1.1 --- a/src/cpu/sparc/vm/stubGenerator_sparc.cpp Fri Mar 13 11:35:17 2009 -0700
1.2 +++ b/src/cpu/sparc/vm/stubGenerator_sparc.cpp Fri Mar 13 18:39:22 2009 -0700
1.3 @@ -900,19 +900,7 @@
1.4 __ align(CodeEntryAlignment);
1.5 StubCodeMark mark(this, "StubRoutines", "partial_subtype_check");
1.6 address start = __ pc();
1.7 - Label loop, miss;
1.8 -
1.9 - // Compare super with sub directly, since super is not in its own SSA.
1.10 - // The compiler used to emit this test, but we fold it in here,
1.11 - // to increase overall code density, with no real loss of speed.
1.12 - { Label L;
1.13 - __ cmp(O1, O2);
1.14 - __ brx(Assembler::notEqual, false, Assembler::pt, L);
1.15 - __ delayed()->nop();
1.16 - __ retl();
1.17 - __ delayed()->addcc(G0,0,O0); // set Z flags, zero result
1.18 - __ bind(L);
1.19 - }
1.20 + Label miss;
1.21
1.22 #if defined(COMPILER2) && !defined(_LP64)
1.23 // Do not use a 'save' because it blows the 64-bit O registers.
1.24 @@ -936,56 +924,12 @@
1.25 Register L2_super = L2;
1.26 Register L3_index = L3;
1.27
1.28 -#ifdef _LP64
1.29 - Register L4_ooptmp = L4;
1.30 -
1.31 - if (UseCompressedOops) {
1.32 - // this must be under UseCompressedOops check, as we rely upon fact
1.33 - // that L4 not clobbered in C2 on 32-bit platforms, where we do explicit save
1.34 - // on stack, see several lines above
1.35 - __ encode_heap_oop(Rsuper, L4_ooptmp);
1.36 - }
1.37 -#endif
1.38 -
1.39 - inc_counter_np(SharedRuntime::_partial_subtype_ctr, L0, L1);
1.40 -
1.41 - __ ld_ptr( Rsub, sizeof(oopDesc) + Klass::secondary_supers_offset_in_bytes(), L3 );
1.42 - __ lduw(L3,arrayOopDesc::length_offset_in_bytes(),L0_ary_len);
1.43 - __ add(L3,arrayOopDesc::base_offset_in_bytes(T_OBJECT),L1_ary_ptr);
1.44 - __ clr(L3_index); // zero index
1.45 - // Load a little early; will load 1 off the end of the array.
1.46 - // Ok for now; revisit if we have other uses of this routine.
1.47 - if (UseCompressedOops) {
1.48 - __ lduw(L1_ary_ptr,0,L2_super);// Will load a little early
1.49 - } else {
1.50 - __ ld_ptr(L1_ary_ptr,0,L2_super);// Will load a little early
1.51 - }
1.52 -
1.53 - assert(heapOopSize != 0, "heapOopSize should be initialized");
1.54 - // The scan loop
1.55 - __ BIND(loop);
1.56 - __ add(L1_ary_ptr, heapOopSize, L1_ary_ptr); // Bump by OOP size
1.57 - __ cmp(L3_index,L0_ary_len);
1.58 - __ br(Assembler::equal,false,Assembler::pn,miss);
1.59 - __ delayed()->inc(L3_index); // Bump index
1.60 -
1.61 - if (UseCompressedOops) {
1.62 -#ifdef _LP64
1.63 - __ subcc(L2_super,L4_ooptmp,Rret); // Check for match; zero in Rret for a hit
1.64 - __ br( Assembler::notEqual, false, Assembler::pt, loop );
1.65 - __ delayed()->lduw(L1_ary_ptr,0,L2_super);// Will load a little early
1.66 -#else
1.67 - ShouldNotReachHere();
1.68 -#endif
1.69 - } else {
1.70 - __ subcc(L2_super,Rsuper,Rret); // Check for match; zero in Rret for a hit
1.71 - __ brx( Assembler::notEqual, false, Assembler::pt, loop );
1.72 - __ delayed()->ld_ptr(L1_ary_ptr,0,L2_super);// Will load a little early
1.73 - }
1.74 -
1.75 - // Got a hit; report success; set cache. Cache load doesn't
1.76 - // happen here; for speed it is directly emitted by the compiler.
1.77 - __ st_ptr( Rsuper, Rsub, sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes() );
1.78 + __ check_klass_subtype_slow_path(Rsub, Rsuper,
1.79 + L0, L1, L2, L3,
1.80 + NULL, &miss);
1.81 +
1.82 + // Match falls through here.
1.83 + __ addcc(G0,0,Rret); // set Z flags, Z result
1.84
1.85 #if defined(COMPILER2) && !defined(_LP64)
1.86 __ ld_ptr(SP,(frame::register_save_words+0)*wordSize,L0);
1.87 @@ -999,7 +943,6 @@
1.88 __ delayed()->restore();
1.89 #endif
1.90
1.91 - // Hit or miss falls through here
1.92 __ BIND(miss);
1.93 __ addcc(G0,1,Rret); // set NZ flags, NZ result
1.94
1.95 @@ -2330,51 +2273,31 @@
1.96 Register super_check_offset,
1.97 Register super_klass,
1.98 Register temp,
1.99 - Label& L_success,
1.100 - Register deccc_hack = noreg) {
1.101 + Label& L_success) {
1.102 assert_different_registers(sub_klass, super_check_offset, super_klass, temp);
1.103
1.104 BLOCK_COMMENT("type_check:");
1.105
1.106 - Label L_miss;
1.107 + Label L_miss, L_pop_to_miss;
1.108
1.109 assert_clean_int(super_check_offset, temp);
1.110
1.111 - // maybe decrement caller's trip count:
1.112 -#define DELAY_SLOT delayed(); \
1.113 - { if (deccc_hack == noreg) __ nop(); else __ deccc(deccc_hack); }
1.114 -
1.115 - // if the pointers are equal, we are done (e.g., String[] elements)
1.116 - __ cmp(sub_klass, super_klass);
1.117 - __ brx(Assembler::equal, true, Assembler::pt, L_success);
1.118 - __ DELAY_SLOT;
1.119 -
1.120 - // check the supertype display:
1.121 - __ ld_ptr(sub_klass, super_check_offset, temp); // query the super type
1.122 - __ cmp(super_klass, temp); // test the super type
1.123 - __ brx(Assembler::equal, true, Assembler::pt, L_success);
1.124 - __ DELAY_SLOT;
1.125 -
1.126 - int sc_offset = (klassOopDesc::header_size() * HeapWordSize +
1.127 - Klass::secondary_super_cache_offset_in_bytes());
1.128 - __ cmp(super_klass, sc_offset);
1.129 - __ brx(Assembler::notEqual, true, Assembler::pt, L_miss);
1.130 - __ delayed()->nop();
1.131 -
1.132 + __ check_klass_subtype_fast_path(sub_klass, super_klass, temp, noreg,
1.133 + &L_success, &L_miss, NULL,
1.134 + super_check_offset);
1.135 +
1.136 + BLOCK_COMMENT("type_check_slow_path:");
1.137 __ save_frame(0);
1.138 - __ mov(sub_klass->after_save(), O1);
1.139 - // mov(super_klass->after_save(), O2); //fill delay slot
1.140 - assert(StubRoutines::Sparc::_partial_subtype_check != NULL, "order of generation");
1.141 - __ call(StubRoutines::Sparc::_partial_subtype_check);
1.142 - __ delayed()->mov(super_klass->after_save(), O2);
1.143 + __ check_klass_subtype_slow_path(sub_klass->after_save(),
1.144 + super_klass->after_save(),
1.145 + L0, L1, L2, L4,
1.146 + NULL, &L_pop_to_miss);
1.147 + __ ba(false, L_success);
1.148 + __ delayed()->restore();
1.149 +
1.150 + __ bind(L_pop_to_miss);
1.151 __ restore();
1.152
1.153 - // Upon return, the condition codes are already set.
1.154 - __ brx(Assembler::equal, true, Assembler::pt, L_success);
1.155 - __ DELAY_SLOT;
1.156 -
1.157 -#undef DELAY_SLOT
1.158 -
1.159 // Fall through on failure!
1.160 __ BIND(L_miss);
1.161 }
1.162 @@ -2411,7 +2334,7 @@
1.163 gen_write_ref_array_pre_barrier(O1, O2);
1.164
1.165 #ifdef ASSERT
1.166 - // We sometimes save a frame (see partial_subtype_check below).
1.167 + // We sometimes save a frame (see generate_type_check below).
1.168 // If this will cause trouble, let's fail now instead of later.
1.169 __ save_frame(0);
1.170 __ restore();
1.171 @@ -2455,41 +2378,39 @@
1.172 // G3, G4, G5 --- current oop, oop.klass, oop.klass.super
1.173 __ align(16);
1.174
1.175 - __ bind(store_element);
1.176 - // deccc(G1_remain); // decrement the count (hoisted)
1.177 + __ BIND(store_element);
1.178 + __ deccc(G1_remain); // decrement the count
1.179 __ store_heap_oop(G3_oop, O1_to, O5_offset); // store the oop
1.180 __ inc(O5_offset, heapOopSize); // step to next offset
1.181 __ brx(Assembler::zero, true, Assembler::pt, do_card_marks);
1.182 __ delayed()->set(0, O0); // return -1 on success
1.183
1.184 // ======== loop entry is here ========
1.185 - __ bind(load_element);
1.186 + __ BIND(load_element);
1.187 __ load_heap_oop(O0_from, O5_offset, G3_oop); // load the oop
1.188 __ br_null(G3_oop, true, Assembler::pt, store_element);
1.189 - __ delayed()->deccc(G1_remain); // decrement the count
1.190 + __ delayed()->nop();
1.191
1.192 __ load_klass(G3_oop, G4_klass); // query the object klass
1.193
1.194 generate_type_check(G4_klass, O3_ckoff, O4_ckval, G5_super,
1.195 // branch to this on success:
1.196 - store_element,
1.197 - // decrement this on success:
1.198 - G1_remain);
1.199 + store_element);
1.200 // ======== end loop ========
1.201
1.202 // It was a real error; we must depend on the caller to finish the job.
1.203 // Register G1 has number of *remaining* oops, O2 number of *total* oops.
1.204 // Emit GC store barriers for the oops we have copied (O2 minus G1),
1.205 // and report their number to the caller.
1.206 - __ bind(fail);
1.207 + __ BIND(fail);
1.208 __ subcc(O2_count, G1_remain, O2_count);
1.209 __ brx(Assembler::zero, false, Assembler::pt, done);
1.210 __ delayed()->not1(O2_count, O0); // report (-1^K) to caller
1.211
1.212 - __ bind(do_card_marks);
1.213 + __ BIND(do_card_marks);
1.214 gen_write_ref_array_post_barrier(O1_to, O2_count, O3); // store check on O1[0..O2]
1.215
1.216 - __ bind(done);
1.217 + __ BIND(done);
1.218 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr, O3, O4);
1.219 __ retl();
1.220 __ delayed()->nop(); // return value in 00
