Matrix processing apparatus
Abstract
Methods, systems, and apparatus, including a system for transforming sparse elements to a dense matrix. The system is configured to receive a request for an output matrix based on sparse elements including sparse elements associated with a first dense matrix and sparse elements associated with a second dense matrix; obtain the sparse elements associated with the first dense matrix fetched by a first group of sparse element access units; obtain the sparse elements associated with the second dense matrix fetched by a second group of sparse element access units; and transform the sparse elements associated with the first dense matrix and the sparse elements associated with the second dense matrix to generate the output dense matrix that includes the sparse elements associated with the first dense matrix and the sparse elements associated with the second dense matrix.
Claims
exact text as granted — not AI-modifiedWhat is claimed is:
1 . (canceled)
2 . A system for transforming data elements into a matrix, the system comprising:
a plurality of data element access units, each data element access unit of the plurality of data element access units comprising a corresponding plurality of processors, and each data element access unit being configured to:
receive respective control signals;
access, based on the respective control signals, multiple data elements of a plurality of data elements corresponding to the data element access unit;
generate an output matrix based on the multiple data element obtained from the plurality of data elements corresponding to the data element access unit; and
output the output matrix to another component of the system.
3 . The system of claim 2 , further comprising:
a sparse-dense transform unit configured to receive instructions corresponding to the respective control signals; and wherein the plurality of data element access units are located in the sparse-dense transform unit.
4 . The system of claim 3 , wherein the sparse-dense transform unit is a multi-dimensional sparse-dense transform unit comprising a row dimension and a column dimension.
5 . The system of claim 4 , wherein the plurality of data element access units are arranged along respective dimensions of the multi-dimensional sparse-dense transform unit.
6 . The system of claim 3 , wherein each data element access unit includes a respective first unit configured to apply a transformation to the multiple data elements to generate the output matrix.
7 . The system of claim 6 , wherein:
each first unit is a concatenation unit; and the transformation is based on a concatenation operation.
8 . The system of claim 6 , wherein:
each first unit is a compress/decompress unit; and the transformation is based on an operation to compress the multiple data elements.
9 . A method for transforming elements into a matrix using a system comprising a plurality of data element access units, each data element access unit of the plurality of data element access units comprising a corresponding plurality of processors, the method comprising:
receiving, by a first data element access unit, a control signal; based on the control signal, accessing, by the first data element access unit, multiple data elements of a plurality of data elements corresponding to the first data element access unit; generating an output matrix based on the multiple data elements obtained from the plurality of data elements corresponding to the first data element access unit; and providing the output matrix to another component of the system.
10 . The method of claim 9 , wherein:
the system includes a sparse-dense transform unit configured to receive instructions; the plurality of data element access units are located in the sparse-dense transform unit; and the method comprises, receiving, by the sparse-dense transform unit, an instruction corresponding to the control signal for the first data element access unit.
11 . The method of claim 10 , wherein the sparse-dense transform unit is a multi-dimensional sparse-dense transform unit comprising a row dimension and a column dimension.
12 . The method of claim 11 , wherein the plurality of data element access units are arranged along respective dimensions of the multi-dimensional sparse-dense transform unit.
13 . The method of claim 10 , wherein:
each of the plurality of data element access units includes a respective first unit configured to transform data elements; and the method comprises, applying, by the respective first unit of the first data element access unit, a transformation to the multiple data elements to generate the output matrix.
14 . The method of claim 13 , wherein:
each first unit is a concatenation unit; and applying the transformation comprises:
applying a concatenation operation to the multiple data elements; and
concatenating the multiple data elements based on the concatenation operation to generate the output matrix.
15 . The method of claim 13 , wherein:
each first unit is a compress/decompress unit; and applying the transformation comprises:
applying a compress operation to the multiple data elements; and
compressing the multiple data elements based on the compress operation to generate the output matrix.
16 . A non-transitory machine-readable storage device comprising instructions that, when executed by a data processing apparatus, cause the data processing apparatus to perform operations comprising:
receiving, by a first data element access unit, a control signal, wherein the first data element access unit comprises a plurality of processors; based on the control signals, accessing, by the first data element access unit, multiple data elements of a plurality of data elements corresponding to the first data element access unit; generating an output matrix based on the multiple data elements obtained from the plurality of data elements corresponding to the first data element access unit; and providing the output matrix to another component of the data processing apparatus.Join the waitlist — get patent alerts
Track US2022391472A1 — get alerts on status changes and closely related new filings.
We store only your email — no account needed. See our privacy policy.