No commit message

luminescent/luminescent/sol.py

@@ -59,6 +59,7 @@ def run(cmd):
     env = '0;'
     # cmd = ["lumi", path]
     gpu_backend = prob["gpu_backend"]
+    run(["julia", "-e", f'println(Base.active_project())'])
     if not gpu_backend:
         cmd = ["julia", "-e", f'{env}using Luminescent;picrun(raw"{path}")']
     else:

luminescent/tiny.py

@@ -0,0 +1,13 @@
+import os
+import luminescent as lumi
+from gdsfactory.generic_tech import LAYER, LAYER_STACK
+import gdsfactory as gf
+import numpy as np
+
+c = gf.components.straight(length=.1, width=0.5, layer=LAYER.WG)
+wavelengths = 1.55
+path = os.path.join("runs", "tiny")
+lumi.make_pic_sim_prob(path, c, wavelengths=wavelengths, keys=[
+                       "2,1"],                       nres=15, approx_2D_mode="TE", gpu="CUDA")  # approx_2D_mode="TE")
+# lumi.solve(path)
+# sol = lumi.lumi_solution()

src/core/monitors.jl

@@ -68,7 +68,7 @@ struct MonitorInstance <: AbstractMonitorInstance
     λmodes
     tags
 end
-@functor MonitorInstance (λmodes,)
+@functor MonitorInstance (λmodes, deltas)
 Base.ndims(m::MonitorInstance) = m.d
 area(m::MonitorInstance) = m.v
 wavelengths(m::MonitorInstance) = keys(m.λmodes)

src/core/update.jl

@@ -55,6 +55,8 @@ function update(u, p, t, dt, field_diffdeltas, field_diffpadvals, source_instanc
     dHdt = -(∇ × E + H ⊙ m) ⊘ μ
     H += dHdt * dt
 
+    @ignore_derivatives gc()
+    # @ignore_derivatives unsafe_free!.((Js, dEdt, dHdt))
     # (; E, H, (Jkeys .=> Jm)..., (Pkeys .=> Pm)...)
     namedtuple([:E => E, :H => H, (Jkeys .=> Jm)..., (Pkeys .=> Pm)...])
 end

src/del.jl

@@ -11,6 +11,7 @@ function diffpad(a, vl, vr=vl; dims=1, diff=diff)
     r = !isnothing(vr)
 
     sz = Tuple(size(a) + (l + r - 1) * sel)
+    # b = similar(a, sz)
     b = Buffer(a, sz)
     b[range.(l * sel + 1, sz - r * sel)...] = diff(a; dims)
     pad!(b, vl, l * sel, 0)

src/gpu.jl

@@ -0,0 +1,3 @@
+
+unsafe_free!(a) = 0
+unsafe_free!(a::CUDA.CuArray) = CUDA.unsafe_free!(a)

src/main.jl

@@ -28,4 +28,6 @@ include("pic/run.jl")
 
 include("ops.jl")
 include("del.jl")
-# include("main.jl")
+# include("main.jl")
+
+include("gpu.jl")

src/pic/run.jl

@@ -92,7 +92,7 @@ function picrun(path; gpuarray=nothing, kw...)
                 frame = frame .>= 0.99maximum(frame)
                 # frame = nothing
                 start = round((bbox[1] - lb) / dl + 1)
-                b = Blob(szd; solid_frac=0.95, morph=false, lsolid=lsolid / dl, lvoid=lvoid / dl, symmetries, F, frame, start)
+                b = Blob(szd; solid_frac=0.99, morph=false, lsolid=lsolid / dl, lvoid=lvoid / dl, symmetries, F, frame, start)
                 display(heatmap(b.frame))
 
                 if !isnothing(sol) && !restart
@@ -182,8 +182,9 @@ function picrun(path; gpuarray=nothing, kw...)
     end
     t0 = time()
     lb3 = (lb..., zmin)
+    # error("not implemented")
+    println("compiling simulation code...")
     if study == "sparams"
-        println("simulating...")
         @unpack S, sols = make_pic_sim_prob(runs, run_probs, lb, dl;
             F, verbose=true, framerate, path)
         plotsols(sols, run_probs, path)

src/pictest.jl

@@ -7,6 +7,8 @@ ENV["JULIA_PKG_PRECOMPILE_AUTO"] = 0
 # picrun(joinpath("runs", "straight");)# gpuarray=cu)
 # picrun(joinpath("runs", "bend_R5"))
 # picrun(joinpath("runs", "mode_converter"))
-picrun(joinpath("runs", "demux"))
+# picrun(joinpath("runs", "demux"))
 # picrun(joinpath("runs", "splitter"))
 
+using CUDA
+picrun(joinpath("runs", "tiny"); gpuarray=cu)

src/sim/setup.jl

@@ -316,7 +316,6 @@ function setup(dl, boundaries, sources, monitors, deltas, mode_deltas;
                      geometry, _geometry, nmax, nmin, ϵeff,
                      is_field_on_lb, is_field_on_ub,
                      u0, dt, kw...) |> pairs |> OrderedDict
-
 end
 update = update
 setup = setup

src/sim/solve.jl

@@ -1,10 +1,18 @@
+function tidy(t, dt)
+    if round(t) > round(t - dt)
+        println("simulation time = $t, took $(timepassed()) seconds")
+    end
+end
+
 function f1(((u,), p, (dt, field_diffdeltas, field_diffpadvals, source_instances)), t)
-    u = update(u, p, t, dt, field_diffdeltas, field_diffpadvals, source_instances)
+    tidy(t, dt)
+    @time u = update(u, p, t, dt, field_diffdeltas, field_diffpadvals, source_instances)
     ((u,), p, (dt, field_diffdeltas, field_diffpadvals, source_instances))
 end
 
 function f2(((u, mf), p, (dt, field_diffdeltas, field_diffpadvals, source_instances), (t0, T, monitor_instances)), t)
-    u = update(u, p, t, dt, field_diffdeltas, field_diffpadvals, source_instances;)
+    tidy(t, dt)
+    @time u = update(u, p, t, dt, field_diffdeltas, field_diffpadvals, source_instances;)
     mf += [[
         begin
             c = dt / T * cispi(-2(t - t0) / λ)
@@ -39,10 +47,12 @@ function solve(prob, ;
     init = (us0, p, (dt, field_diffdeltas, field_diffpadvals, source_instances))
 
     ts = 0:dt:T[1]-F(0.001)
+    nt = -1
     if save_memory
         (u,), = adjoint_reduce(f1, ts, init, ulims)
     else
         (u,), = reduce(ts; init) do us, t
+
             ignore() do
                 if framerate > 0 && t > 0
                     if t % (1 / framerate) < dt
@@ -63,6 +73,7 @@ function solve(prob, ;
     ts = ts[end]+dt:dt:T[2]-F(0.001)
     init = ((u, 0), p, (dt, field_diffdeltas, field_diffpadvals, source_instances), (T[2], durations[2], monitor_instances))
 
+    println("accumulating dft fields")
     if save_memory
         (u, mf), = adjoint_reduce(f2, ts, init, ulims)
     else