remove use of cuda pool, use cuda free async

Co-authored-by: Joseph Heck <j_heck@apple.com>

.circleci/config.yml

@@ -26,9 +26,9 @@ jobs:
           name: Install
           command: |
             xcodebuild -downloadComponent MetalToolchain
-            brew install python@3.9
+            brew install python@3.10
             brew install doxygen
-            python3.9 -m venv env
+            python3.10 -m venv env
             source env/bin/activate
             pip install --upgrade pip
             pip install --upgrade cmake
@@ -140,7 +140,7 @@ jobs:
       - run:
           name: Install Python package
           command: |
-            uv venv --python 3.9
+            uv venv --python 3.10
             uv pip install \
               nanobind==2.4.0 \
               cmake \
@@ -273,7 +273,7 @@ jobs:
     parameters:
       python_version:
         type: string
-        default: "3.9"
+        default: "3.10"
       xcode_version:
         type: string
         default: "26.0.0"
@@ -328,7 +328,7 @@ jobs:
             << parameters.build_env >> MLX_BUILD_STAGE=1 python -m build -w
       - when:
           condition:
-            equal: ["3.9", << parameters.python_version >>]
+            equal: ["3.10", << parameters.python_version >>]
           steps:
             - run:
                 name: Build common package
@@ -351,7 +351,7 @@ jobs:
     parameters:
       python_version:
         type: string
-        default: "3.9"
+        default: "3.10"
       build_env:
         type: string
         default: ""
@@ -387,7 +387,7 @@ jobs:
             bash python/scripts/repair_linux.sh
       - when:
           condition:
-            equal: ["3.9", << parameters.python_version >>]
+            equal: ["3.10", << parameters.python_version >>]
           steps:
             - run:
                 name: Build common package
@@ -484,7 +484,7 @@ workflows:
               ignore: /.*/
           matrix:
             parameters:
-              python_version: ["3.9", "3.10", "3.11", "3.12", "3.13"]
+              python_version: ["3.10", "3.11", "3.12", "3.13", "3.14"]
               macosx_deployment_target: ["13.5", "14.0", "15.0"]
               build_env: ["PYPI_RELEASE=1"]
               xcode_version: ["26.0.0"]
@@ -503,7 +503,7 @@ workflows:
               ignore: /.*/
           matrix:
             parameters:
-              python_version: ["3.9", "3.10", "3.11", "3.12", "3.13"]
+              python_version: ["3.10", "3.11", "3.12", "3.13", "3.14"]
               build_env: ["PYPI_RELEASE=1"]
       - build_cuda_release:
           filters:
@@ -546,13 +546,13 @@ workflows:
       - build_release:
           matrix:
             parameters:
-              python_version: ["3.9", "3.10", "3.11", "3.12", "3.13"]
+              python_version: ["3.10", "3.11", "3.12", "3.13", "3.14"]
               macosx_deployment_target: ["13.5", "14.0", "15.0"]
               xcode_version: ["26.0.0"]
       - build_linux_release:
           matrix:
             parameters:
-              python_version: ["3.9", "3.10", "3.11", "3.12", "3.13"]
+              python_version: ["3.10", "3.11", "3.12", "3.13", "3.14"]
       - build_cuda_release
 
   build_dev_release:
@@ -564,14 +564,14 @@ workflows:
       - build_release:
           matrix:
             parameters:
-              python_version: ["3.9", "3.10", "3.11", "3.12", "3.13"]
+              python_version: ["3.10", "3.11", "3.12", "3.13", "3.14"]
               macosx_deployment_target: ["13.5", "14.0", "15.0"]
               build_env: ["DEV_RELEASE=1"]
               xcode_version: ["26.0.0"]
       - build_linux_release:
           matrix:
             parameters:
-              python_version: ["3.9", "3.10", "3.11", "3.12", "3.13"]
+              python_version: ["3.10", "3.11", "3.12", "3.13", "3.14"]
               build_env: ["DEV_RELEASE=1"]
       - build_cuda_release:
           matrix:

.github/actions/build-cuda-release/action.yml

@@ -0,0 +1,24 @@
+name: 'Build CUDA wheel'
+description: 'Build CUDA wheel'
+
+inputs:
+  nvcc-location:
+    description: 'Location of nvcc compiler'
+    required: true
+
+runs:
+  using: "composite"
+  steps:
+    - name: Build package
+      shell: bash
+      env:
+        MLX_BUILD_STAGE: 2
+        CMAKE_ARGS: -DMLX_BUILD_CUDA=ON -DCMAKE_CUDA_COMPILER=${{ inputs.nvcc-location }}
+      run: |
+        pip install auditwheel build patchelf setuptools
+        python setup.py clean --all
+        python -m build -w
+
+        if [ -f "python/scripts/repair_cuda.sh" ]; then
+          bash python/scripts/repair_cuda.sh
+        fi

.github/actions/build-cuda/action.yml

@@ -0,0 +1,68 @@
+name: 'Build and Test with CUDA'
+description: 'Build and test MLX with CUDA'
+
+inputs:
+  build-type:
+    description: 'Build type (debug, release)'
+    required: false
+    default: 'debug'
+  run-tests:
+    description: 'Whether to run tests'
+    required: false
+    default: 'true'
+  nvcc-location:
+    description: 'Location of nvcc compiler'
+    required: true
+    default: '/usr/local/cuda-12.9/bin/nvcc'
+    # this value is dependent on the CUDA tools installed in the setup-linux workflow
+
+runs:
+  using: "composite"
+  steps:
+    - name: Install Python package
+      shell: bash
+      env:
+        DEBUG: 1
+        CMAKE_ARGS: -DMLX_BUILD_CUDA=ON -DCMAKE_COMPILE_WARNING_AS_ERROR=ON -DCMAKE_CUDA_COMPILER=${{ inputs.nvcc-location }}
+      run: pip install -e ".[dev]" -v
+
+    - name: Check if build actually worked
+      shell: bash
+      run: python -c "import mlx.core"
+
+    - name: Run Python tests - CPU
+      if: inputs.run-tests == 'true'
+      shell: bash
+      env:
+        LOW_MEMORY: 1
+        DEVICE: cpu
+      run: python -m unittest discover python/tests -v
+
+    - name: Run Python tests - GPU
+      if: inputs.run-tests == 'true'
+      shell: bash
+      env:
+        LOW_MEMORY: 1
+        DEVICE: gpu
+      run: python -m tests discover python/tests -v
+
+    - name: Build CPP only
+      if: inputs.build-type == 'debug'
+      shell: bash
+      run: |
+        cmake . -B build \
+          -DMLX_BUILD_CUDA=ON \
+          -DCMAKE_CUDA_COMPILER=${{ inputs.nvcc-location }} \
+          -DCMAKE_BUILD_TYPE=DEBUG
+        cmake --build build -j $(nproc)
+    
+    - name: Run CPP tests
+      if: ${{ inputs.build-type == 'debug' && inputs.run-tests == 'true' }}
+      shell: bash
+      run: ./build/tests/tests -sfe="*fft_tests.cpp,*linalg_tests.cpp"
+
+    - name: Build Python package
+      if: inputs.build-type == 'release'
+      uses: ./.github/actions/build-cuda-release
+      with:
+        nvcc-location: ${{ inputs.nvcc-location }}

.github/actions/build-docs/action.yml

@@ -0,0 +1,38 @@
+name: 'Build Documentation'
+description: 'Build documentation on a mac'
+
+runs:
+  using: "composite"
+  steps:
+    - name: Setup machine
+      uses: ./.github/actions/setup-macos
+
+    - name: Install dependencies
+      shell: sh
+      run: |
+        brew install doxygen
+        uv pip install --upgrade pip cmake
+        uv pip install -r docs/requirements.txt
+        uv pip install . -v
+  
+    - name: Build documentation
+      shell: bash
+      run: |
+        source .venv/bin/activate
+        cd docs
+        doxygen
+        make html O=-W
+    
+    - name: Create artifact tar
+      shell: sh
+      run: tar -cf artifact.tar --cd docs/build/html -L .
+
+    # Do it manually because upload-pages-artifact requires gtar
+    - name: Upload artifact
+      id: upload-artifact
+      uses: actions/upload-artifact@v5
+      with:
+        name: github-pages
+        path: artifact.tar
+        retention-days: 1
+        if-no-files-found: error

.github/actions/build-linux/action.yml

@@ -0,0 +1,78 @@
+name: 'Build and Test on Linux'
+description: 'Build and test MLX on Linux'
+
+inputs:
+  build-type:
+    description: 'Build type'
+    required: false
+    default: 'debug'
+    type: choice
+    options:
+      - debug
+      - release
+  run-tests:
+    description: 'Whether to run tests'
+    required: false
+    default: 'true'
+    type: boolean
+
+runs:
+  using: "composite"
+  steps:
+    - name: Set DEBUG
+      shell: sh
+      if: inputs.build-type == 'debug'
+      run: echo "DEBUG=1" >> $GITHUB_ENV
+
+    - name: Install Python package
+      shell: sh
+      env:
+        CMAKE_ARGS: "-DCMAKE_COMPILE_WARNING_AS_ERROR=ON"
+      run: pip install -e ".[dev]" -v
+    
+    - name: Generate package stubs
+      shell: sh
+      run: |
+        pip install typing_extensions
+        python setup.py generate_stubs
+    
+    - name: Run Python tests
+      if: inputs.run-tests == 'true'
+      shell: bash
+      run: |
+        python -m unittest discover python/tests -v
+        mpirun --bind-to none --allow-run-as-root -host localhost:8 -np 8 python python/tests/mpi_test_distributed.py
+        mlx.launch --verbose -n 8 python/tests/ring_test_distributed.py -v 2> >(tee -a stderr.log >&2)
+        if grep -Fq '[WARN]' stderr.log ; then
+          grep -F '[WARN]' stderr.log
+          echo "Distributed ring test failed";
+          exit 1;
+        fi
+    
+    - name: Build CPP only
+      if: inputs.build-type == 'debug'
+      shell: bash
+      run: |
+        mkdir -p build && cd build
+        cmake .. -DMLX_BUILD_METAL=OFF -DCMAKE_BUILD_TYPE=DEBUG
+        make -j $(nproc)
+    
+    - name: Run CPP tests
+      if: ${{ inputs.build-type == 'debug' && inputs.run-tests == 'true' }}
+      shell: sh
+      run: ./build/tests/tests
+    
+    - name: Build Python package
+      if: inputs.build-type == 'release'
+      shell: bash
+      run: |
+        pip install auditwheel patchelf build
+        python setup.py clean --all
+        MLX_BUILD_STAGE=1 python -m build -w
+        if [ -f "python/scripts/repair_linux.sh" ]; then
+          bash python/scripts/repair_linux.sh
+        fi
+
+        python setup.py clean --all
+        MLX_BUILD_STAGE=2 python -m build -w
+        auditwheel repair dist/mlx_cpu*.whl --plat manylinux_2_35_x86_64

.github/actions/build-macos-release/action.yml

@@ -0,0 +1,22 @@
+name: 'Build macOS release'
+description: 'Build MLX releases macOS'
+
+inputs:
+  macos-target:
+    description: 'macOS build target'
+    required: false
+    default: '15.0'
+
+runs:
+  using: "composite"
+  steps:
+    - name: Build Python package(s)
+      shell: bash
+      env:
+        MACOSX_DEPLOYMENT_TARGET: ${{ inputs.macos-target }}
+      run: |
+        uv pip install build
+        uv run --no-project setup.py clean --all
+        MLX_BUILD_STAGE=1 uv run -m build -w
+        uv run --no-project setup.py clean --all
+        MLX_BUILD_STAGE=2 uv run -m build -w

.github/actions/build-macos/action.yml

@@ -0,0 +1,124 @@
+name: 'Build and Test on macOS'
+description: 'Build and test MLX on macOS'
+
+inputs:
+  build-type:
+    description: 'Build type (debug, release)'
+    required: false
+    default: 'debug'
+    type: choice
+    options:
+      - debug
+      - release
+  run-tests:
+    description: 'Whether to run tests'
+    required: false
+    default: 'true'
+  build-jit:
+    description: 'Whether to build with JIT'
+    required: false
+    default: 'true'
+
+runs:
+  using: "composite"
+  steps:
+    - name: Install dependencies
+      shell: sh
+      env:
+        DEBUG: 1
+        DEV_RELEASE: 1
+      run: |
+        uv pip install --upgrade pip cmake setuptools
+        uv pip install nanobind==2.4.0 \
+          numpy torch tensorflow unittest-xml-reporting
+        uv pip install -e . -v
+
+    - name: Generate package stubs
+      shell: bash
+      run: |
+        uv pip install typing_extensions
+        uv run --no-project setup.py generate_stubs
+
+    - name: Run Python tests
+      if: inputs.run-tests == 'true'
+      shell: bash
+      env:
+        LOW_MEMORY: 1
+      run: |
+        DEVICE=cpu uv run -m xmlrunner discover -v python/tests -o test-results/cpu
+        DEVICE=gpu METAL_DEVICE_WRAPPER_TYPE=1 METAL_DEBUG_ERROR_MODE=0 uv run -m xmlrunner discover -v python/tests -o test-results/gpu
+        mpirun --bind-to none -host localhost:8 -np 8 -x DYLD_LIBRARY_PATH=/opt/homebrew/lib/ python python/tests/mpi_test_distributed.py
+        mlx.launch --verbose -n 8 python/tests/ring_test_distributed.py -v 2> >(tee -a stderr.log >&2)
+        if $(grep "\[WARN\]" stderr.log); then echo "Distributed ring test failed"; exit 1; fi
+    
+    - name: Build example extension
+      if: inputs.run-tests == 'true'
+      shell: bash
+      run: |
+        cd examples/extensions
+        uv pip install -r requirements.txt
+        uv run --no-project setup.py build_ext --inplace
+        uv run --no-project test.py
+    
+    - name: Build CPP only
+      if: inputs.build-type == 'debug'
+      shell: bash
+      run: |
+        mkdir -p build
+        cd build
+        cmake ..
+        make -j $(sysctl -n hw.ncpu)
+    
+    - name: Run CPP tests
+      if: ${{ inputs.build-type == 'debug' && inputs.run-tests == 'true' }}
+      shell: bash
+      env:
+        DEVICE: gpu
+        METAL_DEVICE_WRAPPER_TYPE: 1
+        METAL_DEBUG_ERROR_MODE: 0
+      run: ./build/tests/tests
+    
+    - name: Build small binary with JIT
+      if: inputs.build-jit == 'true'
+      shell: bash
+      run: |
+        mkdir -p build
+        cd build
+        cmake .. -DCMAKE_BUILD_TYPE=MinSizeRel \
+          -DBUILD_SHARED_LIBS=ON \
+          -DMLX_BUILD_CPU=OFF \
+          -DMLX_BUILD_SAFETENSORS=OFF \
+          -DMLX_BUILD_GGUF=OFF \
+          -DMLX_METAL_JIT=ON
+        make -j $(sysctl -n hw.ncpu)
+    
+    - name: Run Python tests with JIT
+      if: ${{ inputs.build-jit == 'true' && inputs.run-tests == 'true' }}
+      shell: bash
+      env:
+        LOW_MEMORY: 1
+        DEVICE: gpu
+        METAL_DEVICE_WRAPPER_TYPE: 1
+        METAL_DEBUG_ERROR_MODE: 0
+      run: |
+        CMAKE_ARGS="-DMLX_METAL_JIT=ON" \
+          uv pip install -e . -v
+        uv run -m xmlrunner discover \
+            -v python/tests \
+            -o test-results/gpu_jit
+
+    - name: Build macOS 13 package
+      if: inputs.build-type == 'release'
+      uses: ./.github/actions/build-macos-release
+      with:
+        macos-target: 13.0
+    - name: Build macOS 14 package
+      if: inputs.build-type == 'release'
+      uses: ./.github/actions/build-macos-release
+      with:
+        macos-target: 14.0
+    - name: Build macOS 15 package
+      if: inputs.build-type == 'release'
+      uses: ./.github/actions/build-macos-release
+      with:
+        macos-target: 15.0

.github/actions/setup-linux/action.yml

@@ -0,0 +1,83 @@
+name: 'Setup Linux Environment'
+description: 'Install dependencies for Linux builds'
+
+inputs:
+  runner-type:
+    description: 'Whether to set this up as a linux or CUDA runner'
+    required: false
+    default: 'linux'
+    type: choice
+    options:
+      - linux
+      - cuda
+  python-version:
+    description: 'Version of python to set up'
+    required: false
+    default: '3.10'
+
+runs:
+  using: "composite"
+  steps:
+    - name: Free disk space
+      shell: sh
+      if: inputs.runner-type == 'linux'
+      run: sudo rm -rf "$AGENT_TOOLSDIRECTORY"
+
+    - name: Install common dependencies
+      env:
+        TZ: Etc/UTC
+      shell: bash
+      run: |
+        sudo apt-get update
+        sudo apt-get install -y libblas-dev liblapack-dev liblapacke-dev tzdata zip
+        sudo apt autoremove -y
+
+    - uses: actions/setup-python@v6
+      with:
+        python-version: ${{ inputs.python-version }}
+        cache: 'pip'
+
+    - name: setup python venv
+      shell: bash
+      run: |
+        python -m venv .venv
+        source .venv/bin/activate
+        echo PATH=$PATH >> $GITHUB_ENV
+        pip install --upgrade pip cmake
+
+    - name: Install MPI
+      if: inputs.runner-type == 'linux'
+      shell: bash
+      run: sudo apt-get install -y openmpi-bin openmpi-common libopenmpi-dev
+
+    - name: Network CUDA installation from packages
+      id: install-cuda
+      if: inputs.runner-type == 'cuda'
+      env:
+        TZ: Etc/UTC
+      shell: bash ## Specific to Ubuntu 22.04 & Architecture x86_64
+      run: |
+        wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-keyring_1.1-1_all.deb
+        sudo dpkg -i cuda-keyring_1.1-1_all.deb
+        sudo apt-get update
+        sudo apt-get install -y libcudnn9-dev-cuda-12 libnccl2 libnccl-dev cuda-toolkit-12-9
+      # Note: This installs CUDA 12.9, which is the latest supported by cuDNN 9.x and works with the NVidia 570 drivers
+      # cuda-toolkit by itself installs version 13 (+) and requires updated drives (580+), which require a reboot to function properly.
+      # Compatibility matrix: https://docs.nvidia.com/deeplearning/cudnn/backend/latest/reference/support-matrix.html
+      # This also drops `nvcc` into `/usr/local/cuda-12.9/bin/nvcc` - but it's *not* on the default PATH
+
+    - name: Package and Driver Report
+      if: inputs.runner-type == 'cuda'
+      shell: bash
+      run: |
+        sudo apt-get install -y ubuntu-drivers-common dkms
+        echo "NVIDIA Driver Packages Available:"
+        sudo ubuntu-drivers list --gpgpu
+        echo "NVIDIA Driver Version:"
+        cat /proc/driver/nvidia/version || echo "nvidia driver not found"
+        echo "Installed NVIDIA and CUDA packages:"
+        dpkg -l | egrep "cuda|nvidia" -i
+        echo "DKMS Status:"
+        dkms status || echo "dkms not found"
+        echo "NVIDIA-SMI Status:"
+        nvidia-smi || echo "nvidia-smi not found"

.github/actions/setup-macos/action.yml

@@ -0,0 +1,31 @@
+name: 'Setup macOS Environment'
+description: 'Install dependencies for macOS builds'
+
+inputs:
+  install-mpi:
+    description: 'Whether to install MPI'
+    required: false
+    default: 'true'
+    type: boolean
+  python-version:
+    description: 'Python version to use'
+    required: false
+    default: '3.10'
+
+runs:
+  using: "composite"
+  steps:
+    - name: Install Homebrew packages
+      shell: sh
+      if: inputs.install-mpi == 'true'
+      run: /opt/homebrew/bin/brew install openmpi
+    
+    - name: Verify MetalToolchain installed
+      shell: bash
+      run: xcodebuild -showComponent MetalToolchain
+    
+    - name: Setup uv
+      uses: astral-sh/setup-uv@v6
+      with:
+          python-version: ${{ inputs.python-version }}
+          activate-environment: true

.github/dependabot.yml

@@ -0,0 +1,6 @@
+version: 2
+updates:
+  - package-ecosystem: "github-actions"
+    directory: "/"
+    schedule:
+      interval: "weekly"

.github/workflows/documentation.yml

@@ -0,0 +1,28 @@
+name: Documentation
+
+on:
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+jobs:
+  build:
+    runs-on: [self-hosted, macos]
+    steps:
+      - uses: actions/checkout@v5
+      - uses: ./.github/actions/build-docs
+      
+  deploy:
+    needs: build
+    permissions:
+      pages: write
+      id-token: write
+    runs-on: ubuntu-latest
+    environment:
+      name: github-pages
+      url: ${{ steps.deployment.outputs.page_url }}
+    steps:
+      - name: Deploy to GitHub Pages
+        id: deployment
+        uses: actions/deploy-pages@v4

.github/workflows/nightly.yml

@@ -0,0 +1,93 @@
+name: Nightly Build
+
+on:
+  schedule:
+    - cron: 33 6 * * 1-5
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+jobs:
+  build_linux_release:
+    strategy:
+      fail-fast: false
+      matrix:
+        python_version: ["3.10", "3.14"]
+    runs-on: ubuntu-22.04
+    steps:
+      - uses: actions/checkout@v5
+      - uses: ./.github/actions/setup-linux
+      - uses: ./.github/actions/build-linux
+        with:
+          build-type: release
+          run-tests: false
+      - name: Upload mlx artifacts
+        uses: actions/upload-artifact@v5
+        with:
+          name: linux-wheels-${{ matrix.python_version }}
+          path: wheelhouse/mlx-*.whl
+          retention-days: 7
+      - name: Upload mlx-cpu artifacts
+        if: matrix.python_version == '3.10'
+        uses: actions/upload-artifact@v5
+        with:
+          name: mlx-cpu
+          path: wheelhouse/mlx_cpu-*.whl
+          retention-days: 7
+  
+  build_linux_with_tests:
+    strategy:
+      fail-fast: false
+      matrix:
+        python_version: ["3.10", "3.11", "3.12", "3.13", "3.14"]
+    runs-on: ubuntu-22.04
+    steps:
+      - uses: actions/checkout@v5
+      - uses: ./.github/actions/setup-linux
+        with:
+          python-version: ${{ matrix.python_version }}
+      - uses: ./.github/actions/build-linux
+
+  build_mac_release:
+    strategy:
+      matrix:
+        python-version: ["3.10", "3.13"]
+        # TODO: 3.14 had issues finding a compatible tensorflow
+    env:
+      MACOSX_DEPLOYMENT_TARGET: "15.0"
+    runs-on: [self-hosted, macos]
+    steps:
+      - uses: actions/checkout@v5
+      - uses: ./.github/actions/setup-macos
+        with:
+          python-version: ${{ matrix.python-version }}
+      - uses: ./.github/actions/build-macos
+
+  build_cuda_with_tests:
+    runs-on: gpu-t4-4-core
+    steps:
+      - uses: actions/checkout@v5
+      - uses: ./.github/actions/setup-linux
+        with:
+          runner-type: 'cuda'
+      - uses: ./.github/actions/build-cuda
+
+  build_cuda_release:
+    runs-on: ubuntu-22-large
+    steps:
+      - uses: actions/checkout@v5
+      - uses: ./.github/actions/setup-linux
+        with:
+          runner-type: 'cuda'
+      - name: Build Python package
+        uses: ./.github/actions/build-cuda-release
+        with:
+          nvcc-location: '/usr/local/cuda-12.9/bin/nvcc'
+      - name: Upload artifacts
+        uses: actions/upload-artifact@v5
+        with:
+          name: mlx-cuda
+          path: wheelhouse/mlx_cuda-*.whl
+          retention-days: 7
+

.github/workflows/pull_request.yml

@@ -1,20 +1,46 @@
-on:
-  pull_request:
-    branches:
-      - main
+name: Build and Test
+
+on: pull_request  
+
+permissions:
+  contents: read
 
 jobs:
   check_lint:
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-22.04
     steps:
-      - uses: actions/checkout@v4
-      - uses: actions/setup-python@v4
+      - uses: actions/checkout@v5
+      - uses: ./.github/actions/setup-linux
+      - uses: pre-commit/action@v3.0.1
+
+  linux_build_and_test:
+    runs-on: ubuntu-22.04
+    steps:
+      - uses: actions/checkout@v5
+      - uses: ./.github/actions/setup-linux
+      - uses: ./.github/actions/build-linux
+
+  mac_build_and_test:
+    runs-on: [self-hosted, macos]
+    needs: check_lint
+    steps:
+      - uses: actions/checkout@v5
+      - uses: ./.github/actions/setup-macos
+      - uses: ./.github/actions/build-macos
+
+  cuda_build_and_test:
+    runs-on: gpu-t4-4-core
+    needs: check_lint
+    steps:
+      - uses: actions/checkout@v5
+      - uses: ./.github/actions/setup-linux
         with:
-          python-version: 3.8
-      - name: Install dependencies
-        run: |
-          python -m pip install --upgrade pip
-          pip install pre-commit black isort clang-format
-      - name: Run lint
-        run: |
-          pre-commit run --all-files
+          runner-type: 'cuda'
+      - uses: ./.github/actions/build-cuda
+
+  build_documentation:
+    runs-on: [self-hosted, macos]
+    needs: check_lint
+    steps:
+      - uses: actions/checkout@v5
+      - uses: ./.github/actions/build-docs

.github/workflows/release.yml

@@ -0,0 +1,188 @@
+name: PyPI Release
+
+on:
+  push:
+    tags:
+      - 'v*'
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+jobs:
+  build_documentation:
+    runs-on: [self-hosted, macos]
+    steps:
+      - uses: actions/checkout@v5
+      - uses: ./.github/actions/build-docs
+    
+  deploy_documentation:
+    needs: build_documentation
+    permissions:
+      pages: write
+      id-token: write
+    runs-on: ubuntu-latest
+    environment:
+      name: github-pages
+      url: ${{ steps.deployment.outputs.page_url }}
+    steps:
+      - name: Deploy to GitHub Pages
+        id: deployment
+        uses: actions/deploy-pages@v4
+
+  build_linux_release:
+    strategy:
+      matrix:
+        python_version: ["3.10", "3.11", "3.12", "3.13", "3.14"]
+    runs-on: ubuntu-22.04
+    env:
+      PYPI_RELEASE: 1
+    steps:
+      - uses: actions/checkout@v5
+      - uses: ./.github/actions/setup-linux
+        with:
+          python-version: ${{ matrix.python_version }}
+      - uses: ./.github/actions/build-linux
+        with:
+          build-type: release
+          run-tests: false
+      - name: Upload MLX artifacts
+        uses: actions/upload-artifact@v5
+        with:
+          name: linux-wheels-${{ matrix.python_version }}
+          path: wheelhouse/mlx-*.whl
+      - name: Upload CPU artifacts
+        if: matrix.python_version == '3.10'
+        uses: actions/upload-artifact@v5
+        with:
+          name: mlx-cpu
+          path: wheelhouse/mlx_cpu-*.whl
+  
+  build_mac_release:
+    strategy:
+      matrix:
+        python-version: ["3.10", "3.11", "3.12", "3.13"]
+        # TODO: 3.14 had issues finding a compatible tensorflow
+    runs-on: [self-hosted, macos]
+    env:
+      PYPI_RELEASE: 1
+    steps:
+      - uses: actions/checkout@v5
+      - uses: ./.github/actions/setup-macos
+        with:
+          python-version: ${{ matrix.python-version }}
+      - uses: ./.github/actions/build-macos
+        with:
+          build-type: release
+      - name: Upload MLX artifacts
+        uses: actions/upload-artifact@v5
+        with:
+          name: mac-wheels-${{ matrix.python-version }}
+          path: dist/mlx-*.whl
+      - name: Upload Metal artifacts
+        if: matrix.python-version == '3.10'
+        uses: actions/upload-artifact@v5
+        with:
+          name: mlx-metal
+          path: dist/mlx_metal-*.whl
+
+  build_cuda_release:
+    runs-on: ubuntu-22-large
+    env:
+      PYPI_RELEASE: 1
+    steps:
+      - uses: actions/checkout@v5
+      - uses: ./.github/actions/setup-linux
+        with:
+          runner-type: 'cuda'
+      - name: Build Python package
+        uses: ./.github/actions/build-cuda-release
+        with:
+          nvcc-location: '/usr/local/cuda-12.9/bin/nvcc'
+      - name: Upload artifacts
+        uses: actions/upload-artifact@v5
+        with:
+          name: mlx-cuda
+          path: wheelhouse/mlx_cuda-*.whl
+
+  pypi-publish:
+    name: Upload release to PyPI
+    runs-on: ubuntu-latest
+    needs: [build_linux_release, build_mac_release]
+    permissions:
+      id-token: write
+    environment:
+      name: pypi
+      url: https://pypi.org/p/mlx
+    steps:
+      - uses: actions/download-artifact@v6
+        with:
+          pattern: linux-wheels-*
+          merge-multiples: true
+          path: artifacts
+      - uses: actions/download-artifact@v6
+        with:
+          pattern: mac-wheels-*
+          merge-multiples: true
+          path: artifacts
+      - name: Display structure of downloaded files
+        run: ls -R artifacts
+      # - name: Publish package distributions to PyPI
+      #  uses: pypa/gh-action-pypi-publish@release/v1
+  
+  pypi-publish-cuda:
+    name: Upload CUDA release to PyPI
+    runs-on: ubuntu-latest
+    needs: build_cuda_release
+    permissions:
+      id-token: write
+    environment:
+      name: pypi
+      url: https://pypi.org/p/mlx-cuda
+    steps:
+      - uses: actions/download-artifact@v6
+        with:
+          name: mlx-cuda
+          path: artifacts
+      - name: Display structure of downloaded files
+        run: ls -R artifacts
+      # - name: Publish package distributions to PyPI
+      #  uses: pypa/gh-action-pypi-publish@release/v1
+
+  pypi-publish-cpu:
+    name: Upload CPU release to PyPI
+    runs-on: ubuntu-latest
+    needs: build_linux_release
+    permissions:
+      id-token: write
+    environment:
+      name: pypi
+      url: https://pypi.org/p/mlx-cpu
+    steps:
+      - uses: actions/download-artifact@v6
+        with:
+          name: mlx-cpu
+          path: artifacts
+      - name: Display structure of downloaded files
+        run: ls -R artifacts
+      # - name: Publish package distributions to PyPI
+      #  uses: pypa/gh-action-pypi-publish@release/v1
+
+  pypi-publish-metal:
+    name: Upload Metal release to PyPI
+    runs-on: ubuntu-latest
+    needs: build_mac_release
+    permissions:
+      id-token: write
+    environment:
+      name: pypi
+      url: https://pypi.org/p/mlx-metal
+    steps:
+      - uses: actions/download-artifact@v6
+        with:
+          name: mlx-metal
+          path: artifacts
+      - name: Display structure of downloaded files
+        run: ls -R artifacts
+      # - name: Publish package distributions to PyPI
+      #  uses: pypa/gh-action-pypi-publish@release/v1

.pre-commit-config.yaml

@@ -1,4 +1,10 @@
 repos:
+-   repo: https://github.com/pre-commit/pre-commit-hooks
+    rev: v6.0.0
+    hooks:
+    -   id: check-yaml
+    # -   id: end-of-file-fixer
+    # -   id: trailing-whitespace
 -   repo: https://github.com/pre-commit/mirrors-clang-format
     rev: v19.1.7
     hooks:

CMakeLists.txt

@@ -26,6 +26,7 @@ set(CMAKE_CXX_STANDARD 17)
 set(CMAKE_CXX_STANDARD_REQUIRED ON)
 set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 set(CMAKE_INSTALL_MESSAGE NEVER)
+set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
 
 # ----------------------------- Configuration -----------------------------
 option(MLX_BUILD_TESTS "Build tests for mlx" ON)
@@ -87,6 +88,11 @@ cmake_policy(SET CMP0135 NEW)
 
 add_library(mlx)
 
+# Supress warnings: note: parameter passing for argument of type
+# ‘std::pair<float, float>’ when C++17 is enabled changed to match C++14 in GCC
+# 10.1
+target_compile_options(mlx PRIVATE -Wno-psabi)
+
 if(MLX_BUILD_CUDA)
   enable_language(CUDA)
 endif()
@@ -173,7 +179,7 @@ if(MLX_BUILD_CPU)
     message(STATUS "Accelerate found ${ACCELERATE_LIBRARY}")
     set(MLX_BUILD_ACCELERATE ON)
   else()
-    message(STATUS "Accelerate or arm neon not found, using default backend.")
+    message(STATUS "Accelerate not found, using default backend.")
     set(MLX_BUILD_ACCELERATE OFF)
   endif()
 

README.md

@@ -110,7 +110,7 @@ Hannun, Jagrit Digani, Angelos Katharopoulos, and Ronan Collobert. If you find
 MLX useful in your research and wish to cite it, please use the following
 BibTex entry:
 
-```
+```text
 @software{mlx2023,
   author = {Awni Hannun and Jagrit Digani and Angelos Katharopoulos and Ronan Collobert},
   title = {{MLX}: Efficient and flexible machine learning on Apple silicon},

benchmarks/python/blas/bench_gemm.py

@@ -142,9 +142,7 @@ def bench_shape(B, M, N, K, np_dtype, transpose="nn"):
     t_b = (0, 1, 2) if transpose[1] == "n" else (0, 2, 1)
 
     c_mlx = a_mx.transpose(t_a) @ b_mx.transpose(t_b)
-    c_npy = a_np.transpose(t_a).astype(np.float32) @ b_np.transpose(t_b).astype(
-        np.float32
-    )
+    c_npy = a_np.transpose(t_a).astype(np_dtype) @ b_np.transpose(t_b).astype(np_dtype)
 
     atol = 1e-5 if np_dtype == np.float32 else 1e-4
 
@@ -163,7 +161,7 @@ def get_gflop_count(B, M, N, K):
 if __name__ == "__main__":
     parser = argparse.ArgumentParser(description="Run gemm benchmarks")
 
-    dtypes = ("float32", "float16")
+    dtypes = ("float32", "float16", "complex64")
     transposes = ("nn", "nt", "tn")
     shapes = (
         (16, 234, 768, 3072),
@@ -187,7 +185,7 @@ if __name__ == "__main__":
                 diff = gflops_mx / gflops_pt - 1.0
 
                 print(
-                    f"{B:3d}, {M:4d}, {N:4d}, {K:4d}, {dtype}, {transpose}, {gflops_pt:05.3f}, {gflops_mx:05.3f}, {100. * diff:+5.2f}%"
+                    f"{B:3d}, {M:4d}, {N:4d}, {K:4d}, {dtype}, {transpose}, {gflops_pt:05.3f}, {gflops_mx:05.3f}, {100.0 * diff:+5.2f}%"
                 )
                 if gflops_pt >= 2.0 * gflops_mx:
                     print("ATTENTION ^^^^^^^")

benchmarks/python/blas/bench_gemv.py

@@ -196,7 +196,7 @@ def bench_with_out_len(ax, out_vec_len, in_vector_lens, dtype, transpose):
 
 
 for transpose in (False, True):
-    for dtype in ("float32", "float16"):
+    for dtype in ("float32", "float16", "complex64"):
         fig, axs = plt.subplots(
             len(in_vec_sizes), 2, figsize=(8.5, 11), layout="constrained"
         )
@@ -215,7 +215,7 @@ for transpose in (False, True):
         fig.suptitle(f"{device_name}: {dtype} {op_name}")
         fig.savefig(
             os.path.join(
-                results_dir, f'{device_name.replace(" ", "_")}_{dtype}_{op_name}.pdf'
+                results_dir, f"{device_name.replace(' ', '_')}_{dtype}_{op_name}.pdf"
             )
         )
         plt.close(fig)

docs/src/install.rst

@@ -16,7 +16,7 @@ silicon computer is
 To install from PyPI your system must meet the following requirements:
 
 - Using an M series chip (Apple silicon)
-- Using a native Python >= 3.9
+- Using a native Python >= 3.10
 - macOS >= 13.5
 
 .. note::
@@ -39,7 +39,7 @@ requirements:
 - Nvidia driver >= 550.54.14
 - CUDA toolkit >= 12.0
 - Linux distribution with glibc >= 2.35
-- Python >= 3.9
+- Python >= 3.10
 
 
 CPU-only (Linux)
@@ -55,7 +55,7 @@ To install the CPU-only package from PyPi your system must meet the following
 requirements:
 
 - Linux distribution with glibc >= 2.35
-- Python >= 3.9
+- Python >= 3.10
 
 
 Troubleshooting

docs/src/python/ops.rst

@@ -112,6 +112,7 @@ Operations
    max
    maximum
    mean
+   median
    meshgrid
    min
    minimum

mlx/array.cpp

@@ -64,7 +64,7 @@ array array::unsafe_weak_copy(const array& other) {
       other.strides(),
       other.flags(),
       [](auto) {});
-  cpy.array_desc_->data_ptr = other.array_desc_->data_ptr;
+  cpy.array_desc_->offset = other.array_desc_->offset;
   return cpy;
 }
 
@@ -141,7 +141,7 @@ bool array::is_tracer() const {
 
 void array::set_data(allocator::Buffer buffer, Deleter d) {
   array_desc_->data = std::make_shared<Data>(buffer, d);
-  array_desc_->data_ptr = buffer.raw_ptr();
+  array_desc_->offset = 0;
   array_desc_->data_size = size();
   array_desc_->flags.contiguous = true;
   array_desc_->flags.row_contiguous = true;
@@ -156,7 +156,7 @@ void array::set_data(
     Flags flags,
     Deleter d) {
   array_desc_->data = std::make_shared<Data>(buffer, d);
-  array_desc_->data_ptr = buffer.raw_ptr();
+  array_desc_->offset = 0;
   array_desc_->data_size = data_size;
   array_desc_->strides = std::move(strides);
   array_desc_->flags = flags;
@@ -172,9 +172,8 @@ void array::copy_shared_buffer(
   array_desc_->strides = strides;
   array_desc_->flags = flags;
   array_desc_->data_size = data_size;
-  auto char_offset = sizeof(char) * itemsize() * offset;
-  array_desc_->data_ptr = static_cast<void*>(
-      static_cast<char*>(other.array_desc_->data_ptr) + char_offset);
+  array_desc_->offset =
+      sizeof(char) * itemsize() * offset + other.array_desc_->offset;
 }
 
 void array::copy_shared_buffer(const array& other) {
@@ -241,8 +240,8 @@ array::ArrayDesc::ArrayDesc(
     std::vector<array> inputs)
     : shape(std::move(shape)),
       dtype(dtype),
-      status(Status::unscheduled),
       primitive(std::move(primitive)),
+      status(Status::unscheduled),
       inputs(std::move(inputs)) {
   init();
 }

mlx/array.h

@@ -349,15 +349,19 @@ class array {
     return array_desc_->data;
   }
 
-  // Return a raw pointer to the arrays data
+  // Return a raw pointer to the arrays data. This function may do a copy if
+  // the underlying buffer is not accessible on the CPU. When accessing the
+  // data for GPU kernels, be sure to use the correct method / function for the
+  // given backend to access the GPU pointer.
   template <typename T>
   T* data() {
-    return static_cast<T*>(array_desc_->data_ptr);
+    return reinterpret_cast<T*>(
+        (static_cast<char*>(buffer().raw_ptr()) + array_desc_->offset));
   }
 
   template <typename T>
   const T* data() const {
-    return static_cast<T*>(array_desc_->data_ptr);
+    return const_cast<array&>(*this).data<T>();
   }
 
   enum Status {
@@ -461,8 +465,8 @@ class array {
     // can share the underlying data buffer.
     std::shared_ptr<Data> data;
 
-    // Properly offset data pointer
-    void* data_ptr{nullptr};
+    // Offset from beginning of data pointer
+    int64_t offset{0};
 
     // The size in elements of the data buffer the array accesses
     size_t data_size;

mlx/backend/common/binary.h

@@ -38,20 +38,20 @@ inline void set_binary_op_output_data(
     const array& a,
     const array& b,
     array& out,
-    BinaryOpType bopt) {
+    BinaryOpType bopt,
+    std::function<allocator::Buffer(size_t)> mallocfn = allocator::malloc) {
   bool b_donatable = is_donatable(b, out);
   bool a_donatable = is_donatable(a, out);
   switch (bopt) {
     case BinaryOpType::ScalarScalar:
-      out.set_data(
-          allocator::malloc(out.itemsize()), 1, a.strides(), a.flags());
+      out.set_data(mallocfn(out.itemsize()), 1, a.strides(), a.flags());
       break;
     case BinaryOpType::ScalarVector:
       if (b_donatable) {
         out.copy_shared_buffer(b);
       } else {
         out.set_data(
-            allocator::malloc(b.data_size() * out.itemsize()),
+            mallocfn(b.data_size() * out.itemsize()),
             b.data_size(),
             b.strides(),
             b.flags());
@@ -62,7 +62,7 @@ inline void set_binary_op_output_data(
         out.copy_shared_buffer(a);
       } else {
         out.set_data(
-            allocator::malloc(a.data_size() * out.itemsize()),
+            mallocfn(a.data_size() * out.itemsize()),
             a.data_size(),
             a.strides(),
             a.flags());
@@ -75,7 +75,7 @@ inline void set_binary_op_output_data(
         out.copy_shared_buffer(b);
       } else {
         out.set_data(
-            allocator::malloc(a.data_size() * out.itemsize()),
+            mallocfn(a.data_size() * out.itemsize()),
             a.data_size(),
             a.strides(),
             a.flags());
@@ -88,7 +88,7 @@ inline void set_binary_op_output_data(
           b_donatable && b.flags().row_contiguous && b.size() == out.size()) {
         out.copy_shared_buffer(b);
       } else {
-        out.set_data(allocator::malloc(out.nbytes()));
+        out.set_data(mallocfn(out.nbytes()));
       }
       break;
   }

mlx/backend/common/compiled.cpp

@@ -114,7 +114,9 @@ void compiled_allocate_outputs(
     const std::vector<array>& inputs,
     std::vector<array>& outputs,
     const std::function<bool(size_t)>& is_constant,
-    bool contiguous) {
+    bool contiguous,
+    const std::function<allocator::Buffer(size_t)>&
+        mallocfn /* = allocator::malloc */) {
   if (contiguous) {
     int o = 0;
     Strides strides;
@@ -140,7 +142,7 @@ void compiled_allocate_outputs(
     }
     for (; o < outputs.size(); ++o) {
       outputs[o].set_data(
-          allocator::malloc(data_size * outputs[o].itemsize()),
+          mallocfn(data_size * outputs[o].itemsize()),
           data_size,
           strides,
           flags);
@@ -163,7 +165,7 @@ void compiled_allocate_outputs(
       }
     }
     for (; o < outputs.size(); ++o) {
-      outputs[o].set_data(allocator::malloc(outputs[o].nbytes()));
+      outputs[o].set_data(mallocfn(outputs[o].nbytes()));
     }
   }
 }

mlx/backend/common/compiled.h

@@ -58,7 +58,9 @@ void compiled_allocate_outputs(
     const std::vector<array>& inputs,
     std::vector<array>& outputs,
     const std::function<bool(size_t)>& is_constant,
-    bool contiguous);
+    bool contiguous,
+    const std::function<allocator::Buffer(size_t)>& mallocfn =
+        allocator::malloc);
 
 // Collapse contiguous dims ignoring scalars and constants.
 std::tuple<bool, Shape, std::vector<Strides>> compiled_collapse_contiguous_dims(

mlx/backend/common/copy.h

@@ -22,7 +22,11 @@ enum class CopyType {
   GeneralGeneral
 };
 
-inline bool set_copy_output_data(const array& in, array& out, CopyType ctype) {
+inline bool set_copy_output_data(
+    const array& in,
+    array& out,
+    CopyType ctype,
+    std::function<allocator::Buffer(size_t)> mallocfn = allocator::malloc) {
   if (ctype == CopyType::Vector) {
     // If the input is donateable, we are doing a vector copy and the types
     // have the same size, then the input buffer can hold the output.
@@ -31,14 +35,14 @@ inline bool set_copy_output_data(const array& in, array& out, CopyType ctype) {
       return true;
     } else {
       out.set_data(
-          allocator::malloc(in.data_size() * out.itemsize()),
+          mallocfn(in.data_size() * out.itemsize()),
           in.data_size(),
           in.strides(),
           in.flags());
       return false;
     }
   } else {
-    out.set_data(allocator::malloc(out.nbytes()));
+    out.set_data(mallocfn(out.nbytes()));
     return false;
   }
 }

mlx/backend/common/matmul.h

@@ -13,7 +13,7 @@ inline std::tuple<Shape, Strides, Strides> collapse_batches(
     const array& a,
     const array& b) {
   if (a.ndim() == 2) {
-    return {{1}, {0}, {0}};
+    return {Shape{1}, Strides{0}, Strides{0}};
   }
 
   Shape A_bshape{a.shape().begin(), a.shape().end() - 2};
@@ -38,7 +38,7 @@ inline std::tuple<Shape, Strides, Strides> collapse_batches(
 inline std::tuple<Shape, Strides, Strides, Strides>
 collapse_batches(const array& a, const array& b, const array& c) {
   if (a.ndim() == 2) {
-    return {{1}, {0}, {0}, {0}};
+    return {Shape{1}, Strides{0}, Strides{0}, Strides{0}};
   }
 
   Shape A_bshape{a.shape().begin(), a.shape().end() - 2};

mlx/backend/common/ternary.h

@@ -11,6 +11,8 @@ namespace mlx::core {
 enum class TernaryOpType {
   ScalarScalarScalar,
   VectorVectorVector,
+  VectorVectorScalar,
+  VectorScalarVector,
   General,
 };
 
@@ -25,6 +27,14 @@ get_ternary_op_type(const array& a, const array& b, const array& c) {
       (a.flags().col_contiguous && b.flags().col_contiguous &&
        c.flags().col_contiguous)) {
     topt = TernaryOpType::VectorVectorVector;
+  } else if (
+      b.data_size() == 1 && a.flags().row_contiguous &&
+      c.flags().row_contiguous) {
+    topt = TernaryOpType::VectorScalarVector;
+  } else if (
+      c.data_size() == 1 && a.flags().row_contiguous &&
+      b.flags().row_contiguous) {
+    topt = TernaryOpType::VectorVectorScalar;
   } else {
     topt = TernaryOpType::General;
   }
@@ -36,7 +46,8 @@ inline void set_ternary_op_output_data(
     const array& b,
     const array& c,
     array& out,
-    TernaryOpType topt) {
+    TernaryOpType topt,
+    std::function<allocator::Buffer(size_t)> mallocfn = allocator::malloc) {
   auto maybe_donate = [&out](const array& x) {
     if (is_donatable(x, out)) {
       out.copy_shared_buffer(x);
@@ -47,24 +58,25 @@ inline void set_ternary_op_output_data(
 
   switch (topt) {
     case TernaryOpType::ScalarScalarScalar:
-      out.set_data(
-          allocator::malloc(out.itemsize()), 1, b.strides(), b.flags());
+      out.set_data(mallocfn(out.itemsize()), 1, b.strides(), b.flags());
       break;
     case TernaryOpType::VectorVectorVector:
       if (!(maybe_donate(a) || maybe_donate(b) || maybe_donate(c))) {
         out.set_data(
-            allocator::malloc(out.itemsize() * b.data_size()),
+            mallocfn(out.itemsize() * b.data_size()),
             b.data_size(),
             b.strides(),
             b.flags());
       }
       break;
+    case TernaryOpType::VectorVectorScalar:
+    case TernaryOpType::VectorScalarVector:
     case TernaryOpType::General:
       // Try to donate an input which is row_contiguous
       if (!((a.flags().row_contiguous && maybe_donate(a)) ||
             (b.flags().row_contiguous && maybe_donate(b)) ||
             (c.flags().row_contiguous && maybe_donate(c)))) {
-        out.set_data(allocator::malloc(out.nbytes()));
+        out.set_data(mallocfn(out.nbytes()));
       }
       break;
   }

mlx/backend/common/unary.h

@@ -7,19 +7,22 @@
 
 namespace mlx::core {
 
-inline void set_unary_output_data(const array& in, array& out) {
+inline void set_unary_output_data(
+    const array& in,
+    array& out,
+    std::function<allocator::Buffer(size_t)> mallocfn = allocator::malloc) {
   if (in.flags().contiguous) {
     if (is_donatable(in, out)) {
       out.copy_shared_buffer(in);
     } else {
       out.set_data(
-          allocator::malloc(in.data_size() * out.itemsize()),
+          mallocfn(in.data_size() * out.itemsize()),
           in.data_size(),
           in.strides(),
           in.flags());
     }
   } else {
-    out.set_data(allocator::malloc(out.nbytes()));
+    out.set_data(mallocfn(out.nbytes()));
   }
 }
 

mlx/backend/cpu/conv.cpp

@@ -996,131 +996,6 @@ void explicit_gemm_conv_1D_cpu(
   encoder.add_temporaries(std::move(temps));
 }
 
-void explicit_gemm_conv_2D_cpu(
-    const array& in,
-    const array& wt,
-    array out,
-    const std::vector<int>& padding_lo,
-    const std::vector<int>& padding_hi,
-    const std::vector<int>& wt_strides,
-    const std::vector<int>& wt_dilation,
-    Stream stream) {
-  const int N = in.shape(0); // Batch size, should be the same as out.shape(0)
-  const int iH = in.shape(1); // Input spatial dim
-  const int iW = in.shape(2); // Input spatial dim
-  const int oH = out.shape(1); // Output spatial dim
-  const int oW = out.shape(2); // Output spatial dim
-  const int O = wt.shape(0); // Out channels
-  const int C = wt.shape(3); // In channels
-  const int wH = wt.shape(1); // Weight spatial dim
-  const int wW = wt.shape(2); // Weight spatial dim
-
-  auto conv_dtype = out.dtype();
-  auto& encoder = cpu::get_command_encoder(stream);
-
-  // Pad input
-  Shape padded_shape = {
-      N,
-      iH + padding_lo[0] + padding_hi[0],
-      iW + padding_lo[1] + padding_hi[1],
-      C};
-  array in_padded(padded_shape, conv_dtype, nullptr, {});
-
-  // Fill with zeros
-  std::vector<array> temps;
-  temps.push_back(array(0, conv_dtype));
-  copy_cpu(temps.back(), in_padded, CopyType::Scalar, stream);
-
-  // Pick input slice from padded
-  size_t data_offset = padding_lo[0] * in_padded.strides()[1] +
-      padding_lo[1] * in_padded.strides()[2];
-  array in_padded_slice(in.shape(), in_padded.dtype(), nullptr, {});
-  in_padded_slice.copy_shared_buffer(
-      in_padded,
-      in_padded.strides(),
-      in_padded.flags(),
-      in_padded_slice.size(),
-      data_offset);
-  temps.push_back(in_padded_slice);
-
-  // Copy input values into the slice
-  copy_cpu_inplace(in, in_padded_slice, CopyType::GeneralGeneral, stream);
-
-  // Make strided view
-  Shape strided_shape = {N, oH, oW, wH, wW, C};
-
-  Strides strided_strides = {
-      in_padded.strides()[0],
-      in_padded.strides()[1] * wt_strides[0],
-      in_padded.strides()[2] * wt_strides[1],
-      in_padded.strides()[1],
-      in_padded.strides()[2],
-      in_padded.strides()[3]};
-  auto flags = in_padded.flags();
-
-  array in_strided_view(strided_shape, in_padded.dtype(), nullptr, {});
-  in_strided_view.copy_shared_buffer(
-      in_padded, strided_strides, flags, in_strided_view.size(), 0);
-
-  // Materialize strided view
-  Shape strided_reshape = {N * oH * oW, wH * wW * C};
-  array in_strided(strided_reshape, in_strided_view.dtype(), nullptr, {});
-  copy_cpu(in_strided_view, in_strided, CopyType::General, stream);
-  temps.push_back(in_strided);
-
-  // Check wt dtype and prepare
-  auto gemm_wt = wt;
-  auto gemm_out = out;
-
-  if (wt.dtype() != float32 || !wt.flags().row_contiguous) {
-    auto ctype =
-        wt.flags().row_contiguous ? CopyType::Vector : CopyType::General;
-    gemm_wt = array(wt.shape(), float32, nullptr, {});
-    copy_cpu(wt, gemm_wt, ctype, stream);
-    temps.push_back(gemm_wt);
-  }
-
-  if (out.dtype() != float32) {
-    gemm_out = array(out.shape(), float32, nullptr, {});
-    gemm_out.set_data(allocator::malloc(gemm_out.nbytes()));
-    temps.push_back(gemm_out);
-  }
-
-  encoder.set_input_array(in_strided);
-  encoder.set_input_array(gemm_wt);
-  encoder.set_output_array(gemm_out);
-
-  encoder.dispatch([in_strided_ptr = in_strided.data<float>(),
-                    gemm_wt_ptr = gemm_wt.data<float>(),
-                    gemm_out_ptr = gemm_out.data<float>(),
-                    strided_reshape = std::move(strided_reshape),
-                    O]() {
-    // Perform gemm
-    cblas_sgemm(
-        CblasRowMajor,
-        CblasNoTrans, // no trans A
-        CblasTrans, // transB
-        strided_reshape[0], // M
-        O, // N
-        strided_reshape[1], // K
-        1.0f, // alpha
-        in_strided_ptr,
-        strided_reshape[1], // lda
-        gemm_wt_ptr,
-        strided_reshape[1], // ldb
-        0.0f, // beta
-        gemm_out_ptr,
-        O // ldc
-    );
-  });
-
-  // Copy results if needed
-  if (out.dtype() != float32) {
-    copy_cpu_inplace(gemm_out, out, CopyType::Vector, stream);
-  }
-  encoder.add_temporaries(std::move(temps));
-}
-
 void explicit_gemm_conv_ND_cpu(
     const array& in,
     const array& wt,

mlx/backend/cpu/eig.cpp

@@ -46,7 +46,6 @@ void eig_impl(
     int info;
     {
       T work;
-      int iwork;
       geev<T>(
           &jobl,
           &jobr,

mlx/backend/cpu/gemms/bnns.cpp

@@ -1,5 +1,4 @@
 // Copyright © 2023-2024 Apple Inc.
-
 #include <Accelerate/Accelerate.h>
 
 #include "mlx/array.h"
@@ -49,9 +48,15 @@ void matmul_bnns(
   size_t K = a_shape[ndim - 1];
 
   BNNSDataType bnns_dtype = to_bnns_dtype<T>();
-
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wdeprecated-declarations"
+  if (beta != 1.0 && beta != 0.0) {
+    // scale the output
+    for (auto i = 0; i < batch_size * M * N; ++i) {
+      out[i] *= beta;
+    }
+    beta = 1.0;
+  }
   const BNNSLayerParametersBroadcastMatMul gemm_params{
       /* float alpha = */ alpha,
       /* float beta = */ beta,

mlx/backend/cpu/gemms/cblas.cpp

@@ -88,4 +88,47 @@ void matmul<double>(
   }
 }
 
+template <>
+void matmul<complex64_t>(
+    const complex64_t* a,
+    const complex64_t* b,
+    complex64_t* out,
+    bool a_transposed,
+    bool b_transposed,
+    size_t lda,
+    size_t ldb,
+    size_t ldc,
+    float alpha,
+    float beta,
+    size_t batch_size,
+    const Shape& a_shape,
+    const Strides& a_strides,
+    const Shape& b_shape,
+    const Strides& b_strides) {
+  auto ndim = a_shape.size();
+  size_t M = a_shape[ndim - 2];
+  size_t N = b_shape[ndim - 1];
+  size_t K = a_shape[ndim - 1];
+  auto calpha = static_cast<complex64_t>(alpha);
+  auto cbeta = static_cast<complex64_t>(beta);
+
+  for (int i = 0; i < batch_size; ++i) {
+    cblas_cgemm(
+        CblasRowMajor,
+        a_transposed ? CblasTrans : CblasNoTrans, // transA
+        b_transposed ? CblasTrans : CblasNoTrans, // transB
+        M,
+        N,
+        K,
+        &calpha,
+        a + elem_to_loc(M * K * i, a_shape, a_strides),
+        lda,
+        b + elem_to_loc(K * N * i, b_shape, b_strides),
+        ldb,
+        &cbeta,
+        out + M * N * i,
+        ldc);
+  }
+}
+
 } // namespace mlx::core

mlx/backend/cpu/masked_mm.cpp

@@ -215,18 +215,18 @@ void BlockMaskedMM::eval_cpu(const std::vector<array>& inputs, array& out) {
 
   encoder.set_input_array(a);
   encoder.set_input_array(b);
-  const void* a_mask_ptr;
-  const void* b_mask_ptr;
-  const void* out_mask_ptr;
+  const void* a_mask_ptr = nullptr;
+  const void* b_mask_ptr = nullptr;
+  const void* out_mask_ptr = nullptr;
   Shape a_mask_shape;
   Shape b_mask_shape;
   Shape out_mask_shape;
   Strides a_mask_strides;
   Strides b_mask_strides;
   Strides out_mask_strides;
-  bool a_mask_bool;
-  bool b_mask_bool;
-  bool out_mask_bool;
+  bool a_mask_bool = false;
+  bool b_mask_bool = false;
+  bool out_mask_bool = false;
   if (has_op_mask) {
     auto& a_mask = inputs[inputs.size() - 2];
     auto& b_mask = inputs[inputs.size() - 1];
@@ -423,7 +423,6 @@ void GatherMM::eval_cpu(const std::vector<array>& inputs, array& out) {
   auto& rhs_indices = inputs[3];
 
   auto batch_shape = get_batch_dims(out.shape());
-  int batch_ndim = batch_shape.size();
 
   auto batch_shape_A = get_batch_dims(a.shape());
   auto batch_strides_A = get_batch_dims(a.strides());

mlx/backend/cpu/matmul.cpp

@@ -91,7 +91,6 @@ void matmul_general(
   auto [b_transposed, ldb, b] = check_transpose(b_pre);
   size_t M = a.shape(-2);
   size_t N = b.shape(-1);
-  size_t K = a.shape(-1);
   if (M == 0 || N == 0) {
     return;
   }
@@ -108,6 +107,9 @@ void matmul_general(
   } else if (out.dtype() == float64) {
     matmul_dispatch<double>(
         a, b, out, a_transposed, b_transposed, lda, ldb, alpha, beta, stream);
+  } else if (out.dtype() == complex64) {
+    matmul_dispatch<complex64_t>(
+        a, b, out, a_transposed, b_transposed, lda, ldb, alpha, beta, stream);
   } else {
     throw std::runtime_error("[Matmul::eval_cpu] Invalid type.");
   }
@@ -128,10 +130,6 @@ void Matmul::eval_cpu(const std::vector<array>& inputs, array& out) {
 }
 
 void AddMM::eval_cpu(const std::vector<array>& inputs, array& out) {
-  if (out.dtype() != float32) {
-    throw std::runtime_error(
-        "[AddMM::eval_cpu] Currently only supports float32.");
-  }
   if (out.size() == 0) {
     out.set_data(allocator::malloc(out.nbytes()));
     return;

mlx/backend/cpu/quantized.cpp

@@ -1,8 +1,11 @@
 // Copyright © 2023 Apple Inc.
 
+#include "mlx/backend/common/unary.h"
 #include "mlx/backend/cpu/copy.h"
 #include "mlx/backend/cpu/encoder.h"
 #include "mlx/backend/cpu/simd/simd.h"
+#include "mlx/backend/cpu/unary.h"
+#include "mlx/backend/cpu/unary_ops.h"
 #include "mlx/fast_primitives.h"
 #include "mlx/primitives.h"
 #include "mlx/utils.h"
@@ -445,7 +448,6 @@ void mxfp4_qmm(
     int K) {
   constexpr int group_size = 32;
   constexpr int pack_factor = get_pack_factor(4, 8);
-  constexpr int bytes_per_pack = get_bytes_per_pack(4);
   constexpr int packs_in_group = group_size / pack_factor;
 
   for (int m = 0; m < M; m++) {
@@ -487,7 +489,6 @@ void mxfp4_qmm_t(
     int K) {
   constexpr int group_size = 32;
   constexpr int pack_factor = get_pack_factor(4, 8);
-  constexpr int bytes_per_pack = get_bytes_per_pack(4);
   constexpr int packs_in_group = group_size / pack_factor;
 
   for (int m = 0; m < M; m++) {
@@ -1104,4 +1105,44 @@ void fast::Quantize::eval_cpu(
   });
 }
 
+void fast::ConvertFP8::eval_cpu(
+    const std::vector<array>& inputs,
+    std::vector<array>& outputs) {
+  auto& in = inputs[0];
+  auto& out = outputs[0];
+  set_unary_output_data(in, out);
+  auto& encoder = cpu::get_command_encoder(stream());
+  encoder.set_input_array(in);
+  encoder.set_output_array(out);
+  encoder.dispatch([in = array::unsafe_weak_copy(in),
+                    out = array::unsafe_weak_copy(out),
+                    to_fp8 = to_fp8_]() mutable {
+    if (to_fp8) {
+      switch (in.dtype()) {
+        case float16:
+          unary_op<float16_t, uint8_t>(in, out, detail::ToFP8());
+          break;
+        case bfloat16:
+          unary_op<bfloat16_t, uint8_t>(in, out, detail::ToFP8());
+          break;
+        default:
+          unary_op<float, uint8_t>(in, out, detail::ToFP8());
+          break;
+      }
+    } else {
+      switch (out.dtype()) {
+        case float16:
+          unary_op<uint8_t, float16_t>(in, out, detail::FromFP8());
+          break;
+        case bfloat16:
+          unary_op<uint8_t, bfloat16_t>(in, out, detail::FromFP8());
+          break;
+        default:
+          unary_op<uint8_t, float>(in, out, detail::FromFP8());
+          break;
+      }
+    }
+  });
+}
+
 } // namespace mlx::core

mlx/backend/cpu/simd/accelerate_simd.h

@@ -1,5 +1,6 @@
 #pragma once
 
+#include <arm_neon.h>
 #include <simd/math.h>
 #include <simd/vector.h>
 
@@ -9,7 +10,7 @@
 
 #include "mlx/backend/cpu/simd/base_simd.h"
 
-// There seems to be a bug in sims/base.h
+// There seems to be a bug in simd/base_simd.h
 // __XROS_2_0 is not defined, the expression evaluates
 // to true instead of false setting the SIMD library
 // higher than it should be even on macOS < 15
@@ -200,6 +201,15 @@ SIMD_DEFAULT_COMPARISONS(<=)
 SIMD_DEFAULT_COMPARISONS(==)
 SIMD_DEFAULT_COMPARISONS(!=)
 
+template <typename T, int N>
+Simd<T, N> clz(Simd<T, N> x) {
+  auto a = *(uint32x4_t*)(&x);
+  auto b = *((uint32x4_t*)(&x) + 1);
+  a = vclzq_u32(a);
+  b = vclzq_u32(b);
+  return asd::make_uint8(a, b);
+}
+
 template <typename T, int N>
 Simd<T, N> atan2(Simd<T, N> a, Simd<T, N> b) {
   return asd::atan2(a.value, b.value);

mlx/backend/cpu/simd/base_simd.h

@@ -171,6 +171,11 @@ DEFAULT_BINARY(&)
 DEFAULT_BINARY(&&)
 DEFAULT_BINARY(||)
 
+template <typename T>
+Simd<T, 1> clz(Simd<T, 1> x_) {
+  return __builtin_clz(x_.value);
+}
+
 template <typename T>
 Simd<T, 1> remainder(Simd<T, 1> a_, Simd<T, 1> b_) {
   T a = a_.value;

mlx/backend/cpu/sort.cpp

@@ -15,6 +15,18 @@ namespace mlx::core {
 
 namespace {
 
+// NaN-aware comparator that places NaNs at the end
+template <typename T>
+bool nan_aware_less(T a, T b) {
+  if constexpr (std::is_floating_point_v<T> || std::is_same_v<T, complex64_t>) {
+    if (std::isnan(a))
+      return false;
+    if (std::isnan(b))
+      return true;
+  }
+  return a < b;
+}
+
 template <typename T>
 struct StridedIterator {
   using iterator_category = std::random_access_iterator_tag;
@@ -27,7 +39,7 @@ struct StridedIterator {
   StridedIterator() = default;
 
   explicit StridedIterator(T* ptr, int64_t stride, difference_type offset = 0)
-      : ptr_(ptr + offset * stride), stride_(stride) {}
+      : stride_(stride), ptr_(ptr + offset * stride) {}
 
   explicit StridedIterator(array& arr, int axis, difference_type offset = 0)
       : StridedIterator(arr.data<T>(), arr.strides()[axis], offset) {}
@@ -130,7 +142,7 @@ void sort(array& out, int axis) {
     StridedIterator st(data_ptr, axis_stride, 0);
     StridedIterator ed(data_ptr, axis_stride, axis_size);
 
-    std::stable_sort(st, ed);
+    std::stable_sort(st, ed, nan_aware_less<T>);
     src_it.step();
   }
 }
@@ -184,6 +196,15 @@ void argsort(const array& in, array& out, int axis) {
     std::stable_sort(st, ed, [data_ptr, in_stride](IdxT a, IdxT b) {
       auto v1 = data_ptr[a * in_stride];
       auto v2 = data_ptr[b * in_stride];
+
+      // Handle NaNs (place them at the end)
+      if (std::is_floating_point<T>::value) {
+        if (std::isnan(v1))
+          return false;
+        if (std::isnan(v2))
+          return true;
+      }
+
       return v1 < v2 || (v1 == v2 && a < b);
     });
   }
@@ -219,7 +240,7 @@ void partition(array& out, int axis, int kth) {
     StridedIterator md(data_ptr, axis_stride, kth);
     StridedIterator ed(data_ptr, axis_stride, axis_size);
 
-    std::nth_element(st, md, ed);
+    std::nth_element(st, md, ed, nan_aware_less<T>);
   }
 }
 
@@ -276,6 +297,15 @@ void argpartition(const array& in, array& out, int axis, int kth) {
     std::nth_element(st, md, ed, [data_ptr, in_stride](IdxT a, IdxT b) {
       auto v1 = data_ptr[a * in_stride];
       auto v2 = data_ptr[b * in_stride];
+
+      // Handle NaNs (place them at the end)
+      if (std::is_floating_point<T>::value) {
+        if (std::isnan(v1))
+          return false;
+        if (std::isnan(v2))
+          return true;
+      }
+
       return v1 < v2 || (v1 == v2 && a < b);
     });
   }

mlx/backend/cpu/svd.cpp

@@ -83,8 +83,6 @@ void svd_impl(
 
     auto jobz = (u_ptr) ? "A" : "N";
 
-    // Will contain the number of singular values after the call has returned.
-    int ns = 0;
     T workspace_dimension = 0;
 
     // Will contain the indices of eigenvectors that failed to converge (not

mlx/backend/cpu/unary.h

@@ -24,9 +24,9 @@ void unary_op(const array& a, array& out, Op) {
   auto ndim = a.ndim();
   if (a.flags().contiguous) {
     auto size = a.data_size();
-    constexpr int N = simd::max_size<T>;
+    constexpr int N = std::min(simd::max_size<T>, simd::max_size<U>);
     while (size >= N) {
-      simd::store(dst, Op{}(simd::load<T, N>(src)));
+      simd::store(dst, simd::Simd<U, N>(Op{}(simd::load<T, N>(src))));
       size -= N;
       src += N;
       dst += N;

mlx/backend/cpu/unary_ops.h

@@ -77,7 +77,8 @@ struct Real {
 struct Sigmoid {
   template <int N, typename T>
   Simd<T, N> operator()(Simd<T, N> x) {
-    return 1.0f / (1.0f + simd::exp(-x));
+    auto y = 1.0f / (1.0f + simd::exp(simd::abs(x)));
+    return simd::select(x < Simd<T, N>{0}, y, Simd<T, N>{1} - y);
   }
   SINGLE()
 };
@@ -107,4 +108,73 @@ struct Square {
   SINGLE()
 };
 
+template <int N>
+Simd<float, N> fp32_from_bits(Simd<uint32_t, N> x) {
+  return *(Simd<float, N>*)(&x);
+}
+template <int N>
+Simd<uint32_t, N> fp32_to_bits(Simd<float, N> x) {
+  return *(Simd<uint32_t, N>*)(&x);
+}
+
+struct ToFP8 {
+  template <typename T, int N>
+  Simd<uint8_t, N> operator()(Simd<T, N> f) {
+    uint32_t fp8_max = 543 << 21;
+    auto denorm_mask = Simd<uint32_t, N>(141 << 23);
+    Simd<uint32_t, N> f_bits;
+    Simd<float, N> f32 = f;
+    f_bits = fp32_to_bits(f32);
+    Simd<uint8_t, N> result = 0u;
+    auto sign = f_bits & 0x80000000;
+    f_bits = f_bits ^ sign;
+
+    auto f_bits_low =
+        fp32_to_bits(fp32_from_bits(f_bits) + fp32_from_bits(denorm_mask));
+    auto result_low = Simd<uint8_t, N>(f_bits_low - denorm_mask);
+
+    auto mant_odd = Simd<uint8_t, N>((f_bits >> 20) & 1);
+    auto f_bits_high = f_bits + (((uint32_t)(7 - 127) << 23) + 0x7FFFF);
+    f_bits_high = f_bits_high + Simd<uint32_t, N>(mant_odd);
+
+    auto result_high = Simd<uint8_t, N>(f_bits_high >> 20);
+    result = select(f_bits < (121 << 23), result_low, result_high);
+
+    auto result_sat = Simd<uint8_t, N>(0x7E);
+    result = select(f_bits >= fp8_max, result_sat, result);
+    return result | Simd<uint8_t, N>(sign >> 24);
+  }
+
+  template <typename T>
+  uint8_t operator()(T x) {
+    return (*this)(Simd<T, 1>(x)).value;
+  }
+};
+
+struct FromFP8 {
+  template <int N>
+  Simd<float, N> operator()(Simd<uint8_t, N> x) {
+    auto w = Simd<uint32_t, N>(x) << 24;
+    auto sign = w & 0x80000000;
+    auto nonsign = w & 0x7FFFFFFF;
+
+    auto renorm_shift = clz(nonsign);
+    renorm_shift = simd::select(
+        renorm_shift > Simd<uint32_t, N>{4},
+        renorm_shift - Simd<uint32_t, N>{4},
+        Simd<uint32_t, N>{0});
+
+    Simd<int32_t, N> inf_nan_mask =
+        (Simd<int32_t, N>(nonsign + 0x01000000) >> 8) & 0x7F800000;
+    auto zero_mask = Simd<int32_t, N>(nonsign - 1) >> 31;
+    auto result = sign |
+        ((((nonsign << renorm_shift >> 4) + ((0x78 - renorm_shift) << 23)) |
+          inf_nan_mask) &
+         ~zero_mask);
+    return fp32_from_bits(result);
+  }
+  float operator()(uint8_t x) {
+    return (*this)(Simd<uint8_t, 1>(x)).value;
+  }
+};
 } // namespace mlx::core::detail

mlx/backend/cuda/CMakeLists.txt

@@ -32,6 +32,7 @@ target_sources(
           ${CMAKE_CURRENT_SOURCE_DIR}/indexing.cpp
           ${CMAKE_CURRENT_SOURCE_DIR}/kernel_utils.cu
           ${CMAKE_CURRENT_SOURCE_DIR}/matmul.cpp
+          ${CMAKE_CURRENT_SOURCE_DIR}/load.cpp
           ${CMAKE_CURRENT_SOURCE_DIR}/layer_norm.cu
           ${CMAKE_CURRENT_SOURCE_DIR}/logsumexp.cu
           ${CMAKE_CURRENT_SOURCE_DIR}/primitives.cpp
@@ -51,12 +52,19 @@ target_sources(
           ${CMAKE_CURRENT_SOURCE_DIR}/ternary.cu
           ${CMAKE_CURRENT_SOURCE_DIR}/utils.cpp
           ${CMAKE_CURRENT_SOURCE_DIR}/quantized/affine_quantize.cu
+          ${CMAKE_CURRENT_SOURCE_DIR}/quantized/fp_quantize.cu
           ${CMAKE_CURRENT_SOURCE_DIR}/quantized/quantized.cpp
+          ${CMAKE_CURRENT_SOURCE_DIR}/quantized/convert_fp8.cu
           ${CMAKE_CURRENT_SOURCE_DIR}/worker.cpp)
 
 add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/binary)
 add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/unary)
 
+# fp4 is not available on < 12.8
+if(CMAKE_CUDA_COMPILER_VERSION VERSION_LESS 12.8.0)
+  target_include_directories(mlx PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/quantized/)
+endif()
+
 if(CMAKE_CUDA_COMPILER_VERSION VERSION_GREATER_EQUAL 12.9.0)
   target_sources(
     mlx PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/gemms/cublas_gemm_batched_12_9.cu)
@@ -170,7 +178,6 @@ target_link_libraries(mlx PRIVATE CUDNN::cudnn_all)
 # Suppress nvcc warnings on MLX headers.
 target_compile_options(mlx PRIVATE $<$<COMPILE_LANGUAGE:CUDA>:-Xcudafe
                                    --diag_suppress=997>)
-
 # Install CCCL headers for JIT.
 install(DIRECTORY ${cccl_SOURCE_DIR}/include/cuda
         DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/cccl)

mlx/backend/cuda/allocator.cpp

@@ -1,6 +1,7 @@
 // Copyright © 2025 Apple Inc.
 
 #include "mlx/backend/cuda/allocator.h"
+#include "mlx/backend/cuda/device.h"
 #include "mlx/backend/cuda/utils.h"
 #include "mlx/utils.h"
 
@@ -30,15 +31,20 @@ SmallSizePool::SmallSizePool() {
   next_free_ = buffer_;
 
   CHECK_CUDA_ERROR(cudaMallocManaged(&data_, small_pool_size));
+
+  int device_count = 0;
+  CHECK_CUDA_ERROR(cudaGetDeviceCount(&device_count));
+  for (int i = 0; i < device_count; ++i) {
 #if CUDART_VERSION >= 13000
     cudaMemLocation loc;
     loc.type = cudaMemLocationTypeDevice;
-  loc.id = 0;
+    loc.id = i;
 #else
-  int loc = 0;
+    int loc = i;
 #endif // CUDART_VERSION >= 13000
     CHECK_CUDA_ERROR(
-      cudaMemAdvise(data_, small_pool_size, cudaMemAdviseSetReadMostly, loc));
+        cudaMemAdvise(data_, small_pool_size, cudaMemAdviseSetAccessedBy, loc));
+  }
 
   auto curr = next_free_;
   for (size_t i = 1; i < num_blocks; ++i) {
@@ -62,6 +68,7 @@ CudaBuffer* SmallSizePool::malloc() {
   next_free_ = next_free_->next;
   b->buf.data = static_cast<char*>(data_) + i * small_block_size;
   b->buf.size = small_block_size;
+  b->buf.device = -1;
   return &b->buf;
 }
 
@@ -83,14 +90,22 @@ CudaAllocator::CudaAllocator()
           page_size,
           [](CudaBuffer* buf) { return buf->size; },
           [this](CudaBuffer* buf) { cuda_free(buf); }) {
-  // TODO: Set memory limit for multi-device.
   size_t free, total;
   CHECK_CUDA_ERROR(cudaMemGetInfo(&free, &total));
-  memory_limit_ = total * 0.8;
+  memory_limit_ = total * 0.95;
   max_pool_size_ = memory_limit_;
+
+  int device_count = 0;
+  CHECK_CUDA_ERROR(cudaGetDeviceCount(&device_count));
+  int curr_device = 0;
+  CHECK_CUDA_ERROR(cudaGetDevice(&curr_device));
+  for (int i = 0; i < device_count; ++i) {
+    free_streams_.emplace_back(
+        cu::device(mlx::core::Device{mlx::core::Device::gpu, i}));
+  }
 }
 
-Buffer CudaAllocator::malloc(size_t size) {
+Buffer CudaAllocator::malloc_impl(size_t size, cudaStream_t stream) {
   // Find available buffer from cache.
   auto orig_size = size;
   std::unique_lock lock(mutex_);
@@ -117,8 +132,17 @@ Buffer CudaAllocator::malloc(size_t size) {
     }
     lock.unlock();
     if (!buf) {
-      buf = new CudaBuffer{nullptr, size};
-      cudaError_t err = cudaMallocManaged(&buf->data, size);
+      int device = -1;
+      if (stream != nullptr) {
+        cudaStreamGetDevice(stream, &device);
+      }
+      buf = new CudaBuffer{nullptr, size, device};
+      cudaError_t err;
+      if (device == -1) {
+        err = cudaMallocManaged(&buf->data, size);
+      } else {
+        err = cudaMallocAsync(&buf->data, size, stream);
+      }
       if (err != cudaSuccess && err != cudaErrorMemoryAllocation) {
         throw std::runtime_error(fmt::format(
             "cudaMallocManaged failed: {}.", cudaGetErrorString(err)));
@@ -136,6 +160,14 @@ Buffer CudaAllocator::malloc(size_t size) {
   return Buffer{buf};
 }
 
+Buffer CudaAllocator::malloc_async(size_t size, cudaStream_t stream) {
+  return malloc_impl(size, stream);
+}
+
+Buffer CudaAllocator::malloc(size_t size) {
+  return malloc_impl(size, nullptr);
+}
+
 void CudaAllocator::free(Buffer buffer) {
   auto* buf = static_cast<CudaBuffer*>(buffer.ptr());
   if (!buf) {
@@ -163,8 +195,12 @@ size_t CudaAllocator::size(Buffer buffer) const {
 void CudaAllocator::cuda_free(CudaBuffer* buf) {
   if (scalar_pool_.in_pool(buf)) {
     scalar_pool_.free(buf);
+  } else {
+    if (buf->device >= 0) {
+      cudaFreeAsync(buf->data, free_streams_[buf->device]);
     } else {
       cudaFree(buf->data);
+    }
     delete buf;
   }
 }
@@ -215,6 +251,16 @@ CudaAllocator& allocator() {
   return *allocator_;
 }
 
+Buffer malloc_async(size_t size, cudaStream_t stream) {
+  auto buffer = allocator().malloc_async(size, stream);
+  if (size && !buffer.ptr()) {
+    std::ostringstream msg;
+    msg << "[malloc_async] Unable to allocate " << size << " bytes.";
+    throw std::runtime_error(msg.str());
+  }
+  return buffer;
+}
+
 } // namespace cu
 
 namespace allocator {
@@ -227,7 +273,19 @@ void* Buffer::raw_ptr() {
   if (!ptr_) {
     return nullptr;
   }
-  return static_cast<cu::CudaBuffer*>(ptr_)->data;
+  auto& cbuf = *static_cast<cu::CudaBuffer*>(ptr_);
+  if (cbuf.device != -1) {
+    // TODO maybe make this async on a i/o stream to avoid synchronizing the
+    // device on malloc/and free
+    void* new_data;
+    CHECK_CUDA_ERROR(cudaMallocManaged(&new_data, cbuf.size));
+    cbuf.device = -1;
+    CHECK_CUDA_ERROR(
+        cudaMemcpy(new_data, cbuf.data, cbuf.size, cudaMemcpyDefault));
+    CHECK_CUDA_ERROR(cudaFree(cbuf.data));
+    cbuf.data = new_data;
+  }
+  return cbuf.data;
 }
 
 } // namespace allocator

mlx/backend/cuda/allocator.h

@@ -4,7 +4,9 @@
 
 #include "mlx/allocator.h"
 #include "mlx/backend/common/buffer_cache.h"
+#include "mlx/backend/cuda/cuda_utils.h"
 
+#include <cuda_runtime.h>
 #include <mutex>
 #include <set>
 #include <utility>
@@ -17,8 +19,14 @@ using allocator::Buffer;
 struct CudaBuffer {
   void* data;
   size_t size;
+  int device; // -1 for managed
 };
 
+template <typename T>
+inline T* gpu_ptr(Buffer buf) {
+  return static_cast<T*>(static_cast<cu::CudaBuffer*>(buf.ptr())->data);
+}
+
 class SmallSizePool {
  private:
   union Block {
@@ -45,6 +53,7 @@ class SmallSizePool {
 class CudaAllocator : public allocator::Allocator {
  public:
   Buffer malloc(size_t size) override;
+  Buffer malloc_async(size_t size, cudaStream_t stream);
   void free(Buffer buffer) override;
   size_t size(Buffer buffer) const override;
 
@@ -58,6 +67,7 @@ class CudaAllocator : public allocator::Allocator {
   void clear_cache();
 
  private:
+  Buffer malloc_impl(size_t size, cudaStream_t stream);
   void cuda_free(CudaBuffer* buf);
 
   CudaAllocator();
@@ -69,9 +79,12 @@ class CudaAllocator : public allocator::Allocator {
   BufferCache<CudaBuffer> buffer_cache_;
   size_t active_memory_{0};
   size_t peak_memory_{0};
+  std::vector<CudaStream> free_streams_;
   SmallSizePool scalar_pool_;
 };
 
 CudaAllocator& allocator();
 
+Buffer malloc_async(size_t size, cudaStream_t stream);
+
 } // namespace mlx::core::cu

mlx/backend/cuda/arange.cu

@@ -41,9 +41,8 @@ void Arange::eval_gpu(const std::vector<array>& inputs, array& out) {
   if (out.size() == 0) {
     return;
   }
-  out.set_data(allocator::malloc(out.nbytes()));
-
   auto& encoder = cu::get_command_encoder(stream());
+  out.set_data(cu::malloc_async(out.nbytes(), encoder.stream()));
   encoder.set_output_array(out);
 
   dispatch_int_float_types(out.dtype(), "Arange", [&](auto type_tag) {
@@ -58,7 +57,7 @@ void Arange::eval_gpu(const std::vector<array>& inputs, array& out) {
           num_blocks,
           block_dims,
           0,
-          out.data<OutType>(),
+          gpu_ptr<OutType>(out),
           out.data_size(),
           static_cast<CTYPE>(start_),
           static_cast<CTYPE>(start_ + step_) - static_cast<CTYPE>(start_));

mlx/backend/cuda/arg_reduce.cu

@@ -140,8 +140,10 @@ void ArgReduce::eval_gpu(const std::vector<array>& inputs, array& out) {
   nvtx3::scoped_range r("ArgReduce::eval_gpu");
   assert(inputs.size() == 1);
   auto& in = inputs[0];
-  out.set_data(allocator::malloc(out.nbytes()));
+
   auto& s = stream();
+  auto& encoder = cu::get_command_encoder(s);
+  out.set_data(cu::malloc_async(out.nbytes(), encoder.stream()));
 
   // Prepare the shapes, strides and axis arguments.
   Shape shape = remove_index(in.shape(), axis_);
@@ -154,7 +156,6 @@ void ArgReduce::eval_gpu(const std::vector<array>& inputs, array& out) {
   int32_t ndim = shape.size();
 
   // ArgReduce.
-  auto& encoder = cu::get_command_encoder(s);
   encoder.set_input_array(in);
   encoder.set_output_array(out);
   dispatch_real_types(in.dtype(), "ArgReduce", [&](auto type_tag) {
@@ -172,8 +173,8 @@ void ArgReduce::eval_gpu(const std::vector<array>& inputs, array& out) {
           num_blocks,
           block_dim(),
           0,
-          in.data<T>(),
-          out.data<uint32_t>(),
+          gpu_ptr<T>(in),
+          gpu_ptr<uint32_t>(out),
           out.size(),
           const_param(shape),
           const_param(in_strides),

mlx/backend/cuda/binary/binary.cuh

@@ -292,9 +292,9 @@ void binary_op_gpu_inplace(
                         {num_blocks_x, num_blocks_y},
                         block_dims,
                         0,
-                        a.data<InType>(),
-                        b.data<InType>(),
-                        out.data<OutType>(),
+                        gpu_ptr<InType>(a),
+                        gpu_ptr<InType>(b),
+                        gpu_ptr<OutType>(out),
                         rest,
                         const_param<dims_constant()>(shape),
                         const_param<dims_constant()>(a_strides),
@@ -310,9 +310,9 @@ void binary_op_gpu_inplace(
                       {num_blocks_x, num_blocks_y},
                       block_dims,
                       0,
-                      a.data<InType>(),
-                      b.data<InType>(),
-                      out.data<OutType>(),
+                      gpu_ptr<InType>(a),
+                      gpu_ptr<InType>(b),
+                      gpu_ptr<OutType>(out),
                       rest,
                       const_param(shape),
                       const_param(a_strides),
@@ -339,9 +339,9 @@ void binary_op_gpu_inplace(
                 num_blocks,
                 block_dims,
                 0,
-                a.data<InType>(),
-                b.data<InType>(),
-                out.data<OutType>(),
+                gpu_ptr<InType>(a),
+                gpu_ptr<InType>(b),
+                gpu_ptr<OutType>(out),
                 out.data_size());
           });
         }
@@ -365,7 +365,11 @@ void binary_op_gpu(
   auto& a = inputs[0];
   auto& b = inputs[1];
   auto bopt = get_binary_op_type(a, b);
-  set_binary_op_output_data(a, b, out, bopt);
+  auto& encoder = cu::get_command_encoder(s);
+
+  set_binary_op_output_data(a, b, out, bopt, [&](auto n) {
+    return cu::malloc_async(n, encoder.stream());
+  });
   binary_op_gpu_inplace<Op>(inputs, out, op, s);
 }
 

mlx/backend/cuda/binary_two.cu

@@ -245,14 +245,18 @@ void binary_two_op_gpu_inplace(
   auto& out_a = outputs[0];
   auto& out_b = outputs[1];
   auto bopt = get_binary_op_type(a, b);
-  set_binary_op_output_data(a, b, out_a, bopt);
-  set_binary_op_output_data(a, b, out_b, bopt);
+  auto& encoder = cu::get_command_encoder(s);
+  set_binary_op_output_data(a, b, out_a, bopt, [&](auto n) {
+    return cu::malloc_async(n, encoder.stream());
+  });
+  set_binary_op_output_data(a, b, out_b, bopt, [&](auto n) {
+    return cu::malloc_async(n, encoder.stream());
+  });
 
   if (out_a.size() == 0) {
     return;
   }
 
-  auto& encoder = cu::get_command_encoder(s);
   encoder.set_input_array(a);
   encoder.set_input_array(b);
   encoder.set_output_array(out_a);
@@ -313,10 +317,10 @@ void binary_two_op_gpu_inplace(
                         {num_blocks_x, num_blocks_y},
                         block_dims,
                         0,
-                        a.data<InType>(),
-                        b.data<InType>(),
-                        out_a.data<OutType>(),
-                        out_b.data<OutType>(),
+                        gpu_ptr<InType>(a),
+                        gpu_ptr<InType>(b),
+                        gpu_ptr<OutType>(out_a),
+                        gpu_ptr<OutType>(out_b),
                         rest,
                         const_param<dims_constant()>(shape),
                         const_param<dims_constant()>(a_strides),
@@ -332,10 +336,10 @@ void binary_two_op_gpu_inplace(
                       {num_blocks_x, num_blocks_y},
                       block_dims,
                       0,
-                      a.data<InType>(),
-                      b.data<InType>(),
-                      out_a.data<OutType>(),
-                      out_b.data<OutType>(),
+                      gpu_ptr<InType>(a),
+                      gpu_ptr<InType>(b),
+                      gpu_ptr<OutType>(out_a),
+                      gpu_ptr<OutType>(out_b),
                       rest,
                       const_param(shape),
                       const_param(a_strides),
@@ -366,10 +370,10 @@ void binary_two_op_gpu_inplace(
                 num_blocks,
                 block_dims,
                 0,
-                a.data<InType>(),
-                b.data<InType>(),
-                out_a.data<OutType>(),
-                out_b.data<OutType>(),
+                gpu_ptr<InType>(a),
+                gpu_ptr<InType>(b),
+                gpu_ptr<OutType>(out_a),
+                gpu_ptr<OutType>(out_b),
                 out_a.data_size());
           });
         }

mlx/backend/cuda/compiled.cpp

@@ -293,8 +293,13 @@ void Compiled::eval_gpu(
     }
   }
 
+  auto& encoder = cu::get_command_encoder(s);
+
   // Put outputs.
-  compiled_allocate_outputs(inputs, outputs, is_constant_, contiguous);
+  compiled_allocate_outputs(
+      inputs, outputs, is_constant_, contiguous, [&](auto n) {
+        return cu::malloc_async(n, encoder.stream());
+      });
   for (auto& x : outputs) {
     args.append(x);
   }
@@ -324,7 +329,6 @@ void Compiled::eval_gpu(
     kernel_name += fmt::format(
         "_strided<{}, {}, {}>", shape.size(), index_type, work_per_thread);
   }
-  auto& encoder = cu::get_command_encoder(s);
   for (const auto& in : inputs) {
     encoder.set_input_array(in);
   }
@@ -332,9 +336,9 @@ void Compiled::eval_gpu(
     encoder.set_output_array(out);
   }
 
-  auto kernel = mod.get_kernel(kernel_name);
+  auto [kernel, max_block_dims] = mod.get_kernel_and_dims(kernel_name);
   auto [num_blocks, block_dims] =
-      get_launch_args(outputs[0], large, work_per_thread);
+      get_launch_args(outputs[0], large, work_per_thread, max_block_dims);
   encoder.add_kernel_node(kernel, num_blocks, block_dims, 0, args.args());
 }
 

mlx/backend/cuda/conv.cpp

@@ -270,17 +270,16 @@ void Convolution::eval_gpu(const std::vector<array>& inputs, array& out_) {
   if (out_.size() == 0) {
     return;
   }
+  auto& s = stream();
+  auto& encoder = cu::get_command_encoder(s);
 
   assert(inputs.size() == 2);
   array in = inputs[0];
   array wt = inputs[1];
   array out = out_;
-  out.set_data(allocator::malloc(out.nbytes()));
+  out.set_data(cu::malloc_async(out.nbytes(), encoder.stream()));
   Dtype dtype = out.dtype();
 
-  auto& s = stream();
-  auto& encoder = cu::get_command_encoder(s);
-
   // Search cache.
   ConvCacheKey cache_key{
       encoder.device().cuda_device(),
@@ -382,15 +381,13 @@ void Convolution::eval_gpu(const std::vector<array>& inputs, array& out_) {
   }
 
   if (op_graph) {
-    // Setup inputs and outputs.
-    register_args(encoder, backend_type, in, wt, out, out_);
-
     // Find a plan for the graph and execute it.
     auto plan = find_cudnn_plan_from_op_graph(
         encoder.device().cudnn_handle(), backend_type, dtype, *op_graph);
-    if (!plan) {
-      throw std::runtime_error("[conv] Unable to find an execution plan.");
-    }
+    if (plan) {
+      // Setup inputs and outputs.
+      register_args(encoder, backend_type, in, wt, out, out_);
+
       auto [x, w, y] = dispatch_args(backend_type, in, wt, out);
       if (encode_cudnn_plan(encoder, *plan, {'x', 'w', 'y'}, x, w, y)) {
         conv_cache().emplace(
@@ -398,6 +395,7 @@ void Convolution::eval_gpu(const std::vector<array>& inputs, array& out_) {
         return;
       }
     }
+  }
 
   // Use fallback kernel for settings not supported by cuDNN.
   gemm_conv(

mlx/backend/cuda/conv/gemm_conv.cu

@@ -86,7 +86,7 @@ array unfold_inputs_nd(
     int mat_N,
     ConvParams<NDIM>& params) {
   array unfolded({mat_M, mat_K}, in.dtype(), nullptr, {});
-  unfolded.set_data(allocator::malloc(unfolded.nbytes()));
+  unfolded.set_data(cu::malloc_async(unfolded.nbytes(), encoder.stream()));
   encoder.add_temporary(unfolded);
 
   int filter_size = params.C;
@@ -118,8 +118,8 @@ array unfold_inputs_nd(
         num_blocks,
         block_dims,
         0,
-        in.data<DataType>(),
-        unfolded.data<DataType>(),
+        gpu_ptr<DataType>(in),
+        gpu_ptr<DataType>(unfolded),
         filter_size,
         out_pixels,
         params);

mlx/backend/cuda/conv/gemm_grouped_conv.cu

@@ -89,7 +89,7 @@ array grouped_unfold_transpose_inputs_nd(
     int mat_N,
     ConvParams<NDIM>& params) {
   array unfolded({mat_M, mat_K * params.groups}, in.dtype(), nullptr, {});
-  unfolded.set_data(allocator::malloc(unfolded.nbytes()));
+  unfolded.set_data(cu::malloc_async(unfolded.nbytes(), encoder.stream()));
   encoder.add_temporary(unfolded);
 
   int filter_size = params.C;
@@ -121,8 +121,8 @@ array grouped_unfold_transpose_inputs_nd(
         num_blocks,
         block_dims,
         0,
-        in.data<DataType>(),
-        unfolded.data<DataType>(),
+        gpu_ptr<DataType>(in),
+        gpu_ptr<DataType>(unfolded),
         filter_size,
         out_pixels,
         params);

mlx/backend/cuda/copy.cu

@@ -5,6 +5,22 @@
 
 namespace mlx::core {
 
+void copy_gpu(const array& in, array& out, CopyType ctype, const Stream& s) {
+  auto& encoder = cu::get_command_encoder(s);
+  bool donated = set_copy_output_data(in, out, ctype, [&](auto n) {
+    return cu::malloc_async(n, encoder.stream());
+  });
+  if (donated && in.dtype() == out.dtype()) {
+    // If the output has the same type as the input then there is nothing to
+    // copy, just use the buffer.
+    return;
+  }
+  if (ctype == CopyType::GeneralGeneral) {
+    ctype = CopyType::General;
+  }
+  copy_gpu_inplace(in, out, ctype, s);
+}
+
 void copy_gpu_inplace(
     const array& in,
     array& out,
@@ -87,11 +103,31 @@ void fill_gpu(const array& in, array& out, const Stream& s) {
   if (out.size() == 0) {
     return;
   }
-  out.set_data(allocator::malloc(out.nbytes()));
   auto& encoder = cu::get_command_encoder(s);
+  out.set_data(cu::malloc_async(out.nbytes(), encoder.stream()));
   encoder.set_input_array(in);
   encoder.set_output_array(out);
   copy_contiguous(encoder, CopyType::Scalar, in, out, 0, 0);
 }
 
+void reshape_gpu(const array& in, array& out, Stream s) {
+  auto [copy_necessary, out_strides] = prepare_reshape(in, out);
+  if (copy_necessary) {
+    auto& encoder = cu::get_command_encoder(s);
+    out.set_data(cu::malloc_async(out.nbytes(), encoder.stream()));
+    copy_gpu_inplace(
+        in,
+        out,
+        in.shape(),
+        in.strides(),
+        make_contiguous_strides(in.shape()),
+        0,
+        0,
+        CopyType::General,
+        s);
+  } else {
+    shared_buffer_reshape(in, out_strides, out);
+  }
+}
+
 } // namespace mlx::core

mlx/backend/cuda/copy/copy_contiguous.cu

@@ -77,8 +77,8 @@ void copy_contiguous(
             num_blocks,
             block_dims,
             0,
-            in.data<InType>() + in_offset,
-            out.data<OutType>() + out_offset,
+            gpu_ptr<InType>(in) + in_offset,
+            gpu_ptr<OutType>(out) + out_offset,
             out.data_size());
       });
     });

mlx/backend/cuda/copy/copy_general.cu

@@ -106,8 +106,8 @@ void copy_general(
             using InType = cuda_type_t<MLX_GET_TYPE(in_type_tag)>;
             using OutType = cuda_type_t<MLX_GET_TYPE(out_type_tag)>;
             using IdxT = std::conditional_t<large(), int64_t, int32_t>;
-            const InType* in_ptr = in.data<InType>() + offset_in;
-            OutType* out_ptr = out.data<OutType>() + offset_out;
+            const InType* in_ptr = gpu_ptr<InType>(in) + offset_in;
+            OutType* out_ptr = gpu_ptr<OutType>(out) + offset_out;
             int ndim = shape.size();
             size_t data_size = 1;
             for (auto& s : shape)

mlx/backend/cuda/copy/copy_general_dynamic.cu

@@ -69,8 +69,8 @@ void copy_general_dynamic(
             using InType = cuda_type_t<MLX_GET_TYPE(in_type_tag)>;
             using OutType = cuda_type_t<MLX_GET_TYPE(out_type_tag)>;
             using IdxT = std::conditional_t<large(), int64_t, int32_t>;
-            const InType* in_ptr = in.data<InType>() + offset_in;
-            OutType* out_ptr = out.data<OutType>() + offset_out;
+            const InType* in_ptr = gpu_ptr<InType>(in) + offset_in;
+            OutType* out_ptr = gpu_ptr<OutType>(out) + offset_out;
             int ndim = shape.size();
             if (ndim <= 3) {
               dispatch_1_2_3(ndim, [&](auto dims_constant) {
@@ -90,8 +90,8 @@ void copy_general_dynamic(
                     const_param<dims_constant()>(shape),
                     const_param<dims_constant()>(strides_in),
                     const_param<dims_constant()>(strides_out),
-                    dynamic_offset_in.data<int64_t>(),
-                    dynamic_offset_out.data<int64_t>());
+                    gpu_ptr<int64_t>(dynamic_offset_in),
+                    gpu_ptr<int64_t>(dynamic_offset_out));
               });
             } else { // ndim >= 4
               auto [num_blocks, block_dims] = get_launch_args(out, large());
@@ -107,8 +107,8 @@ void copy_general_dynamic(
                   const_param(strides_in),
                   const_param(strides_out),
                   ndim,
-                  dynamic_offset_in.data<int64_t>(),
-                  dynamic_offset_out.data<int64_t>());
+                  gpu_ptr<int64_t>(dynamic_offset_in),
+                  gpu_ptr<int64_t>(dynamic_offset_out));
             }
           });
     });

mlx/backend/cuda/copy/copy_general_input.cu

@@ -92,8 +92,8 @@ void copy_general_input(
             using InType = cuda_type_t<MLX_GET_TYPE(in_type_tag)>;
             using OutType = cuda_type_t<MLX_GET_TYPE(out_type_tag)>;
             using IdxT = std::conditional_t<large(), int64_t, int32_t>;
-            const InType* in_ptr = in.data<InType>() + offset_in;
-            OutType* out_ptr = out.data<OutType>() + offset_out;
+            const InType* in_ptr = gpu_ptr<InType>(in) + offset_in;
+            OutType* out_ptr = gpu_ptr<OutType>(out) + offset_out;
             int ndim = shape.size();
             int work_per_thread = 1;
             auto dim0 = ndim > 0 ? shape.back() : 1;

mlx/backend/cuda/cuda_utils.h

@@ -0,0 +1,82 @@
+// Copyright © 2025 Apple Inc.
+
+#pragma once
+
+#include <cublasLt.h>
+#include <cuda.h>
+#include <cuda_runtime.h>
+
+namespace mlx::core {
+
+// Throw exception if the cuda API does not succeed.
+void check_cublas_error(const char* name, cublasStatus_t err);
+void check_cuda_error(const char* name, cudaError_t err);
+void check_cuda_error(const char* name, CUresult err);
+
+// The macro version that prints the command that failed.
+#define CHECK_CUBLAS_ERROR(cmd) check_cublas_error(#cmd, (cmd))
+#define CHECK_CUDA_ERROR(cmd) check_cuda_error(#cmd, (cmd))
+
+// Base class for RAII managed CUDA resources.
+template <typename Handle, cudaError_t (*Destroy)(Handle)>
+class CudaHandle {
+ public:
+  CudaHandle(Handle handle = nullptr) : handle_(handle) {}
+
+  CudaHandle(CudaHandle&& other) : handle_(other.handle_) {
+    assert(this != &other);
+    other.handle_ = nullptr;
+  }
+
+  ~CudaHandle() {
+    reset();
+  }
+
+  CudaHandle(const CudaHandle&) = delete;
+  CudaHandle& operator=(const CudaHandle&) = delete;
+
+  CudaHandle& operator=(CudaHandle&& other) {
+    assert(this != &other);
+    reset();
+    std::swap(handle_, other.handle_);
+    return *this;
+  }
+
+  void reset() {
+    if (handle_ != nullptr) {
+      CHECK_CUDA_ERROR(Destroy(handle_));
+      handle_ = nullptr;
+    }
+  }
+
+  operator Handle() const {
+    return handle_;
+  }
+
+ protected:
+  Handle handle_;
+};
+
+namespace cu {
+class Device;
+}; // namespace cu
+
+// Wrappers of CUDA resources.
+class CudaGraph : public CudaHandle<cudaGraph_t, cudaGraphDestroy> {
+ public:
+  using CudaHandle::CudaHandle;
+  explicit CudaGraph(cu::Device& device);
+  void end_capture(cudaStream_t stream);
+};
+
+class CudaGraphExec : public CudaHandle<cudaGraphExec_t, cudaGraphExecDestroy> {
+ public:
+  void instantiate(cudaGraph_t graph);
+};
+
+class CudaStream : public CudaHandle<cudaStream_t, cudaStreamDestroy> {
+ public:
+  explicit CudaStream(cu::Device& device);
+};
+
+} // namespace mlx::core

mlx/backend/cuda/cudnn_utils.cpp

@@ -133,13 +133,13 @@ bool prepare_cudnn_plan(
     F&& execute) {
   int workspace_size = plan.getWorkspaceSize();
   array workspace(
-      workspace_size > 0 ? allocator::malloc(workspace_size)
+      workspace_size > 0 ? cu::malloc_async(workspace_size, encoder.stream())
                          : allocator::Buffer(nullptr),
       {workspace_size},
       uint8);
 
   auto args = cudnn_frontend::VariantPackBuilder()
-                  .setWorkspacePointer(workspace.data<void>())
+                  .setWorkspacePointer(gpu_ptr<void>(workspace))
                   .setDataPointers(num_args, data_ptrs)
                   .setUids(num_args, uids)
                   .build();
@@ -210,6 +210,9 @@ std::optional<cudnn_frontend::ExecutionPlan> find_cudnn_plan_from_op_graph(
     Dtype dtype,
     cudnn_frontend::OperationGraph& op_graph) {
   auto engine_configs = get_cudnn_engine_configs(backend_type, dtype, op_graph);
+  if (engine_configs.empty()) {
+    return std::nullopt;
+  }
   return find_cudnn_plan_from_engine_configs(handle, engine_configs, op_graph);
 }
 

mlx/backend/cuda/cudnn_utils.h

@@ -3,6 +3,7 @@
 #pragma once
 
 #include "mlx/array.h"
+#include "mlx/backend/cuda/allocator.h"
 #include "mlx/backend/cuda/device/config.h"
 #include "mlx/backend/cuda/utils.h"
 #include "mlx/dtype_utils.h"
@@ -23,7 +24,7 @@ class CommandEncoder;
 // Return pointer alignment of |x|'s data.
 inline uint8_t get_alignment(const array& x) {
   uint8_t alignment = 1;
-  uintptr_t address = reinterpret_cast<uintptr_t>(x.data<void>());
+  uintptr_t address = reinterpret_cast<uintptr_t>(gpu_ptr<void>(x));
   for (; alignment < 32; alignment *= 2) {
     if (address % (alignment * 2)) {
       return alignment;
@@ -56,7 +57,7 @@ inline std::array<T, MAX_NDIM> vector_key(const Vec<T>& vec) {
 
 // Helpers used by get_data_ptrs to get pointers.
 inline void* get_data_ptr(const array& arr) {
-  return const_cast<void*>(arr.data<void>());
+  return const_cast<void*>(gpu_ptr<void>(arr));
 }
 
 template <typename T, typename = std::enable_if_t<std::is_scalar_v<T>>>

mlx/backend/cuda/custom_kernel.cpp

@@ -279,6 +279,7 @@ void CustomKernel::eval_gpu(
     std::vector<array>& outputs) {
   nvtx3::scoped_range r("CustomKernel::eval_gpu");
   auto& s = stream();
+  auto& encoder = cu::get_command_encoder(s);
 
   std::vector<array> copies;
 
@@ -288,7 +289,7 @@ void CustomKernel::eval_gpu(
       copies.emplace_back(init_value_.value(), out.dtype());
       fill_gpu(copies.back(), out, s);
     } else {
-      out.set_data(allocator::malloc(out.nbytes()));
+      out.set_data(cu::malloc_async(out.nbytes(), encoder.stream()));
     }
   }
 
@@ -356,7 +357,6 @@ void CustomKernel::eval_gpu(
   dim3 grid((gx + tx - 1) / tx, (gy + ty - 1) / ty, (gz + tz - 1) / tz);
 
   // Call the kernel
-  auto& encoder = cu::get_command_encoder(s);
   for (const auto& in : checked_inputs) {
     encoder.set_input_array(in);
   }

mlx/backend/cuda/device.cpp

@@ -14,10 +14,6 @@ namespace mlx::core::cu {
 
 namespace {
 
-// Can be tuned with MLX_MAX_OPS_PER_BUFFER
-// This should be less than 255
-constexpr int default_max_nodes_per_graph = 20;
-
 #define CHECK_CUDNN_ERROR(cmd) check_cudnn_error(#cmd, (cmd))
 
 void check_cudnn_error(const char* name, cudnnStatus_t err) {
@@ -68,8 +64,8 @@ Device::~Device() {
 
 void Device::make_current() {
   // We need to set/get current CUDA device very frequently, cache it to reduce
-  // actual calls of CUDA APIs. This function assumes single-thread in host.
-  static int current = 0;
+  // actual calls of CUDA APIs.
+  static thread_local int current = 0;
   if (current != device_) {
     CHECK_CUDA_ERROR(cudaSetDevice(device_));
     current = device_;
@@ -95,6 +91,7 @@ CommandEncoder::CaptureContext::CaptureContext(CommandEncoder& enc) : enc(enc) {
 
 CommandEncoder::CaptureContext::~CaptureContext() {
   if (!use_cuda_graphs()) {
+    enc.node_count_++;
     return;
   }
 
@@ -196,6 +193,7 @@ CommandEncoder::CommandEncoder(Device& d)
     : device_(d),
       stream_(d),
       graph_(d),
+      worker_(d),
       graph_cache_("MLX_CUDA_GRAPH_CACHE_SIZE", /* default_capacity */ 400) {}
 
 void CommandEncoder::add_completed_handler(std::function<void()> task) {
@@ -220,12 +218,6 @@ void CommandEncoder::set_output_array(const array& arr) {
   active_outputs_.push_back(id);
 }
 
-void CommandEncoder::maybe_commit() {
-  if (node_count_ >= env::max_ops_per_buffer(default_max_nodes_per_graph)) {
-    commit();
-  }
-}
-
 void CommandEncoder::add_kernel_node(
     void* func,
     dim3 grid_dim,
@@ -233,6 +225,7 @@ void CommandEncoder::add_kernel_node(
     uint32_t smem_bytes,
     void** params) {
   if (!use_cuda_graphs()) {
+    node_count_++;
     CHECK_CUDA_ERROR(cudaLaunchKernel(
         func, grid_dim, block_dim, params, smem_bytes, stream()));
     return;
@@ -253,6 +246,7 @@ void CommandEncoder::add_kernel_node(
     uint32_t smem_bytes,
     void** params) {
   if (!use_cuda_graphs()) {
+    node_count_++;
     CHECK_CUDA_ERROR(cuLaunchKernel(
         func,
         grid_dim.x,
@@ -295,6 +289,7 @@ void CommandEncoder::add_kernel_node(const CUDA_KERNEL_NODE_PARAMS& params) {
 
 void CommandEncoder::add_graph_node(cudaGraph_t child) {
   if (!use_cuda_graphs()) {
+    node_count_++;
     CudaGraphExec graph_exec;
     graph_exec.instantiate(child);
     device_.make_current();
@@ -306,12 +301,16 @@ void CommandEncoder::add_graph_node(cudaGraph_t child) {
   insert_graph_dependencies(GraphNode{node, 'G'});
 }
 
+int CommandEncoder::get_num_ops() {
+  return node_count_;
+}
+
 void CommandEncoder::commit() {
   nvtx3::scoped_range r("CommandEncoder::commit");
   if (!temporaries_.empty()) {
     add_completed_handler([temporaries = std::move(temporaries_)]() {});
   }
-  if (node_count_ > 0) {
+  if (use_cuda_graphs() && node_count_ > 0) {
     if (!from_nodes_.empty()) {
       CHECK_CUDA_ERROR(cudaGraphAddDependencies(
           graph_,
@@ -354,7 +353,6 @@ void CommandEncoder::commit() {
     CHECK_CUDA_ERROR(cudaGraphLaunch(graph_exec, stream_));
 
     // Reset state
-    node_count_ = 0;
     graph_node_count_ = 0;
     empty_node_count_ = 0;
     from_nodes_.clear();
@@ -366,6 +364,7 @@ void CommandEncoder::commit() {
 
   // Put completion handlers in a batch.
   worker_.commit(stream_);
+  node_count_ = 0;
 }
 
 void CommandEncoder::synchronize() {

mlx/backend/cuda/device.h

@@ -3,6 +3,7 @@
 #pragma once
 
 #include "mlx/array.h"
+#include "mlx/backend/cuda/allocator.h"
 #include "mlx/backend/cuda/lru_cache.h"
 #include "mlx/backend/cuda/worker.h"
 #include "mlx/stream.h"
@@ -83,7 +84,7 @@ class CommandEncoder {
   }
 
   void add_completed_handler(std::function<void()> task);
-  void maybe_commit();
+  int get_num_ops();
   void commit();
 
   Device& device() {
@@ -140,7 +141,7 @@ class Device {
   Device(const Device&) = delete;
   Device& operator=(const Device&) = delete;
 
-  // Make this device the current cuda device, required by some cuda calls.
+  // Make this device the current cuda device, this method is thread-safe.
   void make_current();
 
   CommandEncoder& get_command_encoder(Stream s);

mlx/backend/cuda/device/unary_ops.cuh

@@ -2,6 +2,8 @@
 
 #pragma once
 
+#include <cuda_fp8.h>
+
 #include "mlx/backend/cuda/device/fp16_math.cuh"
 #include "mlx/backend/cuda/device/utils.cuh"
 
@@ -257,8 +259,8 @@ struct Round {
 struct Sigmoid {
   template <typename T>
   __device__ T operator()(T x) {
-    T y = 1 / (1 + exp(-abs(x)));
-    return (x < 0) ? 1 - y : y;
+    T y = 1 / (1 + exp(abs(x)));
+    return (x < 0) ? y : 1 - y;
   }
 };
 
@@ -334,4 +336,17 @@ struct Tanh {
   }
 };
 
+struct ToFP8 {
+  template <typename T>
+  __device__ uint8_t operator()(T x) {
+    return __nv_fp8_e4m3(x).__x;
+  }
+};
+
+struct FromFP8 {
+  __device__ float operator()(uint8_t x) {
+    return float(*(__nv_fp8_e4m3*)(&x));
+  }
+};
+
 } // namespace mlx::core::cu

mlx/backend/cuda/device/utils.cuh

@@ -1,6 +1,6 @@
 // Copyright © 2025 Apple Inc.
 
-// This file must not include any host-only code, utilies that work under both
+// This file must not include any host-only code, utilities that work under both
 // host and device can be put here.
 //
 // See more about the requirements at:
@@ -202,7 +202,7 @@ struct Limits<
   }
 };
 
-// CUDA 11 does not have host side arithmatic operators for half types.
+// CUDA 11 does not have host side arithmetic operators for half types.
 template <typename T>
 struct Limits<
     T,

mlx/backend/cuda/distributed.cu

@@ -15,8 +15,10 @@ void AllReduce::eval_gpu(
   assert(inputs.size() == 1);
   assert(outputs.size() == 1);
 
-  auto set_input_output =
-      [s = stream()](const array& in, array& out) -> std::pair<array, array> {
+  auto& s = stream();
+  auto& encoder = cu::get_command_encoder(s);
+  auto set_input_output = [&](const array& in,
+                              array& out) -> std::pair<array, array> {
     if (!in.flags().row_contiguous) {
       copy_gpu(in, out, CopyType::General, s);
       return {out, out};
@@ -24,19 +26,17 @@ void AllReduce::eval_gpu(
       out.copy_shared_buffer(in);
       return {in, out};
     } else {
-      out.set_data(allocator::malloc(out.nbytes()));
+      out.set_data(cu::malloc_async(out.nbytes(), encoder.stream()));
       return {in, out};
     }
   };
 
   auto [input, output] = set_input_output(inputs[0], outputs[0]);
 
-  auto& encoder = cu::get_command_encoder(stream());
   encoder.set_input_array(input);
   encoder.set_output_array(output);
 
   auto capture = encoder.capture_context();
-  auto& s = stream();
 
   switch (reduce_type_) {
     case Sum:

mlx/backend/cuda/eval.cpp

@@ -5,19 +5,24 @@
 #include "mlx/backend/cuda/device.h"
 #include "mlx/backend/gpu/available.h"
 #include "mlx/primitives.h"
+#include "mlx/scheduler.h"
 
 #include <nvtx3/nvtx3.hpp>
 
 namespace mlx::core::gpu {
 
+// Can be tuned with MLX_MAX_OPS_PER_BUFFER
+constexpr int default_max_nodes_per_graph = 20;
+
 bool is_available() {
   return true;
 }
 
 void new_stream(Stream s) {
-  // Force initalization of CUDA by creating an event, so the CUDA runtime and
-  // our CUDA event pool get destroyed last.
-  cu::CudaEvent(cudaEventDefault);
+  // Force initalization of CUDA, so CUDA runtime get destroyed at last.
+  cudaFree(nullptr);
+  // Make sure CUDA event pool get destroyed after device and stream.
+  cu::CudaEvent::init_pool();
   // Ensure the static stream objects get created.
   cu::get_command_encoder(s);
 }
@@ -35,7 +40,8 @@ void eval(array& arr) {
     arr.primitive().eval_gpu(arr.inputs(), outputs);
   }
 
-  auto& encoder = cu::get_command_encoder(arr.primitive().stream());
+  auto& stream = arr.primitive().stream();
+  auto& encoder = cu::get_command_encoder(stream);
   // Keep used buffers alive until kernel finishes running.
   for (auto& in : arr.inputs()) {
     // Except for the donated one.
@@ -46,7 +52,14 @@ void eval(array& arr) {
   for (auto& s : arr.siblings()) {
     encoder.add_temporary(s);
   }
-  encoder.maybe_commit();
+
+  if (encoder.get_num_ops() >=
+      env::max_ops_per_buffer(default_max_nodes_per_graph)) {
+    scheduler::notify_new_task(stream);
+    encoder.add_completed_handler(
+        [stream]() { scheduler::notify_task_completion(stream); });
+    encoder.commit();
+  }
 }
 
 void finalize(Stream s) {

mlx/backend/cuda/event.cu

@@ -22,11 +22,15 @@ namespace cu {
 namespace {
 
 // Manage cached cudaEvent_t objects.
-struct CudaEventPool {
-  static CudaEventHandle create(int flags) {
-    auto& cache = cache_for(flags);
+class CudaEventPool {
+ public:
+  CudaEventHandle create(Device& d, int flags) {
+    if (!on_creation_thread()) {
+      return CudaEventHandle(d, flags);
+    }
+    auto& cache = cache_for(d, flags);
     if (cache.empty()) {
-      return CudaEventHandle(flags);
+      return CudaEventHandle(d, flags);
     } else {
       CudaEventHandle ret = std::move(cache.back());
       cache.pop_back();
@@ -34,54 +38,89 @@ struct CudaEventPool {
     }
   }
 
-  static void release(CudaEventHandle event) {
-    cache_for(event.flags).push_back(std::move(event));
+  void release(CudaEventHandle event) {
+    if (!on_creation_thread()) {
+      // Event will be destroyed directly instead of getting moved to cache.
+      return;
+    }
+    cache_for(event.device, event.flags).push_back(std::move(event));
   }
 
-  static std::vector<CudaEventHandle>& cache_for(int flags) {
-    static std::map<int, std::vector<CudaEventHandle>> cache;
-    return cache[flags];
+ private:
+  std::vector<CudaEventHandle>& cache_for(Device& d, int flags) {
+    return cache_[d.cuda_device()][flags];
   }
+
+  bool on_creation_thread() {
+    return std::this_thread::get_id() == thread_id_;
+  }
+
+  // The CudaEvent may be created and destroyed on different threads (for
+  // example when waiting on GPU work in CPU stream), we don't want to make
+  // the cache thread-safe as it adds overhead, so we just skip cache when
+  // using events in worker threads.
+  std::thread::id thread_id_{std::this_thread::get_id()};
+
+  // {device: {flags: [events]}}
+  std::map<int, std::map<int, std::vector<CudaEventHandle>>> cache_;
 };
 
+CudaEventPool& cuda_event_pool() {
+  static CudaEventPool pool;
+  return pool;
+}
+
 } // namespace
 
-CudaEventHandle::CudaEventHandle(int flags) : flags(flags) {
+CudaEventHandle::CudaEventHandle(Device& d, int flags)
+    : device(d), flags(flags) {
+  device.make_current();
   CHECK_CUDA_ERROR(cudaEventCreateWithFlags(&handle_, flags));
   assert(handle_ != nullptr);
 }
 
-CudaEvent::CudaEvent(int flags) : event_(CudaEventPool::create(flags)) {}
+CudaEvent::CudaEvent(Device& d, int flags)
+    : event_(cuda_event_pool().create(d, flags)) {}
 
 CudaEvent::~CudaEvent() {
-  CudaEventPool::release(std::move(event_));
+  cuda_event_pool().release(std::move(event_));
 }
 
 void CudaEvent::wait() {
   nvtx3::scoped_range r("cu::CudaEvent::wait");
+  event_.device.make_current();
   cudaEventSynchronize(event_);
 }
 
 void CudaEvent::wait(cudaStream_t stream) {
+  event_.device.make_current();
   cudaStreamWaitEvent(stream, event_);
 }
 
 void CudaEvent::record(cudaStream_t stream) {
+  event_.device.make_current();
   cudaEventRecord(event_, stream);
 }
 
 bool CudaEvent::completed() const {
+  // Note: cudaEventQuery can be safely called from any device.
   return cudaEventQuery(event_) == cudaSuccess;
 }
 
+// static
+void CudaEvent::init_pool() {
+  cuda_event_pool();
+}
+
 // Wraps CudaEvent with a few features:
 // 1. The class can be copied.
 // 2. Make wait/record work with CPU streams.
 // 3. Add checks for waiting on un-recorded event.
 class CopyableCudaEvent {
  public:
-  CopyableCudaEvent()
+  explicit CopyableCudaEvent(Device& d)
       : event_(std::make_shared<CudaEvent>(
+            d,
             cudaEventDisableTiming | cudaEventBlockingSync)) {}
 
   void wait() {
@@ -245,7 +284,7 @@ struct EventImpl {
       nvtx3::mark("Using slow AtomicEvent");
       atomic = std::make_unique<cu::AtomicEvent>();
     } else {
-      cuda = std::make_unique<cu::CopyableCudaEvent>();
+      cuda = std::make_unique<cu::CopyableCudaEvent>(cu::device(s.device));
     }
   }
 };

mlx/backend/cuda/event.h

@@ -13,9 +13,12 @@
 
 namespace mlx::core::cu {
 
+class Device;
+
 // RAII-managed move-only wrapper of cudaEvent_t.
 struct CudaEventHandle : public CudaHandle<cudaEvent_t, cudaEventDestroy> {
-  CudaEventHandle(int flags);
+  CudaEventHandle(Device& d, int flags);
+  Device& device;
   int flags;
 };
 
@@ -23,7 +26,7 @@ struct CudaEventHandle : public CudaHandle<cudaEvent_t, cudaEventDestroy> {
 // on GPU stream in CPU stream, but can not wait on CPU stream.
 class CudaEvent {
  public:
-  explicit CudaEvent(int flags);
+  CudaEvent(Device& d, int flags);
   ~CudaEvent();
 
   CudaEvent(CudaEvent&&) = default;
@@ -40,6 +43,9 @@ class CudaEvent {
   // returns true if record() has not been called.
   bool completed() const;
 
+  // Internal: make sure event pool is initialized.
+  static void init_pool();
+
  private:
   CudaEventHandle event_;
 };

mlx/backend/cuda/gemms/cublas_gemm.cpp

@@ -50,8 +50,10 @@ cublasComputeType_t dtype_to_compute_type(Dtype dtype) {
       return mlx::core::env::enable_tf32() ? CUBLAS_COMPUTE_32F_FAST_TF32
                                            : CUBLAS_COMPUTE_32F;
     case float64:
-    case complex64:
       return CUBLAS_COMPUTE_64F;
+    case complex64:
+      return mlx::core::env::enable_tf32() ? CUBLAS_COMPUTE_32F_FAST_TF32
+                                           : CUBLAS_COMPUTE_32F;
     default:
       throw std::runtime_error(fmt::format(
           "Unsupported dtype in CublasGemm: {}.", dtype_to_string(dtype)));
@@ -126,12 +128,13 @@ CublasGemm::CublasGemm(
       N_(b_cols) {
   heuristic_.state = CUBLAS_STATUS_NOT_INITIALIZED;
 
-  auto scale_type = dtype_to_cublas_type(dtype);
+  scale_type_ = dtype_to_cublas_type(dtype);
   if (dtype == bfloat16 || dtype == float16) {
-    scale_type = CUDA_R_32F;
+    scale_type_ = CUDA_R_32F;
   }
+
   CHECK_CUBLAS_ERROR(cublasLtMatmulDescCreate(
-      &matmul_desc_, dtype_to_compute_type(dtype), scale_type));
+      &matmul_desc_, dtype_to_compute_type(dtype), scale_type_));
   int32_t pointer_mode = CUBLASLT_POINTER_MODE_HOST;
   CHECK_CUBLAS_ERROR(cublasLtMatmulDescSetAttribute(
       matmul_desc_,
@@ -238,7 +241,7 @@ void CublasGemm::set_bias(cu::CommandEncoder& encoder, const array& bias) {
       CUBLASLT_MATMUL_DESC_EPILOGUE,
       &epilogue,
       sizeof(epilogue)));
-  auto* bias_ptr = bias.data<void>();
+  auto* bias_ptr = gpu_ptr<void>(bias);
   CHECK_CUBLAS_ERROR(cublasLtMatmulDescSetAttribute(
       matmul_desc_,
       CUBLASLT_MATMUL_DESC_BIAS_POINTER,
@@ -275,9 +278,9 @@ void CublasGemm::run(
 
   execute(
       encoder,
-      out.data<void>(),
-      a.data<void>(),
-      b.data<void>(),
+      gpu_ptr<void>(out),
+      gpu_ptr<void>(a),
+      gpu_ptr<void>(b),
       nullptr,
       alpha);
 }
@@ -318,10 +321,10 @@ void CublasGemm::run(
 
   execute(
       encoder,
-      out.data<void>(),
-      a.data<void>(),
-      b.data<void>(),
-      c.data<void>(),
+      gpu_ptr<void>(out),
+      gpu_ptr<void>(a),
+      gpu_ptr<void>(b),
+      gpu_ptr<void>(c),
       alpha,
       beta);
 }
@@ -352,28 +355,38 @@ void CublasGemm::execute(
     }
   }
 
+  const void* alpha_ptr = &alpha;
+  const void* beta_ptr = &beta;
+  complex64_t alpha_c, beta_c;
+  if (scale_type_ == CUDA_C_32F) {
+    alpha_c = complex64_t{alpha, 0.0f};
+    beta_c = complex64_t{beta, 0.0f};
+    alpha_ptr = &alpha_c;
+    beta_ptr = &beta_c;
+  }
+
   void* workspace_ptr = nullptr;
   if (heuristic_.workspaceSize > 0) {
     // Ensure workspace is 256-byte aligned
     int nbytes = cuda::ceil_div(heuristic_.workspaceSize, 256) * 256;
     array workspace(
-        allocator::malloc(nbytes),
+        cu::malloc_async(nbytes, encoder.stream()),
         {static_cast<int>(heuristic_.workspaceSize)},
         int8);
     encoder.add_temporary(workspace);
-    workspace_ptr = workspace.data<void>();
+    workspace_ptr = gpu_ptr<void>(workspace);
   }
 
   auto capture = encoder.capture_context();
   CHECK_CUBLAS_ERROR(cublasLtMatmul(
       handle_,
       matmul_desc_,
-      &alpha,
+      alpha_ptr,
       b, // a and b are swapped
       a_desc_,
       a,
       b_desc_,
-      &beta,
+      beta_ptr,
       c ? c : out,
       c ? c_desc_ : out_desc_,
       out,

mlx/backend/cuda/gemms/cublas_gemm.h

@@ -115,6 +115,7 @@ class CublasGemm {
 
   uint64_t M_;
   uint64_t N_;
+  cudaDataType_t scale_type_;
   cublasLtMatmulPreference_t pref_{nullptr};
   cublasLtHandle_t handle_{nullptr};
   cublasLtMatmulDesc_t matmul_desc_{nullptr};

mlx/backend/cuda/gemms/cublas_gemm_batched_12_0.cpp

@@ -25,9 +25,10 @@ void CublasGemm::run_batched(
   for (size_t i = 0; i < nbatch; ++i) {
     execute(
         encoder,
-        out.data<int8_t>() + out.itemsize() * i * batch_shape.back() * M_ * N_,
-        a.data<int8_t>() + a.itemsize() * a_it.loc,
-        b.data<int8_t>() + b.itemsize() * b_it.loc,
+        gpu_ptr<int8_t>(out) +
+            out.itemsize() * i * batch_shape.back() * M_ * N_,
+        gpu_ptr<int8_t>(a) + a.itemsize() * a_it.loc,
+        gpu_ptr<int8_t>(b) + b.itemsize() * b_it.loc,
         nullptr,
         alpha);
     a_it.step();
@@ -60,10 +61,11 @@ void CublasGemm::run_batched(
   for (size_t i = 0; i < nbatch; ++i) {
     execute(
         encoder,
-        out.data<int8_t>() + out.itemsize() * i * batch_shape.back() * M_ * N_,
-        a.data<int8_t>() + a.itemsize() * a_it.loc,
-        b.data<int8_t>() + b.itemsize() * b_it.loc,
-        c.data<int8_t>() + c.itemsize() * c_it.loc,
+        gpu_ptr<int8_t>(out) +
+            out.itemsize() * i * batch_shape.back() * M_ * N_,
+        gpu_ptr<int8_t>(a) + a.itemsize() * a_it.loc,
+        gpu_ptr<int8_t>(b) + b.itemsize() * b_it.loc,
+        gpu_ptr<int8_t>(c) + c.itemsize() * c_it.loc,
         alpha,
         beta);
     a_it.step();

mlx/backend/cuda/gemms/cublas_gemm_batched_12_9.cu

@@ -163,7 +163,7 @@ void CublasGemm::run_batched(
 
   // Launch kernel to set device offsets
   auto pointers = array(
-      allocator::malloc(batch_count * sizeof(void*) * 3),
+      cu::malloc_async(batch_count * sizeof(void*) * 3, encoder.stream()),
       {batch_count * 3},
       uint64);
 
@@ -183,10 +183,10 @@ void CublasGemm::run_batched(
           num_blocks,
           block_dims,
           0,
-          pointers.data<int8_t*>(),
-          a.data<int8_t>(),
-          b.data<int8_t>(),
-          out.data<int8_t>(),
+          gpu_ptr<int8_t*>(pointers),
+          gpu_ptr<int8_t>(a),
+          gpu_ptr<int8_t>(b),
+          gpu_ptr<int8_t>(out),
           item_size,
           const_param<ndim_constant()>(batch_shape),
           const_param<ndim_constant()>(a_batch_strides),
@@ -200,10 +200,10 @@ void CublasGemm::run_batched(
         num_blocks,
         block_dims,
         0,
-        pointers.data<int8_t*>(),
-        a.data<int8_t>(),
-        b.data<int8_t>(),
-        out.data<int8_t>(),
+        gpu_ptr<int8_t*>(pointers),
+        gpu_ptr<int8_t>(a),
+        gpu_ptr<int8_t>(b),
+        gpu_ptr<int8_t>(out),
         item_size,
         const_param(batch_shape),
         const_param(a_batch_strides),
@@ -219,7 +219,7 @@ void CublasGemm::run_batched(
   encoder.set_input_array(b);
   encoder.set_output_array(out);
 
-  auto a_pointers = pointers.data<int8_t*>();
+  auto a_pointers = gpu_ptr<int8_t*>(pointers);
   auto b_pointers = a_pointers + batch_count;
   auto out_pointers = b_pointers + batch_count;
   execute(
@@ -251,7 +251,7 @@ void CublasGemm::run_batched(
 
   // Launch kernel to set device offsets
   auto pointers = array(
-      allocator::malloc(batch_count * sizeof(uint64_t) * 4),
+      cu::malloc_async(batch_count * sizeof(uint64_t) * 4, encoder.stream()),
       {batch_count * 4},
       uint64);
 
@@ -271,11 +271,11 @@ void CublasGemm::run_batched(
           num_blocks,
           block_dims,
           0,
-          pointers.data<int8_t*>(),
-          a.data<int8_t>(),
-          b.data<int8_t>(),
-          c.data<int8_t>(),
-          out.data<int8_t>(),
+          gpu_ptr<int8_t*>(pointers),
+          gpu_ptr<int8_t>(a),
+          gpu_ptr<int8_t>(b),
+          gpu_ptr<int8_t>(c),
+          gpu_ptr<int8_t>(out),
           item_size,
           const_param<ndim_constant()>(batch_shape),
           const_param<ndim_constant()>(a_batch_strides),
@@ -290,11 +290,11 @@ void CublasGemm::run_batched(
         num_blocks,
         block_dims,
         0,
-        pointers.data<int8_t*>(),
-        a.data<int8_t>(),
-        b.data<int8_t>(),
-        c.data<int8_t>(),
-        out.data<int8_t>(),
+        gpu_ptr<int8_t*>(pointers),
+        gpu_ptr<int8_t>(a),
+        gpu_ptr<int8_t>(b),
+        gpu_ptr<int8_t>(c),
+        gpu_ptr<int8_t>(out),
         item_size,
         const_param(batch_shape),
         const_param(a_batch_strides),
@@ -312,7 +312,7 @@ void CublasGemm::run_batched(
   encoder.set_input_array(c);
   encoder.set_output_array(out);
 
-  auto a_pointers = pointers.data<int8_t*>();
+  auto a_pointers = gpu_ptr<int8_t*>(pointers);
   auto b_pointers = a_pointers + batch_count;
   auto c_pointers = b_pointers + batch_count;
   auto out_pointers = c_pointers + batch_count;

mlx/backend/cuda/gemms/gemv.cu

@@ -13,6 +13,37 @@ namespace cg = cooperative_groups;
 
 static constexpr int rows_per_block = 8;
 
+// Accumulator type selection per input element type T.
+template <typename T>
+struct GemvAccType {
+  using type = T;
+};
+
+template <>
+struct GemvAccType<__half> {
+  using type = float;
+};
+
+template <>
+struct GemvAccType<__nv_bfloat16> {
+  using type = float;
+};
+
+template <>
+struct GemvAccType<float> {
+  using type = float;
+};
+
+template <>
+struct GemvAccType<double> {
+  using type = double;
+};
+
+template <>
+struct GemvAccType<cu::complex64_t> {
+  using type = cu::complex64_t;
+};
+
 template <typename T, int rows_per_block, int n_per_thread>
 __device__ void
 gemv_impl(const T* mat, const T* vec, T* out, int rows, int cols) {
@@ -24,7 +55,8 @@ gemv_impl(const T* mat, const T* vec, T* out, int rows, int cols) {
   int row = g_idx.x * rows_per_block + t_idx.y;
 
   if (row < rows) {
-    float sum = 0.0f;
+    using Acc = typename GemvAccType<T>::type;
+    Acc sum = Acc(0);
     for (int col = n_per_thread * warp.thread_rank(); col < cols;
          col += (WARP_SIZE * n_per_thread)) {
       auto local_mat =
@@ -32,12 +64,11 @@ gemv_impl(const T* mat, const T* vec, T* out, int rows, int cols) {
       auto local_vec = unsafe_load_vector<n_per_thread>(vec + col, 0);
 #pragma unroll
       for (int j = 0; j < n_per_thread; ++j) {
-        sum +=
-            static_cast<float>(local_mat[j]) * static_cast<float>(local_vec[j]);
+        sum += static_cast<Acc>(local_mat[j]) * static_cast<Acc>(local_vec[j]);
       }
     }
 
-    sum = cg::reduce(warp, sum, cg::plus<float>{});
+    sum = cg::reduce(warp, sum, cg::plus<Acc>{});
     if (warp.thread_rank() == 0) {
       out[row] = static_cast<T>(sum);
     }
@@ -107,7 +138,7 @@ void gemv(
   encoder.set_input_array(a);
   encoder.set_input_array(b);
   encoder.set_output_array(out);
-  dispatch_float_types(out.dtype(), "gemv", [&](auto type_tag) {
+  dispatch_inexact_types(out.dtype(), "gemv", [&](auto type_tag) {
     using DataType = cuda_type_t<MLX_GET_TYPE(type_tag)>;
     dim3 block_dims{WARP_SIZE, rows_per_block};
     const DataType* mat;
@@ -118,13 +149,13 @@ void gemv(
     auto vec_strides = const_param(b_batch_strides);
 
     if (M == 1) {
-      mat = b.data<DataType>();
-      vec = a.data<DataType>();
+      mat = gpu_ptr<DataType>(b);
+      vec = gpu_ptr<DataType>(a);
       rows = N;
       std::swap(mat_strides, vec_strides);
     } else {
-      mat = a.data<DataType>();
-      vec = b.data<DataType>();
+      mat = gpu_ptr<DataType>(a);
+      vec = gpu_ptr<DataType>(b);
       rows = M;
     }
     uint32_t num_blocks_x = (rows + rows_per_block - 1) / rows_per_block;
@@ -146,7 +177,7 @@ void gemv(
             0,
             mat,
             vec,
-            out.data<DataType>(),
+            gpu_ptr<DataType>(out),
             rows,
             cols);
       } else {
@@ -158,7 +189,7 @@ void gemv(
             0,
             mat,
             vec,
-            out.data<DataType>(),
+            gpu_ptr<DataType>(out),
             rows,
             cols,
             const_param(batch_shape),

mlx/backend/cuda/indexing.cpp

@@ -31,7 +31,7 @@ void append_indices_arg(
     int idx_ndim) {
   SmallVector<const void*> indices(nidx);
   for (int i = 0; i < nidx; ++i) {
-    indices[i] = inputs[i + 1].data<void>();
+    indices[i] = gpu_ptr<void>(inputs[i + 1]);
   }
   args.append(std::move(indices));
   SmallVector<int32_t> indices_shape(nidx * idx_ndim);
@@ -59,7 +59,9 @@ void Gather::eval_gpu(const std::vector<array>& inputs, array& out) {
   assert(inputs.size() > 0);
   const auto& src = inputs[0];
 
-  out.set_data(allocator::malloc(out.nbytes()));
+  auto& s = stream();
+  auto& encoder = cu::get_command_encoder(s);
+  out.set_data(cu::malloc_async(out.nbytes(), encoder.stream()));
   if (out.size() == 0) {
     return;
   }
@@ -80,7 +82,6 @@ void Gather::eval_gpu(const std::vector<array>& inputs, array& out) {
       dtype_to_string(idx_dtype),
       nidx);
 
-  auto& s = stream();
   cu::JitModule& mod = cu::get_jit_module(s.device, module_name, [&]() {
     std::vector<std::string> kernel_names;
     for (int ndim = 0; ndim <= MAX_NDIM; ++ndim) {
@@ -121,7 +122,6 @@ void Gather::eval_gpu(const std::vector<array>& inputs, array& out) {
       idx_ndim,
       large ? "int64_t" : "int32_t");
 
-  auto& encoder = cu::get_command_encoder(s);
   for (const auto& in : inputs) {
     encoder.set_input_array(in);
   }
@@ -239,7 +239,9 @@ void GatherAxis::eval_gpu(const std::vector<array>& inputs, array& out) {
   const auto& src = inputs[0];
   const auto& idx = inputs[1];
 
-  out.set_data(allocator::malloc(out.nbytes()));
+  auto& s = stream();
+  auto& encoder = cu::get_command_encoder(s);
+  out.set_data(cu::malloc_async(out.nbytes(), encoder.stream()));
   if (out.size() == 0) {
     return;
   }
@@ -251,7 +253,6 @@ void GatherAxis::eval_gpu(const std::vector<array>& inputs, array& out) {
       dtype_to_string(out.dtype()),
       dtype_to_string(idx.dtype()));
 
-  auto& s = stream();
   cu::JitModule& mod = cu::get_jit_module(s.device, module_name, [&]() {
     std::vector<std::string> kernel_names;
     for (int ndim = 0; ndim <= MAX_NDIM; ++ndim) {
@@ -312,7 +313,6 @@ void GatherAxis::eval_gpu(const std::vector<array>& inputs, array& out) {
       idx.flags().row_contiguous,
       large ? "int64_t" : "int32_t");
 
-  auto& encoder = cu::get_command_encoder(s);
   for (const auto& in : inputs) {
     encoder.set_input_array(in);
   }

mlx/backend/cuda/jit_module.cpp

@@ -99,6 +99,30 @@ const std::filesystem::path& ptx_cache_dir() {
   return cache;
 }
 
+std::filesystem::path get_ptx_path(
+    const std::filesystem::path& cache_dir,
+    const std::string& module_name) {
+#ifdef _WIN32
+  constexpr int max_file_name_length = 140;
+#else
+  constexpr int max_file_name_length = 245;
+#endif
+
+  if (module_name.size() <= max_file_name_length) {
+    return cache_dir / (module_name + ".ptx");
+  }
+
+  auto ptx_path = cache_dir;
+  int offset = 0;
+  while (module_name.size() - offset > max_file_name_length) {
+    ptx_path /= module_name.substr(offset, max_file_name_length);
+    offset += max_file_name_length;
+  }
+  ptx_path /= module_name.substr(offset) + ".ptx";
+
+  return ptx_path;
+}
+
 // Try to read the cached |ptx| and |ptx_kernels| from |cache_dir|.
 bool read_cached_ptx(
     const std::filesystem::path& cache_dir,
@@ -109,7 +133,7 @@ bool read_cached_ptx(
     return false;
   }
 
-  auto ptx_path = cache_dir / (module_name + ".ptx");
+  auto ptx_path = get_ptx_path(cache_dir, module_name);
   std::error_code error;
   auto ptx_size = std::filesystem::file_size(ptx_path, error);
   if (error) {
@@ -122,7 +146,7 @@ bool read_cached_ptx(
   ptx.resize(ptx_size);
   ptx_file.read(ptx.data(), ptx_size);
 
-  std::ifstream txt_file(cache_dir / (module_name + ".txt"), std::ios::binary);
+  std::ifstream txt_file(ptx_path.replace_extension(".txt"), std::ios::binary);
   std::string line;
   while (std::getline(txt_file, line)) {
     auto tab = line.find('\t');
@@ -144,16 +168,26 @@ void write_cached_ptx(
     return;
   }
 
-  std::ofstream ptx_file(cache_dir / (module_name + ".ptx"), std::ios::binary);
+  auto ptx_path = get_ptx_path(cache_dir, module_name);
+
+  // Ensure that the directory exists
+  auto parent = ptx_path.parent_path();
+  if (parent != cache_dir) {
+    std::filesystem::create_directories(parent);
+  }
+
+  // Write the compiled code and mangled names
+  std::ofstream ptx_file(ptx_path, std::ios::binary);
   if (!ptx.empty()) {
     ptx_file.write(&ptx.front(), ptx.size());
   }
-  std::ofstream txt_file(cache_dir / (module_name + ".txt"), std::ios::binary);
+  std::ofstream txt_file(ptx_path.replace_extension(".txt"), std::ios::binary);
   for (const auto& [name, mangled] : ptx_kernels) {
     txt_file << name << "\t" << mangled << std::endl;
   }
 
-  std::ofstream source_file(cache_dir / (module_name + ".cu"));
+  // Write the generated code
+  std::ofstream source_file(ptx_path.replace_extension(".cu"));
   source_file << source_code;
 }
 
@@ -297,7 +331,8 @@ void load_module(
     const std::string& ptx,
     const std::vector<std::pair<std::string, std::string>>& ptx_kernels,
     CUmodule& module_,
-    std::unordered_map<std::string, std::pair<CUfunction, bool>>& kernels) {
+    std::unordered_map<std::string, std::tuple<CUfunction, bool, uint>>&
+        kernels) {
   // Load module.
   char jit_log[4089] = {};
   CUjit_option options[] = {
@@ -314,7 +349,7 @@ void load_module(
   for (const auto& [name, mangled] : ptx_kernels) {
     CUfunction kernel;
     CHECK_CUDA_ERROR(cuModuleGetFunction(&kernel, module_, mangled.c_str()));
-    kernels[name] = std::make_pair(kernel, false);
+    kernels[name] = std::make_tuple(kernel, false, 0);
   }
 }
 
@@ -358,7 +393,7 @@ JitModule::~JitModule() {
   CHECK_CUDA_ERROR(cuModuleUnload(module_));
 }
 
-CUfunction JitModule::get_kernel(
+std::pair<CUfunction, uint> JitModule::get_kernel_and_dims(
     const std::string& kernel_name,
     std::function<void(CUfunction)> configure_kernel) {
   auto it = kernels_.find(kernel_name);
@@ -369,14 +404,22 @@ CUfunction JitModule::get_kernel(
 
   // If it is the first time we run this kernel then configure it. Do it only
   // once!
-  if (!it->second.second) {
+  auto kernel = std::get<0>(it->second);
+  if (!std::get<1>(it->second)) {
     if (configure_kernel) {
-      configure_kernel(it->second.first);
+      configure_kernel(kernel);
     }
-    it->second.second = true;
+    std::get<1>(it->second) = true;
+    std::get<2>(it->second) = max_occupancy_block_dim(kernel);
   }
 
-  return it->second.first;
+  return {kernel, std::get<2>(it->second)};
+}
+
+CUfunction JitModule::get_kernel(
+    const std::string& kernel_name,
+    std::function<void(CUfunction)> configure_kernel) {
+  return get_kernel_and_dims(kernel_name, std::move(configure_kernel)).first;
 }
 
 std::unordered_map<std::string, JitModule>& get_jit_module_cache() {

mlx/backend/cuda/jit_module.h

@@ -31,7 +31,7 @@ struct KernelArgs {
   }
 
   void append(const array& a) {
-    append(reinterpret_cast<CUdeviceptr>(a.data<void>()));
+    append(reinterpret_cast<CUdeviceptr>(gpu_ptr<void>(a.buffer())));
   }
 
   template <typename T>
@@ -99,10 +99,13 @@ class JitModule {
   CUfunction get_kernel(
       const std::string& kernel_name,
       std::function<void(CUfunction)> configure_kernel = nullptr);
+  std::pair<CUfunction, uint> get_kernel_and_dims(
+      const std::string& kernel_name,
+      std::function<void(CUfunction)> configure_kernel = nullptr);
 
  private:
   CUmodule module_{nullptr};
-  std::unordered_map<std::string, std::pair<CUfunction, bool>> kernels_;
+  std::unordered_map<std::string, std::tuple<CUfunction, bool, uint>> kernels_;
 };
 
 std::unordered_map<std::string, JitModule>& get_jit_module_cache();

mlx/backend/cuda/kernel_utils.cu

@@ -35,12 +35,10 @@ std::tuple<dim3, uint> get_launch_args(
     const Shape& shape,
     const Strides& strides,
     bool large,
-    int work_per_thread) {
+    int work_per_thread /* = 1 */,
+    uint max_block_dim /* = 1024 */) {
   size_t nthreads = cuda::ceil_div(size, work_per_thread);
-  uint block_dim = 1024;
-  if (block_dim > nthreads) {
-    block_dim = nthreads;
-  }
+  uint block_dim = max_block_dim < nthreads ? max_block_dim : nthreads;
   dim3 num_blocks;
   if (large) {
     num_blocks = get_2d_grid_dims(shape, strides, work_per_thread);

mlx/backend/cuda/kernel_utils.cuh

@@ -1,14 +1,15 @@
 // Copyright © 2025 Apple Inc.
 
-// This file includes host-only utilies for writing CUDA kernels, the difference
-// from backend/cuda/device/utils.cuh is that the latter file only include
-// device-only code.
+// This file includes host-only utilities for writing CUDA kernels, the
+// difference from backend/cuda/device/utils.cuh is that the latter file only
+// include device-only code.
 
 #pragma once
 
 #include <type_traits>
 
 #include "mlx/array.h"
+#include "mlx/backend/cuda/allocator.h"
 #include "mlx/backend/cuda/device/utils.cuh"
 
 #include <cuda.h>
@@ -120,19 +121,28 @@ dim3 get_2d_grid_dims(
     size_t divisor);
 std::pair<dim3, dim3> get_grid_and_block(int dim0, int dim1, int dim2);
 
-// Get the num_blocks and block_dims that maximize occupancy for |kernel|,
-// assuming each thread handles |work_per_thread| elements of |arr|.
+// Get the num_blocks and block_dims assuming each thread handles
+// |work_per_thread| elements of |arr|.
 std::tuple<dim3, uint> get_launch_args(
     size_t size,
     const Shape& shape,
     const Strides& strides,
     bool large,
-    int work_per_thread = 1);
+    int work_per_thread = 1,
+    uint max_block_dim = 1024);
 
-inline std::tuple<dim3, uint>
-get_launch_args(const array& arr, bool large, int work_per_thread = 1) {
+inline std::tuple<dim3, uint> get_launch_args(
+    const array& arr,
+    bool large,
+    int work_per_thread = 1,
+    uint max_block_dim = 1024) {
   return get_launch_args(
-      arr.size(), arr.shape(), arr.strides(), large, work_per_thread);
+      arr.size(),
+      arr.shape(),
+      arr.strides(),
+      large,
+      work_per_thread,
+      max_block_dim);
 }
 
 } // namespace mlx::core

mlx/backend/cuda/layer_norm.cu

@@ -230,9 +230,10 @@ void LayerNorm::eval_gpu(
   nvtx3::scoped_range r("LayerNorm::eval_gpu");
   auto& s = stream();
   auto& out = outputs[0];
+  auto& encoder = cu::get_command_encoder(s);
 
   // Make sure that the last dimension is contiguous.
-  auto set_output = [&s, &out](const array& x) {
+  auto set_output = [&s, &out, &encoder](const array& x) {
     bool no_copy = x.flags().contiguous && x.strides()[x.ndim() - 1] == 1;
     if (no_copy && x.ndim() > 1) {
       auto s = x.strides()[x.ndim() - 2];
@@ -243,7 +244,7 @@ void LayerNorm::eval_gpu(
         out.copy_shared_buffer(x);
       } else {
         out.set_data(
-            allocator::malloc(x.data_size() * x.itemsize()),
+            cu::malloc_async(x.data_size() * x.itemsize(), encoder.stream()),
             x.data_size(),
             x.strides(),
             x.flags());
@@ -265,7 +266,6 @@ void LayerNorm::eval_gpu(
   int64_t w_stride = (w.ndim() == 1) ? w.strides()[0] : 0;
   int64_t b_stride = (b.ndim() == 1) ? b.strides()[0] : 0;
 
-  auto& encoder = cu::get_command_encoder(s);
   encoder.set_input_array(x);
   encoder.set_input_array(w);
   encoder.set_input_array(b);
@@ -280,10 +280,10 @@ void LayerNorm::eval_gpu(
           n_rows,
           block_dim(),
           0,
-          x.data<DataType>(),
-          w.data<DataType>(),
-          b.data<DataType>(),
-          out.data<DataType>(),
+          gpu_ptr<DataType>(x),
+          gpu_ptr<DataType>(w),
+          gpu_ptr<DataType>(b),
+          gpu_ptr<DataType>(out),
           eps_,
           axis_size,
           w_stride,
@@ -335,7 +335,7 @@ void LayerNormVJP::eval_gpu(
     gx.copy_shared_buffer(g);
     g_in_gx = true;
   } else {
-    gx.set_data(allocator::malloc(gx.nbytes()));
+    gx.set_data(cu::malloc_async(gx.nbytes(), encoder.stream()));
   }
   if (g_copied && !g_in_gx) {
     encoder.add_temporary(g);
@@ -355,7 +355,7 @@ void LayerNormVJP::eval_gpu(
       g_in_gw = true;
       gw_temp.copy_shared_buffer(g);
     } else {
-      gw_temp.set_data(allocator::malloc(gw_temp.nbytes()));
+      gw_temp.set_data(cu::malloc_async(gw_temp.nbytes(), encoder.stream()));
       encoder.add_temporary(gw_temp);
     }
   }
@@ -393,11 +393,11 @@ void LayerNormVJP::eval_gpu(
                 n_rows,
                 block_dim(),
                 0,
-                x.data<DataType>(),
-                w.data<DataType>(),
-                g.data<DataType>(),
-                gx.data<DataType>(),
-                gw_temp.data<DataType>(),
+                gpu_ptr<DataType>(x),
+                gpu_ptr<DataType>(w),
+                gpu_ptr<DataType>(g),
+                gpu_ptr<DataType>(gx),
+                gpu_ptr<DataType>(gw_temp),
                 eps_,
                 axis_size,
                 w_stride);

mlx/backend/cuda/load.cpp

@@ -0,0 +1,60 @@
+// Copyright © 2023 Apple Inc.
+
+#include <algorithm>
+#include <utility>
+
+#include "mlx/backend/cuda/device.h"
+#include "mlx/backend/cuda/utils.h"
+#include "mlx/primitives.h"
+
+namespace {
+
+template <const uint8_t scalar_size>
+void swap_endianness(uint8_t* data_bytes, size_t N) {
+  struct Elem {
+    uint8_t bytes[scalar_size];
+  };
+
+  Elem* data = reinterpret_cast<Elem*>(data_bytes);
+
+  for (size_t i = 0; i < N; i++) {
+    for (size_t j = 0; j < (scalar_size / 2); j++) {
+      std::swap(data[i].bytes[j], data[i].bytes[scalar_size - j - 1]);
+    }
+  }
+}
+
+} // namespace
+
+namespace mlx::core {
+
+void Load::eval_gpu(const std::vector<array>& inputs, array& out) {
+  auto& encoder = cu::get_command_encoder(stream());
+  auto size = out.size();
+  auto nbytes = size * out.itemsize();
+  out.set_data(cu::malloc_async(nbytes, encoder.stream()));
+  auto out_ptr = malloc(nbytes);
+  reader_->read(static_cast<char*>(out_ptr), nbytes, offset_);
+  if (swap_endianness_) {
+    switch (out.itemsize()) {
+      case 2:
+        swap_endianness<2>(reinterpret_cast<uint8_t*>(out_ptr), size);
+        break;
+      case 4:
+        swap_endianness<4>(reinterpret_cast<uint8_t*>(out_ptr), size);
+        break;
+      case 8:
+        swap_endianness<8>(reinterpret_cast<uint8_t*>(out_ptr), size);
+        break;
+    }
+  }
+  CHECK_CUDA_ERROR(cudaMemcpyAsync(
+      gpu_ptr<void>(out),
+      out_ptr,
+      nbytes,
+      cudaMemcpyDefault,
+      encoder.stream()));
+  CHECK_CUDA_ERROR(cudaLaunchHostFunc(encoder.stream(), free, out_ptr));
+}
+
+} // namespace mlx::core

mlx/backend/cuda/logsumexp.cu

@@ -115,7 +115,7 @@ void LogSumExp::eval_gpu(const std::vector<array>& inputs, array& out) {
 
   auto in = ensure_contiguous(inputs[0]);
   if (in.flags().row_contiguous) {
-    out.set_data(allocator::malloc(out.nbytes()));
+    out.set_data(cu::malloc_async(out.nbytes(), encoder.stream()));
   } else {
     auto n = in.shape(-1);
     auto flags = in.flags();
@@ -130,7 +130,7 @@ void LogSumExp::eval_gpu(const std::vector<array>& inputs, array& out) {
     }
     flags.col_contiguous = col_contig;
     out.set_data(
-        allocator::malloc(in.nbytes() / n),
+        cu::malloc_async(in.nbytes() / n, encoder.stream()),
         in.data_size() / n,
         std::move(strides),
         flags);
@@ -151,8 +151,8 @@ void LogSumExp::eval_gpu(const std::vector<array>& inputs, array& out) {
           n_rows,
           block_dim(),
           0,
-          in.data<DataType>(),
-          out.data<DataType>(),
+          gpu_ptr<DataType>(in),
+          gpu_ptr<DataType>(out),
           axis_size);
     });
   });

mlx/backend/cuda/matmul.cpp

@@ -93,6 +93,10 @@ void gemm_and_bias(
       a_batch_strides.back(),
       b_batch_strides.back());
   if (bias) {
+    if (a.dtype() == complex64) {
+      throw std::runtime_error(
+          "[gemm_and_bias] complex64 bias epilogue isn’t supported in cublasLtMatmul.");
+    }
     gemm.set_bias(encoder, *bias);
   }
   gemm.run(
@@ -117,7 +121,7 @@ void Matmul::eval_gpu(const std::vector<array>& inputs, array& out) {
     return;
   }
 
-  out.set_data(allocator::malloc(out.nbytes()));
+  out.set_data(cu::malloc_async(out.nbytes(), encoder.stream()));
 
   int M = a_pre.shape(-2);
   int N = b_pre.shape(-1);
@@ -157,8 +161,9 @@ void AddMM::eval_gpu(const std::vector<array>& inputs, array& out) {
   /////////////////////////////////////////////////////////////////////////////
   // Dispatch to GEMM with epilogue or AddMM
 
-  if (beta_ == 1 && c.strides(-1) == 1 && c.data_size() == out.shape(-1)) {
-    out.set_data(allocator::malloc(out.nbytes()));
+  if (beta_ == 1 && a.dtype() != complex64 && c.strides(-1) == 1 &&
+      c.data_size() == out.shape(-1)) {
+    out.set_data(cu::malloc_async(out.nbytes(), encoder.stream()));
     gemm_and_bias(
         encoder,
         M,
@@ -182,10 +187,10 @@ void AddMM::eval_gpu(const std::vector<array>& inputs, array& out) {
     auto sty = c.strides()[c.ndim() - 1];
     if (sty == 1 && stx == c.shape(-1)) {
       ldc = stx;
-      out.set_data(allocator::malloc(out.nbytes()));
+      out.set_data(cu::malloc_async(out.nbytes(), encoder.stream()));
     } else if (sty == 1 && stx == 0) {
       ldc = 0;
-      out.set_data(allocator::malloc(out.nbytes()));
+      out.set_data(cu::malloc_async(out.nbytes(), encoder.stream()));
     } else {
       // Copy C into out and set C to out
       ldc = c.shape(-1);

mlx/backend/cuda/primitives.cpp

@@ -28,7 +28,6 @@ NO_GPU(FFT)
 NO_GPU(GatherMM)
 NO_GPU(GatherQMM)
 NO_GPU(Hadamard)
-NO_GPU(Load)
 NO_GPU_MULTI(LUF)
 NO_GPU_MULTI(QRF)
 NO_GPU(QuantizedMatmul)

mlx/backend/cuda/quantized/affine_quantize.cu

@@ -262,10 +262,10 @@ void affine_quantize(
             num_blocks,
             block_dims,
             0,
-            w.data<T>(),
-            wq.data<uint8_t>(),
-            scales.data<T>(),
-            biases.data<T>(),
+            gpu_ptr<T>(w),
+            gpu_ptr<uint8_t>(wq),
+            gpu_ptr<T>(scales),
+            gpu_ptr<T>(biases),
             w.size());
       });
     });
@@ -306,7 +306,7 @@ void affine_dequantize(
   enc.set_input_array(scales);
   enc.set_input_array(biases);
   enc.set_output_array(w);
-  dispatch_float_types(w.dtype(), "affine_quantize", [&](auto type_tag) {
+  dispatch_float_types(w.dtype(), "affine_dequantize", [&](auto type_tag) {
     dispatch_groups(group_size_, [&](auto group_size) {
       dispatch_bits(bits_, [&](auto bits) {
         using T = cuda_type_t<MLX_GET_TYPE(type_tag)>;
@@ -318,10 +318,10 @@ void affine_dequantize(
             num_blocks,
             block_dims,
             0,
-            wq.data<uint8_t>(),
-            scales.data<T>(),
-            biases.data<T>(),
-            w.data<T>(),
+            gpu_ptr<uint8_t>(wq),
+            gpu_ptr<T>(scales),
+            gpu_ptr<T>(biases),
+            gpu_ptr<T>(w),
             w.size());
       });
     });

mlx/backend/cuda/quantized/convert_fp8.cu

@@ -0,0 +1,19 @@
+// Copyright © 2025 Apple Inc.
+#include "mlx/backend/cuda/unary/unary.cuh"
+#include "mlx/fast_primitives.h"
+
+namespace mlx::core {
+void fast::ConvertFP8::eval_gpu(
+    const std::vector<array>& inputs,
+    std::vector<array>& outputs) {
+  nvtx3::scoped_range r("ConvertFP8::eval_gpu");
+  auto& in = inputs[0];
+  auto& out = outputs[0];
+  auto& s = out.primitive().stream();
+  if (to_fp8_) {
+    unary_op_gpu<cu::ToFP8>(inputs, out, name(), s);
+  } else {
+    unary_op_gpu<cu::FromFP8>(inputs, out, name(), s);
+  }
+}
+} // namespace mlx::core

mlx/backend/cuda/quantized/cuda_fp4.h

@@ -0,0 +1,83 @@
+#pragma once
+
+struct __nv_fp8_e8m0 {
+  __device__ __nv_fp8_e8m0(float x) {
+    if (!std::isfinite(x)) {
+      __x = 0xFF;
+      return;
+    }
+    if (x < 0.0f) {
+      __x = 0x00;
+      return;
+    }
+    float le = std::log2f(x);
+    int n = static_cast<int>(std::nearbyintf(le));
+
+    n = n < -127 ? -127 : n;
+    n = n > 127 ? 127 : n;
+    __x = static_cast<uint8_t>(n + 127);
+  }
+
+  __device__ operator float() {
+    if (__x == 0xFF) {
+      return std::numeric_limits<float>::quiet_NaN();
+    }
+    return std::ldexp(1.0f, static_cast<int>(__x) - 127);
+  }
+
+  uint8_t __x{0};
+};
+
+struct __nv_fp4_e2m1 {
+  __device__ __nv_fp4_e2m1(float x) {
+    if (std::isnan(x)) {
+      __x = 0x7;
+      return;
+    }
+
+    const uint8_t sign_bit = (std::signbit(x)) ? 0x8 : 0x0;
+    x = std::abs(x);
+
+    if (x > 5.0f) {
+      __x = 0x7;
+    } else if (x >= 3.5f) {
+      __x = 0x6;
+    } else if (x > 2.5f) {
+      __x = 0x5;
+    } else if (x >= 1.75f) {
+      __x = 0x4;
+    } else if (x > 1.25f) {
+      __x = 0x3;
+    } else if (x >= 0.75f) {
+      __x = 0x2;
+    } else if (x > 0.25f) {
+      __x = 0x1;
+    } else {
+      __x = 0x0;
+    }
+    __x |= sign_bit;
+  }
+
+  __device__ operator float() {
+    static const float LUT[16] = {
+        0.0f,
+        0.5f,
+        1.0f,
+        1.5f,
+        2.0f,
+        3.0f,
+        4.0f,
+        6.0f,
+        -0.0f,
+        -0.5f,
+        -1.0f,
+        -1.5f,
+        -2.0f,
+        -3.0f,
+        -4.0f,
+        -6.0f};
+
+    return LUT[__x];
+  }
+  uint8_t __x{0};
+};

mlx/backend/cuda/quantized/fp_quantize.cu

@@ -0,0 +1,216 @@
+// Copyright © 2025 Apple Inc.
+
+#include "mlx/backend/cuda/device.h"
+#include "mlx/backend/cuda/kernel_utils.cuh"
+#include "mlx/backend/cuda/quantized/quantized.h"
+#include "mlx/dtype_utils.h"
+
+#include <cooperative_groups.h>
+#include <cooperative_groups/reduce.h>
+#include <cuda_fp4.h>
+#include <cuda_fp8.h>
+
+namespace mlx::core {
+namespace cu {
+
+template <int bits>
+struct Quantize {
+  __device__ uint8_t operator()(float x) {
+    if constexpr (bits == 8) {
+      return __nv_fp8_e4m3(x).__x;
+    } else {
+      return __nv_fp4_e2m1(x).__x;
+    }
+  }
+};
+
+template <int bits>
+struct Dequantize {
+  __device__ float operator()(uint8_t x) {
+    if constexpr (bits == 8) {
+      return float(*(__nv_fp8_e4m3*)(&x));
+    } else {
+      return float(*(__nv_fp4_e2m1*)(&x));
+    }
+  }
+};
+
+namespace cg = cooperative_groups;
+
+template <typename T, int group_size, int bits, bool use_mx_scale>
+__global__ void
+fp_quantize(const T* w, uint8_t* out, uint8_t* scales, size_t size) {
+  auto block_size = cg::this_thread_block().dim_threads();
+  auto block_idx = cg::this_thread_block().group_index();
+  auto idx_in_block = cg::this_thread_block().thread_index();
+
+  auto tidx = block_idx.x * block_size.x + idx_in_block.x;
+  auto tidy = block_idx.y * block_size.y + idx_in_block.y;
+
+  auto grid_dim_x =
+      cg::this_grid().dim_blocks().x * cg::this_grid().block_index().x;
+  size_t index = tidx + grid_dim_x * size_t(tidy);
+  if (index >= size) {
+    return;
+  }
+
+  float w_thread = w[index];
+
+  cg::greater<float> max_op;
+  auto warp = cg::tiled_partition<group_size>(cg::this_thread_block());
+
+  float scale = cg::reduce(warp, abs(w_thread), max_op);
+  scale /= bits == 4 ? 6.0f : 448.0f;
+  // Convert to mx scale or nv scale
+  using ScaleType =
+      std::conditional_t<use_mx_scale, __nv_fp8_e8m0, __nv_fp8_e4m3>;
+  auto s = ScaleType(scale);
+  uint8_t q_scale = s.__x;
+  scale = float(s);
+
+  // Write out the scales
+  size_t gindex = index / group_size;
+  if (index % group_size == 0) {
+    scales[gindex] = q_scale;
+  }
+
+  uint8_t output = Quantize<bits>{}(scale == 0 ? 0.0f : w_thread / scale);
+  if (bits == 4) {
+    uint8_t sval = warp.shfl_down(output, 1);
+    output |= sval << bits;
+  }
+  constexpr int pack_factor = bits == 8 ? 1 : 2;
+  if (index % pack_factor == 0) {
+    out[index / pack_factor] = output;
+  }
+}
+
+template <typename T, int group_size, int bits, bool use_mx_scale>
+__global__ void
+fp_dequantize(const uint8_t* w, const uint8_t* scales, T* out, size_t size) {
+  auto block_size = cg::this_thread_block().dim_threads();
+  auto block_idx = cg::this_thread_block().group_index();
+  auto idx_in_block = cg::this_thread_block().thread_index();
+
+  auto tidx = block_idx.x * block_size.x + idx_in_block.x;
+  auto tidy = block_idx.y * block_size.y + idx_in_block.y;
+
+  auto grid_dim_x =
+      cg::this_grid().dim_blocks().x * cg::this_grid().block_index().x;
+
+  constexpr int pack_factor = bits == 8 ? 1 : 2;
+  size_t offset = tidx + grid_dim_x * size_t(tidy);
+  size_t oindex = offset * pack_factor;
+
+  if (oindex >= size) {
+    return;
+  }
+
+  size_t gindex = oindex / group_size;
+  using ScaleType =
+      std::conditional_t<use_mx_scale, __nv_fp8_e8m0, __nv_fp8_e4m3>;
+  auto scale = float(((ScaleType*)(scales))[gindex]);
+
+  out += oindex;
+
+  uint val = w[offset];
+#pragma clang loop unroll(full)
+  for (int i = 0; i < pack_factor; i++) {
+    uint8_t d;
+    if (bits == 4) {
+      d = (val >> (bits * i)) & 0x0f;
+    } else if (bits == 8) {
+      d = val;
+    }
+    out[i] = static_cast<T>(scale * Dequantize<bits>{}(d));
+  }
+}
+
+} // namespace cu
+
+void fp_quantize(
+    const array& w,
+    array& wq,
+    array& scales,
+    int group_size,
+    int bits,
+    cu::CommandEncoder& enc,
+    const Stream& s) {
+  enc.set_input_array(w);
+  enc.set_output_array(wq);
+  enc.set_output_array(scales);
+  dispatch_float_types(w.dtype(), "fp_quantize", [&](auto type_tag) {
+    using T = cuda_type_t<MLX_GET_TYPE(type_tag)>;
+    if constexpr (!std::is_same_v<T, double>) {
+      auto kernel = cu::fp_quantize<T, 32, 4, true>;
+      if (bits == 8) {
+        kernel = cu::fp_quantize<T, 32, 8, true>;
+      } else if (group_size == 16) {
+        kernel = cu::fp_quantize<T, 16, 4, false>;
+      }
+      bool large = w.size() > UINT_MAX;
+      auto [num_blocks, block_dims] =
+          get_launch_args(w.size(), w.shape(), w.strides(), large);
+      enc.add_kernel_node(
+          kernel,
+          num_blocks,
+          block_dims,
+          0,
+          gpu_ptr<T>(w),
+          gpu_ptr<uint8_t>(wq),
+          gpu_ptr<uint8_t>(scales),
+          w.size());
+    } else {
+      throw std::runtime_error(
+          "[Quantize::eval_gpu] Can not quantize input with type float64.");
+    }
+  });
+}
+
+void fp_dequantize(
+    const array& wq,
+    const array& scales,
+    array& w,
+    int group_size,
+    int bits,
+    cu::CommandEncoder& enc,
+    const Stream& s) {
+  constexpr int uint8_per_uint32 = 4;
+  int packs_per_int = 8 / bits;
+
+  size_t size = w.size() / packs_per_int;
+  bool large = size > UINT_MAX;
+  auto grid_shape = w.shape();
+  grid_shape.back() *= uint8_per_uint32;
+
+  enc.set_input_array(wq);
+  enc.set_input_array(scales);
+  enc.set_output_array(w);
+  dispatch_float_types(w.dtype(), "fp_dequantize", [&](auto type_tag) {
+    using T = cuda_type_t<MLX_GET_TYPE(type_tag)>;
+    if constexpr (!std::is_same_v<T, double>) {
+      auto kernel = cu::fp_dequantize<T, 32, 4, true>;
+      if (bits == 8) {
+        kernel = cu::fp_dequantize<T, 32, 8, true>;
+      } else if (group_size == 16) {
+        kernel = cu::fp_dequantize<T, 16, 4, false>;
+      }
+      auto [num_blocks, block_dims] =
+          get_launch_args(size, grid_shape, w.strides(), large);
+      enc.add_kernel_node(
+          kernel,
+          num_blocks,
+          block_dims,
+          0,
+          gpu_ptr<uint8_t>(wq),
+          gpu_ptr<uint8_t>(scales),
+          gpu_ptr<T>(w),
+          w.size());
+    } else {
+      throw std::runtime_error(
+          "[Quantize::eval_gpu] Can not dequantize to output with type float64.");
+    }
+  });
+}
+
+} // namespace mlx::core

mlx/backend/cuda/quantized/quantized.cpp

@@ -57,23 +57,30 @@ void fast::Quantize::eval_gpu(
   if (dequantize_) {
     auto wq = ensure_row_contiguous(inputs[0], enc, s);
     auto scales = ensure_row_contiguous(inputs[1], enc, s);
-    auto biases = ensure_row_contiguous(inputs[2], enc, s);
     auto& w = outputs[0];
 
-    w.set_data(allocator::malloc(w.nbytes()));
+    w.set_data(cu::malloc_async(w.nbytes(), enc.stream()));
 
+    if (mode_ == QuantizationMode::Affine) {
+      auto biases = ensure_row_contiguous(inputs[2], enc, s);
       affine_dequantize(wq, scales, biases, w, group_size_, bits_, enc, s);
+    } else {
+      fp_dequantize(wq, scales, w, group_size_, bits_, enc, s);
+    }
   } else {
     auto w = ensure_row_contiguous(inputs[0], enc, s);
     auto& wq = outputs[0];
     auto& scales = outputs[1];
+
+    wq.set_data(cu::malloc_async(wq.nbytes(), enc.stream()));
+    scales.set_data(cu::malloc_async(scales.nbytes(), enc.stream()));
+    if (mode_ == QuantizationMode::Affine) {
       auto& biases = outputs[2];
-
-    wq.set_data(allocator::malloc(wq.nbytes()));
-    scales.set_data(allocator::malloc(scales.nbytes()));
-    biases.set_data(allocator::malloc(biases.nbytes()));
-
+      biases.set_data(cu::malloc_async(biases.nbytes(), enc.stream()));
       affine_quantize(w, wq, scales, biases, group_size_, bits_, enc, s);
+    } else {
+      fp_quantize(w, wq, scales, group_size_, bits_, enc, s);
+    }
   }
 }
 

mlx/backend/cuda/quantized/quantized.h

@@ -24,4 +24,22 @@ void affine_dequantize(
     cu::CommandEncoder& enc,
     const Stream& s);
 
+void fp_quantize(
+    const array& w,
+    array& wq,
+    array& scales,
+    int group_size,
+    int bits,
+    cu::CommandEncoder& enc,
+    const Stream& s);
+
+void fp_dequantize(
+    const array& wq,
+    const array& scales,
+    array& w,
+    int group_size,
+    int bits,
+    cu::CommandEncoder& enc,
+    const Stream& s);
+
 } // namespace mlx::core

mlx/backend/cuda/random.cu

@@ -143,7 +143,9 @@ void RandomBits::eval_gpu(const std::vector<array>& inputs, array& out) {
 
   uint32_t elems_per_key = out.size() / num_keys;
   uint32_t bytes_per_key = out.itemsize() * elems_per_key;
-  out.set_data(allocator::malloc(out.nbytes()));
+  auto& s = stream();
+  auto& encoder = cu::get_command_encoder(s);
+  out.set_data(cu::malloc_async(out.nbytes(), encoder.stream()));
   if (out.size() == 0) {
     return;
   }
@@ -152,8 +154,6 @@ void RandomBits::eval_gpu(const std::vector<array>& inputs, array& out) {
   uint32_t half_size = out_per_key / 2;
   bool odd = out_per_key % 2;
 
-  auto& s = stream();
-  auto& encoder = cu::get_command_encoder(s);
   encoder.set_input_array(keys);
   encoder.set_output_array(out);
   dim3 grid_dims{num_keys, half_size + odd};
@@ -171,8 +171,8 @@ void RandomBits::eval_gpu(const std::vector<array>& inputs, array& out) {
         grid,
         block,
         0,
-        keys.data<uint32_t>(),
-        out.data<uint8_t>(),
+        gpu_ptr<uint32_t>(keys),
+        gpu_ptr<uint8_t>(out),
         grid_dims,
         odd,
         bytes_per_key);
@@ -182,8 +182,8 @@ void RandomBits::eval_gpu(const std::vector<array>& inputs, array& out) {
         grid,
         block,
         0,
-        keys.data<uint32_t>(),
-        out.data<uint8_t>(),
+        gpu_ptr<uint32_t>(keys),
+        gpu_ptr<uint8_t>(out),
         grid_dims,
         odd,
         bytes_per_key,

mlx/backend/cuda/reduce/all_reduce.cu

@@ -66,7 +66,7 @@ void all_reduce(
     Reduce::ReduceType reduce_type) {
   constexpr int N_READS = 8;
 
-  out.set_data(allocator::malloc(out.nbytes()));
+  out.set_data(cu::malloc_async(out.nbytes(), encoder.stream()));
 
   auto get_args = [](size_t size, int N) {
     int threads = std::min(512UL, (size + N - 1) / N);
@@ -100,14 +100,15 @@ void all_reduce(
   Dtype dt = in.dtype();
 
   // Cub doesn't like const pointers for load (sigh).
-  void* indata = const_cast<void*>(in.data<void>());
+  void* indata = const_cast<void*>(gpu_ptr<void>(in));
 
   // Large array so allocate an intermediate and accumulate there
   std::tie(blocks, threads, block_step) = get_args(insize, N_READS);
   encoder.set_input_array(in);
   if (blocks > 1) {
     array intermediate({blocks}, out.dtype(), nullptr, {});
-    intermediate.set_data(allocator::malloc(intermediate.nbytes()));
+    intermediate.set_data(
+        cu::malloc_async(intermediate.nbytes(), encoder.stream()));
     encoder.add_temporary(intermediate);
     encoder.set_output_array(intermediate);
     dispatch_all_types(dt, [&](auto type_tag) {
@@ -122,14 +123,14 @@ void all_reduce(
             threads,
             0,
             static_cast<T*>(indata),
-            intermediate.data<U>(),
+            gpu_ptr<U>(intermediate),
             block_step,
             insize);
       });
     });
 
     // Set the input for the next step and recalculate the blocks
-    indata = intermediate.data<void>();
+    indata = gpu_ptr<void>(intermediate);
     dt = intermediate.dtype();
     insize = intermediate.size();
     std::tie(blocks, threads, block_step) = get_args(insize, N_READS);
@@ -149,7 +150,7 @@ void all_reduce(
           threads,
           0,
           static_cast<T*>(indata),
-          out.data<U>(),
+          gpu_ptr<U>(out),
           block_step,
           insize);
     });

mlx/backend/cuda/reduce/col_reduce.cu

@@ -181,6 +181,47 @@ col_reduce_looped(T* in, U* out, const __grid_constant__ ColReduceArgs args) {
   }
 }
 
+template <typename T, typename U, typename Op, int N_READS = 4>
+__global__ void col_reduce_small(
+    const T* in,
+    U* out,
+    const __grid_constant__ ColReduceArgs args,
+    size_t total) {
+  Op op;
+  auto grid = cg::this_grid();
+  auto block = cg::this_thread_block();
+
+  const auto idx = grid.thread_rank() * N_READS;
+  const auto before_axis = idx / args.reduction_stride;
+  const auto after_axis = idx % args.reduction_stride;
+  const auto offset =
+      before_axis * args.reduction_stride * args.reduction_size + after_axis;
+
+  if (idx >= total) {
+    return;
+  }
+
+  in += offset;
+  out += idx;
+
+  AlignedVector<U, N_READS> accumulator;
+  for (int i = 0; i < N_READS; i++) {
+    accumulator[i] = ReduceInit<Op, T>::value();
+  }
+
+  for (int i = 0; i < args.reduction_size; i++) {
+    auto values = load_vector<N_READS>(in, 0);
+
+    for (int j = 0; j < N_READS; j++) {
+      accumulator[j] = op(accumulator[j], cast_to<U>(values[j]));
+    }
+
+    in += args.reduction_stride;
+  }
+
+  store_vector(out, 0, accumulator);
+}
+
 } // namespace cu
 
 inline auto output_grid_for_col_reduce(
@@ -206,10 +247,10 @@ void col_reduce_looped(
     Reduce::ReduceType reduce_type,
     const std::vector<int>& axes,
     const ReductionPlan& plan,
-    cu::ColReduceArgs args) {
+    const cu::ColReduceArgs& args) {
   // Allocate data for the output using in's layout to access them as
   // contiguously as possible.
-  allocate_same_layout(out, in, axes);
+  allocate_same_layout(out, in, axes, encoder);
 
   encoder.set_input_array(in);
   encoder.set_output_array(out);
@@ -220,7 +261,7 @@ void col_reduce_looped(
         using T = cuda_type_t<MLX_GET_TYPE(type_tag)>;
         using U = typename cu::ReduceResult<OP, T>::type;
         // Cub doesn't like const pointers for vectorized loads. (sigh)
-        T* indata = const_cast<T*>(in.data<T>());
+        T* indata = const_cast<T*>(gpu_ptr<T>(in));
 
         constexpr int N_READS = 4;
         constexpr int BM = 32;
@@ -230,12 +271,55 @@ void col_reduce_looped(
         auto kernel =
             cu::col_reduce_looped<T, U, OP, reduce_ndim(), BM, BN, N_READS>;
         encoder.add_kernel_node(
-            kernel, grid, blocks, 0, indata, out.data<U>(), args);
+            kernel,
+            grid,
+            blocks,
+            0,
+            indata,
+            gpu_ptr<U>(out),
+            static_cast<cu::ColReduceArgs>(args));
       });
     });
   });
 }
 
+void col_reduce_small(
+    cu::CommandEncoder& encoder,
+    const array& in,
+    array& out,
+    Reduce::ReduceType reduce_type,
+    const std::vector<int>& axes,
+    const ReductionPlan& plan,
+    const cu::ColReduceArgs& args) {
+  // Allocate data for the output using in's layout to access them as
+  // contiguously as possible.
+  allocate_same_layout(out, in, axes, encoder);
+
+  encoder.set_input_array(in);
+  encoder.set_output_array(out);
+  dispatch_all_types(in.dtype(), [&](auto type_tag) {
+    dispatch_reduce_ops(reduce_type, [&](auto reduce_type_tag) {
+      using OP = MLX_GET_TYPE(reduce_type_tag);
+      using T = cuda_type_t<MLX_GET_TYPE(type_tag)>;
+      using U = typename cu::ReduceResult<OP, T>::type;
+
+      constexpr int N_READS = 16 / sizeof(T);
+      auto tmp_grid = get_2d_grid_dims(out.shape(), out.strides());
+      auto [grid, block] = get_grid_and_block(tmp_grid.x, tmp_grid.y, 1);
+      auto kernel = cu::col_reduce_small<T, U, OP, N_READS>;
+      encoder.add_kernel_node(
+          kernel,
+          grid,
+          block,
+          0,
+          gpu_ptr<T>(in),
+          gpu_ptr<U>(out),
+          static_cast<cu::ColReduceArgs>(args),
+          out.size());
+    });
+  });
+}
+
 void col_reduce(
     cu::CommandEncoder& encoder,
     const array& in,
@@ -258,6 +342,13 @@ void col_reduce(
   // Make the args struct to help route to the best kernel
   cu::ColReduceArgs args(in, plan, axes);
 
+  // Small col reduce with a single or contiguous reduction axis
+  if (args.non_col_reductions == 1 && args.reduction_size <= 32 &&
+      args.reduction_stride % (16 / in.itemsize()) == 0) {
+    col_reduce_small(encoder, in, out, reduce_type, axes, plan, args);
+    return;
+  }
+
   // Fallback col reduce
   col_reduce_looped(encoder, in, out, reduce_type, axes, plan, args);
 }

mlx/backend/cuda/reduce/init_reduce.cu

@@ -28,7 +28,7 @@ void init_reduce(
     Reduce::ReduceType reduce_type) {
   // Allocate if needed
   if (out.data_shared_ptr() == nullptr) {
-    out.set_data(allocator::malloc(out.nbytes()));
+    out.set_data(cu::malloc_async(out.nbytes(), encoder.stream()));
   }
 
   encoder.set_output_array(out);
@@ -42,7 +42,7 @@ void init_reduce(
       dim3 block(grid.x < 1024 ? grid.x : 1024, 1, 1);
       grid.x = (grid.x + 1023) / 1024;
       encoder.add_kernel_node(
-          kernel, grid, block, 0, out.data<U>(), out.size());
+          kernel, grid, block, 0, gpu_ptr<U>(out), out.size());
     });
   });
 }

mlx/backend/cuda/reduce/reduce_utils.cuh

@@ -5,6 +5,7 @@
 #include <numeric>
 
 #include "mlx/backend/common/utils.h"
+#include "mlx/backend/cuda/device.h"
 #include "mlx/backend/cuda/device/utils.cuh"
 
 #include <cooperative_groups.h>
@@ -92,9 +93,10 @@ block_reduce(Block block, Warp warp, T (&vals)[N], T* smem, Op op, T init) {
 inline void allocate_same_layout(
     array& out,
     const array& in,
-    const std::vector<int>& axes) {
+    const std::vector<int>& axes,
+    cu::CommandEncoder& encoder) {
   if (in.flags().row_contiguous) {
-    out.set_data(allocator::malloc(out.nbytes()));
+    out.set_data(cu::malloc_async(out.nbytes(), encoder.stream()));
     return;
   }
 
@@ -133,7 +135,7 @@ inline void allocate_same_layout(
   fl.col_contiguous = cc;
   fl.contiguous = true;
   out.set_data(
-      allocator::malloc(out.nbytes()),
+      cu::malloc_async(out.nbytes(), encoder.stream()),
       data_size,
       final_strides,
       fl,

mlx/backend/cuda/reduce/row_reduce.cu

@@ -7,8 +7,6 @@
 
 #include <cooperative_groups.h>
 #include <cooperative_groups/reduce.h>
-#include <cub/block/block_load.cuh>
-#include <cub/block/block_reduce.cuh>
 
 namespace mlx::core {
 
@@ -83,7 +81,8 @@ struct RowReduceArgs {
 };
 
 template <typename T, typename U, typename ReduceOp, int N = 4, int M = 1>
-__global__ void row_reduce_simple(T* in, U* out, size_t n_rows, int size) {
+__global__ void
+row_reduce_simple(const T* in, U* out, size_t n_rows, int size) {
   auto grid = cg::this_grid();
   auto block = cg::this_thread_block();
   auto warp = cg::tiled_partition<WARP_SIZE>(block);
@@ -91,8 +90,8 @@ __global__ void row_reduce_simple(T* in, U* out, size_t n_rows, int size) {
   const U init = cu::ReduceInit<ReduceOp, T>::value();
   ReduceOp op;
 
-  T vals[M][N];
-  U accs[M];
+  AlignedVector<T, N> vals[M];
+  AlignedVector<U, M> accs;
   for (int i = 0; i < M; i++) {
     accs[i] = init;
   }
@@ -101,43 +100,31 @@ __global__ void row_reduce_simple(T* in, U* out, size_t n_rows, int size) {
       min(n_rows - M, static_cast<size_t>(grid.block_rank() * M));
   const size_t full_blocks = size / (block.size() * N);
   const size_t final_offset = full_blocks * (block.size() * N);
-  in += start_row * size;
+  in += start_row * size + block.thread_rank() * N;
   out += start_row;
 
-  if (size % N == 0) {
   for (size_t r = 0; r < full_blocks; r++) {
     for (int k = 0; k < M; k++) {
-        cub::LoadDirectBlockedVectorized<T, N>(
-            block.thread_rank(),
-            in + k * size + r * (block.size() * N),
-            vals[k]);
+      vals[k] = load_vector<N>(in + k * size, 0);
+    }
+    for (int k = 0; k < M; k++) {
       for (int j = 0; j < N; j++) {
         accs[k] = op(accs[k], cast_to<U>(vals[k][j]));
       }
     }
-    }
-  } else {
-    for (size_t r = 0; r < full_blocks; r++) {
-      for (int k = 0; k < M; k++) {
-        cub::LoadDirectBlocked(
-            block.thread_rank(),
-            in + k * size + r * (block.size() * N),
-            vals[k]);
-        for (int j = 0; j < N; j++) {
-          accs[k] = op(accs[k], cast_to<U>(vals[k][j]));
-        }
-      }
-    }
+
+    in += block.size() * N;
   }
 
   if (final_offset < size) {
     for (int k = 0; k < M; k++) {
-      cub::LoadDirectBlocked(
-          block.thread_rank(),
-          in + k * size + final_offset,
-          vals[k],
-          size,
-          cast_to<T>(init));
+      for (int i = 0; i < N; i++) {
+        vals[k][i] = ((final_offset + block.thread_rank() * N + i) < size)
+            ? in[k * size + i]
+            : cast_to<T>(init);
+      }
+    }
+    for (int k = 0; k < M; k++) {
       for (int j = 0; j < N; j++) {
         accs[k] = op(accs[k], cast_to<U>(vals[k][j]));
       }
@@ -145,13 +132,11 @@ __global__ void row_reduce_simple(T* in, U* out, size_t n_rows, int size) {
   }
 
   __shared__ U shared_accumulators[32 * M];
-  block_reduce(block, warp, accs, shared_accumulators, op, init);
+  block_reduce(block, warp, accs.val, shared_accumulators, op, init);
 
   if (block.thread_rank() == 0) {
     if (grid.block_rank() * M + M <= n_rows) {
-      for (int i = 0; i < M; i++) {
-        out[i] = accs[i];
-      }
+      store_vector(out, 0, accs);
     } else {
       short offset = grid.block_rank() * M + M - n_rows;
       for (int i = offset; i < M; i++) {
@@ -161,17 +146,10 @@ __global__ void row_reduce_simple(T* in, U* out, size_t n_rows, int size) {
   }
 }
 
-template <
-    typename T,
-    typename U,
-    typename Op,
-    int NDIM,
-    int BLOCK_DIM,
-    int N_READS = 4>
+template <typename T, typename U, typename Op, int NDIM, int N_READS = 4>
 __global__ void row_reduce_looped(
-    T* in,
+    const T* in,
     U* out,
-    size_t out_size,
     const __grid_constant__ RowReduceArgs args) {
   auto grid = cg::this_grid();
   auto block = cg::this_thread_block();
@@ -185,37 +163,61 @@ __global__ void row_reduce_looped(
   U init = ReduceInit<Op, T>::value();
   total[0] = init;
   LoopedElemToLoc<NDIM, (NDIM > 2)> loop(args.reduce_ndim);
-  size_t full_blocks = args.row_size / (BLOCK_DIM * N_READS);
-  size_t final_offset = full_blocks * BLOCK_DIM * N_READS;
+  const size_t full_blocks = args.row_size / (block.size() * N_READS);
+  const size_t final_offset = full_blocks * (block.size() * N_READS);
 
   in += elem_to_loc(out_idx, args.shape.data(), args.strides.data(), args.ndim);
+  in += block.thread_rank() * N_READS;
+
+  // Unaligned reduce
+  if (final_offset < args.row_size) {
+    bool mask[N_READS];
+    for (int i = 0; i < N_READS; i++) {
+      mask[i] =
+          (final_offset + block.thread_rank() * N_READS + i) < args.row_size;
+    }
 
     for (size_t n = 0; n < args.non_row_reductions; n++) {
+      const T* inlocal = in + loop.location();
+
       for (size_t r = 0; r < full_blocks; r++) {
+        auto vals = load_vector<N_READS>(inlocal, 0);
+        for (int i = 0; i < N_READS; i++) {
+          total[0] = op(total[0], cast_to<U>(vals[i]));
+        }
+        inlocal += block.size() * N_READS;
+      }
+
+      {
         T vals[N_READS];
-      cub::LoadDirectBlockedVectorized<T, N_READS>(
-          block.thread_rank(),
-          in + loop.location() + r * BLOCK_DIM * N_READS,
-          vals);
+        for (int i = 0; i < N_READS; i++) {
+          vals[i] = mask[i] ? inlocal[i] : cast_to<T>(init);
+        }
         for (int i = 0; i < N_READS; i++) {
           total[0] = op(total[0], cast_to<U>(vals[i]));
         }
       }
-    if (final_offset < args.row_size) {
-      T vals[N_READS];
-      cub::LoadDirectBlocked(
-          block.thread_rank(),
-          in + loop.location() + final_offset,
-          vals,
-          args.row_size - final_offset,
-          cast_to<T>(init));
-      for (int i = 0; i < N_READS; i++) {
-        total[0] = op(total[0], cast_to<U>(vals[i]));
-      }
-    }
-    // TODO: Maybe block.sync() here?
+
       loop.next(args.reduce_shape.data(), args.reduce_strides.data());
     }
+  }
+
+  // Aligned case
+  else {
+    for (size_t n = 0; n < args.non_row_reductions; n++) {
+      const T* inlocal = in + loop.location();
+
+      for (size_t r = 0; r < full_blocks; r++) {
+        auto vals = load_vector<N_READS>(inlocal, 0);
+        for (int i = 0; i < N_READS; i++) {
+          total[0] = op(total[0], cast_to<U>(vals[i]));
+        }
+        inlocal += block.size() * N_READS;
+      }
+
+      loop.next(args.reduce_shape.data(), args.reduce_strides.data());
+    }
+  }
 
   __shared__ U shared_accumulators[32];
   block_reduce(block, warp, total, shared_accumulators, op, init);
@@ -234,11 +236,9 @@ void row_reduce_simple(
     Reduce::ReduceType reduce_type,
     const std::vector<int>& axes,
     const ReductionPlan& plan) {
-  constexpr int N_READS = 8;
-
   // Allocate data for the output using in's layout to avoid elem_to_loc in the
   // kernel.
-  allocate_same_layout(out, in, axes);
+  allocate_same_layout(out, in, axes, encoder);
 
   // TODO: If out.size() < 1024 which will be a common case then write this in
   //       2 passes. Something like 32 * out.size() and then do a warp reduce.
@@ -250,14 +250,15 @@ void row_reduce_simple(
       using T = cuda_type_t<MLX_GET_TYPE(type_tag)>;
       using U = typename cu::ReduceResult<OP, T>::type;
 
-      // Cub doesn't like const pointers for vectorized loads. (sigh)
-      T* indata = const_cast<T*>(in.data<T>());
+      constexpr int N_READS = 16 / sizeof(T);
 
       // Calculate the grid and block dims
       size_t reductions = (plan.shape.back() + N_READS - 1) / N_READS;
       dim3 grid = get_2d_grid_dims(out.shape(), out.strides());
-      int threads = std::min(1024UL, reductions);
-      threads = ((threads + WARP_SIZE - 1) / WARP_SIZE) * WARP_SIZE;
+      int warps = (reductions + WARP_SIZE - 1) / WARP_SIZE;
+      warps /= 4;
+      warps = std::max(std::min(warps, 32), 1);
+      int threads = warps * WARP_SIZE;
       dim3 block(threads, 1, 1);
 
       // Pick the kernel
@@ -267,9 +268,10 @@ void row_reduce_simple(
         kernel = cu::row_reduce_simple<T, U, OP, N_READS, 2>;
       }
 
+      T* indata = const_cast<T*>(gpu_ptr<T>(in));
       int size = plan.shape.back();
       encoder.add_kernel_node(
-          kernel, grid, block, 0, indata, out.data<U>(), out.size(), size);
+          kernel, grid, block, 0, indata, gpu_ptr<U>(out), out.size(), size);
     });
   });
 }
@@ -282,11 +284,9 @@ void row_reduce_looped(
     const std::vector<int>& axes,
     const ReductionPlan& plan,
     cu::RowReduceArgs args) {
-  constexpr int N_READS = 8;
-
   // Allocate data for the output using in's layout to access them as
   // contiguously as possible.
-  allocate_same_layout(out, in, axes);
+  allocate_same_layout(out, in, axes, encoder);
 
   encoder.set_input_array(in);
   encoder.set_output_array(out);
@@ -295,34 +295,27 @@ void row_reduce_looped(
       using OP = MLX_GET_TYPE(reduce_type_tag);
       using T = cuda_type_t<MLX_GET_TYPE(type_tag)>;
       using U = typename cu::ReduceResult<OP, T>::type;
-      // Cub doesn't like const pointers for vectorized loads. (sigh)
-      T* indata = const_cast<T*>(in.data<T>());
+
+      constexpr int N_READS = 16 / sizeof(T);
 
       // Calculate the grid and block dims
       args.sort_access_pattern(in, axes);
       dim3 grid = get_2d_grid_dims(out.shape(), out.strides());
       size_t reductions = (args.row_size + N_READS - 1) / N_READS;
-      int threads = std::min(1024UL, reductions);
-      threads = ((threads + WARP_SIZE - 1) / WARP_SIZE) * WARP_SIZE;
+      int warps = (reductions + WARP_SIZE - 1) / WARP_SIZE;
+      warps /= 4;
+      warps = std::max(std::min(warps, 32), 1);
+      int threads = warps * WARP_SIZE;
       dim3 block(threads, 1, 1);
 
       // Pick the kernel
-      auto kernel = cu::row_reduce_looped<T, U, OP, 1, 32, N_READS>;
+      auto kernel = cu::row_reduce_looped<T, U, OP, 1, N_READS>;
       dispatch_reduce_ndim(args.reduce_ndim, [&](auto reduce_ndim) {
-        dispatch_block_dim(threads, [&](auto threads_constant) {
-          kernel = cu::row_reduce_looped<
-              T,
-              U,
-              OP,
-              reduce_ndim.value,
-              threads_constant.value,
-              N_READS>;
-          block.x = threads_constant.value;
-        });
+        kernel = cu::row_reduce_looped<T, U, OP, reduce_ndim.value, N_READS>;
       });
 
       encoder.add_kernel_node(
-          kernel, grid, block, 0, indata, out.data<U>(), out.size(), args);
+          kernel, grid, block, 0, gpu_ptr<T>(in), gpu_ptr<U>(out), args);
     });
   });
 }

mlx/backend/cuda/rms_norm.cu

@@ -176,9 +176,10 @@ void RMSNorm::eval_gpu(
   nvtx3::scoped_range r("RMSNorm::eval_gpu");
   auto& s = stream();
   auto& out = outputs[0];
+  auto& encoder = cu::get_command_encoder(s);
 
   // Make sure that the last dimension is contiguous.
-  auto set_output = [&s, &out](const array& x) {
+  auto set_output = [&s, &out, &encoder](const array& x) {
     bool no_copy = x.flags().contiguous && x.strides()[x.ndim() - 1] == 1;
     if (no_copy && x.ndim() > 1) {
       auto s = x.strides()[x.ndim() - 2];
@@ -189,7 +190,7 @@ void RMSNorm::eval_gpu(
         out.copy_shared_buffer(x);
       } else {
         out.set_data(
-            allocator::malloc(x.data_size() * x.itemsize()),
+            cu::malloc_async(x.data_size() * x.itemsize(), encoder.stream()),
             x.data_size(),
             x.strides(),
             x.flags());
@@ -209,7 +210,6 @@ void RMSNorm::eval_gpu(
   int32_t n_rows = x.data_size() / axis_size;
   int64_t w_stride = (w.ndim() == 1) ? w.strides()[0] : 0;
 
-  auto& encoder = cu::get_command_encoder(s);
   encoder.set_input_array(x);
   encoder.set_input_array(w);
   encoder.set_output_array(out);
@@ -223,9 +223,9 @@ void RMSNorm::eval_gpu(
           n_rows,
           block_dim(),
           0,
-          x.data<DataType>(),
-          w.data<DataType>(),
-          out.data<DataType>(),
+          gpu_ptr<DataType>(x),
+          gpu_ptr<DataType>(w),
+          gpu_ptr<DataType>(out),
           eps_,
           axis_size,
           w_stride);
@@ -274,7 +274,7 @@ void RMSNormVJP::eval_gpu(
     gx.copy_shared_buffer(g);
     g_in_gx = true;
   } else {
-    gx.set_data(allocator::malloc(gx.nbytes()));
+    gx.set_data(cu::malloc_async(gx.nbytes(), encoder.stream()));
   }
   if (g_copied && !g_in_gx) {
     encoder.add_temporary(g);
@@ -292,7 +292,7 @@ void RMSNormVJP::eval_gpu(
     if (!g_in_gx && donate_g) {
       gw_temp.copy_shared_buffer(g);
     } else {
-      gw_temp.set_data(allocator::malloc(gw_temp.nbytes()));
+      gw_temp.set_data(cu::malloc_async(gw_temp.nbytes(), encoder.stream()));
       encoder.add_temporary(gw_temp);
     }
   }
@@ -318,11 +318,11 @@ void RMSNormVJP::eval_gpu(
                 n_rows,
                 block_dim(),
                 0,
-                x.data<DataType>(),
-                w.data<DataType>(),
-                g.data<DataType>(),
-                gx.data<DataType>(),
-                gw_temp.data<DataType>(),
+                gpu_ptr<DataType>(x),
+                gpu_ptr<DataType>(w),
+                gpu_ptr<DataType>(g),
+                gpu_ptr<DataType>(gx),
+                gpu_ptr<DataType>(gw_temp),
                 eps_,
                 axis_size,
                 w_stride);

mlx/backend/cuda/rope.cu

@@ -250,6 +250,7 @@ void RoPE::eval_gpu(
   nvtx3::scoped_range r("RoPE::eval_gpu");
 
   auto& s = stream();
+  auto& encoder = cu::get_command_encoder(s);
   auto& in = inputs[0];
   auto& offset = inputs[1];
   auto& out = outputs[0];
@@ -291,14 +292,14 @@ void RoPE::eval_gpu(
       donated = true;
       out.copy_shared_buffer(in);
     } else {
-      out.set_data(allocator::malloc(out.nbytes()));
+      out.set_data(cu::malloc_async(out.nbytes(), encoder.stream()));
     }
     strides[0] = mat_size;
     strides[1] = in.strides()[ndim - 2];
     strides[2] = in.strides()[ndim - 1];
   } else if (dispatch_ndim == 3) {
     // Handle non-contiguous 3D inputs
-    out.set_data(allocator::malloc(out.nbytes()));
+    out.set_data(cu::malloc_async(out.nbytes(), encoder.stream()));
     strides[0] = in.strides()[ndim - 3];
     strides[1] = in.strides()[ndim - 2];
     strides[2] = in.strides()[ndim - 1];
@@ -319,7 +320,6 @@ void RoPE::eval_gpu(
   bool single = in.flags().row_contiguous && B == 1 && T == 1;
   bool with_freqs = inputs.size() == 3;
 
-  auto& encoder = cu::get_command_encoder(s);
   encoder.set_input_array(donated ? out : in);
   encoder.set_input_array(offset);
   if (with_freqs) {
@@ -340,9 +340,9 @@ void RoPE::eval_gpu(
               grid,
               block,
               0,
-              (donated ? out : in).data<DataType>(),
-              out.data<DataType>(),
-              offset.data<int32_t>(),
+              gpu_ptr<DataType>(donated ? out : in),
+              gpu_ptr<DataType>(out),
+              gpu_ptr<int32_t>(offset),
               scale_,
               std::log2(base_),
               mat_size,
@@ -357,10 +357,10 @@ void RoPE::eval_gpu(
               grid,
               block,
               0,
-              (donated ? out : in).data<DataType>(),
-              out.data<DataType>(),
-              offset.data<int32_t>(),
-              inputs[2].data<float>(),
+              gpu_ptr<DataType>(donated ? out : in),
+              gpu_ptr<DataType>(out),
+              gpu_ptr<int32_t>(offset),
+              gpu_ptr<float>(inputs[2]),
               scale_,
               mat_size,
               dims,
@@ -381,10 +381,10 @@ void RoPE::eval_gpu(
               grid,
               block,
               0,
-              (donated ? out : in).data<DataType>(),
-              out.data<DataType>(),
-              offset.data<int32_t>(),
-              inputs[2].data<float>(),
+              gpu_ptr<DataType>(donated ? out : in),
+              gpu_ptr<DataType>(out),
+              gpu_ptr<int32_t>(offset),
+              gpu_ptr<float>(inputs[2]),
               scale_,
               std::log2(base_),
               strides,
@@ -408,9 +408,9 @@ void RoPE::eval_gpu(
               grid,
               block,
               0,
-              (donated ? out : in).data<DataType>(),
-              out.data<DataType>(),
-              offset.data<int32_t>(),
+              gpu_ptr<DataType>(donated ? out : in),
+              gpu_ptr<DataType>(out),
+              gpu_ptr<int32_t>(offset),
               scale_,
               std::log2(base_),
               strides,