Merge branch 'main' into build-cmake

Author: mgorny

.clang-format

@@ -0,0 +1,36 @@
+---
+BasedOnStyle: LLVM
+AlignAfterOpenBracket: BlockIndent
+BinPackArguments: true
+BinPackParameters: true
+BracedInitializerIndentWidth: 4
+ColumnLimit: 120
+Cpp11BracedListStyle: true
+IndentWidth: 4
+IndentWrappedFunctionNames: true
+PointerAlignment: Left
+SeparateDefinitionBlocks: Always
+Standard: c++17
+StatementMacros:
+  - 'MAKE_PreconditionOptimizer32bit1State'
+  - 'MAKE_PreconditionOptimizer32bit2State'
+  - 'MAKE_PreconditionStatic8bit1State'
+  - 'MAKE_PreconditionStatic8bit2State'
+  - 'MAKE_Optimizer32bit1State'
+  - 'MAKE_optimizerStatic8bit1State'
+  - 'MAKE_optimizerStatic8bit2State'
+  - 'MAKE_OptimizerStatic8bit1StateBlockwise'
+  - 'MAKE_OptimizerStatic8bit2StateBlockwise'
+  - 'MAKE_kQuantizeBlockwise'
+  - 'MAKE_BLOCKWISE8'
+  - 'MAKE_ELEMENTWISE_FUNC'
+  - 'CMAKE_ELEMENTWISE_FUNC'
+  - 'MAKE_FUNC8'
+  - 'MAKE_FUNC32'
+  - 'MAKE_CBLOCKWISE8'
+  - 'MAKE_CFUNC8'
+  - 'MAKE_CFUNC32'
+
+UseTab: Never
+
+...

.git-blame-ignore-revs

@@ -15,3 +15,6 @@ ea7c14f8ef64924f2d0ff80df3cdabf2c7299848
 
 # CHANGELOG: to reverse chron order + mdformat
 4743ff0d43e04e4cc3e5d8b9e7cd016c0defa36d
+
+# Apply clang-format
+4955d136ae083c2be1236d8915913166e1790aad

.github/scripts/build-cuda.sh

@@ -11,14 +11,14 @@ if [[ -v cuda_targets ]]; then
 elif [ "${build_arch}" = "aarch64" ]; then
     build_capability="75;80;90"
 
-    # CUDA 12.8: Add sm100
-    [[ "${cuda_version}" == 12.8.* ]] && build_capability="75;80;90;100"
+    # CUDA 12.8+: Add sm100/sm120
+    [[ "${cuda_version}" == 12.8.* || "${cuda_version}" == 12.9.* ]] && build_capability="75;80;90;100;120"
 else
     # By default, target Maxwell through Hopper.
     build_capability="50;52;60;61;70;75;80;86;89;90"
 
-    # CUDA 12.8: Add sm100 and sm120; remove < sm75 to align with PyTorch 2.7+cu128 minimum
-    [[ "${cuda_version}" == 12.8.* ]] && build_capability="75;80;86;89;90;100;120"
+    # CUDA 12.8+: Add sm100 and sm120; remove < sm75 to align with PyTorch 2.7+cu128 minimum
+    [[ "${cuda_version}" == 12.8.* || "${cuda_version}" == 12.9.* ]] && build_capability="75;80;86;89;90;100;120"
 fi
 
 [[ "${build_os}" = windows-* ]] && python3 -m pip install ninja

.github/scripts/build-rocm.sh

@@ -0,0 +1,21 @@
+#!/bin/bash
+declare build_arch
+declare build_os
+declare rocm_version
+
+set -xeuo pipefail
+bnb_rocm_arch="gfx90a;gfx942;gfx1100"
+if [ "${build_os:0:6}" == ubuntu ]; then
+	image=rocm/dev-ubuntu-22.04:${rocm_version}-complete
+	echo "Using image $image"
+	docker run --rm --platform "linux/$build_arch" -i \
+		-w /src -v "$PWD:/src" "$image" sh -c \
+		"apt-get update \
+      && DEBIAN_FRONTEND=noninteractive apt-get install -y --no-install-recommends cmake \
+      && cmake -DCOMPUTE_BACKEND=hip -DBNB_ROCM_ARCH=\"${bnb_rocm_arch}\" . \
+      && cmake --build ."
+fi
+
+output_dir="output/${build_os}/${build_arch}"
+mkdir -p "${output_dir}"
+(shopt -s nullglob && cp bitsandbytes/*.{so,dylib,dll} "${output_dir}")

.github/workflows/python-package.yml

@@ -72,16 +72,17 @@ jobs:
           - os: windows-latest
             arch: x86_64
         cuda_version:
-          ["11.8.0", "12.0.1", "12.1.1", "12.2.2", "12.3.2", "12.4.1", "12.5.1", "12.6.3", "12.8.1"]
+          ["11.8.0", "12.0.1", "12.1.1", "12.2.2", "12.3.2", "12.4.1", "12.5.1", "12.6.3", "12.8.1", "12.9.1"]
     runs-on: ${{ matrix.os }}
     steps:
       - uses: actions/checkout@v4
         # Windows: We install Cuda on the agent (slow)
-      - uses: Jimver/[email protected].22
+      - uses: Jimver/cuda-toolkit@c35baa1a18fd1fc9dcf47c5bd839bf30559c0bc3 # v0.2.24
         if: startsWith(matrix.os, 'windows')
         id: cuda-toolkit
         with:
-          cuda: ${{ matrix.cuda_version }}
+          # Temporary: Use CUDA 12.9.0 for Windows until 12.9.1 is supported with this action.
+          cuda: ${{ matrix.cuda_version == '12.9.1' && '12.9.0' || matrix.cuda_version }}
           method: "network"
           sub-packages: '["nvcc","cudart","cusparse","cublas","thrust","nvrtc_dev","cublas_dev","cusparse_dev"]'
           linux-local-args: '["--toolkit"]'
@@ -102,10 +103,55 @@ jobs:
           path: output/*
           retention-days: 7
 
+  build-shared-libs-rocm:
+    strategy:
+      matrix:
+        os: [ubuntu-22.04]
+        arch: [x86_64]
+        rocm_version:
+          ["6.1.2", "6.2.4", "6.3.2"]
+    runs-on: ${{ matrix.os }}
+    steps:
+      - uses: actions/checkout@v4
+      - name: Set up Docker multiarch
+        uses: docker/setup-qemu-action@v3
+      - name: Clean up disk space
+        run: |
+          sudo rm -rf \
+              /usr/share/dotnet \
+              /opt/ghc \
+              "/usr/local/share/boost" \
+              "$AGENT_TOOLSDIRECTORY" \
+              /opt/hostedtoolcache \
+              /opt/google/chrome \
+              /opt/microsoft/msedge \
+              /opt/microsoft/powershell \
+              /opt/pipx \
+              /usr/lib/mono \
+              /usr/local/julia* \
+              /usr/local/lib/android \
+              /usr/local/lib/node_modules \
+              /usr/local/share/chromium \
+              /usr/local/share/powershell \
+              /usr/share/swift
+      - name: Build C++
+        run: bash .github/scripts/build-rocm.sh
+        env:
+          build_os: ${{ matrix.os }}
+          build_arch: ${{ matrix.arch }}
+          rocm_version: ${{ matrix.rocm_version }}
+      - name: Upload build artifact
+        uses: actions/upload-artifact@v4
+        with:
+          name: shared_library_rocm_${{ matrix.os }}_${{ matrix.arch }}_${{ matrix.rocm_version }}
+          path: output/*
+          retention-days: 7
+
   build-wheels:
     needs:
       - build-shared-libs
       - build-shared-libs-cuda
+      - build-shared-libs-rocm
     strategy:
       matrix:
         os: [ubuntu-22.04, ubuntu-22.04-arm, windows-latest, macos-latest]
@@ -173,6 +219,7 @@ jobs:
           merge-multiple: true
 
       - name: Inspect tmp directory after downloading artifacts
+
         run: |
           ls -alFR tmp/
           WHEEL_COUNT=$(find tmp/ -type f -name "*.whl" | wc -l)
@@ -210,6 +257,7 @@ jobs:
       - uses: actions/checkout@v4
         with:
           path: repo
+
       - name: Delete old pre-release (if exists)
         run: |
           cd repo && gh release delete continuous-release_main --cleanup-tag -y