초록 ▼

Techniques for adding more complex instructions and their attendant multi-cycle execution units with a single instruction multiple data, stream (SIMD) very long instruction word (VLIW) processing framework are described. In one aspect, an initiation mechanism also acts as a resynchronization mechani

Techniques for adding more complex instructions and their attendant multi-cycle execution units with a single instruction multiple data, stream (SIMD) very long instruction word (VLIW) processing framework are described. In one aspect, an initiation mechanism also acts as a resynchronization mechanism to read the results of multi-cycle execution. This multi-purpose mechanism operates with a short instruction word (SIW) issue of the multi-cycle instruction, in a sequence processor (SP) alone, with a VLIW, and across all processing elements (PEs) individually or as an array of PEs. A number of advantageous floating point instructions are also described.

대표청구항 ▼

Techniques for adding more complex instructions and their attendant multi-cycle execution units with a single instruction multiple data, stream (SIMD) very long instruction word (VLIW) processing framework are described. In one aspect, an initiation mechanism also acts as a resynchronization mechani

Techniques for adding more complex instructions and their attendant multi-cycle execution units with a single instruction multiple data, stream (SIMD) very long instruction word (VLIW) processing framework are described. In one aspect, an initiation mechanism also acts as a resynchronization mechanism to read the results of multi-cycle execution. This multi-purpose mechanism operates with a short instruction word (SIW) issue of the multi-cycle instruction, in a sequence processor (SP) alone, with a VLIW, and across all processing elements (PEs) individually or as an array of PEs. A number of advantageous floating point instructions are also described. t device. 7. The method of claim 1 in which said method is performed by a plurality of host systems. 8. The method of claim 7, in which said performance measure generated for each device is shared among host systems. 9. The method of claim 7, in which a token is used to serialize device address reassignments. 10. The method of claim 7, in which a time stamp is used to prevent multiple reassignments of any one of said device addresses in rapid succession. 11. In an information handling system in which one or more central processing units (CPUs) access one of a plurality of input/output (I/O) devices via one of a plurality of device addresses, and in which I/O requests for a particular device are placed in an I/O queue for that device, each of said device addresses being assigned to at-most one of said devices, and in which system workload is organized into two or more service classes, each of said service classes having one or more processing goals, and in which each of said service classes is assigned an importance index relative to others of said two or more service classes, a method for dynamically managing the assignment of said device addresses to said devices, said method comprising the steps of: generating for each of said devices a performance measure indicating how well requests for that device are being handled; associating with each one of said devices the service classes using said one device; determining that one of said service classes is failing to meet one or more performance goals; determining the positive effect on the performance of said one service class of reassigning one or more device addresses to a receiver device used by said one service class; identifying a one or more donor devices used by service classes of lesser relative importance that said one service class; issuing in one of said one or more central processing units a command to a second processor, instructing said second processor to reassign one or more of said donor addresses to the said receiver device. 12. The method of claim 11 in which the device addresses reassigned to the selected receiver device are not currently assigned to any of said devices. 13. The method of claim 11 in which the device addresses reassigned to the selected receiver device are currently assigned to one or more of said devices. 14. The method of claim 11 in which the performance measure for each device is based on the number of requests in the queue for that device. 15. The method of claim 11 in which said method is performed by a plurality of host systems. 16. The method of claim 15, in which said performance measure generated for each device is shared among host systems. 17. The method of claim 15, in which a token is used to serialize device address reassignments. 18. The method of claim 15, in which a time stamp is used to prevent multiple reassignments of any one of said device addresses in rapid succession. 19. In an information handling system in which a plurality of central processing units (CPUs) access one of a plurality of input/output (I/O) devices via one of a plurality of device addresses, and in which I/O requests for a particular device are placed in an I/O queue for that device, each of said device addresses being assigned to at most one of said devices, and in which is available a plurality of assignment methods for dynamically managing the assignment of said device addresses, and in which a plurality of said assignment methods operate concurrently, a method for contention management among said assignment methods, said contention management method comprising the steps of: designating one of said assignment methods as the dominant assignment method; designating all other of said assignment methods as subordinate assignment methods; operating said dominant assignment method more frequently than any of said subordinate methods; preventing a subordinate assignment method from reassigning a device address, where said device address was r