| File: | build/gcc/omp-offload.cc |
| Warning: | line 767, column 8 Value stored to 'span' during its initialization is never read |
Press '?' to see keyboard shortcuts
Keyboard shortcuts:
| 1 | /* Bits of OpenMP and OpenACC handling that is specific to device offloading |
| 2 | and a lowering pass for OpenACC device directives. |
| 3 | |
| 4 | Copyright (C) 2005-2023 Free Software Foundation, Inc. |
| 5 | |
| 6 | This file is part of GCC. |
| 7 | |
| 8 | GCC is free software; you can redistribute it and/or modify it under |
| 9 | the terms of the GNU General Public License as published by the Free |
| 10 | Software Foundation; either version 3, or (at your option) any later |
| 11 | version. |
| 12 | |
| 13 | GCC is distributed in the hope that it will be useful, but WITHOUT ANY |
| 14 | WARRANTY; without even the implied warranty of MERCHANTABILITY or |
| 15 | FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
| 16 | for more details. |
| 17 | |
| 18 | You should have received a copy of the GNU General Public License |
| 19 | along with GCC; see the file COPYING3. If not see |
| 20 | <http://www.gnu.org/licenses/>. */ |
| 21 | |
| 22 | #include "config.h" |
| 23 | #include "system.h" |
| 24 | #include "coretypes.h" |
| 25 | #include "backend.h" |
| 26 | #include "target.h" |
| 27 | #include "tree.h" |
| 28 | #include "gimple.h" |
| 29 | #include "tree-pass.h" |
| 30 | #include "ssa.h" |
| 31 | #include "cgraph.h" |
| 32 | #include "pretty-print.h" |
| 33 | #include "diagnostic-core.h" |
| 34 | #include "fold-const.h" |
| 35 | #include "internal-fn.h" |
| 36 | #include "langhooks.h" |
| 37 | #include "gimplify.h" |
| 38 | #include "gimple-iterator.h" |
| 39 | #include "gimplify-me.h" |
| 40 | #include "gimple-walk.h" |
| 41 | #include "tree-cfg.h" |
| 42 | #include "tree-into-ssa.h" |
| 43 | #include "tree-nested.h" |
| 44 | #include "stor-layout.h" |
| 45 | #include "common/common-target.h" |
| 46 | #include "omp-general.h" |
| 47 | #include "omp-offload.h" |
| 48 | #include "lto-section-names.h" |
| 49 | #include "gomp-constants.h" |
| 50 | #include "gimple-pretty-print.h" |
| 51 | #include "intl.h" |
| 52 | #include "stringpool.h" |
| 53 | #include "attribs.h" |
| 54 | #include "cfgloop.h" |
| 55 | #include "context.h" |
| 56 | #include "convert.h" |
| 57 | #include "opts.h" |
| 58 | |
| 59 | /* Describe the OpenACC looping structure of a function. The entire |
| 60 | function is held in a 'NULL' loop. */ |
| 61 | |
| 62 | struct oacc_loop |
| 63 | { |
| 64 | oacc_loop *parent; /* Containing loop. */ |
| 65 | |
| 66 | oacc_loop *child; /* First inner loop. */ |
| 67 | |
| 68 | oacc_loop *sibling; /* Next loop within same parent. */ |
| 69 | |
| 70 | location_t loc; /* Location of the loop start. */ |
| 71 | |
| 72 | gcall *marker; /* Initial head marker. */ |
| 73 | |
| 74 | gcall *heads[GOMP_DIM_MAX3]; /* Head marker functions. */ |
| 75 | gcall *tails[GOMP_DIM_MAX3]; /* Tail marker functions. */ |
| 76 | |
| 77 | tree routine; /* Pseudo-loop enclosing a routine. */ |
| 78 | |
| 79 | unsigned mask; /* Partitioning mask. */ |
| 80 | unsigned e_mask; /* Partitioning of element loops (when tiling). */ |
| 81 | unsigned inner; /* Partitioning of inner loops. */ |
| 82 | unsigned flags; /* Partitioning flags. */ |
| 83 | vec<gcall *> ifns; /* Contained loop abstraction functions. */ |
| 84 | tree chunk_size; /* Chunk size. */ |
| 85 | gcall *head_end; /* Final marker of head sequence. */ |
| 86 | }; |
| 87 | |
| 88 | /* Holds offload tables with decls. */ |
| 89 | vec<tree, va_gc> *offload_funcs, *offload_vars; |
| 90 | |
| 91 | /* Return level at which oacc routine may spawn a partitioned loop, or |
| 92 | -1 if it is not a routine (i.e. is an offload fn). */ |
| 93 | |
| 94 | int |
| 95 | oacc_fn_attrib_level (tree attr) |
| 96 | { |
| 97 | tree pos = TREE_VALUE (attr)((tree_check ((attr), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 97, __FUNCTION__, (TREE_LIST)))->list.value); |
| 98 | |
| 99 | if (!TREE_PURPOSE (pos)((tree_check ((pos), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 99, __FUNCTION__, (TREE_LIST)))->list.purpose)) |
| 100 | return -1; |
| 101 | |
| 102 | int ix = 0; |
| 103 | for (ix = 0; ix != GOMP_DIM_MAX3; |
| 104 | ix++, pos = TREE_CHAIN (pos)((contains_struct_check ((pos), (TS_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 104, __FUNCTION__))->common.chain)) |
| 105 | if (!integer_zerop (TREE_PURPOSE (pos)((tree_check ((pos), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 105, __FUNCTION__, (TREE_LIST)))->list.purpose))) |
| 106 | break; |
| 107 | |
| 108 | return ix; |
| 109 | } |
| 110 | |
| 111 | /* Helper function for omp_finish_file routine. Takes decls from V_DECLS and |
| 112 | adds their addresses and sizes to constructor-vector V_CTOR. */ |
| 113 | |
| 114 | static void |
| 115 | add_decls_addresses_to_decl_constructor (vec<tree, va_gc> *v_decls, |
| 116 | vec<constructor_elt, va_gc> *v_ctor) |
| 117 | { |
| 118 | unsigned len = vec_safe_length (v_decls); |
| 119 | for (unsigned i = 0; i < len; i++) |
| 120 | { |
| 121 | tree it = (*v_decls)[i]; |
| 122 | bool is_var = VAR_P (it)(((enum tree_code) (it)->base.code) == VAR_DECL); |
| 123 | bool is_link_var |
| 124 | = is_var |
| 125 | #ifdef ACCEL_COMPILER |
| 126 | && DECL_HAS_VALUE_EXPR_P (it)((tree_check3 ((it), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 126, __FUNCTION__, (VAR_DECL), (PARM_DECL), (RESULT_DECL))) ->decl_common.decl_flag_2) |
| 127 | #endif |
| 128 | && lookup_attribute ("omp declare target link", DECL_ATTRIBUTES (it)((contains_struct_check ((it), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 128, __FUNCTION__))->decl_common.attributes)); |
| 129 | |
| 130 | /* See also omp_finish_file and output_offload_tables in lto-cgraph.cc. */ |
| 131 | if (!in_lto_pglobal_options.x_in_lto_p && !symtab_node::get (it)) |
| 132 | continue; |
| 133 | |
| 134 | tree size = NULL_TREE(tree) __null; |
| 135 | if (is_var) |
| 136 | size = fold_convert (const_ptr_type_node, DECL_SIZE_UNIT (it))fold_convert_loc (((location_t) 0), global_trees[TI_CONST_PTR_TYPE ], ((contains_struct_check ((it), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 136, __FUNCTION__))->decl_common.size_unit)); |
| 137 | |
| 138 | tree addr; |
| 139 | if (!is_link_var) |
| 140 | addr = build_fold_addr_expr (it)build_fold_addr_expr_loc (((location_t) 0), (it)); |
| 141 | else |
| 142 | { |
| 143 | #ifdef ACCEL_COMPILER |
| 144 | /* For "omp declare target link" vars add address of the pointer to |
| 145 | the target table, instead of address of the var. */ |
| 146 | tree value_expr = DECL_VALUE_EXPR (it)(decl_value_expr_lookup ((contains_struct_check ((it), (TS_DECL_WRTL ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 146, __FUNCTION__)))); |
| 147 | tree link_ptr_decl = TREE_OPERAND (value_expr, 0)(*((const_cast<tree*> (tree_operand_check ((value_expr) , (0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 147, __FUNCTION__))))); |
| 148 | varpool_node::finalize_decl (link_ptr_decl); |
| 149 | addr = build_fold_addr_expr (link_ptr_decl)build_fold_addr_expr_loc (((location_t) 0), (link_ptr_decl)); |
| 150 | #else |
| 151 | addr = build_fold_addr_expr (it)build_fold_addr_expr_loc (((location_t) 0), (it)); |
| 152 | #endif |
| 153 | |
| 154 | /* Most significant bit of the size marks "omp declare target link" |
| 155 | vars in host and target tables. */ |
| 156 | unsigned HOST_WIDE_INTlong isize = tree_to_uhwi (size); |
| 157 | isize |= 1ULL << (int_size_in_bytes (const_ptr_type_nodeglobal_trees[TI_CONST_PTR_TYPE]) |
| 158 | * BITS_PER_UNIT(8) - 1); |
| 159 | size = wide_int_to_tree (const_ptr_type_nodeglobal_trees[TI_CONST_PTR_TYPE], isize); |
| 160 | } |
| 161 | |
| 162 | CONSTRUCTOR_APPEND_ELT (v_ctor, NULL_TREE, addr)do { constructor_elt _ce___ = {(tree) __null, addr}; vec_safe_push ((v_ctor), _ce___); } while (0); |
| 163 | if (is_var) |
| 164 | CONSTRUCTOR_APPEND_ELT (v_ctor, NULL_TREE, size)do { constructor_elt _ce___ = {(tree) __null, size}; vec_safe_push ((v_ctor), _ce___); } while (0); |
| 165 | } |
| 166 | } |
| 167 | |
| 168 | /* Return true if DECL is a function for which its references should be |
| 169 | analyzed. */ |
| 170 | |
| 171 | static bool |
| 172 | omp_declare_target_fn_p (tree decl) |
| 173 | { |
| 174 | return (TREE_CODE (decl)((enum tree_code) (decl)->base.code) == FUNCTION_DECL |
| 175 | && lookup_attribute ("omp declare target", DECL_ATTRIBUTES (decl)((contains_struct_check ((decl), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 175, __FUNCTION__))->decl_common.attributes)) |
| 176 | && !lookup_attribute ("omp declare target host", |
| 177 | DECL_ATTRIBUTES (decl)((contains_struct_check ((decl), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 177, __FUNCTION__))->decl_common.attributes)) |
| 178 | && (!flag_openaccglobal_options.x_flag_openacc |
| 179 | || oacc_get_fn_attrib (decl) == NULL_TREE(tree) __null)); |
| 180 | } |
| 181 | |
| 182 | /* Return true if DECL Is a variable for which its initializer references |
| 183 | should be analyzed. */ |
| 184 | |
| 185 | static bool |
| 186 | omp_declare_target_var_p (tree decl) |
| 187 | { |
| 188 | return (VAR_P (decl)(((enum tree_code) (decl)->base.code) == VAR_DECL) |
| 189 | && lookup_attribute ("omp declare target", DECL_ATTRIBUTES (decl)((contains_struct_check ((decl), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 189, __FUNCTION__))->decl_common.attributes)) |
| 190 | && !lookup_attribute ("omp declare target link", |
| 191 | DECL_ATTRIBUTES (decl)((contains_struct_check ((decl), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 191, __FUNCTION__))->decl_common.attributes))); |
| 192 | } |
| 193 | |
| 194 | /* Helper function for omp_discover_implicit_declare_target, called through |
| 195 | walk_tree. Mark referenced FUNCTION_DECLs implicitly as |
| 196 | declare target to. */ |
| 197 | |
| 198 | static tree |
| 199 | omp_discover_declare_target_tgt_fn_r (tree *tp, int *walk_subtrees, void *data) |
| 200 | { |
| 201 | if (TREE_CODE (*tp)((enum tree_code) (*tp)->base.code) == CALL_EXPR |
| 202 | && CALL_EXPR_FN (*tp)(*((const_cast<tree*> (tree_operand_check (((tree_check ((*tp), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 202, __FUNCTION__, (CALL_EXPR)))), (1), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 202, __FUNCTION__))))) |
| 203 | && TREE_CODE (CALL_EXPR_FN (*tp))((enum tree_code) ((*((const_cast<tree*> (tree_operand_check (((tree_check ((*tp), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 203, __FUNCTION__, (CALL_EXPR)))), (1), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 203, __FUNCTION__))))))->base.code) == ADDR_EXPR |
| 204 | && TREE_CODE (TREE_OPERAND (CALL_EXPR_FN (*tp), 0))((enum tree_code) ((*((const_cast<tree*> (tree_operand_check (((*((const_cast<tree*> (tree_operand_check (((tree_check ((*tp), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 204, __FUNCTION__, (CALL_EXPR)))), (1), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 204, __FUNCTION__)))))), (0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 204, __FUNCTION__))))))->base.code) == FUNCTION_DECL |
| 205 | && lookup_attribute ("omp declare variant base", |
| 206 | DECL_ATTRIBUTES (TREE_OPERAND (CALL_EXPR_FN (*tp),((contains_struct_check (((*((const_cast<tree*> (tree_operand_check (((*((const_cast<tree*> (tree_operand_check (((tree_check ((*tp), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 206, __FUNCTION__, (CALL_EXPR)))), (1), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 206, __FUNCTION__)))))), (0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 207, __FUNCTION__)))))), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 207, __FUNCTION__))->decl_common.attributes) |
| 207 | 0))((contains_struct_check (((*((const_cast<tree*> (tree_operand_check (((*((const_cast<tree*> (tree_operand_check (((tree_check ((*tp), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 206, __FUNCTION__, (CALL_EXPR)))), (1), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 206, __FUNCTION__)))))), (0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 207, __FUNCTION__)))))), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 207, __FUNCTION__))->decl_common.attributes))) |
| 208 | { |
| 209 | tree fn = TREE_OPERAND (CALL_EXPR_FN (*tp), 0)(*((const_cast<tree*> (tree_operand_check (((*((const_cast <tree*> (tree_operand_check (((tree_check ((*tp), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 209, __FUNCTION__, (CALL_EXPR)))), (1), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 209, __FUNCTION__)))))), (0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 209, __FUNCTION__))))); |
| 210 | for (tree attr = DECL_ATTRIBUTES (fn)((contains_struct_check ((fn), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 210, __FUNCTION__))->decl_common.attributes); attr; attr = TREE_CHAIN (attr)((contains_struct_check ((attr), (TS_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 210, __FUNCTION__))->common.chain)) |
| 211 | { |
| 212 | attr = lookup_attribute ("omp declare variant base", attr); |
| 213 | if (attr == NULL_TREE(tree) __null) |
| 214 | break; |
| 215 | tree purpose = TREE_PURPOSE (TREE_VALUE (attr))((tree_check ((((tree_check ((attr), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 215, __FUNCTION__, (TREE_LIST)))->list.value)), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 215, __FUNCTION__, (TREE_LIST)))->list.purpose); |
| 216 | if (TREE_CODE (purpose)((enum tree_code) (purpose)->base.code) == FUNCTION_DECL) |
| 217 | omp_discover_declare_target_tgt_fn_r (&purpose, walk_subtrees, data); |
| 218 | } |
| 219 | } |
| 220 | else if (TREE_CODE (*tp)((enum tree_code) (*tp)->base.code) == FUNCTION_DECL) |
| 221 | { |
| 222 | tree decl = *tp; |
| 223 | tree id = get_identifier ("omp declare target")(__builtin_constant_p ("omp declare target") ? get_identifier_with_length (("omp declare target"), strlen ("omp declare target")) : get_identifier ("omp declare target")); |
| 224 | symtab_node *node = symtab_node::get (*tp); |
| 225 | if (node != NULL__null) |
| 226 | { |
| 227 | while (node->alias_target |
| 228 | && TREE_CODE (node->alias_target)((enum tree_code) (node->alias_target)->base.code) == FUNCTION_DECL) |
| 229 | { |
| 230 | if (!omp_declare_target_fn_p (node->decl) |
| 231 | && !lookup_attribute ("omp declare target host", |
| 232 | DECL_ATTRIBUTES (node->decl)((contains_struct_check ((node->decl), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 232, __FUNCTION__))->decl_common.attributes))) |
| 233 | { |
| 234 | node->offloadable = 1; |
| 235 | DECL_ATTRIBUTES (node->decl)((contains_struct_check ((node->decl), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 235, __FUNCTION__))->decl_common.attributes) |
| 236 | = tree_cons (id, NULL_TREE(tree) __null, DECL_ATTRIBUTES (node->decl)((contains_struct_check ((node->decl), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 236, __FUNCTION__))->decl_common.attributes)); |
| 237 | } |
| 238 | node = symtab_node::get (node->alias_target); |
| 239 | } |
| 240 | symtab_node *new_node = node->ultimate_alias_target (); |
| 241 | decl = new_node->decl; |
| 242 | while (node != new_node) |
| 243 | { |
| 244 | if (!omp_declare_target_fn_p (node->decl) |
| 245 | && !lookup_attribute ("omp declare target host", |
| 246 | DECL_ATTRIBUTES (node->decl)((contains_struct_check ((node->decl), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 246, __FUNCTION__))->decl_common.attributes))) |
| 247 | { |
| 248 | node->offloadable = 1; |
| 249 | DECL_ATTRIBUTES (node->decl)((contains_struct_check ((node->decl), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 249, __FUNCTION__))->decl_common.attributes) |
| 250 | = tree_cons (id, NULL_TREE(tree) __null, DECL_ATTRIBUTES (node->decl)((contains_struct_check ((node->decl), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 250, __FUNCTION__))->decl_common.attributes)); |
| 251 | } |
| 252 | gcc_assert (node->alias && node->analyzed)((void)(!(node->alias && node->analyzed) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 252, __FUNCTION__), 0 : 0)); |
| 253 | node = node->get_alias_target (); |
| 254 | } |
| 255 | node->offloadable = 1; |
| 256 | if (ENABLE_OFFLOADING0) |
| 257 | g->have_offload = true; |
| 258 | } |
| 259 | if (omp_declare_target_fn_p (decl) |
| 260 | || lookup_attribute ("omp declare target host", |
| 261 | DECL_ATTRIBUTES (decl)((contains_struct_check ((decl), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 261, __FUNCTION__))->decl_common.attributes))) |
| 262 | return NULL_TREE(tree) __null; |
| 263 | |
| 264 | if (!DECL_EXTERNAL (decl)((contains_struct_check ((decl), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 264, __FUNCTION__))->decl_common.decl_flag_1) && DECL_SAVED_TREE (decl)((tree_check ((decl), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 264, __FUNCTION__, (FUNCTION_DECL)))->function_decl.saved_tree )) |
| 265 | ((vec<tree> *) data)->safe_push (decl); |
| 266 | DECL_ATTRIBUTES (decl)((contains_struct_check ((decl), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 266, __FUNCTION__))->decl_common.attributes) = tree_cons (id, NULL_TREE(tree) __null, |
| 267 | DECL_ATTRIBUTES (decl)((contains_struct_check ((decl), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 267, __FUNCTION__))->decl_common.attributes)); |
| 268 | } |
| 269 | else if (TYPE_P (*tp)(tree_code_type_tmpl <0>::tree_code_type[(int) (((enum tree_code ) (*tp)->base.code))] == tcc_type)) |
| 270 | *walk_subtrees = 0; |
| 271 | else if (TREE_CODE (*tp)((enum tree_code) (*tp)->base.code) == OMP_TARGET) |
| 272 | { |
| 273 | tree c = omp_find_clause (OMP_CLAUSES (*tp)(*((const_cast<tree*> (tree_operand_check (((tree_range_check ((*tp), (OACC_PARALLEL), (OMP_SCAN), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 273, __FUNCTION__))), (1), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 273, __FUNCTION__))))), OMP_CLAUSE_DEVICE); |
| 274 | if (c && OMP_CLAUSE_DEVICE_ANCESTOR (c)((omp_clause_subcode_check ((c), (OMP_CLAUSE_DEVICE), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 274, __FUNCTION__))->base.public_flag)) |
| 275 | *walk_subtrees = 0; |
| 276 | } |
| 277 | return NULL_TREE(tree) __null; |
| 278 | } |
| 279 | |
| 280 | /* Similarly, but ignore references outside of OMP_TARGET regions. */ |
| 281 | |
| 282 | static tree |
| 283 | omp_discover_declare_target_fn_r (tree *tp, int *walk_subtrees, void *data) |
| 284 | { |
| 285 | if (TREE_CODE (*tp)((enum tree_code) (*tp)->base.code) == OMP_TARGET) |
| 286 | { |
| 287 | tree c = omp_find_clause (OMP_CLAUSES (*tp)(*((const_cast<tree*> (tree_operand_check (((tree_range_check ((*tp), (OACC_PARALLEL), (OMP_SCAN), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 287, __FUNCTION__))), (1), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 287, __FUNCTION__))))), OMP_CLAUSE_DEVICE); |
| 288 | if (!c || !OMP_CLAUSE_DEVICE_ANCESTOR (c)((omp_clause_subcode_check ((c), (OMP_CLAUSE_DEVICE), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 288, __FUNCTION__))->base.public_flag)) |
| 289 | walk_tree_without_duplicates (&OMP_TARGET_BODY (*tp),walk_tree_without_duplicates_1 (&(*((const_cast<tree*> (tree_operand_check (((tree_check ((*tp), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 289, __FUNCTION__, (OMP_TARGET)))), (0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 289, __FUNCTION__))))), omp_discover_declare_target_tgt_fn_r , data, __null) |
| 290 | omp_discover_declare_target_tgt_fn_r,walk_tree_without_duplicates_1 (&(*((const_cast<tree*> (tree_operand_check (((tree_check ((*tp), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 289, __FUNCTION__, (OMP_TARGET)))), (0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 289, __FUNCTION__))))), omp_discover_declare_target_tgt_fn_r , data, __null) |
| 291 | data)walk_tree_without_duplicates_1 (&(*((const_cast<tree*> (tree_operand_check (((tree_check ((*tp), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 289, __FUNCTION__, (OMP_TARGET)))), (0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 289, __FUNCTION__))))), omp_discover_declare_target_tgt_fn_r , data, __null); |
| 292 | *walk_subtrees = 0; |
| 293 | } |
| 294 | else if (TYPE_P (*tp)(tree_code_type_tmpl <0>::tree_code_type[(int) (((enum tree_code ) (*tp)->base.code))] == tcc_type)) |
| 295 | *walk_subtrees = 0; |
| 296 | return NULL_TREE(tree) __null; |
| 297 | } |
| 298 | |
| 299 | /* Helper function for omp_discover_implicit_declare_target, called through |
| 300 | walk_tree. Mark referenced FUNCTION_DECLs implicitly as |
| 301 | declare target to. */ |
| 302 | |
| 303 | static tree |
| 304 | omp_discover_declare_target_var_r (tree *tp, int *walk_subtrees, void *data) |
| 305 | { |
| 306 | if (TREE_CODE (*tp)((enum tree_code) (*tp)->base.code) == FUNCTION_DECL) |
| 307 | return omp_discover_declare_target_tgt_fn_r (tp, walk_subtrees, data); |
| 308 | else if (VAR_P (*tp)(((enum tree_code) (*tp)->base.code) == VAR_DECL) |
| 309 | && is_global_var (*tp) |
| 310 | && !omp_declare_target_var_p (*tp)) |
| 311 | { |
| 312 | tree id = get_identifier ("omp declare target")(__builtin_constant_p ("omp declare target") ? get_identifier_with_length (("omp declare target"), strlen ("omp declare target")) : get_identifier ("omp declare target")); |
| 313 | if (lookup_attribute ("omp declare target link", DECL_ATTRIBUTES (*tp)((contains_struct_check ((*tp), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 313, __FUNCTION__))->decl_common.attributes))) |
| 314 | { |
| 315 | error_at (DECL_SOURCE_LOCATION (*tp)((contains_struct_check ((*tp), (TS_DECL_MINIMAL), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 315, __FUNCTION__))->decl_minimal.locus), |
| 316 | "%qD specified both in declare target %<link%> and " |
| 317 | "implicitly in %<to%> clauses", *tp); |
| 318 | DECL_ATTRIBUTES (*tp)((contains_struct_check ((*tp), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 318, __FUNCTION__))->decl_common.attributes) |
| 319 | = remove_attribute ("omp declare target link", DECL_ATTRIBUTES (*tp)((contains_struct_check ((*tp), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 319, __FUNCTION__))->decl_common.attributes)); |
| 320 | } |
| 321 | if (TREE_STATIC (*tp)((*tp)->base.static_flag) && lang_hooks.decls.omp_get_decl_init (*tp)) |
| 322 | ((vec<tree> *) data)->safe_push (*tp); |
| 323 | DECL_ATTRIBUTES (*tp)((contains_struct_check ((*tp), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 323, __FUNCTION__))->decl_common.attributes) = tree_cons (id, NULL_TREE(tree) __null, DECL_ATTRIBUTES (*tp)((contains_struct_check ((*tp), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 323, __FUNCTION__))->decl_common.attributes)); |
| 324 | symtab_node *node = symtab_node::get (*tp); |
| 325 | if (node != NULL__null && !node->offloadable) |
| 326 | { |
| 327 | node->offloadable = 1; |
| 328 | if (ENABLE_OFFLOADING0) |
| 329 | { |
| 330 | g->have_offload = true; |
| 331 | if (is_a <varpool_node *> (node)) |
| 332 | vec_safe_push (offload_vars, node->decl); |
| 333 | } |
| 334 | } |
| 335 | } |
| 336 | else if (TYPE_P (*tp)(tree_code_type_tmpl <0>::tree_code_type[(int) (((enum tree_code ) (*tp)->base.code))] == tcc_type)) |
| 337 | *walk_subtrees = 0; |
| 338 | return NULL_TREE(tree) __null; |
| 339 | } |
| 340 | |
| 341 | /* Perform the OpenMP implicit declare target to discovery. */ |
| 342 | |
| 343 | void |
| 344 | omp_discover_implicit_declare_target (void) |
| 345 | { |
| 346 | cgraph_node *node; |
| 347 | varpool_node *vnode; |
| 348 | auto_vec<tree> worklist; |
| 349 | |
| 350 | FOR_EACH_DEFINED_FUNCTION (node)for ((node) = symtab->first_defined_function (); (node); ( node) = symtab->next_defined_function ((node))) |
| 351 | if (DECL_SAVED_TREE (node->decl)((tree_check ((node->decl), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 351, __FUNCTION__, (FUNCTION_DECL)))->function_decl.saved_tree )) |
| 352 | { |
| 353 | struct cgraph_node *cgn; |
| 354 | if (omp_declare_target_fn_p (node->decl)) |
| 355 | worklist.safe_push (node->decl); |
| 356 | else if (DECL_STRUCT_FUNCTION (node->decl)((tree_check ((node->decl), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 356, __FUNCTION__, (FUNCTION_DECL)))->function_decl.f) |
| 357 | && DECL_STRUCT_FUNCTION (node->decl)((tree_check ((node->decl), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 357, __FUNCTION__, (FUNCTION_DECL)))->function_decl.f)->has_omp_target) |
| 358 | worklist.safe_push (node->decl); |
| 359 | for (cgn = first_nested_function (node); |
| 360 | cgn; cgn = next_nested_function (cgn)) |
| 361 | if (omp_declare_target_fn_p (cgn->decl)) |
| 362 | worklist.safe_push (cgn->decl); |
| 363 | else if (DECL_STRUCT_FUNCTION (cgn->decl)((tree_check ((cgn->decl), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 363, __FUNCTION__, (FUNCTION_DECL)))->function_decl.f) |
| 364 | && DECL_STRUCT_FUNCTION (cgn->decl)((tree_check ((cgn->decl), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 364, __FUNCTION__, (FUNCTION_DECL)))->function_decl.f)->has_omp_target) |
| 365 | worklist.safe_push (cgn->decl); |
| 366 | } |
| 367 | FOR_EACH_VARIABLE (vnode)for ((vnode) = symtab->first_variable (); (vnode); (vnode) = symtab->next_variable ((vnode))) |
| 368 | if (lang_hooks.decls.omp_get_decl_init (vnode->decl) |
| 369 | && omp_declare_target_var_p (vnode->decl)) |
| 370 | worklist.safe_push (vnode->decl); |
| 371 | while (!worklist.is_empty ()) |
| 372 | { |
| 373 | tree decl = worklist.pop (); |
| 374 | if (VAR_P (decl)(((enum tree_code) (decl)->base.code) == VAR_DECL)) |
| 375 | walk_tree_without_duplicates (lang_hooks.decls.omp_get_decl_init (decl),walk_tree_without_duplicates_1 (lang_hooks.decls.omp_get_decl_init (decl), omp_discover_declare_target_var_r, &worklist, __null ) |
| 376 | omp_discover_declare_target_var_r,walk_tree_without_duplicates_1 (lang_hooks.decls.omp_get_decl_init (decl), omp_discover_declare_target_var_r, &worklist, __null ) |
| 377 | &worklist)walk_tree_without_duplicates_1 (lang_hooks.decls.omp_get_decl_init (decl), omp_discover_declare_target_var_r, &worklist, __null ); |
| 378 | else if (omp_declare_target_fn_p (decl)) |
| 379 | walk_tree_without_duplicates (&DECL_SAVED_TREE (decl),walk_tree_without_duplicates_1 (&((tree_check ((decl), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 379, __FUNCTION__, (FUNCTION_DECL)))->function_decl.saved_tree ), omp_discover_declare_target_tgt_fn_r, &worklist, __null ) |
| 380 | omp_discover_declare_target_tgt_fn_r,walk_tree_without_duplicates_1 (&((tree_check ((decl), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 379, __FUNCTION__, (FUNCTION_DECL)))->function_decl.saved_tree ), omp_discover_declare_target_tgt_fn_r, &worklist, __null ) |
| 381 | &worklist)walk_tree_without_duplicates_1 (&((tree_check ((decl), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 379, __FUNCTION__, (FUNCTION_DECL)))->function_decl.saved_tree ), omp_discover_declare_target_tgt_fn_r, &worklist, __null ); |
| 382 | else |
| 383 | walk_tree_without_duplicates (&DECL_SAVED_TREE (decl),walk_tree_without_duplicates_1 (&((tree_check ((decl), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 383, __FUNCTION__, (FUNCTION_DECL)))->function_decl.saved_tree ), omp_discover_declare_target_fn_r, &worklist, __null) |
| 384 | omp_discover_declare_target_fn_r,walk_tree_without_duplicates_1 (&((tree_check ((decl), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 383, __FUNCTION__, (FUNCTION_DECL)))->function_decl.saved_tree ), omp_discover_declare_target_fn_r, &worklist, __null) |
| 385 | &worklist)walk_tree_without_duplicates_1 (&((tree_check ((decl), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 383, __FUNCTION__, (FUNCTION_DECL)))->function_decl.saved_tree ), omp_discover_declare_target_fn_r, &worklist, __null); |
| 386 | } |
| 387 | |
| 388 | lang_hooks.decls.omp_finish_decl_inits (); |
| 389 | } |
| 390 | |
| 391 | |
| 392 | /* Create new symbols containing (address, size) pairs for global variables, |
| 393 | marked with "omp declare target" attribute, as well as addresses for the |
| 394 | functions, which are outlined offloading regions. */ |
| 395 | void |
| 396 | omp_finish_file (void) |
| 397 | { |
| 398 | unsigned num_funcs = vec_safe_length (offload_funcs); |
| 399 | unsigned num_vars = vec_safe_length (offload_vars); |
| 400 | |
| 401 | if (num_funcs == 0 && num_vars == 0) |
| 402 | return; |
| 403 | |
| 404 | if (targetm_common.have_named_sections) |
| 405 | { |
| 406 | vec<constructor_elt, va_gc> *v_f, *v_v; |
| 407 | vec_alloc (v_f, num_funcs); |
| 408 | vec_alloc (v_v, num_vars * 2); |
| 409 | |
| 410 | add_decls_addresses_to_decl_constructor (offload_funcs, v_f); |
| 411 | add_decls_addresses_to_decl_constructor (offload_vars, v_v); |
| 412 | |
| 413 | tree vars_decl_type = build_array_type_nelts (pointer_sized_int_nodeglobal_trees[TI_POINTER_SIZED_TYPE], |
| 414 | vec_safe_length (v_v)); |
| 415 | tree funcs_decl_type = build_array_type_nelts (pointer_sized_int_nodeglobal_trees[TI_POINTER_SIZED_TYPE], |
| 416 | num_funcs); |
| 417 | SET_TYPE_ALIGN (vars_decl_type, TYPE_ALIGN (pointer_sized_int_node))((tree_class_check ((vars_decl_type), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 417, __FUNCTION__))->type_common.align = ffs_hwi ((((tree_class_check ((global_trees[TI_POINTER_SIZED_TYPE]), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 417, __FUNCTION__))->type_common.align) ? ((unsigned)1) << (((tree_class_check ((global_trees[TI_POINTER_SIZED_TYPE]), ( tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 417, __FUNCTION__))->type_common.align) - 1) : 0))); |
| 418 | SET_TYPE_ALIGN (funcs_decl_type, TYPE_ALIGN (pointer_sized_int_node))((tree_class_check ((funcs_decl_type), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 418, __FUNCTION__))->type_common.align = ffs_hwi ((((tree_class_check ((global_trees[TI_POINTER_SIZED_TYPE]), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 418, __FUNCTION__))->type_common.align) ? ((unsigned)1) << (((tree_class_check ((global_trees[TI_POINTER_SIZED_TYPE]), ( tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 418, __FUNCTION__))->type_common.align) - 1) : 0))); |
| 419 | tree ctor_v = build_constructor (vars_decl_type, v_v); |
| 420 | tree ctor_f = build_constructor (funcs_decl_type, v_f); |
| 421 | TREE_CONSTANT (ctor_v)((non_type_check ((ctor_v), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 421, __FUNCTION__))->base.constant_flag) = TREE_CONSTANT (ctor_f)((non_type_check ((ctor_f), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 421, __FUNCTION__))->base.constant_flag) = 1; |
| 422 | TREE_STATIC (ctor_v)((ctor_v)->base.static_flag) = TREE_STATIC (ctor_f)((ctor_f)->base.static_flag) = 1; |
| 423 | tree funcs_decl = build_decl (UNKNOWN_LOCATION((location_t) 0), VAR_DECL, |
| 424 | get_identifier (".offload_func_table")(__builtin_constant_p (".offload_func_table") ? get_identifier_with_length ((".offload_func_table"), strlen (".offload_func_table")) : get_identifier (".offload_func_table")), |
| 425 | funcs_decl_type); |
| 426 | tree vars_decl = build_decl (UNKNOWN_LOCATION((location_t) 0), VAR_DECL, |
| 427 | get_identifier (".offload_var_table")(__builtin_constant_p (".offload_var_table") ? get_identifier_with_length ((".offload_var_table"), strlen (".offload_var_table")) : get_identifier (".offload_var_table")), |
| 428 | vars_decl_type); |
| 429 | TREE_STATIC (funcs_decl)((funcs_decl)->base.static_flag) = TREE_STATIC (vars_decl)((vars_decl)->base.static_flag) = 1; |
| 430 | /* Do not align tables more than TYPE_ALIGN (pointer_sized_int_node), |
| 431 | otherwise a joint table in a binary will contain padding between |
| 432 | tables from multiple object files. */ |
| 433 | DECL_USER_ALIGN (funcs_decl)((contains_struct_check ((funcs_decl), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 433, __FUNCTION__))->base.u.bits.user_align) = DECL_USER_ALIGN (vars_decl)((contains_struct_check ((vars_decl), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 433, __FUNCTION__))->base.u.bits.user_align) = 1; |
| 434 | SET_DECL_ALIGN (funcs_decl, TYPE_ALIGN (funcs_decl_type))(((contains_struct_check ((funcs_decl), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 434, __FUNCTION__))->decl_common.align) = ffs_hwi ((((tree_class_check ((funcs_decl_type), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 434, __FUNCTION__))->type_common.align) ? ((unsigned)1) << (((tree_class_check ((funcs_decl_type), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 434, __FUNCTION__))->type_common.align) - 1) : 0))); |
| 435 | SET_DECL_ALIGN (vars_decl, TYPE_ALIGN (vars_decl_type))(((contains_struct_check ((vars_decl), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 435, __FUNCTION__))->decl_common.align) = ffs_hwi ((((tree_class_check ((vars_decl_type), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 435, __FUNCTION__))->type_common.align) ? ((unsigned)1) << (((tree_class_check ((vars_decl_type), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 435, __FUNCTION__))->type_common.align) - 1) : 0))); |
| 436 | DECL_INITIAL (funcs_decl)((contains_struct_check ((funcs_decl), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 436, __FUNCTION__))->decl_common.initial) = ctor_f; |
| 437 | DECL_INITIAL (vars_decl)((contains_struct_check ((vars_decl), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 437, __FUNCTION__))->decl_common.initial) = ctor_v; |
| 438 | set_decl_section_name (funcs_decl, OFFLOAD_FUNC_TABLE_SECTION_NAME".gnu.offload_funcs"); |
| 439 | set_decl_section_name (vars_decl, OFFLOAD_VAR_TABLE_SECTION_NAME".gnu.offload_vars"); |
| 440 | |
| 441 | varpool_node::finalize_decl (vars_decl); |
| 442 | varpool_node::finalize_decl (funcs_decl); |
| 443 | } |
| 444 | else |
| 445 | { |
| 446 | for (unsigned i = 0; i < num_funcs; i++) |
| 447 | { |
| 448 | tree it = (*offload_funcs)[i]; |
| 449 | /* See also add_decls_addresses_to_decl_constructor |
| 450 | and output_offload_tables in lto-cgraph.cc. */ |
| 451 | if (!in_lto_pglobal_options.x_in_lto_p && !symtab_node::get (it)) |
| 452 | continue; |
| 453 | targetm.record_offload_symbol (it); |
| 454 | } |
| 455 | for (unsigned i = 0; i < num_vars; i++) |
| 456 | { |
| 457 | tree it = (*offload_vars)[i]; |
| 458 | if (!in_lto_pglobal_options.x_in_lto_p && !symtab_node::get (it)) |
| 459 | continue; |
| 460 | #ifdef ACCEL_COMPILER |
| 461 | if (DECL_HAS_VALUE_EXPR_P (it)((tree_check3 ((it), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 461, __FUNCTION__, (VAR_DECL), (PARM_DECL), (RESULT_DECL))) ->decl_common.decl_flag_2) |
| 462 | && lookup_attribute ("omp declare target link", |
| 463 | DECL_ATTRIBUTES (it)((contains_struct_check ((it), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 463, __FUNCTION__))->decl_common.attributes))) |
| 464 | { |
| 465 | tree value_expr = DECL_VALUE_EXPR (it)(decl_value_expr_lookup ((contains_struct_check ((it), (TS_DECL_WRTL ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 465, __FUNCTION__)))); |
| 466 | tree link_ptr_decl = TREE_OPERAND (value_expr, 0)(*((const_cast<tree*> (tree_operand_check ((value_expr) , (0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 466, __FUNCTION__))))); |
| 467 | targetm.record_offload_symbol (link_ptr_decl); |
| 468 | varpool_node::finalize_decl (link_ptr_decl); |
| 469 | } |
| 470 | else |
| 471 | #endif |
| 472 | targetm.record_offload_symbol (it); |
| 473 | } |
| 474 | } |
| 475 | } |
| 476 | |
| 477 | /* Call dim_pos (POS == true) or dim_size (POS == false) builtins for |
| 478 | axis DIM. Return a tmp var holding the result. */ |
| 479 | |
| 480 | static tree |
| 481 | oacc_dim_call (bool pos, int dim, gimple_seq *seq) |
| 482 | { |
| 483 | tree arg = build_int_cst (unsigned_type_nodeinteger_types[itk_unsigned_int], dim); |
| 484 | tree size = create_tmp_var (integer_type_nodeinteger_types[itk_int]); |
| 485 | enum internal_fn fn = pos ? IFN_GOACC_DIM_POS : IFN_GOACC_DIM_SIZE; |
| 486 | gimple *call = gimple_build_call_internal (fn, 1, arg); |
| 487 | |
| 488 | gimple_call_set_lhs (call, size); |
| 489 | gimple_seq_add_stmt (seq, call); |
| 490 | |
| 491 | return size; |
| 492 | } |
| 493 | |
| 494 | /* Find the number of threads (POS = false), or thread number (POS = |
| 495 | true) for an OpenACC region partitioned as MASK. Setup code |
| 496 | required for the calculation is added to SEQ. */ |
| 497 | |
| 498 | static tree |
| 499 | oacc_thread_numbers (bool pos, int mask, gimple_seq *seq) |
| 500 | { |
| 501 | tree res = pos ? NULL_TREE(tree) __null : build_int_cst (unsigned_type_nodeinteger_types[itk_unsigned_int], 1); |
| 502 | unsigned ix; |
| 503 | |
| 504 | /* Start at gang level, and examine relevant dimension indices. */ |
| 505 | for (ix = GOMP_DIM_GANG0; ix != GOMP_DIM_MAX3; ix++) |
| 506 | if (GOMP_DIM_MASK (ix)(1u << (ix)) & mask) |
| 507 | { |
| 508 | if (res) |
| 509 | { |
| 510 | /* We had an outer index, so scale that by the size of |
| 511 | this dimension. */ |
| 512 | tree n = oacc_dim_call (false, ix, seq); |
| 513 | res = fold_build2 (MULT_EXPR, integer_type_node, res, n)fold_build2_loc (((location_t) 0), MULT_EXPR, integer_types[itk_int ], res, n ); |
| 514 | } |
| 515 | if (pos) |
| 516 | { |
| 517 | /* Determine index in this dimension. */ |
| 518 | tree id = oacc_dim_call (true, ix, seq); |
| 519 | if (res) |
| 520 | res = fold_build2 (PLUS_EXPR, integer_type_node, res, id)fold_build2_loc (((location_t) 0), PLUS_EXPR, integer_types[itk_int ], res, id ); |
| 521 | else |
| 522 | res = id; |
| 523 | } |
| 524 | } |
| 525 | |
| 526 | if (res == NULL_TREE(tree) __null) |
| 527 | res = integer_zero_nodeglobal_trees[TI_INTEGER_ZERO]; |
| 528 | |
| 529 | return res; |
| 530 | } |
| 531 | |
| 532 | /* Transform IFN_GOACC_LOOP calls to actual code. See |
| 533 | expand_oacc_for for where these are generated. At the vector |
| 534 | level, we stride loops, such that each member of a warp will |
| 535 | operate on adjacent iterations. At the worker and gang level, |
| 536 | each gang/warp executes a set of contiguous iterations. Chunking |
| 537 | can override this such that each iteration engine executes a |
| 538 | contiguous chunk, and then moves on to stride to the next chunk. */ |
| 539 | |
| 540 | static void |
| 541 | oacc_xform_loop (gcall *call) |
| 542 | { |
| 543 | gimple_stmt_iterator gsi = gsi_for_stmt (call); |
| 544 | enum ifn_goacc_loop_kind code |
| 545 | = (enum ifn_goacc_loop_kind) TREE_INT_CST_LOW (gimple_call_arg (call, 0))((unsigned long) (*tree_int_cst_elt_check ((gimple_call_arg ( call, 0)), (0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 545, __FUNCTION__))); |
| 546 | tree dir = gimple_call_arg (call, 1); |
| 547 | tree range = gimple_call_arg (call, 2); |
| 548 | tree step = gimple_call_arg (call, 3); |
| 549 | tree chunk_size = NULL_TREE(tree) __null; |
| 550 | unsigned mask = (unsigned) TREE_INT_CST_LOW (gimple_call_arg (call, 5))((unsigned long) (*tree_int_cst_elt_check ((gimple_call_arg ( call, 5)), (0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 550, __FUNCTION__))); |
| 551 | tree lhs = gimple_call_lhs (call); |
| 552 | tree type = NULL_TREE(tree) __null; |
| 553 | tree diff_type = TREE_TYPE (range)((contains_struct_check ((range), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 553, __FUNCTION__))->typed.type); |
| 554 | tree r = NULL_TREE(tree) __null; |
| 555 | gimple_seq seq = NULL__null; |
| 556 | bool chunking = false, striding = true; |
| 557 | unsigned outer_mask = mask & (~mask + 1); // Outermost partitioning |
| 558 | unsigned inner_mask = mask & ~outer_mask; // Inner partitioning (if any) |
| 559 | |
| 560 | /* Skip lowering if return value of IFN_GOACC_LOOP call is not used. */ |
| 561 | if (!lhs) |
| 562 | { |
| 563 | gsi_replace_with_seq (&gsi, seq, true); |
| 564 | return; |
| 565 | } |
| 566 | |
| 567 | type = TREE_TYPE (lhs)((contains_struct_check ((lhs), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 567, __FUNCTION__))->typed.type); |
| 568 | |
| 569 | #ifdef ACCEL_COMPILER |
| 570 | chunk_size = gimple_call_arg (call, 4); |
| 571 | if (integer_minus_onep (chunk_size) /* Force static allocation. */ |
| 572 | || integer_zerop (chunk_size)) /* Default (also static). */ |
| 573 | { |
| 574 | /* If we're at the gang level, we want each to execute a |
| 575 | contiguous run of iterations. Otherwise we want each element |
| 576 | to stride. */ |
| 577 | striding = !(outer_mask & GOMP_DIM_MASK (GOMP_DIM_GANG)(1u << (0))); |
| 578 | chunking = false; |
| 579 | } |
| 580 | else |
| 581 | { |
| 582 | /* Chunk of size 1 is striding. */ |
| 583 | striding = integer_onep (chunk_size); |
| 584 | chunking = !striding; |
| 585 | } |
| 586 | #endif |
| 587 | |
| 588 | /* striding=true, chunking=true |
| 589 | -> invalid. |
| 590 | striding=true, chunking=false |
| 591 | -> chunks=1 |
| 592 | striding=false,chunking=true |
| 593 | -> chunks=ceil (range/(chunksize*threads*step)) |
| 594 | striding=false,chunking=false |
| 595 | -> chunk_size=ceil(range/(threads*step)),chunks=1 */ |
| 596 | push_gimplify_context (true); |
| 597 | |
| 598 | switch (code) |
| 599 | { |
| 600 | default: gcc_unreachable ()(fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 600, __FUNCTION__)); |
| 601 | |
| 602 | case IFN_GOACC_LOOP_CHUNKS: |
| 603 | if (!chunking) |
| 604 | r = build_int_cst (type, 1); |
| 605 | else |
| 606 | { |
| 607 | /* chunk_max |
| 608 | = (range - dir) / (chunks * step * num_threads) + dir */ |
| 609 | tree per = oacc_thread_numbers (false, mask, &seq); |
| 610 | per = fold_convert (type, per)fold_convert_loc (((location_t) 0), type, per); |
| 611 | chunk_size = fold_convert (type, chunk_size)fold_convert_loc (((location_t) 0), type, chunk_size); |
| 612 | per = fold_build2 (MULT_EXPR, type, per, chunk_size)fold_build2_loc (((location_t) 0), MULT_EXPR, type, per, chunk_size ); |
| 613 | per = fold_build2 (MULT_EXPR, type, per, step)fold_build2_loc (((location_t) 0), MULT_EXPR, type, per, step ); |
| 614 | r = build2 (MINUS_EXPR, type, range, dir); |
| 615 | r = build2 (PLUS_EXPR, type, r, per); |
| 616 | r = build2 (TRUNC_DIV_EXPR, type, r, per); |
| 617 | } |
| 618 | break; |
| 619 | |
| 620 | case IFN_GOACC_LOOP_STEP: |
| 621 | { |
| 622 | /* If striding, step by the entire compute volume, otherwise |
| 623 | step by the inner volume. */ |
| 624 | unsigned volume = striding ? mask : inner_mask; |
| 625 | |
| 626 | r = oacc_thread_numbers (false, volume, &seq); |
| 627 | r = build2 (MULT_EXPR, type, fold_convert (type, r)fold_convert_loc (((location_t) 0), type, r), step); |
| 628 | } |
| 629 | break; |
| 630 | |
| 631 | case IFN_GOACC_LOOP_OFFSET: |
| 632 | /* Enable vectorization on non-SIMT targets. */ |
| 633 | if (!targetm.simt.vf |
| 634 | && outer_mask == GOMP_DIM_MASK (GOMP_DIM_VECTOR)(1u << (2)) |
| 635 | /* If not -fno-tree-loop-vectorize, hint that we want to vectorize |
| 636 | the loop. */ |
| 637 | && (flag_tree_loop_vectorizeglobal_options.x_flag_tree_loop_vectorize |
| 638 | || !OPTION_SET_P (flag_tree_loop_vectorize)global_options_set.x_flag_tree_loop_vectorize)) |
| 639 | { |
| 640 | basic_block bb = gsi_bb (gsi); |
| 641 | class loop *parent = bb->loop_father; |
| 642 | class loop *body = parent->inner; |
| 643 | |
| 644 | parent->force_vectorize = true; |
| 645 | parent->safelen = INT_MAX2147483647; |
| 646 | |
| 647 | /* "Chunking loops" may have inner loops. */ |
| 648 | if (parent->inner) |
| 649 | { |
| 650 | body->force_vectorize = true; |
| 651 | body->safelen = INT_MAX2147483647; |
| 652 | } |
| 653 | |
| 654 | cfun(cfun + 0)->has_force_vectorize_loops = true; |
| 655 | } |
| 656 | if (striding) |
| 657 | { |
| 658 | r = oacc_thread_numbers (true, mask, &seq); |
| 659 | r = fold_convert (diff_type, r)fold_convert_loc (((location_t) 0), diff_type, r); |
| 660 | } |
| 661 | else |
| 662 | { |
| 663 | tree inner_size = oacc_thread_numbers (false, inner_mask, &seq); |
| 664 | tree outer_size = oacc_thread_numbers (false, outer_mask, &seq); |
| 665 | tree volume = fold_build2 (MULT_EXPR, TREE_TYPE (inner_size),fold_build2_loc (((location_t) 0), MULT_EXPR, ((contains_struct_check ((inner_size), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 665, __FUNCTION__))->typed.type), inner_size, outer_size ) |
| 666 | inner_size, outer_size)fold_build2_loc (((location_t) 0), MULT_EXPR, ((contains_struct_check ((inner_size), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 665, __FUNCTION__))->typed.type), inner_size, outer_size ); |
| 667 | |
| 668 | volume = fold_convert (diff_type, volume)fold_convert_loc (((location_t) 0), diff_type, volume); |
| 669 | if (chunking) |
| 670 | chunk_size = fold_convert (diff_type, chunk_size)fold_convert_loc (((location_t) 0), diff_type, chunk_size); |
| 671 | else |
| 672 | { |
| 673 | tree per = fold_build2 (MULT_EXPR, diff_type, volume, step)fold_build2_loc (((location_t) 0), MULT_EXPR, diff_type, volume , step ); |
| 674 | |
| 675 | chunk_size = build2 (MINUS_EXPR, diff_type, range, dir); |
| 676 | chunk_size = build2 (PLUS_EXPR, diff_type, chunk_size, per); |
| 677 | chunk_size = build2 (TRUNC_DIV_EXPR, diff_type, chunk_size, per); |
| 678 | } |
| 679 | |
| 680 | tree span = build2 (MULT_EXPR, diff_type, chunk_size, |
| 681 | fold_convert (diff_type, inner_size)fold_convert_loc (((location_t) 0), diff_type, inner_size)); |
| 682 | r = oacc_thread_numbers (true, outer_mask, &seq); |
| 683 | r = fold_convert (diff_type, r)fold_convert_loc (((location_t) 0), diff_type, r); |
| 684 | r = build2 (MULT_EXPR, diff_type, r, span); |
| 685 | |
| 686 | tree inner = oacc_thread_numbers (true, inner_mask, &seq); |
| 687 | inner = fold_convert (diff_type, inner)fold_convert_loc (((location_t) 0), diff_type, inner); |
| 688 | r = fold_build2 (PLUS_EXPR, diff_type, r, inner)fold_build2_loc (((location_t) 0), PLUS_EXPR, diff_type, r, inner ); |
| 689 | |
| 690 | if (chunking) |
| 691 | { |
| 692 | tree chunk = fold_convert (diff_type, gimple_call_arg (call, 6))fold_convert_loc (((location_t) 0), diff_type, gimple_call_arg (call, 6)); |
| 693 | tree per |
| 694 | = fold_build2 (MULT_EXPR, diff_type, volume, chunk_size)fold_build2_loc (((location_t) 0), MULT_EXPR, diff_type, volume , chunk_size ); |
| 695 | per = build2 (MULT_EXPR, diff_type, per, chunk); |
| 696 | |
| 697 | r = build2 (PLUS_EXPR, diff_type, r, per); |
| 698 | } |
| 699 | } |
| 700 | r = fold_build2 (MULT_EXPR, diff_type, r, step)fold_build2_loc (((location_t) 0), MULT_EXPR, diff_type, r, step ); |
| 701 | if (type != diff_type) |
| 702 | r = fold_convert (type, r)fold_convert_loc (((location_t) 0), type, r); |
| 703 | break; |
| 704 | |
| 705 | case IFN_GOACC_LOOP_BOUND: |
| 706 | if (striding) |
| 707 | r = range; |
| 708 | else |
| 709 | { |
| 710 | tree inner_size = oacc_thread_numbers (false, inner_mask, &seq); |
| 711 | tree outer_size = oacc_thread_numbers (false, outer_mask, &seq); |
| 712 | tree volume = fold_build2 (MULT_EXPR, TREE_TYPE (inner_size),fold_build2_loc (((location_t) 0), MULT_EXPR, ((contains_struct_check ((inner_size), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 712, __FUNCTION__))->typed.type), inner_size, outer_size ) |
| 713 | inner_size, outer_size)fold_build2_loc (((location_t) 0), MULT_EXPR, ((contains_struct_check ((inner_size), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 712, __FUNCTION__))->typed.type), inner_size, outer_size ); |
| 714 | |
| 715 | volume = fold_convert (diff_type, volume)fold_convert_loc (((location_t) 0), diff_type, volume); |
| 716 | if (chunking) |
| 717 | chunk_size = fold_convert (diff_type, chunk_size)fold_convert_loc (((location_t) 0), diff_type, chunk_size); |
| 718 | else |
| 719 | { |
| 720 | tree per = fold_build2 (MULT_EXPR, diff_type, volume, step)fold_build2_loc (((location_t) 0), MULT_EXPR, diff_type, volume , step ); |
| 721 | |
| 722 | chunk_size = build2 (MINUS_EXPR, diff_type, range, dir); |
| 723 | chunk_size = build2 (PLUS_EXPR, diff_type, chunk_size, per); |
| 724 | chunk_size = build2 (TRUNC_DIV_EXPR, diff_type, chunk_size, per); |
| 725 | } |
| 726 | |
| 727 | tree span = build2 (MULT_EXPR, diff_type, chunk_size, |
| 728 | fold_convert (diff_type, inner_size)fold_convert_loc (((location_t) 0), diff_type, inner_size)); |
| 729 | |
| 730 | r = fold_build2 (MULT_EXPR, diff_type, span, step)fold_build2_loc (((location_t) 0), MULT_EXPR, diff_type, span , step ); |
| 731 | |
| 732 | tree offset = gimple_call_arg (call, 6); |
| 733 | r = build2 (PLUS_EXPR, diff_type, r, |
| 734 | fold_convert (diff_type, offset)fold_convert_loc (((location_t) 0), diff_type, offset)); |
| 735 | r = build2 (integer_onep (dir) ? MIN_EXPR : MAX_EXPR, |
| 736 | diff_type, r, range); |
| 737 | } |
| 738 | if (diff_type != type) |
| 739 | r = fold_convert (type, r)fold_convert_loc (((location_t) 0), type, r); |
| 740 | break; |
| 741 | } |
| 742 | |
| 743 | gimplify_assign (lhs, r, &seq); |
| 744 | |
| 745 | pop_gimplify_context (NULL__null); |
| 746 | |
| 747 | gsi_replace_with_seq (&gsi, seq, true); |
| 748 | } |
| 749 | |
| 750 | /* Transform a GOACC_TILE call. Determines the element loop span for |
| 751 | the specified loop of the nest. This is 1 if we're not tiling. |
| 752 | |
| 753 | GOACC_TILE (collapse_count, loop_no, tile_arg, gwv_tile, gwv_element); */ |
| 754 | |
| 755 | static void |
| 756 | oacc_xform_tile (gcall *call) |
| 757 | { |
| 758 | gimple_stmt_iterator gsi = gsi_for_stmt (call); |
| 759 | unsigned collapse = tree_to_uhwi (gimple_call_arg (call, 0)); |
| 760 | /* Inner loops have higher loop_nos. */ |
| 761 | unsigned loop_no = tree_to_uhwi (gimple_call_arg (call, 1)); |
| 762 | tree tile_size = gimple_call_arg (call, 2); |
| 763 | unsigned e_mask = tree_to_uhwi (gimple_call_arg (call, 4)); |
| 764 | tree lhs = gimple_call_lhs (call); |
| 765 | tree type = TREE_TYPE (lhs)((contains_struct_check ((lhs), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 765, __FUNCTION__))->typed.type); |
| 766 | gimple_seq seq = NULL__null; |
| 767 | tree span = build_int_cst (type, 1); |
Value stored to 'span' during its initialization is never read | |
| 768 | |
| 769 | gcc_assert (!(e_mask((void)(!(!(e_mask & ~((1u << (2)) | (1u << ( 1))))) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 771, __FUNCTION__), 0 : 0)) |
| 770 | & ~(GOMP_DIM_MASK (GOMP_DIM_VECTOR)((void)(!(!(e_mask & ~((1u << (2)) | (1u << ( 1))))) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 771, __FUNCTION__), 0 : 0)) |
| 771 | | GOMP_DIM_MASK (GOMP_DIM_WORKER))))((void)(!(!(e_mask & ~((1u << (2)) | (1u << ( 1))))) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 771, __FUNCTION__), 0 : 0)); |
| 772 | push_gimplify_context (!seen_error ()); |
| 773 | |
| 774 | #ifndef ACCEL_COMPILER |
| 775 | /* Partitioning disabled on host compilers. */ |
| 776 | e_mask = 0; |
| 777 | #endif |
| 778 | if (!e_mask) |
| 779 | /* Not paritioning. */ |
| 780 | span = integer_one_nodeglobal_trees[TI_INTEGER_ONE]; |
| 781 | else if (!integer_zerop (tile_size)) |
| 782 | /* User explicitly specified size. */ |
| 783 | span = tile_size; |
| 784 | else |
| 785 | { |
| 786 | /* Pick a size based on the paritioning of the element loop and |
| 787 | the number of loop nests. */ |
| 788 | tree first_size = NULL_TREE(tree) __null; |
| 789 | tree second_size = NULL_TREE(tree) __null; |
| 790 | |
| 791 | if (e_mask & GOMP_DIM_MASK (GOMP_DIM_VECTOR)(1u << (2))) |
| 792 | first_size = oacc_dim_call (false, GOMP_DIM_VECTOR2, &seq); |
| 793 | if (e_mask & GOMP_DIM_MASK (GOMP_DIM_WORKER)(1u << (1))) |
| 794 | second_size = oacc_dim_call (false, GOMP_DIM_WORKER1, &seq); |
| 795 | |
| 796 | if (!first_size) |
| 797 | { |
| 798 | first_size = second_size; |
| 799 | second_size = NULL_TREE(tree) __null; |
| 800 | } |
| 801 | |
| 802 | if (loop_no + 1 == collapse) |
| 803 | { |
| 804 | span = first_size; |
| 805 | if (!loop_no && second_size) |
| 806 | span = fold_build2 (MULT_EXPR, TREE_TYPE (span),fold_build2_loc (((location_t) 0), MULT_EXPR, ((contains_struct_check ((span), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 806, __FUNCTION__))->typed.type), span, second_size ) |
| 807 | span, second_size)fold_build2_loc (((location_t) 0), MULT_EXPR, ((contains_struct_check ((span), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 806, __FUNCTION__))->typed.type), span, second_size ); |
| 808 | } |
| 809 | else if (loop_no + 2 == collapse) |
| 810 | span = second_size; |
| 811 | else |
| 812 | span = NULL_TREE(tree) __null; |
| 813 | |
| 814 | if (!span) |
| 815 | /* There's no obvious element size for this loop. Options |
| 816 | are 1, first_size or some non-unity constant (32 is my |
| 817 | favourite). We should gather some statistics. */ |
| 818 | span = first_size; |
| 819 | } |
| 820 | |
| 821 | span = fold_convert (type, span)fold_convert_loc (((location_t) 0), type, span); |
| 822 | gimplify_assign (lhs, span, &seq); |
| 823 | |
| 824 | pop_gimplify_context (NULL__null); |
| 825 | |
| 826 | gsi_replace_with_seq (&gsi, seq, true); |
| 827 | } |
| 828 | |
| 829 | /* Default partitioned and minimum partitioned dimensions. */ |
| 830 | |
| 831 | static int oacc_default_dims[GOMP_DIM_MAX3]; |
| 832 | static int oacc_min_dims[GOMP_DIM_MAX3]; |
| 833 | |
| 834 | int |
| 835 | oacc_get_default_dim (int dim) |
| 836 | { |
| 837 | gcc_assert (0 <= dim && dim < GOMP_DIM_MAX)((void)(!(0 <= dim && dim < 3) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 837, __FUNCTION__), 0 : 0)); |
| 838 | return oacc_default_dims[dim]; |
| 839 | } |
| 840 | |
| 841 | int |
| 842 | oacc_get_min_dim (int dim) |
| 843 | { |
| 844 | gcc_assert (0 <= dim && dim < GOMP_DIM_MAX)((void)(!(0 <= dim && dim < 3) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 844, __FUNCTION__), 0 : 0)); |
| 845 | return oacc_min_dims[dim]; |
| 846 | } |
| 847 | |
| 848 | /* Parse the default dimension parameter. This is a set of |
| 849 | :-separated optional compute dimensions. Each specified dimension |
| 850 | is a positive integer. When device type support is added, it is |
| 851 | planned to be a comma separated list of such compute dimensions, |
| 852 | with all but the first prefixed by the colon-terminated device |
| 853 | type. */ |
| 854 | |
| 855 | static void |
| 856 | oacc_parse_default_dims (const char *dims) |
| 857 | { |
| 858 | int ix; |
| 859 | |
| 860 | for (ix = GOMP_DIM_MAX3; ix--;) |
| 861 | { |
| 862 | oacc_default_dims[ix] = -1; |
| 863 | oacc_min_dims[ix] = 1; |
| 864 | } |
| 865 | |
| 866 | #ifndef ACCEL_COMPILER |
| 867 | /* Cannot be overridden on the host. */ |
| 868 | dims = NULL__null; |
| 869 | #endif |
| 870 | if (dims) |
| 871 | { |
| 872 | const char *pos = dims; |
| 873 | |
| 874 | for (ix = 0; *pos && ix != GOMP_DIM_MAX3; ix++) |
| 875 | { |
| 876 | if (ix) |
| 877 | { |
| 878 | if (*pos != ':') |
| 879 | goto malformed; |
| 880 | pos++; |
| 881 | } |
| 882 | |
| 883 | if (*pos != ':') |
| 884 | { |
| 885 | long val; |
| 886 | const char *eptr; |
| 887 | |
| 888 | errno(*__errno_location ()) = 0; |
| 889 | val = strtol (pos, CONST_CAST (char **, &eptr)(const_cast<char **> ((&eptr))), 10); |
| 890 | if (errno(*__errno_location ()) || val <= 0 || (int) val != val) |
| 891 | goto malformed; |
| 892 | pos = eptr; |
| 893 | oacc_default_dims[ix] = (int) val; |
| 894 | } |
| 895 | } |
| 896 | if (*pos) |
| 897 | { |
| 898 | malformed: |
| 899 | error_at (UNKNOWN_LOCATION((location_t) 0), |
| 900 | "%<-fopenacc-dim%> operand is malformed at %qs", pos); |
| 901 | } |
| 902 | } |
| 903 | |
| 904 | /* Allow the backend to validate the dimensions. */ |
| 905 | targetm.goacc.validate_dims (NULL_TREE(tree) __null, oacc_default_dims, -1, 0); |
| 906 | targetm.goacc.validate_dims (NULL_TREE(tree) __null, oacc_min_dims, -2, 0); |
| 907 | } |
| 908 | |
| 909 | /* Validate and update the dimensions for offloaded FN. ATTRS is the |
| 910 | raw attribute. DIMS is an array of dimensions, which is filled in. |
| 911 | LEVEL is the partitioning level of a routine, or -1 for an offload |
| 912 | region itself. USED is the mask of partitioned execution in the |
| 913 | function. */ |
| 914 | |
| 915 | static void |
| 916 | oacc_validate_dims (tree fn, tree attrs, int *dims, int level, unsigned used) |
| 917 | { |
| 918 | tree purpose[GOMP_DIM_MAX3]; |
| 919 | unsigned ix; |
| 920 | tree pos = TREE_VALUE (attrs)((tree_check ((attrs), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 920, __FUNCTION__, (TREE_LIST)))->list.value); |
| 921 | |
| 922 | /* Make sure the attribute creator attached the dimension |
| 923 | information. */ |
| 924 | gcc_assert (pos)((void)(!(pos) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 924, __FUNCTION__), 0 : 0)); |
| 925 | |
| 926 | for (ix = 0; ix != GOMP_DIM_MAX3; ix++) |
| 927 | { |
| 928 | purpose[ix] = TREE_PURPOSE (pos)((tree_check ((pos), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 928, __FUNCTION__, (TREE_LIST)))->list.purpose); |
| 929 | tree val = TREE_VALUE (pos)((tree_check ((pos), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 929, __FUNCTION__, (TREE_LIST)))->list.value); |
| 930 | dims[ix] = val ? TREE_INT_CST_LOW (val)((unsigned long) (*tree_int_cst_elt_check ((val), (0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 930, __FUNCTION__))) : -1; |
| 931 | pos = TREE_CHAIN (pos)((contains_struct_check ((pos), (TS_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 931, __FUNCTION__))->common.chain); |
| 932 | } |
| 933 | |
| 934 | bool check = true; |
| 935 | #ifdef ACCEL_COMPILER |
| 936 | check = false; |
| 937 | #endif |
| 938 | if (check |
| 939 | && warn_openacc_parallelismglobal_options.x_warn_openacc_parallelism |
| 940 | && !lookup_attribute ("oacc kernels", DECL_ATTRIBUTES (fn)((contains_struct_check ((fn), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 940, __FUNCTION__))->decl_common.attributes))) |
| 941 | { |
| 942 | static char const *const axes[] = |
| 943 | /* Must be kept in sync with GOMP_DIM enumeration. */ |
| 944 | { "gang", "worker", "vector" }; |
| 945 | for (ix = level >= 0 ? level : 0; ix != GOMP_DIM_MAX3; ix++) |
| 946 | if (dims[ix] < 0) |
| 947 | ; /* Defaulting axis. */ |
| 948 | else if ((used & GOMP_DIM_MASK (ix)(1u << (ix))) && dims[ix] == 1) |
| 949 | /* There is partitioned execution, but the user requested a |
| 950 | dimension size of 1. They're probably confused. */ |
| 951 | warning_at (DECL_SOURCE_LOCATION (fn)((contains_struct_check ((fn), (TS_DECL_MINIMAL), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 951, __FUNCTION__))->decl_minimal.locus), OPT_Wopenacc_parallelism, |
| 952 | "region contains %s partitioned code but" |
| 953 | " is not %s partitioned", axes[ix], axes[ix]); |
| 954 | else if (!(used & GOMP_DIM_MASK (ix)(1u << (ix))) && dims[ix] != 1) |
| 955 | /* The dimension is explicitly partitioned to non-unity, but |
| 956 | no use is made within the region. */ |
| 957 | warning_at (DECL_SOURCE_LOCATION (fn)((contains_struct_check ((fn), (TS_DECL_MINIMAL), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 957, __FUNCTION__))->decl_minimal.locus), OPT_Wopenacc_parallelism, |
| 958 | "region is %s partitioned but" |
| 959 | " does not contain %s partitioned code", |
| 960 | axes[ix], axes[ix]); |
| 961 | } |
| 962 | |
| 963 | bool changed = targetm.goacc.validate_dims (fn, dims, level, used); |
| 964 | |
| 965 | /* Default anything left to 1 or a partitioned default. */ |
| 966 | for (ix = 0; ix != GOMP_DIM_MAX3; ix++) |
| 967 | if (dims[ix] < 0) |
| 968 | { |
| 969 | /* The OpenACC spec says 'If the [num_gangs] clause is not |
| 970 | specified, an implementation-defined default will be used; |
| 971 | the default may depend on the code within the construct.' |
| 972 | (2.5.6). Thus an implementation is free to choose |
| 973 | non-unity default for a parallel region that doesn't have |
| 974 | any gang-partitioned loops. However, it appears that there |
| 975 | is a sufficient body of user code that expects non-gang |
| 976 | partitioned regions to not execute in gang-redundant mode. |
| 977 | So we (a) don't warn about the non-portability and (b) pick |
| 978 | the minimum permissible dimension size when there is no |
| 979 | partitioned execution. Otherwise we pick the global |
| 980 | default for the dimension, which the user can control. The |
| 981 | same wording and logic applies to num_workers and |
| 982 | vector_length, however the worker- or vector- single |
| 983 | execution doesn't have the same impact as gang-redundant |
| 984 | execution. (If the minimum gang-level partioning is not 1, |
| 985 | the target is probably too confusing.) */ |
| 986 | dims[ix] = (used & GOMP_DIM_MASK (ix)(1u << (ix)) |
| 987 | ? oacc_default_dims[ix] : oacc_min_dims[ix]); |
| 988 | changed = true; |
| 989 | } |
| 990 | |
| 991 | if (changed) |
| 992 | { |
| 993 | /* Replace the attribute with new values. */ |
| 994 | pos = NULL_TREE(tree) __null; |
| 995 | for (ix = GOMP_DIM_MAX3; ix--;) |
| 996 | pos = tree_cons (purpose[ix], |
| 997 | build_int_cst (integer_type_nodeinteger_types[itk_int], dims[ix]), pos); |
| 998 | oacc_replace_fn_attrib (fn, pos); |
| 999 | } |
| 1000 | } |
| 1001 | |
| 1002 | /* Create an empty OpenACC loop structure at LOC. */ |
| 1003 | |
| 1004 | static oacc_loop * |
| 1005 | new_oacc_loop_raw (oacc_loop *parent, location_t loc) |
| 1006 | { |
| 1007 | oacc_loop *loop = XCNEW (oacc_loop)((oacc_loop *) xcalloc (1, sizeof (oacc_loop))); |
| 1008 | |
| 1009 | loop->parent = parent; |
| 1010 | |
| 1011 | if (parent) |
| 1012 | { |
| 1013 | loop->sibling = parent->child; |
| 1014 | parent->child = loop; |
| 1015 | } |
| 1016 | |
| 1017 | loop->loc = loc; |
| 1018 | return loop; |
| 1019 | } |
| 1020 | |
| 1021 | /* Create an outermost, dummy OpenACC loop for offloaded function |
| 1022 | DECL. */ |
| 1023 | |
| 1024 | static oacc_loop * |
| 1025 | new_oacc_loop_outer (tree decl) |
| 1026 | { |
| 1027 | return new_oacc_loop_raw (NULL__null, DECL_SOURCE_LOCATION (decl)((contains_struct_check ((decl), (TS_DECL_MINIMAL), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1027, __FUNCTION__))->decl_minimal.locus)); |
| 1028 | } |
| 1029 | |
| 1030 | /* Start a new OpenACC loop structure beginning at head marker HEAD. |
| 1031 | Link into PARENT loop. Return the new loop. */ |
| 1032 | |
| 1033 | static oacc_loop * |
| 1034 | new_oacc_loop (oacc_loop *parent, gcall *marker) |
| 1035 | { |
| 1036 | oacc_loop *loop = new_oacc_loop_raw (parent, gimple_location (marker)); |
| 1037 | |
| 1038 | loop->marker = marker; |
| 1039 | |
| 1040 | /* TODO: This is where device_type flattening would occur for the loop |
| 1041 | flags. */ |
| 1042 | |
| 1043 | loop->flags = TREE_INT_CST_LOW (gimple_call_arg (marker, 3))((unsigned long) (*tree_int_cst_elt_check ((gimple_call_arg ( marker, 3)), (0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1043, __FUNCTION__))); |
| 1044 | |
| 1045 | tree chunk_size = integer_zero_nodeglobal_trees[TI_INTEGER_ZERO]; |
| 1046 | if (loop->flags & OLF_GANG_STATIC) |
| 1047 | chunk_size = gimple_call_arg (marker, 4); |
| 1048 | loop->chunk_size = chunk_size; |
| 1049 | |
| 1050 | return loop; |
| 1051 | } |
| 1052 | |
| 1053 | /* Create a dummy loop encompassing a call to a openACC routine. |
| 1054 | Extract the routine's partitioning requirements. */ |
| 1055 | |
| 1056 | static void |
| 1057 | new_oacc_loop_routine (oacc_loop *parent, gcall *call, tree decl, tree attrs) |
| 1058 | { |
| 1059 | oacc_loop *loop = new_oacc_loop_raw (parent, gimple_location (call)); |
| 1060 | int level = oacc_fn_attrib_level (attrs); |
| 1061 | |
| 1062 | gcc_assert (level >= 0)((void)(!(level >= 0) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1062, __FUNCTION__), 0 : 0)); |
| 1063 | |
| 1064 | loop->marker = call; |
| 1065 | loop->routine = decl; |
| 1066 | loop->mask = ((GOMP_DIM_MASK (GOMP_DIM_MAX)(1u << (3)) - 1) |
| 1067 | ^ (GOMP_DIM_MASK (level)(1u << (level)) - 1)); |
| 1068 | } |
| 1069 | |
| 1070 | /* Finish off the current OpenACC loop ending at tail marker TAIL. |
| 1071 | Return the parent loop. */ |
| 1072 | |
| 1073 | static oacc_loop * |
| 1074 | finish_oacc_loop (oacc_loop *loop) |
| 1075 | { |
| 1076 | /* If the loop has been collapsed, don't partition it. */ |
| 1077 | if (loop->ifns.is_empty ()) |
| 1078 | loop->mask = loop->flags = 0; |
| 1079 | return loop->parent; |
| 1080 | } |
| 1081 | |
| 1082 | /* Free all OpenACC loop structures within LOOP (inclusive). */ |
| 1083 | |
| 1084 | static void |
| 1085 | free_oacc_loop (oacc_loop *loop) |
| 1086 | { |
| 1087 | if (loop->sibling) |
| 1088 | free_oacc_loop (loop->sibling); |
| 1089 | if (loop->child) |
| 1090 | free_oacc_loop (loop->child); |
| 1091 | |
| 1092 | loop->ifns.release (); |
| 1093 | free (loop); |
| 1094 | } |
| 1095 | |
| 1096 | /* Dump out the OpenACC loop head or tail beginning at FROM. */ |
| 1097 | |
| 1098 | static void |
| 1099 | dump_oacc_loop_part (FILE *file, gcall *from, int depth, |
| 1100 | const char *title, int level) |
| 1101 | { |
| 1102 | enum ifn_unique_kind kind |
| 1103 | = (enum ifn_unique_kind) TREE_INT_CST_LOW (gimple_call_arg (from, 0))((unsigned long) (*tree_int_cst_elt_check ((gimple_call_arg ( from, 0)), (0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1103, __FUNCTION__))); |
| 1104 | |
| 1105 | fprintf (file, "%*s%s-%d:\n", depth * 2, "", title, level); |
| 1106 | for (gimple_stmt_iterator gsi = gsi_for_stmt (from);;) |
| 1107 | { |
| 1108 | gimple *stmt = gsi_stmt (gsi); |
| 1109 | |
| 1110 | if (gimple_call_internal_p (stmt, IFN_UNIQUE)) |
| 1111 | { |
| 1112 | enum ifn_unique_kind k |
| 1113 | = ((enum ifn_unique_kind) TREE_INT_CST_LOW((unsigned long) (*tree_int_cst_elt_check ((gimple_call_arg ( stmt, 0)), (0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1114, __FUNCTION__))) |
| 1114 | (gimple_call_arg (stmt, 0))((unsigned long) (*tree_int_cst_elt_check ((gimple_call_arg ( stmt, 0)), (0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1114, __FUNCTION__)))); |
| 1115 | |
| 1116 | if (k == kind && stmt != from) |
| 1117 | break; |
| 1118 | } |
| 1119 | print_gimple_stmt (file, stmt, depth * 2 + 2); |
| 1120 | |
| 1121 | gsi_next (&gsi); |
| 1122 | while (gsi_end_p (gsi)) |
| 1123 | gsi = gsi_start_bb (single_succ (gsi_bb (gsi))); |
| 1124 | } |
| 1125 | } |
| 1126 | |
| 1127 | /* Dump OpenACC loop LOOP, its children, and its siblings. */ |
| 1128 | |
| 1129 | static void |
| 1130 | dump_oacc_loop (FILE *file, oacc_loop *loop, int depth) |
| 1131 | { |
| 1132 | int ix; |
| 1133 | |
| 1134 | fprintf (file, "%*sLoop %x(%x) %s:%u\n", depth * 2, "", |
| 1135 | loop->flags, loop->mask, |
| 1136 | LOCATION_FILE (loop->loc)((expand_location (loop->loc)).file), LOCATION_LINE (loop->loc)((expand_location (loop->loc)).line)); |
| 1137 | |
| 1138 | if (loop->marker) |
| 1139 | print_gimple_stmt (file, loop->marker, depth * 2); |
| 1140 | |
| 1141 | if (loop->routine) |
| 1142 | fprintf (file, "%*sRoutine %s:%u:%s\n", |
| 1143 | depth * 2, "", DECL_SOURCE_FILE (loop->routine)((expand_location (((contains_struct_check ((loop->routine ), (TS_DECL_MINIMAL), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1143, __FUNCTION__))->decl_minimal.locus))).file), |
| 1144 | DECL_SOURCE_LINE (loop->routine)((expand_location (((contains_struct_check ((loop->routine ), (TS_DECL_MINIMAL), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1144, __FUNCTION__))->decl_minimal.locus))).line), |
| 1145 | IDENTIFIER_POINTER (DECL_NAME (loop->routine))((const char *) (tree_check ((((contains_struct_check ((loop-> routine), (TS_DECL_MINIMAL), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1145, __FUNCTION__))->decl_minimal.name)), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1145, __FUNCTION__, (IDENTIFIER_NODE)))->identifier.id.str )); |
| 1146 | |
| 1147 | for (ix = GOMP_DIM_GANG0; ix != GOMP_DIM_MAX3; ix++) |
| 1148 | if (loop->heads[ix]) |
| 1149 | dump_oacc_loop_part (file, loop->heads[ix], depth, "Head", ix); |
| 1150 | for (ix = GOMP_DIM_MAX3; ix--;) |
| 1151 | if (loop->tails[ix]) |
| 1152 | dump_oacc_loop_part (file, loop->tails[ix], depth, "Tail", ix); |
| 1153 | |
| 1154 | if (loop->child) |
| 1155 | dump_oacc_loop (file, loop->child, depth + 1); |
| 1156 | if (loop->sibling) |
| 1157 | dump_oacc_loop (file, loop->sibling, depth); |
| 1158 | } |
| 1159 | |
| 1160 | void debug_oacc_loop (oacc_loop *); |
| 1161 | |
| 1162 | /* Dump loops to stderr. */ |
| 1163 | |
| 1164 | DEBUG_FUNCTION__attribute__ ((__used__)) void |
| 1165 | debug_oacc_loop (oacc_loop *loop) |
| 1166 | { |
| 1167 | dump_oacc_loop (stderrstderr, loop, 0); |
| 1168 | } |
| 1169 | |
| 1170 | /* Provide diagnostics on OpenACC loop LOOP, its children, and its |
| 1171 | siblings. */ |
| 1172 | |
| 1173 | static void |
| 1174 | inform_oacc_loop (const oacc_loop *loop) |
| 1175 | { |
| 1176 | const char *gang |
| 1177 | = loop->mask & GOMP_DIM_MASK (GOMP_DIM_GANG)(1u << (0)) ? " gang" : ""; |
| 1178 | const char *worker |
| 1179 | = loop->mask & GOMP_DIM_MASK (GOMP_DIM_WORKER)(1u << (1)) ? " worker" : ""; |
| 1180 | const char *vector |
| 1181 | = loop->mask & GOMP_DIM_MASK (GOMP_DIM_VECTOR)(1u << (2)) ? " vector" : ""; |
| 1182 | const char *seq = loop->mask == 0 ? " seq" : ""; |
| 1183 | const dump_user_location_t loc |
| 1184 | = dump_user_location_t::from_location_t (loop->loc); |
| 1185 | dump_printf_loc (MSG_OPTIMIZED_LOCATIONS, loc, |
| 1186 | "assigned OpenACC%s%s%s%s loop parallelism\n", gang, worker, |
| 1187 | vector, seq); |
| 1188 | |
| 1189 | if (loop->child) |
| 1190 | inform_oacc_loop (loop->child); |
| 1191 | if (loop->sibling) |
| 1192 | inform_oacc_loop (loop->sibling); |
| 1193 | } |
| 1194 | |
| 1195 | /* DFS walk of basic blocks BB onwards, creating OpenACC loop |
| 1196 | structures as we go. By construction these loops are properly |
| 1197 | nested. */ |
| 1198 | |
| 1199 | static void |
| 1200 | oacc_loop_discover_walk (oacc_loop *loop, basic_block bb) |
| 1201 | { |
| 1202 | int marker = 0; |
| 1203 | int remaining = 0; |
| 1204 | |
| 1205 | if (bb->flags & BB_VISITED) |
| 1206 | return; |
| 1207 | |
| 1208 | follow: |
| 1209 | bb->flags |= BB_VISITED; |
| 1210 | |
| 1211 | /* Scan for loop markers. */ |
| 1212 | for (gimple_stmt_iterator gsi = gsi_start_bb (bb); !gsi_end_p (gsi); |
| 1213 | gsi_next (&gsi)) |
| 1214 | { |
| 1215 | gimple *stmt = gsi_stmt (gsi); |
| 1216 | |
| 1217 | if (!is_gimple_call (stmt)) |
| 1218 | continue; |
| 1219 | |
| 1220 | gcall *call = as_a <gcall *> (stmt); |
| 1221 | |
| 1222 | /* If this is a routine, make a dummy loop for it. */ |
| 1223 | if (tree decl = gimple_call_fndecl (call)) |
| 1224 | if (tree attrs = oacc_get_fn_attrib (decl)) |
| 1225 | { |
| 1226 | gcc_assert (!marker)((void)(!(!marker) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1226, __FUNCTION__), 0 : 0)); |
| 1227 | new_oacc_loop_routine (loop, call, decl, attrs); |
| 1228 | } |
| 1229 | |
| 1230 | if (!gimple_call_internal_p (call)) |
| 1231 | continue; |
| 1232 | |
| 1233 | switch (gimple_call_internal_fn (call)) |
| 1234 | { |
| 1235 | default: |
| 1236 | break; |
| 1237 | |
| 1238 | case IFN_GOACC_LOOP: |
| 1239 | case IFN_GOACC_TILE: |
| 1240 | /* Record the abstraction function, so we can manipulate it |
| 1241 | later. */ |
| 1242 | loop->ifns.safe_push (call); |
| 1243 | break; |
| 1244 | |
| 1245 | case IFN_UNIQUE: |
| 1246 | enum ifn_unique_kind kind |
| 1247 | = (enum ifn_unique_kind) (TREE_INT_CST_LOW((unsigned long) (*tree_int_cst_elt_check ((gimple_call_arg ( call, 0)), (0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1248, __FUNCTION__))) |
| 1248 | (gimple_call_arg (call, 0))((unsigned long) (*tree_int_cst_elt_check ((gimple_call_arg ( call, 0)), (0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1248, __FUNCTION__)))); |
| 1249 | if (kind == IFN_UNIQUE_OACC_HEAD_MARK |
| 1250 | || kind == IFN_UNIQUE_OACC_TAIL_MARK) |
| 1251 | { |
| 1252 | if (gimple_call_num_args (call) == 2) |
| 1253 | { |
| 1254 | gcc_assert (marker && !remaining)((void)(!(marker && !remaining) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1254, __FUNCTION__), 0 : 0)); |
| 1255 | marker = 0; |
| 1256 | if (kind == IFN_UNIQUE_OACC_TAIL_MARK) |
| 1257 | loop = finish_oacc_loop (loop); |
| 1258 | else |
| 1259 | loop->head_end = call; |
| 1260 | } |
| 1261 | else |
| 1262 | { |
| 1263 | int count = TREE_INT_CST_LOW (gimple_call_arg (call, 2))((unsigned long) (*tree_int_cst_elt_check ((gimple_call_arg ( call, 2)), (0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1263, __FUNCTION__))); |
| 1264 | |
| 1265 | if (!marker) |
| 1266 | { |
| 1267 | if (kind == IFN_UNIQUE_OACC_HEAD_MARK) |
| 1268 | loop = new_oacc_loop (loop, call); |
| 1269 | remaining = count; |
| 1270 | } |
| 1271 | gcc_assert (count == remaining)((void)(!(count == remaining) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1271, __FUNCTION__), 0 : 0)); |
| 1272 | if (remaining) |
| 1273 | { |
| 1274 | remaining--; |
| 1275 | if (kind == IFN_UNIQUE_OACC_HEAD_MARK) |
| 1276 | loop->heads[marker] = call; |
| 1277 | else |
| 1278 | loop->tails[remaining] = call; |
| 1279 | } |
| 1280 | marker++; |
| 1281 | } |
| 1282 | } |
| 1283 | } |
| 1284 | } |
| 1285 | if (remaining || marker) |
| 1286 | { |
| 1287 | bb = single_succ (bb); |
| 1288 | gcc_assert (single_pred_p (bb) && !(bb->flags & BB_VISITED))((void)(!(single_pred_p (bb) && !(bb->flags & BB_VISITED )) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1288, __FUNCTION__), 0 : 0)); |
| 1289 | goto follow; |
| 1290 | } |
| 1291 | |
| 1292 | /* Walk successor blocks. */ |
| 1293 | edge e; |
| 1294 | edge_iterator ei; |
| 1295 | |
| 1296 | FOR_EACH_EDGE (e, ei, bb->succs)for ((ei) = ei_start_1 (&((bb->succs))); ei_cond ((ei) , &(e)); ei_next (&(ei))) |
| 1297 | oacc_loop_discover_walk (loop, e->dest); |
| 1298 | } |
| 1299 | |
| 1300 | /* LOOP is the first sibling. Reverse the order in place and return |
| 1301 | the new first sibling. Recurse to child loops. */ |
| 1302 | |
| 1303 | static oacc_loop * |
| 1304 | oacc_loop_sibling_nreverse (oacc_loop *loop) |
| 1305 | { |
| 1306 | oacc_loop *last = NULL__null; |
| 1307 | do |
| 1308 | { |
| 1309 | if (loop->child) |
| 1310 | loop->child = oacc_loop_sibling_nreverse (loop->child); |
| 1311 | |
| 1312 | oacc_loop *next = loop->sibling; |
| 1313 | loop->sibling = last; |
| 1314 | last = loop; |
| 1315 | loop = next; |
| 1316 | } |
| 1317 | while (loop); |
| 1318 | |
| 1319 | return last; |
| 1320 | } |
| 1321 | |
| 1322 | /* Discover the OpenACC loops marked up by HEAD and TAIL markers for |
| 1323 | the current function. */ |
| 1324 | |
| 1325 | static oacc_loop * |
| 1326 | oacc_loop_discovery () |
| 1327 | { |
| 1328 | /* Clear basic block flags, in particular BB_VISITED which we're going to use |
| 1329 | in the following. */ |
| 1330 | clear_bb_flags (); |
| 1331 | |
| 1332 | oacc_loop *top = new_oacc_loop_outer (current_function_decl); |
| 1333 | oacc_loop_discover_walk (top, ENTRY_BLOCK_PTR_FOR_FN (cfun)(((cfun + 0))->cfg->x_entry_block_ptr)); |
| 1334 | |
| 1335 | /* The siblings were constructed in reverse order, reverse them so |
| 1336 | that diagnostics come out in an unsurprising order. */ |
| 1337 | top = oacc_loop_sibling_nreverse (top); |
| 1338 | |
| 1339 | return top; |
| 1340 | } |
| 1341 | |
| 1342 | /* Transform the abstract internal function markers starting at FROM |
| 1343 | to be for partitioning level LEVEL. Stop when we meet another HEAD |
| 1344 | or TAIL marker. */ |
| 1345 | |
| 1346 | static void |
| 1347 | oacc_loop_xform_head_tail (gcall *from, int level) |
| 1348 | { |
| 1349 | enum ifn_unique_kind kind |
| 1350 | = (enum ifn_unique_kind) TREE_INT_CST_LOW (gimple_call_arg (from, 0))((unsigned long) (*tree_int_cst_elt_check ((gimple_call_arg ( from, 0)), (0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1350, __FUNCTION__))); |
| 1351 | tree replacement = build_int_cst (unsigned_type_nodeinteger_types[itk_unsigned_int], level); |
| 1352 | |
| 1353 | for (gimple_stmt_iterator gsi = gsi_for_stmt (from);;) |
| 1354 | { |
| 1355 | gimple *stmt = gsi_stmt (gsi); |
| 1356 | |
| 1357 | if (gimple_call_internal_p (stmt, IFN_UNIQUE)) |
| 1358 | { |
| 1359 | enum ifn_unique_kind k |
| 1360 | = ((enum ifn_unique_kind) |
| 1361 | TREE_INT_CST_LOW (gimple_call_arg (stmt, 0))((unsigned long) (*tree_int_cst_elt_check ((gimple_call_arg ( stmt, 0)), (0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1361, __FUNCTION__)))); |
| 1362 | |
| 1363 | if (k == IFN_UNIQUE_OACC_FORK |
| 1364 | || k == IFN_UNIQUE_OACC_JOIN |
| 1365 | || k == IFN_UNIQUE_OACC_PRIVATE) |
| 1366 | *gimple_call_arg_ptr (stmt, 2) = replacement; |
| 1367 | else if (k == kind && stmt != from) |
| 1368 | break; |
| 1369 | } |
| 1370 | else if (gimple_call_internal_p (stmt, IFN_GOACC_REDUCTION)) |
| 1371 | *gimple_call_arg_ptr (stmt, 3) = replacement; |
| 1372 | update_stmt (stmt); |
| 1373 | |
| 1374 | gsi_next (&gsi); |
| 1375 | while (gsi_end_p (gsi)) |
| 1376 | gsi = gsi_start_bb (single_succ (gsi_bb (gsi))); |
| 1377 | } |
| 1378 | } |
| 1379 | |
| 1380 | /* Process the discovered OpenACC loops, setting the correct |
| 1381 | partitioning level etc. */ |
| 1382 | |
| 1383 | static void |
| 1384 | oacc_loop_process (oacc_loop *loop, int fn_level) |
| 1385 | { |
| 1386 | if (loop->child) |
| 1387 | oacc_loop_process (loop->child, fn_level); |
| 1388 | |
| 1389 | if (loop->mask && !loop->routine) |
| 1390 | { |
| 1391 | int ix; |
| 1392 | tree mask_arg = build_int_cst (unsigned_type_nodeinteger_types[itk_unsigned_int], loop->mask); |
| 1393 | tree e_mask_arg = build_int_cst (unsigned_type_nodeinteger_types[itk_unsigned_int], loop->e_mask); |
| 1394 | tree chunk_arg = loop->chunk_size; |
| 1395 | gcall *call; |
| 1396 | |
| 1397 | for (ix = 0; loop->ifns.iterate (ix, &call); ix++) |
| 1398 | { |
| 1399 | switch (gimple_call_internal_fn (call)) |
| 1400 | { |
| 1401 | case IFN_GOACC_LOOP: |
| 1402 | { |
| 1403 | bool is_e = gimple_call_arg (call, 5) == integer_minus_one_nodeglobal_trees[TI_INTEGER_MINUS_ONE]; |
| 1404 | gimple_call_set_arg (call, 5, is_e ? e_mask_arg : mask_arg); |
| 1405 | if (!is_e) |
| 1406 | gimple_call_set_arg (call, 4, chunk_arg); |
| 1407 | } |
| 1408 | break; |
| 1409 | |
| 1410 | case IFN_GOACC_TILE: |
| 1411 | gimple_call_set_arg (call, 3, mask_arg); |
| 1412 | gimple_call_set_arg (call, 4, e_mask_arg); |
| 1413 | break; |
| 1414 | |
| 1415 | default: |
| 1416 | gcc_unreachable ()(fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1416, __FUNCTION__)); |
| 1417 | } |
| 1418 | update_stmt (call); |
| 1419 | } |
| 1420 | |
| 1421 | unsigned dim = GOMP_DIM_GANG0; |
| 1422 | unsigned mask = loop->mask | loop->e_mask; |
| 1423 | for (ix = 0; ix != GOMP_DIM_MAX3 && mask; ix++) |
| 1424 | { |
| 1425 | while (!(GOMP_DIM_MASK (dim)(1u << (dim)) & mask)) |
| 1426 | dim++; |
| 1427 | |
| 1428 | oacc_loop_xform_head_tail (loop->heads[ix], dim); |
| 1429 | oacc_loop_xform_head_tail (loop->tails[ix], dim); |
| 1430 | |
| 1431 | mask ^= GOMP_DIM_MASK (dim)(1u << (dim)); |
| 1432 | } |
| 1433 | } |
| 1434 | |
| 1435 | if (loop->sibling) |
| 1436 | oacc_loop_process (loop->sibling, fn_level); |
| 1437 | |
| 1438 | |
| 1439 | /* OpenACC 2.6, 2.9.11. "reduction clause" places a restriction such that |
| 1440 | "The 'reduction' clause may not be specified on an orphaned 'loop' |
| 1441 | construct with the 'gang' clause, or on an orphaned 'loop' construct that |
| 1442 | will generate gang parallelism in a procedure that is compiled with the |
| 1443 | 'routine gang' clause." */ |
| 1444 | if (fn_level == GOMP_DIM_GANG0 |
| 1445 | && (loop->mask & GOMP_DIM_MASK (GOMP_DIM_GANG)(1u << (0))) |
| 1446 | && (loop->flags & OLF_REDUCTION)) |
| 1447 | error_at (loop->loc, |
| 1448 | "gang reduction on an orphan loop"); |
| 1449 | } |
| 1450 | |
| 1451 | /* Walk the OpenACC loop heirarchy checking and assigning the |
| 1452 | programmer-specified partitionings. OUTER_MASK is the partitioning |
| 1453 | this loop is contained within. Return mask of partitioning |
| 1454 | encountered. If any auto loops are discovered, set GOMP_DIM_MAX |
| 1455 | bit. */ |
| 1456 | |
| 1457 | static unsigned |
| 1458 | oacc_loop_fixed_partitions (oacc_loop *loop, unsigned outer_mask) |
| 1459 | { |
| 1460 | unsigned this_mask = loop->mask; |
| 1461 | unsigned mask_all = 0; |
| 1462 | bool noisy = true; |
| 1463 | |
| 1464 | #ifdef ACCEL_COMPILER |
| 1465 | /* When device_type is supported, we want the device compiler to be |
| 1466 | noisy, if the loop parameters are device_type-specific. */ |
| 1467 | noisy = false; |
| 1468 | #endif |
| 1469 | |
| 1470 | if (!loop->routine) |
| 1471 | { |
| 1472 | bool auto_par = (loop->flags & OLF_AUTO) != 0; |
| 1473 | bool seq_par = (loop->flags & OLF_SEQ) != 0; |
| 1474 | bool tiling = (loop->flags & OLF_TILE) != 0; |
| 1475 | |
| 1476 | this_mask = ((loop->flags >> OLF_DIM_BASE) |
| 1477 | & (GOMP_DIM_MASK (GOMP_DIM_MAX)(1u << (3)) - 1)); |
| 1478 | |
| 1479 | /* Apply auto partitioning if this is a non-partitioned regular |
| 1480 | loop, or (no more than) single axis tiled loop. */ |
| 1481 | bool maybe_auto |
| 1482 | = !seq_par && this_mask == (tiling ? this_mask & -this_mask : 0); |
| 1483 | |
| 1484 | if ((this_mask != 0) + auto_par + seq_par > 1) |
| 1485 | { |
| 1486 | if (noisy) |
| 1487 | error_at (loop->loc, |
| 1488 | seq_par |
| 1489 | ? G_("%<seq%> overrides other OpenACC loop specifiers")"%<seq%> overrides other OpenACC loop specifiers" |
| 1490 | : G_("%<auto%> conflicts with other OpenACC loop ""%<auto%> conflicts with other OpenACC loop " "specifiers" |
| 1491 | "specifiers")"%<auto%> conflicts with other OpenACC loop " "specifiers"); |
| 1492 | maybe_auto = false; |
| 1493 | loop->flags &= ~OLF_AUTO; |
| 1494 | if (seq_par) |
| 1495 | { |
| 1496 | loop->flags |
| 1497 | &= ~((GOMP_DIM_MASK (GOMP_DIM_MAX)(1u << (3)) - 1) << OLF_DIM_BASE); |
| 1498 | this_mask = 0; |
| 1499 | } |
| 1500 | } |
| 1501 | |
| 1502 | if (maybe_auto && (loop->flags & OLF_INDEPENDENT)) |
| 1503 | { |
| 1504 | loop->flags |= OLF_AUTO; |
| 1505 | mask_all |= GOMP_DIM_MASK (GOMP_DIM_MAX)(1u << (3)); |
| 1506 | } |
| 1507 | } |
| 1508 | |
| 1509 | if (this_mask & outer_mask) |
| 1510 | { |
| 1511 | const oacc_loop *outer; |
| 1512 | for (outer = loop->parent; outer; outer = outer->parent) |
| 1513 | if ((outer->mask | outer->e_mask) & this_mask) |
| 1514 | break; |
| 1515 | |
| 1516 | if (noisy) |
| 1517 | { |
| 1518 | if (outer) |
| 1519 | { |
| 1520 | error_at (loop->loc, |
| 1521 | loop->routine |
| 1522 | ? G_("routine call uses same OpenACC parallelism""routine call uses same OpenACC parallelism" " as containing loop" |
| 1523 | " as containing loop")"routine call uses same OpenACC parallelism" " as containing loop" |
| 1524 | : G_("inner loop uses same OpenACC parallelism""inner loop uses same OpenACC parallelism" " as containing loop" |
| 1525 | " as containing loop")"inner loop uses same OpenACC parallelism" " as containing loop"); |
| 1526 | inform (outer->loc, "containing loop here"); |
| 1527 | } |
| 1528 | else |
| 1529 | error_at (loop->loc, |
| 1530 | loop->routine |
| 1531 | ? G_("routine call uses OpenACC parallelism disallowed""routine call uses OpenACC parallelism disallowed" " by containing routine" |
| 1532 | " by containing routine")"routine call uses OpenACC parallelism disallowed" " by containing routine" |
| 1533 | : G_("loop uses OpenACC parallelism disallowed""loop uses OpenACC parallelism disallowed" " by containing routine" |
| 1534 | " by containing routine")"loop uses OpenACC parallelism disallowed" " by containing routine"); |
| 1535 | |
| 1536 | if (loop->routine) |
| 1537 | inform (DECL_SOURCE_LOCATION (loop->routine)((contains_struct_check ((loop->routine), (TS_DECL_MINIMAL ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1537, __FUNCTION__))->decl_minimal.locus), |
| 1538 | "routine %qD declared here", loop->routine); |
| 1539 | } |
| 1540 | this_mask &= ~outer_mask; |
| 1541 | } |
| 1542 | else |
| 1543 | { |
| 1544 | unsigned outermost = least_bit_hwi (this_mask); |
| 1545 | |
| 1546 | if (outermost && outermost <= outer_mask) |
| 1547 | { |
| 1548 | if (noisy) |
| 1549 | { |
| 1550 | error_at (loop->loc, |
| 1551 | "incorrectly nested OpenACC loop parallelism"); |
| 1552 | |
| 1553 | const oacc_loop *outer; |
| 1554 | for (outer = loop->parent; |
| 1555 | outer->flags && outer->flags < outermost; |
| 1556 | outer = outer->parent) |
| 1557 | continue; |
| 1558 | inform (outer->loc, "containing loop here"); |
| 1559 | } |
| 1560 | |
| 1561 | this_mask &= ~outermost; |
| 1562 | } |
| 1563 | } |
| 1564 | |
| 1565 | mask_all |= this_mask; |
| 1566 | |
| 1567 | if (loop->flags & OLF_TILE) |
| 1568 | { |
| 1569 | /* When tiling, vector goes to the element loop, and failing |
| 1570 | that we put worker there. The std doesn't contemplate |
| 1571 | specifying all three. We choose to put worker and vector on |
| 1572 | the element loops in that case. */ |
| 1573 | unsigned this_e_mask = this_mask & GOMP_DIM_MASK (GOMP_DIM_VECTOR)(1u << (2)); |
| 1574 | if (!this_e_mask || this_mask & GOMP_DIM_MASK (GOMP_DIM_GANG)(1u << (0))) |
| 1575 | this_e_mask |= this_mask & GOMP_DIM_MASK (GOMP_DIM_WORKER)(1u << (1)); |
| 1576 | |
| 1577 | loop->e_mask = this_e_mask; |
| 1578 | this_mask ^= this_e_mask; |
| 1579 | } |
| 1580 | |
| 1581 | loop->mask = this_mask; |
| 1582 | |
| 1583 | if (dump_file) |
| 1584 | fprintf (dump_file, "Loop %s:%d user specified %d & %d\n", |
| 1585 | LOCATION_FILE (loop->loc)((expand_location (loop->loc)).file), LOCATION_LINE (loop->loc)((expand_location (loop->loc)).line), |
| 1586 | loop->mask, loop->e_mask); |
| 1587 | |
| 1588 | if (loop->child) |
| 1589 | { |
| 1590 | unsigned tmp_mask = outer_mask | this_mask | loop->e_mask; |
| 1591 | loop->inner = oacc_loop_fixed_partitions (loop->child, tmp_mask); |
| 1592 | mask_all |= loop->inner; |
| 1593 | } |
| 1594 | |
| 1595 | if (loop->sibling) |
| 1596 | mask_all |= oacc_loop_fixed_partitions (loop->sibling, outer_mask); |
| 1597 | |
| 1598 | return mask_all; |
| 1599 | } |
| 1600 | |
| 1601 | /* Walk the OpenACC loop heirarchy to assign auto-partitioned loops. |
| 1602 | OUTER_MASK is the partitioning this loop is contained within. |
| 1603 | OUTER_ASSIGN is true if an outer loop is being auto-partitioned. |
| 1604 | Return the cumulative partitioning used by this loop, siblings and |
| 1605 | children. */ |
| 1606 | |
| 1607 | static unsigned |
| 1608 | oacc_loop_auto_partitions (oacc_loop *loop, unsigned outer_mask, |
| 1609 | bool outer_assign) |
| 1610 | { |
| 1611 | bool assign = (loop->flags & OLF_AUTO) && (loop->flags & OLF_INDEPENDENT); |
| 1612 | bool noisy = true; |
| 1613 | bool tiling = loop->flags & OLF_TILE; |
| 1614 | |
| 1615 | #ifdef ACCEL_COMPILER |
| 1616 | /* When device_type is supported, we want the device compiler to be |
| 1617 | noisy, if the loop parameters are device_type-specific. */ |
| 1618 | noisy = false; |
| 1619 | #endif |
| 1620 | |
| 1621 | if (assign && (!outer_assign || loop->inner)) |
| 1622 | { |
| 1623 | /* Allocate outermost and non-innermost loops at the outermost |
| 1624 | non-innermost available level. */ |
| 1625 | unsigned this_mask = GOMP_DIM_MASK (GOMP_DIM_GANG)(1u << (0)); |
| 1626 | |
| 1627 | /* Find the first outermost available partition. */ |
| 1628 | while (this_mask <= outer_mask) |
| 1629 | this_mask <<= 1; |
| 1630 | |
| 1631 | /* Grab two axes if tiling, and we've not assigned anything */ |
| 1632 | if (tiling && !(loop->mask | loop->e_mask)) |
| 1633 | this_mask |= this_mask << 1; |
| 1634 | |
| 1635 | /* Prohibit the innermost partitioning at the moment. */ |
| 1636 | this_mask &= GOMP_DIM_MASK (GOMP_DIM_MAX - 1)(1u << (3 - 1)) - 1; |
| 1637 | |
| 1638 | /* Don't use any dimension explicitly claimed by an inner loop. */ |
| 1639 | this_mask &= ~loop->inner; |
| 1640 | |
| 1641 | if (tiling && !loop->e_mask) |
| 1642 | { |
| 1643 | /* If we got two axes, allocate the inner one to the element |
| 1644 | loop. */ |
| 1645 | loop->e_mask = this_mask & (this_mask << 1); |
| 1646 | this_mask ^= loop->e_mask; |
| 1647 | } |
| 1648 | |
| 1649 | loop->mask |= this_mask; |
| 1650 | } |
| 1651 | |
| 1652 | if (loop->child) |
| 1653 | { |
| 1654 | unsigned tmp_mask = outer_mask | loop->mask | loop->e_mask; |
| 1655 | loop->inner = oacc_loop_auto_partitions (loop->child, tmp_mask, |
| 1656 | outer_assign | assign); |
| 1657 | } |
| 1658 | |
| 1659 | if (assign && (!loop->mask || (tiling && !loop->e_mask) || !outer_assign)) |
| 1660 | { |
| 1661 | /* Allocate the loop at the innermost available level. Note |
| 1662 | that we do this even if we already assigned this loop the |
| 1663 | outermost available level above. That way we'll partition |
| 1664 | this along 2 axes, if they are available. */ |
| 1665 | unsigned this_mask = 0; |
| 1666 | |
| 1667 | /* Determine the outermost partitioning used within this loop. */ |
| 1668 | this_mask = loop->inner | GOMP_DIM_MASK (GOMP_DIM_MAX)(1u << (3)); |
| 1669 | this_mask = least_bit_hwi (this_mask); |
| 1670 | |
| 1671 | /* Pick the partitioning just inside that one. */ |
| 1672 | this_mask >>= 1; |
| 1673 | |
| 1674 | /* And avoid picking one use by an outer loop. */ |
| 1675 | this_mask &= ~outer_mask; |
| 1676 | |
| 1677 | /* If tiling and we failed completely above, grab the next one |
| 1678 | too. Making sure it doesn't hit an outer loop. */ |
| 1679 | if (tiling) |
| 1680 | { |
| 1681 | this_mask &= ~(loop->e_mask | loop->mask); |
| 1682 | unsigned tile_mask = ((this_mask >> 1) |
| 1683 | & ~(outer_mask | loop->e_mask | loop->mask)); |
| 1684 | |
| 1685 | if (tile_mask || loop->mask) |
| 1686 | { |
| 1687 | loop->e_mask |= this_mask; |
| 1688 | this_mask = tile_mask; |
| 1689 | } |
| 1690 | if (!loop->e_mask && noisy) |
| 1691 | warning_at (loop->loc, 0, |
| 1692 | "insufficient partitioning available" |
| 1693 | " to parallelize element loop"); |
| 1694 | } |
| 1695 | |
| 1696 | loop->mask |= this_mask; |
| 1697 | if (!loop->mask && noisy) |
| 1698 | warning_at (loop->loc, 0, |
| 1699 | tiling |
| 1700 | ? G_("insufficient partitioning available""insufficient partitioning available" " to parallelize tile loop" |
| 1701 | " to parallelize tile loop")"insufficient partitioning available" " to parallelize tile loop" |
| 1702 | : G_("insufficient partitioning available""insufficient partitioning available" " to parallelize loop" |
| 1703 | " to parallelize loop")"insufficient partitioning available" " to parallelize loop"); |
| 1704 | } |
| 1705 | |
| 1706 | if (assign && dump_file) |
| 1707 | fprintf (dump_file, "Auto loop %s:%d assigned %d & %d\n", |
| 1708 | LOCATION_FILE (loop->loc)((expand_location (loop->loc)).file), LOCATION_LINE (loop->loc)((expand_location (loop->loc)).line), |
| 1709 | loop->mask, loop->e_mask); |
| 1710 | |
| 1711 | unsigned inner_mask = 0; |
| 1712 | |
| 1713 | if (loop->sibling) |
| 1714 | inner_mask |= oacc_loop_auto_partitions (loop->sibling, |
| 1715 | outer_mask, outer_assign); |
| 1716 | |
| 1717 | inner_mask |= loop->inner | loop->mask | loop->e_mask; |
| 1718 | |
| 1719 | return inner_mask; |
| 1720 | } |
| 1721 | |
| 1722 | /* Walk the OpenACC loop heirarchy to check and assign partitioning |
| 1723 | axes. Return mask of partitioning. */ |
| 1724 | |
| 1725 | static unsigned |
| 1726 | oacc_loop_partition (oacc_loop *loop, unsigned outer_mask) |
| 1727 | { |
| 1728 | unsigned mask_all = oacc_loop_fixed_partitions (loop, outer_mask); |
| 1729 | |
| 1730 | if (mask_all & GOMP_DIM_MASK (GOMP_DIM_MAX)(1u << (3))) |
| 1731 | { |
| 1732 | mask_all ^= GOMP_DIM_MASK (GOMP_DIM_MAX)(1u << (3)); |
| 1733 | mask_all |= oacc_loop_auto_partitions (loop, outer_mask, false); |
| 1734 | } |
| 1735 | return mask_all; |
| 1736 | } |
| 1737 | |
| 1738 | /* Default fork/join early expander. Delete the function calls if |
| 1739 | there is no RTL expander. */ |
| 1740 | |
| 1741 | bool |
| 1742 | default_goacc_fork_join (gcall *ARG_UNUSED (call)call __attribute__ ((__unused__)), |
| 1743 | const int *ARG_UNUSED (dims)dims __attribute__ ((__unused__)), bool is_fork) |
| 1744 | { |
| 1745 | if (is_fork) |
| 1746 | return targetm.have_oacc_fork (); |
| 1747 | else |
| 1748 | return targetm.have_oacc_join (); |
| 1749 | } |
| 1750 | |
| 1751 | /* Default goacc.reduction early expander. |
| 1752 | |
| 1753 | LHS-opt = IFN_REDUCTION (KIND, RES_PTR, VAR, LEVEL, OP, OFFSET) |
| 1754 | If RES_PTR is not integer-zerop: |
| 1755 | SETUP - emit 'LHS = *RES_PTR', LHS = NULL |
| 1756 | TEARDOWN - emit '*RES_PTR = VAR' |
| 1757 | If LHS is not NULL |
| 1758 | emit 'LHS = VAR' */ |
| 1759 | |
| 1760 | void |
| 1761 | default_goacc_reduction (gcall *call) |
| 1762 | { |
| 1763 | unsigned code = (unsigned)TREE_INT_CST_LOW (gimple_call_arg (call, 0))((unsigned long) (*tree_int_cst_elt_check ((gimple_call_arg ( call, 0)), (0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1763, __FUNCTION__))); |
| 1764 | gimple_stmt_iterator gsi = gsi_for_stmt (call); |
| 1765 | tree lhs = gimple_call_lhs (call); |
| 1766 | tree var = gimple_call_arg (call, 2); |
| 1767 | gimple_seq seq = NULL__null; |
| 1768 | |
| 1769 | if (code == IFN_GOACC_REDUCTION_SETUP |
| 1770 | || code == IFN_GOACC_REDUCTION_TEARDOWN) |
| 1771 | { |
| 1772 | /* Setup and Teardown need to copy from/to the receiver object, |
| 1773 | if there is one. */ |
| 1774 | tree ref_to_res = gimple_call_arg (call, 1); |
| 1775 | |
| 1776 | if (!integer_zerop (ref_to_res)) |
| 1777 | { |
| 1778 | tree dst = build_simple_mem_ref (ref_to_res)build_simple_mem_ref_loc (((location_t) 0), ref_to_res); |
| 1779 | tree src = var; |
| 1780 | |
| 1781 | if (code == IFN_GOACC_REDUCTION_SETUP) |
| 1782 | { |
| 1783 | src = dst; |
| 1784 | dst = lhs; |
| 1785 | lhs = NULL__null; |
| 1786 | } |
| 1787 | gimple_seq_add_stmt (&seq, gimple_build_assign (dst, src)); |
| 1788 | } |
| 1789 | } |
| 1790 | |
| 1791 | /* Copy VAR to LHS, if there is an LHS. */ |
| 1792 | if (lhs) |
| 1793 | gimple_seq_add_stmt (&seq, gimple_build_assign (lhs, var)); |
| 1794 | |
| 1795 | gsi_replace_with_seq (&gsi, seq, true); |
| 1796 | } |
| 1797 | |
| 1798 | struct var_decl_rewrite_info |
| 1799 | { |
| 1800 | gimple *stmt; |
| 1801 | hash_map<tree, tree> *adjusted_vars; |
| 1802 | bool avoid_pointer_conversion; |
| 1803 | bool modified; |
| 1804 | }; |
| 1805 | |
| 1806 | /* Helper function for execute_oacc_device_lower. Rewrite VAR_DECLs (by |
| 1807 | themselves or wrapped in various other nodes) according to ADJUSTED_VARS in |
| 1808 | the var_decl_rewrite_info pointed to via DATA. Used as part of coercing |
| 1809 | gang-private variables in OpenACC offload regions to reside in GPU shared |
| 1810 | memory. */ |
| 1811 | |
| 1812 | static tree |
| 1813 | oacc_rewrite_var_decl (tree *tp, int *walk_subtrees, void *data) |
| 1814 | { |
| 1815 | walk_stmt_info *wi = (walk_stmt_info *) data; |
| 1816 | var_decl_rewrite_info *info = (var_decl_rewrite_info *) wi->info; |
| 1817 | |
| 1818 | if (TREE_CODE (*tp)((enum tree_code) (*tp)->base.code) == ADDR_EXPR) |
| 1819 | { |
| 1820 | tree arg = TREE_OPERAND (*tp, 0)(*((const_cast<tree*> (tree_operand_check ((*tp), (0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1820, __FUNCTION__))))); |
| 1821 | tree *new_arg = info->adjusted_vars->get (arg); |
| 1822 | |
| 1823 | if (new_arg) |
| 1824 | { |
| 1825 | if (info->avoid_pointer_conversion) |
| 1826 | { |
| 1827 | *tp = build_fold_addr_expr (*new_arg)build_fold_addr_expr_loc (((location_t) 0), (*new_arg)); |
| 1828 | info->modified = true; |
| 1829 | *walk_subtrees = 0; |
| 1830 | } |
| 1831 | else |
| 1832 | { |
| 1833 | gimple_stmt_iterator gsi = gsi_for_stmt (info->stmt); |
| 1834 | tree repl = build_fold_addr_expr (*new_arg)build_fold_addr_expr_loc (((location_t) 0), (*new_arg)); |
| 1835 | gimple *stmt1 |
| 1836 | = gimple_build_assign (make_ssa_name (TREE_TYPE (repl)((contains_struct_check ((repl), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1836, __FUNCTION__))->typed.type)), repl); |
| 1837 | tree conv = convert_to_pointer (TREE_TYPE (*tp)((contains_struct_check ((*tp), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1837, __FUNCTION__))->typed.type), |
| 1838 | gimple_assign_lhs (stmt1)); |
| 1839 | gimple *stmt2 |
| 1840 | = gimple_build_assign (make_ssa_name (TREE_TYPE (*tp)((contains_struct_check ((*tp), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1840, __FUNCTION__))->typed.type)), conv); |
| 1841 | gsi_insert_before (&gsi, stmt1, GSI_SAME_STMT); |
| 1842 | gsi_insert_before (&gsi, stmt2, GSI_SAME_STMT); |
| 1843 | *tp = gimple_assign_lhs (stmt2); |
| 1844 | info->modified = true; |
| 1845 | *walk_subtrees = 0; |
| 1846 | } |
| 1847 | } |
| 1848 | } |
| 1849 | else if (TREE_CODE (*tp)((enum tree_code) (*tp)->base.code) == COMPONENT_REF || TREE_CODE (*tp)((enum tree_code) (*tp)->base.code) == ARRAY_REF) |
| 1850 | { |
| 1851 | tree *base = &TREE_OPERAND (*tp, 0)(*((const_cast<tree*> (tree_operand_check ((*tp), (0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1851, __FUNCTION__))))); |
| 1852 | |
| 1853 | while (TREE_CODE (*base)((enum tree_code) (*base)->base.code) == COMPONENT_REF |
| 1854 | || TREE_CODE (*base)((enum tree_code) (*base)->base.code) == ARRAY_REF) |
| 1855 | base = &TREE_OPERAND (*base, 0)(*((const_cast<tree*> (tree_operand_check ((*base), (0) , "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1855, __FUNCTION__))))); |
| 1856 | |
| 1857 | if (TREE_CODE (*base)((enum tree_code) (*base)->base.code) != VAR_DECL) |
| 1858 | return NULL__null; |
| 1859 | |
| 1860 | tree *new_decl = info->adjusted_vars->get (*base); |
| 1861 | if (!new_decl) |
| 1862 | return NULL__null; |
| 1863 | |
| 1864 | int base_quals = TYPE_QUALS (TREE_TYPE (*new_decl))((int) ((((tree_class_check ((((contains_struct_check ((*new_decl ), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1864, __FUNCTION__))->typed.type)), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1864, __FUNCTION__))->base.readonly_flag) * TYPE_QUAL_CONST ) | (((tree_class_check ((((contains_struct_check ((*new_decl ), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1864, __FUNCTION__))->typed.type)), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1864, __FUNCTION__))->base.volatile_flag) * TYPE_QUAL_VOLATILE ) | (((tree_class_check ((((contains_struct_check ((*new_decl ), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1864, __FUNCTION__))->typed.type)), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1864, __FUNCTION__))->base.u.bits.atomic_flag) * TYPE_QUAL_ATOMIC ) | (((tree_class_check ((((contains_struct_check ((*new_decl ), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1864, __FUNCTION__))->typed.type)), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1864, __FUNCTION__))->type_common.restrict_flag) * TYPE_QUAL_RESTRICT ) | (((((tree_class_check ((((contains_struct_check ((*new_decl ), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1864, __FUNCTION__))->typed.type)), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1864, __FUNCTION__))->base.u.bits.address_space) & 0xFF ) << 8)))); |
| 1865 | tree field = TREE_OPERAND (*tp, 1)(*((const_cast<tree*> (tree_operand_check ((*tp), (1), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1865, __FUNCTION__))))); |
| 1866 | |
| 1867 | /* Adjust the type of the field. */ |
| 1868 | int field_quals = TYPE_QUALS (TREE_TYPE (field))((int) ((((tree_class_check ((((contains_struct_check ((field ), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1868, __FUNCTION__))->typed.type)), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1868, __FUNCTION__))->base.readonly_flag) * TYPE_QUAL_CONST ) | (((tree_class_check ((((contains_struct_check ((field), ( TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1868, __FUNCTION__))->typed.type)), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1868, __FUNCTION__))->base.volatile_flag) * TYPE_QUAL_VOLATILE ) | (((tree_class_check ((((contains_struct_check ((field), ( TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1868, __FUNCTION__))->typed.type)), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1868, __FUNCTION__))->base.u.bits.atomic_flag) * TYPE_QUAL_ATOMIC ) | (((tree_class_check ((((contains_struct_check ((field), ( TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1868, __FUNCTION__))->typed.type)), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1868, __FUNCTION__))->type_common.restrict_flag) * TYPE_QUAL_RESTRICT ) | (((((tree_class_check ((((contains_struct_check ((field), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1868, __FUNCTION__))->typed.type)), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1868, __FUNCTION__))->base.u.bits.address_space) & 0xFF ) << 8)))); |
| 1869 | if (TREE_CODE (field)((enum tree_code) (field)->base.code) == FIELD_DECL && field_quals != base_quals) |
| 1870 | { |
| 1871 | tree *field_type = &TREE_TYPE (field)((contains_struct_check ((field), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1871, __FUNCTION__))->typed.type); |
| 1872 | while (TREE_CODE (*field_type)((enum tree_code) (*field_type)->base.code) == ARRAY_TYPE) |
| 1873 | field_type = &TREE_TYPE (*field_type)((contains_struct_check ((*field_type), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1873, __FUNCTION__))->typed.type); |
| 1874 | field_quals |= base_quals; |
| 1875 | *field_type = build_qualified_type (*field_type, field_quals); |
| 1876 | } |
| 1877 | |
| 1878 | /* Adjust the type of the component ref itself. */ |
| 1879 | tree comp_type = TREE_TYPE (*tp)((contains_struct_check ((*tp), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1879, __FUNCTION__))->typed.type); |
| 1880 | int comp_quals = TYPE_QUALS (comp_type)((int) ((((tree_class_check ((comp_type), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1880, __FUNCTION__))->base.readonly_flag) * TYPE_QUAL_CONST ) | (((tree_class_check ((comp_type), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1880, __FUNCTION__))->base.volatile_flag) * TYPE_QUAL_VOLATILE ) | (((tree_class_check ((comp_type), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1880, __FUNCTION__))->base.u.bits.atomic_flag) * TYPE_QUAL_ATOMIC ) | (((tree_class_check ((comp_type), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1880, __FUNCTION__))->type_common.restrict_flag) * TYPE_QUAL_RESTRICT ) | (((((tree_class_check ((comp_type), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1880, __FUNCTION__))->base.u.bits.address_space) & 0xFF ) << 8)))); |
| 1881 | if (TREE_CODE (*tp)((enum tree_code) (*tp)->base.code) == COMPONENT_REF && comp_quals != base_quals) |
| 1882 | { |
| 1883 | comp_quals |= base_quals; |
| 1884 | TREE_TYPE (*tp)((contains_struct_check ((*tp), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1884, __FUNCTION__))->typed.type) |
| 1885 | = build_qualified_type (comp_type, comp_quals); |
| 1886 | } |
| 1887 | |
| 1888 | *base = *new_decl; |
| 1889 | info->modified = true; |
| 1890 | } |
| 1891 | else if (TREE_CODE (*tp)((enum tree_code) (*tp)->base.code) == VAR_DECL) |
| 1892 | { |
| 1893 | tree *new_decl = info->adjusted_vars->get (*tp); |
| 1894 | if (new_decl) |
| 1895 | { |
| 1896 | *tp = *new_decl; |
| 1897 | info->modified = true; |
| 1898 | } |
| 1899 | } |
| 1900 | |
| 1901 | return NULL_TREE(tree) __null; |
| 1902 | } |
| 1903 | |
| 1904 | /* Return TRUE if CALL is a call to a builtin atomic/sync operation. */ |
| 1905 | |
| 1906 | static bool |
| 1907 | is_sync_builtin_call (gcall *call) |
| 1908 | { |
| 1909 | tree callee = gimple_call_fndecl (call); |
| 1910 | |
| 1911 | if (callee != NULL_TREE(tree) __null |
| 1912 | && gimple_call_builtin_p (call, BUILT_IN_NORMAL)) |
| 1913 | switch (DECL_FUNCTION_CODE (callee)) |
| 1914 | { |
| 1915 | #undef DEF_SYNC_BUILTIN |
| 1916 | #define DEF_SYNC_BUILTIN(ENUM, NAME, TYPE, ATTRS) case ENUM: |
| 1917 | #include "sync-builtins.def" |
| 1918 | #undef DEF_SYNC_BUILTIN |
| 1919 | return true; |
| 1920 | |
| 1921 | default: |
| 1922 | ; |
| 1923 | } |
| 1924 | |
| 1925 | return false; |
| 1926 | } |
| 1927 | |
| 1928 | /* Main entry point for oacc transformations which run on the device |
| 1929 | compiler after LTO, so we know what the target device is at this |
| 1930 | point (including the host fallback). */ |
| 1931 | |
| 1932 | static unsigned int |
| 1933 | execute_oacc_loop_designation () |
| 1934 | { |
| 1935 | tree attrs = oacc_get_fn_attrib (current_function_decl); |
| 1936 | |
| 1937 | if (!attrs) |
| 1938 | /* Not an offloaded function. */ |
| 1939 | return 0; |
| 1940 | |
| 1941 | /* Parse the default dim argument exactly once. */ |
| 1942 | if ((const void *)flag_openacc_dimsglobal_options.x_flag_openacc_dims != &flag_openacc_dimsglobal_options.x_flag_openacc_dims) |
| 1943 | { |
| 1944 | oacc_parse_default_dims (flag_openacc_dimsglobal_options.x_flag_openacc_dims); |
| 1945 | flag_openacc_dimsglobal_options.x_flag_openacc_dims = (char *)&flag_openacc_dimsglobal_options.x_flag_openacc_dims; |
| 1946 | } |
| 1947 | |
| 1948 | bool is_oacc_parallel |
| 1949 | = (lookup_attribute ("oacc parallel", |
| 1950 | DECL_ATTRIBUTES (current_function_decl)((contains_struct_check ((current_function_decl), (TS_DECL_COMMON ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1950, __FUNCTION__))->decl_common.attributes)) != NULL__null); |
| 1951 | bool is_oacc_kernels |
| 1952 | = (lookup_attribute ("oacc kernels", |
| 1953 | DECL_ATTRIBUTES (current_function_decl)((contains_struct_check ((current_function_decl), (TS_DECL_COMMON ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1953, __FUNCTION__))->decl_common.attributes)) != NULL__null); |
| 1954 | bool is_oacc_serial |
| 1955 | = (lookup_attribute ("oacc serial", |
| 1956 | DECL_ATTRIBUTES (current_function_decl)((contains_struct_check ((current_function_decl), (TS_DECL_COMMON ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1956, __FUNCTION__))->decl_common.attributes)) != NULL__null); |
| 1957 | bool is_oacc_parallel_kernels_parallelized |
| 1958 | = (lookup_attribute ("oacc parallel_kernels_parallelized", |
| 1959 | DECL_ATTRIBUTES (current_function_decl)((contains_struct_check ((current_function_decl), (TS_DECL_COMMON ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1959, __FUNCTION__))->decl_common.attributes)) != NULL__null); |
| 1960 | bool is_oacc_parallel_kernels_gang_single |
| 1961 | = (lookup_attribute ("oacc parallel_kernels_gang_single", |
| 1962 | DECL_ATTRIBUTES (current_function_decl)((contains_struct_check ((current_function_decl), (TS_DECL_COMMON ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1962, __FUNCTION__))->decl_common.attributes)) != NULL__null); |
| 1963 | int fn_level = oacc_fn_attrib_level (attrs); |
| 1964 | bool is_oacc_routine = (fn_level >= 0); |
| 1965 | gcc_checking_assert (is_oacc_parallel((void)(!(is_oacc_parallel + is_oacc_kernels + is_oacc_serial + is_oacc_parallel_kernels_parallelized + is_oacc_parallel_kernels_gang_single + is_oacc_routine == 1) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1971, __FUNCTION__), 0 : 0)) |
| 1966 | + is_oacc_kernels((void)(!(is_oacc_parallel + is_oacc_kernels + is_oacc_serial + is_oacc_parallel_kernels_parallelized + is_oacc_parallel_kernels_gang_single + is_oacc_routine == 1) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1971, __FUNCTION__), 0 : 0)) |
| 1967 | + is_oacc_serial((void)(!(is_oacc_parallel + is_oacc_kernels + is_oacc_serial + is_oacc_parallel_kernels_parallelized + is_oacc_parallel_kernels_gang_single + is_oacc_routine == 1) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1971, __FUNCTION__), 0 : 0)) |
| 1968 | + is_oacc_parallel_kernels_parallelized((void)(!(is_oacc_parallel + is_oacc_kernels + is_oacc_serial + is_oacc_parallel_kernels_parallelized + is_oacc_parallel_kernels_gang_single + is_oacc_routine == 1) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1971, __FUNCTION__), 0 : 0)) |
| 1969 | + is_oacc_parallel_kernels_gang_single((void)(!(is_oacc_parallel + is_oacc_kernels + is_oacc_serial + is_oacc_parallel_kernels_parallelized + is_oacc_parallel_kernels_gang_single + is_oacc_routine == 1) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1971, __FUNCTION__), 0 : 0)) |
| 1970 | + is_oacc_routine((void)(!(is_oacc_parallel + is_oacc_kernels + is_oacc_serial + is_oacc_parallel_kernels_parallelized + is_oacc_parallel_kernels_gang_single + is_oacc_routine == 1) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1971, __FUNCTION__), 0 : 0)) |
| 1971 | == 1)((void)(!(is_oacc_parallel + is_oacc_kernels + is_oacc_serial + is_oacc_parallel_kernels_parallelized + is_oacc_parallel_kernels_gang_single + is_oacc_routine == 1) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1971, __FUNCTION__), 0 : 0)); |
| 1972 | |
| 1973 | bool is_oacc_kernels_parallelized |
| 1974 | = (lookup_attribute ("oacc kernels parallelized", |
| 1975 | DECL_ATTRIBUTES (current_function_decl)((contains_struct_check ((current_function_decl), (TS_DECL_COMMON ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1975, __FUNCTION__))->decl_common.attributes)) != NULL__null); |
| 1976 | if (is_oacc_kernels_parallelized) |
| 1977 | gcc_checking_assert (is_oacc_kernels)((void)(!(is_oacc_kernels) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1977, __FUNCTION__), 0 : 0)); |
| 1978 | |
| 1979 | if (dump_file) |
| 1980 | { |
| 1981 | if (is_oacc_parallel) |
| 1982 | fprintf (dump_file, "Function is OpenACC parallel offload\n"); |
| 1983 | else if (is_oacc_kernels) |
| 1984 | fprintf (dump_file, "Function is %s OpenACC kernels offload\n", |
| 1985 | (is_oacc_kernels_parallelized |
| 1986 | ? "parallelized" : "unparallelized")); |
| 1987 | else if (is_oacc_serial) |
| 1988 | fprintf (dump_file, "Function is OpenACC serial offload\n"); |
| 1989 | else if (is_oacc_parallel_kernels_parallelized) |
| 1990 | fprintf (dump_file, "Function is %s OpenACC kernels offload\n", |
| 1991 | "parallel_kernels_parallelized"); |
| 1992 | else if (is_oacc_parallel_kernels_gang_single) |
| 1993 | fprintf (dump_file, "Function is %s OpenACC kernels offload\n", |
| 1994 | "parallel_kernels_gang_single"); |
| 1995 | else if (is_oacc_routine) |
| 1996 | fprintf (dump_file, "Function is OpenACC routine level %d\n", |
| 1997 | fn_level); |
| 1998 | else |
| 1999 | gcc_unreachable ()(fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 1999, __FUNCTION__)); |
| 2000 | } |
| 2001 | |
| 2002 | /* This doesn't belong into 'pass_oacc_loop_designation' conceptually, but |
| 2003 | it's a convenient place, so... */ |
| 2004 | if (is_oacc_routine) |
| 2005 | { |
| 2006 | tree attr = lookup_attribute ("omp declare target", |
| 2007 | DECL_ATTRIBUTES (current_function_decl)((contains_struct_check ((current_function_decl), (TS_DECL_COMMON ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2007, __FUNCTION__))->decl_common.attributes)); |
| 2008 | gcc_checking_assert (attr)((void)(!(attr) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2008, __FUNCTION__), 0 : 0)); |
| 2009 | tree clauses = TREE_VALUE (attr)((tree_check ((attr), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2009, __FUNCTION__, (TREE_LIST)))->list.value); |
| 2010 | gcc_checking_assert (clauses)((void)(!(clauses) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2010, __FUNCTION__), 0 : 0)); |
| 2011 | |
| 2012 | /* Should this OpenACC routine be discarded? */ |
| 2013 | bool discard = false; |
| 2014 | |
| 2015 | tree clause_nohost = omp_find_clause (clauses, OMP_CLAUSE_NOHOST); |
| 2016 | if (dump_file) |
| 2017 | fprintf (dump_file, |
| 2018 | "OpenACC routine '%s' %s '%s' clause.\n", |
| 2019 | lang_hooks.decl_printable_name (current_function_decl, 2), |
| 2020 | clause_nohost ? "has" : "doesn't have", |
| 2021 | omp_clause_code_name[OMP_CLAUSE_NOHOST]); |
| 2022 | /* Host compiler, 'nohost' clause? */ |
| 2023 | #ifndef ACCEL_COMPILER |
| 2024 | if (clause_nohost) |
| 2025 | discard = true; |
| 2026 | #endif |
| 2027 | |
| 2028 | if (dump_file) |
| 2029 | fprintf (dump_file, |
| 2030 | "OpenACC routine '%s' %sdiscarded.\n", |
| 2031 | lang_hooks.decl_printable_name (current_function_decl, 2), |
| 2032 | discard ? "" : "not "); |
| 2033 | if (discard) |
| 2034 | { |
| 2035 | TREE_ASM_WRITTEN (current_function_decl)((current_function_decl)->base.asm_written_flag) = 1; |
| 2036 | return TODO_discard_function(1 << 23); |
| 2037 | } |
| 2038 | } |
| 2039 | |
| 2040 | /* Unparallelized OpenACC kernels constructs must get launched as 1 x 1 x 1 |
| 2041 | kernels, so remove the parallelism dimensions function attributes |
| 2042 | potentially set earlier on. */ |
| 2043 | if (is_oacc_kernels && !is_oacc_kernels_parallelized) |
| 2044 | { |
| 2045 | oacc_set_fn_attrib (current_function_decl, NULL__null, NULL__null); |
| 2046 | attrs = oacc_get_fn_attrib (current_function_decl); |
| 2047 | } |
| 2048 | |
| 2049 | /* Discover, partition and process the loops. */ |
| 2050 | oacc_loop *loops = oacc_loop_discovery (); |
| 2051 | |
| 2052 | unsigned outer_mask = 0; |
| 2053 | if (is_oacc_routine) |
| 2054 | outer_mask = GOMP_DIM_MASK (fn_level)(1u << (fn_level)) - 1; |
| 2055 | unsigned used_mask = oacc_loop_partition (loops, outer_mask); |
| 2056 | /* OpenACC kernels constructs are special: they currently don't use the |
| 2057 | generic oacc_loop infrastructure and attribute/dimension processing. */ |
| 2058 | if (is_oacc_kernels && is_oacc_kernels_parallelized) |
| 2059 | { |
| 2060 | /* Parallelized OpenACC kernels constructs use gang parallelism. See |
| 2061 | also tree-parloops.cc:create_parallel_loop. */ |
| 2062 | used_mask |= GOMP_DIM_MASK (GOMP_DIM_GANG)(1u << (0)); |
| 2063 | } |
| 2064 | |
| 2065 | int dims[GOMP_DIM_MAX3]; |
| 2066 | oacc_validate_dims (current_function_decl, attrs, dims, fn_level, used_mask); |
| 2067 | |
| 2068 | if (dump_file) |
| 2069 | { |
| 2070 | const char *comma = "Compute dimensions ["; |
| 2071 | for (int ix = 0; ix != GOMP_DIM_MAX3; ix++, comma = ", ") |
| 2072 | fprintf (dump_file, "%s%d", comma, dims[ix]); |
| 2073 | fprintf (dump_file, "]\n"); |
| 2074 | } |
| 2075 | |
| 2076 | /* Verify that for OpenACC 'kernels' decomposed "gang-single" parts we launch |
| 2077 | a single gang only. */ |
| 2078 | if (is_oacc_parallel_kernels_gang_single) |
| 2079 | gcc_checking_assert (dims[GOMP_DIM_GANG] == 1)((void)(!(dims[0] == 1) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2079, __FUNCTION__), 0 : 0)); |
| 2080 | |
| 2081 | oacc_loop_process (loops, fn_level); |
| 2082 | if (dump_file) |
| 2083 | { |
| 2084 | fprintf (dump_file, "OpenACC loops\n"); |
| 2085 | dump_oacc_loop (dump_file, loops, 0); |
| 2086 | fprintf (dump_file, "\n"); |
| 2087 | } |
| 2088 | if (dump_enabled_p ()) |
| 2089 | { |
| 2090 | oacc_loop *l = loops; |
| 2091 | /* OpenACC kernels constructs are special: they currently don't use the |
| 2092 | generic oacc_loop infrastructure. */ |
| 2093 | if (is_oacc_kernels) |
| 2094 | { |
| 2095 | /* Create a fake oacc_loop for diagnostic purposes. */ |
| 2096 | l = new_oacc_loop_raw (NULL__null, |
| 2097 | DECL_SOURCE_LOCATION (current_function_decl)((contains_struct_check ((current_function_decl), (TS_DECL_MINIMAL ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2097, __FUNCTION__))->decl_minimal.locus)); |
| 2098 | l->mask = used_mask; |
| 2099 | } |
| 2100 | else |
| 2101 | { |
| 2102 | /* Skip the outermost, dummy OpenACC loop */ |
| 2103 | l = l->child; |
| 2104 | } |
| 2105 | if (l) |
| 2106 | inform_oacc_loop (l); |
| 2107 | if (is_oacc_kernels) |
| 2108 | free_oacc_loop (l); |
| 2109 | } |
| 2110 | |
| 2111 | free_oacc_loop (loops); |
| 2112 | |
| 2113 | return 0; |
| 2114 | } |
| 2115 | |
| 2116 | static unsigned int |
| 2117 | execute_oacc_device_lower () |
| 2118 | { |
| 2119 | tree attrs = oacc_get_fn_attrib (current_function_decl); |
| 2120 | |
| 2121 | if (!attrs) |
| 2122 | /* Not an offloaded function. */ |
| 2123 | return 0; |
| 2124 | |
| 2125 | int dims[GOMP_DIM_MAX3]; |
| 2126 | for (unsigned i = 0; i < GOMP_DIM_MAX3; i++) |
| 2127 | dims[i] = oacc_get_fn_dim_size (current_function_decl, i); |
| 2128 | |
| 2129 | hash_map<tree, tree> adjusted_vars; |
| 2130 | |
| 2131 | /* Now lower internal loop functions to target-specific code |
| 2132 | sequences. */ |
| 2133 | basic_block bb; |
| 2134 | FOR_ALL_BB_FN (bb, cfun)for (bb = (((cfun + 0))->cfg->x_entry_block_ptr); bb; bb = bb->next_bb) |
| 2135 | for (gimple_stmt_iterator gsi = gsi_start_bb (bb); !gsi_end_p (gsi);) |
| 2136 | { |
| 2137 | gimple *stmt = gsi_stmt (gsi); |
| 2138 | if (!is_gimple_call (stmt)) |
| 2139 | { |
| 2140 | gsi_next (&gsi); |
| 2141 | continue; |
| 2142 | } |
| 2143 | |
| 2144 | gcall *call = as_a <gcall *> (stmt); |
| 2145 | if (!gimple_call_internal_p (call)) |
| 2146 | { |
| 2147 | gsi_next (&gsi); |
| 2148 | continue; |
| 2149 | } |
| 2150 | |
| 2151 | /* Rewind to allow rescan. */ |
| 2152 | gsi_prev (&gsi); |
| 2153 | bool rescan = false, remove = false; |
| 2154 | enum internal_fn ifn_code = gimple_call_internal_fn (call); |
| 2155 | |
| 2156 | switch (ifn_code) |
| 2157 | { |
| 2158 | default: break; |
| 2159 | |
| 2160 | case IFN_GOACC_TILE: |
| 2161 | oacc_xform_tile (call); |
| 2162 | rescan = true; |
| 2163 | break; |
| 2164 | |
| 2165 | case IFN_GOACC_LOOP: |
| 2166 | oacc_xform_loop (call); |
| 2167 | rescan = true; |
| 2168 | break; |
| 2169 | |
| 2170 | case IFN_GOACC_REDUCTION: |
| 2171 | /* Mark the function for SSA renaming. */ |
| 2172 | mark_virtual_operands_for_renaming (cfun(cfun + 0)); |
| 2173 | |
| 2174 | /* If the level is -1, this ended up being an unused |
| 2175 | axis. Handle as a default. */ |
| 2176 | if (integer_minus_onep (gimple_call_arg (call, 3))) |
| 2177 | default_goacc_reduction (call); |
| 2178 | else |
| 2179 | targetm.goacc.reduction (call); |
| 2180 | rescan = true; |
| 2181 | break; |
| 2182 | |
| 2183 | case IFN_UNIQUE: |
| 2184 | { |
| 2185 | enum ifn_unique_kind kind |
| 2186 | = ((enum ifn_unique_kind) |
| 2187 | TREE_INT_CST_LOW (gimple_call_arg (call, 0))((unsigned long) (*tree_int_cst_elt_check ((gimple_call_arg ( call, 0)), (0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2187, __FUNCTION__)))); |
| 2188 | |
| 2189 | switch (kind) |
| 2190 | { |
| 2191 | default: |
| 2192 | break; |
| 2193 | |
| 2194 | case IFN_UNIQUE_OACC_FORK: |
| 2195 | case IFN_UNIQUE_OACC_JOIN: |
| 2196 | if (integer_minus_onep (gimple_call_arg (call, 2))) |
| 2197 | remove = true; |
| 2198 | else if (!targetm.goacc.fork_join |
| 2199 | (call, dims, kind == IFN_UNIQUE_OACC_FORK)) |
| 2200 | remove = true; |
| 2201 | break; |
| 2202 | |
| 2203 | case IFN_UNIQUE_OACC_HEAD_MARK: |
| 2204 | case IFN_UNIQUE_OACC_TAIL_MARK: |
| 2205 | remove = true; |
| 2206 | break; |
| 2207 | |
| 2208 | case IFN_UNIQUE_OACC_PRIVATE: |
| 2209 | { |
| 2210 | dump_flags_t l_dump_flags |
| 2211 | = get_openacc_privatization_dump_flags (); |
| 2212 | |
| 2213 | location_t loc = gimple_location (stmt); |
| 2214 | if (LOCATION_LOCUS (loc)((IS_ADHOC_LOC (loc)) ? get_location_from_adhoc_loc (line_table , loc) : (loc)) == UNKNOWN_LOCATION((location_t) 0)) |
| 2215 | loc = DECL_SOURCE_LOCATION (current_function_decl)((contains_struct_check ((current_function_decl), (TS_DECL_MINIMAL ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2215, __FUNCTION__))->decl_minimal.locus); |
| 2216 | const dump_user_location_t d_u_loc |
| 2217 | = dump_user_location_t::from_location_t (loc); |
| 2218 | |
| 2219 | HOST_WIDE_INTlong level |
| 2220 | = TREE_INT_CST_LOW (gimple_call_arg (call, 2))((unsigned long) (*tree_int_cst_elt_check ((gimple_call_arg ( call, 2)), (0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2220, __FUNCTION__))); |
| 2221 | gcc_checking_assert (level == -1((void)(!(level == -1 || (level >= 0 && level < 3)) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2223, __FUNCTION__), 0 : 0)) |
| 2222 | || (level >= 0((void)(!(level == -1 || (level >= 0 && level < 3)) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2223, __FUNCTION__), 0 : 0)) |
| 2223 | && level < GOMP_DIM_MAX))((void)(!(level == -1 || (level >= 0 && level < 3)) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2223, __FUNCTION__), 0 : 0)); |
| 2224 | for (unsigned i = 3; |
| 2225 | i < gimple_call_num_args (call); |
| 2226 | i++) |
| 2227 | { |
| 2228 | static char const *const axes[] = |
| 2229 | /* Must be kept in sync with GOMP_DIM enumeration. */ |
| 2230 | { "gang", "worker", "vector" }; |
| 2231 | |
| 2232 | tree arg = gimple_call_arg (call, i); |
| 2233 | gcc_checking_assert (TREE_CODE (arg) == ADDR_EXPR)((void)(!(((enum tree_code) (arg)->base.code) == ADDR_EXPR ) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2233, __FUNCTION__), 0 : 0)); |
| 2234 | tree decl = TREE_OPERAND (arg, 0)(*((const_cast<tree*> (tree_operand_check ((arg), (0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2234, __FUNCTION__))))); |
| 2235 | if (dump_enabled_p ()) |
| 2236 | /* PR100695 "Format decoder, quoting in 'dump_printf' etc." */ |
| 2237 | #if __GNUC__4 >= 10 |
| 2238 | # pragma GCC diagnostic push |
| 2239 | # pragma GCC diagnostic ignored "-Wformat" |
| 2240 | #endif |
| 2241 | dump_printf_loc (l_dump_flags, d_u_loc, |
| 2242 | "variable %<%T%> ought to be" |
| 2243 | " adjusted for OpenACC" |
| 2244 | " privatization level: %qs\n", |
| 2245 | decl, |
| 2246 | (level == -1 |
| 2247 | ? "UNKNOWN" : axes[level])); |
| 2248 | #if __GNUC__4 >= 10 |
| 2249 | # pragma GCC diagnostic pop |
| 2250 | #endif |
| 2251 | bool adjusted; |
| 2252 | if (level == -1) |
| 2253 | adjusted = false; |
| 2254 | else if (!targetm.goacc.adjust_private_decl) |
| 2255 | adjusted = false; |
| 2256 | else if (level == GOMP_DIM_VECTOR2) |
| 2257 | { |
| 2258 | /* That's the default behavior. */ |
| 2259 | adjusted = true; |
| 2260 | } |
| 2261 | else |
| 2262 | { |
| 2263 | tree oldtype = TREE_TYPE (decl)((contains_struct_check ((decl), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2263, __FUNCTION__))->typed.type); |
| 2264 | tree newdecl |
| 2265 | = targetm.goacc.adjust_private_decl (loc, decl, |
| 2266 | level); |
| 2267 | adjusted = (TREE_TYPE (newdecl)((contains_struct_check ((newdecl), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2267, __FUNCTION__))->typed.type) != oldtype |
| 2268 | || newdecl != decl); |
| 2269 | if (adjusted) |
| 2270 | adjusted_vars.put (decl, newdecl); |
| 2271 | } |
| 2272 | if (adjusted |
| 2273 | && dump_enabled_p ()) |
| 2274 | /* PR100695 "Format decoder, quoting in 'dump_printf' etc." */ |
| 2275 | #if __GNUC__4 >= 10 |
| 2276 | # pragma GCC diagnostic push |
| 2277 | # pragma GCC diagnostic ignored "-Wformat" |
| 2278 | #endif |
| 2279 | dump_printf_loc (l_dump_flags, d_u_loc, |
| 2280 | "variable %<%T%> adjusted for" |
| 2281 | " OpenACC privatization level:" |
| 2282 | " %qs\n", |
| 2283 | decl, axes[level]); |
| 2284 | #if __GNUC__4 >= 10 |
| 2285 | # pragma GCC diagnostic pop |
| 2286 | #endif |
| 2287 | } |
| 2288 | remove = true; |
| 2289 | } |
| 2290 | break; |
| 2291 | } |
| 2292 | break; |
| 2293 | } |
| 2294 | } |
| 2295 | |
| 2296 | if (gsi_end_p (gsi)) |
| 2297 | /* We rewound past the beginning of the BB. */ |
| 2298 | gsi = gsi_start_bb (bb); |
| 2299 | else |
| 2300 | /* Undo the rewind. */ |
| 2301 | gsi_next (&gsi); |
| 2302 | |
| 2303 | if (remove) |
| 2304 | { |
| 2305 | if (gimple_vdef (call)) |
| 2306 | replace_uses_by (gimple_vdef (call), gimple_vuse (call)); |
| 2307 | if (gimple_call_lhs (call)) |
| 2308 | { |
| 2309 | /* Propagate the data dependency var. */ |
| 2310 | gimple *ass = gimple_build_assign (gimple_call_lhs (call), |
| 2311 | gimple_call_arg (call, 1)); |
| 2312 | gsi_replace (&gsi, ass, false); |
| 2313 | } |
| 2314 | else |
| 2315 | gsi_remove (&gsi, true); |
| 2316 | } |
| 2317 | else if (!rescan) |
| 2318 | /* If not rescanning, advance over the call. */ |
| 2319 | gsi_next (&gsi); |
| 2320 | } |
| 2321 | |
| 2322 | /* Regarding the OpenACC privatization level, we're currently only looking at |
| 2323 | making the gang-private level work. Regarding that, we have the following |
| 2324 | configurations: |
| 2325 | |
| 2326 | - GCN offloading: 'targetm.goacc.adjust_private_decl' does the work (in |
| 2327 | particular, change 'TREE_TYPE', etc.) and there is no |
| 2328 | 'targetm.goacc.expand_var_decl'. |
| 2329 | |
| 2330 | - nvptx offloading: 'targetm.goacc.adjust_private_decl' only sets a |
| 2331 | marker and then 'targetm.goacc.expand_var_decl' does the work. |
| 2332 | |
| 2333 | Eventually (in particular, for worker-private level?), both |
| 2334 | 'targetm.goacc.adjust_private_decl' and 'targetm.goacc.expand_var_decl' |
| 2335 | may need to do things, but that's currently not meant to be addressed, and |
| 2336 | thus not fully worked out and implemented, and thus untested. Hence, |
| 2337 | 'assert' what currently is implemented/tested, only. */ |
| 2338 | |
| 2339 | if (targetm.goacc.expand_var_decl) |
| 2340 | gcc_assert (adjusted_vars.is_empty ())((void)(!(adjusted_vars.is_empty ()) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2340, __FUNCTION__), 0 : 0)); |
| 2341 | |
| 2342 | /* Make adjustments to gang-private local variables if required by the |
| 2343 | target, e.g. forcing them into a particular address space. Afterwards, |
| 2344 | ADDR_EXPR nodes which have adjusted variables as their argument need to |
| 2345 | be modified in one of two ways: |
| 2346 | |
| 2347 | 1. They can be recreated, making a pointer to the variable in the new |
| 2348 | address space, or |
| 2349 | |
| 2350 | 2. The address of the variable in the new address space can be taken, |
| 2351 | converted to the default (original) address space, and the result of |
| 2352 | that conversion subsituted in place of the original ADDR_EXPR node. |
| 2353 | |
| 2354 | Which of these is done depends on the gimple statement being processed. |
| 2355 | At present atomic operations and inline asms use (1), and everything else |
| 2356 | uses (2). At least on AMD GCN, there are atomic operations that work |
| 2357 | directly in the LDS address space. |
| 2358 | |
| 2359 | COMPONENT_REFS, ARRAY_REFS and plain VAR_DECLs are also rewritten to use |
| 2360 | the new decl, adjusting types of appropriate tree nodes as necessary. */ |
| 2361 | |
| 2362 | if (targetm.goacc.adjust_private_decl |
| 2363 | && !adjusted_vars.is_empty ()) |
| 2364 | { |
| 2365 | FOR_ALL_BB_FN (bb, cfun)for (bb = (((cfun + 0))->cfg->x_entry_block_ptr); bb; bb = bb->next_bb) |
| 2366 | for (gimple_stmt_iterator gsi = gsi_start_bb (bb); |
| 2367 | !gsi_end_p (gsi); |
| 2368 | gsi_next (&gsi)) |
| 2369 | { |
| 2370 | gimple *stmt = gsi_stmt (gsi); |
| 2371 | walk_stmt_info wi; |
| 2372 | var_decl_rewrite_info info; |
| 2373 | |
| 2374 | info.avoid_pointer_conversion |
| 2375 | = (is_gimple_call (stmt) |
| 2376 | && is_sync_builtin_call (as_a <gcall *> (stmt))) |
| 2377 | || gimple_code (stmt) == GIMPLE_ASM; |
| 2378 | info.stmt = stmt; |
| 2379 | info.modified = false; |
| 2380 | info.adjusted_vars = &adjusted_vars; |
| 2381 | |
| 2382 | memset (&wi, 0, sizeof (wi)); |
| 2383 | wi.info = &info; |
| 2384 | |
| 2385 | walk_gimple_op (stmt, oacc_rewrite_var_decl, &wi); |
| 2386 | |
| 2387 | if (info.modified) |
| 2388 | update_stmt (stmt); |
| 2389 | } |
| 2390 | } |
| 2391 | |
| 2392 | return 0; |
| 2393 | } |
| 2394 | |
| 2395 | /* Default launch dimension validator. Force everything to 1. A |
| 2396 | backend that wants to provide larger dimensions must override this |
| 2397 | hook. */ |
| 2398 | |
| 2399 | bool |
| 2400 | default_goacc_validate_dims (tree ARG_UNUSED (decl)decl __attribute__ ((__unused__)), int *dims, |
| 2401 | int ARG_UNUSED (fn_level)fn_level __attribute__ ((__unused__)), |
| 2402 | unsigned ARG_UNUSED (used)used __attribute__ ((__unused__))) |
| 2403 | { |
| 2404 | bool changed = false; |
| 2405 | |
| 2406 | for (unsigned ix = 0; ix != GOMP_DIM_MAX3; ix++) |
| 2407 | { |
| 2408 | if (dims[ix] != 1) |
| 2409 | { |
| 2410 | dims[ix] = 1; |
| 2411 | changed = true; |
| 2412 | } |
| 2413 | } |
| 2414 | |
| 2415 | return changed; |
| 2416 | } |
| 2417 | |
| 2418 | /* Default dimension bound is unknown on accelerator and 1 on host. */ |
| 2419 | |
| 2420 | int |
| 2421 | default_goacc_dim_limit (int ARG_UNUSED (axis)axis __attribute__ ((__unused__))) |
| 2422 | { |
| 2423 | #ifdef ACCEL_COMPILER |
| 2424 | return 0; |
| 2425 | #else |
| 2426 | return 1; |
| 2427 | #endif |
| 2428 | } |
| 2429 | |
| 2430 | namespace { |
| 2431 | |
| 2432 | const pass_data pass_data_oacc_loop_designation = |
| 2433 | { |
| 2434 | GIMPLE_PASS, /* type */ |
| 2435 | "oaccloops", /* name */ |
| 2436 | OPTGROUP_OMP, /* optinfo_flags */ |
| 2437 | TV_NONE, /* tv_id */ |
| 2438 | PROP_cfg(1 << 3), /* properties_required */ |
| 2439 | 0 /* Possibly PROP_gimple_eomp. */, /* properties_provided */ |
| 2440 | 0, /* properties_destroyed */ |
| 2441 | 0, /* todo_flags_start */ |
| 2442 | TODO_update_ssa(1 << 11) | TODO_cleanup_cfg(1 << 5), /* todo_flags_finish */ |
| 2443 | }; |
| 2444 | |
| 2445 | class pass_oacc_loop_designation : public gimple_opt_pass |
| 2446 | { |
| 2447 | public: |
| 2448 | pass_oacc_loop_designation (gcc::context *ctxt) |
| 2449 | : gimple_opt_pass (pass_data_oacc_loop_designation, ctxt) |
| 2450 | {} |
| 2451 | |
| 2452 | /* opt_pass methods: */ |
| 2453 | bool gate (function *) final override { return flag_openaccglobal_options.x_flag_openacc; }; |
| 2454 | |
| 2455 | unsigned int execute (function *) final override |
| 2456 | { |
| 2457 | return execute_oacc_loop_designation (); |
| 2458 | } |
| 2459 | |
| 2460 | }; // class pass_oacc_loop_designation |
| 2461 | |
| 2462 | const pass_data pass_data_oacc_device_lower = |
| 2463 | { |
| 2464 | GIMPLE_PASS, /* type */ |
| 2465 | "oaccdevlow", /* name */ |
| 2466 | OPTGROUP_OMP, /* optinfo_flags */ |
| 2467 | TV_NONE, /* tv_id */ |
| 2468 | PROP_cfg(1 << 3), /* properties_required */ |
| 2469 | 0 /* Possibly PROP_gimple_eomp. */, /* properties_provided */ |
| 2470 | 0, /* properties_destroyed */ |
| 2471 | 0, /* todo_flags_start */ |
| 2472 | TODO_update_ssa(1 << 11) | TODO_cleanup_cfg(1 << 5), /* todo_flags_finish */ |
| 2473 | }; |
| 2474 | |
| 2475 | class pass_oacc_device_lower : public gimple_opt_pass |
| 2476 | { |
| 2477 | public: |
| 2478 | pass_oacc_device_lower (gcc::context *ctxt) |
| 2479 | : gimple_opt_pass (pass_data_oacc_device_lower, ctxt) |
| 2480 | {} |
| 2481 | |
| 2482 | /* opt_pass methods: */ |
| 2483 | bool gate (function *) final override { return flag_openaccglobal_options.x_flag_openacc; }; |
| 2484 | |
| 2485 | unsigned int execute (function *) final override |
| 2486 | { |
| 2487 | return execute_oacc_device_lower (); |
| 2488 | } |
| 2489 | |
| 2490 | }; // class pass_oacc_device_lower |
| 2491 | |
| 2492 | } // anon namespace |
| 2493 | |
| 2494 | gimple_opt_pass * |
| 2495 | make_pass_oacc_loop_designation (gcc::context *ctxt) |
| 2496 | { |
| 2497 | return new pass_oacc_loop_designation (ctxt); |
| 2498 | } |
| 2499 | |
| 2500 | gimple_opt_pass * |
| 2501 | make_pass_oacc_device_lower (gcc::context *ctxt) |
| 2502 | { |
| 2503 | return new pass_oacc_device_lower (ctxt); |
| 2504 | } |
| 2505 | |
| 2506 | |
| 2507 | /* Rewrite GOMP_SIMT_ENTER_ALLOC call given by GSI and remove the preceding |
| 2508 | GOMP_SIMT_ENTER call identifying the privatized variables, which are |
| 2509 | turned to structure fields and receive a DECL_VALUE_EXPR accordingly. |
| 2510 | Set *REGIMPLIFY to true, except if no privatized variables were seen. */ |
| 2511 | |
| 2512 | static void |
| 2513 | ompdevlow_adjust_simt_enter (gimple_stmt_iterator *gsi, bool *regimplify) |
| 2514 | { |
| 2515 | gimple *alloc_stmt = gsi_stmt (*gsi); |
| 2516 | tree simtrec = gimple_call_lhs (alloc_stmt); |
| 2517 | tree simduid = gimple_call_arg (alloc_stmt, 0); |
| 2518 | gimple *enter_stmt = SSA_NAME_DEF_STMT (simduid)(tree_check ((simduid), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2518, __FUNCTION__, (SSA_NAME)))->ssa_name.def_stmt; |
| 2519 | gcc_assert (gimple_call_internal_p (enter_stmt, IFN_GOMP_SIMT_ENTER))((void)(!(gimple_call_internal_p (enter_stmt, IFN_GOMP_SIMT_ENTER )) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2519, __FUNCTION__), 0 : 0)); |
| 2520 | tree rectype = lang_hooks.types.make_type (RECORD_TYPE); |
| 2521 | TYPE_ARTIFICIAL (rectype)((tree_class_check ((rectype), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2521, __FUNCTION__))->base.nowarning_flag) = TYPE_NAMELESS (rectype)((tree_class_check ((rectype), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2521, __FUNCTION__))->base.u.bits.nameless_flag) = 1; |
| 2522 | TREE_ADDRESSABLE (rectype)((rectype)->base.addressable_flag) = 1; |
| 2523 | TREE_TYPE (simtrec)((contains_struct_check ((simtrec), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2523, __FUNCTION__))->typed.type) = build_pointer_type (rectype); |
| 2524 | for (unsigned i = 1; i < gimple_call_num_args (enter_stmt); i++) |
| 2525 | { |
| 2526 | tree *argp = gimple_call_arg_ptr (enter_stmt, i); |
| 2527 | if (*argp == null_pointer_nodeglobal_trees[TI_NULL_POINTER]) |
| 2528 | continue; |
| 2529 | gcc_assert (TREE_CODE (*argp) == ADDR_EXPR((void)(!(((enum tree_code) (*argp)->base.code) == ADDR_EXPR && (((enum tree_code) ((*((const_cast<tree*> ( tree_operand_check ((*argp), (0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2530, __FUNCTION__))))))->base.code) == VAR_DECL)) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2530, __FUNCTION__), 0 : 0)) |
| 2530 | && VAR_P (TREE_OPERAND (*argp, 0)))((void)(!(((enum tree_code) (*argp)->base.code) == ADDR_EXPR && (((enum tree_code) ((*((const_cast<tree*> ( tree_operand_check ((*argp), (0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2530, __FUNCTION__))))))->base.code) == VAR_DECL)) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2530, __FUNCTION__), 0 : 0)); |
| 2531 | tree var = TREE_OPERAND (*argp, 0)(*((const_cast<tree*> (tree_operand_check ((*argp), (0) , "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2531, __FUNCTION__))))); |
| 2532 | |
| 2533 | tree field = build_decl (DECL_SOURCE_LOCATION (var)((contains_struct_check ((var), (TS_DECL_MINIMAL), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2533, __FUNCTION__))->decl_minimal.locus), FIELD_DECL, |
| 2534 | DECL_NAME (var)((contains_struct_check ((var), (TS_DECL_MINIMAL), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2534, __FUNCTION__))->decl_minimal.name), TREE_TYPE (var)((contains_struct_check ((var), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2534, __FUNCTION__))->typed.type)); |
| 2535 | SET_DECL_ALIGN (field, DECL_ALIGN (var))(((contains_struct_check ((field), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2535, __FUNCTION__))->decl_common.align) = ffs_hwi ((((contains_struct_check ((var), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2535, __FUNCTION__))->decl_common.align) ? ((unsigned)1) << (((contains_struct_check ((var), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2535, __FUNCTION__))->decl_common.align) - 1) : 0))); |
| 2536 | DECL_USER_ALIGN (field)((contains_struct_check ((field), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2536, __FUNCTION__))->base.u.bits.user_align) = DECL_USER_ALIGN (var)((contains_struct_check ((var), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2536, __FUNCTION__))->base.u.bits.user_align); |
| 2537 | TREE_THIS_VOLATILE (field)((field)->base.volatile_flag) = TREE_THIS_VOLATILE (var)((var)->base.volatile_flag); |
| 2538 | |
| 2539 | insert_field_into_struct (rectype, field); |
| 2540 | |
| 2541 | tree t = build_simple_mem_ref (simtrec)build_simple_mem_ref_loc (((location_t) 0), simtrec); |
| 2542 | t = build3 (COMPONENT_REF, TREE_TYPE (var)((contains_struct_check ((var), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2542, __FUNCTION__))->typed.type), t, field, NULL__null); |
| 2543 | TREE_THIS_VOLATILE (t)((t)->base.volatile_flag) = TREE_THIS_VOLATILE (var)((var)->base.volatile_flag); |
| 2544 | SET_DECL_VALUE_EXPR (var, t)(decl_value_expr_insert ((contains_struct_check ((var), (TS_DECL_WRTL ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2544, __FUNCTION__)), t)); |
| 2545 | DECL_HAS_VALUE_EXPR_P (var)((tree_check3 ((var), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2545, __FUNCTION__, (VAR_DECL), (PARM_DECL), (RESULT_DECL)) ) ->decl_common.decl_flag_2) = 1; |
| 2546 | *regimplify = true; |
| 2547 | } |
| 2548 | layout_type (rectype); |
| 2549 | tree size = TYPE_SIZE_UNIT (rectype)((tree_class_check ((rectype), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2549, __FUNCTION__))->type_common.size_unit); |
| 2550 | tree align = build_int_cst (TREE_TYPE (size)((contains_struct_check ((size), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2550, __FUNCTION__))->typed.type), TYPE_ALIGN_UNIT (rectype)((((tree_class_check ((rectype), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2550, __FUNCTION__))->type_common.align) ? ((unsigned)1) << (((tree_class_check ((rectype), (tcc_type), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2550, __FUNCTION__))->type_common.align) - 1) : 0) / (8) )); |
| 2551 | |
| 2552 | alloc_stmt |
| 2553 | = gimple_build_call_internal (IFN_GOMP_SIMT_ENTER_ALLOC, 2, size, align); |
| 2554 | gimple_call_set_lhs (alloc_stmt, simtrec); |
| 2555 | gsi_replace (gsi, alloc_stmt, false); |
| 2556 | gimple_stmt_iterator enter_gsi = gsi_for_stmt (enter_stmt); |
| 2557 | enter_stmt = gimple_build_assign (simduid, gimple_call_arg (enter_stmt, 0)); |
| 2558 | gsi_replace (&enter_gsi, enter_stmt, false); |
| 2559 | |
| 2560 | use_operand_p use; |
| 2561 | gimple *exit_stmt; |
| 2562 | if (single_imm_use (simtrec, &use, &exit_stmt)) |
| 2563 | { |
| 2564 | gcc_assert (gimple_call_internal_p (exit_stmt, IFN_GOMP_SIMT_EXIT))((void)(!(gimple_call_internal_p (exit_stmt, IFN_GOMP_SIMT_EXIT )) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2564, __FUNCTION__), 0 : 0)); |
| 2565 | gimple_stmt_iterator exit_gsi = gsi_for_stmt (exit_stmt); |
| 2566 | tree clobber = build_clobber (rectype); |
| 2567 | exit_stmt = gimple_build_assign (build_simple_mem_ref (simtrec)build_simple_mem_ref_loc (((location_t) 0), simtrec), clobber); |
| 2568 | gsi_insert_before (&exit_gsi, exit_stmt, GSI_SAME_STMT); |
| 2569 | } |
| 2570 | else |
| 2571 | gcc_checking_assert (has_zero_uses (simtrec))((void)(!(has_zero_uses (simtrec)) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2571, __FUNCTION__), 0 : 0)); |
| 2572 | } |
| 2573 | |
| 2574 | /* Callback for walk_gimple_stmt used to scan for SIMT-privatized variables. */ |
| 2575 | |
| 2576 | static tree |
| 2577 | find_simtpriv_var_op (tree *tp, int *walk_subtrees, void *) |
| 2578 | { |
| 2579 | tree t = *tp; |
| 2580 | |
| 2581 | if (VAR_P (t)(((enum tree_code) (t)->base.code) == VAR_DECL) |
| 2582 | && DECL_HAS_VALUE_EXPR_P (t)((tree_check3 ((t), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2582, __FUNCTION__, (VAR_DECL), (PARM_DECL), (RESULT_DECL)) ) ->decl_common.decl_flag_2) |
| 2583 | && lookup_attribute ("omp simt private", DECL_ATTRIBUTES (t)((contains_struct_check ((t), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2583, __FUNCTION__))->decl_common.attributes))) |
| 2584 | { |
| 2585 | *walk_subtrees = 0; |
| 2586 | return t; |
| 2587 | } |
| 2588 | return NULL_TREE(tree) __null; |
| 2589 | } |
| 2590 | |
| 2591 | /* Cleanup uses of SIMT placeholder internal functions: on non-SIMT targets, |
| 2592 | VF is 1 and LANE is 0; on SIMT targets, VF is folded to a constant, and |
| 2593 | LANE is kept to be expanded to RTL later on. Also cleanup all other SIMT |
| 2594 | internal functions on non-SIMT targets, and likewise some SIMD internal |
| 2595 | functions on SIMT targets. */ |
| 2596 | |
| 2597 | static unsigned int |
| 2598 | execute_omp_device_lower () |
| 2599 | { |
| 2600 | int vf = targetm.simt.vf ? targetm.simt.vf () : 1; |
| 2601 | bool regimplify = false; |
| 2602 | basic_block bb; |
| 2603 | gimple_stmt_iterator gsi; |
| 2604 | bool calls_declare_variant_alt |
| 2605 | = cgraph_node::get (cfun(cfun + 0)->decl)->calls_declare_variant_alt; |
| 2606 | FOR_EACH_BB_FN (bb, cfun)for (bb = ((cfun + 0))->cfg->x_entry_block_ptr->next_bb ; bb != ((cfun + 0))->cfg->x_exit_block_ptr; bb = bb-> next_bb) |
| 2607 | for (gsi = gsi_start_bb (bb); !gsi_end_p (gsi); gsi_next (&gsi)) |
| 2608 | { |
| 2609 | gimple *stmt = gsi_stmt (gsi); |
| 2610 | if (!is_gimple_call (stmt)) |
| 2611 | continue; |
| 2612 | if (!gimple_call_internal_p (stmt)) |
| 2613 | { |
| 2614 | if (calls_declare_variant_alt) |
| 2615 | if (tree fndecl = gimple_call_fndecl (stmt)) |
| 2616 | { |
| 2617 | tree new_fndecl = omp_resolve_declare_variant (fndecl); |
| 2618 | if (new_fndecl != fndecl) |
| 2619 | { |
| 2620 | gimple_call_set_fndecl (stmt, new_fndecl); |
| 2621 | update_stmt (stmt); |
| 2622 | } |
| 2623 | } |
| 2624 | continue; |
| 2625 | } |
| 2626 | tree lhs = gimple_call_lhs (stmt), rhs = NULL_TREE(tree) __null; |
| 2627 | tree type = lhs ? TREE_TYPE (lhs)((contains_struct_check ((lhs), (TS_TYPED), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2627, __FUNCTION__))->typed.type) : integer_type_nodeinteger_types[itk_int]; |
| 2628 | switch (gimple_call_internal_fn (stmt)) |
| 2629 | { |
| 2630 | case IFN_GOMP_TARGET_REV: |
| 2631 | { |
| 2632 | #ifndef ACCEL_COMPILER |
| 2633 | gimple_stmt_iterator gsi2 = gsi; |
| 2634 | gsi_next (&gsi2); |
| 2635 | gcc_assert (!gsi_end_p (gsi2))((void)(!(!gsi_end_p (gsi2)) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2635, __FUNCTION__), 0 : 0)); |
| 2636 | gcc_assert (gimple_call_builtin_p (gsi_stmt (gsi2),((void)(!(gimple_call_builtin_p (gsi_stmt (gsi2), BUILT_IN_GOMP_TARGET )) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2637, __FUNCTION__), 0 : 0)) |
| 2637 | BUILT_IN_GOMP_TARGET))((void)(!(gimple_call_builtin_p (gsi_stmt (gsi2), BUILT_IN_GOMP_TARGET )) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2637, __FUNCTION__), 0 : 0)); |
| 2638 | tree old_decl |
| 2639 | = TREE_OPERAND (gimple_call_arg (gsi_stmt (gsi2), 1), 0)(*((const_cast<tree*> (tree_operand_check ((gimple_call_arg (gsi_stmt (gsi2), 1)), (0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2639, __FUNCTION__))))); |
| 2640 | tree new_decl = gimple_call_arg (gsi_stmt (gsi), 0); |
| 2641 | gimple_call_set_arg (gsi_stmt (gsi2), 1, new_decl); |
| 2642 | update_stmt (gsi_stmt (gsi2)); |
| 2643 | new_decl = TREE_OPERAND (new_decl, 0)(*((const_cast<tree*> (tree_operand_check ((new_decl), ( 0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2643, __FUNCTION__))))); |
| 2644 | unsigned i; |
| 2645 | unsigned num_funcs = vec_safe_length (offload_funcs); |
| 2646 | for (i = 0; i < num_funcs; i++) |
| 2647 | { |
| 2648 | if ((*offload_funcs)[i] == old_decl) |
| 2649 | { |
| 2650 | (*offload_funcs)[i] = new_decl; |
| 2651 | break; |
| 2652 | } |
| 2653 | else if ((*offload_funcs)[i] == new_decl) |
| 2654 | break; /* This can happen due to inlining. */ |
| 2655 | } |
| 2656 | gcc_assert (i < num_funcs)((void)(!(i < num_funcs) ? fancy_abort ("/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2656, __FUNCTION__), 0 : 0)); |
| 2657 | #else |
| 2658 | tree old_decl = TREE_OPERAND (gimple_call_arg (gsi_stmt (gsi), 0),(*((const_cast<tree*> (tree_operand_check ((gimple_call_arg (gsi_stmt (gsi), 0)), (0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2659, __FUNCTION__))))) |
| 2659 | 0)(*((const_cast<tree*> (tree_operand_check ((gimple_call_arg (gsi_stmt (gsi), 0)), (0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2659, __FUNCTION__))))); |
| 2660 | #endif |
| 2661 | /* FIXME: Find a way to actually prevent outputting the empty-body |
| 2662 | old_decl as debug symbol + function in the assembly file. */ |
| 2663 | cgraph_node *node = cgraph_node::get (old_decl); |
| 2664 | node->address_taken = false; |
| 2665 | node->need_lto_streaming = false; |
| 2666 | node->offloadable = false; |
| 2667 | |
| 2668 | unlink_stmt_vdef (stmt); |
| 2669 | } |
| 2670 | break; |
| 2671 | case IFN_GOMP_USE_SIMT: |
| 2672 | rhs = vf == 1 ? integer_zero_nodeglobal_trees[TI_INTEGER_ZERO] : integer_one_nodeglobal_trees[TI_INTEGER_ONE]; |
| 2673 | break; |
| 2674 | case IFN_GOMP_SIMT_ENTER: |
| 2675 | rhs = vf == 1 ? gimple_call_arg (stmt, 0) : NULL_TREE(tree) __null; |
| 2676 | goto simtreg_enter_exit; |
| 2677 | case IFN_GOMP_SIMT_ENTER_ALLOC: |
| 2678 | if (vf != 1) |
| 2679 | ompdevlow_adjust_simt_enter (&gsi, ®implify); |
| 2680 | rhs = vf == 1 ? null_pointer_nodeglobal_trees[TI_NULL_POINTER] : NULL_TREE(tree) __null; |
| 2681 | goto simtreg_enter_exit; |
| 2682 | case IFN_GOMP_SIMT_EXIT: |
| 2683 | simtreg_enter_exit: |
| 2684 | if (vf != 1) |
| 2685 | continue; |
| 2686 | unlink_stmt_vdef (stmt); |
| 2687 | break; |
| 2688 | case IFN_GOMP_SIMT_LANE: |
| 2689 | case IFN_GOMP_SIMT_LAST_LANE: |
| 2690 | rhs = vf == 1 ? build_zero_cst (type) : NULL_TREE(tree) __null; |
| 2691 | break; |
| 2692 | case IFN_GOMP_SIMT_VF: |
| 2693 | rhs = build_int_cst (type, vf); |
| 2694 | break; |
| 2695 | case IFN_GOMP_SIMT_ORDERED_PRED: |
| 2696 | rhs = vf == 1 ? integer_zero_nodeglobal_trees[TI_INTEGER_ZERO] : NULL_TREE(tree) __null; |
| 2697 | if (rhs || !lhs) |
| 2698 | unlink_stmt_vdef (stmt); |
| 2699 | break; |
| 2700 | case IFN_GOMP_SIMT_VOTE_ANY: |
| 2701 | case IFN_GOMP_SIMT_XCHG_BFLY: |
| 2702 | case IFN_GOMP_SIMT_XCHG_IDX: |
| 2703 | rhs = vf == 1 ? gimple_call_arg (stmt, 0) : NULL_TREE(tree) __null; |
| 2704 | break; |
| 2705 | case IFN_GOMP_SIMD_LANE: |
| 2706 | case IFN_GOMP_SIMD_LAST_LANE: |
| 2707 | rhs = vf != 1 ? build_zero_cst (type) : NULL_TREE(tree) __null; |
| 2708 | break; |
| 2709 | case IFN_GOMP_SIMD_VF: |
| 2710 | rhs = vf != 1 ? build_one_cst (type) : NULL_TREE(tree) __null; |
| 2711 | break; |
| 2712 | default: |
| 2713 | continue; |
| 2714 | } |
| 2715 | if (lhs && !rhs) |
| 2716 | continue; |
| 2717 | stmt = lhs ? gimple_build_assign (lhs, rhs) : gimple_build_nop (); |
| 2718 | gsi_replace (&gsi, stmt, false); |
| 2719 | } |
| 2720 | if (regimplify) |
| 2721 | FOR_EACH_BB_REVERSE_FN (bb, cfun)for (bb = ((cfun + 0))->cfg->x_exit_block_ptr->prev_bb ; bb != ((cfun + 0))->cfg->x_entry_block_ptr; bb = bb-> prev_bb) |
| 2722 | for (gsi = gsi_last_bb (bb); !gsi_end_p (gsi); gsi_prev (&gsi)) |
| 2723 | if (walk_gimple_stmt (&gsi, NULL__null, find_simtpriv_var_op, NULL__null)) |
| 2724 | { |
| 2725 | if (gimple_clobber_p (gsi_stmt (gsi))) |
| 2726 | gsi_remove (&gsi, true); |
| 2727 | else |
| 2728 | gimple_regimplify_operands (gsi_stmt (gsi), &gsi); |
| 2729 | } |
| 2730 | if (vf != 1) |
| 2731 | cfun(cfun + 0)->has_force_vectorize_loops = false; |
| 2732 | return 0; |
| 2733 | } |
| 2734 | |
| 2735 | namespace { |
| 2736 | |
| 2737 | const pass_data pass_data_omp_device_lower = |
| 2738 | { |
| 2739 | GIMPLE_PASS, /* type */ |
| 2740 | "ompdevlow", /* name */ |
| 2741 | OPTGROUP_OMP, /* optinfo_flags */ |
| 2742 | TV_NONE, /* tv_id */ |
| 2743 | PROP_cfg(1 << 3), /* properties_required */ |
| 2744 | PROP_gimple_lomp_dev(1 << 16), /* properties_provided */ |
| 2745 | 0, /* properties_destroyed */ |
| 2746 | 0, /* todo_flags_start */ |
| 2747 | TODO_update_ssa(1 << 11), /* todo_flags_finish */ |
| 2748 | }; |
| 2749 | |
| 2750 | class pass_omp_device_lower : public gimple_opt_pass |
| 2751 | { |
| 2752 | public: |
| 2753 | pass_omp_device_lower (gcc::context *ctxt) |
| 2754 | : gimple_opt_pass (pass_data_omp_device_lower, ctxt) |
| 2755 | {} |
| 2756 | |
| 2757 | /* opt_pass methods: */ |
| 2758 | bool gate (function *fun) final override |
| 2759 | { |
| 2760 | return (!(fun->curr_properties & PROP_gimple_lomp_dev(1 << 16)) |
| 2761 | || (flag_openmpglobal_options.x_flag_openmp |
| 2762 | && cgraph_node::get (fun->decl)->calls_declare_variant_alt)); |
| 2763 | } |
| 2764 | unsigned int execute (function *) final override |
| 2765 | { |
| 2766 | return execute_omp_device_lower (); |
| 2767 | } |
| 2768 | |
| 2769 | }; // class pass_expand_omp_ssa |
| 2770 | |
| 2771 | } // anon namespace |
| 2772 | |
| 2773 | gimple_opt_pass * |
| 2774 | make_pass_omp_device_lower (gcc::context *ctxt) |
| 2775 | { |
| 2776 | return new pass_omp_device_lower (ctxt); |
| 2777 | } |
| 2778 | |
| 2779 | /* "omp declare target link" handling pass. */ |
| 2780 | |
| 2781 | namespace { |
| 2782 | |
| 2783 | const pass_data pass_data_omp_target_link = |
| 2784 | { |
| 2785 | GIMPLE_PASS, /* type */ |
| 2786 | "omptargetlink", /* name */ |
| 2787 | OPTGROUP_OMP, /* optinfo_flags */ |
| 2788 | TV_NONE, /* tv_id */ |
| 2789 | PROP_ssa(1 << 5), /* properties_required */ |
| 2790 | 0, /* properties_provided */ |
| 2791 | 0, /* properties_destroyed */ |
| 2792 | 0, /* todo_flags_start */ |
| 2793 | TODO_update_ssa(1 << 11), /* todo_flags_finish */ |
| 2794 | }; |
| 2795 | |
| 2796 | class pass_omp_target_link : public gimple_opt_pass |
| 2797 | { |
| 2798 | public: |
| 2799 | pass_omp_target_link (gcc::context *ctxt) |
| 2800 | : gimple_opt_pass (pass_data_omp_target_link, ctxt) |
| 2801 | {} |
| 2802 | |
| 2803 | /* opt_pass methods: */ |
| 2804 | bool gate (function *fun) final override |
| 2805 | { |
| 2806 | #ifdef ACCEL_COMPILER |
| 2807 | return offloading_function_p (fun->decl); |
| 2808 | #else |
| 2809 | (void) fun; |
| 2810 | return false; |
| 2811 | #endif |
| 2812 | } |
| 2813 | |
| 2814 | unsigned execute (function *) final override; |
| 2815 | }; |
| 2816 | |
| 2817 | /* Callback for walk_gimple_stmt used to scan for link var operands. */ |
| 2818 | |
| 2819 | static tree |
| 2820 | find_link_var_op (tree *tp, int *walk_subtrees, void *) |
| 2821 | { |
| 2822 | tree t = *tp; |
| 2823 | |
| 2824 | if (VAR_P (t)(((enum tree_code) (t)->base.code) == VAR_DECL) |
| 2825 | && DECL_HAS_VALUE_EXPR_P (t)((tree_check3 ((t), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2825, __FUNCTION__, (VAR_DECL), (PARM_DECL), (RESULT_DECL)) ) ->decl_common.decl_flag_2) |
| 2826 | && is_global_var (t) |
| 2827 | && lookup_attribute ("omp declare target link", DECL_ATTRIBUTES (t)((contains_struct_check ((t), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2827, __FUNCTION__))->decl_common.attributes))) |
| 2828 | { |
| 2829 | *walk_subtrees = 0; |
| 2830 | return t; |
| 2831 | } |
| 2832 | |
| 2833 | return NULL_TREE(tree) __null; |
| 2834 | } |
| 2835 | |
| 2836 | unsigned |
| 2837 | pass_omp_target_link::execute (function *fun) |
| 2838 | { |
| 2839 | basic_block bb; |
| 2840 | FOR_EACH_BB_FN (bb, fun)for (bb = (fun)->cfg->x_entry_block_ptr->next_bb; bb != (fun)->cfg->x_exit_block_ptr; bb = bb->next_bb) |
| 2841 | { |
| 2842 | gimple_stmt_iterator gsi; |
| 2843 | for (gsi = gsi_start_bb (bb); !gsi_end_p (gsi); gsi_next (&gsi)) |
| 2844 | { |
| 2845 | if (gimple_call_builtin_p (gsi_stmt (gsi), BUILT_IN_GOMP_TARGET)) |
| 2846 | { |
| 2847 | tree dev = gimple_call_arg (gsi_stmt (gsi), 0); |
| 2848 | tree fn = gimple_call_arg (gsi_stmt (gsi), 1); |
| 2849 | if (POINTER_TYPE_P (TREE_TYPE (fn))(((enum tree_code) (((contains_struct_check ((fn), (TS_TYPED) , "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2849, __FUNCTION__))->typed.type))->base.code) == POINTER_TYPE || ((enum tree_code) (((contains_struct_check ((fn), (TS_TYPED ), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2849, __FUNCTION__))->typed.type))->base.code) == REFERENCE_TYPE )) |
| 2850 | fn = TREE_OPERAND (fn, 0)(*((const_cast<tree*> (tree_operand_check ((fn), (0), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2850, __FUNCTION__))))); |
| 2851 | if (TREE_CODE (dev)((enum tree_code) (dev)->base.code) == INTEGER_CST |
| 2852 | && wi::to_wide (dev) == GOMP_DEVICE_HOST_FALLBACK-2 |
| 2853 | && lookup_attribute ("omp target device_ancestor_nohost", |
| 2854 | DECL_ATTRIBUTES (fn)((contains_struct_check ((fn), (TS_DECL_COMMON), "/buildworker/marxinbox-gcc-clang-static-analyzer/build/gcc/omp-offload.cc" , 2854, __FUNCTION__))->decl_common.attributes)) != NULL_TREE(tree) __null) |
| 2855 | continue; /* ancestor:1 */ |
| 2856 | /* Nullify the second argument of __builtin_GOMP_target_ext. */ |
| 2857 | gimple_call_set_arg (gsi_stmt (gsi), 1, null_pointer_nodeglobal_trees[TI_NULL_POINTER]); |
| 2858 | update_stmt (gsi_stmt (gsi)); |
| 2859 | } |
| 2860 | if (walk_gimple_stmt (&gsi, NULL__null, find_link_var_op, NULL__null)) |
| 2861 | gimple_regimplify_operands (gsi_stmt (gsi), &gsi); |
| 2862 | } |
| 2863 | } |
| 2864 | |
| 2865 | return 0; |
| 2866 | } |
| 2867 | |
| 2868 | } // anon namespace |
| 2869 | |
| 2870 | gimple_opt_pass * |
| 2871 | make_pass_omp_target_link (gcc::context *ctxt) |
| 2872 | { |
| 2873 | return new pass_omp_target_link (ctxt); |
| 2874 | } |