X-Git-Url: http://nsz.repo.hu/git/?a=blobdiff_plain;f=ir%2Fir%2Fircons.c;h=87881b31233f6cc6acacc24cbca9239554508dd7;hb=081ef4f7ca73620500e5494f882ee6095c8bb983;hp=5b85701cb4d3f6df885cfc6ad8befa0c20da2127;hpb=222c1d46a0a3f0bddf3df3d84523287a4a960720;p=libfirm diff --git a/ir/ir/ircons.c b/ir/ir/ircons.c index 5b85701cb..87881b312 100644 --- a/ir/ir/ircons.c +++ b/ir/ir/ircons.c @@ -2386,24 +2386,30 @@ NEW_D_BINOP(Mul) NEW_D_BINOP(Mulh) /** - * Allocate the frag array. + * Allocate a frag array for a node if the current graph state is phase_building. + * + * @param irn the node for which the frag array should be allocated + * @param op the opcode of the (original) node, if does not match opcode of irn, + * nothing is done + * @param frag_store the address of the frag store in irn attributes, if this + * address contains a value != NULL, does nothing */ -static void allocate_frag_arr(ir_node *res, ir_op *op, ir_node ***frag_store) { +void firm_alloc_frag_arr(ir_node *irn, ir_op *op, ir_node ***frag_store) { if (get_opt_precise_exc_context()) { if ((current_ir_graph->phase_state == phase_building) && - (get_irn_op(res) == op) && /* Could be optimized away. */ + (get_irn_op(irn) == op) && /* Could be optimized away. */ !*frag_store) /* Could be a cse where the arr is already set. */ { - *frag_store = new_frag_arr(res); + *frag_store = new_frag_arr(irn); } } -} /* allocate_frag_arr */ +} /* firm_alloc_frag_arr */ ir_node * new_d_Quot(dbg_info *db, ir_node *memop, ir_node *op1, ir_node *op2, ir_mode *mode, op_pin_state state) { ir_node *res; res = new_bd_Quot(db, current_ir_graph->current_block, memop, op1, op2, mode, state); #if PRECISE_EXC_CONTEXT - allocate_frag_arr(res, op_Quot, &res->attr.except.frag_arr); /* Could be optimized away. */ + firm_alloc_frag_arr(res, op_Quot, &res->attr.except.frag_arr); #endif return res; @@ -2414,7 +2420,7 @@ new_d_DivMod(dbg_info *db, ir_node *memop, ir_node *op1, ir_node *op2, ir_mode * ir_node *res; res = new_bd_DivMod(db, current_ir_graph->current_block, memop, op1, op2, mode, state); #if PRECISE_EXC_CONTEXT - allocate_frag_arr(res, op_DivMod, &res->attr.except.frag_arr); /* Could be optimized away. */ + firm_alloc_frag_arr(res, op_DivMod, &res->attr.except.frag_arr); #endif return res; @@ -2425,7 +2431,7 @@ new_d_Div(dbg_info *db, ir_node *memop, ir_node *op1, ir_node *op2, ir_mode *mod ir_node *res; res = new_bd_Div(db, current_ir_graph->current_block, memop, op1, op2, mode, state); #if PRECISE_EXC_CONTEXT - allocate_frag_arr(res, op_Div, &res->attr.except.frag_arr); /* Could be optimized away. */ + firm_alloc_frag_arr(res, op_Div, &res->attr.except.frag_arr); #endif return res; @@ -2436,7 +2442,7 @@ new_d_DivRL(dbg_info *db, ir_node *memop, ir_node *op1, ir_node *op2, ir_mode *m ir_node *res; res = new_bd_DivRL(db, current_ir_graph->current_block, memop, op1, op2, mode, state); #if PRECISE_EXC_CONTEXT - allocate_frag_arr(res, op_Div, &res->attr.except.frag_arr); /* Could be optimized away. */ + firm_alloc_frag_arr(res, op_Div, &res->attr.except.frag_arr); #endif return res; @@ -2447,7 +2453,7 @@ new_d_Mod(dbg_info *db, ir_node *memop, ir_node *op1, ir_node *op2, ir_mode *mod ir_node *res; res = new_bd_Mod(db, current_ir_graph->current_block, memop, op1, op2, mode, state); #if PRECISE_EXC_CONTEXT - allocate_frag_arr(res, op_Mod, &res->attr.except.frag_arr); /* Could be optimized away. */ + firm_alloc_frag_arr(res, op_Mod, &res->attr.except.frag_arr); #endif return res; @@ -2492,7 +2498,7 @@ new_d_Call(dbg_info *db, ir_node *store, ir_node *callee, int arity, ir_node **i res = new_bd_Call(db, current_ir_graph->current_block, store, callee, arity, in, tp); #if PRECISE_EXC_CONTEXT - allocate_frag_arr(res, op_Call, &res->attr.call.exc.frag_arr); /* Could be optimized away. */ + firm_alloc_frag_arr(res, op_Call, &res->attr.call.exc.frag_arr); #endif return res; @@ -2510,7 +2516,7 @@ new_d_Load(dbg_info *db, ir_node *store, ir_node *addr, ir_mode *mode) { res = new_bd_Load(db, current_ir_graph->current_block, store, addr, mode); #if PRECISE_EXC_CONTEXT - allocate_frag_arr(res, op_Load, &res->attr.load.exc.frag_arr); /* Could be optimized away. */ + firm_alloc_frag_arr(res, op_Load, &res->attr.load.exc.frag_arr); #endif return res; @@ -2522,7 +2528,7 @@ new_d_Store(dbg_info *db, ir_node *store, ir_node *addr, ir_node *val) { res = new_bd_Store(db, current_ir_graph->current_block, store, addr, val); #if PRECISE_EXC_CONTEXT - allocate_frag_arr(res, op_Store, &res->attr.store.exc.frag_arr); /* Could be optimized away. */ + firm_alloc_frag_arr(res, op_Store, &res->attr.store.exc.frag_arr); #endif return res; @@ -2535,7 +2541,7 @@ new_d_Alloc(dbg_info *db, ir_node *store, ir_node *size, ir_type *alloc_type, res = new_bd_Alloc(db, current_ir_graph->current_block, store, size, alloc_type, where); #if PRECISE_EXC_CONTEXT - allocate_frag_arr(res, op_Alloc, &res->attr.alloc.exc.frag_arr); /* Could be optimized away. */ + firm_alloc_frag_arr(res, op_Alloc, &res->attr.alloc.exc.frag_arr); #endif return res; @@ -2647,7 +2653,7 @@ ir_node *new_d_CopyB(dbg_info *db,ir_node *store, res = new_bd_CopyB(db, current_ir_graph->current_block, store, dst, src, data_type); #if PRECISE_EXC_CONTEXT - allocate_frag_arr(res, op_CopyB, &res->attr.copyb.exc.frag_arr); + firm_alloc_frag_arr(res, op_CopyB, &res->attr.copyb.exc.frag_arr); #endif return res; } /* new_d_CopyB */ @@ -2669,7 +2675,7 @@ ir_node *new_d_Bound(dbg_info *db,ir_node *store, res = new_bd_Bound(db, current_ir_graph->current_block, store, idx, lower, upper); #if PRECISE_EXC_CONTEXT - allocate_frag_arr(res, op_Bound, &res->attr.bound.exc.frag_arr); + firm_alloc_frag_arr(res, op_Bound, &res->attr.bound.exc.frag_arr); #endif return res; } /* new_d_Bound */ @@ -2892,9 +2898,9 @@ ir_type *get_cur_frame_type(void) { /* call once for each run of the library */ void -init_cons(uninitialized_local_variable_func_t *func) { +firm_init_cons(uninitialized_local_variable_func_t *func) { default_initialize_local_variable = func; -} /* init_cons */ +} /* firm_init_cons */ void irp_finalize_cons(void) {