.. index:: pair: page; Select .. _doxid-dev_guide_op_select: Select ====== General ~~~~~~~ The select operation returns a tensor filled with the elements from the second or the third input, depending on the condition (the first input) value. .. code-block:: cpp dst[i] = cond[i] ? src_0[i] : src_1[i] Broadcasting is supported. Broadcasting rules ------------------ If the auto_broadcast attribute is not none, the select operation takes a two-step broadcast before performing the selection: * Step 1 : Input tensors src_0 and src_1 are broadcasted to dst_shape according to the Numpy `broadcast rules `__. * Step 2 : Then, the cond tensor will be one-way broadcasted to the dst_shape of broadcasted src_0 and src_1. To be more specific, we align the two shapes to the right and compare them from right to left. Each dimension should be either equal or the dimension of cond should be 1. * example : * cond={4, 5}, dst_shape={2, 3, 4, 5} => dst = {2, 3, 4, 5} * cond={3, 1, 5}, dst_shape={2, 3, 4, 5} => dst = {2, 3, 4, 5} * cond={3,5}, dst_shape={2, 3, 4, 5} => dst = invalid_shape Operation attributes ~~~~~~~~~~~~~~~~~~~~ =========================================================================================================================== =========================================================== =========== =============================== ===================== Attribute Name Description Value Type Supported Values Required or Optional =========================================================================================================================== =========================================================== =========== =============================== ===================== :ref:`auto_broadcast ` Specifies rules used for auto-broadcasting of src tensors. string ``none`` , ``numpy`` (default) Optional =========================================================================================================================== =========================================================== =========== =============================== ===================== Execution arguments ~~~~~~~~~~~~~~~~~~~ The inputs and outputs must be provided according to the following index order when constructing an operation. Inputs ------ ====== ============== ===================== Index Argument Name Required or Optional ====== ============== ===================== 0 ``cond`` Required 1 ``src_0`` Required 2 ``src_1`` Required ====== ============== ===================== .. note:: All input shapes should match and no broadcasting is allowed if the ``auto_broadcast`` attribute is set to ``none``, or can be broadcasted according to the broadcasting rules mentioned above if ``auto_broadcast`` attribute is set to ``numpy``. Outputs ------- ====== ============== ===================== Index Argument Name Required or Optional ====== ============== ===================== 0 ``dst`` Required ====== ============== ===================== Supported data types ~~~~~~~~~~~~~~~~~~~~ Select operation supports the following data type combinations. ======== ============== ===== Cond Src_0 / Src_1 Dst ======== ============== ===== boolean f32 f32 boolean bf16 bf16 boolean f16 f16 ======== ============== ===== Implementation Notes ~~~~~~~~~~~~~~~~~~~~ Select supports in-place operations, meaning that ``src_0`` can be used as both input and output. In case of in-place operation, the original ``src_0`` data will be overwritten. Use in-place operations whenever possible for performance.