src/cpu/x86/vm/c1_LinearScan_x86.hpp

Fri, 13 Mar 2009 11:35:17 -0700

author
twisti
date
Fri, 13 Mar 2009 11:35:17 -0700
changeset 1078
c771b7f43bbf
parent 772
9ee9cf798b59
child 1907
c18cbe5936b8
permissions
-rw-r--r--

6378821: bitCount() should use POPC on SPARC processors and AMD+10h
Summary: bitCount() should use POPC on SPARC processors where POPC is implemented directly in hardware.
Reviewed-by: kvn, never

duke@435 1 /*
xdono@772 2 * Copyright 2005-2008 Sun Microsystems, Inc. All Rights Reserved.
duke@435 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
duke@435 4 *
duke@435 5 * This code is free software; you can redistribute it and/or modify it
duke@435 6 * under the terms of the GNU General Public License version 2 only, as
duke@435 7 * published by the Free Software Foundation.
duke@435 8 *
duke@435 9 * This code is distributed in the hope that it will be useful, but WITHOUT
duke@435 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
duke@435 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
duke@435 12 * version 2 for more details (a copy is included in the LICENSE file that
duke@435 13 * accompanied this code).
duke@435 14 *
duke@435 15 * You should have received a copy of the GNU General Public License version
duke@435 16 * 2 along with this work; if not, write to the Free Software Foundation,
duke@435 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
duke@435 18 *
duke@435 19 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
duke@435 20 * CA 95054 USA or visit www.sun.com if you need additional information or
duke@435 21 * have any questions.
duke@435 22 *
duke@435 23 */
duke@435 24
duke@435 25 inline bool LinearScan::is_processed_reg_num(int reg_num) {
never@739 26 #ifndef _LP64
duke@435 27 // rsp and rbp (numbers 6 ancd 7) are ignored
duke@435 28 assert(FrameMap::rsp_opr->cpu_regnr() == 6, "wrong assumption below");
duke@435 29 assert(FrameMap::rbp_opr->cpu_regnr() == 7, "wrong assumption below");
duke@435 30 assert(reg_num >= 0, "invalid reg_num");
duke@435 31
duke@435 32 return reg_num < 6 || reg_num > 7;
never@739 33 #else
never@739 34 // rsp and rbp, r10, r15 (numbers 6 ancd 7) are ignored
never@739 35 assert(FrameMap::r10_opr->cpu_regnr() == 12, "wrong assumption below");
never@739 36 assert(FrameMap::r15_opr->cpu_regnr() == 13, "wrong assumption below");
never@739 37 assert(FrameMap::rsp_opr->cpu_regnrLo() == 14, "wrong assumption below");
never@739 38 assert(FrameMap::rbp_opr->cpu_regnrLo() == 15, "wrong assumption below");
never@739 39 assert(reg_num >= 0, "invalid reg_num");
never@739 40
never@739 41 return reg_num < 12 || reg_num > 15;
never@739 42 #endif // _LP64
duke@435 43 }
duke@435 44
duke@435 45 inline int LinearScan::num_physical_regs(BasicType type) {
duke@435 46 // Intel requires two cpu registers for long,
duke@435 47 // but requires only one fpu register for double
never@739 48 if (LP64_ONLY(false &&) type == T_LONG) {
duke@435 49 return 2;
duke@435 50 }
duke@435 51 return 1;
duke@435 52 }
duke@435 53
duke@435 54
duke@435 55 inline bool LinearScan::requires_adjacent_regs(BasicType type) {
duke@435 56 return false;
duke@435 57 }
duke@435 58
duke@435 59 inline bool LinearScan::is_caller_save(int assigned_reg) {
duke@435 60 assert(assigned_reg >= 0 && assigned_reg < nof_regs, "should call this only for registers");
duke@435 61 return true; // no callee-saved registers on Intel
duke@435 62
duke@435 63 }
duke@435 64
duke@435 65
duke@435 66 inline void LinearScan::pd_add_temps(LIR_Op* op) {
duke@435 67 switch (op->code()) {
duke@435 68 case lir_tan:
duke@435 69 case lir_sin:
duke@435 70 case lir_cos: {
duke@435 71 // The slow path for these functions may need to save and
duke@435 72 // restore all live registers but we don't want to save and
duke@435 73 // restore everything all the time, so mark the xmms as being
duke@435 74 // killed. If the slow path were explicit or we could propagate
duke@435 75 // live register masks down to the assembly we could do better
duke@435 76 // but we don't have any easy way to do that right now. We
duke@435 77 // could also consider not killing all xmm registers if we
duke@435 78 // assume that slow paths are uncommon but it's not clear that
duke@435 79 // would be a good idea.
duke@435 80 if (UseSSE > 0) {
duke@435 81 #ifndef PRODUCT
duke@435 82 if (TraceLinearScanLevel >= 2) {
duke@435 83 tty->print_cr("killing XMMs for trig");
duke@435 84 }
duke@435 85 #endif
duke@435 86 int op_id = op->id();
duke@435 87 for (int xmm = 0; xmm < FrameMap::nof_caller_save_xmm_regs; xmm++) {
duke@435 88 LIR_Opr opr = FrameMap::caller_save_xmm_reg_at(xmm);
duke@435 89 add_temp(reg_num(opr), op_id, noUse, T_ILLEGAL);
duke@435 90 }
duke@435 91 }
duke@435 92 break;
duke@435 93 }
duke@435 94 }
duke@435 95 }
duke@435 96
duke@435 97
duke@435 98 // Implementation of LinearScanWalker
duke@435 99
duke@435 100 inline bool LinearScanWalker::pd_init_regs_for_alloc(Interval* cur) {
duke@435 101 if (allocator()->gen()->is_vreg_flag_set(cur->reg_num(), LIRGenerator::byte_reg)) {
duke@435 102 assert(cur->type() != T_FLOAT && cur->type() != T_DOUBLE, "cpu regs only");
duke@435 103 _first_reg = pd_first_byte_reg;
duke@435 104 _last_reg = pd_last_byte_reg;
duke@435 105 return true;
duke@435 106 } else if ((UseSSE >= 1 && cur->type() == T_FLOAT) || (UseSSE >= 2 && cur->type() == T_DOUBLE)) {
duke@435 107 _first_reg = pd_first_xmm_reg;
duke@435 108 _last_reg = pd_last_xmm_reg;
duke@435 109 return true;
duke@435 110 }
duke@435 111
duke@435 112 return false;
duke@435 113 }
duke@435 114
duke@435 115
duke@435 116 class FpuStackAllocator VALUE_OBJ_CLASS_SPEC {
duke@435 117 private:
duke@435 118 Compilation* _compilation;
duke@435 119 LinearScan* _allocator;
duke@435 120
duke@435 121 LIR_OpVisitState visitor;
duke@435 122
duke@435 123 LIR_List* _lir;
duke@435 124 int _pos;
duke@435 125 FpuStackSim _sim;
duke@435 126 FpuStackSim _temp_sim;
duke@435 127
duke@435 128 bool _debug_information_computed;
duke@435 129
duke@435 130 LinearScan* allocator() { return _allocator; }
duke@435 131 Compilation* compilation() const { return _compilation; }
duke@435 132
duke@435 133 // unified bailout support
duke@435 134 void bailout(const char* msg) const { compilation()->bailout(msg); }
duke@435 135 bool bailed_out() const { return compilation()->bailed_out(); }
duke@435 136
duke@435 137 int pos() { return _pos; }
duke@435 138 void set_pos(int pos) { _pos = pos; }
duke@435 139 LIR_Op* cur_op() { return lir()->instructions_list()->at(pos()); }
duke@435 140 LIR_List* lir() { return _lir; }
duke@435 141 void set_lir(LIR_List* lir) { _lir = lir; }
duke@435 142 FpuStackSim* sim() { return &_sim; }
duke@435 143 FpuStackSim* temp_sim() { return &_temp_sim; }
duke@435 144
duke@435 145 int fpu_num(LIR_Opr opr);
duke@435 146 int tos_offset(LIR_Opr opr);
duke@435 147 LIR_Opr to_fpu_stack_top(LIR_Opr opr, bool dont_check_offset = false);
duke@435 148
duke@435 149 // Helper functions for handling operations
duke@435 150 void insert_op(LIR_Op* op);
duke@435 151 void insert_exchange(int offset);
duke@435 152 void insert_exchange(LIR_Opr opr);
duke@435 153 void insert_free(int offset);
duke@435 154 void insert_free_if_dead(LIR_Opr opr);
duke@435 155 void insert_free_if_dead(LIR_Opr opr, LIR_Opr ignore);
duke@435 156 void insert_copy(LIR_Opr from, LIR_Opr to);
duke@435 157 void do_rename(LIR_Opr from, LIR_Opr to);
duke@435 158 void do_push(LIR_Opr opr);
duke@435 159 void pop_if_last_use(LIR_Op* op, LIR_Opr opr);
duke@435 160 void pop_always(LIR_Op* op, LIR_Opr opr);
duke@435 161 void clear_fpu_stack(LIR_Opr preserve);
duke@435 162 void handle_op1(LIR_Op1* op1);
duke@435 163 void handle_op2(LIR_Op2* op2);
duke@435 164 void handle_opCall(LIR_OpCall* opCall);
duke@435 165 void compute_debug_information(LIR_Op* op);
duke@435 166 void allocate_exception_handler(XHandler* xhandler);
duke@435 167 void allocate_block(BlockBegin* block);
duke@435 168
duke@435 169 #ifndef PRODUCT
duke@435 170 void check_invalid_lir_op(LIR_Op* op);
duke@435 171 #endif
duke@435 172
duke@435 173 // Helper functions for merging of fpu stacks
duke@435 174 void merge_insert_add(LIR_List* instrs, FpuStackSim* cur_sim, int reg);
duke@435 175 void merge_insert_xchg(LIR_List* instrs, FpuStackSim* cur_sim, int slot);
duke@435 176 void merge_insert_pop(LIR_List* instrs, FpuStackSim* cur_sim);
duke@435 177 bool merge_rename(FpuStackSim* cur_sim, FpuStackSim* sux_sim, int start_slot, int change_slot);
duke@435 178 void merge_fpu_stack(LIR_List* instrs, FpuStackSim* cur_sim, FpuStackSim* sux_sim);
duke@435 179 void merge_cleanup_fpu_stack(LIR_List* instrs, FpuStackSim* cur_sim, BitMap& live_fpu_regs);
duke@435 180 bool merge_fpu_stack_with_successors(BlockBegin* block);
duke@435 181
duke@435 182 public:
duke@435 183 LIR_Opr to_fpu_stack(LIR_Opr opr); // used by LinearScan for creation of debug information
duke@435 184
duke@435 185 FpuStackAllocator(Compilation* compilation, LinearScan* allocator);
duke@435 186 void allocate();
duke@435 187 };

mercurial