re PR tree-optimization/65930 (Reduction with sign-change not handled)

2019-10-29  Richard Biener  <rguenther@suse.de>

	PR tree-optimization/65930
	* tree-vect-loop.c (check_reduction_path): Relax single-use
	check allowing out-of-loop uses.
	(vect_is_simple_reduction): SLP reduction chains cannot have
	intermediate stmts used outside of the loop.
	(vect_create_epilog_for_reduction): The adjustment might need
	to be converted.
	(vectorizable_reduction): Annotate live stmts of the reduction
	chain with STMT_VINFO_REDUC_DEF.
	* tree-vect-stms.c (process_use): Remove no longer true asserts.

	* gcc.dg/vect/pr65930-1.c: New testcase.

From-SVN: r277566

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index a9a22934c1267755c2ba89319f2cb286b84179a2..77c8b185c6da674ebcfc8cf8271e1fd78d61c1ec 100644 (file)
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,16 @@
+2019-10-29  Richard Biener  <rguenther@suse.de>
+
+       PR tree-optimization/65930
+       * tree-vect-loop.c (check_reduction_path): Relax single-use
+       check allowing out-of-loop uses.
+       (vect_is_simple_reduction): SLP reduction chains cannot have
+       intermediate stmts used outside of the loop.
+       (vect_create_epilog_for_reduction): The adjustment might need
+       to be converted.
+       (vectorizable_reduction): Annotate live stmts of the reduction
+       chain with STMT_VINFO_REDUC_DEF.
+       * tree-vect-stms.c (process_use): Remove no longer true asserts.
+
 2019-10-29  Richard Sandiford  <richard.sandiford@arm.com>
 
        * calls.c (pass_by_reference): Leave the target to decide whether

diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog
index c1df721a4a7a3b1e088dc2b62c71cbbb2ba34430..f4885b6e7bd41cdd81196734f42048de3c9725f1 100644 (file)
--- a/gcc/testsuite/ChangeLog
+++ b/gcc/testsuite/ChangeLog
@@ -1,3 +1,8 @@
+2019-10-29  Richard Biener  <rguenther@suse.de>
+
+       PR tree-optimization/65930
+       * gcc.dg/vect/pr65930-1.c: New testcase.
+
 2019-10-29  Richard Sandiford  <richard.sandiford@arm.com>
            Kugan Vivekanandarajah  <kugan.vivekanandarajah@linaro.org>
            Prathamesh Kulkarni  <prathamesh.kulkarni@linaro.org>

diff --git a/gcc/testsuite/gcc.dg/vect/pr65930-1.c b/gcc/testsuite/gcc.dg/vect/pr65930-1.c
new file mode 100644 (file)
index 0000000..895fbf8
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/vect/pr65930-1.c
@@ -0,0 +1,26 @@
+/* { dg-require-effective-target vect_int } */
+
+#include "tree-vect.h"
+
+unsigned __attribute__((noipa))
+bar (unsigned int *x)
+{
+  int sum = 4;
+  x = __builtin_assume_aligned (x, __BIGGEST_ALIGNMENT__);
+  for (int i = 0; i < 16; ++i)
+    sum += x[i];
+  return sum;
+}
+
+int
+main()
+{
+  static int a[16] __attribute__((aligned(__BIGGEST_ALIGNMENT__)))
+    = { 1, 3, 5, 8, 9, 10, 17, 18, 23, 29, 30, 55, 42, 2, 3, 1 };
+  check_vect ();
+  if (bar (a) != 260)
+    abort ();
+  return 0;
+}
+
+/* { dg-final { scan-tree-dump "vectorized 1 loops" "vect" } } */

diff --git a/gcc/tree-vect-loop.c b/gcc/tree-vect-loop.c
index d11a1f9f9f54215b9929e6ceb1c78ac17915143b..3f43fe6c3af2d8e5c3521a31726a08e4c3d2e0d2 100644 (file)
--- a/gcc/tree-vect-loop.c
+++ b/gcc/tree-vect-loop.c
@@ -2678,8 +2678,7 @@ pop:
     {
       gimple *use_stmt = USE_STMT (path[i].second);
       tree op = USE_FROM_PTR (path[i].second);
-      if (! has_single_use (op)
-         || ! is_gimple_assign (use_stmt)
+      if (! is_gimple_assign (use_stmt)
          /* The following make sure we can compute the operand index
             easily plus it mostly disallows chaining via COND_EXPR condition
             operands.  */
@@ -2690,7 +2689,21 @@ pop:
          fail = true;
          break;
        }
-      enum tree_code use_code = gimple_assign_rhs_code (use_stmt);
+      /* Check there's only a single stmt the op is used on inside
+         of the loop.  */
+      imm_use_iterator imm_iter;
+      gimple *op_use_stmt;
+      unsigned cnt = 0;
+      FOR_EACH_IMM_USE_STMT (op_use_stmt, imm_iter, op)
+       if (!is_gimple_debug (op_use_stmt)
+           && flow_bb_inside_loop_p (loop, gimple_bb (op_use_stmt)))
+         cnt++;
+      if (cnt != 1)
+       {
+         fail = true;
+         break;
+       }
+      tree_code use_code = gimple_assign_rhs_code (use_stmt);
       if (use_code == MINUS_EXPR)
        {
          use_code = PLUS_EXPR;
@@ -2922,7 +2935,10 @@ vect_is_simple_reduction (loop_vec_info loop_info, stmt_vec_info phi_info,
       for (i = path.length () - 1; i >= 1; --i)
        {
          gimple *stmt = USE_STMT (path[i].second);
-         if (gimple_assign_rhs_code (stmt) != code)
+         if (gimple_assign_rhs_code (stmt) != code
+             /* We can only handle the final value in epilogue
+                generation for reduction chains.  */
+             || (i != 1 && !has_single_use (gimple_assign_lhs (stmt))))
            is_slp_reduc = false;
          stmt_vec_info stmt_info = loop_info->lookup_stmt (stmt);
          STMT_VINFO_REDUC_IDX (stmt_info)
@@ -4119,11 +4135,11 @@ vect_create_epilog_for_reduction (stmt_vec_info stmt_info,
   stmt_vec_info phi_info;
   gimple_stmt_iterator exit_gsi;
   tree vec_dest;
-  tree new_temp = NULL_TREE, new_dest, new_name, new_scalar_dest;
+  tree new_temp = NULL_TREE, new_name, new_scalar_dest;
   gimple *epilog_stmt = NULL;
   gimple *exit_phi;
   tree bitsize;
-  tree expr, def;
+  tree def;
   tree orig_name, scalar_result;
   imm_use_iterator imm_iter, phi_imm_iter;
   use_operand_p use_p, phi_use_p;
@@ -5048,25 +5064,26 @@ vect_create_epilog_for_reduction (stmt_vec_info stmt_info,
   if (adjustment_def)
     {
       gcc_assert (!slp_reduc);
+      gimple_seq stmts = NULL;
       if (nested_in_vect_loop)
        {
           new_phi = new_phis[0];
-         gcc_assert (TREE_CODE (TREE_TYPE (adjustment_def)) == VECTOR_TYPE);
-         expr = build2 (code, vectype, PHI_RESULT (new_phi), adjustment_def);
-         new_dest = vect_create_destination_var (scalar_dest, vectype);
+         gcc_assert (VECTOR_TYPE_P (TREE_TYPE (adjustment_def)));
+         adjustment_def = gimple_convert (&stmts, vectype, adjustment_def);
+         new_temp = gimple_build (&stmts, code, vectype,
+                                  PHI_RESULT (new_phi), adjustment_def);
        }
       else
        {
           new_temp = scalar_results[0];
          gcc_assert (TREE_CODE (TREE_TYPE (adjustment_def)) != VECTOR_TYPE);
-         expr = build2 (code, scalar_type, new_temp, adjustment_def);
-         new_dest = vect_create_destination_var (scalar_dest, scalar_type);
+         adjustment_def = gimple_convert (&stmts, scalar_type, adjustment_def);
+         new_temp = gimple_build (&stmts, code, scalar_type,
+                                  new_temp, adjustment_def);
        }
 
-      epilog_stmt = gimple_build_assign (new_dest, expr);
-      new_temp = make_ssa_name (new_dest, epilog_stmt);
-      gimple_assign_set_lhs (epilog_stmt, new_temp);
-      gsi_insert_before (&exit_gsi, epilog_stmt, GSI_SAME_STMT);
+      epilog_stmt = gimple_seq_last_stmt (stmts);
+      gsi_insert_seq_before (&exit_gsi, stmts, GSI_SAME_STMT);
       if (nested_in_vect_loop)
         {
          stmt_vec_info epilog_stmt_info = loop_vinfo->add_stmt (epilog_stmt);
@@ -5742,6 +5759,10 @@ vectorizable_reduction (stmt_vec_info stmt_info, slp_tree slp_node,
        }
       if (!REDUC_GROUP_FIRST_ELEMENT (def))
        only_slp_reduc_chain = false;
+      /* ???  For epilogue generation live members of the chain need
+         to point back to the PHI for info_for_reduction to work.  */
+      if (STMT_VINFO_LIVE_P (def))
+       STMT_VINFO_REDUC_DEF (def) = phi_info;
       reduc_def = gimple_op (def->stmt, 1 + STMT_VINFO_REDUC_IDX (def));
       reduc_chain_length++;
     }
@@ -7443,10 +7464,10 @@ vectorizable_live_operation (stmt_vec_info stmt_info,
 
   gcc_assert (STMT_VINFO_LIVE_P (stmt_info));
 
-  /* The last stmt of a reduction is live and vectorized via
+  /* If a stmt of a reduction is live, vectorize it via
      vect_create_epilog_for_reduction.  vectorizable_reduction assessed
      validity so just trigger the transform here.  */
-  if (STMT_VINFO_REDUC_DEF (vect_orig_stmt (stmt_info)))
+  if (STMT_VINFO_REDUC_DEF (stmt_info))
     {
       if (!vec_stmt_p)
        return true;

diff --git a/gcc/tree-vect-stmts.c b/gcc/tree-vect-stmts.c
index 36821cec2ee8ee77ab98d93bdf7c042cc6f9d09f..abc827f738ee45afc3990c8605add79466959643 100644 (file)
--- a/gcc/tree-vect-stmts.c
+++ b/gcc/tree-vect-stmts.c
@@ -457,7 +457,6 @@ process_use (stmt_vec_info stmt_vinfo, tree use, loop_vec_info loop_vinfo,
             bool force)
 {
   stmt_vec_info dstmt_vinfo;
-  basic_block bb, def_bb;
   enum vect_def_type dt;
 
   /* case 1: we are only interested in uses that need to be vectorized.  Uses
@@ -473,28 +472,8 @@ process_use (stmt_vec_info stmt_vinfo, tree use, loop_vec_info loop_vinfo,
   if (!dstmt_vinfo)
     return opt_result::success ();
 
-  def_bb = gimple_bb (dstmt_vinfo->stmt);
-
-  /* case 2: A reduction phi (STMT) defined by a reduction stmt (DSTMT_VINFO).
-     DSTMT_VINFO must have already been processed, because this should be the
-     only way that STMT, which is a reduction-phi, was put in the worklist,
-     as there should be no other uses for DSTMT_VINFO in the loop.  So we just
-     check that everything is as expected, and we are done.  */
-  bb = gimple_bb (stmt_vinfo->stmt);
-  if (gimple_code (stmt_vinfo->stmt) == GIMPLE_PHI
-      && STMT_VINFO_DEF_TYPE (stmt_vinfo) == vect_reduction_def
-      && gimple_code (dstmt_vinfo->stmt) != GIMPLE_PHI
-      && STMT_VINFO_DEF_TYPE (dstmt_vinfo) == vect_reduction_def
-      && bb->loop_father == def_bb->loop_father)
-    {
-      if (dump_enabled_p ())
-       dump_printf_loc (MSG_NOTE, vect_location,
-                         "reduc-stmt defining reduc-phi in the same nest.\n");
-      gcc_assert (STMT_VINFO_RELEVANT (dstmt_vinfo) < vect_used_by_reduction);
-      gcc_assert (STMT_VINFO_LIVE_P (dstmt_vinfo)
-                 || STMT_VINFO_RELEVANT (dstmt_vinfo) > vect_unused_in_scope);
-      return opt_result::success ();
-    }
+  basic_block def_bb = gimple_bb (dstmt_vinfo->stmt);
+  basic_block bb = gimple_bb (stmt_vinfo->stmt);
 
   /* case 3a: outer-loop stmt defining an inner-loop stmt:
        outer-loop-header-bb: