gcc/config/nvptx/nvptx.h

   1 /* Target Definitions for NVPTX.
   2    Copyright (C) 2014-2025 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 TARGET_SUPPORTS_WIDE_INT 1
  79
  80 #undef SIZE_TYPE
  81 #define SIZE_TYPE (TARGET_ABI64 ? "long unsigned int" : "unsigned int")
  82 #undef PTRDIFF_TYPE
  83 #define PTRDIFF_TYPE (TARGET_ABI64 ? "long int" : "int")
  84
  85 #define POINTER_SIZE (TARGET_ABI64 ? 64 : 32)
  86 #define Pmode (TARGET_ABI64 ? DImode : SImode)
  87 #define STACK_SIZE_MODE Pmode
  88
  89 /* We always have to maintain the '-msoft-stack' pointer, but the PTX "native"
  90    stack pointer is handled implicitly at function level.  */
  91 #define STACK_SAVEAREA_MODE(LEVEL) \
  92   (TARGET_SOFT_STACK ? Pmode \
  93    : (LEVEL == SAVE_FUNCTION ? VOIDmode \
  94       : Pmode))
  95
  96 #include "nvptx-gen.h"
  97
  98 /* There are no 'TARGET_PTX_3_1' and smaller conditionals: our baseline is
  99    PTX ISA Version 3.1.  */
 100 #define TARGET_PTX_4_1 (ptx_version_option >= PTX_VERSION_4_1)
 101 #define TARGET_PTX_4_2 (ptx_version_option >= PTX_VERSION_4_2)
 102 #define TARGET_PTX_6_0 (ptx_version_option >= PTX_VERSION_6_0)
 103 #define TARGET_PTX_6_3 (ptx_version_option >= PTX_VERSION_6_3)
 104 #define TARGET_PTX_7_0 (ptx_version_option >= PTX_VERSION_7_0)
 105 #define TARGET_PTX_7_3 (ptx_version_option >= PTX_VERSION_7_3)
 106 #define TARGET_PTX_7_8 (ptx_version_option >= PTX_VERSION_7_8)
 107
 108 /* Registers.  Since ptx is a virtual target, we just define a few
 109    hard registers for special purposes and leave pseudos unallocated.
 110    We have to have some available hard registers, to keep gcc setup
 111    happy.  */
 112 #define FIRST_PSEUDO_REGISTER 16
 113 #define FIXED_REGISTERS     { 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0 }
 114 #define CALL_USED_REGISTERS { 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 }
 115
 116 /* Register Classes.  */
 117 enum reg_class             {  NO_REGS,    ALL_REGS,     LIM_REG_CLASSES };
 118 #define REG_CLASS_NAMES    { "NO_REGS",  "ALL_REGS" }
 119 #define REG_CLASS_CONTENTS { { 0x0000 }, { 0xFFFF } }
 120 #define N_REG_CLASSES (int) LIM_REG_CLASSES
 121
 122 #define GENERAL_REGS ALL_REGS
 123 #define REGNO_REG_CLASS(R) ((void)(R), ALL_REGS)
 124 #define BASE_REG_CLASS ALL_REGS
 125 #define INDEX_REG_CLASS NO_REGS
 126
 127 #define REGNO_OK_FOR_BASE_P(X) true
 128 #define REGNO_OK_FOR_INDEX_P(X) false
 129
 130 #define CLASS_MAX_NREGS(class, mode) \
 131   ((GET_MODE_SIZE (mode) + UNITS_PER_WORD - 1) / UNITS_PER_WORD)
 132
 133 #define PROMOTE_MODE(MODE, UNSIGNEDP, TYPE)             \
 134   if ((MODE) == QImode || (MODE) == HImode)             \
 135     {                                                   \
 136       (MODE) = SImode;                                  \
 137       (void)(UNSIGNEDP);                                \
 138       (void)(TYPE);                                     \
 139     }
 140
 141 /* Stack and Calling.  */
 142
 143 #define FRAME_GROWS_DOWNWARD 0
 144 #define STACK_GROWS_DOWNWARD 1
 145
 146 #define NVPTX_RETURN_REGNUM 0
 147 #define STACK_POINTER_REGNUM 1
 148 #define FRAME_POINTER_REGNUM 2
 149 #define ARG_POINTER_REGNUM 3
 150 #define STATIC_CHAIN_REGNUM 4
 151 /* This register points to the shared memory location with the current warp's
 152    soft stack pointer (__nvptx_stacks[tid.y]).  */
 153 #define SOFTSTACK_SLOT_REGNUM 5
 154 /* This register is used to save the previous value of the soft stack pointer
 155    in the prologue and restore it when returning.  */
 156 #define SOFTSTACK_PREV_REGNUM 6
 157
 158 #define REGISTER_NAMES                                                  \
 159   {                                                                     \
 160     "%value", "%stack", "%frame", "%args",                              \
 161     "%chain", "%sspslot", "%sspprev", "%hr7",                           \
 162     "%hr8", "%hr9", "%hr10", "%hr11", "%hr12", "%hr13", "%hr14", "%hr15" \
 163   }
 164
 165 #define FIRST_PARM_OFFSET(FNDECL) ((void)(FNDECL), 0)
 166 #define PUSH_ARGS_REVERSED 1
 167 #define ACCUMULATE_OUTGOING_ARGS 1
 168
 169 /* Avoid using the argument pointer for frame-related things.  */
 170 #define FRAME_POINTER_CFA_OFFSET(FNDECL) ((void)(FNDECL), 0)
 171
 172 #ifdef HOST_WIDE_INT
 173 struct nvptx_args {
 174   tree fntype;
 175   /* Number of arguments passed in registers so far.  */
 176   int count;
 177 };
 178 #endif
 179
 180 #define CUMULATIVE_ARGS struct nvptx_args
 181
 182 #define INIT_CUMULATIVE_ARGS(CUM, FNTYPE, LIBNAME, FNDECL, N_NAMED_ARGS) \
 183   ((CUM).fntype = (FNTYPE), (CUM).count = 0, (void)0)
 184
 185 #define FUNCTION_ARG_REGNO_P(r) 0
 186
 187 #define DEFAULT_PCC_STRUCT_RETURN 0
 188
 189 #define FUNCTION_PROFILER(file, labelno) \
 190   fatal_error (input_location, \
 191                "profiling is not yet implemented for this architecture")
 192
 193 #define TRAMPOLINE_SIZE 32
 194 #define TRAMPOLINE_ALIGNMENT 256
 195 \f
 196 /* We don't run reload, so this isn't actually used, but it still needs to be
 197    defined.  Showing an argp->fp elimination also stops
 198    expand_builtin_setjmp_receiver from generating invalid insns.  */
 199 #define ELIMINABLE_REGS                                 \
 200   {                                                     \
 201     { ARG_POINTER_REGNUM, FRAME_POINTER_REGNUM} \
 202   }
 203
 204 /* Define the offset between two registers, one to be eliminated, and the other
 205    its replacement, at the start of a routine.  */
 206
 207 #define INITIAL_ELIMINATION_OFFSET(FROM, TO, OFFSET) \
 208   ((OFFSET) = 0)
 209 \f
 210 /* Addressing Modes.  */
 211
 212 #define MAX_REGS_PER_ADDRESS 1
 213
 214 #define LEGITIMATE_PIC_OPERAND_P(X) 1
 215 \f
 216
 217 #if defined HOST_WIDE_INT
 218 struct GTY(()) machine_function
 219 {
 220   rtx_expr_list *call_args;  /* Arg list for the current call.  */
 221   bool doing_call; /* Within a CALL_ARGS ... CALL_ARGS_END sequence.  */
 222   bool is_variadic;  /* This call is variadic  */
 223   bool has_variadic;  /* Current function has a variadic call.  */
 224   bool has_chain; /* Current function has outgoing static chain.  */
 225   bool has_softstack; /* Current function has a soft stack frame.  */
 226   bool has_simtreg; /* Current function has an OpenMP SIMD region.  */
 227   int num_args; /* Number of args of current call.  */
 228   int return_mode; /* Return mode of current fn.
 229                       (machine_mode not defined yet.) */
 230   rtx axis_predicate[2]; /* Neutering predicates.  */
 231   int axis_dim[2]; /* Maximum number of threads on each axis, dim[0] is
 232                       vector_length, dim[1] is num_workers.  */
 233   bool axis_dim_init_p;
 234   rtx bcast_partition; /* Register containing the size of each
 235                           vector's partition of share-memory used to
 236                           broadcast state.  */
 237   rtx red_partition; /* Similar to bcast_partition, except for vector
 238                         reductions.  */
 239   rtx sync_bar; /* Synchronization barrier ID for vectors.  */
 240   rtx unisimt_master; /* 'Master lane index' for -muniform-simt.  */
 241   rtx unisimt_predicate; /* Predicate for -muniform-simt.  */
 242   rtx unisimt_outside_simt_predicate; /* Predicate for -muniform-simt.  */
 243   rtx unisimt_location; /* Mask location for -muniform-simt.  */
 244   /* The following two fields hold the maximum size resp. alignment required
 245      for per-lane storage in OpenMP SIMD regions.  */
 246   unsigned HOST_WIDE_INT simt_stack_size;
 247   unsigned HOST_WIDE_INT simt_stack_align;
 248 };
 249 #endif
 250 \f
 251 /* Costs.  */
 252
 253 #define NO_FUNCTION_CSE 1
 254 #define SLOW_BYTE_ACCESS 0
 255 #define BRANCH_COST(speed_p, predictable_p) 6
 256 \f
 257 /* Assembler Format.  */
 258
 259 #undef ASM_DECLARE_FUNCTION_NAME
 260 #define ASM_DECLARE_FUNCTION_NAME(FILE, NAME, DECL)             \
 261   nvptx_declare_function_name (FILE, NAME, DECL)
 262
 263 #undef ASM_DECLARE_FUNCTION_SIZE
 264 #define ASM_DECLARE_FUNCTION_SIZE(STREAM, NAME, DECL) \
 265   nvptx_function_end (STREAM)
 266
 267 #define DWARF2_ASM_LINE_DEBUG_INFO 1
 268
 269 #undef ASM_APP_ON
 270 #define ASM_APP_ON "\t// #APP \n"
 271 #undef ASM_APP_OFF
 272 #define ASM_APP_OFF "\t// #NO_APP \n"
 273
 274 #define DEBUGGER_REGNO(N) N
 275
 276 #define TEXT_SECTION_ASM_OP ""
 277 #define DATA_SECTION_ASM_OP ""
 278
 279 #undef  ASM_GENERATE_INTERNAL_LABEL
 280 #define ASM_GENERATE_INTERNAL_LABEL(LABEL, PREFIX, NUM)         \
 281   do                                                            \
 282     {                                                           \
 283       char *__p;                                                \
 284       __p = stpcpy (&(LABEL)[1], PREFIX);                       \
 285       (LABEL)[0] = '$';                                         \
 286       sprint_ul (__p, (unsigned long) (NUM));                   \
 287     }                                                           \
 288   while (0)
 289
 290 #define ASM_OUTPUT_ALIGN(FILE, POWER)           \
 291   do                                            \
 292     {                                           \
 293       (void) (FILE);                            \
 294       (void) (POWER);                           \
 295     }                                           \
 296   while (0)
 297
 298 #define ASM_OUTPUT_SKIP(FILE, N)                \
 299   nvptx_output_skip (FILE, N)
 300
 301 #undef  ASM_OUTPUT_ASCII
 302 #define ASM_OUTPUT_ASCII(FILE, STR, LENGTH)                     \
 303   nvptx_output_ascii (FILE, STR, LENGTH);
 304
 305 #define ASM_DECLARE_OBJECT_NAME(FILE, NAME, DECL)       \
 306   nvptx_declare_object_name (FILE, NAME, DECL)
 307
 308 #undef  ASM_OUTPUT_ALIGNED_DECL_COMMON
 309 #define ASM_OUTPUT_ALIGNED_DECL_COMMON(FILE, DECL, NAME, SIZE, ALIGN)   \
 310   nvptx_output_aligned_decl (FILE, NAME, DECL, SIZE, ALIGN)
 311
 312 #undef  ASM_OUTPUT_ALIGNED_DECL_LOCAL
 313 #define ASM_OUTPUT_ALIGNED_DECL_LOCAL(FILE, DECL, NAME, SIZE, ALIGN)    \
 314   nvptx_output_aligned_decl (FILE, NAME, DECL, SIZE, ALIGN)
 315
 316 #define CASE_VECTOR_PC_RELATIVE flag_pic
 317 #define JUMP_TABLES_IN_TEXT_SECTION flag_pic
 318
 319 #define ADDR_VEC_ALIGN(VEC) (JUMP_TABLES_IN_TEXT_SECTION ? 5 : 2)
 320
 321 /* Misc.  */
 322
 323 #define DWARF2_LINENO_DEBUGGING_INFO 1
 324
 325 #define CLZ_DEFINED_VALUE_AT_ZERO(MODE, VALUE) \
 326   ((VALUE) = GET_MODE_BITSIZE ((MODE)), 2)
 327 #define CTZ_DEFINED_VALUE_AT_ZERO(MODE, VALUE) \
 328   ((VALUE) = GET_MODE_BITSIZE ((MODE)), 2)
 329
 330 #define SUPPORTS_WEAK 1
 331
 332 #define MAKE_DECL_ONE_ONLY(DECL) \
 333   (DECL_WEAK (DECL) = 1)
 334
 335 /* The documentation states that ASM_OUTPUT_DEF_FROM_DECLS is used in
 336    preference to ASM_OUTPUT_DEF if the tree nodes are available.  However, we
 337    need the tree nodes to emit the prototype, so at this point it's not clear
 338    how we can support ASM_OUTPUT_DEF.  Still, we need to define it, or
 339    ASM_OUTPUT_DEF_FROM_DECLS is ignored.  For now, assert, and once we run
 340    into it possibly improve by somehow emitting the prototype elsewhere, or
 341    emitting a reasonable error message.  */
 342 #define ASM_OUTPUT_DEF(FILE,LABEL1,LABEL2)      \
 343   do                                            \
 344     {                                           \
 345       (void) (FILE);                            \
 346       (void) (LABEL1);                          \
 347       (void) (LABEL2);                          \
 348       gcc_unreachable ();                       \
 349     }                                           \
 350   while (0)
 351 #define ASM_OUTPUT_DEF_FROM_DECLS(STREAM, NAME, VALUE)  \
 352   nvptx_asm_output_def_from_decls (STREAM, NAME, VALUE)
 353
 354 /* ..., but also override other macros to avoid 'gcc/defaults.h'-initialization
 355    due to that dummy 'ASM_OUTPUT_DEF'.  */
 356 #define TARGET_USE_LOCAL_THUNK_ALIAS_P(DECL) TARGET_SUPPORTS_ALIASES
 357 #define TARGET_SUPPORTS_ALIASES (nvptx_alias != 0)
 358
 359 #define NO_DOT_IN_LABEL
 360 #define ASM_COMMENT_START "//"
 361
 362 #define STORE_FLAG_VALUE 1
 363 #define FLOAT_STORE_FLAG_VALUE(MODE) REAL_VALUE_ATOF("1.0", (MODE))
 364
 365 #define CASE_VECTOR_MODE SImode
 366 #define MOVE_MAX 8
 367 #define MOVE_RATIO(SPEED) 4
 368 #define FUNCTION_MODE QImode
 369
 370 /* Implement global constructor, destructor support in a conceptually simpler
 371    way than using 'collect2' (the program): implement the respective
 372    functionality in the nvptx-tools 'ld'.  This however still requires the
 373    compiler-side effects corresponding to 'USE_COLLECT2': the global
 374    constructor, destructor support functions need to have external linkage, and
 375    therefore names that are "unique across the whole link".  Use
 376    '!targetm.have_ctors_dtors' to achieve this (..., and thus don't need to
 377    provide 'targetm.asm_out.constructor', 'targetm.asm_out.destructor').  */
 378 #define TARGET_HAVE_CTORS_DTORS false
 379
 380 /* See 'libgcc/config/nvptx/crt0.c' for wrapping of 'main'.  */
 381 #define HAS_INIT_SECTION
 382
 383 /* The C++ front end insists to link against libstdc++ -- which we don't build.
 384    Tell it to instead link against the innocuous libgcc.  */
 385 #define LIBSTDCXX "gcc"
 386
 387 #endif /* GCC_NVPTX_H */