[CI] Use prebuilt E2E binaries when running on Windows Gen12 (#17335)

Windows Gen12 job, previously took around 42min. With prebuilt binaries
building the e2e tests takes around 12min, and running them took around
6min.

Author: ayylol

.github/workflows/sycl-windows-precommit.yml

@@ -50,9 +50,25 @@ jobs:
     with:
       changes: ${{ needs.detect_changes.outputs.filters }}
 
-  e2e:
+  build-e2e:
     needs: build
     # Continue if build was successful.
+    if: |
+      always()
+      && !cancelled()
+      && needs.build.outputs.build_conclusion == 'success'
+    uses: ./.github/workflows/sycl-windows-run-tests.yml
+    with:
+      name: Build Windows E2E tests
+      runner: '["Windows", "build-e2e"]'
+      sycl_toolchain_archive: ${{ needs.build.outputs.artifact_archive_name }}
+      e2e_testing_mode: build-only
+      extra_lit_opts: --param sycl_build_targets="spir"
+      e2e_binaries_artifact: sycl_windows_e2ebin
+
+  run_prebuilt_e2e_tests:
+    needs: [build, build-e2e]
+    # Continue if build was successful.
     if: |
       always()
       && !cancelled()
@@ -63,10 +79,30 @@ jobs:
         include:
           - name: Intel GEN12 Graphics with Level Zero
             runner: '["Windows","gen12"]'
+    uses: ./.github/workflows/sycl-windows-run-tests.yml
+    with:
+      name: ${{ matrix.name }}
+      runner: ${{ matrix.runner }}
+      sycl_toolchain_archive: ${{ needs.build.outputs.artifact_archive_name }}
+      e2e_testing_mode: run-only
+      e2e_binaries_artifact: sycl_windows_e2ebin
+
+  run_full_e2e_tests:
+    needs: build
+    # Continue if build was successful.
+    if: |
+      always()
+      && !cancelled()
+      && needs.build.outputs.build_conclusion == 'success'
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
           - name: Intel Battlemage Graphics with Level Zero
             runner: '["Windows","bmg"]'
     uses: ./.github/workflows/sycl-windows-run-tests.yml
     with:
       name: ${{ matrix.name }}
       runner: ${{ matrix.runner }}
       sycl_toolchain_archive: ${{ needs.build.outputs.artifact_archive_name }}
+      e2e_testing_mode: full

.github/workflows/sycl-windows-run-tests.yml

@@ -63,6 +63,15 @@ on:
         required: false
         default: "cl"
 
+      e2e_testing_mode:
+        type: string
+        default: "full"
+
+      e2e_binaries_artifact:
+        type: string
+        default: ''
+        required: False
+
       cts_testing_mode:
         description: |
           Testing mode to run SYCL-CTS in, can be either `full`, `build-only`
@@ -143,17 +152,36 @@ jobs:
         sycl-ls
     - run: |
         sycl-ls --verbose
+
+    - name: Download E2E Binaries
+      if: ${{ inputs.tests_selector == 'e2e' && inputs.e2e_testing_mode == 'run-only' }}
+      uses: actions/download-artifact@v4
+      with:
+        name: ${{ inputs.e2e_binaries_artifact }}
+    - name: Extract E2E Binaries
+      if: ${{ inputs.tests_selector == 'e2e' && inputs.e2e_testing_mode == 'run-only' }}
+      shell: bash
+      run: |
+        mkdir build-e2e
+        tar -xf e2e_bin.tar.gz -C build-e2e
+
     - name: Configure E2E with Level Zero target
       if: inputs.tests_selector == 'e2e'
       shell: cmd
       run: |
         mkdir build-e2e
         cmake -GNinja -B build-e2e -S.\llvm\sycl\test-e2e -DSYCL_TEST_E2E_TARGETS="level_zero:gpu" -DCMAKE_CXX_COMPILER="clang++" -DLEVEL_ZERO_LIBS_DIR="D:\\github\\level-zero_win-sdk\\lib" -DLEVEL_ZERO_INCLUDE="D:\\github\\level-zero_win-sdk\\include" -DLLVM_LIT="..\\llvm\\llvm\\utils\\lit\\lit.py"
+
+    - name: Keep track of files after configuring E2E step
+      if: ${{ always() && inputs.tests_selector == 'e2e' && inputs.e2e_testing_mode == 'build-only' }}
+      shell: bash
+      run: ls build-e2e > e2econf_files.txt
+
     - name: Run End-to-End tests
       if: inputs.tests_selector == 'e2e'
       shell: bash {0}
       env:
-        LIT_OPTS: -v --no-progress-bar --show-unsupported --show-pass --show-xfail --max-time 3600 --time-tests --param print_features=True ${{ inputs.extra_lit_opts }}
+        LIT_OPTS: -v --no-progress-bar --show-unsupported --show-pass --show-xfail --max-time 3600 --time-tests --param print_features=True --param test-mode=${{ inputs.e2e_testing_mode }} ${{ inputs.extra_lit_opts }}
       run: |
         # Run E2E tests.
         if [[ ${{inputs.compiler}} == 'icx' ]]; then
@@ -169,6 +197,30 @@ jobs:
         fi
         exit $exit_code
 
+    # Github CI doesn't support containers on Windows, so we cannot guarantee
+    # that paths are the same between building and running systems. To avoid
+    # CMake issues related to absolute paths we reconfigure the build-e2e
+    # folder on the run system.
+    - name: Remove E2E configuration files
+      if: ${{ always() && inputs.tests_selector == 'e2e' && inputs.e2e_testing_mode == 'build-only' }}
+      shell: bash
+      run: |
+        for FILE in $(cat e2econf_files.txt); do rm -r build-e2e/$FILE; done
+        rm e2econf_files.txt
+
+    - name: Pack E2E test binaries
+      if: ${{ always() && inputs.tests_selector == 'e2e' && inputs.e2e_testing_mode == 'build-only' }}
+      shell: bash
+      run: |
+        tar -czf e2e_bin.tar.gz -C build-e2e .
+    - name: Upload E2E test binaries
+      if: ${{ always() && inputs.tests_selector == 'e2e' && inputs.e2e_testing_mode == 'build-only' }}
+      uses: actions/upload-artifact@v4
+      with:
+        name: ${{ inputs.e2e_binaries_artifact }}
+        path: e2e_bin.tar.gz
+        retention-days: 3
+
     - name: Run SYCL CTS Tests
       if: inputs.tests_selector == 'cts'
       uses: ./devops/actions/run-tests/windows/cts

sycl/test-e2e/AOT/cpu.cpp

@@ -6,7 +6,7 @@
 //
 //===---------------------------------------------------------------------===//
 
-// REQUIRES: opencl-aot, cpu
+// REQUIRES: opencl-aot, cpu, opencl-cpu-rt
 
 // CPU AOT targets host isa, so we compile on the run system instead.
 // RUN: %{run-aux} %clangxx -fsycl -fsycl-targets=spir64_x86_64 %S/Inputs/aot.cpp -o %t.out

sycl/test-e2e/AOT/multiple-devices.cpp

@@ -6,7 +6,7 @@
 //
 //===----------------------------------------------------------------------===//
 
-// REQUIRES: opencl-aot, ocloc, cpu, gpu, target-spir
+// REQUIRES: opencl-aot, ocloc, cpu, gpu, target-spir, opencl-cpu-rt
 
 // Produce a fat object for all targets (generic SPIR-V, CPU, GPU)
 // RUN: %clangxx -fsycl -fsycl-targets=spir64,spir64_x86_64,spir64_gen %S/Inputs/aot.cpp -c -o %t.o

sycl/test-e2e/AOT/reqd-sg-size.cpp

@@ -1,7 +1,7 @@
 // This test ensures that a program that has a kernel
 // using various required sub-group sizes can be compiled AOT.
 
-// REQUIRES: ocloc, opencl-aot, any-device-is-cpu
+// REQUIRES: ocloc, opencl-aot, any-device-is-cpu, opencl-cpu-rt
 // RUN: %clangxx -fsycl -fsycl-targets=intel_gpu_tgllp -o %t.tgllp.out %s
 // RUN: %clangxx -fsycl -fsycl-targets=spir64_x86_64 -o %t.x86.out %s
 

sycl/test-e2e/Adapters/cuda_queue_priority.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: gpu, cuda, cuda_dev_kit
+// REQUIRES: target-nvidia, cuda_dev_kit
 // RUN: %{build} %cuda_options -o %t.out
 // RUN: %{run} %t.out
 //

sycl/test-e2e/Adapters/sycl-targets-order.cpp

@@ -5,7 +5,7 @@
 // RUN: %{run-unfiltered-devices} env ONEAPI_DEVICE_SELECTOR="opencl:*" %t-nvptx64-spir64.out
 // RUN: %{run-unfiltered-devices} env ONEAPI_DEVICE_SELECTOR="cuda:*"   %t-nvptx64-spir64.out
 
-// REQUIRES: opencl, target-spir, any-triple-is-nvidia
+// REQUIRES: opencl, target-spir, any-target-is-nvidia
 
 //==------- sycl-targets-order.cpp - SYCL -fsycl-targets order test --------==//
 //

sycl/test-e2e/Basic/interop/make_kernel_subdevice_l0.cpp

@@ -1,5 +1,5 @@
-// REQUIRES: level_zero, opencl, level_zero_dev_kit
-// RUN: %{build} %level_zero_options -lOpenCL -o %t.ze.out
+// REQUIRES: level_zero, opencl, level_zero_dev_kit, opencl_icd
+// RUN: %{build} %level_zero_options %opencl_lib -o %t.ze.out
 // RUN: %{run-unfiltered-devices} env ONEAPI_DEVICE_SELECTOR="level_zero:*" %t.ze.out
 
 #include <cstdlib>

sycl/test-e2e/DeviceImageBackendContent/CUDA_interop_test.cpp

@@ -1,6 +1,6 @@
 // RUN: %{build} %cuda_options -o %t.out
 // RUN: %{run} %t.out
-// REQUIRES: cuda, cuda_dev_kit
+// REQUIRES: target-nvidia, cuda_dev_kit
 
 #include <cuda.h>
 #include <sycl/backend.hpp>

sycl/test-e2e/E2EExpr.py

@@ -14,6 +14,7 @@ class E2EExpr(BooleanExpression):
         "any-target-is-nvidia",
         "any-target-is-amd",
         "any-target-is-native_cpu",
+        "opencl-cpu-rt",
         "spirv-backend",
         "linux",
         "system-linux",

sycl/test-e2e/EnqueueNativeCommand/custom-command-cuda.cpp

@@ -1,6 +1,6 @@
 // RUN: %{build} -Wno-error=deprecated-declarations -o %t.out %cuda_options
 // RUN: %{run} %t.out
-// REQUIRES: cuda, cuda_dev_kit
+// REQUIRES: target-nvidia, cuda_dev_kit
 
 #include <cuda.h>
 

sycl/test-e2e/EnqueueNativeCommand/custom-command-multiple-dev-cuda.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: cuda, cuda_dev_kit
+// REQUIRES: target-nvidia, cuda_dev_kit
 // UNSUPPORTED: cuda
 // UNSUPPORTED-TRACKER: https://github.com/intel/llvm/issues/16951
 // RUN: %{build} -o %t.out %cuda_options

sycl/test-e2e/Graph/NativeCommand/cuda_explicit_usm.cpp

@@ -2,7 +2,7 @@
 // RUN: %{run} %t.out
 // RUN: %if preview-breaking-changes-supported %{ %{build} -fpreview-breaking-changes -o %t2.out %cuda_options %}
 // RUN: %if preview-breaking-changes-supported %{ %{run} %t2.out %}
-// REQUIRES: cuda, cuda_dev_kit
+// REQUIRES: target-nvidia, cuda_dev_kit
 
 #include <cuda.h>
 #include <sycl/backend.hpp>

sycl/test-e2e/Graph/NativeCommand/cuda_multiple_native_commands.cpp

@@ -2,7 +2,7 @@
 // RUN: %{run} %t.out
 // RUN: %if preview-breaking-changes-supported %{ %{build} -fpreview-breaking-changes -o %t2.out %cuda_options %}
 // RUN: %if preview-breaking-changes-supported %{ %{run} %t2.out %}
-// REQUIRES: cuda, cuda_dev_kit
+// REQUIRES: target-nvidia, cuda_dev_kit
 // REQUIRES: aspect-usm_shared_allocations
 
 // Test that when a host-task splits a graph into multiple backend UR

sycl/test-e2e/Graph/NativeCommand/cuda_record_buffer.cpp

@@ -2,7 +2,7 @@
 // RUN: %{run} %t.out
 // RUN: %if preview-breaking-changes-supported %{ %{build} -fpreview-breaking-changes -o %t2.out %cuda_options %}
 // RUN: %if preview-breaking-changes-supported %{ %{run} %t2.out %}
-// REQUIRES: cuda, cuda_dev_kit
+// REQUIRES: target-nvidia, cuda_dev_kit
 
 #include <cuda.h>
 #include <sycl/backend.hpp>

sycl/test-e2e/Graph/NativeCommand/cuda_record_usm.cpp

@@ -2,7 +2,7 @@
 // RUN: %{run} %t.out
 // RUN: %if preview-breaking-changes-supported %{ %{build} -fpreview-breaking-changes -o %t2.out %cuda_options %}
 // RUN: %if preview-breaking-changes-supported %{ %{run} %t2.out %}
-// REQUIRES: cuda, cuda_dev_kit
+// REQUIRES: target-nvidia, cuda_dev_kit
 
 #include <cuda.h>
 #include <sycl/backend.hpp>

sycl/test-e2e/Graph/NativeCommand/opencl_buffer.cpp

@@ -2,7 +2,7 @@
 // RUN: %{run} %t.out
 // RUN: %if preview-breaking-changes-supported %{ %{build} -fpreview-breaking-changes -o %t2.out %threads_lib %opencl_lib %}
 // RUN: %if preview-breaking-changes-supported %{ %{run} %t2.out %}
-// REQUIRES: opencl
+// REQUIRES: opencl, opencl_icd
 
 #include <sycl/backend.hpp>
 #include <sycl/detail/cl.h>

sycl/test-e2e/HostInteropTask/interop-task-cuda-buffer-migrate.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: cuda, cuda_dev_kit
+// REQUIRES: target-nvidia, cuda_dev_kit
 //
 // RUN: %{build} -o %t.out %cuda_options
 // RUN: %{run} %t.out

sycl/test-e2e/HostInteropTask/interop-task-cuda.cpp

@@ -1,6 +1,6 @@
 // RUN: %{build} -o %t.out %cuda_options
 // RUN: %{run} %t.out
-// REQUIRES: cuda, cuda_dev_kit
+// REQUIRES: target-nvidia, cuda_dev_kit
 
 #include <iostream>
 #include <sycl/backend.hpp>

sycl/test-e2e/NewOffloadDriver/cpu.cpp

@@ -6,7 +6,7 @@
 //
 //===---------------------------------------------------------------------===//
 
-// REQUIRES: opencl-aot, cpu
+// REQUIRES: opencl-aot, cpu, opencl-cpu-rt
 
 // CPU AOT targets host isa, so we compile on the run system instead.
 // Test with  `--offload-new-driver`

sycl/test-e2e/OptionalKernelFeatures/is_compatible/is_compatible_several_targets.cpp

@@ -1,8 +1,4 @@
-// REQUIRES: opencl-aot, ocloc
-
-// UNSUPPORTED: windows
-// UNSUPPORTED-TRACKER: https://github.com/intel/llvm/issues/17515
-// There is no CPU device on win yet, so opencl-aot fails to compile the kernel.
+// REQUIRES: opencl-aot, ocloc, opencl-cpu-rt
 
 // RUN: %{run-aux} %clangxx -fsycl -fsycl-targets=spir64_x86_64,spir64_gen -Xsycl-target-backend=spir64_gen %gpu_aot_target_opts %S/Inputs/is_compatible_with_env.cpp -o %t.out
 

sycl/test-e2e/OptionalKernelFeatures/is_compatible/is_compatible_spir64_x86_64.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: opencl-aot, any-device-is-cpu
+// REQUIRES: opencl-aot, opencl-cpu-rt
 
 // RUN: %{run-aux} %clangxx -fsycl -fsycl-targets=spir64_x86_64 %S/Inputs/is_compatible_with_env.cpp -o %t.out
 

sycl/test-e2e/ProgramManager/uneven_kernel_split.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: any-device-is-cpu, gpu, opencl-aot, ocloc, target-spir
+// REQUIRES: any-device-is-cpu, gpu, opencl-aot, ocloc, target-spir, opencl-cpu-rt
 
 // UNSUPPORTED: linux
 // UNSUPPORTED-TRACKER: https://github.com/intel/llvm/issues/17305

sycl/test-e2e/Regression/reduction_resource_leak_dw.cpp

@@ -1,5 +1,5 @@
 // REQUIRES: level_zero, level_zero_dev_kit
-// XFAIL: windows
+// XFAIL: windows && run-mode
 // XFAIL-TRACKER: https://github.com/intel/llvm/issues/16418
 //
 // RUN: %{build} %level_zero_options -o %t.out

sycl/test-e2e/SpecConstants/2020/non_native/multiple-targets.cpp

@@ -1,4 +1,4 @@
-// REQUIRES: opencl-aot, cpu, target-spir
+// REQUIRES: opencl-aot, cpu, target-spir, opencl-cpu-rt
 
 // RUN: %clangxx -fsycl -fsycl-targets=spir64,spir64_x86_64 %S/Inputs/common.cpp -o %t_spv_cpu.out
 // RUN: %{run} %t_spv_cpu.out

sycl/test-e2e/lit.cfg.py

@@ -620,13 +620,17 @@ def open_check_file(file_name):
     if "amdgcn" in sp[1]:
         config.sycl_build_targets.add("target-amd")
 
+cmd = "{} {}".format(config.run_launcher, sycl_ls) if config.run_launcher else sycl_ls
+sycl_ls_output = subprocess.check_output(cmd, text=True, shell=True)
+
+# In contrast to `cpu` feature this is a compile-time feature, which is needed
+# to check if we can build cpu AOT tests.
+if "opencl:cpu" in sycl_ls_output:
+    config.available_features.add("opencl-cpu-rt")
+
 if len(config.sycl_devices) == 1 and config.sycl_devices[0] == "all":
     devices = set()
-    cmd = (
-        "{} {}".format(config.run_launcher, sycl_ls) if config.run_launcher else sycl_ls
-    )
-    sp = subprocess.check_output(cmd, text=True, shell=True)
-    for line in sp.splitlines():
+    for line in sycl_ls_output.splitlines():
         if not line.startswith("["):
             continue
         (backend, device) = line[1:].split("]")[0].split(":")

sycl/test-e2e/syclcompat/kernel/kernel_win.cpp

@@ -2,6 +2,9 @@
 
 // DEFINE: %{sharedflag} = %if cl_options %{/clang:-shared%} %else %{-shared%}
 
-// RUN: %clangxx %{sharedflag} -fsycl %{sycl_target_opts} %S\Inputs\kernel_module.cpp -o %t.dll
-// RUN: %clangxx -DTEST_SHARED_LIB='"%/t.dll"' -fsycl %{sycl_target_opts} %S\Inputs\kernel_function.cpp -o %t.out
+// This test is sensitive to the absolute path of the dll file produced, so we
+// run the test completely on the run system to avoid issues.
+
+// RUN: %{run-aux} %clangxx %{sharedflag} -fsycl %{sycl_target_opts} %S\Inputs\kernel_module.cpp -o %t.dll
+// RUN: %{run-aux} %clangxx -DTEST_SHARED_LIB='"%/t.dll"' -fsycl %{sycl_target_opts} %S\Inputs\kernel_function.cpp -o %t.out
 // RUN: %{run} %t.out