Update CUDA arches for CUDA 11.8 / sm_90 support.

.ci_support/linux_64_cuda_compiler_version11.1cxx_compiler_version10.yaml → .ci_support/linux_64_cuda_compiler_version11.8cxx_compiler_version10.yaml

@@ -7,13 +7,13 @@ channel_targets:
 cuda_compiler:
 - nvcc
 cuda_compiler_version:
-- '11.1'
+- '11.8'
 cxx_compiler:
 - gxx
 cxx_compiler_version:
 - '10'
 docker_image:
-- quay.io/condaforge/linux-anvil-cuda:11.1
+- quay.io/condaforge/linux-anvil-cuda:11.8
 libblas:
 - 3.9 *netlib
 liblapack:

.ci_support/linux_64_cuda_compiler_version11.0cxx_compiler_version9.yaml → .ci_support/linux_aarch64_cuda_compiler_version11.8cxx_compiler_version10.yaml

@@ -1,3 +1,7 @@
+BUILD:
+- aarch64-conda_cos7-linux-gnu
+cdt_arch:
+- aarch64
 cdt_name:
 - cos7
 channel_sources:
@@ -7,13 +11,13 @@ channel_targets:
 cuda_compiler:
 - nvcc
 cuda_compiler_version:
-- '11.0'
+- '11.8'
 cxx_compiler:
 - gxx
 cxx_compiler_version:
-- '9'
+- '10'
 docker_image:
-- quay.io/condaforge/linux-anvil-cuda:11.0
+- quay.io/condaforge/linux-anvil-aarch64-cuda:11.8
 libblas:
 - 3.9 *netlib
 liblapack:
@@ -33,7 +37,7 @@ python:
 - 3.8.* *_cpython
 - 3.9.* *_cpython
 target_platform:
-- linux-64
+- linux-aarch64
 zip_keys:
 - - cxx_compiler_version
   - cuda_compiler_version

.ci_support/linux_64_cuda_compiler_version11.2cxx_compiler_version10.yaml → .ci_support/linux_ppc64le_cuda_compiler_version11.8cxx_compiler_version10.yaml

@@ -7,13 +7,13 @@ channel_targets:
 cuda_compiler:
 - nvcc
 cuda_compiler_version:
-- '11.2'
+- '11.8'
 cxx_compiler:
 - gxx
 cxx_compiler_version:
 - '10'
 docker_image:
-- quay.io/condaforge/linux-anvil-cuda:11.2
+- quay.io/condaforge/linux-anvil-ppc64le-cuda:11.8
 libblas:
 - 3.9 *netlib
 liblapack:
@@ -33,7 +33,7 @@ python:
 - 3.8.* *_cpython
 - 3.9.* *_cpython
 target_platform:
-- linux-64
+- linux-ppc64le
 zip_keys:
 - - cxx_compiler_version
   - cuda_compiler_version

.ci_support/migrations/cuda_112_ppc64le_aarch64.yaml

@@ -0,0 +1,41 @@
+migrator_ts: 1645421738
+__migrator:
+  kind:
+    version
+  migration_number:
+    1
+  build_number:
+    1
+  paused: false
+  override_cbc_keys:
+    - cuda_compiler_stub
+  operation: key_add
+  check_solvable: false
+  primary_key: cuda_compiler_version
+  wait_for_migrators:
+    - aarch64 and ppc64le addition
+  exclude_pinned_pkgs: False
+  commit_message: "Rebuild for cuda for ppc64le and aarch64"
+
+arm_variant_type:              # [aarch64]
+  - sbsa                       # [aarch64]
+
+c_compiler_version:            # [ppc64le or aarch64]
+  - 10                         # [ppc64le or aarch64]
+cxx_compiler_version:          # [ppc64le or aarch64]
+  - 10                         # [ppc64le or aarch64]
+fortran_compiler_version:      # [ppc64le or aarch64]
+  - 10                         # [ppc64le or aarch64]
+
+cuda_compiler_version:         # [ppc64le or aarch64]
+  - 11.2                       # [ppc64le or aarch64]
+
+cudnn:                  # [ppc64le or aarch64]
+  - 8                   # [ppc64le or aarch64]
+
+cdt_name:  # [ppc64le or aarch64]
+  - cos7   # [ppc64le or aarch64]
+
+docker_image:                                           # [os.environ.get("BUILD_PLATFORM", "").startswith("linux") and (ppc64le or aarch64)]
+   - quay.io/condaforge/linux-anvil-ppc64le-cuda:11.2   # [ppc64le and os.environ.get("BUILD_PLATFORM") == "linux-ppc64le"]
+   - quay.io/condaforge/linux-anvil-aarch64-cuda:11.2   # [aarch64 and os.environ.get("BUILD_PLATFORM") == "linux-aarch64"]

recipe/build-lib.sh

@@ -11,9 +11,21 @@ if [ ${cuda_compiler_version} != "None" ]; then
     # docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/index.html#ptxas-options-gpu-name
     # docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/index.html#gpu-feature-list
 
-    # the following are all the x86-relevant gpu arches; for building aarch64-packages, add: 53, 62, 72
-    ARCHES=(52 60 61 70)
-    if [ $(version2int $cuda_compiler_version) -ge $(version2int "11.1") ]; then
+
+    if [ "$(uname -m)" = "aarch64" ]; then
+        # For building aarch64-packages, add arches 53, 62, 72
+        ARCHES=(52 53 60 61 62 70 72)
+    else
+        # Architectures for x86_64
+        ARCHES=(52 60 61 70)
+    fi
+
+    if [ $(version2int $cuda_compiler_version) -ge $(version2int "11.8") ]; then
+        # Hopper support for H100 (sm_90) needs cuda >= 11.8
+        LATEST_ARCH=90
+        # ARCHES does not contain LATEST_ARCH; see usage below
+        ARCHES=( "${ARCHES[@]}" 75 80 86)
+    elif [ $(version2int $cuda_compiler_version) -ge $(version2int "11.1") ]; then
         # Ampere support for GeForce 30 (sm_86) needs cuda >= 11.1
         LATEST_ARCH=86
         # ARCHES does not contain LATEST_ARCH; see usage below

recipe/conda_build_config.yaml

@@ -0,0 +1,16 @@
+c_compiler_version:
+  - 10
+cxx_compiler_version:
+  - 10
+cuda_compiler_version:
+  - 11.8
+fortran_compiler_version:
+  - 10
+cudnn:
+  - 8
+cdt_name:
+  - cos7
+docker_image:
+  - quay.io/condaforge/linux-anvil-cuda:11.8           # [linux64 and cuda_compiler_version != "None"]
+  - quay.io/condaforge/linux-anvil-ppc64le-cuda:11.8   # [linux and ppc64le and cuda_compiler_version != "None"]
+  - quay.io/condaforge/linux-anvil-aarch64-cuda:11.8   # [linux and aarch64 and cuda_compiler_version != "None"]

recipe/meta.yaml

@@ -92,7 +92,7 @@ source:
     - patches/0006-add-bigobj-to-swigfaiss-compile-options-on-windows.patch
 
 build:
-  number: 3
+  number: 4
   skip: true  # [linux and cuda_compiler_version == "10.2"]
   # always crashes
   skip: true  # [win and cuda_compiler_version in ("11.0", "11.1")]