Merge branch 'main' into guangyey/shuffle_down

.github/ci_expected_accuracy/inductor_torchbench_training.csv

@@ -93,7 +93,7 @@ tacotron2,fail_to_run,fail_to_run,fail_to_run,fail_to_run,fail_to_run
 timm_efficientdet,model_fail_to_load,model_fail_to_load,model_fail_to_load,model_fail_to_load,model_fail_to_load
 timm_efficientnet,pass,pass,pass,pass,pass
 timm_nfnet,pass,pass,pass,pass,pass
-timm_regnet,pass,pass,pass,pass,pass
+timm_regnet,pass,fail_accuracy,pass,pass,pass
 timm_resnest,pass,pass,pass,pass,pass
 timm_vision_transformer,pass,pass,pass,pass,pass
 timm_vision_transformer_large,pass_due_to_skip,pass_due_to_skip,pass_due_to_skip,pass_due_to_skip,pass_due_to_skip

.github/workflows/nightly_ondemand.yml

@@ -2,8 +2,10 @@ name: Nightly-OnDemand Tests
 
 on:
   schedule:
-    # GMT+8 21:00 every day
-    - cron: '0 13 * * *'
+    # GMT+8 21:00 every workday
+    - cron: '0 13 * * 0-4'
+    # GMT+8 0:00 Saturday
+    - cron: '0 16 * * 5'
   workflow_dispatch:
     inputs:
       pytorch:
@@ -78,7 +80,7 @@ jobs:
     runs-on: pvc_e2e
     # Don't run on forked repos
     if: github.repository_owner == 'intel'
-    timeout-minutes: 900
+    timeout-minutes: 3600
     env:
       pytorch: ${{ github.event_name == 'schedule' && 'main' || inputs.pytorch }}
       keep_torch_xpu_ops: ${{ github.event_name == 'schedule' && 'false' || inputs.keep_torch_xpu_ops }}
@@ -174,8 +176,10 @@ jobs:
           echo "$GITHUB_ENV"
           rm -rf ../pytorch/inductor_log
           rm -rf /tmp/torchinductor_*
+
+      # Nihglty launch
       - name: Nightly Huggingface FP32/BF16/FP16 Inference & Training Accuracy Test
-        if: github.event_name == 'schedule'
+        if: github.event_name == 'schedule' && github.event.schedule == '0 13 * * 0-4'
         uses: ./.github/actions/inductor-xpu-e2e-test
         with:
           suite: huggingface
@@ -185,7 +189,7 @@ jobs:
           scenario: accuracy
           hf_token: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
       - name: Nightly Torchbench BF16 Training Accuracy Test
-        if: github.event_name == 'schedule'
+        if: github.event_name == 'schedule' && github.event.schedule == '0 13 * * 0-4'
         uses: ./.github/actions/inductor-xpu-e2e-test
         with:
           suite: torchbench
@@ -195,7 +199,7 @@ jobs:
           env_prepare: true
           hf_token: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
       - name: Nightly Timm_models FP16 Training Accuracy Test
-        if: github.event_name == 'schedule'
+        if: github.event_name == 'schedule' && github.event.schedule == '0 13 * * 0-4'
         uses: ./.github/actions/inductor-xpu-e2e-test
         with:
           suite: timm_models
@@ -204,6 +208,38 @@ jobs:
           scenario: accuracy
           env_prepare: true
           hf_token: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
+      # Weekly launch
+      - name: Weekly Huggingface Full Test
+        if: github.event_name == 'schedule' && github.event.schedule == '0 16 * * 5'
+        uses: ./.github/actions/inductor-xpu-e2e-test
+        with:
+          suite: huggingface
+          env_prepare: true
+          dt: float32,bfloat16,float16,amp_bf16,amp_fp16
+          mode: inference,training
+          scenario: accuracy,performance
+          hf_token: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
+      - name: Weekly Torchbench Full Test
+        if: github.event_name == 'schedule' && github.event.schedule == '0 16 * * 5'
+        uses: ./.github/actions/inductor-xpu-e2e-test
+        with:
+          suite: torchbench
+          env_prepare: true
+          dt: float32,bfloat16,float16,amp_bf16,amp_fp16
+          mode: inference,training
+          scenario: accuracy,performance
+          hf_token: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
+      - name: Weekly Timm_models Full Test
+        if: github.event_name == 'schedule' && github.event.schedule == '0 16 * * 5'
+        uses: ./.github/actions/inductor-xpu-e2e-test
+        with:
+          suite: timm_models
+          env_prepare: true
+          dt: float32,bfloat16,float16,amp_bf16,amp_fp16
+          mode: inference,training
+          scenario: accuracy,performance
+          hf_token: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
+      # On-demand launch
       - name: OnDemand Test (${{ inputs.suite }} ${{ inputs.dt }} ${{ inputs.mode }} ${{ inputs.scenario }})
         if: github.event_name != 'schedule'
         uses: ./.github/actions/inductor-xpu-e2e-test
@@ -216,7 +252,7 @@ jobs:
           hf_token: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
       - name: Summarize archieve files
         id: summary
-        if: always()
+        if: ${{ ! cancelled() }}
         run: |
           rm -rf ${{ github.workspace }}/upload_files
           cp -r ${{ github.workspace }}/../pytorch/inductor_log ${{ github.workspace }}/upload_files
@@ -237,14 +273,14 @@ jobs:
             exit 1
           fi
       - name: Upload Inductor XPU E2E Data
-        if: always()
+        if: ${{ ! cancelled() }}
         uses: actions/upload-artifact@v4
         with:
           name: Inductor-XPU-E2E-Data-${{ github.event.pull_request.number || github.sha }}
           path: ${{ github.workspace }}/upload_files
-      
+
   Tests-Failure-And-Report:
-    if: always()
+    if: ${{ ! cancelled() }}
     runs-on: pvc_e2e
     permissions:
       issues: write
@@ -288,6 +324,9 @@ jobs:
             test_type="On-demand"
             test_issue_id=426
             cc_comment="CC @${GITHUB_TRIGGERING_ACTOR}"
+          elif [ "${{ github.event.schedule }}" == "0 16 * * 5" ];then
+            test_type="Weekly"
+            test_issue_id=432
           else
             test_type="Nightly"
             test_issue_id=432

.github/workflows/pull.yml

@@ -127,7 +127,7 @@ jobs:
           env_prepare: true
           hf_token: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
       - name: Summarize archieve files
-        if: always()
+        if: ${{ ! cancelled() }}
         run: |
           rm -rf ${{ github.workspace }}/upload_files
           cp -r ${{ github.workspace }}/../pytorch/inductor_log ${{ github.workspace }}/upload_files
@@ -137,7 +137,7 @@ jobs:
             exit 1
           fi
       - name: Upload Inductor XPU E2E Data
-        if: always()
+        if: ${{ ! cancelled() }}
         uses: actions/upload-artifact@v4
         with:
           name: Inductor-XPU-E2E-Data-${{ github.event.pull_request.number || github.sha }}

cmake/BuildFlags.cmake

@@ -59,10 +59,10 @@ if(CMAKE_CXX_COMPILER_ID STREQUAL "GNU" OR CMAKE_CXX_COMPILER_ID STREQUAL "MSVC"
     set(SYCL_KERNEL_OPTIONS ${SYCL_KERNEL_OPTIONS} -fno-approx-func)
     set(SYCL_KERNEL_OPTIONS ${SYCL_KERNEL_OPTIONS} -Wno-absolute-value)
     set(SYCL_KERNEL_OPTIONS ${SYCL_KERNEL_OPTIONS} -no-ftz)
+    # Equivalent to build option -fpreview-breaking-changes for SYCL compiler.
+    set(SYCL_KERNEL_OPTIONS ${SYCL_KERNEL_OPTIONS} -D__INTEL_PREVIEW_BREAKING_CHANGES)
+    set(SYCL_KERNEL_OPTIONS ${SYCL_KERNEL_OPTIONS} -D_GLIBCXX_USE_CXX11_ABI=${GLIBCXX_USE_CXX11_ABI})
   endif()
-  # TODO: Align with PyTorch and switch to ABI=0 eventually, after
-  # resolving incompatible implementation in SYCL runtime.
-  set(SYCL_KERNEL_OPTIONS ${SYCL_KERNEL_OPTIONS} -D_GLIBCXX_USE_CXX11_ABI=1)
   set(SYCL_FLAGS ${SYCL_FLAGS} ${SYCL_KERNEL_OPTIONS})
 
   set(TORCH_XPU_OPS_FLAGS ${SYCL_HOST_FLAGS})

cmake/Modules/FindSYCL.cmake

@@ -407,7 +407,6 @@ macro(SYCL_LINK_DEVICE_OBJECTS output_file sycl_target)
       OUTPUT ${output_file}
       DEPENDS ${object_files}
       COMMAND ${SYCL_EXECUTABLE}
-      -fsycl
       ${SYCL_device_link_flags}
       -fsycl-link ${object_files}
       -Xs "\"${SYCL_OFFLINE_COMPILER_FLAGS}\""
@@ -471,7 +470,7 @@ macro(SYCL_ADD_LIBRARY sycl_target)
   target_link_libraries(
     ${sycl_target}
     ${SYCL_LINK_LIBRARIES_KEYWORD}
-    ${SYCL_LIBRARIES})
+    ${SYCL_LIBRARY})
 
   set_target_properties(${sycl_target}
     PROPERTIES
@@ -530,7 +529,7 @@ macro(SYCL_ADD_EXECUTABLE sycl_target)
   target_link_libraries(
     ${sycl_target}
     ${SYCL_LINK_LIBRARIES_KEYWORD}
-    ${SYCL_LIBRARIES})
+    ${SYCL_LIBRARY})
 
   set_target_properties(${sycl_target}
     PROPERTIES

cmake/Modules/FindSYCLToolkit.cmake

@@ -25,16 +25,20 @@ This will define the following variables:
 
 #]=======================================================================]
 
-set(SYCLTOOLKIT_FOUND False)
-include(${CMAKE_ROOT}/Modules/FindPackageHandleStandardArgs.cmake)
+include(${TORCH_ROOT}/cmake/Modules/FindSYCLToolkit.cmake)
 
-set(SYCL_ROOT "")
-if(DEFINED ENV{SYCL_ROOT})
-  set(SYCL_ROOT $ENV{SYCL_ROOT})
-elseif(DEFINED ENV{CMPLR_ROOT})
-  set(SYCL_ROOT $ENV{CMPLR_ROOT})
+if(NOT SYCL_FOUND)
+  set(SYCLTOOLKIT_FOUND FALSE)
+  return()
 endif()
 
+if(SYCLTOOLKIT_FOUND)
+  return()
+endif()
+set(SYCLTOOLKIT_FOUND TRUE)
+
+include(${CMAKE_ROOT}/Modules/FindPackageHandleStandardArgs.cmake)
+
 if(WIN32)
   set(SYCL_EXECUTABLE_NAME icx)
 else()
@@ -71,43 +75,6 @@ if(nocmplr)
   set(SYCL_NOT_FOUND_MESSAGE "${SYCL_REASON_FAILURE}")
 endif()
 
-find_file(
-  SYCL_INCLUDE_DIR
-  NAMES include
-  HINTS ${SYCL_ROOT}
-  NO_DEFAULT_PATH
-  )
-
-find_file(
-  SYCL_INCLUDE_SYCL_DIR
-  NAMES sycl
-  HINTS ${SYCL_ROOT}/include
-  NO_DEFAULT_PATH
-  )
-
-list(APPEND SYCL_INCLUDE_DIR ${SYCL_INCLUDE_SYCL_DIR})
-
-find_file(
-  SYCL_LIBRARY_DIR
-  NAMES lib lib64
-  HINTS ${SYCL_ROOT}
-  NO_DEFAULT_PATH
-  )
-
-find_library(
-  SYCL_LIBRARY
-  NAMES sycl
-  HINTS ${SYCL_LIBRARY_DIR}
-  NO_DEFAULT_PATH
-)
-
-if((NOT SYCL_INCLUDE_DIR) OR (NOT SYCL_LIBRARY_DIR) OR (NOT SYCL_LIBRARY))
-  set(SYCLTOOLKIT_FOUND False)
-  set(SYCL_REASON_FAILURE "SYCL sdk is incomplete!!")
-  set(SYCL_NOT_FOUND_MESSAGE "${SYCL_REASON_FAILURE}")
-  return()
-endif()
-
 # Function to write a test case to verify SYCL features.
 
 function(SYCL_CMPLR_TEST_WRITE src)
@@ -202,6 +169,13 @@ set(SYCL_FLAGS "")
 set(SYCL_LINK_FLAGS "")
 list(APPEND SYCL_FLAGS "-fsycl")
 list(APPEND SYCL_LINK_FLAGS "-fsycl")
+if(LINUX)
+  string(REGEX MATCH "libsycl-preview.so" is_abi_neutral ${SYCL_LIBRARY})
+  if(is_abi_neutral)
+    list(APPEND SYCL_FLAGS "-fpreview-breaking-changes")
+    list(APPEND SYCL_LINK_FLAGS "-fpreview-breaking-changes")
+  endif()
+endif()
 
 set(SYCL_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${SYCL_FLAGS}")
 
@@ -249,14 +223,3 @@ endif()
 
 message(DEBUG "The SYCL compiler is ${SYCL_COMPILER}")
 message(DEBUG "The SYCL Flags are ${SYCL_FLAGS}")
-
-# Avoid module variables conflict due to calling find_package recursively
-# e.g. find_package -> add_subdirectory(entering in a sub-project) -> find_package
-# find_package_handle_standard_args(
-#   SYCLToolkit
-#   FOUND_VAR SYCLTOOLKIT_FOUND
-#   REQUIRED_VARS SYCL_INCLUDE_DIR SYCL_LIBRARY_DIR SYCL_LIBRARY SYCL_FLAGS
-#   VERSION_VAR SYCL_LANGUAGE_VERSION
-#   REASON_FAILURE_MESSAGE "${SYCL_REASON_FAILURE}")
-set(SYCLTOOLKIT_FOUND True)
-

cmake/SYCL.cmake

@@ -25,24 +25,6 @@ if(NOT SYCL_VERSION)
   return()
 endif()
 
-find_library(SYCL_LIBRARIES sycl HINTS ${SYCL_LIBRARY_DIR})
-# On Windows, currently there's no sycl.lib. Only sycl7.lib with version suffix,
-# where the current version of the SYCL runtime is 7.
-# Until oneAPI adds support to sycl.lib without the version suffix,
-# sycl_runtime_version needs to be hardcoded and uplifted when SYCL runtime version uplifts.
-# TODO: remove this when sycl.lib is supported on Windows
-if(WIN32)
-  set(sycl_runtime_version 7)
-  find_library(
-    SYCL_LIBRARIES
-    NAMES "sycl${sycl_runtime_version}"
-    HINTS ${SYCL_LIBRARY_DIR}
-  )
-  if(SYCL_LIBRARIES STREQUAL "SYCL_LIBRARIES-NOTFOUND")
-    message(FATAL_ERROR "Cannot find a SYCL library on Windows")
-  endif()
-endif()
-
 set(SYCL_COMPILER_VERSION)
 file(READ ${SYCL_VERSION} version_contents)
 string(REGEX MATCHALL "__SYCL_COMPILER_VERSION +[0-9]+" VERSION_LINE "${version_contents}")

src/ATen/CMakeLists.txt

@@ -1,6 +1,6 @@
 # ATen XPU sources
 
-file(GLOB xpu_cpp "xpu/*.cpp", "native/xpu/*.cpp", "native/sparse/*.cpp")
+file(GLOB xpu_cpp "xpu/*.cpp" "native/xpu/*.cpp" "native/sparse/*.cpp")
 file(GLOB xpu_sycl "native/xpu/sycl/*.cpp")
 
 list(APPEND ATen_XPU_CPP_SRCS ${xpu_cpp})