4

I'm trying to cross=compile a simple program with inline assembly for a variety of ARM processors. It contains the assembly instruction QSUB which according to this document is both a valid ARM and Thumb-2 instruction.

This compiles fine for me when building for a cortex-a8.

I get the compile error in the title when I try building for a cortex-m3 and can't figure out why. I attempted to force thumb mode with the compiler switch -mthumb, and then the error changes to "selected processor does not support thumb mode." I'm a bit confused because cortex-m3 does in fact support thumb mode.

Here's the compile options I'm using with codesourcery's gcc:

arm-none-linux-gnueabi-gcc helloworld.c -o a.out -lm -mthumb -mcpu=cortex-m3
artless noise
  • 21,212
  • 6
  • 68
  • 105
Brandon Yates
  • 2,022
  • 3
  • 22
  • 33
  • 3
    Just something to try, sometimes gcc can be picky about option order, try setting the CPU type _before_ setting thumb mode. (`-mcpu=cortex-m3 -mthumb`) Possibly move both before even helloworld.c – Joachim Isaksson Mar 06 '12 at 18:54
  • 1
    thumb (not thumb2) is the most portable across the arm family. I cant find a concrete list from arm but it appears that the cortex-m4 is an ARMv7E-M and the cortex-m3 is an ARMv7-M. looking at a current ARMv7-M ARM, it shows the QSUB being ARMv7E-M specific not all ARMv7 cores. (older armv7-m arm's did show QSUB as simply an ARMv7-M instruction, very confusing). Note the Cortex-m0 and -m1 are armv6 based and have a very limited thumb2 instruction set... – old_timer Mar 07 '12 at 04:07

2 Answers2

4

In accordance with TI the best gcc flags for cortex-a8 are 

-march=armv7-a -mtune=cortex-a8 -mfpu=neon -ftree-vectorize -ffast-math -mfloat-abi=softfp

http://processors.wiki.ti.com/index.php/Cortex-A8

uta
  • 1,451
  • 10
  • 12
4

QSUB is not mentioned in ARM's "Cortex-M3 Devices Generic User Guide" (Document DUI0552A).

Additionally, TI's "Cortex-M3/M4F Instruction Set, TECHNICAL USER'S MANUAL" indicates that QSUB is not supported on the Cortex-M3.

I think you're just going to have to believe what GCC is telling you...

Michael Burr
  • 333,147
  • 50
  • 533
  • 760
  • Thank you, i just found this independently as well. I guess I'll have to find another way to go about my saturation arithemetic – Brandon Yates Mar 06 '12 at 19:52
  • 2
    @BrandonYates from the linked page: "For the ARMv7-M architecture, they are only available in an ARMv7E-M implementation". ARMv7E-M is implemented in Cortex-M4 (M3 is just ARMv7-M). Maybe check out if the CMSIS-DSP library does what you need; it supports both M3 and M4. – Igor Skochinsky Mar 07 '12 at 13:18