sinc vectorization test

random/sinc.jl

@@ -0,0 +1,28 @@
+# @INFO: there is no SIMD version for sinc (sinpi etc.)
+# We need a sinc kernel that can be SIMD vectorized
+# Below is a sinple comparision of accuracy between a few approaches
+using GLMakie
+
+pix(x) = π * x
+sinc2(x) = x |> pix |> (x -> iszero(x) ? 1.0 : sin(x)/x)
+sinc3(x) = x |> pix |> (x -> x + eps(x)) |> (x -> sin(x)/x)
+
+
+fs = 480e3
+Δt = 1 / fs
+t = -0.5:Δt:0.5
+y_sinc = t .|> sinc;
+y_sinc2 = t .|> sinc2;
+y_sinc3 = t .|> sinc3;
+
+
+@show maximum(abs, y_sinc .- y_sinc2)
+@show maximum(abs, y_sinc .- y_sinc3)
+
+fig = Figure()
+ax = Axis(fig[1, 1])
+lines!(ax, t, abs.(y_sinc - y_sinc2) .|> log10; label="if sinc")
+lines!(ax, t, abs.(y_sinc - y_sinc3) .|> log10; label="eps sinc")
+ylims!(ax, -16, -12)
+xlims!(ax, (t |> extrema)...)
+axislegend(ax)