1 /*
2  * Copyright (c) 2005, 2019, Oracle and/or its affiliates. All rights reserved.
3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4  *
5  * This code is free software; you can redistribute it and/or modify it
6  * under the terms of the GNU General Public License version 2 only, as
7  * published by the Free Software Foundation.
8  *
9  * This code is distributed in the hope that it will be useful, but WITHOUT
10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
12  * version 2 for more details (a copy is included in the LICENSE file that
13  * accompanied this code).
14  *
15  * You should have received a copy of the GNU General Public License version
16  * 2 along with this work; if not, write to the Free Software Foundation,
17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18  *
19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20  * or visit www.oracle.com if you need additional information or have any
21  * questions.
22  *
23  */
24 
25 #ifndef CPU_X86_C1_LINEARSCAN_X86_HPP
26 #define CPU_X86_C1_LINEARSCAN_X86_HPP
27 
is_processed_reg_num(int reg_num)28 inline bool LinearScan::is_processed_reg_num(int reg_num) {
29 #ifndef _LP64
30   // rsp and rbp (numbers 6 ancd 7) are ignored
31   assert(FrameMap::rsp_opr->cpu_regnr() == 6, "wrong assumption below");
32   assert(FrameMap::rbp_opr->cpu_regnr() == 7, "wrong assumption below");
33   assert(reg_num >= 0, "invalid reg_num");
34 #else
35   // rsp and rbp, r10, r15 (numbers [12,15]) are ignored
36   // r12 (number 11) is conditional on compressed oops.
37   assert(FrameMap::r12_opr->cpu_regnr() == 11, "wrong assumption below");
38   assert(FrameMap::r10_opr->cpu_regnr() == 12, "wrong assumption below");
39   assert(FrameMap::r15_opr->cpu_regnr() == 13, "wrong assumption below");
40   assert(FrameMap::rsp_opr->cpu_regnrLo() == 14, "wrong assumption below");
41   assert(FrameMap::rbp_opr->cpu_regnrLo() == 15, "wrong assumption below");
42   assert(reg_num >= 0, "invalid reg_num");
43 #endif // _LP64
44   return reg_num <= FrameMap::last_cpu_reg() || reg_num >= pd_nof_cpu_regs_frame_map;
45 }
46 
num_physical_regs(BasicType type)47 inline int LinearScan::num_physical_regs(BasicType type) {
48   // Intel requires two cpu registers for long,
49   // but requires only one fpu register for double
50   if (LP64_ONLY(false &&) type == T_LONG) {
51     return 2;
52   }
53   return 1;
54 }
55 
56 
requires_adjacent_regs(BasicType type)57 inline bool LinearScan::requires_adjacent_regs(BasicType type) {
58   return false;
59 }
60 
is_caller_save(int assigned_reg)61 inline bool LinearScan::is_caller_save(int assigned_reg) {
62   assert(assigned_reg >= 0 && assigned_reg < nof_regs, "should call this only for registers");
63   return true; // no callee-saved registers on Intel
64 
65 }
66 
67 
pd_add_temps(LIR_Op * op)68 inline void LinearScan::pd_add_temps(LIR_Op* op) {
69   switch (op->code()) {
70     case lir_tan: {
71       // The slow path for these functions may need to save and
72       // restore all live registers but we don't want to save and
73       // restore everything all the time, so mark the xmms as being
74       // killed.  If the slow path were explicit or we could propagate
75       // live register masks down to the assembly we could do better
76       // but we don't have any easy way to do that right now.  We
77       // could also consider not killing all xmm registers if we
78       // assume that slow paths are uncommon but it's not clear that
79       // would be a good idea.
80       if (UseSSE > 0) {
81 #ifndef PRODUCT
82         if (TraceLinearScanLevel >= 2) {
83           tty->print_cr("killing XMMs for trig");
84         }
85 #endif
86         int num_caller_save_xmm_regs = FrameMap::get_num_caller_save_xmms();
87         int op_id = op->id();
88         for (int xmm = 0; xmm < num_caller_save_xmm_regs; xmm++) {
89           LIR_Opr opr = FrameMap::caller_save_xmm_reg_at(xmm);
90           add_temp(reg_num(opr), op_id, noUse, T_ILLEGAL);
91         }
92       }
93       break;
94     }
95     default:
96       break;
97   }
98 }
99 
100 
101 // Implementation of LinearScanWalker
102 
pd_init_regs_for_alloc(Interval * cur)103 inline bool LinearScanWalker::pd_init_regs_for_alloc(Interval* cur) {
104   int last_xmm_reg = pd_last_xmm_reg;
105 #ifdef _LP64
106   if (UseAVX < 3) {
107     last_xmm_reg = pd_first_xmm_reg + (pd_nof_xmm_regs_frame_map / 2) - 1;
108   }
109 #endif
110   if (allocator()->gen()->is_vreg_flag_set(cur->reg_num(), LIRGenerator::byte_reg)) {
111     assert(cur->type() != T_FLOAT && cur->type() != T_DOUBLE, "cpu regs only");
112     _first_reg = pd_first_byte_reg;
113     _last_reg = FrameMap::last_byte_reg();
114     return true;
115   } else if ((UseSSE >= 1 && cur->type() == T_FLOAT) || (UseSSE >= 2 && cur->type() == T_DOUBLE)) {
116     _first_reg = pd_first_xmm_reg;
117     _last_reg = last_xmm_reg;
118     return true;
119   }
120 
121   return false;
122 }
123 
124 
125 class FpuStackAllocator {
126  private:
127   Compilation* _compilation;
128   LinearScan* _allocator;
129 
130   LIR_OpVisitState visitor;
131 
132   LIR_List* _lir;
133   int _pos;
134   FpuStackSim _sim;
135   FpuStackSim _temp_sim;
136 
137   bool _debug_information_computed;
138 
allocator()139   LinearScan*   allocator()                      { return _allocator; }
compilation() const140   Compilation*  compilation() const              { return _compilation; }
141 
142   // unified bailout support
bailout(const char * msg) const143   void          bailout(const char* msg) const   { compilation()->bailout(msg); }
bailed_out() const144   bool          bailed_out() const               { return compilation()->bailed_out(); }
145 
pos()146   int pos() { return _pos; }
set_pos(int pos)147   void set_pos(int pos) { _pos = pos; }
cur_op()148   LIR_Op* cur_op() { return lir()->instructions_list()->at(pos()); }
lir()149   LIR_List* lir() { return _lir; }
set_lir(LIR_List * lir)150   void set_lir(LIR_List* lir) { _lir = lir; }
sim()151   FpuStackSim* sim() { return &_sim; }
temp_sim()152   FpuStackSim* temp_sim() { return &_temp_sim; }
153 
154   int fpu_num(LIR_Opr opr);
155   int tos_offset(LIR_Opr opr);
156   LIR_Opr to_fpu_stack_top(LIR_Opr opr, bool dont_check_offset = false);
157 
158   // Helper functions for handling operations
159   void insert_op(LIR_Op* op);
160   void insert_exchange(int offset);
161   void insert_exchange(LIR_Opr opr);
162   void insert_free(int offset);
163   void insert_free_if_dead(LIR_Opr opr);
164   void insert_free_if_dead(LIR_Opr opr, LIR_Opr ignore);
165   void insert_copy(LIR_Opr from, LIR_Opr to);
166   void do_rename(LIR_Opr from, LIR_Opr to);
167   void do_push(LIR_Opr opr);
168   void pop_if_last_use(LIR_Op* op, LIR_Opr opr);
169   void pop_always(LIR_Op* op, LIR_Opr opr);
170   void clear_fpu_stack(LIR_Opr preserve);
171   void handle_op1(LIR_Op1* op1);
172   void handle_op2(LIR_Op2* op2);
173   void handle_opCall(LIR_OpCall* opCall);
174   void compute_debug_information(LIR_Op* op);
175   void allocate_exception_handler(XHandler* xhandler);
176   void allocate_block(BlockBegin* block);
177 
178 #ifndef PRODUCT
179   void check_invalid_lir_op(LIR_Op* op);
180 #endif
181 
182   // Helper functions for merging of fpu stacks
183   void merge_insert_add(LIR_List* instrs, FpuStackSim* cur_sim, int reg);
184   void merge_insert_xchg(LIR_List* instrs, FpuStackSim* cur_sim, int slot);
185   void merge_insert_pop(LIR_List* instrs, FpuStackSim* cur_sim);
186   bool merge_rename(FpuStackSim* cur_sim, FpuStackSim* sux_sim, int start_slot, int change_slot);
187   void merge_fpu_stack(LIR_List* instrs, FpuStackSim* cur_sim, FpuStackSim* sux_sim);
188   void merge_cleanup_fpu_stack(LIR_List* instrs, FpuStackSim* cur_sim, BitMap& live_fpu_regs);
189   bool merge_fpu_stack_with_successors(BlockBegin* block);
190 
191  public:
192   LIR_Opr to_fpu_stack(LIR_Opr opr); // used by LinearScan for creation of debug information
193 
194   FpuStackAllocator(Compilation* compilation, LinearScan* allocator);
195   void allocate();
196 };
197 
198 #endif // CPU_X86_C1_LINEARSCAN_X86_HPP
199