]>
Commit | Line | Data |
---|---|---|
9cde20cf | 1 | (* Auto-generate ARM Neon intrinsics tests. |
f1717362 | 2 | Copyright (C) 2006-2016 Free Software Foundation, Inc. |
9cde20cf | 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-testgen neon.cmo neon-testgen.ml | |
31 | ||
32 | Run with: | |
33 | cd /path/to/gcc/testsuite/gcc.target/arm/neon | |
34 | /path/to/neon-testgen | |
35 | *) | |
36 | ||
37 | open Neon | |
38 | ||
39 | type c_type_flags = Pointer | Const | |
40 | ||
41 | (* Open a test source file. *) | |
42 | let open_test_file dir name = | |
43 | try | |
44 | open_out (dir ^ "/" ^ name ^ ".c") | |
45 | with Sys_error str -> | |
46 | failwith ("Could not create test source file " ^ name ^ ": " ^ str) | |
47 | ||
48 | (* Emit prologue code to a test source file. *) | |
49 | let emit_prologue chan test_name effective_target compile_test_optim = | |
50 | Printf.fprintf chan "/* Test the `%s' ARM Neon intrinsic. */\n" test_name; | |
51 | Printf.fprintf chan "/* This file was autogenerated by neon-testgen. */\n\n"; | |
52 | Printf.fprintf chan "/* { dg-do assemble } */\n"; | |
53 | Printf.fprintf chan "/* { dg-require-effective-target %s_ok } */\n" | |
54 | effective_target; | |
55 | Printf.fprintf chan "/* { dg-options \"-save-temps %s\" } */\n" compile_test_optim; | |
56 | Printf.fprintf chan "/* { dg-add-options %s } */\n" effective_target; | |
57 | Printf.fprintf chan "\n#include \"arm_neon.h\"\n\n" | |
58 | ||
59 | (* Emit declarations of variables that are going to be passed | |
60 | to an intrinsic, together with one to take a returned value if needed. *) | |
61 | let emit_variables chan c_types features spaces = | |
62 | let emit () = | |
63 | ignore ( | |
64 | List.fold_left (fun arg_number -> fun (flags, ty) -> | |
65 | let pointer_bit = | |
66 | if List.mem Pointer flags then "*" else "" | |
67 | in | |
68 | (* Const arguments to builtins are directly | |
69 | written in as constants. *) | |
70 | if not (List.mem Const flags) then | |
71 | Printf.fprintf chan "%s%s %sarg%d_%s;\n" | |
72 | spaces ty pointer_bit arg_number ty; | |
73 | arg_number + 1) | |
74 | 0 (List.tl c_types)) | |
75 | in | |
76 | match c_types with | |
77 | (_, return_ty) :: tys -> | |
78 | if return_ty <> "void" then begin | |
79 | (* The intrinsic returns a value. We need to do explicit register | |
80 | allocation for vget_low tests or they fail because of copy | |
81 | elimination. *) | |
82 | ((if List.mem Fixed_vector_reg features then | |
83 | Printf.fprintf chan "%sregister %s out_%s asm (\"d18\");\n" | |
84 | spaces return_ty return_ty | |
85 | else if List.mem Fixed_core_reg features then | |
86 | Printf.fprintf chan "%sregister %s out_%s asm (\"r0\");\n" | |
87 | spaces return_ty return_ty | |
88 | else | |
89 | Printf.fprintf chan "%s%s out_%s;\n" spaces return_ty return_ty); | |
90 | emit ()) | |
91 | end else | |
92 | (* The intrinsic does not return a value. *) | |
93 | emit () | |
94 | | _ -> assert false | |
95 | ||
96 | (* Emit code to call an intrinsic. *) | |
97 | let emit_call chan const_valuator c_types name elt_ty = | |
98 | (if snd (List.hd c_types) <> "void" then | |
99 | Printf.fprintf chan " out_%s = " (snd (List.hd c_types)) | |
100 | else | |
101 | Printf.fprintf chan " "); | |
102 | Printf.fprintf chan "%s_%s (" (intrinsic_name name) (string_of_elt elt_ty); | |
103 | let print_arg chan arg_number (flags, ty) = | |
104 | (* If the argument is of const type, then directly write in the | |
105 | constant now. *) | |
106 | if List.mem Const flags then | |
107 | match const_valuator with | |
108 | None -> | |
109 | if List.mem Pointer flags then | |
110 | Printf.fprintf chan "0" | |
111 | else | |
112 | Printf.fprintf chan "1" | |
113 | | Some f -> Printf.fprintf chan "%s" (string_of_int (f arg_number)) | |
114 | else | |
115 | Printf.fprintf chan "arg%d_%s" arg_number ty | |
116 | in | |
117 | let rec print_args arg_number tys = | |
118 | match tys with | |
119 | [] -> () | |
120 | | [ty] -> print_arg chan arg_number ty | |
121 | | ty::tys -> | |
122 | print_arg chan arg_number ty; | |
123 | Printf.fprintf chan ", "; | |
124 | print_args (arg_number + 1) tys | |
125 | in | |
126 | print_args 0 (List.tl c_types); | |
127 | Printf.fprintf chan ");\n" | |
128 | ||
129 | (* Emit epilogue code to a test source file. *) | |
130 | let emit_epilogue chan features regexps = | |
131 | let no_op = List.exists (fun feature -> feature = No_op) features in | |
132 | Printf.fprintf chan "}\n\n"; | |
133 | if not no_op then | |
134 | List.iter (fun regexp -> | |
135 | Printf.fprintf chan | |
136 | "/* { dg-final { scan-assembler \"%s\" } } */\n" regexp) | |
137 | regexps | |
138 | else | |
139 | () | |
140 | ||
141 | ||
142 | (* Check a list of C types to determine which ones are pointers and which | |
143 | ones are const. *) | |
144 | let check_types tys = | |
145 | let tys' = | |
146 | List.map (fun ty -> | |
147 | let len = String.length ty in | |
148 | if len > 2 && String.get ty (len - 2) = ' ' | |
149 | && String.get ty (len - 1) = '*' | |
150 | then ([Pointer], String.sub ty 0 (len - 2)) | |
151 | else ([], ty)) tys | |
152 | in | |
153 | List.map (fun (flags, ty) -> | |
154 | if String.length ty > 6 && String.sub ty 0 6 = "const " | |
155 | then (Const :: flags, String.sub ty 6 ((String.length ty) - 6)) | |
156 | else (flags, ty)) tys' | |
157 | ||
158 | (* Work out what the effective target should be. *) | |
159 | let effective_target features = | |
160 | try | |
161 | match List.find (fun feature -> | |
162 | match feature with Requires_feature _ -> true | |
163 | | Requires_arch _ -> true | |
164 | | Requires_FP_bit 1 -> true | |
165 | | _ -> false) | |
166 | features with | |
167 | Requires_feature "FMA" -> "arm_neonv2" | |
168 | | Requires_feature "CRYPTO" -> "arm_crypto" | |
169 | | Requires_arch 8 -> "arm_v8_neon" | |
170 | | Requires_FP_bit 1 -> "arm_neon_fp16" | |
171 | | _ -> assert false | |
172 | with Not_found -> "arm_neon" | |
173 | ||
174 | (* Work out what the testcase optimization level should be, default to -O0. *) | |
175 | let compile_test_optim features = | |
176 | try | |
177 | match List.find (fun feature -> | |
178 | match feature with Compiler_optim _ -> true | |
179 | | _ -> false) | |
180 | features with | |
181 | Compiler_optim opt -> opt | |
182 | | _ -> assert false | |
183 | with Not_found -> "-O0" | |
184 | ||
185 | (* Given an intrinsic shape, produce a regexp that will match | |
186 | the right-hand sides of instructions generated by an intrinsic of | |
187 | that shape. *) | |
188 | let rec analyze_shape shape = | |
189 | let rec n_things n thing = | |
190 | match n with | |
191 | 0 -> [] | |
192 | | n -> thing :: (n_things (n - 1) thing) | |
193 | in | |
194 | let rec analyze_shape_elt elt = | |
195 | match elt with | |
196 | Dreg -> "\\[dD\\]\\[0-9\\]+" | |
197 | | Qreg -> "\\[qQ\\]\\[0-9\\]+" | |
198 | | Corereg -> "\\[rR\\]\\[0-9\\]+" | |
199 | | Immed -> "#\\[0-9\\]+" | |
200 | | VecArray (1, elt) -> | |
201 | let elt_regexp = analyze_shape_elt elt in | |
202 | "((\\\\\\{" ^ elt_regexp ^ "\\\\\\})|(" ^ elt_regexp ^ "))" | |
203 | | VecArray (n, elt) -> | |
204 | let elt_regexp = analyze_shape_elt elt in | |
205 | let alt1 = elt_regexp ^ "-" ^ elt_regexp in | |
206 | let alt2 = commas (fun x -> x) (n_things n elt_regexp) "" in | |
207 | "\\\\\\{((" ^ alt1 ^ ")|(" ^ alt2 ^ "))\\\\\\}" | |
208 | | (PtrTo elt | CstPtrTo elt) -> | |
209 | "\\\\\\[" ^ (analyze_shape_elt elt) ^ "\\(:\\[0-9\\]+\\)?\\\\\\]" | |
210 | | Element_of_dreg -> (analyze_shape_elt Dreg) ^ "\\\\\\[\\[0-9\\]+\\\\\\]" | |
211 | | Element_of_qreg -> (analyze_shape_elt Qreg) ^ "\\\\\\[\\[0-9\\]+\\\\\\]" | |
212 | | All_elements_of_dreg -> (analyze_shape_elt Dreg) ^ "\\\\\\[\\\\\\]" | |
213 | | Alternatives (elts) -> "(" ^ (String.concat "|" (List.map analyze_shape_elt elts)) ^ ")" | |
214 | in | |
215 | match shape with | |
216 | All (n, elt) -> commas analyze_shape_elt (n_things n elt) "" | |
217 | | Long -> (analyze_shape_elt Qreg) ^ ", " ^ (analyze_shape_elt Dreg) ^ | |
218 | ", " ^ (analyze_shape_elt Dreg) | |
219 | | Long_noreg elt -> (analyze_shape_elt elt) ^ ", " ^ (analyze_shape_elt elt) | |
220 | | Wide -> (analyze_shape_elt Qreg) ^ ", " ^ (analyze_shape_elt Qreg) ^ | |
221 | ", " ^ (analyze_shape_elt Dreg) | |
222 | | Wide_noreg elt -> analyze_shape (Long_noreg elt) | |
223 | | Narrow -> (analyze_shape_elt Dreg) ^ ", " ^ (analyze_shape_elt Qreg) ^ | |
224 | ", " ^ (analyze_shape_elt Qreg) | |
225 | | Use_operands elts -> commas analyze_shape_elt (Array.to_list elts) "" | |
226 | | By_scalar Dreg -> | |
227 | analyze_shape (Use_operands [| Dreg; Dreg; Element_of_dreg |]) | |
228 | | By_scalar Qreg -> | |
229 | analyze_shape (Use_operands [| Qreg; Qreg; Element_of_dreg |]) | |
230 | | By_scalar _ -> assert false | |
231 | | Wide_lane -> | |
232 | analyze_shape (Use_operands [| Qreg; Dreg; Element_of_dreg |]) | |
233 | | Wide_scalar -> | |
234 | analyze_shape (Use_operands [| Qreg; Dreg; Element_of_dreg |]) | |
235 | | Pair_result elt -> | |
236 | let elt_regexp = analyze_shape_elt elt in | |
237 | elt_regexp ^ ", " ^ elt_regexp | |
238 | | Unary_scalar _ -> "FIXME Unary_scalar" | |
239 | | Binary_imm elt -> analyze_shape (Use_operands [| elt; elt; Immed |]) | |
240 | | Narrow_imm -> analyze_shape (Use_operands [| Dreg; Qreg; Immed |]) | |
241 | | Long_imm -> analyze_shape (Use_operands [| Qreg; Dreg; Immed |]) | |
242 | ||
243 | (* Generate tests for one intrinsic. *) | |
244 | let test_intrinsic dir opcode features shape name munge elt_ty = | |
245 | (* Open the test source file. *) | |
246 | let test_name = name ^ (string_of_elt elt_ty) in | |
247 | let chan = open_test_file dir test_name in | |
248 | (* Work out what argument and return types the intrinsic has. *) | |
249 | let c_arity, new_elt_ty = munge shape elt_ty in | |
250 | let c_types = check_types (strings_of_arity c_arity) in | |
251 | (* Extract any constant valuator (a function specifying what constant | |
252 | values are to be written into the intrinsic call) from the features | |
253 | list. *) | |
254 | let const_valuator = | |
255 | try | |
256 | match (List.find (fun feature -> match feature with | |
257 | Const_valuator _ -> true | |
258 | | _ -> false) features) with | |
259 | Const_valuator f -> Some f | |
260 | | _ -> assert false | |
261 | with Not_found -> None | |
262 | in | |
263 | (* Work out what instruction name(s) to expect. *) | |
264 | let insns = get_insn_names features name in | |
265 | let no_suffix = (new_elt_ty = NoElts) in | |
266 | let insns = | |
267 | if no_suffix then insns | |
268 | else List.map (fun insn -> | |
269 | let suffix = string_of_elt_dots new_elt_ty in | |
270 | insn ^ "\\." ^ suffix) insns | |
271 | in | |
272 | (* Construct a regexp to match against the expected instruction name(s). *) | |
273 | let insn_regexp = | |
274 | match insns with | |
275 | [] -> assert false | |
276 | | [insn] -> insn | |
277 | | _ -> | |
278 | let rec calc_regexp insns cur_regexp = | |
279 | match insns with | |
280 | [] -> cur_regexp | |
281 | | [insn] -> cur_regexp ^ "(" ^ insn ^ "))" | |
282 | | insn::insns -> calc_regexp insns (cur_regexp ^ "(" ^ insn ^ ")|") | |
283 | in calc_regexp insns "(" | |
284 | in | |
285 | (* Construct regexps to match against the instructions that this | |
286 | intrinsic expands to. Watch out for any writeback character and | |
287 | comments after the instruction. *) | |
288 | let regexps = List.map (fun regexp -> insn_regexp ^ "\\[ \t\\]+" ^ regexp ^ | |
289 | "!?\\(\\[ \t\\]+@\\[a-zA-Z0-9 \\]+\\)?\\n") | |
290 | (analyze_all_shapes features shape analyze_shape) | |
291 | in | |
292 | let effective_target = effective_target features in | |
293 | let compile_test_optim = compile_test_optim features | |
294 | in | |
295 | (* Emit file and function prologues. *) | |
296 | emit_prologue chan test_name effective_target compile_test_optim; | |
297 | ||
298 | if (compare compile_test_optim "-O0") <> 0 then | |
299 | (* Emit variable declarations. *) | |
300 | emit_variables chan c_types features ""; | |
301 | ||
302 | Printf.fprintf chan "void test_%s (void)\n{\n" test_name; | |
303 | ||
304 | if compare compile_test_optim "-O0" = 0 then | |
305 | (* Emit variable declarations. *) | |
306 | emit_variables chan c_types features " "; | |
307 | ||
308 | Printf.fprintf chan "\n"; | |
309 | (* Emit the call to the intrinsic. *) | |
310 | emit_call chan const_valuator c_types name elt_ty; | |
311 | (* Emit the function epilogue and the DejaGNU scan-assembler directives. *) | |
312 | emit_epilogue chan features regexps; | |
313 | (* Close the test file. *) | |
314 | close_out chan | |
315 | ||
316 | (* Generate tests for one element of the "ops" table. *) | |
317 | let test_intrinsic_group dir (opcode, features, shape, name, munge, types) = | |
318 | List.iter (test_intrinsic dir opcode features shape name munge) types | |
319 | ||
320 | (* Program entry point. *) | |
321 | let _ = | |
322 | let directory = if Array.length Sys.argv <> 1 then Sys.argv.(1) else "." in | |
323 | List.iter (test_intrinsic_group directory) (reinterp @ reinterpq @ ops) | |
324 |