http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50730
Bug #: 50730
Summary: SLP vectorization confused by unrelated DRs
Classification: Unclassified
Product: gcc
Version: 4.7.0
Status: UNCONFIRMED
Keywords: missed-optimization
Severity: normal
Priority: P3
Component: tree-optimization
AssignedTo: unassig...@gcc.gnu.org
ReportedBy: rgue...@gcc.gnu.org
CC: i...@gcc.gnu.org
The SLP vectorizer is confused by the aggregate store in
typedef __complex__ float Value;
struct A {
Value a[16 / sizeof (Value)];
} __attribute__ ((aligned(16)));
A sum(A a, A b)
{
a.a[0]+=b.a[0];
a.a[1]+=b.a[1];
return a;
}
when built with -O3 -fno-tree-sra:
<bb 2>:
D.2130_11 = REALPART_EXPR <a.a[0]>;
D.2131_12 = IMAGPART_EXPR <a.a[0]>;
D.2132_13 = REALPART_EXPR <b.a[0]>;
D.2133_14 = IMAGPART_EXPR <b.a[0]>;
D.2134_15 = D.2130_11 + D.2132_13;
D.2135_16 = D.2131_12 + D.2133_14;
REALPART_EXPR <a.a[0]> = D.2134_15;
IMAGPART_EXPR <a.a[0]> = D.2135_16;
D.2136_17 = REALPART_EXPR <a.a[1]>;
D.2137_18 = IMAGPART_EXPR <a.a[1]>;
D.2138_19 = REALPART_EXPR <b.a[1]>;
D.2139_20 = IMAGPART_EXPR <b.a[1]>;
D.2140_21 = D.2136_17 + D.2138_19;
D.2141_22 = D.2137_18 + D.2139_20;
REALPART_EXPR <a.a[1]> = D.2140_21;
IMAGPART_EXPR <a.a[1]> = D.2141_22;
D.2119 = a;
return D.2119;
but I don't see why SLP couldn't simply stop analyzing the BB when
it encounters a DR that it cannot build or handle. The
previous instructions can be still vectorized.
If you put an unrelated volatile store at the end of any sequence
in a bb-slp testcase this reproduces as well. The above testcase
will probably fail due to unsupported strided stores.
