From cbddb0f25bf9891b1e801671da39d375e9d70591 Mon Sep 17 00:00:00 2001
From: Charles Kawczynski <kawczynski.charles@gmail.com>
Date: Wed, 17 Mar 2021 11:49:55 -0700
Subject: [PATCH] Add perf timings in docs

---
 docs/make.jl                   |  3 ++-
 docs/src/perf.jl               | 48 ++++++++++++++++++++++++++++++++++
 docs/src/performance.md        |  2 +-
 docs/src/performance_timing.md |  8 ++++++
 4 files changed, 59 insertions(+), 2 deletions(-)
 create mode 100644 docs/src/perf.jl
 create mode 100644 docs/src/performance_timing.md

diff --git a/docs/make.jl b/docs/make.jl
index 8a4a1ff..37727ee 100644
--- a/docs/make.jl
+++ b/docs/make.jl
@@ -15,7 +15,8 @@ makedocs(
     modules = [DispatchedTuples],
     pages = Any[
         "Home" => "index.md",
-        "Performance" => "performance.md",
+        "Performance introspection" => "performance.md",
+        "Performance timing" => "performance_timing.md",
         "API" => "api.md",
     ],
 )
diff --git a/docs/src/perf.jl b/docs/src/perf.jl
new file mode 100644
index 0000000..7d69917
--- /dev/null
+++ b/docs/src/perf.jl
@@ -0,0 +1,48 @@
+using DispatchedTuples
+using BenchmarkTools
+
+# (N, N_buf) are low for quick doc builds
+# (N, N_buf) = (5, 3) is reasonable
+N = 0
+N_buf = 1
+
+foo(i) = Symbol(:Foo, i)
+expr(i) = :(struct $(foo(i)) end)
+for i in 1:N+N_buf
+    eval(expr(i))
+end
+
+fooinst(i::Int) = eval(:($(foo(i))))()
+
+footup(n::Int) = ntuple(n) do i
+    (fooinst(i), i)
+end
+
+function perf_dt(N, N_buf)
+    println("********* DispatchedTuple")
+    for n in N_buf:N+N_buf
+        println("--- n = $n")
+        tup = footup(n)
+        @btime footup($n)
+
+        dtup = DispatchedTuple(tup)
+        @btime DispatchedTuple($tup)
+        # @btime $dtup[Foo3()] # always results in 0.028 ns (0 allocations: 0 bytes)
+    end
+end
+
+function perf_ds(N, N_buf)
+    println("********* DispatchedSet")
+    for n in N_buf:N+N_buf
+        println("--- n = $n")
+        tup = footup(n)
+        @btime footup($n)
+
+        dtup = DispatchedSet(tup)
+        @btime DispatchedSet($tup)
+        # @btime $dtup[Foo3()] # always results in 0.028 ns (0 allocations: 0 bytes)
+    end
+end
+
+perf_dt(N, N_buf)
+perf_ds(N, N_buf)
diff --git a/docs/src/performance.md b/docs/src/performance.md
index 1fad614..c1d542b 100644
--- a/docs/src/performance.md
+++ b/docs/src/performance.md
@@ -1,4 +1,4 @@
-# Performance
+# Performance introspection
 
 The performance of `DispatchedTuple`s should scale similar to the performance of ordinary tuples (good with small tuples, but expensive with larger ones).
 
diff --git a/docs/src/performance_timing.md b/docs/src/performance_timing.md
new file mode 100644
index 0000000..e72d98c
--- /dev/null
+++ b/docs/src/performance_timing.md
@@ -0,0 +1,8 @@
+# Performance timing
+
+See the `perf.jl` file for adjusting / experimenting with timings. Note that the output is limited for quick doc builds.
+
+```@example perf
+include("perf.jl")
+nothing
+```