1 /* Target Definitions for NVPTX.
2 Copyright (C) 2014-2023 Free Software Foundation, Inc.
3 Contributed by Bernd Schmidt <bernds@codesourcery.com>
4
5 This file is part of GCC.
6
7 GCC is free software; you can redistribute it and/or modify it
8 under the terms of the GNU General Public License as published
9 by the Free Software Foundation; either version 3, or (at your
10 option) any later version.
11
12 GCC is distributed in the hope that it will be useful, but WITHOUT
13 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
14 or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
15 License 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 #ifndef GCC_NVPTX_H
22 #define GCC_NVPTX_H
23
24 #ifndef NVPTX_OPTS_H
25 #include "config/nvptx/nvptx-opts.h"
26 #endif
27
28 /* Run-time Target. */
29
30 /* Use '--with-arch' for default '-misa'. */
31 #define OPTION_DEFAULT_SPECS \
32 { "arch", "%{!misa=*:-misa=%(VALUE)}" }, \
33
34 /* Assembler supports '-v' option; handle similar to
35 '../../gcc.cc:asm_options', 'HAVE_GNU_AS'. */
36 #define ASM_SPEC "%{v}"
37
38 #define STARTFILE_SPEC "%{mmainkernel:crt0.o%s}"
39
40 #define TARGET_CPU_CPP_BUILTINS() nvptx_cpu_cpp_builtins ()
41
42 /* Avoid the default in ../../gcc.cc, which adds "-pthread", which is not
43 supported for nvptx. */
44 #define GOMP_SELF_SPECS ""
45
46 /* Storage Layout. */
47
48 #define BITS_BIG_ENDIAN 0
49 #define BYTES_BIG_ENDIAN 0
50 #define WORDS_BIG_ENDIAN 0
51
52 /* Chosen such that we won't have to deal with multi-word subregs. */
53 #define UNITS_PER_WORD 8
54
55 /* Alignments in bits. */
56 #define PARM_BOUNDARY 32
57 #define STACK_BOUNDARY 128
58 #define FUNCTION_BOUNDARY 32
59 #define BIGGEST_ALIGNMENT 128
60 #define STRICT_ALIGNMENT 1
61
62 #define MAX_STACK_ALIGNMENT (1024 * 8)
63
64 #define DATA_ALIGNMENT nvptx_data_alignment
65
66 /* Copied from elf.h and other places. We'd otherwise use
67 BIGGEST_ALIGNMENT and fail a number of testcases. */
68 #define MAX_OFILE_ALIGNMENT (32768 * 8)
69
70 /* Type Layout. */
71
72 #define DEFAULT_SIGNED_CHAR 1
73
74 #define SHORT_TYPE_SIZE 16
75 #define INT_TYPE_SIZE 32
76 #define LONG_TYPE_SIZE (TARGET_ABI64 ? 64 : 32)
77 #define LONG_LONG_TYPE_SIZE 64
78 #define FLOAT_TYPE_SIZE 32
79 #define DOUBLE_TYPE_SIZE 64
80 #define LONG_DOUBLE_TYPE_SIZE 64
81 #define TARGET_SUPPORTS_WIDE_INT 1
82
83 #undef SIZE_TYPE
84 #define SIZE_TYPE (TARGET_ABI64 ? "long unsigned int" : "unsigned int")
85 #undef PTRDIFF_TYPE
86 #define PTRDIFF_TYPE (TARGET_ABI64 ? "long int" : "int")
87
88 #define POINTER_SIZE (TARGET_ABI64 ? 64 : 32)
89 #define Pmode (TARGET_ABI64 ? DImode : SImode)
90 #define STACK_SIZE_MODE Pmode
91
92 #include "nvptx-gen.h"
93
94 #define TARGET_PTX_6_0 (ptx_version_option >= PTX_VERSION_6_0)
95 #define TARGET_PTX_6_3 (ptx_version_option >= PTX_VERSION_6_3)
96 #define TARGET_PTX_7_0 (ptx_version_option >= PTX_VERSION_7_0)
97
98 /* Registers. Since ptx is a virtual target, we just define a few
99 hard registers for special purposes and leave pseudos unallocated.
100 We have to have some available hard registers, to keep gcc setup
101 happy. */
102 #define FIRST_PSEUDO_REGISTER 16
103 #define FIXED_REGISTERS { 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0 }
104 #define CALL_USED_REGISTERS { 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 }
105
106 /* Register Classes. */
107 enum reg_class { NO_REGS, ALL_REGS, LIM_REG_CLASSES };
108 #define REG_CLASS_NAMES { "NO_REGS", "ALL_REGS" }
109 #define REG_CLASS_CONTENTS { { 0x0000 }, { 0xFFFF } }
110 #define N_REG_CLASSES (int) LIM_REG_CLASSES
111
112 #define GENERAL_REGS ALL_REGS
113 #define REGNO_REG_CLASS(R) ((void)(R), ALL_REGS)
114 #define BASE_REG_CLASS ALL_REGS
115 #define INDEX_REG_CLASS NO_REGS
116
117 #define REGNO_OK_FOR_BASE_P(X) true
118 #define REGNO_OK_FOR_INDEX_P(X) false
119
120 #define CLASS_MAX_NREGS(class, mode) \
121 ((GET_MODE_SIZE (mode) + UNITS_PER_WORD - 1) / UNITS_PER_WORD)
122
123 #define PROMOTE_MODE(MODE, UNSIGNEDP, TYPE) \
124 if ((MODE) == QImode || (MODE) == HImode) \
125 { \
126 (MODE) = SImode; \
127 (void)(UNSIGNEDP); \
128 (void)(TYPE); \
129 }
130
131 /* Stack and Calling. */
132
133 #define FRAME_GROWS_DOWNWARD 0
134 #define STACK_GROWS_DOWNWARD 1
135
136 #define NVPTX_RETURN_REGNUM 0
137 #define STACK_POINTER_REGNUM 1
138 #define FRAME_POINTER_REGNUM 2
139 #define ARG_POINTER_REGNUM 3
140 #define STATIC_CHAIN_REGNUM 4
141 /* This register points to the shared memory location with the current warp's
142 soft stack pointer (__nvptx_stacks[tid.y]). */
143 #define SOFTSTACK_SLOT_REGNUM 5
144 /* This register is used to save the previous value of the soft stack pointer
145 in the prologue and restore it when returning. */
146 #define SOFTSTACK_PREV_REGNUM 6
147
148 #define REGISTER_NAMES \
149 { \
150 "%value", "%stack", "%frame", "%args", \
151 "%chain", "%sspslot", "%sspprev", "%hr7", \
152 "%hr8", "%hr9", "%hr10", "%hr11", "%hr12", "%hr13", "%hr14", "%hr15" \
153 }
154
155 #define FIRST_PARM_OFFSET(FNDECL) ((void)(FNDECL), 0)
156 #define PUSH_ARGS_REVERSED 1
157 #define ACCUMULATE_OUTGOING_ARGS 1
158
159 /* Avoid using the argument pointer for frame-related things. */
160 #define FRAME_POINTER_CFA_OFFSET(FNDECL) ((void)(FNDECL), 0)
161
162 #ifdef HOST_WIDE_INT
163 struct nvptx_args {
164 tree fntype;
165 /* Number of arguments passed in registers so far. */
166 int count;
167 };
168 #endif
169
170 #define CUMULATIVE_ARGS struct nvptx_args
171
172 #define INIT_CUMULATIVE_ARGS(CUM, FNTYPE, LIBNAME, FNDECL, N_NAMED_ARGS) \
173 ((CUM).fntype = (FNTYPE), (CUM).count = 0, (void)0)
174
175 #define FUNCTION_ARG_REGNO_P(r) 0
176
177 #define DEFAULT_PCC_STRUCT_RETURN 0
178
179 #define FUNCTION_PROFILER(file, labelno) \
180 fatal_error (input_location, \
181 "profiling is not yet implemented for this architecture")
182
183 #define TRAMPOLINE_SIZE 32
184 #define TRAMPOLINE_ALIGNMENT 256
185
186 /* We don't run reload, so this isn't actually used, but it still needs to be
187 defined. Showing an argp->fp elimination also stops
188 expand_builtin_setjmp_receiver from generating invalid insns. */
189 #define ELIMINABLE_REGS \
190 { \
191 { ARG_POINTER_REGNUM, FRAME_POINTER_REGNUM} \
192 }
193
194 /* Define the offset between two registers, one to be eliminated, and the other
195 its replacement, at the start of a routine. */
196
197 #define INITIAL_ELIMINATION_OFFSET(FROM, TO, OFFSET) \
198 ((OFFSET) = 0)
199
200 /* Addressing Modes. */
201
202 #define MAX_REGS_PER_ADDRESS 1
203
204 #define LEGITIMATE_PIC_OPERAND_P(X) 1
205
206
207 #if defined HOST_WIDE_INT
208 struct GTY(()) machine_function
209 {
210 rtx_expr_list *call_args; /* Arg list for the current call. */
211 bool doing_call; /* Within a CALL_ARGS ... CALL_ARGS_END sequence. */
212 bool is_varadic; /* This call is varadic */
213 bool has_varadic; /* Current function has a varadic call. */
214 bool has_chain; /* Current function has outgoing static chain. */
215 bool has_softstack; /* Current function has a soft stack frame. */
216 bool has_simtreg; /* Current function has an OpenMP SIMD region. */
217 int num_args; /* Number of args of current call. */
218 int return_mode; /* Return mode of current fn.
219 (machine_mode not defined yet.) */
220 rtx axis_predicate[2]; /* Neutering predicates. */
221 int axis_dim[2]; /* Maximum number of threads on each axis, dim[0] is
222 vector_length, dim[1] is num_workers. */
223 bool axis_dim_init_p;
224 rtx bcast_partition; /* Register containing the size of each
225 vector's partition of share-memory used to
226 broadcast state. */
227 rtx red_partition; /* Similar to bcast_partition, except for vector
228 reductions. */
229 rtx sync_bar; /* Synchronization barrier ID for vectors. */
230 rtx unisimt_master; /* 'Master lane index' for -muniform-simt. */
231 rtx unisimt_predicate; /* Predicate for -muniform-simt. */
232 rtx unisimt_outside_simt_predicate; /* Predicate for -muniform-simt. */
233 rtx unisimt_location; /* Mask location for -muniform-simt. */
234 /* The following two fields hold the maximum size resp. alignment required
235 for per-lane storage in OpenMP SIMD regions. */
236 unsigned HOST_WIDE_INT simt_stack_size;
237 unsigned HOST_WIDE_INT simt_stack_align;
238 };
239 #endif
240
241 /* Costs. */
242
243 #define NO_FUNCTION_CSE 1
244 #define SLOW_BYTE_ACCESS 0
245 #define BRANCH_COST(speed_p, predictable_p) 6
246
247 /* Assembler Format. */
248
249 #undef ASM_DECLARE_FUNCTION_NAME
250 #define ASM_DECLARE_FUNCTION_NAME(FILE, NAME, DECL) \
251 nvptx_declare_function_name (FILE, NAME, DECL)
252
253 #undef ASM_DECLARE_FUNCTION_SIZE
254 #define ASM_DECLARE_FUNCTION_SIZE(STREAM, NAME, DECL) \
255 nvptx_function_end (STREAM)
256
257 #define DWARF2_ASM_LINE_DEBUG_INFO 1
258
259 #undef ASM_APP_ON
260 #define ASM_APP_ON "\t// #APP \n"
261 #undef ASM_APP_OFF
262 #define ASM_APP_OFF "\t// #NO_APP \n"
263
264 #define DEBUGGER_REGNO(N) N
265
266 #define TEXT_SECTION_ASM_OP ""
267 #define DATA_SECTION_ASM_OP ""
268
269 #undef ASM_GENERATE_INTERNAL_LABEL
270 #define ASM_GENERATE_INTERNAL_LABEL(LABEL, PREFIX, NUM) \
271 do \
272 { \
273 char *__p; \
274 __p = stpcpy (&(LABEL)[1], PREFIX); \
275 (LABEL)[0] = '$'; \
276 sprint_ul (__p, (unsigned long) (NUM)); \
277 } \
278 while (0)
279
280 #define ASM_OUTPUT_ALIGN(FILE, POWER) \
281 do \
282 { \
283 (void) (FILE); \
284 (void) (POWER); \
285 } \
286 while (0)
287
288 #define ASM_OUTPUT_SKIP(FILE, N) \
289 nvptx_output_skip (FILE, N)
290
291 #undef ASM_OUTPUT_ASCII
292 #define ASM_OUTPUT_ASCII(FILE, STR, LENGTH) \
293 nvptx_output_ascii (FILE, STR, LENGTH);
294
295 #define ASM_DECLARE_OBJECT_NAME(FILE, NAME, DECL) \
296 nvptx_declare_object_name (FILE, NAME, DECL)
297
298 #undef ASM_OUTPUT_ALIGNED_DECL_COMMON
299 #define ASM_OUTPUT_ALIGNED_DECL_COMMON(FILE, DECL, NAME, SIZE, ALIGN) \
300 nvptx_output_aligned_decl (FILE, NAME, DECL, SIZE, ALIGN)
301
302 #undef ASM_OUTPUT_ALIGNED_DECL_LOCAL
303 #define ASM_OUTPUT_ALIGNED_DECL_LOCAL(FILE, DECL, NAME, SIZE, ALIGN) \
304 nvptx_output_aligned_decl (FILE, NAME, DECL, SIZE, ALIGN)
305
306 #define CASE_VECTOR_PC_RELATIVE flag_pic
307 #define JUMP_TABLES_IN_TEXT_SECTION flag_pic
308
309 #define ADDR_VEC_ALIGN(VEC) (JUMP_TABLES_IN_TEXT_SECTION ? 5 : 2)
310
311 /* Misc. */
312
313 #define DWARF2_LINENO_DEBUGGING_INFO 1
314
315 #define CLZ_DEFINED_VALUE_AT_ZERO(MODE, VALUE) \
316 ((VALUE) = GET_MODE_BITSIZE ((MODE)), 2)
317 #define CTZ_DEFINED_VALUE_AT_ZERO(MODE, VALUE) \
318 ((VALUE) = GET_MODE_BITSIZE ((MODE)), 2)
319
320 #define SUPPORTS_WEAK 1
321
322 /* The documentation states that ASM_OUTPUT_DEF_FROM_DECLS is used in
323 preference to ASM_OUTPUT_DEF if the tree nodes are available. However, we
324 need the tree nodes to emit the prototype, so at this point it's not clear
325 how we can support ASM_OUTPUT_DEF. Still, we need to define it, or
326 ASM_OUTPUT_DEF_FROM_DECLS is ignored. For now, assert, and once we run
327 into it possibly improve by somehow emitting the prototype elsewhere, or
328 emitting a reasonable error message. */
329 #define ASM_OUTPUT_DEF(FILE,LABEL1,LABEL2) \
330 do \
331 { \
332 (void) (FILE); \
333 (void) (LABEL1); \
334 (void) (LABEL2); \
335 gcc_unreachable (); \
336 } \
337 while (0)
338 #define ASM_OUTPUT_DEF_FROM_DECLS(STREAM, NAME, VALUE) \
339 nvptx_asm_output_def_from_decls (STREAM, NAME, VALUE)
340
341 #define NO_DOT_IN_LABEL
342 #define ASM_COMMENT_START "//"
343
344 #define STORE_FLAG_VALUE 1
345 #define FLOAT_STORE_FLAG_VALUE(MODE) REAL_VALUE_ATOF("1.0", (MODE))
346
347 #define CASE_VECTOR_MODE SImode
348 #define MOVE_MAX 8
349 #define MOVE_RATIO(SPEED) 4
350 #define FUNCTION_MODE QImode
351 #define HAS_INIT_SECTION 1
352
353 /* The C++ front end insists to link against libstdc++ -- which we don't build.
354 Tell it to instead link against the innocuous libgcc. */
355 #define LIBSTDCXX "gcc"
356
357 #endif /* GCC_NVPTX_H */