1 /* Target Definitions for NVPTX. 2 Copyright (C) 2014-2022 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 #define STARTFILE_SPEC "%{mmainkernel:crt0.o%s}" 31 32 #define TARGET_CPU_CPP_BUILTINS() nvptx_cpu_cpp_builtins () 33 34 /* Avoid the default in ../../gcc.cc, which adds "-pthread", which is not 35 supported for nvptx. */ 36 #define GOMP_SELF_SPECS "" 37 38 /* Storage Layout. */ 39 40 #define BITS_BIG_ENDIAN 0 41 #define BYTES_BIG_ENDIAN 0 42 #define WORDS_BIG_ENDIAN 0 43 44 /* Chosen such that we won't have to deal with multi-word subregs. */ 45 #define UNITS_PER_WORD 8 46 47 /* Alignments in bits. */ 48 #define PARM_BOUNDARY 32 49 #define STACK_BOUNDARY 128 50 #define FUNCTION_BOUNDARY 32 51 #define BIGGEST_ALIGNMENT 128 52 #define STRICT_ALIGNMENT 1 53 54 #define MAX_STACK_ALIGNMENT (1024 * 8) 55 56 #define DATA_ALIGNMENT nvptx_data_alignment 57 58 /* Copied from elf.h and other places. We'd otherwise use 59 BIGGEST_ALIGNMENT and fail a number of testcases. */ 60 #define MAX_OFILE_ALIGNMENT (32768 * 8) 61 62 /* Type Layout. */ 63 64 #define DEFAULT_SIGNED_CHAR 1 65 66 #define SHORT_TYPE_SIZE 16 67 #define INT_TYPE_SIZE 32 68 #define LONG_TYPE_SIZE (TARGET_ABI64 ? 64 : 32) 69 #define LONG_LONG_TYPE_SIZE 64 70 #define FLOAT_TYPE_SIZE 32 71 #define DOUBLE_TYPE_SIZE 64 72 #define LONG_DOUBLE_TYPE_SIZE 64 73 #define TARGET_SUPPORTS_WIDE_INT 1 74 75 #undef SIZE_TYPE 76 #define SIZE_TYPE (TARGET_ABI64 ? "long unsigned int" : "unsigned int") 77 #undef PTRDIFF_TYPE 78 #define PTRDIFF_TYPE (TARGET_ABI64 ? "long int" : "int") 79 80 #define POINTER_SIZE (TARGET_ABI64 ? 64 : 32) 81 #define Pmode (TARGET_ABI64 ? DImode : SImode) 82 #define STACK_SIZE_MODE Pmode 83 84 #include "nvptx-gen.h" 85 86 #define TARGET_PTX_6_0 (ptx_version_option >= PTX_VERSION_6_0) 87 #define TARGET_PTX_6_3 (ptx_version_option >= PTX_VERSION_6_3) 88 #define TARGET_PTX_7_0 (ptx_version_option >= PTX_VERSION_7_0) 89 90 /* Registers. Since ptx is a virtual target, we just define a few 91 hard registers for special purposes and leave pseudos unallocated. 92 We have to have some available hard registers, to keep gcc setup 93 happy. */ 94 #define FIRST_PSEUDO_REGISTER 16 95 #define FIXED_REGISTERS { 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0 } 96 #define CALL_USED_REGISTERS { 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 } 97 98 /* Register Classes. */ 99 enum reg_class { NO_REGS, ALL_REGS, LIM_REG_CLASSES }; 100 #define REG_CLASS_NAMES { "NO_REGS", "ALL_REGS" } 101 #define REG_CLASS_CONTENTS { { 0x0000 }, { 0xFFFF } } 102 #define N_REG_CLASSES (int) LIM_REG_CLASSES 103 104 #define GENERAL_REGS ALL_REGS 105 #define REGNO_REG_CLASS(R) ((void)(R), ALL_REGS) 106 #define BASE_REG_CLASS ALL_REGS 107 #define INDEX_REG_CLASS NO_REGS 108 109 #define REGNO_OK_FOR_BASE_P(X) true 110 #define REGNO_OK_FOR_INDEX_P(X) false 111 112 #define CLASS_MAX_NREGS(class, mode) \ 113 ((GET_MODE_SIZE (mode) + UNITS_PER_WORD - 1) / UNITS_PER_WORD) 114 115 #define PROMOTE_MODE(MODE, UNSIGNEDP, TYPE) \ 116 if ((MODE) == QImode || (MODE) == HImode) \ 117 { \ 118 (MODE) = SImode; \ 119 (void)(UNSIGNEDP); \ 120 (void)(TYPE); \ 121 } 122 123 /* Stack and Calling. */ 124 125 #define FRAME_GROWS_DOWNWARD 0 126 #define STACK_GROWS_DOWNWARD 1 127 128 #define NVPTX_RETURN_REGNUM 0 129 #define STACK_POINTER_REGNUM 1 130 #define FRAME_POINTER_REGNUM 2 131 #define ARG_POINTER_REGNUM 3 132 #define STATIC_CHAIN_REGNUM 4 133 /* This register points to the shared memory location with the current warp's 134 soft stack pointer (__nvptx_stacks[tid.y]). */ 135 #define SOFTSTACK_SLOT_REGNUM 5 136 /* This register is used to save the previous value of the soft stack pointer 137 in the prologue and restore it when returning. */ 138 #define SOFTSTACK_PREV_REGNUM 6 139 140 #define REGISTER_NAMES \ 141 { \ 142 "%value", "%stack", "%frame", "%args", \ 143 "%chain", "%sspslot", "%sspprev", "%hr7", \ 144 "%hr8", "%hr9", "%hr10", "%hr11", "%hr12", "%hr13", "%hr14", "%hr15" \ 145 } 146 147 #define FIRST_PARM_OFFSET(FNDECL) ((void)(FNDECL), 0) 148 #define PUSH_ARGS_REVERSED 1 149 #define ACCUMULATE_OUTGOING_ARGS 1 150 151 /* Avoid using the argument pointer for frame-related things. */ 152 #define FRAME_POINTER_CFA_OFFSET(FNDECL) ((void)(FNDECL), 0) 153 154 #ifdef HOST_WIDE_INT 155 struct nvptx_args { 156 tree fntype; 157 /* Number of arguments passed in registers so far. */ 158 int count; 159 }; 160 #endif 161 162 #define CUMULATIVE_ARGS struct nvptx_args 163 164 #define INIT_CUMULATIVE_ARGS(CUM, FNTYPE, LIBNAME, FNDECL, N_NAMED_ARGS) \ 165 ((CUM).fntype = (FNTYPE), (CUM).count = 0, (void)0) 166 167 #define FUNCTION_ARG_REGNO_P(r) 0 168 169 #define DEFAULT_PCC_STRUCT_RETURN 0 170 171 #define FUNCTION_PROFILER(file, labelno) \ 172 fatal_error (input_location, \ 173 "profiling is not yet implemented for this architecture") 174 175 #define TRAMPOLINE_SIZE 32 176 #define TRAMPOLINE_ALIGNMENT 256 177 178 /* We don't run reload, so this isn't actually used, but it still needs to be 179 defined. Showing an argp->fp elimination also stops 180 expand_builtin_setjmp_receiver from generating invalid insns. */ 181 #define ELIMINABLE_REGS \ 182 { \ 183 { ARG_POINTER_REGNUM, FRAME_POINTER_REGNUM} \ 184 } 185 186 /* Define the offset between two registers, one to be eliminated, and the other 187 its replacement, at the start of a routine. */ 188 189 #define INITIAL_ELIMINATION_OFFSET(FROM, TO, OFFSET) \ 190 ((OFFSET) = 0) 191 192 /* Addressing Modes. */ 193 194 #define MAX_REGS_PER_ADDRESS 1 195 196 #define LEGITIMATE_PIC_OPERAND_P(X) 1 197 198 199 #if defined HOST_WIDE_INT 200 struct GTY(()) machine_function 201 { 202 rtx_expr_list *call_args; /* Arg list for the current call. */ 203 bool doing_call; /* Within a CALL_ARGS ... CALL_ARGS_END sequence. */ 204 bool is_varadic; /* This call is varadic */ 205 bool has_varadic; /* Current function has a varadic call. */ 206 bool has_chain; /* Current function has outgoing static chain. */ 207 bool has_softstack; /* Current function has a soft stack frame. */ 208 bool has_simtreg; /* Current function has an OpenMP SIMD region. */ 209 int num_args; /* Number of args of current call. */ 210 int return_mode; /* Return mode of current fn. 211 (machine_mode not defined yet.) */ 212 rtx axis_predicate[2]; /* Neutering predicates. */ 213 int axis_dim[2]; /* Maximum number of threads on each axis, dim[0] is 214 vector_length, dim[1] is num_workers. */ 215 bool axis_dim_init_p; 216 rtx bcast_partition; /* Register containing the size of each 217 vector's partition of share-memory used to 218 broadcast state. */ 219 rtx red_partition; /* Similar to bcast_partition, except for vector 220 reductions. */ 221 rtx sync_bar; /* Synchronization barrier ID for vectors. */ 222 rtx unisimt_master; /* 'Master lane index' for -muniform-simt. */ 223 rtx unisimt_predicate; /* Predicate for -muniform-simt. */ 224 rtx unisimt_outside_simt_predicate; /* Predicate for -muniform-simt. */ 225 rtx unisimt_location; /* Mask location for -muniform-simt. */ 226 /* The following two fields hold the maximum size resp. alignment required 227 for per-lane storage in OpenMP SIMD regions. */ 228 unsigned HOST_WIDE_INT simt_stack_size; 229 unsigned HOST_WIDE_INT simt_stack_align; 230 }; 231 #endif 232 233 /* Costs. */ 234 235 #define NO_FUNCTION_CSE 1 236 #define SLOW_BYTE_ACCESS 0 237 #define BRANCH_COST(speed_p, predictable_p) 6 238 239 /* Assembler Format. */ 240 241 #undef ASM_DECLARE_FUNCTION_NAME 242 #define ASM_DECLARE_FUNCTION_NAME(FILE, NAME, DECL) \ 243 nvptx_declare_function_name (FILE, NAME, DECL) 244 245 #undef ASM_DECLARE_FUNCTION_SIZE 246 #define ASM_DECLARE_FUNCTION_SIZE(STREAM, NAME, DECL) \ 247 nvptx_function_end (STREAM) 248 249 #define DWARF2_ASM_LINE_DEBUG_INFO 1 250 251 #undef ASM_APP_ON 252 #define ASM_APP_ON "\t// #APP \n" 253 #undef ASM_APP_OFF 254 #define ASM_APP_OFF "\t// #NO_APP \n" 255 256 #define DBX_REGISTER_NUMBER(N) N 257 258 #define TEXT_SECTION_ASM_OP "" 259 #define DATA_SECTION_ASM_OP "" 260 261 #undef ASM_GENERATE_INTERNAL_LABEL 262 #define ASM_GENERATE_INTERNAL_LABEL(LABEL, PREFIX, NUM) \ 263 do \ 264 { \ 265 char *__p; \ 266 __p = stpcpy (&(LABEL)[1], PREFIX); \ 267 (LABEL)[0] = '$'; \ 268 sprint_ul (__p, (unsigned long) (NUM)); \ 269 } \ 270 while (0) 271 272 #define ASM_OUTPUT_ALIGN(FILE, POWER) \ 273 do \ 274 { \ 275 (void) (FILE); \ 276 (void) (POWER); \ 277 } \ 278 while (0) 279 280 #define ASM_OUTPUT_SKIP(FILE, N) \ 281 nvptx_output_skip (FILE, N) 282 283 #undef ASM_OUTPUT_ASCII 284 #define ASM_OUTPUT_ASCII(FILE, STR, LENGTH) \ 285 nvptx_output_ascii (FILE, STR, LENGTH); 286 287 #define ASM_DECLARE_OBJECT_NAME(FILE, NAME, DECL) \ 288 nvptx_declare_object_name (FILE, NAME, DECL) 289 290 #undef ASM_OUTPUT_ALIGNED_DECL_COMMON 291 #define ASM_OUTPUT_ALIGNED_DECL_COMMON(FILE, DECL, NAME, SIZE, ALIGN) \ 292 nvptx_output_aligned_decl (FILE, NAME, DECL, SIZE, ALIGN) 293 294 #undef ASM_OUTPUT_ALIGNED_DECL_LOCAL 295 #define ASM_OUTPUT_ALIGNED_DECL_LOCAL(FILE, DECL, NAME, SIZE, ALIGN) \ 296 nvptx_output_aligned_decl (FILE, NAME, DECL, SIZE, ALIGN) 297 298 #define CASE_VECTOR_PC_RELATIVE flag_pic 299 #define JUMP_TABLES_IN_TEXT_SECTION flag_pic 300 301 #define ADDR_VEC_ALIGN(VEC) (JUMP_TABLES_IN_TEXT_SECTION ? 5 : 2) 302 303 /* Misc. */ 304 305 #define DWARF2_LINENO_DEBUGGING_INFO 1 306 307 #define CLZ_DEFINED_VALUE_AT_ZERO(MODE, VALUE) \ 308 ((VALUE) = GET_MODE_BITSIZE ((MODE)), 2) 309 #define CTZ_DEFINED_VALUE_AT_ZERO(MODE, VALUE) \ 310 ((VALUE) = GET_MODE_BITSIZE ((MODE)), 2) 311 312 #define SUPPORTS_WEAK 1 313 314 /* The documentation states that ASM_OUTPUT_DEF_FROM_DECLS is used in 315 preference to ASM_OUTPUT_DEF if the tree nodes are available. However, we 316 need the tree nodes to emit the prototype, so at this point it's not clear 317 how we can support ASM_OUTPUT_DEF. Still, we need to define it, or 318 ASM_OUTPUT_DEF_FROM_DECLS is ignored. For now, assert, and once we run 319 into it possibly improve by somehow emitting the prototype elsewhere, or 320 emitting a reasonable error message. */ 321 #define ASM_OUTPUT_DEF(FILE,LABEL1,LABEL2) \ 322 do \ 323 { \ 324 gcc_unreachable (); \ 325 } \ 326 while (0) 327 #define ASM_OUTPUT_DEF_FROM_DECLS(STREAM, NAME, VALUE) \ 328 nvptx_asm_output_def_from_decls (STREAM, NAME, VALUE) 329 330 #define NO_DOT_IN_LABEL 331 #define ASM_COMMENT_START "//" 332 333 #define STORE_FLAG_VALUE 1 334 #define FLOAT_STORE_FLAG_VALUE(MODE) REAL_VALUE_ATOF("1.0", (MODE)) 335 336 #define CASE_VECTOR_MODE SImode 337 #define MOVE_MAX 8 338 #define MOVE_RATIO(SPEED) 4 339 #define FUNCTION_MODE QImode 340 #define HAS_INIT_SECTION 1 341 342 /* The C++ front end insists to link against libstdc++ -- which we don't build. 343 Tell it to instead link against the innocuous libgcc. */ 344 #define LIBSTDCXX "gcc" 345 346 #endif /* GCC_NVPTX_H */ 347