Mercurial > hg > graal-compiler
changeset 1872:beba40b26a79
Merge
| author | acorn |
|---|---|
| date | Fri, 15 Oct 2010 15:12:04 -0400 |
| parents | 75b0735b4d04 (current diff) ee813f7b46e4 (diff) |
| children | 75ab0162aa84 |
| files | src/cpu/x86/vm/methodHandles_x86.cpp |
| diffstat | 41 files changed, 615 insertions(+), 493 deletions(-) [+] |
line wrap: on
line diff
--- a/src/cpu/sparc/vm/assembler_sparc.cpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/cpu/sparc/vm/assembler_sparc.cpp Fri Oct 15 15:12:04 2010 -0400 @@ -3094,11 +3094,10 @@ void MacroAssembler::check_method_handle_type(Register mtype_reg, Register mh_reg, Register temp_reg, Label& wrong_method_type) { - if (UseCompressedOops) unimplemented("coop"); // field accesses must decode assert_different_registers(mtype_reg, mh_reg, temp_reg); // compare method type against that of the receiver RegisterOrConstant mhtype_offset = delayed_value(java_dyn_MethodHandle::type_offset_in_bytes, temp_reg); - ld_ptr(mh_reg, mhtype_offset, temp_reg); + load_heap_oop(mh_reg, mhtype_offset, temp_reg); cmp(temp_reg, mtype_reg); br(Assembler::notEqual, false, Assembler::pn, wrong_method_type); delayed()->nop(); @@ -3112,16 +3111,15 @@ void MacroAssembler::load_method_handle_vmslots(Register vmslots_reg, Register mh_reg, Register temp_reg) { assert_different_registers(vmslots_reg, mh_reg, temp_reg); - if (UseCompressedOops) unimplemented("coop"); // field accesses must decode // load mh.type.form.vmslots if (java_dyn_MethodHandle::vmslots_offset_in_bytes() != 0) { // hoist vmslots into every mh to avoid dependent load chain - ld( Address(mh_reg, delayed_value(java_dyn_MethodHandle::vmslots_offset_in_bytes, temp_reg)), vmslots_reg); + ld( Address(mh_reg, delayed_value(java_dyn_MethodHandle::vmslots_offset_in_bytes, temp_reg)), vmslots_reg); } else { Register temp2_reg = vmslots_reg; - ld_ptr(Address(mh_reg, delayed_value(java_dyn_MethodHandle::type_offset_in_bytes, temp_reg)), temp2_reg); - ld_ptr(Address(temp2_reg, delayed_value(java_dyn_MethodType::form_offset_in_bytes, temp_reg)), temp2_reg); - ld( Address(temp2_reg, delayed_value(java_dyn_MethodTypeForm::vmslots_offset_in_bytes, temp_reg)), vmslots_reg); + load_heap_oop(Address(mh_reg, delayed_value(java_dyn_MethodHandle::type_offset_in_bytes, temp_reg)), temp2_reg); + load_heap_oop(Address(temp2_reg, delayed_value(java_dyn_MethodType::form_offset_in_bytes, temp_reg)), temp2_reg); + ld( Address(temp2_reg, delayed_value(java_dyn_MethodTypeForm::vmslots_offset_in_bytes, temp_reg)), vmslots_reg); } } @@ -3130,9 +3128,8 @@ assert(mh_reg == G3_method_handle, "caller must put MH object in G3"); assert_different_registers(mh_reg, temp_reg); - if (UseCompressedOops) unimplemented("coop"); // field accesses must decode - // pick out the interpreted side of the handler + // NOTE: vmentry is not an oop! ld_ptr(mh_reg, delayed_value(java_dyn_MethodHandle::vmentry_offset_in_bytes, temp_reg), temp_reg); // off we go... @@ -4653,6 +4650,11 @@ } } +void MacroAssembler::load_heap_oop(Register s1, RegisterOrConstant s2, Register d) { + if (s2.is_constant()) load_heap_oop(s1, s2.as_constant(), d); + else load_heap_oop(s1, s2.as_register(), d); +} + void MacroAssembler::store_heap_oop(Register d, Register s1, Register s2) { if (UseCompressedOops) { assert(s1 != d && s2 != d, "not enough registers");
--- a/src/cpu/sparc/vm/assembler_sparc.hpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/cpu/sparc/vm/assembler_sparc.hpp Fri Oct 15 15:12:04 2010 -0400 @@ -825,6 +825,12 @@ // test if -4096 <= x <= 4095 static bool is_simm13(int x) { return is_simm(x, 13); } + // test if label is in simm16 range in words (wdisp16). + bool is_in_wdisp16_range(Label& L) { + intptr_t d = intptr_t(pc()) - intptr_t(target(L)); + return is_simm(d, 18); + } + enum ASIs { // page 72, v9 ASI_PRIMARY = 0x80, ASI_PRIMARY_LITTLE = 0x88 @@ -2103,6 +2109,7 @@ void load_heap_oop(const Address& s, Register d); void load_heap_oop(Register s1, Register s2, Register d); void load_heap_oop(Register s1, int simm13a, Register d); + void load_heap_oop(Register s1, RegisterOrConstant s2, Register d); void store_heap_oop(Register d, Register s1, Register s2); void store_heap_oop(Register d, Register s1, int simm13a); void store_heap_oop(Register d, const Address& a, int offset = 0); @@ -2225,7 +2232,7 @@ void stop(const char* msg); // prints msg, dumps registers and stops execution void warn(const char* msg); // prints msg, but don't stop void untested(const char* what = ""); - void unimplemented(const char* what = "") { char* b = new char[1024]; sprintf(b, "unimplemented: %s", what); stop(b); } + void unimplemented(const char* what = "") { char* b = new char[1024]; jio_snprintf(b, 1024, "unimplemented: %s", what); stop(b); } void should_not_reach_here() { stop("should not reach here"); } void print_CPU_state();
--- a/src/cpu/sparc/vm/c1_CodeStubs_sparc.cpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/cpu/sparc/vm/c1_CodeStubs_sparc.cpp Fri Oct 15 15:12:04 2010 -0400 @@ -425,8 +425,13 @@ Register pre_val_reg = pre_val()->as_register(); ce->mem2reg(addr(), pre_val(), T_OBJECT, patch_code(), info(), false); - __ br_on_reg_cond(Assembler::rc_z, /*annul*/false, Assembler::pt, - pre_val_reg, _continuation); + if (__ is_in_wdisp16_range(_continuation)) { + __ br_on_reg_cond(Assembler::rc_z, /*annul*/false, Assembler::pt, + pre_val_reg, _continuation); + } else { + __ cmp(pre_val_reg, G0); + __ brx(Assembler::equal, false, Assembler::pn, _continuation); + } __ delayed()->nop(); __ call(Runtime1::entry_for(Runtime1::Runtime1::g1_pre_barrier_slow_id)); @@ -452,8 +457,13 @@ assert(new_val()->is_register(), "Precondition."); Register addr_reg = addr()->as_pointer_register(); Register new_val_reg = new_val()->as_register(); - __ br_on_reg_cond(Assembler::rc_z, /*annul*/false, Assembler::pt, - new_val_reg, _continuation); + if (__ is_in_wdisp16_range(_continuation)) { + __ br_on_reg_cond(Assembler::rc_z, /*annul*/false, Assembler::pt, + new_val_reg, _continuation); + } else { + __ cmp(new_val_reg, G0); + __ brx(Assembler::equal, false, Assembler::pn, _continuation); + } __ delayed()->nop(); __ call(Runtime1::entry_for(Runtime1::Runtime1::g1_post_barrier_slow_id));
--- a/src/cpu/sparc/vm/c1_LIRGenerator_sparc.cpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/cpu/sparc/vm/c1_LIRGenerator_sparc.cpp Fri Oct 15 15:12:04 2010 -0400 @@ -664,7 +664,7 @@ // Use temps to avoid kills LIR_Opr t1 = FrameMap::G1_opr; LIR_Opr t2 = FrameMap::G3_opr; - LIR_Opr addr = new_pointer_register(); + LIR_Opr addr = (type == objectType) ? new_register(T_OBJECT) : new_pointer_register(); // get address of field obj.load_item();
--- a/src/cpu/sparc/vm/c1_LinearScan_sparc.hpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/cpu/sparc/vm/c1_LinearScan_sparc.hpp Fri Oct 15 15:12:04 2010 -0400 @@ -64,7 +64,7 @@ _first_reg = pd_first_callee_saved_reg; _last_reg = pd_last_callee_saved_reg; return true; - } else if (cur->type() == T_INT || cur->type() == T_LONG || cur->type() == T_OBJECT) { + } else if (cur->type() == T_INT || cur->type() == T_LONG || cur->type() == T_OBJECT || cur->type() == T_ADDRESS) { _first_reg = pd_first_cpu_reg; _last_reg = pd_last_allocatable_cpu_reg; return true;
--- a/src/cpu/sparc/vm/methodHandles_sparc.cpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/cpu/sparc/vm/methodHandles_sparc.cpp Fri Oct 15 15:12:04 2010 -0400 @@ -27,6 +27,14 @@ #define __ _masm-> +#ifdef PRODUCT +#define BLOCK_COMMENT(str) /* nothing */ +#else +#define BLOCK_COMMENT(str) __ block_comment(str) +#endif + +#define BIND(label) bind(label); BLOCK_COMMENT(#label ":") + address MethodHandleEntry::start_compiled_entry(MacroAssembler* _masm, address interpreted_entry) { // Just before the actual machine code entry point, allocate space @@ -90,8 +98,8 @@ } // given the MethodType, find out where the MH argument is buried - __ ld_ptr(Address(G5_method_type, __ delayed_value(java_dyn_MethodType::form_offset_in_bytes, O1_scratch)), O0_argslot); - __ ldsw( Address(O0_argslot, __ delayed_value(java_dyn_MethodTypeForm::vmslots_offset_in_bytes, O1_scratch)), O0_argslot); + __ load_heap_oop(Address(G5_method_type, __ delayed_value(java_dyn_MethodType::form_offset_in_bytes, O1_scratch)), O0_argslot); + __ ldsw( Address(O0_argslot, __ delayed_value(java_dyn_MethodTypeForm::vmslots_offset_in_bytes, O1_scratch)), O0_argslot); __ ld_ptr(__ argument_address(O0_argslot), G3_method_handle); __ check_method_handle_type(G5_method_type, G3_method_handle, O1_scratch, wrong_method_type); @@ -105,6 +113,7 @@ static void verify_argslot(MacroAssembler* _masm, Register argslot_reg, Register temp_reg, const char* error_message) { // Verify that argslot lies within (Gargs, FP]. Label L_ok, L_bad; + BLOCK_COMMENT("{ verify_argslot"); #ifdef _LP64 __ add(FP, STACK_BIAS, temp_reg); __ cmp(argslot_reg, temp_reg); @@ -119,6 +128,7 @@ __ bind(L_bad); __ stop(error_message); __ bind(L_ok); + BLOCK_COMMENT("} verify_argslot"); } #endif @@ -175,6 +185,7 @@ // for (temp = sp + size; temp < argslot; temp++) // temp[-size] = temp[0] // argslot -= size; + BLOCK_COMMENT("insert_arg_slots {"); RegisterOrConstant offset = __ regcon_sll_ptr(arg_slots, LogBytesPerWord, temp3_reg); // Keep the stack pointer 2*wordSize aligned. @@ -187,7 +198,7 @@ { Label loop; - __ bind(loop); + __ BIND(loop); // pull one word down each time through the loop __ ld_ptr(Address(temp_reg, 0), temp2_reg); __ st_ptr(temp2_reg, Address(temp_reg, offset)); @@ -199,6 +210,7 @@ // Now move the argslot down, to point to the opened-up space. __ add(argslot_reg, offset, argslot_reg); + BLOCK_COMMENT("} insert_arg_slots"); } @@ -235,6 +247,7 @@ } #endif // ASSERT + BLOCK_COMMENT("remove_arg_slots {"); // Pull up everything shallower than argslot. // Then remove the excess space on the stack. // The stacked return address gets pulled up with everything else. @@ -246,7 +259,7 @@ __ sub(argslot_reg, wordSize, temp_reg); // source pointer for copy { Label loop; - __ bind(loop); + __ BIND(loop); // pull one word up each time through the loop __ ld_ptr(Address(temp_reg, 0), temp2_reg); __ st_ptr(temp2_reg, Address(temp_reg, offset)); @@ -265,29 +278,35 @@ const int TwoWordAlignmentMask = right_n_bits(LogBytesPerWord + 1); RegisterOrConstant masked_offset = __ regcon_andn_ptr(offset, TwoWordAlignmentMask, temp_reg); __ add(SP, masked_offset, SP); + BLOCK_COMMENT("} remove_arg_slots"); } #ifndef PRODUCT extern "C" void print_method_handle(oop mh); void trace_method_handle_stub(const char* adaptername, - oop mh) { -#if 0 - intptr_t* entry_sp, - intptr_t* saved_sp, - intptr_t* saved_bp) { - // called as a leaf from native code: do not block the JVM! - intptr_t* last_sp = (intptr_t*) saved_bp[frame::interpreter_frame_last_sp_offset]; - intptr_t* base_sp = (intptr_t*) saved_bp[frame::interpreter_frame_monitor_block_top_offset]; - printf("MH %s mh="INTPTR_FORMAT" sp=("INTPTR_FORMAT"+"INTX_FORMAT") stack_size="INTX_FORMAT" bp="INTPTR_FORMAT"\n", - adaptername, (intptr_t)mh, (intptr_t)entry_sp, (intptr_t)(saved_sp - entry_sp), (intptr_t)(base_sp - last_sp), (intptr_t)saved_bp); - if (last_sp != saved_sp) - printf("*** last_sp="INTPTR_FORMAT"\n", (intptr_t)last_sp); -#endif - + oopDesc* mh) { printf("MH %s mh="INTPTR_FORMAT"\n", adaptername, (intptr_t) mh); print_method_handle(mh); } +void MethodHandles::trace_method_handle(MacroAssembler* _masm, const char* adaptername) { + if (!TraceMethodHandles) return; + BLOCK_COMMENT("trace_method_handle {"); + // save: Gargs, O5_savedSP + __ save_frame(16); + __ set((intptr_t) adaptername, O0); + __ mov(G3_method_handle, O1); + __ mov(G3_method_handle, L3); + __ mov(Gargs, L4); + __ mov(G5_method_type, L5); + __ call_VM_leaf(L7, CAST_FROM_FN_PTR(address, trace_method_handle_stub)); + + __ mov(L3, G3_method_handle); + __ mov(L4, Gargs); + __ mov(L5, G5_method_type); + __ restore(); + BLOCK_COMMENT("} trace_method_handle"); +} #endif // PRODUCT // which conversion op types are implemented here? @@ -348,18 +367,8 @@ } address interp_entry = __ pc(); - if (UseCompressedOops) __ unimplemented("UseCompressedOops"); -#ifndef PRODUCT - if (TraceMethodHandles) { - // save: Gargs, O5_savedSP - __ save(SP, -16*wordSize, SP); - __ set((intptr_t) entry_name(ek), O0); - __ mov(G3_method_handle, O1); - __ call_VM_leaf(Lscratch, CAST_FROM_FN_PTR(address, trace_method_handle_stub)); - __ restore(SP, 16*wordSize, SP); - } -#endif // PRODUCT + trace_method_handle(_masm, entry_name(ek)); switch ((int) ek) { case _raise_exception: @@ -413,7 +422,7 @@ case _invokestatic_mh: case _invokespecial_mh: { - __ ld_ptr(G3_mh_vmtarget, G5_method); // target is a methodOop + __ load_heap_oop(G3_mh_vmtarget, G5_method); // target is a methodOop __ verify_oop(G5_method); // Same as TemplateTable::invokestatic or invokespecial, // minus the CP setup and profiling: @@ -468,7 +477,7 @@ // minus the CP setup and profiling: __ load_method_handle_vmslots(O0_argslot, G3_method_handle, O1_scratch); Register O1_intf = O1_scratch; - __ ld_ptr(G3_mh_vmtarget, O1_intf); + __ load_heap_oop(G3_mh_vmtarget, O1_intf); __ ldsw(G3_dmh_vmindex, G5_index); __ ld_ptr(__ argument_address(O0_argslot, -1), G3_method_handle); __ null_check(G3_method_handle, oopDesc::klass_offset_in_bytes()); @@ -523,7 +532,7 @@ insert_arg_slots(_masm, arg_slots * stack_move_unit(), arg_mask, O0_argslot, O1_scratch, O2_scratch, G5_index); // Store bound argument into the new stack slot: - __ ld_ptr(G3_bmh_argument, O1_scratch); + __ load_heap_oop(G3_bmh_argument, O1_scratch); if (arg_type == T_OBJECT) { __ st_ptr(O1_scratch, Address(O0_argslot, 0)); } else { @@ -541,12 +550,12 @@ } if (direct_to_method) { - __ ld_ptr(G3_mh_vmtarget, G5_method); // target is a methodOop + __ load_heap_oop(G3_mh_vmtarget, G5_method); // target is a methodOop __ verify_oop(G5_method); __ jump_indirect_to(G5_method_fie, O1_scratch); __ delayed()->nop(); } else { - __ ld_ptr(G3_mh_vmtarget, G3_method_handle); // target is a methodOop + __ load_heap_oop(G3_mh_vmtarget, G3_method_handle); // target is a methodOop __ verify_oop(G3_method_handle); __ jump_to_method_handle_entry(G3_method_handle, O1_scratch); } @@ -556,7 +565,7 @@ case _adapter_retype_only: case _adapter_retype_raw: // Immediately jump to the next MH layer: - __ ld_ptr(G3_mh_vmtarget, G3_method_handle); + __ load_heap_oop(G3_mh_vmtarget, G3_method_handle); __ jump_to_method_handle_entry(G3_method_handle, O1_scratch); // This is OK when all parameter types widen. // It is also OK when a return type narrows. @@ -572,8 +581,8 @@ Address vmarg = __ argument_address(O0_argslot); // What class are we casting to? - __ ld_ptr(G3_amh_argument, G5_klass); // This is a Class object! - __ ld_ptr(Address(G5_klass, java_lang_Class::klass_offset_in_bytes()), G5_klass); + __ load_heap_oop(G3_amh_argument, G5_klass); // This is a Class object! + __ load_heap_oop(Address(G5_klass, java_lang_Class::klass_offset_in_bytes()), G5_klass); Label done; __ ld_ptr(vmarg, O1_scratch); @@ -590,14 +599,14 @@ // If we get here, the type check failed! __ ldsw(G3_amh_vmargslot, O0_argslot); // reload argslot field - __ ld_ptr(G3_amh_argument, O3_scratch); // required class + __ load_heap_oop(G3_amh_argument, O3_scratch); // required class __ ld_ptr(vmarg, O2_scratch); // bad object __ jump_to(AddressLiteral(from_interpreted_entry(_raise_exception)), O0_argslot); __ delayed()->mov(Bytecodes::_checkcast, O1_scratch); // who is complaining? __ bind(done); // Get the new MH: - __ ld_ptr(G3_mh_vmtarget, G3_method_handle); + __ load_heap_oop(G3_mh_vmtarget, G3_method_handle); __ jump_to_method_handle_entry(G3_method_handle, O1_scratch); } break; @@ -676,7 +685,7 @@ __ st(O1_scratch, vmarg); // Get the new MH: - __ ld_ptr(G3_mh_vmtarget, G3_method_handle); + __ load_heap_oop(G3_mh_vmtarget, G3_method_handle); __ jump_to_method_handle_entry(G3_method_handle, O1_scratch); } break; @@ -721,7 +730,7 @@ ShouldNotReachHere(); } - __ ld_ptr(G3_mh_vmtarget, G3_method_handle); + __ load_heap_oop(G3_mh_vmtarget, G3_method_handle); __ jump_to_method_handle_entry(G3_method_handle, O1_scratch); } break; @@ -851,7 +860,7 @@ } } - __ ld_ptr(G3_mh_vmtarget, G3_method_handle); + __ load_heap_oop(G3_mh_vmtarget, G3_method_handle); __ jump_to_method_handle_entry(G3_method_handle, O1_scratch); } break; @@ -895,7 +904,7 @@ __ brx(Assembler::less, false, Assembler::pt, loop); __ delayed()->nop(); // FILLME - __ ld_ptr(G3_mh_vmtarget, G3_method_handle); + __ load_heap_oop(G3_mh_vmtarget, G3_method_handle); __ jump_to_method_handle_entry(G3_method_handle, O1_scratch); } break; @@ -913,7 +922,7 @@ remove_arg_slots(_masm, G5_stack_move, O0_argslot, O1_scratch, O2_scratch, O3_scratch); - __ ld_ptr(G3_mh_vmtarget, G3_method_handle); + __ load_heap_oop(G3_mh_vmtarget, G3_method_handle); __ jump_to_method_handle_entry(G3_method_handle, O1_scratch); } break;
--- a/src/cpu/sparc/vm/stubGenerator_sparc.cpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/cpu/sparc/vm/stubGenerator_sparc.cpp Fri Oct 15 15:12:04 2010 -0400 @@ -2586,6 +2586,8 @@ __ restore(); #endif + assert_clean_int(O2_count, G1); // Make sure 'count' is clean int. + #ifdef ASSERT // caller guarantees that the arrays really are different // otherwise, we would have to make conjoint checks @@ -2600,8 +2602,6 @@ } #endif //ASSERT - assert_clean_int(O2_count, G1); // Make sure 'count' is clean int. - checkcast_copy_entry = __ pc(); // caller can pass a 64-bit byte count here (from generic stub) BLOCK_COMMENT("Entry:");
--- a/src/cpu/sparc/vm/stubRoutines_sparc.hpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/cpu/sparc/vm/stubRoutines_sparc.hpp Fri Oct 15 15:12:04 2010 -0400 @@ -43,7 +43,7 @@ // MethodHandles adapters enum method_handles_platform_dependent_constants { - method_handles_adapters_code_size = 12000 + method_handles_adapters_code_size = 15000 }; class Sparc {
--- a/src/cpu/sparc/vm/templateTable_sparc.cpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/cpu/sparc/vm/templateTable_sparc.cpp Fri Oct 15 15:12:04 2010 -0400 @@ -3273,7 +3273,7 @@ __ sll(Rret, LogBytesPerWord, Rret); __ ld_ptr(Rtemp, Rret, Rret); // get return address - __ ld_ptr(G5_callsite, __ delayed_value(java_dyn_CallSite::target_offset_in_bytes, Rscratch), G3_method_handle); + __ load_heap_oop(G5_callsite, __ delayed_value(java_dyn_CallSite::target_offset_in_bytes, Rscratch), G3_method_handle); __ null_check(G3_method_handle); // Adjust Rret first so Llast_SP can be same as Rret
--- a/src/cpu/x86/vm/assembler_x86.cpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/cpu/x86/vm/assembler_x86.cpp Fri Oct 15 15:12:04 2010 -0400 @@ -7709,9 +7709,14 @@ void MacroAssembler::check_method_handle_type(Register mtype_reg, Register mh_reg, Register temp_reg, Label& wrong_method_type) { - if (UseCompressedOops) unimplemented(); // field accesses must decode + Address type_addr(mh_reg, delayed_value(java_dyn_MethodHandle::type_offset_in_bytes, temp_reg)); // compare method type against that of the receiver - cmpptr(mtype_reg, Address(mh_reg, delayed_value(java_dyn_MethodHandle::type_offset_in_bytes, temp_reg))); + if (UseCompressedOops) { + load_heap_oop(temp_reg, type_addr); + cmpptr(mtype_reg, temp_reg); + } else { + cmpptr(mtype_reg, type_addr); + } jcc(Assembler::notEqual, wrong_method_type); } @@ -7723,15 +7728,14 @@ void MacroAssembler::load_method_handle_vmslots(Register vmslots_reg, Register mh_reg, Register temp_reg) { assert_different_registers(vmslots_reg, mh_reg, temp_reg); - if (UseCompressedOops) unimplemented(); // field accesses must decode // load mh.type.form.vmslots if (java_dyn_MethodHandle::vmslots_offset_in_bytes() != 0) { // hoist vmslots into every mh to avoid dependent load chain movl(vmslots_reg, Address(mh_reg, delayed_value(java_dyn_MethodHandle::vmslots_offset_in_bytes, temp_reg))); } else { Register temp2_reg = vmslots_reg; - movptr(temp2_reg, Address(mh_reg, delayed_value(java_dyn_MethodHandle::type_offset_in_bytes, temp_reg))); - movptr(temp2_reg, Address(temp2_reg, delayed_value(java_dyn_MethodType::form_offset_in_bytes, temp_reg))); + load_heap_oop(temp2_reg, Address(mh_reg, delayed_value(java_dyn_MethodHandle::type_offset_in_bytes, temp_reg))); + load_heap_oop(temp2_reg, Address(temp2_reg, delayed_value(java_dyn_MethodType::form_offset_in_bytes, temp_reg))); movl(vmslots_reg, Address(temp2_reg, delayed_value(java_dyn_MethodTypeForm::vmslots_offset_in_bytes, temp_reg))); } } @@ -7745,9 +7749,8 @@ assert(mh_reg == rcx, "caller must put MH object in rcx"); assert_different_registers(mh_reg, temp_reg); - if (UseCompressedOops) unimplemented(); // field accesses must decode - // pick out the interpreted side of the handler + // NOTE: vmentry is not an oop! movptr(temp_reg, Address(mh_reg, delayed_value(java_dyn_MethodHandle::vmentry_offset_in_bytes, temp_reg))); // off we go... @@ -8238,6 +8241,40 @@ movptr(Address(dst, oopDesc::klass_offset_in_bytes()), src); } +void MacroAssembler::load_heap_oop(Register dst, Address src) { +#ifdef _LP64 + if (UseCompressedOops) { + movl(dst, src); + decode_heap_oop(dst); + } else +#endif + movptr(dst, src); +} + +void MacroAssembler::store_heap_oop(Address dst, Register src) { +#ifdef _LP64 + if (UseCompressedOops) { + assert(!dst.uses(src), "not enough registers"); + encode_heap_oop(src); + movl(dst, src); + } else +#endif + movptr(dst, src); +} + +// Used for storing NULLs. +void MacroAssembler::store_heap_oop_null(Address dst) { +#ifdef _LP64 + if (UseCompressedOops) { + movl(dst, (int32_t)NULL_WORD); + } else { + movslq(dst, (int32_t)NULL_WORD); + } +#else + movl(dst, (int32_t)NULL_WORD); +#endif +} + #ifdef _LP64 void MacroAssembler::store_klass_gap(Register dst, Register src) { if (UseCompressedOops) { @@ -8246,34 +8283,6 @@ } } -void MacroAssembler::load_heap_oop(Register dst, Address src) { - if (UseCompressedOops) { - movl(dst, src); - decode_heap_oop(dst); - } else { - movq(dst, src); - } -} - -void MacroAssembler::store_heap_oop(Address dst, Register src) { - if (UseCompressedOops) { - assert(!dst.uses(src), "not enough registers"); - encode_heap_oop(src); - movl(dst, src); - } else { - movq(dst, src); - } -} - -// Used for storing NULLs. -void MacroAssembler::store_heap_oop_null(Address dst) { - if (UseCompressedOops) { - movl(dst, (int32_t)NULL_WORD); - } else { - movslq(dst, (int32_t)NULL_WORD); - } -} - #ifdef ASSERT void MacroAssembler::verify_heapbase(const char* msg) { assert (UseCompressedOops, "should be compressed");
--- a/src/cpu/x86/vm/assembler_x86.hpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/cpu/x86/vm/assembler_x86.hpp Fri Oct 15 15:12:04 2010 -0400 @@ -1682,24 +1682,24 @@ void load_klass(Register dst, Register src); void store_klass(Register dst, Register src); + void load_heap_oop(Register dst, Address src); + void store_heap_oop(Address dst, Register src); + + // Used for storing NULL. All other oop constants should be + // stored using routines that take a jobject. + void store_heap_oop_null(Address dst); + void load_prototype_header(Register dst, Register src); #ifdef _LP64 void store_klass_gap(Register dst, Register src); - void load_heap_oop(Register dst, Address src); - void store_heap_oop(Address dst, Register src); - // This dummy is to prevent a call to store_heap_oop from // converting a zero (like NULL) into a Register by giving // the compiler two choices it can't resolve void store_heap_oop(Address dst, void* dummy); - // Used for storing NULL. All other oop constants should be - // stored using routines that take a jobject. - void store_heap_oop_null(Address dst); - void encode_heap_oop(Register r); void decode_heap_oop(Register r); void encode_heap_oop_not_null(Register r); @@ -1927,7 +1927,7 @@ void untested() { stop("untested"); } - void unimplemented(const char* what = "") { char* b = new char[1024]; jio_snprintf(b, sizeof(b), "unimplemented: %s", what); stop(b); } + void unimplemented(const char* what = "") { char* b = new char[1024]; jio_snprintf(b, 1024, "unimplemented: %s", what); stop(b); } void should_not_reach_here() { stop("should not reach here"); }
--- a/src/cpu/x86/vm/c1_LIRAssembler_x86.cpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/cpu/x86/vm/c1_LIRAssembler_x86.cpp Fri Oct 15 15:12:04 2010 -0400 @@ -1941,8 +1941,6 @@ __ cmpxchgptr(newval, Address(addr, 0)); } else if (op->code() == lir_cas_int) { __ cmpxchgl(newval, Address(addr, 0)); - } else { - LP64_ONLY(__ cmpxchgq(newval, Address(addr, 0))); } #ifdef _LP64 } else if (op->code() == lir_cas_long) {
--- a/src/cpu/x86/vm/c1_LIRGenerator_x86.cpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/cpu/x86/vm/c1_LIRGenerator_x86.cpp Fri Oct 15 15:12:04 2010 -0400 @@ -765,7 +765,7 @@ ShouldNotReachHere(); } - LIR_Opr addr = new_pointer_register(); + LIR_Opr addr = (type == objectType) ? new_register(T_OBJECT) : new_pointer_register(); LIR_Address* a; if(offset.result()->is_constant()) { a = new LIR_Address(obj.result(),
--- a/src/cpu/x86/vm/methodHandles_x86.cpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/cpu/x86/vm/methodHandles_x86.cpp Fri Oct 15 15:12:04 2010 -0400 @@ -123,11 +123,9 @@ } // given the MethodType, find out where the MH argument is buried - __ movptr(rdx_temp, Address(rax_mtype, - __ delayed_value(java_dyn_MethodType::form_offset_in_bytes, rdi_temp))); + __ load_heap_oop(rdx_temp, Address(rax_mtype, __ delayed_value(java_dyn_MethodType::form_offset_in_bytes, rdi_temp))); Register rdx_vmslots = rdx_temp; - __ movl(rdx_vmslots, Address(rdx_temp, - __ delayed_value(java_dyn_MethodTypeForm::vmslots_offset_in_bytes, rdi_temp))); + __ movl(rdx_vmslots, Address(rdx_temp, __ delayed_value(java_dyn_MethodTypeForm::vmslots_offset_in_bytes, rdi_temp))); __ movptr(rcx_recv, __ argument_address(rdx_vmslots)); trace_method_handle(_masm, "invokeExact"); @@ -154,20 +152,18 @@ rcx_argslot, rbx_temp, rdx_temp); // load up an adapter from the calling type (Java weaves this) - __ movptr(rdx_temp, Address(rax_mtype, - __ delayed_value(java_dyn_MethodType::form_offset_in_bytes, rdi_temp))); + __ load_heap_oop(rdx_temp, Address(rax_mtype, __ delayed_value(java_dyn_MethodType::form_offset_in_bytes, rdi_temp))); Register rdx_adapter = rdx_temp; - // movptr(rdx_adapter, Address(rdx_temp, java_dyn_MethodTypeForm::genericInvoker_offset_in_bytes())); + // __ load_heap_oop(rdx_adapter, Address(rdx_temp, java_dyn_MethodTypeForm::genericInvoker_offset_in_bytes())); // deal with old JDK versions: - __ lea(rdi_temp, Address(rdx_temp, - __ delayed_value(java_dyn_MethodTypeForm::genericInvoker_offset_in_bytes, rdi_temp))); + __ lea(rdi_temp, Address(rdx_temp, __ delayed_value(java_dyn_MethodTypeForm::genericInvoker_offset_in_bytes, rdi_temp))); __ cmpptr(rdi_temp, rdx_temp); Label sorry_no_invoke_generic; - __ jccb(Assembler::below, sorry_no_invoke_generic); + __ jcc(Assembler::below, sorry_no_invoke_generic); - __ movptr(rdx_adapter, Address(rdi_temp, 0)); + __ load_heap_oop(rdx_adapter, Address(rdi_temp, 0)); __ testptr(rdx_adapter, rdx_adapter); - __ jccb(Assembler::zero, sorry_no_invoke_generic); + __ jcc(Assembler::zero, sorry_no_invoke_generic); __ movptr(Address(rcx_argslot, 1 * Interpreter::stackElementSize), rdx_adapter); // As a trusted first argument, pass the type being called, so the adapter knows // the actual types of the arguments and return values. @@ -431,7 +427,6 @@ } address interp_entry = __ pc(); - if (UseCompressedOops) __ unimplemented("UseCompressedOops"); trace_method_handle(_masm, entry_name(ek)); @@ -489,7 +484,7 @@ case _invokespecial_mh: { Register rbx_method = rbx_temp; - __ movptr(rbx_method, rcx_mh_vmtarget); // target is a methodOop + __ load_heap_oop(rbx_method, rcx_mh_vmtarget); // target is a methodOop __ verify_oop(rbx_method); // same as TemplateTable::invokestatic or invokespecial, // minus the CP setup and profiling: @@ -546,8 +541,8 @@ __ load_method_handle_vmslots(rax_argslot, rcx_recv, rdx_temp); Register rdx_intf = rdx_temp; Register rbx_index = rbx_temp; - __ movptr(rdx_intf, rcx_mh_vmtarget); - __ movl(rbx_index, rcx_dmh_vmindex); + __ load_heap_oop(rdx_intf, rcx_mh_vmtarget); + __ movl(rbx_index, rcx_dmh_vmindex); __ movptr(rcx_recv, __ argument_address(rax_argslot, -1)); __ null_check(rcx_recv, oopDesc::klass_offset_in_bytes()); @@ -602,7 +597,7 @@ rax_argslot, rbx_temp, rdx_temp); // store bound argument into the new stack slot: - __ movptr(rbx_temp, rcx_bmh_argument); + __ load_heap_oop(rbx_temp, rcx_bmh_argument); Address prim_value_addr(rbx_temp, java_lang_boxing_object::value_offset_in_bytes(arg_type)); if (arg_type == T_OBJECT) { __ movptr(Address(rax_argslot, 0), rbx_temp); @@ -620,11 +615,11 @@ if (direct_to_method) { Register rbx_method = rbx_temp; - __ movptr(rbx_method, rcx_mh_vmtarget); + __ load_heap_oop(rbx_method, rcx_mh_vmtarget); __ verify_oop(rbx_method); __ jmp(rbx_method_fie); } else { - __ movptr(rcx_recv, rcx_mh_vmtarget); + __ load_heap_oop(rcx_recv, rcx_mh_vmtarget); __ verify_oop(rcx_recv); __ jump_to_method_handle_entry(rcx_recv, rdx_temp); } @@ -634,7 +629,7 @@ case _adapter_retype_only: case _adapter_retype_raw: // immediately jump to the next MH layer: - __ movptr(rcx_recv, rcx_mh_vmtarget); + __ load_heap_oop(rcx_recv, rcx_mh_vmtarget); __ verify_oop(rcx_recv); __ jump_to_method_handle_entry(rcx_recv, rdx_temp); // This is OK when all parameter types widen. @@ -651,13 +646,13 @@ vmarg = __ argument_address(rax_argslot); // What class are we casting to? - __ movptr(rbx_klass, rcx_amh_argument); // this is a Class object! - __ movptr(rbx_klass, Address(rbx_klass, java_lang_Class::klass_offset_in_bytes())); + __ load_heap_oop(rbx_klass, rcx_amh_argument); // this is a Class object! + __ load_heap_oop(rbx_klass, Address(rbx_klass, java_lang_Class::klass_offset_in_bytes())); Label done; __ movptr(rdx_temp, vmarg); __ testptr(rdx_temp, rdx_temp); - __ jccb(Assembler::zero, done); // no cast if null + __ jcc(Assembler::zero, done); // no cast if null __ load_klass(rdx_temp, rdx_temp); // live at this point: @@ -672,14 +667,15 @@ __ movl(rax_argslot, rcx_amh_vmargslot); // reload argslot field __ movptr(rdx_temp, vmarg); - __ pushptr(rcx_amh_argument); // required class - __ push(rdx_temp); // bad object - __ push((int)Bytecodes::_checkcast); // who is complaining? + __ load_heap_oop(rbx_klass, rcx_amh_argument); // required class + __ push(rbx_klass); + __ push(rdx_temp); // bad object + __ push((int)Bytecodes::_checkcast); // who is complaining? __ jump(ExternalAddress(from_interpreted_entry(_raise_exception))); __ bind(done); // get the new MH: - __ movptr(rcx_recv, rcx_mh_vmtarget); + __ load_heap_oop(rcx_recv, rcx_mh_vmtarget); __ jump_to_method_handle_entry(rcx_recv, rdx_temp); } break; @@ -741,7 +737,7 @@ assert(CONV_VMINFO_SHIFT == 0, "preshifted"); // get the new MH: - __ movptr(rcx_recv, rcx_mh_vmtarget); + __ load_heap_oop(rcx_recv, rcx_mh_vmtarget); // (now we are done with the old MH) // original 32-bit vmdata word must be of this form: @@ -816,7 +812,7 @@ ShouldNotReachHere(); } - __ movptr(rcx_recv, rcx_mh_vmtarget); + __ load_heap_oop(rcx_recv, rcx_mh_vmtarget); __ jump_to_method_handle_entry(rcx_recv, rdx_temp); } break; @@ -858,7 +854,7 @@ rax_argslot, rbx_temp, rdx_temp); } - __ movptr(rcx_recv, rcx_mh_vmtarget); + __ load_heap_oop(rcx_recv, rcx_mh_vmtarget); __ jump_to_method_handle_entry(rcx_recv, rdx_temp); } break; @@ -969,7 +965,7 @@ } } - __ movptr(rcx_recv, rcx_mh_vmtarget); + __ load_heap_oop(rcx_recv, rcx_mh_vmtarget); __ jump_to_method_handle_entry(rcx_recv, rdx_temp); } break; @@ -1029,7 +1025,7 @@ __ pop(rdi); // restore temp - __ movptr(rcx_recv, rcx_mh_vmtarget); + __ load_heap_oop(rcx_recv, rcx_mh_vmtarget); __ jump_to_method_handle_entry(rcx_recv, rdx_temp); } break; @@ -1052,7 +1048,7 @@ __ pop(rdi); // restore temp - __ movptr(rcx_recv, rcx_mh_vmtarget); + __ load_heap_oop(rcx_recv, rcx_mh_vmtarget); __ jump_to_method_handle_entry(rcx_recv, rdx_temp); } break; @@ -1103,8 +1099,8 @@ // Check the array type. Register rbx_klass = rbx_temp; - __ movptr(rbx_klass, rcx_amh_argument); // this is a Class object! - __ movptr(rbx_klass, Address(rbx_klass, java_lang_Class::klass_offset_in_bytes())); + __ load_heap_oop(rbx_klass, rcx_amh_argument); // this is a Class object! + __ load_heap_oop(rbx_klass, Address(rbx_klass, java_lang_Class::klass_offset_in_bytes())); Label ok_array_klass, bad_array_klass, bad_array_length; __ check_klass_subtype(rdx_array_klass, rbx_klass, rdi, ok_array_klass); @@ -1186,7 +1182,7 @@ // Arguments are spread. Move to next method handle. UNPUSH_RSI_RDI; - __ movptr(rcx_recv, rcx_mh_vmtarget); + __ load_heap_oop(rcx_recv, rcx_mh_vmtarget); __ jump_to_method_handle_entry(rcx_recv, rdx_temp); __ bind(bad_array_klass);
--- a/src/cpu/x86/vm/stubRoutines_x86_64.hpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/cpu/x86/vm/stubRoutines_x86_64.hpp Fri Oct 15 15:12:04 2010 -0400 @@ -35,7 +35,7 @@ // MethodHandles adapters enum method_handles_platform_dependent_constants { - method_handles_adapters_code_size = 26000 + method_handles_adapters_code_size = 40000 }; class x86 {
--- a/src/cpu/x86/vm/templateTable_x86_32.cpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/cpu/x86/vm/templateTable_x86_32.cpp Fri Oct 15 15:12:04 2010 -0400 @@ -3111,19 +3111,22 @@ // rax: CallSite object (f1) // rbx: unused (f2) + // rcx: receiver address // rdx: flags (unused) + Register rax_callsite = rax; + Register rcx_method_handle = rcx; + if (ProfileInterpreter) { - Label L; // %%% should make a type profile for any invokedynamic that takes a ref argument // profile this call __ profile_call(rsi); } - __ movptr(rcx, Address(rax, __ delayed_value(java_dyn_CallSite::target_offset_in_bytes, rcx))); - __ null_check(rcx); + __ movptr(rcx_method_handle, Address(rax_callsite, __ delayed_value(java_dyn_CallSite::target_offset_in_bytes, rcx))); + __ null_check(rcx_method_handle); __ prepare_to_jump_from_interpreted(); - __ jump_to_method_handle_entry(rcx, rdx); + __ jump_to_method_handle_entry(rcx_method_handle, rdx); } //----------------------------------------------------------------------------------------------------
--- a/src/cpu/x86/vm/templateTable_x86_64.cpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/cpu/x86/vm/templateTable_x86_64.cpp Fri Oct 15 15:12:04 2010 -0400 @@ -3120,17 +3120,19 @@ // rcx: receiver address // rdx: flags (unused) + Register rax_callsite = rax; + Register rcx_method_handle = rcx; + if (ProfileInterpreter) { - Label L; // %%% should make a type profile for any invokedynamic that takes a ref argument // profile this call __ profile_call(r13); } - __ movptr(rcx, Address(rax, __ delayed_value(java_dyn_CallSite::target_offset_in_bytes, rcx))); - __ null_check(rcx); + __ load_heap_oop(rcx_method_handle, Address(rax_callsite, __ delayed_value(java_dyn_CallSite::target_offset_in_bytes, rcx))); + __ null_check(rcx_method_handle); __ prepare_to_jump_from_interpreted(); - __ jump_to_method_handle_entry(rcx, rdx); + __ jump_to_method_handle_entry(rcx_method_handle, rdx); }
--- a/src/cpu/zero/vm/interpreterRT_zero.hpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/cpu/zero/vm/interpreterRT_zero.hpp Fri Oct 15 15:12:04 2010 -0400 @@ -92,15 +92,15 @@ public: SignatureHandlerGenerator(methodHandle method, CodeBuffer* buffer) - : SignatureHandlerGeneratorBase(method, (ffi_cif *) buffer->code_end()), + : SignatureHandlerGeneratorBase(method, (ffi_cif *) buffer->insts_end()), _cb(buffer) { - _cb->set_code_end((address) (cif() + 1)); + _cb->set_insts_end((address) (cif() + 1)); } private: void push(intptr_t value) { - intptr_t *dst = (intptr_t *) _cb->code_end(); - _cb->set_code_end((address) (dst + 1)); + intptr_t *dst = (intptr_t *) _cb->insts_end(); + _cb->set_insts_end((address) (dst + 1)); *dst = value; } };
--- a/src/share/vm/asm/codeBuffer.hpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/share/vm/asm/codeBuffer.hpp Fri Oct 15 15:12:04 2010 -0400 @@ -168,8 +168,8 @@ bool allocates(address pc) const { return pc >= _start && pc < _limit; } bool allocates2(address pc) const { return pc >= _start && pc <= _limit; } - void set_end(address pc) { assert(allocates2(pc),""); _end = pc; } - void set_mark(address pc) { assert(contains2(pc),"not in codeBuffer"); + void set_end(address pc) { assert(allocates2(pc), err_msg("not in CodeBuffer memory: " PTR_FORMAT " <= " PTR_FORMAT " <= " PTR_FORMAT, _start, pc, _limit)); _end = pc; } + void set_mark(address pc) { assert(contains2(pc), "not in codeBuffer"); _mark = pc; } void set_mark_off(int offset) { assert(contains2(offset+_start),"not in codeBuffer"); _mark = offset + _start; }
--- a/src/share/vm/c1/c1_LIRGenerator.cpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/share/vm/c1/c1_LIRGenerator.cpp Fri Oct 15 15:12:04 2010 -0400 @@ -1350,6 +1350,7 @@ addr = ptr; } assert(addr->is_register(), "must be a register at this point"); + assert(addr->type() == T_OBJECT, "addr should point to an object"); LIR_Opr xor_res = new_pointer_register(); LIR_Opr xor_shift_res = new_pointer_register();
--- a/src/share/vm/ci/ciInstanceKlass.cpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/share/vm/ci/ciInstanceKlass.cpp Fri Oct 15 15:12:04 2010 -0400 @@ -471,7 +471,7 @@ ciField* field = fields->at(i); int offset = field->offset_in_bytes(); int size = (field->_type == NULL) ? heapOopSize : field->size_in_bytes(); - assert(last_offset <= offset, "no field overlap"); + assert(last_offset <= offset, err_msg("no field overlap: %d <= %d", last_offset, offset)); if (last_offset > (int)sizeof(oopDesc)) assert((offset - last_offset) < BytesPerLong, "no big holes"); // Note: Two consecutive T_BYTE fields will be separated by wordSize-1
--- a/src/share/vm/ci/ciTypeFlow.cpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/share/vm/ci/ciTypeFlow.cpp Fri Oct 15 15:12:04 2010 -0400 @@ -1945,7 +1945,7 @@ _has_irreducible_entry = false; _osr_bci = osr_bci; _failure_reason = NULL; - assert(start_bci() >= 0 && start_bci() < code_size() , "correct osr_bci argument"); + assert(0 <= start_bci() && start_bci() < code_size() , err_msg("correct osr_bci argument: 0 <= %d < %d", start_bci(), code_size())); _work_list = NULL; _ciblock_count = _methodBlocks->num_blocks();
--- a/src/share/vm/classfile/classFileParser.cpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/share/vm/classfile/classFileParser.cpp Fri Oct 15 15:12:04 2010 -0400 @@ -2702,13 +2702,15 @@ // Adjust the field type from byte to an unmanaged pointer. assert(fac_ptr->nonstatic_byte_count > 0, ""); fac_ptr->nonstatic_byte_count -= 1; - (*fields_ptr)->ushort_at_put(i + instanceKlass::signature_index_offset, - word_sig_index); - fac_ptr->nonstatic_word_count += 1; + + (*fields_ptr)->ushort_at_put(i + instanceKlass::signature_index_offset, word_sig_index); + assert(wordSize == longSize || wordSize == jintSize, "ILP32 or LP64"); + if (wordSize == longSize) fac_ptr->nonstatic_double_count += 1; + else fac_ptr->nonstatic_word_count += 1; FieldAllocationType atype = (FieldAllocationType) (*fields_ptr)->ushort_at(i + instanceKlass::low_offset); assert(atype == NONSTATIC_BYTE, ""); - FieldAllocationType new_atype = NONSTATIC_WORD; + FieldAllocationType new_atype = (wordSize == longSize) ? NONSTATIC_DOUBLE : NONSTATIC_WORD; (*fields_ptr)->ushort_at_put(i + instanceKlass::low_offset, new_atype); found_vmentry = true;
--- a/src/share/vm/code/nmethod.cpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/share/vm/code/nmethod.cpp Fri Oct 15 15:12:04 2010 -0400 @@ -1421,7 +1421,7 @@ } #ifdef SHARK - ((SharkCompiler *) compiler())->free_compiled_method(instructions_begin()); + ((SharkCompiler *) compiler())->free_compiled_method(insts_begin()); #endif // SHARK ((CodeBlob*)(this))->flush();
--- a/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp Fri Oct 15 15:12:04 2010 -0400 @@ -791,7 +791,7 @@ int _worker_i; public: RebuildRSOutOfRegionClosure(G1CollectedHeap* g1, int worker_i = 0) : - _cl(g1->g1_rem_set()->as_HRInto_G1RemSet(), worker_i), + _cl(g1->g1_rem_set(), worker_i), _worker_i(worker_i), _g1h(g1) { } @@ -890,7 +890,7 @@ abandon_cur_alloc_region(); abandon_gc_alloc_regions(); assert(_cur_alloc_region == NULL, "Invariant."); - g1_rem_set()->as_HRInto_G1RemSet()->cleanupHRRS(); + g1_rem_set()->cleanupHRRS(); tear_down_region_lists(); set_used_regions_to_need_zero_fill(); @@ -1506,15 +1506,11 @@ } // Also create a G1 rem set. - if (G1UseHRIntoRS) { - if (mr_bs()->is_a(BarrierSet::CardTableModRef)) { - _g1_rem_set = new HRInto_G1RemSet(this, (CardTableModRefBS*)mr_bs()); - } else { - vm_exit_during_initialization("G1 requires a cardtable mod ref bs."); - return JNI_ENOMEM; - } + if (mr_bs()->is_a(BarrierSet::CardTableModRef)) { + _g1_rem_set = new G1RemSet(this, (CardTableModRefBS*)mr_bs()); } else { - _g1_rem_set = new StupidG1RemSet(this); + vm_exit_during_initialization("G1 requires a cardtable mod ref bs."); + return JNI_ENOMEM; } // Carve out the G1 part of the heap. @@ -2706,8 +2702,7 @@ } size_t G1CollectedHeap::cards_scanned() { - HRInto_G1RemSet* g1_rset = (HRInto_G1RemSet*) g1_rem_set(); - return g1_rset->cardsScanned(); + return g1_rem_set()->cardsScanned(); } void @@ -3850,6 +3845,54 @@ undo_waste() * HeapWordSize / K); } +#ifdef ASSERT +bool G1ParScanThreadState::verify_ref(narrowOop* ref) const { + assert(ref != NULL, "invariant"); + assert(UseCompressedOops, "sanity"); + assert(!has_partial_array_mask(ref), err_msg("ref=" PTR_FORMAT, ref)); + oop p = oopDesc::load_decode_heap_oop(ref); + assert(_g1h->is_in_g1_reserved(p), + err_msg("ref=" PTR_FORMAT " p=" PTR_FORMAT, ref, intptr_t(p))); + return true; +} + +bool G1ParScanThreadState::verify_ref(oop* ref) const { + assert(ref != NULL, "invariant"); + if (has_partial_array_mask(ref)) { + // Must be in the collection set--it's already been copied. + oop p = clear_partial_array_mask(ref); + assert(_g1h->obj_in_cs(p), + err_msg("ref=" PTR_FORMAT " p=" PTR_FORMAT, ref, intptr_t(p))); + } else { + oop p = oopDesc::load_decode_heap_oop(ref); + assert(_g1h->is_in_g1_reserved(p), + err_msg("ref=" PTR_FORMAT " p=" PTR_FORMAT, ref, intptr_t(p))); + } + return true; +} + +bool G1ParScanThreadState::verify_task(StarTask ref) const { + if (ref.is_narrow()) { + return verify_ref((narrowOop*) ref); + } else { + return verify_ref((oop*) ref); + } +} +#endif // ASSERT + +void G1ParScanThreadState::trim_queue() { + StarTask ref; + do { + // Drain the overflow stack first, so other threads can steal. + while (refs()->pop_overflow(ref)) { + deal_with_reference(ref); + } + while (refs()->pop_local(ref)) { + deal_with_reference(ref); + } + } while (!refs()->is_empty()); +} + G1ParClosureSuper::G1ParClosureSuper(G1CollectedHeap* g1, G1ParScanThreadState* par_scan_state) : _g1(g1), _g1_rem(_g1->g1_rem_set()), _cm(_g1->concurrent_mark()), _par_scan_state(par_scan_state) { } @@ -4052,38 +4095,39 @@ : _g1h(g1h), _par_scan_state(par_scan_state), _queues(queues), _terminator(terminator) {} - void do_void() { - G1ParScanThreadState* pss = par_scan_state(); - while (true) { + void do_void(); + +private: + inline bool offer_termination(); +}; + +bool G1ParEvacuateFollowersClosure::offer_termination() { + G1ParScanThreadState* const pss = par_scan_state(); + pss->start_term_time(); + const bool res = terminator()->offer_termination(); + pss->end_term_time(); + return res; +} + +void G1ParEvacuateFollowersClosure::do_void() { + StarTask stolen_task; + G1ParScanThreadState* const pss = par_scan_state(); + pss->trim_queue(); + + do { + while (queues()->steal(pss->queue_num(), pss->hash_seed(), stolen_task)) { + assert(pss->verify_task(stolen_task), "sanity"); + if (stolen_task.is_narrow()) { + pss->push_on_queue((narrowOop*) stolen_task); + } else { + pss->push_on_queue((oop*) stolen_task); + } pss->trim_queue(); - - StarTask stolen_task; - if (queues()->steal(pss->queue_num(), pss->hash_seed(), stolen_task)) { - // slightly paranoid tests; I'm trying to catch potential - // problems before we go into push_on_queue to know where the - // problem is coming from - assert((oop*)stolen_task != NULL, "Error"); - if (stolen_task.is_narrow()) { - assert(UseCompressedOops, "Error"); - narrowOop* p = (narrowOop*) stolen_task; - assert(has_partial_array_mask(p) || - _g1h->is_in_g1_reserved(oopDesc::load_decode_heap_oop(p)), "Error"); - pss->push_on_queue(p); - } else { - oop* p = (oop*) stolen_task; - assert(has_partial_array_mask(p) || _g1h->is_in_g1_reserved(*p), "Error"); - pss->push_on_queue(p); - } - continue; - } - pss->start_term_time(); - if (terminator()->offer_termination()) break; - pss->end_term_time(); } - pss->end_term_time(); - pss->retire_alloc_buffers(); - } -}; + } while (!offer_termination()); + + pss->retire_alloc_buffers(); +} class G1ParTask : public AbstractGangTask { protected: @@ -4182,8 +4226,7 @@ pss.print_termination_stats(i); } - assert(pss.refs_to_scan() == 0, "Task queue should be empty"); - assert(pss.overflowed_refs_to_scan() == 0, "Overflow queue should be empty"); + assert(pss.refs()->is_empty(), "should be empty"); double end_time_ms = os::elapsedTime() * 1000.0; _g1h->g1_policy()->record_gc_worker_end_time(i, end_time_ms); }
--- a/src/share/vm/gc_implementation/g1/g1CollectedHeap.hpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/share/vm/gc_implementation/g1/g1CollectedHeap.hpp Fri Oct 15 15:12:04 2010 -0400 @@ -1651,49 +1651,17 @@ size_t alloc_buffer_waste() const { return _alloc_buffer_waste; } size_t undo_waste() const { return _undo_waste; } +#ifdef ASSERT + bool verify_ref(narrowOop* ref) const; + bool verify_ref(oop* ref) const; + bool verify_task(StarTask ref) const; +#endif // ASSERT + template <class T> void push_on_queue(T* ref) { - assert(ref != NULL, "invariant"); - assert(has_partial_array_mask(ref) || - _g1h->is_in_g1_reserved(oopDesc::load_decode_heap_oop(ref)), "invariant"); -#ifdef ASSERT - if (has_partial_array_mask(ref)) { - oop p = clear_partial_array_mask(ref); - // Verify that we point into the CS - assert(_g1h->obj_in_cs(p), "Should be in CS"); - } -#endif + assert(verify_ref(ref), "sanity"); refs()->push(ref); } - void pop_from_queue(StarTask& ref) { - if (refs()->pop_local(ref)) { - assert((oop*)ref != NULL, "pop_local() returned true"); - assert(UseCompressedOops || !ref.is_narrow(), "Error"); - assert(has_partial_array_mask((oop*)ref) || - _g1h->is_in_g1_reserved(ref.is_narrow() ? oopDesc::load_decode_heap_oop((narrowOop*)ref) - : oopDesc::load_decode_heap_oop((oop*)ref)), - "invariant"); - } else { - StarTask null_task; - ref = null_task; - } - } - - void pop_from_overflow_queue(StarTask& ref) { - StarTask new_ref; - refs()->pop_overflow(new_ref); - assert((oop*)new_ref != NULL, "pop() from a local non-empty stack"); - assert(UseCompressedOops || !new_ref.is_narrow(), "Error"); - assert(has_partial_array_mask((oop*)new_ref) || - _g1h->is_in_g1_reserved(new_ref.is_narrow() ? oopDesc::load_decode_heap_oop((narrowOop*)new_ref) - : oopDesc::load_decode_heap_oop((oop*)new_ref)), - "invariant"); - ref = new_ref; - } - - int refs_to_scan() { return (int)refs()->size(); } - int overflowed_refs_to_scan() { return (int)refs()->overflow_stack()->size(); } - template <class T> void update_rs(HeapRegion* from, T* p, int tid) { if (G1DeferredRSUpdate) { deferred_rs_update(from, p, tid); @@ -1818,59 +1786,15 @@ } } -public: - void trim_queue() { - // I've replicated the loop twice, first to drain the overflow - // queue, second to drain the task queue. This is better than - // having a single loop, which checks both conditions and, inside - // it, either pops the overflow queue or the task queue, as each - // loop is tighter. Also, the decision to drain the overflow queue - // first is not arbitrary, as the overflow queue is not visible - // to the other workers, whereas the task queue is. So, we want to - // drain the "invisible" entries first, while allowing the other - // workers to potentially steal the "visible" entries. - - while (refs_to_scan() > 0 || overflowed_refs_to_scan() > 0) { - while (overflowed_refs_to_scan() > 0) { - StarTask ref_to_scan; - assert((oop*)ref_to_scan == NULL, "Constructed above"); - pop_from_overflow_queue(ref_to_scan); - // We shouldn't have pushed it on the queue if it was not - // pointing into the CSet. - assert((oop*)ref_to_scan != NULL, "Follows from inner loop invariant"); - if (ref_to_scan.is_narrow()) { - assert(UseCompressedOops, "Error"); - narrowOop* p = (narrowOop*)ref_to_scan; - assert(!has_partial_array_mask(p) && - _g1h->is_in_g1_reserved(oopDesc::load_decode_heap_oop(p)), "sanity"); - deal_with_reference(p); - } else { - oop* p = (oop*)ref_to_scan; - assert((has_partial_array_mask(p) && _g1h->is_in_g1_reserved(clear_partial_array_mask(p))) || - _g1h->is_in_g1_reserved(oopDesc::load_decode_heap_oop(p)), "sanity"); - deal_with_reference(p); - } - } - - while (refs_to_scan() > 0) { - StarTask ref_to_scan; - assert((oop*)ref_to_scan == NULL, "Constructed above"); - pop_from_queue(ref_to_scan); - if ((oop*)ref_to_scan != NULL) { - if (ref_to_scan.is_narrow()) { - assert(UseCompressedOops, "Error"); - narrowOop* p = (narrowOop*)ref_to_scan; - assert(!has_partial_array_mask(p) && - _g1h->is_in_g1_reserved(oopDesc::load_decode_heap_oop(p)), "sanity"); - deal_with_reference(p); - } else { - oop* p = (oop*)ref_to_scan; - assert((has_partial_array_mask(p) && _g1h->obj_in_cs(clear_partial_array_mask(p))) || - _g1h->is_in_g1_reserved(oopDesc::load_decode_heap_oop(p)), "sanity"); - deal_with_reference(p); - } - } - } + void deal_with_reference(StarTask ref) { + assert(verify_task(ref), "sanity"); + if (ref.is_narrow()) { + deal_with_reference((narrowOop*)ref); + } else { + deal_with_reference((oop*)ref); } } + +public: + void trim_queue(); };
--- a/src/share/vm/gc_implementation/g1/g1OopClosures.hpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/share/vm/gc_implementation/g1/g1OopClosures.hpp Fri Oct 15 15:12:04 2010 -0400 @@ -25,8 +25,6 @@ class HeapRegion; class G1CollectedHeap; class G1RemSet; -class HRInto_G1RemSet; -class G1RemSet; class ConcurrentMark; class DirtyCardToOopClosure; class CMBitMap;
--- a/src/share/vm/gc_implementation/g1/g1RemSet.cpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/share/vm/gc_implementation/g1/g1RemSet.cpp Fri Oct 15 15:12:04 2010 -0400 @@ -97,13 +97,6 @@ } }; -void -StupidG1RemSet::oops_into_collection_set_do(OopsInHeapRegionClosure* oc, - int worker_i) { - IntoCSRegionClosure rc(_g1, oc); - _g1->heap_region_iterate(&rc); -} - class VerifyRSCleanCardOopClosure: public OopClosure { G1CollectedHeap* _g1; public: @@ -119,8 +112,9 @@ } }; -HRInto_G1RemSet::HRInto_G1RemSet(G1CollectedHeap* g1, CardTableModRefBS* ct_bs) - : G1RemSet(g1), _ct_bs(ct_bs), _g1p(_g1->g1_policy()), +G1RemSet::G1RemSet(G1CollectedHeap* g1, CardTableModRefBS* ct_bs) + : _g1(g1), _conc_refine_cards(0), + _ct_bs(ct_bs), _g1p(_g1->g1_policy()), _cg1r(g1->concurrent_g1_refine()), _traversal_in_progress(false), _cset_rs_update_cl(NULL), @@ -134,7 +128,7 @@ } } -HRInto_G1RemSet::~HRInto_G1RemSet() { +G1RemSet::~G1RemSet() { delete _seq_task; for (uint i = 0; i < n_workers(); i++) { assert(_cset_rs_update_cl[i] == NULL, "it should be"); @@ -277,7 +271,7 @@ // p threads // Then thread t will start at region t * floor (n/p) -HeapRegion* HRInto_G1RemSet::calculateStartRegion(int worker_i) { +HeapRegion* G1RemSet::calculateStartRegion(int worker_i) { HeapRegion* result = _g1p->collection_set(); if (ParallelGCThreads > 0) { size_t cs_size = _g1p->collection_set_size(); @@ -290,7 +284,7 @@ return result; } -void HRInto_G1RemSet::scanRS(OopsInHeapRegionClosure* oc, int worker_i) { +void G1RemSet::scanRS(OopsInHeapRegionClosure* oc, int worker_i) { double rs_time_start = os::elapsedTime(); HeapRegion *startRegion = calculateStartRegion(worker_i); @@ -340,7 +334,7 @@ } }; -void HRInto_G1RemSet::updateRS(DirtyCardQueue* into_cset_dcq, int worker_i) { +void G1RemSet::updateRS(DirtyCardQueue* into_cset_dcq, int worker_i) { double start = os::elapsedTime(); // Apply the given closure to all remaining log entries. RefineRecordRefsIntoCSCardTableEntryClosure into_cset_update_rs_cl(_g1, into_cset_dcq); @@ -439,12 +433,11 @@ } }; -void HRInto_G1RemSet::cleanupHRRS() { +void G1RemSet::cleanupHRRS() { HeapRegionRemSet::cleanup(); } -void -HRInto_G1RemSet::oops_into_collection_set_do(OopsInHeapRegionClosure* oc, +void G1RemSet::oops_into_collection_set_do(OopsInHeapRegionClosure* oc, int worker_i) { #if CARD_REPEAT_HISTO ct_freq_update_histo_and_reset(); @@ -508,8 +501,7 @@ _cset_rs_update_cl[worker_i] = NULL; } -void HRInto_G1RemSet:: -prepare_for_oops_into_collection_set_do() { +void G1RemSet::prepare_for_oops_into_collection_set_do() { #if G1_REM_SET_LOGGING PrintRSClosure cl; _g1->collection_set_iterate(&cl); @@ -581,7 +573,7 @@ // RSet updating, // * the post-write barrier shouldn't be logging updates to young // regions (but there is a situation where this can happen - see - // the comment in HRInto_G1RemSet::concurrentRefineOneCard below - + // the comment in G1RemSet::concurrentRefineOneCard below - // that should not be applicable here), and // * during actual RSet updating, the filtering of cards in young // regions in HeapRegion::oops_on_card_seq_iterate_careful is @@ -601,7 +593,7 @@ } }; -void HRInto_G1RemSet::cleanup_after_oops_into_collection_set_do() { +void G1RemSet::cleanup_after_oops_into_collection_set_do() { guarantee( _cards_scanned != NULL, "invariant" ); _total_cards_scanned = 0; for (uint i = 0; i < n_workers(); ++i) @@ -692,12 +684,12 @@ } }; -void HRInto_G1RemSet::scrub(BitMap* region_bm, BitMap* card_bm) { +void G1RemSet::scrub(BitMap* region_bm, BitMap* card_bm) { ScrubRSClosure scrub_cl(region_bm, card_bm); _g1->heap_region_iterate(&scrub_cl); } -void HRInto_G1RemSet::scrub_par(BitMap* region_bm, BitMap* card_bm, +void G1RemSet::scrub_par(BitMap* region_bm, BitMap* card_bm, int worker_num, int claim_val) { ScrubRSClosure scrub_cl(region_bm, card_bm); _g1->heap_region_par_iterate_chunked(&scrub_cl, worker_num, claim_val); @@ -741,7 +733,7 @@ virtual void do_oop(narrowOop* p) { do_oop_nv(p); } }; -bool HRInto_G1RemSet::concurrentRefineOneCard_impl(jbyte* card_ptr, int worker_i, +bool G1RemSet::concurrentRefineOneCard_impl(jbyte* card_ptr, int worker_i, bool check_for_refs_into_cset) { // Construct the region representing the card. HeapWord* start = _ct_bs->addr_for(card_ptr); @@ -820,7 +812,7 @@ return trigger_cl.value(); } -bool HRInto_G1RemSet::concurrentRefineOneCard(jbyte* card_ptr, int worker_i, +bool G1RemSet::concurrentRefineOneCard(jbyte* card_ptr, int worker_i, bool check_for_refs_into_cset) { // If the card is no longer dirty, nothing to do. if (*card_ptr != CardTableModRefBS::dirty_card_val()) { @@ -995,7 +987,7 @@ } }; -void HRInto_G1RemSet::print_summary_info() { +void G1RemSet::print_summary_info() { G1CollectedHeap* g1 = G1CollectedHeap::heap(); #if CARD_REPEAT_HISTO @@ -1029,30 +1021,26 @@ g1->concurrent_g1_refine()->threads_do(&p); gclog_or_tty->print_cr(""); - if (G1UseHRIntoRS) { - HRRSStatsIter blk; - g1->heap_region_iterate(&blk); - gclog_or_tty->print_cr(" Total heap region rem set sizes = " SIZE_FORMAT "K." - " Max = " SIZE_FORMAT "K.", - blk.total_mem_sz()/K, blk.max_mem_sz()/K); - gclog_or_tty->print_cr(" Static structures = " SIZE_FORMAT "K," - " free_lists = " SIZE_FORMAT "K.", - HeapRegionRemSet::static_mem_size()/K, - HeapRegionRemSet::fl_mem_size()/K); - gclog_or_tty->print_cr(" %d occupied cards represented.", - blk.occupied()); - gclog_or_tty->print_cr(" Max sz region = [" PTR_FORMAT ", " PTR_FORMAT " )" - ", cap = " SIZE_FORMAT "K, occ = " SIZE_FORMAT "K.", - blk.max_mem_sz_region()->bottom(), blk.max_mem_sz_region()->end(), - (blk.max_mem_sz_region()->rem_set()->mem_size() + K - 1)/K, - (blk.max_mem_sz_region()->rem_set()->occupied() + K - 1)/K); - gclog_or_tty->print_cr(" Did %d coarsenings.", - HeapRegionRemSet::n_coarsenings()); - - } + HRRSStatsIter blk; + g1->heap_region_iterate(&blk); + gclog_or_tty->print_cr(" Total heap region rem set sizes = " SIZE_FORMAT "K." + " Max = " SIZE_FORMAT "K.", + blk.total_mem_sz()/K, blk.max_mem_sz()/K); + gclog_or_tty->print_cr(" Static structures = " SIZE_FORMAT "K," + " free_lists = " SIZE_FORMAT "K.", + HeapRegionRemSet::static_mem_size()/K, + HeapRegionRemSet::fl_mem_size()/K); + gclog_or_tty->print_cr(" %d occupied cards represented.", + blk.occupied()); + gclog_or_tty->print_cr(" Max sz region = [" PTR_FORMAT ", " PTR_FORMAT " )" + ", cap = " SIZE_FORMAT "K, occ = " SIZE_FORMAT "K.", + blk.max_mem_sz_region()->bottom(), blk.max_mem_sz_region()->end(), + (blk.max_mem_sz_region()->rem_set()->mem_size() + K - 1)/K, + (blk.max_mem_sz_region()->rem_set()->occupied() + K - 1)/K); + gclog_or_tty->print_cr(" Did %d coarsenings.", HeapRegionRemSet::n_coarsenings()); } -void HRInto_G1RemSet::prepare_for_verify() { +void G1RemSet::prepare_for_verify() { if (G1HRRSFlushLogBuffersOnVerify && (VerifyBeforeGC || VerifyAfterGC) && !_g1->full_collection()) {
--- a/src/share/vm/gc_implementation/g1/g1RemSet.hpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/share/vm/gc_implementation/g1/g1RemSet.hpp Fri Oct 15 15:12:04 2010 -0400 @@ -27,107 +27,18 @@ class G1CollectedHeap; class CardTableModRefBarrierSet; -class HRInto_G1RemSet; class ConcurrentG1Refine; +// A G1RemSet in which each heap region has a rem set that records the +// external heap references into it. Uses a mod ref bs to track updates, +// so that they can be used to update the individual region remsets. + class G1RemSet: public CHeapObj { protected: G1CollectedHeap* _g1; unsigned _conc_refine_cards; size_t n_workers(); -public: - G1RemSet(G1CollectedHeap* g1) : - _g1(g1), _conc_refine_cards(0) - {} - - // Invoke "blk->do_oop" on all pointers into the CS in object in regions - // outside the CS (having invoked "blk->set_region" to set the "from" - // region correctly beforehand.) The "worker_i" param is for the - // parallel case where the number of the worker thread calling this - // function can be helpful in partitioning the work to be done. It - // should be the same as the "i" passed to the calling thread's - // work(i) function. In the sequential case this param will be ingored. - virtual void oops_into_collection_set_do(OopsInHeapRegionClosure* blk, - int worker_i) = 0; - - // Prepare for and cleanup after an oops_into_collection_set_do - // call. Must call each of these once before and after (in sequential - // code) any threads call oops into collection set do. (This offers an - // opportunity to sequential setup and teardown of structures needed by a - // parallel iteration over the CS's RS.) - virtual void prepare_for_oops_into_collection_set_do() = 0; - virtual void cleanup_after_oops_into_collection_set_do() = 0; - - // If "this" is of the given subtype, return "this", else "NULL". - virtual HRInto_G1RemSet* as_HRInto_G1RemSet() { return NULL; } - - // Record, if necessary, the fact that *p (where "p" is in region "from", - // and is, a fortiori, required to be non-NULL) has changed to its new value. - virtual void write_ref(HeapRegion* from, oop* p) = 0; - virtual void write_ref(HeapRegion* from, narrowOop* p) = 0; - virtual void par_write_ref(HeapRegion* from, oop* p, int tid) = 0; - virtual void par_write_ref(HeapRegion* from, narrowOop* p, int tid) = 0; - - // Requires "region_bm" and "card_bm" to be bitmaps with 1 bit per region - // or card, respectively, such that a region or card with a corresponding - // 0 bit contains no part of any live object. Eliminates any remembered - // set entries that correspond to dead heap ranges. - virtual void scrub(BitMap* region_bm, BitMap* card_bm) = 0; - // Like the above, but assumes is called in parallel: "worker_num" is the - // parallel thread id of the current thread, and "claim_val" is the - // value that should be used to claim heap regions. - virtual void scrub_par(BitMap* region_bm, BitMap* card_bm, - int worker_num, int claim_val) = 0; - - // Refine the card corresponding to "card_ptr". If "sts" is non-NULL, - // join and leave around parts that must be atomic wrt GC. (NULL means - // being done at a safepoint.) - // With some implementations of this routine, when check_for_refs_into_cset - // is true, a true result may be returned if the given card contains oops - // that have references into the current collection set. - virtual bool concurrentRefineOneCard(jbyte* card_ptr, int worker_i, - bool check_for_refs_into_cset) { - return false; - } - - // Print any relevant summary info. - virtual void print_summary_info() {} - - // Prepare remebered set for verification. - virtual void prepare_for_verify() {}; -}; - - -// The simplest possible G1RemSet: iterates over all objects in non-CS -// regions, searching for pointers into the CS. -class StupidG1RemSet: public G1RemSet { -public: - StupidG1RemSet(G1CollectedHeap* g1) : G1RemSet(g1) {} - - void oops_into_collection_set_do(OopsInHeapRegionClosure* blk, - int worker_i); - - void prepare_for_oops_into_collection_set_do() {} - void cleanup_after_oops_into_collection_set_do() {} - - // Nothing is necessary in the version below. - void write_ref(HeapRegion* from, oop* p) {} - void write_ref(HeapRegion* from, narrowOop* p) {} - void par_write_ref(HeapRegion* from, oop* p, int tid) {} - void par_write_ref(HeapRegion* from, narrowOop* p, int tid) {} - - void scrub(BitMap* region_bm, BitMap* card_bm) {} - void scrub_par(BitMap* region_bm, BitMap* card_bm, - int worker_num, int claim_val) {} - -}; - -// A G1RemSet in which each heap region has a rem set that records the -// external heap references into it. Uses a mod ref bs to track updates, -// so that they can be used to update the individual region remsets. - -class HRInto_G1RemSet: public G1RemSet { protected: enum SomePrivateConstants { UpdateRStoMergeSync = 0, @@ -175,28 +86,32 @@ // scanned. void cleanupHRRS(); - HRInto_G1RemSet(G1CollectedHeap* g1, CardTableModRefBS* ct_bs); - ~HRInto_G1RemSet(); + G1RemSet(G1CollectedHeap* g1, CardTableModRefBS* ct_bs); + ~G1RemSet(); + // Invoke "blk->do_oop" on all pointers into the CS in objects in regions + // outside the CS (having invoked "blk->set_region" to set the "from" + // region correctly beforehand.) The "worker_i" param is for the + // parallel case where the number of the worker thread calling this + // function can be helpful in partitioning the work to be done. It + // should be the same as the "i" passed to the calling thread's + // work(i) function. In the sequential case this param will be ingored. void oops_into_collection_set_do(OopsInHeapRegionClosure* blk, int worker_i); + // Prepare for and cleanup after an oops_into_collection_set_do + // call. Must call each of these once before and after (in sequential + // code) any threads call oops_into_collection_set_do. (This offers an + // opportunity to sequential setup and teardown of structures needed by a + // parallel iteration over the CS's RS.) void prepare_for_oops_into_collection_set_do(); void cleanup_after_oops_into_collection_set_do(); + void scanRS(OopsInHeapRegionClosure* oc, int worker_i); - template <class T> void scanNewRefsRS_work(OopsInHeapRegionClosure* oc, int worker_i); - void scanNewRefsRS(OopsInHeapRegionClosure* oc, int worker_i) { - if (UseCompressedOops) { - scanNewRefsRS_work<narrowOop>(oc, worker_i); - } else { - scanNewRefsRS_work<oop>(oc, worker_i); - } - } void updateRS(DirtyCardQueue* into_cset_dcq, int worker_i); + HeapRegion* calculateStartRegion(int i); - HRInto_G1RemSet* as_HRInto_G1RemSet() { return this; } - CardTableModRefBS* ct_bs() { return _ct_bs; } size_t cardsScanned() { return _total_cards_scanned; } @@ -219,17 +134,31 @@ bool self_forwarded(oop obj); + // Requires "region_bm" and "card_bm" to be bitmaps with 1 bit per region + // or card, respectively, such that a region or card with a corresponding + // 0 bit contains no part of any live object. Eliminates any remembered + // set entries that correspond to dead heap ranges. void scrub(BitMap* region_bm, BitMap* card_bm); + + // Like the above, but assumes is called in parallel: "worker_num" is the + // parallel thread id of the current thread, and "claim_val" is the + // value that should be used to claim heap regions. void scrub_par(BitMap* region_bm, BitMap* card_bm, int worker_num, int claim_val); - // If check_for_refs_into_cset is true then a true result is returned - // if the card contains oops that have references into the current - // collection set. + // Refine the card corresponding to "card_ptr". If "sts" is non-NULL, + // join and leave around parts that must be atomic wrt GC. (NULL means + // being done at a safepoint.) + // If check_for_refs_into_cset is true, a true result is returned + // if the given card contains oops that have references into the + // current collection set. virtual bool concurrentRefineOneCard(jbyte* card_ptr, int worker_i, bool check_for_refs_into_cset); + // Print any relevant summary info. virtual void print_summary_info(); + + // Prepare remembered set for verification. virtual void prepare_for_verify(); }; @@ -250,13 +179,13 @@ class UpdateRSOopClosure: public OopClosure { HeapRegion* _from; - HRInto_G1RemSet* _rs; + G1RemSet* _rs; int _worker_i; template <class T> void do_oop_work(T* p); public: - UpdateRSOopClosure(HRInto_G1RemSet* rs, int worker_i = 0) : + UpdateRSOopClosure(G1RemSet* rs, int worker_i = 0) : _from(NULL), _rs(rs), _worker_i(worker_i) { guarantee(_rs != NULL, "Requires an HRIntoG1RemSet"); }
--- a/src/share/vm/gc_implementation/g1/g1RemSet.inline.hpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/share/vm/gc_implementation/g1/g1RemSet.inline.hpp Fri Oct 15 15:12:04 2010 -0400 @@ -30,16 +30,18 @@ } } -template <class T> inline void HRInto_G1RemSet::write_ref_nv(HeapRegion* from, T* p) { +template <class T> +inline void G1RemSet::write_ref_nv(HeapRegion* from, T* p) { par_write_ref_nv(from, p, 0); } -inline bool HRInto_G1RemSet::self_forwarded(oop obj) { +inline bool G1RemSet::self_forwarded(oop obj) { bool result = (obj->is_forwarded() && (obj->forwardee()== obj)); return result; } -template <class T> inline void HRInto_G1RemSet::par_write_ref_nv(HeapRegion* from, T* p, int tid) { +template <class T> +inline void G1RemSet::par_write_ref_nv(HeapRegion* from, T* p, int tid) { oop obj = oopDesc::load_decode_heap_oop(p); #ifdef ASSERT // can't do because of races @@ -77,7 +79,7 @@ // Deferred updates to the CSet are either discarded (in the normal case), // or processed (if an evacuation failure occurs) at the end // of the collection. - // See HRInto_G1RemSet::cleanup_after_oops_into_collection_set_do(). + // See G1RemSet::cleanup_after_oops_into_collection_set_do(). } else { #if G1_REM_SET_LOGGING gclog_or_tty->print_cr("Adding " PTR_FORMAT " (" PTR_FORMAT ") to RS" @@ -91,12 +93,14 @@ } } -template <class T> inline void UpdateRSOopClosure::do_oop_work(T* p) { +template <class T> +inline void UpdateRSOopClosure::do_oop_work(T* p) { assert(_from != NULL, "from region must be non-NULL"); _rs->par_write_ref(_from, p, _worker_i); } -template <class T> inline void UpdateRSetImmediate::do_oop_work(T* p) { +template <class T> +inline void UpdateRSetImmediate::do_oop_work(T* p) { assert(_from->is_in_reserved(p), "paranoia"); T heap_oop = oopDesc::load_heap_oop(p); if (!oopDesc::is_null(heap_oop) && !_from->is_survivor()) {
--- a/src/share/vm/gc_implementation/g1/g1_globals.hpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/share/vm/gc_implementation/g1/g1_globals.hpp Fri Oct 15 15:12:04 2010 -0400 @@ -40,9 +40,6 @@ develop(intx, G1PolicyVerbose, 0, \ "The verbosity level on G1 policy decisions") \ \ - develop(bool, G1UseHRIntoRS, true, \ - "Determines whether the 'advanced' HR Into rem set is used.") \ - \ develop(intx, G1MarkingVerboseLevel, 0, \ "Level (0-4) of verboseness of the marking code") \ \
--- a/src/share/vm/gc_implementation/includeDB_gc_g1 Wed Oct 13 11:46:46 2010 -0400 +++ b/src/share/vm/gc_implementation/includeDB_gc_g1 Fri Oct 15 15:12:04 2010 -0400 @@ -310,10 +310,16 @@ heapRegionSeq.inline.hpp heapRegionSeq.hpp +instanceKlass.cpp g1RemSet.inline.hpp + +instanceRefKlass.cpp g1RemSet.inline.hpp + klass.hpp g1OopClosures.hpp memoryService.cpp g1MemoryPool.hpp +objArrayKlass.cpp g1RemSet.inline.hpp + ptrQueue.cpp allocation.hpp ptrQueue.cpp allocation.inline.hpp ptrQueue.cpp mutex.hpp
--- a/src/share/vm/oops/methodOop.cpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/share/vm/oops/methodOop.cpp Fri Oct 15 15:12:04 2010 -0400 @@ -758,7 +758,7 @@ OrderAccess::storestore(); #ifdef SHARK - mh->_from_interpreted_entry = code->instructions_begin(); + mh->_from_interpreted_entry = code->insts_begin(); #else mh->_from_compiled_entry = code->verified_entry_point(); OrderAccess::storestore();
--- a/src/share/vm/oops/oop.inline.hpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/share/vm/oops/oop.inline.hpp Fri Oct 15 15:12:04 2010 -0400 @@ -173,7 +173,7 @@ address base = Universe::narrow_oop_base(); int shift = Universe::narrow_oop_shift(); oop result = (oop)(void*)((uintptr_t)base + ((uintptr_t)v << shift)); - assert(check_obj_alignment(result), "Address not aligned"); + assert(check_obj_alignment(result), err_msg("address not aligned: " PTR_FORMAT, (void*) result)); return result; }
--- a/src/share/vm/opto/library_call.cpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/share/vm/opto/library_call.cpp Fri Oct 15 15:12:04 2010 -0400 @@ -4761,7 +4761,7 @@ Node* cv = generate_checkcast_arraycopy(adr_type, dest_elem_klass, src, src_offset, dest, dest_offset, - copy_length); + ConvI2X(copy_length)); if (cv == NULL) cv = intcon(-1); // failure (no stub available) checked_control = control(); checked_i_o = i_o(); @@ -5206,7 +5206,7 @@ int sco_offset = Klass::super_check_offset_offset_in_bytes() + sizeof(oopDesc); Node* p3 = basic_plus_adr(dest_elem_klass, sco_offset); Node* n3 = new(C, 3) LoadINode(NULL, memory(p3), p3, _gvn.type(p3)->is_ptr()); - Node* check_offset = _gvn.transform(n3); + Node* check_offset = ConvI2X(_gvn.transform(n3)); Node* check_value = dest_elem_klass; Node* src_start = array_element_address(src, src_offset, T_OBJECT);
--- a/src/share/vm/opto/loopTransform.cpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/share/vm/opto/loopTransform.cpp Fri Oct 15 15:12:04 2010 -0400 @@ -2684,7 +2684,14 @@ fill_name, TypeAryPtr::get_array_body_type(t)); call->init_req(TypeFunc::Parms+0, from); call->init_req(TypeFunc::Parms+1, store_value); +#ifdef _LP64 + len = new (C, 2) ConvI2LNode(len); + _igvn.register_new_node_with_optimizer(len); +#endif call->init_req(TypeFunc::Parms+2, len); +#ifdef _LP64 + call->init_req(TypeFunc::Parms+3, C->top()); +#endif call->init_req( TypeFunc::Control, head->init_control()); call->init_req( TypeFunc::I_O , C->top() ) ; // does no i/o call->init_req( TypeFunc::Memory , mem_phi->in(LoopNode::EntryControl) );
--- a/src/share/vm/opto/runtime.cpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/share/vm/opto/runtime.cpp Fri Oct 15 15:12:04 2010 -0400 @@ -646,12 +646,14 @@ const TypeFunc* OptoRuntime::array_fill_Type() { - // create input type (domain) - const Type** fields = TypeTuple::fields(3); - fields[TypeFunc::Parms+0] = TypePtr::NOTNULL; - fields[TypeFunc::Parms+1] = TypeInt::INT; - fields[TypeFunc::Parms+2] = TypeInt::INT; - const TypeTuple *domain = TypeTuple::make(TypeFunc::Parms + 3, fields); + // create input type (domain): pointer, int, size_t + const Type** fields = TypeTuple::fields(3 LP64_ONLY( + 1)); + int argp = TypeFunc::Parms; + fields[argp++] = TypePtr::NOTNULL; + fields[argp++] = TypeInt::INT; + fields[argp++] = TypeX_X; // size in whatevers (size_t) + LP64_ONLY(fields[argp++] = Type::HALF); // other half of long length + const TypeTuple *domain = TypeTuple::make(argp, fields); // create result type fields = TypeTuple::fields(1);
--- a/src/share/vm/prims/methodHandles.cpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/share/vm/prims/methodHandles.cpp Fri Oct 15 15:12:04 2010 -0400 @@ -1568,7 +1568,7 @@ if (ptype != T_INT) { int value_offset = java_lang_boxing_object::value_offset_in_bytes(T_INT); jint value = argument->int_field(value_offset); - int vminfo = adapter_subword_vminfo(ptype); + int vminfo = adapter_unbox_subword_vminfo(ptype); jint subword = truncate_subword_from_vminfo(value, vminfo); if (value != subword) { err = "bound subword value does not fit into the subword type"; @@ -2018,12 +2018,12 @@ assert(src == T_INT || is_subword_type(src), "source is not float"); // Subword-related cases are int -> {boolean,byte,char,short}. ek_opt = _adapter_opt_i2i; - vminfo = adapter_subword_vminfo(dest); + vminfo = adapter_prim_to_prim_subword_vminfo(dest); break; case 2 *4+ 1: if (src == T_LONG && (dest == T_INT || is_subword_type(dest))) { ek_opt = _adapter_opt_l2i; - vminfo = adapter_subword_vminfo(dest); + vminfo = adapter_prim_to_prim_subword_vminfo(dest); } else if (src == T_DOUBLE && dest == T_FLOAT) { ek_opt = _adapter_opt_d2f; } else { @@ -2051,7 +2051,7 @@ switch (type2size[dest]) { case 1: ek_opt = _adapter_opt_unboxi; - vminfo = adapter_subword_vminfo(dest); + vminfo = adapter_unbox_subword_vminfo(dest); break; case 2: ek_opt = _adapter_opt_unboxl;
--- a/src/share/vm/prims/methodHandles.hpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/share/vm/prims/methodHandles.hpp Fri Oct 15 15:12:04 2010 -0400 @@ -226,11 +226,20 @@ } enum { CONV_VMINFO_SIGN_FLAG = 0x80 }; - static int adapter_subword_vminfo(BasicType dest) { - if (dest == T_BOOLEAN) return (BitsPerInt - 1); - if (dest == T_CHAR) return (BitsPerInt - 16); - if (dest == T_BYTE) return (BitsPerInt - 8) | CONV_VMINFO_SIGN_FLAG; - if (dest == T_SHORT) return (BitsPerInt - 16) | CONV_VMINFO_SIGN_FLAG; + // Shift values for prim-to-prim conversions. + static int adapter_prim_to_prim_subword_vminfo(BasicType dest) { + if (dest == T_BOOLEAN) return (BitsPerInt - 1); // boolean is 1 bit + if (dest == T_CHAR) return (BitsPerInt - BitsPerShort); + if (dest == T_BYTE) return (BitsPerInt - BitsPerByte ) | CONV_VMINFO_SIGN_FLAG; + if (dest == T_SHORT) return (BitsPerInt - BitsPerShort) | CONV_VMINFO_SIGN_FLAG; + return 0; // case T_INT + } + // Shift values for unboxing a primitive. + static int adapter_unbox_subword_vminfo(BasicType dest) { + if (dest == T_BOOLEAN) return (BitsPerInt - BitsPerByte ); // implemented as 1 byte + if (dest == T_CHAR) return (BitsPerInt - BitsPerShort); + if (dest == T_BYTE) return (BitsPerInt - BitsPerByte ) | CONV_VMINFO_SIGN_FLAG; + if (dest == T_SHORT) return (BitsPerInt - BitsPerShort) | CONV_VMINFO_SIGN_FLAG; return 0; // case T_INT } // Here is the transformation the i2i adapter must perform:
--- a/src/share/vm/shark/sharkCompiler.hpp Wed Oct 13 11:46:46 2010 -0400 +++ b/src/share/vm/shark/sharkCompiler.hpp Fri Oct 15 15:12:04 2010 -0400 @@ -103,8 +103,7 @@ // Global access public: static SharkCompiler* compiler() { - AbstractCompiler *compiler = - CompileBroker::compiler(CompLevel_fast_compile); + AbstractCompiler *compiler = CompileBroker::compiler(CompLevel_simple); assert(compiler->is_shark() && compiler->is_initialized(), "should be"); return (SharkCompiler *) compiler; }
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/compiler/6987555/Test6987555.java Fri Oct 15 15:12:04 2010 -0400 @@ -0,0 +1,177 @@ +/* + * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This code is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 only, as + * published by the Free Software Foundation. + * + * This code is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * version 2 for more details (a copy is included in the LICENSE file that + * accompanied this code). + * + * You should have received a copy of the GNU General Public License version + * 2 along with this work; if not, write to the Free Software Foundation, + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. + * + * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA + * or visit www.oracle.com if you need additional information or have any + * questions. + * + */ + +/** + * @test + * @bug 6987555 + * @summary JSR 292 unboxing to a boolean value fails on big-endian SPARC + * + * @run main/othervm -Xint -ea -XX:+UnlockExperimentalVMOptions -XX:+EnableMethodHandles -XX:+EnableInvokeDynamic -XX:+UnlockDiagnosticVMOptions -XX:+VerifyMethodHandles Test6987555 + */ + +import java.dyn.*; + +public class Test6987555 { + private static final Class CLASS = Test6987555.class; + private static final String NAME = "foo"; + private static final boolean DEBUG = false; + + public static void main(String[] args) throws Throwable { + testboolean(); + testbyte(); + testchar(); + testshort(); + testint(); + } + + // boolean + static void testboolean() throws Throwable { + doboolean(false); + doboolean(true); + } + static void doboolean(boolean x) throws Throwable { + if (DEBUG) System.out.println("boolean=" + x); + MethodHandle mh1 = MethodHandles.lookup().findStatic(CLASS, NAME, MethodType.methodType(boolean.class, boolean.class)); + MethodHandle mh2 = mh1.asType(MethodType.methodType(boolean.class, Boolean.class)); + boolean a = mh1.<boolean>invokeExact(x); + boolean b = mh2.<boolean>invokeExact(Boolean.valueOf(x)); + assert a == b : a + " != " + b; + } + + // byte + static void testbyte() throws Throwable { + byte[] a = new byte[] { + Byte.MIN_VALUE, + Byte.MIN_VALUE + 1, + -0x0F, + -1, + 0, + 1, + 0x0F, + Byte.MAX_VALUE - 1, + Byte.MAX_VALUE + }; + for (int i = 0; i < a.length; i++) { + dobyte(a[i]); + } + } + static void dobyte(byte x) throws Throwable { + if (DEBUG) System.out.println("byte=" + x); + MethodHandle mh1 = MethodHandles.lookup().findStatic(CLASS, NAME, MethodType.methodType(byte.class, byte.class)); + MethodHandle mh2 = mh1.asType(MethodType.methodType(byte.class, Byte.class)); + byte a = mh1.<byte>invokeExact(x); + byte b = mh2.<byte>invokeExact(Byte.valueOf(x)); + assert a == b : a + " != " + b; + } + + // char + static void testchar() throws Throwable { + char[] a = new char[] { + Character.MIN_VALUE, + Character.MIN_VALUE + 1, + 0x000F, + 0x00FF, + 0x0FFF, + Character.MAX_VALUE - 1, + Character.MAX_VALUE + }; + for (int i = 0; i < a.length; i++) { + dochar(a[i]); + } + } + static void dochar(char x) throws Throwable { + if (DEBUG) System.out.println("char=" + x); + MethodHandle mh1 = MethodHandles.lookup().findStatic(CLASS, NAME, MethodType.methodType(char.class, char.class)); + MethodHandle mh2 = mh1.asType(MethodType.methodType(char.class, Character.class)); + char a = mh1.<char>invokeExact(x); + char b = mh2.<char>invokeExact(Character.valueOf(x)); + assert a == b : a + " != " + b; + } + + // short + static void testshort() throws Throwable { + short[] a = new short[] { + Short.MIN_VALUE, + Short.MIN_VALUE + 1, + -0x0FFF, + -0x00FF, + -0x000F, + -1, + 0, + 1, + 0x000F, + 0x00FF, + 0x0FFF, + Short.MAX_VALUE - 1, + Short.MAX_VALUE + }; + for (int i = 0; i < a.length; i++) { + doshort(a[i]); + } + } + static void doshort(short x) throws Throwable { + if (DEBUG) System.out.println("short=" + x); + MethodHandle mh1 = MethodHandles.lookup().findStatic(CLASS, NAME, MethodType.methodType(short.class, short.class)); + MethodHandle mh2 = mh1.asType(MethodType.methodType(short.class, Short.class)); + short a = mh1.<short>invokeExact(x); + short b = mh2.<short>invokeExact(Short.valueOf(x)); + assert a == b : a + " != " + b; + } + + // int + static void testint() throws Throwable { + int[] a = new int[] { + Integer.MIN_VALUE, + Integer.MIN_VALUE + 1, + -0x00000FFF, + -0x000000FF, + -0x0000000F, + -1, + 0, + 1, + 0x0000000F, + 0x000000FF, + 0x00000FFF, + Integer.MAX_VALUE - 1, + Integer.MAX_VALUE + }; + for (int i = 0; i < a.length; i++) { + doint(a[i]); + } + } + static void doint(int x) throws Throwable { + if (DEBUG) System.out.println("int=" + x); + MethodHandle mh1 = MethodHandles.lookup().findStatic(CLASS, NAME, MethodType.methodType(int.class, int.class)); + MethodHandle mh2 = mh1.asType(MethodType.methodType(int.class, Integer.class)); + int a = mh1.<int>invokeExact(x); + int b = mh2.<int>invokeExact(Integer.valueOf(x)); + assert a == b : a + " != " + b; + } + + public static boolean foo(boolean i) { return i; } + public static byte foo(byte i) { return i; } + public static char foo(char i) { return i; } + public static short foo(short i) { return i; } + public static int foo(int i) { return i; } +}