--- old/src/cpu/x86/vm/sharedRuntime_x86_32.cpp	2016-01-29 14:28:15.185925558 +0100
+++ new/src/cpu/x86/vm/sharedRuntime_x86_32.cpp	2016-01-29 14:28:15.117925561 +0100
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2003, 2015, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2003, 2016, 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
@@ -114,15 +114,20 @@
 
 OopMap* RegisterSaver::save_live_registers(MacroAssembler* masm, int additional_frame_words,
                                            int* total_frame_words, bool verify_fpu, bool save_vectors) {
-  int vect_words = 0;
   int num_xmm_regs = XMMRegisterImpl::number_of_registers;
+  int ymm_bytes = num_xmm_regs * 16;
+  int zmm_bytes = num_xmm_regs * 32;
 #ifdef COMPILER2
   if (save_vectors) {
     assert(UseAVX > 0, "512bit vectors are supported only with EVEX");
     assert(MaxVectorSize == 64, "only 512bit vectors are supported now");
-    // Save upper half of ZMM/YMM registers :
-    vect_words = 8 * 16 / wordSize;
-    additional_frame_words += vect_words;
+    // Save upper half of YMM registers
+    int vect_bytes = ymm_bytes;
+    if (UseAVX > 2) {
+      // Save upper half of ZMM registers as well
+      vect_bytes += zmm_bytes;
+    }
+    additional_frame_words += vect_bytes / wordSize;
   }
 #else
   assert(!save_vectors, "vectors are generated only by C2");
@@ -185,13 +190,14 @@
 
   off = xmm0_off;
   delta = xmm1_off - off;
-  if(UseSSE == 1) {           // Save the XMM state
+  if(UseSSE == 1) {
+    // Save the XMM state
     for (int n = 0; n < num_xmm_regs; n++) {
       __ movflt(Address(rsp, off*wordSize), as_XMMRegister(n));
       off += delta;
     }
   } else if(UseSSE >= 2) {
-    // Save whole 128bit (16 bytes) XMM regiters
+    // Save whole 128bit (16 bytes) XMM registers
     for (int n = 0; n < num_xmm_regs; n++) {
       __ movdqu(Address(rsp, off*wordSize), as_XMMRegister(n));
       off += delta;
@@ -199,13 +205,14 @@
   }
 
   if (save_vectors) {
-    assert(vect_words*wordSize == 128, "");
-    __ subptr(rsp, 128); // Save upper half of YMM registes
+    __ subptr(rsp, ymm_bytes);
+    // Save upper half of YMM registers
     for (int n = 0; n < num_xmm_regs; n++) {
       __ vextractf128h(Address(rsp, n*16), as_XMMRegister(n));
     }
     if (UseAVX > 2) {
-      __ subptr(rsp, 256); // Save upper half of ZMM registes
+      __ subptr(rsp, zmm_bytes);
+      // Save upper half of ZMM registers
       for (int n = 0; n < num_xmm_regs; n++) {
         __ vextractf64x4h(Address(rsp, n*32), as_XMMRegister(n), 1);
       }
@@ -255,50 +262,59 @@
 
 void RegisterSaver::restore_live_registers(MacroAssembler* masm, bool restore_vectors) {
   int num_xmm_regs = XMMRegisterImpl::number_of_registers;
+  int ymm_bytes = num_xmm_regs * 16;
+  int zmm_bytes = num_xmm_regs * 32;
   // Recover XMM & FPU state
   int additional_frame_bytes = 0;
 #ifdef COMPILER2
   if (restore_vectors) {
     assert(UseAVX > 0, "512bit vectors are supported only with EVEX");
     assert(MaxVectorSize == 64, "only 512bit vectors are supported now");
-    additional_frame_bytes = 128;
+    // Save upper half of YMM registers
+    additional_frame_bytes = ymm_bytes;
+    if (UseAVX > 2) {
+      // Save upper half of ZMM registers as well
+      additional_frame_bytes += zmm_bytes;
+    }
   }
 #else
   assert(!restore_vectors, "vectors are generated only by C2");
 #endif
 
-  if (restore_vectors) {
-    assert(additional_frame_bytes == 128, "");
-    if (UseAVX > 2) {
-      // Restore upper half of ZMM registers.
-      for (int n = 0; n < num_xmm_regs; n++) {
-        __ vinsertf64x4h(as_XMMRegister(n), Address(rsp, n*32), 1);
-      }
-      __ addptr(rsp, additional_frame_bytes*2); // Save upper half of ZMM registes
-    }
-    // Restore upper half of YMM registes.
-    for (int n = 0; n < num_xmm_regs; n++) {
-      __ vinsertf128h(as_XMMRegister(n), Address(rsp, n*16));
-    }
-    __ addptr(rsp, additional_frame_bytes); // Save upper half of YMM registes
-  }
-
   int off = xmm0_off;
   int delta = xmm1_off - off;
 
   if (UseSSE == 1) {
+    // Restore XMM registers
+    assert(additional_frame_bytes == 0, "");
     for (int n = 0; n < num_xmm_regs; n++) {
       __ movflt(as_XMMRegister(n), Address(rsp, off*wordSize));
       off += delta;
     }
   } else if (UseSSE >= 2) {
-    // additional_frame_bytes only populated for the restore_vector case, else it is 0
+    // Restore whole 128bit (16 bytes) XMM registers. Do this before restoring YMM and
+    // ZMM because the movdqu instruction zeros the upper part of the XMM register.
     for (int n = 0; n < num_xmm_regs; n++) {
       __ movdqu(as_XMMRegister(n), Address(rsp, off*wordSize+additional_frame_bytes));
       off += delta;
     }
   }
 
+  if (restore_vectors) {
+    if (UseAVX > 2) {
+      // Restore upper half of ZMM registers.
+      for (int n = 0; n < num_xmm_regs; n++) {
+        __ vinsertf64x4h(as_XMMRegister(n), Address(rsp, n*32), 1);
+      }
+      __ addptr(rsp, zmm_bytes);
+    }
+    // Restore upper half of YMM registers.
+    for (int n = 0; n < num_xmm_regs; n++) {
+      __ vinsertf128h(as_XMMRegister(n), Address(rsp, n*16));
+    }
+    __ addptr(rsp, ymm_bytes);
+  }
+
   __ pop_FPU_state();
   __ addptr(rsp, FPU_regs_live*wordSize); // Pop FPU registers
 
@@ -306,7 +322,6 @@
   __ popa();
   // Get the rbp, described implicitly by the frame sender code (no oopMap)
   __ pop(rbp);
-
 }
 
 void RegisterSaver::restore_result_registers(MacroAssembler* masm) {