[Bug target/109747] New: [12/13/14 Regression] SLP cost of constructors is off

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

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109747

            Bug ID: 109747
           Summary: [12/13/14 Regression] SLP cost of constructors is off
           Product: gcc
           Version: 13.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: rguenth at gcc dot gnu.org
  Target Milestone: ---

When r12-7319-g90d693bdc9d718 added the accounting for GPR->XMM moves to the
cost of SLP CTORs I failed to realize that the costing code will pass down
the very same (and full) SLP node from vect_prologue_cost_for_slp but
will generate costs for each individual actual vector that's constructed.

So for the case where the SLP node covers more than one vector the costing
will be off.  That's visible in the costing of the testcase for PR108724
for example which is

void foo(int *a, const int *__restrict b, const int *__restrict c)
{
  for (int i = 0; i < 16; i++) {
    a[i] = b[i] + c[i];
  }
}

and we end up with

_17 8 times unaligned_store (misalign -1) costs 96 in body
node 0x3fb5838 1 times vec_construct costs 100 in prologue
node 0x3fb5838 1 times vec_construct costs 100 in prologue
node 0x3fb5838 1 times vec_construct costs 100 in prologue
node 0x3fb5838 1 times vec_construct costs 100 in prologue
node 0x3fb5838 1 times vec_construct costs 100 in prologue
node 0x3fb5838 1 times vec_construct costs 100 in prologue
node 0x3fb5838 1 times vec_construct costs 100 in prologue
node 0x3fb5838 1 times vec_construct costs 100 in prologue