On 09/20/2016 08:38 AM, Bernd Edlinger wrote:
On 09/20/16 16:20, Kyrill Tkachov wrote:

On 20/09/16 15:11, Bernd Edlinger wrote:
On 09/20/16 13:29, Kyrill Tkachov wrote:
arm bootstrap is now failing:
$SRC/gcc/config/arm/arm.h:2229:40: error: ?: using integer constants in
boolean context [-Werror=int-in-bool-context]
      : (TARGET_VFP_DOUBLE ? (TARGET_FP16 ? 14 : 12) : 0)) \
$SRC/gcc/config/arm/arm-c.c:133:7: note: in expansion of macro
     if (TARGET_ARM_FP)

The full definition of TARGET_ARM_FP is:
#define TARGET_ARM_FP            \
              : (TARGET_VFP_DOUBLE ? (TARGET_FP16 ? 14 : 12) : 0)) \
                : 0)

We want it set to 0 when there's no FP but when FP is available we set
it to a bitmask
to suggest the level that is available. That seems like a legitimate use
to me.

Ok, I see, sorry for that.

I think I will have to suppress the warning if the conditional is in
a macro somehow.

Can you work around that for a few days?

I changed to:
  if (TARGET_ARM_FP != 0)
in my tree to allow the build to proceed but encountered another
bootstrap failure:
In file included from ./tm.h:38:0,
                  from $SRC/gcc/backend.h:28,
                  from $SRC/gcc/regrename.c:23:
$SRC/gcc/regrename.c: In function 'void rename_chains()':
$SRC/gcc/config/arm/arm.h:915:4: error: ?: using integer constants in
boolean context [-Werror=int-in-bool-context]
    (TARGET_ARM     \
$SRC/gcc/regrename.c:484:8: note: in expansion of macro
     || (HARD_FRAME_POINTER_REGNUM && frame_pointer_needed
That looks like a legitimate bug in regrename.c ?
The full condition is:
        if (fixed_regs[reg] || global_regs[reg]
                && reg == HARD_FRAME_POINTER_REGNUM)
            || (HARD_FRAME_POINTER_REGNUM && frame_pointer_needed
                && reg == FRAME_POINTER_REGNUM))

The condition in the second || looks bogus (what use testing if a
register is 'non-zero').

So this looks like a useful catch by the warning, though I'm not sure at
first glance how
to fix it.

I assume HARD_FRAME_POINTER_REGNUM is never zero.
It could be zero. It's just a hard register number. No target has the property that its hard frame pointer register is 0 though :-)

That looks like it should be

  || (frame_pointer_needed && reg == FRAME_POINTER_REGNUM)
I'm not even entirely sure why this is two conditions.

if (fixed_regs[regs] || global_regs[reg]
    || (frame_pointer_needed
        && (reg == HARD_FRAME_POINTER_REGNUM
            || reg == FRAME_POINTER_REGNUM))))


Bernd Schmidt ought to be able to guide us.


Reply via email to