MSP430-elf-objcopy disassembly issue

Krista Hill

Prodigy 70 points

Other Parts Discussed in Thread: MSP430G2231

Hi folks;

I'm practicing my MSP430 assembly language skills and came across something I'm not sure of.

In a program I tried the following to test the value in r6.

cmp r6,#0x00

When I use objdump with the -d option, it produces a simple disassembly.

The disassembly from objdump includes the following:

f816: 03 96 .word 0x9603; ????

In looking at references the instruction appears to be encoded correctly, also the program simulated correctly with msp430-elf-gdb. It's just that objdump was not able to properly decode the instruction. The processor is the MSP430G2231

The situation is strange as a few instructions later another compare instruction is properly handled

f81a: 06 95 cmp r5, r6

Please take a look into objdump. I inserted the program source below.
Much thanks in advance;

; main.S - ex06 - Dec. 2, 2015
         .file "main.S"
#include "asminc.h"

       .data
RamAdx: .fill   1

       .text
_start: mov.w   #WDTPW|WDTHOLD,&WDTCTL
     mov.w   #__stack,SP
         mov.w   #List,r4
       clr     r5
top:     mov.b   @r4,r6
         cmp     r6,#0x00
         jeq     done
       cmp     r5,r6
     jlo     pass
         mov     r6,r5
pass: inc     r4
         jmp     top

done: jmp     done

List:    .byte    0x12,0x24,0x18,0x43,0x16,0x00

         .section ".resetvec","ax",@progbits
         .word    _start

over 9 years ago

0 Krista Hill over 9 years ago

Prodigy 70 points

; asminc.h
#include <msp430.h>

#define PC r0
#define SP r1
#define SR r2
#define CG1 r2
#define CG2 r3

0 George Mock over 9 years ago in reply to Krista Hill

TI__Guru**** 249600 points

Thanks for reporting the problem. I notified the msp430 GCC tools experts at Red Hat.

Thanks and regards,

-George

0 DJ Delorie over 9 years ago

Intellectual 390 points

Try "cmp #0x00,r6" instead - the msp430 disassembler is very generic and doesn't support constants as the destination operand (i.e. "add r6,#0x00" uses the same logic). The examples in the TI documentation for CMP also list the constant first.

0 Krista Hill over 9 years ago in reply to George Mock

Prodigy 70 points

Thanks much;

Krista

0 Krista Hill over 9 years ago in reply to DJ Delorie

Prodigy 70 points

Thanks much for clarifying that point, I see the logic.

0 Krista Hill over 9 years ago in reply to Krista Hill

Prodigy 70 points

Just gave it a try, your fix does work.

Krista

0 Krista Hill over 9 years ago in reply to Krista Hill

Prodigy 70 points

Hold it for a moment, just looked at the instruction set again and see only 8 conditional jump instructions. How can that be? If you remove jump-equal, jump-not equal, and jump-always, that leaves only 5 other conditional jumps.

o There is jump-lower than, jump-higher-or-same, but no jump-higher, and no jump-lower-or-same
o There is a jump-negative, but no jump-positive
o There is jump if greater-or-equal, jump-less-than, but no jump-less-than-or-equal, and no jump greater-than

Of course, to achieve a complement conditional test, it is enough to swap the operands when performing the comparison.
With the above observation in mind, I would think that being able to use the second operand for constant values is a necessary part of the instruction set. I would think that compare is most certainly an important exception.

0 Archaeologist over 9 years ago in reply to Krista Hill

TI__Guru* 84285 points

It's not strictly necessary. Typically you just adjust the constant:

    CMP #1, R12
    JGE case_not_le_zero
    MOV #2, R13 ; if (R12 <= 0) ...
case_not_le_zero:

Code Composer Studio™︎

Code Composer Studio forum

MSP430-elf-objcopy disassembly issue