https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65930
Bug ID: 65930
Summary: Reduction with sign-change not handled
Product: gcc
Version: 6.0
Status: UNCONFIRMED
Keywords: missed-optimization
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: rguenth at gcc dot gnu.org
Blocks: 53947
Target Milestone: ---
Neither
int foo (unsigned int *x)
{
int sum = 0;
for (int i = 0; i < 4; ++i)
sum += x[i*4+0]+ x[i*4 + 1] + x[i*4 + 2] + x[i*4 + 3];
return sum;
}
nor
int bar (unsigned int *x)
{
int sum = 0;
for (int i = 0; i < 16; ++i)
sum += x[i];
return sum;
}
are currently vectorized because
t.c:4:3: note: reduction: not commutative/associative: sum_27 = (int) _26;
though the sign of 'sum' vs the sign of 'x' doesn't really matter here.
It works for
unsigned baz (int *x)
{
unsigned int sum = 0;
for (int i = 0; i < 16; ++i)
sum += x[i];
return sum;
}
because C promotes x[i] instead of sum in this case.
For vectorization we might want to change the add(s) to int - of course
not strictly valid because of undefined overflow issues.
This kind of loop appears in a hot area of x264.
Referenced Bugs:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=53947
[Bug 53947] [meta-bug] vectorizer missed-optimizations
