The concurrent dynamic linking networked graph processor without the program counter

In the concurrent dynamic linking networked graph processor, the operators in the graph processor access the linking states on the switches network among the operators and the operands by themselves. They execute the allocated graphs computing and allocate the new graphs at the same time. The computing in addition to producing the new values to the operands perhaps also access the linking states on the switches network to automatically release the executed graphs no more needed or to change the linking for the next computing without the program counter assignment in the sequential processor after the executing is finished. This mechanism is quite useful to derive the formulas. Not like the equation of the traditional computer program processing in single direction by the stack is fixed and can not be changed ( even it is set as the non-protected mode to be changeable by itself at runtime, what it changes are the sequential instruction codes. ), the sense based graph processor changes the virtual wires directly. The graph processor does not use the instruction code decoder, and it directly matches the symbols linked to the senses to produce the shiftable switch states without the software compiler in the traditional computer. It is a little like the hardware compiler for the FPGA but is not, and the net bitmap producing is concurrently cooperated by the operators inside not outside.

By the new way, the precondition is the total volume of the memory cells for the switch states on the virtual wires must be less than the combinational gate cons of the original decoder. By the old way, the more operators or operands; the more gate levels and gate cons. The old way to address by the decoder decoding has a big problem for the expandability because of the address bandwidth. The new way only needs to add the memory cells and extend the bus for the symbol matching by the fixed bandwidth. In addition to that, using the address decoder only has few linkings at one time that lots of operators are idle; using the concurrent dynamic linking networked graph processor without the program counter has so many linkings at the same time.

Why is the mechanism which the linking states of the computing flow graph can be dynamically modified during runtime so important? Take an instance, the traditional way to input the values into the single directional equation to compute out the answer by the stack has a problem: there is once deviation each time input the value into the triangle, exponential, or logarithmic function. In order to greatly reduce the deviation, of course, it is necessary to replace the formula to become the least function terms before input the parameter values for the precision. Due to the known and unknown terms of a formula are not necessarily the same every time, it is necessary to modify the linking state among the operands and the operators dynamically during the computing. In the other words, the using of the numerical methods should be left to the last, it should be preceded by a good job handling symbolic links. In addition to this, with regard to the graph traveling problem, directly using the virtual wires as the linkages among the nodes can be changed during traveling is basically needed by many graph algorithms too. The graph processor is more suited and more direct to deal with such issues than the traditional numerical addressing stack processor at the hardware level.