Intel PXA270, Optimization Manual

Page: 63 / 144

Intel® PXA27x Processor Family Optimization Guide 4-15
Intel XScale® Microarchitecture & Intel® Wireless MMX™ Technology Optimization
4.3.1.6 Scheduling Data-Processing
Most Intel XScale® Microarchitecture data-processing instructions have a result latency of one
cycle. This means that the current instruction uses the result from the previous data processing
instruction. However, the result latency is two cycles if the current instruction uses the result of the
previous data processing instruction for a shift by immediate. As a result, this code segment would
incur a one-cycle stall for the MOV instruction:
sub r6, r7, r8
add r1, r2, r3
mov r4, r1, LSL #2
This code removes the one-cycle stall:
add r1, r2, r3
sub r6, r7, r8
mov r4, r1, LSL #2
All data processing instructions incur a two-cycle issue penalty and a two-cycle result penalty
when the shifter operand is shifted/rotated by a register or the shifter operand is a register. The next
instruction incur a two-cycle issue penalty and there is no way to avoid such a stall except by
rewriting the assembler instruction. The subtract instruction incurs a one-cycle stall due to the issue
latency of the add instruction as the shifter operand is shifted by a register.
mov r3, #10
mul r4, r2, r3
add r5, r6, r2, LSL r3
sub r7, r8, r2
The issue latency can be avoided by changing the code as:
mov r3, #10
mul r4, r2, r3
add r5, r6, r2, LSL #10
sub r7, r8, r2
4.3.1.7 Scheduling Multiply Instructions
Multiply instructions can cause pipeline stalls due to resource conflicts or result latencies. This
code segment incurs a stall of 0-3 cycles depending on the values in registers R1, R2, R4 and R5
due to resource conflicts:
mul r0, r1, r2
mul r3, r4, r5