gcc/0050-Backport-phiopt-Optimize-x-1024-0-to-int-x-10-PR9769.patch

From c690da762e873d0f5c66ea084e420ba4842354a6 Mon Sep 17 00:00:00 2001
From: Jakub Jelinek <jakub@redhat.com>
Date: Wed, 4 Nov 2020 11:55:29 +0100
Subject: [PATCH 02/35] [Backport] phiopt: Optimize x ? 1024 : 0 to (int) x <<
 10 [PR97690]

Reference: https://gcc.gnu.org/git/gitweb.cgi?p=gcc.git;h=3e190757fa332d327bee27495f37beb01155cfab

The following patch generalizes the x ? 1 : 0 -> (int) x optimization
to handle also left shifts by constant.

During x86_64-linux and i686-linux bootstraps + regtests it triggered
in 1514 unique non-LTO -m64 cases (sort -u on log mentioning
filename, function name and shift count) and 1866 -m32 cases.

Unfortunately, the patch regresses (before the tests have been adjusted):
+FAIL: gcc.dg/tree-ssa/ssa-ccp-11.c scan-tree-dump-times optimized "if " 0
+FAIL: gcc.dg/vect/bb-slp-pattern-2.c -flto -ffat-lto-objects  scan-tree-dump-times slp1 "optimized: basic block" 1
+FAIL: gcc.dg/vect/bb-slp-pattern-2.c scan-tree-dump-times slp1 "optimized: basic block" 1
and in both cases it actually results in worse code.

> > We'd need some optimization that would go through all PHI edges and
> > compute if some use of the phi results don't actually compute a constant
> > across all the PHI edges - 1 & 0 and 0 & 1 is always 0.

> PRE should do this, IMHO only optimizing it at -O2 is fine.

> > Similarly, in the slp vectorization test there is:
> >      a[0] = b[0] ? 1 : 7;

> note this, carefully avoiding the already "optimized" b[0] ? 1 : 0 ...

> So the option is to put : 7 in the 2, 4 an 8 case as well.  The testcase
> wasn't added for any real-world case but is artificial I guess for
> COND_EXPR handling of invariants.

> But yeah, for things like SLP it means we eventually have to
> implement reverse transforms for all of this to make the lanes
> matching.  But that's true anyway for things like x + 1 vs. x + 0
> or x / 3 vs. x / 2 or other simplifications we do.

2020-11-04  Jakub Jelinek  <jakub@redhat.com>

	PR tree-optimization/97690
	* tree-ssa-phiopt.c (conditional_replacement): Also optimize
	cond ? pow2p_cst : 0 as ((type) cond) << cst.

	* gcc.dg/tree-ssa/phi-opt-22.c: New test.
	* gcc.dg/tree-ssa/ssa-ccp-11.c: Use -O2 instead of -O1.
	* gcc.dg/vect/bb-slp-pattern-2.c (foo): Use ? 2 : 7, ? 4 : 7 and
	? 8 : 7 instead of ? 2 : 0, ? 4 : 0, ? 8 : 0.
---
 gcc/testsuite/gcc.dg/tree-ssa/phi-opt-22.c   | 11 ++++++
 gcc/testsuite/gcc.dg/tree-ssa/ssa-ccp-11.c   |  2 +-
 gcc/testsuite/gcc.dg/vect/bb-slp-pattern-2.c |  6 ++--
 gcc/tree-ssa-phiopt.c                        | 38 ++++++++++++++------
 4 files changed, 43 insertions(+), 14 deletions(-)
 create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/phi-opt-22.c

diff --git a/gcc/testsuite/gcc.dg/tree-ssa/phi-opt-22.c b/gcc/testsuite/gcc.dg/tree-ssa/phi-opt-22.c
new file mode 100644
index 000000000..fd3706666
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/tree-ssa/phi-opt-22.c
@@ -0,0 +1,11 @@
+/* PR tree-optimization/97690 */
+/* { dg-do compile } */
+/* { dg-options "-O2 -fdump-tree-phiopt2" } */
+
+int foo (_Bool d) { return d ? 2 : 0; }
+int bar (_Bool d) { return d ? 1 : 0; }
+int baz (_Bool d) { return d ? -__INT_MAX__ - 1 : 0; }
+int qux (_Bool d) { return d ? 1024 : 0; }
+
+/* { dg-final { scan-tree-dump-not "if" "phiopt2" } } */
+/* { dg-final { scan-tree-dump-times " << " 3 "phiopt2" } } */
diff --git a/gcc/testsuite/gcc.dg/tree-ssa/ssa-ccp-11.c b/gcc/testsuite/gcc.dg/tree-ssa/ssa-ccp-11.c
index 36b8e7fc8..d70ea5a01 100644
--- a/gcc/testsuite/gcc.dg/tree-ssa/ssa-ccp-11.c
+++ b/gcc/testsuite/gcc.dg/tree-ssa/ssa-ccp-11.c
@@ -1,5 +1,5 @@
 /* { dg-do compile } */ 
-/* { dg-options "-O1 -fdump-tree-optimized" } */
+/* { dg-options "-O2 -fdump-tree-optimized" } */
 
 /* Test for CPROP across a DAG. */
 
diff --git a/gcc/testsuite/gcc.dg/vect/bb-slp-pattern-2.c b/gcc/testsuite/gcc.dg/vect/bb-slp-pattern-2.c
index d32cb7585..e64f0115a 100644
--- a/gcc/testsuite/gcc.dg/vect/bb-slp-pattern-2.c
+++ b/gcc/testsuite/gcc.dg/vect/bb-slp-pattern-2.c
@@ -13,13 +13,13 @@ foo (short * __restrict__ a, int * __restrict__ b, int stride)
   for (i = 0; i < N/stride; i++, a += stride, b += stride)
    {
      a[0] = b[0] ? 1 : 7;
-     a[1] = b[1] ? 2 : 0;
+     a[1] = b[1] ? 2 : 7;
      a[2] = b[2] ? 3 : 0;
-     a[3] = b[3] ? 4 : 0;
+     a[3] = b[3] ? 4 : 7;
      a[4] = b[4] ? 5 : 0;
      a[5] = b[5] ? 6 : 0;
      a[6] = b[6] ? 7 : 0;
-     a[7] = b[7] ? 8 : 0;
+     a[7] = b[7] ? 8 : 7;
    }
 }
 
diff --git a/gcc/tree-ssa-phiopt.c b/gcc/tree-ssa-phiopt.c
index 591b6435f..85587e8d1 100644
--- a/gcc/tree-ssa-phiopt.c
+++ b/gcc/tree-ssa-phiopt.c
@@ -753,7 +753,9 @@ conditional_replacement (basic_block cond_bb, basic_block middle_bb,
   gimple_stmt_iterator gsi;
   edge true_edge, false_edge;
   tree new_var, new_var2;
-  bool neg;
+  bool neg = false;
+  int shift = 0;
+  tree nonzero_arg;
 
   /* FIXME: Gimplification of complex type is too hard for now.  */
   /* We aren't prepared to handle vectors either (and it is a question
@@ -764,14 +766,22 @@ conditional_replacement (basic_block cond_bb, basic_block middle_bb,
 	   || POINTER_TYPE_P (TREE_TYPE (arg1))))
     return false;
 
-  /* The PHI arguments have the constants 0 and 1, or 0 and -1, then
-     convert it to the conditional.  */
-  if ((integer_zerop (arg0) && integer_onep (arg1))
-      || (integer_zerop (arg1) && integer_onep (arg0)))
-    neg = false;
-  else if ((integer_zerop (arg0) && integer_all_onesp (arg1))
-	   || (integer_zerop (arg1) && integer_all_onesp (arg0)))
+  /* The PHI arguments have the constants 0 and 1, or 0 and -1 or
+     0 and (1 << cst), then convert it to the conditional.  */
+  if (integer_zerop (arg0))
+    nonzero_arg = arg1;
+  else if (integer_zerop (arg1))
+    nonzero_arg = arg0;
+  else
+    return false;
+  if (integer_all_onesp (nonzero_arg))
     neg = true;
+  else if (integer_pow2p (nonzero_arg))
+    {
+      shift = tree_log2 (nonzero_arg);
+      if (shift && POINTER_TYPE_P (TREE_TYPE (nonzero_arg)))
+	return false;
+    }
   else
     return false;
 
@@ -783,12 +793,12 @@ conditional_replacement (basic_block cond_bb, basic_block middle_bb,
      falls through into BB.
 
      There is a single PHI node at the join point (BB) and its arguments
-     are constants (0, 1) or (0, -1).
+     are constants (0, 1) or (0, -1) or (0, (1 << shift)).
 
      So, given the condition COND, and the two PHI arguments, we can
      rewrite this PHI into non-branching code:
 
-       dest = (COND) or dest = COND'
+       dest = (COND) or dest = COND' or dest = (COND) << shift
 
      We use the condition as-is if the argument associated with the
      true edge has the value one or the argument associated with the
@@ -823,6 +833,14 @@ conditional_replacement (basic_block cond_bb, basic_block middle_bb,
       cond = fold_build1_loc (gimple_location (stmt),
                               NEGATE_EXPR, TREE_TYPE (cond), cond);
     }
+  else if (shift)
+    {
+      cond = fold_convert_loc (gimple_location (stmt),
+			       TREE_TYPE (result), cond);
+      cond = fold_build2_loc (gimple_location (stmt),
+			      LSHIFT_EXPR, TREE_TYPE (cond), cond,
+			      build_int_cst (integer_type_node, shift));
+    }
 
   /* Insert our new statements at the end of conditional block before the
      COND_STMT.  */
-- 
2.27.0.windows.1