add kernel for cross function (#535)

* start to implement

Author: KsanaKozlova

dpnp/backend/include/dpnp_iface.hpp

@@ -156,6 +156,19 @@ INP_DLLEXPORT void dpnp_elemwise_absolute_c(void* array1_in, void* result1, size
 template <typename _DataType_input1, typename _DataType_input2, typename _DataType_output>
 INP_DLLEXPORT void dpnp_dot_c(void* array1, void* array2, void* result1, size_t size);
 
+/**
+ * @ingroup BACKEND_API
+ * @brief Custom implementation of cross function
+ *
+ * @param [in]  array1_in  First input array.
+ * @param [in]  array2_in  Second input array.
+ * @param [out] result1 Output array.
+ * @param [in]  size    Number of elements in input arrays.
+ *
+ */
+template <typename _DataType_input1, typename _DataType_input2, typename _DataType_output>
+INP_DLLEXPORT void dpnp_cross_c(void* array1_in, void* array2_in, void* result1, size_t size);
+
 /**
  * @ingroup BACKEND_API
  * @brief Sum of array elements

dpnp/backend/include/dpnp_iface_fptr.hpp

@@ -83,6 +83,7 @@ enum class DPNPFuncName : size_t
     DPNP_FN_COS,                      /**< Used in numpy.cos() implementation  */
     DPNP_FN_COSH,                     /**< Used in numpy.cosh() implementation  */
     DPNP_FN_COV,                      /**< Used in numpy.cov() implementation  */
+    DPNP_FN_CROSS,                    /**< Used in numpy.cross() implementation  */
     DPNP_FN_DEGREES,                  /**< Used in numpy.degrees() implementation  */
     DPNP_FN_DET,                      /**< Used in numpy.linalg.det() implementation  */
     DPNP_FN_DIVIDE,                   /**< Used in numpy.divide() implementation  */

dpnp/backend/kernels/dpnp_krnl_mathematical.cpp

@@ -83,6 +83,25 @@ template void dpnp_elemwise_absolute_c<float>(void* array1_in, void* result1, si
 template void dpnp_elemwise_absolute_c<long>(void* array1_in, void* result1, size_t size);
 template void dpnp_elemwise_absolute_c<int>(void* array1_in, void* result1, size_t size);
 
+template <typename _KernelNameSpecialization1, typename _KernelNameSpecialization2, typename _KernelNameSpecialization3>
+class dpnp_cross_c_kernel;
+
+template <typename _DataType_input1, typename _DataType_input2, typename _DataType_output>
+void dpnp_cross_c(void* array1_in, void* array2_in, void* result1, size_t size)
+{
+    _DataType_input1* array1 = reinterpret_cast<_DataType_input1*>(array1_in);
+    _DataType_input2* array2 = reinterpret_cast<_DataType_input2*>(array2_in);
+    _DataType_output* result = reinterpret_cast<_DataType_output*>(result1);
+
+    result[0] = array1[1] * array2[2] - array1[2] * array2[1];
+
+    result[1] = array1[2] * array2[0] - array1[0] * array2[2];
+
+    result[2] = array1[0] * array2[1] - array1[1] * array2[0];
+
+    return;
+}
+
 template <typename _KernelNameSpecialization1, typename _KernelNameSpecialization2, typename _KernelNameSpecialization3>
 class dpnp_floor_divide_c_kernel;
 
@@ -214,6 +233,23 @@ void func_map_init_mathematical(func_map_t& fmap)
     fmap[DPNPFuncName::DPNP_FN_ABSOLUTE][eft_FLT][eft_FLT] = {eft_FLT, (void*)dpnp_elemwise_absolute_c<float>};
     fmap[DPNPFuncName::DPNP_FN_ABSOLUTE][eft_DBL][eft_DBL] = {eft_DBL, (void*)dpnp_elemwise_absolute_c<double>};
 
+    fmap[DPNPFuncName::DPNP_FN_CROSS][eft_INT][eft_INT] = {eft_INT, (void*)dpnp_cross_c<int, int, int>};
+    fmap[DPNPFuncName::DPNP_FN_CROSS][eft_INT][eft_LNG] = {eft_LNG, (void*)dpnp_cross_c<int, long, long>};
+    fmap[DPNPFuncName::DPNP_FN_CROSS][eft_INT][eft_FLT] = {eft_DBL, (void*)dpnp_cross_c<int, float, double>};
+    fmap[DPNPFuncName::DPNP_FN_CROSS][eft_INT][eft_DBL] = {eft_DBL, (void*)dpnp_cross_c<int, double, double>};
+    fmap[DPNPFuncName::DPNP_FN_CROSS][eft_LNG][eft_INT] = {eft_LNG, (void*)dpnp_cross_c<long, int, long>};
+    fmap[DPNPFuncName::DPNP_FN_CROSS][eft_LNG][eft_LNG] = {eft_LNG, (void*)dpnp_cross_c<long, long, long>};
+    fmap[DPNPFuncName::DPNP_FN_CROSS][eft_LNG][eft_FLT] = {eft_DBL, (void*)dpnp_cross_c<long, float, double>};
+    fmap[DPNPFuncName::DPNP_FN_CROSS][eft_LNG][eft_DBL] = {eft_DBL, (void*)dpnp_cross_c<long, double, double>};
+    fmap[DPNPFuncName::DPNP_FN_CROSS][eft_FLT][eft_INT] = {eft_DBL, (void*)dpnp_cross_c<float, int, double>};
+    fmap[DPNPFuncName::DPNP_FN_CROSS][eft_FLT][eft_LNG] = {eft_DBL, (void*)dpnp_cross_c<float, long, double>};
+    fmap[DPNPFuncName::DPNP_FN_CROSS][eft_FLT][eft_FLT] = {eft_FLT, (void*)dpnp_cross_c<float, float, float>};
+    fmap[DPNPFuncName::DPNP_FN_CROSS][eft_FLT][eft_DBL] = {eft_DBL, (void*)dpnp_cross_c<float, double, double>};
+    fmap[DPNPFuncName::DPNP_FN_CROSS][eft_DBL][eft_INT] = {eft_DBL, (void*)dpnp_cross_c<double, int, double>};
+    fmap[DPNPFuncName::DPNP_FN_CROSS][eft_DBL][eft_LNG] = {eft_DBL, (void*)dpnp_cross_c<double, long, double>};
+    fmap[DPNPFuncName::DPNP_FN_CROSS][eft_DBL][eft_FLT] = {eft_DBL, (void*)dpnp_cross_c<double, float, double>};
+    fmap[DPNPFuncName::DPNP_FN_CROSS][eft_DBL][eft_DBL] = {eft_DBL, (void*)dpnp_cross_c<double, double, double>};
+
     fmap[DPNPFuncName::DPNP_FN_FLOOR_DIVIDE][eft_INT][eft_INT] = {eft_INT, (void*)dpnp_floor_divide_c<int, int, int>};
     fmap[DPNPFuncName::DPNP_FN_FLOOR_DIVIDE][eft_INT][eft_LNG] = {eft_LNG, (void*)dpnp_floor_divide_c<int, long, long>};
     fmap[DPNPFuncName::DPNP_FN_FLOOR_DIVIDE][eft_INT][eft_FLT] = {eft_DBL,

dpnp/dpnp_algo/dpnp_algo.pxd

@@ -56,6 +56,7 @@ cdef extern from "dpnp_iface_fptr.hpp" namespace "DPNPFuncName":  # need this na
         DPNP_FN_COS
         DPNP_FN_COSH
         DPNP_FN_COV
+        DPNP_FN_CROSS
         DPNP_FN_DEGREES
         DPNP_FN_DET
         DPNP_FN_DIVIDE

dpnp/dpnp_algo/dpnp_algo_mathematical.pyx

@@ -136,29 +136,7 @@ cpdef dparray dpnp_copysign(dparray x1, dparray x2):
 
 
 cpdef dparray dpnp_cross(dparray x1, dparray x2):
-
-    types_map = {
-        (dpnp.int32, dpnp.int32): dpnp.int32,
-        (dpnp.int32, dpnp.int64): dpnp.int64,
-        (dpnp.int64, dpnp.int32): dpnp.int64,
-        (dpnp.int64, dpnp.int64): dpnp.int64,
-        (dpnp.float32, dpnp.float32): dpnp.float32,
-    }
-
-    res_type = types_map.get((x1.dtype.type, x2.dtype.type), dpnp.float64)
-
-    cdef dparray result = dparray(3, dtype=res_type)
-
-    cur_res = x1[1] * x2[2] - x1[2] * x2[1]
-    result._setitem_scalar(0, cur_res)
-
-    cur_res = x1[2] * x2[0] - x1[0] * x2[2]
-    result._setitem_scalar(1, cur_res)
-
-    cur_res = x1[0] * x2[1] - x1[1] * x2[0]
-    result._setitem_scalar(2, cur_res)
-
-    return result
+    return call_fptr_2in_1out(DPNP_FN_CROSS, x1, x2, x1.shape)
 
 
 cpdef dparray dpnp_cumprod(dparray x1, bint usenan=False):