Merge remote-tracking branch 'origin/master' into gold/2021

Author: oleksandr-pavlyk

.github/workflows/Windows-IntelLLVM.cmake renamed to .github/workflows/Windows-IntelLLVM_3.22.cmake

@@ -0,0 +1,66 @@
+# Distributed under the OSI-approved BSD 3-Clause License.  See accompanying
+# file Copyright.txt or https://cmake.org/licensing for details.
+
+
+# This module is shared by multiple languages; use include blocker.
+if(__WINDOWS_INTEL_LLVM)
+  return()
+endif()
+set(__WINDOWS_INTEL_LLVM 1)
+
+# Platform/Windows-MSVC adds some linking options icx/ifx do not understand,
+# but that need to be passed to the linker.  Wrap all the linking options from
+# Platform/Windows-MSVC so that the compiler will hand them off to the linker
+# without interpreting them.
+
+# Save original CMAKE_${t}_LINKER_FLAGS_INIT
+foreach(t EXE SHARED MODULE STATIC)
+  set(_saved_cmake_${t}_linker_flags_init ${CMAKE_${t}_LINKER_FLAGS_INIT})
+  set(CMAKE_${t}_LINKER_FLAGS_INIT "")
+endforeach()
+include(Platform/Windows-MSVC)
+# Wrap linker flags from Windows-MSVC
+set(_IntelLLVM_LINKER_WRAPPER_FLAG "/Qoption,link,")
+set(_IntelLLVM_LINKER_WRAPPER_FLAG_SEP ",")
+foreach(t EXE SHARED MODULE STATIC)
+  set(_wrapped_linker_flags "")
+  foreach(flag ${CMAKE_${t}_LINKER_FLAGS_INIT})
+    string(STRIP ${flag} flag)
+    list(APPEND _wrapped_linker_flags "${_IntelLLVM_LINKER_WRAPPER_FLAG}${flag}")
+  endforeach()
+  set(CMAKE_${t}_LINKER_FLAGS_INIT "")
+  list(APPEND CMAKE_${t}_LINKER_FLAGS_INIT
+    ${_saved_cmake_${t}_linker_flags_init} ${_wrapped_linker_flags})
+endforeach()
+
+macro(__windows_compiler_intel lang)
+  __windows_compiler_msvc(${lang})
+
+  set(CMAKE_${lang}_LINKER_WRAPPER_FLAG "${_IntelLLVM_LINKER_WRAPPER_FLAG}")
+  set(CMAKE_${lang}_LINKER_WRAPPER_FLAG_SEP "${_IntelLLVM_LINKER_WRAPPER_FLAG_SEP}")
+  set(CMAKE_${lang}_CREATE_WIN32_EXE "${CMAKE_${lang}_LINKER_WRAPPER_FLAG}/subsystem:windows")
+  set(CMAKE_${lang}_CREATE_CONSOLE_EXE "${CMAKE_${lang}_LINKER_WRAPPER_FLAG}/subsystem:console")
+  set(CMAKE_LINK_DEF_FILE_FLAG "${CMAKE_${lang}_LINKER_WRAPPER_FLAG}/DEF:")
+  set(CMAKE_LIBRARY_PATH_FLAG "${CMAKE_${lang}_LINKER_WRAPPER_FLAG}/LIBPATH:")
+
+  # Features for LINK_LIBRARY generator expression
+  if(MSVC_VERSION GREATER "1900")
+    ## WHOLE_ARCHIVE: Force loading all members of an archive
+    set(CMAKE_LINK_LIBRARY_USING_WHOLE_ARCHIVE "LINKER:/WHOLEARCHIVE:<LIBRARY>")
+    set(CMAKE_LINK_LIBRARY_USING_WHOLE_ARCHIVE_SUPPORTED TRUE)
+  endif()
+
+  set(CMAKE_${lang}_LINK_EXECUTABLE
+    "${_CMAKE_VS_LINK_EXE}<CMAKE_${lang}_COMPILER> ${CMAKE_CL_NOLOGO} <CMAKE_${lang}_LINK_FLAGS> <OBJECTS> ${CMAKE_START_TEMP_FILE} <LINK_FLAGS> <LINK_LIBRARIES> /link /out:<TARGET> /implib:<TARGET_IMPLIB> /pdb:<TARGET_PDB> /version:<TARGET_VERSION_MAJOR>.<TARGET_VERSION_MINOR>${_PLATFORM_LINK_FLAGS} ${CMAKE_END_TEMP_FILE}")
+  set(CMAKE_${lang}_CREATE_SHARED_LIBRARY
+    "${_CMAKE_VS_LINK_DLL}<CMAKE_${lang}_COMPILER> ${CMAKE_CL_NOLOGO} <CMAKE_${lang}_LINK_FLAGS> <OBJECTS> ${CMAKE_START_TEMP_FILE} -LD <LINK_FLAGS> <LINK_LIBRARIES> -link /out:<TARGET> /implib:<TARGET_IMPLIB> /pdb:<TARGET_PDB> /version:<TARGET_VERSION_MAJOR>.<TARGET_VERSION_MINOR>${_PLATFORM_LINK_FLAGS} ${CMAKE_END_TEMP_FILE}")
+  set(CMAKE_${lang}_CREATE_SHARED_MODULE ${CMAKE_${lang}_CREATE_SHARED_LIBRARY})
+  if (NOT "${lang}" STREQUAL "Fortran" OR CMAKE_${lang}_COMPILER_VERSION VERSION_GREATER_EQUAL 2022.1)
+    # The Fortran driver does not support -fuse-ld=llvm-lib before compiler version 2022.1
+    set(CMAKE_${lang}_CREATE_STATIC_LIBRARY
+      "<CMAKE_${lang}_COMPILER> ${CMAKE_CL_NOLOGO} <CMAKE_${lang}_LINK_FLAGS> <OBJECTS> ${CMAKE_START_TEMP_FILE} -fuse-ld=llvm-lib -o <TARGET> <LINK_FLAGS> <LINK_LIBRARIES> ${CMAKE_END_TEMP_FILE}")
+  endif()
+
+  set(CMAKE_DEPFILE_FLAGS_${lang} "-QMD -QMT <DEP_TARGET> -QMF <DEP_FILE>")
+  set(CMAKE_${lang}_DEPFILE_FORMAT gcc)
+endmacro()

.github/workflows/Windows-IntelLLVM_3.26.cmake

@@ -44,7 +44,7 @@ jobs:
         run: conda install conda-build
       - name: Build conda package
         run: |
-          CHANNELS="-c dppy/label/tools -c intel -c main --override-channels"
+          CHANNELS="-c intel -c conda-forge --override-channels"
           VERSIONS="--python ${{ matrix.python }}"
           TEST="--no-test"
           conda build \
@@ -72,7 +72,6 @@ jobs:
           fetch-depth: 0
       - uses: conda-incubator/setup-miniconda@v2
         with:
-          use-only-tar-bz2: true
           auto-activate-base: true
           conda-build-version: "*"
           activate-environment: true
@@ -90,7 +89,7 @@ jobs:
             ${{ runner.os }}-conda-${{ env.CACHE_NUMBER }}-python-${{ matrix.python }}-
             ${{ runner.os }}-conda-${{ env.CACHE_NUMBER }}-
       - name: Build conda package
-        run: conda build --no-test --python ${{ matrix.python }} -c intel -c main --override-channels conda-recipe
+        run: conda build --no-test --python ${{ matrix.python }} -c intel -c conda-forge --override-channels conda-recipe
       - name: Upload artifact
         uses: actions/upload-artifact@v3
         with:
@@ -108,7 +107,7 @@ jobs:
         runner: [ubuntu-20.04]
     continue-on-error: ${{ matrix.experimental }}
     env:
-      CHANNELS: -c intel -c main --override-channels
+      CHANNELS: -c intel -c conda-forge --override-channels
 
     steps:
       - name: Download artifact
@@ -192,7 +191,7 @@ jobs:
     continue-on-error: ${{ matrix.experimental }}
     env:
       workdir: '${{ github.workspace }}'
-      CHANNELS: -c intel -c main --override-channels
+      CHANNELS: -c intel -c conda-forge --override-channels
 
     steps:
       - name: Download artifact
@@ -201,7 +200,6 @@ jobs:
           name: ${{ env.PACKAGE_NAME }} ${{ runner.os }} Python ${{ matrix.python }}
       - uses: conda-incubator/setup-miniconda@v2
         with:
-          use-only-tar-bz2: true
           auto-update-conda: true
           conda-build-version: '*'
           miniconda-version: 'latest'
@@ -355,7 +353,7 @@ jobs:
         runner: [ubuntu-20.04]
     continue-on-error: ${{ matrix.experimental }}
     env:
-      CHANNELS: -c intel -c main --override-channels
+      CHANNELS: -c intel -c conda-forge --override-channels
 
     steps:
       - name: Install conda-build
@@ -408,12 +406,12 @@ jobs:
           conda install -n examples -y ninja $CHANNELS || exit 1
           conda install -n examples -y pybind11 cython scikit-build $CHANNELS || exit 1
           conda install -n examples -y mkl-dpcpp mkl-devel-dpcpp dpcpp_cpp_rt $CHANNELS || exit 1
-          conda create -y -n build_env -c intel gcc_linux-64 gxx_linux-64 dpcpp_linux-64
+          conda create -y -n build_env $CHANNELS gcc_linux-64 gxx_linux-64 dpcpp_linux-64 sysroot_linux-64">=2.28"
       - name: Install dpctl
         shell: bash -l {0}
         run: |
           source $CONDA/etc/profile.d/conda.sh
-          CHANNELS="-c $GITHUB_WORKSPACE/channel -c dppy/label/dev -c intel -c main --override-channels"
+          CHANNELS="-c $GITHUB_WORKSPACE/channel -c dppy/label/dev -c intel -c conda-forge --override-channels"
           export PACKAGE_VERSION=$(python -c "${VER_SCRIPT1} ${VER_SCRIPT2}")
           conda install -n examples -y ${CHANNELS} dpctl=${PACKAGE_VERSION} dpnp">=0.10.1" || exit 1
       - name: Build and run examples of pybind11 extensions
@@ -498,7 +496,7 @@ jobs:
         runner: [ubuntu-20.04]
     continue-on-error: ${{ matrix.experimental }}
     env:
-      CHANNELS: -c intel -c main --override-channels
+      CHANNELS: -c intel -c conda-forge --override-channels
     steps:
       - name: Cache array API tests
         id: cache-array-api-tests

.github/workflows/conda-package.yml

@@ -18,5 +18,6 @@ include dpctl/program/_program_api.h
 include dpctl/tensor/_usmarray.h
 include dpctl/tensor/_usmarray_api.h
 recursive-include dpctl/tensor/include *
+recursive-include dpctl/tensor/libtensor/include *
 include dpctl/tests/input_files/*
 include dpctl/tests/*.pyx

MANIFEST.in

@@ -40,7 +40,6 @@ if(NOT Dpctl_FOUND)
       OUTPUT_STRIP_TRAILING_WHITESPACE
       ERROR_QUIET
       )
-
   endif()
 endif()
 
@@ -49,6 +48,12 @@ find_path(Dpctl_INCLUDE_DIR
   PATHS "${_dpctl_include_dir}" "${PYTHON_INCLUDE_DIR}"
   PATH_SUFFIXES dpctl/include
   )
+get_filename_component(_dpctl_dir ${_dpctl_include_dir} DIRECTORY)
+
+find_path(Dpctl_TENSOR_INCLUDE_DIR
+  kernels utils
+  PATHS "${_dpctl_dir}/tensor/libtensor/include"
+  )
 
 set(Dpctl_INCLUDE_DIRS ${Dpctl_INCLUDE_DIR})
 
@@ -57,8 +62,9 @@ set(Dpctl_INCLUDE_DIRS ${Dpctl_INCLUDE_DIR})
 include(FindPackageHandleStandardArgs)
 find_package_handle_standard_args(Dpctl
                                   REQUIRED_VARS
-                                    Dpctl_INCLUDE_DIR
+                                    Dpctl_INCLUDE_DIR Dpctl_TENSOR_INCLUDE_DIR
                                   VERSION_VAR Dpctl_VERSION
                                   )
 
 mark_as_advanced(Dpctl_INCLUDE_DIR)
+mark_as_advanced(Dpctl_TENSOR_INCLUDE_DIR)

cmake/FindDpctl.cmake

@@ -6,14 +6,15 @@ set "INCLUDE=%BUILD_PREFIX%\include;%INCLUDE%"
 "%PYTHON%" setup.py clean --all
 set "SKBUILD_ARGS=-G Ninja -- -DCMAKE_C_COMPILER:PATH=icx -DCMAKE_CXX_COMPILER:PATH=icx -DCMAKE_VERBOSE_MAKEFILE:BOOL=ON"
 
-FOR %%V IN (14.0.0 14 15.0.0 15 16.0.0 16) DO @(
+FOR %%V IN (14.0.0 14 15.0.0 15 16.0.0 16 17.0.0 17) DO @(
   REM set DIR_HINT if directory exists
   IF EXIST "%BUILD_PREFIX%\Library\lib\clang\%%V\" (
      SET "SYCL_INCLUDE_DIR_HINT=%BUILD_PREFIX%\Library\lib\clang\%%V"
   )
 )
 
-set "PLATFORM_DIR=%PREFIX%\Library\share\cmake-3.22\Modules\Platform"
+set "PATCHED_CMAKE_VERSION=3.26"
+set "PLATFORM_DIR=%PREFIX%\Library\share\cmake-%PATCHED_CMAKE_VERSION%\Modules\Platform"
 set "FN=Windows-IntelLLVM.cmake"
 
 rem Save the original file, and copy patched file to
@@ -22,7 +23,7 @@ if EXIST "%PLATFORM_DIR%" (
   dir "%PLATFORM_DIR%\%FN%"
   copy /Y "%PLATFORM_DIR%\%FN%" .
   if errorlevel 1 exit 1
-  copy /Y .github\workflows\Windows-IntelLLVM.cmake "%PLATFORM_DIR%"
+  copy /Y ".github\workflows\Windows-IntelLLVM_%PATCHED_CMAKE_VERSION%.cmake" "%PLATFORM_DIR%\%FN%"
   if errorlevel 1 exit 1
 )
 
@@ -40,6 +41,6 @@ if NOT "%WHEELS_OUTPUT_FOLDER%"=="" (
 
 if EXIST "%PLATFORM_DIR%" (
   rem copy back
-  copy /Y "%FN%" "%PLATFORM_DIR%"
+  copy /Y "%FN%" "%PLATFORM_DIR%\%FN%"
   if errorlevel 1 exit 1
 )

conda-recipe/bld.bat

@@ -30,6 +30,7 @@ requirements:
         - {{ pin_compatible('numpy', min_pin='x.x', max_pin='x') }}
         - {{ pin_compatible('dpcpp-cpp-rt', min_pin='x.x', max_pin='x') }}  # [py<=39]
         - dpcpp-cpp-rt >=2023.1  # [py>39]
+        - level-zero  # [linux]
 
 test:
     requires:

conda-recipe/meta.yaml

@@ -35,6 +35,13 @@ def print_includes() -> None:
     print("-I " + dpctl.get_include())
 
 
+def print_tensor_includes() -> None:
+    "Prints include flags for dpctl and SyclInterface library"
+    dpctl_dir = _dpctl_dir()
+    libtensor_dir = os.path.join(dpctl_dir, "tensor", "libtensor", "include")
+    print("-I " + libtensor_dir)
+
+
 def print_cmake_dir() -> None:
     "Prints directory with FindDpctl.cmake"
     dpctl_dir = _dpctl_dir()
@@ -75,7 +82,12 @@ def main() -> None:
     parser.add_argument(
         "--includes",
         action="store_true",
-        help="Include flags dpctl headers.",
+        help="Include flags for dpctl headers.",
+    )
+    parser.add_argument(
+        "--tensor-includes",
+        action="store_true",
+        help="Include flags for dpctl libtensor headers.",
     )
     parser.add_argument(
         "--cmakedir",
@@ -128,6 +140,8 @@ def main() -> None:
         return
     if args.includes:
         print_includes()
+    if args.tensor_includes:
+        print_tensor_includes()
     if args.cmakedir:
         print_cmake_dir()
     if args.library:

dpctl/__main__.py

@@ -47,6 +47,7 @@ pybind11_add_module(${python_module_name} MODULE
     ${CMAKE_CURRENT_SOURCE_DIR}/libtensor/source/boolean_reductions.cpp
     ${CMAKE_CURRENT_SOURCE_DIR}/libtensor/source/device_support_queries.cpp
     ${CMAKE_CURRENT_SOURCE_DIR}/libtensor/source/elementwise_functions.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/libtensor/source/sum_reductions.cpp
 )
 set(_clang_prefix "")
 if (WIN32)

dpctl/tensor/CMakeLists.txt

@@ -104,6 +104,7 @@
     sqrt,
     subtract,
 )
+from ._reduction import sum
 
 __all__ = [
     "Device",
@@ -191,4 +192,5 @@
     "multiply",
     "subtract",
     "equal",
+    "sum",
 ]