Merge
authoranoll
Thu, 01 May 2014 21:30:54 -0700
changeset 24339 354b4855663a
parent 24338 bff6fc677861 (current diff)
parent 24329 13ccf2b92d69 (diff)
child 24341 ecca498a5ce9
Merge
--- a/hotspot/agent/src/os/linux/libproc.h	Fri May 02 06:24:39 2014 +0200
+++ b/hotspot/agent/src/os/linux/libproc.h	Thu May 01 21:30:54 2014 -0700
@@ -34,19 +34,7 @@
 #include "libproc_md.h"
 #endif
 
-#if defined(sparc) || defined(sparcv9)
-/*
-  If _LP64 is defined ptrace.h should be taken from /usr/include/asm-sparc64
-  otherwise it should be from /usr/include/asm-sparc
-  These two files define pt_regs structure differently
-*/
-#ifdef _LP64
-#include "asm-sparc64/ptrace.h"
-#else
-#include "asm-sparc/ptrace.h"
-#endif
-
-#endif //sparc or sparcv9
+#include <linux/ptrace.h>
 
 /************************************************************************************
 
--- a/hotspot/src/cpu/sparc/vm/assembler_sparc.hpp	Fri May 02 06:24:39 2014 +0200
+++ b/hotspot/src/cpu/sparc/vm/assembler_sparc.hpp	Thu May 01 21:30:54 2014 -0700
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 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
@@ -123,8 +123,13 @@
     fpop2_op3    = 0x35,
     impdep1_op3  = 0x36,
     aes3_op3     = 0x36,
+    alignaddr_op3  = 0x36,
+    faligndata_op3 = 0x36,
     flog3_op3    = 0x36,
+    edge_op3     = 0x36,
+    fsrc_op3     = 0x36,
     impdep2_op3  = 0x37,
+    stpartialf_op3 = 0x37,
     jmpl_op3     = 0x38,
     rett_op3     = 0x39,
     trap_op3     = 0x3a,
@@ -175,17 +180,23 @@
 
   enum opfs {
     // selected opfs
+    edge8n_opf         = 0x01,
+
     fmovs_opf          = 0x01,
     fmovd_opf          = 0x02,
 
     fnegs_opf          = 0x05,
     fnegd_opf          = 0x06,
 
+    alignaddr_opf      = 0x18,
+
     fadds_opf          = 0x41,
     faddd_opf          = 0x42,
     fsubs_opf          = 0x45,
     fsubd_opf          = 0x46,
 
+    faligndata_opf     = 0x48,
+
     fmuls_opf          = 0x49,
     fmuld_opf          = 0x4a,
     fdivs_opf          = 0x4d,
@@ -348,6 +359,8 @@
     ASI_PRIMARY            = 0x80,
     ASI_PRIMARY_NOFAULT    = 0x82,
     ASI_PRIMARY_LITTLE     = 0x88,
+    // 8x8-bit partial store
+    ASI_PST8_PRIMARY       = 0xC0,
     // Block initializing store
     ASI_ST_BLKINIT_PRIMARY = 0xE2,
     // Most-Recently-Used (MRU) BIS variant
@@ -585,6 +598,9 @@
   // instruction only in VIS1
   static void vis1_only() { assert( VM_Version::has_vis1(), "This instruction only works on SPARC with VIS1"); }
 
+  // instruction only in VIS2
+  static void vis2_only() { assert( VM_Version::has_vis2(), "This instruction only works on SPARC with VIS2"); }
+
   // instruction only in VIS3
   static void vis3_only() { assert( VM_Version::has_vis3(), "This instruction only works on SPARC with VIS3"); }
 
@@ -1164,6 +1180,20 @@
   inline void wrfprs( Register d) { v9_only(); emit_int32( op(arith_op) | rs1(d) | op3(wrreg_op3) | u_field(6, 29, 25)); }
 
 
+  //  VIS1 instructions
+
+  void alignaddr( Register s1, Register s2, Register d ) { vis1_only(); emit_int32( op(arith_op) | rd(d) | op3(alignaddr_op3) | rs1(s1) | opf(alignaddr_opf) | rs2(s2)); }
+
+  void faligndata( FloatRegister s1, FloatRegister s2, FloatRegister d ) { vis1_only(); emit_int32( op(arith_op) | fd(d, FloatRegisterImpl::D) | op3(faligndata_op3) | fs1(s1, FloatRegisterImpl::D) | opf(faligndata_opf) | fs2(s2, FloatRegisterImpl::D)); }
+
+  void fsrc2( FloatRegisterImpl::Width w, FloatRegister s2, FloatRegister d ) { vis1_only(); emit_int32( op(arith_op) | fd(d, w) | op3(fsrc_op3) | opf(0x7A - w) | fs2(s2, w)); }
+
+  void stpartialf( Register s1, Register s2, FloatRegister d, int ia = -1 ) { vis1_only(); emit_int32( op(ldst_op) | fd(d, FloatRegisterImpl::D) | op3(stpartialf_op3) | rs1(s1) | imm_asi(ia) | rs2(s2)); }
+
+  //  VIS2 instructions
+
+  void edge8n( Register s1, Register s2, Register d ) { vis2_only(); emit_int32( op(arith_op) | rd(d) | op3(edge_op3) | rs1(s1) | opf(edge8n_opf) | rs2(s2)); }
+
   // VIS3 instructions
 
   void movstosw( FloatRegister s, Register d ) { vis3_only();  emit_int32( op(arith_op) | rd(d) | op3(mftoi_op3) | opf(mstosw_opf) | fs2(s, FloatRegisterImpl::S)); }
--- a/hotspot/src/cpu/sparc/vm/frame_sparc.hpp	Fri May 02 06:24:39 2014 +0200
+++ b/hotspot/src/cpu/sparc/vm/frame_sparc.hpp	Thu May 01 21:30:54 2014 -0700
@@ -259,8 +259,8 @@
 
   // next two fns read and write Lmonitors value,
  private:
-  BasicObjectLock* interpreter_frame_monitors()           const  { return *interpreter_frame_monitors_addr(); }
-  void interpreter_frame_set_monitors(BasicObjectLock* monitors) {        *interpreter_frame_monitors_addr() = monitors; }
+  BasicObjectLock* interpreter_frame_monitors() const;
+  void interpreter_frame_set_monitors(BasicObjectLock* monitors);
 #else
  public:
   inline interpreterState get_interpreterState() const {
--- a/hotspot/src/cpu/sparc/vm/frame_sparc.inline.hpp	Fri May 02 06:24:39 2014 +0200
+++ b/hotspot/src/cpu/sparc/vm/frame_sparc.inline.hpp	Thu May 01 21:30:54 2014 -0700
@@ -226,6 +226,13 @@
   return (Method**)sp_addr_at( Lmethod->sp_offset_in_saved_window());
 }
 
+inline BasicObjectLock* frame::interpreter_frame_monitors() const {
+  return *interpreter_frame_monitors_addr();
+}
+
+inline void frame::interpreter_frame_set_monitors(BasicObjectLock* monitors) {
+  *interpreter_frame_monitors_addr() = monitors;
+}
 
 // Constant pool cache
 
--- a/hotspot/src/cpu/sparc/vm/stubGenerator_sparc.cpp	Fri May 02 06:24:39 2014 +0200
+++ b/hotspot/src/cpu/sparc/vm/stubGenerator_sparc.cpp	Thu May 01 21:30:54 2014 -0700
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 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
@@ -83,7 +83,7 @@
  private:
 
 #ifdef PRODUCT
-#define inc_counter_np(a,b,c) (0)
+#define inc_counter_np(a,b,c)
 #else
 #define inc_counter_np(counter, t1, t2) \
   BLOCK_COMMENT("inc_counter " #counter); \
@@ -1055,7 +1055,7 @@
                                               Label& L_loop, bool use_prefetch, bool use_bis);
 
   void disjoint_copy_core(Register from, Register to, Register count, int log2_elem_size,
-                          int iter_size, CopyLoopFunc copy_loop_func) {
+                          int iter_size, StubGenerator::CopyLoopFunc copy_loop_func) {
     Label L_copy;
 
     assert(log2_elem_size <= 3, "the following code should be changed");
@@ -1206,7 +1206,7 @@
     __ inc(from, 8);
     __ sllx(O3, left_shift,  O3);
 
-    disjoint_copy_core(from, to, count, log2_elem_size, 16, copy_16_bytes_shift_loop);
+    disjoint_copy_core(from, to, count, log2_elem_size, 16, &StubGenerator::copy_16_bytes_shift_loop);
 
     __ inccc(count, count_dec>>1 ); // + 8 bytes
     __ brx(Assembler::negative, true, Assembler::pn, L_copy_last_bytes);
@@ -2085,7 +2085,7 @@
       __ dec(count, 4);   // The cmp at the beginning guaranty count >= 4
       __ sllx(O3, 32,  O3);
 
-      disjoint_copy_core(from, to, count, 2, 16, copy_16_bytes_loop);
+      disjoint_copy_core(from, to, count, 2, 16, &StubGenerator::copy_16_bytes_loop);
 
       __ br(Assembler::always, false, Assembler::pt, L_copy_4_bytes);
       __ delayed()->inc(count, 4); // restore 'count'
@@ -2366,7 +2366,7 @@
     // count >= 0 (original count - 8)
     __ mov(from, from64);
 
-    disjoint_copy_core(from64, to64, count, 3, 64, copy_64_bytes_loop);
+    disjoint_copy_core(from64, to64, count, 3, 64, &StubGenerator::copy_64_bytes_loop);
 
       // Restore O4(offset0), O5(offset8)
       __ sub(from64, from, offset0);
@@ -3305,9 +3305,12 @@
   }
 
   address generate_aescrypt_encryptBlock() {
+    // required since we read expanded key 'int' array starting first element without alignment considerations
+    assert((arrayOopDesc::base_offset_in_bytes(T_INT) & 7) == 0,
+           "the following code assumes that first element of an int array is aligned to 8 bytes");
     __ align(CodeEntryAlignment);
-    StubCodeMark mark(this, "StubRoutines", "aesencryptBlock");
-    Label L_doLast128bit, L_storeOutput;
+    StubCodeMark mark(this, "StubRoutines", "aescrypt_encryptBlock");
+    Label L_load_misaligned_input, L_load_expanded_key, L_doLast128bit, L_storeOutput, L_store_misaligned_output;
     address start = __ pc();
     Register from = O0; // source byte array
     Register to = O1;   // destination byte array
@@ -3317,15 +3320,33 @@
     // read expanded key length
     __ ldsw(Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)), keylen, 0);
 
-    // load input into F54-F56; F30-F31 used as temp
-    __ ldf(FloatRegisterImpl::S, from, 0, F30);
-    __ ldf(FloatRegisterImpl::S, from, 4, F31);
-    __ fmov(FloatRegisterImpl::D, F30, F54);
-    __ ldf(FloatRegisterImpl::S, from, 8, F30);
-    __ ldf(FloatRegisterImpl::S, from, 12, F31);
-    __ fmov(FloatRegisterImpl::D, F30, F56);
-
-    // load expanded key
+    // Method to address arbitrary alignment for load instructions:
+    // Check last 3 bits of 'from' address to see if it is aligned to 8-byte boundary
+    // If zero/aligned then continue with double FP load instructions
+    // If not zero/mis-aligned then alignaddr will set GSR.align with number of bytes to skip during faligndata
+    // alignaddr will also convert arbitrary aligned 'from' address to nearest 8-byte aligned address
+    // load 3 * 8-byte components (to read 16 bytes input) in 3 different FP regs starting at this aligned address
+    // faligndata will then extract (based on GSR.align value) the appropriate 8 bytes from the 2 source regs
+
+    // check for 8-byte alignment since source byte array may have an arbitrary alignment if offset mod 8 is non-zero
+    __ andcc(from, 7, G0);
+    __ br(Assembler::notZero, true, Assembler::pn, L_load_misaligned_input);
+    __ delayed()->alignaddr(from, G0, from);
+
+    // aligned case: load input into F54-F56
+    __ ldf(FloatRegisterImpl::D, from, 0, F54);
+    __ ldf(FloatRegisterImpl::D, from, 8, F56);
+    __ ba_short(L_load_expanded_key);
+
+    __ BIND(L_load_misaligned_input);
+    __ ldf(FloatRegisterImpl::D, from, 0, F54);
+    __ ldf(FloatRegisterImpl::D, from, 8, F56);
+    __ ldf(FloatRegisterImpl::D, from, 16, F58);
+    __ faligndata(F54, F56, F54);
+    __ faligndata(F56, F58, F56);
+
+    __ BIND(L_load_expanded_key);
+    // Since we load expanded key buffers starting first element, 8-byte alignment is guaranteed
     for ( int i = 0;  i <= 38; i += 2 ) {
       __ ldf(FloatRegisterImpl::D, key, i*4, as_FloatRegister(i));
     }
@@ -3365,8 +3386,7 @@
     __ ldf(FloatRegisterImpl::D, key, 232, F50);
     __ aes_eround01(F52, F54, F56, F58); //round 13
     __ aes_eround23(F46, F54, F56, F60);
-    __ br(Assembler::always, false, Assembler::pt, L_storeOutput);
-    __ delayed()->nop();
+    __ ba_short(L_storeOutput);
 
     __ BIND(L_doLast128bit);
     __ ldf(FloatRegisterImpl::D, key, 160, F48);
@@ -3377,23 +3397,62 @@
     __ aes_eround01_l(F48, F58, F60, F54); //last round
     __ aes_eround23_l(F50, F58, F60, F56);
 
-    // store output into the destination array, F0-F1 used as temp
-    __ fmov(FloatRegisterImpl::D, F54, F0);
-    __ stf(FloatRegisterImpl::S, F0, to, 0);
-    __ stf(FloatRegisterImpl::S, F1, to, 4);
-    __ fmov(FloatRegisterImpl::D, F56, F0);
-    __ stf(FloatRegisterImpl::S, F0, to, 8);
+    // Method to address arbitrary alignment for store instructions:
+    // Check last 3 bits of 'dest' address to see if it is aligned to 8-byte boundary
+    // If zero/aligned then continue with double FP store instructions
+    // If not zero/mis-aligned then edge8n will generate edge mask in result reg (O3 in below case)
+    // Example: If dest address is 0x07 and nearest 8-byte aligned address is 0x00 then edge mask will be 00000001
+    // Compute (8-n) where n is # of bytes skipped by partial store(stpartialf) inst from edge mask, n=7 in this case
+    // We get the value of n from the andcc that checks 'dest' alignment. n is available in O5 in below case.
+    // Set GSR.align to (8-n) using alignaddr
+    // Circular byte shift store values by n places so that the original bytes are at correct position for stpartialf
+    // Set the arbitrarily aligned 'dest' address to nearest 8-byte aligned address
+    // Store (partial) the original first (8-n) bytes starting at the original 'dest' address
+    // Negate the edge mask so that the subsequent stpartialf can store the original (8-n-1)th through 8th bytes at appropriate address
+    // We need to execute this process for both the 8-byte result values
+
+    // check for 8-byte alignment since dest byte array may have arbitrary alignment if offset mod 8 is non-zero
+    __ andcc(to, 7, O5);
+    __ br(Assembler::notZero, true, Assembler::pn, L_store_misaligned_output);
+    __ delayed()->edge8n(to, G0, O3);
+
+    // aligned case: store output into the destination array
+    __ stf(FloatRegisterImpl::D, F54, to, 0);
     __ retl();
-    __ delayed()->stf(FloatRegisterImpl::S, F1, to, 12);
+    __ delayed()->stf(FloatRegisterImpl::D, F56, to, 8);
+
+    __ BIND(L_store_misaligned_output);
+    __ add(to, 8, O4);
+    __ mov(8, O2);
+    __ sub(O2, O5, O2);
+    __ alignaddr(O2, G0, O2);
+    __ faligndata(F54, F54, F54);
+    __ faligndata(F56, F56, F56);
+    __ and3(to, -8, to);
+    __ and3(O4, -8, O4);
+    __ stpartialf(to, O3, F54, Assembler::ASI_PST8_PRIMARY);
+    __ stpartialf(O4, O3, F56, Assembler::ASI_PST8_PRIMARY);
+    __ add(to, 8, to);
+    __ add(O4, 8, O4);
+    __ orn(G0, O3, O3);
+    __ stpartialf(to, O3, F54, Assembler::ASI_PST8_PRIMARY);
+    __ retl();
+    __ delayed()->stpartialf(O4, O3, F56, Assembler::ASI_PST8_PRIMARY);
 
     return start;
   }
 
   address generate_aescrypt_decryptBlock() {
+    assert((arrayOopDesc::base_offset_in_bytes(T_INT) & 7) == 0,
+           "the following code assumes that first element of an int array is aligned to 8 bytes");
+    // required since we read original key 'byte' array as well in the decryption stubs
+    assert((arrayOopDesc::base_offset_in_bytes(T_BYTE) & 7) == 0,
+           "the following code assumes that first element of a byte array is aligned to 8 bytes");
     __ align(CodeEntryAlignment);
-    StubCodeMark mark(this, "StubRoutines", "aesdecryptBlock");
+    StubCodeMark mark(this, "StubRoutines", "aescrypt_decryptBlock");
     address start = __ pc();
-    Label L_expand192bit, L_expand256bit, L_common_transform;
+    Label L_load_misaligned_input, L_load_original_key, L_expand192bit, L_expand256bit, L_reload_misaligned_input;
+    Label L_256bit_transform, L_common_transform, L_store_misaligned_output;
     Register from = O0; // source byte array
     Register to = O1;   // destination byte array
     Register key = O2;  // expanded key array
@@ -3403,15 +3462,29 @@
     // read expanded key array length
     __ ldsw(Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)), keylen, 0);
 
-    // load input into F52-F54; F30,F31 used as temp
-    __ ldf(FloatRegisterImpl::S, from, 0, F30);
-    __ ldf(FloatRegisterImpl::S, from, 4, F31);
-    __ fmov(FloatRegisterImpl::D, F30, F52);
-    __ ldf(FloatRegisterImpl::S, from, 8, F30);
-    __ ldf(FloatRegisterImpl::S, from, 12, F31);
-    __ fmov(FloatRegisterImpl::D, F30, F54);
-
+    // save 'from' since we may need to recheck alignment in case of 256-bit decryption
+    __ mov(from, G1);
+
+    // check for 8-byte alignment since source byte array may have an arbitrary alignment if offset mod 8 is non-zero
+    __ andcc(from, 7, G0);
+    __ br(Assembler::notZero, true, Assembler::pn, L_load_misaligned_input);
+    __ delayed()->alignaddr(from, G0, from);
+
+    // aligned case: load input into F52-F54
+    __ ldf(FloatRegisterImpl::D, from, 0, F52);
+    __ ldf(FloatRegisterImpl::D, from, 8, F54);
+    __ ba_short(L_load_original_key);
+
+    __ BIND(L_load_misaligned_input);
+    __ ldf(FloatRegisterImpl::D, from, 0, F52);
+    __ ldf(FloatRegisterImpl::D, from, 8, F54);
+    __ ldf(FloatRegisterImpl::D, from, 16, F56);
+    __ faligndata(F52, F54, F52);
+    __ faligndata(F54, F56, F54);
+
+    __ BIND(L_load_original_key);
     // load original key from SunJCE expanded decryption key
+    // Since we load original key buffer starting first element, 8-byte alignment is guaranteed
     for ( int i = 0;  i <= 3; i++ ) {
       __ ldf(FloatRegisterImpl::S, original_key, i*4, as_FloatRegister(i));
     }
@@ -3432,8 +3505,7 @@
     // perform 128-bit key specific inverse cipher transformation
     __ fxor(FloatRegisterImpl::D, F42, F54, F54);
     __ fxor(FloatRegisterImpl::D, F40, F52, F52);
-    __ br(Assembler::always, false, Assembler::pt, L_common_transform);
-    __ delayed()->nop();
+    __ ba_short(L_common_transform);
 
     __ BIND(L_expand192bit);
 
@@ -3457,8 +3529,7 @@
     __ aes_dround01(F44, F52, F54, F56);
     __ aes_dround23(F42, F56, F58, F54);
     __ aes_dround01(F40, F56, F58, F52);
-    __ br(Assembler::always, false, Assembler::pt, L_common_transform);
-    __ delayed()->nop();
+    __ ba_short(L_common_transform);
 
     __ BIND(L_expand256bit);
 
@@ -3478,14 +3549,31 @@
     __ aes_kexpand2(F50, F56, F58);
 
     for ( int i = 0;  i <= 6; i += 2 ) {
-      __ fmov(FloatRegisterImpl::D, as_FloatRegister(58-i), as_FloatRegister(i));
+      __ fsrc2(FloatRegisterImpl::D, as_FloatRegister(58-i), as_FloatRegister(i));
     }
 
-    // load input into F52-F54
+    // reload original 'from' address
+    __ mov(G1, from);
+
+    // re-check 8-byte alignment
+    __ andcc(from, 7, G0);
+    __ br(Assembler::notZero, true, Assembler::pn, L_reload_misaligned_input);
+    __ delayed()->alignaddr(from, G0, from);
+
+    // aligned case: load input into F52-F54
     __ ldf(FloatRegisterImpl::D, from, 0, F52);
     __ ldf(FloatRegisterImpl::D, from, 8, F54);
+    __ ba_short(L_256bit_transform);
+
+    __ BIND(L_reload_misaligned_input);
+    __ ldf(FloatRegisterImpl::D, from, 0, F52);
+    __ ldf(FloatRegisterImpl::D, from, 8, F54);
+    __ ldf(FloatRegisterImpl::D, from, 16, F56);
+    __ faligndata(F52, F54, F52);
+    __ faligndata(F54, F56, F54);
 
     // perform 256-bit key specific inverse cipher transformation
+    __ BIND(L_256bit_transform);
     __ fxor(FloatRegisterImpl::D, F0, F54, F54);
     __ fxor(FloatRegisterImpl::D, F2, F52, F52);
     __ aes_dround23(F4, F52, F54, F58);
@@ -3515,43 +3603,71 @@
       }
     }
 
-    // store output to destination array, F0-F1 used as temp
-    __ fmov(FloatRegisterImpl::D, F52, F0);
-    __ stf(FloatRegisterImpl::S, F0, to, 0);
-    __ stf(FloatRegisterImpl::S, F1, to, 4);
-    __ fmov(FloatRegisterImpl::D, F54, F0);
-    __ stf(FloatRegisterImpl::S, F0, to, 8);
+    // check for 8-byte alignment since dest byte array may have arbitrary alignment if offset mod 8 is non-zero
+    __ andcc(to, 7, O5);
+    __ br(Assembler::notZero, true, Assembler::pn, L_store_misaligned_output);
+    __ delayed()->edge8n(to, G0, O3);
+
+    // aligned case: store output into the destination array
+    __ stf(FloatRegisterImpl::D, F52, to, 0);
     __ retl();
-    __ delayed()->stf(FloatRegisterImpl::S, F1, to, 12);
+    __ delayed()->stf(FloatRegisterImpl::D, F54, to, 8);
+
+    __ BIND(L_store_misaligned_output);
+    __ add(to, 8, O4);
+    __ mov(8, O2);
+    __ sub(O2, O5, O2);
+    __ alignaddr(O2, G0, O2);
+    __ faligndata(F52, F52, F52);
+    __ faligndata(F54, F54, F54);
+    __ and3(to, -8, to);
+    __ and3(O4, -8, O4);
+    __ stpartialf(to, O3, F52, Assembler::ASI_PST8_PRIMARY);
+    __ stpartialf(O4, O3, F54, Assembler::ASI_PST8_PRIMARY);
+    __ add(to, 8, to);
+    __ add(O4, 8, O4);
+    __ orn(G0, O3, O3);
+    __ stpartialf(to, O3, F52, Assembler::ASI_PST8_PRIMARY);
+    __ retl();
+    __ delayed()->stpartialf(O4, O3, F54, Assembler::ASI_PST8_PRIMARY);
 
     return start;
   }
 
   address generate_cipherBlockChaining_encryptAESCrypt() {
+    assert((arrayOopDesc::base_offset_in_bytes(T_INT) & 7) == 0,
+           "the following code assumes that first element of an int array is aligned to 8 bytes");
+    assert((arrayOopDesc::base_offset_in_bytes(T_BYTE) & 7) == 0,
+           "the following code assumes that first element of a byte array is aligned to 8 bytes");
     __ align(CodeEntryAlignment);
     StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_encryptAESCrypt");
-    Label L_cbcenc128, L_cbcenc192, L_cbcenc256;
+    Label L_cbcenc128, L_load_misaligned_input_128bit, L_128bit_transform, L_store_misaligned_output_128bit;
+    Label L_check_loop_end_128bit, L_cbcenc192, L_load_misaligned_input_192bit, L_192bit_transform;
+    Label L_store_misaligned_output_192bit, L_check_loop_end_192bit, L_cbcenc256, L_load_misaligned_input_256bit;
+    Label L_256bit_transform, L_store_misaligned_output_256bit, L_check_loop_end_256bit;
     address start = __ pc();
-    Register from = O0; // source byte array
-    Register to = O1;   // destination byte array
-    Register key = O2;  // expanded key array
-    Register rvec = O3; // init vector
-    const Register len_reg = O4; // cipher length
-    const Register keylen = O5;  // reg for storing expanded key array length
-
-    // save cipher len to return in the end
-    __ mov(len_reg, L1);
+    Register from = I0; // source byte array
+    Register to = I1;   // destination byte array
+    Register key = I2;  // expanded key array
+    Register rvec = I3; // init vector
+    const Register len_reg = I4; // cipher length
+    const Register keylen = I5;  // reg for storing expanded key array length
+
+    // save cipher len before save_frame, to return in the end
+    __ mov(O4, L0);
+    __ save_frame(0);
 
     // read expanded key length
     __ ldsw(Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)), keylen, 0);
 
-    // load init vector
+    // load initial vector, 8-byte alignment is guranteed
     __ ldf(FloatRegisterImpl::D, rvec, 0, F60);
     __ ldf(FloatRegisterImpl::D, rvec, 8, F62);
+    // load key, 8-byte alignment is guranteed
     __ ldx(key,0,G1);
-    __ ldx(key,8,G2);
-
-    // start loading expanded key
+    __ ldx(key,8,G5);
+
+    // start loading expanded key, 8-byte alignment is guranteed
     for ( int i = 0, j = 16;  i <= 38; i += 2, j += 8 ) {
       __ ldf(FloatRegisterImpl::D, key, j, as_FloatRegister(i));
     }
@@ -3571,15 +3687,35 @@
     }
 
     // 256-bit original key size
-    __ br(Assembler::always, false, Assembler::pt, L_cbcenc256);
-    __ delayed()->nop();
+    __ ba_short(L_cbcenc256);
 
     __ align(OptoLoopAlignment);
     __ BIND(L_cbcenc128);
+    // check for 8-byte alignment since source byte array may have an arbitrary alignment if offset mod 8 is non-zero
+    __ andcc(from, 7, G0);
+    __ br(Assembler::notZero, true, Assembler::pn, L_load_misaligned_input_128bit);
+    __ delayed()->mov(from, L1); // save original 'from' address before alignaddr
+
+    // aligned case: load input into G3 and G4
     __ ldx(from,0,G3);
     __ ldx(from,8,G4);
+    __ ba_short(L_128bit_transform);
+
+    __ BIND(L_load_misaligned_input_128bit);
+    // can clobber F48, F50 and F52 as they are not used in 128 and 192-bit key encryption
+    __ alignaddr(from, G0, from);
+    __ ldf(FloatRegisterImpl::D, from, 0, F48);
+    __ ldf(FloatRegisterImpl::D, from, 8, F50);
+    __ ldf(FloatRegisterImpl::D, from, 16, F52);
+    __ faligndata(F48, F50, F48);
+    __ faligndata(F50, F52, F50);
+    __ movdtox(F48, G3);
+    __ movdtox(F50, G4);
+    __ mov(L1, from);
+
+    __ BIND(L_128bit_transform);
     __ xor3(G1,G3,G3);
-    __ xor3(G2,G4,G4);
+    __ xor3(G5,G4,G4);
     __ movxtod(G3,F56);
     __ movxtod(G4,F58);
     __ fxor(FloatRegisterImpl::D, F60, F56, F60);
@@ -3598,24 +3734,81 @@
       }
     }
 
+    // check for 8-byte alignment since dest byte array may have arbitrary alignment if offset mod 8 is non-zero
+    __ andcc(to, 7, L1);
+    __ br(Assembler::notZero, true, Assembler::pn, L_store_misaligned_output_128bit);
+    __ delayed()->edge8n(to, G0, L2);
+
+    // aligned case: store output into the destination array
     __ stf(FloatRegisterImpl::D, F60, to, 0);
     __ stf(FloatRegisterImpl::D, F62, to, 8);
+    __ ba_short(L_check_loop_end_128bit);
+
+    __ BIND(L_store_misaligned_output_128bit);
+    __ add(to, 8, L3);
+    __ mov(8, L4);
+    __ sub(L4, L1, L4);
+    __ alignaddr(L4, G0, L4);
+    // save cipher text before circular right shift
+    // as it needs to be stored as iv for next block (see code before next retl)
+    __ movdtox(F60, L6);
+    __ movdtox(F62, L7);
+    __ faligndata(F60, F60, F60);
+    __ faligndata(F62, F62, F62);
+    __ mov(to, L5);
+    __ and3(to, -8, to);
+    __ and3(L3, -8, L3);
+    __ stpartialf(to, L2, F60, Assembler::ASI_PST8_PRIMARY);
+    __ stpartialf(L3, L2, F62, Assembler::ASI_PST8_PRIMARY);
+    __ add(to, 8, to);
+    __ add(L3, 8, L3);
+    __ orn(G0, L2, L2);
+    __ stpartialf(to, L2, F60, Assembler::ASI_PST8_PRIMARY);
+    __ stpartialf(L3, L2, F62, Assembler::ASI_PST8_PRIMARY);
+    __ mov(L5, to);
+    __ movxtod(L6, F60);
+    __ movxtod(L7, F62);
+
+    __ BIND(L_check_loop_end_128bit);
     __ add(from, 16, from);
     __ add(to, 16, to);
     __ subcc(len_reg, 16, len_reg);
     __ br(Assembler::notEqual, false, Assembler::pt, L_cbcenc128);
     __ delayed()->nop();
+    // re-init intial vector for next block, 8-byte alignment is guaranteed
     __ stf(FloatRegisterImpl::D, F60, rvec, 0);
     __ stf(FloatRegisterImpl::D, F62, rvec, 8);
+    __ restore();
     __ retl();
-    __ delayed()->mov(L1, O0);
+    __ delayed()->mov(L0, O0);
 
     __ align(OptoLoopAlignment);
     __ BIND(L_cbcenc192);
+    // check for 8-byte alignment since source byte array may have an arbitrary alignment if offset mod 8 is non-zero
+    __ andcc(from, 7, G0);
+    __ br(Assembler::notZero, true, Assembler::pn, L_load_misaligned_input_192bit);
+    __ delayed()->mov(from, L1); // save original 'from' address before alignaddr
+
+    // aligned case: load input into G3 and G4
     __ ldx(from,0,G3);
     __ ldx(from,8,G4);
+    __ ba_short(L_192bit_transform);
+
+    __ BIND(L_load_misaligned_input_192bit);
+    // can clobber F48, F50 and F52 as they are not used in 128 and 192-bit key encryption
+    __ alignaddr(from, G0, from);
+    __ ldf(FloatRegisterImpl::D, from, 0, F48);
+    __ ldf(FloatRegisterImpl::D, from, 8, F50);
+    __ ldf(FloatRegisterImpl::D, from, 16, F52);
+    __ faligndata(F48, F50, F48);
+    __ faligndata(F50, F52, F50);
+    __ movdtox(F48, G3);
+    __ movdtox(F50, G4);
+    __ mov(L1, from);
+
+    __ BIND(L_192bit_transform);
     __ xor3(G1,G3,G3);
-    __ xor3(G2,G4,G4);
+    __ xor3(G5,G4,G4);
     __ movxtod(G3,F56);
     __ movxtod(G4,F58);
     __ fxor(FloatRegisterImpl::D, F60, F56, F60);
@@ -3634,24 +3827,81 @@
       }
     }
 
+    // check for 8-byte alignment since dest byte array may have arbitrary alignment if offset mod 8 is non-zero
+    __ andcc(to, 7, L1);
+    __ br(Assembler::notZero, true, Assembler::pn, L_store_misaligned_output_192bit);
+    __ delayed()->edge8n(to, G0, L2);
+
+    // aligned case: store output into the destination array
     __ stf(FloatRegisterImpl::D, F60, to, 0);
     __ stf(FloatRegisterImpl::D, F62, to, 8);
+    __ ba_short(L_check_loop_end_192bit);
+
+    __ BIND(L_store_misaligned_output_192bit);
+    __ add(to, 8, L3);
+    __ mov(8, L4);
+    __ sub(L4, L1, L4);
+    __ alignaddr(L4, G0, L4);
+    __ movdtox(F60, L6);
+    __ movdtox(F62, L7);
+    __ faligndata(F60, F60, F60);
+    __ faligndata(F62, F62, F62);
+    __ mov(to, L5);
+    __ and3(to, -8, to);
+    __ and3(L3, -8, L3);
+    __ stpartialf(to, L2, F60, Assembler::ASI_PST8_PRIMARY);
+    __ stpartialf(L3, L2, F62, Assembler::ASI_PST8_PRIMARY);
+    __ add(to, 8, to);
+    __ add(L3, 8, L3);
+    __ orn(G0, L2, L2);
+    __ stpartialf(to, L2, F60, Assembler::ASI_PST8_PRIMARY);
+    __ stpartialf(L3, L2, F62, Assembler::ASI_PST8_PRIMARY);
+    __ mov(L5, to);
+    __ movxtod(L6, F60);
+    __ movxtod(L7, F62);
+
+    __ BIND(L_check_loop_end_192bit);
     __ add(from, 16, from);
     __ subcc(len_reg, 16, len_reg);
     __ add(to, 16, to);
     __ br(Assembler::notEqual, false, Assembler::pt, L_cbcenc192);
     __ delayed()->nop();
+    // re-init intial vector for next block, 8-byte alignment is guaranteed
     __ stf(FloatRegisterImpl::D, F60, rvec, 0);
     __ stf(FloatRegisterImpl::D, F62, rvec, 8);
+    __ restore();
     __ retl();
-    __ delayed()->mov(L1, O0);
+    __ delayed()->mov(L0, O0);
 
     __ align(OptoLoopAlignment);
     __ BIND(L_cbcenc256);
+    // check for 8-byte alignment since source byte array may have an arbitrary alignment if offset mod 8 is non-zero
+    __ andcc(from, 7, G0);
+    __ br(Assembler::notZero, true, Assembler::pn, L_load_misaligned_input_256bit);
+    __ delayed()->mov(from, L1); // save original 'from' address before alignaddr
+
+    // aligned case: load input into G3 and G4
     __ ldx(from,0,G3);
     __ ldx(from,8,G4);
+    __ ba_short(L_256bit_transform);
+
+    __ BIND(L_load_misaligned_input_256bit);
+    // cannot clobber F48, F50 and F52. F56, F58 can be used though
+    __ alignaddr(from, G0, from);
+    __ movdtox(F60, L2); // save F60 before overwriting
+    __ ldf(FloatRegisterImpl::D, from, 0, F56);
+    __ ldf(FloatRegisterImpl::D, from, 8, F58);
+    __ ldf(FloatRegisterImpl::D, from, 16, F60);
+    __ faligndata(F56, F58, F56);
+    __ faligndata(F58, F60, F58);
+    __ movdtox(F56, G3);
+    __ movdtox(F58, G4);
+    __ mov(L1, from);
+    __ movxtod(L2, F60);
+
+    __ BIND(L_256bit_transform);
     __ xor3(G1,G3,G3);
-    __ xor3(G2,G4,G4);
+    __ xor3(G5,G4,G4);
     __ movxtod(G3,F56);
     __ movxtod(G4,F58);
     __ fxor(FloatRegisterImpl::D, F60, F56, F60);
@@ -3670,26 +3920,69 @@
       }
     }
 
+    // check for 8-byte alignment since dest byte array may have arbitrary alignment if offset mod 8 is non-zero
+    __ andcc(to, 7, L1);
+    __ br(Assembler::notZero, true, Assembler::pn, L_store_misaligned_output_256bit);
+    __ delayed()->edge8n(to, G0, L2);
+
+    // aligned case: store output into the destination array
     __ stf(FloatRegisterImpl::D, F60, to, 0);
     __ stf(FloatRegisterImpl::D, F62, to, 8);
+    __ ba_short(L_check_loop_end_256bit);
+
+    __ BIND(L_store_misaligned_output_256bit);
+    __ add(to, 8, L3);
+    __ mov(8, L4);
+    __ sub(L4, L1, L4);
+    __ alignaddr(L4, G0, L4);
+    __ movdtox(F60, L6);
+    __ movdtox(F62, L7);
+    __ faligndata(F60, F60, F60);
+    __ faligndata(F62, F62, F62);
+    __ mov(to, L5);
+    __ and3(to, -8, to);
+    __ and3(L3, -8, L3);
+    __ stpartialf(to, L2, F60, Assembler::ASI_PST8_PRIMARY);
+    __ stpartialf(L3, L2, F62, Assembler::ASI_PST8_PRIMARY);
+    __ add(to, 8, to);
+    __ add(L3, 8, L3);
+    __ orn(G0, L2, L2);
+    __ stpartialf(to, L2, F60, Assembler::ASI_PST8_PRIMARY);
+    __ stpartialf(L3, L2, F62, Assembler::ASI_PST8_PRIMARY);
+    __ mov(L5, to);
+    __ movxtod(L6, F60);
+    __ movxtod(L7, F62);
+
+    __ BIND(L_check_loop_end_256bit);
     __ add(from, 16, from);
     __ subcc(len_reg, 16, len_reg);
     __ add(to, 16, to);
     __ br(Assembler::notEqual, false, Assembler::pt, L_cbcenc256);
     __ delayed()->nop();
+    // re-init intial vector for next block, 8-byte alignment is guaranteed
     __ stf(FloatRegisterImpl::D, F60, rvec, 0);
     __ stf(FloatRegisterImpl::D, F62, rvec, 8);
+    __ restore();
     __ retl();
-    __ delayed()->mov(L1, O0);
+    __ delayed()->mov(L0, O0);
 
     return start;
   }
 
   address generate_cipherBlockChaining_decryptAESCrypt_Parallel() {
+    assert((arrayOopDesc::base_offset_in_bytes(T_INT) & 7) == 0,
+           "the following code assumes that first element of an int array is aligned to 8 bytes");
+    assert((arrayOopDesc::base_offset_in_bytes(T_BYTE) & 7) == 0,
+           "the following code assumes that first element of a byte array is aligned to 8 bytes");
     __ align(CodeEntryAlignment);
     StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt");
     Label L_cbcdec_end, L_expand192bit, L_expand256bit, L_dec_first_block_start;
     Label L_dec_first_block128, L_dec_first_block192, L_dec_next2_blocks128, L_dec_next2_blocks192, L_dec_next2_blocks256;
+    Label L_load_misaligned_input_first_block, L_transform_first_block, L_load_misaligned_next2_blocks128, L_transform_next2_blocks128;
+    Label L_load_misaligned_next2_blocks192, L_transform_next2_blocks192, L_load_misaligned_next2_blocks256, L_transform_next2_blocks256;
+    Label L_store_misaligned_output_first_block, L_check_decrypt_end, L_store_misaligned_output_next2_blocks128;
+    Label L_check_decrypt_loop_end128, L_store_misaligned_output_next2_blocks192, L_check_decrypt_loop_end192;
+    Label L_store_misaligned_output_next2_blocks256, L_check_decrypt_loop_end256;
     address start = __ pc();
     Register from = I0; // source byte array
     Register to = I1;   // destination byte array
@@ -3704,11 +3997,12 @@
     __ save_frame(0); //args are read from I* registers since we save the frame in the beginning
 
     // load original key from SunJCE expanded decryption key
+    // Since we load original key buffer starting first element, 8-byte alignment is guaranteed
     for ( int i = 0;  i <= 3; i++ ) {
       __ ldf(FloatRegisterImpl::S, original_key, i*4, as_FloatRegister(i));
     }
 
-    // load initial vector
+    // load initial vector, 8-byte alignment is guaranteed
     __ ldx(rvec,0,L0);
     __ ldx(rvec,8,L1);
 
@@ -3733,11 +4027,10 @@
     __ movdtox(F42,L3);
 
     __ and3(len_reg, 16, L4);
-    __ br_null(L4, false, Assembler::pt, L_dec_next2_blocks128);
-    __ delayed()->nop();
-
-    __ br(Assembler::always, false, Assembler::pt, L_dec_first_block_start);
-    __ delayed()->nop();
+    __ br_null_short(L4, Assembler::pt, L_dec_next2_blocks128);
+    __ nop();
+
+    __ ba_short(L_dec_first_block_start);
 
     __ BIND(L_expand192bit);
     // load rest of the 192-bit key
@@ -3758,11 +4051,10 @@
     __ movdtox(F50,L3);
 
     __ and3(len_reg, 16, L4);
-    __ br_null(L4, false, Assembler::pt, L_dec_next2_blocks192);
-    __ delayed()->nop();
-
-    __ br(Assembler::always, false, Assembler::pt, L_dec_first_block_start);
-    __ delayed()->nop();
+    __ br_null_short(L4, Assembler::pt, L_dec_next2_blocks192);
+    __ nop();
+
+    __ ba_short(L_dec_first_block_start);
 
     __ BIND(L_expand256bit);
     // load rest of the 256-bit key
@@ -3785,12 +4077,32 @@
     __ movdtox(F58,L3);
 
     __ and3(len_reg, 16, L4);
-    __ br_null(L4, false, Assembler::pt, L_dec_next2_blocks256);
-    __ delayed()->nop();
+    __ br_null_short(L4, Assembler::pt, L_dec_next2_blocks256);
 
     __ BIND(L_dec_first_block_start);
+    // check for 8-byte alignment since source byte array may have an arbitrary alignment if offset mod 8 is non-zero
+    __ andcc(from, 7, G0);
+    __ br(Assembler::notZero, true, Assembler::pn, L_load_misaligned_input_first_block);
+    __ delayed()->mov(from, G1); // save original 'from' address before alignaddr
+
+    // aligned case: load input into L4 and L5
     __ ldx(from,0,L4);
     __ ldx(from,8,L5);
+    __ ba_short(L_transform_first_block);
+
+    __ BIND(L_load_misaligned_input_first_block);
+    __ alignaddr(from, G0, from);
+    // F58, F60, F62 can be clobbered
+    __ ldf(FloatRegisterImpl::D, from, 0, F58);
+    __ ldf(FloatRegisterImpl::D, from, 8, F60);
+    __ ldf(FloatRegisterImpl::D, from, 16, F62);
+    __ faligndata(F58, F60, F58);
+    __ faligndata(F60, F62, F60);
+    __ movdtox(F58, L4);
+    __ movdtox(F60, L5);
+    __ mov(G1, from);
+
+    __ BIND(L_transform_first_block);
     __ xor3(L2,L4,G1);
     __ movxtod(G1,F60);
     __ xor3(L3,L5,G1);
@@ -3833,9 +4145,36 @@
     __ fxor(FloatRegisterImpl::D, F56, F60, F60);
     __ fxor(FloatRegisterImpl::D, F58, F62, F62);
 
+    // check for 8-byte alignment since dest byte array may have arbitrary alignment if offset mod 8 is non-zero
+    __ andcc(to, 7, G1);
+    __ br(Assembler::notZero, true, Assembler::pn, L_store_misaligned_output_first_block);
+    __ delayed()->edge8n(to, G0, G2);
+
+    // aligned case: store output into the destination array
     __ stf(FloatRegisterImpl::D, F60, to, 0);
     __ stf(FloatRegisterImpl::D, F62, to, 8);
-
+    __ ba_short(L_check_decrypt_end);
+
+    __ BIND(L_store_misaligned_output_first_block);
+    __ add(to, 8, G3);
+    __ mov(8, G4);
+    __ sub(G4, G1, G4);
+    __ alignaddr(G4, G0, G4);
+    __ faligndata(F60, F60, F60);
+    __ faligndata(F62, F62, F62);
+    __ mov(to, G1);
+    __ and3(to, -8, to);
+    __ and3(G3, -8, G3);
+    __ stpartialf(to, G2, F60, Assembler::ASI_PST8_PRIMARY);
+    __ stpartialf(G3, G2, F62, Assembler::ASI_PST8_PRIMARY);
+    __ add(to, 8, to);
+    __ add(G3, 8, G3);
+    __ orn(G0, G2, G2);
+    __ stpartialf(to, G2, F60, Assembler::ASI_PST8_PRIMARY);
+    __ stpartialf(G3, G2, F62, Assembler::ASI_PST8_PRIMARY);
+    __ mov(G1, to);
+
+    __ BIND(L_check_decrypt_end);
     __ add(from, 16, from);
     __ add(to, 16, to);
     __ subcc(len_reg, 16, len_reg);
@@ -3852,17 +4191,44 @@
     __ BIND(L_dec_next2_blocks128);
     __ nop();
 
-    // F40:F42 used for first 16-bytes
+    // check for 8-byte alignment since source byte array may have an arbitrary alignment if offset mod 8 is non-zero
+    __ andcc(from, 7, G0);
+    __ br(Assembler::notZero, true, Assembler::pn, L_load_misaligned_next2_blocks128);
+    __ delayed()->mov(from, G1); // save original 'from' address before alignaddr
+
+    // aligned case: load input into G4, G5, L4 and L5
     __ ldx(from,0,G4);
     __ ldx(from,8,G5);
+    __ ldx(from,16,L4);
+    __ ldx(from,24,L5);
+    __ ba_short(L_transform_next2_blocks128);
+
+    __ BIND(L_load_misaligned_next2_blocks128);
+    __ alignaddr(from, G0, from);
+    // F40, F42, F58, F60, F62 can be clobbered
+    __ ldf(FloatRegisterImpl::D, from, 0, F40);
+    __ ldf(FloatRegisterImpl::D, from, 8, F42);
+    __ ldf(FloatRegisterImpl::D, from, 16, F60);
+    __ ldf(FloatRegisterImpl::D, from, 24, F62);
+    __ ldf(FloatRegisterImpl::D, from, 32, F58);
+    __ faligndata(F40, F42, F40);
+    __ faligndata(F42, F60, F42);
+    __ faligndata(F60, F62, F60);
+    __ faligndata(F62, F58, F62);
+    __ movdtox(F40, G4);
+    __ movdtox(F42, G5);
+    __ movdtox(F60, L4);
+    __ movdtox(F62, L5);
+    __ mov(G1, from);
+
+    __ BIND(L_transform_next2_blocks128);
+    // F40:F42 used for first 16-bytes
     __ xor3(L2,G4,G1);
     __ movxtod(G1,F40);
     __ xor3(L3,G5,G1);
     __ movxtod(G1,F42);
 
     // F60:F62 used for next 16-bytes
-    __ ldx(from,16,L4);
-    __ ldx(from,24,L5);
     __ xor3(L2,L4,G1);
     __ movxtod(G1,F60);
     __ xor3(L3,L5,G1);
@@ -3891,9 +4257,6 @@
     __ fxor(FloatRegisterImpl::D, F46, F40, F40);
     __ fxor(FloatRegisterImpl::D, F44, F42, F42);
 
-    __ stf(FloatRegisterImpl::D, F40, to, 0);
-    __ stf(FloatRegisterImpl::D, F42, to, 8);
-
     __ movxtod(G4,F56);
     __ movxtod(G5,F58);
     __ mov(L4,L0);
@@ -3901,32 +4264,93 @@
     __ fxor(FloatRegisterImpl::D, F56, F60, F60);
     __ fxor(FloatRegisterImpl::D, F58, F62, F62);
 
+    // For mis-aligned store of 32 bytes of result we can do:
+    // Circular right-shift all 4 FP registers so that 'head' and 'tail'
+    // parts that need to be stored starting at mis-aligned address are in a FP reg
+    // the other 3 FP regs can thus be stored using regular store
+    // we then use the edge + partial-store mechanism to store the 'head' and 'tail' parts
+
+    // check for 8-byte alignment since dest byte array may have arbitrary alignment if offset mod 8 is non-zero
+    __ andcc(to, 7, G1);
+    __ br(Assembler::notZero, true, Assembler::pn, L_store_misaligned_output_next2_blocks128);
+    __ delayed()->edge8n(to, G0, G2);
+
+    // aligned case: store output into the destination array
+    __ stf(FloatRegisterImpl::D, F40, to, 0);
+    __ stf(FloatRegisterImpl::D, F42, to, 8);
     __ stf(FloatRegisterImpl::D, F60, to, 16);
     __ stf(FloatRegisterImpl::D, F62, to, 24);
-
+    __ ba_short(L_check_decrypt_loop_end128);
+
+    __ BIND(L_store_misaligned_output_next2_blocks128);
+    __ mov(8, G4);
+    __ sub(G4, G1, G4);
+    __ alignaddr(G4, G0, G4);
+    __ faligndata(F40, F42, F56); // F56 can be clobbered
+    __ faligndata(F42, F60, F42);
+    __ faligndata(F60, F62, F60);
+    __ faligndata(F62, F40, F40);
+    __ mov(to, G1);
+    __ and3(to, -8, to);
+    __ stpartialf(to, G2, F40, Assembler::ASI_PST8_PRIMARY);
+    __ stf(FloatRegisterImpl::D, F56, to, 8);
+    __ stf(FloatRegisterImpl::D, F42, to, 16);
+    __ stf(FloatRegisterImpl::D, F60, to, 24);
+    __ add(to, 32, to);
+    __ orn(G0, G2, G2);
+    __ stpartialf(to, G2, F40, Assembler::ASI_PST8_PRIMARY);
+    __ mov(G1, to);
+
+    __ BIND(L_check_decrypt_loop_end128);
     __ add(from, 32, from);
     __ add(to, 32, to);
     __ subcc(len_reg, 32, len_reg);
     __ br(Assembler::notEqual, false, Assembler::pt, L_dec_next2_blocks128);
     __ delayed()->nop();
-    __ br(Assembler::always, false, Assembler::pt, L_cbcdec_end);
-    __ delayed()->nop();
+    __ ba_short(L_cbcdec_end);
 
     __ align(OptoLoopAlignment);
     __ BIND(L_dec_next2_blocks192);
     __ nop();
 
-    // F48:F50 used for first 16-bytes
+    // check for 8-byte alignment since source byte array may have an arbitrary alignment if offset mod 8 is non-zero
+    __ andcc(from, 7, G0);
+    __ br(Assembler::notZero, true, Assembler::pn, L_load_misaligned_next2_blocks192);
+    __ delayed()->mov(from, G1); // save original 'from' address before alignaddr
+
+    // aligned case: load input into G4, G5, L4 and L5
     __ ldx(from,0,G4);
     __ ldx(from,8,G5);
+    __ ldx(from,16,L4);
+    __ ldx(from,24,L5);
+    __ ba_short(L_transform_next2_blocks192);
+
+    __ BIND(L_load_misaligned_next2_blocks192);
+    __ alignaddr(from, G0, from);
+    // F48, F50, F52, F60, F62 can be clobbered
+    __ ldf(FloatRegisterImpl::D, from, 0, F48);
+    __ ldf(FloatRegisterImpl::D, from, 8, F50);
+    __ ldf(FloatRegisterImpl::D, from, 16, F60);
+    __ ldf(FloatRegisterImpl::D, from, 24, F62);
+    __ ldf(FloatRegisterImpl::D, from, 32, F52);
+    __ faligndata(F48, F50, F48);
+    __ faligndata(F50, F60, F50);
+    __ faligndata(F60, F62, F60);
+    __ faligndata(F62, F52, F62);
+    __ movdtox(F48, G4);
+    __ movdtox(F50, G5);
+    __ movdtox(F60, L4);
+    __ movdtox(F62, L5);
+    __ mov(G1, from);
+
+    __ BIND(L_transform_next2_blocks192);
+    // F48:F50 used for first 16-bytes
     __ xor3(L2,G4,G1);
     __ movxtod(G1,F48);
     __ xor3(L3,G5,G1);
     __ movxtod(G1,F50);
 
     // F60:F62 used for next 16-bytes
-    __ ldx(from,16,L4);
-    __ ldx(from,24,L5);
     __ xor3(L2,L4,G1);
     __ movxtod(G1,F60);
     __ xor3(L3,L5,G1);
@@ -3955,9 +4379,6 @@
     __ fxor(FloatRegisterImpl::D, F54, F48, F48);
     __ fxor(FloatRegisterImpl::D, F52, F50, F50);
 
-    __ stf(FloatRegisterImpl::D, F48, to, 0);
-    __ stf(FloatRegisterImpl::D, F50, to, 8);
-
     __ movxtod(G4,F56);
     __ movxtod(G5,F58);
     __ mov(L4,L0);
@@ -3965,32 +4386,87 @@
     __ fxor(FloatRegisterImpl::D, F56, F60, F60);
     __ fxor(FloatRegisterImpl::D, F58, F62, F62);
 
+    // check for 8-byte alignment since dest byte array may have arbitrary alignment if offset mod 8 is non-zero
+    __ andcc(to, 7, G1);
+    __ br(Assembler::notZero, true, Assembler::pn, L_store_misaligned_output_next2_blocks192);
+    __ delayed()->edge8n(to, G0, G2);
+
+    // aligned case: store output into the destination array
+    __ stf(FloatRegisterImpl::D, F48, to, 0);
+    __ stf(FloatRegisterImpl::D, F50, to, 8);
     __ stf(FloatRegisterImpl::D, F60, to, 16);
     __ stf(FloatRegisterImpl::D, F62, to, 24);
-
+    __ ba_short(L_check_decrypt_loop_end192);
+
+    __ BIND(L_store_misaligned_output_next2_blocks192);
+    __ mov(8, G4);
+    __ sub(G4, G1, G4);
+    __ alignaddr(G4, G0, G4);
+    __ faligndata(F48, F50, F56); // F56 can be clobbered
+    __ faligndata(F50, F60, F50);
+    __ faligndata(F60, F62, F60);
+    __ faligndata(F62, F48, F48);
+    __ mov(to, G1);
+    __ and3(to, -8, to);
+    __ stpartialf(to, G2, F48, Assembler::ASI_PST8_PRIMARY);
+    __ stf(FloatRegisterImpl::D, F56, to, 8);
+    __ stf(FloatRegisterImpl::D, F50, to, 16);
+    __ stf(FloatRegisterImpl::D, F60, to, 24);
+    __ add(to, 32, to);
+    __ orn(G0, G2, G2);
+    __ stpartialf(to, G2, F48, Assembler::ASI_PST8_PRIMARY);
+    __ mov(G1, to);
+
+    __ BIND(L_check_decrypt_loop_end192);
     __ add(from, 32, from);
     __ add(to, 32, to);
     __ subcc(len_reg, 32, len_reg);
     __ br(Assembler::notEqual, false, Assembler::pt, L_dec_next2_blocks192);
     __ delayed()->nop();
-    __ br(Assembler::always, false, Assembler::pt, L_cbcdec_end);
-    __ delayed()->nop();
+    __ ba_short(L_cbcdec_end);
 
     __ align(OptoLoopAlignment);
     __ BIND(L_dec_next2_blocks256);
     __ nop();
 
-    // F0:F2 used for first 16-bytes
+    // check for 8-byte alignment since source byte array may have an arbitrary alignment if offset mod 8 is non-zero
+    __ andcc(from, 7, G0);
+    __ br(Assembler::notZero, true, Assembler::pn, L_load_misaligned_next2_blocks256);
+    __ delayed()->mov(from, G1); // save original 'from' address before alignaddr
+
+    // aligned case: load input into G4, G5, L4 and L5
     __ ldx(from,0,G4);
     __ ldx(from,8,G5);
+    __ ldx(from,16,L4);
+    __ ldx(from,24,L5);
+    __ ba_short(L_transform_next2_blocks256);
+
+    __ BIND(L_load_misaligned_next2_blocks256);
+    __ alignaddr(from, G0, from);
+    // F0, F2, F4, F60, F62 can be clobbered
+    __ ldf(FloatRegisterImpl::D, from, 0, F0);
+    __ ldf(FloatRegisterImpl::D, from, 8, F2);
+    __ ldf(FloatRegisterImpl::D, from, 16, F60);
+    __ ldf(FloatRegisterImpl::D, from, 24, F62);
+    __ ldf(FloatRegisterImpl::D, from, 32, F4);
+    __ faligndata(F0, F2, F0);
+    __ faligndata(F2, F60, F2);
+    __ faligndata(F60, F62, F60);
+    __ faligndata(F62, F4, F62);
+    __ movdtox(F0, G4);
+    __ movdtox(F2, G5);
+    __ movdtox(F60, L4);
+    __ movdtox(F62, L5);
+    __ mov(G1, from);
+
+    __ BIND(L_transform_next2_blocks256);
+    // F0:F2 used for first 16-bytes
     __ xor3(L2,G4,G1);
     __ movxtod(G1,F0);
     __ xor3(L3,G5,G1);
     __ movxtod(G1,F2);
 
     // F60:F62 used for next 16-bytes
-    __ ldx(from,16,L4);
-    __ ldx(from,24,L5);
     __ xor3(L2,L4,G1);
     __ movxtod(G1,F60);
     __ xor3(L3,L5,G1);
@@ -4043,9 +4519,6 @@
     __ fxor(FloatRegisterImpl::D, F6, F0, F0);
     __ fxor(FloatRegisterImpl::D, F4, F2, F2);
 
-    __ stf(FloatRegisterImpl::D, F0, to, 0);
-    __ stf(FloatRegisterImpl::D, F2, to, 8);
-
     __ movxtod(G4,F56);
     __ movxtod(G5,F58);
     __ mov(L4,L0);
@@ -4053,9 +4526,38 @@
     __ fxor(FloatRegisterImpl::D, F56, F60, F60);
     __ fxor(FloatRegisterImpl::D, F58, F62, F62);
 
+    // check for 8-byte alignment since dest byte array may have arbitrary alignment if offset mod 8 is non-zero
+    __ andcc(to, 7, G1);
+    __ br(Assembler::notZero, true, Assembler::pn, L_store_misaligned_output_next2_blocks256);
+    __ delayed()->edge8n(to, G0, G2);
+
+    // aligned case: store output into the destination array
+    __ stf(FloatRegisterImpl::D, F0, to, 0);
+    __ stf(FloatRegisterImpl::D, F2, to, 8);
     __ stf(FloatRegisterImpl::D, F60, to, 16);
     __ stf(FloatRegisterImpl::D, F62, to, 24);
-
+    __ ba_short(L_check_decrypt_loop_end256);
+
+    __ BIND(L_store_misaligned_output_next2_blocks256);
+    __ mov(8, G4);
+    __ sub(G4, G1, G4);
+    __ alignaddr(G4, G0, G4);
+    __ faligndata(F0, F2, F56); // F56 can be clobbered
+    __ faligndata(F2, F60, F2);
+    __ faligndata(F60, F62, F60);
+    __ faligndata(F62, F0, F0);
+    __ mov(to, G1);
+    __ and3(to, -8, to);
+    __ stpartialf(to, G2, F0, Assembler::ASI_PST8_PRIMARY);
+    __ stf(FloatRegisterImpl::D, F56, to, 8);
+    __ stf(FloatRegisterImpl::D, F2, to, 16);
+    __ stf(FloatRegisterImpl::D, F60, to, 24);
+    __ add(to, 32, to);
+    __ orn(G0, G2, G2);
+    __ stpartialf(to, G2, F0, Assembler::ASI_PST8_PRIMARY);
+    __ mov(G1, to);
+
+    __ BIND(L_check_decrypt_loop_end256);
     __ add(from, 32, from);
     __ add(to, 32, to);
     __ subcc(len_reg, 32, len_reg);
@@ -4063,6 +4565,7 @@
     __ delayed()->nop();
 
     __ BIND(L_cbcdec_end);
+    // re-init intial vector for next block, 8-byte alignment is guaranteed
     __ stx(L0, rvec, 0);
     __ stx(L1, rvec, 8);
     __ restore();
--- a/hotspot/src/cpu/sparc/vm/stubRoutines_sparc.hpp	Fri May 02 06:24:39 2014 +0200
+++ b/hotspot/src/cpu/sparc/vm/stubRoutines_sparc.hpp	Thu May 01 21:30:54 2014 -0700
@@ -41,7 +41,7 @@
 enum /* platform_dependent_constants */ {
   // %%%%%%%% May be able to shrink this a lot
   code_size1 = 20000,           // simply increase if too small (assembler will crash if too small)
-  code_size2 = 20000            // simply increase if too small (assembler will crash if too small)
+  code_size2 = 22000            // simply increase if too small (assembler will crash if too small)
 };
 
 class Sparc {
--- a/hotspot/src/cpu/sparc/vm/vm_version_sparc.cpp	Fri May 02 06:24:39 2014 +0200
+++ b/hotspot/src/cpu/sparc/vm/vm_version_sparc.cpp	Thu May 01 21:30:54 2014 -0700
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 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
@@ -266,9 +266,9 @@
   if (!has_vis1()) // Drop to 0 if no VIS1 support
     UseVIS = 0;
 
-  // T2 and above should have support for AES instructions
+  // SPARC T4 and above should have support for AES instructions
   if (has_aes()) {
-    if (UseVIS > 0) { // AES intrinsics use FXOR instruction which is VIS1
+    if (UseVIS > 2) { // AES intrinsics use MOVxTOd/MOVdTOx which are VIS3
       if (FLAG_IS_DEFAULT(UseAES)) {
         FLAG_SET_DEFAULT(UseAES, true);
       }
@@ -282,7 +282,7 @@
       }
     } else {
         if (UseAES || UseAESIntrinsics) {
-          warning("SPARC AES intrinsics require VIS1 instruction support. Intrinsics will be disabled.");
+          warning("SPARC AES intrinsics require VIS3 instruction support. Intrinsics will be disabled.");
           if (UseAES) {
             FLAG_SET_DEFAULT(UseAES, false);
           }
--- a/hotspot/src/os_cpu/linux_sparc/vm/atomic_linux_sparc.inline.hpp	Fri May 02 06:24:39 2014 +0200
+++ b/hotspot/src/os_cpu/linux_sparc/vm/atomic_linux_sparc.inline.hpp	Thu May 01 21:30:54 2014 -0700
@@ -78,12 +78,12 @@
   __asm__ volatile(
     "1: \n\t"
     " ldx    [%2], %%o2\n\t"
-    " add    %0, %%o2, %%o3\n\t"
+    " add    %1, %%o2, %%o3\n\t"
     " casx   [%2], %%o2, %%o3\n\t"
     " cmp    %%o2, %%o3\n\t"
     " bne    %%xcc, 1b\n\t"
     "  nop\n\t"
-    " add    %0, %%o2, %0\n\t"
+    " add    %1, %%o2, %0\n\t"
     : "=r" (rv)
     : "r" (add_value), "r" (dest)
     : "memory", "o2", "o3");
--- a/hotspot/src/os_cpu/linux_sparc/vm/os_linux_sparc.cpp	Fri May 02 06:24:39 2014 +0200
+++ b/hotspot/src/os_cpu/linux_sparc/vm/os_linux_sparc.cpp	Thu May 01 21:30:54 2014 -0700
@@ -302,29 +302,30 @@
   if (context == NULL) return;
 
   ucontext_t *uc = (ucontext_t*)context;
+  sigcontext* sc = (sigcontext*)context;
   intptr_t *sp = (intptr_t *)os::Linux::ucontext_get_sp(uc);
 
   st->print_cr("Register to memory mapping:");
   st->cr();
 
   // this is only for the "general purpose" registers
-  st->print("G1="); print_location(st, SIG_REGS(sc).u_regs[CON__G1]);
-  st->print("G2="); print_location(st, SIG_REGS(sc).u_regs[CON__G2]);
-  st->print("G3="); print_location(st, SIG_REGS(sc).u_regs[CON__G3]);
-  st->print("G4="); print_location(st, SIG_REGS(sc).u_regs[CON__G4]);
-  st->print("G5="); print_location(st, SIG_REGS(sc).u_regs[CON__G5]);
-  st->print("G6="); print_location(st, SIG_REGS(sc).u_regs[CON__G6]);
-  st->print("G7="); print_location(st, SIG_REGS(sc).u_regs[CON__G7]);
+  st->print("G1="); print_location(st, SIG_REGS(sc).u_regs[CON_G1]);
+  st->print("G2="); print_location(st, SIG_REGS(sc).u_regs[CON_G2]);
+  st->print("G3="); print_location(st, SIG_REGS(sc).u_regs[CON_G3]);
+  st->print("G4="); print_location(st, SIG_REGS(sc).u_regs[CON_G4]);
+  st->print("G5="); print_location(st, SIG_REGS(sc).u_regs[CON_G5]);
+  st->print("G6="); print_location(st, SIG_REGS(sc).u_regs[CON_G6]);
+  st->print("G7="); print_location(st, SIG_REGS(sc).u_regs[CON_G7]);
   st->cr();
 
-  st->print("O0="); print_location(st, SIG_REGS(sc).u_regs[CON__O0]);
-  st->print("O1="); print_location(st, SIG_REGS(sc).u_regs[CON__O1]);
-  st->print("O2="); print_location(st, SIG_REGS(sc).u_regs[CON__O2]);
-  st->print("O3="); print_location(st, SIG_REGS(sc).u_regs[CON__O3]);
-  st->print("O4="); print_location(st, SIG_REGS(sc).u_regs[CON__O4]);
-  st->print("O5="); print_location(st, SIG_REGS(sc).u_regs[CON__O5]);
-  st->print("O6="); print_location(st, SIG_REGS(sc).u_regs[CON__O6]);
-  st->print("O7="); print_location(st, SIG_REGS(sc).u_regs[CON__O7]);
+  st->print("O0="); print_location(st, SIG_REGS(sc).u_regs[CON_O0]);
+  st->print("O1="); print_location(st, SIG_REGS(sc).u_regs[CON_O1]);
+  st->print("O2="); print_location(st, SIG_REGS(sc).u_regs[CON_O2]);
+  st->print("O3="); print_location(st, SIG_REGS(sc).u_regs[CON_O3]);
+  st->print("O4="); print_location(st, SIG_REGS(sc).u_regs[CON_O4]);
+  st->print("O5="); print_location(st, SIG_REGS(sc).u_regs[CON_O5]);
+  st->print("O6="); print_location(st, SIG_REGS(sc).u_regs[CON_O6]);
+  st->print("O7="); print_location(st, SIG_REGS(sc).u_regs[CON_O7]);
   st->cr();
 
   st->print("L0="); print_location(st, sp[L0->sp_offset_in_saved_window()]);
@@ -516,7 +517,7 @@
   if (nativeInstruction_at(*pc)->is_ic_miss_trap()) {
 #ifdef ASSERT
 #ifdef TIERED
-    CodeBlob* cb = CodeCache::find_blob_unsafe(pc);
+    CodeBlob* cb = CodeCache::find_blob_unsafe(*pc);
     assert(cb->is_compiled_by_c2(), "Wrong compiler");
 #endif // TIERED
 #endif // ASSERT
--- a/hotspot/src/share/vm/classfile/vmSymbols.hpp	Fri May 02 06:24:39 2014 +0200
+++ b/hotspot/src/share/vm/classfile/vmSymbols.hpp	Thu May 01 21:30:54 2014 -0700
@@ -775,7 +775,7 @@
   /* java/lang/ref/Reference */                                                                                         \
   do_intrinsic(_Reference_get,            java_lang_ref_Reference, get_name,    void_object_signature, F_R)             \
                                                                                                                         \
-  /* support for com.sum.crypto.provider.AESCrypt and some of its callers */                                            \
+  /* support for com.sun.crypto.provider.AESCrypt and some of its callers */                                            \
   do_class(com_sun_crypto_provider_aescrypt,      "com/sun/crypto/provider/AESCrypt")                                   \
   do_intrinsic(_aescrypt_encryptBlock, com_sun_crypto_provider_aescrypt, encryptBlock_name, byteArray_int_byteArray_int_signature, F_R)   \
   do_intrinsic(_aescrypt_decryptBlock, com_sun_crypto_provider_aescrypt, decryptBlock_name, byteArray_int_byteArray_int_signature, F_R)   \
--- a/hotspot/src/share/vm/opto/runtime.cpp	Fri May 02 06:24:39 2014 +0200
+++ b/hotspot/src/share/vm/opto/runtime.cpp	Thu May 01 21:30:54 2014 -0700
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1998, 2013, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1998, 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
@@ -868,7 +868,7 @@
   return TypeFunc::make(domain, range);
 }
 
-// for cipherBlockChaining calls of aescrypt encrypt/decrypt, four pointers and a length, returning void
+// for cipherBlockChaining calls of aescrypt encrypt/decrypt, four pointers and a length, returning int
 const TypeFunc* OptoRuntime::cipherBlockChaining_aescrypt_Type() {
   // create input type (domain)
   int num_args      = 5;
--- a/hotspot/test/compiler/7184394/TestAESBase.java	Fri May 02 06:24:39 2014 +0200
+++ b/hotspot/test/compiler/7184394/TestAESBase.java	Thu May 01 21:30:54 2014 -0700
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2012, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2012, 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
@@ -40,9 +40,20 @@
   int msgSize = Integer.getInteger("msgSize", 646);
   boolean checkOutput = Boolean.getBoolean("checkOutput");
   boolean noReinit = Boolean.getBoolean("noReinit");
+  boolean testingMisalignment;
+  private static final int ALIGN = 8;
+  int encInputOffset = Integer.getInteger("encInputOffset", 0) % ALIGN;
+  int encOutputOffset = Integer.getInteger("encOutputOffset", 0) % ALIGN;
+  int decOutputOffset = Integer.getInteger("decOutputOffset", 0) % ALIGN;
+  int lastChunkSize = Integer.getInteger("lastChunkSize", 32);
   int keySize = Integer.getInteger("keySize", 128);
+  int inputLength;
+  int encodeLength;
+  int decodeLength;
+  int decodeMsgSize;
   String algorithm = System.getProperty("algorithm", "AES");
   String mode = System.getProperty("mode", "CBC");
+  String paddingStr = System.getProperty("paddingStr", "PKCS5Padding");
   byte[] input;
   byte[] encode;
   byte[] expectedEncode;
@@ -51,7 +62,6 @@
   Random random = new Random(0);
   Cipher cipher;
   Cipher dCipher;
-  String paddingStr = "PKCS5Padding";
   AlgorithmParameters algParams;
   SecretKey key;
 
@@ -67,7 +77,10 @@
 
   public void prepare() {
     try {
-    System.out.println("\nalgorithm=" + algorithm + ", mode=" + mode + ", msgSize=" + msgSize + ", keySize=" + keySize + ", noReinit=" + noReinit + ", checkOutput=" + checkOutput);
+    System.out.println("\nalgorithm=" + algorithm + ", mode=" + mode + ", paddingStr=" + paddingStr + ", msgSize=" + msgSize + ", keySize=" + keySize + ", noReinit=" + noReinit + ", checkOutput=" + checkOutput + ", encInputOffset=" + encInputOffset + ", encOutputOffset=" + encOutputOffset + ", decOutputOffset=" + decOutputOffset + ", lastChunkSize=" +lastChunkSize );
+
+      if (encInputOffset % ALIGN != 0 || encOutputOffset % ALIGN != 0 || decOutputOffset % ALIGN !=0 )
+        testingMisalignment = true;
 
       int keyLenBytes = (keySize == 0 ? 16 : keySize/8);
       byte keyBytes[] = new byte[keyLenBytes];
@@ -81,10 +94,6 @@
         System.out.println("Algorithm: " + key.getAlgorithm() + "("
                            + key.getEncoded().length * 8 + "bit)");
       }
-      input = new byte[msgSize];
-      for (int i=0; i<input.length; i++) {
-        input[i] = (byte) (i & 0xff);
-      }
 
       cipher = Cipher.getInstance(algorithm + "/" + mode + "/" + paddingStr, "SunJCE");
       dCipher = Cipher.getInstance(algorithm + "/" + mode + "/" + paddingStr, "SunJCE");
@@ -103,10 +112,35 @@
         childShowCipher();
       }
 
+      inputLength = msgSize + encInputOffset;
+      if (testingMisalignment) {
+        encodeLength = cipher.getOutputSize(msgSize - lastChunkSize) + encOutputOffset;
+        encodeLength += cipher.getOutputSize(lastChunkSize);
+        decodeLength = dCipher.getOutputSize(encodeLength - lastChunkSize) + decOutputOffset;
+        decodeLength += dCipher.getOutputSize(lastChunkSize);
+      } else {
+        encodeLength = cipher.getOutputSize(msgSize) + encOutputOffset;
+        decodeLength = dCipher.getOutputSize(encodeLength) + decOutputOffset;
+      }
+
+      input = new byte[inputLength];
+      for (int i=encInputOffset, j=0; i<inputLength; i++, j++) {
+        input[i] = (byte) (j & 0xff);
+      }
+
       // do one encode and decode in preparation
-      // this will also create the encode buffer and decode buffer
-      encode = cipher.doFinal(input);
-      decode = dCipher.doFinal(encode);
+      encode = new byte[encodeLength];
+      decode = new byte[decodeLength];
+      if (testingMisalignment) {
+        decodeMsgSize = cipher.update(input, encInputOffset, (msgSize - lastChunkSize), encode, encOutputOffset);
+        decodeMsgSize += cipher.doFinal(input, (encInputOffset + msgSize - lastChunkSize), lastChunkSize, encode, (encOutputOffset + decodeMsgSize));
+
+        int tempSize = dCipher.update(encode, encOutputOffset, (decodeMsgSize - lastChunkSize), decode, decOutputOffset);
+        dCipher.doFinal(encode, (encOutputOffset + decodeMsgSize - lastChunkSize), lastChunkSize, decode, (decOutputOffset + tempSize));
+      } else {
+        decodeMsgSize = cipher.doFinal(input, encInputOffset, msgSize, encode, encOutputOffset);
+        dCipher.doFinal(encode, encOutputOffset, decodeMsgSize, decode, decOutputOffset);
+      }
       if (checkOutput) {
         expectedEncode = (byte[]) encode.clone();
         expectedDecode = (byte[]) decode.clone();
--- a/hotspot/test/compiler/7184394/TestAESDecode.java	Fri May 02 06:24:39 2014 +0200
+++ b/hotspot/test/compiler/7184394/TestAESDecode.java	Thu May 01 21:30:54 2014 -0700
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2012, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2012, 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
@@ -33,14 +33,15 @@
   public void run() {
     try {
       if (!noReinit) dCipher.init(Cipher.DECRYPT_MODE, key, algParams);
+      decode = new byte[decodeLength];
+      if (testingMisalignment) {
+        int tempSize = dCipher.update(encode, encOutputOffset, (decodeMsgSize - lastChunkSize), decode, decOutputOffset);
+        dCipher.doFinal(encode, (encOutputOffset + decodeMsgSize - lastChunkSize), lastChunkSize, decode, (decOutputOffset + tempSize));
+      } else {
+        dCipher.doFinal(encode, encOutputOffset, decodeMsgSize, decode, decOutputOffset);
+      }
       if (checkOutput) {
-        // checked version creates new output buffer each time
-        decode = dCipher.doFinal(encode, 0, encode.length);
         compareArrays(decode, expectedDecode);
-      } else {
-        // non-checked version outputs to existing encode buffer for maximum speed
-        decode = new byte[dCipher.getOutputSize(encode.length)];
-        dCipher.doFinal(encode, 0, encode.length, decode);
       }
     }
     catch (Exception e) {
--- a/hotspot/test/compiler/7184394/TestAESEncode.java	Fri May 02 06:24:39 2014 +0200
+++ b/hotspot/test/compiler/7184394/TestAESEncode.java	Thu May 01 21:30:54 2014 -0700
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2012, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2012, 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
@@ -33,14 +33,15 @@
   public void run() {
     try {
       if (!noReinit) cipher.init(Cipher.ENCRYPT_MODE, key, algParams);
+      encode = new byte[encodeLength];
+      if (testingMisalignment) {
+        int tempSize = cipher.update(input, encInputOffset, (msgSize - lastChunkSize), encode, encOutputOffset);
+        cipher.doFinal(input, (encInputOffset + msgSize - lastChunkSize), lastChunkSize, encode, (encOutputOffset + tempSize));
+      } else {
+        cipher.doFinal(input, encInputOffset, msgSize, encode, encOutputOffset);
+      }
       if (checkOutput) {
-        // checked version creates new output buffer each time
-        encode = cipher.doFinal(input, 0, msgSize);
         compareArrays(encode, expectedEncode);
-      } else {
-        // non-checked version outputs to existing encode buffer for maximum speed
-        encode = new byte[cipher.getOutputSize(msgSize)];
-        cipher.doFinal(input, 0, msgSize, encode);
       }
     }
     catch (Exception e) {
--- a/hotspot/test/compiler/7184394/TestAESMain.java	Fri May 02 06:24:39 2014 +0200
+++ b/hotspot/test/compiler/7184394/TestAESMain.java	Thu May 01 21:30:54 2014 -0700
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2012, 2014 Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2012, 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
@@ -28,7 +28,19 @@
  * @summary add intrinsics to use AES instructions
  *
  * @run main/othervm/timeout=600 -Xbatch -DcheckOutput=true -Dmode=CBC TestAESMain
+ * @run main/othervm/timeout=600 -Xbatch -DcheckOutput=true -Dmode=CBC -DencInputOffset=1 TestAESMain
+ * @run main/othervm/timeout=600 -Xbatch -DcheckOutput=true -Dmode=CBC -DencOutputOffset=1 TestAESMain
+ * @run main/othervm/timeout=600 -Xbatch -DcheckOutput=true -Dmode=CBC -DdecOutputOffset=1 TestAESMain
+ * @run main/othervm/timeout=600 -Xbatch -DcheckOutput=true -Dmode=CBC -DencInputOffset=1 -DencOutputOffset=1 TestAESMain
+ * @run main/othervm/timeout=600 -Xbatch -DcheckOutput=true -Dmode=CBC -DencInputOffset=1 -DencOutputOffset=1 -DdecOutputOffset=1 TestAESMain
+ * @run main/othervm/timeout=600 -Xbatch -DcheckOutput=true -Dmode=CBC -DencInputOffset=1 -DencOutputOffset=1 -DdecOutputOffset=1 -DpaddingStr=NoPadding -DmsgSize=640 TestAESMain
  * @run main/othervm/timeout=600 -Xbatch -DcheckOutput=true -Dmode=ECB TestAESMain
+ * @run main/othervm/timeout=600 -Xbatch -DcheckOutput=true -Dmode=ECB -DencInputOffset=1 TestAESMain
+ * @run main/othervm/timeout=600 -Xbatch -DcheckOutput=true -Dmode=ECB -DencOutputOffset=1 TestAESMain
+ * @run main/othervm/timeout=600 -Xbatch -DcheckOutput=true -Dmode=ECB -DdecOutputOffset=1 TestAESMain
+ * @run main/othervm/timeout=600 -Xbatch -DcheckOutput=true -Dmode=ECB -DencInputOffset=1 -DencOutputOffset=1 TestAESMain
+ * @run main/othervm/timeout=600 -Xbatch -DcheckOutput=true -Dmode=ECB -DencInputOffset=1 -DencOutputOffset=1 -DdecOutputOffset=1 TestAESMain
+ * @run main/othervm/timeout=600 -Xbatch -DcheckOutput=true -Dmode=ECB -DencInputOffset=1 -DencOutputOffset=1 -DdecOutputOffset=1 -DpaddingStr=NoPadding -DmsgSize=640 TestAESMain
  *
  * @author Tom Deneau
  */
@@ -36,12 +48,13 @@
 public class TestAESMain {
   public static void main(String[] args) {
     int iters = (args.length > 0 ? Integer.valueOf(args[0]) : 1000000);
+    int warmupIters = (args.length > 1 ? Integer.valueOf(args[1]) : 20000);
     System.out.println(iters + " iterations");
     TestAESEncode etest = new TestAESEncode();
     etest.prepare();
-    // warm-up for 20K iterations
+    // warm-up
     System.out.println("Starting encryption warm-up");
-    for (int i=0; i<20000; i++) {
+    for (int i=0; i<warmupIters; i++) {
       etest.run();
     }
     System.out.println("Finished encryption warm-up");
@@ -54,9 +67,9 @@
 
     TestAESDecode dtest = new TestAESDecode();
     dtest.prepare();
-    // warm-up for 20K iterations
+    // warm-up
     System.out.println("Starting decryption warm-up");
-    for (int i=0; i<20000; i++) {
+    for (int i=0; i<warmupIters; i++) {
       dtest.run();
     }
     System.out.println("Finished decryption warm-up");