ViT tiny [256, 256, 3, 4]/reverse/CUDA/DisablePadAll`	`0.010433047` s	`0.010435377` s	`1.00`
`VGG11 bn=true [224, 224, 3, 4]/reverse/CUDA/NoOpt`	`0.007182739` s	`0.007157338` s	`1.00`
`FNO [64, 64, 1, 4]/reverse/CUDA/DefaultAfterEnzyme`	`0.003252567` s	`0.003126927` s	`1.04`
`DGCNN [3, 128, 256]/reverse/CUDA/DefaultAfterEnzyme`	`0.00303201` s	`0.003024562` s	`1.00`
`DGCNN [3, 128, 256]/reverse/CUDA/DefaultAll`	`0.003520211` s	`0.003537342` s	`1.00`
`heat_3d [128, 128, 128, 256]/primal/CUDA/Default`	`0.013342762` s	`0.013134791` s	`1.02`
`mvt [4096]/primal/CUDA/Default_manual_vectorized`	`0.000107014` s	`0.000110445` s	`0.97`
`VGG11 bn=true [224, 224, 3, 4]/reverse/CUDA/DefaultAll`	`0.007069624` s	`0.007038347` s	`1.00`
`doitgen [256, 1024, 512]/primal/CUDA/Default`	`0.002241755` s	`0.002241956` s	`1.00`
`DGCNN [3, 128, 256]/primal/CUDA/Default`	`0.000974305` s	`0.000971581` s	`1.00`
`VGG11 bn=true [224, 224, 3, 4]/primal/CUDA/NoOpt`	`0.001939534` s	`0.001926944` s	`1.01`
`bloch_rf [1024 spins]/reverse/CUDA/Default_NoBatching_Checkpointing`	`0.261730497` s	`0.257782645` s	`1.02`
`DeepONet ([64, 1024], [1, 128])/reverse/CUDA/NoOpt`	`0.000634165` s	`0.000644931` s	`0.98`
`syrk [2048]/primal/CUDA/Default_manual_vectorized`	`0.000254876` s	`0.000255133` s	`1.00`
`gemmver [2048]/primal/CUDA/Default`	`0.000044986` s	`0.00004434` s	`1.01`
`NewtonSchulz [256 x 256]/primal/CUDA/StructuredTensors`	`0.001436332` s	`0.001411867` s	`1.02`
`DGCNN [3, 128, 256]/reverse/CUDA/NoOpt`	`0.006571116` s	`0.006575345` s	`1.00`
`bloch_rf [128 spins]/reverse/CUDA/Default_Checkpointing`	`0.147163896` s	`0.151287582` s	`0.97`
`bicg [2048, 4096]/primal/CUDA/Default`	`0.000055485` s	`0.00006569` s	`0.84`
`DGCNN [3, 128, 256]/reverse/CUDA/DefaultBeforeEnzyme`	`0.006508961` s	`0.006538191` s	`1.00`
`bicg [2048, 4096]/primal/CUDA/Default_manual_vectorized`	`0.000053565` s	`0.000054109` s	`0.99`
`2mm [2048]/primal/CUDA/Default`	`0.000462545` s	`0.000462548` s	`1.00`
`jacobi_1d [2048, 1024]/primal/CUDA/Default`	`0.020627327` s	`0.020134235` s	`1.02`
`heat_3d [128, 128, 128, 256]/primal/CUDA/Default_manual_vectorized`	`0.012847875` s	`0.013123787` s	`0.98`
`bloch_rf [16384 spins]/reverse/CUDA/Default_NoBatching`	`0.083654162` s	`0.082095426` s	`1.02`
`jacobi_2d [512, 512, 1024]/primal/CUDA/Default`	`0.020403413` s	`0.020061832` s	`1.02`
`correlation [2048, 2048]/primal/CUDA/Default_manual_vectorized`	`0.000268405` s	`0.000268627` s	`1.00`
`bloch_rf [8192 spins]/reverse/CUDA/Default_NoBatching`	`0.072826295` s	`0.07273783` s	`1.00`
`bloch_rf [1024 spins]/reverse/CUDA/Default_Checkpointing`	`0.147529895` s	`0.145460278` s	`1.01`
`DeepONet ([64, 1024], [1, 128])/reverse/CUDA/DefaultBeforeEnzyme`	`0.000521865` s	`0.000581555` s	`0.90`
`2mm [2048]/primal/CUDA/Default_manual_vectorized`	`0.000465735` s	`0.000465898` s	`1.00`
`ViT tiny [256, 256, 3, 4]/reverse/CUDA/NoOpt`	`0.01163171` s	`0.010929207` s	`1.06`
`FNO [64, 64, 1, 4]/reverse/CUDA/DefaultBeforeEnzyme`	`0.003194712` s	`0.003147606` s	`1.01`
`bloch_rf [1024 spins]/reverse/CUDA/Default`	`0.036278792` s	`0.035616641` s	`1.02`
`NewtonSchulz [1024 x 1024]/primal/CUDA/Default`	`0.001847829` s	`0.001829829` s	`1.01`
`atax [2048]/primal/CUDA/Default`	`0.000029872` s	`0.000025562` s	`1.17`
`gesummv [4096]/primal/CUDA/Default`	`0.000108406` s	`0.000106736` s	`1.02`
`ViT tiny [256, 256, 3, 4]/reverse/CUDA/DefaultAll`	`0.010504191` s	`0.010640459` s	`0.99`
`syr2k [2048]/primal/CUDA/Default_manual_vectorized`	`0.0004996` s	`0.000499725` s	`1.00`
`DGCNN [3, 128, 256]/reverse/CUDA/DisableTransposeReshapeBeforeEnzyme`	`0.014071049` s	`0.014599092` s	`0.96`
`bloch_rf [128 spins]/reverse/CUDA/Default_NoBatching`	`0.076232127` s	`0.07602372` s	`1.00`
`FNO [64, 64, 1, 4]/reverse/CUDA/NoOpt`	`0.003336913` s	`0.00336687` s	`0.99`
`bloch_rf [8192 spins]/reverse/CUDA/Default`	`0.035577678` s	`0.034350543` s	`1.04`
`FNO [64, 64, 1, 4]/primal/CUDA/NoOpt`	`0.001149151` s	`0.001131692` s	`1.02`
`DeepONet ([64, 1024], [1, 128])/reverse/CUDA/DefaultAfterEnzyme`	`0.000517677` s	`0.000545863` s	`0.95`
`NewtonSchulz [256 x 256]/primal/CUDA/StructuredTensors (Only Detection)`	`0.001369076` s	`0.001349102` s	`1.01`
`DeepONet ([64, 1024], [1, 128])/primal/CUDA/NoOpt`	`0.00021077` s	`0.000212637` s	`0.99`
`mvt [4096]/primal/CUDA/Default`	`0.000107007` s	`0.000109972` s	`0.97`
`syrk [2048]/primal/CUDA/Default`	`0.000255953` s	`0.000256032` s	`1.00`
`fdtd_2d [1024, 2048, 256]/primal/CUDA/Default`	`0.025073715` s	`0.025553858` s	`0.98`
`NewtonSchulz [1024 x 1024]/primal/CUDA/StructuredTensors (Only Detection)`	`0.005458217` s	`0.005679814` s	`0.96`
`bloch_rf [128 spins]/reverse/CUDA/Default_NoBatching_Checkpointing`	`0.415549826` s	`0.399026405` s	`1.04`
`FNO [64, 64, 1, 4]/primal/CUDA/Default`	`0.00108504` s	`0.00107467` s	`1.01`
`ViT tiny [256, 256, 3, 4]/primal/TPU/Default`	`0.000217708` s	`0.000217818` s	`1.00`
`NewtonSchulz [4096 x 4096]/primal/TPU/Default`	`0.024508124` s	`0.024705897` s	`0.99`
`bloch_rf [1024 spins]/reverse/TPU/Default_Checkpointing`	`0.003438832` s	`0.003438864` s	`1.00`
`FNO [64, 64, 1, 4]/reverse/TPU/DefaultAll`	`0.003093378` s	`0.003093037` s	`1.00`
`DGCNN [3, 128, 256]/reverse/TPU/DisableTransposeReshapeAll`	`0.005004825` s	`0.005005214` s	`1.00`
`syrk [2048]/primal/TPU/Default_manual_vectorized`	`0.000031548` s	`0.000031535` s	`1.00`
`bloch_rf [16384 spins]/reverse/TPU/Default_NoBatching`	`0.047648354` s	`0.047636619` s	`1.00`
`ViT tiny [256, 256, 3, 4]/primal/TPU/NoOpt`	`0.000586544` s	`0.000586181` s	`1.00`
`FNO [64, 64, 1, 4]/primal/TPU/NoOpt`	`0.001138321` s	`0.001138033` s	`1.00`
`gemm [2048, 4096]/primal/TPU/Default`	`0.000072625` s	`0.000072509` s	`1.00`
`3mm [256, 1024, 2048, 4096]/primal/TPU/Default_manual_vectorized`	`0.000016375` s	`0.000016401` s	`1.00`
`atax [2048]/primal/TPU/Default_manual_vectorized`	`0.000024208` s	`0.000024146` s	`1.00`
`ViT tiny [256, 256, 3, 4]/reverse/TPU/DisableScatterGatherAll`	`0.001698505` s	`0.001698623` s	`1.00`
`syr2k [2048]/primal/TPU/Default`	`0.000058343` s	`0.000058219` s	`1.00`
`2mm [2048]/primal/TPU/Default_manual_vectorized`	`0.000074958` s	`0.000074929` s	`1.00`
`FNO [64, 64, 1, 4]/primal/TPU/Default`	`0.000953001` s	`0.000953054` s	`1.00`
`VGG11 bn=true [224, 224, 3, 4]/reverse/TPU/DefaultBeforeEnzyme`	`0.004179579` s	`0.004179874` s	`1.00`
`DeepONet ([64, 1024], [1, 128])/primal/TPU/Default`	`0.000006047` s	`0.000006087` s	`0.99`
`doitgen [256, 1024, 512]/primal/TPU/Default_manual_vectorized`	`0.001088949` s	`0.001086157` s	`1.00`
`ViT tiny [256, 256, 3, 4]/reverse/TPU/DisableScatterGatherPadAll`	`0.001697239` s	`0.001697514` s	`1.00`
`doitgen [256, 1024, 512]/primal/TPU/Default`	`0.001579798` s	`0.001579097` s	`1.00`
`DGCNN [3, 128, 256]/primal/TPU/DisableTransposeReshape`	`0.002865498` s	`0.002865397` s	`1.00`
`bloch_rf [1024 spins]/reverse/TPU/Default`	`0.001323008` s	`0.001323031` s	`1.00`
`FNO [64, 64, 1, 4]/reverse/TPU/DefaultAfterEnzyme`	`0.003085456` s	`0.003085515` s	`1.00`
`bloch_rf [128 spins]/reverse/TPU/Default`	`0.000694924` s	`0.000694866` s	`1.00`
`mvt [4096]/primal/TPU/Default`	`0.000045023` s	`0.000045082` s	`1.00`
`bloch_rf [16384 spins]/reverse/TPU/Default`	`0.026236881` s	`0.026240626` s	`1.00`
`gemmver [2048]/primal/TPU/Default`	`0.000036403` s	`0.00003654` s	`1.00`
`FNO [64, 64, 1, 4]/reverse/TPU/NoOpt`	`0.002959794` s	`0.002960119` s	`1.00`
`syr2k [2048]/primal/TPU/Default_manual_vectorized`	`0.000057948` s	`0.000057998` s	`1.00`
`bloch_rf [8192 spins]/reverse/TPU/Default_NoBatching`	`0.033746967` s	`0.033754872` s	`1.00`
`bloch_rf [8192 spins]/reverse/TPU/Default_Checkpointing`	`0.019418776` s	`0.0194229` s	`1.00`
`fdtd_2d [1024, 2048, 256]/primal/TPU/Default`	`0.018833155` s	`0.018833106` s	`1.00`
`covariance [2048, 2048]/primal/TPU/Default_manual_vectorized`	`0.000047606` s	`0.000047646` s	`1.00`
`heat_3d [128, 128, 128, 256]/primal/TPU/Default`	`0.26175386` s	`0.261753897` s	`1.00`
`bloch_rf [128 spins]/reverse/TPU/Default_NoBatching_Checkpointing`	`0.017732083` s	`0.01791452` s	`0.99`
`NewtonSchulz [256 x 256]/primal/TPU/StructuredTensors (Only Detection)`	`0.000019731` s	`0.000019735` s	`1.00`
`gemmver [2048]/primal/TPU/Default_manual_vectorized`	`0.000037325` s	`0.000037471` s	`1.00`
`VGG11 bn=true [224, 224, 3, 4]/reverse/TPU/NoOpt`	`0.004056119` s	`0.004055074` s	`1.00`
`DGCNN [3, 128, 256]/reverse/TPU/DefaultBeforeEnzyme`	`0.004752407` s	`0.004751814` s	`1.00`
`bicg [2048, 4096]/primal/TPU/Default_manual_vectorized`	`0.000023459` s	`0.000023456` s	`1.00`
`NewtonSchulz [1024 x 1024]/primal/TPU/StructuredTensors`	`0.00022266` s	`0.0002227` s	`1.00`
`ViT tiny [256, 256, 3, 4]/reverse/TPU/DisablePadAll`	`0.00169829` s	`0.001697866` s	`1.00`
`bloch_rf [1024 spins]/reverse/TPU/Default_NoBatching_Checkpointing`	`0.018718537` s	`0.019394475` s	`0.97`
`3mm [256, 1024, 2048, 4096]/primal/TPU/Default`	`0.000016374` s	`0.000016327` s	`1.00`
`jacobi_1d [2048, 1024]/primal/TPU/Default`	`0.008143506` s	`0.008144584` s	`1.00`
`DGCNN [3, 128, 256]/reverse/TPU/DefaultAfterEnzyme`	`0.004670825` s	`0.00467102` s	`1.00`
`ViT tiny [256, 256, 3, 4]/reverse/TPU/DisableTransposeReshapeAll`	`0.00143745` s	`0.001437695` s	`1.00`
`syrk [2048]/primal/TPU/Default`	`0.000030314` s	`0.000030265` s	`1.00`
`correlation [2048, 2048]/primal/TPU/Default_manual_vectorized`	`0.000053189` s	`0.000053206` s	`1.00`
`bloch_rf [128 spins]/reverse/TPU/Default_Checkpointing`	`0.001658265` s	`0.001658484` s	`1.00`
`DGCNN [3, 128, 256]/primal/TPU/NoOpt`	`0.002867399` s	`0.002867601` s	`1.00`
`bloch_rf [16384 spins]/reverse/TPU/Default_NoBatching_Checkpointing`	`0.035670511` s	`0.03566848` s	`1.00`
`NewtonSchulz [1024 x 1024]/primal/TPU/StructuredTensors (Only Detection)`	`0.000222784` s	`0.000222713` s	`1.00`
`VGG11 bn=true [224, 224, 3, 4]/primal/TPU/Default`	`0.000813696` s	`0.000813646` s	`1.00`
`NewtonSchulz [256 x 256]/primal/TPU/Default`	`0.000018987` s	`0.000018965` s	`1.00`
`jacobi_1d [2048, 1024]/primal/TPU/Default_manual_vectorized`	`0.005707607` s	`0.005708007` s	`1.00`
`FNO [64, 64, 1, 4]/reverse/TPU/DefaultBeforeEnzyme`	`0.003093487` s	`0.003093686` s	`1.00`
`bloch_rf [8192 spins]/reverse/TPU/Default_NoBatching_Checkpointing`	`0.024888531` s	`0.024891187` s	`1.00`
`DeepONet ([64, 1024], [1, 128])/reverse/TPU/DefaultBeforeEnzyme`	`0.000027333` s	`0.00002735` s	`1.00`
`VGG11 bn=true [224, 224, 3, 4]/reverse/TPU/DefaultAfterEnzyme`	`0.004180059` s	`0.004180592` s	`1.00`
`gesummv [4096]/primal/TPU/Default`	`0.000087436` s	`0.000087412` s	`1.00`
`covariance [2048, 2048]/primal/TPU/Default`	`0.000051871` s	`0.000051776` s	`1.00`
`gesummv [4096]/primal/TPU/Default_manual_vectorized`	`0.000087393` s	`0.000087481` s	`1.00`
`ViT tiny [256, 256, 3, 4]/reverse/TPU/NoOpt`	`0.002020693` s	`0.002020436` s	`1.00`
`DGCNN [3, 128, 256]/reverse/TPU/NoOpt`	`0.005226905` s	`0.005227528` s	`1.00`
`NewtonSchulz [1024 x 1024]/primal/TPU/Default`	`0.00022076` s	`0.00022078` s	`1.00`
`bicg [2048, 4096]/primal/TPU/Default`	`0.00002348` s	`0.000023427` s	`1.00`
`fdtd_2d [1024, 2048, 256]/primal/TPU/Default_manual_vectorized`	`0.027117804` s	`0.027117819` s	`1.00`
`jacobi_2d [512, 512, 1024]/primal/TPU/Default_manual_vectorized`	`0.02207921` s	`0.022079209` s	`1.00`
`jacobi_2d [512, 512, 1024]/primal/TPU/Default`	`0.0266761` s	`0.026676079` s	`1.00`
`gemm [2048, 4096]/primal/TPU/Default_manual_vectorized`	`0.000072719` s	`0.000072801` s	`1.00`
`ViT tiny [256, 256, 3, 4]/reverse/TPU/DefaultAll`	`0.001698392` s	`0.001698055` s	`1.00`
`NewtonSchulz [4096 x 4096]/primal/TPU/StructuredTensors (Only Detection)`	`0.024553534` s	`0.02476967` s	`0.99`
`2mm [2048]/primal/TPU/Default`	`0.000086594` s	`0.000086593` s	`1.00`
`NewtonSchulz [256 x 256]/primal/TPU/StructuredTensors`	`0.000019743` s	`0.000019746` s	`1.00`
`bloch_rf [16384 spins]/reverse/TPU/Default_Checkpointing`	`0.085420627` s	`0.08541824` s	`1.00`
`DGCNN [3, 128, 256]/primal/TPU/Default`	`0.002349152` s	`0.002349392` s	`1.00`
`DeepONet ([64, 1024], [1, 128])/reverse/TPU/DefaultAfterEnzyme`	`0.000027326` s	`0.00002731` s	`1.00`
`DeepONet ([64, 1024], [1, 128])/primal/TPU/NoOpt`	`0.000006327` s	`0.000006281` s	`1.01`
`atax [2048]/primal/TPU/Default`	`0.000024192` s	`0.000024112` s	`1.00`
`DGCNN [3, 128, 256]/reverse/TPU/DisableTransposeReshapeAfterEnzyme`	`0.005179774` s	`0.005180388` s	`1.00`
`DGCNN [3, 128, 256]/reverse/TPU/DefaultAll`	`0.004673113` s	`0.004672252` s	`1.00`
`DeepONet ([64, 1024], [1, 128])/reverse/TPU/NoOpt`	`0.000027424` s	`0.000027524` s	`1.00`
`VGG11 bn=true [224, 224, 3, 4]/reverse/TPU/DefaultAll`	`0.004180245` s	`0.004180778` s	`1.00`
`DeepONet ([64, 1024], [1, 128])/reverse/TPU/DefaultAll`	`0.000027259` s	`0.000027323` s	`1.00`
`VGG11 bn=true [224, 224, 3, 4]/primal/TPU/NoOpt`	`0.000861191` s	`0.000861071` s	`1.00`
`DGCNN [3, 128, 256]/reverse/TPU/DisableTransposeReshapeBeforeEnzyme`	`0.005004828` s	`0.005004994` s	`1.00`
`mvt [4096]/primal/TPU/Default_manual_vectorized`	`0.000045565` s	`0.000045534` s	`1.00`
`bloch_rf [128 spins]/reverse/TPU/Default_NoBatching`	`0.006970248` s	`0.006971372` s	`1.00`
`bloch_rf [1024 spins]/reverse/TPU/Default_NoBatching`	`0.007860035` s	`0.007846746` s	`1.00`
`NewtonSchulz [4096 x 4096]/primal/TPU/StructuredTensors`	`0.024573923` s	`0.024616768` s	`1.00`
`bloch_rf [8192 spins]/reverse/TPU/Default`	`0.016541352` s	`0.016542545` s	`1.00`
`heat_3d [128, 128, 128, 256]/primal/TPU/Default_manual_vectorized`	`0.261753842` s	`0.261753882` s	`1.00`
`correlation [2048, 2048]/primal/TPU/Default`	`0.000055706` s	`0.000055745` s	`1.00`

This comment was automatically generated by workflow using github-action-benchmark.

benchmark/oceananigans/abernathey_channel.jl

jlk9 · 2026-02-06T14:28:53Z

Additionally, checkpointing is set for this for loop:
https://github.com/EnzymeAD/Reactant.jl/blob/9227d57f248250440ef77a25d954d9a3b629e1a2/benchmark/oceananigans/abernathey_channel.jl#L280C1-L286C4

To enable/disable checkpointing you need to change the checkpointing = true line there.

wsmoses · 2026-02-11T05:21:42Z

- Run `import Pkg; Pkg.add("Chairmarks")` to install the Chairmarks package.

wsmoses · 2026-02-11T13:30:59Z

@jlk9 seems like a setup error?


ERROR: LoadError: MethodError: no method matching Duplicated(::Reactant.TracedRArray{Float64, 2}, ::Matrix{Float64})
The type `Duplicated` exists, but no method is defined for this combination of argument types when trying to construct it.

Closest candidates are:
  Duplicated(::T1, ::T1) where T1
   @ EnzymeCore ~/.julia/packages/EnzymeCore/RpjpI/src/EnzymeCore.jl:68
  Duplicated(::T1, ::T1, ::Bool) where T1
   @ EnzymeCore ~/.julia/packages/EnzymeCore/RpjpI/src/EnzymeCore.jl:68
  Duplicated(::T1, ::T1) where T1<:SubArray
   @ EnzymeCore ~/.julia/packages/EnzymeCore/RpjpI/src/EnzymeCore.jl:69
  ...

Stacktrace:
  [1] macro expansion
    @ /__w/Reactant.jl/Reactant.jl/src/utils.jl:0 [inlined]
  [2] call_with_reactant(::Reactant.EnsureReturnType{Union{}}, ::Type{Duplicated}, ::Reactant.TracedRArray{Float64, 2}, ::Matrix{Float64})
    @ Reactant /__w/Reactant.jl/Reactant.jl/src/utils.jl:1177
  [3] call_with_reactant(::typeof(Main.AbernatheyChannel.differentiate_tracer_error),

avik-pal · 2026-02-11T13:59:40Z

That's a screwup from me

wsmoses · 2026-02-11T14:38:09Z

benchmark/oceananigans/abernathey_channel.jl

+
+function loop!(model)
+    Δt = model.clock.last_Δt
+    @trace mincut = true checkpointing = true track_numbers = false for i in 1:100


potentially for a follow up, @avik-pal we should try testing checkpointing on vs off, and also optimizations on vs off

jlk9 · 2026-02-12T04:23:34Z

@avik-pal I think the benchmark has too many timesteps in the autodiff run. Locally having 10 steps in loop! passes but 100 (what it is in this PR now) fails. I'm testing if 10 passes on the reduced benchmark.

github-actions bot reviewed Feb 6, 2026

View reviewed changes

giordano reviewed Feb 6, 2026

View reviewed changes

benchmark/oceananigans/abernathey_channel.jl Outdated Show resolved Hide resolved

avik-pal force-pushed the jlk9/oceananigans-benchmark branch 4 times, most recently from f6ba7a4 to cb73cf3 Compare February 11, 2026 03:03

avik-pal force-pushed the jlk9/oceananigans-benchmark branch 4 times, most recently from da19e01 to 0c7728a Compare February 11, 2026 11:48

wsmoses reviewed Feb 11, 2026

View reviewed changes

avik-pal force-pushed the jlk9/oceananigans-benchmark branch 2 times, most recently from fa36b03 to 3e6cdda Compare February 11, 2026 15:06

avik-pal and others added 10 commits February 11, 2026 17:56

refactor: run benchmarks in separate environments

c46b933

fix: dont run reactant on main process

47c175e

Adding an Oceananigans benchmark

3306a3a

Trimming Project.toml for Oceananigans benchmark

dd9173d

fix: merge projects

44020de

perf: setup runbenchmarks.jl

a62d82d

perf: setup to use profiler

f7d22d1

fix: missing dep

c60051b

feat: tflops

e8efabd

fix: array type

de628d6

perf: only run oceananigans (drop me)

c8e8494

avik-pal force-pushed the jlk9/oceananigans-benchmark branch from 3e6cdda to c8e8494 Compare February 11, 2026 22:57

jlk9 mentioned this pull request Feb 12, 2026

Testing reduced Oceananigans script. DO NOT MERGE #2425

Merged

	function estimate_tracer_error(model, initial_temperature, initial_salinity, u_wind_stress, v_wind_stress, temp_flux, Δz, mld)
	run_reentrant_channel_model!(model, initial_temperature, initial_salinity, u_wind_stress, v_wind_stress, temp_flux)

	function differentiate_tracer_error(model, Tᵢ, Sᵢ, u_wind_stress, v_wind_stress, temp_flux, Δz, mld,
	dmodel, dTᵢ, dSᵢ, du_wind_stress, dv_wind_stress, dtemp_flux, dΔz, dmld)

	dedν = autodiff(set_strong_zero(Enzyme.ReverseWithPrimal),
	estimate_tracer_error, Active,
	Duplicated(model, dmodel),
	Duplicated(Tᵢ, dTᵢ),
	Duplicated(Sᵢ, dSᵢ),
	Duplicated(u_wind_stress, du_wind_stress),
	Duplicated(v_wind_stress, dv_wind_stress),
	Duplicated(temp_flux, dtemp_flux),
	Duplicated(Δz, dΔz),
	Duplicated(mld, dmld))

	grid = make_grid(architecture, Nx, Ny, Nz, z_faces)
	model = build_model(grid, Δt₀, parameters)
	T_flux = T_flux_init(model.grid, parameters)

	Tᵢ, Sᵢ = temperature_salinity_init(model.grid, parameters)
	mld = Field{Center, Center, Nothing}(model.grid) # Not used for now
	Δz = Reactant.ConcreteRArray(Δz)

	dmodel = Enzyme.make_zero(model)
	dTᵢ = Field{Center, Center, Center}(model.grid)
	dSᵢ = Field{Center, Center, Center}(model.grid)
	du_wind_stress = Field{Face, Center, Nothing}(model.grid)
	dv_wind_stress = Field{Center, Face, Nothing}(model.grid)
	dT_flux = Field{Center, Center, Nothing}(model.grid)
	dmld = Field{Center, Center, Nothing}(model.grid)
	dΔz = Enzyme.make_zero(Δz)

	rdifferentiate_tracer_error = @compile raise_first=true raise=true sync=true differentiate_tracer_error(model, Tᵢ, Sᵢ, u_wind_stress, v_wind_stress, T_flux, Δz, mld,
	dmodel, dTᵢ, dSᵢ, du_wind_stress, dv_wind_stress, dT_flux, dΔz, dmld)

Conversation

jlk9 commented Feb 6, 2026

Uh oh!

github-actions bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

jlk9 commented Feb 6, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

github-actions bot left a comment

Choose a reason for hiding this comment

Reactant.jl Benchmarks

Uh oh!

Uh oh!

jlk9 commented Feb 6, 2026

Uh oh!

wsmoses commented Feb 11, 2026

Uh oh!

wsmoses commented Feb 11, 2026

Uh oh!

avik-pal commented Feb 11, 2026

Uh oh!

wsmoses Feb 11, 2026

Choose a reason for hiding this comment

Uh oh!

jlk9 commented Feb 12, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants