[Bug target/117688] [15 Regression] RISC-V: Wrong code for .SAT_SUB

[kristerw at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117688

--- Comment #1 from Krister Walfridsson <kristerw at gcc dot gnu.org> ---
Interestingly, the function is generated correctly if x is passed as a function
argument:

__attribute__ ((noipa)) void
foo2 (int8_t x)
{
  int8_t minus; 
  _Bool overflow = __builtin_sub_overflow (x, y, &minus);
  result = overflow ? (x < 0 ? 0x80 : 0x7f) : minus;
}

The difference in the generated code is that GCC generates foo as if x were
unsigned, so code which for foo2 is generated as:

        srai    a3,a0,63
        xori    a4,a3,127
        and     a4,a4,a5

is for foo miscompiled as:

        andi    a4,a5,127