CTOR roots are not explicitely represented so we have to make sure to materialize permutes on SLP graph entries to them.
Bootstrapped and tested on x86_64-unknown-linux-gnu, pushed. 2021-07-28 Richard Biener <rguent...@suse.de> PR tree-optimization/101615 * tree-vect-slp.c (vect_optimize_slp): Materialize permutes at CTOR SLP graph entries. * gcc.dg/vect/bb-slp-pr101615-2.c: New testcase. --- gcc/testsuite/gcc.dg/vect/bb-slp-pr101615-2.c | 23 +++++++++++++++++++ gcc/tree-vect-slp.c | 12 ++++++++++ 2 files changed, 35 insertions(+) create mode 100644 gcc/testsuite/gcc.dg/vect/bb-slp-pr101615-2.c diff --git a/gcc/testsuite/gcc.dg/vect/bb-slp-pr101615-2.c b/gcc/testsuite/gcc.dg/vect/bb-slp-pr101615-2.c new file mode 100644 index 00000000000..ac89883de22 --- /dev/null +++ b/gcc/testsuite/gcc.dg/vect/bb-slp-pr101615-2.c @@ -0,0 +1,23 @@ +/* { dg-do run } */ +/* { dg-additional-options "-O3 -w -Wno-psabi" } */ + +#include "tree-vect.h" + +int res[6] = { 5, 7, 11, 3, 3, 3 }; +int a[6] = {5, 5, 8}; +int c; + +int main() +{ + check_vect (); + for (int b = 0; b <= 4; b++) + for (; c <= 4; c++) { + a[0] |= 1; + for (int e = 0; e <= 4; e++) + a[e + 1] |= 3; + } + for (int d = 0; d < 6; d++) + if (a[d] != res[d]) + __builtin_abort (); + return 0; +} diff --git a/gcc/tree-vect-slp.c b/gcc/tree-vect-slp.c index 07cc24a60e1..a554c24e0fb 100644 --- a/gcc/tree-vect-slp.c +++ b/gcc/tree-vect-slp.c @@ -3715,6 +3715,18 @@ vect_optimize_slp (vec_info *vinfo) vertices[idx].perm_out = perms.length () - 1; } + /* In addition to the above we have to mark outgoing permutes facing + non-reduction graph entries that are not represented as to be + materialized. */ + for (slp_instance instance : vinfo->slp_instances) + if (SLP_INSTANCE_KIND (instance) == slp_inst_kind_ctor) + { + /* Just setting perm_out isn't enough for the propagation to + pick this up. */ + vertices[SLP_INSTANCE_TREE (instance)->vertex].perm_in = 0; + vertices[SLP_INSTANCE_TREE (instance)->vertex].perm_out = 0; + } + /* Propagate permutes along the graph and compute materialization points. */ bool changed; bool do_materialization = false; -- 2.26.2