https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68961
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|NEW |ASSIGNED
Assignee|unassigned at gcc dot gnu.org |rguenth at gcc dot
gnu.org
--- Comment #8 from Richard Biener <rguenth at gcc dot gnu.org> ---
Btw, the testcase seems to be really "special" given exact register overlap
between return value and incoming args (if you'd look at the vectorizers choice
to say this is profitable to vectorize). Making it fairer like with
long double
pack (double x, double a, double aa)
{
union u_ld u;
u.d[0] = a;
u.d[1] = aa;
return u.ld;
}
produces without SLP
pack:
mfvsrd 10,2
fmr 2,3
mtvsrd 1,10
blr
and with
pack:
xxpermdi 0,3,2,0
addi 9,1,-16
xxpermdi 0,0,0,2
stxvd2x 0,0,9
lfd 1,-16(1)
lfd 2,-8(1)
blr
to that would be the thing to compare cost-wise. Currently we have
t.c:9:11: note: Cost model analysis:
Vector inside of basic block cost: 1
Vector prologue cost: 0
Vector epilogue cost: 0
Scalar cost of basic block: 2
so for some reason the vector build is not accounted for.
Ah, I see why. Mine.
