Fri, 13 Mar 2009 11:35:17 -0700
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@631 | 2 | * Copyright 1997-2008 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 | #include "incls/_precompiled.incl" |
duke@435 | 26 | #include "incls/_vtableStubs_sparc.cpp.incl" |
duke@435 | 27 | |
duke@435 | 28 | // machine-dependent part of VtableStubs: create vtableStub of correct size and |
duke@435 | 29 | // initialize its code |
duke@435 | 30 | |
duke@435 | 31 | #define __ masm-> |
duke@435 | 32 | |
duke@435 | 33 | |
duke@435 | 34 | #ifndef PRODUCT |
duke@435 | 35 | extern "C" void bad_compiled_vtable_index(JavaThread* thread, oopDesc* receiver, int index); |
duke@435 | 36 | #endif |
duke@435 | 37 | |
duke@435 | 38 | |
duke@435 | 39 | // Used by compiler only; may use only caller saved, non-argument registers |
duke@435 | 40 | // NOTE: %%%% if any change is made to this stub make sure that the function |
duke@435 | 41 | // pd_code_size_limit is changed to ensure the correct size for VtableStub |
duke@435 | 42 | VtableStub* VtableStubs::create_vtable_stub(int vtable_index) { |
duke@435 | 43 | const int sparc_code_length = VtableStub::pd_code_size_limit(true); |
duke@435 | 44 | VtableStub* s = new(sparc_code_length) VtableStub(true, vtable_index); |
duke@435 | 45 | ResourceMark rm; |
duke@435 | 46 | CodeBuffer cb(s->entry_point(), sparc_code_length); |
duke@435 | 47 | MacroAssembler* masm = new MacroAssembler(&cb); |
duke@435 | 48 | |
duke@435 | 49 | #ifndef PRODUCT |
duke@435 | 50 | if (CountCompiledCalls) { |
duke@435 | 51 | Address ctr(G5, SharedRuntime::nof_megamorphic_calls_addr()); |
duke@435 | 52 | __ sethi(ctr); |
duke@435 | 53 | __ ld(ctr, G3_scratch); |
duke@435 | 54 | __ inc(G3_scratch); |
duke@435 | 55 | __ st(G3_scratch, ctr); |
duke@435 | 56 | } |
duke@435 | 57 | #endif /* PRODUCT */ |
duke@435 | 58 | |
duke@435 | 59 | assert(VtableStub::receiver_location() == O0->as_VMReg(), "receiver expected in O0"); |
duke@435 | 60 | |
duke@435 | 61 | // get receiver klass |
duke@435 | 62 | address npe_addr = __ pc(); |
coleenp@548 | 63 | __ load_klass(O0, G3_scratch); |
duke@435 | 64 | |
duke@435 | 65 | // set methodOop (in case of interpreted method), and destination address |
duke@435 | 66 | int entry_offset = instanceKlass::vtable_start_offset() + vtable_index*vtableEntry::size(); |
duke@435 | 67 | #ifndef PRODUCT |
duke@435 | 68 | if (DebugVtables) { |
duke@435 | 69 | Label L; |
duke@435 | 70 | // check offset vs vtable length |
duke@435 | 71 | __ ld(G3_scratch, instanceKlass::vtable_length_offset()*wordSize, G5); |
duke@435 | 72 | __ cmp(G5, vtable_index*vtableEntry::size()); |
duke@435 | 73 | __ br(Assembler::greaterUnsigned, false, Assembler::pt, L); |
duke@435 | 74 | __ delayed()->nop(); |
duke@435 | 75 | __ set(vtable_index, O2); |
duke@435 | 76 | __ call_VM(noreg, CAST_FROM_FN_PTR(address, bad_compiled_vtable_index), O0, O2); |
duke@435 | 77 | __ bind(L); |
duke@435 | 78 | } |
duke@435 | 79 | #endif |
duke@435 | 80 | int v_off = entry_offset*wordSize + vtableEntry::method_offset_in_bytes(); |
duke@435 | 81 | if( __ is_simm13(v_off) ) { |
duke@435 | 82 | __ ld_ptr(G3, v_off, G5_method); |
duke@435 | 83 | } else { |
duke@435 | 84 | __ set(v_off,G5); |
duke@435 | 85 | __ ld_ptr(G3, G5, G5_method); |
duke@435 | 86 | } |
duke@435 | 87 | |
duke@435 | 88 | #ifndef PRODUCT |
duke@435 | 89 | if (DebugVtables) { |
duke@435 | 90 | Label L; |
duke@435 | 91 | __ br_notnull(G5_method, false, Assembler::pt, L); |
duke@435 | 92 | __ delayed()->nop(); |
duke@435 | 93 | __ stop("Vtable entry is ZERO"); |
duke@435 | 94 | __ bind(L); |
duke@435 | 95 | } |
duke@435 | 96 | #endif |
duke@435 | 97 | |
duke@435 | 98 | address ame_addr = __ pc(); // if the vtable entry is null, the method is abstract |
duke@435 | 99 | // NOTE: for vtable dispatches, the vtable entry will never be null. |
duke@435 | 100 | |
duke@435 | 101 | __ ld_ptr(G5_method, in_bytes(methodOopDesc::from_compiled_offset()), G3_scratch); |
duke@435 | 102 | |
duke@435 | 103 | // jump to target (either compiled code or c2iadapter) |
duke@435 | 104 | __ JMP(G3_scratch, 0); |
duke@435 | 105 | // load methodOop (in case we call c2iadapter) |
duke@435 | 106 | __ delayed()->nop(); |
duke@435 | 107 | |
duke@435 | 108 | masm->flush(); |
jrose@1058 | 109 | |
jrose@1058 | 110 | if (PrintMiscellaneous && (WizardMode || Verbose)) { |
jrose@1058 | 111 | tty->print_cr("vtable #%d at "PTR_FORMAT"[%d] left over: %d", |
jrose@1058 | 112 | vtable_index, s->entry_point(), |
jrose@1058 | 113 | (int)(s->code_end() - s->entry_point()), |
jrose@1058 | 114 | (int)(s->code_end() - __ pc())); |
jrose@1058 | 115 | } |
jrose@1058 | 116 | guarantee(__ pc() <= s->code_end(), "overflowed buffer"); |
jrose@1058 | 117 | |
duke@435 | 118 | s->set_exception_points(npe_addr, ame_addr); |
duke@435 | 119 | return s; |
duke@435 | 120 | } |
duke@435 | 121 | |
duke@435 | 122 | |
duke@435 | 123 | // NOTE: %%%% if any change is made to this stub make sure that the function |
duke@435 | 124 | // pd_code_size_limit is changed to ensure the correct size for VtableStub |
jrose@1058 | 125 | VtableStub* VtableStubs::create_itable_stub(int itable_index) { |
duke@435 | 126 | const int sparc_code_length = VtableStub::pd_code_size_limit(false); |
jrose@1058 | 127 | VtableStub* s = new(sparc_code_length) VtableStub(false, itable_index); |
duke@435 | 128 | ResourceMark rm; |
duke@435 | 129 | CodeBuffer cb(s->entry_point(), sparc_code_length); |
duke@435 | 130 | MacroAssembler* masm = new MacroAssembler(&cb); |
duke@435 | 131 | |
duke@435 | 132 | Register G3_klassOop = G3_scratch; |
duke@435 | 133 | Register G5_interface = G5; // Passed in as an argument |
duke@435 | 134 | Label search; |
duke@435 | 135 | |
duke@435 | 136 | // Entry arguments: |
duke@435 | 137 | // G5_interface: Interface |
duke@435 | 138 | // O0: Receiver |
duke@435 | 139 | assert(VtableStub::receiver_location() == O0->as_VMReg(), "receiver expected in O0"); |
duke@435 | 140 | |
duke@435 | 141 | // get receiver klass (also an implicit null-check) |
duke@435 | 142 | address npe_addr = __ pc(); |
coleenp@548 | 143 | __ load_klass(O0, G3_klassOop); |
duke@435 | 144 | __ verify_oop(G3_klassOop); |
duke@435 | 145 | |
duke@435 | 146 | // Push a new window to get some temp registers. This chops the head of all |
duke@435 | 147 | // my 64-bit %o registers in the LION build, but this is OK because no longs |
duke@435 | 148 | // are passed in the %o registers. Instead, longs are passed in G1 and G4 |
duke@435 | 149 | // and so those registers are not available here. |
duke@435 | 150 | __ save(SP,-frame::register_save_words*wordSize,SP); |
duke@435 | 151 | |
duke@435 | 152 | #ifndef PRODUCT |
duke@435 | 153 | if (CountCompiledCalls) { |
duke@435 | 154 | Address ctr(L0, SharedRuntime::nof_megamorphic_calls_addr()); |
duke@435 | 155 | __ sethi(ctr); |
duke@435 | 156 | __ ld(ctr, L1); |
duke@435 | 157 | __ inc(L1); |
duke@435 | 158 | __ st(L1, ctr); |
duke@435 | 159 | } |
duke@435 | 160 | #endif /* PRODUCT */ |
duke@435 | 161 | |
jrose@1058 | 162 | Label throw_icce; |
duke@435 | 163 | |
jrose@1058 | 164 | Register L5_method = L5; |
jrose@1058 | 165 | __ lookup_interface_method(// inputs: rec. class, interface, itable index |
jrose@1058 | 166 | G3_klassOop, G5_interface, itable_index, |
jrose@1058 | 167 | // outputs: method, scan temp. reg |
jrose@1058 | 168 | L5_method, L2, L3, |
jrose@1058 | 169 | throw_icce); |
duke@435 | 170 | |
duke@435 | 171 | #ifndef PRODUCT |
duke@435 | 172 | if (DebugVtables) { |
duke@435 | 173 | Label L01; |
jrose@1058 | 174 | __ bpr(Assembler::rc_nz, false, Assembler::pt, L5_method, L01); |
duke@435 | 175 | __ delayed()->nop(); |
duke@435 | 176 | __ stop("methodOop is null"); |
duke@435 | 177 | __ bind(L01); |
jrose@1058 | 178 | __ verify_oop(L5_method); |
duke@435 | 179 | } |
duke@435 | 180 | #endif |
duke@435 | 181 | |
duke@435 | 182 | // If the following load is through a NULL pointer, we'll take an OS |
duke@435 | 183 | // exception that should translate into an AbstractMethodError. We need the |
duke@435 | 184 | // window count to be correct at that time. |
jrose@1058 | 185 | __ restore(L5_method, 0, G5_method); |
jrose@1058 | 186 | // Restore registers *before* the AME point. |
duke@435 | 187 | |
duke@435 | 188 | address ame_addr = __ pc(); // if the vtable entry is null, the method is abstract |
duke@435 | 189 | __ ld_ptr(G5_method, in_bytes(methodOopDesc::from_compiled_offset()), G3_scratch); |
duke@435 | 190 | |
duke@435 | 191 | // G5_method: methodOop |
duke@435 | 192 | // O0: Receiver |
duke@435 | 193 | // G3_scratch: entry point |
duke@435 | 194 | __ JMP(G3_scratch, 0); |
duke@435 | 195 | __ delayed()->nop(); |
duke@435 | 196 | |
dcubed@451 | 197 | __ bind(throw_icce); |
dcubed@451 | 198 | Address icce(G3_scratch, StubRoutines::throw_IncompatibleClassChangeError_entry()); |
dcubed@451 | 199 | __ jump_to(icce, 0); |
dcubed@451 | 200 | __ delayed()->restore(); |
dcubed@451 | 201 | |
duke@435 | 202 | masm->flush(); |
dcubed@451 | 203 | |
jrose@1058 | 204 | if (PrintMiscellaneous && (WizardMode || Verbose)) { |
jrose@1058 | 205 | tty->print_cr("itable #%d at "PTR_FORMAT"[%d] left over: %d", |
jrose@1058 | 206 | itable_index, s->entry_point(), |
jrose@1058 | 207 | (int)(s->code_end() - s->entry_point()), |
jrose@1058 | 208 | (int)(s->code_end() - __ pc())); |
jrose@1058 | 209 | } |
dcubed@451 | 210 | guarantee(__ pc() <= s->code_end(), "overflowed buffer"); |
dcubed@451 | 211 | |
duke@435 | 212 | s->set_exception_points(npe_addr, ame_addr); |
duke@435 | 213 | return s; |
duke@435 | 214 | } |
duke@435 | 215 | |
duke@435 | 216 | |
duke@435 | 217 | int VtableStub::pd_code_size_limit(bool is_vtable_stub) { |
dcubed@451 | 218 | if (TraceJumps || DebugVtables || CountCompiledCalls || VerifyOops) return 1000; |
duke@435 | 219 | else { |
duke@435 | 220 | const int slop = 2*BytesPerInstWord; // sethi;add (needed for long offsets) |
duke@435 | 221 | if (is_vtable_stub) { |
coleenp@548 | 222 | // ld;ld;ld,jmp,nop |
coleenp@548 | 223 | const int basic = 5*BytesPerInstWord + |
kvn@1077 | 224 | // shift;add for load_klass (only shift with zero heap based) |
kvn@1077 | 225 | (UseCompressedOops ? |
kvn@1077 | 226 | ((Universe::narrow_oop_base() == NULL) ? BytesPerInstWord : 2*BytesPerInstWord) : 0); |
duke@435 | 227 | return basic + slop; |
duke@435 | 228 | } else { |
jrose@1058 | 229 | const int basic = (28 LP64_ONLY(+ 6)) * BytesPerInstWord + |
kvn@1077 | 230 | // shift;add for load_klass (only shift with zero heap based) |
kvn@1077 | 231 | (UseCompressedOops ? |
kvn@1077 | 232 | ((Universe::narrow_oop_base() == NULL) ? BytesPerInstWord : 2*BytesPerInstWord) : 0); |
duke@435 | 233 | return (basic + slop); |
duke@435 | 234 | } |
duke@435 | 235 | } |
duke@435 | 236 | } |
duke@435 | 237 | |
duke@435 | 238 | |
duke@435 | 239 | int VtableStub::pd_code_alignment() { |
duke@435 | 240 | // UltraSPARC cache line size is 8 instructions: |
duke@435 | 241 | const unsigned int icache_line_size = 32; |
duke@435 | 242 | return icache_line_size; |
duke@435 | 243 | } |