[Bug target/78192] New: extract from vector registers to int results in wrong data order

[acsawdey at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78192

            Bug ID: 78192
           Summary: extract from vector registers to int results in wrong
                    data order
           Product: gcc
           Version: 7.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: meissner at gcc dot gnu.org
          Reporter: acsawdey at gcc dot gnu.org
                CC: wschmidt at gcc dot gnu.org
  Target Milestone: ---
            Target: rs6000 ppc64le power9

Tested with trunk 241701, running on power8 ppc64le compiling for power9
ppc64le.

gcc.c-torture/execute/simd-1.c is failing and it appears that the extraction
from vector to int registers is not happening in the right order.

      7 typedef int __attribute__((mode(SI))) __attribute__((vector_size (16)))
vecint;
      8 typedef int __attribute__((mode(SI))) siint;
      9
     10 vecint i = { 150, 100, 150, 200 };
     11 vecint j = { 10, 13, 20, 30 };
     12 vecint k;
     13
     14 union {
     15   vecint v;
     16   siint i[4];
     17 } res;
     18
     19 /* This should go away once we can use == and != on vector types.  */
     20 void
     21 verify (siint a1, siint a2, siint a3, siint a4,
     22         siint b1, siint b2, siint b3, siint b4)
     23 {
     24   if (a1 != b1
     25       || a2 != b2
     26       || a3 != b3
     27       || a4 != b4)
     28     abort ();
     29 }
     30
     31 int
     32 main ()
     33 {
     34   k = i + j;
     35   res.v = k;

   0x00000000100006f0 <main+44>:        lxv     vs63,0(r9)
   0x00000000100006f4 <main+48>:        lxv     vs62,16(r9)
   0x00000000100006f8 <main+52>:        vadduwm v0,v31,v30
   0x00000000100006fc <main+56>:        nop
   0x0000000010000700 <main+60>:        addi    r30,r2,-32400
   0x0000000010000704 <main+64>:        stxv    vs32,0(r30)
   0x0000000010000708 <main+68>:        nop
   0x000000001000070c <main+72>:        addi    r31,r2,-32384
   0x0000000010000710 <main+76>:        stxv    vs32,0(r31)
   0x0000000010000714 <main+80>:        xxextractuw vs33,vs32,12
   0x0000000010000718 <main+84>:        mfvrwz  r6,v1
   0x000000001000071c <main+88>:        mfvrwz  r5,v0
   0x0000000010000720 <main+92>:        xxextractuw vs33,vs32,4
   0x0000000010000724 <main+96>:        mfvrwz  r4,v1
   0x0000000010000728 <main+100>:       xxextractuw vs32,vs32,0
   0x000000001000072c <main+104>:       mfvrwz  r3,v0
   0x0000000010000730 <main+108>:       li      r10,230
   0x0000000010000734 <main+112>:       li      r9,170
   0x0000000010000738 <main+116>:       li      r8,113
   0x000000001000073c <main+120>:       li      r7,160
   0x0000000010000740 <main+124>:       extsw   r6,r6
   0x0000000010000744 <main+128>:       extsw   r5,r5
   0x0000000010000748 <main+132>:       extsw   r4,r4
   0x000000001000074c <main+136>:       extsw   r3,r3
   0x0000000010000750 <main+140>:       bl      0x10000684 <verify+8>

at 0x00000000100006fc the vector register results look ok:
(gdb) p $vs32
$22 = {uint128 = 0x000000e6000000aa00000071000000a0, v2_double =
{2.3978552445812627e-312, 4.8805903200600379e-312}, v4_float = {2.24207754e-43,
1.58346726e-43, 2.38220739e-43, 3.22298647e-43}, v4_int32 = {
    160, 113, 170, 230}, v8_int16 = {160, 0, 113, 0, 170, 0, 230, 0}, v16_int8
= {-96, 0, 0, 0, 113, 0, 0, 0, -86, 0, 0, 0, -26, 0, 0, 0}}
(gdb) p $vs62
$23 = {uint128 = 0x0000001e000000140000000d0000000a, v2_double =
{2.7585945287489197e-313, 6.3659873738839482e-313}, v4_float = {1.40129846e-44,
1.821688e-44, 2.80259693e-44, 4.20389539e-44}, v4_int32 = {10,
    13, 20, 30}, v8_int16 = {10, 0, 13, 0, 20, 0, 30, 0}, v16_int8 = {10, 0, 0,
0, 13, 0, 0, 0, 20, 0, 0, 0, 30, 0, 0, 0}}
(gdb) p $vs63
$24 = {uint128 = 0x000000c8000000960000006400000096, v2_double =
{2.1219957917063708e-312, 4.2439915826716431e-312}, v4_float = {2.1019477e-43,
1.40129846e-43, 2.1019477e-43, 2.80259693e-43}, v4_int32 = {150,
    100, 150, 200}, v8_int16 = {150, 0, 100, 0, 150, 0, 200, 0}, v16_int8 =
{-106, 0, 0, 0, 100, 0, 0, 0, -106, 0, 0, 0, -56, 0, 0, 0}}

at 0x0000000010000730 the extraction into gprs is incorrect:
(gdb) p $r3
$14 = 230
(gdb) p $r4
$15 = 170
(gdb) p $r5
$16 = 170
(gdb) p $r6
$17 = 160