160 likes | 170 Views
Interesting Points of the SPARC Processor. Saunders Roesser Summer 2002 CS-585-2. Outline. Quick History Register Windowing Processor Traps Memory Model Conclusion. Quick History. Processor Design based off of the RISC I and II designs as the University of California, Berkeley
E N D
Interesting Points of the SPARC Processor Saunders Roesser Summer 2002 CS-585-2
Outline • Quick History • Register Windowing • Processor Traps • Memory Model • Conclusion
Quick History • Processor Design based off of the RISC I and II designs as the University of California, Berkeley • First SPARC Processor appeared in 1987 in a SUN MS SUN-4 Computer • 1989 – SPARC International Organization Setup
Register Windowing • Technique for handling register management • Only a small amount of the total physical register are available at any one time. • Virtual Window slides up and down the register stack to show groups of registers.
Register Windowing • Two types of registers • General Purpose Registers • Control/Status Registers • Each Processor Part has it’s own set of registers: • Instruction Unit (IU) • Co-Processor (CP) • Floating Point Unit (FPU)
Instruction Unit • May have between 40-520 Registers • 8 Global registers • Divide the rest into Sets • Set contains: • 8 local registers • 8 in registers • Window Contains: • 1 Set (8 local and 8 in) • Plus 8 Out Registers which equal the 8 In Registers of the Next Window
Traps • Error Handling Mechanism in the SPARC architecture • Handle errors, such as overflow, I/O, and instruction errors.
Handling Traps • Current Window Point is decremented • Information on the trap is stored in the local registers of the window.
Types of Traps • Precise trap • Error occurs before the instruction is executed • Deferred Trap • Error occurs after the instruction has executed • Can occur after several more instruction have occurred • Interrupting Traps • Error occurred from an event, such as I/O
Memory Model • Total Store Ordering • Store, Flush, and Load happen in order they were issued • Partial Store Ordering • Issuing order can happen out of order to increase speed. • Hardware Implementation dependent • As long as it appears correct to software • Same model for one processor or mutiprocessor
Memory • Memory is stored in: • Half Word – 2 byte boundaries • Word – 4 byte boundaries • Double Word – 8 byte boundaries • Big Endian Architecture
Conclusion • SPARC has been the most successful implementation of a RISC architecture • Closest competitors have 10 years less experience • Processor is hardwired • PC is read by the CALL or JMPL instructions, not by numbers.