AArch64: update costing for MLA by invariant

When determining issue rates we currently discount non-constant MLA accumulators
for Advanced SIMD but don't do it for the latency.

This means the costs for Advanced SIMD with a constant accumulator are wrong and
results in us costing SVE and Advanced SIMD the same.  This can cauze us to
vectorize with Advanced SIMD instead of SVE in some cases.

This patch adds the same discount for SVE and Scalar as we do for issue rate.

This gives a 5% improvement in fotonik3d_r in SPECCPU 2017 on large
Neoverse cores.

gcc/ChangeLog:

	* config/aarch64/aarch64.cc (aarch64_multiply_add_p): Update handling
	of constants.
	(aarch64_adjust_stmt_cost): Use it.
	(aarch64_vector_costs::count_ops): Likewise.
	(aarch64_vector_costs::add_stmt_cost): Pass vinfo to
	aarch64_adjust_stmt_cost.

1 files changed, 15 insertions, 9 deletions

diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc
index 5b8d8fa..53fbecb 100644
--- a/gcc/config/aarch64/aarch64.cc
+++ b/gcc/config/aarch64/aarch64.cc
@@ -16410,10 +16410,6 @@ aarch64_multiply_add_p (vec_info *vinfo, stmt_vec_info stmt_info,
   if (code != PLUS_EXPR && code != MINUS_EXPR)
     return false;
 
-  if (CONSTANT_CLASS_P (gimple_assign_rhs1 (assign))
-      || CONSTANT_CLASS_P (gimple_assign_rhs2 (assign)))
-    return false;
-
   for (int i = 1; i < 3; ++i)
     {
       tree rhs = gimple_op (assign, i);
@@ -16441,7 +16437,8 @@ aarch64_multiply_add_p (vec_info *vinfo, stmt_vec_info stmt_info,
 	    return false;
 	  def_stmt_info = vinfo->lookup_def (rhs);
 	  if (!def_stmt_info
-	      || STMT_VINFO_DEF_TYPE (def_stmt_info) == vect_external_def)
+	      || STMT_VINFO_DEF_TYPE (def_stmt_info) == vect_external_def
+	      || STMT_VINFO_DEF_TYPE (def_stmt_info) == vect_constant_def)
 	    return false;
 	}
 
@@ -16721,8 +16718,9 @@ aarch64_sve_adjust_stmt_cost (class vec_info *vinfo, vect_cost_for_stmt kind,
    and which when vectorized would operate on vector type VECTYPE.  Add the
    cost of any embedded operations.  */
 static fractional_cost
-aarch64_adjust_stmt_cost (vect_cost_for_stmt kind, stmt_vec_info stmt_info,
-			  tree vectype, fractional_cost stmt_cost)
+aarch64_adjust_stmt_cost (vec_info *vinfo, vect_cost_for_stmt kind,
+			  stmt_vec_info stmt_info, tree vectype,
+			  unsigned vec_flags, fractional_cost stmt_cost)
 {
   if (vectype)
     {
@@ -16745,6 +16743,14 @@ aarch64_adjust_stmt_cost (vect_cost_for_stmt kind, stmt_vec_info stmt_info,
 	  break;
 	}
 
+      gassign *assign = dyn_cast<gassign *> (STMT_VINFO_STMT (stmt_info));
+      if (assign && !vect_is_reduction (stmt_info))
+	{
+	  /* For MLA we need to reduce the cost since MLA is 1 instruction.  */
+	  if (aarch64_multiply_add_p (vinfo, stmt_info, vec_flags))
+	    return 0;
+	}
+
       if (kind == vector_stmt || kind == vec_to_scalar)
 	if (tree cmp_type = vect_embedded_comparison_type (stmt_info))
 	  {
@@ -17060,8 +17066,8 @@ aarch64_vector_costs::add_stmt_cost (int count, vect_cost_for_stmt kind,
     {
       /* Account for any extra "embedded" costs that apply additively
 	 to the base cost calculated above.  */
-      stmt_cost = aarch64_adjust_stmt_cost (kind, stmt_info, vectype,
-					    stmt_cost);
+      stmt_cost = aarch64_adjust_stmt_cost (m_vinfo, kind, stmt_info,
+					    vectype, m_vec_flags, stmt_cost);
 
       /* If we're recording a nonzero vector loop body cost for the
 	 innermost loop, also estimate the operations that would need