System and method for high throughput in multiple computations

Abstract

Device, circuit and method are configured to enhance throughout of processing of vast amount of data such as video stream. In some embodiment frequently used data blocks are stored in a fast RAM of the processor. In another embodiment received stream of data is divided to plurality of data portions and is streamed concurrently to streaming multiprocessors of a graphic processing unit (GPU) and is processed concurrently before the entire stream is loaded.

1 . A method for enhancing graphical data throughput exchanged between graphical data source and a graphical processing unit (GPU) via a streaming multiprocessor unit that comprises a processing core unit (PCU), a register file unit, multiple cache units, shared memory unit, unified cache unit and interface cache unit, the method comprising:
 transferring stream of graphical data via interface cache unit and via the multiple cache units and via the unified cache unit to the register file unit;   transferring a second stream of graphical data from the register file unit to the processing core unit (PCU); and   storing and receiving frequently used portions of data in shared memory unit, via register file unit.   
     
     
         2 . The method of  claim 1  wherein the register file unit is configured to direct data processed by the PCU to the shared memory unit as long as it is capable of receiving more data, based on the level of frequent use of that data. 
     
     
         3 . The method of  claim 2  wherein the level of frequent use is determined by the PCU. 
     
     
         4 . A streaming multiprocessor unit for enhancing graphical data throughput comprising:
 a processing core unit (PCU) configured to process graphical data;   a register file unit, configured to provide graphical data from the PCU and to receive and temporary store processed graphical data from the PCU;   multiple cache units, configured to provide graphical data from the register file unit and to receive and temporary store processed graphical data from the register file unit;   shared memory unit configured to provide graphical data from the register file unit and to receive and temporary store processed graphical data from the register file unit;   unified cache unit configured to provide graphical data from the register file unit and to receive and temporary store processed graphical data from the register file unit; and   interface cache unit, configured to receive graphical data for graphical processing at high pace, to provide the graphical data to at least one of shared memory unit and unified cache unit, to receive processed graphical data from the unified cache unit, and to provide the processed graphical data to external processing units.   
     
     
         5 . The streaming multiprocessor unit of  claim 4  wherein at least some of the graphical data elements are stored, before and/or after processing by the PCU in the shared memory unit, based on a priority figure that is associated with the probability of their close call by the PCU. 
     
     
         6 . The streaming multiprocessor unit of  claim 5  wherein the priority figure is higher as the probability is higher.