Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Refactor stencil kernels to show mem bandwidth #1968

Merged
merged 1 commit into from
Sep 3, 2024
Merged

Refactor stencil kernels to show mem bandwidth #1968

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
19 changes: 15 additions & 4 deletions test/Operators/finitedifference/benchmark_stencils.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,11 +6,22 @@ include("benchmark_stencils_utils.jl")

#! format: off
@testset "Benchmark operators" begin
# benchmark_operators_column(Float64; z_elems = 63, helem = 30, Nq = 4, compile = true)
benchmark_operators_column(Float64; z_elems = 63, helem = 30, Nq = 4)
# column_benchmark_arrays(device, z_elems = 63, bm.float_type)
# sphere_benchmark_arrays(device, z_elems = 63, helem = 30, Nq = 4, bm.float_type)

# benchmark_operators_sphere(Float64; z_elems = 63, helem = 30, Nq = 4, compile = true)
benchmark_operators_sphere(Float64; z_elems = 63, helem = 30, Nq = 4)
bm = Benchmark(;float_type = Float64, device_name)
# benchmark_operators_column(bm; z_elems = 63, helem = 30, Nq = 4, compile = true)
(;t_min) = benchmark_operators_column(bm; z_elems = 63, helem = 30, Nq = 4)
test_results_column(t_min)

bm = Benchmark(;float_type = Float64, device_name)
# benchmark_operators_sphere(bm; z_elems = 63, helem = 30, Nq = 4, compile = true)
(;t_min) = benchmark_operators_sphere(bm; z_elems = 63, helem = 30, Nq = 4)
test_results_sphere(t_min)

bm = Benchmark(;float_type = Float32, device_name)
# benchmark_operators_sphere(bm; z_elems = 63, helem = 30, Nq = 4, compile = true)
(;t_min) = benchmark_operators_sphere(bm; z_elems = 63, helem = 30, Nq = 4)
end
#! format: on

Expand Down
22 changes: 22 additions & 0 deletions test/Operators/finitedifference/benchmark_stencils_climacore_kernels.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,107 +4,126 @@ function op_GradientF2C!(c, f, bcs = (;))
@. c.∇x = ∇f(f.y)
return nothing
end
n_reads_writes(::Type{typeof(op_GradientF2C!)}) = 2 # 1 write + 1 read (0 metric terms)
function op_GradientC2F!(c, f, bcs)
∇f = Operators.GradientC2F(bcs)
@. f.∇x = ∇f(c.y)
return nothing
end
n_reads_writes(::Type{typeof(op_GradientC2F!)}) = 2 # 1 write + 1 read (0 metric terms)
#### Divergences
function op_DivergenceF2C!(c, f, bcs = (;))
div = Operators.DivergenceF2C(bcs)
@. c.x = div(Geometry.WVector(f.y))
return nothing
end
n_reads_writes(::Type{typeof(op_DivergenceF2C!)}) = 3 # 1 write + 2 reads (1 metric term)
function op_DivergenceC2F!(c, f, bcs)
div = Operators.DivergenceC2F(bcs)
@. f.x = div(Geometry.WVector(c.y))
return nothing
end
n_reads_writes(::Type{typeof(op_DivergenceC2F!)}) = 3 # 1 write + 2 reads (1 metric term)
#### Interpolations
function op_InterpolateF2C!(c, f, bcs = (;))
interp = Operators.InterpolateF2C(bcs)
@. c.x = interp(f.y)
return nothing
end
n_reads_writes(::Type{typeof(op_InterpolateF2C!)}) = 2 # 1 write + 1 reads (0 metric terms)
function op_InterpolateC2F!(c, f, bcs)
interp = Operators.InterpolateC2F(bcs)
@. f.x = interp(c.y)
return nothing
end
n_reads_writes(::Type{typeof(op_InterpolateC2F!)}) = 2 # 1 write + 1 reads (0 metric terms)
function op_LeftBiasedC2F!(c, f, bcs)
interp = Operators.LeftBiasedC2F(bcs)
@. f.x = interp(c.y)
return nothing
end
n_reads_writes(::Type{typeof(op_LeftBiasedC2F!)}) = 2 # 1 write + 1 reads (0 metric terms)
function op_LeftBiasedF2C!(c, f, bcs = (;))
interp = Operators.LeftBiasedF2C(bcs)
@. c.x = interp(f.y)
return nothing
end
n_reads_writes(::Type{typeof(op_LeftBiasedF2C!)}) = 2 # 1 write + 1 reads (0 metric terms)
function op_RightBiasedC2F!(c, f, bcs)
interp = Operators.RightBiasedC2F(bcs)
@. f.x = interp(c.y)
return nothing
end
n_reads_writes(::Type{typeof(op_RightBiasedC2F!)}) = 2 # 1 write + 1 reads (0 metric terms)
function op_RightBiasedF2C!(c, f, bcs = (;))
interp = Operators.RightBiasedF2C(bcs)
@. c.x = interp(f.y)
return nothing
end
n_reads_writes(::Type{typeof(op_RightBiasedF2C!)}) = 2 # 1 write + 1 reads (0 metric terms)
#### Curl
function op_CurlC2F!(c, f, bcs = (;))
curl = Operators.CurlC2F(bcs)
@. f.curluₕ = curl(c.uₕ)
return nothing
end
n_reads_writes(::Type{typeof(op_CurlC2F!)}) = -1 # todo
#### Mixed/adaptive
function op_UpwindBiasedProductC2F!(c, f, bcs = (;))
upwind = Operators.UpwindBiasedProductC2F(bcs)
@. f.contra3 = upwind(f.w, c.x)
return nothing
end
n_reads_writes(::Type{typeof(op_UpwindBiasedProductC2F!)}) = -1 # todo
function op_Upwind3rdOrderBiasedProductC2F!(c, f, bcs = (;))
upwind = Operators.Upwind3rdOrderBiasedProductC2F(bcs)
@. f.contra3 = upwind(f.w, c.x)
return nothing
end
n_reads_writes(::Type{typeof(op_Upwind3rdOrderBiasedProductC2F!)}) = -1 # todo
#### Simple composed (non-exhaustive due to combinatorial explosion)
function op_divgrad_CC!(c, f, bcs)
grad = Operators.GradientC2F(bcs.inner)
div = Operators.DivergenceF2C(bcs.outer)
@. c.y = div(grad(c.x))
return nothing
end
n_reads_writes(::Type{typeof(op_divgrad_CC!)}) = 3 # 1 write, 2 reads (1 metric term)
function op_divgrad_FF!(c, f, bcs)
grad = Operators.GradientF2C(bcs.inner)
div = Operators.DivergenceC2F(bcs.outer)
@. f.y = div(grad(f.x))
return nothing
end
n_reads_writes(::Type{typeof(op_divgrad_FF!)}) = 3 # 1 write, 2 reads (1 metric term)
function op_div_interp_CC!(c, f, bcs)
interp = Operators.InterpolateC2F(bcs.inner)
div = Operators.DivergenceF2C(bcs.outer)
@. c.y = div(interp(c.contra3))
return nothing
end
n_reads_writes(::Type{typeof(op_div_interp_CC!)}) = -1 # todo
function op_div_interp_FF!(c, f, bcs)
interp = Operators.InterpolateF2C(bcs.inner)
div = Operators.DivergenceC2F(bcs.outer)
@. f.y = div(interp(f.contra3))
return nothing
end
n_reads_writes(::Type{typeof(op_div_interp_FF!)}) = -1 # todo
function op_divgrad_uₕ!(c, f, bcs)
grad = Operators.GradientC2F(bcs.inner)
div = Operators.DivergenceF2C(bcs.outer)
@. c.uₕ2 = div(f.y * grad(c.uₕ))
return nothing
end
n_reads_writes(::Type{typeof(op_divgrad_uₕ!)}) = -1 # todo
function op_divUpwind3rdOrderBiasedProductC2F!(c, f, bcs)
upwind = Operators.Upwind3rdOrderBiasedProductC2F(bcs.inner)
divf2c = Operators.DivergenceF2C(bcs.outer)
@. c.y = divf2c(upwind(f.w, c.x))
return nothing
end
n_reads_writes(::Type{typeof(op_divUpwind3rdOrderBiasedProductC2F!)}) = -1 # todo

function op_broadcast_example0!(c, f, bcs)
Fields.bycolumn(axes(f.ᶠu³)) do colidx
Expand All @@ -113,6 +132,7 @@ function op_broadcast_example0!(c, f, bcs)
end
return nothing
end
n_reads_writes(::Type{typeof(op_broadcast_example0!)}) = 3 # 1 write, 2 reads (0 metric term)

function op_broadcast_example1!(c, f, bcs)
Fields.bycolumn(axes(f.ᶠu³)) do colidx
Expand All @@ -121,12 +141,14 @@ function op_broadcast_example1!(c, f, bcs)
end
return nothing
end
n_reads_writes(::Type{typeof(op_broadcast_example1!)}) = 4 # 1 write, 3 reads (1 metric term)

function op_broadcast_example2!(c, f, bcs)
CT3 = Geometry.Contravariant3Vector
@. f.ᶠu³ = f.ᶠuₕ³ + CT3(f.ᶠw)
return nothing
end
n_reads_writes(::Type{typeof(op_broadcast_example2!)}) = 4 # 1 write, 3 reads (1 metric term)

#=
#####
Expand Down
Loading
Loading