This discussion has been locked.
You can no longer post new replies to this discussion. If you have a question you can start a new discussion

armcc problem?

Note: This was originally posted on 20th April 2009 at http://forums.arm.com

Hello,

We're using armcc here version:
  ARM/Thumb C/C++ Compiler, RVCT3.1 [Build 739]

And it seems to have an issue. Here's some simple c-code:

  main () {
    volatile unsigned int tmp32;

    printf ("GPIO: 1_reset: before\n");

    // touch tb_gpio
    *(unsigned int *)(0x50000004) = 0x12345678;
    tmp32 = *((volatile unsigned int *)(0x50000004));
    printf ("GPIO: tmp32 = 0x%0x\n", tmp32);
    printf ("GPIO: tmp32 = 0x%0x\n", tmp32);
    if (tmp32 != 0x12345678) printf("FAIL: 1st tb_gpio WRV failed!\n");

Here's the resulting output:

  # GPIO: 1_reset: before 
  # GPIO: tmp32 = 0x12345678
  # GPIO: tmp32 = 0x0
  # FAIL: 1st tb_gpio WRV failed!

Why is the 2nd printf of tmp32 different? Here's the disassembled code, and it seems that the first printf is printing not tmp32 which is in r4, but the literal that was written, which is in r6:

       0x00040082:    4e2b        +N      LDR      r6,[pc,#172]  ; [0x40130] = 0x12345678
        0x00040084:    f04f45a0    O..E    MOV      r5,#0x50000000
        0x00040088:    606e        n`      STR      r6,[r5,#4]
        0x0004008a:    686c        lh      LDR      r4,[r5,#4]
        0x0004008c:    4631        1F      MOV      r1,r6
        0x0004008e:    a029        ).      ADR      r0,{pc}+0xa6 ; 0x40134
        0x00040090:    f000fcf6    ....    BL       __0printf  ; 0x40a80
        0x00040094:    4621        !F      MOV      r1,r4
        0x00040096:    a027        '.      ADR      r0,{pc}+0x9e ; 0x40134
        0x00040098:    f000fcf2    ....    BL       __0printf  ; 0x40a80

This make sense? Is this an armcc bug? If so, what next?

Cheers,
Ric Peregrino
Dust Networks
  • Note: This was originally posted on 22nd April 2009 at http://forums.arm.com

    Downloaded and tried again with:

    [rperegrino@hdev02 1_reset]$ armcc --vsn
    ARM/Thumb C/C++ Compiler, RVCT3.1 [Build 794]
    For support see [url="http://www.arm.com/support/"]http://www.arm.com/support/[/url]
    Software supplied by: ARM Limited

    and got same result. Here's the invocation of the compiler and linker:

    armcc -g -DVLOGTIME -DTIME -O3 -Otime --no_inline --cpu Cortex-M3 -c 1_reset.c -
    o 1_reset.o
    armcc -g -DVLOGTIME -DTIME -O3 -Otime --no_inline --cpu Cortex-M3 -c clib.c -o c
    lib.o
    "clib.c", line 66: Warning:  #167-D: argument of type "unsigned char *" is incom
    patible with parameter of type "char *"
          sendchar(&tempch);
                   ^
    clib.c: 1 warning, 0 errors
    armcc -g -DVLOGTIME -DTIME -O3 -Otime --no_inline --cpu Cortex-M3 -c exceptions.
    c -o exceptions.o
    armlink  --info totals --symbols --inline \
                    --scatter cm3_code_on_codebus.scat \
                    --entry=__main --map \
                    --keep=exceptions.o\(exceptions_area\) \
                    1_reset.o  clib.o exceptions.o\
                    --callgraph --elf -o 1_reset.elf > link.txt
  • Note: This was originally posted on 24th April 2009 at http://forums.arm.com

    Can you try to make tmp32 global and check the assembly? Also please make sure you can actually write and read back from 0x50000004 location.

    The move from r6 to r1 may be because it does not think tmp32 to be volatile. In fact it did not even assign a memory location to tmp32.

    Move from r4 to r1 looks correct to me but i am surprised to see 0  in the second line of output. That makes me think that the location 0x50000004 wasnt written to properly or is not read properly.


    Another thing you can try is:
    volatile unsigned int *loc = (volatile unsigned int *)0x50000004;
    main () {
    unsigned int tmp32;

    printf ("GPIO: 1_reset: before\n");

    // touch tb_gpio
    *loc = 0x12345678;
    tmp32 = *loc;
    printf ("GPIO: tmp32 = 0x%0x\n", tmp32);
    *loc = 0x87654321;
    tmp32 = *loc;
    printf ("GPIO: tmp32 = 0x%0x\n", tmp32);
    if (tmp32 != 0x12345678) printf("FAIL: 1st tb_gpio WRV failed!\n");

    Please share the findings.

    Samar
  • Note: This was originally posted on 21st April 2009 at http://forums.arm.com

    That looks like a compiler bug to me - although I'm not a C spec expert...

    Can you try patching RVCT to the latest patch level (3.1 Build 794) and see if the issue still occurs? There were some known issues with volatiles in builds at high optimization levels which have since been fixed.

    The patch download can be found here: [url="http://www.arm.com/support/downloads/RVDS_31.html"]http://www.arm.com/support/downloads/RVDS_31.html[/url]

    If you still have a problem can you also post the compiler and linker command lines you used?
  • Note: This was originally posted on 22nd April 2009 at http://forums.arm.com

    Hi Ric, 

    My usual fixes for compiler bugs when no patch is available it to try a lower optimization level for the file in which the bug occurs, or try restructuring the code to see if it "goes away".

    Splitting the known buggy function out into a separate file can reduce the impact of the lower optimization level, because the other code can still be compiled at -O3.

    Raising the issue formally with ARM support (email support@arm.com) may be able to provide more information.