Intel PXA270, Руководство по оптимизации

Страница: 24 / 144

2-4 Intel® PXA27x Processor Family Optimization Guide
Microarchitecture Overview
2.2.3.2 Instruction Decode (ID) Pipestage
The ID pipestage accepts an instruction word from the IFU and sends register decode information
to the RF pipestage. The ID is able to accept a new instruction word from the IFU on every clock
cycle in which there is no stall. The ID pipestage is responsible for:
•
General instruction decoding (extracting the opcode, operand addresses, destination addresses
and the offset).
•
Detecting undefined instructions and generating an exception.
•
Dynamic expansion of complex instructions into sequence of simple instructions. Complex
instructions are defined as ones that take more than one clock cycle to issue, such as LDM,
STM, and SWP.
2.2.3.3 Register File / Shifter (RF) Pipestage
The main function of the RF pipestage is to read and write to the register file unit (RFU). It
provides source data to:
•
X1 for ALU operations
•
MAC for multiply operations
•
Data cache for memory writes
•
Coprocessor interface
The ID unit decodes the instruction and specifies the registers accessed in the RFU. Based on this
information, the RFU determines if it needs to stall the pipeline due to a register dependency. A
register dependency occurs when a previous instruction is about to modify a register value that has
not been returned to the RFU and the current instruction needs to access that same register. If no
dependencies exist, the RFU selects the appropriate data from the register file and passes it to the
next pipestage. When a register dependency does exist, the RFU keeps track of the unavailable
register. The RFU stops stalling the pipe when the result is returned.
The ARM* architecture specifies one of the operands for data processing instructions as the shifter
operand. A 32-bit shift can be performed on a value before it is used as an input to the ALU. This
shifter is located in the second half of the RF pipestage.
2.2.3.4 Execute (X1) Pipestages
The X1 pipestage performs these functions:
•
ALU calculations – the ALU performs arithmetic and logic operations, as required for data
processing instructions and load/store index calculations.
•
Determine conditional instruction executions – the instruction’s condition is compared to the
CPSR prior to execution of each instruction. Any instruction with a false condition is
cancelled and does not cause any architectural state changes, including modifications of
registers, memory, and PSR.
•
Branch target determinations – the X1 pipestage flushes all instructions in the previous
pipestages and sends the branch target address to the BTB if a branch is mispredicted by the
BTB. The flushing of these instructions restarts the pipeline.