From ac6cd62a351a8f1f3637a2552c74eb5eb51cfdda Mon Sep 17 00:00:00 2001
From: Richard Biener <rguenther@suse.de>
Date: Tue, 3 Sep 2024 09:23:20 +0200
Subject: [PATCH] tree-optimization/116575 - avoid ICE with SLP mask_load_lane
The following avoids performing re-discovery with single lanes in
the attempt to for the use of mask_load_lane as rediscovery will
fail since a single lane of a mask load will appear permuted which
isn't supported.
PR tree-optimization/116575
* tree-vect-slp.cc (vect_analyze_slp): Properly compute
the mask argument for vect_load/store_lanes_supported.
When the load is masked for now avoid rediscovery.
* gcc.dg/vect/pr116575.c: New testcase.
---
gcc/testsuite/gcc.dg/vect/pr116575.c | 15 +++++++++++++++
gcc/tree-vect-slp.cc | 19 +++++++++++++++++--
2 files changed, 32 insertions(+), 2 deletions(-)
create mode 100644 gcc/testsuite/gcc.dg/vect/pr116575.c
diff --git a/gcc/testsuite/gcc.dg/vect/pr116575.c b/gcc/testsuite/gcc.dg/vect/pr116575.c
new file mode 100644
index 000000000000..2047041ca64b
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/vect/pr116575.c
@@ -0,0 +1,15 @@
+/* { dg-do compile } */
+
+int a;
+float *b, *c;
+void d(char * __restrict e)
+{
+ for (; a; a++, b += 4, c += 4)
+ if (*e++) {
+ float *f = c;
+ f[0] = b[0];
+ f[1] = b[1];
+ f[2] = b[2];
+ f[3] = b[3];
+ }
+}
diff --git a/gcc/tree-vect-slp.cc b/gcc/tree-vect-slp.cc
index 2302d91fd23f..1342913affa1 100644
--- a/gcc/tree-vect-slp.cc
+++ b/gcc/tree-vect-slp.cc
@@ -4720,11 +4720,16 @@ vect_analyze_slp (vec_info *vinfo, unsigned max_tree_size)
}
}
+ gimple *rep = STMT_VINFO_STMT (SLP_TREE_REPRESENTATIVE (slp_root));
+ bool masked = (is_gimple_call (rep)
+ && gimple_call_internal_p (rep)
+ && internal_fn_mask_index
+ (gimple_call_internal_fn (rep)) != -1);
/* If the loads and stores can use load/store-lanes force re-discovery
with single lanes. */
if (loads_permuted
&& !slp_root->ldst_lanes
- && vect_store_lanes_supported (vectype, group_size, false)
+ && vect_store_lanes_supported (vectype, group_size, masked)
!= IFN_LAST)
{
bool can_use_lanes = true;
@@ -4734,13 +4739,23 @@ vect_analyze_slp (vec_info *vinfo, unsigned max_tree_size)
{
stmt_vec_info stmt_vinfo = DR_GROUP_FIRST_ELEMENT
(SLP_TREE_REPRESENTATIVE (load_node));
+ rep = STMT_VINFO_STMT (stmt_vinfo);
+ masked = (is_gimple_call (rep)
+ && gimple_call_internal_p (rep)
+ && internal_fn_mask_index
+ (gimple_call_internal_fn (rep)));
/* Use SLP for strided accesses (or if we can't
load-lanes). */
if (STMT_VINFO_STRIDED_P (stmt_vinfo)
|| compare_step_with_zero (vinfo, stmt_vinfo) <= 0
|| vect_load_lanes_supported
(STMT_VINFO_VECTYPE (stmt_vinfo),
- DR_GROUP_SIZE (stmt_vinfo), false) == IFN_LAST)
+ DR_GROUP_SIZE (stmt_vinfo), masked) == IFN_LAST
+ /* ??? During SLP re-discovery with a single lane
+ a masked grouped load will appear permuted and
+ discovery will fail. We have to rework this
+ on the discovery side - for now avoid ICEing. */
+ || masked)
{
can_use_lanes = false;
break;
--
GitLab