Enable oneAPI and ROCm (#9 )

Co-authored-by: Anton Reinhard <anton.reinhard@proton.me> Reviewed-on: #9
Seed Randomness, Fix tests (#8 )
2024-05-08 19:26:18 +02:00 · 2024-05-08 18:04:48 +02:00 · 2024-05-08 14:00:25 +02:00 · 2024-05-08 12:03:27 +02:00 · 2024-01-03 16:38:32 +01:00 · 2023-12-18 14:31:52 +01:00
202 changed files with 10120 additions and 1439 deletions
--- a/.gitattributes
+++ b/.gitattributes
@ -1,2 +1,3 @@
 input/AB->ABBBBBBBBB.txt filter=lfs diff=lfs merge=lfs -text
 input/AB->ABBBBBBB.txt filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
--- a/.gitea/workflows/julia-package-ci.yml
+++ b/.gitea/workflows/julia-package-ci.yml
@ -7,64 +7,7 @@ env:
  JULIA_DEPOT_PATH: './.julia'

 jobs:
-  prepare:
-    runs-on: ubuntu-22.04
-
-    steps:
-      - name: Checkout repository
-        uses: actions/checkout@v3
-        with:
-          fetch-depth: 0
-
-      - name: Setup Julia environment
-        uses: https://github.com/julia-actions/setup-julia@v1.9.2
-        with:
-          version: '1.9.2'
-
-      # needed for the file hashing, should be removed when ${{ hashFiles('**/Project.toml') }} is supported in gitea
-      - name: Setup go environment
-        uses: actions/setup-go@v3
-        with:
-          go-version: '1.20'
-
-      - name: Hash files
-        uses: https://gitea.com/actions/go-hashfiles@v0.0.1
-        id: get-hash
-        with: 
-          patterns: |-
-            **/Project.toml
-
-      - name: Restore Cache
-        uses: actions/cache/restore@v3
-        id: cache-restore
-        with:
-          path: |
-            .julia/artifacts
-            .julia/packages
-            .julia/registries
-          key: julia-${{ steps.get-hash.outputs.hash }}
-
-      - name: Check cache hit
-        if: steps.cache-restore.outputs.cache-hit == 'true'
-        run: exit 0
-
-      - name: Install dependencies
-        run: |
-          julia --project=./        -e 'import Pkg; Pkg.instantiate(); Pkg.precompile()'
-          julia --project=examples/ -e 'import Pkg; Pkg.develop(Pkg.PackageSpec(path=pwd())); Pkg.instantiate(); Pkg.precompile()'
-          julia --project=docs/     -e 'import Pkg; Pkg.develop(Pkg.PackageSpec(path=pwd())); Pkg.instantiate(); Pkg.precompile()'
-
-      - name: Cache Julia packages
-        uses: actions/cache/save@v3
-        with:
-          path: |
-            .julia/artifacts
-            .julia/packages
-            .julia/registries
-          key: julia-${{ steps.get-hash.outputs.hash }}
-
  test:
-    needs: prepare
    runs-on: ubuntu-22.04

    steps:
@ -74,37 +17,14 @@ jobs:
          fetch-depth: 0

      - name: Setup Julia environment
-        uses: https://github.com/julia-actions/setup-julia@v1.9.2
+        uses: https://github.com/julia-actions/setup-julia@v2
        with:
-          version: '1.9.2'
+          version: '1.10'

-      # needed for the file hashing, should be removed when ${{ hashFiles('**/Project.toml') }} is supported in gitea
-      - name: Setup go environment
-        uses: actions/setup-go@v3
-        with:
-          go-version: '1.20'
-
-      - name: Hash files
-        uses: https://gitea.com/actions/go-hashfiles@v0.0.1
-        id: get-hash
-        with: 
-          patterns: |-
-            **/Project.toml  
-
-      - name: Restore cached Julia packages
-        uses: actions/cache/restore@v3
-        with:
-          path: |
-            .julia/artifacts
-            .julia/packages
-            .julia/registries
-          key: julia-${{ steps.get-hash.outputs.hash }}
-
-      - name: Install dependencies
+      - name: Instantiate
        run: |
-          julia --project=./        -e 'import Pkg; Pkg.instantiate(); Pkg.precompile()'
-          julia --project=examples/ -e 'import Pkg; Pkg.develop(Pkg.PackageSpec(path=pwd())); Pkg.instantiate(); Pkg.precompile()'
-          julia --project=docs/     -e 'import Pkg; Pkg.develop(Pkg.PackageSpec(path=pwd())); Pkg.instantiate(); Pkg.precompile()'
+          julia --project=./ -e 'using Pkg; Pkg.instantiate()'
+          julia --project=./ -e 'using Pkg; Pkg.add(url="https://github.com/QEDjl-project/QEDprocesses.jl/")'

      - name: Format check
        run: |
@ -114,19 +34,21 @@ jobs:
          if out == ""
            exit(0)
          else
-            @error "Some files have not been formatted !!!"
+            @error "Some files have not been formatted!!!"
            write(stdout, out)
            exit(1)
          end'

      - name: Run tests
-        run: julia --project=./ -t 4 -e 'import Pkg; Pkg.test()' -O0
+        run: julia --project=./ -t 4 -e 'using Pkg; Pkg.test()' -O0

      - name: Run examples
-        run: julia --project=examples/ -t 4 -e 'include("examples/import_bench.jl")' -O3
+        run: |
+          julia --project=examples/ -e 'using Pkg; Pkg.develop(Pkg.PackageSpec(path=pwd())); Pkg.instantiate(); Pkg.precompile()'
+          julia --project=examples/ -t 4 -e 'include("examples/import_bench.jl")' -O3
+          julia --project=examples/ -t 4 -e 'include("examples/ab5.jl")' -O3

  docs:
-    needs: prepare
    runs-on: ubuntu-22.04

    steps:
@ -136,40 +58,14 @@ jobs:
          fetch-depth: 0

      - name: Setup Julia environment
-        uses: https://github.com/julia-actions/setup-julia@v1.9.2
+        uses: https://github.com/julia-actions/setup-julia@v2
        with:
-          version: '1.9.2'
-
-      # needed for the file hashing, should be removed when ${{ hashFiles('**/Project.toml') }} is supported in gitea
-      - name: Setup go environment
-        uses: actions/setup-go@v3
-        with:
-          go-version: '1.20'
-
-      - name: Hash files
-        uses: https://gitea.com/actions/go-hashfiles@v0.0.1
-        id: get-hash
-        with: 
-          patterns: |-
-            **/Project.toml  
-
-      - name: Restore cached Julia packages
-        uses: actions/cache/restore@v3
-        with:
-          path: |
-            .julia/artifacts
-            .julia/packages
-            .julia/registries
-          key: julia-${{ steps.get-hash.outputs.hash }}
-
-      - name: Install dependencies
-        run: |
-          julia --project=./        -e 'import Pkg; Pkg.instantiate(); Pkg.precompile()'
-          julia --project=examples/ -e 'import Pkg; Pkg.develop(Pkg.PackageSpec(path=pwd())); Pkg.instantiate(); Pkg.precompile()'
-          julia --project=docs/     -e 'import Pkg; Pkg.develop(Pkg.PackageSpec(path=pwd())); Pkg.instantiate(); Pkg.precompile()'
+          version: '1.10'

      - name: Build docs
-        run: julia --project=docs/ docs/make.jl
+        run: |
+          julia --project=docs/ -e 'using Pkg; Pkg.develop(Pkg.PackageSpec(path=pwd())); Pkg.instantiate(); Pkg.precompile()'
+          julia --project=docs/ docs/make.jl

      - name: Upload artifacts
        uses: actions/upload-artifact@v3
--- a/.gitignore
+++ b/.gitignore
@ -5,6 +5,7 @@

 # Files generated by invoking Julia with --track-allocation
 *.mem
+*.pb.gz

 # System-specific files and directories generated by the BinaryProvider and BinDeps packages
 # They contain absolute paths specific to the host computer, and so should not be committed
@ -28,3 +29,8 @@ Manifest.toml
 .vscode
 .julia
 **/.ipynb_checkpoints/
+*.bkp
+
+*.sif
+
+data/hemera_temp
--- a/Project.toml
+++ b/Project.toml
@ -4,19 +4,25 @@ authors = ["Anton Reinhard <anton.reinhard@proton.me>"]
 version = "0.1.0"

 [deps]
+AMDGPU = "21141c5a-9bdb-4563-92ae-f87d6854732e"
 AccurateArithmetic = "22286c92-06ac-501d-9306-4abd417d9753"
 CUDA = "052768ef-5323-5732-b1bb-66c8b64840ba"
+Combinatorics = "861a8166-3701-5b0c-9a16-15d98fcdc6aa"
 DataStructures = "864edb3b-99cc-5e75-8d2d-829cb0a9cfe8"
 ForwardDiff = "f6369f11-7733-5829-9624-2563aa707210"
 JuliaFormatter = "98e50ef6-434e-11e9-1051-2b60c6c9e899"
 KernelAbstractions = "63c18a36-062a-441e-b654-da1e3ab1ce7c"
 NumaAllocators = "21436f30-1b4a-4f08-87af-e26101bb5379"
 QEDbase = "10e22c08-3ccb-4172-bfcf-7d7aa3d04d93"
+QEDprocesses = "46de9c38-1bb3-4547-a1ec-da24d767fdad"
 Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
 Roots = "f2b01f46-fcfa-551c-844a-d8ac1e96c665"
+StaticArrays = "90137ffa-7385-5640-81b9-e52037218182"
 UUIDs = "cf7118a7-6976-5b1a-9a39-7adc72f591a4"
+oneAPI = "8f75cd03-7ff8-4ecb-9b8f-daf728133b1b"

 [extras]
+CUDA_Runtime_jll = "76a88914-d11a-5bdc-97e0-2f5a05c973a2"
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"

 [targets]
--- a/README.md
+++ b/README.md
@ -50,8 +50,8 @@ Problems:

 For graphs AB->AB^n:
 - Number of Sums should always be 1
- Number of ComputeTaskS2 should always be (n+1)!
- Number of ComputeTaskU should always be (n+3)
+- Number of ComputeTaskABC_S2 should always be (n+1)!
+- Number of ComputeTaskABC_U should always be (n+3)

 Times are from my home machine: AMD Ryzen 7900X3D, 64GB DDR5 RAM @ 6000MHz (not necessarily up to date, check Jupyter Notebooks in `notebooks/` instead)

@ -59,9 +59,9 @@ Times are from my home machine: AMD Ryzen 7900X3D, 64GB DDR5 RAM @ 6000MHz (not
 $ julia --project examples/import_bench.jl
 AB->AB:
 Graph:
-  Nodes: Total: 34, DataTask: 19, ComputeTaskP: 4, 
-         ComputeTaskS2: 2, ComputeTaskV: 4, ComputeTaskU: 4, 
-         ComputeTaskSum: 1
+  Nodes: Total: 34, DataTask: 19, ComputeTaskABC_P: 4, 
+         ComputeTaskABC_S2: 2, ComputeTaskABC_V: 4, ComputeTaskABC_U: 4, 
+         ComputeTaskABC_Sum: 1
  Edges: 37
  Total Compute Effort: 185
  Total Data Transfer: 102
@ -71,9 +71,9 @@ Graph:

 AB->ABBB:
 Graph:
-  Nodes: Total: 280, DataTask: 143, ComputeTaskP: 6, 
-         ComputeTaskS2: 24, ComputeTaskV: 64, ComputeTaskU: 6, 
-         ComputeTaskSum: 1, ComputeTaskS1: 36
+  Nodes: Total: 280, DataTask: 143, ComputeTaskABC_P: 6, 
+         ComputeTaskABC_S2: 24, ComputeTaskABC_V: 64, ComputeTaskABC_U: 6, 
+         ComputeTaskABC_Sum: 1, ComputeTaskABC_S1: 36
  Edges: 385
  Total Compute Effort: 2007
  Total Data Transfer: 828
@ -83,9 +83,9 @@ Graph:

 AB->ABBBBB:
 Graph:
-  Nodes: Total: 7854, DataTask: 3931, ComputeTaskP: 8, 
-         ComputeTaskS2: 720, ComputeTaskV: 1956, ComputeTaskU: 8, 
-         ComputeTaskSum: 1, ComputeTaskS1: 1230
+  Nodes: Total: 7854, DataTask: 3931, ComputeTaskABC_P: 8, 
+         ComputeTaskABC_S2: 720, ComputeTaskABC_V: 1956, ComputeTaskABC_U: 8, 
+         ComputeTaskABC_Sum: 1, ComputeTaskABC_S1: 1230
  Edges: 11241
  Total Compute Effort: 58789
  Total Data Transfer: 23244
@ -95,9 +95,9 @@ Graph:

 AB->ABBBBBBB:
 Graph:
-  Nodes: Total: 438436, DataTask: 219223, ComputeTaskP: 10, 
-         ComputeTaskS2: 40320, ComputeTaskV: 109600, ComputeTaskU: 10, 
-         ComputeTaskSum: 1, ComputeTaskS1: 69272
+  Nodes: Total: 438436, DataTask: 219223, ComputeTaskABC_P: 10, 
+         ComputeTaskABC_S2: 40320, ComputeTaskABC_V: 109600, ComputeTaskABC_U: 10, 
+         ComputeTaskABC_Sum: 1, ComputeTaskABC_S1: 69272
  Edges: 628665
  Total Compute Effort: 3288131
  Total Data Transfer: 1297700
@ -107,7 +107,7 @@ Graph:

 AB->ABBBBBBBBB:
 Graph:
-  Nodes: Total: 39456442, DataTask: 19728227, ComputeTaskS1: 6235290, ComputeTaskP: 12, ComputeTaskU: 12, ComputeTaskV: 9864100, ComputeTaskS2: 3628800, ComputeTaskSum: 1
+  Nodes: Total: 39456442, DataTask: 19728227, ComputeTaskABC_S1: 6235290, ComputeTaskABC_P: 12, ComputeTaskABC_U: 12, ComputeTaskABC_V: 9864100, ComputeTaskABC_S2: 3628800, ComputeTaskABC_Sum: 1
  Edges: 56578129
  Total Compute Effort: 295923153
  Total Data Transfer: 175407750
@ -116,9 +116,9 @@ Graph:

 ABAB->ABAB:
 Graph:
-  Nodes: Total: 3218, DataTask: 1613, ComputeTaskP: 8, 
-         ComputeTaskS2: 288, ComputeTaskV: 796, ComputeTaskU: 8, 
-         ComputeTaskSum: 1, ComputeTaskS1: 504
+  Nodes: Total: 3218, DataTask: 1613, ComputeTaskABC_P: 8, 
+         ComputeTaskABC_S2: 288, ComputeTaskABC_V: 796, ComputeTaskABC_U: 8, 
+         ComputeTaskABC_Sum: 1, ComputeTaskABC_S1: 504
  Edges: 4581
  Total Compute Effort: 24009
  Total Data Transfer: 9494
@ -128,9 +128,9 @@ Graph:

 ABAB->ABC:
 Graph:
-  Nodes: Total: 817, DataTask: 412, ComputeTaskP: 7,
-         ComputeTaskS2: 72, ComputeTaskV: 198, ComputeTaskU: 7,
-         ComputeTaskSum: 1, ComputeTaskS1: 120
+  Nodes: Total: 817, DataTask: 412, ComputeTaskABC_P: 7,
+         ComputeTaskABC_S2: 72, ComputeTaskABC_V: 198, ComputeTaskABC_U: 7,
+         ComputeTaskABC_Sum: 1, ComputeTaskABC_S1: 120
  Edges: 1151
  Total Compute Effort: 6028
  Total Data Transfer: 2411
--- a/data/evaluate.jl
+++ b/data/evaluate.jl
@ -0,0 +1,170 @@
+using CSV
+using DataFrames
+using Plots
+using StatsPlots
+using LaTeXStrings
+
+if (length(ARGS) < 1)
+    println("Please use with \"input_file.csv\"")
+end
+
+processes = [
+    "QED Process: 'ke->ke'",
+    "QED Process: 'ke->kke'",
+    "QED Process: 'ke->kkke'",
+    #"QED Process: 'ke->kkkke'",
+    #"QED Process: 'ke->kkkkke'",
+    #"QED Process: 'ke->kkkkkke'",
+    #"QED Process: 'ke->kkkkkkke'",
+    "ABC Process: 'AB->AB'",
+    "ABC Process: 'AB->ABBB'",
+    #"ABC Process: 'AB->ABBBBB'",
+]
+
+function proc_to_n(str::AbstractString)
+    parts = split(str, "'")
+    parts = split(parts[2], "->")
+    k_count = count(c -> c == 'k', parts[2])
+    return k_count
+end
+
+function beautify_title(str::AbstractString)
+    parts = split(str, "'")
+
+    preprefix = parts[1]
+    infix = parts[2]
+    sufsuffix = parts[3]
+
+    parts = split(infix, "->")
+
+    prefix = parts[1]
+    suffix = parts[2]
+
+    k_count = count(c -> c == 'k', suffix)
+    B_count = count(c -> c == 'B', suffix)
+
+    if k_count == 1 || B_count == 1
+        new_suffix = suffix
+    elseif k_count >= 1
+        new_suffix = replace(suffix, r"k+" => "k^$k_count")
+    elseif B_count >= 1
+        new_suffix = replace(suffix, r"B+" => "B^$B_count")
+    end
+
+    return preprefix * L"%$prefix \rightarrow %$new_suffix" * sufsuffix
+end
+
+input_file = ARGS[1]
+df = CSV.read(input_file, DataFrame)
+n_inputs = df[:, "n_inputs"][1]
+
+# plotting with threads as x axis
+for process_name in processes
+    title_string = "$(beautify_title(process_name)), $n_inputs samples"
+    println("$title_string")
+
+    process_no_opt = process_name * " not optimized"
+    process_red = process_name * " reduced"
+    process_tape_no_opt = process_name * " not optimized tape"
+    process_tape_red = process_name * " reduced tape"
+
+    df_no_opt = filter(:process_name => x -> x == process_no_opt, df)
+    df_red = filter(:process_name => x -> x == process_red, df)
+    df_tape_no_opt = filter(:process_name => x -> x == process_tape_no_opt, df)
+    df_tape_red = filter(:process_name => x -> x == process_tape_red, df)
+
+    @df df_no_opt scatter(:cpu_threads, :cpu_time, label = "unoptimized function", markershape = :circle)
+    @df df_red scatter!(:cpu_threads, :cpu_time, label = "reduced function", markershape = :rect)
+    @df df_tape_no_opt scatter!(:cpu_threads, :cpu_time, label = "unoptimized tape", markershape = :utriangle)
+    @df df_tape_red scatter!(:cpu_threads, :cpu_time, label = "reduced tape", markershape = :star)
+
+    plot!(
+        title = title_string,
+        yscale = :linear,
+        legend = :outerright,
+        legendcolumns = 1,
+        legend_font_pointsize = 10,
+        size = (800, 600),
+        ylabel = "execution time (s)",
+        xlabel = "threads (#)",
+    )
+
+    savefig("$(process_name)_time.pdf")
+
+
+    @df df_no_opt scatter(:cpu_threads, :cpu_rate, label = "unoptimized function", markershape = :circle)
+    @df df_red scatter!(:cpu_threads, :cpu_rate, label = "reduced function", markershape = :rect)
+    @df df_tape_no_opt scatter!(:cpu_threads, :cpu_rate, label = "unoptimized tape", markershape = :utriangle)
+    @df df_tape_red scatter!(:cpu_threads, :cpu_rate, label = "reduced tape", markershape = :star)
+
+    plot!(
+        title = "Sample rate, " * title_string,
+        yscale = :log10,
+        legend = :outerright,
+        legendcolumns = 1,
+        legend_font_pointsize = 10,
+        size = (800, 600),
+        ylabel = "rate (" * L"s^{-1}" * ")",
+        xlabel = "threads (#)",
+    )
+
+    savefig("$(process_name)_rate.pdf")
+
+
+    @df df_no_opt scatter(:cpu_threads, :cpu_gflops, label = "unoptimized function", markershape = :circle)
+    @df df_red scatter!(:cpu_threads, :cpu_gflops, label = "reduced function", markershape = :rect)
+    @df df_tape_no_opt scatter!(:cpu_threads, :cpu_gflops, label = "unoptimized tape", markershape = :utriangle)
+    @df df_tape_red scatter!(:cpu_threads, :cpu_gflops, label = "reduced tape", markershape = :star)
+
+    plot!(
+        title = "CPU performance, " * title_string,
+        yscale = :linear,
+        legend = :outerright,
+        legendcolumns = 1,
+        legend_font_pointsize = 10,
+        size = (800, 600),
+        ylabel = "CPU performance (GFLOPS)",
+        xlabel = "threads (#)",
+    )
+
+    savefig("$(process_name)_performance.pdf")
+
+end
+
+# plotting with process size as x axis
+THREADS = [1, 2, 4, 8, 16]
+
+for threads in THREADS
+    title_string = "$threads threads, $n_inputs samples"
+
+    df_filt = filter(:cpu_threads => x -> x == threads, df)
+    df_filt = filter(:process_name => x -> proc_to_n(x) >= 1, df_filt)
+
+    df_filt.process_size = @. proc_to_n(df_filt.process_name)
+
+    df_no_opt = filter(:process_name => x -> match(r" not optimized$", x) !== nothing, df_filt)
+    df_red = filter(:process_name => x -> match(r" reduced$", x) !== nothing, df_filt)
+    df_tape_no_opt = filter(:process_name => x -> match(r" not optimized tape$", x) !== nothing, df_filt)
+    df_tape_red = filter(:process_name => x -> match(r" reduced tape$", x) !== nothing, df_filt)
+
+
+    @df df_no_opt scatter(:process_size, :graph_gen_time, label = "graph generation time")
+    @df df_red scatter!(:process_size, :optimization_time, label = "optimization time")
+    @df df_no_opt scatter!(:process_size, :function_generation_time, label = "unoptimized function generation time")
+    @df df_tape_no_opt scatter!(:process_size, :function_generation_time, label = "unoptimized tape generation time")
+    @df df_red scatter!(:process_size, :function_generation_time, label = "reduced function generation time")
+    @df df_tape_red scatter!(:process_size, :function_generation_time, label = "reduced tape generation time")
+
+    plot!(
+        title = "function generation times, " * title_string,
+        yscale = :log10,
+        legend = :outerbottom,
+        legendcolumns = 2,
+        legend_font_pointsize = 10,
+        size = (800, 600),
+        ylabel = "time (s)",
+        xlabel = "process size (#)",
+    )
+
+    savefig("gen_times_$(threads)_threads.pdf")
+end
--- a/data/evaluate_cpu_gpu_exec.jl
+++ b/data/evaluate_cpu_gpu_exec.jl
@ -0,0 +1,143 @@
+using CSV
+using DataFrames
+using Plots
+using StatsPlots
+using LaTeXStrings
+
+if (length(ARGS) < 1)
+    println("Please use with \"input_file.csv\"")
+end
+
+processes = [
+    "QED Process: 'ke->ke'",
+    "QED Process: 'ke->kke'",
+    "QED Process: 'ke->kkke'",
+    "QED Process: 'ke->kkkke'",
+    "QED Process: 'ke->kkkkke'",
+    #"QED Process: 'ke->kkkkkke'",
+    #"QED Process: 'ke->kkkkkkke'",
+    "ABC Process: 'AB->AB'",
+    "ABC Process: 'AB->ABBB'",
+    "ABC Process: 'AB->ABBBBB'",
+]
+
+function proc_to_n(str::AbstractString)
+    parts = split(str, "'")
+    parts = split(parts[2], "->")
+    k_count = count(c -> c == 'k', parts[2])
+    return k_count
+end
+
+function abc_proc_to_n(str::AbstractString)
+    parts = split(str, "'")
+    parts = split(parts[2], "->")
+    b_count = count(c -> c == 'B', parts[2])
+    return b_count
+end
+
+function beautify_title(str::AbstractString)
+    parts = split(str, "'")
+
+    preprefix = parts[1]
+    infix = parts[2]
+    sufsuffix = parts[3]
+
+    parts = split(infix, "->")
+
+    prefix = parts[1]
+    suffix = parts[2]
+
+    k_count = count(c -> c == 'k', suffix)
+    B_count = count(c -> c == 'B', suffix)
+
+    if k_count == 1 || B_count == 1
+        new_suffix = suffix
+    elseif k_count >= 1
+        new_suffix = replace(suffix, r"k+" => "k^$k_count")
+    elseif B_count >= 1
+        new_suffix = replace(suffix, r"B+" => "B^$B_count")
+    end
+
+    return preprefix * L"%$prefix \rightarrow %$new_suffix" * sufsuffix
+end
+
+input_file = ARGS[1]
+df = CSV.read(input_file, DataFrame)
+n_inputs = df[:, "n_inputs"][1]
+
+
+
+title_string = "QED N-Photon Compton Scattering\nCalculate 1,048,576 (\$2^{20}\$) Matrix Elements"
+
+df_filt = filter(:process_name => x -> proc_to_n(x) >= 1, df)
+
+df_filt.process_size = @. proc_to_n(df_filt.process_name)
+
+df_red = filter(:process_name => x -> match(r" reduced$", x) !== nothing, df_filt)
+
+@df df_red scatter(
+    :process_size,
+    :cpu_time,
+    yerror = :cpu_std,
+    label = "CPU execution time, 32 threads (s)",
+    markersize = 6,
+)
+@df df_red scatter!(
+    :process_size,
+    :gpu_time,
+    yerror = :gpu_std,
+    label = "GPU execution time, A100 80GB (s)",
+    markersize = 6,
+)
+
+plot!(
+    #title = title_string,
+    yscale = :log10,
+    legend = :outerbottom,
+    legendcolumns = 2,
+    legend_font_pointsize = 10,
+    size = (800, 600),
+    ylabel = "time (s)",
+    xlabel = "process size (#)",
+)
+
+savefig("cpu_vs_gpu_qed.pdf")
+
+
+
+
+title_string = "\$AB\\rightarrow AB^n\$ ABC Processes\nCalculate 1,048,576 (\$2^{20}\$) Matrix Elements"
+
+df_filt = filter(:process_name => x -> abc_proc_to_n(x) >= 1, df)
+
+df_filt.process_size = @. abc_proc_to_n(df_filt.process_name)
+
+df_red = filter(:process_name => x -> match(r" reduced$", x) !== nothing, df_filt)
+
+@df df_red scatter(
+    :process_size,
+    :cpu_time,
+    yerror = :cpu_std,
+    label = "CPU execution time, 32 threads (s)",
+    markersize = 6,
+)
+@df df_red scatter!(
+    :process_size,
+    :gpu_time,
+    yerror = :gpu_std,
+    label = "GPU execution time, A100 80GB (s)",
+    markersize = 6,
+)
+
+plot!(
+    #title = title_string,
+    yscale = :log10,
+    legend = :outerbottom,
+    legendcolumns = 2,
+    legend_font_pointsize = 10,
+    size = (800, 600),
+    ylabel = "time (s)",
+    xlabel = "process size (#)",
+)
+
+savefig("cpu_vs_gpu_abc.pdf")
--- a/data/evaluate_full_node_bench.jl
+++ b/data/evaluate_full_node_bench.jl
@ -0,0 +1,212 @@
+using CSV
+using DataFrames
+using Plots
+using StatsPlots
+using LaTeXStrings
+
+if (length(ARGS) < 1)
+    println("Please use with \"input_file.csv\"")
+end
+
+processes = [
+    "QED Process: 'ke->ke'",
+    "QED Process: 'ke->kke'",
+    "QED Process: 'ke->kkke'",
+    "QED Process: 'ke->kkkke'",
+    "QED Process: 'ke->kkkkke'",
+]
+
+function proc_to_n(str::AbstractString)
+    parts = split(str, "'")
+    parts = split(parts[2], "->")
+    k_count = count(c -> c == 'k', parts[2])
+    return k_count
+end
+
+function beautify_title(str::AbstractString)
+    parts = split(str, "'")
+
+    preprefix = parts[1]
+    infix = parts[2]
+    sufsuffix = parts[3]
+
+    parts = split(infix, "->")
+
+    prefix = parts[1]
+    suffix = parts[2]
+
+    k_count = count(c -> c == 'k', suffix)
+    B_count = count(c -> c == 'B', suffix)
+
+    if k_count == 1 || B_count == 1
+        new_suffix = suffix
+    elseif k_count >= 1
+        new_suffix = replace(suffix, r"k+" => "k^$k_count")
+    elseif B_count >= 1
+        new_suffix = replace(suffix, r"B+" => "B^$B_count")
+    end
+
+    return "QED Compton Scattering Process " * L"%$prefix \rightarrow %$new_suffix" * sufsuffix
+end
+
+input_file = ARGS[1]
+df = CSV.read(input_file, DataFrame)
+n_inputs = df[:, "n_inputs"][1]
+gpus = df.gpu_devices[1]
+cpus = df.cpu_threads[1]
+
+power = Int(round(log2(n_inputs)))
+
+chunk_sizes = [
+    "\$2^{10}\$",
+    "\$2^{11}\$",
+    "\$2^{12}\$",
+    "\$2^{13}\$",
+    "\$2^{14}\$",
+    "\$2^{15}\$",
+    "\$2^{16}\$",
+    "\$2^{17}\$",
+    "\$2^{18}\$",
+    "\$2^{19}\$",
+    "\$2^{20}\$",
+]
+
+best_times = Vector{Float64}()
+best_times_std = Vector{Float64}()
+
+# plotting with threads as x axis
+for process_name in processes
+    df_filt = filter(:process_name => x -> x == process_name, df)
+
+    df_filt.cpu_ratio = df_filt.cpu_chunks ./ (df_filt.cpu_chunks .+ df_filt.gpu_chunks) .* 100.0
+    df_filt.gpu_ratio = df_filt.gpu_chunks ./ (df_filt.cpu_chunks .+ df_filt.gpu_chunks) .* 100.0
+
+    push!(best_times, minimum(df_filt.time))
+
+    bar(chunk_sizes, df_filt.cpu_ratio, label = "workload completed by \$$(cpus)\$ CPU threads (%)")
+    bar!(
+        chunk_sizes,
+        [100 for _ in chunk_sizes],
+        label = "workload completed by $(gpus) GPUs (%)",
+        fillto = df_filt.cpu_ratio,
+    )
+
+    plot!(
+        #title = "$(beautify_title(process_name))\nComputing $(n_inputs) (\$2^{$(power)}\$) Matrix Elements",
+        yscale = :linear,
+        #xticks = [1024 4096 16384 65536 262144 1048576],
+        ylim = (0, 105),
+        legend = :outerbottom,
+        legendcolumns = 1,
+        legend_font_pointsize = 10,
+        #size = (800, 600),
+        ylabel = "contribution (%)",
+        xlabel = "chunk size (#)",
+    )
+
+    savefig("full_node_chunk_size_$(proc_to_n(process_name))k_ratio.pdf")
+
+
+    scatter(
+        chunk_sizes,
+        df_filt.rate,
+        label = "total execution rate (\$s^{-1}\$)",
+        #title = "$(beautify_title(process_name))\nComputing $(n_inputs) (\$2^{$(power)}\$) Matrix Elements",
+        ylim = (0, :auto),
+        #yscale = :log10,
+        #xticks = [1024 4096 16384 65536 262144 1048576],
+        legend = :outerbottom,
+        legendcolumns = 1,
+        legend_font_pointsize = 10,
+        #size = (800, 600),
+        ylabel = "rate (\$s^{-1}\$)",
+        xlabel = "chunk size (#)",
+        markersize = 7,
+    )
+
+    savefig("full_node_chunk_size_$(proc_to_n(process_name))k_rate.pdf")
+
+
+    scatter(
+        chunk_sizes,
+        df_filt.time,
+        yerror = df_filt.std,
+        label = "total execution time (s)",
+        #title = "$(beautify_title(process_name))\nComputing $(n_inputs) (\$2^{$(power)}\$) Matrix Elements",
+        #xticks = [1024 4096 16384 65536 262144 1048576],
+        ylim = (0, maximum(df_filt.time) * 1.05),
+        legend = :outerbottom,
+        legendcolumns = 1,
+        legend_font_pointsize = 10,
+        #size = (800, 600),
+        ylabel = "time (s)",
+        xlabel = "chunk size (#)",
+        markersize = 7,
+    )
+
+    savefig("full_node_chunk_size_$(proc_to_n(process_name))k_time.pdf")
+
+end
+
+
+# plotting with process size as x axis
+A100_rates = [2.530045276927587e9, 1.16972304616864e9, 2.0002725972692013e8, 3.495722925446318e7, 4.792187095617111e6]
+CPU_32threads_rates =
+    [3.2691139045711152e7, 1.1578342663759507e7, 3.1670680975577887e6, 731037.7069429948, 115001.5594731802]
+
+theory_rates = (A100_rates .+ CPU_32threads_rates) .* 4
+
+scatter(
+    proc_to_n.(processes),
+    best_times,
+    label = "full node best achieved time (s)",
+    #title = "QED N-Photon Compton Scattering\nComputing $(n_inputs) (\$2^{$(power)}\$) Matrix Elements",
+    ylim = (0, maximum(best_times) * 1.05),
+    legend = :outerbottom,
+    legendcolumns = 1,
+    legend_font_pointsize = 10,
+    #size = (800, 600),
+    ylabel = "time (s)",
+    xlabel = "process size (#)",
+    markersize = 7,
+)
+
+savefig("full_node_process_best_time.pdf")
+
+
+scatter(
+    proc_to_n.(processes),
+    (n_inputs ./ best_times),
+    label = "full node best achieved rate (\$s^{-1}\$)",
+    #title = "QED N-Photon Compton Scattering\nComputing $(n_inputs) (\$2^{$(power)}\$) Matrix Elements",
+    ylim = (0, maximum(n_inputs ./ best_times) * 1.05),
+    legend = :outerbottom,
+    legendcolumns = 1,
+    legend_font_pointsize = 10,
+    #size = (800, 600),
+    ylabel = "rate (\$s^{-1}\$)",
+    xlabel = "process size (#)",
+    markersize = 7,
+)
+
+savefig("full_node_process_best_rate.pdf")
+
+
+
+scatter(
+    proc_to_n.(processes),
+    [(n_inputs ./ best_times) theory_rates],
+    label = ["full node best achieved rate (\$s^{-1}\$)" "theoretical rate from previous benchmarks (\$s^{-1}\$)"],
+    #title = "QED N-Photon Compton Scattering\nComputing $(n_inputs) (\$2^{$(power)}\$) Matrix Elements",
+    #ylim = (0, max(maximum(n_inputs ./ best_times), maximum(theory_rates)) * 1.05),
+    yscale = :log10,
+    legend = :outerbottom,
+    legendcolumns = 1,
+    legend_font_pointsize = 10,
+    #size = (800, 600),
+    ylabel = "rate (\$s^{-1}\$)",
+    xlabel = "process size (#)",
+    markersize = 7,
+)
+
+savefig("full_node_process_best_rate_plus_theory.pdf")
--- a/data/evaluate_gen.jl
+++ b/data/evaluate_gen.jl
@ -0,0 +1,232 @@
+using CSV
+using DataFrames
+using Plots
+using StatsPlots
+using LaTeXStrings
+
+if (length(ARGS) < 1)
+    println("Please use with \"input_file.csv\"")
+end
+
+function proc_to_n(str::AbstractString)
+    parts = split(str, "->")
+    k_count = count(c -> c == 'k', parts[2])
+    return k_count
+end
+
+input_file = ARGS[1]
+df = CSV.read(input_file, DataFrame)
+
+# plotting with process size as x axis
+THREADS = [1]
+
+for threads in THREADS
+    title_string = "n-photon Compton diagram generation"
+
+    df_filt = filter(:cpu_threads => x -> x == threads, df)
+    df_filt = filter(:process_name => x -> proc_to_n(x) >= 1, df_filt)
+
+    # ns -> s
+    df_filt.graph_gen_mean = @. df_filt.graph_gen_mean / 1e9
+    df_filt.graph_gen_std = @. df_filt.graph_gen_std / 1e9
+
+    # B -> MB (not MiB since the log scale is base 10)
+    df_filt.graph_mem = @. df_filt.graph_mem / 1e6
+    df_filt.graph_mem_reduced = @. df_filt.graph_mem_reduced / 1e6
+
+    df_filt.process_size = @. proc_to_n(df_filt.process_name)
+    l = length(df_filt.process_size)
+
+    println(df_filt[!, :process_size])
+    println(df_filt[!, :graph_mem])
+    println(df_filt[!, :graph_mem_reduced])
+
+
+    @df df_filt scatter(:process_size, :graph_mem, label = "unreduced graph", markersize = 7)
+    scatter!(
+        df_filt[!, :process_size],
+        df_filt[!, :graph_mem_reduced],
+        label = "reduced graph",
+        markershape = :square,
+        markersize = 7,
+    )
+
+    plot!(
+        title = "n-photon Compton diagram memory footprint",
+        yscale = :log10,
+        legend = :outerbottom,
+        minorgrid = true,
+        xticks = :process_size,
+        #yticks = [1e-3, 1e-1, 1e1, 1e3],
+        xgrid = false,
+        xminorticks = false,
+        legendcolumns = 1,
+        legend_font_pointsize = 12,
+        fontsize = 12,
+        size = (800, 600),
+        ylabel = "memory footprint (MB)",
+        xlabel = "process size (#)",
+    )
+
+    savefig("gen_memory_$(threads).pdf")
+
+    @df df_filt scatter(
+        :process_size,
+        :graph_gen_mean,
+        yerror = :graph_gen_std,
+        label = "graph generation time",
+        markersize = 7,
+    )
+    scatter!(
+        df_filt[!, :process_size],
+        df_filt[!, :graph_elapsed_reduce],
+        label = "graph reduction time",
+        markershape = :square,
+        markersize = 7,
+    )
+
+    plot!(
+        title = title_string,
+        yscale = :log10,
+        legend = :outerbottom,
+        minorgrid = true,
+        xticks = :process_size,
+        yticks = [1e-3, 1e-1, 1e1, 1e3],
+        xgrid = false,
+        xminorticks = false,
+        legendcolumns = 1,
+        legend_font_pointsize = 12,
+        fontsize = 12,
+        size = (800, 600),
+        ylabel = "time (s)",
+        xlabel = "process size (#)",
+    )
+
+    savefig("gen_times_$(threads)_threads.pdf")
+    exit(0)
+
+    # graph size
+    title_string = "n-photon Compton unreduced graph size"
+
+    @df df_filt scatter(:process_size, :graph_nodes, label = "nodes", markershape = :circle)
+    @df df_filt scatter!(:process_size, :graph_edges, label = "edges", markershape = :square)
+    @df df_filt scatter!(:process_size, :graph_u_nodes, label = "U-nodes", markershape = :star)
+    @df df_filt scatter!(:process_size, :graph_v_nodes, label = "V-nodes", markershape = :utriangle)
+    @df df_filt scatter!(:process_size[2:end], :graph_s1_nodes[2:end], label = "S1-nodes", markershape = :x)
+    @df df_filt scatter!(:process_size, :graph_s2_nodes, label = "S2-nodes", markershape = :diamond)
+
+    plot!(
+        title = title_string,
+        yscale = :log10,
+        legend = :outerbottom,
+        yminorgrid = true,
+        xticks = :process_size,
+        yticks = [1e1, 1e3, 1e5, 1e7],
+        xgrid = false,
+        xminorticks = false,
+        legendcolumns = 2,
+        legend_font_pointsize = 10,
+        size = (800, 600),
+        ylabel = "(#)",
+        xlabel = "process size (#)",
+    )
+
+    savefig("compton_graph_size_unreduced.pdf")
+
+
+    # graph size
+    title_string = "n-photon Compton reduced graph size"
+
+    @df df_filt scatter(:process_size, :graph_nodes_reduced, label = "nodes", markershape = :circle)
+    @df df_filt scatter!(:process_size, :graph_edges_reduced, label = "edges", markershape = :square)
+    @df df_filt scatter!(:process_size, :graph_u_nodes_reduced, label = "U-nodes", markershape = :star)
+    @df df_filt scatter!(:process_size, :graph_v_nodes_reduced, label = "V-nodes", markershape = :utriangle)
+    @df df_filt scatter!(:process_size[2:end], :graph_s1_nodes_reduced[2:end], label = "S1-nodes", markershape = :x)
+    @df df_filt scatter!(:process_size, :graph_s2_nodes_reduced, label = "S2-nodes", markershape = :diamond)
+
+    plot!(
+        title = title_string,
+        yscale = :log10,
+        legend = :outerbottom,
+        yminorgrid = true,
+        xticks = :process_size,
+        yticks = [1e1, 1e2, 1e3, 1e4, 1e5, 1e6],
+        xgrid = false,
+        xminorticks = false,
+        legendcolumns = 2,
+        legend_font_pointsize = 10,
+        size = (800, 600),
+        ylabel = "(#)",
+        xlabel = "process size (#)",
+    )
+
+    savefig("compton_graph_size_reduced.pdf")
+
+
+    # graph size versus
+    title_string = "n-photon Compton graph sizes"
+
+    @df df_filt scatter(:process_size, :graph_nodes, label = "nodes", markershape = :circle)
+    @df df_filt scatter!(:process_size, :graph_edges, label = "edges", markershape = :square)
+    @df df_filt scatter!(:process_size, :graph_nodes_reduced, label = "nodes (after reduction)", markershape = :star)
+    @df df_filt scatter!(
+        :process_size,
+        :graph_edges_reduced,
+        label = "edges (after reduction)",
+        markershape = :utriangle,
+    )
+
+    plot!(
+        title = title_string,
+        yscale = :log10,
+        legend = :outerbottom,
+        yminorgrid = true,
+        xticks = :process_size,
+        yticks = [1e1, 1e2, 1e3, 1e4, 1e5, 1e6],
+        xgrid = false,
+        xminorticks = false,
+        legendcolumns = 2,
+        legend_font_pointsize = 10,
+        size = (800, 600),
+        ylabel = "(#)",
+        xlabel = "process size (#)",
+    )
+
+    savefig("compton_graph_size_versus.pdf")
+
+end
+
+# for a specific process, plot times with threads as x
+process = "ke->kkkkkkkke"
+title_string = "n-photon Compton diagram generation times, $process"
+
+df_filt = filter(:process_name => x -> x == process, df)
+df_filt.graph_gen_mean = @. df_filt.graph_gen_mean / 1e9
+df_filt.graph_gen_std = @. df_filt.graph_gen_std / 1e9
+
+@df df_filt scatter(
+    :cpu_threads,
+    :graph_gen_mean,
+    yerror = :graph_gen_std,
+    label = "graph generation time",
+    markersize = 7,
+)
+
+plot!(
+    title = title_string,
+    yscale = :linear,
+    legend = :outerbottom,
+    minorgrid = true,
+    xticks = :cpu_threads,
+    #yticks = [1e-3, 1e-2, 1e-1, 1e-0, 1e1],
+    ylim = (0, max(df_filt[!, :graph_gen_mean]...) * 1.1),
+    xgrid = false,
+    xminorticks = false,
+    legendcolumns = 1,
+    legend_font_pointsize = 10,
+    size = (800, 600),
+    ylabel = "time (s)",
+    xlabel = "process size (#)",
+)
+
+savefig("gen_times_$(process).pdf")
--- a/data/evaluate_gen_one_sided_comparison.jl
+++ b/data/evaluate_gen_one_sided_comparison.jl
@ -0,0 +1,52 @@
+using CSV
+using DataFrames
+using Plots
+using StatsPlots
+using LaTeXStrings
+
+if (length(ARGS) < 2)
+    println("Please use with \"input_file.csv\" \"input_file_onesided.csv\"")
+end
+
+function proc_to_n(str::AbstractString)
+    parts = split(str, "->")
+    k_count = count(c -> c == 'k', parts[2])
+    return k_count
+end
+
+input_file = ARGS[1]
+input_file_onesided = ARGS[2]
+df = CSV.read(input_file, DataFrame)
+df2 = CSV.read(input_file_onesided, DataFrame)
+
+
+df_filt = filter(:process_name => x -> proc_to_n(x) >= 1 && proc_to_n(x) <= 7, df)
+df_filt.process_size = @. proc_to_n(df_filt.process_name)
+
+df_filt2 = filter(:process_name => x -> proc_to_n(x) >= 1 && proc_to_n(x) <= 7, df2)
+df_filt2.process_size = @. proc_to_n(df_filt2.process_name)
+
+
+# graph size
+title_string = "n-photon Compton reduced graph size"
+
+@df df_filt scatter(:process_size, :graph_nodes_reduced, label = "nodes, two-sided generation", markershape = :circle)
+@df df_filt2 scatter!(:process_size, :graph_nodes_reduced, label = "nodes, one-sided generation", markershape = :square)
+
+plot!(
+    title = title_string,
+    yscale = :log10,
+    legend = :outerbottom,
+    yminorgrid = true,
+    xticks = :process_size,
+    yticks = [1e1, 1e2, 1e3, 1e4, 1e5, 1e6],
+    xgrid = false,
+    xminorticks = false,
+    legendcolumns = 2,
+    legend_font_pointsize = 10,
+    size = (800, 600),
+    ylabel = "(#)",
+    xlabel = "process size (#)",
+)
+
+savefig("compton_diagram_gen_comparison.pdf")
--- a/data/evaluate_gpu.jl
+++ b/data/evaluate_gpu.jl
@ -0,0 +1,130 @@
+using CSV
+using DataFrames
+using Plots
+using StatsPlots
+using LaTeXStrings
+
+if (length(ARGS) < 1)
+    println("Please use with \"input_file.csv\"")
+end
+
+processes = [
+    "QED Process: 'ke->ke'",
+    "QED Process: 'ke->kke'",
+    "QED Process: 'ke->kkke'",
+    "QED Process: 'ke->kkkke'",
+    "QED Process: 'ke->kkkkke'",
+    "ABC Process: 'AB->AB'",
+    "ABC Process: 'AB->ABBB'",
+    "ABC Process: 'AB->ABBBBB'",
+]
+
+function proc_to_n(str::AbstractString)
+    parts = split(str, "'")
+    parts = split(parts[2], "->")
+    k_count = count(c -> c == 'k', parts[2])
+    return k_count
+end
+
+function beautify_title(str::AbstractString)
+    parts = split(str, "'")
+
+    preprefix = parts[1]
+    infix = parts[2]
+    sufsuffix = parts[3]
+
+    parts = split(infix, "->")
+
+    prefix = parts[1]
+    suffix = parts[2]
+
+    k_count = count(c -> c == 'k', suffix)
+    B_count = count(c -> c == 'B', suffix)
+
+    if k_count == 1 || B_count == 1
+        new_suffix = suffix
+    elseif k_count >= 1
+        new_suffix = replace(suffix, r"k+" => "k^$k_count")
+    elseif B_count >= 1
+        new_suffix = replace(suffix, r"B+" => "B^$B_count")
+    end
+
+    return preprefix * L"%$prefix \rightarrow %$new_suffix" * sufsuffix
+end
+
+input_file = ARGS[1]
+df = CSV.read(input_file, DataFrame)
+n_inputs = df[:, "n_inputs"][1]
+gpu_name = df[:, "gpu_name"][1]
+if (gpu_name == "")
+    println("Results file did not execute everything on GPU! (or didn't write gpu name)")
+    exit(0)
+end
+
+# plotting with process size as x axis
+title_string = "GPU $gpu_name, $n_inputs samples"
+
+df_filt = filter(:process_name => x -> proc_to_n(x) >= 1, df)
+df_filt.gpu_rate = df_filt.gpu_rate
+df_filt.gpu_time = df_filt.gpu_time
+df_filt.gpu_gflops = df_filt.gpu_gflops
+
+df_filt.process_size = @. proc_to_n(df_filt.process_name)
+
+df_no_opt = filter(:process_name => x -> match(r" not optimized$", x) !== nothing, df_filt)
+df_red = filter(:process_name => x -> match(r" reduced$", x) !== nothing, df_filt)
+
+@df df_no_opt scatter(:process_size, :gpu_rate, label = "unoptimized function execution rate", markersize = 7)
+@df df_red scatter!(:process_size, :gpu_rate, label = "reduced function execution rate", markersize = 7)
+
+plot!(
+    #title = title_string * ", sample rate",
+    yscale = :log10,
+    legend = :outerbottom,
+    xticks = [1, 2, 3, 4, 5],
+    legendcolumns = 2,
+    legend_font_pointsize = 10,
+    size = (800, 600),
+    ylabel = "rate (" * L"s^{-1}" * ")",
+    xlabel = "process size (#)",
+)
+
+savefig("gpu_rate_$(gpu_name).pdf")
+
+
+
+@df df_no_opt scatter(:process_size, :gpu_time, label = "unoptimized function execution time", markersize = 7)
+@df df_red scatter!(:process_size, :gpu_time, label = "reduced function execution time", markersize = 7)
+
+plot!(
+    #title = title_string * ", execution time",
+    yscale = :log10,
+    legend = :outerbottom,
+    xticks = [1, 2, 3, 4, 5],
+    legendcolumns = 2,
+    legend_font_pointsize = 10,
+    size = (800, 600),
+    ylabel = "time (s)",
+    xlabel = "process size (#)",
+)
+
+savefig("gpu_times_$(gpu_name).pdf")
+
+
+
+@df df_no_opt scatter(:process_size, :gpu_gflops, label = "unoptimized function", markersize = 7)
+@df df_red scatter!(:process_size, :gpu_gflops, label = "reduced function", markersize = 7)
+
+plot!(
+    #title = title_string * ", GFLOPS",
+    yscale = :linear,
+    legend = :outerbottom,
+    xticks = [1, 2, 3, 4, 5],
+    legendcolumns = 2,
+    legend_font_pointsize = 10,
+    size = (800, 600),
+    ylabel = "performance (GFLOPS)",
+    xlabel = "process size (#)",
+)
+
+savefig("gpu_perf_$(gpu_name).pdf")
--- a/data/evaluate_reduce_bench.jl
+++ b/data/evaluate_reduce_bench.jl
@ -0,0 +1,279 @@
+using Plots
+using StatsPlots
+using CSV
+using DataFrames
+using LaTeXStrings
+
+
+if (length(ARGS) < 2)
+    println("Please use with \"input_file.csv\" \"input_file_gpu.csv\"")
+end
+
+
+function proc_to_n(str::AbstractString)
+    parts = split(str, "'")
+    parts = split(parts[2], "->")
+    k_count = count(c -> c == 'k', parts[2])
+    return k_count
+end
+
+function beautify_title(str::AbstractString)
+    parts = split(str, "->")
+
+    prefix = parts[1]
+    suffix = parts[2]
+
+    k_count = count(c -> c == 'k', suffix)
+    B_count = count(c -> c == 'B', suffix)
+
+    if k_count == 1 || B_count == 1
+        new_suffix = suffix
+    elseif k_count >= 1
+        new_suffix = replace(suffix, r"k+" => "k^$k_count")
+    elseif B_count >= 1
+        new_suffix = replace(suffix, r"B+" => "B^$B_count")
+    end
+
+    return L"%$prefix \rightarrow %$new_suffix"
+end
+
+processes = ["ke->ke", "ke->kke", "ke->kkke", "ke->kkkke", "ke->kkkkke", "AB->AB", "AB->ABBB", "AB->ABBBBB"]
+
+input_file = ARGS[1]
+input_file_gpu = ARGS[2]
+df = CSV.read(input_file, DataFrame)
+df_gpu = CSV.read(input_file_gpu, DataFrame)
+n_inputs = 2^20
+
+#=
+for process in processes
+    df_filt = filter(:process => x -> x == process, df)
+    df_filt_gpu = filter(:process => x -> x == process, df_gpu)
+
+    # add dummy factors to get the numbers in similar orders of magnitude
+    df_filt.cumulative_optimization_time = df_filt.cumulative_optimization_time .* 1e4
+    df_filt_gpu.cumulative_optimization_time = df_filt_gpu.cumulative_optimization_time .* 1e4
+    df_filt_gpu.gpu_t = df_filt_gpu.gpu_t .* 1e3
+
+    cpu = !isempty(df_filt)
+    gpu = !isempty(df_filt_gpu)
+
+    ymax = 0.0
+
+    if cpu
+        @df df_filt scatter(
+            :operations,
+            :cumulative_optimization_time,
+            label = "Cumulative Optimization Time (x10000) (s)",
+            markersize = 4,
+        )
+        ymax = max(df_filt[!, :cpu_st_t]..., df_filt[!, :cumulative_optimization_time]...) * 1.1 * 1e4
+        @df df_filt scatter!(
+            :operations,
+            :cpu_st_t,
+            label = "Single-Threaded Execution (s)",
+            markersize = 4,
+            markershape = :square,
+        )
+    end
+
+    if gpu
+        if !cpu
+            @df df_filt_gpu scatter(
+                :operations,
+                :cumulative_optimization_time,
+                label = "Cumulative Optimization Time (x10000) (s)",
+                markersize = 4,
+            )
+            ymax = max(df_filt_gpu[!, :gpu_t]..., df_filt_gpu[!, :cumulative_optimization_time]...) * 1.1 * 1e4
+        end
+
+        @df df_filt_gpu scatter!(
+            :operations,
+            :gpu_t,
+            label = "GPU Execution (x1000) (s)",
+            markersize = 4,
+            markershape = :diamond,
+        )
+    end
+
+    if cpu || gpu
+        plot!(
+            title = ("$(beautify_title(process)) Reduction Progression ($(n_inputs) Inputs)"),
+            xscale = :linear,
+            yscale = :linear,
+            #ylim = (0, ymax),
+            legend = :outerbottom,
+            minorgrid = true,
+            xticks = :process_size,
+            #yticks = [1e-3, 1e-1, 1e1, 1e3],
+            xgrid = false,
+            xminorticks = false,
+            legendcolumns = 1,
+            legend_font_pointsize = 12,
+            fontsize = 12,
+            size = (800, 600),
+            xlabel = "optimizer steps (#)",
+            ylabel = "time (s)",
+        )
+
+        savefig("$(String(process))_reduction_bench.pdf")
+    end
+end
+
+
+# ABC vs QED
+AB_process = "AB->ABBB"
+abc_label = "\$A + B \\rightarrow A + 3B\$"
+
+QED_process = "ke->kkkke"
+qed_label = "\$e^- + \\gamma \\rightarrow e^- + 4\\gamma\$"
+
+df_filt_AB = filter(:process => x -> x == AB_process, df)
+df_filt_QED = filter(:process => x -> x == QED_process, df)
+
+max_AB = max(df_filt_AB[!, :operations]...)
+max_QED = max(df_filt_QED[!, :operations]...)
+
+df_filt_AB.reduction_progress = df_filt_AB.operations ./ max_AB .* 100.0
+df_filt_QED.reduction_progress = df_filt_QED.operations ./ max_QED .* 100.0
+
+df_filt_AB.relative_performance = df_filt_AB.cpu_st_t ./ df_filt_AB[!, :cpu_st_t][1] .* 100.0
+df_filt_QED.relative_performance = df_filt_QED.cpu_st_t ./ df_filt_QED[!, :cpu_st_t][1] .* 100.0
+
+@df df_filt_AB scatter(:reduction_progress, :relative_performance, label = abc_label, markersize = 4)
+@df df_filt_QED scatter!(:reduction_progress, :relative_performance, label = qed_label, markersize = 4)
+
+plot!(
+    #title = ("Relative Performance of $(beautify_title(QED_process)) versus $(beautify_title(AB_process)) on CPU"),
+    xscale = :linear,
+    yscale = :linear,
+    #ylim = (0, ymax),
+    legend = :outerbottom,
+    minorgrid = true,
+    xticks = :process_size,
+    #yticks = [1e-3, 1e-1, 1e1, 1e3],
+    xgrid = false,
+    xminorticks = false,
+    legendcolumns = 1,
+    legend_font_pointsize = 12,
+    fontsize = 12,
+    size = (800, 600),
+    xlabel = "reduction progress (%)",
+    ylabel = "relative time taken (%)",
+)
+
+savefig("reduction_bench_relative.pdf")
+
+
+# ABC vs QED on GPU
+AB_process = "AB->ABBB"
+abc_label = "\$A + B \\rightarrow A + 3B\$"
+
+QED_process = "ke->kkkke"
+qed_label = "\$e^- + \\gamma \\rightarrow e^- + 4\\gamma\$"
+
+df_filt_AB = filter(:process => x -> x == AB_process, df_gpu)
+df_filt_QED = filter(:process => x -> x == QED_process, df_gpu)
+
+max_AB = max(df_filt_AB[!, :operations]...)
+max_QED = max(df_filt_QED[!, :operations]...)
+
+df_filt_AB.reduction_progress = df_filt_AB.operations ./ max_AB .* 100.0
+df_filt_QED.reduction_progress = df_filt_QED.operations ./ max_QED .* 100.0
+
+df_filt_AB.relative_performance = df_filt_AB.gpu_t ./ df_filt_AB[!, :gpu_t][1] .* 100.0
+df_filt_QED.relative_performance = df_filt_QED.gpu_t ./ df_filt_QED[!, :gpu_t][1] .* 100.0
+
+df_filt_AB.relative_std = df_filt_AB.gpu_s ./ df_filt_AB[!, :gpu_t][1] .* 100.0
+df_filt_QED.relative_std = df_filt_QED.gpu_s ./ df_filt_QED[!, :gpu_t][1] .* 100.0
+
+@df df_filt_AB scatter(
+    :reduction_progress,
+    :relative_performance,
+    yerror = :relative_std,
+    label = abc_label,
+    markersize = 4,
+)
+@df df_filt_QED scatter!(
+    :reduction_progress,
+    :relative_performance,
+    yerror = :relative_std,
+    label = qed_label,
+    markersize = 4,
+)
+
+plot!(
+    #title = "Relative Performance of $(beautify_title(QED_process)) versus $(beautify_title(AB_process)) on GPU (A100)",
+    xscale = :linear,
+    yscale = :linear,
+    #ylim = (0, ymax),
+    legend = :outerbottom,
+    minorgrid = true,
+    xticks = :process_size,
+    #yticks = [1e-3, 1e-1, 1e1, 1e3],
+    xgrid = false,
+    xminorticks = false,
+    legendcolumns = 1,
+    legend_font_pointsize = 12,
+    fontsize = 12,
+    size = (800, 600),
+    xlabel = "reduction progress (%)",
+    ylabel = "relative time taken (%)",
+)
+
+savefig("reduction_bench_relative_gpu.pdf")
+=#
+
+QED_process = "ke->kkke"
+QED_label = "\$\\e^- + \\gamma \\rightarrow \\e^- + 3\\gamma\$"
+
+df_filt_QED_GPU = filter(:process => x -> x == QED_process, df_gpu)
+df_filt_QED_CPU = filter(:process => x -> x == QED_process, df)
+
+max_QED = max(df_filt_QED_CPU[!, :operations]...)
+
+df_filt_QED_GPU.reduction_progress = df_filt_QED_GPU.operations ./ max_QED .* 100.0
+df_filt_QED_CPU.reduction_progress = df_filt_QED_CPU.operations ./ max_QED .* 100.0
+
+df_filt_QED_GPU.relative_performance = df_filt_QED_GPU.gpu_t ./ df_filt_QED_GPU[!, :gpu_t][1] .* 100.0
+df_filt_QED_CPU.relative_performance = df_filt_QED_CPU.cpu_st_t ./ df_filt_QED_CPU[!, :cpu_st_t][1] .* 100.0
+
+df_filt_QED_GPU.relative_std = df_filt_QED_GPU.gpu_s ./ df_filt_QED_GPU[!, :gpu_t][1] .* 100.0
+df_filt_QED_CPU.relative_std = df_filt_QED_CPU.cpu_st_s ./ df_filt_QED_CPU[!, :cpu_st_t][1] .* 100.0
+
+@df df_filt_QED_CPU scatter(
+    :reduction_progress,
+    :relative_performance,
+    yerror = :relative_std,
+    label = "CPU relative time taken (%)",
+    markersize = 4,
+)
+@df df_filt_QED_GPU scatter!(
+    :reduction_progress,
+    :relative_performance,
+    yerror = :relative_std,
+    label = "GPU relative time taken (%)",
+    markersize = 4,
+)
+
+plot!(
+    #title = "Relative Performance of $(beautify_title(QED_process)) on CPU versus GPU",
+    xscale = :linear,
+    yscale = :linear,
+    ylim = (0, :auto),
+    legend = :outerbottom,
+    minorgrid = true,
+    xticks = :process_size,
+    #yticks = [1e-3, 1e-1, 1e1, 1e3],
+    xgrid = false,
+    xminorticks = false,
+    legendcolumns = 1,
+    legend_font_pointsize = 12,
+    fontsize = 12,
+    size = (800, 600),
+    xlabel = "reduction progress (%)",
+    ylabel = "relative time taken (%)",
+)
+
+savefig("reduction_bench_relative_cpu_vs_gpu.pdf")
--- a/data/results.zip
+++ b/data/results.zip
--- a/docs/Project.toml
+++ b/docs/Project.toml
@ -2,3 +2,4 @@
 Documenter = "e30172f5-a6a5-5a46-863b-614d45cd2de4"
 DocumenterTools = "35a29f4d-8980-5a13-9543-d66fff28ecb8"
 MetagraphOptimization = "3e869610-d48d-4942-ba70-c1b702a33ca4"
+QEDprocesses = "46de9c38-1bb3-4547-a1ec-da24d767fdad"
--- a/docs/src/QED-Bhabha-DAG.drawio
+++ b/docs/src/QED-Bhabha-DAG.drawio
@ -0,0 +1,259 @@
+<mxfile host="Electron" modified="2023-11-25T19:36:18.149Z" agent="Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) draw.io/21.6.1 Chrome/114.0.5735.289 Electron/25.9.4 Safari/537.36" etag="hBUSDG3lmnElLEv2sh-Z" version="21.6.1" type="device">
+  <diagram name="Page-1" id="pzz-rsbNjEFeZeQIA-38">
+    <mxGraphModel dx="1430" dy="853" grid="1" gridSize="10" guides="1" tooltips="1" connect="1" arrows="1" fold="1" page="1" pageScale="1" pageWidth="500" pageHeight="900" math="0" shadow="0">
+      <root>
+        <mxCell id="0" />
+        <mxCell id="1" parent="0" />
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-19" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-6" target="KG5lhhUBjoQ79gfvQvIC-18" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-6" value="&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#336600&quot;&gt;base_state&lt;/font&gt;&lt;/b&gt;&amp;nbsp;-&amp;gt;&lt;br&gt;&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#cc00cc&quot;&gt;AdjointBiSpinor&lt;/font&gt;&lt;/b&gt;" style="rounded=1;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#b3b3b3;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="190" y="600" width="120" height="60" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-20" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-7" target="KG5lhhUBjoQ79gfvQvIC-18" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-7" value="&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#336600&quot;&gt;base_state&lt;/font&gt;&lt;/b&gt;&amp;nbsp;-&amp;gt;&lt;br&gt;&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#cc00cc&quot;&gt;BiSpinor&lt;/font&gt;&lt;/b&gt;" style="rounded=1;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#b3b3b3;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="310" y="600" width="120" height="60" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-28" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-8" target="KG5lhhUBjoQ79gfvQvIC-26" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-8" value="&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#336600&quot;&gt;base_state&lt;/font&gt;&lt;/b&gt;&amp;nbsp;-&amp;gt;&lt;br&gt;&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#cc00cc&quot;&gt;AdjointBiSpinor&lt;/font&gt;&lt;/b&gt;" style="rounded=1;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#b3b3b3;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="550" y="600" width="120" height="60" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-29" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-9" target="KG5lhhUBjoQ79gfvQvIC-26" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-9" value="&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#336600&quot;&gt;base_state&lt;/font&gt;&lt;/b&gt;&amp;nbsp;-&amp;gt;&lt;br&gt;&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#cc00cc&quot;&gt;BiSpinor&lt;/font&gt;&lt;/b&gt;" style="rounded=1;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#b3b3b3;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="430" y="600" width="120" height="60" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-21" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-14" target="KG5lhhUBjoQ79gfvQvIC-6" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-14" value="U(p&lt;sub&gt;1&lt;/sub&gt;)&lt;br style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font color=&quot;#0000ff&quot;&gt;&lt;b&gt;IncomingAntiFermion&lt;/b&gt;&lt;/font&gt;&lt;br&gt;Outer Edge" style="ellipse;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#9AC7BF;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="190" y="700" width="120" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-22" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-15" target="KG5lhhUBjoQ79gfvQvIC-7" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-15" value="U(e&lt;sub style=&quot;border-color: var(--border-color);&quot;&gt;1&lt;/sub&gt;)&lt;br style=&quot;border-color: var(--border-color);&quot;&gt;&lt;b&gt;&lt;font color=&quot;#0000ff&quot;&gt;IncomingFermion&lt;br&gt;&lt;/font&gt;&lt;/b&gt;Outer Edge" style="ellipse;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#9AC7BF;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="310" y="700" width="120" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-23" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-16" target="KG5lhhUBjoQ79gfvQvIC-9" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-16" value="U(p&lt;span style=&quot;font-size: 10px;&quot;&gt;&lt;sub&gt;2&lt;/sub&gt;&lt;/span&gt;)&lt;br style=&quot;border-color: var(--border-color);&quot;&gt;&lt;b&gt;&lt;font color=&quot;#0000ff&quot;&gt;OutgoingAntiFermion&lt;br&gt;&lt;/font&gt;&lt;/b&gt;Outer Edge" style="ellipse;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#9AC7BF;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="430" y="700" width="120" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-24" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-17" target="KG5lhhUBjoQ79gfvQvIC-8" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-17" value="U(e&lt;sub style=&quot;border-color: var(--border-color);&quot;&gt;2&lt;/sub&gt;)&lt;br style=&quot;border-color: var(--border-color);&quot;&gt;&lt;b&gt;&lt;font color=&quot;#0000ff&quot;&gt;OutgoingFermion&lt;br&gt;&lt;/font&gt;&lt;/b&gt;Outer Edge" style="ellipse;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#9AC7BF;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="550" y="700" width="120" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-30" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-18" target="KG5lhhUBjoQ79gfvQvIC-25" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-18" value="V(e&lt;sub&gt;1&lt;/sub&gt;, p&lt;sub&gt;1&lt;/sub&gt;)&lt;br&gt;Result Particle:&lt;br&gt;q: &lt;font color=&quot;#0000ff&quot;&gt;&lt;b&gt;OutgoingPhoton&lt;br&gt;&lt;/b&gt;&lt;/font&gt;" style="ellipse;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#9AC7BF;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="240" y="480" width="140" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-36" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-25" target="KG5lhhUBjoQ79gfvQvIC-35" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-25" value="&lt;b&gt;&lt;font color=&quot;#336600&quot;&gt;vertex&lt;/font&gt;&lt;/b&gt; -&amp;gt;&lt;br&gt;AdjointBiSpinor * &lt;br&gt;DiracMatrix *&lt;br&gt;BiSpinor&lt;br&gt;=&amp;nbsp;&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font color=&quot;#cc00cc&quot; style=&quot;border-color: var(--border-color);&quot;&gt;Complex&lt;br&gt;&lt;/font&gt;&lt;/b&gt;" style="rounded=1;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#b3b3b3;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="250" y="360" width="120" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-32" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-26" target="KG5lhhUBjoQ79gfvQvIC-27" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-26" value="V(e&lt;sub&gt;2&lt;/sub&gt;, p&lt;sub&gt;2&lt;/sub&gt;)&lt;br&gt;Result Particle:&lt;br&gt;q&#39;: &lt;b&gt;&lt;font color=&quot;#0000ff&quot;&gt;IncomingPhoton&lt;/font&gt;&lt;/b&gt;" style="ellipse;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#9AC7BF;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="480" y="480" width="140" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-37" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-27" target="KG5lhhUBjoQ79gfvQvIC-35" edge="1">
+          <mxGeometry relative="1" as="geometry">
+            <Array as="points">
+              <mxPoint x="550" y="340" />
+              <mxPoint x="420" y="340" />
+            </Array>
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-27" value="&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#336600&quot;&gt;vertex&lt;/font&gt;&lt;/b&gt;&amp;nbsp;-&amp;gt;&lt;br&gt;AdjointBiSpinor * &lt;br&gt;DiracMatrix *&lt;br&gt;BiSpinor&lt;br&gt;=&amp;nbsp;&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font color=&quot;#cc00cc&quot; style=&quot;border-color: var(--border-color);&quot;&gt;Complex&lt;/font&gt;&lt;/b&gt;" style="rounded=1;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#b3b3b3;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="490" y="360" width="120" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-39" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-35" target="KG5lhhUBjoQ79gfvQvIC-38" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-35" value="S2(q, q&#39;)&lt;br&gt;q == -q&#39;&lt;br&gt;&lt;b&gt;&lt;font color=&quot;#0000ff&quot;&gt;No Result Particle&lt;/font&gt;&lt;/b&gt;" style="ellipse;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#9AC7BF;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="360" y="240" width="120" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-71" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-38" target="KG5lhhUBjoQ79gfvQvIC-70" edge="1">
+          <mxGeometry relative="1" as="geometry">
+            <Array as="points">
+              <mxPoint x="670" y="170" />
+            </Array>
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-38" value="&lt;b style=&quot;&quot;&gt;&lt;font color=&quot;#336600&quot;&gt;inner_edge&lt;/font&gt;&lt;/b&gt; -&amp;gt;&lt;br&gt;Complex * propagator(q) * Complex&amp;nbsp;&lt;br&gt;=&amp;nbsp;&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#cc00cc&quot;&gt;Complex&lt;br&gt;&lt;/font&gt;&lt;/b&gt;" style="rounded=1;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#b3b3b3;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="350" y="140" width="140" height="60" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-40" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-41" target="KG5lhhUBjoQ79gfvQvIC-57" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-41" value="&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#336600&quot;&gt;base_state&lt;/font&gt;&lt;/b&gt;&amp;nbsp;-&amp;gt;&lt;br&gt;&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#cc00cc&quot;&gt;AdjointBiSpinor&lt;/font&gt;&lt;/b&gt;" style="rounded=1;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#b3b3b3;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="670" y="600" width="120" height="60" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-68" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-43" target="KG5lhhUBjoQ79gfvQvIC-61" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-43" value="&lt;b style=&quot;&quot;&gt;&lt;font color=&quot;#336600&quot;&gt;base_state&lt;/font&gt;&lt;/b&gt; -&amp;gt;&lt;br&gt;&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#cc00cc&quot;&gt;AdjointBiSpinor&lt;/font&gt;&lt;/b&gt;" style="rounded=1;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#b3b3b3;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="1030" y="600" width="120" height="60" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-67" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-45" target="KG5lhhUBjoQ79gfvQvIC-57" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-45" value="&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#336600&quot;&gt;base_state&lt;/font&gt;&lt;/b&gt;&amp;nbsp;-&amp;gt;&lt;br&gt;&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#cc00cc&quot;&gt;BiSpinor&lt;/font&gt;&lt;/b&gt;" style="rounded=1;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#b3b3b3;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="790" y="600" width="120" height="60" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-46" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-47" target="KG5lhhUBjoQ79gfvQvIC-61" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-47" value="&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#336600&quot;&gt;base_state&lt;/font&gt;&lt;/b&gt;&amp;nbsp;-&amp;gt;&lt;br&gt;&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#cc00cc&quot;&gt;BiSpinor&lt;/font&gt;&lt;/b&gt;" style="rounded=1;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#b3b3b3;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="910" y="600" width="120" height="60" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-48" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-49" target="KG5lhhUBjoQ79gfvQvIC-41" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-49" value="U(p&lt;sub style=&quot;border-color: var(--border-color);&quot;&gt;1&lt;/sub&gt;)&lt;br style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#0000ff&quot;&gt;&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;IncomingAntiFermion&lt;/b&gt;&lt;/font&gt;&lt;br style=&quot;border-color: var(--border-color);&quot;&gt;Outer Edge" style="ellipse;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#9AC7BF;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="670" y="700" width="120" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-69" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-51" target="KG5lhhUBjoQ79gfvQvIC-43" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-51" value="U(e&lt;sub&gt;2&lt;/sub&gt;)&lt;br style=&quot;border-color: var(--border-color);&quot;&gt;&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#0000ff&quot;&gt;OutgoingFermion&lt;br style=&quot;border-color: var(--border-color);&quot;&gt;&lt;/font&gt;&lt;/b&gt;Outer Edge" style="ellipse;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#9AC7BF;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="1030" y="700" width="120" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-52" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-53" target="KG5lhhUBjoQ79gfvQvIC-47" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-53" value="U(e&lt;sub&gt;1&lt;/sub&gt;)&lt;br style=&quot;border-color: var(--border-color);&quot;&gt;&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#0000ff&quot;&gt;IncomingFermion&lt;br style=&quot;border-color: var(--border-color);&quot;&gt;&lt;/font&gt;&lt;/b&gt;Outer Edge" style="ellipse;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#9AC7BF;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="910" y="700" width="120" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-54" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-55" target="KG5lhhUBjoQ79gfvQvIC-45" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-55" value="U(p&lt;sub style=&quot;border-color: var(--border-color);&quot;&gt;2&lt;/sub&gt;)&lt;br style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font color=&quot;#0000ff&quot;&gt;&lt;b&gt;OutgoingAntiFermion&lt;br&gt;&lt;/b&gt;&lt;/font&gt;Outer Edge" style="ellipse;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#9AC7BF;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="790" y="700" width="120" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-56" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-57" target="KG5lhhUBjoQ79gfvQvIC-59" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-57" value="&lt;font style=&quot;font-size: 12px;&quot;&gt;V(p&lt;sub&gt;1&lt;/sub&gt;, p&lt;sub&gt;2&lt;/sub&gt;)&lt;/font&gt;&lt;br&gt;Result Particle:&lt;br&gt;q: &lt;b&gt;&lt;font color=&quot;#0000ff&quot;&gt;IncomingPhoton&lt;/font&gt;&lt;/b&gt;" style="ellipse;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#9AC7BF;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="720" y="480" width="140" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-58" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-59" target="KG5lhhUBjoQ79gfvQvIC-65" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-59" value="&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#336600&quot;&gt;vertex&lt;/font&gt;&lt;/b&gt;&amp;nbsp;-&amp;gt;&lt;br&gt;AdjointBiSpinor * &lt;br&gt;DiracMatrix *&lt;br&gt;BiSpinor&lt;br&gt;=&amp;nbsp;&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#cc00cc&quot;&gt;Complex&lt;/font&gt;&lt;/b&gt;" style="rounded=1;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#b3b3b3;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="730" y="360" width="120" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-60" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-61" target="KG5lhhUBjoQ79gfvQvIC-63" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-61" value="&lt;font style=&quot;font-size: 12px;&quot;&gt;V(e&lt;sub&gt;1&lt;/sub&gt;, e&lt;sub&gt;2&lt;/sub&gt;)&lt;/font&gt;&lt;br&gt;Result Particle:&lt;br&gt;q&#39;: &lt;font color=&quot;#0000ff&quot;&gt;&lt;b&gt;OutgoingPhoton&lt;/b&gt;&lt;/font&gt;" style="ellipse;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#9AC7BF;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="960" y="480" width="140" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-62" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-63" target="KG5lhhUBjoQ79gfvQvIC-65" edge="1">
+          <mxGeometry relative="1" as="geometry">
+            <Array as="points">
+              <mxPoint x="1030" y="340" />
+              <mxPoint x="900" y="340" />
+            </Array>
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-63" value="&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#336600&quot;&gt;vertex&lt;/font&gt;&lt;/b&gt;&amp;nbsp;-&amp;gt;&lt;br&gt;AdjointBiSpinor * &lt;br&gt;DiracMatrix * &lt;br&gt;BiSpinor &lt;br&gt;=&amp;nbsp;&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#cc00cc&quot;&gt;Complex&lt;br&gt;&lt;/font&gt;&lt;/b&gt;" style="rounded=1;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#b3b3b3;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="970" y="360" width="120" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-64" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-65" target="KG5lhhUBjoQ79gfvQvIC-66" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-65" value="S2(q, q&#39;)&lt;br&gt;q == -q&#39;&lt;br&gt;&lt;b&gt;&lt;font color=&quot;#0000ff&quot;&gt;No Result Particle&lt;/font&gt;&lt;/b&gt;" style="ellipse;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#9AC7BF;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="840" y="240" width="120" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-72" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-66" target="KG5lhhUBjoQ79gfvQvIC-70" edge="1">
+          <mxGeometry relative="1" as="geometry">
+            <Array as="points">
+              <mxPoint x="670" y="170" />
+            </Array>
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-66" value="&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#336600&quot;&gt;inner_edge&lt;/font&gt;&lt;/b&gt;&amp;nbsp;-&amp;gt;&lt;br&gt;Complex * propagator(q) * Complex&lt;br&gt;=&amp;nbsp;&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font color=&quot;#cc00cc&quot;&gt;Complex&lt;/font&gt;&lt;/b&gt;" style="rounded=1;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#b3b3b3;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="830" y="140" width="140" height="60" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-70" value="&lt;span style=&quot;font-size: 32px;&quot;&gt;diff()&lt;/span&gt;" style="ellipse;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#ffcd28;strokeColor=#d79b00;gradientColor=#ffa500;" parent="1" vertex="1">
+          <mxGeometry x="610" y="20" width="120" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-73" value="&lt;b&gt;&lt;font style=&quot;font-size: 22px;&quot;&gt;ValueAcc&lt;/font&gt;&lt;/b&gt;" style="text;html=1;strokeColor=none;fillColor=none;align=center;verticalAlign=middle;whiteSpace=wrap;rounded=0;" parent="1" vertex="1">
+          <mxGeometry x="10" y="155" width="120" height="30" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-74" value="&lt;b&gt;&lt;font style=&quot;font-size: 22px;&quot;&gt;ValueAcc&lt;/font&gt;&lt;/b&gt;" style="text;html=1;strokeColor=none;fillColor=none;align=center;verticalAlign=middle;whiteSpace=wrap;rounded=0;" parent="1" vertex="1">
+          <mxGeometry x="10" y="385" width="120" height="30" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-75" value="&lt;b&gt;&lt;font style=&quot;font-size: 22px;&quot;&gt;ValueAcc&lt;/font&gt;&lt;/b&gt;" style="text;html=1;strokeColor=none;fillColor=none;align=center;verticalAlign=middle;whiteSpace=wrap;rounded=0;" parent="1" vertex="1">
+          <mxGeometry x="10" y="615" width="120" height="30" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-76" value="&lt;b&gt;&lt;font style=&quot;font-size: 22px;&quot;&gt;Particle Propagation&lt;/font&gt;&lt;/b&gt;" style="text;html=1;strokeColor=none;fillColor=none;align=center;verticalAlign=middle;whiteSpace=wrap;rounded=0;" parent="1" vertex="1">
+          <mxGeometry x="10" y="725" width="120" height="30" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-77" value="&lt;b&gt;&lt;font style=&quot;font-size: 22px;&quot;&gt;Particle Propagation&lt;/font&gt;&lt;/b&gt;" style="text;html=1;strokeColor=none;fillColor=none;align=center;verticalAlign=middle;whiteSpace=wrap;rounded=0;" parent="1" vertex="1">
+          <mxGeometry x="10" y="505" width="120" height="30" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-78" value="&lt;b&gt;&lt;font style=&quot;font-size: 22px;&quot;&gt;Particle Propagation&lt;/font&gt;&lt;/b&gt;" style="text;html=1;strokeColor=none;fillColor=none;align=center;verticalAlign=middle;whiteSpace=wrap;rounded=0;" parent="1" vertex="1">
+          <mxGeometry x="10" y="265" width="120" height="30" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-79" value="" style="endArrow=none;dashed=1;html=1;dashPattern=1 3;strokeWidth=2;rounded=0;" parent="1" edge="1">
+          <mxGeometry width="50" height="50" relative="1" as="geometry">
+            <mxPoint y="220" as="sourcePoint" />
+            <mxPoint x="1160" y="220" as="targetPoint" />
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-80" value="" style="endArrow=none;dashed=1;html=1;dashPattern=1 3;strokeWidth=2;rounded=0;" parent="1" edge="1">
+          <mxGeometry width="50" height="50" relative="1" as="geometry">
+            <mxPoint y="340" as="sourcePoint" />
+            <mxPoint x="1160" y="340" as="targetPoint" />
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-81" value="" style="endArrow=none;dashed=1;html=1;dashPattern=1 3;strokeWidth=2;rounded=0;" parent="1" edge="1">
+          <mxGeometry width="50" height="50" relative="1" as="geometry">
+            <mxPoint y="460" as="sourcePoint" />
+            <mxPoint x="1160" y="460" as="targetPoint" />
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-82" value="" style="endArrow=none;dashed=1;html=1;dashPattern=1 3;strokeWidth=2;rounded=0;" parent="1" edge="1">
+          <mxGeometry width="50" height="50" relative="1" as="geometry">
+            <mxPoint y="120" as="sourcePoint" />
+            <mxPoint x="1160" y="120" as="targetPoint" />
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-83" value="" style="endArrow=none;dashed=1;html=1;dashPattern=1 3;strokeWidth=2;rounded=0;" parent="1" edge="1">
+          <mxGeometry width="50" height="50" relative="1" as="geometry">
+            <mxPoint y="580" as="sourcePoint" />
+            <mxPoint x="1160" y="580" as="targetPoint" />
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-84" value="" style="endArrow=none;dashed=1;html=1;dashPattern=1 3;strokeWidth=2;rounded=0;" parent="1" edge="1">
+          <mxGeometry width="50" height="50" relative="1" as="geometry">
+            <mxPoint y="680" as="sourcePoint" />
+            <mxPoint x="1160" y="680" as="targetPoint" />
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-85" value="" style="endArrow=none;dashed=1;html=1;dashPattern=1 3;strokeWidth=2;rounded=0;" parent="1" edge="1">
+          <mxGeometry width="50" height="50" relative="1" as="geometry">
+            <mxPoint y="800" as="sourcePoint" />
+            <mxPoint x="1160" y="800" as="targetPoint" />
+          </mxGeometry>
+        </mxCell>
+      </root>
+    </mxGraphModel>
+  </diagram>
+</mxfile>
--- a/docs/src/QED-Compton-DAG.drawio
+++ b/docs/src/QED-Compton-DAG.drawio
@ -0,0 +1,259 @@
+<mxfile host="Electron" modified="2023-11-25T19:38:46.724Z" agent="Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) draw.io/21.6.1 Chrome/114.0.5735.289 Electron/25.9.4 Safari/537.36" etag="GgWoS9drZkpaqTCZiHI_" version="21.6.1" type="device">
+  <diagram name="Page-1" id="pzz-rsbNjEFeZeQIA-38">
+    <mxGraphModel dx="1430" dy="853" grid="1" gridSize="10" guides="1" tooltips="1" connect="1" arrows="1" fold="1" page="1" pageScale="1" pageWidth="500" pageHeight="900" math="0" shadow="0">
+      <root>
+        <mxCell id="0" />
+        <mxCell id="1" parent="0" />
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-19" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-6" target="KG5lhhUBjoQ79gfvQvIC-18" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-6" value="&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#336600&quot;&gt;base_state&lt;/font&gt;&lt;/b&gt;&amp;nbsp;-&amp;gt;&lt;br&gt;&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font color=&quot;#cc00cc&quot;&gt;LorentzVector&lt;/font&gt;&lt;/b&gt;" style="rounded=1;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#b3b3b3;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="190" y="600" width="120" height="60" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-20" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-7" target="KG5lhhUBjoQ79gfvQvIC-18" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-7" value="&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#336600&quot;&gt;base_state&lt;/font&gt;&lt;/b&gt;&amp;nbsp;-&amp;gt;&lt;br&gt;&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#cc00cc&quot;&gt;BiSpinor&lt;/font&gt;&lt;/b&gt;" style="rounded=1;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#b3b3b3;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="310" y="600" width="120" height="60" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-28" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-8" target="KG5lhhUBjoQ79gfvQvIC-26" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-8" value="&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#336600&quot;&gt;base_state&lt;/font&gt;&lt;/b&gt;&amp;nbsp;-&amp;gt;&lt;br&gt;&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#cc00cc&quot;&gt;AdjointBiSpinor&lt;/font&gt;&lt;/b&gt;" style="rounded=1;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#b3b3b3;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="550" y="600" width="120" height="60" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-29" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-9" target="KG5lhhUBjoQ79gfvQvIC-26" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-9" value="&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#336600&quot;&gt;base_state&lt;/font&gt;&lt;/b&gt;&amp;nbsp;-&amp;gt;&lt;br&gt;&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#cc00cc&quot;&gt;LorentzVector&lt;/font&gt;&lt;/b&gt;" style="rounded=1;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#b3b3b3;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="430" y="600" width="120" height="60" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-21" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-14" target="KG5lhhUBjoQ79gfvQvIC-6" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-14" value="U(γ&lt;sub style=&quot;border-color: var(--border-color);&quot;&gt;1&lt;/sub&gt;)&lt;br style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font color=&quot;#0000ff&quot;&gt;&lt;b&gt;IncomingPhoton&lt;/b&gt;&lt;/font&gt;&lt;br&gt;Outer Edge" style="ellipse;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#9AC7BF;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="190" y="700" width="120" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-22" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-15" target="KG5lhhUBjoQ79gfvQvIC-7" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-15" value="U(p&lt;sub style=&quot;border-color: var(--border-color);&quot;&gt;1&lt;/sub&gt;)&lt;br style=&quot;border-color: var(--border-color);&quot;&gt;&lt;b&gt;&lt;font color=&quot;#0000ff&quot;&gt;IncomingFermion&lt;br&gt;&lt;/font&gt;&lt;/b&gt;Outer Edge" style="ellipse;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#9AC7BF;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="310" y="700" width="120" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-23" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-16" target="KG5lhhUBjoQ79gfvQvIC-9" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-16" value="U(γ&lt;sub style=&quot;border-color: var(--border-color);&quot;&gt;2&lt;/sub&gt;)&lt;br style=&quot;border-color: var(--border-color);&quot;&gt;&lt;b&gt;&lt;font color=&quot;#0000ff&quot;&gt;OutgoingPhoton&lt;br&gt;&lt;/font&gt;&lt;/b&gt;Outer Edge" style="ellipse;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#9AC7BF;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="430" y="700" width="120" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-24" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-17" target="KG5lhhUBjoQ79gfvQvIC-8" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-17" value="U(p&lt;sub style=&quot;border-color: var(--border-color);&quot;&gt;2&lt;/sub&gt;)&lt;br style=&quot;border-color: var(--border-color);&quot;&gt;&lt;b&gt;&lt;font color=&quot;#0000ff&quot;&gt;OutgoingFermion&lt;br&gt;&lt;/font&gt;&lt;/b&gt;Outer Edge" style="ellipse;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#9AC7BF;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="550" y="700" width="120" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-30" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-18" target="KG5lhhUBjoQ79gfvQvIC-25" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-18" value="V(γ&lt;sub&gt;1&lt;/sub&gt;, p&lt;sub&gt;1&lt;/sub&gt;)&lt;br&gt;Result Particle:&lt;br&gt;q: &lt;font color=&quot;#0000ff&quot;&gt;&lt;b&gt;OutgoingFermion&lt;/b&gt;&lt;/font&gt;" style="ellipse;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#9AC7BF;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="240" y="480" width="140" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-36" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-25" target="KG5lhhUBjoQ79gfvQvIC-35" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-25" value="&lt;b&gt;&lt;font color=&quot;#336600&quot;&gt;vertex&lt;/font&gt;&lt;/b&gt; -&amp;gt;&lt;br&gt;LorentzVector * DiracMatrix * &lt;br&gt;BiSpinor&lt;br&gt;=&amp;nbsp;&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font color=&quot;#cc00cc&quot; style=&quot;border-color: var(--border-color);&quot;&gt;BiSpinor&lt;/font&gt;&lt;/b&gt;" style="rounded=1;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#b3b3b3;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="250" y="360" width="120" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-32" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-26" target="KG5lhhUBjoQ79gfvQvIC-27" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-26" value="V(γ&lt;sub&gt;2&lt;/sub&gt;, p&lt;sub&gt;2&lt;/sub&gt;)&lt;br&gt;Result Particle:&lt;br&gt;q&#39;: &lt;b&gt;&lt;font color=&quot;#0000ff&quot;&gt;IncomingFermion&lt;/font&gt;&lt;/b&gt;" style="ellipse;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#9AC7BF;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="480" y="480" width="140" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-37" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-27" target="KG5lhhUBjoQ79gfvQvIC-35" edge="1">
+          <mxGeometry relative="1" as="geometry">
+            <Array as="points">
+              <mxPoint x="550" y="340" />
+              <mxPoint x="420" y="340" />
+            </Array>
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-27" value="&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#336600&quot;&gt;vertex&lt;/font&gt;&lt;/b&gt;&amp;nbsp;-&amp;gt;&lt;br&gt;AdjointBiSpinor * LorentzVector * &lt;br&gt;DiracMatrix &lt;br&gt;=&amp;nbsp;&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font color=&quot;#cc00cc&quot; style=&quot;border-color: var(--border-color);&quot;&gt;AdjointBiSpinor&lt;/font&gt;&lt;/b&gt;" style="rounded=1;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#b3b3b3;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="490" y="360" width="120" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-39" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-35" target="KG5lhhUBjoQ79gfvQvIC-38" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-35" value="S2(q, q&#39;)&lt;br&gt;q == -q&#39;&lt;br&gt;&lt;b&gt;&lt;font color=&quot;#0000ff&quot;&gt;No Result Particle&lt;/font&gt;&lt;/b&gt;" style="ellipse;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#9AC7BF;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="360" y="240" width="120" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-71" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-38" target="KG5lhhUBjoQ79gfvQvIC-70" edge="1">
+          <mxGeometry relative="1" as="geometry">
+            <Array as="points">
+              <mxPoint x="670" y="170" />
+            </Array>
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-38" value="&lt;b style=&quot;&quot;&gt;&lt;font color=&quot;#336600&quot;&gt;inner_edge&lt;/font&gt;&lt;/b&gt; -&amp;gt;&lt;br&gt;AdjointBiSpinor * propagator(q) * BiSpinor =&amp;nbsp;&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#cc00cc&quot;&gt;ComplexF64&lt;/font&gt;&lt;/b&gt;" style="rounded=1;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#b3b3b3;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="350" y="140" width="140" height="60" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-40" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-41" target="KG5lhhUBjoQ79gfvQvIC-57" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-41" value="&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#336600&quot;&gt;base_state&lt;/font&gt;&lt;/b&gt;&amp;nbsp;-&amp;gt;&lt;br&gt;&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#cc00cc&quot;&gt;LorentzVector&lt;/font&gt;&lt;/b&gt;" style="rounded=1;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#b3b3b3;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="670" y="600" width="120" height="60" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-68" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-43" target="KG5lhhUBjoQ79gfvQvIC-61" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-43" value="&lt;b style=&quot;&quot;&gt;&lt;font color=&quot;#336600&quot;&gt;base_state&lt;/font&gt;&lt;/b&gt; -&amp;gt;&lt;br&gt;&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#cc00cc&quot;&gt;BiSpinor&lt;/font&gt;&lt;/b&gt;" style="rounded=1;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#b3b3b3;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="1030" y="600" width="120" height="60" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-67" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-45" target="KG5lhhUBjoQ79gfvQvIC-57" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-45" value="&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#336600&quot;&gt;base_state&lt;/font&gt;&lt;/b&gt;&amp;nbsp;-&amp;gt;&lt;br&gt;&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#cc00cc&quot;&gt;AdjointBiSpinor&lt;/font&gt;&lt;/b&gt;" style="rounded=1;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#b3b3b3;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="790" y="600" width="120" height="60" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-46" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-47" target="KG5lhhUBjoQ79gfvQvIC-61" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-47" value="&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#336600&quot;&gt;base_state&lt;/font&gt;&lt;/b&gt;&amp;nbsp;-&amp;gt;&lt;br&gt;&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#cc00cc&quot;&gt;LorentzVector&lt;/font&gt;&lt;/b&gt;" style="rounded=1;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#b3b3b3;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="910" y="600" width="120" height="60" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-48" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-49" target="KG5lhhUBjoQ79gfvQvIC-41" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-49" value="U(γ&lt;sub style=&quot;border-color: var(--border-color);&quot;&gt;1&lt;/sub&gt;)&lt;br style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#0000ff&quot;&gt;&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;IncomingPhoton&lt;/b&gt;&lt;/font&gt;&lt;br style=&quot;border-color: var(--border-color);&quot;&gt;Outer Edge" style="ellipse;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#9AC7BF;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="670" y="700" width="120" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-69" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-51" target="KG5lhhUBjoQ79gfvQvIC-43" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-51" value="U(p&lt;sub style=&quot;border-color: var(--border-color);&quot;&gt;1&lt;/sub&gt;)&lt;br style=&quot;border-color: var(--border-color);&quot;&gt;&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#0000ff&quot;&gt;IncomingFermion&lt;br style=&quot;border-color: var(--border-color);&quot;&gt;&lt;/font&gt;&lt;/b&gt;Outer Edge" style="ellipse;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#9AC7BF;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="1030" y="700" width="120" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-52" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-53" target="KG5lhhUBjoQ79gfvQvIC-47" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-53" value="U(γ&lt;sub style=&quot;border-color: var(--border-color);&quot;&gt;2&lt;/sub&gt;)&lt;br style=&quot;border-color: var(--border-color);&quot;&gt;&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#0000ff&quot;&gt;OutgoingPhoton&lt;br style=&quot;border-color: var(--border-color);&quot;&gt;&lt;/font&gt;&lt;/b&gt;Outer Edge" style="ellipse;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#9AC7BF;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="910" y="700" width="120" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-54" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-55" target="KG5lhhUBjoQ79gfvQvIC-45" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-55" value="U(p&lt;sub style=&quot;border-color: var(--border-color);&quot;&gt;2&lt;/sub&gt;)&lt;br style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font color=&quot;#0000ff&quot;&gt;&lt;b&gt;OutgoingFermion&lt;br&gt;&lt;/b&gt;&lt;/font&gt;Outer Edge" style="ellipse;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#9AC7BF;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="790" y="700" width="120" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-56" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-57" target="KG5lhhUBjoQ79gfvQvIC-59" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-57" value="V(γ&lt;sub&gt;1&lt;/sub&gt;, p&lt;span style=&quot;font-size: 10px;&quot;&gt;&lt;sub&gt;2&lt;/sub&gt;&lt;/span&gt;)&lt;br&gt;Result Particle:&lt;br&gt;q: &lt;b&gt;&lt;font color=&quot;#0000ff&quot;&gt;IncomingFermion&lt;/font&gt;&lt;/b&gt;" style="ellipse;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#9AC7BF;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="720" y="480" width="140" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-58" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-59" target="KG5lhhUBjoQ79gfvQvIC-65" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-59" value="&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#336600&quot;&gt;vertex&lt;/font&gt;&lt;/b&gt;&amp;nbsp;-&amp;gt;&lt;br&gt;AdjointBiSpinor * LorentzVector * &lt;br&gt;DiracMatrix&amp;nbsp;&lt;br&gt;=&amp;nbsp;&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#cc00cc&quot;&gt;AdjointBiSpinor&lt;/font&gt;&lt;/b&gt;" style="rounded=1;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#b3b3b3;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="730" y="360" width="120" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-60" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-61" target="KG5lhhUBjoQ79gfvQvIC-63" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-61" value="V(γ&lt;sub&gt;2&lt;/sub&gt;, p&lt;span style=&quot;font-size: 10px;&quot;&gt;&lt;sub&gt;1&lt;/sub&gt;&lt;/span&gt;)&lt;br&gt;Result Particle:&lt;br&gt;q&#39;: &lt;b&gt;&lt;font color=&quot;#0000ff&quot;&gt;OutgoingFermion&lt;/font&gt;&lt;/b&gt;" style="ellipse;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#9AC7BF;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="960" y="480" width="140" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-62" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-63" target="KG5lhhUBjoQ79gfvQvIC-65" edge="1">
+          <mxGeometry relative="1" as="geometry">
+            <Array as="points">
+              <mxPoint x="1030" y="340" />
+              <mxPoint x="900" y="340" />
+            </Array>
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-63" value="&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#336600&quot;&gt;vertex&lt;/font&gt;&lt;/b&gt;&amp;nbsp;-&amp;gt;&lt;br&gt;LorentzVector * &lt;br&gt;DiracMatrix * &lt;br&gt;BiSpinor &lt;br&gt;=&amp;nbsp;&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#cc00cc&quot;&gt;BiSpinor&lt;/font&gt;&lt;/b&gt;" style="rounded=1;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#b3b3b3;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="970" y="360" width="120" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-64" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-65" target="KG5lhhUBjoQ79gfvQvIC-66" edge="1">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-65" value="S2(q, q&#39;)&lt;br&gt;q == -q&#39;&lt;br&gt;&lt;b&gt;&lt;font color=&quot;#0000ff&quot;&gt;No Result Particle&lt;/font&gt;&lt;/b&gt;" style="ellipse;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#9AC7BF;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="840" y="240" width="120" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-72" style="edgeStyle=orthogonalEdgeStyle;rounded=0;orthogonalLoop=1;jettySize=auto;html=1;strokeWidth=2;" parent="1" source="KG5lhhUBjoQ79gfvQvIC-66" target="KG5lhhUBjoQ79gfvQvIC-70" edge="1">
+          <mxGeometry relative="1" as="geometry">
+            <Array as="points">
+              <mxPoint x="670" y="170" />
+            </Array>
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-66" value="&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font style=&quot;border-color: var(--border-color);&quot; color=&quot;#336600&quot;&gt;inner_edge&lt;/font&gt;&lt;/b&gt;&amp;nbsp;-&amp;gt;&lt;br&gt;AdjointBiSpinor * propagator(q) * BiSpinor =&amp;nbsp;&lt;b style=&quot;border-color: var(--border-color);&quot;&gt;&lt;font color=&quot;#cc00cc&quot;&gt;ComplexF64&lt;/font&gt;&lt;/b&gt;" style="rounded=1;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#f5f5f5;gradientColor=#b3b3b3;strokeColor=#666666;" parent="1" vertex="1">
+          <mxGeometry x="830" y="140" width="140" height="60" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-70" value="&lt;font style=&quot;font-size: 32px;&quot;&gt;Σ&lt;/font&gt;" style="ellipse;whiteSpace=wrap;html=1;strokeWidth=2;fillColor=#ffcd28;strokeColor=#d79b00;gradientColor=#ffa500;" parent="1" vertex="1">
+          <mxGeometry x="610" y="20" width="120" height="80" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-73" value="&lt;b&gt;&lt;font style=&quot;font-size: 22px;&quot;&gt;ValueAcc&lt;/font&gt;&lt;/b&gt;" style="text;html=1;strokeColor=none;fillColor=none;align=center;verticalAlign=middle;whiteSpace=wrap;rounded=0;" parent="1" vertex="1">
+          <mxGeometry x="10" y="155" width="120" height="30" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-74" value="&lt;b&gt;&lt;font style=&quot;font-size: 22px;&quot;&gt;ValueAcc&lt;/font&gt;&lt;/b&gt;" style="text;html=1;strokeColor=none;fillColor=none;align=center;verticalAlign=middle;whiteSpace=wrap;rounded=0;" parent="1" vertex="1">
+          <mxGeometry x="10" y="385" width="120" height="30" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-75" value="&lt;b&gt;&lt;font style=&quot;font-size: 22px;&quot;&gt;ValueAcc&lt;/font&gt;&lt;/b&gt;" style="text;html=1;strokeColor=none;fillColor=none;align=center;verticalAlign=middle;whiteSpace=wrap;rounded=0;" parent="1" vertex="1">
+          <mxGeometry x="10" y="615" width="120" height="30" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-76" value="&lt;b&gt;&lt;font style=&quot;font-size: 22px;&quot;&gt;Particle Propagation&lt;/font&gt;&lt;/b&gt;" style="text;html=1;strokeColor=none;fillColor=none;align=center;verticalAlign=middle;whiteSpace=wrap;rounded=0;" parent="1" vertex="1">
+          <mxGeometry x="10" y="725" width="120" height="30" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-77" value="&lt;b&gt;&lt;font style=&quot;font-size: 22px;&quot;&gt;Particle Propagation&lt;/font&gt;&lt;/b&gt;" style="text;html=1;strokeColor=none;fillColor=none;align=center;verticalAlign=middle;whiteSpace=wrap;rounded=0;" parent="1" vertex="1">
+          <mxGeometry x="10" y="505" width="120" height="30" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-78" value="&lt;b&gt;&lt;font style=&quot;font-size: 22px;&quot;&gt;Particle Propagation&lt;/font&gt;&lt;/b&gt;" style="text;html=1;strokeColor=none;fillColor=none;align=center;verticalAlign=middle;whiteSpace=wrap;rounded=0;" parent="1" vertex="1">
+          <mxGeometry x="10" y="265" width="120" height="30" as="geometry" />
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-79" value="" style="endArrow=none;dashed=1;html=1;dashPattern=1 3;strokeWidth=2;rounded=0;" parent="1" edge="1">
+          <mxGeometry width="50" height="50" relative="1" as="geometry">
+            <mxPoint y="220" as="sourcePoint" />
+            <mxPoint x="1160" y="220" as="targetPoint" />
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-80" value="" style="endArrow=none;dashed=1;html=1;dashPattern=1 3;strokeWidth=2;rounded=0;" parent="1" edge="1">
+          <mxGeometry width="50" height="50" relative="1" as="geometry">
+            <mxPoint y="340" as="sourcePoint" />
+            <mxPoint x="1160" y="340" as="targetPoint" />
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-81" value="" style="endArrow=none;dashed=1;html=1;dashPattern=1 3;strokeWidth=2;rounded=0;" parent="1" edge="1">
+          <mxGeometry width="50" height="50" relative="1" as="geometry">
+            <mxPoint y="460" as="sourcePoint" />
+            <mxPoint x="1160" y="460" as="targetPoint" />
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-82" value="" style="endArrow=none;dashed=1;html=1;dashPattern=1 3;strokeWidth=2;rounded=0;" parent="1" edge="1">
+          <mxGeometry width="50" height="50" relative="1" as="geometry">
+            <mxPoint y="120" as="sourcePoint" />
+            <mxPoint x="1160" y="120" as="targetPoint" />
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-83" value="" style="endArrow=none;dashed=1;html=1;dashPattern=1 3;strokeWidth=2;rounded=0;" parent="1" edge="1">
+          <mxGeometry width="50" height="50" relative="1" as="geometry">
+            <mxPoint y="580" as="sourcePoint" />
+            <mxPoint x="1160" y="580" as="targetPoint" />
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-84" value="" style="endArrow=none;dashed=1;html=1;dashPattern=1 3;strokeWidth=2;rounded=0;" parent="1" edge="1">
+          <mxGeometry width="50" height="50" relative="1" as="geometry">
+            <mxPoint y="680" as="sourcePoint" />
+            <mxPoint x="1160" y="680" as="targetPoint" />
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="KG5lhhUBjoQ79gfvQvIC-85" value="" style="endArrow=none;dashed=1;html=1;dashPattern=1 3;strokeWidth=2;rounded=0;" parent="1" edge="1">
+          <mxGeometry width="50" height="50" relative="1" as="geometry">
+            <mxPoint y="800" as="sourcePoint" />
+            <mxPoint x="1160" y="800" as="targetPoint" />
+          </mxGeometry>
+        </mxCell>
+      </root>
+    </mxGraphModel>
+  </diagram>
+</mxfile>
--- a/docs/src/lib/internals/code_gen.md
+++ b/docs/src/lib/internals/code_gen.md
@ -1,8 +1,23 @@
 # Code Generation

-## Main
+## Types
 ```@autodocs
 Modules = [MetagraphOptimization]
-Pages = ["code_gen/main.jl"]
+Pages = ["code_gen/type.jl"]
+Order = [:type, :constant, :function]
+```
+
+## Function Generation
+Implementations for generation of a callable function. A function generated this way cannot immediately be called. One Julia World Age has to pass before this is possible, which happens when the global Julia scope advances. If the DAG and therefore the generated function becomes too large, use the tape machine instead, since compiling large functions becomes infeasible.
+```@autodocs
+Modules = [MetagraphOptimization]
+Pages = ["code_gen/function.jl"]
+Order = [:function]
+```
+
+## Tape Machine
+```@autodocs
+Modules = [MetagraphOptimization]
+Pages = ["code_gen/tabe_machine.jl"]
 Order = [:function]
 ```
--- a/docs/src/lib/internals/models.md
+++ b/docs/src/lib/internals/models.md
@ -69,4 +69,58 @@ Order = [:function]

 ## QED-Model

-*To be added*
+### Feynman Diagrams
+```@autodocs
+Modules = [MetagraphOptimization]
+Pages = ["models/qed/diagrams.jl"]
+Order   = [:type, :function, :constant]
+```
+
+### Types
+```@autodocs
+Modules = [MetagraphOptimization]
+Pages = ["models/qed/types.jl"]
+Order   = [:type, :constant]
+```
+
+### Particle
+```@autodocs
+Modules = [MetagraphOptimization]
+Pages = ["models/qed/particle.jl"]
+Order   = [:type, :constant, :function]
+```
+
+### Parse
+```@autodocs
+Modules = [MetagraphOptimization]
+Pages = ["models/qed/parse.jl"]
+Order   = [:function]
+```
+
+### Properties
+```@autodocs
+Modules = [MetagraphOptimization]
+Pages = ["models/qed/properties.jl"]
+Order   = [:function]
+```
+
+### Create
+```@autodocs
+Modules = [MetagraphOptimization]
+Pages = ["models/qed/create.jl"]
+Order = [:function]
+```
+
+### Compute
+```@autodocs
+Modules = [MetagraphOptimization]
+Pages = ["models/qed/compute.jl"]
+Order = [:function]
+```
+
+### Print
+```@autodocs
+Modules = [MetagraphOptimization]
+Pages = ["models/qed/print.jl"]
+Order = [:function]
+```
--- a/docs/src/lib/internals/scheduler.md
+++ b/docs/src/lib/internals/scheduler.md
@ -7,6 +7,13 @@ Pages = ["scheduler/interface.jl"]
 Order   = [:type, :function]
 ```

+## Types
+```@autodocs
+Modules = [MetagraphOptimization]
+Pages = ["scheduler/type.jl"]
+Order   = [:type, :function]
+```
+
 ## Greedy
 ```@autodocs
 Modules = [MetagraphOptimization]
--- a/docs/src/lib/internals/task.md
+++ b/docs/src/lib/internals/task.md
@ -34,10 +34,3 @@ Modules = [MetagraphOptimization]
 Pages = ["task/properties.jl"]
 Order   = [:function]
 ```
-
-## Print
-```@autodocs
-Modules = [MetagraphOptimization]
-Pages = ["task/print.jl"]
-Order   = [:function]
-```
--- a/docs/src/structure_qed.drawio
+++ b/docs/src/structure_qed.drawio
--- a/examples/Project.toml
+++ b/examples/Project.toml
@ -1,3 +1,10 @@
 [deps]
 BenchmarkTools = "6e4b80f9-dd63-53aa-95a3-0cdb28fa8baf"
+CSV = "336ed68f-0bac-5ca0-87d4-7b16caf5d00b"
+CUDA = "052768ef-5323-5732-b1bb-66c8b64840ba"
+DataFrames = "a93c6f00-e57d-5684-b7b6-d8193f3e46c0"
 MetagraphOptimization = "3e869610-d48d-4942-ba70-c1b702a33ca4"
+Plots = "91a5bcdd-55d7-5caf-9e0b-520d859cae80"
+QEDbase = "10e22c08-3ccb-4172-bfcf-7d7aa3d04d93"
+QEDprocesses = "46de9c38-1bb3-4547-a1ec-da24d767fdad"
+StatsPlots = "f3b207a7-027a-5e70-b257-86293d7955fd"
--- a/examples/full_node_bench.jl
+++ b/examples/full_node_bench.jl
@ -0,0 +1,249 @@
+using MetagraphOptimization
+using CUDA
+using UUIDs
+using DataFrames
+using CSV
+using Random
+using BenchmarkTools
+using Dates
+
+using Base.Threads
+
+
+function log(x...)
+    println(now(), " ", join(x, " ")...)
+    flush(stdout)
+    return nothing
+end
+
+results_filename = "full_node_bench.csv"
+
+df = DataFrame(
+    process_name = String[],
+    cpu_threads = Int[],
+    gpu_devices = Int[],
+    n_inputs = Int[],
+    chunk_size = Int[],
+    time = Float64[],
+    std = Float64[],
+    rate = Float64[],
+    cpu_chunks = Float64[],
+    gpu_chunks = Float64[],
+    memory_est = Float64[],
+)
+
+# if they exist, read existing results and append new ones
+if isfile(results_filename)
+    df = CSV.read(results_filename, DataFrame)
+end
+
+nInputs = 2^26
+
+lck = ReentrantLock()
+
+progress = 1
+cpu_chunks = 0
+gpu_chunks = 0
+
+chunkSizes = [1024, 4096, 16384, 65536, 262144, 1048576]    # 2^10 to 2^20
+
+function cpu_worker(compute_func, inputs, chunk_size)
+    global progress
+    global cpu_chunks
+    global lck
+    quit = false
+    work_start = 0
+    work_end = 0
+    while true
+        lock(lck) do
+            if progress >= nInputs
+                quit = true
+            else
+                work_start = progress
+                progress = progress + chunk_size
+                work_end = min(progress - 1, nInputs)
+                cpu_chunks = cpu_chunks + 1
+                #log("CPU Worker $(Threads.threadid()) computing $(cpu_chunks)th cpu chunk ($work_start, $work_end)")
+            end
+        end
+        if quit
+            break
+        end
+
+        for i in work_start:work_end
+            compute_func(inputs[i])
+        end
+    end
+
+    #log("CPU Worker on $(Threads.threadid()) finished!")
+
+    return nothing
+end
+
+# called with a specific device selected
+function gpu_worker(kernel!, inputs, chunk_size)
+    global progress
+    global gpu_chunks
+    global lck
+    cuOutputs = CuVector{ComplexF64}()
+    resize!(cuOutputs, chunk_size)
+
+    quit = false
+    work_start = 0
+    work_end = 0
+    while true
+        lock(lck) do
+            if progress >= nInputs
+                quit = true
+            else
+                work_start = progress
+                progress = progress + chunk_size
+                work_end = min(progress - 1, nInputs)
+                gpu_chunks = gpu_chunks + 1
+                #log("GPU Worker $(CUDA.device()) computing $(gpu_chunks)th gpu chunk ($work_start, $work_end)")
+            end
+        end
+        if quit
+            break
+        end
+
+        cuInputs = CuVector(inputs[work_start:work_end])
+        ts = 32
+        bs = Int(chunk_size / 32)
+        @cuda threads = ts blocks = bs always_inline = true kernel!(cuInputs, cuOutputs, chunk_size)
+        CUDA.device_synchronize()
+    end
+
+    #log("GPU Worker on Device $(CUDA.device()) finished!")
+
+    return nothing
+end
+
+cpu_gpu_ratio = Vector{Tuple{Int, Int}}()
+
+function full_compute(compute_func, kernel!, inputs, chunk_size)
+    global progress
+    progress = 1
+    global cpu_chunks
+    cpu_chunks = 0
+    global gpu_chunks
+    gpu_chunks = 0
+
+    tasks = Vector()
+
+    for dev in CUDA.devices()
+        t = Threads.@spawn device!(dev) do
+            gpu_worker(kernel!, inputs, chunk_size)
+            return nothing
+        end
+        push!(tasks, t)
+    end
+
+    for i in 1:(Threads.nthreads() - length(CUDA.devices()))
+        t = Threads.@spawn cpu_worker(compute_func, inputs, chunk_size)
+        push!(tasks, t)
+    end
+
+    for t in tasks
+        wait(t)
+    end
+
+    push!(cpu_gpu_ratio, (cpu_chunks, gpu_chunks))
+    return nothing
+end
+
+function bench(compute_function, kernel!, inputs, chunk_size)
+    global cpu_gpu_ratio
+    empty!(cpu_gpu_ratio)
+
+    bench = @benchmark begin
+        full_compute($compute_function, $kernel!, $inputs, $chunk_size)
+    end gcsample = true seconds = 60
+
+    time = median(bench.times) / 1e9
+    s = std(bench.times) / 1e9
+    rate = length(inputs) / time
+
+    med_cpu_chunks = median(getindex.(cpu_gpu_ratio, 1))
+    med_gpu_chunks = median(getindex.(cpu_gpu_ratio, 2))
+    mem_estimate = bench.memory
+
+    log("CPU/GPU ratios: $(cpu_gpu_ratio)")
+
+    return (time, rate, s, med_cpu_chunks, med_gpu_chunks, mem_estimate)
+end
+
+function full_node_bench(process::MetagraphOptimization.AbstractProcessDescription, func, kernel!, chunk_size, inputs)
+    process_name = string(process)
+    log("\n--- Benchmarking $(process_name) on $(nInputs) with chunk size $(chunk_size) ---")
+
+    log("Available Cuda Devices:")
+    display.(CUDA.devices())
+
+    log("Benchmarking full node...")
+    (time, rate, s, med_cpu_chunks, med_gpu_chunks, mem_estimate) = bench(func, kernel!, inputs, chunk_size)
+    log(
+        "Benchmarking complete with median time $(time), $(med_cpu_chunks) cpu chunks, and $(med_gpu_chunks) gpu chunks.",
+    )
+
+    push!(
+        df,
+        Dict(
+            :process_name => process_name,
+            :cpu_threads => Threads.nthreads() - length(CUDA.devices()),
+            :gpu_devices => length(CUDA.devices()),
+            :n_inputs => nInputs,
+            :chunk_size => chunk_size,
+            :time => time,
+            :std => s,
+            :rate => rate,
+            :cpu_chunks => med_cpu_chunks,
+            :gpu_chunks => med_gpu_chunks,
+            :memory_est => mem_estimate,
+        ),
+    )
+
+    return nothing
+end
+
+# use "mock" machine that only uses cpu for compilation
+machine = Machine(
+    [
+        MetagraphOptimization.NumaNode(
+            0,
+            1,
+            MetagraphOptimization.default_strategy(MetagraphOptimization.NumaNode),
+            -1.0,
+            UUIDs.uuid1(),
+        ),
+    ],
+    [-1.0;;],
+)
+
+optimizer = ReductionOptimizer()
+processes = ["ke->ke", "ke->kke", "ke->kkke", "ke->kkkke", "ke->kkkkke"]
+
+for proc in processes
+    process = parse_process(proc, QEDModel())
+    graph = gen_graph(process)
+    optimize_to_fixpoint!(optimizer, graph)
+    compute_func = get_compute_function(graph, process, machine)
+    kernel! = get_cuda_kernel(graph, process, machine)
+
+    log("Generating $nInputs inputs with $(Threads.nthreads()) threads...")
+    inputs = Vector{typeof(gen_process_input(process))}()
+    resize!(inputs, nInputs)
+    procs = Vector{typeof(process)}()
+    for i in 1:Threads.nthreads()
+        push!(procs, copy(process))
+    end
+
+    @inbounds Threads.@threads for i in eachindex(inputs)
+        inputs[i] = gen_process_input(procs[Threads.nthreads()])
+    end
+
+    for chunk_size in chunkSizes
+        full_node_bench(process, compute_func, kernel!, chunk_size, inputs)
+        CSV.write(results_filename, df)
+    end
+end;
--- a/examples/import_bench.jl
+++ b/examples/import_bench.jl
@ -34,9 +34,10 @@ function import_bench()
    bench_txt("AB->ABBB.txt")
    bench_txt("AB->ABBBBB.txt")
    bench_txt("AB->ABBBBBBB.txt")
-    #bench_txt("AB->ABBBBBBBBB.txt")
+    bench_txt("AB->ABBBBBBBBB.txt")
    bench_txt("ABAB->ABAB.txt")
-    return bench_txt("ABAB->ABC.txt")
+    bench_txt("ABAB->ABC.txt")
+    return nothing
 end

 import_bench()
--- a/examples/qed_bench.jl
+++ b/examples/qed_bench.jl
@ -0,0 +1,272 @@
+using MetagraphOptimization
+using LIKWID
+using CUDA
+using UUIDs
+using DataFrames
+using CSV
+using Random
+using BenchmarkTools
+using Dates
+
+DISABLE_GPU = false
+
+function log(x...)
+    println(now(), " ", join(x, " ")...)
+    return flush(stdout)
+end
+
+results_filename = "bench_results_$(Threads.nthreads()).csv"
+
+df = DataFrame(
+    process_name = String[],
+    graph_gen_time = Float64[],
+    optimization_time = Float64[],
+    function_generation_time = Float64[],
+    graph_nodes = Int[],
+    graph_edges = Int[],
+    graph_mem = Float64[],
+    cpu_threads = Int[],
+    n_inputs = Int[],
+    nflops_likwid = Int[],
+    cpu_time = Float64[],
+    cpu_std = Float64[],
+    cpu_rate = Float64[],
+    cpu_gflops = Float64[],
+    gpu_name = String[],
+    gpu_time = Float64[],
+    gpu_std = Float64[],
+    gpu_rate = Float64[],
+    gpu_gflops = Float64[],
+)
+
+# if they exist, read existing results and append new ones
+if isfile(results_filename)
+    df = CSV.read(results_filename, DataFrame)
+end
+
+nInputs = 2^20
+
+function cpu_bench(compute_function, inputs)
+    bench = @benchmark begin
+        @inbounds Threads.@threads for i in eachindex($inputs)
+            @invokelatest $compute_function($inputs[i])
+        end
+    end gcsample = true samples = 20 evals = 1
+
+    time = median(bench.times) / 1e9
+    s = std(bench.times) / 1e9
+    rate = length(inputs) / time
+
+    return (time, rate, s)
+end
+
+function gpu_bench(kernel!, inputs)
+    n = length(inputs)
+    outputs = CuVector{ComplexF64}()
+    resize!(outputs, n)
+    ts = 32
+    bs = Int(n / ts)
+    bench = @benchmark begin
+        @cuda threads = $ts blocks = $bs always_inline = true $kernel!($inputs, $outputs, $n)
+        CUDA.device_synchronize()
+    end gcsample = true samples = 20 evals = 1
+
+    time = median(bench.times) / 1e9
+    s = std(bench.times) / 1e9
+    rate = length(inputs) / time
+
+    return (time, rate, s)
+end
+
+function bench_process(
+    process::MetagraphOptimization.AbstractProcessDescription,
+    process_name::String,
+    graph::DAG,
+    func,
+    kernel!,
+    gen_time::Float64,
+    opt_time::Float64,
+    func_time::Float64;
+    use_likwid = false,
+    use_gpu = true,
+)
+    log("\n--- Benchmarking $(process_name) ---")
+    if DISABLE_GPU
+        use_gpu = false
+    end
+
+    graph_props = GraphProperties(graph)
+    NFLOPs = graph_props.computeEffort
+    nflops_likwid = 0
+    if use_likwid
+        input = gen_process_input(process)
+        func(input) # compile first
+
+        # get rid of annoying output to console
+        oldstd = stdout
+        redirect_stdout(devnull)
+        _, events = @perfmon "FLOPS_DP" func(input)
+        redirect_stdout(oldstd) # recover original stdout
+
+        NFLOPs = first(events["FLOPS_DP"])["RETIRED_SSE_AVX_FLOPS_ALL"]
+        nflops_likwid = NFLOPs
+    end
+
+    log("Generating $nInputs inputs with $(Threads.nthreads()) threads...")
+
+    inputs = Vector{typeof(gen_process_input(process))}()
+    resize!(inputs, nInputs)
+    processes = Vector{typeof(process)}()
+    for i in 1:Threads.nthreads()
+        push!(processes, copy(process))
+    end
+
+    @inbounds Threads.@threads for i in eachindex(inputs)
+        inputs[i] = gen_process_input(processes[Threads.nthreads()])
+    end
+
+    log("Benchmarking CPU with $(Threads.nthreads()) threads...")
+    (time_cpu, rate_cpu, std_cpu) = cpu_bench(func, inputs)
+    flops_cpu = (rate_cpu * NFLOPs) / 10^9
+
+    time_gpu = 0.0
+    std_gpu = 0.0
+    rate_gpu = 0.0
+    flops_gpu = 0.0
+    gpu_name = "none"
+    if use_gpu
+        log("Benchmarking GPU...")
+        gpu_name = "$(name(first(CUDA.devices())))"
+        cuInputs = CuArray(inputs)
+        (time_gpu, rate_gpu, std_gpu) = gpu_bench(kernel!, cuInputs)
+        flops_gpu = (rate_gpu * NFLOPs) / 10^9
+    else
+        log("Skipping GPU...")
+    end
+
+    log("\nBenchmark Summary for $(process):")
+
+    if use_likwid
+        log("Measured FLOPS by LIKWID: $NFLOPs")
+    else
+        log("Total graph compute effort: $NFLOPs")
+    end
+    log("Total input size: $(bytes_to_human_readable(Base.summarysize(inputs)))")
+    log("CPU, $(Threads.nthreads()) threads")
+    log("  Time:  $time_cpu")
+    log("  Rate:  $rate_cpu")
+    log("  GFLOPS: $flops_cpu")
+    if use_gpu
+        log("GPU, $gpu_name")
+        log("  Time:  $time_gpu")
+        log("  Rate:  $rate_gpu")
+        log("  GFLOPS: $flops_gpu")
+    end
+
+    if (process_name != "warmup")
+        push!(
+            df,
+            Dict(
+                :process_name => process_name,
+                :graph_gen_time => gen_time,
+                :optimization_time => opt_time,
+                :function_generation_time => func_time,
+                :graph_nodes => graph_props.noNodes,
+                :graph_edges => graph_props.noEdges,
+                :graph_mem => MetagraphOptimization.mem(graph),
+                :cpu_threads => Threads.nthreads(),
+                :n_inputs => nInputs,
+                :nflops_likwid => nflops_likwid,
+                :cpu_time => time_cpu,
+                :cpu_std => std_cpu,
+                :cpu_rate => rate_cpu,
+                :cpu_gflops => flops_cpu,
+                :gpu_name => gpu_name,
+                :gpu_time => time_gpu,
+                :gpu_std => std_gpu,
+                :gpu_rate => rate_gpu,
+                :gpu_gflops => flops_gpu,
+            ),
+        )
+    end
+
+    return nothing
+end
+
+# use "mock" machine that only uses cpu
+machine = Machine(
+    [
+        MetagraphOptimization.NumaNode(
+            0,
+            1,
+            MetagraphOptimization.default_strategy(MetagraphOptimization.NumaNode),
+            -1.0,
+            UUIDs.uuid1(),
+        ),
+    ],
+    [-1.0;;],
+)
+
+# sadly cannot put these in functions because the world age must increase after the function is created which happens only in the global scope
+
+## -- WARMUP TO COMPILE FUNCTIONS first
+#=
+optimizer = RandomWalkOptimizer(MersenneTwister(0))
+
+# 2-photon compton
+process = parse_process("ke->kke", QEDModel())
+gen_time = @elapsed graph = gen_graph(process)
+opt_time = @elapsed optimize!(optimizer, graph, 200)
+func_gen_time = @elapsed compute_func = get_compute_function(graph, process, machine)
+kernel! = get_cuda_kernel(graph, process, machine)
+bench_process(process, "warmup", graph, compute_func, kernel!, gen_time, opt_time, func_gen_time)
+
+optimizer = ReductionOptimizer()
+
+# AB->AB^3
+process = parse_process("AB->ABBB", ABCModel())
+gen_time = @elapsed graph = parse_dag("input/AB->ABBB.txt", ABCModel())
+opt_time = @elapsed optimize_to_fixpoint!(optimizer, graph)
+func_gen_time = @elapsed compute_func = get_compute_function(graph, process, machine)
+kernel! = get_cuda_kernel(graph, process, machine)
+bench_process(process, "warmup", graph, compute_func, kernel!, gen_time, opt_time, func_gen_time)
+=#
+## -- WARMUP END
+
+optimizer = ReductionOptimizer()
+
+processes = ["ke->ke", "ke->kke", "ke->kkke", "ke->kkkke", "ke->kkkkke"]
+
+for process_str in processes
+    # compton
+    process = parse_process(process_str, QEDModel())
+    gen_time = @elapsed graph = gen_graph(process)
+    func_gen_time = @elapsed compute_func = get_compute_function(graph, process, machine)
+    kernel! = get_cuda_kernel(graph, process, machine)
+    bench_process(process, "$process not optimized", graph, compute_func, kernel!, gen_time, 0.0, func_gen_time)
+
+    opt_time = @elapsed optimize_to_fixpoint!(optimizer, graph)
+    func_gen_time = @elapsed compute_func = get_compute_function(graph, process, machine)
+    kernel! = get_cuda_kernel(graph, process, machine)
+    bench_process(process, "$process reduced", graph, compute_func, kernel!, gen_time, opt_time, func_gen_time)
+
+    CSV.write(results_filename, df)
+end
+
+processes = ["AB->AB", "AB->ABBB", "AB->ABBBBB", "AB->ABBBBBBB"]
+
+for process_str in processes
+    # AB->AB
+    process = parse_process(process_str, ABCModel())
+    gen_time = @elapsed graph = parse_dag("input/$(process_str).txt", ABCModel())
+    func_gen_time = @elapsed compute_func = get_compute_function(graph, process, machine)
+    kernel! = get_cuda_kernel(graph, process, machine)
+    bench_process(process, "$process not optimized", graph, compute_func, kernel!, gen_time, 0.0, func_gen_time)
+
+    opt_time = @elapsed optimize_to_fixpoint!(optimizer, graph)
+    func_gen_time = @elapsed compute_func = get_compute_function(graph, process, machine)
+    kernel! = get_cuda_kernel(graph, process, machine)
+    bench_process(process, "$process reduced", graph, compute_func, kernel!, gen_time, opt_time, func_gen_time)
+
+    CSV.write(results_filename, df)
+end
--- a/examples/qed_bench_reduction_steps.jl
+++ b/examples/qed_bench_reduction_steps.jl
@ -0,0 +1,163 @@
+using MetagraphOptimization
+using CUDA
+using UUIDs
+using BenchmarkTools
+using DataFrames
+using CSV
+
+results_filename = "bench_results_reduction_steps.csv"
+
+df = DataFrame(
+    threads = Int[],
+    process = String[],
+    operations = Int[],
+    cumulative_optimization_time = Float64[],
+    graph_nodes = Int[],
+    graph_edges = Int[],
+    graph_ce = Float64[],
+    graph_dt = Float64[],
+    graph_ci = Float64[],
+    gen_func_t = Float64[],
+    cpu_compile_t = Float64[],
+    cpu_st_t = Float64[],
+    cpu_mt_t = Float64[],
+    gpu_compile_t = Float64[],
+    gpu_t = Float64[],
+)
+
+# if they exist, read existing results and append new ones
+if isfile(results_filename)
+    df = CSV.read(results_filename, DataFrame)
+end
+
+function bench(func, inputs)
+    compile_time = @elapsed func(inputs[1])
+
+    single_thread = @benchmark $func.($inputs)
+    multi_threaded = @benchmark Threads.@threads for i in eachindex($inputs)
+        $func($inputs[i])
+    end
+
+    return (
+        cpu_compile_time = compile_time,
+        gpu_compile_time = 0.0,
+        cpu_single_thread_time = mean(single_thread.times) / 1e9,
+        cpu_multi_thread_time = mean(multi_threaded.times) / 1e9,
+        gpu_time = 0.0,
+    )
+end
+
+# preparation of machine
+machine = Machine(
+    [
+        MetagraphOptimization.NumaNode(
+            0,
+            1,
+            MetagraphOptimization.default_strategy(MetagraphOptimization.NumaNode),
+            -1.0,
+            UUIDs.uuid1(),
+        ),
+    ],
+    [-1.0;;],
+)
+
+# bench and produce data
+n_inputs = 50_000
+optimizer = ReductionOptimizer()
+processes = [("ke->kke", 5), ("ke->ke", 1), ("ke->kke", 1), ("ke->kkke", 1), ("ke->kkkke", 1), ("ke->kkkkke", 1)]
+
+for (process_str, STEPSIZE) in processes
+    n = 0
+    opt_time_cum = 0
+
+    process = parse_process(process_str, QEDModel())
+    graph = gen_graph(process)
+    inputs = [gen_process_input(process) for _ in 1:n_inputs]
+
+    get_compute_function(graph, process, machine)
+
+    while true
+        func_gen_time = @elapsed func = get_compute_function(graph, process, machine)
+        res = bench(func, inputs)
+
+        graph_properties = get_properties(graph)
+        push!(
+            df,
+            (
+                Threads.nthreads(),
+                process_str,
+                n,
+                opt_time_cum,
+                graph_properties.noNodes,
+                graph_properties.noEdges,
+                graph_properties.computeEffort,
+                graph_properties.data,
+                graph_properties.computeIntensity,
+                func_gen_time,
+                res.cpu_compile_time,
+                res.cpu_single_thread_time,
+                res.cpu_multi_thread_time,
+                res.gpu_compile_time,
+                res.gpu_time,
+            ),
+        )
+        CSV.write(results_filename, df)
+
+        if fixpoint_reached(optimizer, graph)
+            break
+        end
+
+        opt_time_cum += @elapsed optimize!(optimizer, graph, STEPSIZE)
+        n += STEPSIZE
+    end
+end
+
+CSV.write(results_filename, df)
+
+for (process_str, STEPSIZE) in [("AB->AB", 1), ("AB->ABBB", 1), ("AB->ABBBBB", 1)]
+    n = 0
+    opt_time_cum = 0
+
+    process = parse_process(process_str, ABCModel())
+    graph = parse_dag("input/$process_str.txt", ABCModel())
+    inputs = [gen_process_input(process) for _ in 1:n_inputs]
+
+    get_compute_function(graph, process, machine)
+
+    while true
+        func_gen_time = @elapsed func = get_compute_function(graph, process, machine)
+        res = bench(func, inputs)
+
+        graph_properties = get_properties(graph)
+        push!(
+            df,
+            (
+                Threads.nthreads(),
+                process_str,
+                n,
+                opt_time_cum,
+                graph_properties.noNodes,
+                graph_properties.noEdges,
+                graph_properties.computeEffort,
+                graph_properties.data,
+                graph_properties.computeIntensity,
+                func_gen_time,
+                res.cpu_compile_time,
+                res.cpu_single_thread_time,
+                res.cpu_multi_thread_time,
+                res.gpu_compile_time,
+                res.gpu_time,
+            ),
+        )
+        CSV.write(results_filename, df)
+
+        if fixpoint_reached(optimizer, graph)
+            break
+        end
+
+        opt_time_cum += @elapsed optimize!(optimizer, graph, STEPSIZE)
+        n += STEPSIZE
+    end
+end
+
+CSV.write(results_filename, df)
--- a/examples/qed_bench_reduction_steps_gpu.jl
+++ b/examples/qed_bench_reduction_steps_gpu.jl
@ -0,0 +1,208 @@
+using MetagraphOptimization
+using CUDA
+using UUIDs
+using BenchmarkTools
+using DataFrames
+using CSV
+using Dates
+
+results_filename = "bench_results_reduction_steps_gpu.csv"
+
+df = DataFrame(
+    threads = Int[],
+    process = String[],
+    operations = Int[],
+    cumulative_optimization_time = Float64[],
+    graph_nodes = Int[],
+    graph_edges = Int[],
+    graph_ce = Float64[],
+    graph_dt = Float64[],
+    graph_ci = Float64[],
+    cpu_st_t = Float64[],
+    cpu_st_s = Float64[],
+    cpu_mt_t = Float64[],
+    cpu_mt_s = Float64[],
+    cpu_mem = Float64[],
+    gpu_t = Float64[],
+    gpu_s = Float64[],
+    gpu_mem = Float64[],
+)
+
+# if they exist, read existing results and append new ones
+if isfile(results_filename)
+    df = CSV.read(results_filename, DataFrame)
+end
+
+function log(x...)
+    println(now(), " ", join(x, " ")...)
+    return flush(stdout)
+end
+
+function bench(func, kernel!, inputs)
+    # gpu part
+    n = length(inputs)
+    cu_inputs = CuVector(inputs)
+    cu_outputs = CuVector{ComplexF64}()
+    resize!(cu_outputs, n)
+    ts = 32
+    bs = Int(n / ts)
+    bench = @benchmark begin
+        @cuda threads = $ts blocks = $bs always_inline = true $kernel!($cu_inputs, $cu_outputs, $n)
+        CUDA.device_synchronize()
+    end gcsample = true samples = 20 evals = 1
+
+    gpu_time = median(bench.times) / 1e9
+    gpu_std = std(bench.times) / 1e9
+    gpu_mem = bench.memory
+
+    # cpu part
+    single_thread = @benchmark $func.($inputs)
+    multi_threaded = @benchmark Threads.@threads for i in eachindex($inputs)
+        $func($inputs[i])
+    end
+
+    cpu_st_time = median(single_thread.times) / 1e9
+    cpu_st_std = std(single_thread.times) / 1e9
+    cpu_mt_time = median(multi_threaded.times) / 1e9
+    cpu_mt_std = std(multi_threaded.times) / 1e9
+    cpu_mem = std(single_thread.times)
+
+
+    return (
+        cpu_single_thread_time = cpu_st_time,
+        cpu_single_thread_std = cpu_st_std,
+        cpu_multi_thread_time = cpu_mt_time,
+        cpu_multi_thread_std = cpu_mt_std,
+        cpu_mem = cpu_mem,
+        gpu_time = gpu_time,
+        gpu_std = gpu_std,
+        gpu_mem = gpu_mem,
+    )
+end
+
+log("Available CUDA devices:")
+for dev in CUDA.devices()
+    display(dev)
+end
+
+# preparation of machine
+machine = Machine(
+    [
+        MetagraphOptimization.NumaNode(
+            0,
+            1,
+            MetagraphOptimization.default_strategy(MetagraphOptimization.NumaNode),
+            -1.0,
+            UUIDs.uuid1(),
+        ),
+    ],
+    [-1.0;;],
+)
+
+
+# bench and produce data
+n_inputs = 2^16
+optimizer = ReductionOptimizer()
+processes = [("ke->ke", 1), ("ke->kke", 1), ("ke->kkke", 1), ("ke->kkkke", 5)]
+
+for (process_str, STEPSIZE) in processes
+    n = 0
+    opt_time_cum = 0
+
+    process = parse_process(process_str, QEDModel())
+    graph = gen_graph(process)
+    inputs = Vector([gen_process_input(process) for _ in 1:n_inputs])
+
+    get_compute_function(graph, process, machine)
+
+    while true
+        func = get_compute_function(graph, process, machine)
+        kernel! = get_cuda_kernel(graph, process, machine)
+        res = bench(func, kernel!, inputs)
+
+        graph_properties = get_properties(graph)
+        push!(
+            df,
+            (
+                Threads.nthreads(),
+                process_str,
+                n,
+                opt_time_cum,
+                graph_properties.noNodes,
+                graph_properties.noEdges,
+                graph_properties.computeEffort,
+                graph_properties.data,
+                graph_properties.computeIntensity,
+                res.cpu_single_thread_time,
+                res.cpu_single_thread_std,
+                res.cpu_multi_thread_time,
+                res.cpu_multi_thread_std,
+                res.cpu_mem,
+                res.gpu_time,
+                res.gpu_std,
+                res.gpu_mem,
+            ),
+        )
+        CSV.write(results_filename, df)
+
+        if fixpoint_reached(optimizer, graph)
+            break
+        end
+
+        opt_time_cum += @elapsed optimize!(optimizer, graph, STEPSIZE)
+        n += STEPSIZE
+    end
+end
+
+CSV.write(results_filename, df)
+
+for (process_str, STEPSIZE) in [("AB->AB", 1), ("AB->ABBB", 1), ("AB->ABBBBB", 1)]
+    n = 0
+    opt_time_cum = 0
+
+    process = parse_process(process_str, ABCModel())
+    graph = parse_dag("input/$process_str.txt", ABCModel())
+    inputs = Vector([gen_process_input(process) for _ in 1:n_inputs])
+
+    get_compute_function(graph, process, machine)
+
+    while true
+        func = get_compute_function(graph, process, machine)
+        kernel! = get_cuda_kernel(graph, process, machine)
+        res = bench(func, kernel!, inputs)
+
+        graph_properties = get_properties(graph)
+        push!(
+            df,
+            (
+                Threads.nthreads(),
+                process_str,
+                n,
+                opt_time_cum,
+                graph_properties.noNodes,
+                graph_properties.noEdges,
+                graph_properties.computeEffort,
+                graph_properties.data,
+                graph_properties.computeIntensity,
+                res.cpu_single_thread_time,
+                res.cpu_single_thread_std,
+                res.cpu_multi_thread_time,
+                res.cpu_multi_thread_std,
+                res.cpu_mem,
+                res.gpu_time,
+                res.gpu_std,
+                res.gpu_mem,
+            ),
+        )
+        CSV.write(results_filename, df)
+
+        if fixpoint_reached(optimizer, graph)
+            break
+        end
+
+        opt_time_cum += @elapsed optimize!(optimizer, graph, STEPSIZE)
+        n += STEPSIZE
+    end
+end
+
+CSV.write(results_filename, df)
--- a/examples/qed_bench_tape.jl
+++ b/examples/qed_bench_tape.jl
@ -0,0 +1,232 @@
+using MetagraphOptimization
+using LIKWID
+using UUIDs
+using DataFrames
+using CSV
+using Random
+using BenchmarkTools
+using Dates
+
+function log(x...)
+    println(now(), " ", join(x, " ")...)
+    return flush(stdout)
+end
+
+results_filename = "bench_results_tape_$(Threads.nthreads()).csv"
+
+df = DataFrame(
+    process_name = String[],
+    graph_gen_time = Float64[],
+    optimization_time = Float64[],
+    function_generation_time = Float64[],
+    graph_nodes = Int[],
+    graph_edges = Int[],
+    graph_mem = Float64[],
+    cpu_threads = Int[],
+    n_inputs = Int[],
+    nflops_likwid = Int[],
+    cpu_time = Float64[],
+    cpu_rate = Float64[],
+    cpu_gflops = Float64[],
+    cpu_std = Float64[],
+    gpu_name = String[],
+    gpu_time = Float64[],
+    gpu_std = Float64[],
+    gpu_rate = Float64[],
+    gpu_gflops = Float64[],
+)
+
+# if they exist, read existing results and append new ones
+if isfile(results_filename)
+    df = CSV.read(results_filename, DataFrame)
+end
+
+nInputs = 1_000_000
+
+# use "mock" machine that only uses cpu
+machine = Machine(
+    [
+        MetagraphOptimization.NumaNode(
+            0,
+            1,
+            MetagraphOptimization.default_strategy(MetagraphOptimization.NumaNode),
+            -1.0,
+            UUIDs.uuid1(),
+        ),
+    ],
+    [-1.0;;],
+)
+
+
+function cpu_bench(tape, inputs)
+    bench = @benchmark begin
+        @inbounds Threads.@threads for i in eachindex($inputs)
+            execute_tape($tape, $inputs[i])
+        end
+    end gcsample = true seconds = 300
+
+    time = mean(bench.times) / 1e9
+    s = std(bench.times) / 1e9
+    rate = length(inputs) / time
+
+    return (time, rate, s)
+end
+
+function bench_process(
+    process::MetagraphOptimization.AbstractProcessDescription,
+    process_name::String,
+    graph::DAG,
+    gen_time::Float64,
+    opt_time::Float64,
+    io::IO = stdout;
+    use_likwid = false,
+)
+    log("\n--- Benchmarking $(process_name) ---")
+
+    func_time = @elapsed tape = gen_tape(graph, process, machine)
+
+    graph_props = GraphProperties(graph)
+    NFLOPs = graph_props.computeEffort
+    nflops_likwid = 0
+    if use_likwid
+        input = gen_process_input(process)
+
+        # get rid of annoying output to console
+        oldstd = stdout
+        redirect_stdout(devnull)
+        _, events = @perfmon "FLOPS_DP" execute_tape(tape, input)
+        redirect_stdout(oldstd) # recover original stdout
+
+        NFLOPs = first(events["FLOPS_DP"])["RETIRED_SSE_AVX_FLOPS_ALL"]
+        nflops_likwid = NFLOPs
+    end
+
+    log("Generating $nInputs inputs with $(Threads.nthreads()) threads...")
+
+    inputs = Vector{typeof(gen_process_input(process))}()
+    resize!(inputs, nInputs)
+    processes = Vector{typeof(process)}()
+    for i in 1:Threads.nthreads()
+        push!(processes, copy(process))
+    end
+
+    @inbounds Threads.@threads for i in eachindex(inputs)
+        inputs[i] = gen_process_input(processes[Threads.nthreads()])
+    end
+
+    log("Benchmarking CPU with $(Threads.nthreads()) threads...")
+    (time_cpu, rate_cpu, std_cpu) = cpu_bench(tape, inputs)
+    flops_cpu = (rate_cpu * NFLOPs) / 10^9
+
+    log("\nBenchmark Summary for $(process):")
+
+    if use_likwid
+        log("Measured FLOPS by LIKWID: $NFLOPs")
+    else
+        log("Total graph compute effort: $NFLOPs")
+    end
+    log("Total input size: $(bytes_to_human_readable(Base.summarysize(inputs)))")
+    log("CPU, $(Threads.nthreads()) threads")
+    log("  Time:  $time_cpu")
+    log("  Rate:  $rate_cpu")
+    log("  GFLOPS: $flops_cpu")
+
+    if (process_name != "warmup")
+        push!(
+            df,
+            Dict(
+                :process_name => process_name,
+                :graph_gen_time => gen_time,
+                :optimization_time => opt_time,
+                :function_generation_time => func_time,
+                :graph_nodes => graph_props.noNodes,
+                :graph_edges => graph_props.noEdges,
+                :graph_mem => MetagraphOptimization.mem(graph),
+                :cpu_threads => Threads.nthreads(),
+                :n_inputs => nInputs,
+                :nflops_likwid => nflops_likwid,
+                :cpu_time => time_cpu,
+                :cpu_std => std_cpu,
+                :cpu_rate => rate_cpu,
+                :cpu_gflops => flops_cpu,
+                :gpu_name => "none",
+                :gpu_time => 0.0,
+                :gpu_std => 0.0,
+                :gpu_rate => 0.0,
+                :gpu_gflops => 0.0,
+            ),
+        )
+    end
+
+    return nothing
+end
+
+function bench_qed(process_string::String, skip_unoptimized = false)
+    optimizer = ReductionOptimizer()
+
+    process = parse_process(process_string, QEDModel())
+    gen_time = @elapsed graph = gen_graph(process)
+    opt_time = 0.0
+    if !skip_unoptimized
+        bench_process(process, "$process not optimized tape", graph, gen_time, opt_time)
+    end
+
+    opt_time = @elapsed optimize_to_fixpoint!(optimizer, graph)
+    bench_process(process, "$process reduced tape", graph, gen_time, opt_time)
+
+    return nothing
+end
+
+function bench_abc(process_string::String)
+    optimizer = ReductionOptimizer()
+
+    process = parse_process(process_string, ABCModel())
+    gen_time = @elapsed graph = parse_dag("input/$process_string.txt", ABCModel())
+    bench_process(process, "$process not optimized tape", graph, gen_time, 0.0)
+
+    opt_time = @elapsed optimize_to_fixpoint!(optimizer, graph)
+    bench_process(process, "$process reduced tape", graph, gen_time, opt_time)
+
+    return nothing
+end
+
+# sadly cannot put these in functions because the world age must increase after the function is created which happens only in the global scope
+
+## -- WARMUP TO COMPILE FUNCTIONS first
+optimizer = ReductionOptimizer()
+
+process = parse_process("ke->kke", QEDModel())
+gen_time = @elapsed graph = gen_graph(process)
+opt_time = @elapsed optimize_to_fixpoint!(optimizer, graph)
+bench_process(process, "warmup", graph, gen_time, opt_time)
+
+# AB->AB^3
+process = parse_process("AB->ABBB", ABCModel())
+gen_time = @elapsed graph = parse_dag("input/AB->ABBB.txt", ABCModel())
+opt_time = @elapsed optimize_to_fixpoint!(optimizer, graph)
+bench_process(process, "warmup", graph, gen_time, opt_time)
+
+## -- WARMUP END
+
+# compton
+bench_qed("ke->ke")
+CSV.write(results_filename, df)
+bench_qed("ke->kke")
+CSV.write(results_filename, df)
+bench_qed("ke->kkke")
+CSV.write(results_filename, df)
+bench_qed("ke->kkkke")
+CSV.write(results_filename, df)
+bench_qed("ke->kkkkke")
+CSV.write(results_filename, df)
+bench_qed("ke->kkkkkke")
+CSV.write(results_filename, df)
+bench_qed("ke->kkkkkkke")
+CSV.write(results_filename, df)
+
+bench_abc("AB->AB")
+CSV.write(results_filename, df)
+bench_abc("AB->ABBB")
+CSV.write(results_filename, df)
+bench_abc("AB->ABBBBB")
+CSV.write(results_filename, df)
--- a/examples/qed_gen_bench.jl
+++ b/examples/qed_gen_bench.jl
@ -0,0 +1,144 @@
+using MetagraphOptimization
+using DataFrames
+using CSV
+using BenchmarkTools
+using StatsBase
+
+results_filename = "qed_gen_results_$(Threads.nthreads()).csv"
+
+df = DataFrame(
+    process_name = String[],
+    cpu_threads = Int[],
+    graph_gen_samples = Int[],
+    graph_gen_mean = Float64[],
+    graph_gen_std = Float64[],
+    graph_gen_median = Float64[],
+    graph_nodes = Int[],
+    graph_data_nodes = Int[],
+    graph_u_nodes = Int[],
+    graph_v_nodes = Int[],
+    graph_s1_nodes = Int[],
+    graph_s2_nodes = Int[],
+    graph_edges = Int[],
+    graph_nodes_reduced = Int[],
+    graph_data_nodes_reduced = Int[],
+    graph_u_nodes_reduced = Int[],
+    graph_v_nodes_reduced = Int[],
+    graph_s1_nodes_reduced = Int[],
+    graph_s2_nodes_reduced = Int[],
+    graph_edges_reduced = Int[],
+    graph_mem = Float64[],
+    graph_mem_reduced = Float64[],
+    graph_elapsed_reduce = Float64[],
+)
+
+function bench_process(process::AbstractString; warmup = false, optimize = true)
+    println("Benchmarking $process...")
+    model = QEDModel()
+
+    proc = parse_process(process, model)
+
+    gen_bench = @benchmark gen_graph($proc) gcsample = true seconds = 5
+
+    graph = gen_graph(proc)
+
+    props = GraphProperties(graph)
+    node_dict = countmap(typeof.(graph.nodes))
+    graph_size = Base.summarysize(graph)
+
+    reduce_elapsed = -1.0
+    node_dict_reduced = Dict()
+    graph_size_reduced = -1.0
+    props_reduced = GraphProperties()
+    if optimize
+        reduce_elapsed = @elapsed optimize_to_fixpoint!(ReductionOptimizer(), graph)
+
+        props_reduced = GraphProperties(graph)
+        node_dict_reduced = countmap(typeof.(graph.nodes))
+        graph_size_reduced = Base.summarysize(graph)
+    end
+
+    if warmup
+        return nothing
+    end
+
+    push!(
+        df,
+        Dict(
+            :process_name => process,
+            :cpu_threads => Threads.nthreads(),
+            :graph_gen_samples => length(gen_bench.times),
+            :graph_gen_mean => mean(gen_bench.times),
+            :graph_gen_std => std(gen_bench.times),
+            :graph_gen_median => median(gen_bench.times),
+            :graph_nodes => props.noNodes,
+            :graph_data_nodes => get(node_dict, DataTaskNode{DataTask}, 0),
+            :graph_u_nodes => get(node_dict, ComputeTaskNode{ComputeTaskQED_U}, 0),
+            :graph_v_nodes => get(node_dict, ComputeTaskNode{ComputeTaskQED_V}, 0),
+            :graph_s1_nodes => get(node_dict, ComputeTaskNode{ComputeTaskQED_S1}, 0),
+            :graph_s2_nodes => get(node_dict, ComputeTaskNode{ComputeTaskQED_S2}, 0),
+            :graph_edges => props.noEdges,
+            :graph_nodes_reduced => props_reduced.noNodes,
+            :graph_data_nodes_reduced => get(node_dict_reduced, DataTaskNode{DataTask}, 0),
+            :graph_u_nodes_reduced => get(node_dict_reduced, ComputeTaskNode{ComputeTaskQED_U}, 0),
+            :graph_v_nodes_reduced => get(node_dict_reduced, ComputeTaskNode{ComputeTaskQED_V}, 0),
+            :graph_s1_nodes_reduced => get(node_dict_reduced, ComputeTaskNode{ComputeTaskQED_S1}, 0),
+            :graph_s2_nodes_reduced => get(node_dict_reduced, ComputeTaskNode{ComputeTaskQED_S2}, 0),
+            :graph_edges_reduced => props_reduced.noEdges,
+            :graph_mem => graph_size,
+            :graph_mem_reduced => graph_size_reduced,
+            :graph_elapsed_reduce => reduce_elapsed,
+        ),
+    )
+    return nothing
+end
+
+processes = [
+    ("ke->ke", true),
+    ("ke->kke", true),
+    ("ke->kkke", true),
+    ("ke->kkkke", true),
+    ("ke->kkkkke", true),
+    ("ke->kkkkkke", true),
+    ("ke->kkkkkkke", true),
+    #("ke->kkkkkkkke", false),
+    #("ke->kkkkkkkkke", false),
+]
+
+df = DataFrame(
+    process_name = String[],
+    cpu_threads = Int[],
+    graph_gen_samples = Int[],
+    graph_gen_mean = Float64[],
+    graph_gen_std = Float64[],
+    graph_gen_median = Float64[],
+    graph_nodes = Int[],
+    graph_data_nodes = Int[],
+    graph_u_nodes = Int[],
+    graph_v_nodes = Int[],
+    graph_s1_nodes = Int[],
+    graph_s2_nodes = Int[],
+    graph_edges = Int[],
+    graph_nodes_reduced = Int[],
+    graph_data_nodes_reduced = Int[],
+    graph_u_nodes_reduced = Int[],
+    graph_v_nodes_reduced = Int[],
+    graph_s1_nodes_reduced = Int[],
+    graph_s2_nodes_reduced = Int[],
+    graph_edges_reduced = Int[],
+    graph_mem = Float64[],
+    graph_mem_reduced = Float64[],
+    graph_elapsed_reduce = Float64[],
+)
+
+# if they exist, read existing results and append new ones
+if isfile(results_filename)
+    df = CSV.read(results_filename, DataFrame)
+end
+
+bench_process("ke->kke", warmup = true)
+
+for (process, opt) in processes
+    bench_process(process, optimize = opt)
+    CSV.write(results_filename, df)
+end
--- a/experiments/CUDA_container.def
+++ b/experiments/CUDA_container.def
@ -0,0 +1,63 @@
+Bootstrap: docker
+From: nvidia/cuda:12.3.1-devel-ubuntu20.04
+
+%labels
+    Requires CUDA driver 470.57+.
+
+%environment
+    export LANG=C
+
+%runscript
+    nvidia-smi
+    ./run.sh
+
+%post
+    . /.singularity.d/env/10-docker*.sh
+
+    apt-get update
+    apt-get install -y pciutils
+    DEBIAN_FRONTEND='noninteractive' apt-get -y -o Dpkg::Options::='--force-confdef' -o Dpkg::Options::='--force-confold' install build-essential cuda-compat-12-3 libibverbs-dev ibverbs-utils gcc wget git libcap2-bin
+    apt-get -y autoremove; apt-get -y clean
+
+    cd /tmp
+
+    # install slurm
+    : ${SLURM_VERSION:=17-02-11-1}
+    wget https://github.com/SchedMD/slurm/archive/slurm-${SLURM_VERSION}.tar.gz
+    tar -xf slurm-${SLURM_VERSION}.tar.gz
+        cd slurm-slurm-${SLURM_VERSION}
+        ./configure --prefix=/usr/ --sysconfdir=/etc/slurm --localstatedir=/var --disable-debug
+        make -C contribs/pmi2 -j$(nproc) install
+    cd ..
+    rm -rf slurm-*
+
+    # install julia
+    cd ~
+    wget https://julialang-s3.julialang.org/bin/linux/x64/1.9/julia-1.9.4-linux-x86_64.tar.gz
+    tar zxvf julia-1.9.4-linux-x86_64.tar.gz
+    mv julia-1.9.4/ /opt/julia-1.9.4
+    #mkdir /usr/local/bin
+    ln -s /opt/julia-1.9.4/bin/julia /usr/local/bin/julia
+
+    #Add nvidia driver paths to the environment variables
+    echo "\n #Nvidia driver paths \n" >> /environment
+    echo 'export PATH="/nvbin:$PATH"' >> /environment
+    echo 'export LD_LIBRARY_PATH="/nvlib:$LD_LIBRARY_PATH"' >> /environment
+
+    #Add CUDA paths
+    echo "\n #Cuda paths \n" >> /environment
+    echo 'export CPATH="/usr/local/cuda/include:$CPATH"' >> /environment
+    echo 'export PATH="/usr/local/cuda/bin:$PATH"' >> /environment
+    echo 'export LD_LIBRARY_PATH="/usr/local/cuda/lib64:$LD_LIBRARY_PATH"' >> /environment
+    echo 'export CUDA_HOME="/usr/local/cuda"' >> /environment
+
+    # install likwid
+    VERSION=5.3.0
+    wget http://ftp.fau.de/pub/likwid/likwid-$VERSION.tar.gz
+    tar -xaf likwid-$VERSION.tar.gz
+    cd likwid-$VERSION
+    # accessdaemon doesn't work because of permissions
+    sed -i 's/ACCESSMODE = accessdaemon/ACCESSMODE = perf_event/g' config.mk
+    make -j4
+    make -j4 install 
+    echo 'export LD_LIBRARY_PATH="/usr/local/lib:$LD_LIBRARY_PATH"' >> /environment
--- a/experiments/cluster/diagram_bench_hemera.sh
+++ b/experiments/cluster/diagram_bench_hemera.sh
@ -0,0 +1,22 @@
+#!/bin/bash
+#SBATCH --array=1-32
+#SBATCH --job-name=qed_bench
+#SBATCH --partition=intel
+#SBATCH --time=16:00:00
+#SBATCH --nodes=1
+#SBATCH --ntasks=1
+#SBATCH --cpus-per-task=32
+#SBATCH --mem=16GB
+#SBATCH --output=simulation-%A-%a.out
+#SBATCH --error=simulation-%A-%a.err
+
+cd $HOME/repos/metagraph_optimization
+
+module load singularity
+module load git 
+
+printf "Current git commit hash: " > results/git.txt
+git rev-parse HEAD >> results/git.txt
+git status >> results/git.txt
+
+singularity exec experiments/CUDA_container.sif ./experiments/run_qed_exec.sh $SLURM_ARRAY_TASK_ID
--- a/experiments/cluster/diagram_bench_hemera_a100.sh
+++ b/experiments/cluster/diagram_bench_hemera_a100.sh
@ -0,0 +1,24 @@
+#!/bin/bash
+#SBATCH --job-name=qed_bench
+#SBATCH --partition=casus_a100
+#SBATCH --account=casus
+#SBATCH --time=8:00:00
+#SBATCH --nodes=1
+#SBATCH --ntasks=1
+#SBATCH --cpus-per-task=32
+#SBATCH --gres=gpu:1
+#SBATCH --mem=256GB
+#SBATCH --output=simulation-%A-%a.out
+#SBATCH --error=simulation-%A-%a.err
+
+cd $HOME/repos/metagraph_optimization
+
+module load singularity
+module load git
+module load cuda/12.1
+
+printf "Current git commit hash: " > results/git.txt
+git rev-parse HEAD >> results/git.txt
+git status >> results/git.txt
+
+singularity exec --nv experiments/CUDA_container.sif ./experiments/run_qed_exec.sh 32
--- a/experiments/cluster/full_node_hemera.sh
+++ b/experiments/cluster/full_node_hemera.sh
@ -0,0 +1,24 @@
+#!/bin/bash
+#SBATCH --job-name=qed_bench
+#SBATCH --partition=casus_a100
+#SBATCH --account=casus
+#SBATCH --time=8:00:00
+#SBATCH --nodes=1
+#SBATCH --ntasks=1
+#SBATCH --cpus-per-task=128
+#SBATCH --gres=gpu:4
+#SBATCH --mem=2048GB
+#SBATCH --output=simulation-%A-%a.out
+#SBATCH --error=simulation-%A-%a.err
+
+cd $HOME/repos/metagraph_optimization
+
+module load singularity
+module load git
+module load cuda/12.1
+
+printf "Current git commit hash: " > results/git_reduce_bench_gpu.txt
+git rev-parse HEAD >> results/git_reduce_bench_gpu.txt
+git status >> results/git_reduce_bench_gpu.txt
+
+singularity exec --nv experiments/CUDA_container.sif ./experiments/full_node.sh
--- a/experiments/cluster/gen_diagram_hemera.sh
+++ b/experiments/cluster/gen_diagram_hemera.sh
@ -0,0 +1,22 @@
+#!/bin/bash
+#SBATCH --array=1-8
+#SBATCH --job-name=qed_diag_gen
+#SBATCH --partition=intel
+#SBATCH --time=4:00:00
+#SBATCH --nodes=1
+#SBATCH --ntasks=1
+#SBATCH --cpus-per-task=8
+#SBATCH --mem=64GB
+#SBATCH --output=simulation-%A-%a.out
+#SBATCH --error=simulation-%A-%a.err
+
+cd $HOME/repos/metagraph_optimization
+
+module load singularity
+module load git 
+
+printf "Current git commit hash: " > results/git.txt
+git rev-parse HEAD >> results/git.txt
+git status >> results/git.txt
+
+singularity exec experiments/CUDA_container.sif ./experiments/run_gen_diagram.sh $SLURM_ARRAY_TASK_ID
--- a/experiments/cluster/reduce_bench_hemera.sh
+++ b/experiments/cluster/reduce_bench_hemera.sh
@ -0,0 +1,21 @@
+#!/bin/bash
+#SBATCH --job-name=qed_bench
+#SBATCH --partition=intel
+#SBATCH --time=48:00:00
+#SBATCH --nodes=1
+#SBATCH --ntasks=1
+#SBATCH --cpus-per-task=32
+#SBATCH --mem=24GB
+#SBATCH --output=simulation-%A-%a.out
+#SBATCH --error=simulation-%A-%a.err
+
+cd $HOME/repos/metagraph_optimization
+
+module load singularity
+module load git 
+
+printf "Current git commit hash: " > results/git_reduce_bench.txt
+git rev-parse HEAD >> results/git_reduce_bench.txt
+git status >> results/git_reduce_bench.txt
+
+singularity exec experiments/CUDA_container.sif ./experiments/run_reduce_bench.sh
--- a/experiments/cluster/reduce_bench_hemera_gpu.sh
+++ b/experiments/cluster/reduce_bench_hemera_gpu.sh
@ -0,0 +1,24 @@
+#!/bin/bash
+#SBATCH --job-name=qed_bench
+#SBATCH --partition=casus_a100
+#SBATCH --account=casus
+#SBATCH --time=16:00:00
+#SBATCH --nodes=1
+#SBATCH --ntasks=1
+#SBATCH --cpus-per-task=32
+#SBATCH --gres=gpu:1
+#SBATCH --mem=256GB
+#SBATCH --output=simulation-%A-%a.out
+#SBATCH --error=simulation-%A-%a.err
+
+cd $HOME/repos/metagraph_optimization
+
+module load singularity
+module load git 
+module load cuda/12.1
+
+printf "Current git commit hash: " > results/git_reduce_bench_gpu.txt
+git rev-parse HEAD >> results/git_reduce_bench_gpu.txt
+git status >> results/git_reduce_bench_gpu.txt
+
+singularity exec --nv experiments/CUDA_container.sif ./experiments/run_reduce_bench_gpu.sh
--- a/experiments/full_node.sh
+++ b/experiments/full_node.sh
@ -0,0 +1,25 @@
+#!/bin/bash
+
+SCRIPT_DIR=$( cd -- "$( dirname -- "${BASH_SOURCE[0]}" )" &> /dev/null && pwd )
+LOG_FILE="$SCRIPT_DIR/../julia_full_node.log"
+
+cd $SCRIPT_DIR/..
+
+echo "Writing system info..."
+
+# collect some information of the used node and system
+uname -a > results/system_full_node.txt
+julia --version > results/julia_full_node.txt
+lscpu > results/cpu_full_node.txt
+nvidia-smi > results/cuda_gpu_full_node.txt
+lsblk > results/storage_full_node.txt
+lspci > results/pci_full_node.txt
+
+#echo "Initiating julia..."
+#julia --threads=8 --project=./ -e 'using Pkg; Pkg.instantiate(); Pkg.add(url="https://github.com/QEDjl-project/QEDprocesses.jl/")' >> $LOG_FILE 2>&1 || exit 1    # need current dev version of QEDprocesses
+#julia --threads=8 --project=./ -e 'using Pkg; Pkg.instantiate(); Pkg.add(url="https://github.com/AntonReinhard/QEDbase.jl/tree/fix_bs_multiplication")' >> $LOG_FILE 2>&1 || exit 1 # need a specific fix for abs*bs multiplication for gpu
+#julia --threads=8 -e 'using Pkg; Pkg.add("CSV"); Pkg.add("DataFrames"); Pkg.add("CUDA"); Pkg.add("Random"); Pkg.add("BenchmarkTools"); Pkg.add("Dates")' >> $LOG_FILE 2>&1 || exit 1        # add requirements for the bench script
+#julia --project -e 'using CUDA; CUDA.set_runtime_version!(VersionNumber("12.1"))' >> $LOG_FILE 2>&1 || echo "Failed to set CUDA version number"
+
+echo "Benchmarking Full Node 128 Threads + *GPUs*"
+julia --project -O3 --threads=128 examples/full_node_bench.jl >> $LOG_FILE 2>&1 || echo "-- Something went wrong, check logs --"
--- a/experiments/run_gen_diagram.sh
+++ b/experiments/run_gen_diagram.sh
@ -0,0 +1,27 @@
+#!/bin/bash
+
+# first arg = number of threads
+i=$1
+
+SCRIPT_DIR=$( cd -- "$( dirname -- "${BASH_SOURCE[0]}" )" &> /dev/null && pwd )
+LOG_FILE="$SCRIPT_DIR/../julia.log"
+
+cd $SCRIPT_DIR/..
+
+echo "Writing system info..."
+
+# collect some information of the used node and system
+uname -a > results/system.txt
+julia --version > results/julia.txt
+lscpu > results/cpu.txt
+lsblk > results/storage.txt
+lspci > results/pci.txt
+
+echo "Initiating julia..."
+julia --threads=8 --project=./ -e 'using Pkg; Pkg.instantiate(); Pkg.add(url="https://github.com/QEDjl-project/QEDprocesses.jl/")' >> $LOG_FILE 2>&1 || exit 1    # need current dev version of QEDprocesses
+julia --threads=8 --project=./ -e 'using Pkg; Pkg.instantiate(); Pkg.add(url="https://github.com/AntonReinhard/QEDbase.jl/tree/fix_bs_multiplication")' >> $LOG_FILE 2>&1 || exit 1 # need a specific fix for abs*bs multiplication for gpu
+julia --threads=8 -e 'using Pkg; Pkg.add("CSV"); Pkg.add("DataFrames"); Pkg.add("BenchmarkTools"); Pkg.add("StatsBase")' >> $LOG_FILE 2>&1 || exit 1        # add requirements for the bench script
+
+echo "Benchmarking with $i threads..."
+
+julia --project -O3 --threads=$i examples/qed_gen_bench.jl >> $LOG_FILE 2>&1 || echo "-- Something went wrong, check logs --"
--- a/experiments/run_qed_exec.sh
+++ b/experiments/run_qed_exec.sh
@ -0,0 +1,31 @@
+#!/bin/bash
+
+# first arg = number of threads
+i=$1
+
+SCRIPT_DIR=$( cd -- "$( dirname -- "${BASH_SOURCE[0]}" )" &> /dev/null && pwd )
+LOG_FILE="$SCRIPT_DIR/../julia_$i.log"
+
+cd $SCRIPT_DIR/..
+
+echo "Writing system info..."
+
+# collect some information of the used node and system
+uname -a > results/system_$i.txt
+julia --version > results/julia_$i.txt
+lscpu > results/cpu_$i.txt
+nvidia-smi > results/cuda_gpu_$i.txt
+lsblk > results/storage_$i.txt
+lspci > results/pci_$i.txt
+
+echo "Initiating julia..."
+#julia --threads=8 --project=./ -e 'using Pkg; Pkg.instantiate(); Pkg.add(url="https://github.com/QEDjl-project/QEDprocesses.jl/")' >> $LOG_FILE 2>&1 || exit 1    # need current dev version of QEDprocesses
+#julia --threads=8 --project=./ -e 'using Pkg; Pkg.instantiate(); Pkg.add(url="https://github.com/AntonReinhard/QEDbase.jl/tree/fix_bs_multiplication")' >> $LOG_FILE 2>&1 || exit 1 # need a specific fix for abs*bs multiplication for gpu
+#julia --threads=8 -e 'using Pkg; Pkg.add("CSV"); Pkg.add("DataFrames"); Pkg.add("LIKWID"); Pkg.add("CUDA"); Pkg.add("Random"); Pkg.add("BenchmarkTools"); Pkg.add("Dates")' >> $LOG_FILE 2>&1 || exit 1        # add requirements for the bench script
+#julia --project -e 'using CUDA; CUDA.set_runtime_version!(VersionNumber("12.1"))' >> $LOG_FILE 2>&1 || echo "Failed to set CUDA version number"
+
+echo "Benchmarking $i Threads"
+julia --project -O3 --threads=$i examples/qed_bench.jl >> $LOG_FILE 2>&1 || echo "-- Something went wrong, check logs --"
+
+echo "Benchmarking Tape variant $i Threads"
+julia --project -O3 --threads=$i examples/qed_bench_tape.jl >> $LOG_FILE 2>&1 || echo "-- Something went wrong, check logs --"
--- a/experiments/run_reduce_bench.sh
+++ b/experiments/run_reduce_bench.sh
@ -0,0 +1,24 @@
+#!/bin/bash
+
+SCRIPT_DIR=$( cd -- "$( dirname -- "${BASH_SOURCE[0]}" )" &> /dev/null && pwd )
+LOG_FILE="$SCRIPT_DIR/../julia_bench_reduce.log"
+
+cd $SCRIPT_DIR/..
+
+echo "Writing system info..."
+
+# collect some information of the used node and system
+uname -a > results/system_bench_reduce.txt
+julia --version > results/julia_bench_reduce.txt
+lscpu > results/cpu_bench_reduce.txt
+nvidia-smi > results/cuda_gpu_bench_reduce.txt
+lsblk > results/storage_bench_reduce.txt
+lspci > results/pci_bench_reduce.txt
+
+#echo "Initiating julia..."
+#julia --threads=8 --project=./ -e 'using Pkg; Pkg.instantiate(); Pkg.add(url="https://github.com/QEDjl-project/QEDprocesses.jl/")' >> $LOG_FILE 2>&1 || exit 1    # need current dev version of QEDprocesses
+#julia --threads=8 --project=./ -e 'using Pkg; Pkg.instantiate(); Pkg.add(url="https://github.com/AntonReinhard/QEDbase.jl/tree/fix_bs_multiplication")' >> $LOG_FILE 2>&1 || exit 1 # need a specific fix for abs*bs multiplication for gpu
+#julia --threads=8 -e 'using Pkg; Pkg.add("CSV"); Pkg.add("DataFrames"); Pkg.add("LIKWID"); Pkg.add("CUDA"); Pkg.add("Random"); Pkg.add("BenchmarkTools"); Pkg.add("Dates")' >> $LOG_FILE 2>&1 || exit 1        # add requirements for the bench script
+
+echo "Benchmarking Reduction 32 Threads"
+julia --project -O3 --threads=32 examples/qed_bench_reduction_steps.jl >> $LOG_FILE 2>&1 || echo "-- Something went wrong, check logs --"
--- a/experiments/run_reduce_bench_gpu.sh
+++ b/experiments/run_reduce_bench_gpu.sh
@ -0,0 +1,25 @@
+#!/bin/bash
+
+SCRIPT_DIR=$( cd -- "$( dirname -- "${BASH_SOURCE[0]}" )" &> /dev/null && pwd )
+LOG_FILE="$SCRIPT_DIR/../julia_bench_reduce_gpu.log"
+
+cd $SCRIPT_DIR/..
+
+echo "Writing system info..."
+
+# collect some information of the used node and system
+uname -a > results/system_bench_reduce_gpu.txt
+julia --version > results/julia_bench_reduce_gpu.txt
+lscpu > results/cpu_bench_reduce_gpu.txt
+nvidia-smi > results/cuda_gpu_bench_reduce_gpu.txt
+lsblk > results/storage_bench_reduce_gpu.txt
+lspci > results/pci_bench_reduce_gpu.txt
+
+#echo "Initiating julia..."
+#julia --threads=8 --project=./ -e 'using Pkg; Pkg.instantiate(); Pkg.add(url="https://github.com/QEDjl-project/QEDprocesses.jl/")' >> $LOG_FILE 2>&1 || exit 1    # need current dev version of QEDprocesses
+#julia --threads=8 --project=./ -e 'using Pkg; Pkg.instantiate(); Pkg.add(url="https://github.com/AntonReinhard/QEDbase.jl/tree/fix_bs_multiplication")' >> $LOG_FILE 2>&1 || exit 1 # need a specific fix for abs*bs multiplication for gpu
+#julia --threads=8 -e 'using Pkg; Pkg.add("CSV"); Pkg.add("DataFrames"); Pkg.add("LIKWID"); Pkg.add("CUDA"); Pkg.add("Random"); Pkg.add("BenchmarkTools"); Pkg.add("Dates")' >> $LOG_FILE 2>&1 || exit 1        # add requirements for the bench script
+#julia --project -e 'using CUDA; CUDA.set_runtime_version!(VersionNumber("12.1"))' >> $LOG_FILE 2>&1 || echo "Failed to set CUDA version number"
+
+echo "Benchmarking Reduction 32 Threads, *GPU*"
+julia --project -O3 --threads=32 examples/qed_bench_reduction_steps_gpu.jl >> $LOG_FILE 2>&1 || echo "-- Something went wrong, check logs --"
--- a/images/AB->ABBBBB_reduction_bench.pdf
+++ b/images/AB->ABBBBB_reduction_bench.pdf
--- a/images/AB->ABBB_reduction_bench.pdf
+++ b/images/AB->ABBB_reduction_bench.pdf
--- a/images/README.md
+++ b/images/README.md
@ -0,0 +1,3 @@
+# Images
+
+In this folder we collect benchmark results in pdf form which may be useful in the future.
--- a/images/compton_diagram_gen_comparison.pdf
+++ b/images/compton_diagram_gen_comparison.pdf
--- a/images/compton_graph_size_reduced.pdf
+++ b/images/compton_graph_size_reduced.pdf
--- a/images/compton_graph_size_unreduced.pdf
+++ b/images/compton_graph_size_unreduced.pdf
--- a/images/compton_graph_size_versus.pdf
+++ b/images/compton_graph_size_versus.pdf
--- a/images/cpu_vs_gpu_abc.pdf
+++ b/images/cpu_vs_gpu_abc.pdf
--- a/images/cpu_vs_gpu_qed.pdf
+++ b/images/cpu_vs_gpu_qed.pdf
--- a/images/full_node/1bil_size/full_node_chunk_size_1k_rate.pdf
+++ b/images/full_node/1bil_size/full_node_chunk_size_1k_rate.pdf
--- a/images/full_node/1bil_size/full_node_chunk_size_1k_ratio.pdf
+++ b/images/full_node/1bil_size/full_node_chunk_size_1k_ratio.pdf
--- a/images/full_node/1bil_size/full_node_chunk_size_1k_time.pdf
+++ b/images/full_node/1bil_size/full_node_chunk_size_1k_time.pdf
--- a/images/full_node/1bil_size/full_node_chunk_size_2k_rate.pdf
+++ b/images/full_node/1bil_size/full_node_chunk_size_2k_rate.pdf
--- a/images/full_node/1bil_size/full_node_chunk_size_2k_ratio.pdf
+++ b/images/full_node/1bil_size/full_node_chunk_size_2k_ratio.pdf
--- a/images/full_node/1bil_size/full_node_chunk_size_2k_time.pdf
+++ b/images/full_node/1bil_size/full_node_chunk_size_2k_time.pdf
--- a/images/full_node/1bil_size/full_node_chunk_size_3k_rate.pdf
+++ b/images/full_node/1bil_size/full_node_chunk_size_3k_rate.pdf
--- a/images/full_node/1bil_size/full_node_chunk_size_3k_ratio.pdf
+++ b/images/full_node/1bil_size/full_node_chunk_size_3k_ratio.pdf
--- a/images/full_node/1bil_size/full_node_chunk_size_3k_time.pdf
+++ b/images/full_node/1bil_size/full_node_chunk_size_3k_time.pdf
--- a/images/full_node/1bil_size/full_node_chunk_size_4k_rate.pdf
+++ b/images/full_node/1bil_size/full_node_chunk_size_4k_rate.pdf
--- a/images/full_node/1bil_size/full_node_chunk_size_4k_ratio.pdf
+++ b/images/full_node/1bil_size/full_node_chunk_size_4k_ratio.pdf
--- a/images/full_node/1bil_size/full_node_chunk_size_4k_time.pdf
+++ b/images/full_node/1bil_size/full_node_chunk_size_4k_time.pdf
--- a/images/full_node/1bil_size/full_node_chunk_size_5k_rate.pdf
+++ b/images/full_node/1bil_size/full_node_chunk_size_5k_rate.pdf
--- a/images/full_node/1bil_size/full_node_chunk_size_5k_ratio.pdf
+++ b/images/full_node/1bil_size/full_node_chunk_size_5k_ratio.pdf
--- a/images/full_node/1bil_size/full_node_chunk_size_5k_time.pdf
+++ b/images/full_node/1bil_size/full_node_chunk_size_5k_time.pdf
--- a/images/full_node/1bil_size/full_node_process_best_rate.pdf
+++ b/images/full_node/1bil_size/full_node_process_best_rate.pdf
--- a/images/full_node/1bil_size/full_node_process_best_rate_plus_theory.pdf
+++ b/images/full_node/1bil_size/full_node_process_best_rate_plus_theory.pdf
--- a/images/full_node/1bil_size/full_node_process_best_time.pdf
+++ b/images/full_node/1bil_size/full_node_process_best_time.pdf
--- a/images/full_node/64mil_size/full_node_chunk_size_1k_rate.pdf
+++ b/images/full_node/64mil_size/full_node_chunk_size_1k_rate.pdf
--- a/images/full_node/64mil_size/full_node_chunk_size_1k_ratio.pdf
+++ b/images/full_node/64mil_size/full_node_chunk_size_1k_ratio.pdf
--- a/images/full_node/64mil_size/full_node_chunk_size_1k_time.pdf
+++ b/images/full_node/64mil_size/full_node_chunk_size_1k_time.pdf
--- a/images/full_node/64mil_size/full_node_chunk_size_2k_rate.pdf
+++ b/images/full_node/64mil_size/full_node_chunk_size_2k_rate.pdf
--- a/images/full_node/64mil_size/full_node_chunk_size_2k_ratio.pdf
+++ b/images/full_node/64mil_size/full_node_chunk_size_2k_ratio.pdf
--- a/images/full_node/64mil_size/full_node_chunk_size_2k_time.pdf
+++ b/images/full_node/64mil_size/full_node_chunk_size_2k_time.pdf
--- a/images/full_node/64mil_size/full_node_chunk_size_3k_rate.pdf
+++ b/images/full_node/64mil_size/full_node_chunk_size_3k_rate.pdf
--- a/images/full_node/64mil_size/full_node_chunk_size_3k_ratio.pdf
+++ b/images/full_node/64mil_size/full_node_chunk_size_3k_ratio.pdf
--- a/images/full_node/64mil_size/full_node_chunk_size_3k_time.pdf
+++ b/images/full_node/64mil_size/full_node_chunk_size_3k_time.pdf
--- a/images/full_node/64mil_size/full_node_chunk_size_4k_rate.pdf
+++ b/images/full_node/64mil_size/full_node_chunk_size_4k_rate.pdf
--- a/images/full_node/64mil_size/full_node_chunk_size_4k_ratio.pdf
+++ b/images/full_node/64mil_size/full_node_chunk_size_4k_ratio.pdf
--- a/images/full_node/64mil_size/full_node_chunk_size_4k_time.pdf
+++ b/images/full_node/64mil_size/full_node_chunk_size_4k_time.pdf
--- a/images/full_node/64mil_size/full_node_chunk_size_5k_rate.pdf
+++ b/images/full_node/64mil_size/full_node_chunk_size_5k_rate.pdf
--- a/images/full_node/64mil_size/full_node_chunk_size_5k_ratio.pdf
+++ b/images/full_node/64mil_size/full_node_chunk_size_5k_ratio.pdf
--- a/images/full_node/64mil_size/full_node_chunk_size_5k_time.pdf
+++ b/images/full_node/64mil_size/full_node_chunk_size_5k_time.pdf
--- a/images/full_node/64mil_size/full_node_process_best_rate.pdf
+++ b/images/full_node/64mil_size/full_node_process_best_rate.pdf
--- a/images/full_node/64mil_size/full_node_process_best_rate_plus_theory.pdf
+++ b/images/full_node/64mil_size/full_node_process_best_rate_plus_theory.pdf
--- a/images/full_node/64mil_size/full_node_process_best_time.pdf
+++ b/images/full_node/64mil_size/full_node_process_best_time.pdf
--- a/images/full_node/README.md
+++ b/images/full_node/README.md
@ -0,0 +1,3 @@
+# Full Node Benchmarks
+
+Done using `experiments/full_node.sh` on hemera, using 128 threads and 4 A100 GPUs.
--- a/images/fuse_reduce_split_cdplot_greedy.pdf
+++ b/images/fuse_reduce_split_cdplot_greedy.pdf
--- a/images/gen_memory.pdf
+++ b/images/gen_memory.pdf
--- a/images/gen_times.pdf
+++ b/images/gen_times.pdf
--- a/images/gpu_rate_NVIDIA
+++ b/images/gpu_rate_NVIDIA
--- a/images/gpu_times_NVIDIA
+++ b/images/gpu_times_NVIDIA
--- a/images/ke->kke_reduction_bench.pdf
+++ b/images/ke->kke_reduction_bench.pdf
--- a/images/ke->kkke_reduction_bench_lin.pdf
+++ b/images/ke->kkke_reduction_bench_lin.pdf
--- a/images/ke->kkkke_reduction_bench.pdf
+++ b/images/ke->kkkke_reduction_bench.pdf
--- a/images/ke->kkkkke_reduction_bench.pdf
+++ b/images/ke->kkkkke_reduction_bench.pdf
--- a/images/optim_plots/README.md
+++ b/images/optim_plots/README.md
@ -0,0 +1,5 @@
+# Optimizer Plots
+
+Plots of FusionOptimizer, ReductionOptimizer, SplitOptimizer, RandomWalkOptimizer, and GreedyOptimizer, executed on a system with 32 threads and an A30 GPU.
+
+Benchmarked using `notebooks/optimizers.ipynb`.
--- a/images/optim_plots/qed_k3_cdplot_fuse.pdf
+++ b/images/optim_plots/qed_k3_cdplot_fuse.pdf
--- a/images/optim_plots/qed_k3_cdplot_greedy.pdf
+++ b/images/optim_plots/qed_k3_cdplot_greedy.pdf
--- a/images/optim_plots/qed_k3_cdplot_random.pdf
+++ b/images/optim_plots/qed_k3_cdplot_random.pdf
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
rubydragon	d888713e97	Enable oneAPI and ROCm (#9 ) All checks were successful MetagraphOptimization_CI / docs (push) Successful in 8m30s Details MetagraphOptimization_CI / test (push) Successful in 22m2s Details Co-authored-by: Anton Reinhard <anton.reinhard@proton.me> Reviewed-on: #9	2024-05-08 19:26:18 +02:00
rubydragon	38e7ff3b90	Seed Randomness, Fix tests (#8 ) All checks were successful MetagraphOptimization_CI / docs (push) Successful in 7m34s Details MetagraphOptimization_CI / test (push) Successful in 20m49s Details Seeded randomness in all places, however, multithreaded randomness still exists. Disabled some tests that are failing, will add issues and fix later. These are related to (likely) precision problems in the ABC model, which is not priority, and the Node Fusion, which will be fundamentally reworked anyways. Co-authored-by: Anton Reinhard <anton.reinhard@proton.me> Reviewed-on: #8	2024-05-08 18:04:48 +02:00
rubydragon	7d7782f97f	Add Workaround for Trie implementation for Julia version 1.10+ (#7 ) Some checks failed MetagraphOptimization_CI / docs (push) Successful in 8m32s Details MetagraphOptimization_CI / test (push) Failing after 12m1s Details Co-authored-by: Anton Reinhard <anton.reinhard@proton.me> Reviewed-on: #7	2024-05-08 14:00:25 +02:00
rubydragon	87dbaf2c32	experiments (#1 ) All checks were successful MetagraphOptimization_CI / docs (push) Successful in 10m41s Details MetagraphOptimization_CI / test (push) Successful in 30m40s Details Co-authored-by: Anton Reinhard <anton.reinhard@proton.me> Reviewed-on: #1	2024-05-08 12:03:27 +02:00
Anton Reinhard	82ed774b7e	Tape Machine (#30 ) Adds a tape machine way of executing the code. The tape machine is a series of FunctionCall objects, which can either be called one by one, or be used to generate expressions to make up a function. Reviewed-on: Rubydragon/MetagraphOptimization.jl#30 Co-authored-by: Anton Reinhard <anton.reinhard@proton.me> Co-committed-by: Anton Reinhard <anton.reinhard@proton.me>	2024-01-03 16:38:32 +01:00
Anton Reinhard	92e0eeaaef	heterogeneity (#27 ) Prepare things to work with heterogeneity, make things work on GPU Reviewed-on: Rubydragon/MetagraphOptimization.jl#27 Co-authored-by: Anton Reinhard <anton.reinhard@proton.me> Co-committed-by: Anton Reinhard <anton.reinhard@proton.me>	2023-12-18 14:31:52 +01:00
Anton Reinhard	c90346e948	Add QED Model (#25 ) Reviewed-on: Rubydragon/MetagraphOptimization.jl#25 Co-authored-by: Anton Reinhard <anton.reinhard@proton.me> Co-committed-by: Anton Reinhard <anton.reinhard@proton.me>	2023-12-07 02:54:15 +01:00
Anton Reinhard	938bf216e5	Improve actions workflow by removing prepare step (#23 ) Reviewed-on: Rubydragon/MetagraphOptimization.jl#23 Co-authored-by: Anton Reinhard <anton.reinhard@proton.me> Co-committed-by: Anton Reinhard <anton.reinhard@proton.me>	2023-11-24 19:20:05 +01:00