User manual

3.5 The 16-BIS/32-BIS Concept
3.5.1 16-BIS/32-BIS Advantages
www.ti.com The 16-BIS/32-BIS Concept
The key idea behind 16-BIS is that of a super-reduced instruction set. Essentially, the ARM926EJ
processor has two instruction sets:
• ARM mode or 32-BIS: the standard 32-bit instruction set
• Thumb mode or 16-BIS: a 16-bit instruction set
The 16-bit instruction length (16-BIS) allows the 16-BIS to approach twice the density of standard 32-BIS
code while retaining most of the 32-BIS’s performance advantage over a traditional 16-bit processor using
16-bit registers. This is possible because 16-BIS code operates on the same 32-bit register set as 32-BIS
code. 16-bit code can provide up to 65% of the code size of the 32-bit code and 160% of the performance
of an equivalent 32-BIS processor connected to a 16-bit memory system.
16-bit instructions operate with the standard 32-bit register configuration, allowing excellent
inter-operability between 32-BIS and 16-BIS states. Each 16-bit instruction has a corresponding 32-bit
instruction with the same effect on the processor model. The major advantage of a 32-bit architecture over
a 16-bit architecture is its ability to manipulate 32-bit integers with single instructions, and to address a
large address space efficiently. When processing 32-bit data, a 16-bit architecture takes at least two
instructions to perform the same task as a single 32-bit instruction. However, not all of the code in a
program processes 32-bit data (for example, code that performs character string handling), and some
instructions (like branches) do not process any data at all. If a 16-bit architecture only has 16-bit
instructions, and a 32-bit architecture only has 32-bit instructions, then the 16-bit architecture has better
code density overall, and has better than one half of the performance of the 32-bit architecture. Clearly,
32-bit performance comes at the cost of code density. The 16-bit instruction breaks this constraint by
implementing a 16-bit instruction length on a 32-bit architecture, making the processing of 32-bit data
efficient with compact instruction coding. This provides far better performance than a 16-bit architecture,
with better code density than a 32-bit architecture. The 16-BIS also has a major advantage over other
32-bit architectures with 16-bit instructions. The advantage is the ability to switch back to full 32-bit code
and execute at full speed. Thus, critical loops for applications such as fast interrupts and DSP algorithms
can be coded using the full 32-BIS and linked with 16-BIS code. The overhead of switching from 16-bit
code to 32-bit code is folded into sub-routine entry time. Various portions of a system can be optimized for
speed or for code density by switching between 16-BIS and 32-BIS execution, as appropriate.
SPRUEP9A – May 2008 ARM Core 23
Submit Documentation Feedback