use ctk_prev build and add arch=all to nvcc

Author: lijinf2

.github/workflows/build-wheel.yml

@@ -453,20 +453,17 @@ jobs:
           path: ${{ env.CUDA_CORE_CYTHON_TESTS_DIR }}/test_*${{ env.PY_EXT_SUFFIX }}
           if-no-files-found: error
 
-      - name: Set up mini CTK for test binaries
+      # Note: This overwrites CUDA_PATH etc
+      - name: Set up mini CTK
         uses: ./.github/actions/fetch_ctk
         continue-on-error: false
         with:
           host-platform: ${{ inputs.host-platform }}
           cuda-version: ${{ inputs.prev-cuda-version }}
-          cuda-path: "./cuda_toolkit_test_binaries"
+          cuda-path: "./cuda_toolkit_prev"
 
       - name: Build cuda.core test binaries
-        run: |
-          TEST_CTK="$(realpath ./cuda_toolkit_test_binaries)"
-          export PATH="${TEST_CTK}/bin:${PATH}"
-          export CUDA_HOME="${TEST_CTK}"
-          bash ${{ env.CUDA_CORE_TEST_BINARIES_DIR }}/build_test_binaries.sh
+        run: bash ${{ env.CUDA_CORE_TEST_BINARIES_DIR }}/build_test_binaries.sh
 
       - name: Upload cuda.core test binaries
         uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a  # v7.0.1
@@ -475,15 +472,6 @@ jobs:
           path: ${{ env.CUDA_CORE_TEST_BINARIES_DIR }}/*.o
           if-no-files-found: error
 
-      # Note: This overwrites CUDA_PATH etc
-      - name: Set up mini CTK
-        uses: ./.github/actions/fetch_ctk
-        continue-on-error: false
-        with:
-          host-platform: ${{ inputs.host-platform }}
-          cuda-version: ${{ inputs.prev-cuda-version }}
-          cuda-path: "./cuda_toolkit_prev"
-
       - name: Download cuda.bindings build artifacts from the prior branch
         env:
           GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}

cuda_core/tests/test_binaries/build_test_binaries.sh

@@ -5,9 +5,7 @@
 
 set -euo pipefail
 
-# Build .o test fixtures. Invoked at CI build stage with the oldest test-matrix
-# CTK (prev-cuda-version, currently 12.9.x) so nvJitLink on 12.9/13.0/13.3
-# test jobs can consume the embedded device code.
+# Build .o test fixtures. Invoked at CI build stage
 
 SCRIPTPATH=$(dirname "$(realpath "$0")")
 
@@ -16,13 +14,7 @@ if [[ "${OS:-}" == "Windows_NT" ]]; then
     NVCC_EXTRA_FLAGS+=(-Xcompiler /Zc:preprocessor)
 fi
 
-GENCODE=()
-for cc in 70 75 80 89 90 120; do
-    if nvcc --list-gpu-code | grep -qx "sm_${cc}"; then
-        GENCODE+=(-gencode "arch=compute_${cc},code=sm_${cc}")
-    fi
-done
-
-nvcc -dc "${NVCC_EXTRA_FLAGS[@]}" "${GENCODE[@]}" -o "${SCRIPTPATH}/saxpy.o" "${SCRIPTPATH}/saxpy.cu"
+nvcc -dc "${NVCC_EXTRA_FLAGS[@]}" -arch=all-major \
+    -o "${SCRIPTPATH}/saxpy.o" "${SCRIPTPATH}/saxpy.cu"
 
 ls -lah "${SCRIPTPATH}/saxpy.o"