gcc/config/arm/neon-gen.ml

   1 (* Auto-generate ARM Neon intrinsics header file.
   2    Copyright (C) 2006, 2007, 2009 Free Software Foundation, Inc.
   3    Contributed by CodeSourcery.
   4
   5    This file is part of GCC.
   6
   7    GCC is free software; you can redistribute it and/or modify it under
   8    the terms of the GNU General Public License as published by the Free
   9    Software Foundation; either version 3, or (at your option) any later
  10    version.
  11
  12    GCC is distributed in the hope that it will be useful, but WITHOUT ANY
  13    WARRANTY; without even the implied warranty of MERCHANTABILITY or
  14    FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  15    for more details.
  16
  17    You should have received a copy of the GNU General Public License
  18    along with GCC; see the file COPYING3.  If not see
  19    <http://www.gnu.org/licenses/>.
  20
  21    This is an O'Caml program.  The O'Caml compiler is available from:
  22
  23      http://caml.inria.fr/
  24
  25    Or from your favourite OS's friendly packaging system. Tested with version
  26    3.09.2, though other versions will probably work too.
  27
  28    Compile with:
  29      ocamlc -c neon.ml
  30      ocamlc -o neon-gen neon.cmo neon-gen.ml
  31
  32    Run with:
  33      ./neon-gen > arm_neon.h
  34 *)
  35
  36 open Neon
  37
  38 (* The format codes used in the following functions are documented at:
  39      http://caml.inria.fr/pub/docs/manual-ocaml/libref/Format.html\
  40      #6_printflikefunctionsforprettyprinting
  41    (one line, remove the backslash.)
  42 *)
  43
  44 (* Following functions can be used to approximate GNU indentation style.  *)
  45 let start_function () =
  46   Format.printf "@[<v 0>";
  47   ref 0
  48
  49 let end_function nesting =
  50   match !nesting with
  51     0 -> Format.printf "@;@;@]"
  52   | _ -> failwith ("Bad nesting (ending function at level "
  53                    ^ (string_of_int !nesting) ^ ")")
  54
  55 let open_braceblock nesting =
  56   begin match !nesting with
  57     0 -> Format.printf "@,@<0>{@[<v 2>@,"
  58   | _ -> Format.printf "@,@[<v 2>  @<0>{@[<v 2>@,"
  59   end;
  60   incr nesting
  61
  62 let close_braceblock nesting =
  63   decr nesting;
  64   match !nesting with
  65     0 -> Format.printf "@]@,@<0>}"
  66   | _ -> Format.printf "@]@,@<0>}@]"
  67
  68 let print_function arity fnname body =
  69   let ffmt = start_function () in
  70   Format.printf "__extension__ static __inline ";
  71   let inl = "__attribute__ ((__always_inline__))" in
  72   begin match arity with
  73     Arity0 ret ->
  74       Format.printf "%s %s@,%s (void)" (string_of_vectype ret) inl fnname
  75   | Arity1 (ret, arg0) ->
  76       Format.printf "%s %s@,%s (%s __a)" (string_of_vectype ret) inl fnname
  77                                         (string_of_vectype arg0)
  78   | Arity2 (ret, arg0, arg1) ->
  79       Format.printf "%s %s@,%s (%s __a, %s __b)"
  80         (string_of_vectype ret) inl fnname (string_of_vectype arg0)
  81         (string_of_vectype arg1)
  82   | Arity3 (ret, arg0, arg1, arg2) ->
  83       Format.printf "%s %s@,%s (%s __a, %s __b, %s __c)"
  84         (string_of_vectype ret) inl fnname (string_of_vectype arg0)
  85         (string_of_vectype arg1) (string_of_vectype arg2)
  86   | Arity4 (ret, arg0, arg1, arg2, arg3) ->
  87       Format.printf "%s %s@,%s (%s __a, %s __b, %s __c, %s __d)"
  88         (string_of_vectype ret) inl fnname (string_of_vectype arg0)
  89         (string_of_vectype arg1) (string_of_vectype arg2)
  90         (string_of_vectype arg3)
  91   end;
  92   open_braceblock ffmt;
  93   let rec print_lines = function
  94     [] -> ()
  95   | [line] -> Format.printf "%s" line
  96   | line::lines -> Format.printf "%s@," line; print_lines lines in
  97   print_lines body;
  98   close_braceblock ffmt;
  99   end_function ffmt
 100
 101 let return_by_ptr features = List.mem ReturnPtr features
 102
 103 let union_string num elts base =
 104   let itype = inttype_for_array num elts in
 105   let iname = string_of_inttype itype
 106   and sname = string_of_vectype (T_arrayof (num, elts)) in
 107   Printf.sprintf "union { %s __i; %s __o; } %s" sname iname base
 108
 109 let rec signed_ctype = function
 110     T_uint8x8 | T_poly8x8 -> T_int8x8
 111   | T_uint8x16 | T_poly8x16 -> T_int8x16
 112   | T_uint16x4 | T_poly16x4 -> T_int16x4
 113   | T_uint16x8 | T_poly16x8 -> T_int16x8
 114   | T_uint32x2 -> T_int32x2
 115   | T_uint32x4 -> T_int32x4
 116   | T_uint64x1 -> T_int64x1
 117   | T_uint64x2 -> T_int64x2
 118   (* Cast to types defined by mode in arm.c, not random types pulled in from
 119      the <stdint.h> header in use. This fixes incompatible pointer errors when
 120      compiling with C++.  *)
 121   | T_uint8 | T_int8 -> T_intQI
 122   | T_uint16 | T_int16 -> T_intHI
 123   | T_uint32 | T_int32 -> T_intSI
 124   | T_uint64 | T_int64 -> T_intDI
 125   | T_float32 -> T_floatSF
 126   | T_poly8 -> T_intQI
 127   | T_poly16 -> T_intHI
 128   | T_arrayof (n, elt) -> T_arrayof (n, signed_ctype elt)
 129   | T_ptrto elt -> T_ptrto (signed_ctype elt)
 130   | T_const elt -> T_const (signed_ctype elt)
 131   | x -> x
 132
 133 let add_cast ctype cval =
 134   let stype = signed_ctype ctype in
 135   if ctype <> stype then
 136     Printf.sprintf "(%s) %s" (string_of_vectype stype) cval
 137   else
 138     cval
 139
 140 let cast_for_return to_ty = "(" ^ (string_of_vectype to_ty) ^ ")"
 141
 142 (* Return a tuple of a list of declarations to go at the start of the function,
 143    and a list of statements needed to return THING.  *)
 144 let return arity return_by_ptr thing =
 145   match arity with
 146     Arity0 (ret) | Arity1 (ret, _) | Arity2 (ret, _, _) | Arity3 (ret, _, _, _)
 147   | Arity4 (ret, _, _, _, _) ->
 148     match ret with
 149       T_arrayof (num, vec) ->
 150         if return_by_ptr then
 151           let sname = string_of_vectype ret in
 152           [Printf.sprintf "%s __rv;" sname],
 153           [thing ^ ";"; "return __rv;"]
 154         else
 155           let uname = union_string num vec "__rv" in
 156           [uname ^ ";"], ["__rv.__o = " ^ thing ^ ";"; "return __rv.__i;"]
 157     | T_void -> [], [thing ^ ";"]
 158     | _ ->
 159         [], ["return " ^ (cast_for_return ret) ^ thing ^ ";"]
 160
 161 let rec element_type ctype =
 162   match ctype with
 163     T_arrayof (_, v) -> element_type v
 164   | _ -> ctype
 165
 166 let params return_by_ptr ps =
 167   let pdecls = ref [] in
 168   let ptype t p =
 169     match t with
 170       T_arrayof (num, elts) ->
 171         let uname = union_string num elts (p ^ "u") in
 172         let decl = Printf.sprintf "%s = { %s };" uname p in
 173         pdecls := decl :: !pdecls;
 174         p ^ "u.__o"
 175     | _ -> add_cast t p in
 176   let plist = match ps with
 177     Arity0 _ -> []
 178   | Arity1 (_, t1) -> [ptype t1 "__a"]
 179   | Arity2 (_, t1, t2) -> [ptype t1 "__a"; ptype t2 "__b"]
 180   | Arity3 (_, t1, t2, t3) -> [ptype t1 "__a"; ptype t2 "__b"; ptype t3 "__c"]
 181   | Arity4 (_, t1, t2, t3, t4) ->
 182       [ptype t1 "__a"; ptype t2 "__b"; ptype t3 "__c"; ptype t4 "__d"] in
 183   match ps with
 184     Arity0 ret | Arity1 (ret, _) | Arity2 (ret, _, _) | Arity3 (ret, _, _, _)
 185   | Arity4 (ret, _, _, _, _) ->
 186       if return_by_ptr then
 187         !pdecls, add_cast (T_ptrto (element_type ret)) "&__rv.val[0]" :: plist
 188       else
 189         !pdecls, plist
 190
 191 let modify_params features plist =
 192   let is_flipped =
 193     List.exists (function Flipped _ -> true | _ -> false) features in
 194   if is_flipped then
 195     match plist with
 196       [ a; b ] -> [ b; a ]
 197     | _ ->
 198       failwith ("Don't know how to flip args " ^ (String.concat ", " plist))
 199   else
 200     plist
 201
 202 (* !!! Decide whether to add an extra information word based on the shape
 203    form.  *)
 204 let extra_word shape features paramlist bits =
 205   let use_word =
 206     match shape with
 207       All _ | Long | Long_noreg _ | Wide | Wide_noreg _ | Narrow
 208     | By_scalar _ | Wide_scalar | Wide_lane | Binary_imm _ | Long_imm
 209     | Narrow_imm -> true
 210     | _ -> List.mem InfoWord features
 211   in
 212     if use_word then
 213       paramlist @ [string_of_int bits]
 214     else
 215       paramlist
 216
 217 (* Bit 0 represents signed (1) vs unsigned (0), or float (1) vs poly (0).
 218    Bit 1 represents floats & polynomials (1), or ordinary integers (0).
 219    Bit 2 represents rounding (1) vs none (0).  *)
 220 let infoword_value elttype features =
 221   let bits01 =
 222     match elt_class elttype with
 223       Signed | ConvClass (Signed, _) | ConvClass (_, Signed) -> 0b001
 224     | Poly -> 0b010
 225     | Float -> 0b011
 226     | _ -> 0b000
 227   and rounding_bit = if List.mem Rounding features then 0b100 else 0b000 in
 228   bits01 lor rounding_bit
 229
 230 (* "Cast" type operations will throw an exception in mode_of_elt (actually in
 231    elt_width, called from there). Deal with that here, and generate a suffix
 232    with multiple modes (<to><from>).  *)
 233 let rec mode_suffix elttype shape =
 234   try
 235     let mode = mode_of_elt elttype shape in
 236     string_of_mode mode
 237   with MixedMode (dst, src) ->
 238     let dstmode = mode_of_elt dst shape
 239     and srcmode = mode_of_elt src shape in
 240     string_of_mode dstmode ^ string_of_mode srcmode
 241
 242 let print_variant opcode features shape name (ctype, asmtype, elttype) =
 243   let bits = infoword_value elttype features in
 244   let modesuf = mode_suffix elttype shape in
 245   let return_by_ptr = return_by_ptr features in
 246   let pdecls, paramlist = params return_by_ptr ctype in
 247   let paramlist' = modify_params features paramlist in
 248   let paramlist'' = extra_word shape features paramlist' bits in
 249   let parstr = String.concat ", " paramlist'' in
 250   let builtin = Printf.sprintf "__builtin_neon_%s%s (%s)"
 251                   (builtin_name features name) modesuf parstr in
 252   let rdecls, stmts = return ctype return_by_ptr builtin in
 253   let body = pdecls @ rdecls @ stmts
 254   and fnname = (intrinsic_name name) ^ "_" ^ (string_of_elt elttype) in
 255   print_function ctype fnname body
 256
 257 (* When this function processes the element types in the ops table, it rewrites
 258    them in a list of tuples (a,b,c):
 259      a : C type as an "arity", e.g. Arity1 (T_poly8x8, T_poly8x8)
 260      b : Asm type : a single, processed element type, e.g. P16. This is the
 261          type which should be attached to the asm opcode.
 262      c : Variant type : the unprocessed type for this variant (e.g. in add
 263          instructions which don't care about the sign, b might be i16 and c
 264          might be s16.)
 265 *)
 266
 267 let print_op (opcode, features, shape, name, munge, types) =
 268   let sorted_types = List.sort compare types in
 269   let munged_types = List.map
 270     (fun elt -> let c, asm = munge shape elt in c, asm, elt) sorted_types in
 271   List.iter
 272     (fun variant -> print_variant opcode features shape name variant)
 273     munged_types
 274
 275 let print_ops ops =
 276   List.iter print_op ops
 277
 278 (* Output type definitions. Table entries are:
 279      cbase : "C" name for the type.
 280      abase : "ARM" base name for the type (i.e. int in int8x8_t).
 281      esize : element size.
 282      enum : element count.
 283 *)
 284
 285 let deftypes () =
 286   let typeinfo = [
 287     (* Doubleword vector types.  *)
 288     "__builtin_neon_qi", "int", 8, 8;
 289     "__builtin_neon_hi", "int", 16, 4;
 290     "__builtin_neon_si", "int", 32, 2;
 291     "__builtin_neon_di", "int", 64, 1;
 292     "__builtin_neon_sf", "float", 32, 2;
 293     "__builtin_neon_poly8", "poly", 8, 8;
 294     "__builtin_neon_poly16", "poly", 16, 4;
 295     "__builtin_neon_uqi", "uint", 8, 8;
 296     "__builtin_neon_uhi", "uint", 16, 4;
 297     "__builtin_neon_usi", "uint", 32, 2;
 298     "__builtin_neon_udi", "uint", 64, 1;
 299
 300     (* Quadword vector types.  *)
 301     "__builtin_neon_qi", "int", 8, 16;
 302     "__builtin_neon_hi", "int", 16, 8;
 303     "__builtin_neon_si", "int", 32, 4;
 304     "__builtin_neon_di", "int", 64, 2;
 305     "__builtin_neon_sf", "float", 32, 4;
 306     "__builtin_neon_poly8", "poly", 8, 16;
 307     "__builtin_neon_poly16", "poly", 16, 8;
 308     "__builtin_neon_uqi", "uint", 8, 16;
 309     "__builtin_neon_uhi", "uint", 16, 8;
 310     "__builtin_neon_usi", "uint", 32, 4;
 311     "__builtin_neon_udi", "uint", 64, 2
 312   ] in
 313   List.iter
 314     (fun (cbase, abase, esize, enum) ->
 315       let attr =
 316         match enum with
 317           1 -> ""
 318         | _ -> Printf.sprintf "\t__attribute__ ((__vector_size__ (%d)))"
 319                               (esize * enum / 8) in
 320       Format.printf "typedef %s %s%dx%d_t%s;@\n" cbase abase esize enum attr)
 321     typeinfo;
 322   Format.print_newline ();
 323   (* Extra types not in <stdint.h>.  *)
 324   Format.printf "typedef float float32_t;\n";
 325   Format.printf "typedef __builtin_neon_poly8 poly8_t;\n";
 326   Format.printf "typedef __builtin_neon_poly16 poly16_t;\n"
 327
 328 (* Output structs containing arrays, for load & store instructions etc.  *)
 329
 330 let arrtypes () =
 331   let typeinfo = [
 332     "int", 8;    "int", 16;
 333     "int", 32;   "int", 64;
 334     "uint", 8;   "uint", 16;
 335     "uint", 32;  "uint", 64;
 336     "float", 32; "poly", 8;
 337     "poly", 16
 338   ] in
 339   let writestruct elname elsize regsize arrsize =
 340     let elnum = regsize / elsize in
 341     let structname =
 342       Printf.sprintf "%s%dx%dx%d_t" elname elsize elnum arrsize in
 343     let sfmt = start_function () in
 344     Format.printf "typedef struct %s" structname;
 345     open_braceblock sfmt;
 346     Format.printf "%s%dx%d_t val[%d];" elname elsize elnum arrsize;
 347     close_braceblock sfmt;
 348     Format.printf " %s;" structname;
 349     end_function sfmt;
 350   in
 351     for n = 2 to 4 do
 352       List.iter
 353         (fun (elname, elsize) ->
 354           writestruct elname elsize 64 n;
 355           writestruct elname elsize 128 n)
 356         typeinfo
 357     done
 358
 359 let print_lines = List.iter (fun s -> Format.printf "%s@\n" s)
 360
 361 (* Do it.  *)
 362
 363 let _ =
 364   print_lines [
 365 "/* ARM NEON intrinsics include file. This file is generated automatically";
 366 "   using neon-gen.ml.  Please do not edit manually.";
 367 "";
 368 "   Copyright (C) 2006, 2007, 2009 Free Software Foundation, Inc.";
 369 "   Contributed by CodeSourcery.";
 370 "";
 371 "   This file is part of GCC.";
 372 "";
 373 "   GCC is free software; you can redistribute it and/or modify it";
 374 "   under the terms of the GNU General Public License as published";
 375 "   by the Free Software Foundation; either version 3, or (at your";
 376 "   option) any later version.";
 377 "";
 378 "   GCC is distributed in the hope that it will be useful, but WITHOUT";
 379 "   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY";
 380 "   or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public";
 381 "   License for more details.";
 382 "";
 383 "   Under Section 7 of GPL version 3, you are granted additional";
 384 "   permissions described in the GCC Runtime Library Exception, version";
 385 "   3.1, as published by the Free Software Foundation.";
 386 "";
 387 "   You should have received a copy of the GNU General Public License and";
 388 "   a copy of the GCC Runtime Library Exception along with this program;";
 389 "   see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see";
 390 "   <http://www.gnu.org/licenses/>.  */";
 391 "";
 392 "#ifndef _GCC_ARM_NEON_H";
 393 "#define _GCC_ARM_NEON_H 1";
 394 "";
 395 "#ifndef __ARM_NEON__";
 396 "#error You must enable NEON instructions (e.g. -mfloat-abi=softfp -mfpu=neon) to use arm_neon.h";
 397 "#else";
 398 "";
 399 "#ifdef __cplusplus";
 400 "extern \"C\" {";
 401 "#endif";
 402 "";
 403 "#include <stdint.h>";
 404 ""];
 405   deftypes ();
 406   arrtypes ();
 407   Format.print_newline ();
 408   print_ops ops;
 409   Format.print_newline ();
 410   print_ops reinterp;
 411   print_lines [
 412 "#ifdef __cplusplus";
 413 "}";
 414 "#endif";
 415 "#endif";
 416 "#endif"]