On Thu, 24 Nov 2016, Richard Biener wrote:
>
> I am testing the following patch for an optimization regression where
> a loop made dead by final value replacement was made used again by
> DOM 20 passes later. The real issue here is that we do not get rid
> of dead loops until very late so this patch makes sure to do that.
> We could schedule it later (but better no later than unrolling
> as that might expose a pretty inefficient way of removing a dead loop).
>
> Bootstrap and regtest running on x86_64-unknown-linux-gnu.
As expected some testcases need adjustment, thus applied as follows.
Richard.
2016-11-24 Richard Biener <rguent...@suse.de>
PR tree-optimization/78343
* passes.def: Add CD-DCE pass after loop splitting.
* tree-ssa-dce.c (find_obviously_necessary_stmts): Move
SCEV init/finalize ...
(perform_tree_ssa_dce): ... here. Deal with being
executed inside the loop pipeline in aggressive mode.
* gcc.dg/tree-ssa/sccp-2.c: New testcase.
* gcc.dg/autopar/uns-outer-6.c: Adjust.
* gcc.dg/tree-ssa/20030808-1.c: Likewise.
* gcc.dg/tree-ssa/20040305-1.c: Likewise.
* gcc.dg/vect/pr38529.c: Likewise.
diff --git a/gcc/passes.def b/gcc/passes.def
index 2a470a7..2fa682b 100644
--- a/gcc/passes.def
+++ b/gcc/passes.def
@@ -271,6 +271,9 @@ along with GCC; see the file COPYING3. If not see
NEXT_PASS (pass_tree_unswitch);
NEXT_PASS (pass_scev_cprop);
NEXT_PASS (pass_loop_split);
+ /* All unswitching, final value replacement and splitting can expose
+ empty loops. Remove them now. */
+ NEXT_PASS (pass_cd_dce);
NEXT_PASS (pass_record_bounds);
NEXT_PASS (pass_loop_distribution);
NEXT_PASS (pass_copy_prop);
diff --git a/gcc/testsuite/gcc.dg/autopar/uns-outer-6.c
b/gcc/testsuite/gcc.dg/autopar/uns-outer-6.c
index dc2870b..5af60b0 100644
--- a/gcc/testsuite/gcc.dg/autopar/uns-outer-6.c
+++ b/gcc/testsuite/gcc.dg/autopar/uns-outer-6.c
@@ -25,7 +25,7 @@ parloop (int N)
for (i = 0; i < N; i++)
{
for (j = 0; j < N; j++)
- y[i]=x[i][j];
+ y[i] += x[i][j];
sum += y[i];
}
g_sum = sum;
@@ -46,6 +46,10 @@ main (void)
/* Check that outer loop is parallelized. */
+/* This fails because we have
+ FAILED: data dependencies exist across iterations
+
+
/* { dg-final { scan-tree-dump-times "parallelizing outer loop" 1 "parloops2"
} } */
/* { dg-final { scan-tree-dump-times "parallelizing inner loop" 0 "parloops2"
} } */
/* { dg-final { scan-tree-dump-times "loopfn" 4 "optimized" } } */
diff --git a/gcc/testsuite/gcc.dg/tree-ssa/20030808-1.c
b/gcc/testsuite/gcc.dg/tree-ssa/20030808-1.c
index 7cc5404..cda86a7 100644
--- a/gcc/testsuite/gcc.dg/tree-ssa/20030808-1.c
+++ b/gcc/testsuite/gcc.dg/tree-ssa/20030808-1.c
@@ -33,8 +33,8 @@ delete_dead_jumptables ()
/* There should be no loads of ->code. If any exist, then we failed to
optimize away all the IF statements and the statements feeding
their conditions. */
-/* { dg-final { scan-tree-dump-times "->code" 0 "cddce2"} } */
+/* { dg-final { scan-tree-dump-times "->code" 0 "cddce3"} } */
/* There should be no IF statements. */
-/* { dg-final { scan-tree-dump-times "if " 0 "cddce2"} } */
+/* { dg-final { scan-tree-dump-times "if " 0 "cddce3"} } */
diff --git a/gcc/testsuite/gcc.dg/tree-ssa/20040305-1.c
b/gcc/testsuite/gcc.dg/tree-ssa/20040305-1.c
index 501e28c..d1a9af8 100644
--- a/gcc/testsuite/gcc.dg/tree-ssa/20040305-1.c
+++ b/gcc/testsuite/gcc.dg/tree-ssa/20040305-1.c
@@ -27,4 +27,4 @@ void foo(int edx, int eax)
/* After cddce we should have two IF statements remaining as the other
two tests can be threaded. */
-/* { dg-final { scan-tree-dump-times "if " 2 "cddce2"} } */
+/* { dg-final { scan-tree-dump-times "if " 2 "cddce3"} } */
diff --git a/gcc/testsuite/gcc.dg/tree-ssa/sccp-2.c
b/gcc/testsuite/gcc.dg/tree-ssa/sccp-2.c
new file mode 100644
index 0000000..099b281
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/tree-ssa/sccp-2.c
@@ -0,0 +1,15 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -fdump-tree-optimized" } */
+
+unsigned int
+test(unsigned int quant)
+{
+ unsigned int sum = 0;
+ for (unsigned int i = 0; i < quant; ++i)
+ sum += quant;
+ return sum;
+}
+
+/* A single basic-block should remain (computing and
+ returning quant * quant). */
+/* { dg-final { scan-tree-dump-times "bb" 1 "optimized" } } */
diff --git a/gcc/testsuite/gcc.dg/vect/pr38529.c
b/gcc/testsuite/gcc.dg/vect/pr38529.c
index 171adeb..9b5919d 100644
--- a/gcc/testsuite/gcc.dg/vect/pr38529.c
+++ b/gcc/testsuite/gcc.dg/vect/pr38529.c
@@ -11,7 +11,3 @@ void foo()
for (j = 0; j < 17; ++j)
a[i] = 0;
}
-
-/* { dg-final { scan-tree-dump-times "OUTER LOOP VECTORIZED" 1 "vect" } } */
-
-
diff --git a/gcc/tree-ssa-dce.c b/gcc/tree-ssa-dce.c
index 7b9814e..50b5eef 100644
--- a/gcc/tree-ssa-dce.c
+++ b/gcc/tree-ssa-dce.c
@@ -400,7 +400,6 @@ find_obviously_necessary_stmts (bool aggressive)
if (aggressive)
{
struct loop *loop;
- scev_initialize ();
if (mark_irreducible_loops ())
FOR_EACH_BB_FN (bb, cfun)
{
@@ -423,7 +422,6 @@ find_obviously_necessary_stmts (bool aggressive)
fprintf (dump_file, "can not prove finiteness of loop %i\n",
loop->num);
mark_control_dependent_edges_necessary (loop->latch, false);
}
- scev_finalize ();
}
}
@@ -1567,9 +1565,13 @@ perform_tree_ssa_dce (bool aggressive)
/* Preheaders are needed for SCEV to work.
Simple lateches and recorded exits improve chances that loop will
proved to be finite in testcases such as in loop-15.c and loop-24.c */
- if (aggressive)
- loop_optimizer_init (LOOPS_NORMAL
- | LOOPS_HAVE_RECORDED_EXITS);
+ bool in_loop_pipeline = scev_initialized_p ();
+ if (aggressive && ! in_loop_pipeline)
+ {
+ scev_initialize ();
+ loop_optimizer_init (LOOPS_NORMAL
+ | LOOPS_HAVE_RECORDED_EXITS);
+ }
tree_dce_init (aggressive);
@@ -1588,8 +1590,11 @@ perform_tree_ssa_dce (bool aggressive)
find_obviously_necessary_stmts (aggressive);
- if (aggressive)
- loop_optimizer_finalize ();
+ if (aggressive && ! in_loop_pipeline)
+ {
+ loop_optimizer_finalize ();
+ scev_finalize ();
+ }
longest_chain = 0;
total_chain = 0;
@@ -1623,7 +1628,7 @@ perform_tree_ssa_dce (bool aggressive)
if (something_changed)
{
free_numbers_of_iterations_estimates (cfun);
- if (scev_initialized_p ())
+ if (in_loop_pipeline)
scev_reset ();
return TODO_update_ssa | TODO_cleanup_cfg;
}
