[main] Upgrade to CUDA 12.9

.ci_support/linux_aarch64_channel_targetsconda-forge_maincuda_compiler_version12.9is_rcFalse.yaml → .ci_support/linux_aarch64_c_compiler_version14channel_targetsconda-forge_maincuda_compiler_version12.9cxx_compiler_version14is_rcFalse.yaml

@@ -3,7 +3,7 @@ blas_impl:
 c_compiler:
 - gcc
 c_compiler_version:
-- '13'
+- '14'
 c_stdlib:
 - sysroot
 c_stdlib_version:
@@ -23,7 +23,7 @@ cudnn:
 cxx_compiler:
 - gxx
 cxx_compiler_version:
-- '13'
+- '14'
 docker_image:
 - quay.io/condaforge/linux-anvil-x86_64:alma9
 github_actions_labels:

.ci_support/migrations/cuda129.yaml

@@ -7,40 +7,51 @@ __migrator:
   build_number:
     1
   paused: false
-  use_local: true
   override_cbc_keys:
     - cuda_compiler_stub
   check_solvable: false
   primary_key: cuda_compiler_version
   ordering:
     cuda_compiler_version:
-      - 11.8
       - 12.4
       - 12.6
       - 12.8
       - None
       - 12.9
+      # to allow manual opt-in for CUDA 11.8, see
+      # https://github.com/conda-forge/conda-forge-pinning-feedstock/pull/7472
+      # must be last due to how cuda_compiler ordering in that migrator works
+      - 11.8
   commit_message: |
     Upgrade to CUDA 12.9
 
-    With CUDA 12.9, the following new architectures are added `sm_100`, `sm_101` and `sm_120`.
-    To build for these architectures, maintainers will need to add these to list of architectures
-    that their package builds for.
+    CUDA 12.8 added support for architectures `sm_100`, `sm_101` and `sm_120`,
+    while CUDA 12.9 further added `sm_103` and `sm_121`. To build for these,
+    maintainers will need to modify their existing list of specified architectures
+    (e.g. `CMAKE_CUDA_ARCHITECTURES`, `TORCH_CUDA_ARCH_LIST`, etc.)
+    for their package. A good balance between broad support and storage
+    footprint (resp. compilation time) is to add `sm_100` and `sm_120`.
+
+    Since CUDA 12.8, the conda-forge nvcc package now sets `CUDAARCHS` and
+    `TORCH_CUDA_ARCH_LIST` in its activation script to a string containing all
+    of the supported real architectures plus the virtual architecture of the
+    latest. Recipes for packages who use these variables to control their build
+    but do not want to build for all supported architectures will need to override
+    these variables in their build script.
 
     ref: https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html#new-features
 
 cuda_compiler_version:         # [((linux and (x86_64 or aarch64)) or win64) and os.environ.get("CF_CUDA_ENABLED", "False") == "True"]
-  - 12.9                       # [(linux and (x86_64 or aarch64)) and os.environ.get("CF_CUDA_ENABLED", "False") == "True"]
-  - 12.8                       # [win64 and os.environ.get("CF_CUDA_ENABLED", "False") == "True"]
+  - 12.9                       # [((linux and (x86_64 or aarch64)) or win64) and os.environ.get("CF_CUDA_ENABLED", "False") == "True"]
+
+cuda_compiler_version_min:     # [((linux and (x86_64 or aarch64)) or win64) and os.environ.get("CF_CUDA_ENABLED", "False") == "True"]
+  - 12.9                       # [((linux and (x86_64 or aarch64)) or win64) and os.environ.get("CF_CUDA_ENABLED", "False") == "True"]
 
-c_compiler_version:            # [linux and os.environ.get("CF_CUDA_ENABLED", "False") == "True"]
-  - 14                         # [(linux and x86_64) and os.environ.get("CF_CUDA_ENABLED", "False") == "True"]
-  - 13                         # [(linux and aarch64) and os.environ.get("CF_CUDA_ENABLED", "False") == "True"]
+c_compiler_version:            # [(linux and (x86_64 or aarch64)) and os.environ.get("CF_CUDA_ENABLED", "False") == "True"]
+  - 14                         # [(linux and (x86_64 or aarch64)) and os.environ.get("CF_CUDA_ENABLED", "False") == "True"]
 
-cxx_compiler_version:          # [linux and os.environ.get("CF_CUDA_ENABLED", "False") == "True"]
-  - 14                         # [(linux and x86_64) and os.environ.get("CF_CUDA_ENABLED", "False") == "True"]
-  - 13                         # [(linux and aarch64) and os.environ.get("CF_CUDA_ENABLED", "False") == "True"]
+cxx_compiler_version:          # [(linux and (x86_64 or aarch64)) and os.environ.get("CF_CUDA_ENABLED", "False") == "True"]
+  - 14                         # [(linux and (x86_64 or aarch64)) and os.environ.get("CF_CUDA_ENABLED", "False") == "True"]
 
-fortran_compiler_version:      # [linux and os.environ.get("CF_CUDA_ENABLED", "False") == "True"]
-  - 14                         # [(linux and x86_64) and os.environ.get("CF_CUDA_ENABLED", "False") == "True"]
-  - 13                         # [(linux and aarch64) and os.environ.get("CF_CUDA_ENABLED", "False") == "True"]
+fortran_compiler_version:      # [(linux and (x86_64 or aarch64)) and os.environ.get("CF_CUDA_ENABLED", "False") == "True"]
+  - 14                         # [(linux and (x86_64 or aarch64)) and os.environ.get("CF_CUDA_ENABLED", "False") == "True"]

.ci_support/win_64_channel_targetsconda-forge_maincuda_compiler_version12.8is_rcFalse.yaml → .ci_support/win_64_channel_targetsconda-forge_maincuda_compiler_version12.9is_rcFalse.yaml

@@ -11,7 +11,7 @@ channel_targets:
 cuda_compiler:
 - cuda-nvcc
 cuda_compiler_version:
-- '12.8'
+- '12.9'
 cudnn:
 - '9'
 cxx_compiler:

.github/workflows/conda-build.yml

@@ -49,25 +49,25 @@ jobs:
             runs_on: ['cirun-openstack-gpu-2xlarge--${{ github.run_id }}-linux_64_blas_implmklchannel_targetscond_h38f93959', 'linux', 'x64', 'self-hosted']
             DOCKER_IMAGE: quay.io/condaforge/linux-anvil-x86_64:alma9
             CONDA_FORGE_DOCKER_RUN_ARGS: "--gpus all"
-          - CONFIG: linux_aarch64_channel_targetsconda-forge_maincuda_compiler_version12.9is_rcFalse
-            SHORT_CONFIG: linux_aarch64_channel_targetsconda-forge_heefc8d83
+          - CONFIG: linux_aarch64_c_compiler_version13channel_targetsconda-forge_maincuda_compiler_versionNonecxx_compiler_version13is_rcFalse
+            SHORT_CONFIG: linux_aarch64_c_compiler_version13channe_hd91c52e4
             UPLOAD_PACKAGES: True
             os: ubuntu
-            runs_on: ['cirun-openstack-gpu-2xlarge--${{ github.run_id }}-linux_aarch64_channel_targetsconda-forge_heefc8d83', 'linux', 'x64', 'self-hosted']
+            runs_on: ['cirun-openstack-gpu-2xlarge--${{ github.run_id }}-linux_aarch64_c_compiler_version13channe_hd91c52e4', 'linux', 'x64', 'self-hosted']
             DOCKER_IMAGE: quay.io/condaforge/linux-anvil-x86_64:alma9
             CONDA_FORGE_DOCKER_RUN_ARGS: "--gpus all"
-          - CONFIG: linux_aarch64_channel_targetsconda-forge_maincuda_compiler_versionNoneis_rcFalse
-            SHORT_CONFIG: linux_aarch64_channel_targetsconda-forge_h56c2c839
+          - CONFIG: linux_aarch64_c_compiler_version14channel_targetsconda-forge_maincuda_compiler_version12.9cxx_compiler_version14is_rcFalse
+            SHORT_CONFIG: linux_aarch64_c_compiler_version14channe_hecba00a8
             UPLOAD_PACKAGES: True
             os: ubuntu
-            runs_on: ['cirun-openstack-gpu-2xlarge--${{ github.run_id }}-linux_aarch64_channel_targetsconda-forge_h56c2c839', 'linux', 'x64', 'self-hosted']
+            runs_on: ['cirun-openstack-gpu-2xlarge--${{ github.run_id }}-linux_aarch64_c_compiler_version14channe_hecba00a8', 'linux', 'x64', 'self-hosted']
             DOCKER_IMAGE: quay.io/condaforge/linux-anvil-x86_64:alma9
             CONDA_FORGE_DOCKER_RUN_ARGS: "--gpus all"
-          - CONFIG: win_64_channel_targetsconda-forge_maincuda_compiler_version12.8is_rcFalse
-            SHORT_CONFIG: win_64_channel_targetsconda-forge_maincu_hca575dce
+          - CONFIG: win_64_channel_targetsconda-forge_maincuda_compiler_version12.9is_rcFalse
+            SHORT_CONFIG: win_64_channel_targetsconda-forge_maincu_hda40b8c4
             UPLOAD_PACKAGES: True
             os: windows
-            runs_on: ['cirun-azure-windows-4xlarge--${{ github.run_id }}-win_64_channel_targetsconda-forge_maincu_hca575dce', 'windows', 'x64', 'self-hosted']
+            runs_on: ['cirun-azure-windows-4xlarge--${{ github.run_id }}-win_64_channel_targetsconda-forge_maincu_hda40b8c4', 'windows', 'x64', 'self-hosted']
           - CONFIG: win_64_channel_targetsconda-forge_maincuda_compiler_versionNoneis_rcFalse
             SHORT_CONFIG: win_64_channel_targetsconda-forge_maincu_hc68ac914
             UPLOAD_PACKAGES: True

recipe/meta.yaml

@@ -1,6 +1,6 @@
 # if you wish to build release candidate number X, append the version string with ".rcX"
 {% set version = "2.7.1" %}
-{% set build = 3 %}
+{% set build = 4 %}
 
 # Use a higher build number for the CUDA variant, to ensure that it's
 # preferred by conda's solver, and it's preferentially