[PR69123] fix VTA dataflow oscillation

Here are two patches related with PR69123, an infinite dataflow loop in
VTA.

The first non-comment hunk in var-tracking.c:drop_overlapping_mem_locs
is what fixes the problem, but the other changes in the first patch fix
similar problems that might cause other such oscillations.

The second patch adds some more information to detailed vartrack dumps,
avoiding short-circuiting of dataflow set compares and dumping added and
removed locations for variables present in both sets.

The patches are largely independent, but they were successfully
regstrapped together on x86_64-linux-gnu and i686-linux-gnu.  They were
only compile-tested separately.

Ok to install?


[PR69123] fix handling of MEMs in VTA to avoid dataflow oscillation

From: Alexandre Oliva <aoliva@redhat.com>

The problem arises because we used to drop overwritten MEMs from loc
lists of VALUEs, but not of other onepart variables, and it just so
happens that, by doing so, block 6 in the testcase has no D#5 in its
output in the first pass, because the MEM holding its (previous) value
was correctly dropped from value 88:88, but gains it in the second
pass because D#5 has the MEM location incoming directly in its loc
list, rather than indirectly in a VALUE.

This incorrect binding enables other blocks to believe they have a
tentative binding for D#5 in some cycles, but others, still operating
on the early conclusion, believe there isn't, and they oscillate from
that.

Since we check for escaping MEMs in clobbers, we won't lose anything
relevant by dropping call-clobbered or overwritten MEMs in all onepart
variables, and this ensures the loc intersection operation in onepart
vars won't let a MEM through that wasn't present in earlier
iterations.

for  gcc/ChangeLog

	PR bootstrap/69123
	* var-tracking.c (drop_overlapping_mem_locs): Operate on all
	onepart vars.  Fix typo in comment.  Fix reversed condition in
	unshare test.
	(dataflow_set_remove_mem_locs): Operate on all onepart vars.

for gcc/testsuite/ChangeLog

	PR bootstrap/69123
	* gcc.dg/pr69123.c: New.
---
 gcc/testsuite/gcc.dg/pr69123.c |   95 ++++++++++++++++++++++++++++++++++++++++
 gcc/var-tracking.c             |   12 +++--
 2 files changed, 101 insertions(+), 6 deletions(-)
 create mode 100644 gcc/testsuite/gcc.dg/pr69123.c

On January 9, 2016 5:08:51 AM GMT+01:00, Alexandre Oliva <aoliva@redhat.com> wrote:
>Here are two patches related with PR69123, an infinite dataflow loop in
>VTA.
>
>The first non-comment hunk in var-tracking.c:drop_overlapping_mem_locs
>is what fixes the problem, but the other changes in the first patch fix
>similar problems that might cause other such oscillations.
>
>The second patch adds some more information to detailed vartrack dumps,
>avoiding short-circuiting of dataflow set compares and dumping added
>and
>removed locations for variables present in both sets.
>
>The patches are largely independent, but they were successfully
>regstrapped together on x86_64-linux-gnu and i686-linux-gnu.  They were
>only compile-tested separately.
>
>Ok to install?

OK.

Thanks,
Richard.

>
>[PR69123] fix handling of MEMs in VTA to avoid dataflow oscillation
>
>From: Alexandre Oliva <aoliva@redhat.com>
>
>The problem arises because we used to drop overwritten MEMs from loc
>lists of VALUEs, but not of other onepart variables, and it just so
>happens that, by doing so, block 6 in the testcase has no D#5 in its
>output in the first pass, because the MEM holding its (previous) value
>was correctly dropped from value 88:88, but gains it in the second
>pass because D#5 has the MEM location incoming directly in its loc
>list, rather than indirectly in a VALUE.
>
>This incorrect binding enables other blocks to believe they have a
>tentative binding for D#5 in some cycles, but others, still operating
>on the early conclusion, believe there isn't, and they oscillate from
>that.
>
>Since we check for escaping MEMs in clobbers, we won't lose anything
>relevant by dropping call-clobbered or overwritten MEMs in all onepart
>variables, and this ensures the loc intersection operation in onepart
>vars won't let a MEM through that wasn't present in earlier
>iterations.
>
>for  gcc/ChangeLog
>
>	PR bootstrap/69123
>	* var-tracking.c (drop_overlapping_mem_locs): Operate on all
>	onepart vars.  Fix typo in comment.  Fix reversed condition in
>	unshare test.
>	(dataflow_set_remove_mem_locs): Operate on all onepart vars.
>
>for gcc/testsuite/ChangeLog
>
>	PR bootstrap/69123
>	* gcc.dg/pr69123.c: New.
>---
>gcc/testsuite/gcc.dg/pr69123.c |   95
>++++++++++++++++++++++++++++++++++++++++
> gcc/var-tracking.c             |   12 +++--
> 2 files changed, 101 insertions(+), 6 deletions(-)
> create mode 100644 gcc/testsuite/gcc.dg/pr69123.c
>
>diff --git a/gcc/testsuite/gcc.dg/pr69123.c
>b/gcc/testsuite/gcc.dg/pr69123.c
>new file mode 100644
>index 0000000..0546e20
>--- /dev/null
>+++ b/gcc/testsuite/gcc.dg/pr69123.c
>@@ -0,0 +1,95 @@
>+/* { dg-do compile } */
>+/* { dg-options "-O3 -g" } */
>+
>+/* This was reduced from gcc/tree-vect-slp.c by H.J.Lu.  */
>+
>+struct xxx_def;
>+typedef xxx_def *xxx;
>+
>+union rtxxx
>+{
>+  const char *rt_str;
>+  xxx rt_xxx;
>+};
>+
>+struct xxx_def {
>+  union u {
>+    rtxxx fld[1];
>+  } u;
>+};
>+
>+extern xxx bar (void);
>+extern int foo1 (xxx);
>+
>+static inline xxx
>+foo2 (xxx arg0, xxx arg1)
>+{
>+  xxx rt;
>+  rt = bar ();
>+  (((rt)->u.fld[0]).rt_xxx) = arg0;
>+  (((rt)->u.fld[1]).rt_xxx) = arg1;
>+  return rt;
>+}
>+
>+static inline xxx
>+foo4 (const char *arg0 )
>+{
>+  xxx rt;
>+  rt = bar ();
>+  (((rt)->u.fld[0]).rt_str) = arg0;
>+  (((rt)->u.fld[1]).rt_xxx) = (xxx) 0;
>+  return rt;
>+}
>+
>+extern xxx foo5 (long);
>+
>+struct address_cost_data
>+{
>+  unsigned costs[2][2][2][2];
>+};
>+
>+void
>+get_address_cost (address_cost_data *data)
>+{
>+  unsigned acost;
>+  long i;
>+  long rat, off = 0;
>+  unsigned sym_p, var_p, off_p, rat_p;
>+  xxx addr, base;
>+  xxx reg0, reg1;
>+
>+  reg1 = bar ();
>+  addr = foo2 (reg1, (xxx) 0);
>+  rat = 1;
>+  acost = 0;
>+  reg0 = bar ();
>+  reg1 = bar ();
>+
>+  for (i = 0; i < 16; i++)
>+    {
>+      sym_p = i & 1;
>+      var_p = (i >> 1) & 1;
>+      off_p = (i >> 2) & 1;
>+      rat_p = (i >> 3) & 1;
>+
>+      addr = reg0;
>+      if (rat_p)
>+	addr = foo2 (addr, foo5 (rat)) ;
>+
>+      if (var_p)
>+	addr = foo2 (addr, reg1);
>+
>+      if (sym_p)
>+	base = foo4 ("");
>+      else if (off_p)
>+	base = foo5 (off);
>+      else
>+	base = (xxx) 0;
>+
>+      if (base)
>+	addr = foo2 (addr, base);
>+
>+      acost = foo1 (addr);
>+      data->costs[sym_p][var_p][off_p][rat_p] = acost;
>+    }
>+}
>diff --git a/gcc/var-tracking.c b/gcc/var-tracking.c
>index 634ebe0..a8931f3 100644
>--- a/gcc/var-tracking.c
>+++ b/gcc/var-tracking.c
>@@ -2224,7 +2224,7 @@ struct overlapping_mems
> };
> 
> /* Remove all MEMs that overlap with COMS->LOC from the location list
>-   of a hash table entry for a value.  COMS->ADDR must be a
>+   of a hash table entry for a onepart variable.  COMS->ADDR must be a
>    canonicalized form of COMS->LOC's address, and COMS->LOC must be
>    canonicalized itself.  */
> 
>@@ -2235,7 +2235,7 @@ drop_overlapping_mem_locs (variable **slot,
>overlapping_mems *coms)
>   rtx mloc = coms->loc, addr = coms->addr;
>   variable *var = *slot;
> 
>-  if (var->onepart == ONEPART_VALUE)
>+  if (var->onepart != NOT_ONEPART)
>     {
>       location_chain *loc, **locp;
>       bool changed = false;
>@@ -4682,11 +4682,11 @@ dataflow_set_preserve_mem_locs (variable
>**slot, dataflow_set *set)
> 	{
> 	  for (loc = var->var_part[0].loc_chain; loc; loc = loc->next)
> 	    {
>-	      /* We want to remove dying MEMs that doesn't refer to DECL.  */
>+	      /* We want to remove dying MEMs that don't refer to DECL.  */
> 	      if (GET_CODE (loc->loc) == MEM
> 		  && (MEM_EXPR (loc->loc) != decl
> 		      || INT_MEM_OFFSET (loc->loc) != 0)
>-		  && !mem_dies_at_call (loc->loc))
>+		  && mem_dies_at_call (loc->loc))
> 		break;
> 	      /* We want to move here MEMs that do refer to DECL.  */
> 	      else if (GET_CODE (loc->loc) == VALUE
>@@ -4769,14 +4769,14 @@ dataflow_set_preserve_mem_locs (variable
>**slot, dataflow_set *set)
> }
> 
> /* Remove all MEMs from the location list of a hash table entry for a
>-   value.  */
>+   onepart variable.  */
> 
> int
> dataflow_set_remove_mem_locs (variable **slot, dataflow_set *set)
> {
>   variable *var = *slot;
> 
>-  if (var->onepart == ONEPART_VALUE)
>+  if (var->onepart != NOT_ONEPART)
>     {
>       location_chain *loc, **locp;
>       bool changed = false;
>
>
>[PR69123] make dataflow_set_different details more verbose
>
>From: Alexandre Oliva <aoliva@redhat.com>
>
>for  gcc/ChangeLog
>
>	PR bootstrap/69123
>	* var-tracking.c (dump_onepart_variable_differences): New.
>	(dataflow_set_different): If a detailed dump is requested,
>	delay early returns and dump differences between onepart
>	variables present before and after, and added variables.
>---
>gcc/var-tracking.c |  113
>+++++++++++++++++++++++++++++++++++++++++++++++-----
> 1 file changed, 103 insertions(+), 10 deletions(-)
>
>diff --git a/gcc/var-tracking.c b/gcc/var-tracking.c
>index a5cca2b..634ebe0 100644
>--- a/gcc/var-tracking.c
>+++ b/gcc/var-tracking.c
>@@ -4921,6 +4921,63 @@ onepart_variable_different_p (variable *var1,
>variable *var2)
>   return lc1 != lc2;
> }
> 
>+/* Return true if one-part variables VAR1 and VAR2 are different.
>+   They must be in canonical order.  */
>+
>+static void
>+dump_onepart_variable_differences (variable *var1, variable *var2)
>+{
>+  location_chain *lc1, *lc2;
>+
>+  gcc_assert (var1 != var2);
>+  gcc_assert (dump_file);
>+  gcc_assert (dv_as_opaque (var1->dv) == dv_as_opaque (var2->dv));
>+  gcc_assert (var1->n_var_parts == 1
>+	      && var2->n_var_parts == 1);
>+
>+  lc1 = var1->var_part[0].loc_chain;
>+  lc2 = var2->var_part[0].loc_chain;
>+
>+  gcc_assert (lc1 && lc2);
>+
>+  while (lc1 && lc2)
>+    {
>+      switch (loc_cmp (lc1->loc, lc2->loc))
>+	{
>+	case -1:
>+	  fprintf (dump_file, "removed: ");
>+	  print_rtl_single (dump_file, lc1->loc);
>+	  lc1 = lc1->next;
>+	  continue;
>+	case 0:
>+	  break;
>+	case 1:
>+	  fprintf (dump_file, "added: ");
>+	  print_rtl_single (dump_file, lc2->loc);
>+	  lc2 = lc2->next;
>+	  continue;
>+	default:
>+	  gcc_unreachable ();
>+	}
>+      lc1 = lc1->next;
>+      lc2 = lc2->next;
>+    }
>+
>+  while (lc1)
>+    {
>+      fprintf (dump_file, "removed: ");
>+      print_rtl_single (dump_file, lc1->loc);
>+      lc1 = lc1->next;
>+    }
>+
>+  while (lc2)
>+    {
>+      fprintf (dump_file, "added: ");
>+      print_rtl_single (dump_file, lc2->loc);
>+      lc2 = lc2->next;
>+    }
>+}
>+
> /* Return true if variables VAR1 and VAR2 are different.  */
> 
> static bool
>@@ -4964,19 +5021,32 @@ dataflow_set_different (dataflow_set *old_set,
>dataflow_set *new_set)
> {
>   variable_iterator_type hi;
>   variable *var1;
>+  bool diffound = false;
>+  bool details = (dump_file && (dump_flags & TDF_DETAILS));
>+
>+#define RETRUE					\
>+  do						\
>+    {						\
>+      if (!details)				\
>+	return true;				\
>+      else					\
>+	diffound = true;			\
>+    }						\
>+  while (0)
> 
>   if (old_set->vars == new_set->vars)
>     return false;
> 
>   if (shared_hash_htab (old_set->vars)->elements ()
>       != shared_hash_htab (new_set->vars)->elements ())
>-    return true;
>+    RETRUE;
> 
>   FOR_EACH_HASH_TABLE_ELEMENT (*shared_hash_htab (old_set->vars),
> 			       var1, variable, hi)
>     {
>       variable_table_type *htab = shared_hash_htab (new_set->vars);
>variable *var2 = htab->find_with_hash (var1->dv, dv_htab_hash
>(var1->dv));
>+
>       if (!var2)
> 	{
> 	  if (dump_file && (dump_flags & TDF_DETAILS))
>@@ -4984,26 +5054,49 @@ dataflow_set_different (dataflow_set *old_set,
>dataflow_set *new_set)
>	      fprintf (dump_file, "dataflow difference found: removal of:\n");
> 	      dump_var (var1);
> 	    }
>-	  return true;
>+	  RETRUE;
> 	}
>-
>-      if (variable_different_p (var1, var2))
>+      else if (variable_different_p (var1, var2))
> 	{
>-	  if (dump_file && (dump_flags & TDF_DETAILS))
>+	  if (details)
> 	    {
> 	      fprintf (dump_file, "dataflow difference found: "
> 		       "old and new follow:\n");
> 	      dump_var (var1);
>+	      if (dv_onepart_p (var1->dv))
>+		dump_onepart_variable_differences (var1, var2);
> 	      dump_var (var2);
> 	    }
>-	  return true;
>+	  RETRUE;
> 	}
>     }
> 
>-  /* No need to traverse the second hashtab, if both have the same
>number
>-     of elements and the second one had all entries found in the first
>one,
>-     then it can't have any extra entries.  */
>-  return false;
>+  /* There's no need to traverse the second hashtab unless we want to
>+     print the details.  If both have the same number of elements and
>+     the second one had all entries found in the first one, then the
>+     second can't have any extra entries.  */
>+  if (!details)
>+    return diffound;
>+
>+  FOR_EACH_HASH_TABLE_ELEMENT (*shared_hash_htab (new_set->vars),
>+			       var1, variable, hi)
>+    {
>+      variable_table_type *htab = shared_hash_htab (old_set->vars);
>+      variable *var2 = htab->find_with_hash (var1->dv, dv_htab_hash
>(var1->dv));
>+      if (!var2)
>+	{
>+	  if (details)
>+	    {
>+	      fprintf (dump_file, "dataflow difference found: addition
>of:\n");
>+	      dump_var (var1);
>+	    }
>+	  RETRUE;
>+	}
>+    }
>+
>+#undef RETRUE
>+
>+  return diffound;
> }
> 
> /* Free the contents of dataflow set SET.  */

On Sat, Jan 09, 2016 at 02:08:51AM -0200, Alexandre Oliva wrote:
> Here are two patches related with PR69123, an infinite dataflow loop in
> VTA.
> 
> The first non-comment hunk in var-tracking.c:drop_overlapping_mem_locs
> is what fixes the problem, but the other changes in the first patch fix
> similar problems that might cause other such oscillations.
> 
> The second patch adds some more information to detailed vartrack dumps,
> avoiding short-circuiting of dataflow set compares and dumping added and
> removed locations for variables present in both sets.
> 
> The patches are largely independent, but they were successfully
> regstrapped together on x86_64-linux-gnu and i686-linux-gnu.  They were
> only compile-tested separately.
> 
> Ok to install?

Ok, thanks.

	Jakub