TABLE OF CONTENTS Chapters 6 8 explored software tools that are essential to the application programmer. This chapter returns to architectural concepts important to application programming. Efficient usage of the Itanium stacked register file and main memory resources, especially the stack, is critical to producing high-performance software. In this chapter, we examine how Itanium software utilizes these resources to minimize usage of main memory.
The Itanium architecture has a load/store instruction set. Because memory references have a longer latency than register instructions, they create stalls in the hardware pipelines. Therefore, memory references should be avoided whenever possible.
A large percentage of memory reference instructions are related to function calls. Calls require that arguments be passed, registers saved, return addresses be placed on the user stack and, later, these parameters restored from the stack on the return from a function. All of these actions generate a large amount of memory traffic. In other words, modular programming makes extensive demands of system resources.
If there were a way of eliminating, or at least minimizing, the number of main memory accesses due to function calls, application program efficiency would greatly benefit. The Itanium processor uses an ingenious design to eliminate much of this memory traffic. Its large general register file permits implementation of an on-chip stack. The Itanium processor allocates up to 96 registers for use as a register stack. Accessing information in this on-chip stack greatly reduces the overhead of the function call process.
This method makes use of the...
