From 8a645b2ba0ce84611927e553c0be72ebafdc341b Mon Sep 17 00:00:00 2001
From: Christian Guinard <28689358+christiangnrd@users.noreply.github.com>
Date: Mon, 24 Mar 2025 14:54:05 -0300
Subject: [PATCH 1/3] Actually test with 2 threads

---
 .github/workflows/CI-CPU.yml | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/.github/workflows/CI-CPU.yml b/.github/workflows/CI-CPU.yml
index 4efc362..e163592 100644
--- a/.github/workflows/CI-CPU.yml
+++ b/.github/workflows/CI-CPU.yml
@@ -49,6 +49,8 @@ jobs:
       - uses: julia-actions/cache@v2
       - uses: julia-actions/julia-buildpkg@v1
       - uses: julia-actions/julia-runtest@v1
+        env:
+          JULIA_NUM_THREADS: ${{ matrix.env.JULIA_NUM_THREADS }}
   docs:
     name: Documentation
     runs-on: ubuntu-latest
@@ -58,4 +60,4 @@ jobs:
       - uses: julia-actions/julia-docdeploy@latest
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-          DOCUMENTER_KEY: ${{ secrets.DOCUMENTER_KEY }}
\ No newline at end of file
+          DOCUMENTER_KEY: ${{ secrets.DOCUMENTER_KEY }}

From f58f7bbc311b348f552d7f80fff3269aaa430ed5 Mon Sep 17 00:00:00 2001
From: Christian Guinard <28689358+christiangnrd@users.noreply.github.com>
Date: Mon, 24 Mar 2025 14:55:02 -0300
Subject: [PATCH 2/3] only macos-13 runners support macos x64 and officially
 test macos aarch64

---
 .github/workflows/CI-CPU.yml | 12 ++++++++++++
 1 file changed, 12 insertions(+)

diff --git a/.github/workflows/CI-CPU.yml b/.github/workflows/CI-CPU.yml
index e163592..a9848a3 100644
--- a/.github/workflows/CI-CPU.yml
+++ b/.github/workflows/CI-CPU.yml
@@ -31,15 +31,27 @@ jobs:
           - ubuntu-latest
           - macOS-latest
           - windows-latest
+          - macOS-13
         arch:
           - x64
           - x86
+          - aarch64
         env:
           - JULIA_NUM_THREADS: 1
           - JULIA_NUM_THREADS: 2
         exclude:
+          - os: macOS-13
+            arch: x86
+          - os: macOS-13
+            arch: aarch64
           - os: macOS-latest
             arch: x86
+          - os: macOS-latest
+            arch: x64
+          - os: ubuntu-latest
+            arch: aarch64
+          - os: windows-latest
+            arch: aarch64
     steps:
       - uses: actions/checkout@v4
       - uses: julia-actions/setup-julia@v2

From 6b2e348f3dba25a79003eaf731beb29d22df40e4 Mon Sep 17 00:00:00 2001
From: Christian Guinard <28689358+christiangnrd@users.noreply.github.com>
Date: Mon, 24 Mar 2025 18:17:41 -0300
Subject: [PATCH 3/3] Fix multi-thread CPU mapreduce

---
 src/reduce/reduce.jl | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/src/reduce/reduce.jl b/src/reduce/reduce.jl
index 88e535b..06d775b 100644
--- a/src/reduce/reduce.jl
+++ b/src/reduce/reduce.jl
@@ -274,12 +274,12 @@ function _mapreduce_impl(
             if num_tasks <= 1
                 return Base.mapreduce(f, op, src; init=init)
             end
-            return OMT.tmapreduce(
-                f, op, src, init=init,
+            return op(init, OMT.tmapreduce(
+                f, op, src; init=neutral,
                 scheduler=scheduler,
                 outputtype=typeof(init),
                 nchunks=num_tasks,
-            )
+            ))
         else
             # FIXME: waiting on OhMyThreads.jl for n-dimensional reduction
             return Base.mapreduce(f, op, src; init=init, dims=dims)