From dd6beb0a4a118212f3b9045ee480eb0b5c05a4cb Mon Sep 17 00:00:00 2001
From: Cheng <git@zcbenz.com>
Date: Thu, 24 Jul 2025 15:29:08 +0900
Subject: [PATCH 1/4] Detect ccache

---
 CMakeLists.txt | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 9e67e4bf2..1ed41aea5 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -41,6 +41,7 @@ option(MLX_BUILD_GGUF "Include support for GGUF format" ON)
 option(MLX_BUILD_SAFETENSORS "Include support for safetensors format" ON)
 option(MLX_BUILD_BLAS_FROM_SOURCE "Build OpenBLAS from source code" OFF)
 option(MLX_METAL_JIT "Use JIT compilation for Metal kernels" OFF)
+option(MLX_USE_CCACHE "Use CCache for compilation cache when available" ON)
 option(BUILD_SHARED_LIBS "Build mlx as a shared library" OFF)
 
 # --------------------- Processor tests -------------------------
@@ -68,6 +69,15 @@ else()
   set(MLX_BUILD_METAL OFF)
 endif()
 
+if(MLX_USE_CCACHE)
+  find_program(CCACHE_PROGRAM ccache)
+  if(CCACHE_PROGRAM)
+    set(CMAKE_C_COMPILER_LAUNCHER "${CCACHE_PROGRAM}")
+    set(CMAKE_CXX_COMPILER_LAUNCHER "${CCACHE_PROGRAM}")
+    set(CMAKE_CUDA_COMPILER_LAUNCHER "${CCACHE_PROGRAM}")
+  endif()
+endif()
+
 # ----------------------------- Lib -----------------------------
 
 include(FetchContent)

From 1ae2f817bfe15f6301e49256c60ae267223fcaff Mon Sep 17 00:00:00 2001
From: Cheng <git@zcbenz.com>
Date: Thu, 24 Jul 2025 15:39:09 +0900
Subject: [PATCH 2/4] Use ccache in CI

---
 .circleci/config.yml | 24 ++++++++++++++++++++++--
 setup.py             |  3 +++
 2 files changed, 25 insertions(+), 2 deletions(-)

diff --git a/.circleci/config.yml b/.circleci/config.yml
index f33e80f0c..fb2d50e0d 100644
--- a/.circleci/config.yml
+++ b/.circleci/config.yml
@@ -212,22 +212,42 @@ jobs:
       resource_class: gpu.nvidia.small.gen2
     steps:
       - checkout
+      - restore_cache:
+          keys:
+            - mlx-v1-cuda-{{ arch }}-
       - run:
-          name: Install Python package
+          name: Install dependencies
           command: |
             sudo apt-get update
             sudo apt-get install libcudnn9-dev-cuda-12
             sudo apt-get install libblas-dev liblapack-dev liblapacke-dev
+            curl -sL https://github.com/ccache/ccache/releases/download/v4.11.3/ccache-4.11.3-linux-x86_64.tar.xz | tar xJf -
+            sudo mv ccache-4.11.3-linux-x86_64/ccache /usr/bin/ccache
+            rm -rf ccache-4.11.3-linux-x86_64
+      - run:
+          name: Install Python package
+          command: |
             python3 -m venv env
             source env/bin/activate
             CMAKE_ARGS="-DMLX_BUILD_CUDA=ON -DCMAKE_CUDA_COMPILER=`which nvcc`" \
-              pip install -e ".[dev]"
+              pip install -e ".[dev]" -v
       - run:
           name: Run Python tests
           command: |
             source env/bin/activate
             LOW_MEMORY=1 DEVICE=cpu python -m unittest discover python/tests -v
             LOW_MEMORY=1 DEVICE=gpu python -m tests discover python/tests -v
+      - run:
+          name: CCache report
+          command: |
+            ccache --show-stats
+            ccache --zero-stats
+            ccache --max-size 400MB
+            ccache --cleanup
+      - save_cache:
+          key: mlx-v1-cuda-{{ arch }}-{{ epoch }}
+          paths:
+            - /home/circleci/.cache/ccache
 
   build_release:
     parameters:
diff --git a/setup.py b/setup.py
index 8a5517112..87159a0aa 100644
--- a/setup.py
+++ b/setup.py
@@ -113,6 +113,9 @@ class CMakeBuild(build_ext):
         if "CMAKE_BUILD_PARALLEL_LEVEL" not in os.environ:
             build_args += [f"-j{os.cpu_count()}"]
 
+        # Avoid cache miss when building from temporary dirs.
+        os.environ["CCACHE_BASEDIR"] = os.path.abspath(self.build_temp)
+
         subprocess.run(
             ["cmake", ext.sourcedir, *cmake_args], cwd=build_temp, check=True
         )

From f1394d3365a3e880d4a740d70a2f51b5e8a7eaac Mon Sep 17 00:00:00 2001
From: Cheng <git@zcbenz.com>
Date: Fri, 25 Jul 2025 08:15:03 +0900
Subject: [PATCH 3/4] Separate cache for different images

---
 .circleci/config.yml | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/.circleci/config.yml b/.circleci/config.yml
index fb2d50e0d..4cb35e00c 100644
--- a/.circleci/config.yml
+++ b/.circleci/config.yml
@@ -214,7 +214,7 @@ jobs:
       - checkout
       - restore_cache:
           keys:
-            - mlx-v1-cuda-{{ arch }}-
+            - cuda-<< parameters.image_date >>-{{ arch }}-
       - run:
           name: Install dependencies
           command: |
@@ -245,7 +245,7 @@ jobs:
             ccache --max-size 400MB
             ccache --cleanup
       - save_cache:
-          key: mlx-v1-cuda-{{ arch }}-{{ epoch }}
+          key: cuda-<< parameters.image_date >>-{{ arch }}-{{ epoch }}
           paths:
             - /home/circleci/.cache/ccache
 

From 5a1a9ee1cd20756a5dd7e328e01304dc1f192e1f Mon Sep 17 00:00:00 2001
From: Cheng <git@zcbenz.com>
Date: Fri, 25 Jul 2025 08:25:36 +0900
Subject: [PATCH 4/4] Test both 12.2 and 12.9 for PRs

---
 .circleci/config.yml | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/.circleci/config.yml b/.circleci/config.yml
index 4cb35e00c..0fea8804c 100644
--- a/.circleci/config.yml
+++ b/.circleci/config.yml
@@ -575,6 +575,9 @@ workflows:
           requires: [ hold ]
       - cuda_build_and_test:
           requires: [ hold ]
+          matrix:
+            parameters:
+              image_date: ["2023.11.1", "2025.05.1"]
   nightly_build:
     when:
       and: