FilesLast commitLast update
4 months ago
1 month ago
1 month ago
1 month ago
1 month ago
1 month ago
1 month ago
11 days ago
README

CompareAndBitpack

English | 简体中文

Product Support Status

Product Supported
Ascend 950PR&950DT products √
Atlas A3 products √
Atlas A2 products √
Atlas 200I/500 A2 inference products √
Atlas inference products √
Atlas training products √

Function Description

  • Operator function: compares each element in the input tensor x with the threshold, and packs the comparison result (1 if greater, otherwise 0) into a bit-field output of the uint8 type. The comparison results of every 8 consecutive input elements are packed into one uint8 value.

Parameter Description

Parameter Name Input/Output/Attribute Description Data Type Data Format
x Input Input tensor to be compared, with at least 1 dimension. FLOAT, FLOAT16, DOUBLE, INT8, INT16, INT32, INT64, BOOL ND
threshold Input Scalar threshold used for comparison with the elements in x. FLOAT, FLOAT16, DOUBLE, INT8, INT16, INT32, INT64, BOOL ND
y Output Packed bit-field output tensor, whose innermost dimension is 1/8 of the innermost dimension of the input. UINT8 ND

Constraint Description

  • threshold must be a scalar
  • x must be a tensor with at least one dimension (vector or higher), and scalar inputs are not supported
  • The size of the innermost dimension (the last dimension) of x must be divisible by 8
  • The data types of x and threshold must be the same

Invocation Description

Invocation Method Sample Code Description
Graph-mode API test_geir_compare_and_bit_pack Invokes the CompareAndBitpack operator through the operator IR API.