850428 - strh instruction

850428 - strh instruction

• Proram

  a very good sample code u gave, thx

  and about what u said, alex, let's don't make things more complicated than what they really are! when the writer may use some words exactly in place of others, like "horrible" in place of "awful", it must be an option for the reader not to encounter this word in the book because of any reason like a horrible background it may have about this word. i didn't mean to omit such an essential word! using "the" is inevitable. i just need the compiler not to decide for me in using ldr, ldrh, str and strh when it may use ldrb and strb in place of them.

  when u write a code including a packed struct with a char following with an __int64 member the compiler generates 8 ldrb's when u access the second member as far as it "sees" that it's the member of this struct. so it must not be a weird expectation that the compiler do the same in all cases with responsibility of the programmer himself/herself in not having the same performance.

  how can i ask developers of the embedded visual C++ read this text

  thx

  
  

• Daniel_Bowen

  Yes, I see, but then even the regular Win32 compiler does not do this. On the x86 platform (Win32) if you attempt reading from an improperly aligned address, you get an alignment fault handled for you internally (which indeed slows things down a bit). On Win16 platform it would have caused the application to crash. Since addresses are usually dynamic (other than in trivial test scenarios), how do you propose compiler deal with it Test the address alignment in runtime This would impose a penalty since usually people do not read unaligned data without a dire need
  
  

• Jason Holding

  I'm not sure how do you imagine your code compiled without the use of LDR, but even if you do, you are out of luck. Compiler is not *that* flexible. Besides wouldn't it be kind of like asking a writer to produce a novel without the word "the" explaining it by your personal dislike of the word.

  What is the alingment exception you are referring to

  
  
  

• eslam nader

  Could you possibly include a code snippet that demonstrates the problem as you see it I am having a hard time visualizing it - it'd be easier to speak about something I can "touch".
  
  

• Manash

  He's referring to the fact that 16- and 32-bit values on ARM devices must be aligned on 2- and 4-byte boundries. If you have a 1-byte-aligned struct, or point an int pointer into an array of bytes, you get an exception as soon as you try to refer to the value, if it's not aligned.
  
  char foo[4] = {1, 0, 0, 0};
  int* bar = foo;
  *bar++; // exception if memory location of foo isn't evenly divisible by 4
  
  He wants the compiler to load ints as multiple bytes, instead of just wrapping an #ifdef around it and doing it manually.
  
  char foo[4] = {1, 0, 0, 0};
  #if defined (ARM)
  int baz = foo[0] | (foo[1] << 8) | (foo[2] << 16) | (foo[3] << 24);
  int* bar = &baz;
  #else
  int* bar = foo;
  #endif
  *bar++;
  
  Of course, the MS compiler doesn't properly pack to the byte even when you set packing to 1 byte anyway, but that's a completely different gripe.
  
  
• 850428 - strh instruction