The PR97330 fix caused some missed sinking of loads out of loops
the following patch re-instantiates.
Bootstrapped and tested on x86_64-unknown-linux-gnu, pushed.
2022-05-17 Richard Biener <rguent...@suse.de>
PR tree-optimization/105618
* tree-ssa-sink.cc (statement_sink_location): For virtual
PHI uses ignore those defining the used virtual operand.
* gcc.dg/tree-ssa/ssa-sink-19.c: New testcase.
---
gcc/testsuite/gcc.dg/tree-ssa/ssa-sink-19.c | 21 +++++++++++++++++++++
gcc/tree-ssa-sink.cc | 3 +++
2 files changed, 24 insertions(+)
create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/ssa-sink-19.c
diff --git a/gcc/testsuite/gcc.dg/tree-ssa/ssa-sink-19.c
b/gcc/testsuite/gcc.dg/tree-ssa/ssa-sink-19.c
new file mode 100644
index 00000000000..e98d13fe85b
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/tree-ssa/ssa-sink-19.c
@@ -0,0 +1,21 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -fdump-tree-sink1-details -fdump-tree-cddce2-details" } */
+
+static int b=4;
+int c;
+
+int
+main()
+{
+ int e[5] = {1,1,1,1,1};
+ for (; b >= 0; b--) {
+ c = e[b];
+ }
+ return 0;
+}
+
+/* We should sink e[b] out of the loop which is possible after
+ applying store motion to c and b. */
+/* { dg-final { scan-tree-dump "Sinking # VUSE" "sink1" } } */
+/* And remove the loop after final value replacement. */
+/* { dg-final { scan-tree-dump "fix_loop_structure: removing loop" "cddce2" }
} */
diff --git a/gcc/tree-ssa-sink.cc b/gcc/tree-ssa-sink.cc
index 1c226406feb..8ce4403ddc8 100644
--- a/gcc/tree-ssa-sink.cc
+++ b/gcc/tree-ssa-sink.cc
@@ -390,6 +390,9 @@ statement_sink_location (gimple *stmt, basic_block frombb,
with the use. */
if (gimple_code (use_stmt) == GIMPLE_PHI)
{
+ /* If the PHI defines the virtual operand, ignore it. */
+ if (gimple_phi_result (use_stmt) == gimple_vuse (stmt))
+ continue;
/* In case the PHI node post-dominates the current insert
location we can disregard it. But make sure it is not
dominating it as well as can happen in a CFG cycle. */
--
2.35.3
