https://gcc.gnu.org/bugzilla/show_bug.cgi?id=128846

            Bug ID: 128846
           Summary: [17 regression] gcc.dg/tls/pr24428.c etc. FAIL on
                    Solaris/i386
           Product: gcc
           Version: 17.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: ro at gcc dot gnu.org
                CC: rguenth at gcc dot gnu.org
  Target Milestone: ---
            Target: i386-pc-solaris2.11

This patch

commit a9b19ded78ca2dd8f66b355b05ebbbb149c2d320
Author: Richard Biener <[email protected]>
Date:   Thu Oct 1 09:21:30 2026 +0200

    tree-optimization/127686 - mis-applied -O2 vector cost model heuristic

caused two tests to regress on 32-bit Solaris/x86, as identified in a reghunt:

FAIL: gcc.dg/tls/pr24428-2.c execution test
FAIL: gcc.dg/tls/pr24428.c execution test

E.g. for gcc.dg/tls/pr24428.c:

Thread 2 received signal SIGSEGV, Segmentation fault.
[Switching to thread 2 (Thread 1 (LWP 1))]
main () at /vol/gcc/src/hg/master/local/gcc/testsuite/gcc.dg/tls/pr24428.c:11
11          thrtest[i] = 1.0;
1: x/i $pc
=> 0x8050c70 <main+32>: movaps %xmm0,(%eax)
(gdb) p/x $eax
$1 = 0xfe5e0238
(gdb) p/x $xmm0
$2 = {v8_bfloat16 = {0x0, 0x0, 0x0, 0x3ff0, 0x0, 0x0, 0x0, 0x3ff0}, v8_half = {
    0x0, 0x0, 0x0, 0x3ff0, 0x0, 0x0, 0x0, 0x3ff0}, v4_float = {0x0, 0x3ff00000, 
    0x0, 0x3ff00000}, v2_double = {0x3ff0000000000000, 0x3ff0000000000000}, 
  v16_int8 = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0xf0, 0x3f, 0x0, 0x0, 0x0, 0x0, 
    0x0, 0x0, 0xf0, 0x3f}, v8_int16 = {0x0, 0x0, 0x0, 0x3ff0, 0x0, 0x0, 0x0, 
    0x3ff0}, v4_int32 = {0x0, 0x3ff00000, 0x0, 0x3ff00000}, v2_int64 = {
    0x3ff0000000000000, 0x3ff0000000000000}, 
  uint128 = 0x3ff00000000000003ff0000000000000}

The source needs to be 16-byte aligned, but isn't.

Reverting the patch locally fixed those two failures with no other changes.

Reply via email to