From 03005bace5816d40af3fe98123b63d804782e4df Mon Sep 17 00:00:00 2001
From: wheeheee <104880306+wheeheee@users.noreply.github.com>
Date: Tue, 3 Feb 2026 14:21:57 +0800
Subject: [PATCH 1/4] change iteration order

generally more performant to iterate from left to right since Julia is column-major
---
 src/plan.jl | 22 ++++++++--------------
 1 file changed, 8 insertions(+), 14 deletions(-)

diff --git a/src/plan.jl b/src/plan.jl
index efa8333..2193934 100644
--- a/src/plan.jl
+++ b/src/plan.jl
@@ -145,10 +145,8 @@ function LinearAlgebra.mul!(y::AbstractArray{U,N}, p::FFTAPlan_cx{T,1}, x::Abstr
     end
     Rpre = CartesianIndices(size(x)[1:p.region[]-1])
     Rpost = CartesianIndices(size(x)[p.region[]+1:end])
-    for Ipre in Rpre
-        for Ipost in Rpost
-            @views fft!(y[Ipre,:,Ipost], x[Ipre,:,Ipost], 1, 1, p.dir, p.callgraph[1][1].type, p.callgraph[1], 1)
-        end
+    for Ipost in Rpost, Ipre in Rpre
+        @views fft!(y[Ipre,:,Ipost], x[Ipre,:,Ipost], 1, 1, p.dir, p.callgraph[1][1].type, p.callgraph[1], 1)
     end
 end
 
@@ -186,17 +184,13 @@ function _mul_loop(
     rows::Int,
     cols::Int
 )
-    for I1 in R1
-        for I2 in R2
-            for I3 in R3
-                for k in 1:cols
-                    @views fft!(y_tmp[:,k],  x[I1,:,I2,k,I3], 1, 1, p.dir, p.callgraph[1][1].type, p.callgraph[1], 1)
-                end
+    for I3 in R3, I2 in R2, I1 in R1
+        for k in 1:cols
+            @views fft!(y_tmp[:,k],  x[I1,:,I2,k,I3], 1, 1, p.dir, p.callgraph[1][1].type, p.callgraph[1], 1)
+        end
 
-                for k in 1:rows
-                    @views fft!(y[I1,k,I2,:,I3], y_tmp[k,:], 1, 1, p.dir, p.callgraph[2][1].type, p.callgraph[2], 1)
-                end
-            end
+        for k in 1:rows
+            @views fft!(y[I1,k,I2,:,I3], y_tmp[k,:], 1, 1, p.dir, p.callgraph[2][1].type, p.callgraph[2], 1)
         end
     end
 end

From fa3aa9006a32f3d2a75545d373702a8d3baa8452 Mon Sep 17 00:00:00 2001
From: wheeheee <104880306+wheeheee@users.noreply.github.com>
Date: Wed, 4 Feb 2026 00:45:20 +0800
Subject: [PATCH 2/4] another `_mul_loop!` for function barrier

---
 src/plan.jl | 13 +++++++++++--
 1 file changed, 11 insertions(+), 2 deletions(-)

diff --git a/src/plan.jl b/src/plan.jl
index 2193934..8a72045 100644
--- a/src/plan.jl
+++ b/src/plan.jl
@@ -145,6 +145,15 @@ function LinearAlgebra.mul!(y::AbstractArray{U,N}, p::FFTAPlan_cx{T,1}, x::Abstr
     end
     Rpre = CartesianIndices(size(x)[1:p.region[]-1])
     Rpost = CartesianIndices(size(x)[p.region[]+1:end])
+    _mul_loop!(y, x, Rpre, Rpost, p)
+end
+
+function _mul_loop!(
+    y::AbstractArray{U,N},
+    x::AbstractArray{T,N},
+    Rpre::CartesianIndices,
+    Rpost::CartesianIndices,
+    p::FFTAPlan_cx{T,1}) where {T,U,N}
     for Ipost in Rpost, Ipre in Rpre
         @views fft!(y[Ipre,:,Ipost], x[Ipre,:,Ipost], 1, 1, p.dir, p.callgraph[1][1].type, p.callgraph[1], 1)
     end
@@ -170,10 +179,10 @@ function LinearAlgebra.mul!(y::AbstractArray{U,N}, p::FFTAPlan_cx{T,2}, x::Abstr
     # Introduce function barrier here since the variables used in the loop ranges aren't inferred. This
     # is partly because the region field of the plan is abstractly typed but even if that wasn't the case,
     # it might be a bit tricky to construct the Rxs in an inferred way.
-    _mul_loop(y_tmp, y, x, p, R1, R2, R3, rows, cols)
+    _mul_loop!(y_tmp, y, x, p, R1, R2, R3, rows, cols)
 end
 
-function _mul_loop(
+function _mul_loop!(
     y_tmp::AbstractArray,
     y::AbstractArray,
     x::AbstractArray,

From e9410afa7f70c26f0675ffee1e9d2f063c1a804b Mon Sep 17 00:00:00 2001
From: wheeheee <104880306+wheeheee@users.noreply.github.com>
Date: Wed, 4 Feb 2026 00:45:41 +0800
Subject: [PATCH 3/4] add some common files to .gitignore

---
 .gitignore | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/.gitignore b/.gitignore
index 0708ed4..d32b7cf 100644
--- a/.gitignore
+++ b/.gitignore
@@ -1,2 +1,4 @@
-/Manifest.toml
-docs/build
\ No newline at end of file
+Manifest.toml
+Manifest-v1*.toml
+docs/build
+settings.json
\ No newline at end of file

From de3eb54c65c5b97dd9288f64162adb40c2a2d67f Mon Sep 17 00:00:00 2001
From: wheeheee <104880306+wheeheee@users.noreply.github.com>
Date: Wed, 4 Feb 2026 09:14:04 +0800
Subject: [PATCH 4/4] Revert "add some common files to .gitignore"

This reverts commit e9410afa7f70c26f0675ffee1e9d2f063c1a804b.
---
 .gitignore | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

diff --git a/.gitignore b/.gitignore
index d32b7cf..0708ed4 100644
--- a/.gitignore
+++ b/.gitignore
@@ -1,4 +1,2 @@
-Manifest.toml
-Manifest-v1*.toml
-docs/build
-settings.json
\ No newline at end of file
+/Manifest.toml
+docs/build
\ No newline at end of file