changeset 6830:3d02fef8050a

Merge
author roland
date Fri, 01 Aug 2014 15:57:00 -0700
parents e55980631e8c 2c26420129d9
children fcb6c71678c9
files
diffstat 37 files changed, 543 insertions(+), 231 deletions(-) [+]
line wrap: on
line diff
--- a/src/cpu/ppc/vm/assembler_ppc.hpp	Thu Jul 31 16:39:57 2014 -0700
+++ b/src/cpu/ppc/vm/assembler_ppc.hpp	Fri Aug 01 15:57:00 2014 -0700
@@ -298,6 +298,7 @@
     LWZ_OPCODE   = (32u << OPCODE_SHIFT),
     LWZX_OPCODE  = (31u << OPCODE_SHIFT |  23u << 1),
     LWZU_OPCODE  = (33u << OPCODE_SHIFT),
+    LWBRX_OPCODE = (31u << OPCODE_SHIFT |  534 << 1),
 
     LHA_OPCODE   = (42u << OPCODE_SHIFT),
     LHAX_OPCODE  = (31u << OPCODE_SHIFT | 343u << 1),
@@ -306,6 +307,7 @@
     LHZ_OPCODE   = (40u << OPCODE_SHIFT),
     LHZX_OPCODE  = (31u << OPCODE_SHIFT | 279u << 1),
     LHZU_OPCODE  = (41u << OPCODE_SHIFT),
+    LHBRX_OPCODE = (31u << OPCODE_SHIFT |  790 << 1),
 
     LBZ_OPCODE   = (34u << OPCODE_SHIFT),
     LBZX_OPCODE  = (31u << OPCODE_SHIFT |  87u << 1),
@@ -1364,11 +1366,17 @@
   inline void lwax( Register d, Register s1, Register s2);
   inline void lwa(  Register d, int si16,    Register s1);
 
+  // 4 bytes reversed
+  inline void lwbrx( Register d, Register s1, Register s2);
+
   // 2 bytes
   inline void lhzx( Register d, Register s1, Register s2);
   inline void lhz(  Register d, int si16,    Register s1);
   inline void lhzu( Register d, int si16,    Register s1);
 
+  // 2 bytes reversed
+  inline void lhbrx( Register d, Register s1, Register s2);
+
   // 2 bytes
   inline void lhax( Register d, Register s1, Register s2);
   inline void lha(  Register d, int si16,    Register s1);
@@ -1858,10 +1866,12 @@
   inline void lwz(  Register d, int si16);
   inline void lwax( Register d, Register s2);
   inline void lwa(  Register d, int si16);
+  inline void lwbrx(Register d, Register s2);
   inline void lhzx( Register d, Register s2);
   inline void lhz(  Register d, int si16);
   inline void lhax( Register d, Register s2);
   inline void lha(  Register d, int si16);
+  inline void lhbrx(Register d, Register s2);
   inline void lbzx( Register d, Register s2);
   inline void lbz(  Register d, int si16);
   inline void ldx(  Register d, Register s2);
--- a/src/cpu/ppc/vm/assembler_ppc.inline.hpp	Thu Jul 31 16:39:57 2014 -0700
+++ b/src/cpu/ppc/vm/assembler_ppc.inline.hpp	Fri Aug 01 15:57:00 2014 -0700
@@ -263,10 +263,14 @@
 inline void Assembler::lwax( Register d, Register s1, Register s2) { emit_int32(LWAX_OPCODE | rt(d) | ra0mem(s1) | rb(s2));}
 inline void Assembler::lwa(  Register d, int si16,    Register s1) { emit_int32(LWA_OPCODE  | rt(d) | ds(si16)   | ra0mem(s1));}
 
+inline void Assembler::lwbrx( Register d, Register s1, Register s2) { emit_int32(LWBRX_OPCODE | rt(d) | ra0mem(s1) | rb(s2));}
+
 inline void Assembler::lhzx( Register d, Register s1, Register s2) { emit_int32(LHZX_OPCODE | rt(d) | ra0mem(s1) | rb(s2));}
 inline void Assembler::lhz(  Register d, int si16,    Register s1) { emit_int32(LHZ_OPCODE  | rt(d) | d1(si16)   | ra0mem(s1));}
 inline void Assembler::lhzu( Register d, int si16,    Register s1) { assert(d != s1, "according to ibm manual"); emit_int32(LHZU_OPCODE | rt(d) | d1(si16) | rta0mem(s1));}
 
+inline void Assembler::lhbrx( Register d, Register s1, Register s2) { emit_int32(LHBRX_OPCODE | rt(d) | ra0mem(s1) | rb(s2));}
+
 inline void Assembler::lhax( Register d, Register s1, Register s2) { emit_int32(LHAX_OPCODE | rt(d) | ra0mem(s1) | rb(s2));}
 inline void Assembler::lha(  Register d, int si16,    Register s1) { emit_int32(LHA_OPCODE  | rt(d) | d1(si16)   | ra0mem(s1));}
 inline void Assembler::lhau( Register d, int si16,    Register s1) { assert(d != s1, "according to ibm manual"); emit_int32(LHAU_OPCODE | rt(d) | d1(si16) | rta0mem(s1));}
@@ -736,10 +740,12 @@
 inline void Assembler::lwz(  Register d, int si16   ) { emit_int32( LWZ_OPCODE  | rt(d) | d1(si16));}
 inline void Assembler::lwax( Register d, Register s2) { emit_int32( LWAX_OPCODE | rt(d) | rb(s2));}
 inline void Assembler::lwa(  Register d, int si16   ) { emit_int32( LWA_OPCODE  | rt(d) | ds(si16));}
+inline void Assembler::lwbrx(Register d, Register s2) { emit_int32( LWBRX_OPCODE| rt(d) | rb(s2));}
 inline void Assembler::lhzx( Register d, Register s2) { emit_int32( LHZX_OPCODE | rt(d) | rb(s2));}
 inline void Assembler::lhz(  Register d, int si16   ) { emit_int32( LHZ_OPCODE  | rt(d) | d1(si16));}
 inline void Assembler::lhax( Register d, Register s2) { emit_int32( LHAX_OPCODE | rt(d) | rb(s2));}
 inline void Assembler::lha(  Register d, int si16   ) { emit_int32( LHA_OPCODE  | rt(d) | d1(si16));}
+inline void Assembler::lhbrx(Register d, Register s2) { emit_int32( LHBRX_OPCODE| rt(d) | rb(s2));}
 inline void Assembler::lbzx( Register d, Register s2) { emit_int32( LBZX_OPCODE | rt(d) | rb(s2));}
 inline void Assembler::lbz(  Register d, int si16   ) { emit_int32( LBZ_OPCODE  | rt(d) | d1(si16));}
 inline void Assembler::ld(   Register d, int si16   ) { emit_int32( LD_OPCODE   | rt(d) | ds(si16));}
--- a/src/cpu/ppc/vm/interp_masm_ppc_64.cpp	Thu Jul 31 16:39:57 2014 -0700
+++ b/src/cpu/ppc/vm/interp_masm_ppc_64.cpp	Fri Aug 01 15:57:00 2014 -0700
@@ -119,9 +119,15 @@
     // Call the Interpreter::remove_activation_preserving_args_entry()
     // func to get the address of the same-named entrypoint in the
     // generated interpreter code.
+#if defined(ABI_ELFv2)
+    call_c(CAST_FROM_FN_PTR(address,
+                            Interpreter::remove_activation_preserving_args_entry),
+           relocInfo::none);
+#else
     call_c(CAST_FROM_FN_PTR(FunctionDescriptor*,
                             Interpreter::remove_activation_preserving_args_entry),
            relocInfo::none);
+#endif
 
     // Jump to Interpreter::_remove_activation_preserving_args_entry.
     mtctr(R3_RET);
@@ -331,29 +337,40 @@
 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp(int         bcp_offset,
                                                           Register    Rdst,
                                                           signedOrNot is_signed) {
+#if defined(VM_LITTLE_ENDIAN)
+  if (bcp_offset) {
+    load_const_optimized(Rdst, bcp_offset);
+    lhbrx(Rdst, R14_bcp, Rdst);
+  } else {
+    lhbrx(Rdst, R14_bcp);
+  }
+  if (is_signed == Signed) {
+    extsh(Rdst, Rdst);
+  }
+#else
   // Read Java big endian format.
   if (is_signed == Signed) {
     lha(Rdst, bcp_offset, R14_bcp);
   } else {
     lhz(Rdst, bcp_offset, R14_bcp);
   }
-#if 0
-  assert(Rtmp != Rdst, "need separate temp register");
-  Register Rfirst = Rtmp;
-  lbz(Rfirst, bcp_offset, R14_bcp); // first byte
-  lbz(Rdst, bcp_offset+1, R14_bcp); // second byte
-
-  // Rdst = ((Rfirst<<8) & 0xFF00) | (Rdst &~ 0xFF00)
-  rldimi(/*RA=*/Rdst, /*RS=*/Rfirst, /*sh=*/8, /*mb=*/48);
-  if (is_signed == Signed) {
-    extsh(Rdst, Rdst);
-  }
 #endif
 }
 
 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp(int         bcp_offset,
                                                           Register    Rdst,
                                                           signedOrNot is_signed) {
+#if defined(VM_LITTLE_ENDIAN)
+  if (bcp_offset) {
+    load_const_optimized(Rdst, bcp_offset);
+    lwbrx(Rdst, R14_bcp, Rdst);
+  } else {
+    lwbrx(Rdst, R14_bcp);
+  }
+  if (is_signed == Signed) {
+    extsw(Rdst, Rdst);
+  }
+#else
   // Read Java big endian format.
   if (bcp_offset & 3) { // Offset unaligned?
     load_const_optimized(Rdst, bcp_offset);
@@ -369,18 +386,26 @@
       lwz(Rdst, bcp_offset, R14_bcp);
     }
   }
+#endif
 }
 
+
 // Load the constant pool cache index from the bytecode stream.
 //
 // Kills / writes:
 //   - Rdst, Rscratch
 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register Rdst, int bcp_offset, size_t index_size) {
   assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
+  // Cache index is always in the native format, courtesy of Rewriter.
   if (index_size == sizeof(u2)) {
-    get_2_byte_integer_at_bcp(bcp_offset, Rdst, Unsigned);
+    lhz(Rdst, bcp_offset, R14_bcp);
   } else if (index_size == sizeof(u4)) {
-    get_4_byte_integer_at_bcp(bcp_offset, Rdst, Signed);
+    if (bcp_offset & 3) {
+      load_const_optimized(Rdst, bcp_offset);
+      lwax(Rdst, R14_bcp, Rdst);
+    } else {
+      lwa(Rdst, bcp_offset, R14_bcp);
+    }
     assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
     nand(Rdst, Rdst, Rdst); // convert to plain index
   } else if (index_size == sizeof(u1)) {
@@ -397,6 +422,29 @@
   add(cache, R27_constPoolCache, cache);
 }
 
+// Load 4-byte signed or unsigned integer in Java format (that is, big-endian format)
+// from (Rsrc)+offset.
+void InterpreterMacroAssembler::get_u4(Register Rdst, Register Rsrc, int offset,
+                                       signedOrNot is_signed) {
+#if defined(VM_LITTLE_ENDIAN)
+  if (offset) {
+    load_const_optimized(Rdst, offset);
+    lwbrx(Rdst, Rdst, Rsrc);
+  } else {
+    lwbrx(Rdst, Rsrc);
+  }
+  if (is_signed == Signed) {
+    extsw(Rdst, Rdst);
+  }
+#else
+  if (is_signed == Signed) {
+    lwa(Rdst, offset, Rsrc);
+  } else {
+    lwz(Rdst, offset, Rsrc);
+  }
+#endif
+}
+
 // Load object from cpool->resolved_references(index).
 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result, Register index) {
   assert_different_registers(result, index);
--- a/src/cpu/ppc/vm/interp_masm_ppc_64.hpp	Thu Jul 31 16:39:57 2014 -0700
+++ b/src/cpu/ppc/vm/interp_masm_ppc_64.hpp	Fri Aug 01 15:57:00 2014 -0700
@@ -130,6 +130,7 @@
 
   void get_cache_and_index_at_bcp(Register cache, int bcp_offset, size_t index_size = sizeof(u2));
 
+  void get_u4(Register Rdst, Register Rsrc, int offset, signedOrNot is_signed);
 
   // common code
 
--- a/src/cpu/ppc/vm/templateInterpreter_ppc.cpp	Thu Jul 31 16:39:57 2014 -0700
+++ b/src/cpu/ppc/vm/templateInterpreter_ppc.cpp	Fri Aug 01 15:57:00 2014 -0700
@@ -176,8 +176,12 @@
   const Register size  = R12_scratch2;
   __ get_cache_and_index_at_bcp(cache, 1, index_size);
 
-  // Big Endian (get least significant byte of 64 bit value):
+  // Get least significant byte of 64 bit value:
+#if defined(VM_LITTLE_ENDIAN)
+  __ lbz(size, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()), cache);
+#else
   __ lbz(size, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()) + 7, cache);
+#endif
   __ sldi(size, size, Interpreter::logStackElementSize);
   __ add(R15_esp, R15_esp, size);
   __ dispatch_next(state, step);
@@ -858,7 +862,9 @@
   // Our signature handlers copy required arguments to the C stack
   // (outgoing C args), R3_ARG1 to R10_ARG8, and FARG1 to FARG13.
   __ mr(R3_ARG1, R18_locals);
+#if !defined(ABI_ELFv2)
   __ ld(signature_handler_fd, 0, signature_handler_fd);
+#endif
 
   __ call_stub(signature_handler_fd);
 
@@ -1020,8 +1026,13 @@
   // native result across the call. No oop is present.
 
   __ mr(R3_ARG1, R16_thread);
+#if defined(ABI_ELFv2)
+  __ call_c(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans),
+            relocInfo::none);
+#else
   __ call_c(CAST_FROM_FN_PTR(FunctionDescriptor*, JavaThread::check_special_condition_for_native_trans),
             relocInfo::none);
+#endif
 
   __ bind(sync_check_done);
 
--- a/src/cpu/ppc/vm/templateTable_ppc_64.cpp	Thu Jul 31 16:39:57 2014 -0700
+++ b/src/cpu/ppc/vm/templateTable_ppc_64.cpp	Fri Aug 01 15:57:00 2014 -0700
@@ -189,8 +189,12 @@
       assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
       assert(load_bc_into_bc_reg, "we use bc_reg as temp");
       __ get_cache_and_index_at_bcp(Rtemp /* dst = cache */, 1);
-      // Big Endian: ((*(cache+indices))>>((1+byte_no)*8))&0xFF
+      // ((*(cache+indices))>>((1+byte_no)*8))&0xFF:
+#if defined(VM_LITTLE_ENDIAN)
+      __ lbz(Rnew_bc, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + 1 + byte_no, Rtemp);
+#else
       __ lbz(Rnew_bc, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + 7 - (1 + byte_no), Rtemp);
+#endif
       __ cmpwi(CCR0, Rnew_bc, 0);
       __ li(Rnew_bc, (unsigned int)(unsigned char)new_bc);
       __ beq(CCR0, L_patch_done);
@@ -1839,8 +1843,8 @@
   __ clrrdi(Rdef_offset_addr, Rdef_offset_addr, log2_long((jlong)BytesPerInt));
 
   // Load lo & hi.
-  __ lwz(Rlow_byte, BytesPerInt, Rdef_offset_addr);
-  __ lwz(Rhigh_byte, BytesPerInt * 2, Rdef_offset_addr);
+  __ get_u4(Rlow_byte, Rdef_offset_addr, BytesPerInt, InterpreterMacroAssembler::Unsigned);
+  __ get_u4(Rhigh_byte, Rdef_offset_addr, 2 *BytesPerInt, InterpreterMacroAssembler::Unsigned);
 
   // Check for default case (=index outside [low,high]).
   __ cmpw(CCR0, R17_tos, Rlow_byte);
@@ -1854,12 +1858,17 @@
   __ profile_switch_case(Rindex, Rhigh_byte /* scratch */, Rscratch1, Rscratch2);
   __ sldi(Rindex, Rindex, LogBytesPerInt);
   __ addi(Rindex, Rindex, 3 * BytesPerInt);
+#if defined(VM_LITTLE_ENDIAN)
+  __ lwbrx(Roffset, Rdef_offset_addr, Rindex);
+  __ extsw(Roffset, Roffset);
+#else
   __ lwax(Roffset, Rdef_offset_addr, Rindex);
+#endif
   __ b(Ldispatch);
 
   __ bind(Ldefault_case);
   __ profile_switch_default(Rhigh_byte, Rscratch1);
-  __ lwa(Roffset, 0, Rdef_offset_addr);
+  __ get_u4(Roffset, Rdef_offset_addr, 0, InterpreterMacroAssembler::Signed);
 
   __ bind(Ldispatch);
 
@@ -1875,12 +1884,11 @@
 // Table switch using linear search through cases.
 // Bytecode stream format:
 // Bytecode (1) | 4-byte padding | default offset (4) | count (4) | value/offset pair1 (8) | value/offset pair2 (8) | ...
-// Note: Everything is big-endian format here. So on little endian machines, we have to revers offset and count and cmp value.
+// Note: Everything is big-endian format here.
 void TemplateTable::fast_linearswitch() {
   transition(itos, vtos);
 
-  Label Lloop_entry, Lsearch_loop, Lfound, Lcontinue_execution, Ldefault_case;
-
+  Label Lloop_entry, Lsearch_loop, Lcontinue_execution, Ldefault_case;
   Register Rcount           = R3_ARG1,
            Rcurrent_pair    = R4_ARG2,
            Rdef_offset_addr = R5_ARG3, // Is going to contain address of default offset.
@@ -1894,47 +1902,40 @@
   __ clrrdi(Rdef_offset_addr, Rdef_offset_addr, log2_long((jlong)BytesPerInt));
 
   // Setup loop counter and limit.
-  __ lwz(Rcount, BytesPerInt, Rdef_offset_addr);    // Load count.
+  __ get_u4(Rcount, Rdef_offset_addr, BytesPerInt, InterpreterMacroAssembler::Unsigned);
   __ addi(Rcurrent_pair, Rdef_offset_addr, 2 * BytesPerInt); // Rcurrent_pair now points to first pair.
 
-  // Set up search loop.
+  __ mtctr(Rcount);
   __ cmpwi(CCR0, Rcount, 0);
-  __ beq(CCR0, Ldefault_case);
-
-  __ mtctr(Rcount);
-
-  // linear table search
-  __ bind(Lsearch_loop);
-
-  __ lwz(Rvalue, 0, Rcurrent_pair);
-  __ lwa(Roffset, 1 * BytesPerInt, Rcurrent_pair);
-
-  __ cmpw(CCR0, Rvalue, Rcmp_value);
-  __ beq(CCR0, Lfound);
-
-  __ addi(Rcurrent_pair, Rcurrent_pair, 2 * BytesPerInt);
-  __ bdnz(Lsearch_loop);
-
-  // default case
+  __ bne(CCR0, Lloop_entry);
+
+  // Default case
   __ bind(Ldefault_case);
-
-  __ lwa(Roffset, 0, Rdef_offset_addr);
+  __ get_u4(Roffset, Rdef_offset_addr, 0, InterpreterMacroAssembler::Signed);
   if (ProfileInterpreter) {
     __ profile_switch_default(Rdef_offset_addr, Rcount/* scratch */);
-    __ b(Lcontinue_execution);
   }
-
-  // Entry found, skip Roffset bytecodes and continue.
-  __ bind(Lfound);
+  __ b(Lcontinue_execution);
+
+  // Next iteration
+  __ bind(Lsearch_loop);
+  __ bdz(Ldefault_case);
+  __ addi(Rcurrent_pair, Rcurrent_pair, 2 * BytesPerInt);
+  __ bind(Lloop_entry);
+  __ get_u4(Rvalue, Rcurrent_pair, 0, InterpreterMacroAssembler::Unsigned);
+  __ cmpw(CCR0, Rvalue, Rcmp_value);
+  __ bne(CCR0, Lsearch_loop);
+
+  // Found, load offset.
+  __ get_u4(Roffset, Rcurrent_pair, BytesPerInt, InterpreterMacroAssembler::Signed);
+  // Calculate case index and profile
+  __ mfctr(Rcurrent_pair);
   if (ProfileInterpreter) {
-    // Calc the num of the pair we hit. Careful, Rcurrent_pair points 2 ints
-    // beyond the actual current pair due to the auto update load above!
-    __ sub(Rcurrent_pair, Rcurrent_pair, Rdef_offset_addr);
-    __ addi(Rcurrent_pair, Rcurrent_pair, - 2 * BytesPerInt);
-    __ srdi(Rcurrent_pair, Rcurrent_pair, LogBytesPerInt + 1);
+    __ sub(Rcurrent_pair, Rcount, Rcurrent_pair);
     __ profile_switch_case(Rcurrent_pair, Rcount /*scratch*/, Rdef_offset_addr/*scratch*/, Rscratch);
-    __ bind(Lcontinue_execution);
   }
+
+  __ bind(Lcontinue_execution);
   __ add(R14_bcp, Roffset, R14_bcp);
   __ dispatch_next(vtos);
 }
@@ -1990,7 +1991,7 @@
 
   // initialize i & j
   __ li(Ri,0);
-  __ lwz(Rj, -BytesPerInt, Rarray);
+  __ get_u4(Rj, Rarray, -BytesPerInt, InterpreterMacroAssembler::Unsigned);
 
   // and start.
   Label entry;
@@ -2007,7 +2008,11 @@
     //   i = h;
     // }
     __ sldi(Rscratch, Rh, log_entry_size);
+#if defined(VM_LITTLE_ENDIAN)
+    __ lwbrx(Rscratch, Rscratch, Rarray);
+#else
     __ lwzx(Rscratch, Rscratch, Rarray);
+#endif
 
     // if (key < current value)
     //   Rh = Rj
@@ -2039,20 +2044,20 @@
   // Ri = value offset
   __ sldi(Ri, Ri, log_entry_size);
   __ add(Ri, Ri, Rarray);
-  __ lwz(Rscratch, 0, Ri);
+  __ get_u4(Rscratch, Ri, 0, InterpreterMacroAssembler::Unsigned);
 
   Label not_found;
   // Ri = offset offset
   __ cmpw(CCR0, Rkey, Rscratch);
   __ beq(CCR0, not_found);
   // entry not found -> j = default offset
-  __ lwz(Rj, -2 * BytesPerInt, Rarray);
+  __ get_u4(Rj, Rarray, -2 * BytesPerInt, InterpreterMacroAssembler::Unsigned);
   __ b(default_case);
 
   __ bind(not_found);
   // entry found -> j = offset
   __ profile_switch_case(Rh, Rj, Rscratch, Rkey);
-  __ lwz(Rj, BytesPerInt, Ri);
+  __ get_u4(Rj, Ri, BytesPerInt, InterpreterMacroAssembler::Unsigned);
 
   if (ProfileInterpreter) {
     __ b(continue_execution);
@@ -2147,8 +2152,11 @@
 
   assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
   // We are resolved if the indices offset contains the current bytecode.
-  // Big Endian:
+#if defined(VM_LITTLE_ENDIAN)
+  __ lbz(Rscratch, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + byte_no + 1, Rcache);
+#else
   __ lbz(Rscratch, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()) + 7 - (byte_no + 1), Rcache);
+#endif
   // Acquire by cmp-br-isync (see below).
   __ cmpdi(CCR0, Rscratch, (int)bytecode());
   __ beq(CCR0, Lresolved);
--- a/src/share/vm/opto/callnode.cpp	Thu Jul 31 16:39:57 2014 -0700
+++ b/src/share/vm/opto/callnode.cpp	Fri Aug 01 15:57:00 2014 -0700
@@ -688,7 +688,7 @@
     return new MachProjNode(this,proj->_con,RegMask::Empty,MachProjNode::unmatched_proj);
 
   case TypeFunc::Parms+1:       // For LONG & DOUBLE returns
-    assert(tf()->_range->field_at(TypeFunc::Parms+1) == Type::HALF, "");
+    assert(tf()->range()->field_at(TypeFunc::Parms+1) == Type::HALF, "");
     // 2nd half of doubles and longs
     return new MachProjNode(this,proj->_con, RegMask::Empty, (uint)OptoReg::Bad);
 
--- a/src/share/vm/opto/cfgnode.cpp	Thu Jul 31 16:39:57 2014 -0700
+++ b/src/share/vm/opto/cfgnode.cpp	Fri Aug 01 15:57:00 2014 -0700
@@ -108,6 +108,7 @@
 
         rreq++;                 // One more input to Region
       } // Found a region to merge into Region
+      igvn->_worklist.push(r);
       // Clobber pointer to the now dead 'r'
       region->set_req(i, phase->C->top());
     }
@@ -449,6 +450,7 @@
   // Remove TOP or NULL input paths. If only 1 input path remains, this Region
   // degrades to a copy.
   bool add_to_worklist = false;
+  bool modified = false;
   int cnt = 0;                  // Count of values merging
   DEBUG_ONLY( int cnt_orig = req(); ) // Save original inputs count
   int del_it = 0;               // The last input path we delete
@@ -459,6 +461,7 @@
       // Remove useless control copy inputs
       if( n->is_Region() && n->as_Region()->is_copy() ) {
         set_req(i, n->nonnull_req());
+        modified = true;
         i--;
         continue;
       }
@@ -466,12 +469,14 @@
         Node *call = n->in(0);
         if (call->is_Call() && call->as_Call()->entry_point() == OptoRuntime::rethrow_stub()) {
           set_req(i, call->in(0));
+          modified = true;
           i--;
           continue;
         }
       }
       if( phase->type(n) == Type::TOP ) {
         set_req(i, NULL);       // Ignore TOP inputs
+        modified = true;
         i--;
         continue;
       }
@@ -691,7 +696,7 @@
     }
   }
 
-  return NULL;
+  return modified ? this : NULL;
 }
 
 
--- a/src/share/vm/opto/compile.cpp	Thu Jul 31 16:39:57 2014 -0700
+++ b/src/share/vm/opto/compile.cpp	Fri Aug 01 15:57:00 2014 -0700
@@ -1039,6 +1039,7 @@
 
   _node_note_array = NULL;
   _default_node_notes = NULL;
+  DEBUG_ONLY( _modified_nodes = NULL; ) // Used in Optimize()
 
   _immutable_memory = NULL; // filled in at first inquiry
 
@@ -1247,6 +1248,18 @@
     }
   }
 }
+void Compile::record_modified_node(Node* n) {
+  if (_modified_nodes != NULL && !_inlining_incrementally &&
+      n->outcnt() != 0 && !n->is_Con()) {
+    _modified_nodes->push(n);
+  }
+}
+
+void Compile::remove_modified_node(Node* n) {
+  if (_modified_nodes != NULL) {
+    _modified_nodes->remove(n);
+  }
+}
 #endif
 
 #ifndef PRODUCT
@@ -2035,6 +2048,9 @@
   // Iterative Global Value Numbering, including ideal transforms
   // Initialize IterGVN with types and values from parse-time GVN
   PhaseIterGVN igvn(initial_gvn());
+#ifdef ASSERT
+  _modified_nodes = new (comp_arena()) Unique_Node_List(comp_arena());
+#endif
   {
     NOT_PRODUCT( TracePhase t2("iterGVN", &_t_iterGVN, TimeCompiler); )
     igvn.optimize();
@@ -2197,6 +2213,7 @@
     }
   }
 
+  DEBUG_ONLY( _modified_nodes = NULL; )
  } // (End scope of igvn; run destructor if necessary for asserts.)
 
   process_print_inlining();
@@ -4031,6 +4048,7 @@
   int j = 0;
   int identical = 0;
   int i = 0;
+  bool modified = false;
   for (; i < _expensive_nodes->length()-1; i++) {
     assert(j <= i, "can't write beyond current index");
     if (_expensive_nodes->at(i)->Opcode() == _expensive_nodes->at(i+1)->Opcode()) {
@@ -4043,20 +4061,23 @@
       identical = 0;
     } else {
       Node* n = _expensive_nodes->at(i);
-      igvn.hash_delete(n);
-      n->set_req(0, NULL);
+      igvn.replace_input_of(n, 0, NULL);
       igvn.hash_insert(n);
+      modified = true;
     }
   }
   if (identical > 0) {
     _expensive_nodes->at_put(j++, _expensive_nodes->at(i));
   } else if (_expensive_nodes->length() >= 1) {
     Node* n = _expensive_nodes->at(i);
-    igvn.hash_delete(n);
-    n->set_req(0, NULL);
+    igvn.replace_input_of(n, 0, NULL);
     igvn.hash_insert(n);
+    modified = true;
   }
   _expensive_nodes->trunc_to(j);
+  if (modified) {
+    igvn.optimize();
+  }
 }
 
 void Compile::add_expensive_node(Node * n) {
--- a/src/share/vm/opto/compile.hpp	Thu Jul 31 16:39:57 2014 -0700
+++ b/src/share/vm/opto/compile.hpp	Fri Aug 01 15:57:00 2014 -0700
@@ -344,6 +344,8 @@
   VectorSet             _dead_node_list;        // Set of dead nodes
   uint                  _dead_node_count;       // Number of dead nodes; VectorSet::Size() is O(N).
                                                 // So use this to keep count and make the call O(1).
+  DEBUG_ONLY( Unique_Node_List* _modified_nodes; )  // List of nodes which inputs were modified
+
   debug_only(static int _debug_idx;)            // Monotonic counter (not reset), use -XX:BreakAtNode=<idx>
   Arena                 _node_arena;            // Arena for new-space Nodes
   Arena                 _old_arena;             // Arena for old-space Nodes, lifetime during xform
@@ -766,6 +768,11 @@
   void         print_missing_nodes();
 #endif
 
+  // Record modified nodes to check that they are put on IGVN worklist
+  void         record_modified_node(Node* n) NOT_DEBUG_RETURN;
+  void         remove_modified_node(Node* n) NOT_DEBUG_RETURN;
+  DEBUG_ONLY( Unique_Node_List*   modified_nodes() const { return _modified_nodes; } )
+
   // Constant table
   ConstantTable&   constant_table() { return _constant_table; }
 
--- a/src/share/vm/opto/divnode.cpp	Thu Jul 31 16:39:57 2014 -0700
+++ b/src/share/vm/opto/divnode.cpp	Fri Aug 01 15:57:00 2014 -0700
@@ -479,7 +479,10 @@
 
   if (i == 0) return NULL;      // Dividing by zero constant does not idealize
 
-  set_req(0,NULL);              // Dividing by a not-zero constant; no faulting
+  if (in(0) != NULL) {
+    phase->igvn_rehash_node_delayed(this);
+    set_req(0, NULL);           // Dividing by a not-zero constant; no faulting
+  }
 
   // Dividing by MININT does not optimize as a power-of-2 shift.
   if( i == min_jint ) return NULL;
@@ -578,7 +581,10 @@
 
   if (l == 0) return NULL;      // Dividing by zero constant does not idealize
 
-  set_req(0,NULL);              // Dividing by a not-zero constant; no faulting
+  if (in(0) != NULL) {
+    phase->igvn_rehash_node_delayed(this);
+    set_req(0, NULL);           // Dividing by a not-zero constant; no faulting
+  }
 
   // Dividing by MINLONG does not optimize as a power-of-2 shift.
   if( l == min_jlong ) return NULL;
--- a/src/share/vm/opto/graphKit.cpp	Thu Jul 31 16:39:57 2014 -0700
+++ b/src/share/vm/opto/graphKit.cpp	Fri Aug 01 15:57:00 2014 -0700
@@ -2084,9 +2084,9 @@
 void GraphKit::round_double_arguments(ciMethod* dest_method) {
   // (Note:  TypeFunc::make has a cache that makes this fast.)
   const TypeFunc* tf    = TypeFunc::make(dest_method);
-  int             nargs = tf->_domain->_cnt - TypeFunc::Parms;
+  int             nargs = tf->domain()->cnt() - TypeFunc::Parms;
   for (int j = 0; j < nargs; j++) {
-    const Type *targ = tf->_domain->field_at(j + TypeFunc::Parms);
+    const Type *targ = tf->domain()->field_at(j + TypeFunc::Parms);
     if( targ->basic_type() == T_DOUBLE ) {
       // If any parameters are doubles, they must be rounded before
       // the call, dstore_rounding does gvn.transform
@@ -2188,10 +2188,10 @@
     return;
   }
   const TypeFunc* tf    = TypeFunc::make(dest_method);
-  int             nargs = tf->_domain->_cnt - TypeFunc::Parms;
+  int             nargs = tf->domain()->cnt() - TypeFunc::Parms;
   int skip = Bytecodes::has_receiver(bc) ? 1 : 0;
   for (int j = skip, i = 0; j < nargs && i < TypeProfileArgsLimit; j++) {
-    const Type *targ = tf->_domain->field_at(j + TypeFunc::Parms);
+    const Type *targ = tf->domain()->field_at(j + TypeFunc::Parms);
     if (targ->basic_type() == T_OBJECT || targ->basic_type() == T_ARRAY) {
       bool maybe_null = true;
       ciKlass* better_type = NULL;
--- a/src/share/vm/opto/loopPredicate.cpp	Thu Jul 31 16:39:57 2014 -0700
+++ b/src/share/vm/opto/loopPredicate.cpp	Fri Aug 01 15:57:00 2014 -0700
@@ -107,8 +107,7 @@
     rgn = new RegionNode(1);
     rgn->add_req(uncommon_proj);
     register_control(rgn, loop, uncommon_proj);
-    _igvn.hash_delete(call);
-    call->set_req(0, rgn);
+    _igvn.replace_input_of(call, 0, rgn);
     // When called from beautify_loops() idom is not constructed yet.
     if (_idom != NULL) {
       set_idom(call, rgn, dom_depth(rgn));
@@ -166,8 +165,7 @@
 
   if (new_entry == NULL) {
     // Attach if_cont to iff
-    _igvn.hash_delete(iff);
-    iff->set_req(0, if_cont);
+    _igvn.replace_input_of(iff, 0, if_cont);
     if (_idom != NULL) {
       set_idom(iff, if_cont, dom_depth(iff));
     }
@@ -194,8 +192,7 @@
     rgn = new RegionNode(1);
     register_new_node_with_optimizer(rgn);
     rgn->add_req(uncommon_proj);
-    hash_delete(call);
-    call->set_req(0, rgn);
+    replace_input_of(call, 0, rgn);
   } else {
     // Find region's edge corresponding to uncommon_proj
     for (; proj_index < rgn->req(); proj_index++)
--- a/src/share/vm/opto/loopTransform.cpp	Thu Jul 31 16:39:57 2014 -0700
+++ b/src/share/vm/opto/loopTransform.cpp	Fri Aug 01 15:57:00 2014 -0700
@@ -924,15 +924,13 @@
   if( bol->outcnt() != 1 ) {
     bol = bol->clone();
     register_new_node(bol,main_end->in(CountedLoopEndNode::TestControl));
-    _igvn.hash_delete(main_end);
-    main_end->set_req(CountedLoopEndNode::TestValue, bol);
+    _igvn.replace_input_of(main_end, CountedLoopEndNode::TestValue, bol);
   }
   // Need only 1 user of 'cmp' because I will be hacking the loop bounds.
   if( cmp->outcnt() != 1 ) {
     cmp = cmp->clone();
     register_new_node(cmp,main_end->in(CountedLoopEndNode::TestControl));
-    _igvn.hash_delete(bol);
-    bol->set_req(1, cmp);
+    _igvn.replace_input_of(bol, 1, cmp);
   }
 
   //------------------------------
@@ -1118,8 +1116,7 @@
     Node* pre_bol = pre_end->in(CountedLoopEndNode::TestValue)->as_Bool();
     BoolNode* new_bol0 = new BoolNode(pre_bol->in(1), new_test);
     register_new_node( new_bol0, pre_head->in(0) );
-    _igvn.hash_delete(pre_end);
-    pre_end->set_req(CountedLoopEndNode::TestValue, new_bol0);
+    _igvn.replace_input_of(pre_end, CountedLoopEndNode::TestValue, new_bol0);
     // Modify main loop guard condition
     assert(min_iff->in(CountedLoopEndNode::TestValue) == min_bol, "guard okay");
     BoolNode* new_bol1 = new BoolNode(min_bol->in(1), new_test);
@@ -1130,8 +1127,7 @@
     BoolNode* main_bol = main_end->in(CountedLoopEndNode::TestValue)->as_Bool();
     BoolNode* new_bol2 = new BoolNode(main_bol->in(1), new_test);
     register_new_node( new_bol2, main_end->in(CountedLoopEndNode::TestControl) );
-    _igvn.hash_delete(main_end);
-    main_end->set_req(CountedLoopEndNode::TestValue, new_bol2);
+    _igvn.replace_input_of(main_end, CountedLoopEndNode::TestValue, new_bol2);
   }
 
   // Flag main loop
@@ -1346,8 +1342,7 @@
         Node* bol2 = loop_end->in(1)->clone();
         bol2->set_req(1, cmp2);
         register_new_node(bol2, ctrl2);
-        _igvn.hash_delete(loop_end);
-        loop_end->set_req(1, bol2);
+        _igvn.replace_input_of(loop_end, 1, bol2);
       }
       // Step 3: Find the min-trip test guaranteed before a 'main' loop.
       // Make it a 1-trip test (means at least 2 trips).
@@ -1356,8 +1351,7 @@
       // can edit it's inputs directly.  Hammer in the new limit for the
       // minimum-trip guard.
       assert(opaq->outcnt() == 1, "");
-      _igvn.hash_delete(opaq);
-      opaq->set_req(1, new_limit);
+      _igvn.replace_input_of(opaq, 1, new_limit);
     }
 
     // Adjust max trip count. The trip count is intentionally rounded
@@ -1407,8 +1401,7 @@
     register_new_node( cmp2, ctrl2 );
     Node *bol2 = new BoolNode( cmp2, loop_end->test_trip() );
     register_new_node( bol2, ctrl2 );
-    _igvn.hash_delete(loop_end);
-    loop_end->set_req(CountedLoopEndNode::TestValue, bol2);
+    _igvn.replace_input_of(loop_end, CountedLoopEndNode::TestValue, bol2);
 
     // Step 3: Find the min-trip test guaranteed before a 'main' loop.
     // Make it a 1-trip test (means at least 2 trips).
@@ -1997,8 +1990,7 @@
                                  : (Node*)new MaxINode(pre_limit, orig_limit);
     register_new_node(pre_limit, pre_ctrl);
   }
-  _igvn.hash_delete(pre_opaq);
-  pre_opaq->set_req(1, pre_limit);
+  _igvn.replace_input_of(pre_opaq, 1, pre_limit);
 
   // Note:: we are making the main loop limit no longer precise;
   // need to round up based on stride.
--- a/src/share/vm/opto/loopnode.cpp	Thu Jul 31 16:39:57 2014 -0700
+++ b/src/share/vm/opto/loopnode.cpp	Fri Aug 01 15:57:00 2014 -0700
@@ -133,7 +133,7 @@
   // Return earliest legal location
   assert(early == find_non_split_ctrl(early), "unexpected early control");
 
-  if (n->is_expensive()) {
+  if (n->is_expensive() && !_verify_only && !_verify_me) {
     assert(n->in(0), "should have control input");
     early = get_early_ctrl_for_expensive(n, early);
   }
@@ -226,8 +226,7 @@
   }
 
   if (ctl != n->in(0)) {
-    _igvn.hash_delete(n);
-    n->set_req(0, ctl);
+    _igvn.replace_input_of(n, 0, ctl);
     _igvn.hash_insert(n);
   }
 
@@ -521,8 +520,7 @@
     assert(check_iff->in(1)->Opcode() == Op_Conv2B &&
            check_iff->in(1)->in(1)->Opcode() == Op_Opaque1, "");
     Node* opq = check_iff->in(1)->in(1);
-    _igvn.hash_delete(opq);
-    opq->set_req(1, bol);
+    _igvn.replace_input_of(opq, 1, bol);
     // Update ctrl.
     set_ctrl(opq, check_iff->in(0));
     set_ctrl(check_iff->in(1), check_iff->in(0));
@@ -690,7 +688,7 @@
   incr->set_req(2,stride);
   incr = _igvn.register_new_node_with_optimizer(incr);
   set_early_ctrl( incr );
-  _igvn.hash_delete(phi);
+  _igvn.rehash_node_delayed(phi);
   phi->set_req_X( LoopNode::LoopBackControl, incr, &_igvn );
 
   // If phi type is more restrictive than Int, raise to
@@ -743,8 +741,8 @@
     iffalse = iff2;
     iftrue  = ift2;
   } else {
-    _igvn.hash_delete(iffalse);
-    _igvn.hash_delete(iftrue);
+    _igvn.rehash_node_delayed(iffalse);
+    _igvn.rehash_node_delayed(iftrue);
     iffalse->set_req_X( 0, le, &_igvn );
     iftrue ->set_req_X( 0, le, &_igvn );
   }
@@ -1257,6 +1255,7 @@
       _head->del_req(i);
     }
   }
+  igvn.rehash_node_delayed(_head);
   // Transform landing pad
   igvn.register_new_node_with_optimizer(landing_pad, _head);
   // Insert landing pad into the header
@@ -1397,7 +1396,7 @@
   igvn.register_new_node_with_optimizer(r, _head);
   // Plug region into end of loop _head, followed by hot_tail
   while( _head->req() > 3 ) _head->del_req( _head->req()-1 );
-  _head->set_req(2, r);
+  igvn.replace_input_of(_head, 2, r);
   if( hot_idx ) _head->add_req(hot_tail);
 
   // Split all the Phis up between '_head' loop and the Region 'r'
@@ -1419,7 +1418,7 @@
       igvn.register_new_node_with_optimizer(phi, n);
       // Add the merge phi to the old Phi
       while( n->req() > 3 ) n->del_req( n->req()-1 );
-      n->set_req(2, phi);
+      igvn.replace_input_of(n, 2, phi);
       if( hot_idx ) n->add_req(hot_phi);
     }
   }
@@ -1495,13 +1494,14 @@
   if( fall_in_cnt > 1 ) {
     // Since I am just swapping inputs I do not need to update def-use info
     Node *tmp = _head->in(1);
+    igvn.rehash_node_delayed(_head);
     _head->set_req( 1, _head->in(fall_in_cnt) );
     _head->set_req( fall_in_cnt, tmp );
     // Swap also all Phis
     for (DUIterator_Fast imax, i = _head->fast_outs(imax); i < imax; i++) {
       Node* phi = _head->fast_out(i);
       if( phi->is_Phi() ) {
-        igvn.hash_delete(phi); // Yank from hash before hacking edges
+        igvn.rehash_node_delayed(phi); // Yank from hash before hacking edges
         tmp = phi->in(1);
         phi->set_req( 1, phi->in(fall_in_cnt) );
         phi->set_req( fall_in_cnt, tmp );
@@ -2905,6 +2905,7 @@
           uint k = 0;             // Probably cfg->in(0)
           while( cfg->in(k) != m ) k++; // But check incase cfg is a Region
           cfg->set_req( k, if_t ); // Now point to NeverBranch
+          _igvn._worklist.push(cfg);
 
           // Now create the never-taken loop exit
           Node *if_f = new CProjNode( iff, 1 );
--- a/src/share/vm/opto/loopopts.cpp	Thu Jul 31 16:39:57 2014 -0700
+++ b/src/share/vm/opto/loopopts.cpp	Fri Aug 01 15:57:00 2014 -0700
@@ -2574,7 +2574,7 @@
   new_head->set_unswitch_count(head->unswitch_count()); // Preserve
   _igvn.register_new_node_with_optimizer(new_head);
   assert(first_not_peeled->in(0) == last_peel, "last_peel <- first_not_peeled");
-  first_not_peeled->set_req(0, new_head);
+  _igvn.replace_input_of(first_not_peeled, 0, new_head);
   set_loop(new_head, loop);
   loop->_body.push(new_head);
   not_peel.set(new_head->_idx);
--- a/src/share/vm/opto/macro.cpp	Thu Jul 31 16:39:57 2014 -0700
+++ b/src/share/vm/opto/macro.cpp	Fri Aug 01 15:57:00 2014 -0700
@@ -855,6 +855,7 @@
     int start = jvms->debug_start();
     int end   = jvms->debug_end();
     sfpt->replace_edges_in_range(res, sobj, start, end);
+    _igvn._worklist.push(sfpt);
     safepoints_done.append_if_missing(sfpt); // keep it for rollback
   }
   return true;
@@ -1775,6 +1776,7 @@
       Node *pf_region = new RegionNode(3);
       Node *pf_phi_rawmem = new PhiNode( pf_region, Type::MEMORY,
                                              TypeRawPtr::BOTTOM );
+      transform_later(pf_region);
 
       // Generate several prefetch instructions.
       uint lines = (length != NULL) ? AllocatePrefetchLines : AllocateInstancePrefetchLines;
--- a/src/share/vm/opto/memnode.cpp	Thu Jul 31 16:39:57 2014 -0700
+++ b/src/share/vm/opto/memnode.cpp	Fri Aug 01 15:57:00 2014 -0700
@@ -1471,6 +1471,7 @@
 
   Node* ctrl    = in(MemNode::Control);
   Node* address = in(MemNode::Address);
+  bool progress = false;
 
   // Skip up past a SafePoint control.  Cannot do this for Stores because
   // pointer stores & cardmarks must stay on the same side of a SafePoint.
@@ -1478,6 +1479,7 @@
       phase->C->get_alias_index(phase->type(address)->is_ptr()) != Compile::AliasIdxRaw ) {
     ctrl = ctrl->in(0);
     set_req(MemNode::Control,ctrl);
+    progress = true;
   }
 
   intptr_t ignore = 0;
@@ -1490,6 +1492,7 @@
         && all_controls_dominate(base, phase->C->start())) {
       // A method-invariant, non-null address (constant or 'this' argument).
       set_req(MemNode::Control, NULL);
+      progress = true;
     }
   }
 
@@ -1550,7 +1553,7 @@
     }
   }
 
-  return NULL;                  // No further progress
+  return progress ? this : NULL;
 }
 
 // Helper to recognize certain Klass fields which are invariant across
@@ -2944,6 +2947,7 @@
     return NULL;
   }
 
+  bool progress = false;
   // Eliminate volatile MemBars for scalar replaced objects.
   if (can_reshape && req() == (Precedent+1)) {
     bool eliminate = false;
@@ -2966,6 +2970,7 @@
           phase->is_IterGVN()->_worklist.push(my_mem); // remove dead node later
           my_mem = NULL;
         }
+        progress = true;
       }
       if (my_mem != NULL && my_mem->is_Mem()) {
         const TypeOopPtr* t_oop = my_mem->in(MemNode::Address)->bottom_type()->isa_oopptr();
@@ -2995,7 +3000,7 @@
       return new ConINode(TypeInt::ZERO);
     }
   }
-  return NULL;
+  return progress ? this : NULL;
 }
 
 //------------------------------Value------------------------------------------
@@ -3497,6 +3502,7 @@
   // if it redundantly stored the same value (or zero to fresh memory).
 
   // In any case, wire it in:
+  phase->igvn_rehash_node_delayed(this);
   set_req(i, new_st);
 
   // The caller may now kill the old guy.
--- a/src/share/vm/opto/node.cpp	Thu Jul 31 16:39:57 2014 -0700
+++ b/src/share/vm/opto/node.cpp	Fri Aug 01 15:57:00 2014 -0700
@@ -620,6 +620,7 @@
   *(address*)this = badAddress;  // smash the C++ vtbl, probably
   _in = _out = (Node**) badAddress;
   _max = _cnt = _outmax = _outcnt = 0;
+  compile->remove_modified_node(this);
 #endif
 }
 
@@ -765,6 +766,7 @@
   if (n != NULL) n->del_out((Node *)this);
   _in[idx] = in(--_cnt);  // Compact the array
   _in[_cnt] = NULL;       // NULL out emptied slot
+  Compile::current()->record_modified_node(this);
 }
 
 //------------------------------del_req_ordered--------------------------------
@@ -780,6 +782,7 @@
     Copy::conjoint_words_to_lower((HeapWord*)&_in[idx+1], (HeapWord*)&_in[idx], ((_cnt-idx-1)*sizeof(Node*)));
   }
   _in[--_cnt] = NULL;   // NULL out emptied slot
+  Compile::current()->record_modified_node(this);
 }
 
 //------------------------------ins_req----------------------------------------
@@ -1297,6 +1300,7 @@
       // Done with outputs.
       igvn->hash_delete(dead);
       igvn->_worklist.remove(dead);
+      igvn->C->remove_modified_node(dead);
       igvn->set_type(dead, Type::TOP);
       if (dead->is_macro()) {
         igvn->C->remove_macro_node(dead);
--- a/src/share/vm/opto/node.hpp	Thu Jul 31 16:39:57 2014 -0700
+++ b/src/share/vm/opto/node.hpp	Fri Aug 01 15:57:00 2014 -0700
@@ -398,6 +398,7 @@
     if (*p != NULL)  (*p)->del_out((Node *)this);
     (*p) = n;
     if (n != NULL)      n->add_out((Node *)this);
+    Compile::current()->record_modified_node(this);
   }
   // Light version of set_req() to init inputs after node creation.
   void init_req( uint i, Node *n ) {
@@ -409,6 +410,7 @@
     assert( _in[i] == NULL, "sanity");
     _in[i] = n;
     if (n != NULL)      n->add_out((Node *)this);
+    Compile::current()->record_modified_node(this);
   }
   // Find first occurrence of n among my edges:
   int find_edge(Node* n);
--- a/src/share/vm/opto/output.cpp	Thu Jul 31 16:39:57 2014 -0700
+++ b/src/share/vm/opto/output.cpp	Fri Aug 01 15:57:00 2014 -0700
@@ -785,9 +785,10 @@
     // grow downwards in all implementations.
     // (If, on some machine, the interpreter's Java locals or stack
     // were to grow upwards, the embedded doubles would be word-swapped.)
-    jint   *dp = (jint*)&d;
-    array->append(new ConstantIntValue(dp[1]));
-    array->append(new ConstantIntValue(dp[0]));
+    jlong_accessor acc;
+    acc.long_value = jlong_cast(d);
+    array->append(new ConstantIntValue(acc.words[1]));
+    array->append(new ConstantIntValue(acc.words[0]));
 #endif
     break;
   }
@@ -804,9 +805,10 @@
     // grow downwards in all implementations.
     // (If, on some machine, the interpreter's Java locals or stack
     // were to grow upwards, the embedded doubles would be word-swapped.)
-    jint *dp = (jint*)&d;
-    array->append(new ConstantIntValue(dp[1]));
-    array->append(new ConstantIntValue(dp[0]));
+    jlong_accessor acc;
+    acc.long_value = d;
+    array->append(new ConstantIntValue(acc.words[1]));
+    array->append(new ConstantIntValue(acc.words[0]));
 #endif
     break;
   }
--- a/src/share/vm/opto/parse1.cpp	Thu Jul 31 16:39:57 2014 -0700
+++ b/src/share/vm/opto/parse1.cpp	Fri Aug 01 15:57:00 2014 -0700
@@ -575,12 +575,13 @@
       decrement_age();
     }
   }
-  if (depth() == 1) {
+
+  if (depth() == 1 && !failing()) {
     // Add check to deoptimize the nmethod if RTM state was changed
     rtm_deopt();
   }
 
-  // Check for bailouts during method entry.
+  // Check for bailouts during method entry or RTM state check setup.
   if (failing()) {
     if (log)  log->done("parse");
     C->set_default_node_notes(caller_nn);
--- a/src/share/vm/opto/phaseX.cpp	Thu Jul 31 16:39:57 2014 -0700
+++ b/src/share/vm/opto/phaseX.cpp	Fri Aug 01 15:57:00 2014 -0700
@@ -933,9 +933,32 @@
   for (int i = 0; i < _verify_window_size; i++) {
     _verify_window[i] = NULL;
   }
+#ifdef ASSERT
+  // Verify that all modified nodes are on _worklist
+  Unique_Node_List* modified_list = C->modified_nodes();
+  while (modified_list != NULL && modified_list->size()) {
+    Node* n = modified_list->pop();
+    if (n->outcnt() != 0 && !n->is_Con() && !_worklist.member(n)) {
+      n->dump();
+      assert(false, "modified node is not on IGVN._worklist");
+    }
+  }
+#endif
 }
 
 void PhaseIterGVN::verify_PhaseIterGVN() {
+#ifdef ASSERT
+  // Verify nodes with changed inputs.
+  Unique_Node_List* modified_list = C->modified_nodes();
+  while (modified_list != NULL && modified_list->size()) {
+    Node* n = modified_list->pop();
+    if (n->outcnt() != 0 && !n->is_Con()) { // skip dead and Con nodes
+      n->dump();
+      assert(false, "modified node was not processed by IGVN.transform_old()");
+    }
+  }
+#endif
+
   C->verify_graph_edges();
   if( VerifyOpto && allow_progress() ) {
     // Must turn off allow_progress to enable assert and break recursion
@@ -964,6 +987,14 @@
                   (int) _verify_counter, (int) _verify_full_passes);
     }
   }
+
+#ifdef ASSERT
+  while (modified_list->size()) {
+    Node* n = modified_list->pop();
+    n->dump();
+    assert(false, "VerifyIterativeGVN: new modified node was added");
+  }
+#endif
 }
 #endif /* PRODUCT */
 
@@ -1066,6 +1097,7 @@
   Node* k = n;
   DEBUG_ONLY(dead_loop_check(k);)
   DEBUG_ONLY(bool is_new = (k->outcnt() == 0);)
+  C->remove_modified_node(k);
   Node* i = k->Ideal(this, /*can_reshape=*/true);
   assert(i != k || is_new || i->outcnt() > 0, "don't return dead nodes");
 #ifndef PRODUCT
@@ -1107,6 +1139,7 @@
     DEBUG_ONLY(dead_loop_check(k);)
     // Try idealizing again
     DEBUG_ONLY(is_new = (k->outcnt() == 0);)
+    C->remove_modified_node(k);
     i = k->Ideal(this, /*can_reshape=*/true);
     assert(i != k || is_new || (i->outcnt() > 0), "don't return dead nodes");
 #ifndef PRODUCT
@@ -1259,6 +1292,7 @@
       _stack.pop();
       // Remove dead node from iterative worklist
       _worklist.remove(dead);
+      C->remove_modified_node(dead);
       // Constant node that has no out-edges and has only one in-edge from
       // root is usually dead. However, sometimes reshaping walk makes
       // it reachable by adding use edges. So, we will NOT count Con nodes
@@ -1288,7 +1322,7 @@
   for (DUIterator_Last imin, i = old->last_outs(imin); i >= imin; ) {
     Node* use = old->last_out(i);  // for each use...
     // use might need re-hashing (but it won't if it's a new node)
-    bool is_in_table = _table.hash_delete( use );
+    rehash_node_delayed(use);
     // Update use-def info as well
     // We remove all occurrences of old within use->in,
     // so as to avoid rehashing any node more than once.
@@ -1300,11 +1334,6 @@
         ++num_edges;
       }
     }
-    // Insert into GVN hash table if unique
-    // If a duplicate, 'use' will be cleaned up when pulled off worklist
-    if( is_in_table ) {
-      hash_find_insert(use);
-    }
     i -= num_edges;    // we deleted 1 or more copies of this edge
   }
 
--- a/src/share/vm/opto/phaseX.hpp	Thu Jul 31 16:39:57 2014 -0700
+++ b/src/share/vm/opto/phaseX.hpp	Fri Aug 01 15:57:00 2014 -0700
@@ -311,6 +311,9 @@
                                const Type* limit_type) const
   { ShouldNotCallThis(); return NULL; }
 
+  // Delayed node rehash if this is an IGVN phase
+  virtual void igvn_rehash_node_delayed(Node* n) {}
+
 #ifndef PRODUCT
   void dump_old2new_map() const;
   void dump_new( uint new_lidx ) const;
@@ -488,6 +491,10 @@
     _worklist.push(n);
   }
 
+  void igvn_rehash_node_delayed(Node* n) {
+    rehash_node_delayed(n);
+  }
+
   // Replace ith edge of "n" with "in"
   void replace_input_of(Node* n, int i, Node* in) {
     rehash_node_delayed(n);
--- a/src/share/vm/opto/rootnode.cpp	Thu Jul 31 16:39:57 2014 -0700
+++ b/src/share/vm/opto/rootnode.cpp	Fri Aug 01 15:57:00 2014 -0700
@@ -35,10 +35,12 @@
 //------------------------------Ideal------------------------------------------
 // Remove dead inputs
 Node *RootNode::Ideal(PhaseGVN *phase, bool can_reshape) {
+  bool modified = false;
   for( uint i = 1; i < req(); i++ ) { // For all inputs
     // Check for and remove dead inputs
     if( phase->type(in(i)) == Type::TOP ) {
       del_req(i--);             // Delete TOP inputs
+      modified = true;
     }
   }
 
@@ -56,7 +58,7 @@
   // If we want to get the rest of the win later, we should pattern match
   // simple recursive call trees to closed-form solutions.
 
-  return NULL;                  // No further opportunities exposed
+  return modified ? this : NULL;
 }
 
 //=============================================================================
--- a/src/share/vm/opto/type.cpp	Thu Jul 31 16:39:57 2014 -0700
+++ b/src/share/vm/opto/type.cpp	Fri Aug 01 15:57:00 2014 -0700
@@ -5087,11 +5087,11 @@
 // Dump Function Type
 #ifndef PRODUCT
 void TypeFunc::dump2( Dict &d, uint depth, outputStream *st ) const {
-  if( _range->_cnt <= Parms )
+  if( _range->cnt() <= Parms )
     st->print("void");
   else {
     uint i;
-    for (i = Parms; i < _range->_cnt-1; i++) {
+    for (i = Parms; i < _range->cnt()-1; i++) {
       _range->field_at(i)->dump2(d,depth,st);
       st->print("/");
     }
@@ -5104,9 +5104,9 @@
     return;
   }
   d.Insert((void*)this,(void*)this);    // Stop recursion
-  if (Parms < _domain->_cnt)
+  if (Parms < _domain->cnt())
     _domain->field_at(Parms)->dump2(d,depth-1,st);
-  for (uint i = Parms+1; i < _domain->_cnt; i++) {
+  for (uint i = Parms+1; i < _domain->cnt(); i++) {
     st->print(", ");
     _domain->field_at(i)->dump2(d,depth-1,st);
   }
--- a/src/share/vm/opto/type.hpp	Thu Jul 31 16:39:57 2014 -0700
+++ b/src/share/vm/opto/type.hpp	Fri Aug 01 15:57:00 2014 -0700
@@ -609,16 +609,16 @@
 // signature types.
 class TypeTuple : public Type {
   TypeTuple( uint cnt, const Type **fields ) : Type(Tuple), _cnt(cnt), _fields(fields) { }
+
+  const uint          _cnt;              // Count of fields
+  const Type ** const _fields;           // Array of field types
+
 public:
   virtual bool eq( const Type *t ) const;
   virtual int  hash() const;             // Type specific hashing
   virtual bool singleton(void) const;    // TRUE if type is a singleton
   virtual bool empty(void) const;        // TRUE if type is vacuous
 
-public:
-  const uint          _cnt;              // Count of fields
-  const Type ** const _fields;           // Array of field types
-
   // Accessors:
   uint cnt() const { return _cnt; }
   const Type* field_at(uint i) const {
@@ -1447,6 +1447,10 @@
   virtual int  hash() const;             // Type specific hashing
   virtual bool singleton(void) const;    // TRUE if type is a singleton
   virtual bool empty(void) const;        // TRUE if type is vacuous
+
+  const TypeTuple* const _domain;     // Domain of inputs
+  const TypeTuple* const _range;      // Range of results
+
 public:
   // Constants are shared among ADLC and VM
   enum { Control    = AdlcVMDeps::Control,
@@ -1457,8 +1461,6 @@
          Parms      = AdlcVMDeps::Parms
   };
 
-  const TypeTuple* const _domain;     // Domain of inputs
-  const TypeTuple* const _range;      // Range of results
 
   // Accessors:
   const TypeTuple* domain() const { return _domain; }
--- a/src/share/vm/runtime/deoptimization.cpp	Thu Jul 31 16:39:57 2014 -0700
+++ b/src/share/vm/runtime/deoptimization.cpp	Fri Aug 01 15:57:00 2014 -0700
@@ -661,7 +661,7 @@
              (iframe->interpreter_frame_expression_stack_size() == (next_mask_expression_stack_size -
                                                                     top_frame_expression_stack_adjustment))) ||
             (is_top_frame && (exec_mode == Unpack_exception) && iframe->interpreter_frame_expression_stack_size() == 0) ||
-            (is_top_frame && (exec_mode == Unpack_uncommon_trap || exec_mode == Unpack_reexecute) &&
+            (is_top_frame && (exec_mode == Unpack_uncommon_trap || exec_mode == Unpack_reexecute || el->should_reexecute()) &&
              (iframe->interpreter_frame_expression_stack_size() == mask.expression_stack_size() + cur_invoke_parameter_size))
             )) {
         ttyLocker ttyl;
--- a/src/share/vm/runtime/sharedRuntimeMath.hpp	Thu Jul 31 16:39:57 2014 -0700
+++ b/src/share/vm/runtime/sharedRuntimeMath.hpp	Fri Aug 01 15:57:00 2014 -0700
@@ -27,20 +27,51 @@
 
 #include <math.h>
 
-// VM_LITTLE_ENDIAN is #defined appropriately in the Makefiles
-// [jk] this is not 100% correct because the float word order may different
-// from the byte order (e.g. on ARM FPA)
+// Used to access the lower/higher 32 bits of a double
+typedef union {
+    double d;
+    struct {
 #ifdef VM_LITTLE_ENDIAN
-# define __HI(x) *(1+(int*)&x)
-# define __LO(x) *(int*)&x
+      int lo;
+      int hi;
 #else
-# define __HI(x) *(int*)&x
-# define __LO(x) *(1+(int*)&x)
+      int hi;
+      int lo;
 #endif
+    } split;
+} DoubleIntConv;
+
+static inline int high(double d) {
+  DoubleIntConv x;
+  x.d = d;
+  return x.split.hi;
+}
+
+static inline int low(double d) {
+  DoubleIntConv x;
+  x.d = d;
+  return x.split.lo;
+}
+
+static inline void set_high(double* d, int high) {
+  DoubleIntConv conv;
+  conv.d = *d;
+  conv.split.hi = high;
+  *d = conv.d;
+}
+
+static inline void set_low(double* d, int low) {
+  DoubleIntConv conv;
+  conv.d = *d;
+  conv.split.lo = low;
+  *d = conv.d;
+}
 
 static double copysignA(double x, double y) {
-  __HI(x) = (__HI(x)&0x7fffffff)|(__HI(y)&0x80000000);
-  return x;
+  DoubleIntConv convX;
+  convX.d = x;
+  convX.split.hi = (convX.split.hi & 0x7fffffff) | (high(y) & 0x80000000);
+  return convX.d;
 }
 
 /*
@@ -67,30 +98,32 @@
 hugeX  = 1.0e+300,
 tiny   = 1.0e-300;
 
-static double scalbnA (double x, int n) {
+static double scalbnA(double x, int n) {
   int  k,hx,lx;
-  hx = __HI(x);
-  lx = __LO(x);
+  hx = high(x);
+  lx = low(x);
   k = (hx&0x7ff00000)>>20;              /* extract exponent */
   if (k==0) {                           /* 0 or subnormal x */
     if ((lx|(hx&0x7fffffff))==0) return x; /* +-0 */
     x *= two54;
-    hx = __HI(x);
+    hx = high(x);
     k = ((hx&0x7ff00000)>>20) - 54;
     if (n< -50000) return tiny*x;       /*underflow*/
   }
   if (k==0x7ff) return x+x;             /* NaN or Inf */
   k = k+n;
-  if (k >  0x7fe) return hugeX*copysignA(hugeX,x); /* overflow  */
-  if (k > 0)                            /* normal result */
-    {__HI(x) = (hx&0x800fffff)|(k<<20); return x;}
+  if (k > 0x7fe) return hugeX*copysignA(hugeX,x); /* overflow  */
+  if (k > 0) {                          /* normal result */
+    set_high(&x, (hx&0x800fffff)|(k<<20));
+    return x;
+  }
   if (k <= -54) {
     if (n > 50000)      /* in case integer overflow in n+k */
       return hugeX*copysignA(hugeX,x);  /*overflow*/
     else return tiny*copysignA(tiny,x); /*underflow*/
   }
   k += 54;                              /* subnormal result */
-  __HI(x) = (hx&0x800fffff)|(k<<20);
+  set_high(&x, (hx&0x800fffff)|(k<<20));
   return x*twom54;
 }
 
--- a/src/share/vm/runtime/sharedRuntimeTrans.cpp	Thu Jul 31 16:39:57 2014 -0700
+++ b/src/share/vm/runtime/sharedRuntimeTrans.cpp	Fri Aug 01 15:57:00 2014 -0700
@@ -40,6 +40,7 @@
 // generated; can not figure out how to turn down optimization for one
 // file in the IDE on Windows
 #ifdef WIN32
+# pragma warning( disable: 4748 ) // /GS can not protect parameters and local variables from local buffer overrun because optimizations are disabled in function
 # pragma optimize ( "", off )
 #endif
 
@@ -114,8 +115,8 @@
   int k,hx,i,j;
   unsigned lx;
 
-  hx = __HI(x);               /* high word of x */
-  lx = __LO(x);               /* low  word of x */
+  hx = high(x);               /* high word of x */
+  lx = low(x);                /* low  word of x */
 
   k=0;
   if (hx < 0x00100000) {                   /* x < 2**-1022  */
@@ -123,13 +124,13 @@
       return -two54/zero;             /* log(+-0)=-inf */
     if (hx<0) return (x-x)/zero;   /* log(-#) = NaN */
     k -= 54; x *= two54; /* subnormal number, scale up x */
-    hx = __HI(x);             /* high word of x */
+    hx = high(x);             /* high word of x */
   }
   if (hx >= 0x7ff00000) return x+x;
   k += (hx>>20)-1023;
   hx &= 0x000fffff;
   i = (hx+0x95f64)&0x100000;
-  __HI(x) = hx|(i^0x3ff00000);        /* normalize x or x/2 */
+  set_high(&x, hx|(i^0x3ff00000)); /* normalize x or x/2 */
   k += (i>>20);
   f = x-1.0;
   if((0x000fffff&(2+hx))<3) {  /* |f| < 2**-20 */
@@ -208,8 +209,8 @@
   int i,k,hx;
   unsigned lx;
 
-  hx = __HI(x);       /* high word of x */
-  lx = __LO(x);       /* low word of x */
+  hx = high(x);       /* high word of x */
+  lx = low(x);        /* low word of x */
 
   k=0;
   if (hx < 0x00100000) {                  /* x < 2**-1022  */
@@ -217,14 +218,14 @@
       return -two54/zero;             /* log(+-0)=-inf */
     if (hx<0) return (x-x)/zero;        /* log(-#) = NaN */
     k -= 54; x *= two54; /* subnormal number, scale up x */
-    hx = __HI(x);                /* high word of x */
+    hx = high(x);                /* high word of x */
   }
   if (hx >= 0x7ff00000) return x+x;
   k += (hx>>20)-1023;
   i  = ((unsigned)k&0x80000000)>>31;
   hx = (hx&0x000fffff)|((0x3ff-i)<<20);
   y  = (double)(k+i);
-  __HI(x) = hx;
+  set_high(&x, hx);
   z  = y*log10_2lo + ivln10*__ieee754_log(x);
   return  z+y*log10_2hi;
 }
@@ -319,14 +320,14 @@
   int k=0,xsb;
   unsigned hx;
 
-  hx  = __HI(x);        /* high word of x */
+  hx  = high(x);                /* high word of x */
   xsb = (hx>>31)&1;             /* sign bit of x */
   hx &= 0x7fffffff;             /* high word of |x| */
 
   /* filter out non-finite argument */
   if(hx >= 0x40862E42) {                        /* if |x|>=709.78... */
     if(hx>=0x7ff00000) {
-      if(((hx&0xfffff)|__LO(x))!=0)
+      if(((hx&0xfffff)|low(x))!=0)
         return x+x;             /* NaN */
       else return (xsb==0)? x:0.0;      /* exp(+-inf)={inf,0} */
     }
@@ -357,10 +358,10 @@
   if(k==0)      return one-((x*c)/(c-2.0)-x);
   else          y = one-((lo-(x*c)/(2.0-c))-hi);
   if(k >= -1021) {
-    __HI(y) += (k<<20); /* add k to y's exponent */
+    set_high(&y, high(y) + (k<<20)); /* add k to y's exponent */
     return y;
   } else {
-    __HI(y) += ((k+1000)<<20);/* add k to y's exponent */
+    set_high(&y, high(y) + ((k+1000)<<20)); /* add k to y's exponent */
     return y*twom1000;
   }
 }
@@ -447,8 +448,8 @@
   unsigned lx,ly;
 
   i0 = ((*(int*)&one)>>29)^1; i1=1-i0;
-  hx = __HI(x); lx = __LO(x);
-  hy = __HI(y); ly = __LO(y);
+  hx = high(x); lx = low(x);
+  hy = high(y); ly = low(y);
   ix = hx&0x7fffffff;  iy = hy&0x7fffffff;
 
   /* y==zero: x**0 = 1 */
@@ -548,14 +549,14 @@
     u = ivln2_h*t;      /* ivln2_h has 21 sig. bits */
     v = t*ivln2_l-w*ivln2;
     t1 = u+v;
-    __LO(t1) = 0;
+    set_low(&t1, 0);
     t2 = v-(t1-u);
   } else {
     double ss,s2,s_h,s_l,t_h,t_l;
     n = 0;
     /* take care subnormal number */
     if(ix<0x00100000)
-      {ax *= two53; n -= 53; ix = __HI(ax); }
+      {ax *= two53; n -= 53; ix = high(ax); }
     n  += ((ix)>>20)-0x3ff;
     j  = ix&0x000fffff;
     /* determine interval */
@@ -563,17 +564,17 @@
     if(j<=0x3988E) k=0;         /* |x|<sqrt(3/2) */
     else if(j<0xBB67A) k=1;     /* |x|<sqrt(3)   */
     else {k=0;n+=1;ix -= 0x00100000;}
-    __HI(ax) = ix;
+    set_high(&ax, ix);
 
     /* compute ss = s_h+s_l = (x-1)/(x+1) or (x-1.5)/(x+1.5) */
     u = ax-bp[k];               /* bp[0]=1.0, bp[1]=1.5 */
     v = one/(ax+bp[k]);
     ss = u*v;
     s_h = ss;
-    __LO(s_h) = 0;
+    set_low(&s_h, 0);
     /* t_h=ax+bp[k] High */
     t_h = zeroX;
-    __HI(t_h)=((ix>>1)|0x20000000)+0x00080000+(k<<18);
+    set_high(&t_h, ((ix>>1)|0x20000000)+0x00080000+(k<<18));
     t_l = ax - (t_h-bp[k]);
     s_l = v*((u-s_h*t_h)-s_h*t_l);
     /* compute log(ax) */
@@ -582,32 +583,32 @@
     r += s_l*(s_h+ss);
     s2  = s_h*s_h;
     t_h = 3.0+s2+r;
-    __LO(t_h) = 0;
+    set_low(&t_h, 0);
     t_l = r-((t_h-3.0)-s2);
     /* u+v = ss*(1+...) */
     u = s_h*t_h;
     v = s_l*t_h+t_l*ss;
     /* 2/(3log2)*(ss+...) */
     p_h = u+v;
-    __LO(p_h) = 0;
+    set_low(&p_h, 0);
     p_l = v-(p_h-u);
     z_h = cp_h*p_h;             /* cp_h+cp_l = 2/(3*log2) */
     z_l = cp_l*p_h+p_l*cp+dp_l[k];
     /* log2(ax) = (ss+..)*2/(3*log2) = n + dp_h + z_h + z_l */
     t = (double)n;
     t1 = (((z_h+z_l)+dp_h[k])+t);
-    __LO(t1) = 0;
+    set_low(&t1, 0);
     t2 = z_l-(((t1-t)-dp_h[k])-z_h);
   }
 
   /* split up y into y1+y2 and compute (y1+y2)*(t1+t2) */
   y1  = y;
-  __LO(y1) = 0;
+  set_low(&y1, 0);
   p_l = (y-y1)*t1+y*t2;
   p_h = y1*t1;
   z = p_l+p_h;
-  j = __HI(z);
-  i = __LO(z);
+  j = high(z);
+  i = low(z);
   if (j>=0x40900000) {                          /* z >= 1024 */
     if(((j-0x40900000)|i)!=0)                   /* if z > 1024 */
       return s*hugeX*hugeX;                     /* overflow */
@@ -631,13 +632,13 @@
     n = j+(0x00100000>>(k+1));
     k = ((n&0x7fffffff)>>20)-0x3ff;     /* new k for n */
     t = zeroX;
-    __HI(t) = (n&~(0x000fffff>>k));
+    set_high(&t, (n&~(0x000fffff>>k)));
     n = ((n&0x000fffff)|0x00100000)>>(20-k);
     if(j<0) n = -n;
     p_h -= t;
   }
   t = p_l+p_h;
-  __LO(t) = 0;
+  set_low(&t, 0);
   u = t*lg2_h;
   v = (p_l-(t-p_h))*lg2+t*lg2_l;
   z = u+v;
@@ -646,10 +647,10 @@
   t1  = z - t*(P1+t*(P2+t*(P3+t*(P4+t*P5))));
   r  = (z*t1)/(t1-two)-(w+z*w);
   z  = one-(r-z);
-  j  = __HI(z);
+  j  = high(z);
   j += (n<<20);
   if((j>>20)<=0) z = scalbnA(z,n);       /* subnormal output */
-  else __HI(z) += (n<<20);
+  else set_high(&z, high(z) + (n<<20));
   return s*z;
 }
 
--- a/src/share/vm/runtime/sharedRuntimeTrig.cpp	Thu Jul 31 16:39:57 2014 -0700
+++ b/src/share/vm/runtime/sharedRuntimeTrig.cpp	Fri Aug 01 15:57:00 2014 -0700
@@ -519,7 +519,7 @@
 {
         double z,r,v;
         int ix;
-        ix = __HI(x)&0x7fffffff;        /* high word of x */
+        ix = high(x)&0x7fffffff;                /* high word of x */
         if(ix<0x3e400000)                       /* |x| < 2**-27 */
            {if((int)x==0) return x;}            /* generate inexact */
         z       =  x*x;
@@ -574,9 +574,9 @@
 
 static double __kernel_cos(double x, double y)
 {
-  double a,h,z,r,qx;
+  double a,h,z,r,qx=0;
   int ix;
-  ix = __HI(x)&0x7fffffff;      /* ix = |x|'s high word*/
+  ix = high(x)&0x7fffffff;              /* ix = |x|'s high word*/
   if(ix<0x3e400000) {                   /* if x < 2**27 */
     if(((int)x)==0) return one;         /* generate inexact */
   }
@@ -588,8 +588,8 @@
     if(ix > 0x3fe90000) {               /* x > 0.78125 */
       qx = 0.28125;
     } else {
-      __HI(qx) = ix-0x00200000; /* x/4 */
-      __LO(qx) = 0;
+      set_high(&qx, ix-0x00200000); /* x/4 */
+      set_low(&qx, 0);
     }
     h = 0.5*z-qx;
     a = one-qx;
@@ -654,11 +654,11 @@
 {
   double z,r,v,w,s;
   int ix,hx;
-  hx = __HI(x);   /* high word of x */
+  hx = high(x);           /* high word of x */
   ix = hx&0x7fffffff;     /* high word of |x| */
   if(ix<0x3e300000) {                     /* x < 2**-28 */
     if((int)x==0) {                       /* generate inexact */
-      if (((ix | __LO(x)) | (iy + 1)) == 0)
+      if (((ix | low(x)) | (iy + 1)) == 0)
         return one / fabsd(x);
       else {
         if (iy == 1)
@@ -667,10 +667,10 @@
           double a, t;
 
           z = w = x + y;
-          __LO(z) = 0;
+          set_low(&z, 0);
           v = y - (z - x);
           t = a = -one / w;
-          __LO(t) = 0;
+          set_low(&t, 0);
           s = one + t * z;
           return t + a * (s + t * v);
         }
@@ -705,10 +705,10 @@
     /*  compute -1.0/(x+r) accurately */
     double a,t;
     z  = w;
-    __LO(z) = 0;
+    set_low(&z, 0);
     v  = r-(z - x);     /* z+v = r+x */
     t = a  = -1.0/w;    /* a = -1.0/w */
-    __LO(t) = 0;
+    set_low(&t, 0);
     s  = 1.0+t*z;
     return t+a*(s+t*v);
   }
@@ -757,7 +757,7 @@
   int n, ix;
 
   /* High word of x. */
-  ix = __HI(x);
+  ix = high(x);
 
   /* |x| ~< pi/4 */
   ix &= 0x7fffffff;
@@ -815,7 +815,7 @@
   int n, ix;
 
   /* High word of x. */
-  ix = __HI(x);
+  ix = high(x);
 
   /* |x| ~< pi/4 */
   ix &= 0x7fffffff;
@@ -872,7 +872,7 @@
   int n, ix;
 
   /* High word of x. */
-  ix = __HI(x);
+  ix = high(x);
 
   /* |x| ~< pi/4 */
   ix &= 0x7fffffff;
--- a/src/share/vm/utilities/globalDefinitions.hpp	Thu Jul 31 16:39:57 2014 -0700
+++ b/src/share/vm/utilities/globalDefinitions.hpp	Fri Aug 01 15:57:00 2014 -0700
@@ -558,6 +558,27 @@
   return fabs(value);
 }
 
+//----------------------------------------------------------------------------------------------------
+// Special casts
+// Cast floats into same-size integers and vice-versa w/o changing bit-pattern
+typedef union {
+  jfloat f;
+  jint i;
+} FloatIntConv;
+
+typedef union {
+  jdouble d;
+  jlong l;
+  julong ul;
+} DoubleLongConv;
+
+inline jint    jint_cast    (jfloat  x)  { return ((FloatIntConv*)&x)->i; }
+inline jfloat  jfloat_cast  (jint    x)  { return ((FloatIntConv*)&x)->f; }
+
+inline jlong   jlong_cast   (jdouble x)  { return ((DoubleLongConv*)&x)->l;  }
+inline julong  julong_cast  (jdouble x)  { return ((DoubleLongConv*)&x)->ul; }
+inline jdouble jdouble_cast (jlong   x)  { return ((DoubleLongConv*)&x)->d;  }
+
 inline jint low (jlong value)                    { return jint(value); }
 inline jint high(jlong value)                    { return jint(value >> 32); }
 
--- a/src/share/vm/utilities/globalDefinitions_gcc.hpp	Thu Jul 31 16:39:57 2014 -0700
+++ b/src/share/vm/utilities/globalDefinitions_gcc.hpp	Fri Aug 01 15:57:00 2014 -0700
@@ -167,17 +167,6 @@
 typedef uint32_t juint;
 typedef uint64_t julong;
 
-//----------------------------------------------------------------------------------------------------
-// Special (possibly not-portable) casts
-// Cast floats into same-size integers and vice-versa w/o changing bit-pattern
-// %%%%%% These seem like standard C++ to me--how about factoring them out? - Ungar
-
-inline jint    jint_cast   (jfloat  x)           { return *(jint*   )&x; }
-inline jlong   jlong_cast  (jdouble x)           { return *(jlong*  )&x; }
-inline julong  julong_cast (jdouble x)           { return *(julong* )&x; }
-
-inline jfloat  jfloat_cast (jint    x)           { return *(jfloat* )&x; }
-inline jdouble jdouble_cast(jlong   x)           { return *(jdouble*)&x; }
 
 //----------------------------------------------------------------------------------------------------
 // Constant for jlong (specifying an long long canstant is C++ compiler specific)
--- a/src/share/vm/utilities/globalDefinitions_sparcWorks.hpp	Thu Jul 31 16:39:57 2014 -0700
+++ b/src/share/vm/utilities/globalDefinitions_sparcWorks.hpp	Fri Aug 01 15:57:00 2014 -0700
@@ -183,15 +183,6 @@
 typedef unsigned int       juint;
 typedef unsigned long long julong;
 
-//----------------------------------------------------------------------------------------------------
-// Special (possibly not-portable) casts
-// Cast floats into same-size integers and vice-versa w/o changing bit-pattern
-
-inline jint    jint_cast   (jfloat  x)           { return *(jint*   )&x; }
-inline jlong   jlong_cast  (jdouble x)           { return *(jlong*  )&x; }
-
-inline jfloat  jfloat_cast (jint    x)           { return *(jfloat* )&x; }
-inline jdouble jdouble_cast(jlong   x)           { return *(jdouble*)&x; }
 
 //----------------------------------------------------------------------------------------------------
 // Constant for jlong (specifying an long long constant is C++ compiler specific)
--- a/src/share/vm/utilities/globalDefinitions_visCPP.hpp	Thu Jul 31 16:39:57 2014 -0700
+++ b/src/share/vm/utilities/globalDefinitions_visCPP.hpp	Fri Aug 01 15:57:00 2014 -0700
@@ -116,16 +116,6 @@
 typedef unsigned int     juint;
 typedef unsigned __int64 julong;
 
-//----------------------------------------------------------------------------------------------------
-// Special (possibly not-portable) casts
-// Cast floats into same-size integers and vice-versa w/o changing bit-pattern
-
-inline jint    jint_cast   (jfloat  x)           { return *(jint*   )&x; }
-inline jlong   jlong_cast  (jdouble x)           { return *(jlong*  )&x; }
-
-inline jfloat  jfloat_cast (jint    x)           { return *(jfloat* )&x; }
-inline jdouble jdouble_cast(jlong   x)           { return *(jdouble*)&x; }
-
 
 //----------------------------------------------------------------------------------------------------
 // Non-standard stdlib-like stuff:
--- a/src/share/vm/utilities/globalDefinitions_xlc.hpp	Thu Jul 31 16:39:57 2014 -0700
+++ b/src/share/vm/utilities/globalDefinitions_xlc.hpp	Fri Aug 01 15:57:00 2014 -0700
@@ -114,16 +114,6 @@
 typedef uint32_t juint;
 typedef uint64_t julong;
 
-//----------------------------------------------------------------------------------------------------
-// Special (possibly not-portable) casts
-// Cast floats into same-size integers and vice-versa w/o changing bit-pattern
-// %%%%%% These seem like standard C++ to me--how about factoring them out? - Ungar
-
-inline jint    jint_cast   (jfloat  x)           { return *(jint*   )&x; }
-inline jlong   jlong_cast  (jdouble x)           { return *(jlong*  )&x; }
-
-inline jfloat  jfloat_cast (jint    x)           { return *(jfloat* )&x; }
-inline jdouble jdouble_cast(jlong   x)           { return *(jdouble*)&x; }
 
 //----------------------------------------------------------------------------------------------------
 // Constant for jlong (specifying an long long canstant is C++ compiler specific)
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/test/compiler/osr/TestOSRWithNonEmptyStack.java	Fri Aug 01 15:57:00 2014 -0700
@@ -0,0 +1,127 @@
+/*
+ * Copyright (c) 2014, Oracle and/or its affiliates. All rights reserved.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This code is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.
+ *
+ * This code is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+ * or visit www.oracle.com if you need additional information or have any
+ * questions.
+ */
+
+import java.lang.reflect.Constructor;
+import java.lang.reflect.Method;
+
+import jdk.internal.org.objectweb.asm.ClassWriter;
+import jdk.internal.org.objectweb.asm.Label;
+import jdk.internal.org.objectweb.asm.MethodVisitor;
+import static jdk.internal.org.objectweb.asm.Opcodes.*;
+
+/**
+ * @test
+ * @bug 8051344
+ * @summary Force OSR compilation with non-empty stack at the OSR entry point.
+ * @compile -XDignore.symbol.file TestOSRWithNonEmptyStack.java
+ * @run main/othervm -XX:CompileOnly=TestCase.test TestOSRWithNonEmptyStack
+ */
+public class TestOSRWithNonEmptyStack extends ClassLoader {
+    private static final int CLASS_FILE_VERSION = 52;
+    private static final String CLASS_NAME = "TestCase";
+    private static final String METHOD_NAME = "test";
+    private static final int ITERATIONS = 1_000_000;
+
+    private static byte[] generateTestClass() {
+        ClassWriter cw = new ClassWriter(ClassWriter.COMPUTE_FRAMES);
+
+        cw.visit(TestOSRWithNonEmptyStack.CLASS_FILE_VERSION, ACC_PUBLIC,
+                TestOSRWithNonEmptyStack.CLASS_NAME, null, "java/lang/Object",
+                null);
+
+        TestOSRWithNonEmptyStack.generateConstructor(cw);
+        TestOSRWithNonEmptyStack.generateTestMethod(cw);
+
+        cw.visitEnd();
+        return cw.toByteArray();
+    }
+
+    private static void generateConstructor(ClassWriter classWriter) {
+        MethodVisitor mv = classWriter.visitMethod(ACC_PUBLIC, "<init>", "()V",
+                null, null);
+
+        mv.visitCode();
+
+        mv.visitVarInsn(ALOAD, 0);
+        mv.visitMethodInsn(INVOKESPECIAL, "java/lang/Object", "<init>", "()V",
+                false);
+        mv.visitInsn(RETURN);
+
+        mv.visitMaxs(0, 0);
+        mv.visitEnd();
+    }
+
+    private static void generateTestMethod(ClassWriter classWriter) {
+        MethodVisitor mv = classWriter.visitMethod(ACC_PUBLIC,
+                TestOSRWithNonEmptyStack.METHOD_NAME, "()V", null, null);
+        Label osrEntryPoint = new Label();
+
+        mv.visitCode();
+        // Push 'this' into stack before OSR entry point to bail out compilation
+        mv.visitVarInsn(ALOAD, 0);
+        // Setup loop counter
+        mv.visitInsn(ICONST_0);
+        mv.visitVarInsn(ISTORE, 1);
+        // Begin loop
+        mv.visitLabel(osrEntryPoint);
+        // Increment loop counter
+        mv.visitVarInsn(ILOAD, 1);
+        mv.visitInsn(ICONST_1);
+        mv.visitInsn(IADD);
+        // Duplicate it for loop condition check
+        mv.visitInsn(DUP);
+        mv.visitVarInsn(ISTORE, 1);
+        // Check loop condition
+        mv.visitLdcInsn(TestOSRWithNonEmptyStack.ITERATIONS);
+        mv.visitJumpInsn(IF_ICMPLT, osrEntryPoint);
+        // Pop 'this'.
+        mv.visitInsn(POP);
+        mv.visitInsn(RETURN);
+
+        mv.visitMaxs(0, 0);
+        mv.visitEnd();
+    }
+
+    private void run() {
+        byte[] bytecode = TestOSRWithNonEmptyStack.generateTestClass();
+
+        try {
+            Class klass = defineClass(TestOSRWithNonEmptyStack.CLASS_NAME,
+                    bytecode, 0, bytecode.length);
+
+            Constructor ctor = klass.getConstructor();
+            Method method = klass.getDeclaredMethod(
+                    TestOSRWithNonEmptyStack.METHOD_NAME);
+
+            Object testCase = ctor.newInstance();
+            method.invoke(testCase);
+        } catch (Exception e) {
+            throw new RuntimeException(
+                    "Test bug: generated class should be valid.", e);
+        }
+    }
+
+    public static void main(String args[]) {
+        new TestOSRWithNonEmptyStack().run();
+    }
+}