[Bug target/123271] [13/14/15/16 Regression] -ftrapv fails to detect overflow on ARM32 since GCC 10 since r10-3995

[rearnsha at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123271

--- Comment #5 from Richard Earnshaw <rearnsha at gcc dot gnu.org> ---
(In reply to Jakub Jelinek from comment #3)

> Unless ARM has also conditional trap instruction that can trap if CC_V
> CC_REGNUM, in that case just the first one should be renamed and
> define_expand added which would behave similarly to subv<mode>4 except
> instead of conditionally jumping to the last operand it would conditionally
> trap.

We don't have conditional trap instructions, but we do have a trap instruction
pattern.  We could do better than a libcall by open-coding a trap pattern as 

   adds    r0, r1, r2
   bvs     Ltrap
   ...

Ltrap:
   <trap-bit-pattern>

Though that risks common-code making it hard to identify which arithmetic
instruction actually trapped.

Slightly more expensive, but probably faster than a libcall would be to have a
pattern that jumped over a trap instruction:

   adds    r0, r1, r2
   bvc     LnotrapN
   <trap-bit-pattern>
LnotrapN:
   ...

which could be coded as a single pattern to limit the code-commoning.