Thu, 19 Mar 2009 09:13:24 -0700
Merge
1.1 --- a/agent/src/share/classes/sun/jvm/hotspot/debugger/Debugger.java Wed Mar 18 11:37:48 2009 -0400 1.2 +++ b/agent/src/share/classes/sun/jvm/hotspot/debugger/Debugger.java Thu Mar 19 09:13:24 2009 -0700 1.3 @@ -118,9 +118,9 @@ 1.4 public long getJIntSize(); 1.5 public long getJLongSize(); 1.6 public long getJShortSize(); 1.7 - public long getHeapBase(); 1.8 public long getHeapOopSize(); 1.9 - public long getLogMinObjAlignmentInBytes(); 1.10 + public long getNarrowOopBase(); 1.11 + public int getNarrowOopShift(); 1.12 1.13 public ReadResult readBytesFromProcess(long address, long numBytes) 1.14 throws DebuggerException;
2.1 --- a/agent/src/share/classes/sun/jvm/hotspot/debugger/DebuggerBase.java Wed Mar 18 11:37:48 2009 -0400 2.2 +++ b/agent/src/share/classes/sun/jvm/hotspot/debugger/DebuggerBase.java Thu Mar 19 09:13:24 2009 -0700 2.3 @@ -56,8 +56,8 @@ 2.4 // heap data. 2.5 protected long oopSize; 2.6 protected long heapOopSize; 2.7 - protected long heapBase; // heap base for compressed oops. 2.8 - protected long logMinObjAlignmentInBytes; // Used to decode compressed oops. 2.9 + protected long narrowOopBase; // heap base for compressed oops. 2.10 + protected int narrowOopShift; // shift to decode compressed oops. 2.11 // Should be initialized if desired by calling initCache() 2.12 private PageCache cache; 2.13 2.14 @@ -159,10 +159,10 @@ 2.15 javaPrimitiveTypesConfigured = true; 2.16 } 2.17 2.18 - public void putHeapConst(long heapBase, long heapOopSize, long logMinObjAlignmentInBytes) { 2.19 - this.heapBase = heapBase; 2.20 + public void putHeapConst(long heapOopSize, long narrowOopBase, int narrowOopShift) { 2.21 this.heapOopSize = heapOopSize; 2.22 - this.logMinObjAlignmentInBytes = logMinObjAlignmentInBytes; 2.23 + this.narrowOopBase = narrowOopBase; 2.24 + this.narrowOopShift = narrowOopShift; 2.25 } 2.26 2.27 /** May be called by subclasses if desired to initialize the page 2.28 @@ -459,7 +459,7 @@ 2.29 long value = readCInteger(address, getHeapOopSize(), true); 2.30 if (value != 0) { 2.31 // See oop.inline.hpp decode_heap_oop 2.32 - value = (long)(heapBase + (long)(value << logMinObjAlignmentInBytes)); 2.33 + value = (long)(narrowOopBase + (long)(value << narrowOopShift)); 2.34 } 2.35 return value; 2.36 } 2.37 @@ -545,10 +545,10 @@ 2.38 return heapOopSize; 2.39 } 2.40 2.41 - public long getHeapBase() { 2.42 - return heapBase; 2.43 + public long getNarrowOopBase() { 2.44 + return narrowOopBase; 2.45 } 2.46 - public long getLogMinObjAlignmentInBytes() { 2.47 - return logMinObjAlignmentInBytes; 2.48 + public int getNarrowOopShift() { 2.49 + return narrowOopShift; 2.50 } 2.51 }
3.1 --- a/agent/src/share/classes/sun/jvm/hotspot/debugger/JVMDebugger.java Wed Mar 18 11:37:48 2009 -0400 3.2 +++ b/agent/src/share/classes/sun/jvm/hotspot/debugger/JVMDebugger.java Thu Mar 19 09:13:24 2009 -0700 3.3 @@ -42,5 +42,5 @@ 3.4 long jintSize, 3.5 long jlongSize, 3.6 long jshortSize); 3.7 - public void putHeapConst(long heapBase, long heapOopSize, long logMinObjAlignment); 3.8 + public void putHeapConst(long heapOopSize, long narrowOopBase, int narrowOopShift); 3.9 }
4.1 --- a/agent/src/share/classes/sun/jvm/hotspot/debugger/remote/RemoteDebugger.java Wed Mar 18 11:37:48 2009 -0400 4.2 +++ b/agent/src/share/classes/sun/jvm/hotspot/debugger/remote/RemoteDebugger.java Thu Mar 19 09:13:24 2009 -0700 4.3 @@ -65,9 +65,10 @@ 4.4 public long getJIntSize() throws RemoteException; 4.5 public long getJLongSize() throws RemoteException; 4.6 public long getJShortSize() throws RemoteException; 4.7 - public long getHeapBase() throws RemoteException; 4.8 public long getHeapOopSize() throws RemoteException; 4.9 - public long getLogMinObjAlignmentInBytes() throws RemoteException; 4.10 + public long getNarrowOopBase() throws RemoteException; 4.11 + public int getNarrowOopShift() throws RemoteException; 4.12 + 4.13 public boolean areThreadsEqual(long addrOrId1, boolean isAddress1, 4.14 long addrOrId2, boolean isAddress2) throws RemoteException; 4.15 public int getThreadHashCode(long addrOrId, boolean isAddress) throws RemoteException;
5.1 --- a/agent/src/share/classes/sun/jvm/hotspot/debugger/remote/RemoteDebuggerClient.java Wed Mar 18 11:37:48 2009 -0400 5.2 +++ b/agent/src/share/classes/sun/jvm/hotspot/debugger/remote/RemoteDebuggerClient.java Thu Mar 19 09:13:24 2009 -0700 5.3 @@ -85,9 +85,9 @@ 5.4 jlongSize = remoteDebugger.getJLongSize(); 5.5 jshortSize = remoteDebugger.getJShortSize(); 5.6 javaPrimitiveTypesConfigured = true; 5.7 - heapBase = remoteDebugger.getHeapBase(); 5.8 + narrowOopBase = remoteDebugger.getNarrowOopBase(); 5.9 + narrowOopShift = remoteDebugger.getNarrowOopShift(); 5.10 heapOopSize = remoteDebugger.getHeapOopSize(); 5.11 - logMinObjAlignmentInBytes = remoteDebugger.getLogMinObjAlignmentInBytes(); 5.12 } 5.13 catch (RemoteException e) { 5.14 throw new DebuggerException(e);
6.1 --- a/agent/src/share/classes/sun/jvm/hotspot/debugger/remote/RemoteDebuggerServer.java Wed Mar 18 11:37:48 2009 -0400 6.2 +++ b/agent/src/share/classes/sun/jvm/hotspot/debugger/remote/RemoteDebuggerServer.java Thu Mar 19 09:13:24 2009 -0700 6.3 @@ -114,17 +114,18 @@ 6.4 return debugger.getJShortSize(); 6.5 } 6.6 6.7 - public long getHeapBase() throws RemoteException { 6.8 - return debugger.getHeapBase(); 6.9 - } 6.10 - 6.11 public long getHeapOopSize() throws RemoteException { 6.12 return debugger.getHeapOopSize(); 6.13 } 6.14 6.15 - public long getLogMinObjAlignmentInBytes() throws RemoteException { 6.16 - return debugger.getLogMinObjAlignmentInBytes(); 6.17 + public long getNarrowOopBase() throws RemoteException { 6.18 + return debugger.getNarrowOopBase(); 6.19 } 6.20 + 6.21 + public int getNarrowOopShift() throws RemoteException { 6.22 + return debugger.getNarrowOopShift(); 6.23 + } 6.24 + 6.25 public boolean areThreadsEqual(long addrOrId1, boolean isAddress1, 6.26 long addrOrId2, boolean isAddress2) throws RemoteException { 6.27 ThreadProxy t1 = getThreadProxy(addrOrId1, isAddress1);
7.1 --- a/agent/src/share/classes/sun/jvm/hotspot/memory/Universe.java Wed Mar 18 11:37:48 2009 -0400 7.2 +++ b/agent/src/share/classes/sun/jvm/hotspot/memory/Universe.java Thu Mar 19 09:13:24 2009 -0700 7.3 @@ -53,7 +53,8 @@ 7.4 // system obj array klass object 7.5 private static sun.jvm.hotspot.types.OopField systemObjArrayKlassObjField; 7.6 7.7 - private static AddressField heapBaseField; 7.8 + private static AddressField narrowOopBaseField; 7.9 + private static CIntegerField narrowOopShiftField; 7.10 7.11 static { 7.12 VM.registerVMInitializedObserver(new Observer() { 7.13 @@ -86,7 +87,8 @@ 7.14 7.15 systemObjArrayKlassObjField = type.getOopField("_systemObjArrayKlassObj"); 7.16 7.17 - heapBaseField = type.getAddressField("_heap_base"); 7.18 + narrowOopBaseField = type.getAddressField("_narrow_oop._base"); 7.19 + narrowOopShiftField = type.getCIntegerField("_narrow_oop._shift"); 7.20 } 7.21 7.22 public Universe() { 7.23 @@ -100,14 +102,18 @@ 7.24 } 7.25 } 7.26 7.27 - public static long getHeapBase() { 7.28 - if (heapBaseField.getValue() == null) { 7.29 + public static long getNarrowOopBase() { 7.30 + if (narrowOopBaseField.getValue() == null) { 7.31 return 0; 7.32 } else { 7.33 - return heapBaseField.getValue().minus(null); 7.34 + return narrowOopBaseField.getValue().minus(null); 7.35 } 7.36 } 7.37 7.38 + public static int getNarrowOopShift() { 7.39 + return (int)narrowOopShiftField.getValue(); 7.40 + } 7.41 + 7.42 /** Returns "TRUE" iff "p" points into the allocated area of the heap. */ 7.43 public boolean isIn(Address p) { 7.44 return heap().isIn(p);
8.1 --- a/agent/src/share/classes/sun/jvm/hotspot/runtime/VM.java Wed Mar 18 11:37:48 2009 -0400 8.2 +++ b/agent/src/share/classes/sun/jvm/hotspot/runtime/VM.java Thu Mar 19 09:13:24 2009 -0700 8.3 @@ -342,8 +342,8 @@ 8.4 throw new RuntimeException("Attempt to initialize VM twice"); 8.5 } 8.6 soleInstance = new VM(db, debugger, debugger.getMachineDescription().isBigEndian()); 8.7 - debugger.putHeapConst(Universe.getHeapBase(), soleInstance.getHeapOopSize(), 8.8 - soleInstance.logMinObjAlignmentInBytes); 8.9 + debugger.putHeapConst(soleInstance.getHeapOopSize(), Universe.getNarrowOopBase(), 8.10 + Universe.getNarrowOopShift()); 8.11 for (Iterator iter = vmInitializedObservers.iterator(); iter.hasNext(); ) { 8.12 ((Observer) iter.next()).update(null, null); 8.13 }
9.1 --- a/make/windows/get_msc_ver.sh Wed Mar 18 11:37:48 2009 -0400 9.2 +++ b/make/windows/get_msc_ver.sh Thu Mar 19 09:13:24 2009 -0700 9.3 @@ -29,6 +29,7 @@ 9.4 # cl version 13.10.3077 returns "MSC_VER=1310" 9.5 # cl version 14.00.30701 returns "MSC_VER=1399" (OLD_MSSDK version) 9.6 # cl version 14.00.40310.41 returns "MSC_VER=1400" 9.7 +# cl version 15.00.21022.8 returns "MSC_VER=1500" 9.8 9.9 # Note that we currently do not have a way to set HotSpotMksHome in 9.10 # the batch build, but so far this has not seemed to be a problem. The
10.1 --- a/make/windows/makefiles/compile.make Wed Mar 18 11:37:48 2009 -0400 10.2 +++ b/make/windows/makefiles/compile.make Thu Mar 19 09:13:24 2009 -0700 10.3 @@ -170,11 +170,9 @@ 10.4 # Manifest Tool - used in VS2005 and later to adjust manifests stored 10.5 # as resources inside build artifacts. 10.6 MT=mt.exe 10.7 -!if "$(BUILDARCH)" == "i486" 10.8 -# VS2005 on x86 restricts the use of certain libc functions without this 10.9 +# VS2005 and later restricts the use of certain libc functions without this 10.10 CPP_FLAGS=$(CPP_FLAGS) /D _CRT_SECURE_NO_DEPRECATE 10.11 !endif 10.12 -!endif 10.13 10.14 !if "$(COMPILER_NAME)" == "VS2008" 10.15 PRODUCT_OPT_OPTION = /O2 /Oy- 10.16 @@ -185,11 +183,9 @@ 10.17 # Manifest Tool - used in VS2005 and later to adjust manifests stored 10.18 # as resources inside build artifacts. 10.19 MT=mt.exe 10.20 -!if "$(BUILDARCH)" == "i486" 10.21 -# VS2005 on x86 restricts the use of certain libc functions without this 10.22 +# VS2005 and later restricts the use of certain libc functions without this 10.23 CPP_FLAGS=$(CPP_FLAGS) /D _CRT_SECURE_NO_DEPRECATE 10.24 !endif 10.25 -!endif 10.26 10.27 # Compile for space above time. 10.28 !if "$(Variant)" == "kernel"
11.1 --- a/make/windows/makefiles/sa.make Wed Mar 18 11:37:48 2009 -0400 11.2 +++ b/make/windows/makefiles/sa.make Thu Mar 19 09:13:24 2009 -0700 11.3 @@ -89,9 +89,11 @@ 11.4 SA_CFLAGS = /nologo $(MS_RUNTIME_OPTION) /W3 $(GX_OPTION) /Od /D "WIN32" /D "WIN64" /D "_WINDOWS" /D "_DEBUG" /D "_CONSOLE" /D "_MBCS" /YX /FD /c 11.5 !elseif "$(BUILDARCH)" == "amd64" 11.6 SA_CFLAGS = /nologo $(MS_RUNTIME_OPTION) /W3 $(GX_OPTION) /Od /D "WIN32" /D "WIN64" /D "_WINDOWS" /D "_DEBUG" /D "_CONSOLE" /D "_MBCS" /YX /FD /c 11.7 +!if "$(COMPILER_NAME)" == "VS2005" 11.8 # On amd64, VS2005 compiler requires bufferoverflowU.lib on the link command line, 11.9 # otherwise we get missing __security_check_cookie externals at link time. 11.10 SA_LINK_FLAGS = bufferoverflowU.lib 11.11 +!endif 11.12 !else 11.13 SA_CFLAGS = /nologo $(MS_RUNTIME_OPTION) /W3 /Gm $(GX_OPTION) /ZI /Od /D "WIN32" /D "_WINDOWS" /D "_DEBUG" /D "_CONSOLE" /D "_MBCS" /YX /FD /GZ /c 11.14 !endif
12.1 --- a/make/windows/makefiles/sanity.make Wed Mar 18 11:37:48 2009 -0400 12.2 +++ b/make/windows/makefiles/sanity.make Thu Mar 19 09:13:24 2009 -0700 12.3 @@ -27,9 +27,9 @@ 12.4 all: checkCL checkLink 12.5 12.6 checkCL: 12.7 - @ if "$(MSC_VER)" NEQ "1310" if "$(MSC_VER)" NEQ "1399" if "$(MSC_VER)" NEQ "1400" \ 12.8 + @ if "$(MSC_VER)" NEQ "1310" if "$(MSC_VER)" NEQ "1399" if "$(MSC_VER)" NEQ "1400" if "$(MSC_VER)" NEQ "1500" \ 12.9 echo *** WARNING *** unrecognized cl.exe version $(MSC_VER) ($(RAW_MSC_VER)). Use FORCE_MSC_VER to override automatic detection. 12.10 12.11 checkLink: 12.12 - @ if "$(LINK_VER)" NEQ "710" if "$(LINK_VER)" NEQ "800" \ 12.13 + @ if "$(LINK_VER)" NEQ "710" if "$(LINK_VER)" NEQ "800" if "$(LINK_VER)" NEQ "900" \ 12.14 echo *** WARNING *** unrecognized link.exe version $(LINK_VER) ($(RAW_LINK_VER)). Use FORCE_LINK_VER to override automatic detection.
13.1 --- a/src/cpu/sparc/vm/assembler_sparc.cpp Wed Mar 18 11:37:48 2009 -0400 13.2 +++ b/src/cpu/sparc/vm/assembler_sparc.cpp Thu Mar 19 09:13:24 2009 -0700 13.3 @@ -2767,6 +2767,268 @@ 13.4 } 13.5 13.6 13.7 +void MacroAssembler::check_klass_subtype(Register sub_klass, 13.8 + Register super_klass, 13.9 + Register temp_reg, 13.10 + Register temp2_reg, 13.11 + Label& L_success) { 13.12 + Label L_failure, L_pop_to_failure; 13.13 + check_klass_subtype_fast_path(sub_klass, super_klass, 13.14 + temp_reg, temp2_reg, 13.15 + &L_success, &L_failure, NULL); 13.16 + Register sub_2 = sub_klass; 13.17 + Register sup_2 = super_klass; 13.18 + if (!sub_2->is_global()) sub_2 = L0; 13.19 + if (!sup_2->is_global()) sup_2 = L1; 13.20 + 13.21 + save_frame_and_mov(0, sub_klass, sub_2, super_klass, sup_2); 13.22 + check_klass_subtype_slow_path(sub_2, sup_2, 13.23 + L2, L3, L4, L5, 13.24 + NULL, &L_pop_to_failure); 13.25 + 13.26 + // on success: 13.27 + restore(); 13.28 + ba(false, L_success); 13.29 + delayed()->nop(); 13.30 + 13.31 + // on failure: 13.32 + bind(L_pop_to_failure); 13.33 + restore(); 13.34 + bind(L_failure); 13.35 +} 13.36 + 13.37 + 13.38 +void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass, 13.39 + Register super_klass, 13.40 + Register temp_reg, 13.41 + Register temp2_reg, 13.42 + Label* L_success, 13.43 + Label* L_failure, 13.44 + Label* L_slow_path, 13.45 + RegisterConstant super_check_offset, 13.46 + Register instanceof_hack) { 13.47 + int sc_offset = (klassOopDesc::header_size() * HeapWordSize + 13.48 + Klass::secondary_super_cache_offset_in_bytes()); 13.49 + int sco_offset = (klassOopDesc::header_size() * HeapWordSize + 13.50 + Klass::super_check_offset_offset_in_bytes()); 13.51 + 13.52 + bool must_load_sco = (super_check_offset.constant_or_zero() == -1); 13.53 + bool need_slow_path = (must_load_sco || 13.54 + super_check_offset.constant_or_zero() == sco_offset); 13.55 + 13.56 + assert_different_registers(sub_klass, super_klass, temp_reg); 13.57 + if (super_check_offset.is_register()) { 13.58 + assert_different_registers(sub_klass, super_klass, 13.59 + super_check_offset.as_register()); 13.60 + } else if (must_load_sco) { 13.61 + assert(temp2_reg != noreg, "supply either a temp or a register offset"); 13.62 + } 13.63 + 13.64 + Label L_fallthrough; 13.65 + int label_nulls = 0; 13.66 + if (L_success == NULL) { L_success = &L_fallthrough; label_nulls++; } 13.67 + if (L_failure == NULL) { L_failure = &L_fallthrough; label_nulls++; } 13.68 + if (L_slow_path == NULL) { L_slow_path = &L_fallthrough; label_nulls++; } 13.69 + assert(label_nulls <= 1 || instanceof_hack != noreg || 13.70 + (L_slow_path == &L_fallthrough && label_nulls <= 2 && !need_slow_path), 13.71 + "at most one NULL in the batch, usually"); 13.72 + 13.73 + // Support for the instanceof hack, which uses delay slots to 13.74 + // set a destination register to zero or one. 13.75 + bool do_bool_sets = (instanceof_hack != noreg); 13.76 +#define BOOL_SET(bool_value) \ 13.77 + if (do_bool_sets && bool_value >= 0) \ 13.78 + set(bool_value, instanceof_hack) 13.79 +#define DELAYED_BOOL_SET(bool_value) \ 13.80 + if (do_bool_sets && bool_value >= 0) \ 13.81 + delayed()->set(bool_value, instanceof_hack); \ 13.82 + else delayed()->nop() 13.83 + // Hacked ba(), which may only be used just before L_fallthrough. 13.84 +#define FINAL_JUMP(label, bool_value) \ 13.85 + if (&(label) == &L_fallthrough) { \ 13.86 + BOOL_SET(bool_value); \ 13.87 + } else { \ 13.88 + ba((do_bool_sets && bool_value >= 0), label); \ 13.89 + DELAYED_BOOL_SET(bool_value); \ 13.90 + } 13.91 + 13.92 + // If the pointers are equal, we are done (e.g., String[] elements). 13.93 + // This self-check enables sharing of secondary supertype arrays among 13.94 + // non-primary types such as array-of-interface. Otherwise, each such 13.95 + // type would need its own customized SSA. 13.96 + // We move this check to the front of the fast path because many 13.97 + // type checks are in fact trivially successful in this manner, 13.98 + // so we get a nicely predicted branch right at the start of the check. 13.99 + cmp(super_klass, sub_klass); 13.100 + brx(Assembler::equal, do_bool_sets, Assembler::pn, *L_success); 13.101 + DELAYED_BOOL_SET(1); 13.102 + 13.103 + // Check the supertype display: 13.104 + if (must_load_sco) { 13.105 + // The super check offset is always positive... 13.106 + lduw(super_klass, sco_offset, temp2_reg); 13.107 + super_check_offset = RegisterConstant(temp2_reg); 13.108 + } 13.109 + ld_ptr(sub_klass, super_check_offset, temp_reg); 13.110 + cmp(super_klass, temp_reg); 13.111 + 13.112 + // This check has worked decisively for primary supers. 13.113 + // Secondary supers are sought in the super_cache ('super_cache_addr'). 13.114 + // (Secondary supers are interfaces and very deeply nested subtypes.) 13.115 + // This works in the same check above because of a tricky aliasing 13.116 + // between the super_cache and the primary super display elements. 13.117 + // (The 'super_check_addr' can address either, as the case requires.) 13.118 + // Note that the cache is updated below if it does not help us find 13.119 + // what we need immediately. 13.120 + // So if it was a primary super, we can just fail immediately. 13.121 + // Otherwise, it's the slow path for us (no success at this point). 13.122 + 13.123 + if (super_check_offset.is_register()) { 13.124 + brx(Assembler::equal, do_bool_sets, Assembler::pn, *L_success); 13.125 + delayed(); if (do_bool_sets) BOOL_SET(1); 13.126 + // if !do_bool_sets, sneak the next cmp into the delay slot: 13.127 + cmp(super_check_offset.as_register(), sc_offset); 13.128 + 13.129 + if (L_failure == &L_fallthrough) { 13.130 + brx(Assembler::equal, do_bool_sets, Assembler::pt, *L_slow_path); 13.131 + delayed()->nop(); 13.132 + BOOL_SET(0); // fallthrough on failure 13.133 + } else { 13.134 + brx(Assembler::notEqual, do_bool_sets, Assembler::pn, *L_failure); 13.135 + DELAYED_BOOL_SET(0); 13.136 + FINAL_JUMP(*L_slow_path, -1); // -1 => vanilla delay slot 13.137 + } 13.138 + } else if (super_check_offset.as_constant() == sc_offset) { 13.139 + // Need a slow path; fast failure is impossible. 13.140 + if (L_slow_path == &L_fallthrough) { 13.141 + brx(Assembler::equal, do_bool_sets, Assembler::pt, *L_success); 13.142 + DELAYED_BOOL_SET(1); 13.143 + } else { 13.144 + brx(Assembler::notEqual, false, Assembler::pn, *L_slow_path); 13.145 + delayed()->nop(); 13.146 + FINAL_JUMP(*L_success, 1); 13.147 + } 13.148 + } else { 13.149 + // No slow path; it's a fast decision. 13.150 + if (L_failure == &L_fallthrough) { 13.151 + brx(Assembler::equal, do_bool_sets, Assembler::pt, *L_success); 13.152 + DELAYED_BOOL_SET(1); 13.153 + BOOL_SET(0); 13.154 + } else { 13.155 + brx(Assembler::notEqual, do_bool_sets, Assembler::pn, *L_failure); 13.156 + DELAYED_BOOL_SET(0); 13.157 + FINAL_JUMP(*L_success, 1); 13.158 + } 13.159 + } 13.160 + 13.161 + bind(L_fallthrough); 13.162 + 13.163 +#undef final_jump 13.164 +#undef bool_set 13.165 +#undef DELAYED_BOOL_SET 13.166 +#undef final_jump 13.167 +} 13.168 + 13.169 + 13.170 +void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass, 13.171 + Register super_klass, 13.172 + Register count_temp, 13.173 + Register scan_temp, 13.174 + Register scratch_reg, 13.175 + Register coop_reg, 13.176 + Label* L_success, 13.177 + Label* L_failure) { 13.178 + assert_different_registers(sub_klass, super_klass, 13.179 + count_temp, scan_temp, scratch_reg, coop_reg); 13.180 + 13.181 + Label L_fallthrough, L_loop; 13.182 + int label_nulls = 0; 13.183 + if (L_success == NULL) { L_success = &L_fallthrough; label_nulls++; } 13.184 + if (L_failure == NULL) { L_failure = &L_fallthrough; label_nulls++; } 13.185 + assert(label_nulls <= 1, "at most one NULL in the batch"); 13.186 + 13.187 + // a couple of useful fields in sub_klass: 13.188 + int ss_offset = (klassOopDesc::header_size() * HeapWordSize + 13.189 + Klass::secondary_supers_offset_in_bytes()); 13.190 + int sc_offset = (klassOopDesc::header_size() * HeapWordSize + 13.191 + Klass::secondary_super_cache_offset_in_bytes()); 13.192 + 13.193 + // Do a linear scan of the secondary super-klass chain. 13.194 + // This code is rarely used, so simplicity is a virtue here. 13.195 + 13.196 +#ifndef PRODUCT 13.197 + int* pst_counter = &SharedRuntime::_partial_subtype_ctr; 13.198 + inc_counter((address) pst_counter, count_temp, scan_temp); 13.199 +#endif 13.200 + 13.201 + // We will consult the secondary-super array. 13.202 + ld_ptr(sub_klass, ss_offset, scan_temp); 13.203 + 13.204 + // Compress superclass if necessary. 13.205 + Register search_key = super_klass; 13.206 + bool decode_super_klass = false; 13.207 + if (UseCompressedOops) { 13.208 + if (coop_reg != noreg) { 13.209 + encode_heap_oop_not_null(super_klass, coop_reg); 13.210 + search_key = coop_reg; 13.211 + } else { 13.212 + encode_heap_oop_not_null(super_klass); 13.213 + decode_super_klass = true; // scarce temps! 13.214 + } 13.215 + // The superclass is never null; it would be a basic system error if a null 13.216 + // pointer were to sneak in here. Note that we have already loaded the 13.217 + // Klass::super_check_offset from the super_klass in the fast path, 13.218 + // so if there is a null in that register, we are already in the afterlife. 13.219 + } 13.220 + 13.221 + // Load the array length. (Positive movl does right thing on LP64.) 13.222 + lduw(scan_temp, arrayOopDesc::length_offset_in_bytes(), count_temp); 13.223 + 13.224 + // Check for empty secondary super list 13.225 + tst(count_temp); 13.226 + 13.227 + // Top of search loop 13.228 + bind(L_loop); 13.229 + br(Assembler::equal, false, Assembler::pn, *L_failure); 13.230 + delayed()->add(scan_temp, heapOopSize, scan_temp); 13.231 + assert(heapOopSize != 0, "heapOopSize should be initialized"); 13.232 + 13.233 + // Skip the array header in all array accesses. 13.234 + int elem_offset = arrayOopDesc::base_offset_in_bytes(T_OBJECT); 13.235 + elem_offset -= heapOopSize; // the scan pointer was pre-incremented also 13.236 + 13.237 + // Load next super to check 13.238 + if (UseCompressedOops) { 13.239 + // Don't use load_heap_oop; we don't want to decode the element. 13.240 + lduw( scan_temp, elem_offset, scratch_reg ); 13.241 + } else { 13.242 + ld_ptr( scan_temp, elem_offset, scratch_reg ); 13.243 + } 13.244 + 13.245 + // Look for Rsuper_klass on Rsub_klass's secondary super-class-overflow list 13.246 + cmp(scratch_reg, search_key); 13.247 + 13.248 + // A miss means we are NOT a subtype and need to keep looping 13.249 + brx(Assembler::notEqual, false, Assembler::pn, L_loop); 13.250 + delayed()->deccc(count_temp); // decrement trip counter in delay slot 13.251 + 13.252 + // Falling out the bottom means we found a hit; we ARE a subtype 13.253 + if (decode_super_klass) decode_heap_oop(super_klass); 13.254 + 13.255 + // Success. Cache the super we found and proceed in triumph. 13.256 + st_ptr(super_klass, sub_klass, sc_offset); 13.257 + 13.258 + if (L_success != &L_fallthrough) { 13.259 + ba(false, *L_success); 13.260 + delayed()->nop(); 13.261 + } 13.262 + 13.263 + bind(L_fallthrough); 13.264 +} 13.265 + 13.266 + 13.267 + 13.268 + 13.269 void MacroAssembler::biased_locking_enter(Register obj_reg, Register mark_reg, 13.270 Register temp_reg, 13.271 Label& done, Label* slow_case, 13.272 @@ -4316,7 +4578,13 @@ 13.273 13.274 void MacroAssembler::encode_heap_oop(Register src, Register dst) { 13.275 assert (UseCompressedOops, "must be compressed"); 13.276 + assert (Universe::heap() != NULL, "java heap should be initialized"); 13.277 + assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong"); 13.278 verify_oop(src); 13.279 + if (Universe::narrow_oop_base() == NULL) { 13.280 + srlx(src, LogMinObjAlignmentInBytes, dst); 13.281 + return; 13.282 + } 13.283 Label done; 13.284 if (src == dst) { 13.285 // optimize for frequent case src == dst 13.286 @@ -4338,26 +4606,39 @@ 13.287 13.288 void MacroAssembler::encode_heap_oop_not_null(Register r) { 13.289 assert (UseCompressedOops, "must be compressed"); 13.290 + assert (Universe::heap() != NULL, "java heap should be initialized"); 13.291 + assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong"); 13.292 verify_oop(r); 13.293 - sub(r, G6_heapbase, r); 13.294 + if (Universe::narrow_oop_base() != NULL) 13.295 + sub(r, G6_heapbase, r); 13.296 srlx(r, LogMinObjAlignmentInBytes, r); 13.297 } 13.298 13.299 void MacroAssembler::encode_heap_oop_not_null(Register src, Register dst) { 13.300 assert (UseCompressedOops, "must be compressed"); 13.301 + assert (Universe::heap() != NULL, "java heap should be initialized"); 13.302 + assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong"); 13.303 verify_oop(src); 13.304 - sub(src, G6_heapbase, dst); 13.305 - srlx(dst, LogMinObjAlignmentInBytes, dst); 13.306 + if (Universe::narrow_oop_base() == NULL) { 13.307 + srlx(src, LogMinObjAlignmentInBytes, dst); 13.308 + } else { 13.309 + sub(src, G6_heapbase, dst); 13.310 + srlx(dst, LogMinObjAlignmentInBytes, dst); 13.311 + } 13.312 } 13.313 13.314 // Same algorithm as oops.inline.hpp decode_heap_oop. 13.315 void MacroAssembler::decode_heap_oop(Register src, Register dst) { 13.316 assert (UseCompressedOops, "must be compressed"); 13.317 - Label done; 13.318 + assert (Universe::heap() != NULL, "java heap should be initialized"); 13.319 + assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong"); 13.320 sllx(src, LogMinObjAlignmentInBytes, dst); 13.321 - bpr(rc_nz, true, Assembler::pt, dst, done); 13.322 - delayed() -> add(dst, G6_heapbase, dst); // annuled if not taken 13.323 - bind(done); 13.324 + if (Universe::narrow_oop_base() != NULL) { 13.325 + Label done; 13.326 + bpr(rc_nz, true, Assembler::pt, dst, done); 13.327 + delayed() -> add(dst, G6_heapbase, dst); // annuled if not taken 13.328 + bind(done); 13.329 + } 13.330 verify_oop(dst); 13.331 } 13.332 13.333 @@ -4366,8 +4647,11 @@ 13.334 // pd_code_size_limit. 13.335 // Also do not verify_oop as this is called by verify_oop. 13.336 assert (UseCompressedOops, "must be compressed"); 13.337 + assert (Universe::heap() != NULL, "java heap should be initialized"); 13.338 + assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong"); 13.339 sllx(r, LogMinObjAlignmentInBytes, r); 13.340 - add(r, G6_heapbase, r); 13.341 + if (Universe::narrow_oop_base() != NULL) 13.342 + add(r, G6_heapbase, r); 13.343 } 13.344 13.345 void MacroAssembler::decode_heap_oop_not_null(Register src, Register dst) { 13.346 @@ -4375,14 +4659,17 @@ 13.347 // pd_code_size_limit. 13.348 // Also do not verify_oop as this is called by verify_oop. 13.349 assert (UseCompressedOops, "must be compressed"); 13.350 + assert (Universe::heap() != NULL, "java heap should be initialized"); 13.351 + assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong"); 13.352 sllx(src, LogMinObjAlignmentInBytes, dst); 13.353 - add(dst, G6_heapbase, dst); 13.354 + if (Universe::narrow_oop_base() != NULL) 13.355 + add(dst, G6_heapbase, dst); 13.356 } 13.357 13.358 void MacroAssembler::reinit_heapbase() { 13.359 if (UseCompressedOops) { 13.360 // call indirectly to solve generation ordering problem 13.361 - Address base(G6_heapbase, (address)Universe::heap_base_addr()); 13.362 + Address base(G6_heapbase, (address)Universe::narrow_oop_base_addr()); 13.363 load_ptr_contents(base, G6_heapbase); 13.364 } 13.365 }
14.1 --- a/src/cpu/sparc/vm/assembler_sparc.hpp Wed Mar 18 11:37:48 2009 -0400 14.2 +++ b/src/cpu/sparc/vm/assembler_sparc.hpp Thu Mar 19 09:13:24 2009 -0700 14.3 @@ -2327,6 +2327,46 @@ 14.4 Register temp_reg, Register temp2_reg, 14.5 Label& no_such_interface); 14.6 14.7 + // Test sub_klass against super_klass, with fast and slow paths. 14.8 + 14.9 + // The fast path produces a tri-state answer: yes / no / maybe-slow. 14.10 + // One of the three labels can be NULL, meaning take the fall-through. 14.11 + // If super_check_offset is -1, the value is loaded up from super_klass. 14.12 + // No registers are killed, except temp_reg and temp2_reg. 14.13 + // If super_check_offset is not -1, temp2_reg is not used and can be noreg. 14.14 + void check_klass_subtype_fast_path(Register sub_klass, 14.15 + Register super_klass, 14.16 + Register temp_reg, 14.17 + Register temp2_reg, 14.18 + Label* L_success, 14.19 + Label* L_failure, 14.20 + Label* L_slow_path, 14.21 + RegisterConstant super_check_offset = RegisterConstant(-1), 14.22 + Register instanceof_hack = noreg); 14.23 + 14.24 + // The rest of the type check; must be wired to a corresponding fast path. 14.25 + // It does not repeat the fast path logic, so don't use it standalone. 14.26 + // The temp_reg can be noreg, if no temps are available. 14.27 + // It can also be sub_klass or super_klass, meaning it's OK to kill that one. 14.28 + // Updates the sub's secondary super cache as necessary. 14.29 + void check_klass_subtype_slow_path(Register sub_klass, 14.30 + Register super_klass, 14.31 + Register temp_reg, 14.32 + Register temp2_reg, 14.33 + Register temp3_reg, 14.34 + Register temp4_reg, 14.35 + Label* L_success, 14.36 + Label* L_failure); 14.37 + 14.38 + // Simplified, combined version, good for typical uses. 14.39 + // Falls through on failure. 14.40 + void check_klass_subtype(Register sub_klass, 14.41 + Register super_klass, 14.42 + Register temp_reg, 14.43 + Register temp2_reg, 14.44 + Label& L_success); 14.45 + 14.46 + 14.47 // Stack overflow checking 14.48 14.49 // Note: this clobbers G3_scratch
15.1 --- a/src/cpu/sparc/vm/c1_LIRAssembler_sparc.cpp Wed Mar 18 11:37:48 2009 -0400 15.2 +++ b/src/cpu/sparc/vm/c1_LIRAssembler_sparc.cpp Thu Mar 19 09:13:24 2009 -0700 15.3 @@ -2393,23 +2393,11 @@ 15.4 15.5 // get instance klass 15.6 load(k_RInfo, objArrayKlass::element_klass_offset_in_bytes() + sizeof(oopDesc), k_RInfo, T_OBJECT, NULL); 15.7 - // get super_check_offset 15.8 - load(k_RInfo, sizeof(oopDesc) + Klass::super_check_offset_offset_in_bytes(), Rtmp1, T_INT, NULL); 15.9 - // See if we get an immediate positive hit 15.10 - __ ld_ptr(klass_RInfo, Rtmp1, FrameMap::O7_oop_opr->as_register()); 15.11 - __ cmp(k_RInfo, O7); 15.12 - __ br(Assembler::equal, false, Assembler::pn, done); 15.13 - __ delayed()->nop(); 15.14 - // check for immediate negative hit 15.15 - __ cmp(Rtmp1, sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes()); 15.16 - __ br(Assembler::notEqual, false, Assembler::pn, *stub->entry()); 15.17 - __ delayed()->nop(); 15.18 - // check for self 15.19 - __ cmp(klass_RInfo, k_RInfo); 15.20 - __ br(Assembler::equal, false, Assembler::pn, done); 15.21 - __ delayed()->nop(); 15.22 - 15.23 - // assert(sub.is_same(FrameMap::G3_RInfo) && super.is_same(FrameMap::G1_RInfo), "incorrect call setup"); 15.24 + // perform the fast part of the checking logic 15.25 + __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, O7, &done, stub->entry(), NULL); 15.26 + 15.27 + // call out-of-line instance of __ check_klass_subtype_slow_path(...): 15.28 + assert(klass_RInfo == G3 && k_RInfo == G1, "incorrect call setup"); 15.29 __ call(Runtime1::entry_for(Runtime1::slow_subtype_check_id), relocInfo::runtime_call_type); 15.30 __ delayed()->nop(); 15.31 __ cmp(G3, 0); 15.32 @@ -2493,58 +2481,30 @@ 15.33 __ delayed()->nop(); 15.34 __ bind(done); 15.35 } else { 15.36 + bool need_slow_path = true; 15.37 if (k->is_loaded()) { 15.38 - load(klass_RInfo, k->super_check_offset(), Rtmp1, T_OBJECT, NULL); 15.39 - 15.40 - if (sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes() != k->super_check_offset()) { 15.41 - // See if we get an immediate positive hit 15.42 - __ cmp(Rtmp1, k_RInfo ); 15.43 - __ br(Assembler::notEqual, false, Assembler::pn, *stub->entry()); 15.44 - __ delayed()->nop(); 15.45 - } else { 15.46 - // See if we get an immediate positive hit 15.47 - assert_different_registers(Rtmp1, k_RInfo, klass_RInfo); 15.48 - __ cmp(Rtmp1, k_RInfo ); 15.49 - __ br(Assembler::equal, false, Assembler::pn, done); 15.50 - // check for self 15.51 - __ delayed()->cmp(klass_RInfo, k_RInfo); 15.52 - __ br(Assembler::equal, false, Assembler::pn, done); 15.53 - __ delayed()->nop(); 15.54 - 15.55 - // assert(sub.is_same(FrameMap::G3_RInfo) && super.is_same(FrameMap::G1_RInfo), "incorrect call setup"); 15.56 - __ call(Runtime1::entry_for(Runtime1::slow_subtype_check_id), relocInfo::runtime_call_type); 15.57 - __ delayed()->nop(); 15.58 - __ cmp(G3, 0); 15.59 - __ br(Assembler::equal, false, Assembler::pn, *stub->entry()); 15.60 - __ delayed()->nop(); 15.61 - } 15.62 - __ bind(done); 15.63 + if (k->super_check_offset() != sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes()) 15.64 + need_slow_path = false; 15.65 + // perform the fast part of the checking logic 15.66 + __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, noreg, 15.67 + (need_slow_path ? &done : NULL), 15.68 + stub->entry(), NULL, 15.69 + RegisterConstant(k->super_check_offset())); 15.70 } else { 15.71 - assert_different_registers(Rtmp1, klass_RInfo, k_RInfo); 15.72 - 15.73 - load(k_RInfo, sizeof(oopDesc) + Klass::super_check_offset_offset_in_bytes(), Rtmp1, T_INT, NULL); 15.74 - // See if we get an immediate positive hit 15.75 - load(klass_RInfo, Rtmp1, FrameMap::O7_oop_opr, T_OBJECT); 15.76 - __ cmp(k_RInfo, O7); 15.77 - __ br(Assembler::equal, false, Assembler::pn, done); 15.78 - __ delayed()->nop(); 15.79 - // check for immediate negative hit 15.80 - __ cmp(Rtmp1, sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes()); 15.81 - __ br(Assembler::notEqual, false, Assembler::pn, *stub->entry()); 15.82 - // check for self 15.83 - __ delayed()->cmp(klass_RInfo, k_RInfo); 15.84 - __ br(Assembler::equal, false, Assembler::pn, done); 15.85 - __ delayed()->nop(); 15.86 - 15.87 - // assert(sub.is_same(FrameMap::G3_RInfo) && super.is_same(FrameMap::G1_RInfo), "incorrect call setup"); 15.88 + // perform the fast part of the checking logic 15.89 + __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, O7, 15.90 + &done, stub->entry(), NULL); 15.91 + } 15.92 + if (need_slow_path) { 15.93 + // call out-of-line instance of __ check_klass_subtype_slow_path(...): 15.94 + assert(klass_RInfo == G3 && k_RInfo == G1, "incorrect call setup"); 15.95 __ call(Runtime1::entry_for(Runtime1::slow_subtype_check_id), relocInfo::runtime_call_type); 15.96 __ delayed()->nop(); 15.97 __ cmp(G3, 0); 15.98 __ br(Assembler::equal, false, Assembler::pn, *stub->entry()); 15.99 __ delayed()->nop(); 15.100 - __ bind(done); 15.101 } 15.102 - 15.103 + __ bind(done); 15.104 } 15.105 __ mov(obj, dst); 15.106 } else if (code == lir_instanceof) { 15.107 @@ -2582,58 +2542,32 @@ 15.108 __ set(0, dst); 15.109 __ bind(done); 15.110 } else { 15.111 + bool need_slow_path = true; 15.112 if (k->is_loaded()) { 15.113 - assert_different_registers(Rtmp1, klass_RInfo, k_RInfo); 15.114 - load(klass_RInfo, k->super_check_offset(), Rtmp1, T_OBJECT, NULL); 15.115 - 15.116 - if (sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes() != k->super_check_offset()) { 15.117 - // See if we get an immediate positive hit 15.118 - __ cmp(Rtmp1, k_RInfo ); 15.119 - __ br(Assembler::equal, true, Assembler::pt, done); 15.120 - __ delayed()->set(1, dst); 15.121 - __ set(0, dst); 15.122 - __ bind(done); 15.123 - } else { 15.124 - // See if we get an immediate positive hit 15.125 - assert_different_registers(Rtmp1, k_RInfo, klass_RInfo); 15.126 - __ cmp(Rtmp1, k_RInfo ); 15.127 - __ br(Assembler::equal, true, Assembler::pt, done); 15.128 - __ delayed()->set(1, dst); 15.129 - // check for self 15.130 - __ cmp(klass_RInfo, k_RInfo); 15.131 - __ br(Assembler::equal, true, Assembler::pt, done); 15.132 - __ delayed()->set(1, dst); 15.133 - 15.134 - // assert(sub.is_same(FrameMap::G3_RInfo) && super.is_same(FrameMap::G1_RInfo), "incorrect call setup"); 15.135 - __ call(Runtime1::entry_for(Runtime1::slow_subtype_check_id), relocInfo::runtime_call_type); 15.136 - __ delayed()->nop(); 15.137 - __ mov(G3, dst); 15.138 - __ bind(done); 15.139 - } 15.140 + if (k->super_check_offset() != sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes()) 15.141 + need_slow_path = false; 15.142 + // perform the fast part of the checking logic 15.143 + __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, O7, noreg, 15.144 + (need_slow_path ? &done : NULL), 15.145 + (need_slow_path ? &done : NULL), NULL, 15.146 + RegisterConstant(k->super_check_offset()), 15.147 + dst); 15.148 } else { 15.149 assert(dst != klass_RInfo && dst != k_RInfo, "need 3 registers"); 15.150 - 15.151 - load(k_RInfo, sizeof(oopDesc) + Klass::super_check_offset_offset_in_bytes(), dst, T_INT, NULL); 15.152 - // See if we get an immediate positive hit 15.153 - load(klass_RInfo, dst, FrameMap::O7_oop_opr, T_OBJECT); 15.154 - __ cmp(k_RInfo, O7); 15.155 - __ br(Assembler::equal, true, Assembler::pt, done); 15.156 - __ delayed()->set(1, dst); 15.157 - // check for immediate negative hit 15.158 - __ cmp(dst, sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes()); 15.159 - __ br(Assembler::notEqual, true, Assembler::pt, done); 15.160 - __ delayed()->set(0, dst); 15.161 - // check for self 15.162 - __ cmp(klass_RInfo, k_RInfo); 15.163 - __ br(Assembler::equal, true, Assembler::pt, done); 15.164 - __ delayed()->set(1, dst); 15.165 - 15.166 - // assert(sub.is_same(FrameMap::G3_RInfo) && super.is_same(FrameMap::G1_RInfo), "incorrect call setup"); 15.167 + // perform the fast part of the checking logic 15.168 + __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, O7, dst, 15.169 + &done, &done, NULL, 15.170 + RegisterConstant(-1), 15.171 + dst); 15.172 + } 15.173 + if (need_slow_path) { 15.174 + // call out-of-line instance of __ check_klass_subtype_slow_path(...): 15.175 + assert(klass_RInfo == G3 && k_RInfo == G1, "incorrect call setup"); 15.176 __ call(Runtime1::entry_for(Runtime1::slow_subtype_check_id), relocInfo::runtime_call_type); 15.177 __ delayed()->nop(); 15.178 __ mov(G3, dst); 15.179 - __ bind(done); 15.180 } 15.181 + __ bind(done); 15.182 } 15.183 } else { 15.184 ShouldNotReachHere();
16.1 --- a/src/cpu/sparc/vm/c1_Runtime1_sparc.cpp Wed Mar 18 11:37:48 2009 -0400 16.2 +++ b/src/cpu/sparc/vm/c1_Runtime1_sparc.cpp Thu Mar 19 09:13:24 2009 -0700 16.3 @@ -714,38 +714,19 @@ 16.4 // sub : G3, argument, destroyed 16.5 // super: G1, argument, not changed 16.6 // raddr: O7, blown by call 16.7 - Label loop, miss; 16.8 + Label miss; 16.9 16.10 __ save_frame(0); // Blow no registers! 16.11 16.12 - __ ld_ptr( G3, sizeof(oopDesc) + Klass::secondary_supers_offset_in_bytes(), L3 ); 16.13 - __ lduw(L3,arrayOopDesc::length_offset_in_bytes(),L0); // length in l0 16.14 - __ add(L3,arrayOopDesc::base_offset_in_bytes(T_OBJECT),L1); // ptr into array 16.15 - __ clr(L4); // Index 16.16 - // Load a little early; will load 1 off the end of the array. 16.17 - // Ok for now; revisit if we have other uses of this routine. 16.18 - __ ld_ptr(L1,0,L2); // Will load a little early 16.19 - 16.20 - // The scan loop 16.21 - __ bind(loop); 16.22 - __ add(L1,wordSize,L1); // Bump by OOP size 16.23 - __ cmp(L4,L0); 16.24 - __ br(Assembler::equal,false,Assembler::pn,miss); 16.25 - __ delayed()->inc(L4); // Bump index 16.26 - __ subcc(L2,G1,L3); // Check for match; zero in L3 for a hit 16.27 - __ brx( Assembler::notEqual, false, Assembler::pt, loop ); 16.28 - __ delayed()->ld_ptr(L1,0,L2); // Will load a little early 16.29 - 16.30 - // Got a hit; report success; set cache 16.31 - __ st_ptr( G1, G3, sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes() ); 16.32 + __ check_klass_subtype_slow_path(G3, G1, L0, L1, L2, L4, NULL, &miss); 16.33 16.34 __ mov(1, G3); 16.35 - __ ret(); // Result in G5 is ok; flags set 16.36 + __ ret(); // Result in G5 is 'true' 16.37 __ delayed()->restore(); // free copy or add can go here 16.38 16.39 __ bind(miss); 16.40 __ mov(0, G3); 16.41 - __ ret(); // Result in G5 is ok; flags set 16.42 + __ ret(); // Result in G5 is 'false' 16.43 __ delayed()->restore(); // free copy or add can go here 16.44 } 16.45
17.1 --- a/src/cpu/sparc/vm/interp_masm_sparc.cpp Wed Mar 18 11:37:48 2009 -0400 17.2 +++ b/src/cpu/sparc/vm/interp_masm_sparc.cpp Thu Mar 19 09:13:24 2009 -0700 17.3 @@ -866,65 +866,18 @@ 17.4 Register Rtmp2, 17.5 Register Rtmp3, 17.6 Label &ok_is_subtype ) { 17.7 - Label not_subtype, loop; 17.8 + Label not_subtype; 17.9 17.10 // Profile the not-null value's klass. 17.11 profile_typecheck(Rsub_klass, Rtmp1); 17.12 17.13 - // Load the super-klass's check offset into Rtmp1 17.14 - ld( Rsuper_klass, sizeof(oopDesc) + Klass::super_check_offset_offset_in_bytes(), Rtmp1 ); 17.15 - // Load from the sub-klass's super-class display list, or a 1-word cache of 17.16 - // the secondary superclass list, or a failing value with a sentinel offset 17.17 - // if the super-klass is an interface or exceptionally deep in the Java 17.18 - // hierarchy and we have to scan the secondary superclass list the hard way. 17.19 - ld_ptr( Rsub_klass, Rtmp1, Rtmp2 ); 17.20 - // See if we get an immediate positive hit 17.21 - cmp( Rtmp2, Rsuper_klass ); 17.22 - brx( Assembler::equal, false, Assembler::pt, ok_is_subtype ); 17.23 - // In the delay slot, check for immediate negative hit 17.24 - delayed()->cmp( Rtmp1, sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes() ); 17.25 - br( Assembler::notEqual, false, Assembler::pt, not_subtype ); 17.26 - // In the delay slot, check for self 17.27 - delayed()->cmp( Rsub_klass, Rsuper_klass ); 17.28 - brx( Assembler::equal, false, Assembler::pt, ok_is_subtype ); 17.29 - 17.30 - // Now do a linear scan of the secondary super-klass chain. 17.31 - delayed()->ld_ptr( Rsub_klass, sizeof(oopDesc) + Klass::secondary_supers_offset_in_bytes(), Rtmp2 ); 17.32 - 17.33 - // compress superclass 17.34 - if (UseCompressedOops) encode_heap_oop(Rsuper_klass); 17.35 - 17.36 - // Rtmp2 holds the objArrayOop of secondary supers. 17.37 - ld( Rtmp2, arrayOopDesc::length_offset_in_bytes(), Rtmp1 );// Load the array length 17.38 - // Check for empty secondary super list 17.39 - tst(Rtmp1); 17.40 - 17.41 - // Top of search loop 17.42 - bind( loop ); 17.43 - br( Assembler::equal, false, Assembler::pn, not_subtype ); 17.44 - delayed()->nop(); 17.45 - 17.46 - // load next super to check 17.47 - if (UseCompressedOops) { 17.48 - lduw( Rtmp2, arrayOopDesc::base_offset_in_bytes(T_OBJECT), Rtmp3); 17.49 - // Bump array pointer forward one oop 17.50 - add( Rtmp2, 4, Rtmp2 ); 17.51 - } else { 17.52 - ld_ptr( Rtmp2, arrayOopDesc::base_offset_in_bytes(T_OBJECT), Rtmp3); 17.53 - // Bump array pointer forward one oop 17.54 - add( Rtmp2, wordSize, Rtmp2); 17.55 - } 17.56 - // Look for Rsuper_klass on Rsub_klass's secondary super-class-overflow list 17.57 - cmp( Rtmp3, Rsuper_klass ); 17.58 - // A miss means we are NOT a subtype and need to keep looping 17.59 - brx( Assembler::notEqual, false, Assembler::pt, loop ); 17.60 - delayed()->deccc( Rtmp1 ); // dec trip counter in delay slot 17.61 - // Falling out the bottom means we found a hit; we ARE a subtype 17.62 - if (UseCompressedOops) decode_heap_oop(Rsuper_klass); 17.63 - br( Assembler::always, false, Assembler::pt, ok_is_subtype ); 17.64 - // Update the cache 17.65 - delayed()->st_ptr( Rsuper_klass, Rsub_klass, 17.66 - sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes() ); 17.67 + check_klass_subtype_fast_path(Rsub_klass, Rsuper_klass, 17.68 + Rtmp1, Rtmp2, 17.69 + &ok_is_subtype, ¬_subtype, NULL); 17.70 + 17.71 + check_klass_subtype_slow_path(Rsub_klass, Rsuper_klass, 17.72 + Rtmp1, Rtmp2, Rtmp3, /*hack:*/ noreg, 17.73 + &ok_is_subtype, NULL); 17.74 17.75 bind(not_subtype); 17.76 profile_typecheck_failed(Rtmp1);
18.1 --- a/src/cpu/sparc/vm/sparc.ad Wed Mar 18 11:37:48 2009 -0400 18.2 +++ b/src/cpu/sparc/vm/sparc.ad Thu Mar 19 09:13:24 2009 -0700 18.3 @@ -547,7 +547,11 @@ 18.4 int v_off = entry_offset*wordSize + vtableEntry::method_offset_in_bytes(); 18.5 int klass_load_size; 18.6 if (UseCompressedOops) { 18.7 - klass_load_size = 3*BytesPerInstWord; // see MacroAssembler::load_klass() 18.8 + assert(Universe::heap() != NULL, "java heap should be initialized"); 18.9 + if (Universe::narrow_oop_base() == NULL) 18.10 + klass_load_size = 2*BytesPerInstWord; // see MacroAssembler::load_klass() 18.11 + else 18.12 + klass_load_size = 3*BytesPerInstWord; 18.13 } else { 18.14 klass_load_size = 1*BytesPerInstWord; 18.15 } 18.16 @@ -1601,9 +1605,11 @@ 18.17 st->print_cr("\nUEP:"); 18.18 #ifdef _LP64 18.19 if (UseCompressedOops) { 18.20 + assert(Universe::heap() != NULL, "java heap should be initialized"); 18.21 st->print_cr("\tLDUW [R_O0 + oopDesc::klass_offset_in_bytes],R_G5\t! Inline cache check - compressed klass"); 18.22 st->print_cr("\tSLL R_G5,3,R_G5"); 18.23 - st->print_cr("\tADD R_G5,R_G6_heap_base,R_G5"); 18.24 + if (Universe::narrow_oop_base() != NULL) 18.25 + st->print_cr("\tADD R_G5,R_G6_heap_base,R_G5"); 18.26 } else { 18.27 st->print_cr("\tLDX [R_O0 + oopDesc::klass_offset_in_bytes],R_G5\t! Inline cache check"); 18.28 } 18.29 @@ -2502,7 +2508,11 @@ 18.30 __ load_klass(O0, G3_scratch); 18.31 int klass_load_size; 18.32 if (UseCompressedOops) { 18.33 - klass_load_size = 3*BytesPerInstWord; 18.34 + assert(Universe::heap() != NULL, "java heap should be initialized"); 18.35 + if (Universe::narrow_oop_base() == NULL) 18.36 + klass_load_size = 2*BytesPerInstWord; 18.37 + else 18.38 + klass_load_size = 3*BytesPerInstWord; 18.39 } else { 18.40 klass_load_size = 1*BytesPerInstWord; 18.41 } 18.42 @@ -9005,6 +9015,33 @@ 18.43 ins_pipe(long_memory_op); 18.44 %} 18.45 18.46 + 18.47 +//---------- Population Count Instructions ------------------------------------- 18.48 + 18.49 +instruct popCountI(iRegI dst, iRegI src) %{ 18.50 + predicate(UsePopCountInstruction); 18.51 + match(Set dst (PopCountI src)); 18.52 + 18.53 + format %{ "POPC $src, $dst" %} 18.54 + ins_encode %{ 18.55 + __ popc($src$$Register, $dst$$Register); 18.56 + %} 18.57 + ins_pipe(ialu_reg); 18.58 +%} 18.59 + 18.60 +// Note: Long.bitCount(long) returns an int. 18.61 +instruct popCountL(iRegI dst, iRegL src) %{ 18.62 + predicate(UsePopCountInstruction); 18.63 + match(Set dst (PopCountL src)); 18.64 + 18.65 + format %{ "POPC $src, $dst" %} 18.66 + ins_encode %{ 18.67 + __ popc($src$$Register, $dst$$Register); 18.68 + %} 18.69 + ins_pipe(ialu_reg); 18.70 +%} 18.71 + 18.72 + 18.73 // ============================================================================ 18.74 //------------Bytes reverse-------------------------------------------------- 18.75
19.1 --- a/src/cpu/sparc/vm/stubGenerator_sparc.cpp Wed Mar 18 11:37:48 2009 -0400 19.2 +++ b/src/cpu/sparc/vm/stubGenerator_sparc.cpp Thu Mar 19 09:13:24 2009 -0700 19.3 @@ -900,19 +900,7 @@ 19.4 __ align(CodeEntryAlignment); 19.5 StubCodeMark mark(this, "StubRoutines", "partial_subtype_check"); 19.6 address start = __ pc(); 19.7 - Label loop, miss; 19.8 - 19.9 - // Compare super with sub directly, since super is not in its own SSA. 19.10 - // The compiler used to emit this test, but we fold it in here, 19.11 - // to increase overall code density, with no real loss of speed. 19.12 - { Label L; 19.13 - __ cmp(O1, O2); 19.14 - __ brx(Assembler::notEqual, false, Assembler::pt, L); 19.15 - __ delayed()->nop(); 19.16 - __ retl(); 19.17 - __ delayed()->addcc(G0,0,O0); // set Z flags, zero result 19.18 - __ bind(L); 19.19 - } 19.20 + Label miss; 19.21 19.22 #if defined(COMPILER2) && !defined(_LP64) 19.23 // Do not use a 'save' because it blows the 64-bit O registers. 19.24 @@ -936,56 +924,12 @@ 19.25 Register L2_super = L2; 19.26 Register L3_index = L3; 19.27 19.28 -#ifdef _LP64 19.29 - Register L4_ooptmp = L4; 19.30 - 19.31 - if (UseCompressedOops) { 19.32 - // this must be under UseCompressedOops check, as we rely upon fact 19.33 - // that L4 not clobbered in C2 on 32-bit platforms, where we do explicit save 19.34 - // on stack, see several lines above 19.35 - __ encode_heap_oop(Rsuper, L4_ooptmp); 19.36 - } 19.37 -#endif 19.38 - 19.39 - inc_counter_np(SharedRuntime::_partial_subtype_ctr, L0, L1); 19.40 - 19.41 - __ ld_ptr( Rsub, sizeof(oopDesc) + Klass::secondary_supers_offset_in_bytes(), L3 ); 19.42 - __ lduw(L3,arrayOopDesc::length_offset_in_bytes(),L0_ary_len); 19.43 - __ add(L3,arrayOopDesc::base_offset_in_bytes(T_OBJECT),L1_ary_ptr); 19.44 - __ clr(L3_index); // zero index 19.45 - // Load a little early; will load 1 off the end of the array. 19.46 - // Ok for now; revisit if we have other uses of this routine. 19.47 - if (UseCompressedOops) { 19.48 - __ lduw(L1_ary_ptr,0,L2_super);// Will load a little early 19.49 - } else { 19.50 - __ ld_ptr(L1_ary_ptr,0,L2_super);// Will load a little early 19.51 - } 19.52 - 19.53 - assert(heapOopSize != 0, "heapOopSize should be initialized"); 19.54 - // The scan loop 19.55 - __ BIND(loop); 19.56 - __ add(L1_ary_ptr, heapOopSize, L1_ary_ptr); // Bump by OOP size 19.57 - __ cmp(L3_index,L0_ary_len); 19.58 - __ br(Assembler::equal,false,Assembler::pn,miss); 19.59 - __ delayed()->inc(L3_index); // Bump index 19.60 - 19.61 - if (UseCompressedOops) { 19.62 -#ifdef _LP64 19.63 - __ subcc(L2_super,L4_ooptmp,Rret); // Check for match; zero in Rret for a hit 19.64 - __ br( Assembler::notEqual, false, Assembler::pt, loop ); 19.65 - __ delayed()->lduw(L1_ary_ptr,0,L2_super);// Will load a little early 19.66 -#else 19.67 - ShouldNotReachHere(); 19.68 -#endif 19.69 - } else { 19.70 - __ subcc(L2_super,Rsuper,Rret); // Check for match; zero in Rret for a hit 19.71 - __ brx( Assembler::notEqual, false, Assembler::pt, loop ); 19.72 - __ delayed()->ld_ptr(L1_ary_ptr,0,L2_super);// Will load a little early 19.73 - } 19.74 - 19.75 - // Got a hit; report success; set cache. Cache load doesn't 19.76 - // happen here; for speed it is directly emitted by the compiler. 19.77 - __ st_ptr( Rsuper, Rsub, sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes() ); 19.78 + __ check_klass_subtype_slow_path(Rsub, Rsuper, 19.79 + L0, L1, L2, L3, 19.80 + NULL, &miss); 19.81 + 19.82 + // Match falls through here. 19.83 + __ addcc(G0,0,Rret); // set Z flags, Z result 19.84 19.85 #if defined(COMPILER2) && !defined(_LP64) 19.86 __ ld_ptr(SP,(frame::register_save_words+0)*wordSize,L0); 19.87 @@ -999,7 +943,6 @@ 19.88 __ delayed()->restore(); 19.89 #endif 19.90 19.91 - // Hit or miss falls through here 19.92 __ BIND(miss); 19.93 __ addcc(G0,1,Rret); // set NZ flags, NZ result 19.94 19.95 @@ -2330,51 +2273,31 @@ 19.96 Register super_check_offset, 19.97 Register super_klass, 19.98 Register temp, 19.99 - Label& L_success, 19.100 - Register deccc_hack = noreg) { 19.101 + Label& L_success) { 19.102 assert_different_registers(sub_klass, super_check_offset, super_klass, temp); 19.103 19.104 BLOCK_COMMENT("type_check:"); 19.105 19.106 - Label L_miss; 19.107 + Label L_miss, L_pop_to_miss; 19.108 19.109 assert_clean_int(super_check_offset, temp); 19.110 19.111 - // maybe decrement caller's trip count: 19.112 -#define DELAY_SLOT delayed(); \ 19.113 - { if (deccc_hack == noreg) __ nop(); else __ deccc(deccc_hack); } 19.114 - 19.115 - // if the pointers are equal, we are done (e.g., String[] elements) 19.116 - __ cmp(sub_klass, super_klass); 19.117 - __ brx(Assembler::equal, true, Assembler::pt, L_success); 19.118 - __ DELAY_SLOT; 19.119 - 19.120 - // check the supertype display: 19.121 - __ ld_ptr(sub_klass, super_check_offset, temp); // query the super type 19.122 - __ cmp(super_klass, temp); // test the super type 19.123 - __ brx(Assembler::equal, true, Assembler::pt, L_success); 19.124 - __ DELAY_SLOT; 19.125 - 19.126 - int sc_offset = (klassOopDesc::header_size() * HeapWordSize + 19.127 - Klass::secondary_super_cache_offset_in_bytes()); 19.128 - __ cmp(super_klass, sc_offset); 19.129 - __ brx(Assembler::notEqual, true, Assembler::pt, L_miss); 19.130 - __ delayed()->nop(); 19.131 - 19.132 + __ check_klass_subtype_fast_path(sub_klass, super_klass, temp, noreg, 19.133 + &L_success, &L_miss, NULL, 19.134 + super_check_offset); 19.135 + 19.136 + BLOCK_COMMENT("type_check_slow_path:"); 19.137 __ save_frame(0); 19.138 - __ mov(sub_klass->after_save(), O1); 19.139 - // mov(super_klass->after_save(), O2); //fill delay slot 19.140 - assert(StubRoutines::Sparc::_partial_subtype_check != NULL, "order of generation"); 19.141 - __ call(StubRoutines::Sparc::_partial_subtype_check); 19.142 - __ delayed()->mov(super_klass->after_save(), O2); 19.143 + __ check_klass_subtype_slow_path(sub_klass->after_save(), 19.144 + super_klass->after_save(), 19.145 + L0, L1, L2, L4, 19.146 + NULL, &L_pop_to_miss); 19.147 + __ ba(false, L_success); 19.148 + __ delayed()->restore(); 19.149 + 19.150 + __ bind(L_pop_to_miss); 19.151 __ restore(); 19.152 19.153 - // Upon return, the condition codes are already set. 19.154 - __ brx(Assembler::equal, true, Assembler::pt, L_success); 19.155 - __ DELAY_SLOT; 19.156 - 19.157 -#undef DELAY_SLOT 19.158 - 19.159 // Fall through on failure! 19.160 __ BIND(L_miss); 19.161 } 19.162 @@ -2411,7 +2334,7 @@ 19.163 gen_write_ref_array_pre_barrier(O1, O2); 19.164 19.165 #ifdef ASSERT 19.166 - // We sometimes save a frame (see partial_subtype_check below). 19.167 + // We sometimes save a frame (see generate_type_check below). 19.168 // If this will cause trouble, let's fail now instead of later. 19.169 __ save_frame(0); 19.170 __ restore(); 19.171 @@ -2455,41 +2378,39 @@ 19.172 // G3, G4, G5 --- current oop, oop.klass, oop.klass.super 19.173 __ align(16); 19.174 19.175 - __ bind(store_element); 19.176 - // deccc(G1_remain); // decrement the count (hoisted) 19.177 + __ BIND(store_element); 19.178 + __ deccc(G1_remain); // decrement the count 19.179 __ store_heap_oop(G3_oop, O1_to, O5_offset); // store the oop 19.180 __ inc(O5_offset, heapOopSize); // step to next offset 19.181 __ brx(Assembler::zero, true, Assembler::pt, do_card_marks); 19.182 __ delayed()->set(0, O0); // return -1 on success 19.183 19.184 // ======== loop entry is here ======== 19.185 - __ bind(load_element); 19.186 + __ BIND(load_element); 19.187 __ load_heap_oop(O0_from, O5_offset, G3_oop); // load the oop 19.188 __ br_null(G3_oop, true, Assembler::pt, store_element); 19.189 - __ delayed()->deccc(G1_remain); // decrement the count 19.190 + __ delayed()->nop(); 19.191 19.192 __ load_klass(G3_oop, G4_klass); // query the object klass 19.193 19.194 generate_type_check(G4_klass, O3_ckoff, O4_ckval, G5_super, 19.195 // branch to this on success: 19.196 - store_element, 19.197 - // decrement this on success: 19.198 - G1_remain); 19.199 + store_element); 19.200 // ======== end loop ======== 19.201 19.202 // It was a real error; we must depend on the caller to finish the job. 19.203 // Register G1 has number of *remaining* oops, O2 number of *total* oops. 19.204 // Emit GC store barriers for the oops we have copied (O2 minus G1), 19.205 // and report their number to the caller. 19.206 - __ bind(fail); 19.207 + __ BIND(fail); 19.208 __ subcc(O2_count, G1_remain, O2_count); 19.209 __ brx(Assembler::zero, false, Assembler::pt, done); 19.210 __ delayed()->not1(O2_count, O0); // report (-1^K) to caller 19.211 19.212 - __ bind(do_card_marks); 19.213 + __ BIND(do_card_marks); 19.214 gen_write_ref_array_post_barrier(O1_to, O2_count, O3); // store check on O1[0..O2] 19.215 19.216 - __ bind(done); 19.217 + __ BIND(done); 19.218 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr, O3, O4); 19.219 __ retl(); 19.220 __ delayed()->nop(); // return value in 00 19.221 @@ -2942,14 +2863,15 @@ 19.222 StubRoutines::_atomic_add_ptr_entry = StubRoutines::_atomic_add_entry; 19.223 StubRoutines::_fence_entry = generate_fence(); 19.224 #endif // COMPILER2 !=> _LP64 19.225 - 19.226 - StubRoutines::Sparc::_partial_subtype_check = generate_partial_subtype_check(); 19.227 } 19.228 19.229 19.230 void generate_all() { 19.231 // Generates all stubs and initializes the entry points 19.232 19.233 + // Generate partial_subtype_check first here since its code depends on 19.234 + // UseZeroBaseCompressedOops which is defined after heap initialization. 19.235 + StubRoutines::Sparc::_partial_subtype_check = generate_partial_subtype_check(); 19.236 // These entry points require SharedInfo::stack0 to be set up in non-core builds 19.237 StubRoutines::_throw_AbstractMethodError_entry = generate_throw_exception("AbstractMethodError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_AbstractMethodError), false); 19.238 StubRoutines::_throw_IncompatibleClassChangeError_entry= generate_throw_exception("IncompatibleClassChangeError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_IncompatibleClassChangeError), false);
20.1 --- a/src/cpu/sparc/vm/vm_version_sparc.cpp Wed Mar 18 11:37:48 2009 -0400 20.2 +++ b/src/cpu/sparc/vm/vm_version_sparc.cpp Thu Mar 19 09:13:24 2009 -0700 20.3 @@ -1,5 +1,5 @@ 20.4 /* 20.5 - * Copyright 1997-2008 Sun Microsystems, Inc. All Rights Reserved. 20.6 + * Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. 20.7 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 20.8 * 20.9 * This code is free software; you can redistribute it and/or modify it 20.10 @@ -72,6 +72,9 @@ 20.11 FLAG_SET_ERGO(bool, UseCompressedOops, false); 20.12 } 20.13 } 20.14 + // 32-bit oops don't make sense for the 64-bit VM on sparc 20.15 + // since the 32-bit VM has the same registers and smaller objects. 20.16 + Universe::set_narrow_oop_shift(LogMinObjAlignmentInBytes); 20.17 #endif // _LP64 20.18 #ifdef COMPILER2 20.19 // Indirect branch is the same cost as direct 20.20 @@ -89,16 +92,26 @@ 20.21 #endif 20.22 } 20.23 20.24 + // Use hardware population count instruction if available. 20.25 + if (has_hardware_popc()) { 20.26 + if (FLAG_IS_DEFAULT(UsePopCountInstruction)) { 20.27 + UsePopCountInstruction = true; 20.28 + } 20.29 + } 20.30 + 20.31 char buf[512]; 20.32 - jio_snprintf(buf, sizeof(buf), "%s%s%s%s%s%s%s%s%s", 20.33 + jio_snprintf(buf, sizeof(buf), "%s%s%s%s%s%s%s%s%s%s%s%s", 20.34 (has_v8() ? ", has_v8" : ""), 20.35 (has_v9() ? ", has_v9" : ""), 20.36 + (has_hardware_popc() ? ", popc" : ""), 20.37 (has_vis1() ? ", has_vis1" : ""), 20.38 (has_vis2() ? ", has_vis2" : ""), 20.39 (is_ultra3() ? ", is_ultra3" : ""), 20.40 (is_sun4v() ? ", is_sun4v" : ""), 20.41 (is_niagara1() ? ", is_niagara1" : ""), 20.42 - (!has_hardware_int_muldiv() ? ", no-muldiv" : ""), 20.43 + (is_niagara1_plus() ? ", is_niagara1_plus" : ""), 20.44 + (!has_hardware_mul32() ? ", no-mul32" : ""), 20.45 + (!has_hardware_div32() ? ", no-div32" : ""), 20.46 (!has_hardware_fsmuld() ? ", no-fsmuld" : "")); 20.47 20.48 // buf is started with ", " or is empty
21.1 --- a/src/cpu/sparc/vm/vm_version_sparc.hpp Wed Mar 18 11:37:48 2009 -0400 21.2 +++ b/src/cpu/sparc/vm/vm_version_sparc.hpp Thu Mar 19 09:13:24 2009 -0700 21.3 @@ -1,5 +1,5 @@ 21.4 /* 21.5 - * Copyright 1997-2008 Sun Microsystems, Inc. All Rights Reserved. 21.6 + * Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. 21.7 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 21.8 * 21.9 * This code is free software; you can redistribute it and/or modify it 21.10 @@ -25,34 +25,38 @@ 21.11 class VM_Version: public Abstract_VM_Version { 21.12 protected: 21.13 enum Feature_Flag { 21.14 - v8_instructions = 0, 21.15 - hardware_int_muldiv = 1, 21.16 - hardware_fsmuld = 2, 21.17 - v9_instructions = 3, 21.18 - vis1_instructions = 4, 21.19 - vis2_instructions = 5, 21.20 - sun4v_instructions = 6 21.21 + v8_instructions = 0, 21.22 + hardware_mul32 = 1, 21.23 + hardware_div32 = 2, 21.24 + hardware_fsmuld = 3, 21.25 + hardware_popc = 4, 21.26 + v9_instructions = 5, 21.27 + vis1_instructions = 6, 21.28 + vis2_instructions = 7, 21.29 + sun4v_instructions = 8 21.30 }; 21.31 21.32 enum Feature_Flag_Set { 21.33 - unknown_m = 0, 21.34 - all_features_m = -1, 21.35 + unknown_m = 0, 21.36 + all_features_m = -1, 21.37 21.38 - v8_instructions_m = 1 << v8_instructions, 21.39 - hardware_int_muldiv_m = 1 << hardware_int_muldiv, 21.40 - hardware_fsmuld_m = 1 << hardware_fsmuld, 21.41 - v9_instructions_m = 1 << v9_instructions, 21.42 - vis1_instructions_m = 1 << vis1_instructions, 21.43 - vis2_instructions_m = 1 << vis2_instructions, 21.44 - sun4v_m = 1 << sun4v_instructions, 21.45 + v8_instructions_m = 1 << v8_instructions, 21.46 + hardware_mul32_m = 1 << hardware_mul32, 21.47 + hardware_div32_m = 1 << hardware_div32, 21.48 + hardware_fsmuld_m = 1 << hardware_fsmuld, 21.49 + hardware_popc_m = 1 << hardware_popc, 21.50 + v9_instructions_m = 1 << v9_instructions, 21.51 + vis1_instructions_m = 1 << vis1_instructions, 21.52 + vis2_instructions_m = 1 << vis2_instructions, 21.53 + sun4v_m = 1 << sun4v_instructions, 21.54 21.55 - generic_v8_m = v8_instructions_m | hardware_int_muldiv_m | hardware_fsmuld_m, 21.56 - generic_v9_m = generic_v8_m | v9_instructions_m | vis1_instructions_m, 21.57 - ultra3_m = generic_v9_m | vis2_instructions_m, 21.58 + generic_v8_m = v8_instructions_m | hardware_mul32_m | hardware_div32_m | hardware_fsmuld_m, 21.59 + generic_v9_m = generic_v8_m | v9_instructions_m, 21.60 + ultra3_m = generic_v9_m | vis1_instructions_m | vis2_instructions_m, 21.61 21.62 // Temporary until we have something more accurate 21.63 - niagara1_unique_m = sun4v_m, 21.64 - niagara1_m = generic_v9_m | niagara1_unique_m 21.65 + niagara1_unique_m = sun4v_m, 21.66 + niagara1_m = generic_v9_m | niagara1_unique_m 21.67 }; 21.68 21.69 static int _features; 21.70 @@ -62,7 +66,7 @@ 21.71 static int determine_features(); 21.72 static int platform_features(int features); 21.73 21.74 - static bool is_niagara1(int features) { return (features & niagara1_m) == niagara1_m; } 21.75 + static bool is_niagara1(int features) { return (features & sun4v_m) != 0; } 21.76 21.77 static int maximum_niagara1_processor_count() { return 32; } 21.78 // Returns true if the platform is in the niagara line and 21.79 @@ -76,8 +80,10 @@ 21.80 // Instruction support 21.81 static bool has_v8() { return (_features & v8_instructions_m) != 0; } 21.82 static bool has_v9() { return (_features & v9_instructions_m) != 0; } 21.83 - static bool has_hardware_int_muldiv() { return (_features & hardware_int_muldiv_m) != 0; } 21.84 + static bool has_hardware_mul32() { return (_features & hardware_mul32_m) != 0; } 21.85 + static bool has_hardware_div32() { return (_features & hardware_div32_m) != 0; } 21.86 static bool has_hardware_fsmuld() { return (_features & hardware_fsmuld_m) != 0; } 21.87 + static bool has_hardware_popc() { return (_features & hardware_popc_m) != 0; } 21.88 static bool has_vis1() { return (_features & vis1_instructions_m) != 0; } 21.89 static bool has_vis2() { return (_features & vis2_instructions_m) != 0; } 21.90
22.1 --- a/src/cpu/sparc/vm/vtableStubs_sparc.cpp Wed Mar 18 11:37:48 2009 -0400 22.2 +++ b/src/cpu/sparc/vm/vtableStubs_sparc.cpp Thu Mar 19 09:13:24 2009 -0700 22.3 @@ -221,13 +221,15 @@ 22.4 if (is_vtable_stub) { 22.5 // ld;ld;ld,jmp,nop 22.6 const int basic = 5*BytesPerInstWord + 22.7 - // shift;add for load_klass 22.8 - (UseCompressedOops ? 2*BytesPerInstWord : 0); 22.9 + // shift;add for load_klass (only shift with zero heap based) 22.10 + (UseCompressedOops ? 22.11 + ((Universe::narrow_oop_base() == NULL) ? BytesPerInstWord : 2*BytesPerInstWord) : 0); 22.12 return basic + slop; 22.13 } else { 22.14 const int basic = (28 LP64_ONLY(+ 6)) * BytesPerInstWord + 22.15 - // shift;add for load_klass 22.16 - (UseCompressedOops ? 2*BytesPerInstWord : 0); 22.17 + // shift;add for load_klass (only shift with zero heap based) 22.18 + (UseCompressedOops ? 22.19 + ((Universe::narrow_oop_base() == NULL) ? BytesPerInstWord : 2*BytesPerInstWord) : 0); 22.20 return (basic + slop); 22.21 } 22.22 }
23.1 --- a/src/cpu/x86/vm/assembler_x86.cpp Wed Mar 18 11:37:48 2009 -0400 23.2 +++ b/src/cpu/x86/vm/assembler_x86.cpp Thu Mar 19 09:13:24 2009 -0700 23.3 @@ -727,7 +727,7 @@ 23.4 } 23.5 23.6 #ifdef _LP64 23.7 - assert(false, "fix locate_operand"); 23.8 + assert(which == narrow_oop_operand && !is_64bit, "instruction is not a movl adr, imm32"); 23.9 #else 23.10 assert(which == imm_operand, "instruction has only an imm field"); 23.11 #endif // LP64 23.12 @@ -2193,6 +2193,25 @@ 23.13 emit_byte(0x58 | encode); 23.14 } 23.15 23.16 +void Assembler::popcntl(Register dst, Address src) { 23.17 + assert(VM_Version::supports_popcnt(), "must support"); 23.18 + InstructionMark im(this); 23.19 + emit_byte(0xF3); 23.20 + prefix(src, dst); 23.21 + emit_byte(0x0F); 23.22 + emit_byte(0xB8); 23.23 + emit_operand(dst, src); 23.24 +} 23.25 + 23.26 +void Assembler::popcntl(Register dst, Register src) { 23.27 + assert(VM_Version::supports_popcnt(), "must support"); 23.28 + emit_byte(0xF3); 23.29 + int encode = prefix_and_encode(dst->encoding(), src->encoding()); 23.30 + emit_byte(0x0F); 23.31 + emit_byte(0xB8); 23.32 + emit_byte(0xC0 | encode); 23.33 +} 23.34 + 23.35 void Assembler::popf() { 23.36 emit_byte(0x9D); 23.37 } 23.38 @@ -3224,12 +3243,6 @@ 23.39 emit_byte(0xF1); 23.40 } 23.41 23.42 -void Assembler::mov_literal32(Register dst, int32_t imm32, RelocationHolder const& rspec, int format) { 23.43 - InstructionMark im(this); 23.44 - int encode = prefix_and_encode(dst->encoding()); 23.45 - emit_byte(0xB8 | encode); 23.46 - emit_data((int)imm32, rspec, format); 23.47 -} 23.48 23.49 #ifndef _LP64 23.50 23.51 @@ -3249,6 +3262,12 @@ 23.52 emit_data((int)imm32, rspec, 0); 23.53 } 23.54 23.55 +void Assembler::mov_literal32(Register dst, int32_t imm32, RelocationHolder const& rspec) { 23.56 + InstructionMark im(this); 23.57 + int encode = prefix_and_encode(dst->encoding()); 23.58 + emit_byte(0xB8 | encode); 23.59 + emit_data((int)imm32, rspec, 0); 23.60 +} 23.61 23.62 void Assembler::popa() { // 32bit 23.63 emit_byte(0x61); 23.64 @@ -3857,6 +3876,37 @@ 23.65 emit_data64(imm64, rspec); 23.66 } 23.67 23.68 +void Assembler::mov_narrow_oop(Register dst, int32_t imm32, RelocationHolder const& rspec) { 23.69 + InstructionMark im(this); 23.70 + int encode = prefix_and_encode(dst->encoding()); 23.71 + emit_byte(0xB8 | encode); 23.72 + emit_data((int)imm32, rspec, narrow_oop_operand); 23.73 +} 23.74 + 23.75 +void Assembler::mov_narrow_oop(Address dst, int32_t imm32, RelocationHolder const& rspec) { 23.76 + InstructionMark im(this); 23.77 + prefix(dst); 23.78 + emit_byte(0xC7); 23.79 + emit_operand(rax, dst, 4); 23.80 + emit_data((int)imm32, rspec, narrow_oop_operand); 23.81 +} 23.82 + 23.83 +void Assembler::cmp_narrow_oop(Register src1, int32_t imm32, RelocationHolder const& rspec) { 23.84 + InstructionMark im(this); 23.85 + int encode = prefix_and_encode(src1->encoding()); 23.86 + emit_byte(0x81); 23.87 + emit_byte(0xF8 | encode); 23.88 + emit_data((int)imm32, rspec, narrow_oop_operand); 23.89 +} 23.90 + 23.91 +void Assembler::cmp_narrow_oop(Address src1, int32_t imm32, RelocationHolder const& rspec) { 23.92 + InstructionMark im(this); 23.93 + prefix(src1); 23.94 + emit_byte(0x81); 23.95 + emit_operand(rax, src1, 4); 23.96 + emit_data((int)imm32, rspec, narrow_oop_operand); 23.97 +} 23.98 + 23.99 void Assembler::movdq(XMMRegister dst, Register src) { 23.100 // table D-1 says MMX/SSE2 23.101 NOT_LP64(assert(VM_Version::supports_sse2() || VM_Version::supports_mmx(), "")); 23.102 @@ -4049,6 +4099,25 @@ 23.103 addq(rsp, 16 * wordSize); 23.104 } 23.105 23.106 +void Assembler::popcntq(Register dst, Address src) { 23.107 + assert(VM_Version::supports_popcnt(), "must support"); 23.108 + InstructionMark im(this); 23.109 + emit_byte(0xF3); 23.110 + prefixq(src, dst); 23.111 + emit_byte(0x0F); 23.112 + emit_byte(0xB8); 23.113 + emit_operand(dst, src); 23.114 +} 23.115 + 23.116 +void Assembler::popcntq(Register dst, Register src) { 23.117 + assert(VM_Version::supports_popcnt(), "must support"); 23.118 + emit_byte(0xF3); 23.119 + int encode = prefixq_and_encode(dst->encoding(), src->encoding()); 23.120 + emit_byte(0x0F); 23.121 + emit_byte(0xB8); 23.122 + emit_byte(0xC0 | encode); 23.123 +} 23.124 + 23.125 void Assembler::popq(Address dst) { 23.126 InstructionMark im(this); 23.127 prefixq(dst); 23.128 @@ -7217,6 +7286,225 @@ 23.129 } 23.130 23.131 23.132 +void MacroAssembler::check_klass_subtype(Register sub_klass, 23.133 + Register super_klass, 23.134 + Register temp_reg, 23.135 + Label& L_success) { 23.136 + Label L_failure; 23.137 + check_klass_subtype_fast_path(sub_klass, super_klass, temp_reg, &L_success, &L_failure, NULL); 23.138 + check_klass_subtype_slow_path(sub_klass, super_klass, temp_reg, noreg, &L_success, NULL); 23.139 + bind(L_failure); 23.140 +} 23.141 + 23.142 + 23.143 +void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass, 23.144 + Register super_klass, 23.145 + Register temp_reg, 23.146 + Label* L_success, 23.147 + Label* L_failure, 23.148 + Label* L_slow_path, 23.149 + RegisterConstant super_check_offset) { 23.150 + assert_different_registers(sub_klass, super_klass, temp_reg); 23.151 + bool must_load_sco = (super_check_offset.constant_or_zero() == -1); 23.152 + if (super_check_offset.is_register()) { 23.153 + assert_different_registers(sub_klass, super_klass, 23.154 + super_check_offset.as_register()); 23.155 + } else if (must_load_sco) { 23.156 + assert(temp_reg != noreg, "supply either a temp or a register offset"); 23.157 + } 23.158 + 23.159 + Label L_fallthrough; 23.160 + int label_nulls = 0; 23.161 + if (L_success == NULL) { L_success = &L_fallthrough; label_nulls++; } 23.162 + if (L_failure == NULL) { L_failure = &L_fallthrough; label_nulls++; } 23.163 + if (L_slow_path == NULL) { L_slow_path = &L_fallthrough; label_nulls++; } 23.164 + assert(label_nulls <= 1, "at most one NULL in the batch"); 23.165 + 23.166 + int sc_offset = (klassOopDesc::header_size() * HeapWordSize + 23.167 + Klass::secondary_super_cache_offset_in_bytes()); 23.168 + int sco_offset = (klassOopDesc::header_size() * HeapWordSize + 23.169 + Klass::super_check_offset_offset_in_bytes()); 23.170 + Address super_check_offset_addr(super_klass, sco_offset); 23.171 + 23.172 + // Hacked jcc, which "knows" that L_fallthrough, at least, is in 23.173 + // range of a jccb. If this routine grows larger, reconsider at 23.174 + // least some of these. 23.175 +#define local_jcc(assembler_cond, label) \ 23.176 + if (&(label) == &L_fallthrough) jccb(assembler_cond, label); \ 23.177 + else jcc( assembler_cond, label) /*omit semi*/ 23.178 + 23.179 + // Hacked jmp, which may only be used just before L_fallthrough. 23.180 +#define final_jmp(label) \ 23.181 + if (&(label) == &L_fallthrough) { /*do nothing*/ } \ 23.182 + else jmp(label) /*omit semi*/ 23.183 + 23.184 + // If the pointers are equal, we are done (e.g., String[] elements). 23.185 + // This self-check enables sharing of secondary supertype arrays among 23.186 + // non-primary types such as array-of-interface. Otherwise, each such 23.187 + // type would need its own customized SSA. 23.188 + // We move this check to the front of the fast path because many 23.189 + // type checks are in fact trivially successful in this manner, 23.190 + // so we get a nicely predicted branch right at the start of the check. 23.191 + cmpptr(sub_klass, super_klass); 23.192 + local_jcc(Assembler::equal, *L_success); 23.193 + 23.194 + // Check the supertype display: 23.195 + if (must_load_sco) { 23.196 + // Positive movl does right thing on LP64. 23.197 + movl(temp_reg, super_check_offset_addr); 23.198 + super_check_offset = RegisterConstant(temp_reg); 23.199 + } 23.200 + Address super_check_addr(sub_klass, super_check_offset, Address::times_1, 0); 23.201 + cmpptr(super_klass, super_check_addr); // load displayed supertype 23.202 + 23.203 + // This check has worked decisively for primary supers. 23.204 + // Secondary supers are sought in the super_cache ('super_cache_addr'). 23.205 + // (Secondary supers are interfaces and very deeply nested subtypes.) 23.206 + // This works in the same check above because of a tricky aliasing 23.207 + // between the super_cache and the primary super display elements. 23.208 + // (The 'super_check_addr' can address either, as the case requires.) 23.209 + // Note that the cache is updated below if it does not help us find 23.210 + // what we need immediately. 23.211 + // So if it was a primary super, we can just fail immediately. 23.212 + // Otherwise, it's the slow path for us (no success at this point). 23.213 + 23.214 + if (super_check_offset.is_register()) { 23.215 + local_jcc(Assembler::equal, *L_success); 23.216 + cmpl(super_check_offset.as_register(), sc_offset); 23.217 + if (L_failure == &L_fallthrough) { 23.218 + local_jcc(Assembler::equal, *L_slow_path); 23.219 + } else { 23.220 + local_jcc(Assembler::notEqual, *L_failure); 23.221 + final_jmp(*L_slow_path); 23.222 + } 23.223 + } else if (super_check_offset.as_constant() == sc_offset) { 23.224 + // Need a slow path; fast failure is impossible. 23.225 + if (L_slow_path == &L_fallthrough) { 23.226 + local_jcc(Assembler::equal, *L_success); 23.227 + } else { 23.228 + local_jcc(Assembler::notEqual, *L_slow_path); 23.229 + final_jmp(*L_success); 23.230 + } 23.231 + } else { 23.232 + // No slow path; it's a fast decision. 23.233 + if (L_failure == &L_fallthrough) { 23.234 + local_jcc(Assembler::equal, *L_success); 23.235 + } else { 23.236 + local_jcc(Assembler::notEqual, *L_failure); 23.237 + final_jmp(*L_success); 23.238 + } 23.239 + } 23.240 + 23.241 + bind(L_fallthrough); 23.242 + 23.243 +#undef local_jcc 23.244 +#undef final_jmp 23.245 +} 23.246 + 23.247 + 23.248 +void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass, 23.249 + Register super_klass, 23.250 + Register temp_reg, 23.251 + Register temp2_reg, 23.252 + Label* L_success, 23.253 + Label* L_failure, 23.254 + bool set_cond_codes) { 23.255 + assert_different_registers(sub_klass, super_klass, temp_reg); 23.256 + if (temp2_reg != noreg) 23.257 + assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg); 23.258 +#define IS_A_TEMP(reg) ((reg) == temp_reg || (reg) == temp2_reg) 23.259 + 23.260 + Label L_fallthrough; 23.261 + int label_nulls = 0; 23.262 + if (L_success == NULL) { L_success = &L_fallthrough; label_nulls++; } 23.263 + if (L_failure == NULL) { L_failure = &L_fallthrough; label_nulls++; } 23.264 + assert(label_nulls <= 1, "at most one NULL in the batch"); 23.265 + 23.266 + // a couple of useful fields in sub_klass: 23.267 + int ss_offset = (klassOopDesc::header_size() * HeapWordSize + 23.268 + Klass::secondary_supers_offset_in_bytes()); 23.269 + int sc_offset = (klassOopDesc::header_size() * HeapWordSize + 23.270 + Klass::secondary_super_cache_offset_in_bytes()); 23.271 + Address secondary_supers_addr(sub_klass, ss_offset); 23.272 + Address super_cache_addr( sub_klass, sc_offset); 23.273 + 23.274 + // Do a linear scan of the secondary super-klass chain. 23.275 + // This code is rarely used, so simplicity is a virtue here. 23.276 + // The repne_scan instruction uses fixed registers, which we must spill. 23.277 + // Don't worry too much about pre-existing connections with the input regs. 23.278 + 23.279 + assert(sub_klass != rax, "killed reg"); // killed by mov(rax, super) 23.280 + assert(sub_klass != rcx, "killed reg"); // killed by lea(rcx, &pst_counter) 23.281 + 23.282 + // Get super_klass value into rax (even if it was in rdi or rcx). 23.283 + bool pushed_rax = false, pushed_rcx = false, pushed_rdi = false; 23.284 + if (super_klass != rax || UseCompressedOops) { 23.285 + if (!IS_A_TEMP(rax)) { push(rax); pushed_rax = true; } 23.286 + mov(rax, super_klass); 23.287 + } 23.288 + if (!IS_A_TEMP(rcx)) { push(rcx); pushed_rcx = true; } 23.289 + if (!IS_A_TEMP(rdi)) { push(rdi); pushed_rdi = true; } 23.290 + 23.291 +#ifndef PRODUCT 23.292 + int* pst_counter = &SharedRuntime::_partial_subtype_ctr; 23.293 + ExternalAddress pst_counter_addr((address) pst_counter); 23.294 + NOT_LP64( incrementl(pst_counter_addr) ); 23.295 + LP64_ONLY( lea(rcx, pst_counter_addr) ); 23.296 + LP64_ONLY( incrementl(Address(rcx, 0)) ); 23.297 +#endif //PRODUCT 23.298 + 23.299 + // We will consult the secondary-super array. 23.300 + movptr(rdi, secondary_supers_addr); 23.301 + // Load the array length. (Positive movl does right thing on LP64.) 23.302 + movl(rcx, Address(rdi, arrayOopDesc::length_offset_in_bytes())); 23.303 + // Skip to start of data. 23.304 + addptr(rdi, arrayOopDesc::base_offset_in_bytes(T_OBJECT)); 23.305 + 23.306 + // Scan RCX words at [RDI] for an occurrence of RAX. 23.307 + // Set NZ/Z based on last compare. 23.308 +#ifdef _LP64 23.309 + // This part is tricky, as values in supers array could be 32 or 64 bit wide 23.310 + // and we store values in objArrays always encoded, thus we need to encode 23.311 + // the value of rax before repne. Note that rax is dead after the repne. 23.312 + if (UseCompressedOops) { 23.313 + encode_heap_oop_not_null(rax); 23.314 + // The superclass is never null; it would be a basic system error if a null 23.315 + // pointer were to sneak in here. Note that we have already loaded the 23.316 + // Klass::super_check_offset from the super_klass in the fast path, 23.317 + // so if there is a null in that register, we are already in the afterlife. 23.318 + repne_scanl(); 23.319 + } else 23.320 +#endif // _LP64 23.321 + repne_scan(); 23.322 + 23.323 + // Unspill the temp. registers: 23.324 + if (pushed_rdi) pop(rdi); 23.325 + if (pushed_rcx) pop(rcx); 23.326 + if (pushed_rax) pop(rax); 23.327 + 23.328 + if (set_cond_codes) { 23.329 + // Special hack for the AD files: rdi is guaranteed non-zero. 23.330 + assert(!pushed_rdi, "rdi must be left non-NULL"); 23.331 + // Also, the condition codes are properly set Z/NZ on succeed/failure. 23.332 + } 23.333 + 23.334 + if (L_failure == &L_fallthrough) 23.335 + jccb(Assembler::notEqual, *L_failure); 23.336 + else jcc(Assembler::notEqual, *L_failure); 23.337 + 23.338 + // Success. Cache the super we found and proceed in triumph. 23.339 + movptr(super_cache_addr, super_klass); 23.340 + 23.341 + if (L_success != &L_fallthrough) { 23.342 + jmp(*L_success); 23.343 + } 23.344 + 23.345 +#undef IS_A_TEMP 23.346 + 23.347 + bind(L_fallthrough); 23.348 +} 23.349 + 23.350 + 23.351 void MacroAssembler::ucomisd(XMMRegister dst, AddressLiteral src) { 23.352 ucomisd(dst, as_Address(src)); 23.353 } 23.354 @@ -7710,14 +7998,21 @@ 23.355 void MacroAssembler::load_prototype_header(Register dst, Register src) { 23.356 #ifdef _LP64 23.357 if (UseCompressedOops) { 23.358 + assert (Universe::heap() != NULL, "java heap should be initialized"); 23.359 movl(dst, Address(src, oopDesc::klass_offset_in_bytes())); 23.360 - movq(dst, Address(r12_heapbase, dst, Address::times_8, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes())); 23.361 + if (Universe::narrow_oop_shift() != 0) { 23.362 + assert(Address::times_8 == LogMinObjAlignmentInBytes && 23.363 + Address::times_8 == Universe::narrow_oop_shift(), "decode alg wrong"); 23.364 + movq(dst, Address(r12_heapbase, dst, Address::times_8, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes())); 23.365 + } else { 23.366 + movq(dst, Address(dst, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes())); 23.367 + } 23.368 } else 23.369 #endif 23.370 - { 23.371 - movptr(dst, Address(src, oopDesc::klass_offset_in_bytes())); 23.372 - movptr(dst, Address(dst, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes())); 23.373 - } 23.374 + { 23.375 + movptr(dst, Address(src, oopDesc::klass_offset_in_bytes())); 23.376 + movptr(dst, Address(dst, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes())); 23.377 + } 23.378 } 23.379 23.380 void MacroAssembler::store_klass(Register dst, Register src) { 23.381 @@ -7760,11 +8055,20 @@ 23.382 // Algorithm must match oop.inline.hpp encode_heap_oop. 23.383 void MacroAssembler::encode_heap_oop(Register r) { 23.384 assert (UseCompressedOops, "should be compressed"); 23.385 + assert (Universe::heap() != NULL, "java heap should be initialized"); 23.386 + if (Universe::narrow_oop_base() == NULL) { 23.387 + verify_oop(r, "broken oop in encode_heap_oop"); 23.388 + if (Universe::narrow_oop_shift() != 0) { 23.389 + assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong"); 23.390 + shrq(r, LogMinObjAlignmentInBytes); 23.391 + } 23.392 + return; 23.393 + } 23.394 #ifdef ASSERT 23.395 if (CheckCompressedOops) { 23.396 Label ok; 23.397 push(rscratch1); // cmpptr trashes rscratch1 23.398 - cmpptr(r12_heapbase, ExternalAddress((address)Universe::heap_base_addr())); 23.399 + cmpptr(r12_heapbase, ExternalAddress((address)Universe::narrow_oop_base_addr())); 23.400 jcc(Assembler::equal, ok); 23.401 stop("MacroAssembler::encode_heap_oop: heap base corrupted?"); 23.402 bind(ok); 23.403 @@ -7780,6 +8084,7 @@ 23.404 23.405 void MacroAssembler::encode_heap_oop_not_null(Register r) { 23.406 assert (UseCompressedOops, "should be compressed"); 23.407 + assert (Universe::heap() != NULL, "java heap should be initialized"); 23.408 #ifdef ASSERT 23.409 if (CheckCompressedOops) { 23.410 Label ok; 23.411 @@ -7790,12 +8095,18 @@ 23.412 } 23.413 #endif 23.414 verify_oop(r, "broken oop in encode_heap_oop_not_null"); 23.415 - subq(r, r12_heapbase); 23.416 - shrq(r, LogMinObjAlignmentInBytes); 23.417 + if (Universe::narrow_oop_base() != NULL) { 23.418 + subq(r, r12_heapbase); 23.419 + } 23.420 + if (Universe::narrow_oop_shift() != 0) { 23.421 + assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong"); 23.422 + shrq(r, LogMinObjAlignmentInBytes); 23.423 + } 23.424 } 23.425 23.426 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) { 23.427 assert (UseCompressedOops, "should be compressed"); 23.428 + assert (Universe::heap() != NULL, "java heap should be initialized"); 23.429 #ifdef ASSERT 23.430 if (CheckCompressedOops) { 23.431 Label ok; 23.432 @@ -7809,18 +8120,32 @@ 23.433 if (dst != src) { 23.434 movq(dst, src); 23.435 } 23.436 - subq(dst, r12_heapbase); 23.437 - shrq(dst, LogMinObjAlignmentInBytes); 23.438 + if (Universe::narrow_oop_base() != NULL) { 23.439 + subq(dst, r12_heapbase); 23.440 + } 23.441 + if (Universe::narrow_oop_shift() != 0) { 23.442 + assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong"); 23.443 + shrq(dst, LogMinObjAlignmentInBytes); 23.444 + } 23.445 } 23.446 23.447 void MacroAssembler::decode_heap_oop(Register r) { 23.448 assert (UseCompressedOops, "should be compressed"); 23.449 + assert (Universe::heap() != NULL, "java heap should be initialized"); 23.450 + if (Universe::narrow_oop_base() == NULL) { 23.451 + if (Universe::narrow_oop_shift() != 0) { 23.452 + assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong"); 23.453 + shlq(r, LogMinObjAlignmentInBytes); 23.454 + } 23.455 + verify_oop(r, "broken oop in decode_heap_oop"); 23.456 + return; 23.457 + } 23.458 #ifdef ASSERT 23.459 if (CheckCompressedOops) { 23.460 Label ok; 23.461 push(rscratch1); 23.462 cmpptr(r12_heapbase, 23.463 - ExternalAddress((address)Universe::heap_base_addr())); 23.464 + ExternalAddress((address)Universe::narrow_oop_base_addr())); 23.465 jcc(Assembler::equal, ok); 23.466 stop("MacroAssembler::decode_heap_oop: heap base corrupted?"); 23.467 bind(ok); 23.468 @@ -7844,32 +8169,76 @@ 23.469 23.470 void MacroAssembler::decode_heap_oop_not_null(Register r) { 23.471 assert (UseCompressedOops, "should only be used for compressed headers"); 23.472 + assert (Universe::heap() != NULL, "java heap should be initialized"); 23.473 // Cannot assert, unverified entry point counts instructions (see .ad file) 23.474 // vtableStubs also counts instructions in pd_code_size_limit. 23.475 // Also do not verify_oop as this is called by verify_oop. 23.476 - assert(Address::times_8 == LogMinObjAlignmentInBytes, "decode alg wrong"); 23.477 - leaq(r, Address(r12_heapbase, r, Address::times_8, 0)); 23.478 + if (Universe::narrow_oop_base() == NULL) { 23.479 + if (Universe::narrow_oop_shift() != 0) { 23.480 + assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong"); 23.481 + shlq(r, LogMinObjAlignmentInBytes); 23.482 + } 23.483 + } else { 23.484 + assert (Address::times_8 == LogMinObjAlignmentInBytes && 23.485 + Address::times_8 == Universe::narrow_oop_shift(), "decode alg wrong"); 23.486 + leaq(r, Address(r12_heapbase, r, Address::times_8, 0)); 23.487 + } 23.488 } 23.489 23.490 void MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) { 23.491 assert (UseCompressedOops, "should only be used for compressed headers"); 23.492 + assert (Universe::heap() != NULL, "java heap should be initialized"); 23.493 // Cannot assert, unverified entry point counts instructions (see .ad file) 23.494 // vtableStubs also counts instructions in pd_code_size_limit. 23.495 // Also do not verify_oop as this is called by verify_oop. 23.496 - assert(Address::times_8 == LogMinObjAlignmentInBytes, "decode alg wrong"); 23.497 - leaq(dst, Address(r12_heapbase, src, Address::times_8, 0)); 23.498 + if (Universe::narrow_oop_shift() != 0) { 23.499 + assert (Address::times_8 == LogMinObjAlignmentInBytes && 23.500 + Address::times_8 == Universe::narrow_oop_shift(), "decode alg wrong"); 23.501 + leaq(dst, Address(r12_heapbase, src, Address::times_8, 0)); 23.502 + } else if (dst != src) { 23.503 + movq(dst, src); 23.504 + } 23.505 } 23.506 23.507 void MacroAssembler::set_narrow_oop(Register dst, jobject obj) { 23.508 - assert(oop_recorder() != NULL, "this assembler needs an OopRecorder"); 23.509 + assert (UseCompressedOops, "should only be used for compressed headers"); 23.510 + assert (Universe::heap() != NULL, "java heap should be initialized"); 23.511 + assert (oop_recorder() != NULL, "this assembler needs an OopRecorder"); 23.512 int oop_index = oop_recorder()->find_index(obj); 23.513 RelocationHolder rspec = oop_Relocation::spec(oop_index); 23.514 - mov_literal32(dst, oop_index, rspec, narrow_oop_operand); 23.515 + mov_narrow_oop(dst, oop_index, rspec); 23.516 +} 23.517 + 23.518 +void MacroAssembler::set_narrow_oop(Address dst, jobject obj) { 23.519 + assert (UseCompressedOops, "should only be used for compressed headers"); 23.520 + assert (Universe::heap() != NULL, "java heap should be initialized"); 23.521 + assert (oop_recorder() != NULL, "this assembler needs an OopRecorder"); 23.522 + int oop_index = oop_recorder()->find_index(obj); 23.523 + RelocationHolder rspec = oop_Relocation::spec(oop_index); 23.524 + mov_narrow_oop(dst, oop_index, rspec); 23.525 +} 23.526 + 23.527 +void MacroAssembler::cmp_narrow_oop(Register dst, jobject obj) { 23.528 + assert (UseCompressedOops, "should only be used for compressed headers"); 23.529 + assert (Universe::heap() != NULL, "java heap should be initialized"); 23.530 + assert (oop_recorder() != NULL, "this assembler needs an OopRecorder"); 23.531 + int oop_index = oop_recorder()->find_index(obj); 23.532 + RelocationHolder rspec = oop_Relocation::spec(oop_index); 23.533 + Assembler::cmp_narrow_oop(dst, oop_index, rspec); 23.534 +} 23.535 + 23.536 +void MacroAssembler::cmp_narrow_oop(Address dst, jobject obj) { 23.537 + assert (UseCompressedOops, "should only be used for compressed headers"); 23.538 + assert (Universe::heap() != NULL, "java heap should be initialized"); 23.539 + assert (oop_recorder() != NULL, "this assembler needs an OopRecorder"); 23.540 + int oop_index = oop_recorder()->find_index(obj); 23.541 + RelocationHolder rspec = oop_Relocation::spec(oop_index); 23.542 + Assembler::cmp_narrow_oop(dst, oop_index, rspec); 23.543 } 23.544 23.545 void MacroAssembler::reinit_heapbase() { 23.546 if (UseCompressedOops) { 23.547 - movptr(r12_heapbase, ExternalAddress((address)Universe::heap_base_addr())); 23.548 + movptr(r12_heapbase, ExternalAddress((address)Universe::narrow_oop_base_addr())); 23.549 } 23.550 } 23.551 #endif // _LP64
24.1 --- a/src/cpu/x86/vm/assembler_x86.hpp Wed Mar 18 11:37:48 2009 -0400 24.2 +++ b/src/cpu/x86/vm/assembler_x86.hpp Thu Mar 19 09:13:24 2009 -0700 24.3 @@ -578,20 +578,25 @@ 24.4 24.5 // These are all easily abused and hence protected 24.6 24.7 - void mov_literal32(Register dst, int32_t imm32, RelocationHolder const& rspec, int format = 0); 24.8 - 24.9 // 32BIT ONLY SECTION 24.10 #ifndef _LP64 24.11 // Make these disappear in 64bit mode since they would never be correct 24.12 void cmp_literal32(Register src1, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY 24.13 void cmp_literal32(Address src1, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY 24.14 24.15 + void mov_literal32(Register dst, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY 24.16 void mov_literal32(Address dst, int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY 24.17 24.18 void push_literal32(int32_t imm32, RelocationHolder const& rspec); // 32BIT ONLY 24.19 #else 24.20 // 64BIT ONLY SECTION 24.21 void mov_literal64(Register dst, intptr_t imm64, RelocationHolder const& rspec); // 64BIT ONLY 24.22 + 24.23 + void cmp_narrow_oop(Register src1, int32_t imm32, RelocationHolder const& rspec); 24.24 + void cmp_narrow_oop(Address src1, int32_t imm32, RelocationHolder const& rspec); 24.25 + 24.26 + void mov_narrow_oop(Register dst, int32_t imm32, RelocationHolder const& rspec); 24.27 + void mov_narrow_oop(Address dst, int32_t imm32, RelocationHolder const& rspec); 24.28 #endif // _LP64 24.29 24.30 // These are unique in that we are ensured by the caller that the 32bit 24.31 @@ -1219,6 +1224,14 @@ 24.32 void popq(Address dst); 24.33 #endif 24.34 24.35 + void popcntl(Register dst, Address src); 24.36 + void popcntl(Register dst, Register src); 24.37 + 24.38 +#ifdef _LP64 24.39 + void popcntq(Register dst, Address src); 24.40 + void popcntq(Register dst, Register src); 24.41 +#endif 24.42 + 24.43 // Prefetches (SSE, SSE2, 3DNOW only) 24.44 24.45 void prefetchnta(Address src); 24.46 @@ -1647,6 +1660,9 @@ 24.47 void decode_heap_oop_not_null(Register dst, Register src); 24.48 24.49 void set_narrow_oop(Register dst, jobject obj); 24.50 + void set_narrow_oop(Address dst, jobject obj); 24.51 + void cmp_narrow_oop(Register dst, jobject obj); 24.52 + void cmp_narrow_oop(Address dst, jobject obj); 24.53 24.54 // if heap base register is used - reinit it with the correct value 24.55 void reinit_heapbase(); 24.56 @@ -1791,6 +1807,40 @@ 24.57 Register scan_temp, 24.58 Label& no_such_interface); 24.59 24.60 + // Test sub_klass against super_klass, with fast and slow paths. 24.61 + 24.62 + // The fast path produces a tri-state answer: yes / no / maybe-slow. 24.63 + // One of the three labels can be NULL, meaning take the fall-through. 24.64 + // If super_check_offset is -1, the value is loaded up from super_klass. 24.65 + // No registers are killed, except temp_reg. 24.66 + void check_klass_subtype_fast_path(Register sub_klass, 24.67 + Register super_klass, 24.68 + Register temp_reg, 24.69 + Label* L_success, 24.70 + Label* L_failure, 24.71 + Label* L_slow_path, 24.72 + RegisterConstant super_check_offset = RegisterConstant(-1)); 24.73 + 24.74 + // The rest of the type check; must be wired to a corresponding fast path. 24.75 + // It does not repeat the fast path logic, so don't use it standalone. 24.76 + // The temp_reg and temp2_reg can be noreg, if no temps are available. 24.77 + // Updates the sub's secondary super cache as necessary. 24.78 + // If set_cond_codes, condition codes will be Z on success, NZ on failure. 24.79 + void check_klass_subtype_slow_path(Register sub_klass, 24.80 + Register super_klass, 24.81 + Register temp_reg, 24.82 + Register temp2_reg, 24.83 + Label* L_success, 24.84 + Label* L_failure, 24.85 + bool set_cond_codes = false); 24.86 + 24.87 + // Simplified, combined version, good for typical uses. 24.88 + // Falls through on failure. 24.89 + void check_klass_subtype(Register sub_klass, 24.90 + Register super_klass, 24.91 + Register temp_reg, 24.92 + Label& L_success); 24.93 + 24.94 //---- 24.95 void set_word_if_not_zero(Register reg); // sets reg to 1 if not zero, otherwise 0 24.96
25.1 --- a/src/cpu/x86/vm/c1_LIRAssembler_x86.cpp Wed Mar 18 11:37:48 2009 -0400 25.2 +++ b/src/cpu/x86/vm/c1_LIRAssembler_x86.cpp Thu Mar 19 09:13:24 2009 -0700 25.3 @@ -1598,18 +1598,9 @@ 25.4 25.5 // get instance klass 25.6 __ movptr(k_RInfo, Address(k_RInfo, objArrayKlass::element_klass_offset_in_bytes() + sizeof(oopDesc))); 25.7 - // get super_check_offset 25.8 - __ movl(Rtmp1, Address(k_RInfo, sizeof(oopDesc) + Klass::super_check_offset_offset_in_bytes())); 25.9 - // See if we get an immediate positive hit 25.10 - __ cmpptr(k_RInfo, Address(klass_RInfo, Rtmp1, Address::times_1)); 25.11 - __ jcc(Assembler::equal, done); 25.12 - // check for immediate negative hit 25.13 - __ cmpl(Rtmp1, sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes()); 25.14 - __ jcc(Assembler::notEqual, *stub->entry()); 25.15 - // check for self 25.16 - __ cmpptr(klass_RInfo, k_RInfo); 25.17 - __ jcc(Assembler::equal, done); 25.18 - 25.19 + // perform the fast part of the checking logic 25.20 + __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, &done, stub->entry(), NULL); 25.21 + // call out-of-line instance of __ check_klass_subtype_slow_path(...): 25.22 __ push(klass_RInfo); 25.23 __ push(k_RInfo); 25.24 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id))); 25.25 @@ -1735,17 +1726,9 @@ 25.26 } 25.27 __ bind(done); 25.28 } else { 25.29 - __ movl(Rtmp1, Address(k_RInfo, sizeof(oopDesc) + Klass::super_check_offset_offset_in_bytes())); 25.30 - // See if we get an immediate positive hit 25.31 - __ cmpptr(k_RInfo, Address(klass_RInfo, Rtmp1, Address::times_1)); 25.32 - __ jcc(Assembler::equal, done); 25.33 - // check for immediate negative hit 25.34 - __ cmpl(Rtmp1, sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes()); 25.35 - __ jcc(Assembler::notEqual, *stub->entry()); 25.36 - // check for self 25.37 - __ cmpptr(klass_RInfo, k_RInfo); 25.38 - __ jcc(Assembler::equal, done); 25.39 - 25.40 + // perform the fast part of the checking logic 25.41 + __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, &done, stub->entry(), NULL); 25.42 + // call out-of-line instance of __ check_klass_subtype_slow_path(...): 25.43 __ push(klass_RInfo); 25.44 __ push(k_RInfo); 25.45 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id))); 25.46 @@ -1821,23 +1804,15 @@ 25.47 __ pop(dst); 25.48 __ jmp(done); 25.49 } 25.50 - } else { 25.51 -#else 25.52 - { // YUCK 25.53 + } 25.54 + else // next block is unconditional if LP64: 25.55 #endif // LP64 25.56 + { 25.57 assert(dst != klass_RInfo && dst != k_RInfo, "need 3 registers"); 25.58 25.59 - __ movl(dst, Address(k_RInfo, sizeof(oopDesc) + Klass::super_check_offset_offset_in_bytes())); 25.60 - // See if we get an immediate positive hit 25.61 - __ cmpptr(k_RInfo, Address(klass_RInfo, dst, Address::times_1)); 25.62 - __ jcc(Assembler::equal, one); 25.63 - // check for immediate negative hit 25.64 - __ cmpl(dst, sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes()); 25.65 - __ jcc(Assembler::notEqual, zero); 25.66 - // check for self 25.67 - __ cmpptr(klass_RInfo, k_RInfo); 25.68 - __ jcc(Assembler::equal, one); 25.69 - 25.70 + // perform the fast part of the checking logic 25.71 + __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, dst, &one, &zero, NULL); 25.72 + // call out-of-line instance of __ check_klass_subtype_slow_path(...): 25.73 __ push(klass_RInfo); 25.74 __ push(k_RInfo); 25.75 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
26.1 --- a/src/cpu/x86/vm/c1_Runtime1_x86.cpp Wed Mar 18 11:37:48 2009 -0400 26.2 +++ b/src/cpu/x86/vm/c1_Runtime1_x86.cpp Thu Mar 19 09:13:24 2009 -0700 26.3 @@ -1354,6 +1354,13 @@ 26.4 26.5 case slow_subtype_check_id: 26.6 { 26.7 + // Typical calling sequence: 26.8 + // __ push(klass_RInfo); // object klass or other subclass 26.9 + // __ push(sup_k_RInfo); // array element klass or other superclass 26.10 + // __ call(slow_subtype_check); 26.11 + // Note that the subclass is pushed first, and is therefore deepest. 26.12 + // Previous versions of this code reversed the names 'sub' and 'super'. 26.13 + // This was operationally harmless but made the code unreadable. 26.14 enum layout { 26.15 rax_off, SLOT2(raxH_off) 26.16 rcx_off, SLOT2(rcxH_off) 26.17 @@ -1361,9 +1368,10 @@ 26.18 rdi_off, SLOT2(rdiH_off) 26.19 // saved_rbp_off, SLOT2(saved_rbpH_off) 26.20 return_off, SLOT2(returnH_off) 26.21 - sub_off, SLOT2(subH_off) 26.22 - super_off, SLOT2(superH_off) 26.23 - framesize 26.24 + sup_k_off, SLOT2(sup_kH_off) 26.25 + klass_off, SLOT2(superH_off) 26.26 + framesize, 26.27 + result_off = klass_off // deepest argument is also the return value 26.28 }; 26.29 26.30 __ set_info("slow_subtype_check", dont_gc_arguments); 26.31 @@ -1373,19 +1381,14 @@ 26.32 __ push(rax); 26.33 26.34 // This is called by pushing args and not with C abi 26.35 - __ movptr(rsi, Address(rsp, (super_off) * VMRegImpl::stack_slot_size)); // super 26.36 - __ movptr(rax, Address(rsp, (sub_off ) * VMRegImpl::stack_slot_size)); // sub 26.37 - 26.38 - __ movptr(rdi,Address(rsi,sizeof(oopDesc) + Klass::secondary_supers_offset_in_bytes())); 26.39 - // since size is postive movl does right thing on 64bit 26.40 - __ movl(rcx, Address(rdi, arrayOopDesc::length_offset_in_bytes())); 26.41 - __ addptr(rdi, arrayOopDesc::base_offset_in_bytes(T_OBJECT)); 26.42 + __ movptr(rsi, Address(rsp, (klass_off) * VMRegImpl::stack_slot_size)); // subclass 26.43 + __ movptr(rax, Address(rsp, (sup_k_off) * VMRegImpl::stack_slot_size)); // superclass 26.44 26.45 Label miss; 26.46 - __ repne_scan(); 26.47 - __ jcc(Assembler::notEqual, miss); 26.48 - __ movptr(Address(rsi,sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes()), rax); 26.49 - __ movptr(Address(rsp, (super_off) * VMRegImpl::stack_slot_size), 1); // result 26.50 + __ check_klass_subtype_slow_path(rsi, rax, rcx, rdi, NULL, &miss); 26.51 + 26.52 + // fallthrough on success: 26.53 + __ movptr(Address(rsp, (result_off) * VMRegImpl::stack_slot_size), 1); // result 26.54 __ pop(rax); 26.55 __ pop(rcx); 26.56 __ pop(rsi); 26.57 @@ -1393,7 +1396,7 @@ 26.58 __ ret(0); 26.59 26.60 __ bind(miss); 26.61 - __ movptr(Address(rsp, (super_off) * VMRegImpl::stack_slot_size), NULL_WORD); // result 26.62 + __ movptr(Address(rsp, (result_off) * VMRegImpl::stack_slot_size), NULL_WORD); // result 26.63 __ pop(rax); 26.64 __ pop(rcx); 26.65 __ pop(rsi);
27.1 --- a/src/cpu/x86/vm/interp_masm_x86_32.cpp Wed Mar 18 11:37:48 2009 -0400 27.2 +++ b/src/cpu/x86/vm/interp_masm_x86_32.cpp Thu Mar 19 09:13:24 2009 -0700 27.3 @@ -219,47 +219,16 @@ 27.4 // Resets EDI to locals. Register sub_klass cannot be any of the above. 27.5 void InterpreterMacroAssembler::gen_subtype_check( Register Rsub_klass, Label &ok_is_subtype ) { 27.6 assert( Rsub_klass != rax, "rax, holds superklass" ); 27.7 - assert( Rsub_klass != rcx, "rcx holds 2ndary super array length" ); 27.8 - assert( Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr" ); 27.9 - Label not_subtype, loop; 27.10 + assert( Rsub_klass != rcx, "used as a temp" ); 27.11 + assert( Rsub_klass != rdi, "used as a temp, restored from locals" ); 27.12 27.13 // Profile the not-null value's klass. 27.14 - profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, rdi 27.15 + profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi 27.16 27.17 - // Load the super-klass's check offset into ECX 27.18 - movl( rcx, Address(rax, sizeof(oopDesc) + Klass::super_check_offset_offset_in_bytes() ) ); 27.19 - // Load from the sub-klass's super-class display list, or a 1-word cache of 27.20 - // the secondary superclass list, or a failing value with a sentinel offset 27.21 - // if the super-klass is an interface or exceptionally deep in the Java 27.22 - // hierarchy and we have to scan the secondary superclass list the hard way. 27.23 - // See if we get an immediate positive hit 27.24 - cmpptr( rax, Address(Rsub_klass,rcx,Address::times_1) ); 27.25 - jcc( Assembler::equal,ok_is_subtype ); 27.26 + // Do the check. 27.27 + check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx 27.28 27.29 - // Check for immediate negative hit 27.30 - cmpl( rcx, sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes() ); 27.31 - jcc( Assembler::notEqual, not_subtype ); 27.32 - // Check for self 27.33 - cmpptr( Rsub_klass, rax ); 27.34 - jcc( Assembler::equal, ok_is_subtype ); 27.35 - 27.36 - // Now do a linear scan of the secondary super-klass chain. 27.37 - movptr( rdi, Address(Rsub_klass, sizeof(oopDesc) + Klass::secondary_supers_offset_in_bytes()) ); 27.38 - // EDI holds the objArrayOop of secondary supers. 27.39 - movl( rcx, Address(rdi, arrayOopDesc::length_offset_in_bytes()));// Load the array length 27.40 - // Skip to start of data; also clear Z flag incase ECX is zero 27.41 - addptr( rdi, arrayOopDesc::base_offset_in_bytes(T_OBJECT) ); 27.42 - // Scan ECX words at [EDI] for occurance of EAX 27.43 - // Set NZ/Z based on last compare 27.44 - repne_scan(); 27.45 - restore_locals(); // Restore EDI; Must not blow flags 27.46 - // Not equal? 27.47 - jcc( Assembler::notEqual, not_subtype ); 27.48 - // Must be equal but missed in cache. Update cache. 27.49 - movptr( Address(Rsub_klass, sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes()), rax ); 27.50 - jmp( ok_is_subtype ); 27.51 - 27.52 - bind(not_subtype); 27.53 + // Profile the failure of the check. 27.54 profile_typecheck_failed(rcx); // blows rcx 27.55 } 27.56
28.1 --- a/src/cpu/x86/vm/interp_masm_x86_64.cpp Wed Mar 18 11:37:48 2009 -0400 28.2 +++ b/src/cpu/x86/vm/interp_masm_x86_64.cpp Thu Mar 19 09:13:24 2009 -0700 28.3 @@ -232,65 +232,13 @@ 28.4 assert(Rsub_klass != rcx, "rcx holds 2ndary super array length"); 28.5 assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr"); 28.6 28.7 - Label not_subtype, not_subtype_pop, loop; 28.8 + // Profile the not-null value's klass. 28.9 + profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi 28.10 28.11 - // Profile the not-null value's klass. 28.12 - profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, rdi 28.13 + // Do the check. 28.14 + check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx 28.15 28.16 - // Load the super-klass's check offset into rcx 28.17 - movl(rcx, Address(rax, sizeof(oopDesc) + 28.18 - Klass::super_check_offset_offset_in_bytes())); 28.19 - // Load from the sub-klass's super-class display list, or a 1-word 28.20 - // cache of the secondary superclass list, or a failing value with a 28.21 - // sentinel offset if the super-klass is an interface or 28.22 - // exceptionally deep in the Java hierarchy and we have to scan the 28.23 - // secondary superclass list the hard way. See if we get an 28.24 - // immediate positive hit 28.25 - cmpptr(rax, Address(Rsub_klass, rcx, Address::times_1)); 28.26 - jcc(Assembler::equal,ok_is_subtype); 28.27 - 28.28 - // Check for immediate negative hit 28.29 - cmpl(rcx, sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes()); 28.30 - jcc( Assembler::notEqual, not_subtype ); 28.31 - // Check for self 28.32 - cmpptr(Rsub_klass, rax); 28.33 - jcc(Assembler::equal, ok_is_subtype); 28.34 - 28.35 - // Now do a linear scan of the secondary super-klass chain. 28.36 - movptr(rdi, Address(Rsub_klass, sizeof(oopDesc) + 28.37 - Klass::secondary_supers_offset_in_bytes())); 28.38 - // rdi holds the objArrayOop of secondary supers. 28.39 - // Load the array length 28.40 - movl(rcx, Address(rdi, arrayOopDesc::length_offset_in_bytes())); 28.41 - // Skip to start of data; also clear Z flag incase rcx is zero 28.42 - addptr(rdi, arrayOopDesc::base_offset_in_bytes(T_OBJECT)); 28.43 - // Scan rcx words at [rdi] for occurance of rax 28.44 - // Set NZ/Z based on last compare 28.45 - 28.46 - // this part is kind tricky, as values in supers array could be 32 or 64 bit wide 28.47 - // and we store values in objArrays always encoded, thus we need to encode value 28.48 - // before repne 28.49 - if (UseCompressedOops) { 28.50 - push(rax); 28.51 - encode_heap_oop(rax); 28.52 - repne_scanl(); 28.53 - // Not equal? 28.54 - jcc(Assembler::notEqual, not_subtype_pop); 28.55 - // restore heap oop here for movq 28.56 - pop(rax); 28.57 - } else { 28.58 - repne_scan(); 28.59 - jcc(Assembler::notEqual, not_subtype); 28.60 - } 28.61 - // Must be equal but missed in cache. Update cache. 28.62 - movptr(Address(Rsub_klass, sizeof(oopDesc) + 28.63 - Klass::secondary_super_cache_offset_in_bytes()), rax); 28.64 - jmp(ok_is_subtype); 28.65 - 28.66 - bind(not_subtype_pop); 28.67 - // restore heap oop here for miss 28.68 - if (UseCompressedOops) pop(rax); 28.69 - bind(not_subtype); 28.70 + // Profile the failure of the check. 28.71 profile_typecheck_failed(rcx); // blows rcx 28.72 } 28.73
29.1 --- a/src/cpu/x86/vm/interpreterRT_x86_64.cpp Wed Mar 18 11:37:48 2009 -0400 29.2 +++ b/src/cpu/x86/vm/interpreterRT_x86_64.cpp Thu Mar 19 09:13:24 2009 -0700 29.3 @@ -349,7 +349,7 @@ 29.4 29.5 if (_num_args < Argument::n_float_register_parameters_c-1) { 29.6 *_reg_args++ = from_obj; 29.7 - *_fp_identifiers |= (0x01 << (_num_args*2)); // mark as float 29.8 + *_fp_identifiers |= (intptr_t)(0x01 << (_num_args*2)); // mark as float 29.9 _num_args++; 29.10 } else { 29.11 *_to++ = from_obj; 29.12 @@ -364,7 +364,7 @@ 29.13 29.14 if (_num_args < Argument::n_float_register_parameters_c-1) { 29.15 *_reg_args++ = from_obj; 29.16 - *_fp_identifiers |= (0x3 << (_num_args*2)); // mark as double 29.17 + *_fp_identifiers |= (intptr_t)(0x3 << (_num_args*2)); // mark as double 29.18 _num_args++; 29.19 } else { 29.20 *_to++ = from_obj;
30.1 --- a/src/cpu/x86/vm/stubGenerator_x86_32.cpp Wed Mar 18 11:37:48 2009 -0400 30.2 +++ b/src/cpu/x86/vm/stubGenerator_x86_32.cpp Thu Mar 19 09:13:24 2009 -0700 30.3 @@ -1310,81 +1310,51 @@ 30.4 Address& super_check_offset_addr, 30.5 Address& super_klass_addr, 30.6 Register temp, 30.7 - Label* L_success_ptr, Label* L_failure_ptr) { 30.8 + Label* L_success, Label* L_failure) { 30.9 BLOCK_COMMENT("type_check:"); 30.10 30.11 Label L_fallthrough; 30.12 - bool fall_through_on_success = (L_success_ptr == NULL); 30.13 - if (fall_through_on_success) { 30.14 - L_success_ptr = &L_fallthrough; 30.15 - } else { 30.16 - L_failure_ptr = &L_fallthrough; 30.17 - } 30.18 - Label& L_success = *L_success_ptr; 30.19 - Label& L_failure = *L_failure_ptr; 30.20 +#define LOCAL_JCC(assembler_con, label_ptr) \ 30.21 + if (label_ptr != NULL) __ jcc(assembler_con, *(label_ptr)); \ 30.22 + else __ jcc(assembler_con, L_fallthrough) /*omit semi*/ 30.23 30.24 + // The following is a strange variation of the fast path which requires 30.25 + // one less register, because needed values are on the argument stack. 30.26 + // __ check_klass_subtype_fast_path(sub_klass, *super_klass*, temp, 30.27 + // L_success, L_failure, NULL); 30.28 assert_different_registers(sub_klass, temp); 30.29 30.30 - // a couple of useful fields in sub_klass: 30.31 - int ss_offset = (klassOopDesc::header_size() * HeapWordSize + 30.32 - Klass::secondary_supers_offset_in_bytes()); 30.33 int sc_offset = (klassOopDesc::header_size() * HeapWordSize + 30.34 Klass::secondary_super_cache_offset_in_bytes()); 30.35 - Address secondary_supers_addr(sub_klass, ss_offset); 30.36 - Address super_cache_addr( sub_klass, sc_offset); 30.37 30.38 // if the pointers are equal, we are done (e.g., String[] elements) 30.39 __ cmpptr(sub_klass, super_klass_addr); 30.40 - __ jcc(Assembler::equal, L_success); 30.41 + LOCAL_JCC(Assembler::equal, L_success); 30.42 30.43 // check the supertype display: 30.44 __ movl2ptr(temp, super_check_offset_addr); 30.45 Address super_check_addr(sub_klass, temp, Address::times_1, 0); 30.46 __ movptr(temp, super_check_addr); // load displayed supertype 30.47 __ cmpptr(temp, super_klass_addr); // test the super type 30.48 - __ jcc(Assembler::equal, L_success); 30.49 + LOCAL_JCC(Assembler::equal, L_success); 30.50 30.51 // if it was a primary super, we can just fail immediately 30.52 __ cmpl(super_check_offset_addr, sc_offset); 30.53 - __ jcc(Assembler::notEqual, L_failure); 30.54 + LOCAL_JCC(Assembler::notEqual, L_failure); 30.55 30.56 - // Now do a linear scan of the secondary super-klass chain. 30.57 - // This code is rarely used, so simplicity is a virtue here. 30.58 - inc_counter_np(SharedRuntime::_partial_subtype_ctr); 30.59 - { 30.60 - // The repne_scan instruction uses fixed registers, which we must spill. 30.61 - // (We need a couple more temps in any case.) 30.62 - __ push(rax); 30.63 - __ push(rcx); 30.64 - __ push(rdi); 30.65 - assert_different_registers(sub_klass, rax, rcx, rdi); 30.66 + // The repne_scan instruction uses fixed registers, which will get spilled. 30.67 + // We happen to know this works best when super_klass is in rax. 30.68 + Register super_klass = temp; 30.69 + __ movptr(super_klass, super_klass_addr); 30.70 + __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg, 30.71 + L_success, L_failure); 30.72 30.73 - __ movptr(rdi, secondary_supers_addr); 30.74 - // Load the array length. 30.75 - __ movl(rcx, Address(rdi, arrayOopDesc::length_offset_in_bytes())); 30.76 - // Skip to start of data. 30.77 - __ addptr(rdi, arrayOopDesc::base_offset_in_bytes(T_OBJECT)); 30.78 - // Scan rcx words at [edi] for occurance of rax, 30.79 - // Set NZ/Z based on last compare 30.80 - __ movptr(rax, super_klass_addr); 30.81 - __ repne_scan(); 30.82 + __ bind(L_fallthrough); 30.83 30.84 - // Unspill the temp. registers: 30.85 - __ pop(rdi); 30.86 - __ pop(rcx); 30.87 - __ pop(rax); 30.88 - } 30.89 - __ jcc(Assembler::notEqual, L_failure); 30.90 + if (L_success == NULL) { BLOCK_COMMENT("L_success:"); } 30.91 + if (L_failure == NULL) { BLOCK_COMMENT("L_failure:"); } 30.92 30.93 - // Success. Cache the super we found and proceed in triumph. 30.94 - __ movptr(temp, super_klass_addr); // note: rax, is dead 30.95 - __ movptr(super_cache_addr, temp); 30.96 - 30.97 - if (!fall_through_on_success) 30.98 - __ jmp(L_success); 30.99 - 30.100 - // Fall through on failure! 30.101 - __ bind(L_fallthrough); 30.102 +#undef LOCAL_JCC 30.103 } 30.104 30.105 //
31.1 --- a/src/cpu/x86/vm/stubGenerator_x86_64.cpp Wed Mar 18 11:37:48 2009 -0400 31.2 +++ b/src/cpu/x86/vm/stubGenerator_x86_64.cpp Thu Mar 19 09:13:24 2009 -0700 31.3 @@ -2091,66 +2091,9 @@ 31.4 31.5 Label L_miss; 31.6 31.7 - // a couple of useful fields in sub_klass: 31.8 - int ss_offset = (klassOopDesc::header_size() * HeapWordSize + 31.9 - Klass::secondary_supers_offset_in_bytes()); 31.10 - int sc_offset = (klassOopDesc::header_size() * HeapWordSize + 31.11 - Klass::secondary_super_cache_offset_in_bytes()); 31.12 - Address secondary_supers_addr(sub_klass, ss_offset); 31.13 - Address super_cache_addr( sub_klass, sc_offset); 31.14 - 31.15 - // if the pointers are equal, we are done (e.g., String[] elements) 31.16 - __ cmpptr(super_klass, sub_klass); 31.17 - __ jcc(Assembler::equal, L_success); 31.18 - 31.19 - // check the supertype display: 31.20 - Address super_check_addr(sub_klass, super_check_offset, Address::times_1, 0); 31.21 - __ cmpptr(super_klass, super_check_addr); // test the super type 31.22 - __ jcc(Assembler::equal, L_success); 31.23 - 31.24 - // if it was a primary super, we can just fail immediately 31.25 - __ cmpl(super_check_offset, sc_offset); 31.26 - __ jcc(Assembler::notEqual, L_miss); 31.27 - 31.28 - // Now do a linear scan of the secondary super-klass chain. 31.29 - // The repne_scan instruction uses fixed registers, which we must spill. 31.30 - // (We need a couple more temps in any case.) 31.31 - // This code is rarely used, so simplicity is a virtue here. 31.32 - inc_counter_np(SharedRuntime::_partial_subtype_ctr); 31.33 - { 31.34 - __ push(rax); 31.35 - __ push(rcx); 31.36 - __ push(rdi); 31.37 - assert_different_registers(sub_klass, super_klass, rax, rcx, rdi); 31.38 - 31.39 - __ movptr(rdi, secondary_supers_addr); 31.40 - // Load the array length. 31.41 - __ movl(rcx, Address(rdi, arrayOopDesc::length_offset_in_bytes())); 31.42 - // Skip to start of data. 31.43 - __ addptr(rdi, arrayOopDesc::base_offset_in_bytes(T_OBJECT)); 31.44 - // Scan rcx words at [rdi] for occurance of rax 31.45 - // Set NZ/Z based on last compare 31.46 - __ movptr(rax, super_klass); 31.47 - if (UseCompressedOops) { 31.48 - // Compare against compressed form. Don't need to uncompress because 31.49 - // looks like orig rax is restored in popq below. 31.50 - __ encode_heap_oop(rax); 31.51 - __ repne_scanl(); 31.52 - } else { 31.53 - __ repne_scan(); 31.54 - } 31.55 - 31.56 - // Unspill the temp. registers: 31.57 - __ pop(rdi); 31.58 - __ pop(rcx); 31.59 - __ pop(rax); 31.60 - 31.61 - __ jcc(Assembler::notEqual, L_miss); 31.62 - } 31.63 - 31.64 - // Success. Cache the super we found and proceed in triumph. 31.65 - __ movptr(super_cache_addr, super_klass); // note: rax is dead 31.66 - __ jmp(L_success); 31.67 + __ check_klass_subtype_fast_path(sub_klass, super_klass, noreg, &L_success, &L_miss, NULL, 31.68 + super_check_offset); 31.69 + __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg, &L_success, NULL); 31.70 31.71 // Fall through on failure! 31.72 __ BIND(L_miss);
32.1 --- a/src/cpu/x86/vm/vm_version_x86.cpp Wed Mar 18 11:37:48 2009 -0400 32.2 +++ b/src/cpu/x86/vm/vm_version_x86.cpp Thu Mar 19 09:13:24 2009 -0700 32.3 @@ -284,7 +284,7 @@ 32.4 } 32.5 32.6 char buf[256]; 32.7 - jio_snprintf(buf, sizeof(buf), "(%u cores per cpu, %u threads per core) family %d model %d stepping %d%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s", 32.8 + jio_snprintf(buf, sizeof(buf), "(%u cores per cpu, %u threads per core) family %d model %d stepping %d%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s", 32.9 cores_per_cpu(), threads_per_core(), 32.10 cpu_family(), _model, _stepping, 32.11 (supports_cmov() ? ", cmov" : ""), 32.12 @@ -297,6 +297,7 @@ 32.13 (supports_ssse3()? ", ssse3": ""), 32.14 (supports_sse4_1() ? ", sse4.1" : ""), 32.15 (supports_sse4_2() ? ", sse4.2" : ""), 32.16 + (supports_popcnt() ? ", popcnt" : ""), 32.17 (supports_mmx_ext() ? ", mmxext" : ""), 32.18 (supports_3dnow() ? ", 3dnow" : ""), 32.19 (supports_3dnow2() ? ", 3dnowext" : ""), 32.20 @@ -410,6 +411,13 @@ 32.21 } 32.22 } 32.23 32.24 + // Use population count instruction if available. 32.25 + if (supports_popcnt()) { 32.26 + if (FLAG_IS_DEFAULT(UsePopCountInstruction)) { 32.27 + UsePopCountInstruction = true; 32.28 + } 32.29 + } 32.30 + 32.31 assert(0 <= ReadPrefetchInstr && ReadPrefetchInstr <= 3, "invalid value"); 32.32 assert(0 <= AllocatePrefetchInstr && AllocatePrefetchInstr <= 3, "invalid value"); 32.33
33.1 --- a/src/cpu/x86/vm/vm_version_x86.hpp Wed Mar 18 11:37:48 2009 -0400 33.2 +++ b/src/cpu/x86/vm/vm_version_x86.hpp Thu Mar 19 09:13:24 2009 -0700 33.3 @@ -70,7 +70,9 @@ 33.4 dca : 1, 33.5 sse4_1 : 1, 33.6 sse4_2 : 1, 33.7 - : 11; 33.8 + : 2, 33.9 + popcnt : 1, 33.10 + : 8; 33.11 } bits; 33.12 }; 33.13 33.14 @@ -179,7 +181,8 @@ 33.15 CPU_SSSE3 = (1 << 9), 33.16 CPU_SSE4A = (1 << 10), 33.17 CPU_SSE4_1 = (1 << 11), 33.18 - CPU_SSE4_2 = (1 << 12) 33.19 + CPU_SSE4_2 = (1 << 12), 33.20 + CPU_POPCNT = (1 << 13) 33.21 } cpuFeatureFlags; 33.22 33.23 // cpuid information block. All info derived from executing cpuid with 33.24 @@ -290,6 +293,8 @@ 33.25 result |= CPU_SSE4_1; 33.26 if (_cpuid_info.std_cpuid1_ecx.bits.sse4_2 != 0) 33.27 result |= CPU_SSE4_2; 33.28 + if (_cpuid_info.std_cpuid1_ecx.bits.popcnt != 0) 33.29 + result |= CPU_POPCNT; 33.30 return result; 33.31 } 33.32 33.33 @@ -379,6 +384,7 @@ 33.34 static bool supports_ssse3() { return (_cpuFeatures & CPU_SSSE3)!= 0; } 33.35 static bool supports_sse4_1() { return (_cpuFeatures & CPU_SSE4_1) != 0; } 33.36 static bool supports_sse4_2() { return (_cpuFeatures & CPU_SSE4_2) != 0; } 33.37 + static bool supports_popcnt() { return (_cpuFeatures & CPU_POPCNT) != 0; } 33.38 // 33.39 // AMD features 33.40 //
34.1 --- a/src/cpu/x86/vm/x86_32.ad Wed Mar 18 11:37:48 2009 -0400 34.2 +++ b/src/cpu/x86/vm/x86_32.ad Thu Mar 19 09:13:24 2009 -0700 34.3 @@ -1483,16 +1483,20 @@ 34.4 // main source block for now. In future, we can generalize this by 34.5 // adding a syntax that specifies the sizes of fields in an order, 34.6 // so that the adlc can build the emit functions automagically 34.7 - enc_class OpcP %{ // Emit opcode 34.8 - emit_opcode(cbuf,$primary); 34.9 - %} 34.10 - 34.11 - enc_class OpcS %{ // Emit opcode 34.12 - emit_opcode(cbuf,$secondary); 34.13 - %} 34.14 - 34.15 - enc_class Opcode(immI d8 ) %{ // Emit opcode 34.16 - emit_opcode(cbuf,$d8$$constant); 34.17 + 34.18 + // Emit primary opcode 34.19 + enc_class OpcP %{ 34.20 + emit_opcode(cbuf, $primary); 34.21 + %} 34.22 + 34.23 + // Emit secondary opcode 34.24 + enc_class OpcS %{ 34.25 + emit_opcode(cbuf, $secondary); 34.26 + %} 34.27 + 34.28 + // Emit opcode directly 34.29 + enc_class Opcode(immI d8) %{ 34.30 + emit_opcode(cbuf, $d8$$constant); 34.31 %} 34.32 34.33 enc_class SizePrefix %{ 34.34 @@ -1688,26 +1692,15 @@ 34.35 Register Reax = as_Register(EAX_enc); // super class 34.36 Register Recx = as_Register(ECX_enc); // killed 34.37 Register Resi = as_Register(ESI_enc); // sub class 34.38 - Label hit, miss; 34.39 + Label miss; 34.40 34.41 MacroAssembler _masm(&cbuf); 34.42 - // Compare super with sub directly, since super is not in its own SSA. 34.43 - // The compiler used to emit this test, but we fold it in here, 34.44 - // to allow platform-specific tweaking on sparc. 34.45 - __ cmpptr(Reax, Resi); 34.46 - __ jcc(Assembler::equal, hit); 34.47 -#ifndef PRODUCT 34.48 - __ incrementl(ExternalAddress((address)&SharedRuntime::_partial_subtype_ctr)); 34.49 -#endif //PRODUCT 34.50 - __ movptr(Redi,Address(Resi,sizeof(oopDesc) + Klass::secondary_supers_offset_in_bytes())); 34.51 - __ movl(Recx,Address(Redi,arrayOopDesc::length_offset_in_bytes())); 34.52 - __ addptr(Redi,arrayOopDesc::base_offset_in_bytes(T_OBJECT)); 34.53 - __ repne_scan(); 34.54 - __ jcc(Assembler::notEqual, miss); 34.55 - __ movptr(Address(Resi,sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes()),Reax); 34.56 - __ bind(hit); 34.57 - if( $primary ) 34.58 - __ xorptr(Redi,Redi); 34.59 + __ check_klass_subtype_slow_path(Resi, Reax, Recx, Redi, 34.60 + NULL, &miss, 34.61 + /*set_cond_codes:*/ true); 34.62 + if ($primary) { 34.63 + __ xorptr(Redi, Redi); 34.64 + } 34.65 __ bind(miss); 34.66 %} 34.67 34.68 @@ -6387,6 +6380,67 @@ 34.69 %} 34.70 34.71 34.72 +//---------- Population Count Instructions ------------------------------------- 34.73 + 34.74 +instruct popCountI(eRegI dst, eRegI src) %{ 34.75 + predicate(UsePopCountInstruction); 34.76 + match(Set dst (PopCountI src)); 34.77 + 34.78 + format %{ "POPCNT $dst, $src" %} 34.79 + ins_encode %{ 34.80 + __ popcntl($dst$$Register, $src$$Register); 34.81 + %} 34.82 + ins_pipe(ialu_reg); 34.83 +%} 34.84 + 34.85 +instruct popCountI_mem(eRegI dst, memory mem) %{ 34.86 + predicate(UsePopCountInstruction); 34.87 + match(Set dst (PopCountI (LoadI mem))); 34.88 + 34.89 + format %{ "POPCNT $dst, $mem" %} 34.90 + ins_encode %{ 34.91 + __ popcntl($dst$$Register, $mem$$Address); 34.92 + %} 34.93 + ins_pipe(ialu_reg); 34.94 +%} 34.95 + 34.96 +// Note: Long.bitCount(long) returns an int. 34.97 +instruct popCountL(eRegI dst, eRegL src, eRegI tmp, eFlagsReg cr) %{ 34.98 + predicate(UsePopCountInstruction); 34.99 + match(Set dst (PopCountL src)); 34.100 + effect(KILL cr, TEMP tmp, TEMP dst); 34.101 + 34.102 + format %{ "POPCNT $dst, $src.lo\n\t" 34.103 + "POPCNT $tmp, $src.hi\n\t" 34.104 + "ADD $dst, $tmp" %} 34.105 + ins_encode %{ 34.106 + __ popcntl($dst$$Register, $src$$Register); 34.107 + __ popcntl($tmp$$Register, HIGH_FROM_LOW($src$$Register)); 34.108 + __ addl($dst$$Register, $tmp$$Register); 34.109 + %} 34.110 + ins_pipe(ialu_reg); 34.111 +%} 34.112 + 34.113 +// Note: Long.bitCount(long) returns an int. 34.114 +instruct popCountL_mem(eRegI dst, memory mem, eRegI tmp, eFlagsReg cr) %{ 34.115 + predicate(UsePopCountInstruction); 34.116 + match(Set dst (PopCountL (LoadL mem))); 34.117 + effect(KILL cr, TEMP tmp, TEMP dst); 34.118 + 34.119 + format %{ "POPCNT $dst, $mem\n\t" 34.120 + "POPCNT $tmp, $mem+4\n\t" 34.121 + "ADD $dst, $tmp" %} 34.122 + ins_encode %{ 34.123 + //__ popcntl($dst$$Register, $mem$$Address$$first); 34.124 + //__ popcntl($tmp$$Register, $mem$$Address$$second); 34.125 + __ popcntl($dst$$Register, Address::make_raw($mem$$base, $mem$$index, $mem$$scale, $mem$$disp, false)); 34.126 + __ popcntl($tmp$$Register, Address::make_raw($mem$$base, $mem$$index, $mem$$scale, $mem$$disp + 4, false)); 34.127 + __ addl($dst$$Register, $tmp$$Register); 34.128 + %} 34.129 + ins_pipe(ialu_reg); 34.130 +%} 34.131 + 34.132 + 34.133 //----------Load/Store/Move Instructions--------------------------------------- 34.134 //----------Load Instructions-------------------------------------------------- 34.135 // Load Byte (8bit signed) 34.136 @@ -12501,15 +12555,12 @@ 34.137 effect( KILL rcx, KILL cr ); 34.138 34.139 ins_cost(1100); // slightly larger than the next version 34.140 - format %{ "CMPL EAX,ESI\n\t" 34.141 - "JEQ,s hit\n\t" 34.142 - "MOV EDI,[$sub+Klass::secondary_supers]\n\t" 34.143 + format %{ "MOV EDI,[$sub+Klass::secondary_supers]\n\t" 34.144 "MOV ECX,[EDI+arrayKlass::length]\t# length to scan\n\t" 34.145 "ADD EDI,arrayKlass::base_offset\t# Skip to start of data; set NZ in case count is zero\n\t" 34.146 "REPNE SCASD\t# Scan *EDI++ for a match with EAX while CX-- != 0\n\t" 34.147 "JNE,s miss\t\t# Missed: EDI not-zero\n\t" 34.148 "MOV [$sub+Klass::secondary_super_cache],$super\t# Hit: update cache\n\t" 34.149 - "hit:\n\t" 34.150 "XOR $result,$result\t\t Hit: EDI zero\n\t" 34.151 "miss:\t" %} 34.152 34.153 @@ -12523,9 +12574,7 @@ 34.154 effect( KILL rcx, KILL result ); 34.155 34.156 ins_cost(1000); 34.157 - format %{ "CMPL EAX,ESI\n\t" 34.158 - "JEQ,s miss\t# Actually a hit; we are done.\n\t" 34.159 - "MOV EDI,[$sub+Klass::secondary_supers]\n\t" 34.160 + format %{ "MOV EDI,[$sub+Klass::secondary_supers]\n\t" 34.161 "MOV ECX,[EDI+arrayKlass::length]\t# length to scan\n\t" 34.162 "ADD EDI,arrayKlass::base_offset\t# Skip to start of data; set NZ in case count is zero\n\t" 34.163 "REPNE SCASD\t# Scan *EDI++ for a match with EAX while CX-- != 0\n\t"
35.1 --- a/src/cpu/x86/vm/x86_64.ad Wed Mar 18 11:37:48 2009 -0400 35.2 +++ b/src/cpu/x86/vm/x86_64.ad Thu Mar 19 09:13:24 2009 -0700 35.3 @@ -326,7 +326,6 @@ 35.4 R9, R9_H, 35.5 R10, R10_H, 35.6 R11, R11_H, 35.7 - R12, R12_H, 35.8 R13, R13_H, 35.9 R14, R14_H); 35.10 35.11 @@ -340,7 +339,6 @@ 35.12 R9, R9_H, 35.13 R10, R10_H, 35.14 R11, R11_H, 35.15 - R12, R12_H, 35.16 R13, R13_H, 35.17 R14, R14_H); 35.18 35.19 @@ -354,7 +352,6 @@ 35.20 R9, R9_H, 35.21 R10, R10_H, 35.22 R11, R11_H, 35.23 - R12, R12_H, 35.24 R13, R13_H, 35.25 R14, R14_H); 35.26 35.27 @@ -444,9 +441,6 @@ 35.28 // Singleton class for RDX long register 35.29 reg_class long_rdx_reg(RDX, RDX_H); 35.30 35.31 -// Singleton class for R12 long register 35.32 -reg_class long_r12_reg(R12, R12_H); 35.33 - 35.34 // Class for all int registers (except RSP) 35.35 reg_class int_reg(RAX, 35.36 RDX, 35.37 @@ -1842,7 +1836,9 @@ 35.38 { 35.39 if (UseCompressedOops) { 35.40 st->print_cr("movl rscratch1, [j_rarg0 + oopDesc::klass_offset_in_bytes() #%d]\t", oopDesc::klass_offset_in_bytes()); 35.41 - st->print_cr("leaq rscratch1, [r12_heapbase, r, Address::times_8, 0]"); 35.42 + if (Universe::narrow_oop_shift() != 0) { 35.43 + st->print_cr("leaq rscratch1, [r12_heapbase, r, Address::times_8, 0]"); 35.44 + } 35.45 st->print_cr("cmpq rax, rscratch1\t # Inline cache check"); 35.46 } else { 35.47 st->print_cr("cmpq rax, [j_rarg0 + oopDesc::klass_offset_in_bytes() #%d]\t" 35.48 @@ -1891,7 +1887,11 @@ 35.49 uint MachUEPNode::size(PhaseRegAlloc* ra_) const 35.50 { 35.51 if (UseCompressedOops) { 35.52 - return OptoBreakpoint ? 19 : 20; 35.53 + if (Universe::narrow_oop_shift() == 0) { 35.54 + return OptoBreakpoint ? 15 : 16; 35.55 + } else { 35.56 + return OptoBreakpoint ? 19 : 20; 35.57 + } 35.58 } else { 35.59 return OptoBreakpoint ? 11 : 12; 35.60 } 35.61 @@ -2575,45 +2575,13 @@ 35.62 Register Rrax = as_Register(RAX_enc); // super class 35.63 Register Rrcx = as_Register(RCX_enc); // killed 35.64 Register Rrsi = as_Register(RSI_enc); // sub class 35.65 - Label hit, miss, cmiss; 35.66 + Label miss; 35.67 + const bool set_cond_codes = true; 35.68 35.69 MacroAssembler _masm(&cbuf); 35.70 - // Compare super with sub directly, since super is not in its own SSA. 35.71 - // The compiler used to emit this test, but we fold it in here, 35.72 - // to allow platform-specific tweaking on sparc. 35.73 - __ cmpptr(Rrax, Rrsi); 35.74 - __ jcc(Assembler::equal, hit); 35.75 -#ifndef PRODUCT 35.76 - __ lea(Rrcx, ExternalAddress((address)&SharedRuntime::_partial_subtype_ctr)); 35.77 - __ incrementl(Address(Rrcx, 0)); 35.78 -#endif //PRODUCT 35.79 - __ movptr(Rrdi, Address(Rrsi, 35.80 - sizeof(oopDesc) + 35.81 - Klass::secondary_supers_offset_in_bytes())); 35.82 - __ movl(Rrcx, Address(Rrdi, arrayOopDesc::length_offset_in_bytes())); 35.83 - __ addptr(Rrdi, arrayOopDesc::base_offset_in_bytes(T_OBJECT)); 35.84 - if (UseCompressedOops) { 35.85 - __ encode_heap_oop(Rrax); 35.86 - __ repne_scanl(); 35.87 - __ jcc(Assembler::notEqual, cmiss); 35.88 - __ decode_heap_oop(Rrax); 35.89 - __ movptr(Address(Rrsi, 35.90 - sizeof(oopDesc) + 35.91 - Klass::secondary_super_cache_offset_in_bytes()), 35.92 - Rrax); 35.93 - __ jmp(hit); 35.94 - __ bind(cmiss); 35.95 - __ decode_heap_oop(Rrax); 35.96 - __ jmp(miss); 35.97 - } else { 35.98 - __ repne_scan(); 35.99 - __ jcc(Assembler::notEqual, miss); 35.100 - __ movptr(Address(Rrsi, 35.101 - sizeof(oopDesc) + 35.102 - Klass::secondary_super_cache_offset_in_bytes()), 35.103 - Rrax); 35.104 - } 35.105 - __ bind(hit); 35.106 + __ check_klass_subtype_slow_path(Rrsi, Rrax, Rrcx, Rrdi, 35.107 + NULL, &miss, 35.108 + /*set_cond_codes:*/ true); 35.109 if ($primary) { 35.110 __ xorptr(Rrdi, Rrdi); 35.111 } 35.112 @@ -4906,15 +4874,6 @@ 35.113 interface(REG_INTER); 35.114 %} 35.115 35.116 - 35.117 -operand r12RegL() %{ 35.118 - constraint(ALLOC_IN_RC(long_r12_reg)); 35.119 - match(RegL); 35.120 - 35.121 - format %{ %} 35.122 - interface(REG_INTER); 35.123 -%} 35.124 - 35.125 operand rRegN() %{ 35.126 constraint(ALLOC_IN_RC(int_reg)); 35.127 match(RegN); 35.128 @@ -5289,21 +5248,6 @@ 35.129 %} 35.130 %} 35.131 35.132 -// Indirect Narrow Oop Plus Offset Operand 35.133 -operand indNarrowOopOffset(rRegN src, immL32 off) %{ 35.134 - constraint(ALLOC_IN_RC(ptr_reg)); 35.135 - match(AddP (DecodeN src) off); 35.136 - 35.137 - op_cost(10); 35.138 - format %{"[R12 + $src << 3 + $off] (compressed oop addressing)" %} 35.139 - interface(MEMORY_INTER) %{ 35.140 - base(0xc); // R12 35.141 - index($src); 35.142 - scale(0x3); 35.143 - disp($off); 35.144 - %} 35.145 -%} 35.146 - 35.147 // Indirect Memory Times Scale Plus Positive Index Register Plus Offset Operand 35.148 operand indPosIndexScaleOffset(any_RegP reg, immL32 off, rRegI idx, immI2 scale) 35.149 %{ 35.150 @@ -5321,6 +5265,158 @@ 35.151 %} 35.152 %} 35.153 35.154 +// Indirect Narrow Oop Plus Offset Operand 35.155 +// Note: x86 architecture doesn't support "scale * index + offset" without a base 35.156 +// we can't free r12 even with Universe::narrow_oop_base() == NULL. 35.157 +operand indCompressedOopOffset(rRegN reg, immL32 off) %{ 35.158 + predicate(UseCompressedOops && (Universe::narrow_oop_shift() != 0)); 35.159 + constraint(ALLOC_IN_RC(ptr_reg)); 35.160 + match(AddP (DecodeN reg) off); 35.161 + 35.162 + op_cost(10); 35.163 + format %{"[R12 + $reg << 3 + $off] (compressed oop addressing)" %} 35.164 + interface(MEMORY_INTER) %{ 35.165 + base(0xc); // R12 35.166 + index($reg); 35.167 + scale(0x3); 35.168 + disp($off); 35.169 + %} 35.170 +%} 35.171 + 35.172 +// Indirect Memory Operand 35.173 +operand indirectNarrow(rRegN reg) 35.174 +%{ 35.175 + predicate(Universe::narrow_oop_shift() == 0); 35.176 + constraint(ALLOC_IN_RC(ptr_reg)); 35.177 + match(DecodeN reg); 35.178 + 35.179 + format %{ "[$reg]" %} 35.180 + interface(MEMORY_INTER) %{ 35.181 + base($reg); 35.182 + index(0x4); 35.183 + scale(0x0); 35.184 + disp(0x0); 35.185 + %} 35.186 +%} 35.187 + 35.188 +// Indirect Memory Plus Short Offset Operand 35.189 +operand indOffset8Narrow(rRegN reg, immL8 off) 35.190 +%{ 35.191 + predicate(Universe::narrow_oop_shift() == 0); 35.192 + constraint(ALLOC_IN_RC(ptr_reg)); 35.193 + match(AddP (DecodeN reg) off); 35.194 + 35.195 + format %{ "[$reg + $off (8-bit)]" %} 35.196 + interface(MEMORY_INTER) %{ 35.197 + base($reg); 35.198 + index(0x4); 35.199 + scale(0x0); 35.200 + disp($off); 35.201 + %} 35.202 +%} 35.203 + 35.204 +// Indirect Memory Plus Long Offset Operand 35.205 +operand indOffset32Narrow(rRegN reg, immL32 off) 35.206 +%{ 35.207 + predicate(Universe::narrow_oop_shift() == 0); 35.208 + constraint(ALLOC_IN_RC(ptr_reg)); 35.209 + match(AddP (DecodeN reg) off); 35.210 + 35.211 + format %{ "[$reg + $off (32-bit)]" %} 35.212 + interface(MEMORY_INTER) %{ 35.213 + base($reg); 35.214 + index(0x4); 35.215 + scale(0x0); 35.216 + disp($off); 35.217 + %} 35.218 +%} 35.219 + 35.220 +// Indirect Memory Plus Index Register Plus Offset Operand 35.221 +operand indIndexOffsetNarrow(rRegN reg, rRegL lreg, immL32 off) 35.222 +%{ 35.223 + predicate(Universe::narrow_oop_shift() == 0); 35.224 + constraint(ALLOC_IN_RC(ptr_reg)); 35.225 + match(AddP (AddP (DecodeN reg) lreg) off); 35.226 + 35.227 + op_cost(10); 35.228 + format %{"[$reg + $off + $lreg]" %} 35.229 + interface(MEMORY_INTER) %{ 35.230 + base($reg); 35.231 + index($lreg); 35.232 + scale(0x0); 35.233 + disp($off); 35.234 + %} 35.235 +%} 35.236 + 35.237 +// Indirect Memory Plus Index Register Plus Offset Operand 35.238 +operand indIndexNarrow(rRegN reg, rRegL lreg) 35.239 +%{ 35.240 + predicate(Universe::narrow_oop_shift() == 0); 35.241 + constraint(ALLOC_IN_RC(ptr_reg)); 35.242 + match(AddP (DecodeN reg) lreg); 35.243 + 35.244 + op_cost(10); 35.245 + format %{"[$reg + $lreg]" %} 35.246 + interface(MEMORY_INTER) %{ 35.247 + base($reg); 35.248 + index($lreg); 35.249 + scale(0x0); 35.250 + disp(0x0); 35.251 + %} 35.252 +%} 35.253 + 35.254 +// Indirect Memory Times Scale Plus Index Register 35.255 +operand indIndexScaleNarrow(rRegN reg, rRegL lreg, immI2 scale) 35.256 +%{ 35.257 + predicate(Universe::narrow_oop_shift() == 0); 35.258 + constraint(ALLOC_IN_RC(ptr_reg)); 35.259 + match(AddP (DecodeN reg) (LShiftL lreg scale)); 35.260 + 35.261 + op_cost(10); 35.262 + format %{"[$reg + $lreg << $scale]" %} 35.263 + interface(MEMORY_INTER) %{ 35.264 + base($reg); 35.265 + index($lreg); 35.266 + scale($scale); 35.267 + disp(0x0); 35.268 + %} 35.269 +%} 35.270 + 35.271 +// Indirect Memory Times Scale Plus Index Register Plus Offset Operand 35.272 +operand indIndexScaleOffsetNarrow(rRegN reg, immL32 off, rRegL lreg, immI2 scale) 35.273 +%{ 35.274 + predicate(Universe::narrow_oop_shift() == 0); 35.275 + constraint(ALLOC_IN_RC(ptr_reg)); 35.276 + match(AddP (AddP (DecodeN reg) (LShiftL lreg scale)) off); 35.277 + 35.278 + op_cost(10); 35.279 + format %{"[$reg + $off + $lreg << $scale]" %} 35.280 + interface(MEMORY_INTER) %{ 35.281 + base($reg); 35.282 + index($lreg); 35.283 + scale($scale); 35.284 + disp($off); 35.285 + %} 35.286 +%} 35.287 + 35.288 +// Indirect Memory Times Scale Plus Positive Index Register Plus Offset Operand 35.289 +operand indPosIndexScaleOffsetNarrow(rRegN reg, immL32 off, rRegI idx, immI2 scale) 35.290 +%{ 35.291 + constraint(ALLOC_IN_RC(ptr_reg)); 35.292 + predicate(Universe::narrow_oop_shift() == 0 && n->in(2)->in(3)->in(1)->as_Type()->type()->is_long()->_lo >= 0); 35.293 + match(AddP (AddP (DecodeN reg) (LShiftL (ConvI2L idx) scale)) off); 35.294 + 35.295 + op_cost(10); 35.296 + format %{"[$reg + $off + $idx << $scale]" %} 35.297 + interface(MEMORY_INTER) %{ 35.298 + base($reg); 35.299 + index($idx); 35.300 + scale($scale); 35.301 + disp($off); 35.302 + %} 35.303 +%} 35.304 + 35.305 + 35.306 //----------Special Memory Operands-------------------------------------------- 35.307 // Stack Slot Operand - This operand is used for loading and storing temporary 35.308 // values on the stack where a match requires a value to 35.309 @@ -5488,7 +5584,10 @@ 35.310 35.311 opclass memory(indirect, indOffset8, indOffset32, indIndexOffset, indIndex, 35.312 indIndexScale, indIndexScaleOffset, indPosIndexScaleOffset, 35.313 - indNarrowOopOffset); 35.314 + indCompressedOopOffset, 35.315 + indirectNarrow, indOffset8Narrow, indOffset32Narrow, 35.316 + indIndexOffsetNarrow, indIndexNarrow, indIndexScaleNarrow, 35.317 + indIndexScaleOffsetNarrow, indPosIndexScaleOffsetNarrow); 35.318 35.319 //----------PIPELINE----------------------------------------------------------- 35.320 // Rules which define the behavior of the target architectures pipeline. 35.321 @@ -6234,9 +6333,7 @@ 35.322 ins_cost(125); // XXX 35.323 format %{ "movl $dst, $mem\t# compressed ptr" %} 35.324 ins_encode %{ 35.325 - Address addr = build_address($mem$$base, $mem$$index, $mem$$scale, $mem$$disp); 35.326 - Register dst = as_Register($dst$$reg); 35.327 - __ movl(dst, addr); 35.328 + __ movl($dst$$Register, $mem$$Address); 35.329 %} 35.330 ins_pipe(ialu_reg_mem); // XXX 35.331 %} 35.332 @@ -6262,9 +6359,7 @@ 35.333 ins_cost(125); // XXX 35.334 format %{ "movl $dst, $mem\t# compressed klass ptr" %} 35.335 ins_encode %{ 35.336 - Address addr = build_address($mem$$base, $mem$$index, $mem$$scale, $mem$$disp); 35.337 - Register dst = as_Register($dst$$reg); 35.338 - __ movl(dst, addr); 35.339 + __ movl($dst$$Register, $mem$$Address); 35.340 %} 35.341 ins_pipe(ialu_reg_mem); // XXX 35.342 %} 35.343 @@ -6418,6 +6513,102 @@ 35.344 ins_pipe(ialu_reg_reg_fat); 35.345 %} 35.346 35.347 +instruct leaPPosIdxScaleOff(rRegP dst, indPosIndexScaleOffset mem) 35.348 +%{ 35.349 + match(Set dst mem); 35.350 + 35.351 + ins_cost(110); 35.352 + format %{ "leaq $dst, $mem\t# ptr posidxscaleoff" %} 35.353 + opcode(0x8D); 35.354 + ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem)); 35.355 + ins_pipe(ialu_reg_reg_fat); 35.356 +%} 35.357 + 35.358 +// Load Effective Address which uses Narrow (32-bits) oop 35.359 +instruct leaPCompressedOopOffset(rRegP dst, indCompressedOopOffset mem) 35.360 +%{ 35.361 + predicate(UseCompressedOops && (Universe::narrow_oop_shift() != 0)); 35.362 + match(Set dst mem); 35.363 + 35.364 + ins_cost(110); 35.365 + format %{ "leaq $dst, $mem\t# ptr compressedoopoff32" %} 35.366 + opcode(0x8D); 35.367 + ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem)); 35.368 + ins_pipe(ialu_reg_reg_fat); 35.369 +%} 35.370 + 35.371 +instruct leaP8Narrow(rRegP dst, indOffset8Narrow mem) 35.372 +%{ 35.373 + predicate(Universe::narrow_oop_shift() == 0); 35.374 + match(Set dst mem); 35.375 + 35.376 + ins_cost(110); // XXX 35.377 + format %{ "leaq $dst, $mem\t# ptr off8narrow" %} 35.378 + opcode(0x8D); 35.379 + ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem)); 35.380 + ins_pipe(ialu_reg_reg_fat); 35.381 +%} 35.382 + 35.383 +instruct leaP32Narrow(rRegP dst, indOffset32Narrow mem) 35.384 +%{ 35.385 + predicate(Universe::narrow_oop_shift() == 0); 35.386 + match(Set dst mem); 35.387 + 35.388 + ins_cost(110); 35.389 + format %{ "leaq $dst, $mem\t# ptr off32narrow" %} 35.390 + opcode(0x8D); 35.391 + ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem)); 35.392 + ins_pipe(ialu_reg_reg_fat); 35.393 +%} 35.394 + 35.395 +instruct leaPIdxOffNarrow(rRegP dst, indIndexOffsetNarrow mem) 35.396 +%{ 35.397 + predicate(Universe::narrow_oop_shift() == 0); 35.398 + match(Set dst mem); 35.399 + 35.400 + ins_cost(110); 35.401 + format %{ "leaq $dst, $mem\t# ptr idxoffnarrow" %} 35.402 + opcode(0x8D); 35.403 + ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem)); 35.404 + ins_pipe(ialu_reg_reg_fat); 35.405 +%} 35.406 + 35.407 +instruct leaPIdxScaleNarrow(rRegP dst, indIndexScaleNarrow mem) 35.408 +%{ 35.409 + predicate(Universe::narrow_oop_shift() == 0); 35.410 + match(Set dst mem); 35.411 + 35.412 + ins_cost(110); 35.413 + format %{ "leaq $dst, $mem\t# ptr idxscalenarrow" %} 35.414 + opcode(0x8D); 35.415 + ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem)); 35.416 + ins_pipe(ialu_reg_reg_fat); 35.417 +%} 35.418 + 35.419 +instruct leaPIdxScaleOffNarrow(rRegP dst, indIndexScaleOffsetNarrow mem) 35.420 +%{ 35.421 + predicate(Universe::narrow_oop_shift() == 0); 35.422 + match(Set dst mem); 35.423 + 35.424 + ins_cost(110); 35.425 + format %{ "leaq $dst, $mem\t# ptr idxscaleoffnarrow" %} 35.426 + opcode(0x8D); 35.427 + ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem)); 35.428 + ins_pipe(ialu_reg_reg_fat); 35.429 +%} 35.430 + 35.431 +instruct leaPPosIdxScaleOffNarrow(rRegP dst, indPosIndexScaleOffsetNarrow mem) 35.432 +%{ 35.433 + predicate(Universe::narrow_oop_shift() == 0); 35.434 + match(Set dst mem); 35.435 + 35.436 + ins_cost(110); 35.437 + format %{ "leaq $dst, $mem\t# ptr posidxscaleoffnarrow" %} 35.438 + opcode(0x8D); 35.439 + ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem)); 35.440 + ins_pipe(ialu_reg_reg_fat); 35.441 +%} 35.442 + 35.443 instruct loadConI(rRegI dst, immI src) 35.444 %{ 35.445 match(Set dst src); 35.446 @@ -6528,8 +6719,7 @@ 35.447 effect(KILL cr); 35.448 format %{ "xorq $dst, $src\t# compressed NULL ptr" %} 35.449 ins_encode %{ 35.450 - Register dst = $dst$$Register; 35.451 - __ xorq(dst, dst); 35.452 + __ xorq($dst$$Register, $dst$$Register); 35.453 %} 35.454 ins_pipe(ialu_reg); 35.455 %} 35.456 @@ -6541,11 +6731,10 @@ 35.457 format %{ "movl $dst, $src\t# compressed ptr" %} 35.458 ins_encode %{ 35.459 address con = (address)$src$$constant; 35.460 - Register dst = $dst$$Register; 35.461 if (con == NULL) { 35.462 ShouldNotReachHere(); 35.463 } else { 35.464 - __ set_narrow_oop(dst, (jobject)$src$$constant); 35.465 + __ set_narrow_oop($dst$$Register, (jobject)$src$$constant); 35.466 } 35.467 %} 35.468 ins_pipe(ialu_reg_fat); // XXX 35.469 @@ -6794,12 +6983,25 @@ 35.470 ins_pipe(ialu_mem_reg); 35.471 %} 35.472 35.473 +instruct storeImmP0(memory mem, immP0 zero) 35.474 +%{ 35.475 + predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL)); 35.476 + match(Set mem (StoreP mem zero)); 35.477 + 35.478 + ins_cost(125); // XXX 35.479 + format %{ "movq $mem, R12\t# ptr (R12_heapbase==0)" %} 35.480 + ins_encode %{ 35.481 + __ movq($mem$$Address, r12); 35.482 + %} 35.483 + ins_pipe(ialu_mem_reg); 35.484 +%} 35.485 + 35.486 // Store NULL Pointer, mark word, or other simple pointer constant. 35.487 instruct storeImmP(memory mem, immP31 src) 35.488 %{ 35.489 match(Set mem (StoreP mem src)); 35.490 35.491 - ins_cost(125); // XXX 35.492 + ins_cost(150); // XXX 35.493 format %{ "movq $mem, $src\t# ptr" %} 35.494 opcode(0xC7); /* C7 /0 */ 35.495 ins_encode(REX_mem_wide(mem), OpcP, RM_opc_mem(0x00, mem), Con32(src)); 35.496 @@ -6814,14 +7016,55 @@ 35.497 ins_cost(125); // XXX 35.498 format %{ "movl $mem, $src\t# compressed ptr" %} 35.499 ins_encode %{ 35.500 - Address addr = build_address($mem$$base, $mem$$index, $mem$$scale, $mem$$disp); 35.501 - Register src = as_Register($src$$reg); 35.502 - __ movl(addr, src); 35.503 + __ movl($mem$$Address, $src$$Register); 35.504 %} 35.505 ins_pipe(ialu_mem_reg); 35.506 %} 35.507 35.508 +instruct storeImmN0(memory mem, immN0 zero) 35.509 +%{ 35.510 + predicate(Universe::narrow_oop_base() == NULL); 35.511 + match(Set mem (StoreN mem zero)); 35.512 + 35.513 + ins_cost(125); // XXX 35.514 + format %{ "movl $mem, R12\t# compressed ptr (R12_heapbase==0)" %} 35.515 + ins_encode %{ 35.516 + __ movl($mem$$Address, r12); 35.517 + %} 35.518 + ins_pipe(ialu_mem_reg); 35.519 +%} 35.520 + 35.521 +instruct storeImmN(memory mem, immN src) 35.522 +%{ 35.523 + match(Set mem (StoreN mem src)); 35.524 + 35.525 + ins_cost(150); // XXX 35.526 + format %{ "movl $mem, $src\t# compressed ptr" %} 35.527 + ins_encode %{ 35.528 + address con = (address)$src$$constant; 35.529 + if (con == NULL) { 35.530 + __ movl($mem$$Address, (int32_t)0); 35.531 + } else { 35.532 + __ set_narrow_oop($mem$$Address, (jobject)$src$$constant); 35.533 + } 35.534 + %} 35.535 + ins_pipe(ialu_mem_imm); 35.536 +%} 35.537 + 35.538 // Store Integer Immediate 35.539 +instruct storeImmI0(memory mem, immI0 zero) 35.540 +%{ 35.541 + predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL)); 35.542 + match(Set mem (StoreI mem zero)); 35.543 + 35.544 + ins_cost(125); // XXX 35.545 + format %{ "movl $mem, R12\t# int (R12_heapbase==0)" %} 35.546 + ins_encode %{ 35.547 + __ movl($mem$$Address, r12); 35.548 + %} 35.549 + ins_pipe(ialu_mem_reg); 35.550 +%} 35.551 + 35.552 instruct storeImmI(memory mem, immI src) 35.553 %{ 35.554 match(Set mem (StoreI mem src)); 35.555 @@ -6834,6 +7077,19 @@ 35.556 %} 35.557 35.558 // Store Long Immediate 35.559 +instruct storeImmL0(memory mem, immL0 zero) 35.560 +%{ 35.561 + predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL)); 35.562 + match(Set mem (StoreL mem zero)); 35.563 + 35.564 + ins_cost(125); // XXX 35.565 + format %{ "movq $mem, R12\t# long (R12_heapbase==0)" %} 35.566 + ins_encode %{ 35.567 + __ movq($mem$$Address, r12); 35.568 + %} 35.569 + ins_pipe(ialu_mem_reg); 35.570 +%} 35.571 + 35.572 instruct storeImmL(memory mem, immL32 src) 35.573 %{ 35.574 match(Set mem (StoreL mem src)); 35.575 @@ -6846,6 +7102,19 @@ 35.576 %} 35.577 35.578 // Store Short/Char Immediate 35.579 +instruct storeImmC0(memory mem, immI0 zero) 35.580 +%{ 35.581 + predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL)); 35.582 + match(Set mem (StoreC mem zero)); 35.583 + 35.584 + ins_cost(125); // XXX 35.585 + format %{ "movw $mem, R12\t# short/char (R12_heapbase==0)" %} 35.586 + ins_encode %{ 35.587 + __ movw($mem$$Address, r12); 35.588 + %} 35.589 + ins_pipe(ialu_mem_reg); 35.590 +%} 35.591 + 35.592 instruct storeImmI16(memory mem, immI16 src) 35.593 %{ 35.594 predicate(UseStoreImmI16); 35.595 @@ -6859,6 +7128,19 @@ 35.596 %} 35.597 35.598 // Store Byte Immediate 35.599 +instruct storeImmB0(memory mem, immI0 zero) 35.600 +%{ 35.601 + predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL)); 35.602 + match(Set mem (StoreB mem zero)); 35.603 + 35.604 + ins_cost(125); // XXX 35.605 + format %{ "movb $mem, R12\t# short/char (R12_heapbase==0)" %} 35.606 + ins_encode %{ 35.607 + __ movb($mem$$Address, r12); 35.608 + %} 35.609 + ins_pipe(ialu_mem_reg); 35.610 +%} 35.611 + 35.612 instruct storeImmB(memory mem, immI8 src) 35.613 %{ 35.614 match(Set mem (StoreB mem src)); 35.615 @@ -6898,6 +7180,19 @@ 35.616 %} 35.617 35.618 // Store CMS card-mark Immediate 35.619 +instruct storeImmCM0_reg(memory mem, immI0 zero) 35.620 +%{ 35.621 + predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL)); 35.622 + match(Set mem (StoreCM mem zero)); 35.623 + 35.624 + ins_cost(125); // XXX 35.625 + format %{ "movb $mem, R12\t# CMS card-mark byte 0 (R12_heapbase==0)" %} 35.626 + ins_encode %{ 35.627 + __ movb($mem$$Address, r12); 35.628 + %} 35.629 + ins_pipe(ialu_mem_reg); 35.630 +%} 35.631 + 35.632 instruct storeImmCM0(memory mem, immI0 src) 35.633 %{ 35.634 match(Set mem (StoreCM mem src)); 35.635 @@ -6931,6 +7226,19 @@ 35.636 %} 35.637 35.638 // Store immediate Float value (it is faster than store from XMM register) 35.639 +instruct storeF0(memory mem, immF0 zero) 35.640 +%{ 35.641 + predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL)); 35.642 + match(Set mem (StoreF mem zero)); 35.643 + 35.644 + ins_cost(25); // XXX 35.645 + format %{ "movl $mem, R12\t# float 0. (R12_heapbase==0)" %} 35.646 + ins_encode %{ 35.647 + __ movl($mem$$Address, r12); 35.648 + %} 35.649 + ins_pipe(ialu_mem_reg); 35.650 +%} 35.651 + 35.652 instruct storeF_imm(memory mem, immF src) 35.653 %{ 35.654 match(Set mem (StoreF mem src)); 35.655 @@ -6957,6 +7265,7 @@ 35.656 // Store immediate double 0.0 (it is faster than store from XMM register) 35.657 instruct storeD0_imm(memory mem, immD0 src) 35.658 %{ 35.659 + predicate(!UseCompressedOops || (Universe::narrow_oop_base() != NULL)); 35.660 match(Set mem (StoreD mem src)); 35.661 35.662 ins_cost(50); 35.663 @@ -6966,6 +7275,19 @@ 35.664 ins_pipe(ialu_mem_imm); 35.665 %} 35.666 35.667 +instruct storeD0(memory mem, immD0 zero) 35.668 +%{ 35.669 + predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL)); 35.670 + match(Set mem (StoreD mem zero)); 35.671 + 35.672 + ins_cost(25); // XXX 35.673 + format %{ "movq $mem, R12\t# double 0. (R12_heapbase==0)" %} 35.674 + ins_encode %{ 35.675 + __ movq($mem$$Address, r12); 35.676 + %} 35.677 + ins_pipe(ialu_mem_reg); 35.678 +%} 35.679 + 35.680 instruct storeSSI(stackSlotI dst, rRegI src) 35.681 %{ 35.682 match(Set dst src); 35.683 @@ -7077,6 +7399,56 @@ 35.684 ins_pipe( ialu_mem_reg ); 35.685 %} 35.686 35.687 + 35.688 +//---------- Population Count Instructions ------------------------------------- 35.689 + 35.690 +instruct popCountI(rRegI dst, rRegI src) %{ 35.691 + predicate(UsePopCountInstruction); 35.692 + match(Set dst (PopCountI src)); 35.693 + 35.694 + format %{ "popcnt $dst, $src" %} 35.695 + ins_encode %{ 35.696 + __ popcntl($dst$$Register, $src$$Register); 35.697 + %} 35.698 + ins_pipe(ialu_reg); 35.699 +%} 35.700 + 35.701 +instruct popCountI_mem(rRegI dst, memory mem) %{ 35.702 + predicate(UsePopCountInstruction); 35.703 + match(Set dst (PopCountI (LoadI mem))); 35.704 + 35.705 + format %{ "popcnt $dst, $mem" %} 35.706 + ins_encode %{ 35.707 + __ popcntl($dst$$Register, $mem$$Address); 35.708 + %} 35.709 + ins_pipe(ialu_reg); 35.710 +%} 35.711 + 35.712 +// Note: Long.bitCount(long) returns an int. 35.713 +instruct popCountL(rRegI dst, rRegL src) %{ 35.714 + predicate(UsePopCountInstruction); 35.715 + match(Set dst (PopCountL src)); 35.716 + 35.717 + format %{ "popcnt $dst, $src" %} 35.718 + ins_encode %{ 35.719 + __ popcntq($dst$$Register, $src$$Register); 35.720 + %} 35.721 + ins_pipe(ialu_reg); 35.722 +%} 35.723 + 35.724 +// Note: Long.bitCount(long) returns an int. 35.725 +instruct popCountL_mem(rRegI dst, memory mem) %{ 35.726 + predicate(UsePopCountInstruction); 35.727 + match(Set dst (PopCountL (LoadL mem))); 35.728 + 35.729 + format %{ "popcnt $dst, $mem" %} 35.730 + ins_encode %{ 35.731 + __ popcntq($dst$$Register, $mem$$Address); 35.732 + %} 35.733 + ins_pipe(ialu_reg); 35.734 +%} 35.735 + 35.736 + 35.737 //----------MemBar Instructions----------------------------------------------- 35.738 // Memory barrier flavors 35.739 35.740 @@ -7192,9 +7564,7 @@ 35.741 effect(KILL cr); 35.742 format %{ "encode_heap_oop_not_null $dst,$src" %} 35.743 ins_encode %{ 35.744 - Register s = $src$$Register; 35.745 - Register d = $dst$$Register; 35.746 - __ encode_heap_oop_not_null(d, s); 35.747 + __ encode_heap_oop_not_null($dst$$Register, $src$$Register); 35.748 %} 35.749 ins_pipe(ialu_reg_long); 35.750 %} 35.751 @@ -7224,7 +7594,11 @@ 35.752 ins_encode %{ 35.753 Register s = $src$$Register; 35.754 Register d = $dst$$Register; 35.755 - __ decode_heap_oop_not_null(d, s); 35.756 + if (s != d) { 35.757 + __ decode_heap_oop_not_null(d, s); 35.758 + } else { 35.759 + __ decode_heap_oop_not_null(d); 35.760 + } 35.761 %} 35.762 ins_pipe(ialu_reg_long); 35.763 %} 35.764 @@ -11389,8 +11763,9 @@ 35.765 35.766 // This will generate a signed flags result. This should be OK since 35.767 // any compare to a zero should be eq/neq. 35.768 -instruct testP_reg_mem(rFlagsReg cr, memory op, immP0 zero) 35.769 -%{ 35.770 +instruct testP_mem(rFlagsReg cr, memory op, immP0 zero) 35.771 +%{ 35.772 + predicate(!UseCompressedOops || (Universe::narrow_oop_base() != NULL)); 35.773 match(Set cr (CmpP (LoadP op) zero)); 35.774 35.775 ins_cost(500); // XXX 35.776 @@ -11401,13 +11776,24 @@ 35.777 ins_pipe(ialu_cr_reg_imm); 35.778 %} 35.779 35.780 +instruct testP_mem_reg0(rFlagsReg cr, memory mem, immP0 zero) 35.781 +%{ 35.782 + predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL)); 35.783 + match(Set cr (CmpP (LoadP mem) zero)); 35.784 + 35.785 + format %{ "cmpq R12, $mem\t# ptr (R12_heapbase==0)" %} 35.786 + ins_encode %{ 35.787 + __ cmpq(r12, $mem$$Address); 35.788 + %} 35.789 + ins_pipe(ialu_cr_reg_mem); 35.790 +%} 35.791 35.792 instruct compN_rReg(rFlagsRegU cr, rRegN op1, rRegN op2) 35.793 %{ 35.794 match(Set cr (CmpN op1 op2)); 35.795 35.796 format %{ "cmpl $op1, $op2\t# compressed ptr" %} 35.797 - ins_encode %{ __ cmpl(as_Register($op1$$reg), as_Register($op2$$reg)); %} 35.798 + ins_encode %{ __ cmpl($op1$$Register, $op2$$Register); %} 35.799 ins_pipe(ialu_cr_reg_reg); 35.800 %} 35.801 35.802 @@ -11415,11 +11801,30 @@ 35.803 %{ 35.804 match(Set cr (CmpN src (LoadN mem))); 35.805 35.806 - ins_cost(500); // XXX 35.807 - format %{ "cmpl $src, mem\t# compressed ptr" %} 35.808 + format %{ "cmpl $src, $mem\t# compressed ptr" %} 35.809 ins_encode %{ 35.810 - Address adr = build_address($mem$$base, $mem$$index, $mem$$scale, $mem$$disp); 35.811 - __ cmpl(as_Register($src$$reg), adr); 35.812 + __ cmpl($src$$Register, $mem$$Address); 35.813 + %} 35.814 + ins_pipe(ialu_cr_reg_mem); 35.815 +%} 35.816 + 35.817 +instruct compN_rReg_imm(rFlagsRegU cr, rRegN op1, immN op2) %{ 35.818 + match(Set cr (CmpN op1 op2)); 35.819 + 35.820 + format %{ "cmpl $op1, $op2\t# compressed ptr" %} 35.821 + ins_encode %{ 35.822 + __ cmp_narrow_oop($op1$$Register, (jobject)$op2$$constant); 35.823 + %} 35.824 + ins_pipe(ialu_cr_reg_imm); 35.825 +%} 35.826 + 35.827 +instruct compN_mem_imm(rFlagsRegU cr, memory mem, immN src) 35.828 +%{ 35.829 + match(Set cr (CmpN src (LoadN mem))); 35.830 + 35.831 + format %{ "cmpl $mem, $src\t# compressed ptr" %} 35.832 + ins_encode %{ 35.833 + __ cmp_narrow_oop($mem$$Address, (jobject)$src$$constant); 35.834 %} 35.835 ins_pipe(ialu_cr_reg_mem); 35.836 %} 35.837 @@ -11432,15 +11837,27 @@ 35.838 ins_pipe(ialu_cr_reg_imm); 35.839 %} 35.840 35.841 -instruct testN_reg_mem(rFlagsReg cr, memory mem, immN0 zero) 35.842 -%{ 35.843 +instruct testN_mem(rFlagsReg cr, memory mem, immN0 zero) 35.844 +%{ 35.845 + predicate(Universe::narrow_oop_base() != NULL); 35.846 match(Set cr (CmpN (LoadN mem) zero)); 35.847 35.848 ins_cost(500); // XXX 35.849 format %{ "testl $mem, 0xffffffff\t# compressed ptr" %} 35.850 ins_encode %{ 35.851 - Address addr = build_address($mem$$base, $mem$$index, $mem$$scale, $mem$$disp); 35.852 - __ cmpl(addr, (int)0xFFFFFFFF); 35.853 + __ cmpl($mem$$Address, (int)0xFFFFFFFF); 35.854 + %} 35.855 + ins_pipe(ialu_cr_reg_mem); 35.856 +%} 35.857 + 35.858 +instruct testN_mem_reg0(rFlagsReg cr, memory mem, immN0 zero) 35.859 +%{ 35.860 + predicate(Universe::narrow_oop_base() == NULL); 35.861 + match(Set cr (CmpN (LoadN mem) zero)); 35.862 + 35.863 + format %{ "cmpl R12, $mem\t# compressed ptr (R12_heapbase==0)" %} 35.864 + ins_encode %{ 35.865 + __ cmpl(r12, $mem$$Address); 35.866 %} 35.867 ins_pipe(ialu_cr_reg_mem); 35.868 %} 35.869 @@ -11472,7 +11889,6 @@ 35.870 %{ 35.871 match(Set cr (CmpL op1 (LoadL op2))); 35.872 35.873 - ins_cost(500); // XXX 35.874 format %{ "cmpq $op1, $op2" %} 35.875 opcode(0x3B); /* Opcode 3B /r */ 35.876 ins_encode(REX_reg_mem_wide(op1, op2), OpcP, reg_mem(op1, op2)); 35.877 @@ -11733,15 +12149,12 @@ 35.878 effect(KILL rcx, KILL cr); 35.879 35.880 ins_cost(1100); // slightly larger than the next version 35.881 - format %{ "cmpq rax, rsi\n\t" 35.882 - "jeq,s hit\n\t" 35.883 - "movq rdi, [$sub + (sizeof(oopDesc) + Klass::secondary_supers_offset_in_bytes())]\n\t" 35.884 + format %{ "movq rdi, [$sub + (sizeof(oopDesc) + Klass::secondary_supers_offset_in_bytes())]\n\t" 35.885 "movl rcx, [rdi + arrayOopDesc::length_offset_in_bytes()]\t# length to scan\n\t" 35.886 "addq rdi, arrayOopDex::base_offset_in_bytes(T_OBJECT)\t# Skip to start of data; set NZ in case count is zero\n\t" 35.887 "repne scasq\t# Scan *rdi++ for a match with rax while rcx--\n\t" 35.888 "jne,s miss\t\t# Missed: rdi not-zero\n\t" 35.889 "movq [$sub + (sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes())], $super\t# Hit: update cache\n\t" 35.890 - "hit:\n\t" 35.891 "xorq $result, $result\t\t Hit: rdi zero\n\t" 35.892 "miss:\t" %} 35.893 35.894 @@ -11756,13 +12169,10 @@ 35.895 rdi_RegP result) 35.896 %{ 35.897 match(Set cr (CmpP (PartialSubtypeCheck sub super) zero)); 35.898 - predicate(!UseCompressedOops); // decoding oop kills condition codes 35.899 effect(KILL rcx, KILL result); 35.900 35.901 ins_cost(1000); 35.902 - format %{ "cmpq rax, rsi\n\t" 35.903 - "jeq,s miss\t# Actually a hit; we are done.\n\t" 35.904 - "movq rdi, [$sub + (sizeof(oopDesc) + Klass::secondary_supers_offset_in_bytes())]\n\t" 35.905 + format %{ "movq rdi, [$sub + (sizeof(oopDesc) + Klass::secondary_supers_offset_in_bytes())]\n\t" 35.906 "movl rcx, [rdi + arrayOopDesc::length_offset_in_bytes()]\t# length to scan\n\t" 35.907 "addq rdi, arrayOopDex::base_offset_in_bytes(T_OBJECT)\t# Skip to start of data; set NZ in case count is zero\n\t" 35.908 "repne scasq\t# Scan *rdi++ for a match with rax while cx-- != 0\n\t"
36.1 --- a/src/os/linux/vm/os_linux.cpp Wed Mar 18 11:37:48 2009 -0400 36.2 +++ b/src/os/linux/vm/os_linux.cpp Thu Mar 19 09:13:24 2009 -0700 36.3 @@ -2582,7 +2582,7 @@ 36.4 #define SHM_HUGETLB 04000 36.5 #endif 36.6 36.7 -char* os::reserve_memory_special(size_t bytes) { 36.8 +char* os::reserve_memory_special(size_t bytes, char* req_addr) { 36.9 assert(UseLargePages, "only for large pages"); 36.10 36.11 key_t key = IPC_PRIVATE;
37.1 --- a/src/os/solaris/dtrace/generateJvmOffsets.cpp Wed Mar 18 11:37:48 2009 -0400 37.2 +++ b/src/os/solaris/dtrace/generateJvmOffsets.cpp Thu Mar 19 09:13:24 2009 -0700 37.3 @@ -249,6 +249,10 @@ 37.4 37.5 printf("\n"); 37.6 37.7 + GEN_OFFS(NarrowOopStruct, _base); 37.8 + GEN_OFFS(NarrowOopStruct, _shift); 37.9 + printf("\n"); 37.10 + 37.11 GEN_VALUE(SIZE_HeapBlockHeader, sizeof(HeapBlock::Header)); 37.12 GEN_SIZE(oopDesc); 37.13 GEN_SIZE(constantPoolOopDesc);
38.1 --- a/src/os/solaris/dtrace/jhelper.d Wed Mar 18 11:37:48 2009 -0400 38.2 +++ b/src/os/solaris/dtrace/jhelper.d Thu Mar 19 09:13:24 2009 -0700 38.3 @@ -46,7 +46,10 @@ 38.4 extern pointer __1cJCodeCacheF_heap_; 38.5 extern pointer __1cIUniverseP_methodKlassObj_; 38.6 extern pointer __1cIUniverseO_collectedHeap_; 38.7 -extern pointer __1cIUniverseK_heap_base_; 38.8 +extern pointer __1cIUniverseL_narrow_oop_; 38.9 +#ifdef _LP64 38.10 +extern pointer UseCompressedOops; 38.11 +#endif 38.12 38.13 extern pointer __1cHnmethodG__vtbl_; 38.14 extern pointer __1cKBufferBlobG__vtbl_; 38.15 @@ -56,6 +59,7 @@ 38.16 #define copyin_uint16(ADDR) *(uint16_t*) copyin((pointer) (ADDR), sizeof(uint16_t)) 38.17 #define copyin_uint32(ADDR) *(uint32_t*) copyin((pointer) (ADDR), sizeof(uint32_t)) 38.18 #define copyin_int32(ADDR) *(int32_t*) copyin((pointer) (ADDR), sizeof(int32_t)) 38.19 +#define copyin_uint8(ADDR) *(uint8_t*) copyin((pointer) (ADDR), sizeof(uint8_t)) 38.20 38.21 #define SAME(x) x 38.22 #define copyin_offset(JVM_CONST) JVM_CONST = \ 38.23 @@ -132,6 +136,9 @@ 38.24 copyin_offset(SIZE_oopDesc); 38.25 copyin_offset(SIZE_constantPoolOopDesc); 38.26 38.27 + copyin_offset(OFFSET_NarrowOopStruct_base); 38.28 + copyin_offset(OFFSET_NarrowOopStruct_shift); 38.29 + 38.30 /* 38.31 * The PC to translate is in arg0. 38.32 */ 38.33 @@ -151,9 +158,19 @@ 38.34 38.35 this->Universe_methodKlassOop = copyin_ptr(&``__1cIUniverseP_methodKlassObj_); 38.36 this->CodeCache_heap_address = copyin_ptr(&``__1cJCodeCacheF_heap_); 38.37 - this->Universe_heap_base = copyin_ptr(&``__1cIUniverseK_heap_base_); 38.38 38.39 /* Reading volatile values */ 38.40 +#ifdef _LP64 38.41 + this->Use_Compressed_Oops = copyin_uint8(&``UseCompressedOops); 38.42 +#else 38.43 + this->Use_Compressed_Oops = 0; 38.44 +#endif 38.45 + 38.46 + this->Universe_narrow_oop_base = copyin_ptr(&``__1cIUniverseL_narrow_oop_ + 38.47 + OFFSET_NarrowOopStruct_base); 38.48 + this->Universe_narrow_oop_shift = copyin_int32(&``__1cIUniverseL_narrow_oop_ + 38.49 + OFFSET_NarrowOopStruct_shift); 38.50 + 38.51 this->CodeCache_low = copyin_ptr(this->CodeCache_heap_address + 38.52 OFFSET_CodeHeap_memory + OFFSET_VirtualSpace_low); 38.53 38.54 @@ -295,7 +312,7 @@ 38.55 38.56 dtrace:helper:ustack: 38.57 /!this->done && this->vtbl == this->BufferBlob_vtbl && 38.58 -this->Universe_heap_base == NULL && 38.59 +this->Use_Compressed_Oops == 0 && 38.60 this->methodOopPtr > this->heap_start && this->methodOopPtr < this->heap_end/ 38.61 { 38.62 MARK_LINE; 38.63 @@ -306,7 +323,7 @@ 38.64 38.65 dtrace:helper:ustack: 38.66 /!this->done && this->vtbl == this->BufferBlob_vtbl && 38.67 -this->Universe_heap_base != NULL && 38.68 +this->Use_Compressed_Oops != 0 && 38.69 this->methodOopPtr > this->heap_start && this->methodOopPtr < this->heap_end/ 38.70 { 38.71 MARK_LINE; 38.72 @@ -314,8 +331,8 @@ 38.73 * Read compressed pointer and decode heap oop, same as oop.inline.hpp 38.74 */ 38.75 this->cklass = copyin_uint32(this->methodOopPtr + OFFSET_oopDesc_metadata); 38.76 - this->klass = (uint64_t)((uintptr_t)this->Universe_heap_base + 38.77 - ((uintptr_t)this->cklass << 3)); 38.78 + this->klass = (uint64_t)((uintptr_t)this->Universe_narrow_oop_base + 38.79 + ((uintptr_t)this->cklass << this->Universe_narrow_oop_shift)); 38.80 this->methodOop = this->klass == this->Universe_methodKlassOop; 38.81 this->done = !this->methodOop; 38.82 }
39.1 --- a/src/os/solaris/dtrace/libjvm_db.c Wed Mar 18 11:37:48 2009 -0400 39.2 +++ b/src/os/solaris/dtrace/libjvm_db.c Thu Mar 19 09:13:24 2009 -0700 39.3 @@ -146,13 +146,17 @@ 39.4 uint64_t BufferBlob_vtbl; 39.5 uint64_t RuntimeStub_vtbl; 39.6 39.7 + uint64_t Use_Compressed_Oops_address; 39.8 uint64_t Universe_methodKlassObj_address; 39.9 + uint64_t Universe_narrow_oop_base_address; 39.10 + uint64_t Universe_narrow_oop_shift_address; 39.11 uint64_t CodeCache_heap_address; 39.12 - uint64_t Universe_heap_base_address; 39.13 39.14 /* Volatiles */ 39.15 + uint8_t Use_Compressed_Oops; 39.16 uint64_t Universe_methodKlassObj; 39.17 - uint64_t Universe_heap_base; 39.18 + uint64_t Universe_narrow_oop_base; 39.19 + uint32_t Universe_narrow_oop_shift; 39.20 uint64_t CodeCache_low; 39.21 uint64_t CodeCache_high; 39.22 uint64_t CodeCache_segmap_low; 39.23 @@ -279,8 +283,11 @@ 39.24 if (strcmp("_methodKlassObj", vmp->fieldName) == 0) { 39.25 J->Universe_methodKlassObj_address = vmp->address; 39.26 } 39.27 - if (strcmp("_heap_base", vmp->fieldName) == 0) { 39.28 - J->Universe_heap_base_address = vmp->address; 39.29 + if (strcmp("_narrow_oop._base", vmp->fieldName) == 0) { 39.30 + J->Universe_narrow_oop_base_address = vmp->address; 39.31 + } 39.32 + if (strcmp("_narrow_oop._shift", vmp->fieldName) == 0) { 39.33 + J->Universe_narrow_oop_shift_address = vmp->address; 39.34 } 39.35 } 39.36 CHECK_FAIL(err); 39.37 @@ -298,14 +305,39 @@ 39.38 return -1; 39.39 } 39.40 39.41 +static int find_symbol(jvm_agent_t* J, const char *name, uint64_t* valuep) { 39.42 + psaddr_t sym_addr; 39.43 + int err; 39.44 + 39.45 + err = ps_pglobal_lookup(J->P, LIBJVM_SO, name, &sym_addr); 39.46 + if (err != PS_OK) goto fail; 39.47 + *valuep = sym_addr; 39.48 + return PS_OK; 39.49 + 39.50 + fail: 39.51 + return err; 39.52 +} 39.53 + 39.54 static int read_volatiles(jvm_agent_t* J) { 39.55 uint64_t ptr; 39.56 int err; 39.57 39.58 + err = find_symbol(J, "UseCompressedOops", &J->Use_Compressed_Oops_address); 39.59 + if (err == PS_OK) { 39.60 + err = ps_pread(J->P, J->Use_Compressed_Oops_address, &J->Use_Compressed_Oops, sizeof(uint8_t)); 39.61 + CHECK_FAIL(err); 39.62 + } else { 39.63 + J->Use_Compressed_Oops = 0; 39.64 + } 39.65 + 39.66 err = read_pointer(J, J->Universe_methodKlassObj_address, &J->Universe_methodKlassObj); 39.67 CHECK_FAIL(err); 39.68 - err = read_pointer(J, J->Universe_heap_base_address, &J->Universe_heap_base); 39.69 + 39.70 + err = read_pointer(J, J->Universe_narrow_oop_base_address, &J->Universe_narrow_oop_base); 39.71 CHECK_FAIL(err); 39.72 + err = ps_pread(J->P, J->Universe_narrow_oop_shift_address, &J->Universe_narrow_oop_shift, sizeof(uint32_t)); 39.73 + CHECK_FAIL(err); 39.74 + 39.75 err = read_pointer(J, J->CodeCache_heap_address + OFFSET_CodeHeap_memory + 39.76 OFFSET_VirtualSpace_low, &J->CodeCache_low); 39.77 CHECK_FAIL(err); 39.78 @@ -374,19 +406,6 @@ 39.79 return -1; 39.80 } 39.81 39.82 -static int find_symbol(jvm_agent_t* J, const char *name, uint64_t* valuep) { 39.83 - psaddr_t sym_addr; 39.84 - int err; 39.85 - 39.86 - err = ps_pglobal_lookup(J->P, LIBJVM_SO, name, &sym_addr); 39.87 - if (err != PS_OK) goto fail; 39.88 - *valuep = sym_addr; 39.89 - return PS_OK; 39.90 - 39.91 - fail: 39.92 - return err; 39.93 -} 39.94 - 39.95 static int find_jlong_constant(jvm_agent_t* J, const char *name, uint64_t* valuep) { 39.96 psaddr_t sym_addr; 39.97 int err = ps_pglobal_lookup(J->P, LIBJVM_SO, name, &sym_addr); 39.98 @@ -458,14 +477,14 @@ 39.99 static int is_methodOop(jvm_agent_t* J, uint64_t methodOopPtr) { 39.100 uint64_t klass; 39.101 int err; 39.102 - // If heap_base is nonnull, this was a compressed oop. 39.103 - if (J->Universe_heap_base != NULL) { 39.104 + // If UseCompressedOops, this was a compressed oop. 39.105 + if (J->Use_Compressed_Oops != 0) { 39.106 uint32_t cklass; 39.107 err = read_compressed_pointer(J, methodOopPtr + OFFSET_oopDesc_metadata, 39.108 &cklass); 39.109 // decode heap oop, same as oop.inline.hpp 39.110 - klass = (uint64_t)((uintptr_t)J->Universe_heap_base + 39.111 - ((uintptr_t)cklass << 3)); 39.112 + klass = (uint64_t)((uintptr_t)J->Universe_narrow_oop_base + 39.113 + ((uintptr_t)cklass << J->Universe_narrow_oop_shift)); 39.114 } else { 39.115 err = read_pointer(J, methodOopPtr + OFFSET_oopDesc_metadata, &klass); 39.116 }
40.1 --- a/src/os/solaris/vm/os_solaris.cpp Wed Mar 18 11:37:48 2009 -0400 40.2 +++ b/src/os/solaris/vm/os_solaris.cpp Thu Mar 19 09:13:24 2009 -0700 40.3 @@ -3220,7 +3220,7 @@ 40.4 return true; 40.5 } 40.6 40.7 -char* os::reserve_memory_special(size_t bytes) { 40.8 +char* os::reserve_memory_special(size_t bytes, char* addr) { 40.9 assert(UseLargePages && UseISM, "only for ISM large pages"); 40.10 40.11 size_t size = bytes; 40.12 @@ -4451,6 +4451,9 @@ 40.13 int_fnP_thread_t os::Solaris::_thr_suspend_mutator; 40.14 int_fnP_thread_t os::Solaris::_thr_continue_mutator; 40.15 40.16 +// (Static) wrapper for getisax(2) call. 40.17 +os::Solaris::getisax_func_t os::Solaris::_getisax = 0; 40.18 + 40.19 // (Static) wrappers for the liblgrp API 40.20 os::Solaris::lgrp_home_func_t os::Solaris::_lgrp_home; 40.21 os::Solaris::lgrp_init_func_t os::Solaris::_lgrp_init; 40.22 @@ -4465,16 +4468,19 @@ 40.23 // (Static) wrapper for meminfo() call. 40.24 os::Solaris::meminfo_func_t os::Solaris::_meminfo = 0; 40.25 40.26 -static address resolve_symbol(const char *name) { 40.27 - address addr; 40.28 - 40.29 - addr = (address) dlsym(RTLD_DEFAULT, name); 40.30 +static address resolve_symbol_lazy(const char* name) { 40.31 + address addr = (address) dlsym(RTLD_DEFAULT, name); 40.32 if(addr == NULL) { 40.33 // RTLD_DEFAULT was not defined on some early versions of 2.5.1 40.34 addr = (address) dlsym(RTLD_NEXT, name); 40.35 - if(addr == NULL) { 40.36 - fatal(dlerror()); 40.37 - } 40.38 + } 40.39 + return addr; 40.40 +} 40.41 + 40.42 +static address resolve_symbol(const char* name) { 40.43 + address addr = resolve_symbol_lazy(name); 40.44 + if(addr == NULL) { 40.45 + fatal(dlerror()); 40.46 } 40.47 return addr; 40.48 } 40.49 @@ -4673,15 +4679,26 @@ 40.50 } 40.51 40.52 void os::Solaris::misc_sym_init() { 40.53 - address func = (address)dlsym(RTLD_DEFAULT, "meminfo"); 40.54 - if(func == NULL) { 40.55 - func = (address) dlsym(RTLD_NEXT, "meminfo"); 40.56 - } 40.57 + address func; 40.58 + 40.59 + // getisax 40.60 + func = resolve_symbol_lazy("getisax"); 40.61 + if (func != NULL) { 40.62 + os::Solaris::_getisax = CAST_TO_FN_PTR(getisax_func_t, func); 40.63 + } 40.64 + 40.65 + // meminfo 40.66 + func = resolve_symbol_lazy("meminfo"); 40.67 if (func != NULL) { 40.68 os::Solaris::set_meminfo(CAST_TO_FN_PTR(meminfo_func_t, func)); 40.69 } 40.70 } 40.71 40.72 +uint_t os::Solaris::getisax(uint32_t* array, uint_t n) { 40.73 + assert(_getisax != NULL, "_getisax not set"); 40.74 + return _getisax(array, n); 40.75 +} 40.76 + 40.77 // Symbol doesn't exist in Solaris 8 pset.h 40.78 #ifndef PS_MYID 40.79 #define PS_MYID -3 40.80 @@ -4716,6 +4733,10 @@ 40.81 40.82 Solaris::initialize_system_info(); 40.83 40.84 + // Initialize misc. symbols as soon as possible, so we can use them 40.85 + // if we need them. 40.86 + Solaris::misc_sym_init(); 40.87 + 40.88 int fd = open("/dev/zero", O_RDWR); 40.89 if (fd < 0) { 40.90 fatal1("os::init: cannot open /dev/zero (%s)", strerror(errno)); 40.91 @@ -4857,7 +4878,6 @@ 40.92 } 40.93 } 40.94 40.95 - Solaris::misc_sym_init(); 40.96 Solaris::signal_sets_init(); 40.97 Solaris::init_signal_mem(); 40.98 Solaris::install_signal_handlers();
41.1 --- a/src/os/solaris/vm/os_solaris.hpp Wed Mar 18 11:37:48 2009 -0400 41.2 +++ b/src/os/solaris/vm/os_solaris.hpp Thu Mar 19 09:13:24 2009 -0700 41.3 @@ -1,5 +1,5 @@ 41.4 /* 41.5 - * Copyright 1997-2008 Sun Microsystems, Inc. All Rights Reserved. 41.6 + * Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. 41.7 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 41.8 * 41.9 * This code is free software; you can redistribute it and/or modify it 41.10 @@ -72,6 +72,8 @@ 41.11 LGRP_VIEW_OS /* what's available to operating system */ 41.12 } lgrp_view_t; 41.13 41.14 + typedef uint_t (*getisax_func_t)(uint32_t* array, uint_t n); 41.15 + 41.16 typedef lgrp_id_t (*lgrp_home_func_t)(idtype_t idtype, id_t id); 41.17 typedef lgrp_cookie_t (*lgrp_init_func_t)(lgrp_view_t view); 41.18 typedef int (*lgrp_fini_func_t)(lgrp_cookie_t cookie); 41.19 @@ -87,6 +89,8 @@ 41.20 const uint_t info_req[], int info_count, 41.21 uint64_t outdata[], uint_t validity[]); 41.22 41.23 + static getisax_func_t _getisax; 41.24 + 41.25 static lgrp_home_func_t _lgrp_home; 41.26 static lgrp_init_func_t _lgrp_init; 41.27 static lgrp_fini_func_t _lgrp_fini; 41.28 @@ -283,6 +287,9 @@ 41.29 } 41.30 static lgrp_cookie_t lgrp_cookie() { return _lgrp_cookie; } 41.31 41.32 + static bool supports_getisax() { return _getisax != NULL; } 41.33 + static uint_t getisax(uint32_t* array, uint_t n); 41.34 + 41.35 static void set_meminfo(meminfo_func_t func) { _meminfo = func; } 41.36 static int meminfo (const uint64_t inaddr[], int addr_count, 41.37 const uint_t info_req[], int info_count,
42.1 --- a/src/os/windows/vm/os_windows.cpp Wed Mar 18 11:37:48 2009 -0400 42.2 +++ b/src/os/windows/vm/os_windows.cpp Thu Mar 19 09:13:24 2009 -0700 42.3 @@ -2595,7 +2595,7 @@ 42.4 return true; 42.5 } 42.6 42.7 -char* os::reserve_memory_special(size_t bytes) { 42.8 +char* os::reserve_memory_special(size_t bytes, char* addr) { 42.9 42.10 if (UseLargePagesIndividualAllocation) { 42.11 if (TracePageSizes && Verbose) { 42.12 @@ -2615,7 +2615,7 @@ 42.13 "use -XX:-UseLargePagesIndividualAllocation to turn off"); 42.14 return NULL; 42.15 } 42.16 - p_buf = (char *) VirtualAlloc(NULL, 42.17 + p_buf = (char *) VirtualAlloc(addr, 42.18 size_of_reserve, // size of Reserve 42.19 MEM_RESERVE, 42.20 PAGE_EXECUTE_READWRITE);
43.1 --- a/src/os_cpu/linux_sparc/vm/globals_linux_sparc.hpp Wed Mar 18 11:37:48 2009 -0400 43.2 +++ b/src/os_cpu/linux_sparc/vm/globals_linux_sparc.hpp Thu Mar 19 09:13:24 2009 -0700 43.3 @@ -30,5 +30,7 @@ 43.4 define_pd_global(uintx, JVMInvokeMethodSlack, 12288); 43.5 define_pd_global(intx, CompilerThreadStackSize, 0); 43.6 43.7 +// Only used on 64 bit platforms 43.8 +define_pd_global(uintx, HeapBaseMinAddress, 4*G); 43.9 // Only used on 64 bit Windows platforms 43.10 define_pd_global(bool, UseVectoredExceptions, false);
44.1 --- a/src/os_cpu/linux_x86/vm/globals_linux_x86.hpp Wed Mar 18 11:37:48 2009 -0400 44.2 +++ b/src/os_cpu/linux_x86/vm/globals_linux_x86.hpp Thu Mar 19 09:13:24 2009 -0700 44.3 @@ -43,5 +43,7 @@ 44.4 44.5 define_pd_global(uintx, JVMInvokeMethodSlack, 8192); 44.6 44.7 +// Only used on 64 bit platforms 44.8 +define_pd_global(uintx, HeapBaseMinAddress, 2*G); 44.9 // Only used on 64 bit Windows platforms 44.10 define_pd_global(bool, UseVectoredExceptions, false);
45.1 --- a/src/os_cpu/solaris_sparc/vm/globals_solaris_sparc.hpp Wed Mar 18 11:37:48 2009 -0400 45.2 +++ b/src/os_cpu/solaris_sparc/vm/globals_solaris_sparc.hpp Thu Mar 19 09:13:24 2009 -0700 45.3 @@ -30,5 +30,9 @@ 45.4 define_pd_global(uintx, JVMInvokeMethodSlack, 12288); 45.5 define_pd_global(intx, CompilerThreadStackSize, 0); 45.6 45.7 +// Only used on 64 bit platforms 45.8 +define_pd_global(uintx, HeapBaseMinAddress, 4*G); 45.9 // Only used on 64 bit Windows platforms 45.10 define_pd_global(bool, UseVectoredExceptions, false); 45.11 + 45.12 +
46.1 --- a/src/os_cpu/solaris_sparc/vm/vm_version_solaris_sparc.cpp Wed Mar 18 11:37:48 2009 -0400 46.2 +++ b/src/os_cpu/solaris_sparc/vm/vm_version_solaris_sparc.cpp Thu Mar 19 09:13:24 2009 -0700 46.3 @@ -1,5 +1,5 @@ 46.4 /* 46.5 - * Copyright 2006 Sun Microsystems, Inc. All Rights Reserved. 46.6 + * Copyright 2006-2009 Sun Microsystems, Inc. All Rights Reserved. 46.7 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 46.8 * 46.9 * This code is free software; you can redistribute it and/or modify it 46.10 @@ -25,58 +25,107 @@ 46.11 # include "incls/_precompiled.incl" 46.12 # include "incls/_vm_version_solaris_sparc.cpp.incl" 46.13 46.14 +# include <sys/auxv.h> 46.15 +# include <sys/auxv_SPARC.h> 46.16 # include <sys/systeminfo.h> 46.17 46.18 +// We need to keep these here as long as we have to build on Solaris 46.19 +// versions before 10. 46.20 +#ifndef SI_ARCHITECTURE_32 46.21 +#define SI_ARCHITECTURE_32 516 /* basic 32-bit SI_ARCHITECTURE */ 46.22 +#endif 46.23 + 46.24 +#ifndef SI_ARCHITECTURE_64 46.25 +#define SI_ARCHITECTURE_64 517 /* basic 64-bit SI_ARCHITECTURE */ 46.26 +#endif 46.27 + 46.28 +static void do_sysinfo(int si, const char* string, int* features, int mask) { 46.29 + char tmp; 46.30 + size_t bufsize = sysinfo(si, &tmp, 1); 46.31 + 46.32 + // All SI defines used below must be supported. 46.33 + guarantee(bufsize != -1, "must be supported"); 46.34 + 46.35 + char* buf = (char*) malloc(bufsize); 46.36 + 46.37 + if (buf == NULL) 46.38 + return; 46.39 + 46.40 + if (sysinfo(si, buf, bufsize) == bufsize) { 46.41 + // Compare the string. 46.42 + if (strcmp(buf, string) == 0) { 46.43 + *features |= mask; 46.44 + } 46.45 + } 46.46 + 46.47 + free(buf); 46.48 +} 46.49 + 46.50 int VM_Version::platform_features(int features) { 46.51 - // We determine what sort of hardware we have via sysinfo(SI_ISALIST, ...). 46.52 - // This isn't the best of all possible ways because there's not enough 46.53 - // detail in the isa list it returns, but it's a bit less arcane than 46.54 - // generating assembly code and an illegal instruction handler. We used 46.55 - // to generate a getpsr trap, but that's even more arcane. 46.56 - // 46.57 - // Another possibility would be to use sysinfo(SI_PLATFORM, ...), but 46.58 - // that would require more knowledge here than is wise. 46.59 + // getisax(2), SI_ARCHITECTURE_32, and SI_ARCHITECTURE_64 are 46.60 + // supported on Solaris 10 and later. 46.61 + if (os::Solaris::supports_getisax()) { 46.62 +#ifndef PRODUCT 46.63 + if (PrintMiscellaneous && Verbose) 46.64 + tty->print_cr("getisax(2) supported."); 46.65 +#endif 46.66 46.67 - // isalist spec via 'man isalist' as of 01-Aug-2001 46.68 + // Check 32-bit architecture. 46.69 + do_sysinfo(SI_ARCHITECTURE_32, "sparc", &features, v8_instructions_m); 46.70 46.71 - char tmp; 46.72 - size_t bufsize = sysinfo(SI_ISALIST, &tmp, 1); 46.73 - char* buf = (char*)malloc(bufsize); 46.74 + // Check 64-bit architecture. 46.75 + do_sysinfo(SI_ARCHITECTURE_64, "sparcv9", &features, generic_v9_m); 46.76 46.77 - if (buf != NULL) { 46.78 - if (sysinfo(SI_ISALIST, buf, bufsize) == bufsize) { 46.79 - // Figure out what kind of sparc we have 46.80 - char *sparc_string = strstr(buf, "sparc"); 46.81 - if (sparc_string != NULL) { features |= v8_instructions_m; 46.82 - if (sparc_string[5] == 'v') { 46.83 - if (sparc_string[6] == '8') { 46.84 - if (sparc_string[7] == '-') features |= hardware_int_muldiv_m; 46.85 - else if (sparc_string[7] == 'p') features |= generic_v9_m; 46.86 - else features |= generic_v8_m; 46.87 - } else if (sparc_string[6] == '9') features |= generic_v9_m; 46.88 + // Extract valid instruction set extensions. 46.89 + uint_t av; 46.90 + uint_t avn = os::Solaris::getisax(&av, 1); 46.91 + assert(avn == 1, "should only return one av"); 46.92 + 46.93 + if (av & AV_SPARC_MUL32) features |= hardware_mul32_m; 46.94 + if (av & AV_SPARC_DIV32) features |= hardware_div32_m; 46.95 + if (av & AV_SPARC_FSMULD) features |= hardware_fsmuld_m; 46.96 + if (av & AV_SPARC_V8PLUS) features |= v9_instructions_m; 46.97 + if (av & AV_SPARC_POPC) features |= hardware_popc_m; 46.98 + if (av & AV_SPARC_VIS) features |= vis1_instructions_m; 46.99 + if (av & AV_SPARC_VIS2) features |= vis2_instructions_m; 46.100 + } else { 46.101 + // getisax(2) failed, use the old legacy code. 46.102 +#ifndef PRODUCT 46.103 + if (PrintMiscellaneous && Verbose) 46.104 + tty->print_cr("getisax(2) not supported."); 46.105 +#endif 46.106 + 46.107 + char tmp; 46.108 + size_t bufsize = sysinfo(SI_ISALIST, &tmp, 1); 46.109 + char* buf = (char*) malloc(bufsize); 46.110 + 46.111 + if (buf != NULL) { 46.112 + if (sysinfo(SI_ISALIST, buf, bufsize) == bufsize) { 46.113 + // Figure out what kind of sparc we have 46.114 + char *sparc_string = strstr(buf, "sparc"); 46.115 + if (sparc_string != NULL) { features |= v8_instructions_m; 46.116 + if (sparc_string[5] == 'v') { 46.117 + if (sparc_string[6] == '8') { 46.118 + if (sparc_string[7] == '-') { features |= hardware_mul32_m; 46.119 + features |= hardware_div32_m; 46.120 + } else if (sparc_string[7] == 'p') features |= generic_v9_m; 46.121 + else features |= generic_v8_m; 46.122 + } else if (sparc_string[6] == '9') features |= generic_v9_m; 46.123 + } 46.124 + } 46.125 + 46.126 + // Check for visualization instructions 46.127 + char *vis = strstr(buf, "vis"); 46.128 + if (vis != NULL) { features |= vis1_instructions_m; 46.129 + if (vis[3] == '2') features |= vis2_instructions_m; 46.130 } 46.131 } 46.132 - 46.133 - // Check for visualization instructions 46.134 - char *vis = strstr(buf, "vis"); 46.135 - if (vis != NULL) { features |= vis1_instructions_m; 46.136 - if (vis[3] == '2') features |= vis2_instructions_m; 46.137 - } 46.138 + free(buf); 46.139 } 46.140 - free(buf); 46.141 } 46.142 46.143 - bufsize = sysinfo(SI_MACHINE, &tmp, 1); 46.144 - buf = (char*)malloc(bufsize); 46.145 - 46.146 - if (buf != NULL) { 46.147 - if (sysinfo(SI_MACHINE, buf, bufsize) == bufsize) { 46.148 - if (strstr(buf, "sun4v") != NULL) { 46.149 - features |= sun4v_m; 46.150 - } 46.151 - } 46.152 - free(buf); 46.153 - } 46.154 + // Determine the machine type. 46.155 + do_sysinfo(SI_MACHINE, "sun4v", &features, sun4v_m); 46.156 46.157 return features; 46.158 }
47.1 --- a/src/os_cpu/solaris_x86/vm/globals_solaris_x86.hpp Wed Mar 18 11:37:48 2009 -0400 47.2 +++ b/src/os_cpu/solaris_x86/vm/globals_solaris_x86.hpp Thu Mar 19 09:13:24 2009 -0700 47.3 @@ -46,5 +46,7 @@ 47.4 47.5 define_pd_global(intx, CompilerThreadStackSize, 0); 47.6 47.7 +// Only used on 64 bit platforms 47.8 +define_pd_global(uintx, HeapBaseMinAddress, 256*M); 47.9 // Only used on 64 bit Windows platforms 47.10 define_pd_global(bool, UseVectoredExceptions, false);
48.1 --- a/src/os_cpu/windows_x86/vm/globals_windows_x86.hpp Wed Mar 18 11:37:48 2009 -0400 48.2 +++ b/src/os_cpu/windows_x86/vm/globals_windows_x86.hpp Thu Mar 19 09:13:24 2009 -0700 48.3 @@ -45,5 +45,7 @@ 48.4 48.5 define_pd_global(uintx, JVMInvokeMethodSlack, 8192); 48.6 48.7 +// Only used on 64 bit platforms 48.8 +define_pd_global(uintx, HeapBaseMinAddress, 2*G); 48.9 // Only used on 64 bit Windows platforms 48.10 define_pd_global(bool, UseVectoredExceptions, false);
49.1 --- a/src/os_cpu/windows_x86/vm/unwind_windows_x86.hpp Wed Mar 18 11:37:48 2009 -0400 49.2 +++ b/src/os_cpu/windows_x86/vm/unwind_windows_x86.hpp Thu Mar 19 09:13:24 2009 -0700 49.3 @@ -68,6 +68,9 @@ 49.4 PVOID HandlerData; 49.5 } DISPATCHER_CONTEXT, *PDISPATCHER_CONTEXT; 49.6 49.7 +#if MSC_VER < 1500 49.8 + 49.9 +/* Not needed for VS2008 compiler, comes from winnt.h. */ 49.10 typedef EXCEPTION_DISPOSITION (*PEXCEPTION_ROUTINE) ( 49.11 IN PEXCEPTION_RECORD ExceptionRecord, 49.12 IN ULONG64 EstablisherFrame, 49.13 @@ -75,4 +78,6 @@ 49.14 IN OUT PDISPATCHER_CONTEXT DispatcherContext 49.15 ); 49.16 49.17 +#endif 49.18 + 49.19 #endif // AMD64
50.1 --- a/src/share/vm/adlc/adlc.hpp Wed Mar 18 11:37:48 2009 -0400 50.2 +++ b/src/share/vm/adlc/adlc.hpp Thu Mar 19 09:13:24 2009 -0700 50.3 @@ -44,7 +44,7 @@ 50.4 #error "Something is wrong with the detection of MSC_VER in the makefiles" 50.5 #endif 50.6 50.7 -#if _MSC_VER >= 1400 && !defined(_WIN64) 50.8 +#if _MSC_VER >= 1400 50.9 #define strdup _strdup 50.10 #endif 50.11
51.1 --- a/src/share/vm/asm/assembler.cpp Wed Mar 18 11:37:48 2009 -0400 51.2 +++ b/src/share/vm/asm/assembler.cpp Thu Mar 19 09:13:24 2009 -0700 51.3 @@ -321,16 +321,19 @@ 51.4 bool MacroAssembler::needs_explicit_null_check(intptr_t offset) { 51.5 // Exception handler checks the nmethod's implicit null checks table 51.6 // only when this method returns false. 51.7 - if (UseCompressedOops) { 51.8 +#ifdef _LP64 51.9 + if (UseCompressedOops && Universe::narrow_oop_base() != NULL) { 51.10 + assert (Universe::heap() != NULL, "java heap should be initialized"); 51.11 // The first page after heap_base is unmapped and 51.12 // the 'offset' is equal to [heap_base + offset] for 51.13 // narrow oop implicit null checks. 51.14 - uintptr_t heap_base = (uintptr_t)Universe::heap_base(); 51.15 - if ((uintptr_t)offset >= heap_base) { 51.16 + uintptr_t base = (uintptr_t)Universe::narrow_oop_base(); 51.17 + if ((uintptr_t)offset >= base) { 51.18 // Normalize offset for the next check. 51.19 - offset = (intptr_t)(pointer_delta((void*)offset, (void*)heap_base, 1)); 51.20 + offset = (intptr_t)(pointer_delta((void*)offset, (void*)base, 1)); 51.21 } 51.22 } 51.23 +#endif 51.24 return offset < 0 || os::vm_page_size() <= offset; 51.25 } 51.26
52.1 --- a/src/share/vm/ci/ciMethodBlocks.cpp Wed Mar 18 11:37:48 2009 -0400 52.2 +++ b/src/share/vm/ci/ciMethodBlocks.cpp Thu Mar 19 09:13:24 2009 -0700 52.3 @@ -284,6 +284,11 @@ 52.4 // 52.5 int ex_start = handler->start(); 52.6 int ex_end = handler->limit(); 52.7 + // ensure a block at the start of exception range and start of following code 52.8 + (void) make_block_at(ex_start); 52.9 + if (ex_end < _code_size) 52.10 + (void) make_block_at(ex_end); 52.11 + 52.12 if (eb->is_handler()) { 52.13 // Extend old handler exception range to cover additional range. 52.14 int old_ex_start = eb->ex_start_bci(); 52.15 @@ -295,10 +300,6 @@ 52.16 eb->clear_exception_handler(); // Reset exception information 52.17 } 52.18 eb->set_exception_range(ex_start, ex_end); 52.19 - // ensure a block at the start of exception range and start of following code 52.20 - (void) make_block_at(ex_start); 52.21 - if (ex_end < _code_size) 52.22 - (void) make_block_at(ex_end); 52.23 } 52.24 } 52.25
53.1 --- a/src/share/vm/classfile/vmSymbols.hpp Wed Mar 18 11:37:48 2009 -0400 53.2 +++ b/src/share/vm/classfile/vmSymbols.hpp Thu Mar 19 09:13:24 2009 -0700 53.3 @@ -284,6 +284,7 @@ 53.4 template(value_name, "value") \ 53.5 template(frontCacheEnabled_name, "frontCacheEnabled") \ 53.6 template(stringCacheEnabled_name, "stringCacheEnabled") \ 53.7 + template(bitCount_name, "bitCount") \ 53.8 \ 53.9 /* non-intrinsic name/signature pairs: */ \ 53.10 template(register_method_name, "register") \ 53.11 @@ -304,6 +305,7 @@ 53.12 template(double_long_signature, "(D)J") \ 53.13 template(double_double_signature, "(D)D") \ 53.14 template(int_float_signature, "(I)F") \ 53.15 + template(long_int_signature, "(J)I") \ 53.16 template(long_long_signature, "(J)J") \ 53.17 template(long_double_signature, "(J)D") \ 53.18 template(byte_signature, "B") \ 53.19 @@ -507,6 +509,10 @@ 53.20 do_name( doubleToLongBits_name, "doubleToLongBits") \ 53.21 do_intrinsic(_longBitsToDouble, java_lang_Double, longBitsToDouble_name, long_double_signature, F_S) \ 53.22 do_name( longBitsToDouble_name, "longBitsToDouble") \ 53.23 + \ 53.24 + do_intrinsic(_bitCount_i, java_lang_Integer, bitCount_name, int_int_signature, F_S) \ 53.25 + do_intrinsic(_bitCount_l, java_lang_Long, bitCount_name, long_int_signature, F_S) \ 53.26 + \ 53.27 do_intrinsic(_reverseBytes_i, java_lang_Integer, reverseBytes_name, int_int_signature, F_S) \ 53.28 do_name( reverseBytes_name, "reverseBytes") \ 53.29 do_intrinsic(_reverseBytes_l, java_lang_Long, reverseBytes_name, long_long_signature, F_S) \ 53.30 @@ -696,7 +702,6 @@ 53.31 do_signature(putShort_raw_signature, "(JS)V") \ 53.32 do_signature(getChar_raw_signature, "(J)C") \ 53.33 do_signature(putChar_raw_signature, "(JC)V") \ 53.34 - do_signature(getInt_raw_signature, "(J)I") \ 53.35 do_signature(putInt_raw_signature, "(JI)V") \ 53.36 do_alias(getLong_raw_signature, /*(J)J*/ long_long_signature) \ 53.37 do_alias(putLong_raw_signature, /*(JJ)V*/ long_long_void_signature) \ 53.38 @@ -713,7 +718,7 @@ 53.39 do_intrinsic(_getByte_raw, sun_misc_Unsafe, getByte_name, getByte_raw_signature, F_RN) \ 53.40 do_intrinsic(_getShort_raw, sun_misc_Unsafe, getShort_name, getShort_raw_signature, F_RN) \ 53.41 do_intrinsic(_getChar_raw, sun_misc_Unsafe, getChar_name, getChar_raw_signature, F_RN) \ 53.42 - do_intrinsic(_getInt_raw, sun_misc_Unsafe, getInt_name, getInt_raw_signature, F_RN) \ 53.43 + do_intrinsic(_getInt_raw, sun_misc_Unsafe, getInt_name, long_int_signature, F_RN) \ 53.44 do_intrinsic(_getLong_raw, sun_misc_Unsafe, getLong_name, getLong_raw_signature, F_RN) \ 53.45 do_intrinsic(_getFloat_raw, sun_misc_Unsafe, getFloat_name, getFloat_raw_signature, F_RN) \ 53.46 do_intrinsic(_getDouble_raw, sun_misc_Unsafe, getDouble_name, getDouble_raw_signature, F_RN) \
54.1 --- a/src/share/vm/gc_implementation/g1/concurrentMark.cpp Wed Mar 18 11:37:48 2009 -0400 54.2 +++ b/src/share/vm/gc_implementation/g1/concurrentMark.cpp Thu Mar 19 09:13:24 2009 -0700 54.3 @@ -107,7 +107,7 @@ 54.4 #ifndef PRODUCT 54.5 bool CMBitMapRO::covers(ReservedSpace rs) const { 54.6 // assert(_bm.map() == _virtual_space.low(), "map inconsistency"); 54.7 - assert(((size_t)_bm.size() * (1 << _shifter)) == _bmWordSize, 54.8 + assert(((size_t)_bm.size() * (size_t)(1 << _shifter)) == _bmWordSize, 54.9 "size inconsistency"); 54.10 return _bmStartWord == (HeapWord*)(rs.base()) && 54.11 _bmWordSize == rs.size()>>LogHeapWordSize;
55.1 --- a/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp Wed Mar 18 11:37:48 2009 -0400 55.2 +++ b/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp Thu Mar 19 09:13:24 2009 -0700 55.3 @@ -1422,9 +1422,34 @@ 55.4 // Reserve the maximum. 55.5 PermanentGenerationSpec* pgs = collector_policy()->permanent_generation(); 55.6 // Includes the perm-gen. 55.7 + 55.8 + const size_t total_reserved = max_byte_size + pgs->max_size(); 55.9 + char* addr = Universe::preferred_heap_base(total_reserved, Universe::UnscaledNarrowOop); 55.10 + 55.11 ReservedSpace heap_rs(max_byte_size + pgs->max_size(), 55.12 HeapRegion::GrainBytes, 55.13 - false /*ism*/); 55.14 + false /*ism*/, addr); 55.15 + 55.16 + if (UseCompressedOops) { 55.17 + if (addr != NULL && !heap_rs.is_reserved()) { 55.18 + // Failed to reserve at specified address - the requested memory 55.19 + // region is taken already, for example, by 'java' launcher. 55.20 + // Try again to reserver heap higher. 55.21 + addr = Universe::preferred_heap_base(total_reserved, Universe::ZeroBasedNarrowOop); 55.22 + ReservedSpace heap_rs0(total_reserved, HeapRegion::GrainBytes, 55.23 + false /*ism*/, addr); 55.24 + if (addr != NULL && !heap_rs0.is_reserved()) { 55.25 + // Failed to reserve at specified address again - give up. 55.26 + addr = Universe::preferred_heap_base(total_reserved, Universe::HeapBasedNarrowOop); 55.27 + assert(addr == NULL, ""); 55.28 + ReservedSpace heap_rs1(total_reserved, HeapRegion::GrainBytes, 55.29 + false /*ism*/, addr); 55.30 + heap_rs = heap_rs1; 55.31 + } else { 55.32 + heap_rs = heap_rs0; 55.33 + } 55.34 + } 55.35 + } 55.36 55.37 if (!heap_rs.is_reserved()) { 55.38 vm_exit_during_initialization("Could not reserve enough space for object heap");
56.1 --- a/src/share/vm/gc_implementation/g1/heapRegionRemSet.cpp Wed Mar 18 11:37:48 2009 -0400 56.2 +++ b/src/share/vm/gc_implementation/g1/heapRegionRemSet.cpp Thu Mar 19 09:13:24 2009 -0700 56.3 @@ -508,7 +508,7 @@ 56.4 typedef PosParPRT* PosParPRTPtr; 56.5 if (_max_fine_entries == 0) { 56.6 assert(_mod_max_fine_entries_mask == 0, "Both or none."); 56.7 - _max_fine_entries = (1 << G1LogRSRegionEntries); 56.8 + _max_fine_entries = (size_t)(1 << G1LogRSRegionEntries); 56.9 _mod_max_fine_entries_mask = _max_fine_entries - 1; 56.10 #if SAMPLE_FOR_EVICTION 56.11 assert(_fine_eviction_sample_size == 0
57.1 --- a/src/share/vm/gc_implementation/parallelScavenge/parallelScavengeHeap.cpp Wed Mar 18 11:37:48 2009 -0400 57.2 +++ b/src/share/vm/gc_implementation/parallelScavenge/parallelScavengeHeap.cpp Thu Mar 19 09:13:24 2009 -0700 57.3 @@ -104,12 +104,38 @@ 57.4 og_min_size, og_max_size, 57.5 yg_min_size, yg_max_size); 57.6 57.7 + const size_t total_reserved = pg_max_size + og_max_size + yg_max_size; 57.8 + char* addr = Universe::preferred_heap_base(total_reserved, Universe::UnscaledNarrowOop); 57.9 + 57.10 // The main part of the heap (old gen + young gen) can often use a larger page 57.11 // size than is needed or wanted for the perm gen. Use the "compound 57.12 // alignment" ReservedSpace ctor to avoid having to use the same page size for 57.13 // all gens. 57.14 + 57.15 ReservedHeapSpace heap_rs(pg_max_size, pg_align, og_max_size + yg_max_size, 57.16 - og_align); 57.17 + og_align, addr); 57.18 + 57.19 + if (UseCompressedOops) { 57.20 + if (addr != NULL && !heap_rs.is_reserved()) { 57.21 + // Failed to reserve at specified address - the requested memory 57.22 + // region is taken already, for example, by 'java' launcher. 57.23 + // Try again to reserver heap higher. 57.24 + addr = Universe::preferred_heap_base(total_reserved, Universe::ZeroBasedNarrowOop); 57.25 + ReservedHeapSpace heap_rs0(pg_max_size, pg_align, og_max_size + yg_max_size, 57.26 + og_align, addr); 57.27 + if (addr != NULL && !heap_rs0.is_reserved()) { 57.28 + // Failed to reserve at specified address again - give up. 57.29 + addr = Universe::preferred_heap_base(total_reserved, Universe::HeapBasedNarrowOop); 57.30 + assert(addr == NULL, ""); 57.31 + ReservedHeapSpace heap_rs1(pg_max_size, pg_align, og_max_size + yg_max_size, 57.32 + og_align, addr); 57.33 + heap_rs = heap_rs1; 57.34 + } else { 57.35 + heap_rs = heap_rs0; 57.36 + } 57.37 + } 57.38 + } 57.39 + 57.40 os::trace_page_sizes("ps perm", pg_min_size, pg_max_size, pg_page_sz, 57.41 heap_rs.base(), pg_max_size); 57.42 os::trace_page_sizes("ps main", og_min_size + yg_min_size,
58.1 --- a/src/share/vm/includeDB_core Wed Mar 18 11:37:48 2009 -0400 58.2 +++ b/src/share/vm/includeDB_core Thu Mar 19 09:13:24 2009 -0700 58.3 @@ -4598,6 +4598,7 @@ 58.4 vm_version_<arch>.hpp globals_extension.hpp 58.5 vm_version_<arch>.hpp vm_version.hpp 58.6 58.7 +vm_version_<os_arch>.cpp os.hpp 58.8 vm_version_<os_arch>.cpp vm_version_<arch>.hpp 58.9 58.10 vmreg.cpp assembler.hpp
59.1 --- a/src/share/vm/memory/blockOffsetTable.hpp Wed Mar 18 11:37:48 2009 -0400 59.2 +++ b/src/share/vm/memory/blockOffsetTable.hpp Thu Mar 19 09:13:24 2009 -0700 59.3 @@ -235,7 +235,7 @@ 59.4 }; 59.5 59.6 static size_t power_to_cards_back(uint i) { 59.7 - return 1 << (LogBase * i); 59.8 + return (size_t)(1 << (LogBase * i)); 59.9 } 59.10 static size_t power_to_words_back(uint i) { 59.11 return power_to_cards_back(i) * N_words;
60.1 --- a/src/share/vm/memory/genCollectedHeap.cpp Wed Mar 18 11:37:48 2009 -0400 60.2 +++ b/src/share/vm/memory/genCollectedHeap.cpp Thu Mar 19 09:13:24 2009 -0700 60.3 @@ -218,6 +218,31 @@ 60.4 heap_address -= total_reserved; 60.5 } else { 60.6 heap_address = NULL; // any address will do. 60.7 + if (UseCompressedOops) { 60.8 + heap_address = Universe::preferred_heap_base(total_reserved, Universe::UnscaledNarrowOop); 60.9 + *_total_reserved = total_reserved; 60.10 + *_n_covered_regions = n_covered_regions; 60.11 + *heap_rs = ReservedHeapSpace(total_reserved, alignment, 60.12 + UseLargePages, heap_address); 60.13 + 60.14 + if (heap_address != NULL && !heap_rs->is_reserved()) { 60.15 + // Failed to reserve at specified address - the requested memory 60.16 + // region is taken already, for example, by 'java' launcher. 60.17 + // Try again to reserver heap higher. 60.18 + heap_address = Universe::preferred_heap_base(total_reserved, Universe::ZeroBasedNarrowOop); 60.19 + *heap_rs = ReservedHeapSpace(total_reserved, alignment, 60.20 + UseLargePages, heap_address); 60.21 + 60.22 + if (heap_address != NULL && !heap_rs->is_reserved()) { 60.23 + // Failed to reserve at specified address again - give up. 60.24 + heap_address = Universe::preferred_heap_base(total_reserved, Universe::HeapBasedNarrowOop); 60.25 + assert(heap_address == NULL, ""); 60.26 + *heap_rs = ReservedHeapSpace(total_reserved, alignment, 60.27 + UseLargePages, heap_address); 60.28 + } 60.29 + } 60.30 + return heap_address; 60.31 + } 60.32 } 60.33 60.34 *_total_reserved = total_reserved;
61.1 --- a/src/share/vm/memory/universe.cpp Wed Mar 18 11:37:48 2009 -0400 61.2 +++ b/src/share/vm/memory/universe.cpp Thu Mar 19 09:13:24 2009 -0700 61.3 @@ -99,7 +99,8 @@ 61.4 size_t Universe::_heap_used_at_last_gc = 0; 61.5 61.6 CollectedHeap* Universe::_collectedHeap = NULL; 61.7 -address Universe::_heap_base = NULL; 61.8 + 61.9 +NarrowOopStruct Universe::_narrow_oop = { NULL, 0, true }; 61.10 61.11 61.12 void Universe::basic_type_classes_do(void f(klassOop)) { 61.13 @@ -729,6 +730,53 @@ 61.14 return JNI_OK; 61.15 } 61.16 61.17 +// Choose the heap base address and oop encoding mode 61.18 +// when compressed oops are used: 61.19 +// Unscaled - Use 32-bits oops without encoding when 61.20 +// NarrowOopHeapBaseMin + heap_size < 4Gb 61.21 +// ZeroBased - Use zero based compressed oops with encoding when 61.22 +// NarrowOopHeapBaseMin + heap_size < 32Gb 61.23 +// HeapBased - Use compressed oops with heap base + encoding. 61.24 + 61.25 +// 4Gb 61.26 +static const uint64_t NarrowOopHeapMax = (uint64_t(max_juint) + 1); 61.27 +// 32Gb 61.28 +static const uint64_t OopEncodingHeapMax = NarrowOopHeapMax << LogMinObjAlignmentInBytes; 61.29 + 61.30 +char* Universe::preferred_heap_base(size_t heap_size, NARROW_OOP_MODE mode) { 61.31 +#ifdef _LP64 61.32 + if (UseCompressedOops) { 61.33 + assert(mode == UnscaledNarrowOop || 61.34 + mode == ZeroBasedNarrowOop || 61.35 + mode == HeapBasedNarrowOop, "mode is invalid"); 61.36 + 61.37 + const size_t total_size = heap_size + HeapBaseMinAddress; 61.38 + if (total_size <= OopEncodingHeapMax && (mode != HeapBasedNarrowOop)) { 61.39 + if (total_size <= NarrowOopHeapMax && (mode == UnscaledNarrowOop) && 61.40 + (Universe::narrow_oop_shift() == 0)) { 61.41 + // Use 32-bits oops without encoding and 61.42 + // place heap's top on the 4Gb boundary 61.43 + return (char*)(NarrowOopHeapMax - heap_size); 61.44 + } else { 61.45 + // Can't reserve with NarrowOopShift == 0 61.46 + Universe::set_narrow_oop_shift(LogMinObjAlignmentInBytes); 61.47 + if (mode == UnscaledNarrowOop || 61.48 + mode == ZeroBasedNarrowOop && total_size <= NarrowOopHeapMax) { 61.49 + // Use zero based compressed oops with encoding and 61.50 + // place heap's top on the 32Gb boundary in case 61.51 + // total_size > 4Gb or failed to reserve below 4Gb. 61.52 + return (char*)(OopEncodingHeapMax - heap_size); 61.53 + } 61.54 + } 61.55 + } else { 61.56 + // Can't reserve below 32Gb. 61.57 + Universe::set_narrow_oop_shift(LogMinObjAlignmentInBytes); 61.58 + } 61.59 + } 61.60 +#endif 61.61 + return NULL; // also return NULL (don't care) for 32-bit VM 61.62 +} 61.63 + 61.64 jint Universe::initialize_heap() { 61.65 61.66 if (UseParallelGC) { 61.67 @@ -773,6 +821,8 @@ 61.68 if (status != JNI_OK) { 61.69 return status; 61.70 } 61.71 + 61.72 +#ifdef _LP64 61.73 if (UseCompressedOops) { 61.74 // Subtract a page because something can get allocated at heap base. 61.75 // This also makes implicit null checking work, because the 61.76 @@ -780,8 +830,49 @@ 61.77 // See needs_explicit_null_check. 61.78 // Only set the heap base for compressed oops because it indicates 61.79 // compressed oops for pstack code. 61.80 - Universe::_heap_base = Universe::heap()->base() - os::vm_page_size(); 61.81 + if (PrintCompressedOopsMode) { 61.82 + tty->cr(); 61.83 + tty->print("heap address: "PTR_FORMAT, Universe::heap()->base()); 61.84 + } 61.85 + if ((uint64_t)Universe::heap()->reserved_region().end() > OopEncodingHeapMax) { 61.86 + // Can't reserve heap below 32Gb. 61.87 + Universe::set_narrow_oop_base(Universe::heap()->base() - os::vm_page_size()); 61.88 + Universe::set_narrow_oop_shift(LogMinObjAlignmentInBytes); 61.89 + if (PrintCompressedOopsMode) { 61.90 + tty->print(", Compressed Oops with base: "PTR_FORMAT, Universe::narrow_oop_base()); 61.91 + } 61.92 + } else { 61.93 + Universe::set_narrow_oop_base(0); 61.94 + if (PrintCompressedOopsMode) { 61.95 + tty->print(", zero based Compressed Oops"); 61.96 + } 61.97 +#ifdef _WIN64 61.98 + if (!Universe::narrow_oop_use_implicit_null_checks()) { 61.99 + // Don't need guard page for implicit checks in indexed addressing 61.100 + // mode with zero based Compressed Oops. 61.101 + Universe::set_narrow_oop_use_implicit_null_checks(true); 61.102 + } 61.103 +#endif // _WIN64 61.104 + if((uint64_t)Universe::heap()->reserved_region().end() > NarrowOopHeapMax) { 61.105 + // Can't reserve heap below 4Gb. 61.106 + Universe::set_narrow_oop_shift(LogMinObjAlignmentInBytes); 61.107 + } else { 61.108 + assert(Universe::narrow_oop_shift() == 0, "use unscaled narrow oop"); 61.109 + if (PrintCompressedOopsMode) { 61.110 + tty->print(", 32-bits Oops"); 61.111 + } 61.112 + } 61.113 + } 61.114 + if (PrintCompressedOopsMode) { 61.115 + tty->cr(); 61.116 + tty->cr(); 61.117 + } 61.118 } 61.119 + assert(Universe::narrow_oop_base() == (Universe::heap()->base() - os::vm_page_size()) || 61.120 + Universe::narrow_oop_base() == NULL, "invalid value"); 61.121 + assert(Universe::narrow_oop_shift() == LogMinObjAlignmentInBytes || 61.122 + Universe::narrow_oop_shift() == 0, "invalid value"); 61.123 +#endif 61.124 61.125 // We will never reach the CATCH below since Exceptions::_throw will cause 61.126 // the VM to exit if an exception is thrown during initialization
62.1 --- a/src/share/vm/memory/universe.hpp Wed Mar 18 11:37:48 2009 -0400 62.2 +++ b/src/share/vm/memory/universe.hpp Thu Mar 19 09:13:24 2009 -0700 62.3 @@ -90,6 +90,19 @@ 62.4 methodOop get_methodOop(); 62.5 }; 62.6 62.7 +// For UseCompressedOops. 62.8 +struct NarrowOopStruct { 62.9 + // Base address for oop-within-java-object materialization. 62.10 + // NULL if using wide oops or zero based narrow oops. 62.11 + address _base; 62.12 + // Number of shift bits for encoding/decoding narrow oops. 62.13 + // 0 if using wide oops or zero based unscaled narrow oops, 62.14 + // LogMinObjAlignmentInBytes otherwise. 62.15 + int _shift; 62.16 + // Generate code with implicit null checks for narrow oops. 62.17 + bool _use_implicit_null_checks; 62.18 +}; 62.19 + 62.20 62.21 class Universe: AllStatic { 62.22 // Ugh. Universe is much too friendly. 62.23 @@ -181,9 +194,9 @@ 62.24 62.25 // The particular choice of collected heap. 62.26 static CollectedHeap* _collectedHeap; 62.27 - // Base address for oop-within-java-object materialization. 62.28 - // NULL if using wide oops. Doubles as heap oop null value. 62.29 - static address _heap_base; 62.30 + 62.31 + // For UseCompressedOops. 62.32 + static struct NarrowOopStruct _narrow_oop; 62.33 62.34 // array of dummy objects used with +FullGCAlot 62.35 debug_only(static objArrayOop _fullgc_alot_dummy_array;) 62.36 @@ -328,8 +341,25 @@ 62.37 static CollectedHeap* heap() { return _collectedHeap; } 62.38 62.39 // For UseCompressedOops 62.40 - static address heap_base() { return _heap_base; } 62.41 - static address* heap_base_addr() { return &_heap_base; } 62.42 + static address* narrow_oop_base_addr() { return &_narrow_oop._base; } 62.43 + static address narrow_oop_base() { return _narrow_oop._base; } 62.44 + static int narrow_oop_shift() { return _narrow_oop._shift; } 62.45 + static void set_narrow_oop_base(address base) { _narrow_oop._base = base; } 62.46 + static void set_narrow_oop_shift(int shift) { _narrow_oop._shift = shift; } 62.47 + static bool narrow_oop_use_implicit_null_checks() { return _narrow_oop._use_implicit_null_checks; } 62.48 + static void set_narrow_oop_use_implicit_null_checks(bool use) { _narrow_oop._use_implicit_null_checks = use; } 62.49 + // Narrow Oop encoding mode: 62.50 + // 0 - Use 32-bits oops without encoding when 62.51 + // NarrowOopHeapBaseMin + heap_size < 4Gb 62.52 + // 1 - Use zero based compressed oops with encoding when 62.53 + // NarrowOopHeapBaseMin + heap_size < 32Gb 62.54 + // 2 - Use compressed oops with heap base + encoding. 62.55 + enum NARROW_OOP_MODE { 62.56 + UnscaledNarrowOop = 0, 62.57 + ZeroBasedNarrowOop = 1, 62.58 + HeapBasedNarrowOop = 2 62.59 + }; 62.60 + static char* preferred_heap_base(size_t heap_size, NARROW_OOP_MODE mode); 62.61 62.62 // Historic gc information 62.63 static size_t get_heap_capacity_at_last_gc() { return _heap_capacity_at_last_gc; }
63.1 --- a/src/share/vm/oops/oop.inline.hpp Wed Mar 18 11:37:48 2009 -0400 63.2 +++ b/src/share/vm/oops/oop.inline.hpp Thu Mar 19 09:13:24 2009 -0700 63.3 @@ -148,10 +148,11 @@ 63.4 63.5 inline narrowOop oopDesc::encode_heap_oop_not_null(oop v) { 63.6 assert(!is_null(v), "oop value can never be zero"); 63.7 - address heap_base = Universe::heap_base(); 63.8 - uint64_t pd = (uint64_t)(pointer_delta((void*)v, (void*)heap_base, 1)); 63.9 + address base = Universe::narrow_oop_base(); 63.10 + int shift = Universe::narrow_oop_shift(); 63.11 + uint64_t pd = (uint64_t)(pointer_delta((void*)v, (void*)base, 1)); 63.12 assert(OopEncodingHeapMax > pd, "change encoding max if new encoding"); 63.13 - uint64_t result = pd >> LogMinObjAlignmentInBytes; 63.14 + uint64_t result = pd >> shift; 63.15 assert((result & CONST64(0xffffffff00000000)) == 0, "narrow oop overflow"); 63.16 return (narrowOop)result; 63.17 } 63.18 @@ -162,8 +163,9 @@ 63.19 63.20 inline oop oopDesc::decode_heap_oop_not_null(narrowOop v) { 63.21 assert(!is_null(v), "narrow oop value can never be zero"); 63.22 - address heap_base = Universe::heap_base(); 63.23 - return (oop)(void*)((uintptr_t)heap_base + ((uintptr_t)v << LogMinObjAlignmentInBytes)); 63.24 + address base = Universe::narrow_oop_base(); 63.25 + int shift = Universe::narrow_oop_shift(); 63.26 + return (oop)(void*)((uintptr_t)base + ((uintptr_t)v << shift)); 63.27 } 63.28 63.29 inline oop oopDesc::decode_heap_oop(narrowOop v) {
64.1 --- a/src/share/vm/opto/addnode.cpp Wed Mar 18 11:37:48 2009 -0400 64.2 +++ b/src/share/vm/opto/addnode.cpp Thu Mar 19 09:13:24 2009 -0700 64.3 @@ -756,7 +756,13 @@ 64.4 if ( eti == NULL ) { 64.5 // there must be one pointer among the operands 64.6 guarantee(tptr == NULL, "must be only one pointer operand"); 64.7 - tptr = et->isa_oopptr(); 64.8 + if (UseCompressedOops && Universe::narrow_oop_shift() == 0) { 64.9 + // 32-bits narrow oop can be the base of address expressions 64.10 + tptr = et->make_ptr()->isa_oopptr(); 64.11 + } else { 64.12 + // only regular oops are expected here 64.13 + tptr = et->isa_oopptr(); 64.14 + } 64.15 guarantee(tptr != NULL, "non-int operand must be pointer"); 64.16 if (tptr->higher_equal(tp->add_offset(tptr->offset()))) 64.17 tp = tptr; // Set more precise type for bailout
65.1 --- a/src/share/vm/opto/classes.hpp Wed Mar 18 11:37:48 2009 -0400 65.2 +++ b/src/share/vm/opto/classes.hpp Thu Mar 19 09:13:24 2009 -0700 65.3 @@ -184,6 +184,8 @@ 65.4 macro(Parm) 65.5 macro(PartialSubtypeCheck) 65.6 macro(Phi) 65.7 +macro(PopCountI) 65.8 +macro(PopCountL) 65.9 macro(PowD) 65.10 macro(PrefetchRead) 65.11 macro(PrefetchWrite)
66.1 --- a/src/share/vm/opto/compile.cpp Wed Mar 18 11:37:48 2009 -0400 66.2 +++ b/src/share/vm/opto/compile.cpp Thu Mar 19 09:13:24 2009 -0700 66.3 @@ -2081,7 +2081,7 @@ 66.4 66.5 #ifdef _LP64 66.6 case Op_CastPP: 66.7 - if (n->in(1)->is_DecodeN() && UseImplicitNullCheckForNarrowOop) { 66.8 + if (n->in(1)->is_DecodeN() && Universe::narrow_oop_use_implicit_null_checks()) { 66.9 Compile* C = Compile::current(); 66.10 Node* in1 = n->in(1); 66.11 const Type* t = n->bottom_type(); 66.12 @@ -2136,7 +2136,7 @@ 66.13 new_in2 = in2->in(1); 66.14 } else if (in2->Opcode() == Op_ConP) { 66.15 const Type* t = in2->bottom_type(); 66.16 - if (t == TypePtr::NULL_PTR && UseImplicitNullCheckForNarrowOop) { 66.17 + if (t == TypePtr::NULL_PTR && Universe::narrow_oop_use_implicit_null_checks()) { 66.18 new_in2 = ConNode::make(C, TypeNarrowOop::NULL_PTR); 66.19 // 66.20 // This transformation together with CastPP transformation above
67.1 --- a/src/share/vm/opto/connode.cpp Wed Mar 18 11:37:48 2009 -0400 67.2 +++ b/src/share/vm/opto/connode.cpp Thu Mar 19 09:13:24 2009 -0700 67.3 @@ -433,7 +433,7 @@ 67.4 // If not converting int->oop, throw away cast after constant propagation 67.5 Node *CastPPNode::Ideal_DU_postCCP( PhaseCCP *ccp ) { 67.6 const Type *t = ccp->type(in(1)); 67.7 - if (!t->isa_oop_ptr() || in(1)->is_DecodeN()) { 67.8 + if (!t->isa_oop_ptr() || (in(1)->is_DecodeN() && Universe::narrow_oop_use_implicit_null_checks())) { 67.9 return NULL; // do not transform raw pointers or narrow oops 67.10 } 67.11 return ConstraintCastNode::Ideal_DU_postCCP(ccp);
68.1 --- a/src/share/vm/opto/connode.hpp Wed Mar 18 11:37:48 2009 -0400 68.2 +++ b/src/share/vm/opto/connode.hpp Thu Mar 19 09:13:24 2009 -0700 68.3 @@ -1,5 +1,5 @@ 68.4 /* 68.5 - * Copyright 1997-2008 Sun Microsystems, Inc. All Rights Reserved. 68.6 + * Copyright 1997-2009 Sun Microsystems, Inc. All Rights Reserved. 68.7 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 68.8 * 68.9 * This code is free software; you can redistribute it and/or modify it 68.10 @@ -635,3 +635,23 @@ 68.11 virtual uint ideal_reg() const { return Op_RegL; } 68.12 virtual const Type* Value( PhaseTransform *phase ) const; 68.13 }; 68.14 + 68.15 +//---------- PopCountINode ----------------------------------------------------- 68.16 +// Population count (bit count) of an integer. 68.17 +class PopCountINode : public Node { 68.18 +public: 68.19 + PopCountINode(Node* in1) : Node(0, in1) {} 68.20 + virtual int Opcode() const; 68.21 + const Type* bottom_type() const { return TypeInt::INT; } 68.22 + virtual uint ideal_reg() const { return Op_RegI; } 68.23 +}; 68.24 + 68.25 +//---------- PopCountLNode ----------------------------------------------------- 68.26 +// Population count (bit count) of a long. 68.27 +class PopCountLNode : public Node { 68.28 +public: 68.29 + PopCountLNode(Node* in1) : Node(0, in1) {} 68.30 + virtual int Opcode() const; 68.31 + const Type* bottom_type() const { return TypeInt::INT; } 68.32 + virtual uint ideal_reg() const { return Op_RegI; } 68.33 +};
69.1 --- a/src/share/vm/opto/graphKit.cpp Wed Mar 18 11:37:48 2009 -0400 69.2 +++ b/src/share/vm/opto/graphKit.cpp Thu Mar 19 09:13:24 2009 -0700 69.3 @@ -2277,7 +2277,7 @@ 69.4 r_not_subtype->init_req(1, _gvn.transform( new (C, 1) IfTrueNode (iff2) ) ); 69.5 set_control( _gvn.transform( new (C, 1) IfFalseNode(iff2) ) ); 69.6 69.7 - // Check for self. Very rare to get here, but its taken 1/3 the time. 69.8 + // Check for self. Very rare to get here, but it is taken 1/3 the time. 69.9 // No performance impact (too rare) but allows sharing of secondary arrays 69.10 // which has some footprint reduction. 69.11 Node *cmp3 = _gvn.transform( new (C, 3) CmpPNode( subklass, superklass ) ); 69.12 @@ -2286,11 +2286,27 @@ 69.13 r_ok_subtype->init_req(2, _gvn.transform( new (C, 1) IfTrueNode ( iff3 ) ) ); 69.14 set_control( _gvn.transform( new (C, 1) IfFalseNode( iff3 ) ) ); 69.15 69.16 + // -- Roads not taken here: -- 69.17 + // We could also have chosen to perform the self-check at the beginning 69.18 + // of this code sequence, as the assembler does. This would not pay off 69.19 + // the same way, since the optimizer, unlike the assembler, can perform 69.20 + // static type analysis to fold away many successful self-checks. 69.21 + // Non-foldable self checks work better here in second position, because 69.22 + // the initial primary superclass check subsumes a self-check for most 69.23 + // types. An exception would be a secondary type like array-of-interface, 69.24 + // which does not appear in its own primary supertype display. 69.25 + // Finally, we could have chosen to move the self-check into the 69.26 + // PartialSubtypeCheckNode, and from there out-of-line in a platform 69.27 + // dependent manner. But it is worthwhile to have the check here, 69.28 + // where it can be perhaps be optimized. The cost in code space is 69.29 + // small (register compare, branch). 69.30 + 69.31 // Now do a linear scan of the secondary super-klass array. Again, no real 69.32 // performance impact (too rare) but it's gotta be done. 69.33 - // (The stub also contains the self-check of subklass == superklass. 69.34 // Since the code is rarely used, there is no penalty for moving it 69.35 - // out of line, and it can only improve I-cache density.) 69.36 + // out of line, and it can only improve I-cache density. 69.37 + // The decision to inline or out-of-line this final check is platform 69.38 + // dependent, and is found in the AD file definition of PartialSubtypeCheck. 69.39 Node* psc = _gvn.transform( 69.40 new (C, 3) PartialSubtypeCheckNode(control(), subklass, superklass) ); 69.41
70.1 --- a/src/share/vm/opto/lcm.cpp Wed Mar 18 11:37:48 2009 -0400 70.2 +++ b/src/share/vm/opto/lcm.cpp Thu Mar 19 09:13:24 2009 -0700 70.3 @@ -158,7 +158,14 @@ 70.4 continue; // Give up if offset is beyond page size 70.5 // cannot reason about it; is probably not implicit null exception 70.6 } else { 70.7 - const TypePtr* tptr = base->bottom_type()->is_ptr(); 70.8 + const TypePtr* tptr; 70.9 + if (UseCompressedOops && Universe::narrow_oop_shift() == 0) { 70.10 + // 32-bits narrow oop can be the base of address expressions 70.11 + tptr = base->bottom_type()->make_ptr(); 70.12 + } else { 70.13 + // only regular oops are expected here 70.14 + tptr = base->bottom_type()->is_ptr(); 70.15 + } 70.16 // Give up if offset is not a compile-time constant 70.17 if( offset == Type::OffsetBot || tptr->_offset == Type::OffsetBot ) 70.18 continue;
71.1 --- a/src/share/vm/opto/library_call.cpp Wed Mar 18 11:37:48 2009 -0400 71.2 +++ b/src/share/vm/opto/library_call.cpp Thu Mar 19 09:13:24 2009 -0700 71.3 @@ -1,5 +1,5 @@ 71.4 /* 71.5 - * Copyright 1999-2008 Sun Microsystems, Inc. All Rights Reserved. 71.6 + * Copyright 1999-2009 Sun Microsystems, Inc. All Rights Reserved. 71.7 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 71.8 * 71.9 * This code is free software; you can redistribute it and/or modify it 71.10 @@ -221,6 +221,7 @@ 71.11 bool inline_unsafe_CAS(BasicType type); 71.12 bool inline_unsafe_ordered_store(BasicType type); 71.13 bool inline_fp_conversions(vmIntrinsics::ID id); 71.14 + bool inline_bitCount(vmIntrinsics::ID id); 71.15 bool inline_reverseBytes(vmIntrinsics::ID id); 71.16 }; 71.17 71.18 @@ -314,6 +315,11 @@ 71.19 if (!JDK_Version::is_gte_jdk14x_version()) return NULL; 71.20 break; 71.21 71.22 + case vmIntrinsics::_bitCount_i: 71.23 + case vmIntrinsics::_bitCount_l: 71.24 + if (!UsePopCountInstruction) return NULL; 71.25 + break; 71.26 + 71.27 default: 71.28 break; 71.29 } 71.30 @@ -617,6 +623,10 @@ 71.31 case vmIntrinsics::_longBitsToDouble: 71.32 return inline_fp_conversions(intrinsic_id()); 71.33 71.34 + case vmIntrinsics::_bitCount_i: 71.35 + case vmIntrinsics::_bitCount_l: 71.36 + return inline_bitCount(intrinsic_id()); 71.37 + 71.38 case vmIntrinsics::_reverseBytes_i: 71.39 case vmIntrinsics::_reverseBytes_l: 71.40 return inline_reverseBytes((vmIntrinsics::ID) intrinsic_id()); 71.41 @@ -1714,6 +1724,27 @@ 71.42 } 71.43 } 71.44 71.45 +//----------------------------inline_bitCount_int/long----------------------- 71.46 +// inline int Integer.bitCount(int) 71.47 +// inline int Long.bitCount(long) 71.48 +bool LibraryCallKit::inline_bitCount(vmIntrinsics::ID id) { 71.49 + assert(id == vmIntrinsics::_bitCount_i || id == vmIntrinsics::_bitCount_l, "not bitCount"); 71.50 + if (id == vmIntrinsics::_bitCount_i && !Matcher::has_match_rule(Op_PopCountI)) return false; 71.51 + if (id == vmIntrinsics::_bitCount_l && !Matcher::has_match_rule(Op_PopCountL)) return false; 71.52 + _sp += arg_size(); // restore stack pointer 71.53 + switch (id) { 71.54 + case vmIntrinsics::_bitCount_i: 71.55 + push(_gvn.transform(new (C, 2) PopCountINode(pop()))); 71.56 + break; 71.57 + case vmIntrinsics::_bitCount_l: 71.58 + push(_gvn.transform(new (C, 2) PopCountLNode(pop_pair()))); 71.59 + break; 71.60 + default: 71.61 + ShouldNotReachHere(); 71.62 + } 71.63 + return true; 71.64 +} 71.65 + 71.66 //----------------------------inline_reverseBytes_int/long------------------- 71.67 // inline Integer.reverseBytes(int) 71.68 // inline Long.reverseBytes(long)
72.1 --- a/src/share/vm/opto/matcher.cpp Wed Mar 18 11:37:48 2009 -0400 72.2 +++ b/src/share/vm/opto/matcher.cpp Thu Mar 19 09:13:24 2009 -0700 72.3 @@ -1481,8 +1481,13 @@ 72.4 const Type* mach_at = mach->adr_type(); 72.5 // DecodeN node consumed by an address may have different type 72.6 // then its input. Don't compare types for such case. 72.7 - if (m->adr_type() != mach_at && m->in(MemNode::Address)->is_AddP() && 72.8 - m->in(MemNode::Address)->in(AddPNode::Address)->is_DecodeN()) { 72.9 + if (m->adr_type() != mach_at && 72.10 + (m->in(MemNode::Address)->is_DecodeN() || 72.11 + m->in(MemNode::Address)->is_AddP() && 72.12 + m->in(MemNode::Address)->in(AddPNode::Address)->is_DecodeN() || 72.13 + m->in(MemNode::Address)->is_AddP() && 72.14 + m->in(MemNode::Address)->in(AddPNode::Address)->is_AddP() && 72.15 + m->in(MemNode::Address)->in(AddPNode::Address)->in(AddPNode::Address)->is_DecodeN())) { 72.16 mach_at = m->adr_type(); 72.17 } 72.18 if (m->adr_type() != mach_at) {
73.1 --- a/src/share/vm/runtime/arguments.cpp Wed Mar 18 11:37:48 2009 -0400 73.2 +++ b/src/share/vm/runtime/arguments.cpp Thu Mar 19 09:13:24 2009 -0700 73.3 @@ -1211,7 +1211,9 @@ 73.4 if (UseLargePages && UseCompressedOops) { 73.5 // Cannot allocate guard pages for implicit checks in indexed addressing 73.6 // mode, when large pages are specified on windows. 73.7 - FLAG_SET_DEFAULT(UseImplicitNullCheckForNarrowOop, false); 73.8 + // This flag could be switched ON if narrow oop base address is set to 0, 73.9 + // see code in Universe::initialize_heap(). 73.10 + Universe::set_narrow_oop_use_implicit_null_checks(false); 73.11 } 73.12 #endif // _WIN64 73.13 } else {
74.1 --- a/src/share/vm/runtime/globals.hpp Wed Mar 18 11:37:48 2009 -0400 74.2 +++ b/src/share/vm/runtime/globals.hpp Thu Mar 19 09:13:24 2009 -0700 74.3 @@ -303,11 +303,14 @@ 74.4 "Use 32-bit object references in 64-bit VM. " \ 74.5 "lp64_product means flag is always constant in 32 bit VM") \ 74.6 \ 74.7 - lp64_product(bool, CheckCompressedOops, trueInDebug, \ 74.8 - "generate checks in encoding/decoding code") \ 74.9 - \ 74.10 - product(bool, UseImplicitNullCheckForNarrowOop, true, \ 74.11 - "generate implicit null check in indexed addressing mode.") \ 74.12 + notproduct(bool, CheckCompressedOops, true, \ 74.13 + "generate checks in encoding/decoding code in debug VM") \ 74.14 + \ 74.15 + product_pd(uintx, HeapBaseMinAddress, \ 74.16 + "OS specific low limit for heap base address") \ 74.17 + \ 74.18 + diagnostic(bool, PrintCompressedOopsMode, false, \ 74.19 + "Print compressed oops base address and encoding mode") \ 74.20 \ 74.21 /* UseMembar is theoretically a temp flag used for memory barrier \ 74.22 * removal testing. It was supposed to be removed before FCS but has \ 74.23 @@ -2169,6 +2172,9 @@ 74.24 diagnostic(bool, PrintIntrinsics, false, \ 74.25 "prints attempted and successful inlining of intrinsics") \ 74.26 \ 74.27 + product(bool, UsePopCountInstruction, false, \ 74.28 + "Use population count instruction") \ 74.29 + \ 74.30 diagnostic(ccstrlist, DisableIntrinsic, "", \ 74.31 "do not expand intrinsics whose (internal) names appear here") \ 74.32 \
75.1 --- a/src/share/vm/runtime/os.hpp Wed Mar 18 11:37:48 2009 -0400 75.2 +++ b/src/share/vm/runtime/os.hpp Thu Mar 19 09:13:24 2009 -0700 75.3 @@ -243,7 +243,7 @@ 75.4 75.5 static char* non_memory_address_word(); 75.6 // reserve, commit and pin the entire memory region 75.7 - static char* reserve_memory_special(size_t size); 75.8 + static char* reserve_memory_special(size_t size, char* addr = NULL); 75.9 static bool release_memory_special(char* addr, size_t bytes); 75.10 static bool large_page_init(); 75.11 static size_t large_page_size();
76.1 --- a/src/share/vm/runtime/virtualspace.cpp Wed Mar 18 11:37:48 2009 -0400 76.2 +++ b/src/share/vm/runtime/virtualspace.cpp Thu Mar 19 09:13:24 2009 -0700 76.3 @@ -109,6 +109,7 @@ 76.4 const size_t prefix_align, 76.5 const size_t suffix_size, 76.6 const size_t suffix_align, 76.7 + char* requested_address, 76.8 const size_t noaccess_prefix) 76.9 { 76.10 assert(prefix_size != 0, "sanity"); 76.11 @@ -131,7 +132,7 @@ 76.12 const bool try_reserve_special = UseLargePages && 76.13 prefix_align == os::large_page_size(); 76.14 if (!os::can_commit_large_page_memory() && try_reserve_special) { 76.15 - initialize(size, prefix_align, true, NULL, noaccess_prefix); 76.16 + initialize(size, prefix_align, true, requested_address, noaccess_prefix); 76.17 return; 76.18 } 76.19 76.20 @@ -146,7 +147,13 @@ 76.21 noaccess_prefix == prefix_align, "noaccess prefix wrong"); 76.22 76.23 // Optimistically try to reserve the exact size needed. 76.24 - char* addr = os::reserve_memory(size, NULL, prefix_align); 76.25 + char* addr; 76.26 + if (requested_address != 0) { 76.27 + addr = os::attempt_reserve_memory_at(size, 76.28 + requested_address-noaccess_prefix); 76.29 + } else { 76.30 + addr = os::reserve_memory(size, NULL, prefix_align); 76.31 + } 76.32 if (addr == NULL) return; 76.33 76.34 // Check whether the result has the needed alignment (unlikely unless 76.35 @@ -206,12 +213,8 @@ 76.36 char* base = NULL; 76.37 76.38 if (special) { 76.39 - // It's not hard to implement reserve_memory_special() such that it can 76.40 - // allocate at fixed address, but there seems no use of this feature 76.41 - // for now, so it's not implemented. 76.42 - assert(requested_address == NULL, "not implemented"); 76.43 76.44 - base = os::reserve_memory_special(size); 76.45 + base = os::reserve_memory_special(size, requested_address); 76.46 76.47 if (base != NULL) { 76.48 // Check alignment constraints 76.49 @@ -372,7 +375,8 @@ 76.50 bool large, char* requested_address) : 76.51 ReservedSpace(size, alignment, large, 76.52 requested_address, 76.53 - UseCompressedOops && UseImplicitNullCheckForNarrowOop ? 76.54 + (UseCompressedOops && (Universe::narrow_oop_base() != NULL) && 76.55 + Universe::narrow_oop_use_implicit_null_checks()) ? 76.56 lcm(os::vm_page_size(), alignment) : 0) { 76.57 // Only reserved space for the java heap should have a noaccess_prefix 76.58 // if using compressed oops. 76.59 @@ -382,9 +386,12 @@ 76.60 ReservedHeapSpace::ReservedHeapSpace(const size_t prefix_size, 76.61 const size_t prefix_align, 76.62 const size_t suffix_size, 76.63 - const size_t suffix_align) : 76.64 + const size_t suffix_align, 76.65 + char* requested_address) : 76.66 ReservedSpace(prefix_size, prefix_align, suffix_size, suffix_align, 76.67 - UseCompressedOops && UseImplicitNullCheckForNarrowOop ? 76.68 + requested_address, 76.69 + (UseCompressedOops && (Universe::narrow_oop_base() != NULL) && 76.70 + Universe::narrow_oop_use_implicit_null_checks()) ? 76.71 lcm(os::vm_page_size(), prefix_align) : 0) { 76.72 protect_noaccess_prefix(prefix_size+suffix_size); 76.73 }
77.1 --- a/src/share/vm/runtime/virtualspace.hpp Wed Mar 18 11:37:48 2009 -0400 77.2 +++ b/src/share/vm/runtime/virtualspace.hpp Thu Mar 19 09:13:24 2009 -0700 77.3 @@ -73,7 +73,8 @@ 77.4 const size_t noaccess_prefix = 0); 77.5 ReservedSpace(const size_t prefix_size, const size_t prefix_align, 77.6 const size_t suffix_size, const size_t suffix_align, 77.7 - const size_t noaccess_prefix); 77.8 + char* requested_address, 77.9 + const size_t noaccess_prefix = 0); 77.10 77.11 // Accessors 77.12 char* base() const { return _base; } 77.13 @@ -121,7 +122,8 @@ 77.14 ReservedHeapSpace(size_t size, size_t forced_base_alignment, 77.15 bool large, char* requested_address); 77.16 ReservedHeapSpace(const size_t prefix_size, const size_t prefix_align, 77.17 - const size_t suffix_size, const size_t suffix_align); 77.18 + const size_t suffix_size, const size_t suffix_align, 77.19 + char* requested_address); 77.20 }; 77.21 77.22 // VirtualSpace is data structure for committing a previously reserved address range in smaller chunks.
78.1 --- a/src/share/vm/runtime/vmStructs.cpp Wed Mar 18 11:37:48 2009 -0400 78.2 +++ b/src/share/vm/runtime/vmStructs.cpp Thu Mar 19 09:13:24 2009 -0700 78.3 @@ -263,7 +263,9 @@ 78.4 static_field(Universe, _bootstrapping, bool) \ 78.5 static_field(Universe, _fully_initialized, bool) \ 78.6 static_field(Universe, _verify_count, int) \ 78.7 - static_field(Universe, _heap_base, address) \ 78.8 + static_field(Universe, _narrow_oop._base, address) \ 78.9 + static_field(Universe, _narrow_oop._shift, int) \ 78.10 + static_field(Universe, _narrow_oop._use_implicit_null_checks, bool) \ 78.11 \ 78.12 /**********************************************************************************/ \ 78.13 /* Generation and Space hierarchies */ \
79.1 --- a/src/share/vm/runtime/vm_version.cpp Wed Mar 18 11:37:48 2009 -0400 79.2 +++ b/src/share/vm/runtime/vm_version.cpp Thu Mar 19 09:13:24 2009 -0700 79.3 @@ -163,9 +163,11 @@ 79.4 #elif _MSC_VER == 1200 79.5 #define HOTSPOT_BUILD_COMPILER "MS VC++ 6.0" 79.6 #elif _MSC_VER == 1310 79.7 - #define HOTSPOT_BUILD_COMPILER "MS VC++ 7.1" 79.8 + #define HOTSPOT_BUILD_COMPILER "MS VC++ 7.1 (VS2003)" 79.9 #elif _MSC_VER == 1400 79.10 - #define HOTSPOT_BUILD_COMPILER "MS VC++ 8.0" 79.11 + #define HOTSPOT_BUILD_COMPILER "MS VC++ 8.0 (VS2005)" 79.12 + #elif _MSC_VER == 1500 79.13 + #define HOTSPOT_BUILD_COMPILER "MS VC++ 9.0 (VS2008)" 79.14 #else 79.15 #define HOTSPOT_BUILD_COMPILER "unknown MS VC++:" XSTR(_MSC_VER) 79.16 #endif
80.1 --- a/src/share/vm/utilities/globalDefinitions_visCPP.hpp Wed Mar 18 11:37:48 2009 -0400 80.2 +++ b/src/share/vm/utilities/globalDefinitions_visCPP.hpp Thu Mar 19 09:13:24 2009 -0700 80.3 @@ -162,7 +162,7 @@ 80.4 } 80.5 80.6 // Visual Studio 2005 deprecates POSIX names - use ISO C++ names instead 80.7 -#if _MSC_VER >= 1400 && !defined(_WIN64) 80.8 +#if _MSC_VER >= 1400 80.9 #define open _open 80.10 #define close _close 80.11 #define read _read
81.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000 81.2 +++ b/test/compiler/6378821/Test6378821.java Thu Mar 19 09:13:24 2009 -0700 81.3 @@ -0,0 +1,75 @@ 81.4 +/* 81.5 + * Copyright 2009 Sun Microsystems, Inc. All Rights Reserved. 81.6 + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 81.7 + * 81.8 + * This code is free software; you can redistribute it and/or modify it 81.9 + * under the terms of the GNU General Public License version 2 only, as 81.10 + * published by the Free Software Foundation. 81.11 + * 81.12 + * This code is distributed in the hope that it will be useful, but WITHOUT 81.13 + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 81.14 + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 81.15 + * version 2 for more details (a copy is included in the LICENSE file that 81.16 + * accompanied this code). 81.17 + * 81.18 + * You should have received a copy of the GNU General Public License version 81.19 + * 2 along with this work; if not, write to the Free Software Foundation, 81.20 + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 81.21 + * 81.22 + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, 81.23 + * CA 95054 USA or visit www.sun.com if you need additional information or 81.24 + * have any questions. 81.25 + */ 81.26 + 81.27 +/** 81.28 + * @test 81.29 + * @bug 6378821 81.30 + * @summary where available, bitCount() should use POPC on SPARC processors and AMD+10h 81.31 + * 81.32 + * @run main/othervm -Xcomp -XX:CompileOnly=Test6378821.fcomp Test6378821 81.33 + */ 81.34 + 81.35 +public class Test6378821 { 81.36 + static final int[] ia = new int[] { 0x12345678 }; 81.37 + static final long[] la = new long[] { 0x12345678abcdefL }; 81.38 + 81.39 + public static void main(String [] args) { 81.40 + // Resolve the class and the method. 81.41 + Integer.bitCount(1); 81.42 + Long.bitCount(1); 81.43 + 81.44 + sub(ia[0]); 81.45 + sub(la[0]); 81.46 + sub(ia); 81.47 + sub(la); 81.48 + } 81.49 + 81.50 + static void check(int i, int expected, int result) { 81.51 + if (result != expected) { 81.52 + throw new InternalError("Wrong population count for " + i + ": " + result + " != " + expected); 81.53 + } 81.54 + } 81.55 + 81.56 + static void check(long l, int expected, int result) { 81.57 + if (result != expected) { 81.58 + throw new InternalError("Wrong population count for " + l + ": " + result + " != " + expected); 81.59 + } 81.60 + } 81.61 + 81.62 + static void sub(int i) { check(i, fint(i), fcomp(i) ); } 81.63 + static void sub(int[] ia) { check(ia[0], fint(ia), fcomp(ia)); } 81.64 + static void sub(long l) { check(l, fint(l), fcomp(l) ); } 81.65 + static void sub(long[] la) { check(la[0], fint(la), fcomp(la)); } 81.66 + 81.67 + static int fint (int i) { return Integer.bitCount(i); } 81.68 + static int fcomp(int i) { return Integer.bitCount(i); } 81.69 + 81.70 + static int fint (int[] ia) { return Integer.bitCount(ia[0]); } 81.71 + static int fcomp(int[] ia) { return Integer.bitCount(ia[0]); } 81.72 + 81.73 + static int fint (long l) { return Long.bitCount(l); } 81.74 + static int fcomp(long l) { return Long.bitCount(l); } 81.75 + 81.76 + static int fint (long[] la) { return Long.bitCount(la[0]); } 81.77 + static int fcomp(long[] la) { return Long.bitCount(la[0]); } 81.78 +}