Skip to content

ci: add ROCm 7.14 support and support more architectures, drop ROCm 6.2 and 6.3 #2007

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
matthewdouglas merged 1 commit into from
Jul 16, 2026
+83 −18
Merged

ci: add ROCm 7.14 support and support more architectures, drop ROCm 6.2 and 6.3 #2007

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
62 changes: 51 additions & 11 deletions .github/scripts/build-rocm.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,16 +4,48 @@ set -xeuo pipefail
: "${RUNNER_OS:?RUNNER_OS must be set (Linux/Windows)}"
: "${ROCM_VERSION:?ROCM_VERSION must be set}"

rocm_version_at_least() {
local required_version="$1"
local current_major current_minor required_major required_minor

IFS=. read -r current_major current_minor _ <<< "${ROCM_VERSION}"
IFS=. read -r required_major required_minor _ <<< "${required_version}"

if ((current_major > required_major)); then
return 0
fi
if ((current_major < required_major)); then
return 1
fi
if ((current_minor >= required_minor)); then
return 0
fi
return 1
}

bnb_rocm_arch="gfx90a;gfx942;gfx1100;gfx1101;gfx1102;gfx1103"

# ROCm 6.4+ - Add RDNA4 and RDNA3.5 targets. Note we assume >=6.4.4.
[[ "${ROCM_VERSION}" == 6.4.* || "${ROCM_VERSION}" == 7.* ]] && bnb_rocm_arch="${bnb_rocm_arch};gfx1150;gfx1151;gfx1152;gfx1153;gfx1200;gfx1201"
if rocm_version_at_least "6.4"; then
bnb_rocm_arch="${bnb_rocm_arch};gfx1150;gfx1151;gfx1152;gfx1153;gfx1200;gfx1201"
fi

# ROCm 7.0+ - Add gfx950
[[ "${ROCM_VERSION}" == 7.* ]] && bnb_rocm_arch="${bnb_rocm_arch};gfx950"
if rocm_version_at_least "7.0"; then
bnb_rocm_arch="${bnb_rocm_arch};gfx950"
fi

# ROCm 7.14+ - Add CDNA1 and RDNA2 targets.
if rocm_version_at_least "7.14"; then
bnb_rocm_arch="${bnb_rocm_arch};gfx908;gfx1030;gfx1031;gfx1032;gfx1033;gfx1034;gfx1035;gfx1036"
fi

if [ "${RUNNER_OS}" == "Linux" ]; then
image=rocm/dev-ubuntu-22.04:${ROCM_VERSION}-complete
image_suffix="complete"
if rocm_version_at_least "7.14"; then
image_suffix="full"
fi
image=rocm/dev-ubuntu-22.04:${ROCM_VERSION}-${image_suffix}
echo "Using image $image"
docker run --rm -i \
-w /src -v "$PWD:/src" "$image" sh -c \
Expand All @@ -23,15 +55,23 @@ if [ "${RUNNER_OS}" == "Linux" ]; then
else
bnb_rocm_arch="gfx1100;gfx1101;gfx1102;gfx1150;gfx1151;gfx1200;gfx1201"

# Install ROCm SDK wheels from repo.radeon.com.
rocm_base_url="https://repo.radeon.com/rocm/windows/rocm-rel-${ROCM_VERSION}"
pip install \
"${rocm_base_url}/rocm_sdk_core-${ROCM_VERSION}-py3-none-win_amd64.whl" \
"${rocm_base_url}/rocm_sdk_devel-${ROCM_VERSION}-py3-none-win_amd64.whl" \
"${rocm_base_url}/rocm_sdk_libraries_custom-${ROCM_VERSION}-py3-none-win_amd64.whl" \
"${rocm_base_url}/rocm-${ROCM_VERSION}.tar.gz"
if rocm_version_at_least "7.14"; then
# Add RDNA2 and additional RDNA3.5 targets.
bnb_rocm_arch="${bnb_rocm_arch};gfx1030;gfx1031;gfx1032;gfx1033;gfx1034;gfx1035;gfx1036;gfx1152;gfx1153"

pip install --index-url https://repo.amd.com/rocm/whl-multi-arch/ "rocm[libraries,devel]==${ROCM_VERSION}"
else
# Install ROCm SDK wheels from repo.radeon.com.
rocm_base_url="https://repo.radeon.com/rocm/windows/rocm-rel-${ROCM_VERSION}"
pip install \
"${rocm_base_url}/rocm_sdk_core-${ROCM_VERSION}-py3-none-win_amd64.whl" \
"${rocm_base_url}/rocm_sdk_devel-${ROCM_VERSION}-py3-none-win_amd64.whl" \
"${rocm_base_url}/rocm_sdk_libraries_custom-${ROCM_VERSION}-py3-none-win_amd64.whl" \
"${rocm_base_url}/rocm-${ROCM_VERSION}.tar.gz"

fi

# Expand the devel tarball
# Expand the devel tarball.
rocm-sdk init

ROCM_PATH="$(rocm-sdk path --root | tr '\\' '/')"
Expand Down
4 changes: 3 additions & 1 deletion .github/workflows/python-package.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -118,10 +118,12 @@ jobs:
strategy:
matrix:
os: [ubuntu-22.04]
rocm_version: ["6.2.4", "6.3.4", "6.4.4", "7.0.2", "7.1.1", "7.2.4"]
rocm_version: ["6.4.4", "7.0.2", "7.1.1", "7.2.4", "7.14.0"]
include:
- os: windows-2025
rocm_version: "7.2.1"
- os: windows-2025
rocm_version: "7.14.0"
runs-on: ${{ matrix.os }}
steps:
- uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
Expand Down
35 changes: 29 additions & 6 deletions docs/source/installation.mdx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -241,13 +241,13 @@ The currently distributed `bitsandbytes` are built with the following configurat

| **OS** | **ROCm** | **Targets**
|--------------------|----------|---------------------------------------------------------------------|
| **Linux x86-64** | 6.2.4 | CDNA: gfx90a, gfx942 / RDNA: gfx1100, gfx1101, gfx1102, gfx1103
| **Linux x86-64** | 6.3.4 | CDNA: gfx90a, gfx942 / RDNA: gfx1100, gfx1101, gfx1102, gfx1103
| **Linux x86-64** | 6.4.4 | CDNA: gfx90a, gfx942 / RDNA: gfx1100, gfx1101, gfx1102, gfx1103, gfx1150, gfx1151, gfx1152, gfx1153, gfx1200, gfx1201
| **Linux x86-64** | 7.0.2 | CDNA: gfx90a, gfx942, gfx950 / RDNA: gfx1100, gfx1101, gfx1102, gfx1103, gfx1150, gfx1151, gfx1152, gfx1153, gfx1200, gfx1201
| **Linux x86-64** | 7.1.1 | CDNA: gfx90a, gfx942, gfx950 / RDNA: gfx1100, gfx1101, gfx1102, gfx1103, gfx1150, gfx1151, gfx1152, gfx1153, gfx1200, gfx1201
| **Linux x86-64** | 7.2.4 | CDNA: gfx90a, gfx942, gfx950 / RDNA: gfx1100, gfx1101, gfx1102, gfx1103, gfx1150, gfx1151, gfx1152, gfx1153, gfx1200, gfx1201
| **Linux x86-64** | 7.14.0 | CDNA: gfx908, gfx90a, gfx942, gfx950 / RDNA: gfx1030, gfx1031, gfx1032, gfx1033, gfx1034, gfx1035, gfx1036, gfx1100, gfx1101, gfx1102, gfx1103, gfx1150, gfx1151, gfx1152, gfx1153, gfx1200, gfx1201
| **Windows x86-64** | 7.2.1 | RDNA: gfx1100, gfx1101, gfx1102, gfx1150, gfx1151, gfx1200, gfx1201
| **Windows x86-64** | 7.14.0 | RDNA: gfx1030, gfx1031, gfx1032, gfx1033, gfx1034, gfx1035, gfx1036, gfx1100, gfx1101, gfx1102, gfx1150, gfx1151, gfx1152, gfx1153, gfx1200, gfx1201

Use `pip` or `uv` to install the latest release:

Expand All @@ -257,14 +257,14 @@ pip install bitsandbytes

### Compile from Source[[rocm-compile]]

bitsandbytes can be compiled from ROCm 6.2 - ROCm 7.2.4. See the `CMakeLists.txt` for additional options.
bitsandbytes can be compiled from ROCm 6.4 - ROCm 7.14.0. See the `CMakeLists.txt` for additional options.

<hfoptions id="rocm-source">
<hfoption id="Linux">

To compile from source, you need CMake >= **3.31.6** and Python >= **3.10** installed. Make sure you have a compiler installed to compile C++ (`gcc`, `make`, headers, etc.).

You should also have a ROCm installation (system-wide or via Docker). The current minimum supported version is **6.2**.
You should also have a ROCm installation (system-wide or via Docker). The current minimum supported version is **6.4**.

```bash
# Install bitsandbytes from source
Expand All @@ -284,10 +284,33 @@ pip install -e . # `-e` for "editable" install, when developing BNB (otherwise

Compilation on Windows requires Visual Studio 2022 with C++ support, CMake, Ninja, and Python >= **3.10**.

Instead of a system-wide ROCm installation, you can use the pip-installable ROCm SDK wheels from [repo.radeon.com](https://repo.radeon.com/rocm/windows/):
Instead of a system-wide ROCm installation, use the pip-installable ROCm SDK.

For ROCm 7.14.0, install the current multi-architecture SDK from [repo.amd.com](https://repo.amd.com/rocm/whl-multi-arch/):

```bash
pip install ninja cmake
# Replace gfx1100 with your GPU architecture. To support multiple GPU
# architectures, add multiple device extras (for example, device-gfx1100,device-gfx1151).
pip install --index-url https://repo.amd.com/rocm/whl-multi-arch/ \
"rocm[libraries,devel,device-gfx1100]==7.14.0"

# Expand the devel tarball and link the installed device files.
rocm-sdk init

# Set ROCM_PATH and activate Visual Studio environment, then build
export ROCM_PATH="$(rocm-sdk path --root)"
export PATH="${ROCM_PATH}/bin:${PATH}"
git clone https://github.com/bitsandbytes-foundation/bitsandbytes.git && cd bitsandbytes/
# Use the same GPU architecture selected above. Separate multiple architectures with semicolons.
cmake -G Ninja -DCOMPUTE_BACKEND=hip -DBNB_ROCM_ARCH="gfx1100" -DCMAKE_BUILD_TYPE=Release -DCMAKE_HIP_COMPILER_ROCM_ROOT="${ROCM_PATH}" -S .
cmake --build . --config Release
pip install .
```

For the legacy ROCm 7.2.1 SDK, use the direct wheels from [repo.radeon.com](https://repo.radeon.com/rocm/windows/):

```bash
# Install ROCm SDK wheels (adjust version as needed)
pip install ninja cmake
pip install \
https://repo.radeon.com/rocm/windows/rocm-rel-7.2.1/rocm_sdk_core-7.2.1-py3-none-win_amd64.whl \
Expand Down
Loading