https://gcc.gnu.org/bugzilla/show_bug.cgi?id=128728

            Bug ID: 128728
           Summary: [17 Regression] wrong code at -O2 for upper bits of a
                    product of two 3-bit bit-fields
           Product: gcc
           Version: 17.0
            Status: UNCONFIRMED
          Keywords: wrong-code
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: shimizu2486 at gmail dot com
  Target Milestone: ---
            Target: aarch64-linux-gnu loongarch64-linux-gnu
                    riscv64-linux-gnu

Live reproducer: https://godbolt.org/z/n1rbf1nE8

>From -O2 up, p >> 4 of an unsigned char product of two 3-bit bit-fields comes
out one too small.

$ cat t1.c
#include <stdio.h>
struct S { int v : 3; };
static volatile int A = -3, B = 2;
int main(void) {
  struct S a = { A }, b = { B };
  unsigned char p = b.v * (unsigned)a.v;
  printf("%d\n", (struct S){ p >> 4 }.v);
}

a.v is -3 and b.v is 2, so p is 250 and p >> 4 is 15, which is -1 as a 3-bit
bit-field. The program should print -1.

$ aarch64-linux-gnu-gcc -O0 -static t1.c -o t1 && qemu-aarch64 ./t1
-1                                  (correct)
$ aarch64-linux-gnu-gcc -O2 -static t1.c -o t1 && qemu-aarch64 ./t1
-2                                  (wrong)

  aarch64 trunk (372c32f911fa)   wrong
  aarch64 16.2.0                 correct
  aarch64 15.3.0                 correct
  aarch64 14.4.0                 correct

-O0 and -O1 are correct. -O2, -O3 and -Os are wrong.
-fno-expensive-optimizations makes -O2 print -1.
loongarch64, riscv64 trunk are wrong too. x86_64 trunk prints -1.

Reply via email to