mikeash.com pyblog/friday-qa-2011-12-30-disassembling-the-assembly-part-3-arm-edition.html comments

wasim - 2016-02-27 09:03:50

Sat, 27 Feb 2016 09:03:50 GMT

very useful and helpful.

lldong - 2014-07-21 15:59:43

Mon, 21 Jul 2014 15:59:43 GMT

Why apple choose mov r7, r7 as a marker. Can r7 be replaced by other register?

Matt Galloway - 2012-02-07 17:44:37

Tue, 07 Feb 2012 17:44:37 GMT



mov r4, r0

mov r0, r4

I stumbled across this also whilst trying to work out why the compiler was doing something odd - http://stackoverflow.com/questions/9151028/why-do-these-simple-methods-compile-differently/

The optimiser clearly is doing strange things and quite odd that it didn't manage to clean up the redundant mov. I guess these things happen though.

You should file a bug report for it as well.

Gwynne Raskind - 2012-01-02 23:49:18

Mon, 02 Jan 2012 23:49:18 GMT

Steve: Sorry, I forgot to mention that you'll have to change that line to #import <Foundation/Foundation.h> to build for armv7.

Steve Nicholson - 2012-01-02 22:19:17

Mon, 02 Jan 2012 22:19:17 GMT

When I try to use your command line to compile test.m from Part 1, it bombs immediately with test.m:3:13: fatal error: 'Cocoa/Cocoa.h' file not found. Which seems to make sense since iOS programs don't use Cocoa.h. Are you sure you used the identical test.m?

Owen Shepherd - 2012-01-01 13:02:51

Sun, 01 Jan 2012 13:02:51 GMT

Passing floating point values in the integer registers is actually less efficient. However, GCC has historically had no support for the VFP version of AAPCS - that is, it always uses the version of the ABI designed for platforms without an FPU.

In any case, calls to NSLog and other variadic functions will always use the base version of AAPCS, as stated in the specification.

Also: In my experience (I've done kernel level work on ARM), the ARMv7 Architecture reference manual (Or, as I like to say, the ARM ARM) contains pretty much everything that the 6 x86 PDFs do. The SystemV x86_64 ABI (as used by everything but Windows and EFI) is specified in a separate document found on x86-64.org, so is really no different, and pretty much all the ARM floating point and vector operations are specified inside the ARM ARM.

Vladimir Pouzanov - 2011-12-31 11:38:00

Sat, 31 Dec 2011 11:38:00 GMT

Nice article, thanks!

Ken Thomases - 2011-12-31 00:43:42

Sat, 31 Dec 2011 00:43:42 GMT

Enjoyed the article. Minor error: in the discussion of MyFPFunction, it's loading 0.5, not 5.0.