Use preferred mode for doloop IV [PR61837]

Currently, doloop.xx variable is using the type as niter which may be
shorter than word size.  For some targets, it would be better to use
word size type.  For example, on 64bit system, to access 32bit value,
subreg maybe used.  Then using 64bit type maybe better for niter if
it can be present in both 32bit and 64bit.

This patch add target hook to query preferred mode for doloop IV,
and update mode accordingly.

gcc/ChangeLog:

2021-07-29  Jiufu Guo  <guojiufu@linux.ibm.com>

	PR target/61837
	* config/rs6000/rs6000.c (TARGET_PREFERRED_DOLOOP_MODE): New hook.
	(rs6000_preferred_doloop_mode): New hook.
	* doc/tm.texi: Regenerate.
	* doc/tm.texi.in: Add hook preferred_doloop_mode.
	* target.def (preferred_doloop_mode): New hook.
	* targhooks.c (default_preferred_doloop_mode): New hook.
	* targhooks.h (default_preferred_doloop_mode): New hook.
	* tree-ssa-loop-ivopts.c (compute_doloop_base_on_mode): New function.
	(add_iv_candidate_for_doloop): Call targetm.preferred_doloop_mode
	and compute_doloop_base_on_mode.

gcc/testsuite/ChangeLog:

2021-07-29  Jiufu Guo  <guojiufu@linux.ibm.com>

	PR target/61837
	* gcc.target/powerpc/pr61837.c: New test.

1 files changed, 67 insertions, 2 deletions

diff --git a/gcc/tree-ssa-loop-ivopts.c b/gcc/tree-ssa-loop-ivopts.c
index 12a8a49..47b867f 100644
--- a/gcc/tree-ssa-loop-ivopts.c
+++ b/gcc/tree-ssa-loop-ivopts.c
@@ -5657,6 +5657,59 @@ relate_compare_use_with_all_cands (struct ivopts_data *data)
     }
 }
 
+/* If PREFERRED_MODE is suitable and profitable, use the preferred
+   PREFERRED_MODE to compute doloop iv base from niter: base = niter + 1.  */
+
+static tree
+compute_doloop_base_on_mode (machine_mode preferred_mode, tree niter,
+			     const widest_int &iterations_max)
+{
+  tree ntype = TREE_TYPE (niter);
+  tree pref_type = lang_hooks.types.type_for_mode (preferred_mode, 1);
+  if (!pref_type)
+    return fold_build2 (PLUS_EXPR, ntype, unshare_expr (niter),
+			build_int_cst (ntype, 1));
+
+  gcc_assert (TREE_CODE (pref_type) == INTEGER_TYPE);
+
+  int prec = TYPE_PRECISION (ntype);
+  int pref_prec = TYPE_PRECISION (pref_type);
+
+  tree base;
+
+  /* Check if the PREFERRED_MODED is able to present niter.  */
+  if (pref_prec > prec
+      || wi::ltu_p (iterations_max,
+		    widest_int::from (wi::max_value (pref_prec, UNSIGNED),
+				      UNSIGNED)))
+    {
+      /* No wrap, it is safe to use preferred type after niter + 1.  */
+      if (wi::ltu_p (iterations_max,
+		     widest_int::from (wi::max_value (prec, UNSIGNED),
+				       UNSIGNED)))
+	{
+	  /* This could help to optimize "-1 +1" pair when niter looks
+	     like "n-1": n is in original mode.  "base = (n - 1) + 1"
+	     in PREFERRED_MODED: it could be base = (PREFERRED_TYPE)n.  */
+	  base = fold_build2 (PLUS_EXPR, ntype, unshare_expr (niter),
+			      build_int_cst (ntype, 1));
+	  base = fold_convert (pref_type, base);
+	}
+
+      /* To avoid wrap, convert niter to preferred type before plus 1.  */
+      else
+	{
+	  niter = fold_convert (pref_type, niter);
+	  base = fold_build2 (PLUS_EXPR, pref_type, unshare_expr (niter),
+			      build_int_cst (pref_type, 1));
+	}
+    }
+  else
+    base = fold_build2 (PLUS_EXPR, ntype, unshare_expr (niter),
+			build_int_cst (ntype, 1));
+  return base;
+}
+
 /* Add one doloop dedicated IV candidate:
      - Base is (may_be_zero ? 1 : (niter + 1)).
      - Step is -1.  */
@@ -5688,8 +5741,20 @@ add_iv_candidate_for_doloop (struct ivopts_data *data)
 	return;
     }
 
-  tree base = fold_build2 (PLUS_EXPR, ntype, unshare_expr (niter),
-			   build_int_cst (ntype, 1));
+  machine_mode mode = TYPE_MODE (ntype);
+  machine_mode pref_mode = targetm.preferred_doloop_mode (mode);
+
+  tree base;
+  if (mode != pref_mode)
+    {
+      base = compute_doloop_base_on_mode (pref_mode, niter, niter_desc->max);
+      ntype = TREE_TYPE (base);
+    }
+  else
+    base = fold_build2 (PLUS_EXPR, ntype, unshare_expr (niter),
+			build_int_cst (ntype, 1));
+
+
   add_candidate (data, base, build_int_cst (ntype, -1), true, NULL, NULL, true);
 }