Compiler-driven dynamic memory allocation methodology for scratch-pad based embedded systems

An algorithm for performing dynamic memory allocation in a scratch-pad based embedded system, said algorithm comprising the steps of:

• partitioning an application program stored within the scratch-pad based embedded system into a plurality of regions;
• representing the plurality of regions with a data structure having a plurality of nodes, wherein each node represents one of a procedure, loop and variable of the application program;
• assigning a timestamp to a starting point of each procedure and loop node of the plurality of nodes to form a timestamp order;
• and
• determining a sequence of memory transfers at the starting point of each procedure and loop node between one type of memory and a scratch-pad memory of the embedded system in accordance with the timestamp order.

In an embedded system having a processor running a compiler and a memory means storing an application program, a method comprising the computer-implemented steps of:

• partitioning by the compiler the application program stored within the memory means into a plurality of regions;
• representing the plurality of regions with a data structure having a plurality of nodes, wherein each node represents one of a procedure, loop and variable of the application program;
• assigning a timestamp to a starting point of each procedure and loop node of the plurality of nodes to form a timestamp order;
• and
• determining a sequence of memory transfers at the starting point of each procedure and loop node between one type of memory and a scratch-pad memory of the embedded system in accordance with the timestamp order.

An embedded system comprising:

• means for partitioning an application program into a plurality of regions;
• means for representing the plurality of regions with a data structure having a plurality of nodes, wherein each node represents one of a procedure, loop and variable of the application program;
• means for assigning a timestamp to a starting point of each procedure and loop node of the plurality of nodes to form a timestamp order;
• and
• means for determining a sequence of memory transfers at the starting point of each procedure and loop node between one type of memory and a scratch-pad memory of the embedded system in accordance with the timestamp order.