Hi experts!
I want to use ldr/str or ldm/stm to copy memory not aligned 4bytes.
I know their input address should be aligned by 4 bytes.
but is there any solution to use ldr/str or ldm/stm though src or dst isn't aliged 4byte by modifing following code?
void wordcopy(unsigned char *dst, unsigned char *src, int num) {loop: LDR r3,[r1]; STR r3,[r0]; SUBS r2,r2,#1; BNE loop; BX lr;}
Thanks Martin, jensbauer, yasuhikokoumoto for your kind response!
It's very helpful!
To enhance Martin's reply slightly:
In the very beginning of your routine, make a check on the lowest 2 bits on both source and destination.
If the bits match on both source and destination, use a 32-bit copy with a ramp-up.
If bit 0 on source differs from bit 0 on destination, you may benefit from byte-copy; you also have the option to use shifts, but the code will be large.
If bit 1 on source differs from bit 1 on destination, you may benefit from halfword-copy, but the penalty for copying misaligned halfwords is not so bad, so you may want to do a full 32-bit copy there.
Hint: Use the EOR instruction followed by a LSLS#30 to check if the low two bits match.
Hello,
I hope the __rt_memcpy function of the following link will help you.
http://research.microsoft.com/en-us/um/redmond/projects/invisible/src/crt/md/arm/_memfunc.s.htm
Best Regards,
Yasuhiko Koumoto.
LDM/STM only support aligned addresses.
LDR/STR can be used with unaligned addresses, as long as the target addresses are marked as Normal and that alignment checking isn't enabled.
You didn't mention which processor/architecture you are targeting. For A class, the translation tables that define whether an address is Normal or Device. For R and M class, it's the MPU (if present).
The other option is to test in the passed in pointers. Do a number of byte load/stores until you get an aligned address, then do LDM/STM, and finish with more byte load/stores if needed.
EDIT: Typo
View all questions in Cortex-A / A-Profile forum