Merge master into include-dpctl-tensor

.github/workflows/conda-package.yml

@@ -20,7 +20,7 @@ env:
   test-env-name: 'test'
   rerun-tests-on-failure: 'true'
   rerun-tests-max-attempts: 2
-  rerun-tests-timeout: 40
+  rerun-tests-timeout: 45
 
 jobs:
   build:

.github/workflows/openssf-scorecard.yml

@@ -72,6 +72,6 @@ jobs:
 
       # Upload the results to GitHub's code scanning dashboard.
       - name: "Upload to code-scanning"
-        uses: github/codeql-action/upload-sarif@19b2f06db2b6f5108140aeb04014ef02b648f789 # v4.31.11
+        uses: github/codeql-action/upload-sarif@9e907b5e64f6b83e7804b09294d44122997950d6 # v4.32.3
         with:
           sarif_file: results.sarif

.pre-commit-config.yaml

@@ -127,7 +127,7 @@ repos:
     hooks:
     -   id: actionlint
 -   repo: https://github.com/BlankSpruce/gersemi
-    rev: 0.25.1
+    rev: 0.25.4
     hooks:
     -   id: gersemi
         exclude: "dpnp/backend/cmake/Modules/"

CHANGELOG.md

@@ -45,6 +45,13 @@ Also, that release drops support for Python 3.9, making Python 3.10 the minimum
 * Updated `dpnp.fix` to reuse `dpnp.trunc` internally [#2722](https://github.com/IntelPython/dpnp/pull/2722)
 * Changed the build scripts and documentation due to `python setup.py develop` deprecation notice [#2716](https://github.com/IntelPython/dpnp/pull/2716)
 * Clarified behavior on repeated `axes` in `dpnp.tensordot` and `dpnp.linalg.tensordot` functions [#2733](https://github.com/IntelPython/dpnp/pull/2733)
+* Improved documentation of `file` argument in `dpnp.fromfile` [#2745](https://github.com/IntelPython/dpnp/pull/2745)
+* Aligned `dpnp.trim_zeros` with NumPy 2.4 to support a tuple of integers passed with `axis` keyword [#2746](https://github.com/IntelPython/dpnp/pull/2746)
+* Aligned `strides` property of `dpnp.ndarray` with NumPy and CuPy implementations [#2747](https://github.com/IntelPython/dpnp/pull/2747)
+* Extended `dpnp.nan_to_num` to support broadcasting of `nan`, `posinf`, and `neginf` keywords [#2754](https://github.com/IntelPython/dpnp/pull/2754)
+* Changed `dpnp.partition` implementation to reuse `dpnp.sort` where it brings the performance benefit [#2766](https://github.com/IntelPython/dpnp/pull/2766)
+* `dpnp` uses pybind11 3.0.2 [#27734](https://github.com/IntelPython/dpnp/pull/2773)
+* Modified CMake files for the extension to explicitly mark DPC++ compiler and dpctl headers as system ones and so to suppress the build warning generated inside them [#2770](https://github.com/IntelPython/dpnp/pull/2770)
 
 ### Deprecated
 
@@ -59,6 +66,7 @@ Also, that release drops support for Python 3.9, making Python 3.10 the minimum
 * Removed the obsolete interface from DPNP to Numba JIT [#2647](https://github.com/IntelPython/dpnp/pull/2647)
 * Removed the `newshape` parameter from `dpnp.reshape`, which has been deprecated since dpnp 0.17.0. Pass it positionally or use `shape=` on newer versions [#2670](https://github.com/IntelPython/dpnp/pull/2670)
 * Removed unused `pytest` configuration from `pyproject.toml` [#2729](https://github.com/IntelPython/dpnp/pull/2729)
+* Dropped a conda run dependency on `onemkl-sycl-stats` package [#2771](https://github.com/IntelPython/dpnp/pull/2771)
 
 ### Fixed
 

CMakeLists.txt

@@ -60,7 +60,16 @@ set(CMAKE_FIND_ROOT_PATH_MODE_LIBRARY BOTH)
 # 8. Paths stored in the CMake System Package Registry
 # 9. Paths specified by the PATHS option (assumed hard-coded guesses)
 set(path_to_cmake_dir ${CMAKE_SOURCE_DIR}/dpnp/backend/cmake/Modules)
-find_package(IntelSYCL REQUIRED PATHS ${path_to_cmake_dir})
+# TODO: use the commented logic once the compiler resolves CMake issue CMPLRLLVM-73484
+# find_package(IntelSYCL REQUIRED PATHS ${path_to_cmake_dir})
+find_package(IntelSYCL QUIET)
+if(SYCL_LIBRARY_FOUND)
+    find_package(IntelSYCL REQUIRED)
+else()
+    # compiler CMake might have an issue and can't find SYCL_LIBRARY properly
+    # then use vendored CMake with fixed logic
+    find_package(IntelSYCL REQUIRED PATHS ${path_to_cmake_dir} NO_DEFAULT_PATH)
+endif()
 find_package(TBB REQUIRED PATHS ${path_to_cmake_dir})
 
 set(MKL_ARCH "intel64")
@@ -86,8 +95,8 @@ find_package(Python 3.10...<3.15 REQUIRED COMPONENTS Development.Module NumPy)
 include(FetchContent)
 FetchContent_Declare(
     pybind11
-    URL https://github.com/pybind/pybind11/archive/refs/tags/v3.0.1.tar.gz
-    URL_HASH SHA256=741633da746b7c738bb71f1854f957b9da660bcd2dce68d71949037f0969d0ca
+    URL https://github.com/pybind/pybind11/archive/refs/tags/v3.0.2.tar.gz
+    URL_HASH SHA256=2f20a0af0b921815e0e169ea7fec63909869323581b89d7de1553468553f6a2d
     FIND_PACKAGE_ARGS NAMES pybind11
 )
 FetchContent_MakeAvailable(pybind11)

conda-recipe/bld.bat

@@ -13,7 +13,7 @@ if DEFINED OVERRIDE_INTEL_IPO (
   set "CMAKE_ARGS=%CMAKE_ARGS% -DCMAKE_INTERPROCEDURAL_OPTIMIZATION:BOOL=FALSE"
 )
 
-FOR %%V IN (17.0.0 17 18.0.0 18 19.0.0 19 20.0.0 20 21.0.0 21) DO @(
+FOR %%V IN (17.0.0 17 18.0.0 18 19.0.0 19 20.0.0 20 21.0.0 21 22.0.0 22) 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/meta.yaml

@@ -1,4 +1,4 @@
-{% set max_compiler_and_mkl_version = environ.get("MAX_BUILD_CMPL_MKL_VERSION", "2026.0a0") %}
+{% set max_compiler_and_mkl_version = environ.get("MAX_BUILD_CMPL_MKL_VERSION", "2027.0a0") %}
 {% set required_compiler_and_mkl_version = "2025.0" %}
 {% set required_dpctl_version = "0.22.0*" %}
 
@@ -49,7 +49,6 @@ requirements:
       - {{ pin_compatible('onemkl-sycl-dft', min_pin='x.x', max_pin='x') }}
       - {{ pin_compatible('onemkl-sycl-lapack', min_pin='x.x', max_pin='x') }}
       - {{ pin_compatible('onemkl-sycl-rng', min_pin='x.x', max_pin='x') }}
-      - {{ pin_compatible('onemkl-sycl-stats', min_pin='x.x', max_pin='x') }}
       - {{ pin_compatible('onemkl-sycl-vm', min_pin='x.x', max_pin='x') }}
       - numpy
       - intel-gpu-ocl-icd-system

doc/reference/set.rst

@@ -23,7 +23,6 @@ Boolean operations
    :toctree: generated/
    :nosignatures:
 
-   in1d
    intersect1d
    isin
    setdiff1d

dpnp/backend/cmake/Modules/IntelSYCLConfig.cmake

@@ -329,7 +329,7 @@ if(SYCL_COMPILER)
       )
   #TODO Make an input file to configure and update the lib current version
   if(WIN32)
-    set(sycl_lib_suffix "8")
+    set(sycl_lib_suffix "9")
   else()
     set(sycl_lib_suffix "")
   endif()

dpnp/backend/extensions/blas/CMakeLists.txt

@@ -70,7 +70,13 @@ target_include_directories(
         ${CMAKE_SOURCE_DIR}/dpnp/backend/include
         ${CMAKE_SOURCE_DIR}/dpctl/tensor/libtensor/include
 )
-target_include_directories(${python_module_name} PUBLIC ${Dpctl_INCLUDE_DIR})
+
+# treat below headers as system to suppress the warnings there during the build
+target_include_directories(
+    ${python_module_name}
+    SYSTEM
+    PRIVATE ${SYCL_INCLUDE_DIR} ${Dpctl_INCLUDE_DIRS} ${Dpctl_TENSOR_INCLUDE_DIR}
+)
 
 if(WIN32)
     target_compile_options(

dpnp/backend/extensions/fft/CMakeLists.txt

@@ -63,7 +63,13 @@ target_include_directories(
         ${CMAKE_SOURCE_DIR}/dpnp/backend/include
         ${CMAKE_SOURCE_DIR}/dpctl/tensor/libtensor/include
 )
-target_include_directories(${python_module_name} PUBLIC ${Dpctl_INCLUDE_DIR})
+
+# treat below headers as system to suppress the warnings there during the build
+target_include_directories(
+    ${python_module_name}
+    SYSTEM
+    PRIVATE ${SYCL_INCLUDE_DIR} ${Dpctl_INCLUDE_DIRS} ${Dpctl_TENSOR_INCLUDE_DIR}
+)
 
 if(WIN32)
     target_compile_options(

dpnp/backend/extensions/indexing/CMakeLists.txt

@@ -67,7 +67,13 @@ target_include_directories(
         ${CMAKE_SOURCE_DIR}/dpnp/backend/include
         ${CMAKE_SOURCE_DIR}/dpctl/tensor/libtensor/include
 )
-target_include_directories(${python_module_name} PUBLIC ${Dpctl_INCLUDE_DIR})
+
+# treat below headers as system to suppress the warnings there during the build
+target_include_directories(
+    ${python_module_name}
+    SYSTEM
+    PRIVATE ${SYCL_INCLUDE_DIR} ${Dpctl_INCLUDE_DIRS} ${Dpctl_TENSOR_INCLUDE_DIR}
+)
 
 if(WIN32)
     target_compile_options(

dpnp/backend/extensions/lapack/CMakeLists.txt

@@ -80,6 +80,11 @@ set_target_properties(
     PROPERTIES CMAKE_POSITION_INDEPENDENT_CODE ON
 )
 
+target_include_directories(
+    ${python_module_name}
+    PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/../common
+)
+
 target_include_directories(
     ${python_module_name}
     PRIVATE
@@ -88,7 +93,13 @@ target_include_directories(
         ${CMAKE_SOURCE_DIR}/dpnp/backend/include
         ${CMAKE_SOURCE_DIR}/dpctl/tensor/libtensor/include
 )
-target_include_directories(${python_module_name} PUBLIC ${Dpctl_INCLUDE_DIR})
+
+# treat below headers as system to suppress the warnings there during the build
+target_include_directories(
+    ${python_module_name}
+    SYSTEM
+    PRIVATE ${SYCL_INCLUDE_DIR} ${Dpctl_INCLUDE_DIRS} ${Dpctl_TENSOR_INCLUDE_DIR}
+)
 
 if(WIN32)
     target_compile_options(

dpnp/backend/extensions/statistics/CMakeLists.txt

@@ -72,7 +72,13 @@ target_include_directories(
         ${CMAKE_SOURCE_DIR}/dpnp/backend/include
         ${CMAKE_SOURCE_DIR}/dpctl/tensor/libtensor/include
 )
-target_include_directories(${python_module_name} PUBLIC ${Dpctl_INCLUDE_DIR})
+
+# treat below headers as system to suppress the warnings there during the build
+target_include_directories(
+    ${python_module_name}
+    SYSTEM
+    PRIVATE ${SYCL_INCLUDE_DIR} ${Dpctl_INCLUDE_DIRS} ${Dpctl_TENSOR_INCLUDE_DIR}
+)
 
 if(WIN32)
     target_compile_options(

dpnp/backend/extensions/ufunc/CMakeLists.txt

@@ -90,7 +90,13 @@ target_include_directories(
         ${CMAKE_SOURCE_DIR}/dpnp/backend/include
         ${CMAKE_SOURCE_DIR}/dpctl/tensor/libtensor/include
 )
-target_include_directories(${python_module_name} PUBLIC ${Dpctl_INCLUDE_DIR})
+
+# treat below headers as system to suppress the warnings there during the build
+target_include_directories(
+    ${python_module_name}
+    SYSTEM
+    PRIVATE ${SYCL_INCLUDE_DIR} ${Dpctl_INCLUDE_DIRS} ${Dpctl_TENSOR_INCLUDE_DIR}
+)
 
 if(_dpnp_sycl_targets)
     # make fat binary

dpnp/backend/extensions/vm/CMakeLists.txt

@@ -112,7 +112,13 @@ target_include_directories(
         ${CMAKE_SOURCE_DIR}/dpnp/backend/include
         ${CMAKE_SOURCE_DIR}/dpctl/tensor/libtensor/include
 )
-target_include_directories(${python_module_name} PUBLIC ${Dpctl_INCLUDE_DIR})
+
+# treat below headers as system to suppress the warnings there during the build
+target_include_directories(
+    ${python_module_name}
+    SYSTEM
+    PRIVATE ${SYCL_INCLUDE_DIR} ${Dpctl_INCLUDE_DIRS} ${Dpctl_TENSOR_INCLUDE_DIR}
+)
 
 if(WIN32)
     target_compile_options(

dpnp/backend/extensions/window/CMakeLists.txt

@@ -68,7 +68,13 @@ target_include_directories(
         ${CMAKE_SOURCE_DIR}/dpnp/backend/include
         ${CMAKE_SOURCE_DIR}/dpctl/tensor/libtensor/include
 )
-target_include_directories(${python_module_name} PUBLIC ${Dpctl_INCLUDE_DIR})
+
+# treat below headers as system to suppress the warnings there during the build
+target_include_directories(
+    ${python_module_name}
+    SYSTEM
+    PRIVATE ${SYCL_INCLUDE_DIR} ${Dpctl_INCLUDE_DIRS} ${Dpctl_TENSOR_INCLUDE_DIR}
+)
 
 if(WIN32)
     target_compile_options(

dpnp/backend/kernels/dpnp_krnl_sorting.cpp

@@ -70,90 +70,27 @@ DPCTLSyclEventRef dpnp_partition_c(DPCTLSyclQueueRef q_ref,
 
     sycl::queue q = *(reinterpret_cast<sycl::queue *>(q_ref));
 
-    if (ndim == 1) // 1d array with C-contiguous data
-    {
-        _DataType *arr = static_cast<_DataType *>(array1_in);
-        _DataType *result = static_cast<_DataType *>(result1);
+    _DataType *arr = static_cast<_DataType *>(array1_in);
+    _DataType *result = static_cast<_DataType *>(result1);
 
-        auto policy = oneapi::dpl::execution::make_device_policy<
-            dpnp_partition_c_kernel<_DataType>>(q);
+    auto policy = oneapi::dpl::execution::make_device_policy<
+        dpnp_partition_c_kernel<_DataType>>(q);
 
-        // fill the result array with data from input one
-        q.memcpy(result, arr, size * sizeof(_DataType)).wait();
+    // fill the result array with data from input one
+    q.memcpy(result, arr, size * sizeof(_DataType)).wait();
 
-        // make a partial sorting such that:
+    // note, a loop for a multidemension input array (size_ > 1) is an
+    // experimental and it isn't tested properly as for now
+    for (size_t i = 0; i < size_; i++) {
+        _DataType *bufptr = result + i * shape_[0];
+
+        // for every slice it makes a partial sorting such that:
         // 1. result[0 <= i < kth]    <= result[kth]
         // 2. result[kth <= i < size] >= result[kth]
         // event-blocking call, no need for wait()
-        std::nth_element(policy, result, result + kth, result + size,
+        std::nth_element(policy, bufptr, bufptr + kth, bufptr + size,
                          dpnp_less_comp());
-        return event_ref;
-    }
-
-    DPNPC_ptr_adapter<_DataType> input1_ptr(q_ref, array1_in, size, true);
-    DPNPC_ptr_adapter<_DataType> input2_ptr(q_ref, array2_in, size, true);
-    DPNPC_ptr_adapter<_DataType> result1_ptr(q_ref, result1, size, true, true);
-    _DataType *arr = input1_ptr.get_ptr();
-    _DataType *arr2 = input2_ptr.get_ptr();
-    _DataType *result = result1_ptr.get_ptr();
-
-    auto arr_to_result_event = q.memcpy(result, arr, size * sizeof(_DataType));
-    arr_to_result_event.wait();
-
-    _DataType *matrix = new _DataType[shape_[ndim - 1]];
-
-    for (size_t i = 0; i < size_; ++i) {
-        size_t ind_begin = i * shape_[ndim - 1];
-        size_t ind_end = (i + 1) * shape_[ndim - 1] - 1;
-
-        for (size_t j = ind_begin; j < ind_end + 1; ++j) {
-            size_t ind = j - ind_begin;
-            matrix[ind] = arr2[j];
-        }
-        std::partial_sort(matrix, matrix + shape_[ndim - 1],
-                          matrix + shape_[ndim - 1], dpnp_less_comp());
-        for (size_t j = ind_begin; j < ind_end + 1; ++j) {
-            size_t ind = j - ind_begin;
-            arr2[j] = matrix[ind];
-        }
     }
-
-    shape_elem_type *shape = reinterpret_cast<shape_elem_type *>(
-        sycl::malloc_shared(ndim * sizeof(shape_elem_type), q));
-    auto memcpy_event = q.memcpy(shape, shape_, ndim * sizeof(shape_elem_type));
-
-    memcpy_event.wait();
-
-    sycl::range<2> gws(size_, kth + 1);
-    auto kernel_parallel_for_func = [=](sycl::id<2> global_id) {
-        size_t j = global_id[0];
-        size_t k = global_id[1];
-
-        _DataType val = arr2[j * shape[ndim - 1] + k];
-
-        for (size_t i = 0; i < static_cast<size_t>(shape[ndim - 1]); ++i) {
-            if (result[j * shape[ndim - 1] + i] == val) {
-                _DataType change_val1 = result[j * shape[ndim - 1] + i];
-                _DataType change_val2 = result[j * shape[ndim - 1] + k];
-                result[j * shape[ndim - 1] + k] = change_val1;
-                result[j * shape[ndim - 1] + i] = change_val2;
-            }
-        }
-    };
-
-    auto kernel_func = [&](sycl::handler &cgh) {
-        cgh.depends_on({memcpy_event});
-        cgh.parallel_for<class dpnp_partition_c_kernel<_DataType>>(
-            gws, kernel_parallel_for_func);
-    };
-
-    auto event = q.submit(kernel_func);
-
-    event.wait();
-
-    delete[] matrix;
-    sycl::free(shape, q);
-
     return event_ref;
 }