Email: Password: Remember Me | Create Account (Free)

Back to Subject List

Old thread has been locked -- no new posts accepted in this thread
???
03/06/09 17:47
Read: times


 		 #163175 - 67% of your loop was your loop
Responding to: ???'s previous message
When using the simulator, it may be better to look at the cycle count instead of the time figure.

You did 257 iterations in 67us. That is 260ns / iteration.
Your clock speed was 58MHz (17ns/tick) so 260ns represents 15 clock cycles/iteration.

The load instruction will take 3 clock cycles. The store instruction will take 2 clock cycles. An unrolled loop would end up with 5 clock cycles/byte, resulting or 22us at 58MHz (86ns/byte) or 17.8us st 72MHz (max speed of LPC23xx) or 69ns/byte. Some ARM chips may be slower in case the memory subsystem requires extra wait cycles. That is the reason why really fast processors adds one or more levels of data and code caches - standard big-size memory can't keep up at higher radio frequencies.

In this case, your loop seem to have spent 10 clock cycles just to figure out what to do. I'm not sure why the compiler didn't do a better loop - did you use optimization? I don't have access to the ARM compiler right now, so I can't play with it.

List of 76 messages in thread
TopicAuthorDate
NXP P89LPC936 Auxilary RAM            01/01/70 00:00      
   Compiler Cannot Save the Day At Runtime            01/01/70 00:00      
      I found it.....            01/01/70 00:00      
         Incorrect!            01/01/70 00:00      
            To be fair,...            01/01/70 00:00      
               Its good info to know...            01/01/70 00:00      
   how is it done in C?            01/01/70 00:00      
   Maybe you should try doing it ASM!            01/01/70 00:00      
      You should start a new thread on that!            01/01/70 00:00      
      Not in this case!            01/01/70 00:00      
         A compiler should support ALL MCU features ...            01/01/70 00:00      
            A compiler should translate the language            01/01/70 00:00      
               Still C - just not strict.            01/01/70 00:00      
            I didn't say that!            01/01/70 00:00      
               It is not that clean and clear            01/01/70 00:00      
                  Deviants are deviant            01/01/70 00:00      
                     Keil and SDCC need no macros ...            01/01/70 00:00      
                        Are you sure?            01/01/70 00:00      
                     That's probably correct, but ...            01/01/70 00:00      
                        A C compiler can map the hardware quite well            01/01/70 00:00      
                           compiler vendors are looking at the new processors            01/01/70 00:00      
                              I once mentioned that ...            01/01/70 00:00      
                           It's not what I'd choose, but it is a matter of perefernce            01/01/70 00:00      
                              As flash gets bigger, the code steps do too.            01/01/70 00:00      
                              Still not needed for other architectures            01/01/70 00:00      
                                 We will have to agree to disagree ... I guess            01/01/70 00:00      
                                    Which single-clocker is cheaper than an ARM?            01/01/70 00:00      
                                       RE: Which single-clocker is cheaper than an ARM?            01/01/70 00:00      
                                       differs considerably from the classic microcontroller?            01/01/70 00:00      
                                       Horses for courses            01/01/70 00:00      
                                          Always start each project by scanning the market            01/01/70 00:00      
                                             On that we can agree            01/01/70 00:00      
                                          Maybe, but what are they comparing?            01/01/70 00:00      
                                             Did you actually look?            01/01/70 00:00      
                                                Yes, I did.            01/01/70 00:00      
                                       it's a tradeoff            01/01/70 00:00      
                                          Is it 2006 already?            01/01/70 00:00      
                                             Really?            01/01/70 00:00      
                                                that's outright silly            01/01/70 00:00      
                                                   Maybe ... which is why it is not yet the case            01/01/70 00:00      
                                                      the eyes of the beholder            01/01/70 00:00      
                                                         Look at it from another viewpoint for a moment            01/01/70 00:00      
                                          RE: I'm not the one to ask about IC prices            01/01/70 00:00      
                                             doesn't mean I'm totally in the dark            01/01/70 00:00      
                                                Richard Erlacher has left the planet            01/01/70 00:00      
                                                   Maybe Andy's the one who's lost            01/01/70 00:00      
                                                      I cannot believe you even looked at ARM            01/01/70 00:00      
                                                         It's all relative            01/01/70 00:00      
                                                            Price doesn't directly follow processor size            01/01/70 00:00      
                                                            What about performance?            01/01/70 00:00      
                                                         Be specific please.            01/01/70 00:00      
                                                            You can use your own supplier            01/01/70 00:00      
                                                               Your test simulates as 41.04us            01/01/70 00:00      
                                                                  RE: ARM compiles do not like byte-addressing            01/01/70 00:00      
                                                                     A typo on my part            01/01/70 00:00      
                                                                        Case of full unroll            01/01/70 00:00      
                                                                           ... and what does THAT ARM MCU cost?            01/01/70 00:00      
                                                                              WHO CARES            01/01/70 00:00      
                                                                                 Absolutely true!            01/01/70 00:00      
                                                                              $5 or lower            01/01/70 00:00      
                                                                                 My Simulation times were wrong.            01/01/70 00:00      
                                                                                    67% of your loop was your loop            01/01/70 00:00      
                                                                                    re-written the loop in C            01/01/70 00:00      
                                                                                       Try without loop counter            01/01/70 00:00      
                                                                                          You're right!            01/01/70 00:00      
                                                                                             Similar trick with ARM7 would require 66.67MHz            01/01/70 00:00      
                                                                                             60ns with no store instruction.            01/01/70 00:00      
                                                                                                25% speedup            01/01/70 00:00      
                                                                                                   Actually, I found the for loop faster.            01/01/70 00:00      
                                                                                                      Compiler setting?            01/01/70 00:00      
                                                                                                         I made NO optimisation            01/01/70 00:00      
                                                                                                            Avoid variable aliasing if you like high optimization levels            01/01/70 00:00      
                                                                                                               The RealView compiler is very competent            01/01/70 00:00      
                                                                                                                  Yes, caching            01/01/70 00:00      
                                                                                                                  Yes, caching            01/01/70 00:00      
                                                                                                It's much easier with the 8-bit single-clocker            01/01/70 00:00      

Back to Subject List