add tensor linalg extension (#2799)

This PR migrates the `_tensor_linalg_impl` extension to
`dpctl_ext.tensor` and extends `dpctl_ext.tensor` Python API with
`dpctl.tensor` functions `matmul`, `matrix_transpose`, `tensordot`, and
`vecdot`

Author: ndgrigorian

dpctl_ext/tensor/CMakeLists.txt

@@ -166,6 +166,10 @@ set(_sorting_sources
     ${CMAKE_CURRENT_SOURCE_DIR}/libtensor/source/sorting/searchsorted.cpp
     ${CMAKE_CURRENT_SOURCE_DIR}/libtensor/source/sorting/topk.cpp
 )
+set(_linalg_sources
+    ${CMAKE_CURRENT_SOURCE_DIR}/libtensor/source/elementwise_functions/elementwise_functions_type_utils.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/libtensor/source/linalg_functions/dot.cpp
+)
 set(_tensor_accumulation_impl_sources
     ${CMAKE_CURRENT_SOURCE_DIR}/libtensor/source/tensor_accumulation.cpp
     ${_accumulator_sources}
@@ -182,6 +186,10 @@ set(_tensor_sorting_impl_sources
     ${CMAKE_CURRENT_SOURCE_DIR}/libtensor/source/tensor_sorting.cpp
     ${_sorting_sources}
 )
+set(_tensor_linalg_impl_sources
+    ${CMAKE_CURRENT_SOURCE_DIR}/libtensor/source/tensor_linalg.cpp
+    ${_linalg_sources}
+)
 
 set(_static_lib_trgt simplify_iteration_space)
 
@@ -228,6 +236,12 @@ add_sycl_to_target(TARGET ${python_module_name} SOURCES ${_tensor_sorting_impl_s
 target_link_libraries(${python_module_name} PRIVATE ${_static_lib_trgt})
 list(APPEND _py_trgts ${python_module_name})
 
+set(python_module_name _tensor_linalg_impl)
+pybind11_add_module(${python_module_name} MODULE ${_tensor_linalg_impl_sources})
+add_sycl_to_target(TARGET ${python_module_name} SOURCES ${_tensor_linalg_impl_sources})
+target_link_libraries(${python_module_name} PRIVATE ${_static_lib_trgt})
+list(APPEND _py_trgts ${python_module_name})
+
 set(_clang_prefix "")
 if(WIN32)
     set(_clang_prefix "/clang:")
@@ -245,7 +259,7 @@ list(
     ${_elementwise_sources}
     ${_reduction_sources}
     ${_sorting_sources}
-    # ${_linalg_sources}
+    ${_linalg_sources}
     ${_accumulator_sources}
 )
 

dpctl_ext/tensor/__init__.py

@@ -107,6 +107,12 @@
     take,
     take_along_axis,
 )
+from ._linear_algebra_functions import (
+    matmul,
+    matrix_transpose,
+    tensordot,
+    vecdot,
+)
 from ._manipulation_functions import (
     broadcast_arrays,
     broadcast_to,
@@ -216,6 +222,8 @@
     "min",
     "moveaxis",
     "permute_dims",
+    "matmul",
+    "matrix_transpose",
     "negative",
     "nonzero",
     "ones",
@@ -251,6 +259,7 @@
     "take_along_axis",
     "tan",
     "tanh",
+    "tensordot",
     "tile",
     "top_k",
     "to_numpy",
@@ -262,6 +271,7 @@
     "unique_inverse",
     "unique_values",
     "unstack",
+    "vecdot",
     "where",
     "zeros",
     "zeros_like",