RISC-V: Support segment intrinsics

Add segment load/store intrinsics:
riscv-non-isa/rvv-intrinsic-doc#198

gcc/ChangeLog:

	* config/riscv/riscv-vector-builtins-bases.cc (fold_fault_load):
	New function.
	(class vlseg): New class.
	(class vsseg): Ditto.
	(class vlsseg): Ditto.
	(class vssseg): Ditto.
	(class seg_indexed_load): Ditto.
	(class seg_indexed_store): Ditto.
	(class vlsegff): Ditto.
	(BASE): Ditto.
	* config/riscv/riscv-vector-builtins-bases.h: Ditto.
	* config/riscv/riscv-vector-builtins-functions.def (vlseg):
	Ditto.
	(vsseg): Ditto.
	(vlsseg): Ditto.
	(vssseg): Ditto.
	(vluxseg): Ditto.
	(vloxseg): Ditto.
	(vsuxseg): Ditto.
	(vsoxseg): Ditto.
	(vlsegff): Ditto.
	* config/riscv/riscv-vector-builtins-shapes.cc (struct
	seg_loadstore_def): Ditto.
	(struct seg_indexed_loadstore_def): Ditto.
	(struct seg_fault_load_def): Ditto.
	(SHAPE): Ditto.
	* config/riscv/riscv-vector-builtins-shapes.h: Ditto.
	* config/riscv/riscv-vector-builtins.cc
	(function_builder::append_nf): New function.
	* config/riscv/riscv-vector-builtins.def (vfloat32m1x2_t):
	Change ptr from double into float.
	(vfloat32m1x3_t): Ditto.
	(vfloat32m1x4_t): Ditto.
	(vfloat32m1x5_t): Ditto.
	(vfloat32m1x6_t): Ditto.
	(vfloat32m1x7_t): Ditto.
	(vfloat32m1x8_t): Ditto.
	(vfloat32m2x2_t): Ditto.
	(vfloat32m2x3_t): Ditto.
	(vfloat32m2x4_t): Ditto.
	(vfloat32m4x2_t): Ditto.
	* config/riscv/riscv-vector-builtins.h: Add segment intrinsics.
	* config/riscv/riscv-vsetvl.cc (fault_first_load_p): Adapt for
	segment ff load.
	* config/riscv/riscv.md: Add segment instructions.
	* config/riscv/vector-iterators.md: Support segment intrinsics.
	* config/riscv/vector.md (@pred_unit_strided_load<mode>): New
	pattern.
	(@pred_unit_strided_store<mode>): Ditto.
	(@pred_strided_load<mode>): Ditto.
	(@pred_strided_store<mode>): Ditto.
	(@pred_fault_load<mode>): Ditto.
	(@pred_indexed_<order>load<V1T:mode><V1I:mode>): Ditto.
	(@pred_indexed_<order>load<V2T:mode><V2I:mode>): Ditto.
	(@pred_indexed_<order>load<V4T:mode><V4I:mode>): Ditto.
	(@pred_indexed_<order>load<V8T:mode><V8I:mode>): Ditto.
	(@pred_indexed_<order>load<V16T:mode><V16I:mode>): Ditto.
	(@pred_indexed_<order>load<V32T:mode><V32I:mode>): Ditto.
	(@pred_indexed_<order>load<V64T:mode><V64I:mode>): Ditto.
	(@pred_indexed_<order>store<V1T:mode><V1I:mode>): Ditto.
	(@pred_indexed_<order>store<V2T:mode><V2I:mode>): Ditto.
	(@pred_indexed_<order>store<V4T:mode><V4I:mode>): Ditto.
	(@pred_indexed_<order>store<V8T:mode><V8I:mode>): Ditto.
	(@pred_indexed_<order>store<V16T:mode><V16I:mode>): Ditto.
	(@pred_indexed_<order>store<V32T:mode><V32I:mode>): Ditto.
	(@pred_indexed_<order>store<V64T:mode><V64I:mode>): Ditto.

Signed-off-by: Ju-Zhe Zhong <juzhe.zhong@rivai.ai>

gcc/config/riscv/riscv-vector-builtins-bases.cc

@@ -58,6 +58,54 @@ enum lst_type
   LST_INDEXED,
 };
 
+/* Helper function to fold vleff and vlsegff.  */
+static gimple *
+fold_fault_load (gimple_folder &f)
+{
+  /* fold fault_load (const *base, size_t *new_vl, size_t vl)
+
+     ====> fault_load (const *base, size_t vl)
+	   new_vl = MEM_REF[read_vl ()].  */
+
+  auto_vec<tree> vargs (gimple_call_num_args (f.call) - 1);
+
+  for (unsigned i = 0; i < gimple_call_num_args (f.call); i++)
+    {
+      /* Exclude size_t *new_vl argument.  */
+      if (i == gimple_call_num_args (f.call) - 2)
+	continue;
+
+      vargs.quick_push (gimple_call_arg (f.call, i));
+    }
+
+  gimple *repl = gimple_build_call_vec (gimple_call_fn (f.call), vargs);
+  gimple_call_set_lhs (repl, f.lhs);
+
+  /* Handle size_t *new_vl by read_vl.  */
+  tree new_vl = gimple_call_arg (f.call, gimple_call_num_args (f.call) - 2);
+  if (integer_zerop (new_vl))
+    {
+      /* This case happens when user passes the nullptr to new_vl argument.
+	 In this case, we just need to ignore the new_vl argument and return
+	 fault_load instruction directly. */
+      return repl;
+    }
+
+  tree tmp_var = create_tmp_var (size_type_node, "new_vl");
+  tree decl = get_read_vl_decl ();
+  gimple *g = gimple_build_call (decl, 0);
+  gimple_call_set_lhs (g, tmp_var);
+  tree indirect
+    = fold_build2 (MEM_REF, size_type_node,
+		   gimple_call_arg (f.call, gimple_call_num_args (f.call) - 2),
+		   build_int_cst (build_pointer_type (size_type_node), 0));
+  gassign *assign = gimple_build_assign (indirect, tmp_var);
+
+  gsi_insert_after (f.gsi, assign, GSI_SAME_STMT);
+  gsi_insert_after (f.gsi, g, GSI_SAME_STMT);
+  return repl;
+}
+
 /* Implements vsetvl<mode> && vsetvlmax<mode>.  */
 template<bool VLMAX_P>
 class vsetvl : public function_base
@@ -1655,49 +1703,7 @@ class vleff : public function_base
 
   gimple *fold (gimple_folder &f) const override
   {
-    /* fold vleff (const *base, size_t *new_vl, size_t vl)
-
-       ====> vleff (const *base, size_t vl)
-	     new_vl = MEM_REF[read_vl ()].  */
-
-    auto_vec<tree> vargs (gimple_call_num_args (f.call) - 1);
-
-    for (unsigned i = 0; i < gimple_call_num_args (f.call); i++)
-      {
-	/* Exclude size_t *new_vl argument.  */
-	if (i == gimple_call_num_args (f.call) - 2)
-	  continue;
-
-	vargs.quick_push (gimple_call_arg (f.call, i));
-      }
-
-    gimple *repl = gimple_build_call_vec (gimple_call_fn (f.call), vargs);
-    gimple_call_set_lhs (repl, f.lhs);
-
-    /* Handle size_t *new_vl by read_vl.  */
-    tree new_vl = gimple_call_arg (f.call, gimple_call_num_args (f.call) - 2);
-    if (integer_zerop (new_vl))
-      {
-	/* This case happens when user passes the nullptr to new_vl argument.
-	   In this case, we just need to ignore the new_vl argument and return
-	   vleff instruction directly. */
-	return repl;
-      }
-
-    tree tmp_var = create_tmp_var (size_type_node, "new_vl");
-    tree decl = get_read_vl_decl ();
-    gimple *g = gimple_build_call (decl, 0);
-    gimple_call_set_lhs (g, tmp_var);
-    tree indirect
-      = fold_build2 (MEM_REF, size_type_node,
-		     gimple_call_arg (f.call,
-				      gimple_call_num_args (f.call) - 2),
-		     build_int_cst (build_pointer_type (size_type_node), 0));
-    gassign *assign = gimple_build_assign (indirect, tmp_var);
-
-    gsi_insert_after (f.gsi, assign, GSI_SAME_STMT);
-    gsi_insert_after (f.gsi, g, GSI_SAME_STMT);
-    return repl;
+    return fold_fault_load (f);
   }
 
   rtx expand (function_expander &e) const override
@@ -1722,6 +1728,166 @@ class vlenb : public function_base
   }
 };
 
+/* Implements vlseg.v.  */
+class vlseg : public function_base
+{
+public:
+  unsigned int call_properties (const function_instance &) const override
+  {
+    return CP_READ_MEMORY;
+  }
+
+  bool can_be_overloaded_p (enum predication_type_index pred) const override
+  {
+    return pred != PRED_TYPE_none && pred != PRED_TYPE_mu;
+  }
+
+  rtx expand (function_expander &e) const override
+  {
+    return e.use_exact_insn (
+      code_for_pred_unit_strided_load (e.vector_mode ()));
+  }
+};
+
+/* Implements vsseg.v.  */
+class vsseg : public function_base
+{
+public:
+  bool apply_tail_policy_p () const override { return false; }
+  bool apply_mask_policy_p () const override { return false; }
+
+  unsigned int call_properties (const function_instance &) const override
+  {
+    return CP_WRITE_MEMORY;
+  }
+
+  bool can_be_overloaded_p (enum predication_type_index) const override
+  {
+    return true;
+  }
+
+  rtx expand (function_expander &e) const override
+  {
+    return e.use_exact_insn (
+      code_for_pred_unit_strided_store (e.vector_mode ()));
+  }
+};
+
+/* Implements vlsseg.v.  */
+class vlsseg : public function_base
+{
+public:
+  unsigned int call_properties (const function_instance &) const override
+  {
+    return CP_READ_MEMORY;
+  }
+
+  bool can_be_overloaded_p (enum predication_type_index pred) const override
+  {
+    return pred != PRED_TYPE_none && pred != PRED_TYPE_mu;
+  }
+
+  rtx expand (function_expander &e) const override
+  {
+    return e.use_exact_insn (
+      code_for_pred_strided_load (e.vector_mode ()));
+  }
+};
+
+/* Implements vssseg.v.  */
+class vssseg : public function_base
+{
+public:
+  bool apply_tail_policy_p () const override { return false; }
+  bool apply_mask_policy_p () const override { return false; }
+
+  unsigned int call_properties (const function_instance &) const override
+  {
+    return CP_WRITE_MEMORY;
+  }
+
+  bool can_be_overloaded_p (enum predication_type_index) const override
+  {
+    return true;
+  }
+
+  rtx expand (function_expander &e) const override
+  {
+    return e.use_exact_insn (
+      code_for_pred_strided_store (e.vector_mode ()));
+  }
+};
+
+template<int UNSPEC>
+class seg_indexed_load : public function_base
+{
+public:
+  unsigned int call_properties (const function_instance &) const override
+  {
+    return CP_READ_MEMORY;
+  }
+
+  bool can_be_overloaded_p (enum predication_type_index) const override
+  {
+    return true;
+  }
+
+  rtx expand (function_expander &e) const override
+  {
+    return e.use_exact_insn (
+      code_for_pred_indexed_load (UNSPEC, e.vector_mode (), e.index_mode ()));
+  }
+};
+
+template<int UNSPEC>
+class seg_indexed_store : public function_base
+{
+public:
+  bool apply_tail_policy_p () const override { return false; }
+  bool apply_mask_policy_p () const override { return false; }
+
+  unsigned int call_properties (const function_instance &) const override
+  {
+    return CP_WRITE_MEMORY;
+  }
+
+  bool can_be_overloaded_p (enum predication_type_index) const override
+  {
+    return true;
+  }
+
+  rtx expand (function_expander &e) const override
+  {
+    return e.use_exact_insn (
+      code_for_pred_indexed_store (UNSPEC, e.vector_mode (), e.index_mode ()));
+  }
+};
+
+/* Implements vlsegff.v.  */
+class vlsegff : public function_base
+{
+public:
+  unsigned int call_properties (const function_instance &) const override
+  {
+    return CP_READ_MEMORY | CP_WRITE_CSR;
+  }
+
+  bool can_be_overloaded_p (enum predication_type_index pred) const override
+  {
+    return pred != PRED_TYPE_none && pred != PRED_TYPE_mu;
+  }
+
+  gimple *fold (gimple_folder &f) const override
+  {
+    return fold_fault_load (f);
+  }
+
+  rtx expand (function_expander &e) const override
+  {
+    return e.use_exact_insn (code_for_pred_fault_load (e.vector_mode ()));
+  }
+};
+
 static CONSTEXPR const vsetvl<false> vsetvl_obj;
 static CONSTEXPR const vsetvl<true> vsetvlmax_obj;
 static CONSTEXPR const loadstore<false, LST_UNIT_STRIDE, false> vle_obj;
@@ -1933,6 +2099,15 @@ static CONSTEXPR const vget vget_obj;
 static CONSTEXPR const read_vl read_vl_obj;
 static CONSTEXPR const vleff vleff_obj;
 static CONSTEXPR const vlenb vlenb_obj;
+static CONSTEXPR const vlseg vlseg_obj;
+static CONSTEXPR const vsseg vsseg_obj;
+static CONSTEXPR const vlsseg vlsseg_obj;
+static CONSTEXPR const vssseg vssseg_obj;
+static CONSTEXPR const seg_indexed_load<UNSPEC_UNORDERED> vluxseg_obj;
+static CONSTEXPR const seg_indexed_load<UNSPEC_ORDERED> vloxseg_obj;
+static CONSTEXPR const seg_indexed_store<UNSPEC_UNORDERED> vsuxseg_obj;
+static CONSTEXPR const seg_indexed_store<UNSPEC_ORDERED> vsoxseg_obj;
+static CONSTEXPR const vlsegff vlsegff_obj;
 
 /* Declare the function base NAME, pointing it to an instance
    of class <NAME>_obj.  */
@@ -2150,5 +2325,14 @@ BASE (vget)
 BASE (read_vl)
 BASE (vleff)
 BASE (vlenb)
+BASE (vlseg)
+BASE (vsseg)
+BASE (vlsseg)
+BASE (vssseg)
+BASE (vluxseg)
+BASE (vloxseg)
+BASE (vsuxseg)
+BASE (vsoxseg)
+BASE (vlsegff)
 
 } // end namespace riscv_vector

gcc/config/riscv/riscv-vector-builtins-bases.h

@@ -241,6 +241,15 @@ extern const function_base *const vget;
 extern const function_base *const read_vl;
 extern const function_base *const vleff;
 extern const function_base *const vlenb;
+extern const function_base *const vlseg;
+extern const function_base *const vsseg;
+extern const function_base *const vlsseg;
+extern const function_base *const vssseg;
+extern const function_base *const vluxseg;
+extern const function_base *const vloxseg;
+extern const function_base *const vsuxseg;
+extern const function_base *const vsoxseg;
+extern const function_base *const vlsegff;
 }
 
 } // end namespace riscv_vector

gcc/config/riscv/riscv-vector-builtins-functions.def

@@ -536,5 +536,26 @@ DEF_RVV_FUNCTION (vget, vget, none_preds, all_v_vget_lmul4_x2_ops)
 // Tuple types
 DEF_RVV_FUNCTION (vset, vset, none_preds, all_v_vset_tuple_ops)
 DEF_RVV_FUNCTION (vget, vget, none_preds, all_v_vget_tuple_ops)
+DEF_RVV_FUNCTION (vlseg, seg_loadstore, full_preds, tuple_v_scalar_const_ptr_ops)
+DEF_RVV_FUNCTION (vsseg, seg_loadstore, none_m_preds, tuple_v_scalar_ptr_ops)
+DEF_RVV_FUNCTION (vlsseg, seg_loadstore, full_preds, tuple_v_scalar_const_ptr_ptrdiff_ops)
+DEF_RVV_FUNCTION (vssseg, seg_loadstore, none_m_preds, tuple_v_scalar_ptr_ptrdiff_ops)
+DEF_RVV_FUNCTION (vluxseg, seg_indexed_loadstore, full_preds, tuple_v_scalar_const_ptr_eew8_index_ops)
+DEF_RVV_FUNCTION (vluxseg, seg_indexed_loadstore, full_preds, tuple_v_scalar_const_ptr_eew16_index_ops)
+DEF_RVV_FUNCTION (vluxseg, seg_indexed_loadstore, full_preds, tuple_v_scalar_const_ptr_eew32_index_ops)
+DEF_RVV_FUNCTION (vluxseg, seg_indexed_loadstore, full_preds, tuple_v_scalar_const_ptr_eew64_index_ops)
+DEF_RVV_FUNCTION (vloxseg, seg_indexed_loadstore, full_preds, tuple_v_scalar_const_ptr_eew8_index_ops)
+DEF_RVV_FUNCTION (vloxseg, seg_indexed_loadstore, full_preds, tuple_v_scalar_const_ptr_eew16_index_ops)
+DEF_RVV_FUNCTION (vloxseg, seg_indexed_loadstore, full_preds, tuple_v_scalar_const_ptr_eew32_index_ops)
+DEF_RVV_FUNCTION (vloxseg, seg_indexed_loadstore, full_preds, tuple_v_scalar_const_ptr_eew64_index_ops)
+DEF_RVV_FUNCTION (vsuxseg, seg_indexed_loadstore, none_m_preds, tuple_v_scalar_ptr_eew8_index_ops)
+DEF_RVV_FUNCTION (vsuxseg, seg_indexed_loadstore, none_m_preds, tuple_v_scalar_ptr_eew16_index_ops)
+DEF_RVV_FUNCTION (vsuxseg, seg_indexed_loadstore, none_m_preds, tuple_v_scalar_ptr_eew32_index_ops)
+DEF_RVV_FUNCTION (vsuxseg, seg_indexed_loadstore, none_m_preds, tuple_v_scalar_ptr_eew64_index_ops)
+DEF_RVV_FUNCTION (vsoxseg, seg_indexed_loadstore, none_m_preds, tuple_v_scalar_ptr_eew8_index_ops)
+DEF_RVV_FUNCTION (vsoxseg, seg_indexed_loadstore, none_m_preds, tuple_v_scalar_ptr_eew16_index_ops)
+DEF_RVV_FUNCTION (vsoxseg, seg_indexed_loadstore, none_m_preds, tuple_v_scalar_ptr_eew32_index_ops)
+DEF_RVV_FUNCTION (vsoxseg, seg_indexed_loadstore, none_m_preds, tuple_v_scalar_ptr_eew64_index_ops)
+DEF_RVV_FUNCTION (vlsegff, seg_fault_load, full_preds, tuple_v_scalar_const_ptr_size_ptr_ops)
 
 #undef DEF_RVV_FUNCTION