Inline asm with MSP430-GCC

Andrew G

Other Parts Discussed in Thread: MSP430G2553

As a workaround for https://e2e.ti.com/support/development_tools/compiler/f/343/t/487754 I wrote this inline assembly:

inline bool BOOL(unsigned int v) {
  bool b;
  asm("clr %0\n"
      "cmp #0, %1\n"
      "jz $+2\n"
      "mov #1, %0" : "=r" (b) : "r" (v));
  return b;
}

int main() {
  volatile unsigned int v = 100;
  volatile bool i = BOOL(v);
}

/opt/msp430/bin/msp430-elf-gcc -mmcu=msp430g2553 -O3 main.cc -o main.elf

However, this doesn't work as expected:

0000c150 <main>:
    c150:       04 12           push    r4              ;
    c152:       04 41           mov     r1,     r4      ;
    c154:       21 82           sub     #4,     r1      ;r2 As==10
    c156:       b4 40 64 00     mov     #100,   -2(r4)  ;#0x0064, 0xfffe
    c15a:       fe ff 
    c15c:       1c 44 fe ff     mov     -2(r4), r12     ;
    c160:       0c 43           clr     r12             ;
    c162:       0c 93           cmp     #0,     r12     ;r3 As==00
    c164:       00 24           jz      $+2             ;abs 0xc166
    c166:       1c 43           mov     #1,     r12     ;r3 As==01
    c168:       c4 4c fd ff     mov.b   r12,    -3(r4)  ; 0xfffd
    c16c:       0c 43           clr     r12             ;
    c16e:       21 52           add     #4,     r1      ;r2 As==10
    c170:       34 41           pop     r4              ;
    c172:       30 41           ret

gcc reuses the same register (r12) as both the input and the output of my BOOL function. Is this a bug in gcc or am I doing something wrong?

over 9 years ago

0 Andrew G over 9 years ago

Intellectual 900 points

Answering my own question. For this to work right, the input v also needs to be listed as the output:

inline bool BOOL(uint16_t v) {
  bool b;
  asm("clr %0\n"
      "cmp #0, %1\n"
      "jz $+2\n"
      "mov #1, %0\n" : "=r" (b), "=r" (v) : "1" (v));
  return b;
}

This may be a bug in gcc. The cmp instruction does not change its operands, hence listing the second register as an output should not be necessary.

0 George Mock over 9 years ago in reply to Andrew G

TI__Guru**** 251500 points

Thanks for letting us know. -George

0 DJ Delorie over 9 years ago

Intellectual 390 points

GCC assumes that all input operands are read before any output operands are written. In your case, this is not true, so you need to mark the output operand as "earlyclobber" by prefixing it with & - like "=&r"

0 Andrew G over 9 years ago in reply to DJ Delorie

Intellectual 900 points

That seems to do the trick, thank you.

0 Andrew G over 9 years ago in reply to DJ Delorie

Intellectual 900 points

I have another question about inline asm if you don't mind.

Let's say I want to use memory addressing instead of registers in my asm:

int a = 0;

int main() {
a = 5;
asm("inc %0" : "=m" (a));
}

This doesn't work (clobbers a = 5), because the asm statement only specifies outputs. But the following generates a warning:

asm("inc %0" : "=m" (a) : "0" (a));

test.cc:5:37: warning: matching constraint does not allow a register
asm("inc %0" : "=m" (a) : "0" (a));

What's the right way to do this?

0 Andrew G over 9 years ago in reply to Andrew G

Intellectual 900 points

Looks like what I need is asm("inc %0" : "+m" (a)). Should have googled before posting.

0 DJ Delorie over 9 years ago in reply to Andrew G

Intellectual 390 points

That's a right way. GCC is warning you that if the parameters don't match the 'm' constraint, there's nothing gcc can do about it. A better way would be to use "g" or "rm" instead of "m" so that gcc can reload values into a register if needed, assuming the instruction can handle either a register or a memory operand. In your example, gcc would prefer a register, but if you take out the "a=5;" it would prefer a memory operand.

Code Composer Studio™︎

Code Composer Studio forum

Inline asm with MSP430-GCC