Merge pull request #1907 from IntelPython/backport-to-maintenance-0.1…

…8.x/gh-1893

CMakeLists.txt

@@ -29,6 +29,7 @@ option(DPCTL_TARGET_CUDA
     "Build DPCTL to target CUDA devices"
     OFF
 )
+option(DPCTL_WITH_REDIST "Build DPCTL assuming DPC++ redistributable is installed into Python prefix" OFF)
 
 find_package(IntelSYCL REQUIRED PATHS ${CMAKE_SOURCE_DIR}/cmake NO_DEFAULT_PATH)
 

conda-recipe/bld.bat

@@ -10,7 +10,7 @@ if DEFINED OVERRIDE_INTEL_IPO (
   set "CMAKE_ARGS=%CMAKE_ARGS% -DCMAKE_INTERPROCEDURAL_OPTIMIZATION:BOOL=FALSE"
 )
 
-FOR %%V IN (14.0.0 14 15.0.0 15 16.0.0 16 17.0.0 17) DO @(
+FOR %%V IN (17.0.0 17 18.0.0 18 19.0.0 19) 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"

conda-recipe/build.sh

@@ -1,7 +1,7 @@
 #!/bin/bash
 
 # This is necessary to help DPC++ find Intel libraries such as SVML, IRNG, etc in build prefix
-export LD_LIBRARY_PATH="$LD_LIBRARY_PATH:${BUILD_PREFIX}/lib"
+export LIBRARY_PATH="$LIBRARY_PATH:${BUILD_PREFIX}/lib"
 
 # Intel LLVM must cooperate with compiler and sysroot from conda
 echo "--gcc-toolchain=${BUILD_PREFIX} --sysroot=${BUILD_PREFIX}/${HOST}/sysroot -target ${HOST}" > icpx_for_conda.cfg
@@ -22,7 +22,7 @@ export CMAKE_GENERATOR=Ninja
 # Make CMake verbose
 export VERBOSE=1
 
-CMAKE_ARGS="${CMAKE_ARGS} -DDPCTL_LEVEL_ZERO_INCLUDE_DIR=${PREFIX}/include/level_zero"
+CMAKE_ARGS="${CMAKE_ARGS} -DDPCTL_LEVEL_ZERO_INCLUDE_DIR=${PREFIX}/include/level_zero -DDPCTL_WITH_REDIST=ON"
 
 # -wnx flags mean: --wheel --no-isolation --skip-dependency-check
 ${PYTHON} -m build -w -n -x
@@ -43,7 +43,3 @@ ${PYTHON} -m pip install dist/dpctl*.whl \
 if [[ -d "${WHEELS_OUTPUT_FOLDER}" ]]; then
     cp dist/dpctl*.whl "${WHEELS_OUTPUT_FOLDER[@]}"
 fi
-
-# need to create this folder so ensure that .dpctl-post-link.sh can work correctly
-mkdir -p $PREFIX/etc/OpenCL/vendors
-echo "dpctl creates symbolic link to system installed /etc/OpenCL/vendors/intel.icd as a work-around." > $PREFIX/etc/OpenCL/vendors/.dpctl_readme

conda-recipe/meta.yaml

@@ -29,7 +29,8 @@ requirements:
         - pip >=24.0
         - level-zero-devel >=1.16
         - pybind11 >=2.12
-        - {{ pin_compatible('dpcpp-cpp-rt', min_pin='x.x', max_pin='x') }}
+        - {{ pin_compatible('intel-sycl-rt', min_pin='x.x', max_pin='x') }}
+        - {{ pin_compatible('intel-cmplr-lib-rt', min_pin='x.x', max_pin='x') }}
         # Ensure we are using latest version of setuptools, since we don't need
         # editable environments for release.
         - setuptools >=69
@@ -51,7 +52,7 @@ requirements:
         - tomli # [py<311]
     run:
         - python
-        - {{ pin_compatible('dpcpp-cpp-rt', min_pin='x.x', max_pin='x') }}
+        - {{ pin_compatible('intel-sycl-rt', min_pin='x.x', max_pin='x') }}
         - {{ pin_compatible('intel-cmplr-lib-rt', min_pin='x.x', max_pin='x') }}
         - numpy
 

dpctl/CMakeLists.txt

@@ -105,7 +105,7 @@ set(CMAKE_INSTALL_RPATH "$ORIGIN")
 
 function(build_dpctl_ext _trgt _src _dest)
     set(options SYCL)
-    cmake_parse_arguments(BUILD_DPCTL_EXT "${options}" "" "" ${ARGN})
+    cmake_parse_arguments(BUILD_DPCTL_EXT "${options}" "RELATIVE_PATH" "" ${ARGN})
     add_cython_target(${_trgt} ${_src} CXX OUTPUT_VAR _generated_src)
     set(_cythonize_trgt "${_trgt}_cythonize_pyx")
     Python_add_library(${_trgt} MODULE WITH_SOABI ${_generated_src})
@@ -146,6 +146,14 @@ function(build_dpctl_ext _trgt _src _dest)
     get_filename_component(_generated_src_dir_dir ${_generated_src_dir} DIRECTORY)
     # TODO: do not set directory if we did not generate header
     target_include_directories(${_trgt} INTERFACE ${_generated_src_dir_dir})
+    set(_rpath_value "$ORIGIN")
+    if (BUILD_DPCTL_EXT_RELATIVE_PATH)
+        set(_rpath_value "${_rpath_value}/${BUILD_DPCTL_EXT_RELATIVE_PATH}")
+    endif()
+    if (DPCTL_WITH_REDIST)
+        set(_rpath_value "${_rpath_value}:${_rpath_value}/../../..")
+    endif()
+    set_target_properties(${_trgt} PROPERTIES INSTALL_RPATH ${_rpath_value})
 
     install(TARGETS ${_trgt}
         LIBRARY DESTINATION ${_dest})

dpctl/memory/CMakeLists.txt

@@ -1,7 +1,7 @@
 
 set(_cy_file ${CMAKE_CURRENT_SOURCE_DIR}/_memory.pyx)
 get_filename_component(_trgt ${_cy_file} NAME_WLE)
-build_dpctl_ext(${_trgt} ${_cy_file} "dpctl/memory" SYCL)
+build_dpctl_ext(${_trgt} ${_cy_file} "dpctl/memory" SYCL RELATIVE_PATH "..")
 # _memory include _opaque_smart_ptr.hpp
 target_include_directories(${_trgt} PRIVATE ${CMAKE_CURRENT_SOURCE_DIR})
 target_link_libraries(DpctlCAPI INTERFACE ${_trgt}_headers)

dpctl/program/CMakeLists.txt

@@ -2,6 +2,6 @@
 file(GLOB _cython_sources *.pyx)
 foreach(_cy_file ${_cython_sources})
     get_filename_component(_trgt ${_cy_file} NAME_WLE)
-    build_dpctl_ext(${_trgt} ${_cy_file} "dpctl/program")
+    build_dpctl_ext(${_trgt} ${_cy_file} "dpctl/program" RELATIVE_PATH "..")
     target_link_libraries(DpctlCAPI INTERFACE ${_trgt}_headers)
 endforeach()

dpctl/tensor/CMakeLists.txt

@@ -1,7 +1,7 @@
 file(GLOB _cython_sources *.pyx)
 foreach(_cy_file ${_cython_sources})
     get_filename_component(_trgt ${_cy_file} NAME_WLE)
-    build_dpctl_ext(${_trgt} ${_cy_file} "dpctl/tensor")
+    build_dpctl_ext(${_trgt} ${_cy_file} "dpctl/tensor" RELATIVE_PATH "..")
     target_include_directories(${_trgt} PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/include)
     target_link_libraries(DpctlCAPI INTERFACE ${_trgt}_headers)
 endforeach()
@@ -284,5 +284,12 @@ foreach(python_module_name ${_py_trgts})
     # TODO: update source so they refernece individual libraries instead of
     #   dpctl4pybind11.hpp. It will allow to simplify dependency tree
     target_link_libraries(${python_module_name} PRIVATE DpctlCAPI)
+    if (DPCTL_WITH_REDIST)
+        set_target_properties(
+            ${python_module_name}
+            PROPERTIES
+                INSTALL_RPATH "$ORIGIN/../../../.."
+        )
+    endif()
     install(TARGETS ${python_module_name} DESTINATION "dpctl/tensor")
 endforeach()

dpctl/utils/CMakeLists.txt

@@ -2,71 +2,71 @@
 file(GLOB _cython_sources *.pyx)
 foreach(_cy_file ${_cython_sources})
     get_filename_component(_trgt ${_cy_file} NAME_WLE)
-    build_dpctl_ext(${_trgt} ${_cy_file} "dpctl/utils")
+    build_dpctl_ext(${_trgt} ${_cy_file} "dpctl/utils" RELATIVE_PATH "..")
 endforeach()
 
+set(_pybind11_targets)
+
 set(python_module_name _device_queries)
 set(_module_src ${CMAKE_CURRENT_SOURCE_DIR}/src/device_queries.cpp)
 pybind11_add_module(${python_module_name} MODULE
   ${_module_src}
 )
 add_sycl_to_target(TARGET ${python_module_name} SOURCES ${_module_src})
-if(DPCTL_GENERATE_COVERAGE)
-     if(DPCTL_GENERATE_COVERAGE_FOR_PYBIND11_EXTENSIONS)
-         target_compile_options(${python_module_name}
-             PRIVATE -fprofile-instr-generate -fcoverage-mapping
-         )
-     endif()
-     target_link_options(${python_module_name}
-         PRIVATE -fprofile-instr-generate -fcoverage-mapping
-     )
-endif()
-if(_dpctl_sycl_targets)
-    # make fat binary
-    target_compile_options(
-        ${python_module_name}
-        PRIVATE
-        -fsycl-targets=${_dpctl_sycl_targets}
-    )
-    target_link_options(
-        ${python_module_name}
-        PRIVATE
-        -fsycl-targets=${_dpctl_sycl_targets}
-    )
-endif()
-target_link_libraries(${python_module_name} PRIVATE DpctlCAPI)
-install(TARGETS ${python_module_name} DESTINATION "dpctl/utils")
+list(APPEND _pybind11_targets ${python_module_name})
 
 
 set(python_module_name _seq_order_keeper)
 set(_module_src ${CMAKE_CURRENT_SOURCE_DIR}/src/order_keeper.cpp)
 pybind11_add_module(${python_module_name} MODULE
   ${_module_src}
 )
-target_include_directories(${python_module_name} PRIVATE ${CUMAKE_CURRENT_SOURCE_DIR}/src)
+target_include_directories(${python_module_name} PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/src)
 add_sycl_to_target(TARGET ${python_module_name} SOURCES ${_module_src})
-if(DPCTL_GENERATE_COVERAGE)
-     if(DPCTL_GENERATE_COVERAGE_FOR_PYBIND11_EXTENSIONS)
-         target_compile_options(${python_module_name}
-             PRIVATE -fprofile-instr-generate -fcoverage-mapping
-         )
-     endif()
-     target_link_options(${python_module_name}
-         PRIVATE -fprofile-instr-generate -fcoverage-mapping
-     )
-endif()
-if(_dpctl_sycl_targets)
-    # make fat binary
-    target_compile_options(
-        ${python_module_name}
-        PRIVATE
-        -fsycl-targets=${_dpctl_sycl_targets}
-    )
-    target_link_options(
-        ${python_module_name}
+list(APPEND _pybind11_targets ${python_module_name})
+
+set(_linker_options "LINKER:${DPCTL_LDFLAGS}")
+foreach(python_module_name ${_pybind11_targets})
+    target_compile_options(${python_module_name} PRIVATE -fno-sycl-id-queries-fit-in-int)
+    target_link_options(${python_module_name} PRIVATE -fsycl-device-code-split=per_kernel)
+    target_include_directories(${python_module_name}
         PRIVATE
-        -fsycl-targets=${_dpctl_sycl_targets}
+        ${CMAKE_CURRENT_SOURCE_DIR}/libtensor/include
+        ${CMAKE_CURRENT_SOURCE_DIR}/libtensor/source/
     )
-endif()
-target_link_libraries(${python_module_name} PRIVATE DpctlCAPI)
-install(TARGETS ${python_module_name} DESTINATION "dpctl/utils")
+    target_link_options(${python_module_name} PRIVATE ${_linker_options})
+    if(DPCTL_GENERATE_COVERAGE)
+        if(DPCTL_GENERATE_COVERAGE_FOR_PYBIND11_EXTENSIONS)
+            target_compile_options(${python_module_name}
+                PRIVATE -fprofile-instr-generate -fcoverage-mapping
+            )
+	endif()
+        target_link_options(${python_module_name}
+            PRIVATE -fprofile-instr-generate -fcoverage-mapping
+        )
+    endif()
+    if(_dpctl_sycl_targets)
+        # make fat binary
+        target_compile_options(
+            ${python_module_name}
+            PRIVATE
+            -fsycl-targets=${_dpctl_sycl_targets}
+        )
+        target_link_options(
+            ${python_module_name}
+            PRIVATE
+            -fsycl-targets=${_dpctl_sycl_targets}
+        )
+    endif()
+    # TODO: update source so they refernece individual libraries instead of
+    #   dpctl4pybind11.hpp. It will allow to simplify dependency tree
+    target_link_libraries(${python_module_name} PRIVATE DpctlCAPI)
+    if (DPCTL_WITH_REDIST)
+        set_target_properties(
+            ${python_module_name}
+            PROPERTIES
+                INSTALL_RPATH "$ORIGIN/../../../.."
+        )
+    endif()
+    install(TARGETS ${python_module_name} DESTINATION "dpctl/utils")
+endforeach()

libsyclinterface/CMakeLists.txt

@@ -314,6 +314,11 @@ target_include_directories(DPCTLSyclInterfaceHeaders INTERFACE
     ${CMAKE_CURRENT_SOURCE_DIR}/include
 )
 
+if (DPCTL_WITH_REDIST)
+   cmake_path(RELATIVE_PATH CMAKE_SOURCE_DIR BASE_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR} OUTPUT_VARIABLE _relative_path)
+   set_target_properties(DPCTLSyclInterface PROPERTIES INSTALL_RPATH "$ORIGIN/${_relative_path}/../../")
+endif()
+
 install(TARGETS
     DPCTLSyclInterface
     LIBRARY

libsyclinterface/cmake/modules/FindIntelSyclCompiler.cmake

@@ -119,7 +119,7 @@ if(${clangxx_result} MATCHES "0")
     if("x${CMAKE_SYSTEM_NAME}" STREQUAL "xWindows")
         find_file(
             IntelSyclCompiler_SYCL_LIBRARY
-            NAMES "sycl.lib" "sycl6.lib" "sycl7.lib"
+            NAMES "sycl.lib" "sycl6.lib" "sycl7.lib" "sycl8.lib"
             PATHS ${IntelSyclCompiler_LIBRARY_DIR}
         )
         find_file(