Linalg to desc 1 (#813)

Author: shssf

dpnp/dpnp_algo/dpnp_algo_statistics.pyx

@@ -138,6 +138,7 @@ cpdef dparray dpnp_cov(dparray array1):
 
 
 cpdef dparray _dpnp_max(dparray input, _axis_, output_shape):
+    cdef dparray_shape_type input_shape = input.shape
     cdef DPNPFuncType param1_type = dpnp_dtype_to_DPNPFuncType(input.dtype)
 
     cdef DPNPFuncData kernel_data = get_dpnp_function_ptr(DPNP_FN_MAX, param1_type, param1_type)
@@ -157,7 +158,7 @@ cpdef dparray _dpnp_max(dparray input, _axis_, output_shape):
             axis_.push_back(shape_it)
         axis_size = len(axis)
 
-    func(input.get_data(), result.get_data(), < size_t * > input._dparray_shape.data(), input.ndim, < size_t * > axis_.data(), axis_size)
+    func(input.get_data(), result.get_data(), < size_t * > input_shape.data(), input.ndim, < size_t * > axis_.data(), axis_size)
 
     dpnp_array = dpnp.array(result, dtype=input.dtype)
     dpnp_result_array = dpnp_array.reshape(output_shape)
@@ -194,6 +195,7 @@ cpdef dparray dpnp_max(dparray input, axis):
 
 
 cpdef dparray _dpnp_mean(dparray input):
+    cdef dparray_shape_type input_shape = input.shape
     cdef DPNPFuncType param1_type = dpnp_dtype_to_DPNPFuncType(input.dtype)
 
     cdef DPNPFuncData kernel_data = get_dpnp_function_ptr(DPNP_FN_MEAN, param1_type, param1_type)
@@ -207,7 +209,7 @@ cpdef dparray _dpnp_mean(dparray input):
     cdef dparray_shape_type axis
     cdef Py_ssize_t axis_size = 0
 
-    func(input.get_data(), result.get_data(), < size_t * > input._dparray_shape.data(), input.ndim, < size_t * > axis.data(), axis_size)
+    func(input.get_data(), result.get_data(), < size_t * > input_shape.data(), input.ndim, < size_t * > axis.data(), axis_size)
 
     return result
 
@@ -336,6 +338,7 @@ cpdef dparray dpnp_median(utils.dpnp_descriptor array1):
 
 
 cpdef dparray _dpnp_min(dparray input, _axis_, output_shape):
+    cdef dparray_shape_type input_shape = input.shape
     cdef DPNPFuncType param1_type = dpnp_dtype_to_DPNPFuncType(input.dtype)
 
     cdef DPNPFuncData kernel_data = get_dpnp_function_ptr(DPNP_FN_MIN, param1_type, param1_type)
@@ -357,7 +360,7 @@ cpdef dparray _dpnp_min(dparray input, _axis_, output_shape):
             axis_.push_back(shape_it)
         axis_size = len(axis)
 
-    func(input.get_data(), result.get_data(), < size_t * > input._dparray_shape.data(), input.ndim, < size_t * > axis_.data(), axis_size)
+    func(input.get_data(), result.get_data(), < size_t * > input_shape.data(), input.ndim, < size_t * > axis_.data(), axis_size)
 
     dpnp_array = dpnp.array(result, dtype=input.dtype)
     dpnp_result_array = dpnp_array.reshape(output_shape)

dpnp/linalg/dpnp_algo_linalg.pyx

@@ -33,7 +33,7 @@ and the rest of the library
 """
 
 import dpnp
-from dpnp.dpnp_utils cimport *
+cimport dpnp.dpnp_utils as utils
 from dpnp.dpnp_algo cimport *
 from dpnp.dparray cimport dparray, dparray_shape_type
 import numpy
@@ -63,24 +63,19 @@ ctypedef void(*custom_linalg_1in_3out_shape_t)(void * , void * , void * , void *
 ctypedef void(*custom_linalg_2in_1out_func_ptr_t)(void * , void * , void * , size_t )
 
 
-cpdef dparray dpnp_cholesky(dparray input):
-    if input.dtype == dpnp.int32 or input.dtype == dpnp.int64:
-        input_ = input.astype(dpnp.float64)
-    else:
-        input_ = input
-
+cpdef utils.dpnp_descriptor dpnp_cholesky(utils.dpnp_descriptor input_):
     size_ = input_.shape[-1]
 
     cdef DPNPFuncType param1_type = dpnp_dtype_to_DPNPFuncType(input_.dtype)
 
     cdef DPNPFuncData kernel_data = get_dpnp_function_ptr(DPNP_FN_CHOLESKY, param1_type, param1_type)
 
-    result_type = dpnp_DPNPFuncType_to_dtype( < size_t > kernel_data.return_type)
-    cdef dparray result = dparray(input_.shape, dtype=result_type)
+    # ceate result array with type given by FPTR data
+    cdef utils.dpnp_descriptor result = utils.create_output_descriptor(input_.shape, kernel_data.return_type, None)
 
     cdef custom_linalg_1in_1out_with_2size_func_ptr_t_ func = <custom_linalg_1in_1out_with_2size_func_ptr_t_ > kernel_data.ptr
 
-    func(input_.get_data(), result.get_data(), input.size, size_)
+    func(input_.get_data(), result.get_data(), input_.size, size_)
 
     return result
 
@@ -108,30 +103,31 @@ cpdef dparray dpnp_cond(dparray input, p):
     return ret
 
 
-cpdef dparray dpnp_det(dparray input):
+cpdef utils.dpnp_descriptor dpnp_det(utils.dpnp_descriptor input):
+    cdef dparray_shape_type input_shape = input.shape
     cdef size_t n = input.shape[-1]
     cdef size_t size_out = 1
     if input.ndim != 2:
         output_shape = tuple((list(input.shape))[:-2])
         for i in range(len(output_shape)):
             size_out *= output_shape[i]
 
+    cdef dparray_shape_type result_shape = (size_out,)
+    if size_out > 1:
+        result_shape = output_shape
+
     cdef DPNPFuncType param1_type = dpnp_dtype_to_DPNPFuncType(input.dtype)
 
     cdef DPNPFuncData kernel_data = get_dpnp_function_ptr(DPNP_FN_DET, param1_type, param1_type)
 
-    result_type = dpnp_DPNPFuncType_to_dtype( < size_t > kernel_data.return_type)
-    cdef dparray result = dparray(size_out, dtype=result_type)
+    # ceate result array with type given by FPTR data
+    cdef utils.dpnp_descriptor result = utils.create_output_descriptor(result_shape, kernel_data.return_type, None)
 
     cdef custom_linalg_1in_1out_func_ptr_t func = <custom_linalg_1in_1out_func_ptr_t > kernel_data.ptr
 
-    func(input.get_data(), result.get_data(), < size_t * > input._dparray_shape.data(), input.ndim)
+    func(input.get_data(), result.get_data(), < size_t * > input_shape.data(), input.ndim)
 
-    if size_out > 1:
-        dpnp_result = result.reshape(output_shape)
-        return dpnp_result
-    else:
-        return result
+    return result
 
 
 cpdef tuple dpnp_eig(dparray x1):
@@ -154,17 +150,16 @@ cpdef tuple dpnp_eig(dparray x1):
     return (res_val, res_vec)
 
 
-cpdef dparray dpnp_eigvals(dparray input):
+cpdef utils.dpnp_descriptor dpnp_eigvals(utils.dpnp_descriptor input):
     cdef dparray_shape_type input_shape = input.shape
 
     cdef size_t size = 0 if input_shape.empty() else input_shape.front()
 
     cdef DPNPFuncType param1_type = dpnp_dtype_to_DPNPFuncType(input.dtype)
     cdef DPNPFuncData kernel_data = get_dpnp_function_ptr(DPNP_FN_EIGVALS, param1_type, param1_type)
 
-    result_type = dpnp_DPNPFuncType_to_dtype( < size_t > kernel_data.return_type)
-
-    cdef dparray res_val = dparray((size,), dtype=result_type)
+    # ceate result array with type given by FPTR data
+    cdef utils.dpnp_descriptor res_val = utils.create_output_descriptor((size,), kernel_data.return_type, None)
 
     cdef custom_linalg_1in_1out_with_size_func_ptr_t_ func = <custom_linalg_1in_1out_with_size_func_ptr_t_ > kernel_data.ptr
     # call FPTR function
@@ -175,6 +170,8 @@ cpdef dparray dpnp_eigvals(dparray input):
 
 cpdef dparray dpnp_inv(dparray input_):
     cdef dparray input = input_.astype(dpnp.float64)
+    cdef dparray_shape_type input_shape = input.shape
+
     cdef DPNPFuncType param1_type = dpnp_dtype_to_DPNPFuncType(input.dtype)
 
     cdef DPNPFuncData kernel_data = get_dpnp_function_ptr(DPNP_FN_INV, param1_type, param1_type)
@@ -183,23 +180,24 @@ cpdef dparray dpnp_inv(dparray input_):
 
     cdef custom_linalg_1in_1out_func_ptr_t func = <custom_linalg_1in_1out_func_ptr_t > kernel_data.ptr
 
-    func(input.get_data(), result.get_data(), < size_t * > input._dparray_shape.data(), input.ndim)
+    func(input.get_data(), result.get_data(), < size_t * > input_shape.data(), input.ndim)
 
     dpnp_result = result.reshape(input.shape)
     return dpnp_result
 
 
-cpdef dparray dpnp_matrix_rank(dparray input):
+cpdef utils.dpnp_descriptor dpnp_matrix_rank(utils.dpnp_descriptor input):
+    cdef dparray_shape_type input_shape = input.shape
     cdef DPNPFuncType param1_type = dpnp_dtype_to_DPNPFuncType(input.dtype)
 
     cdef DPNPFuncData kernel_data = get_dpnp_function_ptr(DPNP_FN_MATRIX_RANK, param1_type, param1_type)
 
-    result_type = dpnp_DPNPFuncType_to_dtype( < size_t > kernel_data.return_type)
-    cdef dparray result = dparray((1,), dtype=result_type)
+    # ceate result array with type given by FPTR data
+    cdef utils.dpnp_descriptor result = utils.create_output_descriptor((1,), kernel_data.return_type, None)
 
     cdef custom_linalg_1in_1out_func_ptr_t func = <custom_linalg_1in_1out_func_ptr_t > kernel_data.ptr
 
-    func(input.get_data(), result.get_data(), < size_t * > input._dparray_shape.data(), input.ndim)
+    func(input.get_data(), result.get_data(), < size_t * > input_shape.data(), input.ndim)
 
     return result
 

dpnp/linalg/dpnp_iface_linalg.py

@@ -88,13 +88,17 @@ def cholesky(input):
         matrix object if `input` is a matrix object.
     """
 
-    if not use_origin_backend(input):
-        if not isinstance(input, dparray):
-            pass
-        elif input.shape[-1] != input.shape[-2]:
+    x1_desc = dpnp.get_dpnp_descriptor(input)
+    if x1_desc:
+        if x1_desc.shape[-1] != x1_desc.shape[-2]:
             pass
         else:
-            return dpnp_cholesky(input)
+            if input.dtype == dpnp.int32 or input.dtype == dpnp.int64:
+                # TODO memory copy. needs to move into DPNPC
+                input_ = dpnp.get_dpnp_descriptor(input.astype(dpnp.float64))
+            else:
+                input_ = x1_desc
+            return dpnp_cholesky(input_).get_pyobj()
 
     return call_origin(numpy.linalg.cholesky, input)
 
@@ -140,11 +144,11 @@ def det(input):
     det : (...) array_like
         Determinant of `input`.
     """
-    is_input_dparray = isinstance(input, dparray)
 
-    if not use_origin_backend(input) and is_input_dparray:
-        if input.shape[-1] == input.shape[-2]:
-            result_obj = dpnp_det(input)
+    x1_desc = dpnp.get_dpnp_descriptor(input)
+    if x1_desc:
+        if x1_desc.shape[-1] == x1_desc.shape[-2]:
+            result_obj = dpnp_det(x1_desc).get_pyobj()
             result = dpnp.convert_single_elem_array_to_scalar(result_obj)
 
             return result
@@ -188,11 +192,10 @@ def eigvals(input):
         real for real matrices.
     """
 
-    is_input_dparray = isinstance(input, dparray)
-
-    if (not use_origin_backend(input) and is_input_dparray):
-        if (input.size > 0):
-            return dpnp_eigvals(input)
+    x1_desc = dpnp.get_dpnp_descriptor(input)
+    if x1_desc:
+        if x1_desc.size > 0:
+            return dpnp_eigvals(x1_desc).get_pyobj()
 
     return call_origin(numpy.linalg.eigvals, input)
 
@@ -276,18 +279,17 @@ def matrix_rank(input, tol=None, hermitian=False):
 
     """
 
-    is_input_dparray = isinstance(input, dparray)
-
-    if not use_origin_backend(input) and is_input_dparray:
+    x1_desc = dpnp.get_dpnp_descriptor(input)
+    if x1_desc:
         if tol is not None:
-            checker_throw_value_error("matrix_rank", "tol", type(tol), None)
-        if hermitian is not False:
-            checker_throw_value_error("matrix_rank", "hermitian", hermitian, False)
-
-        result_obj = dpnp_matrix_rank(input)
-        result = dpnp.convert_single_elem_array_to_scalar(result_obj)
-
-        return result
+            pass
+        elif hermitian:
+            pass
+        else:
+            result_obj = dpnp_matrix_rank(x1_desc).get_pyobj()
+            result = dpnp.convert_single_elem_array_to_scalar(result_obj)
+    
+            return result
 
     return call_origin(numpy.linalg.matrix_rank, input, tol, hermitian)