[bug #58862] avr/wdt.h LLVM11 error for atmega328p

[avrs-admin]

Fri 31 Jul 2020 01:32:44 PM CEST

Error message:

avr\include\avr/wdt.h:419:12: error: value '64' out of range for constraint 'I'
: "I" (_SFR_IO_ADDR(_WD_CONTROL_REG)),
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~
avr\include\avr/sfr_defs.h:183:27: note: expanded from macro '_SFR_IO_ADDR'
#define _SFR_IO_ADDR(sfr) (_SFR_MEM_ADDR(sfr) - __SFR_OFFSET)
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

---

I guess that wdt_enable()/wdt_disable() should have conditional preprocessor code instead of conditional inline.

This issue was migrated from https://savannah.nongnu.org/bugs/?58862

[KOLANICH]

I guess that the quickest fix is to use & 0b111111 for "runtime dispatch" currently present to calm down the compiler. It seems that that branch is not expected to be taken anyway on those MCUs.

[sprintersb]

The problem is in clang. Even with __builtin_constant_p it compiles a block with a non-matching constraint, but it's the very purpose of __builtin_constant_p to skip compilation of if (0) blocks. Test case:

#include <avr/io.h>

#define SFR (*(volatile char*) (0x40 + __SFR_OFFSET))

void func (void)
{
    if (__builtin_constant_p (_SFR_IO_REG_P (SFR))
        && _SFR_IO_REG_P (SFR))
    {
        __asm volatile ("in __tmp_reg__,%0"
                        :: "I" (_SFR_IO_ADDR (SFR)));
    }
}

Notice that _SFR_IO_REG_P (SFR) is a compile-time constant that evaluates to 0. Even simpler test case:

void func (void)
{
    if (0)
        __asm volatile ("in __tmp_reg__,%0" :: "I" (64));
}

Closed. This is a clang/llvm issue.

[UltraBlackLinux]

@sprintersb I've also just run into this issue, but I don't really understand what's going on with this problem at all.
Would you mind creating an issue over at llvm regarding this problem? I think I'd grossly misrepresent this problem if I opened an issue there myself.

I myself kinda depend on clang to get language servers working, and tooling that does exist for arduino often doesn't properly interface with the software I'm using. clang and it's compile_commands.json seems to be the only good way to get reliable code completion and syntax highlighting within my editor.