Enable linux ppc64le

.ci_support/linux_ppc64le_c_compiler_version12channel_targetsconda-forge_maincuda_compilercuda-nvcccuda_compiler_version12.4cxx_compiler_version12is_rcFalse.yaml

@@ -0,0 +1,82 @@
+blas_impl:
+- generic
+c_compiler:
+- gcc
+c_compiler_version:
+- '12'
+c_stdlib:
+- sysroot
+c_stdlib_version:
+- '2.17'
+cdt_name:
+- conda
+channel_sources:
+- conda-forge
+channel_targets:
+- conda-forge main
+cuda_compiler:
+- cuda-nvcc
+cuda_compiler_version:
+- '12.4'
+cxx_compiler:
+- gxx
+cxx_compiler_version:
+- '12'
+docker_image:
+- quay.io/condaforge/linux-anvil-x86_64:alma9
+github_actions_labels:
+- cirun-openstack-gpu-2xlarge
+is_rc:
+- 'False'
+libabseil:
+- '20250127'
+libblas:
+- 3.9.* *netlib
+libcblas:
+- 3.9.* *netlib
+liblapack:
+- 3.9.* *netlib
+libprotobuf:
+- 5.29.3
+libtorch:
+- '2.5'
+megabuild:
+- 'true'
+mkl:
+- '2024'
+nccl:
+- '2'
+numpy:
+- '2.0'
+- '2.0'
+- '2.0'
+- '2'
+- '2.0'
+orc:
+- 2.1.1
+pin_run_as_build:
+  python:
+    min_pin: x.x
+    max_pin: x.x
+python:
+- 3.10.* *_cpython
+- 3.11.* *_cpython
+- 3.12.* *_cpython
+- 3.13.* *_cp313
+- 3.9.* *_cpython
+pytorch:
+- '2.5'
+target_platform:
+- linux-ppc64le
+zip_keys:
+- - c_compiler_version
+  - cxx_compiler_version
+  - cuda_compiler
+  - cuda_compiler_version
+  - docker_image
+- - channel_targets
+  - is_rc
+- - python
+  - numpy
+zlib:
+- '1'

.ci_support/linux_ppc64le_c_compiler_version13channel_targetsconda-forge_maincuda_compilerNonecuda_compiler_versionNonecxx_compiler_version13is_rcFalse.yaml

@@ -0,0 +1,82 @@
+blas_impl:
+- generic
+c_compiler:
+- gcc
+c_compiler_version:
+- '13'
+c_stdlib:
+- sysroot
+c_stdlib_version:
+- '2.17'
+cdt_name:
+- conda
+channel_sources:
+- conda-forge
+channel_targets:
+- conda-forge main
+cuda_compiler:
+- None
+cuda_compiler_version:
+- None
+cxx_compiler:
+- gxx
+cxx_compiler_version:
+- '13'
+docker_image:
+- quay.io/condaforge/linux-anvil-x86_64:alma9
+github_actions_labels:
+- cirun-openstack-gpu-2xlarge
+is_rc:
+- 'False'
+libabseil:
+- '20250127'
+libblas:
+- 3.9.* *netlib
+libcblas:
+- 3.9.* *netlib
+liblapack:
+- 3.9.* *netlib
+libprotobuf:
+- 5.29.3
+libtorch:
+- '2.5'
+megabuild:
+- 'true'
+mkl:
+- '2024'
+nccl:
+- '2'
+numpy:
+- '2.0'
+- '2.0'
+- '2.0'
+- '2'
+- '2.0'
+orc:
+- 2.1.1
+pin_run_as_build:
+  python:
+    min_pin: x.x
+    max_pin: x.x
+python:
+- 3.10.* *_cpython
+- 3.11.* *_cpython
+- 3.12.* *_cpython
+- 3.13.* *_cp313
+- 3.9.* *_cpython
+pytorch:
+- '2.5'
+target_platform:
+- linux-ppc64le
+zip_keys:
+- - c_compiler_version
+  - cxx_compiler_version
+  - cuda_compiler
+  - cuda_compiler_version
+  - docker_image
+- - channel_targets
+  - is_rc
+- - python
+  - numpy
+zlib:
+- '1'

.github/workflows/conda-build.yml

@@ -57,6 +57,18 @@ jobs:
             runs_on: ['cirun-openstack-gpu-2xlarge--${{ github.run_id }}-linux_aarch64_c_compiler_version13c_h94771c6c09', 'linux', 'x64', 'self-hosted']
             DOCKER_IMAGE: quay.io/condaforge/linux-anvil-x86_64:alma9
             CONDA_FORGE_DOCKER_RUN_ARGS: "--gpus all"
+          - CONFIG: linux_ppc64le_c_compiler_version12channel_targetsconda-forge_maincuda_compilercuda-nvcccuda_compiler_version12.4cxx_compiler_version12is_rcFalse
+            UPLOAD_PACKAGES: True
+            os: ubuntu
+            runs_on: ['cirun-openstack-gpu-2xlarge--${{ github.run_id }}-linux_ppc64le_c_compiler_version12c_ha7f0e70c1b', 'linux', 'x64', 'self-hosted']
+            DOCKER_IMAGE: quay.io/condaforge/linux-anvil-x86_64:alma9
+            CONDA_FORGE_DOCKER_RUN_ARGS: "--gpus all"
+          - CONFIG: linux_ppc64le_c_compiler_version13channel_targetsconda-forge_maincuda_compilerNonecuda_compiler_versionNonecxx_compiler_version13is_rcFalse
+            UPLOAD_PACKAGES: True
+            os: ubuntu
+            runs_on: ['cirun-openstack-gpu-2xlarge--${{ github.run_id }}-linux_ppc64le_c_compiler_version13c_hcf0949fca5', 'linux', 'x64', 'self-hosted']
+            DOCKER_IMAGE: quay.io/condaforge/linux-anvil-x86_64:alma9
+            CONDA_FORGE_DOCKER_RUN_ARGS: "--gpus all"
           - CONFIG: win_64_channel_targetsconda-forge_maincuda_compilerNonecuda_compiler_versionNoneis_rcFalse
             UPLOAD_PACKAGES: True
             os: windows

conda-forge.yml

@@ -10,6 +10,7 @@ bot:
   - v2.5.x
 build_platform:
   linux_aarch64: linux_64
+  linux_ppc64le: linux_64
   osx_arm64: osx_64
 conda_build:
   pkg_format: '2'
@@ -26,5 +27,6 @@ github_actions:
 provider:
   linux_64: github_actions
   linux_aarch64: azure
+  linux_ppc64le: azure
   win_64: github_actions
 test: native_and_emulated

recipe/build.sh

@@ -178,7 +178,9 @@ elif [[ ${cuda_compiler_version} != "None" ]]; then
     # with no NVIDIA GPUs.
     export USE_MKLDNN=1
     export USE_CUDA=1
-    export USE_CUFILE=1
+    if [[ "$target_platform" != "linux-ppc64le" ]]; then
+      export USE_CUFILE=1
+    fi
     # PyTorch has multiple different bits of logic finding CUDA, override
     # all of them.
     export CUDAToolkit_BIN_DIR=${BUILD_PREFIX}/bin
@@ -193,6 +195,9 @@ elif [[ ${cuda_compiler_version} != "None" ]]; then
         linux-aarch64)
             export CUDAToolkit_TARGET_DIR=${PREFIX}/targets/sbsa-linux
             ;;
+        linux-ppc64le)
+            export CUDAToolkit_TARGET_DIR=${PREFIX}/targets/sbsa-linux
+            ;;
         *)
             echo "unknown CUDA arch, edit build.sh"
             exit 1

recipe/meta.yaml

@@ -1,6 +1,6 @@
 # if you wish to build release candidate number X, append the version string with ".rcX"
 {% set version = "2.6.0" %}
-{% set build = 2 %}
+{% set build = 3 %}
 
 # Use a higher build number for the CUDA variant, to ensure that it's
 # preferred by conda's solver, and it's preferentially
@@ -122,7 +122,7 @@ requirements:
     - rsync     # [unix]
   host:
     # GPU requirements
-    - cudnn                           # [cuda_compiler_version != "None"]
+    - cudnn                           # [cuda_compiler_version != "None" and not ppc64le]
     - nccl                            # [cuda_compiler_version != "None" and linux]
     - magma                           # [cuda_compiler_version != "None"]
     - cuda-version {{ cuda_compiler_version }}  # [cuda_compiler_version != "None"]
@@ -135,10 +135,10 @@ requirements:
     - cuda-nvtx-dev
     - cuda-nvml-dev
     - cuda-profiler-api
-    - cusparselt
+    - cusparselt     # [not ppc64le]
     - libcublas-dev
-    - libcudss-dev
-    - libcufile-dev  # [linux]
+    - libcudss-dev   # [not ppc64le]
+    - libcufile-dev  # [linux and not ppc64le]
     - libcufft-dev
     - libcurand-dev
     - libcusolver-dev
@@ -273,7 +273,7 @@ outputs:
       host:
         - {{ pin_subpackage('libtorch', exact=True) }}
         # GPU requirements
-        - cudnn                           # [cuda_compiler_version != "None"]
+        - cudnn                           # [cuda_compiler_version != "None" and not ppc64le]
         - nccl                            # [cuda_compiler_version != "None" and linux]
         - cuda-version {{ cuda_compiler_version }}  # [cuda_compiler_version != "None"]
         - nvtx-c                          # [cuda_compiler_version != "None"]
@@ -286,10 +286,10 @@ outputs:
         - cuda-nvtx-dev
         - cuda-nvml-dev
         - cuda-profiler-api
-        - cusparselt
+        - cusparselt     # [not ppc64le]
         - libcublas-dev
-        - libcudss-dev
-        - libcufile-dev  # [linux]
+        - libcudss-dev   # [not ppc64le]
+        - libcufile-dev  # [linux and not ppc64le]
         - libcufft-dev
         - libcurand-dev
         - libcusolver-dev