[Bug target/95798] [10/11 Regression] Initialization code --- suboptimal

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95798

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
            Summary|Initialization code ---     |[10/11 Regression]
                   |suboptimal                  |Initialization code ---
                   |                            |suboptimal
   Last reconfirmed|                            |2020-06-22
                 CC|                            |jakub at gcc dot gnu.org,
                   |                            |rdapp at gcc dot gnu.org
   Target Milestone|---                         |10.2
     Ever confirmed|0                           |1

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
The 9 -> 10 regression started with
r10-2806-gdf7d46d925c7baca7bf9961aee900876d8aef225
since which the IL is much larger and the resulting code less efficient.
The testcase as written is just weird, it is an expensive check whether the
program is called with multiple of 128 arguments >= 1024 arguments (otherwise
it invokes UB).

Adjusted testcase that is more meaningful:
void bar (unsigned long long *, int);

void
foo (int y, unsigned long long z)
{
  unsigned long long x[1024];
  unsigned long long i = y % 127;
  __builtin_memset (x, -1, sizeof (x));
  x[i] = 0;
  x[i + 1] = 0;
  x[i + 2] = 0;
  x[i + 3] = 0;
  x[i + 4] = 0;
  x[i + 5] = 0;
  x[i + 6] = 0;
  x[i + 7] = 0;
  bar (x, y);
}