AN910 Datasheet, PDF (26/51 Pages) STMicroelectronics – ST7 AND ST9 PERFORMANCE BENCHMARKING

ST7 AND ST9 PERFORMANCE BENCHMARKING
7.2 REMARKS ON SOME PARAMETERS
7.2.1 Instruction processing
Only two different instruction processings exist:
• standard processing: current instruction is completely processed before next one is fetched
• prefetch mechanism: some next opcodes are prefetched as current instruction is processed
The prefetch mechanism is best described as a queue rather than as a pipeline. Queue
logic fetches program information and positions it for execution, but instructions are executed
sequentially. A typical pipelined CPU executes more than one instruction at the same time.
The queue size is given, but performance is not precised because no value is given by
databooks. Nevertheless, general statistics on instruction processing mechanisms give an
usual average 20%-25% gain for one stage, and this gain is not more than 25%-30% for two
stages. Additional stages without complex mechanisms do not give higher gain. Anyway, the
instruction processing mechanism has a leading role in general performance.
7.2.2 Average ICT/CPI and IL
The average ICT (Instruction Cycle Time) is a currently used parameter. But it is linked to the
frequency f, then we prefer the average CPI (Clock Per Instruction) to describe the
instruction set. On the other hand, to compare MCU core performance, the frequency has
to be considered, and so the average ICT is used in result analysis (section 4). Charts with
ICT and IL ranges are presented at the end of this section (see <Italic>7.4 Instruction Cycle
Time chart<Italic end> and <Italic>7.5 Instruction Length chart<Italic end>).
Remark that the average ICT (in µs) is the inverse of the MIPS parameter (Million Instruction
Per Second), and so we have the formula:
f
1
MIPS =
=
CPI ICT
(f is in MHz and ICT is in µs)
The average ICT/CPI and average IL have been calculated considering all available
instructions and all possible addressing modes, favouring mostly used ones in the test
routines. Ranges are presented instead of decimal values, to take the subjectivity of the
calculation into account. Thus the values can be considered as reliable.
26/51