aarch64: Generalise require_immediate_lane_index

require_immediate_lane_index previously hard-coded the assumption
that the group size is determined by the argument immediately before
the index.  However, for SME, there are cases where it should be
determined by an earlier argument instead.

gcc/
	* config/aarch64/aarch64-sve-builtins.h:
	(function_checker::require_immediate_lane_index): Add an argument
	for the index of the indexed vector argument.
	* config/aarch64/aarch64-sve-builtins.cc
	(function_checker::require_immediate_lane_index): Likewise.
	* config/aarch64/aarch64-sve-builtins-shapes.cc
	(ternary_bfloat_lane_base::check): Update accordingly.
	(ternary_qq_lane_base::check): Likewise.
	(binary_lane_def::check): Likewise.
	(binary_long_lane_def::check): Likewise.
	(ternary_lane_def::check): Likewise.
	(ternary_lane_rotate_def::check): Likewise.
	(ternary_long_lane_def::check): Likewise.
	(ternary_qq_lane_rotate_def::check): Likewise.

diff --git a/gcc/config/aarch64/aarch64-sve-builtins-shapes.cc b/gcc/config/aarch64/aarch64-sve-builtins-shapes.cc
index af816c4c9e705d9cc4bce5cc50481cb27e6a03a7..1646afc7a0d6b68d4e2840138d65d3ef328e4a06 100644 (file)
--- a/gcc/config/aarch64/aarch64-sve-builtins-shapes.cc
+++ b/gcc/config/aarch64/aarch64-sve-builtins-shapes.cc
@@ -941,7 +941,7 @@ struct ternary_bfloat_lane_base
   bool
   check (function_checker &c) const override
   {
-    return c.require_immediate_lane_index (3, N);
+    return c.require_immediate_lane_index (3, 2, N);
   }
 };
 
@@ -956,7 +956,7 @@ struct ternary_qq_lane_base
   bool
   check (function_checker &c) const override
   {
-    return c.require_immediate_lane_index (3, 4);
+    return c.require_immediate_lane_index (3, 0);
   }
 };
 
@@ -1123,7 +1123,7 @@ struct binary_lane_def : public overloaded_base<0>
   bool
   check (function_checker &c) const override
   {
-    return c.require_immediate_lane_index (2);
+    return c.require_immediate_lane_index (2, 1);
   }
 };
 SHAPE (binary_lane)
@@ -1162,7 +1162,7 @@ struct binary_long_lane_def : public overloaded_base<0>
   bool
   check (function_checker &c) const override
   {
-    return c.require_immediate_lane_index (2);
+    return c.require_immediate_lane_index (2, 1);
   }
 };
 SHAPE (binary_long_lane)
@@ -2817,7 +2817,7 @@ struct ternary_lane_def : public overloaded_base<0>
   bool
   check (function_checker &c) const override
   {
-    return c.require_immediate_lane_index (3);
+    return c.require_immediate_lane_index (3, 2);
   }
 };
 SHAPE (ternary_lane)
@@ -2845,7 +2845,7 @@ struct ternary_lane_rotate_def : public overloaded_base<0>
   bool
   check (function_checker &c) const override
   {
-    return (c.require_immediate_lane_index (3, 2)
+    return (c.require_immediate_lane_index (3, 2, 2)
            && c.require_immediate_one_of (4, 0, 90, 180, 270));
   }
 };
@@ -2868,7 +2868,7 @@ struct ternary_long_lane_def
   bool
   check (function_checker &c) const override
   {
-    return c.require_immediate_lane_index (3);
+    return c.require_immediate_lane_index (3, 2);
   }
 };
 SHAPE (ternary_long_lane)
@@ -2965,7 +2965,7 @@ struct ternary_qq_lane_rotate_def : public overloaded_base<0>
   bool
   check (function_checker &c) const override
   {
-    return (c.require_immediate_lane_index (3, 4)
+    return (c.require_immediate_lane_index (3, 0)
            && c.require_immediate_one_of (4, 0, 90, 180, 270));
   }
 };

diff --git a/gcc/config/aarch64/aarch64-sve-builtins.cc b/gcc/config/aarch64/aarch64-sve-builtins.cc
index c5aaf1bef173a1db809975c710f40bdc5015a2d4..e6ac81f6b52529d3f9da936313519c780140833d 100644 (file)
--- a/gcc/config/aarch64/aarch64-sve-builtins.cc
+++ b/gcc/config/aarch64/aarch64-sve-builtins.cc
@@ -2440,20 +2440,26 @@ function_checker::require_immediate_enum (unsigned int rel_argno, tree type)
   return false;
 }
 
-/* Check that argument REL_ARGNO is suitable for indexing argument
-   REL_ARGNO - 1, in groups of GROUP_SIZE elements.  REL_ARGNO counts
-   from the end of the predication arguments.  */
+/* The intrinsic conceptually divides vector argument REL_VEC_ARGNO into
+   groups of GROUP_SIZE elements.  Return true if argument REL_ARGNO is
+   a suitable constant index for selecting one of these groups.  The
+   selection happens within a 128-bit quadword, rather than the whole vector.
+
+   REL_ARGNO and REL_VEC_ARGNO count from the end of the predication
+   arguments.  */
 bool
 function_checker::require_immediate_lane_index (unsigned int rel_argno,
+                                               unsigned int rel_vec_argno,
                                                unsigned int group_size)
 {
   unsigned int argno = m_base_arg + rel_argno;
   if (!argument_exists_p (argno))
     return true;
 
-  /* Get the type of the previous argument.  tree_argument_type wants a
-     1-based number, whereas ARGNO is 0-based.  */
-  machine_mode mode = TYPE_MODE (type_argument_type (m_fntype, argno));
+  /* Get the type of the vector argument.  tree_argument_type wants a
+     1-based number, whereas VEC_ARGNO is 0-based.  */
+  unsigned int vec_argno = m_base_arg + rel_vec_argno;
+  machine_mode mode = TYPE_MODE (type_argument_type (m_fntype, vec_argno + 1));
   gcc_assert (VECTOR_MODE_P (mode));
   unsigned int nlanes = 128 / (group_size * GET_MODE_UNIT_BITSIZE (mode));
   return require_immediate_range (rel_argno, 0, nlanes - 1);

diff --git a/gcc/config/aarch64/aarch64-sve-builtins.h b/gcc/config/aarch64/aarch64-sve-builtins.h
index 1ac561dab4748a0fd4dfd9b50ad211a03fe8585e..2ca5b208efa5709a9f4e066ed43075c4c884bf77 100644 (file)
--- a/gcc/config/aarch64/aarch64-sve-builtins.h
+++ b/gcc/config/aarch64/aarch64-sve-builtins.h
@@ -463,7 +463,8 @@ public:
   bool require_immediate_either_or (unsigned int, HOST_WIDE_INT,
                                    HOST_WIDE_INT);
   bool require_immediate_enum (unsigned int, tree);
-  bool require_immediate_lane_index (unsigned int, unsigned int = 1);
+  bool require_immediate_lane_index (unsigned int, unsigned int,
+                                    unsigned int = 1);
   bool require_immediate_one_of (unsigned int, HOST_WIDE_INT, HOST_WIDE_INT,
                                 HOST_WIDE_INT, HOST_WIDE_INT);
   bool require_immediate_range (unsigned int, HOST_WIDE_INT, HOST_WIDE_INT);