diff options
Diffstat (limited to 'gcc-4.9/gcc/c-family/c-omp.c')
-rw-r--r-- | gcc-4.9/gcc/c-family/c-omp.c | 1028 |
1 files changed, 1028 insertions, 0 deletions
diff --git a/gcc-4.9/gcc/c-family/c-omp.c b/gcc-4.9/gcc/c-family/c-omp.c new file mode 100644 index 000000000..dd0a45d96 --- /dev/null +++ b/gcc-4.9/gcc/c-family/c-omp.c @@ -0,0 +1,1028 @@ +/* This file contains routines to construct GNU OpenMP constructs, + called from parsing in the C and C++ front ends. + + Copyright (C) 2005-2014 Free Software Foundation, Inc. + Contributed by Richard Henderson <rth@redhat.com>, + Diego Novillo <dnovillo@redhat.com>. + +This file is part of GCC. + +GCC is free software; you can redistribute it and/or modify it under +the terms of the GNU General Public License as published by the Free +Software Foundation; either version 3, or (at your option) any later +version. + +GCC is distributed in the hope that it will be useful, but WITHOUT ANY +WARRANTY; without even the implied warranty of MERCHANTABILITY or +FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License +for more details. + +You should have received a copy of the GNU General Public License +along with GCC; see the file COPYING3. If not see +<http://www.gnu.org/licenses/>. */ + +#include "config.h" +#include "system.h" +#include "coretypes.h" +#include "tree.h" +#include "c-common.h" +#include "c-pragma.h" +#include "gimple-expr.h" +#include "langhooks.h" + + +/* Complete a #pragma omp master construct. STMT is the structured-block + that follows the pragma. LOC is the l*/ + +tree +c_finish_omp_master (location_t loc, tree stmt) +{ + tree t = add_stmt (build1 (OMP_MASTER, void_type_node, stmt)); + SET_EXPR_LOCATION (t, loc); + return t; +} + +/* Complete a #pragma omp taskgroup construct. STMT is the structured-block + that follows the pragma. LOC is the l*/ + +tree +c_finish_omp_taskgroup (location_t loc, tree stmt) +{ + tree t = add_stmt (build1 (OMP_TASKGROUP, void_type_node, stmt)); + SET_EXPR_LOCATION (t, loc); + return t; +} + +/* Complete a #pragma omp critical construct. STMT is the structured-block + that follows the pragma, NAME is the identifier in the pragma, or null + if it was omitted. LOC is the location of the #pragma. */ + +tree +c_finish_omp_critical (location_t loc, tree body, tree name) +{ + tree stmt = make_node (OMP_CRITICAL); + TREE_TYPE (stmt) = void_type_node; + OMP_CRITICAL_BODY (stmt) = body; + OMP_CRITICAL_NAME (stmt) = name; + SET_EXPR_LOCATION (stmt, loc); + return add_stmt (stmt); +} + +/* Complete a #pragma omp ordered construct. STMT is the structured-block + that follows the pragma. LOC is the location of the #pragma. */ + +tree +c_finish_omp_ordered (location_t loc, tree stmt) +{ + tree t = build1 (OMP_ORDERED, void_type_node, stmt); + SET_EXPR_LOCATION (t, loc); + return add_stmt (t); +} + + +/* Complete a #pragma omp barrier construct. LOC is the location of + the #pragma. */ + +void +c_finish_omp_barrier (location_t loc) +{ + tree x; + + x = builtin_decl_explicit (BUILT_IN_GOMP_BARRIER); + x = build_call_expr_loc (loc, x, 0); + add_stmt (x); +} + + +/* Complete a #pragma omp taskwait construct. LOC is the location of the + pragma. */ + +void +c_finish_omp_taskwait (location_t loc) +{ + tree x; + + x = builtin_decl_explicit (BUILT_IN_GOMP_TASKWAIT); + x = build_call_expr_loc (loc, x, 0); + add_stmt (x); +} + + +/* Complete a #pragma omp taskyield construct. LOC is the location of the + pragma. */ + +void +c_finish_omp_taskyield (location_t loc) +{ + tree x; + + x = builtin_decl_explicit (BUILT_IN_GOMP_TASKYIELD); + x = build_call_expr_loc (loc, x, 0); + add_stmt (x); +} + + +/* Complete a #pragma omp atomic construct. For CODE OMP_ATOMIC + the expression to be implemented atomically is LHS opcode= RHS. + For OMP_ATOMIC_READ V = LHS, for OMP_ATOMIC_CAPTURE_{NEW,OLD} LHS + opcode= RHS with the new or old content of LHS returned. + LOC is the location of the atomic statement. The value returned + is either error_mark_node (if the construct was erroneous) or an + OMP_ATOMIC* node which should be added to the current statement + tree with add_stmt. */ + +tree +c_finish_omp_atomic (location_t loc, enum tree_code code, + enum tree_code opcode, tree lhs, tree rhs, + tree v, tree lhs1, tree rhs1, bool swapped, bool seq_cst) +{ + tree x, type, addr, pre = NULL_TREE; + + if (lhs == error_mark_node || rhs == error_mark_node + || v == error_mark_node || lhs1 == error_mark_node + || rhs1 == error_mark_node) + return error_mark_node; + + /* ??? According to one reading of the OpenMP spec, complex type are + supported, but there are no atomic stores for any architecture. + But at least icc 9.0 doesn't support complex types here either. + And lets not even talk about vector types... */ + type = TREE_TYPE (lhs); + if (!INTEGRAL_TYPE_P (type) + && !POINTER_TYPE_P (type) + && !SCALAR_FLOAT_TYPE_P (type)) + { + error_at (loc, "invalid expression type for %<#pragma omp atomic%>"); + return error_mark_node; + } + + /* ??? Validate that rhs does not overlap lhs. */ + + /* Take and save the address of the lhs. From then on we'll reference it + via indirection. */ + addr = build_unary_op (loc, ADDR_EXPR, lhs, 0); + if (addr == error_mark_node) + return error_mark_node; + addr = save_expr (addr); + if (TREE_CODE (addr) != SAVE_EXPR + && (TREE_CODE (addr) != ADDR_EXPR + || TREE_CODE (TREE_OPERAND (addr, 0)) != VAR_DECL)) + { + /* Make sure LHS is simple enough so that goa_lhs_expr_p can recognize + it even after unsharing function body. */ + tree var = create_tmp_var_raw (TREE_TYPE (addr), NULL); + DECL_CONTEXT (var) = current_function_decl; + addr = build4 (TARGET_EXPR, TREE_TYPE (addr), var, addr, NULL, NULL); + } + lhs = build_indirect_ref (loc, addr, RO_NULL); + + if (code == OMP_ATOMIC_READ) + { + x = build1 (OMP_ATOMIC_READ, type, addr); + SET_EXPR_LOCATION (x, loc); + OMP_ATOMIC_SEQ_CST (x) = seq_cst; + return build_modify_expr (loc, v, NULL_TREE, NOP_EXPR, + loc, x, NULL_TREE); + } + + /* There are lots of warnings, errors, and conversions that need to happen + in the course of interpreting a statement. Use the normal mechanisms + to do this, and then take it apart again. */ + if (swapped) + { + rhs = build2_loc (loc, opcode, TREE_TYPE (lhs), rhs, lhs); + opcode = NOP_EXPR; + } + bool save = in_late_binary_op; + in_late_binary_op = true; + x = build_modify_expr (loc, lhs, NULL_TREE, opcode, loc, rhs, NULL_TREE); + in_late_binary_op = save; + if (x == error_mark_node) + return error_mark_node; + if (TREE_CODE (x) == COMPOUND_EXPR) + { + pre = TREE_OPERAND (x, 0); + gcc_assert (TREE_CODE (pre) == SAVE_EXPR); + x = TREE_OPERAND (x, 1); + } + gcc_assert (TREE_CODE (x) == MODIFY_EXPR); + rhs = TREE_OPERAND (x, 1); + + /* Punt the actual generation of atomic operations to common code. */ + if (code == OMP_ATOMIC) + type = void_type_node; + x = build2 (code, type, addr, rhs); + SET_EXPR_LOCATION (x, loc); + OMP_ATOMIC_SEQ_CST (x) = seq_cst; + + /* Generally it is hard to prove lhs1 and lhs are the same memory + location, just diagnose different variables. */ + if (rhs1 + && TREE_CODE (rhs1) == VAR_DECL + && TREE_CODE (lhs) == VAR_DECL + && rhs1 != lhs) + { + if (code == OMP_ATOMIC) + error_at (loc, "%<#pragma omp atomic update%> uses two different variables for memory"); + else + error_at (loc, "%<#pragma omp atomic capture%> uses two different variables for memory"); + return error_mark_node; + } + + if (code != OMP_ATOMIC) + { + /* Generally it is hard to prove lhs1 and lhs are the same memory + location, just diagnose different variables. */ + if (lhs1 && TREE_CODE (lhs1) == VAR_DECL && TREE_CODE (lhs) == VAR_DECL) + { + if (lhs1 != lhs) + { + error_at (loc, "%<#pragma omp atomic capture%> uses two different variables for memory"); + return error_mark_node; + } + } + x = build_modify_expr (loc, v, NULL_TREE, NOP_EXPR, + loc, x, NULL_TREE); + if (rhs1 && rhs1 != lhs) + { + tree rhs1addr = build_unary_op (loc, ADDR_EXPR, rhs1, 0); + if (rhs1addr == error_mark_node) + return error_mark_node; + x = omit_one_operand_loc (loc, type, x, rhs1addr); + } + if (lhs1 && lhs1 != lhs) + { + tree lhs1addr = build_unary_op (loc, ADDR_EXPR, lhs1, 0); + if (lhs1addr == error_mark_node) + return error_mark_node; + if (code == OMP_ATOMIC_CAPTURE_OLD) + x = omit_one_operand_loc (loc, type, x, lhs1addr); + else + { + x = save_expr (x); + x = omit_two_operands_loc (loc, type, x, x, lhs1addr); + } + } + } + else if (rhs1 && rhs1 != lhs) + { + tree rhs1addr = build_unary_op (loc, ADDR_EXPR, rhs1, 0); + if (rhs1addr == error_mark_node) + return error_mark_node; + x = omit_one_operand_loc (loc, type, x, rhs1addr); + } + + if (pre) + x = omit_one_operand_loc (loc, type, x, pre); + return x; +} + + +/* Complete a #pragma omp flush construct. We don't do anything with + the variable list that the syntax allows. LOC is the location of + the #pragma. */ + +void +c_finish_omp_flush (location_t loc) +{ + tree x; + + x = builtin_decl_explicit (BUILT_IN_SYNC_SYNCHRONIZE); + x = build_call_expr_loc (loc, x, 0); + add_stmt (x); +} + + +/* Check and canonicalize #pragma omp for increment expression. + Helper function for c_finish_omp_for. */ + +static tree +check_omp_for_incr_expr (location_t loc, tree exp, tree decl) +{ + tree t; + + if (!INTEGRAL_TYPE_P (TREE_TYPE (exp)) + || TYPE_PRECISION (TREE_TYPE (exp)) < TYPE_PRECISION (TREE_TYPE (decl))) + return error_mark_node; + + if (exp == decl) + return build_int_cst (TREE_TYPE (exp), 0); + + switch (TREE_CODE (exp)) + { + CASE_CONVERT: + t = check_omp_for_incr_expr (loc, TREE_OPERAND (exp, 0), decl); + if (t != error_mark_node) + return fold_convert_loc (loc, TREE_TYPE (exp), t); + break; + case MINUS_EXPR: + t = check_omp_for_incr_expr (loc, TREE_OPERAND (exp, 0), decl); + if (t != error_mark_node) + return fold_build2_loc (loc, MINUS_EXPR, + TREE_TYPE (exp), t, TREE_OPERAND (exp, 1)); + break; + case PLUS_EXPR: + t = check_omp_for_incr_expr (loc, TREE_OPERAND (exp, 0), decl); + if (t != error_mark_node) + return fold_build2_loc (loc, PLUS_EXPR, + TREE_TYPE (exp), t, TREE_OPERAND (exp, 1)); + t = check_omp_for_incr_expr (loc, TREE_OPERAND (exp, 1), decl); + if (t != error_mark_node) + return fold_build2_loc (loc, PLUS_EXPR, + TREE_TYPE (exp), TREE_OPERAND (exp, 0), t); + break; + case COMPOUND_EXPR: + { + /* cp_build_modify_expr forces preevaluation of the RHS to make + sure that it is evaluated before the lvalue-rvalue conversion + is applied to the LHS. Reconstruct the original expression. */ + tree op0 = TREE_OPERAND (exp, 0); + if (TREE_CODE (op0) == TARGET_EXPR + && !VOID_TYPE_P (TREE_TYPE (op0))) + { + tree op1 = TREE_OPERAND (exp, 1); + tree temp = TARGET_EXPR_SLOT (op0); + if (TREE_CODE_CLASS (TREE_CODE (op1)) == tcc_binary + && TREE_OPERAND (op1, 1) == temp) + { + op1 = copy_node (op1); + TREE_OPERAND (op1, 1) = TARGET_EXPR_INITIAL (op0); + return check_omp_for_incr_expr (loc, op1, decl); + } + } + break; + } + default: + break; + } + + return error_mark_node; +} + +/* If the OMP_FOR increment expression in INCR is of pointer type, + canonicalize it into an expression handled by gimplify_omp_for() + and return it. DECL is the iteration variable. */ + +static tree +c_omp_for_incr_canonicalize_ptr (location_t loc, tree decl, tree incr) +{ + if (POINTER_TYPE_P (TREE_TYPE (decl)) + && TREE_OPERAND (incr, 1)) + { + tree t = fold_convert_loc (loc, + sizetype, TREE_OPERAND (incr, 1)); + + if (TREE_CODE (incr) == POSTDECREMENT_EXPR + || TREE_CODE (incr) == PREDECREMENT_EXPR) + t = fold_build1_loc (loc, NEGATE_EXPR, sizetype, t); + t = fold_build_pointer_plus (decl, t); + incr = build2 (MODIFY_EXPR, void_type_node, decl, t); + } + return incr; +} + +/* Validate and emit code for the OpenMP directive #pragma omp for. + DECLV is a vector of iteration variables, for each collapsed loop. + INITV, CONDV and INCRV are vectors containing initialization + expressions, controlling predicates and increment expressions. + BODY is the body of the loop and PRE_BODY statements that go before + the loop. */ + +tree +c_finish_omp_for (location_t locus, enum tree_code code, tree declv, + tree initv, tree condv, tree incrv, tree body, tree pre_body) +{ + location_t elocus; + bool fail = false; + int i; + + if (code == CILK_SIMD + && !c_check_cilk_loop (locus, TREE_VEC_ELT (declv, 0))) + fail = true; + + gcc_assert (TREE_VEC_LENGTH (declv) == TREE_VEC_LENGTH (initv)); + gcc_assert (TREE_VEC_LENGTH (declv) == TREE_VEC_LENGTH (condv)); + gcc_assert (TREE_VEC_LENGTH (declv) == TREE_VEC_LENGTH (incrv)); + for (i = 0; i < TREE_VEC_LENGTH (declv); i++) + { + tree decl = TREE_VEC_ELT (declv, i); + tree init = TREE_VEC_ELT (initv, i); + tree cond = TREE_VEC_ELT (condv, i); + tree incr = TREE_VEC_ELT (incrv, i); + + elocus = locus; + if (EXPR_HAS_LOCATION (init)) + elocus = EXPR_LOCATION (init); + + /* Validate the iteration variable. */ + if (!INTEGRAL_TYPE_P (TREE_TYPE (decl)) + && TREE_CODE (TREE_TYPE (decl)) != POINTER_TYPE) + { + error_at (elocus, "invalid type for iteration variable %qE", decl); + fail = true; + } + + /* In the case of "for (int i = 0...)", init will be a decl. It should + have a DECL_INITIAL that we can turn into an assignment. */ + if (init == decl) + { + elocus = DECL_SOURCE_LOCATION (decl); + + init = DECL_INITIAL (decl); + if (init == NULL) + { + error_at (elocus, "%qE is not initialized", decl); + init = integer_zero_node; + fail = true; + } + + init = build_modify_expr (elocus, decl, NULL_TREE, NOP_EXPR, + /* FIXME diagnostics: This should + be the location of the INIT. */ + elocus, + init, + NULL_TREE); + } + if (init != error_mark_node) + { + gcc_assert (TREE_CODE (init) == MODIFY_EXPR); + gcc_assert (TREE_OPERAND (init, 0) == decl); + } + + if (cond == NULL_TREE) + { + error_at (elocus, "missing controlling predicate"); + fail = true; + } + else + { + bool cond_ok = false; + + if (EXPR_HAS_LOCATION (cond)) + elocus = EXPR_LOCATION (cond); + + if (TREE_CODE (cond) == LT_EXPR + || TREE_CODE (cond) == LE_EXPR + || TREE_CODE (cond) == GT_EXPR + || TREE_CODE (cond) == GE_EXPR + || TREE_CODE (cond) == NE_EXPR + || TREE_CODE (cond) == EQ_EXPR) + { + tree op0 = TREE_OPERAND (cond, 0); + tree op1 = TREE_OPERAND (cond, 1); + + /* 2.5.1. The comparison in the condition is computed in + the type of DECL, otherwise the behavior is undefined. + + For example: + long n; int i; + i < n; + + according to ISO will be evaluated as: + (long)i < n; + + We want to force: + i < (int)n; */ + if (TREE_CODE (op0) == NOP_EXPR + && decl == TREE_OPERAND (op0, 0)) + { + TREE_OPERAND (cond, 0) = TREE_OPERAND (op0, 0); + TREE_OPERAND (cond, 1) + = fold_build1_loc (elocus, NOP_EXPR, TREE_TYPE (decl), + TREE_OPERAND (cond, 1)); + } + else if (TREE_CODE (op1) == NOP_EXPR + && decl == TREE_OPERAND (op1, 0)) + { + TREE_OPERAND (cond, 1) = TREE_OPERAND (op1, 0); + TREE_OPERAND (cond, 0) + = fold_build1_loc (elocus, NOP_EXPR, TREE_TYPE (decl), + TREE_OPERAND (cond, 0)); + } + + if (decl == TREE_OPERAND (cond, 0)) + cond_ok = true; + else if (decl == TREE_OPERAND (cond, 1)) + { + TREE_SET_CODE (cond, + swap_tree_comparison (TREE_CODE (cond))); + TREE_OPERAND (cond, 1) = TREE_OPERAND (cond, 0); + TREE_OPERAND (cond, 0) = decl; + cond_ok = true; + } + + if (TREE_CODE (cond) == NE_EXPR + || TREE_CODE (cond) == EQ_EXPR) + { + if (!INTEGRAL_TYPE_P (TREE_TYPE (decl))) + cond_ok = false; + else if (operand_equal_p (TREE_OPERAND (cond, 1), + TYPE_MIN_VALUE (TREE_TYPE (decl)), + 0)) + TREE_SET_CODE (cond, TREE_CODE (cond) == NE_EXPR + ? GT_EXPR : LE_EXPR); + else if (operand_equal_p (TREE_OPERAND (cond, 1), + TYPE_MAX_VALUE (TREE_TYPE (decl)), + 0)) + TREE_SET_CODE (cond, TREE_CODE (cond) == NE_EXPR + ? LT_EXPR : GE_EXPR); + else if (code != CILK_SIMD) + cond_ok = false; + } + } + + if (!cond_ok) + { + error_at (elocus, "invalid controlling predicate"); + fail = true; + } + } + + if (incr == NULL_TREE) + { + error_at (elocus, "missing increment expression"); + fail = true; + } + else + { + bool incr_ok = false; + + if (EXPR_HAS_LOCATION (incr)) + elocus = EXPR_LOCATION (incr); + + /* Check all the valid increment expressions: v++, v--, ++v, --v, + v = v + incr, v = incr + v and v = v - incr. */ + switch (TREE_CODE (incr)) + { + case POSTINCREMENT_EXPR: + case PREINCREMENT_EXPR: + case POSTDECREMENT_EXPR: + case PREDECREMENT_EXPR: + if (TREE_OPERAND (incr, 0) != decl) + break; + + incr_ok = true; + incr = c_omp_for_incr_canonicalize_ptr (elocus, decl, incr); + break; + + case COMPOUND_EXPR: + if (TREE_CODE (TREE_OPERAND (incr, 0)) != SAVE_EXPR + || TREE_CODE (TREE_OPERAND (incr, 1)) != MODIFY_EXPR) + break; + incr = TREE_OPERAND (incr, 1); + /* FALLTHRU */ + case MODIFY_EXPR: + if (TREE_OPERAND (incr, 0) != decl) + break; + if (TREE_OPERAND (incr, 1) == decl) + break; + if (TREE_CODE (TREE_OPERAND (incr, 1)) == PLUS_EXPR + && (TREE_OPERAND (TREE_OPERAND (incr, 1), 0) == decl + || TREE_OPERAND (TREE_OPERAND (incr, 1), 1) == decl)) + incr_ok = true; + else if ((TREE_CODE (TREE_OPERAND (incr, 1)) == MINUS_EXPR + || (TREE_CODE (TREE_OPERAND (incr, 1)) + == POINTER_PLUS_EXPR)) + && TREE_OPERAND (TREE_OPERAND (incr, 1), 0) == decl) + incr_ok = true; + else + { + tree t = check_omp_for_incr_expr (elocus, + TREE_OPERAND (incr, 1), + decl); + if (t != error_mark_node) + { + incr_ok = true; + t = build2 (PLUS_EXPR, TREE_TYPE (decl), decl, t); + incr = build2 (MODIFY_EXPR, void_type_node, decl, t); + } + } + break; + + default: + break; + } + if (!incr_ok) + { + error_at (elocus, "invalid increment expression"); + fail = true; + } + } + + TREE_VEC_ELT (initv, i) = init; + TREE_VEC_ELT (incrv, i) = incr; + } + + if (fail) + return NULL; + else + { + tree t = make_node (code); + + TREE_TYPE (t) = void_type_node; + OMP_FOR_INIT (t) = initv; + OMP_FOR_COND (t) = condv; + OMP_FOR_INCR (t) = incrv; + OMP_FOR_BODY (t) = body; + OMP_FOR_PRE_BODY (t) = pre_body; + + SET_EXPR_LOCATION (t, locus); + return add_stmt (t); + } +} + +/* Right now we have 14 different combined constructs, this + function attempts to split or duplicate clauses for combined + constructs. CODE is the innermost construct in the combined construct, + and MASK allows to determine which constructs are combined together, + as every construct has at least one clause that no other construct + has (except for OMP_SECTIONS, but that can be only combined with parallel). + Combined constructs are: + #pragma omp parallel for + #pragma omp parallel sections + #pragma omp parallel for simd + #pragma omp for simd + #pragma omp distribute simd + #pragma omp distribute parallel for + #pragma omp distribute parallel for simd + #pragma omp teams distribute + #pragma omp teams distribute parallel for + #pragma omp teams distribute parallel for simd + #pragma omp target teams + #pragma omp target teams distribute + #pragma omp target teams distribute parallel for + #pragma omp target teams distribute parallel for simd */ + +void +c_omp_split_clauses (location_t loc, enum tree_code code, + omp_clause_mask mask, tree clauses, tree *cclauses) +{ + tree next, c; + enum c_omp_clause_split s; + int i; + + for (i = 0; i < C_OMP_CLAUSE_SPLIT_COUNT; i++) + cclauses[i] = NULL; + /* Add implicit nowait clause on + #pragma omp parallel {for,for simd,sections}. */ + if ((mask & (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_NUM_THREADS)) != 0) + switch (code) + { + case OMP_FOR: + case OMP_SIMD: + cclauses[C_OMP_CLAUSE_SPLIT_FOR] + = build_omp_clause (loc, OMP_CLAUSE_NOWAIT); + break; + case OMP_SECTIONS: + cclauses[C_OMP_CLAUSE_SPLIT_SECTIONS] + = build_omp_clause (loc, OMP_CLAUSE_NOWAIT); + break; + default: + break; + } + + for (; clauses ; clauses = next) + { + next = OMP_CLAUSE_CHAIN (clauses); + + switch (OMP_CLAUSE_CODE (clauses)) + { + /* First the clauses that are unique to some constructs. */ + case OMP_CLAUSE_DEVICE: + case OMP_CLAUSE_MAP: + s = C_OMP_CLAUSE_SPLIT_TARGET; + break; + case OMP_CLAUSE_NUM_TEAMS: + case OMP_CLAUSE_THREAD_LIMIT: + s = C_OMP_CLAUSE_SPLIT_TEAMS; + break; + case OMP_CLAUSE_DIST_SCHEDULE: + s = C_OMP_CLAUSE_SPLIT_DISTRIBUTE; + break; + case OMP_CLAUSE_COPYIN: + case OMP_CLAUSE_NUM_THREADS: + case OMP_CLAUSE_PROC_BIND: + s = C_OMP_CLAUSE_SPLIT_PARALLEL; + break; + case OMP_CLAUSE_ORDERED: + case OMP_CLAUSE_SCHEDULE: + case OMP_CLAUSE_NOWAIT: + s = C_OMP_CLAUSE_SPLIT_FOR; + break; + case OMP_CLAUSE_SAFELEN: + case OMP_CLAUSE_LINEAR: + case OMP_CLAUSE_ALIGNED: + s = C_OMP_CLAUSE_SPLIT_SIMD; + break; + /* Duplicate this to all of distribute, for and simd. */ + case OMP_CLAUSE_COLLAPSE: + if (code == OMP_SIMD) + { + c = build_omp_clause (OMP_CLAUSE_LOCATION (clauses), + OMP_CLAUSE_COLLAPSE); + OMP_CLAUSE_COLLAPSE_EXPR (c) + = OMP_CLAUSE_COLLAPSE_EXPR (clauses); + OMP_CLAUSE_CHAIN (c) = cclauses[C_OMP_CLAUSE_SPLIT_SIMD]; + cclauses[C_OMP_CLAUSE_SPLIT_SIMD] = c; + } + if ((mask & (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_SCHEDULE)) != 0) + { + if ((mask & (OMP_CLAUSE_MASK_1 + << PRAGMA_OMP_CLAUSE_DIST_SCHEDULE)) != 0) + { + c = build_omp_clause (OMP_CLAUSE_LOCATION (clauses), + OMP_CLAUSE_COLLAPSE); + OMP_CLAUSE_COLLAPSE_EXPR (c) + = OMP_CLAUSE_COLLAPSE_EXPR (clauses); + OMP_CLAUSE_CHAIN (c) = cclauses[C_OMP_CLAUSE_SPLIT_FOR]; + cclauses[C_OMP_CLAUSE_SPLIT_FOR] = c; + s = C_OMP_CLAUSE_SPLIT_DISTRIBUTE; + } + else + s = C_OMP_CLAUSE_SPLIT_FOR; + } + else + s = C_OMP_CLAUSE_SPLIT_DISTRIBUTE; + break; + /* Private clause is supported on all constructs but target, + it is enough to put it on the innermost one. For + #pragma omp {for,sections} put it on parallel though, + as that's what we did for OpenMP 3.1. */ + case OMP_CLAUSE_PRIVATE: + switch (code) + { + case OMP_SIMD: s = C_OMP_CLAUSE_SPLIT_SIMD; break; + case OMP_FOR: case OMP_SECTIONS: + case OMP_PARALLEL: s = C_OMP_CLAUSE_SPLIT_PARALLEL; break; + case OMP_DISTRIBUTE: s = C_OMP_CLAUSE_SPLIT_DISTRIBUTE; break; + case OMP_TEAMS: s = C_OMP_CLAUSE_SPLIT_TEAMS; break; + default: gcc_unreachable (); + } + break; + /* Firstprivate clause is supported on all constructs but + target and simd. Put it on the outermost of those and + duplicate on parallel. */ + case OMP_CLAUSE_FIRSTPRIVATE: + if ((mask & (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_NUM_THREADS)) + != 0) + { + if ((mask & ((OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_NUM_TEAMS) + | (OMP_CLAUSE_MASK_1 + << PRAGMA_OMP_CLAUSE_DIST_SCHEDULE))) != 0) + { + c = build_omp_clause (OMP_CLAUSE_LOCATION (clauses), + OMP_CLAUSE_FIRSTPRIVATE); + OMP_CLAUSE_DECL (c) = OMP_CLAUSE_DECL (clauses); + OMP_CLAUSE_CHAIN (c) = cclauses[C_OMP_CLAUSE_SPLIT_PARALLEL]; + cclauses[C_OMP_CLAUSE_SPLIT_PARALLEL] = c; + if ((mask & (OMP_CLAUSE_MASK_1 + << PRAGMA_OMP_CLAUSE_NUM_TEAMS)) != 0) + s = C_OMP_CLAUSE_SPLIT_TEAMS; + else + s = C_OMP_CLAUSE_SPLIT_DISTRIBUTE; + } + else + /* This must be + #pragma omp parallel{, for{, simd}, sections}. */ + s = C_OMP_CLAUSE_SPLIT_PARALLEL; + } + else if ((mask & (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_NUM_TEAMS)) + != 0) + { + /* This must be #pragma omp {,target }teams distribute. */ + gcc_assert (code == OMP_DISTRIBUTE); + s = C_OMP_CLAUSE_SPLIT_TEAMS; + } + else if ((mask & (OMP_CLAUSE_MASK_1 + << PRAGMA_OMP_CLAUSE_DIST_SCHEDULE)) != 0) + { + /* This must be #pragma omp distribute simd. */ + gcc_assert (code == OMP_SIMD); + s = C_OMP_CLAUSE_SPLIT_TEAMS; + } + else + { + /* This must be #pragma omp for simd. */ + gcc_assert (code == OMP_SIMD); + s = C_OMP_CLAUSE_SPLIT_FOR; + } + break; + /* Lastprivate is allowed on for, sections and simd. In + parallel {for{, simd},sections} we actually want to put it on + parallel rather than for or sections. */ + case OMP_CLAUSE_LASTPRIVATE: + if (code == OMP_FOR || code == OMP_SECTIONS) + { + if ((mask & (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_NUM_THREADS)) + != 0) + s = C_OMP_CLAUSE_SPLIT_PARALLEL; + else + s = C_OMP_CLAUSE_SPLIT_FOR; + break; + } + gcc_assert (code == OMP_SIMD); + if ((mask & (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_SCHEDULE)) != 0) + { + c = build_omp_clause (OMP_CLAUSE_LOCATION (clauses), + OMP_CLAUSE_LASTPRIVATE); + OMP_CLAUSE_DECL (c) = OMP_CLAUSE_DECL (clauses); + if ((mask & (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_NUM_THREADS)) + != 0) + s = C_OMP_CLAUSE_SPLIT_PARALLEL; + else + s = C_OMP_CLAUSE_SPLIT_FOR; + OMP_CLAUSE_CHAIN (c) = cclauses[s]; + cclauses[s] = c; + } + s = C_OMP_CLAUSE_SPLIT_SIMD; + break; + /* Shared and default clauses are allowed on private and teams. */ + case OMP_CLAUSE_SHARED: + case OMP_CLAUSE_DEFAULT: + if (code == OMP_TEAMS) + { + s = C_OMP_CLAUSE_SPLIT_TEAMS; + break; + } + if ((mask & (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_NUM_TEAMS)) + != 0) + { + c = build_omp_clause (OMP_CLAUSE_LOCATION (clauses), + OMP_CLAUSE_CODE (clauses)); + if (OMP_CLAUSE_CODE (clauses) == OMP_CLAUSE_SHARED) + OMP_CLAUSE_DECL (c) = OMP_CLAUSE_DECL (clauses); + else + OMP_CLAUSE_DEFAULT_KIND (c) + = OMP_CLAUSE_DEFAULT_KIND (clauses); + OMP_CLAUSE_CHAIN (c) = cclauses[C_OMP_CLAUSE_SPLIT_TEAMS]; + cclauses[C_OMP_CLAUSE_SPLIT_TEAMS] = c; + + } + s = C_OMP_CLAUSE_SPLIT_PARALLEL; + break; + /* Reduction is allowed on simd, for, parallel, sections and teams. + Duplicate it on all of them, but omit on for or sections if + parallel is present. */ + case OMP_CLAUSE_REDUCTION: + if (code == OMP_SIMD) + { + c = build_omp_clause (OMP_CLAUSE_LOCATION (clauses), + OMP_CLAUSE_REDUCTION); + OMP_CLAUSE_DECL (c) = OMP_CLAUSE_DECL (clauses); + OMP_CLAUSE_REDUCTION_CODE (c) + = OMP_CLAUSE_REDUCTION_CODE (clauses); + OMP_CLAUSE_REDUCTION_PLACEHOLDER (c) + = OMP_CLAUSE_REDUCTION_PLACEHOLDER (clauses); + OMP_CLAUSE_CHAIN (c) = cclauses[C_OMP_CLAUSE_SPLIT_SIMD]; + cclauses[C_OMP_CLAUSE_SPLIT_SIMD] = c; + } + if ((mask & (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_SCHEDULE)) != 0) + { + if ((mask & (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_NUM_TEAMS)) + != 0) + { + c = build_omp_clause (OMP_CLAUSE_LOCATION (clauses), + OMP_CLAUSE_REDUCTION); + OMP_CLAUSE_DECL (c) = OMP_CLAUSE_DECL (clauses); + OMP_CLAUSE_REDUCTION_CODE (c) + = OMP_CLAUSE_REDUCTION_CODE (clauses); + OMP_CLAUSE_REDUCTION_PLACEHOLDER (c) + = OMP_CLAUSE_REDUCTION_PLACEHOLDER (clauses); + OMP_CLAUSE_CHAIN (c) = cclauses[C_OMP_CLAUSE_SPLIT_PARALLEL]; + cclauses[C_OMP_CLAUSE_SPLIT_PARALLEL] = c; + s = C_OMP_CLAUSE_SPLIT_TEAMS; + } + else if ((mask & (OMP_CLAUSE_MASK_1 + << PRAGMA_OMP_CLAUSE_NUM_THREADS)) != 0) + s = C_OMP_CLAUSE_SPLIT_PARALLEL; + else + s = C_OMP_CLAUSE_SPLIT_FOR; + } + else if (code == OMP_SECTIONS) + s = C_OMP_CLAUSE_SPLIT_PARALLEL; + else + s = C_OMP_CLAUSE_SPLIT_TEAMS; + break; + case OMP_CLAUSE_IF: + /* FIXME: This is currently being discussed. */ + if ((mask & (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_NUM_THREADS)) + != 0) + s = C_OMP_CLAUSE_SPLIT_PARALLEL; + else + s = C_OMP_CLAUSE_SPLIT_TARGET; + break; + default: + gcc_unreachable (); + } + OMP_CLAUSE_CHAIN (clauses) = cclauses[s]; + cclauses[s] = clauses; + } +} + + +/* qsort callback to compare #pragma omp declare simd clauses. */ + +static int +c_omp_declare_simd_clause_cmp (const void *p, const void *q) +{ + tree a = *(const tree *) p; + tree b = *(const tree *) q; + if (OMP_CLAUSE_CODE (a) != OMP_CLAUSE_CODE (b)) + { + if (OMP_CLAUSE_CODE (a) > OMP_CLAUSE_CODE (b)) + return -1; + return 1; + } + if (OMP_CLAUSE_CODE (a) != OMP_CLAUSE_SIMDLEN + && OMP_CLAUSE_CODE (a) != OMP_CLAUSE_INBRANCH + && OMP_CLAUSE_CODE (a) != OMP_CLAUSE_NOTINBRANCH) + { + int c = tree_to_shwi (OMP_CLAUSE_DECL (a)); + int d = tree_to_shwi (OMP_CLAUSE_DECL (b)); + if (c < d) + return 1; + if (c > d) + return -1; + } + return 0; +} + +/* Change PARM_DECLs in OMP_CLAUSE_DECL of #pragma omp declare simd + CLAUSES on FNDECL into argument indexes and sort them. */ + +tree +c_omp_declare_simd_clauses_to_numbers (tree parms, tree clauses) +{ + tree c; + vec<tree> clvec = vNULL; + + for (c = clauses; c; c = OMP_CLAUSE_CHAIN (c)) + { + if (OMP_CLAUSE_CODE (c) != OMP_CLAUSE_SIMDLEN + && OMP_CLAUSE_CODE (c) != OMP_CLAUSE_INBRANCH + && OMP_CLAUSE_CODE (c) != OMP_CLAUSE_NOTINBRANCH) + { + tree decl = OMP_CLAUSE_DECL (c); + tree arg; + int idx; + for (arg = parms, idx = 0; arg; + arg = TREE_CHAIN (arg), idx++) + if (arg == decl) + break; + if (arg == NULL_TREE) + { + error_at (OMP_CLAUSE_LOCATION (c), + "%qD is not an function argument", decl); + continue; + } + OMP_CLAUSE_DECL (c) = build_int_cst (integer_type_node, idx); + } + clvec.safe_push (c); + } + if (!clvec.is_empty ()) + { + unsigned int len = clvec.length (), i; + clvec.qsort (c_omp_declare_simd_clause_cmp); + clauses = clvec[0]; + for (i = 0; i < len; i++) + OMP_CLAUSE_CHAIN (clvec[i]) = (i < len - 1) ? clvec[i + 1] : NULL_TREE; + } + clvec.release (); + return clauses; +} + +/* Change argument indexes in CLAUSES of FNDECL back to PARM_DECLs. */ + +void +c_omp_declare_simd_clauses_to_decls (tree fndecl, tree clauses) +{ + tree c; + + for (c = clauses; c; c = OMP_CLAUSE_CHAIN (c)) + if (OMP_CLAUSE_CODE (c) != OMP_CLAUSE_SIMDLEN + && OMP_CLAUSE_CODE (c) != OMP_CLAUSE_INBRANCH + && OMP_CLAUSE_CODE (c) != OMP_CLAUSE_NOTINBRANCH) + { + int idx = tree_to_shwi (OMP_CLAUSE_DECL (c)), i; + tree arg; + for (arg = DECL_ARGUMENTS (fndecl), i = 0; arg; + arg = TREE_CHAIN (arg), i++) + if (i == idx) + break; + gcc_assert (arg); + OMP_CLAUSE_DECL (c) = arg; + } +} + +/* True if OpenMP sharing attribute of DECL is predetermined. */ + +enum omp_clause_default_kind +c_omp_predetermined_sharing (tree decl) +{ + /* Variables with const-qualified type having no mutable member + are predetermined shared. */ + if (TREE_READONLY (decl)) + return OMP_CLAUSE_DEFAULT_SHARED; + + return OMP_CLAUSE_DEFAULT_UNSPECIFIED; +} |