Modest performance improvements + some suggestions for future profiling from Claude

src/KineticForces/PitchIntegration.jl

@@ -96,17 +96,13 @@ function integrate_pitch_gar_quadgk(
     # Split domain at trapped/passing boundary so Gauss-Kronrod resolves
     # the kink in leff = ell + n*q (circulating) → ell (trapped).
     bobmax_clip = clamp(bobmax, lambda_min, lambda_max)
-    segments = if lambda_min < bobmax_clip < lambda_max
-        (lambda_min, bobmax_clip, lambda_max)
-    else
-        (lambda_min, lambda_max)
-    end
-
-    # In-place quadgk! buffer; copy the result out so the returned vector is
-    # owned by the caller.
     buf = zeros(ComplexF64, nqty)
     kernel! = (out, λ) -> _pitch_gar_kernel_quadgk!(out, λ, params)
-    I, _ = quadgk!(kernel!, buf, segments...; atol=pitch_atol, rtol=pitch_rtol)
+    I, _ = if lambda_min < bobmax_clip < lambda_max
+        quadgk!(kernel!, buf, lambda_min, bobmax_clip, lambda_max; atol=pitch_atol, rtol=pitch_rtol)
+    else
+        quadgk!(kernel!, buf, lambda_min, lambda_max; atol=pitch_atol, rtol=pitch_rtol)
+    end
     return copy(I)
 end
 
@@ -195,15 +191,13 @@ function integrate_pitch_gar_quadgk_wt(
     lambda_max = last(fbnce.cache.x)
 
     bobmax_clip = clamp(bobmax, lambda_min, lambda_max)
-    segments = if lambda_min < bobmax_clip < lambda_max
-        (lambda_min, bobmax_clip, lambda_max)
-    else
-        (lambda_min, lambda_max)
-    end
-
     buf = zeros(ComplexF64, 2 * nqty)
     kernel! = (out, λ) -> _pitch_gar_kernel_quadgk_wt!(out, λ, params)
-    I, _ = quadgk!(kernel!, buf, segments...; atol=pitch_atol, rtol=pitch_rtol)
+    I, _ = if lambda_min < bobmax_clip < lambda_max
+        quadgk!(kernel!, buf, lambda_min, bobmax_clip, lambda_max; atol=pitch_atol, rtol=pitch_rtol)
+    else
+        quadgk!(kernel!, buf, lambda_min, lambda_max; atol=pitch_atol, rtol=pitch_rtol)
+    end
     return copy(I)
 end
 

src/KineticForces/Torque.jl

@@ -30,7 +30,7 @@ Imaginary component is proportional to the kinetic energy Im(T) = 2*n*dW_k.
 """
 function tpsi!(tpsi_var::Ref{ComplexF64}, psi::Float64, n::Int, l::Int,
               zi::Int, mi::Int, wdfac::Float64, divxfac::Float64,
-              electron::Bool, method::String, equil, intr::KineticForcesInternal,
+              electron::Bool, method::String, equil::Equilibrium.PlasmaEquilibrium, intr::KineticForcesInternal,
               kinetic_profiles::Equilibrium.KineticProfileSplines;
               op_wmats::Union{Nothing,Array{ComplexF64,3}}=nothing,
               rex_override::Union{Nothing,Float64}=nothing,
@@ -889,7 +889,7 @@ function compute_kinetic_matrices_at_psi!(
     ktmat::Array{ComplexF64,3},
     psi::Float64, n::Int, l::Int,
     zi::Int, mi::Int, wdfac::Float64, _divxfac::Float64,
-    electron::Bool, equil, intr::KineticForcesInternal,
+    electron::Bool, equil::Equilibrium.PlasmaEquilibrium, intr::KineticForcesInternal,
     kinetic_profiles::Equilibrium.KineticProfileSplines;
     nutype::String="harmonic", f0type::String="maxwellian", nufac::Float64=1.0,
     atol_xlmda::Float64=1e-9, rtol_xlmda::Float64=1e-6)

src/LocalStability/Ballooning.jl

@@ -1180,7 +1180,7 @@ function integrate_ballooning_ode(ode_coefficient_spline; theta_k::Float64=0.0)
         problem_left,
         DP5();
         reltol=TOLERANCE,
-        abstol=TOLERANCE^2,
+        abstol=TOLERANCE,
         dtmin=MINIMUM_STEP,
         adaptive=true,
         save_everystep=false,
@@ -1201,7 +1201,7 @@ function integrate_ballooning_ode(ode_coefficient_spline; theta_k::Float64=0.0)
         problem_right,
         DP5();
         reltol=TOLERANCE,
-        abstol=TOLERANCE^2,
+        abstol=TOLERANCE,
         dtmin=MINIMUM_STEP,
         adaptive=true,
         save_everystep=false,

src/Vacuum/Kernel2D.jl

@@ -639,12 +639,11 @@ according to equations (36)-(42) of Chance 1997. Replaces `green` from Fortran c
     gamma_prefactor::Float64=2 * sqrt(π) * gamma(0.5 - n),
     uselegacygreenfunction::Bool=false
 )
-
     x_obs2 = x_obs^2
     x_source2 = x_source^2
     x_minus2 = (x_obs - x_source)^2
     x_multiple = x_obs * x_source
-    ζ = (z_obs - z_source)
+    ζ = z_obs - z_source
     ζ2 = ζ^2
 
     ρ2 = x_minus2 + ζ2
@@ -655,49 +654,36 @@ according to equations (36)-(42) of Chance 1997. Replaces `green` from Fortran c
     R = sqrt(R2)
     R5 = R4 * R
 
-    # Argument of Legendre function 𝘴 [Chance Phys. Plasmas 1997 2161 eq. 42]
-    s = (x_obs2 + x_source2 + ζ2) / R2
+    S = x_obs2 + x_source2 + ζ2
+    a = x_obs2 - x_source2
+    D = a + ζ2            # x_obs2 - x_source2 + ζ2
+    E = ζ2 - a             # x_source2 - x_obs2 + ζ2
+    fourXmult = 4 * x_multiple
+    twoXobsD  = 2 * x_obs * D
+    xSourceE  = x_source * E
+
+    s = S / R2
 
-    # Legendre functions for
-    # P⁰ = p0, P¹ = p1, Pⁿ = pn, Pⁿ⁺¹ = pnp1
     legendre = acquire!(pool, Float64, n + 2)
     if uselegacygreenfunction
         Pn_minus_half_1997!(legendre, s, n)
     else
         Pn_minus_half_2007!(legendre, s, n)
     end
 
-    p0 = legendre[1]
-    p1 = legendre[2]
-    pnp1 = legendre[end]
-    pn = legendre[end-1]
+    p0, p1, pnp1, pn = @inbounds legendre[1], legendre[2], legendre[end], legendre[end-1]
 
-    # Green's function 2π𝒢ⁿ = G_n [Chance Phys. Plasmas 1997 2161 eq. 40]
     gg = gamma_prefactor / R
     G_n = gg * pn
+    grad_gg = gg / (2π * R4)
 
-    # Gradient factor [Chance Phys. Plasmas 1997 2161 eq. 44]
-    # NOTE: Paper has erroneous extra factor of 2π
-    grad_gg = gg / R4 / 2π
-
-    # Derivatives of Green's function [Chance Phys. Plasmas 1997 2161 eq. 36-38]
-    # ∂Gⁿ/∂X' using chain rule: ∂Gⁿ/∂X' = (∂Gⁿ/∂R)(∂R/∂X') + (∂Gⁿ/∂s)(∂s/∂X')
-    xterm1 = (n * (x_obs2 + x_source2 + ζ2) * (x_obs2 - x_source2 + ζ2) - x_source2*(x_source2-x_obs2+ζ2)) * pn
-    xterm2 = (2.0 * x_source * x_obs * (x_obs2-x_source2+ζ2)) * pnp1
-    dG_dX = grad_gg * (xterm1 + xterm2) / x_source
-
-    # ∂Gⁿ/∂Z' using chain rule
-    zterm1 = (2.0 * n + 1.0) * (x_obs2 + x_source2 + ζ2) * pn
-    zterm2 = 4.0 * x_multiple * pnp1
-    dG_dZ = grad_gg * (zterm1 + zterm2) * ζ
+    dG_dX = grad_gg * ( (n * S * D - x_source * xSourceE) * pn / x_source + twoXobsD * pnp1 )
+    dG_dZ = grad_gg * ((2n + 1) * S * pn + fourXmult * pnp1) * ζ
 
-    # Coupling term 𝒥 ∇'𝒢ⁿ∇'ℒ [Chance Phys. Plasmas 1997 2161 eq. 51]
-    # Jacobian factor from coordinate transformation
     coupling_n = -x_source * (dz_dtheta * dG_dX - dx_dtheta * dG_dZ)
 
-    # Special case for n=0: coupling_0 = 1/(2π) 𝒥 ∇'𝒢⁰∇'ℒ
-    dG_dX0_R5 = ((2.0 * x_obs * (x_obs2-x_source2+ζ2)) * p1 - x_source * (x_source2-x_obs2+ζ2) * p0)
-    dG_dZ0_R5 = ζ * ((x_obs2 + x_source2 + ζ2) * p0 + 4.0 * x_multiple * p1)
+    dG_dX0_R5 = twoXobsD * p1 - xSourceE * p0
+    dG_dZ0_R5 = ζ * (S * p0 + fourXmult * p1)
     coupling_0 = -x_source * (dz_dtheta * dG_dX0_R5 - dx_dtheta * dG_dZ0_R5) / R5
     return G_n, coupling_n, coupling_0
 end