[Bug target/122695] New: LoongArch: Wrong code with __builtin_shuffle and -mlasx

[xry111 at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122695

            Bug ID: 122695
           Summary: LoongArch: Wrong code with __builtin_shuffle and
                    -mlasx
           Product: gcc
           Version: 16.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: xry111 at gcc dot gnu.org
  Target Milestone: ---

Test case 1:

[[gnu::vector_size (32)]] char a, b, c;

[[gnu::noipa]] void
test (void)
{
  a = __builtin_shuffle (a, a, c);
}

int
main (void)
{
  a = (typeof (a)){} + 5;
  c = (typeof (a)){} + 64;
  test ();
  if (a[0] != 5)
    __builtin_trap ();
}

results:

test:
        la.local        $r12,.LANCHOR0
        xvld    $xr0,$r12,0
        xvld    $xr2,$r12,32
        xvpermi.d       $xr1,$xr0,68
        xvpermi.d       $xr0,$xr0,238
        xvshuf.b        $xr0,$xr0,$xr1,$xr2
        xvst    $xr0,$r12,0
        jr      $r1

It invokes unpredictable result of xvshuf.b, producing wrong result on at least
Loongson 3C5000L-LL.

Test case 2:

[[gnu::vector_size(32)]] short a, b, c;

[[gnu::noipa]] void
test (void)
{
  a = __builtin_shuffle(a, b, c) + c;
}

int
main (void)
{
  a = (typeof (a)){} + 1;
  b = (typeof (b)){} + 2;
  c = (typeof (c)){} + 128;
  test ();
  if (a[0] != 129)
    __builtin_trap ();
}

results:

test:
    la.local    $r12,.LANCHOR0
    xvrepli.h   $xr3,31
...
    xvand.v $xr0,$xr0,$xr3
...
    xvadd.h $xr0,$xr1,$xr0
    xvst    $xr0,$r12,32
    jr  $r1 

i.e. it's now __builtin_shuffle (a, b, c) + (c & 31).