8149655: PPC64: Implement CompactString intrinsics
authormdoerr
Fri, 19 Feb 2016 11:09:59 +0100
changeset 36303 6241574f5982
parent 36302 23a79c43ba92
child 36304 3a742df23055
8149655: PPC64: Implement CompactString intrinsics Reviewed-by: goetz, kvn
hotspot/src/cpu/ppc/vm/globals_ppc.hpp
hotspot/src/cpu/ppc/vm/macroAssembler_ppc.cpp
hotspot/src/cpu/ppc/vm/macroAssembler_ppc.hpp
hotspot/src/cpu/ppc/vm/ppc.ad
hotspot/src/cpu/ppc/vm/stubGenerator_ppc.cpp
hotspot/src/cpu/ppc/vm/vm_version_ppc.cpp
hotspot/src/cpu/ppc/vm/vm_version_ppc.hpp
hotspot/test/compiler/intrinsics/string/TestStringIntrinsics2.java
--- a/hotspot/src/cpu/ppc/vm/globals_ppc.hpp	Fri Feb 19 20:45:26 2016 +0300
+++ b/hotspot/src/cpu/ppc/vm/globals_ppc.hpp	Fri Feb 19 11:09:59 2016 +0100
@@ -1,6 +1,6 @@
 /*
- * Copyright (c) 2002, 2015, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2012, 2015 SAP SE. All rights reserved.
+ * Copyright (c) 2002, 2016, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2012, 2016 SAP SE. 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
@@ -75,8 +75,7 @@
 
 define_pd_global(uintx, TypeProfileLevel, 111);
 
-// No performance work done here yet.
-define_pd_global(bool, CompactStrings, false);
+define_pd_global(bool, CompactStrings, true);
 
 // Platform dependent flag handling: flags only defined on this platform.
 #define ARCH_FLAGS(develop, product, diagnostic, experimental, notproduct, range, constraint)  \
--- a/hotspot/src/cpu/ppc/vm/macroAssembler_ppc.cpp	Fri Feb 19 20:45:26 2016 +0300
+++ b/hotspot/src/cpu/ppc/vm/macroAssembler_ppc.cpp	Fri Feb 19 11:09:59 2016 +0100
@@ -45,6 +45,9 @@
 #include "gc/g1/g1SATBCardTableModRefBS.hpp"
 #include "gc/g1/heapRegion.hpp"
 #endif // INCLUDE_ALL_GCS
+#ifdef COMPILER2
+#include "opto/intrinsicnode.hpp"
+#endif
 
 #ifdef PRODUCT
 #define BLOCK_COMMENT(str) // nothing
@@ -3168,6 +3171,553 @@
 
 /////////////////////////////////////////// String intrinsics ////////////////////////////////////////////
 
+#ifdef COMPILER2
+// Intrinsics for CompactStrings
+
+// Compress char[] to byte[] by compressing 16 bytes at once.
+void MacroAssembler::string_compress_16(Register src, Register dst, Register cnt,
+                                        Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5,
+                                        Label& Lfailure) {
+
+  const Register tmp0 = R0;
+  assert_different_registers(src, dst, cnt, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5);
+  Label Lloop, Lslow;
+
+  // Check if cnt >= 8 (= 16 bytes)
+  lis(tmp1, 0xFF);                // tmp1 = 0x00FF00FF00FF00FF
+  srwi_(tmp2, cnt, 3);
+  beq(CCR0, Lslow);
+  ori(tmp1, tmp1, 0xFF);
+  rldimi(tmp1, tmp1, 32, 0);
+  mtctr(tmp2);
+
+  // 2x unrolled loop
+  bind(Lloop);
+  ld(tmp2, 0, src);               // _0_1_2_3 (Big Endian)
+  ld(tmp4, 8, src);               // _4_5_6_7
+
+  orr(tmp0, tmp2, tmp4);
+  rldicl(tmp3, tmp2, 6*8, 64-24); // _____1_2
+  rldimi(tmp2, tmp2, 2*8, 2*8);   // _0_2_3_3
+  rldicl(tmp5, tmp4, 6*8, 64-24); // _____5_6
+  rldimi(tmp4, tmp4, 2*8, 2*8);   // _4_6_7_7
+
+  andc_(tmp0, tmp0, tmp1);
+  bne(CCR0, Lfailure);            // Not latin1.
+  addi(src, src, 16);
+
+  rlwimi(tmp3, tmp2, 0*8, 24, 31);// _____1_3
+  srdi(tmp2, tmp2, 3*8);          // ____0_2_
+  rlwimi(tmp5, tmp4, 0*8, 24, 31);// _____5_7
+  srdi(tmp4, tmp4, 3*8);          // ____4_6_
+
+  orr(tmp2, tmp2, tmp3);          // ____0123
+  orr(tmp4, tmp4, tmp5);          // ____4567
+
+  stw(tmp2, 0, dst);
+  stw(tmp4, 4, dst);
+  addi(dst, dst, 8);
+  bdnz(Lloop);
+
+  bind(Lslow);                    // Fallback to slow version
+}
+
+// Compress char[] to byte[]. cnt must be positive int.
+void MacroAssembler::string_compress(Register src, Register dst, Register cnt, Register tmp, Label& Lfailure) {
+  Label Lloop;
+  mtctr(cnt);
+
+  bind(Lloop);
+  lhz(tmp, 0, src);
+  cmplwi(CCR0, tmp, 0xff);
+  bgt(CCR0, Lfailure);            // Not latin1.
+  addi(src, src, 2);
+  stb(tmp, 0, dst);
+  addi(dst, dst, 1);
+  bdnz(Lloop);
+}
+
+// Inflate byte[] to char[] by inflating 16 bytes at once.
+void MacroAssembler::string_inflate_16(Register src, Register dst, Register cnt,
+                                       Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5) {
+  const Register tmp0 = R0;
+  assert_different_registers(src, dst, cnt, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5);
+  Label Lloop, Lslow;
+
+  // Check if cnt >= 8
+  srwi_(tmp2, cnt, 3);
+  beq(CCR0, Lslow);
+  lis(tmp1, 0xFF);                // tmp1 = 0x00FF00FF
+  ori(tmp1, tmp1, 0xFF);
+  mtctr(tmp2);
+
+  // 2x unrolled loop
+  bind(Lloop);
+  lwz(tmp2, 0, src);              // ____0123 (Big Endian)
+  lwz(tmp4, 4, src);              // ____4567
+  addi(src, src, 8);
+
+  rldicl(tmp3, tmp2, 7*8, 64-8);  // _______2
+  rlwimi(tmp2, tmp2, 3*8, 16, 23);// ____0113
+  rldicl(tmp5, tmp4, 7*8, 64-8);  // _______6
+  rlwimi(tmp4, tmp4, 3*8, 16, 23);// ____4557
+
+  andc(tmp0, tmp2, tmp1);         // ____0_1_
+  rlwimi(tmp2, tmp3, 2*8, 0, 23); // _____2_3
+  andc(tmp3, tmp4, tmp1);         // ____4_5_
+  rlwimi(tmp4, tmp5, 2*8, 0, 23); // _____6_7
+
+  rldimi(tmp2, tmp0, 3*8, 0*8);   // _0_1_2_3
+  rldimi(tmp4, tmp3, 3*8, 0*8);   // _4_5_6_7
+
+  std(tmp2, 0, dst);
+  std(tmp4, 8, dst);
+  addi(dst, dst, 16);
+  bdnz(Lloop);
+
+  bind(Lslow);                    // Fallback to slow version
+}
+
+// Inflate byte[] to char[]. cnt must be positive int.
+void MacroAssembler::string_inflate(Register src, Register dst, Register cnt, Register tmp) {
+  Label Lloop;
+  mtctr(cnt);
+
+  bind(Lloop);
+  lbz(tmp, 0, src);
+  addi(src, src, 1);
+  sth(tmp, 0, dst);
+  addi(dst, dst, 2);
+  bdnz(Lloop);
+}
+
+void MacroAssembler::string_compare(Register str1, Register str2,
+                                    Register cnt1, Register cnt2,
+                                    Register tmp1, Register result, int ae) {
+  const Register tmp0 = R0,
+                 diff = tmp1;
+
+  assert_different_registers(str1, str2, cnt1, cnt2, tmp0, tmp1, result);
+  Label Ldone, Lslow, Lloop, Lreturn_diff;
+
+  // Note: Making use of the fact that compareTo(a, b) == -compareTo(b, a)
+  // we interchange str1 and str2 in the UL case and negate the result.
+  // Like this, str1 is always latin1 encoded, except for the UU case.
+  // In addition, we need 0 (or sign which is 0) extend.
+
+  if (ae == StrIntrinsicNode::UU) {
+    srwi(cnt1, cnt1, 1);
+  } else {
+    clrldi(cnt1, cnt1, 32);
+  }
+
+  if (ae != StrIntrinsicNode::LL) {
+    srwi(cnt2, cnt2, 1);
+  } else {
+    clrldi(cnt2, cnt2, 32);
+  }
+
+  // See if the lengths are different, and calculate min in cnt1.
+  // Save diff in case we need it for a tie-breaker.
+  subf_(diff, cnt2, cnt1); // diff = cnt1 - cnt2
+  // if (diff > 0) { cnt1 = cnt2; }
+  if (VM_Version::has_isel()) {
+    isel(cnt1, CCR0, Assembler::greater, /*invert*/ false, cnt2);
+  } else {
+    Label Lskip;
+    blt(CCR0, Lskip);
+    mr(cnt1, cnt2);
+    bind(Lskip);
+  }
+
+  // Rename registers
+  Register chr1 = result;
+  Register chr2 = tmp0;
+
+  // Compare multiple characters in fast loop (only implemented for same encoding).
+  int stride1 = 8, stride2 = 8;
+  if (ae == StrIntrinsicNode::LL || ae == StrIntrinsicNode::UU) {
+    int log2_chars_per_iter = (ae == StrIntrinsicNode::LL) ? 3 : 2;
+    Label Lfastloop, Lskipfast;
+
+    srwi_(tmp0, cnt1, log2_chars_per_iter);
+    beq(CCR0, Lskipfast);
+    rldicl(cnt2, cnt1, 0, 64 - log2_chars_per_iter); // Remaining characters.
+    li(cnt1, 1 << log2_chars_per_iter); // Initialize for failure case: Rescan characters from current iteration.
+    mtctr(tmp0);
+
+    bind(Lfastloop);
+    ld(chr1, 0, str1);
+    ld(chr2, 0, str2);
+    cmpd(CCR0, chr1, chr2);
+    bne(CCR0, Lslow);
+    addi(str1, str1, stride1);
+    addi(str2, str2, stride2);
+    bdnz(Lfastloop);
+    mr(cnt1, cnt2); // Remaining characters.
+    bind(Lskipfast);
+  }
+
+  // Loop which searches the first difference character by character.
+  cmpwi(CCR0, cnt1, 0);
+  beq(CCR0, Lreturn_diff);
+  bind(Lslow);
+  mtctr(cnt1);
+
+  switch (ae) {
+    case StrIntrinsicNode::LL: stride1 = 1; stride2 = 1; break;
+    case StrIntrinsicNode::UL: // fallthru (see comment above)
+    case StrIntrinsicNode::LU: stride1 = 1; stride2 = 2; break;
+    case StrIntrinsicNode::UU: stride1 = 2; stride2 = 2; break;
+    default: ShouldNotReachHere(); break;
+  }
+
+  bind(Lloop);
+  if (stride1 == 1) { lbz(chr1, 0, str1); } else { lhz(chr1, 0, str1); }
+  if (stride2 == 1) { lbz(chr2, 0, str2); } else { lhz(chr2, 0, str2); }
+  subf_(result, chr2, chr1); // result = chr1 - chr2
+  bne(CCR0, Ldone);
+  addi(str1, str1, stride1);
+  addi(str2, str2, stride2);
+  bdnz(Lloop);
+
+  // If strings are equal up to min length, return the length difference.
+  bind(Lreturn_diff);
+  mr(result, diff);
+
+  // Otherwise, return the difference between the first mismatched chars.
+  bind(Ldone);
+  if (ae == StrIntrinsicNode::UL) {
+    neg(result, result); // Negate result (see note above).
+  }
+}
+
+void MacroAssembler::array_equals(bool is_array_equ, Register ary1, Register ary2,
+                                  Register limit, Register tmp1, Register result, bool is_byte) {
+  const Register tmp0 = R0;
+  assert_different_registers(ary1, ary2, limit, tmp0, tmp1, result);
+  Label Ldone, Lskiploop, Lloop, Lfastloop, Lskipfast;
+  bool limit_needs_shift = false;
+
+  if (is_array_equ) {
+    const int length_offset = arrayOopDesc::length_offset_in_bytes();
+    const int base_offset   = arrayOopDesc::base_offset_in_bytes(is_byte ? T_BYTE : T_CHAR);
+
+    // Return true if the same array.
+    cmpd(CCR0, ary1, ary2);
+    beq(CCR0, Lskiploop);
+
+    // Return false if one of them is NULL.
+    cmpdi(CCR0, ary1, 0);
+    cmpdi(CCR1, ary2, 0);
+    li(result, 0);
+    cror(CCR0, Assembler::equal, CCR1, Assembler::equal);
+    beq(CCR0, Ldone);
+
+    // Load the lengths of arrays.
+    lwz(limit, length_offset, ary1);
+    lwz(tmp0, length_offset, ary2);
+
+    // Return false if the two arrays are not equal length.
+    cmpw(CCR0, limit, tmp0);
+    bne(CCR0, Ldone);
+
+    // Load array addresses.
+    addi(ary1, ary1, base_offset);
+    addi(ary2, ary2, base_offset);
+  } else {
+    limit_needs_shift = !is_byte;
+    li(result, 0); // Assume not equal.
+  }
+
+  // Rename registers
+  Register chr1 = tmp0;
+  Register chr2 = tmp1;
+
+  // Compare 8 bytes per iteration in fast loop.
+  const int log2_chars_per_iter = is_byte ? 3 : 2;
+
+  srwi_(tmp0, limit, log2_chars_per_iter + (limit_needs_shift ? 1 : 0));
+  beq(CCR0, Lskipfast);
+  mtctr(tmp0);
+
+  bind(Lfastloop);
+  ld(chr1, 0, ary1);
+  ld(chr2, 0, ary2);
+  addi(ary1, ary1, 8);
+  addi(ary2, ary2, 8);
+  cmpd(CCR0, chr1, chr2);
+  bne(CCR0, Ldone);
+  bdnz(Lfastloop);
+
+  bind(Lskipfast);
+  rldicl_(limit, limit, limit_needs_shift ? 64 - 1 : 0, 64 - log2_chars_per_iter); // Remaining characters.
+  beq(CCR0, Lskiploop);
+  mtctr(limit);
+
+  // Character by character.
+  bind(Lloop);
+  if (is_byte) {
+    lbz(chr1, 0, ary1);
+    lbz(chr2, 0, ary2);
+    addi(ary1, ary1, 1);
+    addi(ary2, ary2, 1);
+  } else {
+    lhz(chr1, 0, ary1);
+    lhz(chr2, 0, ary2);
+    addi(ary1, ary1, 2);
+    addi(ary2, ary2, 2);
+  }
+  cmpw(CCR0, chr1, chr2);
+  bne(CCR0, Ldone);
+  bdnz(Lloop);
+
+  bind(Lskiploop);
+  li(result, 1); // All characters are equal.
+  bind(Ldone);
+}
+
+void MacroAssembler::string_indexof(Register result, Register haystack, Register haycnt,
+                                    Register needle, ciTypeArray* needle_values, Register needlecnt, int needlecntval,
+                                    Register tmp1, Register tmp2, Register tmp3, Register tmp4, int ae) {
+
+  // Ensure 0<needlecnt<=haycnt in ideal graph as prerequisite!
+  Label L_TooShort, L_Found, L_NotFound, L_End;
+  Register last_addr = haycnt, // Kill haycnt at the beginning.
+  addr      = tmp1,
+  n_start   = tmp2,
+  ch1       = tmp3,
+  ch2       = R0;
+
+  assert(ae != StrIntrinsicNode::LU, "Invalid encoding");
+  const int h_csize = (ae == StrIntrinsicNode::LL) ? 1 : 2;
+  const int n_csize = (ae == StrIntrinsicNode::UU) ? 2 : 1;
+
+  // **************************************************************************************************
+  // Prepare for main loop: optimized for needle count >=2, bail out otherwise.
+  // **************************************************************************************************
+
+  // Compute last haystack addr to use if no match gets found.
+  clrldi(haycnt, haycnt, 32);         // Ensure positive int is valid as 64 bit value.
+  addi(addr, haystack, -h_csize);     // Accesses use pre-increment.
+  if (needlecntval == 0) { // variable needlecnt
+   cmpwi(CCR6, needlecnt, 2);
+   clrldi(needlecnt, needlecnt, 32);  // Ensure positive int is valid as 64 bit value.
+   blt(CCR6, L_TooShort);             // Variable needlecnt: handle short needle separately.
+  }
+
+  if (n_csize == 2) { lwz(n_start, 0, needle); } else { lhz(n_start, 0, needle); } // Load first 2 characters of needle.
+
+  if (needlecntval == 0) { // variable needlecnt
+   subf(ch1, needlecnt, haycnt);      // Last character index to compare is haycnt-needlecnt.
+   addi(needlecnt, needlecnt, -2);    // Rest of needle.
+  } else { // constant needlecnt
+  guarantee(needlecntval != 1, "IndexOf with single-character needle must be handled separately");
+  assert((needlecntval & 0x7fff) == needlecntval, "wrong immediate");
+   addi(ch1, haycnt, -needlecntval);  // Last character index to compare is haycnt-needlecnt.
+   if (needlecntval > 3) { li(needlecnt, needlecntval - 2); } // Rest of needle.
+  }
+
+  if (h_csize == 2) { slwi(ch1, ch1, 1); } // Scale to number of bytes.
+
+  if (ae ==StrIntrinsicNode::UL) {
+   srwi(tmp4, n_start, 1*8);          // ___0
+   rlwimi(n_start, tmp4, 2*8, 0, 23); // _0_1
+  }
+
+  add(last_addr, haystack, ch1);      // Point to last address to compare (haystack+2*(haycnt-needlecnt)).
+
+  // Main Loop (now we have at least 2 characters).
+  Label L_OuterLoop, L_InnerLoop, L_FinalCheck, L_Comp1, L_Comp2;
+  bind(L_OuterLoop); // Search for 1st 2 characters.
+  Register addr_diff = tmp4;
+   subf(addr_diff, addr, last_addr);  // Difference between already checked address and last address to check.
+   addi(addr, addr, h_csize);         // This is the new address we want to use for comparing.
+   srdi_(ch2, addr_diff, h_csize);
+   beq(CCR0, L_FinalCheck);           // 2 characters left?
+   mtctr(ch2);                        // num of characters / 2
+  bind(L_InnerLoop);                  // Main work horse (2x unrolled search loop)
+   if (h_csize == 2) {                // Load 2 characters of haystack (ignore alignment).
+    lwz(ch1, 0, addr);
+    lwz(ch2, 2, addr);
+   } else {
+    lhz(ch1, 0, addr);
+    lhz(ch2, 1, addr);
+   }
+   cmpw(CCR0, ch1, n_start);          // Compare 2 characters (1 would be sufficient but try to reduce branches to CompLoop).
+   cmpw(CCR1, ch2, n_start);
+   beq(CCR0, L_Comp1);                // Did we find the needle start?
+   beq(CCR1, L_Comp2);
+   addi(addr, addr, 2 * h_csize);
+   bdnz(L_InnerLoop);
+  bind(L_FinalCheck);
+   andi_(addr_diff, addr_diff, h_csize); // Remaining characters not covered by InnerLoop: (num of characters) & 1.
+   beq(CCR0, L_NotFound);
+   if (h_csize == 2) { lwz(ch1, 0, addr); } else { lhz(ch1, 0, addr); } // One position left at which we have to compare.
+   cmpw(CCR1, ch1, n_start);
+   beq(CCR1, L_Comp1);
+  bind(L_NotFound);
+   li(result, -1);                    // not found
+   b(L_End);
+
+   // **************************************************************************************************
+   // Special Case: unfortunately, the variable needle case can be called with needlecnt<2
+   // **************************************************************************************************
+  if (needlecntval == 0) {           // We have to handle these cases separately.
+  Label L_OneCharLoop;
+  bind(L_TooShort);
+   mtctr(haycnt);
+   if (n_csize == 2) { lhz(n_start, 0, needle); } else { lbz(n_start, 0, needle); } // First character of needle
+  bind(L_OneCharLoop);
+   if (h_csize == 2) { lhzu(ch1, 2, addr); } else { lbzu(ch1, 1, addr); }
+   cmpw(CCR1, ch1, n_start);
+   beq(CCR1, L_Found);               // Did we find the one character needle?
+   bdnz(L_OneCharLoop);
+   li(result, -1);                   // Not found.
+   b(L_End);
+  }
+
+  // **************************************************************************************************
+  // Regular Case Part II: compare rest of needle (first 2 characters have been compared already)
+  // **************************************************************************************************
+
+  // Compare the rest
+  bind(L_Comp2);
+   addi(addr, addr, h_csize);        // First comparison has failed, 2nd one hit.
+  bind(L_Comp1);                     // Addr points to possible needle start.
+  if (needlecntval != 2) {           // Const needlecnt==2?
+   if (needlecntval != 3) {
+    if (needlecntval == 0) { beq(CCR6, L_Found); } // Variable needlecnt==2?
+    Register n_ind = tmp4,
+             h_ind = n_ind;
+    li(n_ind, 2 * n_csize);          // First 2 characters are already compared, use index 2.
+    mtctr(needlecnt);                // Decremented by 2, still > 0.
+   Label L_CompLoop;
+   bind(L_CompLoop);
+    if (ae ==StrIntrinsicNode::UL) {
+      h_ind = ch1;
+      sldi(h_ind, n_ind, 1);
+    }
+    if (n_csize == 2) { lhzx(ch2, needle, n_ind); } else { lbzx(ch2, needle, n_ind); }
+    if (h_csize == 2) { lhzx(ch1, addr, h_ind); } else { lbzx(ch1, addr, h_ind); }
+    cmpw(CCR1, ch1, ch2);
+    bne(CCR1, L_OuterLoop);
+    addi(n_ind, n_ind, n_csize);
+    bdnz(L_CompLoop);
+   } else { // No loop required if there's only one needle character left.
+    if (n_csize == 2) { lhz(ch2, 2 * 2, needle); } else { lbz(ch2, 2 * 1, needle); }
+    if (h_csize == 2) { lhz(ch1, 2 * 2, addr); } else { lbz(ch1, 2 * 1, addr); }
+    cmpw(CCR1, ch1, ch2);
+    bne(CCR1, L_OuterLoop);
+   }
+  }
+  // Return index ...
+  bind(L_Found);
+   subf(result, haystack, addr);     // relative to haystack, ...
+   if (h_csize == 2) { srdi(result, result, 1); } // in characters.
+  bind(L_End);
+} // string_indexof
+
+void MacroAssembler::string_indexof_char(Register result, Register haystack, Register haycnt,
+                                         Register needle, jchar needleChar, Register tmp1, Register tmp2, bool is_byte) {
+  assert_different_registers(haystack, haycnt, needle, tmp1, tmp2);
+
+  Label L_InnerLoop, L_FinalCheck, L_Found1, L_Found2, L_NotFound, L_End;
+  Register addr = tmp1,
+           ch1 = tmp2,
+           ch2 = R0;
+
+  const int h_csize = is_byte ? 1 : 2;
+
+//4:
+   srwi_(tmp2, haycnt, 1);   // Shift right by exact_log2(UNROLL_FACTOR).
+   mr(addr, haystack);
+   beq(CCR0, L_FinalCheck);
+   mtctr(tmp2);              // Move to count register.
+//8:
+  bind(L_InnerLoop);         // Main work horse (2x unrolled search loop).
+   if (!is_byte) {
+    lhz(ch1, 0, addr);
+    lhz(ch2, 2, addr);
+   } else {
+    lbz(ch1, 0, addr);
+    lbz(ch2, 1, addr);
+   }
+   (needle != R0) ? cmpw(CCR0, ch1, needle) : cmplwi(CCR0, ch1, (unsigned int)needleChar);
+   (needle != R0) ? cmpw(CCR1, ch2, needle) : cmplwi(CCR1, ch2, (unsigned int)needleChar);
+   beq(CCR0, L_Found1);      // Did we find the needle?
+   beq(CCR1, L_Found2);
+   addi(addr, addr, 2 * h_csize);
+   bdnz(L_InnerLoop);
+//16:
+  bind(L_FinalCheck);
+   andi_(R0, haycnt, 1);
+   beq(CCR0, L_NotFound);
+   if (!is_byte) { lhz(ch1, 0, addr); } else { lbz(ch1, 0, addr); } // One position left at which we have to compare.
+   (needle != R0) ? cmpw(CCR1, ch1, needle) : cmplwi(CCR1, ch1, (unsigned int)needleChar);
+   beq(CCR1, L_Found1);
+//21:
+  bind(L_NotFound);
+   li(result, -1);           // Not found.
+   b(L_End);
+
+  bind(L_Found2);
+   addi(addr, addr, h_csize);
+//24:
+  bind(L_Found1);            // Return index ...
+   subf(result, haystack, addr); // relative to haystack, ...
+   if (!is_byte) { srdi(result, result, 1); } // in characters.
+  bind(L_End);
+} // string_indexof_char
+
+
+void MacroAssembler::has_negatives(Register src, Register cnt, Register result,
+                                   Register tmp1, Register tmp2) {
+  const Register tmp0 = R0;
+  assert_different_registers(src, result, cnt, tmp0, tmp1, tmp2);
+  Label Lfastloop, Lslow, Lloop, Lnoneg, Ldone;
+
+  // Check if cnt >= 8 (= 16 bytes)
+  lis(tmp1, (int)(short)0x8080);  // tmp1 = 0x8080808080808080
+  srwi_(tmp2, cnt, 4);
+  li(result, 1);                  // Assume there's a negative byte.
+  beq(CCR0, Lslow);
+  ori(tmp1, tmp1, 0x8080);
+  rldimi(tmp1, tmp1, 32, 0);
+  mtctr(tmp2);
+
+  // 2x unrolled loop
+  bind(Lfastloop);
+  ld(tmp2, 0, src);
+  ld(tmp0, 8, src);
+
+  orr(tmp0, tmp2, tmp0);
+
+  and_(tmp0, tmp0, tmp1);
+  bne(CCR0, Ldone);               // Found negative byte.
+  addi(src, src, 16);
+
+  bdnz(Lfastloop);
+
+  bind(Lslow);                    // Fallback to slow version
+  rldicl_(tmp0, cnt, 0, 64-4);
+  beq(CCR0, Lnoneg);
+  mtctr(tmp0);
+  bind(Lloop);
+  lbz(tmp0, 0, src);
+  addi(src, src, 1);
+  andi_(tmp0, tmp0, 0x80);
+  bne(CCR0, Ldone);               // Found negative byte.
+  bdnz(Lloop);
+  bind(Lnoneg);
+  li(result, 0);
+
+  bind(Ldone);
+}
+
+
+// Intrinsics for non-CompactStrings
+
 // Search for a single jchar in an jchar[].
 //
 // Assumes that result differs from all other registers.
@@ -3613,6 +4163,8 @@
   bind(Ldone_false);
 }
 
+#endif // Compiler2
+
 // Helpers for Intrinsic Emitters
 //
 // Revert the byte order of a 32bit value in a register
--- a/hotspot/src/cpu/ppc/vm/macroAssembler_ppc.hpp	Fri Feb 19 20:45:26 2016 +0300
+++ b/hotspot/src/cpu/ppc/vm/macroAssembler_ppc.hpp	Fri Feb 19 11:09:59 2016 +0100
@@ -1,6 +1,6 @@
 /*
- * Copyright (c) 2002, 2015, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2012, 2015 SAP SE. All rights reserved.
+ * Copyright (c) 2002, 2016, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2012, 2016 SAP SE. 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
@@ -679,6 +679,39 @@
 
   void clear_memory_doubleword(Register base_ptr, Register cnt_dwords, Register tmp = R0);
 
+#ifdef COMPILER2
+  // Intrinsics for CompactStrings
+  // Compress char[] to byte[] by compressing 16 bytes at once.
+  void string_compress_16(Register src, Register dst, Register cnt,
+                          Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5,
+                          Label& Lfailure);
+
+  // Compress char[] to byte[]. cnt must be positive int.
+  void string_compress(Register src, Register dst, Register cnt, Register tmp, Label& Lfailure);
+
+  // Inflate byte[] to char[] by inflating 16 bytes at once.
+  void string_inflate_16(Register src, Register dst, Register cnt,
+                         Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5);
+
+  // Inflate byte[] to char[]. cnt must be positive int.
+  void string_inflate(Register src, Register dst, Register cnt, Register tmp);
+
+  void string_compare(Register str1, Register str2, Register cnt1, Register cnt2,
+                      Register tmp1, Register result, int ae);
+
+  void array_equals(bool is_array_equ, Register ary1, Register ary2,
+                    Register limit, Register tmp1, Register result, bool is_byte);
+
+  void string_indexof(Register result, Register haystack, Register haycnt,
+                      Register needle, ciTypeArray* needle_values, Register needlecnt, int needlecntval,
+                      Register tmp1, Register tmp2, Register tmp3, Register tmp4, int ae);
+
+  void string_indexof_char(Register result, Register haystack, Register haycnt,
+                           Register needle, jchar needleChar, Register tmp1, Register tmp2, bool is_byte);
+
+  void has_negatives(Register src, Register cnt, Register result, Register tmp1, Register tmp2);
+
+  // Intrinsics for non-CompactStrings
   // Needle of length 1.
   void string_indexof_1(Register result, Register haystack, Register haycnt,
                         Register needle, jchar needleChar,
@@ -694,6 +727,7 @@
                           Register tmp5_reg);
   void char_arrays_equalsImm(Register str1_reg, Register str2_reg, int cntval, Register result_reg,
                              Register tmp1_reg, Register tmp2_reg);
+#endif
 
   // Emitters for BigInteger.multiplyToLen intrinsic.
   inline void multiply64(Register dest_hi, Register dest_lo,
--- a/hotspot/src/cpu/ppc/vm/ppc.ad	Fri Feb 19 20:45:26 2016 +0300
+++ b/hotspot/src/cpu/ppc/vm/ppc.ad	Fri Feb 19 11:09:59 2016 +0100
@@ -1,6 +1,6 @@
 //
 // Copyright (c) 2011, 2016, Oracle and/or its affiliates. All rights reserved.
-// Copyright (c) 2012, 2015 SAP SE. All rights reserved.
+// Copyright (c) 2012, 2016 SAP SE. 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
@@ -2024,13 +2024,13 @@
     return (UsePopCountInstruction && VM_Version::has_popcntw());
 
   case Op_StrComp:
-    return SpecialStringCompareTo && !CompactStrings;
+    return SpecialStringCompareTo;
   case Op_StrEquals:
-    return SpecialStringEquals && !CompactStrings;
+    return SpecialStringEquals;
   case Op_StrIndexOf:
-    return SpecialStringIndexOf && !CompactStrings;
+    return SpecialStringIndexOf;
   case Op_StrIndexOfChar:
-    return SpecialStringIndexOf && !CompactStrings;
+    return SpecialStringIndexOf;
   }
 
   return true;  // Per default match rules are supported.
@@ -11022,6 +11022,584 @@
   ins_pipe(pipe_class_default);
 %}
 
+instruct string_compareL(rarg1RegP str1, rarg2RegP str2, rarg3RegI cnt1, rarg4RegI cnt2, iRegIdst result,
+                         iRegIdst tmp, regCTR ctr, flagsRegCR0 cr0) %{
+  predicate(((StrCompNode*)n)->encoding() == StrIntrinsicNode::LL);
+  match(Set result (StrComp (Binary str1 cnt1) (Binary str2 cnt2)));
+  effect(TEMP_DEF result, USE_KILL str1, USE_KILL str2, USE_KILL cnt1, USE_KILL cnt2, KILL ctr, KILL cr0, TEMP tmp);
+  ins_cost(300);
+  format %{ "String Compare byte[] $str1,$cnt1,$str2,$cnt2 -> $result \t// KILL $tmp" %}
+  ins_encode %{
+    // TODO: PPC port $archOpcode(ppc64Opcode_compound);
+    __ string_compare($str1$$Register, $str2$$Register,
+                      $cnt1$$Register, $cnt2$$Register,
+                      $tmp$$Register,
+                      $result$$Register, StrIntrinsicNode::LL);
+  %}
+  ins_pipe(pipe_class_default);
+%}
+
+instruct string_compareU(rarg1RegP str1, rarg2RegP str2, rarg3RegI cnt1, rarg4RegI cnt2, iRegIdst result,
+                         iRegIdst tmp, regCTR ctr, flagsRegCR0 cr0) %{
+  predicate(((StrCompNode*)n)->encoding() == StrIntrinsicNode::UU);
+  match(Set result (StrComp (Binary str1 cnt1) (Binary str2 cnt2)));
+  effect(TEMP_DEF result, USE_KILL str1, USE_KILL str2, USE_KILL cnt1, USE_KILL cnt2, KILL ctr, KILL cr0, TEMP tmp);
+  ins_cost(300);
+  format %{ "String Compare char[] $str1,$cnt1,$str2,$cnt2 -> $result \t// KILL $tmp" %}
+  ins_encode %{
+    // TODO: PPC port $archOpcode(ppc64Opcode_compound);
+    __ string_compare($str1$$Register, $str2$$Register,
+                      $cnt1$$Register, $cnt2$$Register,
+                      $tmp$$Register,
+                      $result$$Register, StrIntrinsicNode::UU);
+  %}
+  ins_pipe(pipe_class_default);
+%}
+
+instruct string_compareLU(rarg1RegP str1, rarg2RegP str2, rarg3RegI cnt1, rarg4RegI cnt2, iRegIdst result,
+                          iRegIdst tmp, regCTR ctr, flagsRegCR0 cr0) %{
+  predicate(((StrCompNode*)n)->encoding() == StrIntrinsicNode::LU);
+  match(Set result (StrComp (Binary str1 cnt1) (Binary str2 cnt2)));
+  effect(TEMP_DEF result, USE_KILL str1, USE_KILL str2, USE_KILL cnt1, USE_KILL cnt2, KILL ctr, KILL cr0, TEMP tmp);
+  ins_cost(300);
+  format %{ "String Compare byte[] $str1,$cnt1,$str2,$cnt2 -> $result \t// KILL $tmp" %}
+  ins_encode %{
+    // TODO: PPC port $archOpcode(ppc64Opcode_compound);
+    __ string_compare($str1$$Register, $str2$$Register,
+                      $cnt1$$Register, $cnt2$$Register,
+                      $tmp$$Register,
+                      $result$$Register, StrIntrinsicNode::LU);
+  %}
+  ins_pipe(pipe_class_default);
+%}
+
+instruct string_compareUL(rarg1RegP str1, rarg2RegP str2, rarg3RegI cnt1, rarg4RegI cnt2, iRegIdst result,
+                          iRegIdst tmp, regCTR ctr, flagsRegCR0 cr0) %{
+  predicate(((StrCompNode*)n)->encoding() == StrIntrinsicNode::UL);
+  match(Set result (StrComp (Binary str1 cnt1) (Binary str2 cnt2)));
+  effect(TEMP_DEF result, USE_KILL str1, USE_KILL str2, USE_KILL cnt1, USE_KILL cnt2, KILL ctr, KILL cr0, TEMP tmp);
+  ins_cost(300);
+  format %{ "String Compare byte[] $str1,$cnt1,$str2,$cnt2 -> $result \t// KILL $tmp" %}
+  ins_encode %{
+    // TODO: PPC port $archOpcode(ppc64Opcode_compound);
+    __ string_compare($str2$$Register, $str1$$Register,
+                      $cnt2$$Register, $cnt1$$Register,
+                      $tmp$$Register,
+                      $result$$Register, StrIntrinsicNode::UL);
+  %}
+  ins_pipe(pipe_class_default);
+%}
+
+instruct string_equalsL(rarg1RegP str1, rarg2RegP str2, rarg3RegI cnt, iRegIdst result,
+                        iRegIdst tmp, regCTR ctr, flagsRegCR0 cr0) %{
+  predicate(((StrEqualsNode*)n)->encoding() == StrIntrinsicNode::LL);
+  match(Set result (StrEquals (Binary str1 str2) cnt));
+  effect(TEMP_DEF result, USE_KILL str1, USE_KILL str2, USE_KILL cnt, TEMP tmp, KILL ctr, KILL cr0);
+  ins_cost(300);
+  format %{ "String Equals byte[] $str1,$str2,$cnt -> $result \t// KILL $tmp" %}
+  ins_encode %{
+    // TODO: PPC port $archOpcode(ppc64Opcode_compound);
+    __ array_equals(false, $str1$$Register, $str2$$Register,
+                    $cnt$$Register, $tmp$$Register,
+                    $result$$Register, true /* byte */);
+  %}
+  ins_pipe(pipe_class_default);
+%}
+
+instruct string_equalsU(rarg1RegP str1, rarg2RegP str2, rarg3RegI cnt, iRegIdst result,
+                        iRegIdst tmp, regCTR ctr, flagsRegCR0 cr0) %{
+  predicate(((StrEqualsNode*)n)->encoding() == StrIntrinsicNode::UU);
+  match(Set result (StrEquals (Binary str1 str2) cnt));
+  effect(TEMP_DEF result, USE_KILL str1, USE_KILL str2, USE_KILL cnt, TEMP tmp, KILL ctr, KILL cr0);
+  ins_cost(300);
+  format %{ "String Equals char[]  $str1,$str2,$cnt -> $result \t// KILL $tmp" %}
+  ins_encode %{
+    // TODO: PPC port $archOpcode(ppc64Opcode_compound);
+    __ array_equals(false, $str1$$Register, $str2$$Register,
+                    $cnt$$Register, $tmp$$Register,
+                    $result$$Register, false /* byte */);
+  %}
+  ins_pipe(pipe_class_default);
+%}
+
+instruct array_equalsB(rarg1RegP ary1, rarg2RegP ary2, iRegIdst result,
+                       iRegIdst tmp1, iRegIdst tmp2, regCTR ctr, flagsRegCR0 cr0, flagsRegCR0 cr1) %{
+  predicate(((AryEqNode*)n)->encoding() == StrIntrinsicNode::LL);
+  match(Set result (AryEq ary1 ary2));
+  effect(TEMP_DEF result, USE_KILL ary1, USE_KILL ary2, TEMP tmp1, TEMP tmp2, KILL ctr, KILL cr0, KILL cr1);
+  ins_cost(300);
+  format %{ "Array Equals $ary1,$ary2 -> $result \t// KILL $tmp1,$tmp2" %}
+  ins_encode %{
+    // TODO: PPC port $archOpcode(ppc64Opcode_compound);
+    __ array_equals(true, $ary1$$Register, $ary2$$Register,
+                    $tmp1$$Register, $tmp2$$Register,
+                    $result$$Register, true /* byte */);
+  %}
+  ins_pipe(pipe_class_default);
+%}
+
+instruct array_equalsC(rarg1RegP ary1, rarg2RegP ary2, iRegIdst result,
+                       iRegIdst tmp1, iRegIdst tmp2, regCTR ctr, flagsRegCR0 cr0, flagsRegCR0 cr1) %{
+  predicate(((AryEqNode*)n)->encoding() == StrIntrinsicNode::UU);
+  match(Set result (AryEq ary1 ary2));
+  effect(TEMP_DEF result, USE_KILL ary1, USE_KILL ary2, TEMP tmp1, TEMP tmp2, KILL ctr, KILL cr0, KILL cr1);
+  ins_cost(300);
+  format %{ "Array Equals $ary1,$ary2 -> $result \t// KILL $tmp1,$tmp2" %}
+  ins_encode %{
+    // TODO: PPC port $archOpcode(ppc64Opcode_compound);
+    __ array_equals(true, $ary1$$Register, $ary2$$Register,
+                    $tmp1$$Register, $tmp2$$Register,
+                    $result$$Register, false /* byte */);
+  %}
+  ins_pipe(pipe_class_default);
+%}
+
+instruct indexOf_imm1_char_U(iRegIdst result, iRegPsrc haystack, iRegIsrc haycnt,
+                             immP needleImm, immL offsetImm, immI_1 needlecntImm,
+                             iRegIdst tmp1, iRegIdst tmp2,
+                             flagsRegCR0 cr0, flagsRegCR1 cr1, regCTR ctr) %{
+  match(Set result (StrIndexOf (Binary haystack haycnt) (Binary (AddP needleImm offsetImm) needlecntImm)));
+  effect(TEMP tmp1, TEMP tmp2, KILL cr0, KILL cr1, KILL ctr);
+  // Required for EA: check if it is still a type_array.
+  predicate(((StrIndexOfNode*)n)->encoding() == StrIntrinsicNode::UU);
+  ins_cost(150);
+
+  format %{ "String IndexOf CSCL1 $haystack[0..$haycnt], $needleImm+$offsetImm[0..$needlecntImm]"
+            "-> $result \t// KILL $haycnt, $tmp1, $tmp2, $cr0, $cr1" %}
+
+  ins_encode %{
+    // TODO: PPC port $archOpcode(ppc64Opcode_compound);
+    immPOper *needleOper = (immPOper *)$needleImm;
+    const TypeOopPtr *t = needleOper->type()->isa_oopptr();
+    ciTypeArray* needle_values = t->const_oop()->as_type_array();  // Pointer to live char *
+    jchar chr;
+#ifdef VM_LITTLE_ENDIAN
+    chr = (((jchar)(unsigned char)needle_values->element_value(1).as_byte()) << 8) |
+           ((jchar)(unsigned char)needle_values->element_value(0).as_byte());
+#else
+    chr = (((jchar)(unsigned char)needle_values->element_value(0).as_byte()) << 8) |
+           ((jchar)(unsigned char)needle_values->element_value(1).as_byte());
+#endif
+    __ string_indexof_char($result$$Register,
+                           $haystack$$Register, $haycnt$$Register,
+                           R0, chr,
+                           $tmp1$$Register, $tmp2$$Register, false /*is_byte*/);
+  %}
+  ins_pipe(pipe_class_compare);
+%}
+
+instruct indexOf_imm1_char_L(iRegIdst result, iRegPsrc haystack, iRegIsrc haycnt,
+                             immP needleImm, immL offsetImm, immI_1 needlecntImm,
+                             iRegIdst tmp1, iRegIdst tmp2,
+                             flagsRegCR0 cr0, flagsRegCR1 cr1, regCTR ctr) %{
+  match(Set result (StrIndexOf (Binary haystack haycnt) (Binary (AddP needleImm offsetImm) needlecntImm)));
+  effect(TEMP tmp1, TEMP tmp2, KILL cr0, KILL cr1, KILL ctr);
+  // Required for EA: check if it is still a type_array.
+  predicate(((StrIndexOfNode*)n)->encoding() == StrIntrinsicNode::LL);
+  ins_cost(150);
+
+  format %{ "String IndexOf CSCL1 $haystack[0..$haycnt], $needleImm+$offsetImm[0..$needlecntImm]"
+            "-> $result \t// KILL $haycnt, $tmp1, $tmp2, $cr0, $cr1" %}
+
+  ins_encode %{
+    // TODO: PPC port $archOpcode(ppc64Opcode_compound);
+    immPOper *needleOper = (immPOper *)$needleImm;
+    const TypeOopPtr *t = needleOper->type()->isa_oopptr();
+    ciTypeArray* needle_values = t->const_oop()->as_type_array();  // Pointer to live char *
+    jchar chr = (jchar)needle_values->element_value(0).as_byte();
+    __ string_indexof_char($result$$Register,
+                           $haystack$$Register, $haycnt$$Register,
+                           R0, chr,
+                           $tmp1$$Register, $tmp2$$Register, true /*is_byte*/);
+  %}
+  ins_pipe(pipe_class_compare);
+%}
+
+instruct indexOf_imm1_char_UL(iRegIdst result, iRegPsrc haystack, iRegIsrc haycnt,
+                              immP needleImm, immL offsetImm, immI_1 needlecntImm,
+                              iRegIdst tmp1, iRegIdst tmp2,
+                              flagsRegCR0 cr0, flagsRegCR1 cr1, regCTR ctr) %{
+  match(Set result (StrIndexOf (Binary haystack haycnt) (Binary (AddP needleImm offsetImm) needlecntImm)));
+  effect(TEMP tmp1, TEMP tmp2, KILL cr0, KILL cr1, KILL ctr);
+  // Required for EA: check if it is still a type_array.
+  predicate(((StrIndexOfNode*)n)->encoding() == StrIntrinsicNode::UL);
+  ins_cost(150);
+
+  format %{ "String IndexOf CSCL1 $haystack[0..$haycnt], $needleImm+$offsetImm[0..$needlecntImm]"
+            "-> $result \t// KILL $haycnt, $tmp1, $tmp2, $cr0, $cr1" %}
+
+  ins_encode %{
+    // TODO: PPC port $archOpcode(ppc64Opcode_compound);
+    immPOper *needleOper = (immPOper *)$needleImm;
+    const TypeOopPtr *t = needleOper->type()->isa_oopptr();
+    ciTypeArray* needle_values = t->const_oop()->as_type_array();  // Pointer to live char *
+    jchar chr = (jchar)needle_values->element_value(0).as_byte();
+    __ string_indexof_char($result$$Register,
+                           $haystack$$Register, $haycnt$$Register,
+                           R0, chr,
+                           $tmp1$$Register, $tmp2$$Register, false /*is_byte*/);
+  %}
+  ins_pipe(pipe_class_compare);
+%}
+
+instruct indexOf_imm1_U(iRegIdst result, iRegPsrc haystack, iRegIsrc haycnt,
+                        rscratch2RegP needle, immI_1 needlecntImm,
+                        iRegIdst tmp1, iRegIdst tmp2,
+                        flagsRegCR0 cr0, flagsRegCR1 cr1, regCTR ctr) %{
+  match(Set result (StrIndexOf (Binary haystack haycnt) (Binary needle needlecntImm)));
+  effect(USE_KILL needle, TEMP tmp1, TEMP tmp2, KILL cr0, KILL cr1, KILL ctr);
+  // Required for EA: check if it is still a type_array.
+  predicate(((StrIndexOfNode*)n)->encoding() == StrIntrinsicNode::UU &&
+            n->in(3)->in(1)->bottom_type()->is_aryptr()->const_oop() &&
+            n->in(3)->in(1)->bottom_type()->is_aryptr()->const_oop()->is_type_array());
+  ins_cost(180);
+
+  format %{ "String IndexOf SCL1 $haystack[0..$haycnt], $needle[0..$needlecntImm]"
+            " -> $result \t// KILL $haycnt, $needle, $tmp1, $tmp2, $cr0, $cr1" %}
+  ins_encode %{
+    // TODO: PPC port $archOpcode(ppc64Opcode_compound);
+    Node *ndl = in(operand_index($needle));  // The node that defines needle.
+    ciTypeArray* needle_values = ndl->bottom_type()->is_aryptr()->const_oop()->as_type_array();
+    guarantee(needle_values, "sanity");
+    jchar chr;
+#ifdef VM_LITTLE_ENDIAN
+    chr = (((jchar)(unsigned char)needle_values->element_value(1).as_byte()) << 8) |
+           ((jchar)(unsigned char)needle_values->element_value(0).as_byte());
+#else
+    chr = (((jchar)(unsigned char)needle_values->element_value(0).as_byte()) << 8) |
+           ((jchar)(unsigned char)needle_values->element_value(1).as_byte());
+#endif
+    __ string_indexof_char($result$$Register,
+                           $haystack$$Register, $haycnt$$Register,
+                           R0, chr,
+                           $tmp1$$Register, $tmp2$$Register, false /*is_byte*/);
+  %}
+  ins_pipe(pipe_class_compare);
+%}
+
+instruct indexOf_imm1_L(iRegIdst result, iRegPsrc haystack, iRegIsrc haycnt,
+                        rscratch2RegP needle, immI_1 needlecntImm,
+                        iRegIdst tmp1, iRegIdst tmp2,
+                        flagsRegCR0 cr0, flagsRegCR1 cr1, regCTR ctr) %{
+  match(Set result (StrIndexOf (Binary haystack haycnt) (Binary needle needlecntImm)));
+  effect(USE_KILL needle, TEMP tmp1, TEMP tmp2, KILL cr0, KILL cr1, KILL ctr);
+  // Required for EA: check if it is still a type_array.
+  predicate(((StrIndexOfNode*)n)->encoding() == StrIntrinsicNode::LL &&
+            n->in(3)->in(1)->bottom_type()->is_aryptr()->const_oop() &&
+            n->in(3)->in(1)->bottom_type()->is_aryptr()->const_oop()->is_type_array());
+  ins_cost(180);
+
+  format %{ "String IndexOf SCL1 $haystack[0..$haycnt], $needle[0..$needlecntImm]"
+            " -> $result \t// KILL $haycnt, $needle, $tmp1, $tmp2, $cr0, $cr1" %}
+  ins_encode %{
+    // TODO: PPC port $archOpcode(ppc64Opcode_compound);
+    Node *ndl = in(operand_index($needle));  // The node that defines needle.
+    ciTypeArray* needle_values = ndl->bottom_type()->is_aryptr()->const_oop()->as_type_array();
+    guarantee(needle_values, "sanity");
+    jchar chr = (jchar)needle_values->element_value(0).as_byte();
+    __ string_indexof_char($result$$Register,
+                           $haystack$$Register, $haycnt$$Register,
+                           R0, chr,
+                           $tmp1$$Register, $tmp2$$Register, true /*is_byte*/);
+  %}
+  ins_pipe(pipe_class_compare);
+%}
+
+instruct indexOf_imm1_UL(iRegIdst result, iRegPsrc haystack, iRegIsrc haycnt,
+                         rscratch2RegP needle, immI_1 needlecntImm,
+                         iRegIdst tmp1, iRegIdst tmp2,
+                         flagsRegCR0 cr0, flagsRegCR1 cr1, regCTR ctr) %{
+  match(Set result (StrIndexOf (Binary haystack haycnt) (Binary needle needlecntImm)));
+  effect(USE_KILL needle, TEMP tmp1, TEMP tmp2, KILL cr0, KILL cr1, KILL ctr);
+  // Required for EA: check if it is still a type_array.
+  predicate(((StrIndexOfNode*)n)->encoding() == StrIntrinsicNode::UL &&
+            n->in(3)->in(1)->bottom_type()->is_aryptr()->const_oop() &&
+            n->in(3)->in(1)->bottom_type()->is_aryptr()->const_oop()->is_type_array());
+  ins_cost(180);
+
+  format %{ "String IndexOf SCL1 $haystack[0..$haycnt], $needle[0..$needlecntImm]"
+            " -> $result \t// KILL $haycnt, $needle, $tmp1, $tmp2, $cr0, $cr1" %}
+  ins_encode %{
+    // TODO: PPC port $archOpcode(ppc64Opcode_compound);
+    Node *ndl = in(operand_index($needle));  // The node that defines needle.
+    ciTypeArray* needle_values = ndl->bottom_type()->is_aryptr()->const_oop()->as_type_array();
+    guarantee(needle_values, "sanity");
+    jchar chr = (jchar)needle_values->element_value(0).as_byte();
+    __ string_indexof_char($result$$Register,
+                           $haystack$$Register, $haycnt$$Register,
+                           R0, chr,
+                           $tmp1$$Register, $tmp2$$Register, false /*is_byte*/);
+  %}
+  ins_pipe(pipe_class_compare);
+%}
+
+instruct indexOfChar_U(iRegIdst result, iRegPsrc haystack, iRegIsrc haycnt,
+                       iRegIsrc ch, iRegIdst tmp1, iRegIdst tmp2,
+                       flagsRegCR0 cr0, flagsRegCR1 cr1, regCTR ctr) %{
+  match(Set result (StrIndexOfChar (Binary haystack haycnt) ch));
+  effect(TEMP tmp1, TEMP tmp2, KILL cr0, KILL cr1, KILL ctr);
+  predicate(CompactStrings);
+  ins_cost(180);
+
+  format %{ "String IndexOfChar $haystack[0..$haycnt], $ch"
+            " -> $result \t// KILL $haycnt, $tmp1, $tmp2, $cr0, $cr1" %}
+  ins_encode %{
+    // TODO: PPC port $archOpcode(ppc64Opcode_compound);
+    __ string_indexof_char($result$$Register,
+                           $haystack$$Register, $haycnt$$Register,
+                           $ch$$Register, 0 /* this is not used if the character is already in a register */,
+                           $tmp1$$Register, $tmp2$$Register, false /*is_byte*/);
+  %}
+  ins_pipe(pipe_class_compare);
+%}
+
+instruct indexOf_imm_U(iRegIdst result, iRegPsrc haystack, rscratch1RegI haycnt,
+                       iRegPsrc needle, uimmI15 needlecntImm,
+                       iRegIdst tmp1, iRegIdst tmp2, iRegIdst tmp3, iRegIdst tmp4, iRegIdst tmp5,
+                       flagsRegCR0 cr0, flagsRegCR1 cr1, flagsRegCR6 cr6, regCTR ctr) %{
+  match(Set result (StrIndexOf (Binary haystack haycnt) (Binary needle needlecntImm)));
+  effect(USE_KILL haycnt, /* better: TDEF haycnt, */ TEMP_DEF result,
+         TEMP tmp1, TEMP tmp2, TEMP tmp3, TEMP tmp4, TEMP tmp5, KILL cr0, KILL cr1, KILL cr6, KILL ctr);
+  // Required for EA: check if it is still a type_array.
+  predicate(((StrIndexOfNode*)n)->encoding() == StrIntrinsicNode::UU &&
+            n->in(3)->in(1)->bottom_type()->is_aryptr()->const_oop() &&
+            n->in(3)->in(1)->bottom_type()->is_aryptr()->const_oop()->is_type_array());
+  ins_cost(250);
+
+  format %{ "String IndexOf SCL $haystack[0..$haycnt], $needle[0..$needlecntImm]"
+            " -> $result \t// KILL $haycnt, $tmp1, $tmp2, $tmp3, $tmp4, $tmp5, $cr0, $cr1" %}
+  ins_encode %{
+    // TODO: PPC port $archOpcode(ppc64Opcode_compound);
+    Node *ndl = in(operand_index($needle));  // The node that defines needle.
+    ciTypeArray* needle_values = ndl->bottom_type()->is_aryptr()->const_oop()->as_type_array();
+
+    __ string_indexof($result$$Register,
+                      $haystack$$Register, $haycnt$$Register,
+                      $needle$$Register, needle_values, $tmp5$$Register, $needlecntImm$$constant,
+                      $tmp1$$Register, $tmp2$$Register, $tmp3$$Register, $tmp4$$Register, StrIntrinsicNode::UU);
+  %}
+  ins_pipe(pipe_class_compare);
+%}
+
+instruct indexOf_imm_L(iRegIdst result, iRegPsrc haystack, rscratch1RegI haycnt,
+                       iRegPsrc needle, uimmI15 needlecntImm,
+                       iRegIdst tmp1, iRegIdst tmp2, iRegIdst tmp3, iRegIdst tmp4, iRegIdst tmp5,
+                       flagsRegCR0 cr0, flagsRegCR1 cr1, flagsRegCR6 cr6, regCTR ctr) %{
+  match(Set result (StrIndexOf (Binary haystack haycnt) (Binary needle needlecntImm)));
+  effect(USE_KILL haycnt, /* better: TDEF haycnt, */ TEMP_DEF result,
+         TEMP tmp1, TEMP tmp2, TEMP tmp3, TEMP tmp4, TEMP tmp5, KILL cr0, KILL cr1, KILL cr6, KILL ctr);
+  // Required for EA: check if it is still a type_array.
+  predicate(((StrIndexOfNode*)n)->encoding() == StrIntrinsicNode::LL &&
+            n->in(3)->in(1)->bottom_type()->is_aryptr()->const_oop() &&
+            n->in(3)->in(1)->bottom_type()->is_aryptr()->const_oop()->is_type_array());
+  ins_cost(250);
+
+  format %{ "String IndexOf SCL $haystack[0..$haycnt], $needle[0..$needlecntImm]"
+            " -> $result \t// KILL $haycnt, $tmp1, $tmp2, $tmp3, $tmp4, $tmp5, $cr0, $cr1" %}
+  ins_encode %{
+    // TODO: PPC port $archOpcode(ppc64Opcode_compound);
+    Node *ndl = in(operand_index($needle));  // The node that defines needle.
+    ciTypeArray* needle_values = ndl->bottom_type()->is_aryptr()->const_oop()->as_type_array();
+
+    __ string_indexof($result$$Register,
+                      $haystack$$Register, $haycnt$$Register,
+                      $needle$$Register, needle_values, $tmp5$$Register, $needlecntImm$$constant,
+                      $tmp1$$Register, $tmp2$$Register, $tmp3$$Register, $tmp4$$Register, StrIntrinsicNode::LL);
+  %}
+  ins_pipe(pipe_class_compare);
+%}
+
+instruct indexOf_imm_UL(iRegIdst result, iRegPsrc haystack, rscratch1RegI haycnt,
+                        iRegPsrc needle, uimmI15 needlecntImm,
+                        iRegIdst tmp1, iRegIdst tmp2, iRegIdst tmp3, iRegIdst tmp4, iRegIdst tmp5,
+                        flagsRegCR0 cr0, flagsRegCR1 cr1, flagsRegCR6 cr6, regCTR ctr) %{
+  match(Set result (StrIndexOf (Binary haystack haycnt) (Binary needle needlecntImm)));
+  effect(USE_KILL haycnt, /* better: TDEF haycnt, */ TEMP_DEF result,
+         TEMP tmp1, TEMP tmp2, TEMP tmp3, TEMP tmp4, TEMP tmp5, KILL cr0, KILL cr1, KILL cr6, KILL ctr);
+  // Required for EA: check if it is still a type_array.
+  predicate(((StrIndexOfNode*)n)->encoding() == StrIntrinsicNode::UL &&
+            n->in(3)->in(1)->bottom_type()->is_aryptr()->const_oop() &&
+            n->in(3)->in(1)->bottom_type()->is_aryptr()->const_oop()->is_type_array());
+  ins_cost(250);
+
+  format %{ "String IndexOf SCL $haystack[0..$haycnt], $needle[0..$needlecntImm]"
+            " -> $result \t// KILL $haycnt, $tmp1, $tmp2, $tmp3, $tmp4, $tmp5, $cr0, $cr1" %}
+  ins_encode %{
+    // TODO: PPC port $archOpcode(ppc64Opcode_compound);
+    Node *ndl = in(operand_index($needle));  // The node that defines needle.
+    ciTypeArray* needle_values = ndl->bottom_type()->is_aryptr()->const_oop()->as_type_array();
+
+    __ string_indexof($result$$Register,
+                      $haystack$$Register, $haycnt$$Register,
+                      $needle$$Register, needle_values, $tmp5$$Register, $needlecntImm$$constant,
+                      $tmp1$$Register, $tmp2$$Register, $tmp3$$Register, $tmp4$$Register, StrIntrinsicNode::UL);
+  %}
+  ins_pipe(pipe_class_compare);
+%}
+
+instruct indexOf_U(iRegIdst result, iRegPsrc haystack, rscratch1RegI haycnt, iRegPsrc needle, rscratch2RegI needlecnt,
+                   iRegLdst tmp1, iRegLdst tmp2, iRegLdst tmp3, iRegLdst tmp4,
+                   flagsRegCR0 cr0, flagsRegCR1 cr1, flagsRegCR6 cr6, regCTR ctr) %{
+  match(Set result (StrIndexOf (Binary haystack haycnt) (Binary needle needlecnt)));
+  effect(USE_KILL haycnt, USE_KILL needlecnt, /*better: TDEF haycnt, TDEF needlecnt,*/
+         TEMP_DEF result,
+         TEMP tmp1, TEMP tmp2, TEMP tmp3, TEMP tmp4, KILL cr0, KILL cr1, KILL cr6, KILL ctr);
+  predicate(((StrIndexOfNode*)n)->encoding() == StrIntrinsicNode::UU);
+  ins_cost(300);
+
+  format %{ "String IndexOf $haystack[0..$haycnt], $needle[0..$needlecnt]"
+             " -> $result \t// KILL $haycnt, $needlecnt, $tmp1, $tmp2, $tmp3, $tmp4, $cr0, $cr1" %}
+  ins_encode %{
+    // TODO: PPC port $archOpcode(ppc64Opcode_compound);
+    __ string_indexof($result$$Register,
+                      $haystack$$Register, $haycnt$$Register,
+                      $needle$$Register, NULL, $needlecnt$$Register, 0,  // needlecnt not constant.
+                      $tmp1$$Register, $tmp2$$Register, $tmp3$$Register, $tmp4$$Register, StrIntrinsicNode::UU);
+  %}
+  ins_pipe(pipe_class_compare);
+%}
+
+instruct indexOf_L(iRegIdst result, iRegPsrc haystack, rscratch1RegI haycnt, iRegPsrc needle, rscratch2RegI needlecnt,
+                   iRegLdst tmp1, iRegLdst tmp2, iRegLdst tmp3, iRegLdst tmp4,
+                   flagsRegCR0 cr0, flagsRegCR1 cr1, flagsRegCR6 cr6, regCTR ctr) %{
+  match(Set result (StrIndexOf (Binary haystack haycnt) (Binary needle needlecnt)));
+  effect(USE_KILL haycnt, USE_KILL needlecnt, /*better: TDEF haycnt, TDEF needlecnt,*/
+         TEMP_DEF result,
+         TEMP tmp1, TEMP tmp2, TEMP tmp3, TEMP tmp4, KILL cr0, KILL cr1, KILL cr6, KILL ctr);
+  predicate(((StrIndexOfNode*)n)->encoding() == StrIntrinsicNode::LL);
+  ins_cost(300);
+
+  format %{ "String IndexOf $haystack[0..$haycnt], $needle[0..$needlecnt]"
+             " -> $result \t// KILL $haycnt, $needlecnt, $tmp1, $tmp2, $tmp3, $tmp4, $cr0, $cr1" %}
+  ins_encode %{
+    // TODO: PPC port $archOpcode(ppc64Opcode_compound);
+    __ string_indexof($result$$Register,
+                      $haystack$$Register, $haycnt$$Register,
+                      $needle$$Register, NULL, $needlecnt$$Register, 0,  // needlecnt not constant.
+                      $tmp1$$Register, $tmp2$$Register, $tmp3$$Register, $tmp4$$Register, StrIntrinsicNode::LL);
+  %}
+  ins_pipe(pipe_class_compare);
+%}
+
+instruct indexOf_UL(iRegIdst result, iRegPsrc haystack, rscratch1RegI haycnt, iRegPsrc needle, rscratch2RegI needlecnt,
+                    iRegLdst tmp1, iRegLdst tmp2, iRegLdst tmp3, iRegLdst tmp4,
+                    flagsRegCR0 cr0, flagsRegCR1 cr1, flagsRegCR6 cr6, regCTR ctr) %{
+  match(Set result (StrIndexOf (Binary haystack haycnt) (Binary needle needlecnt)));
+  effect(USE_KILL haycnt, USE_KILL needlecnt, /*better: TDEF haycnt, TDEF needlecnt,*/
+         TEMP_DEF result,
+         TEMP tmp1, TEMP tmp2, TEMP tmp3, TEMP tmp4, KILL cr0, KILL cr1, KILL cr6, KILL ctr);
+  predicate(((StrIndexOfNode*)n)->encoding() == StrIntrinsicNode::UL);
+  ins_cost(300);
+
+  format %{ "String IndexOf $haystack[0..$haycnt], $needle[0..$needlecnt]"
+             " -> $result \t// KILL $haycnt, $needlecnt, $tmp1, $tmp2, $tmp3, $tmp4, $cr0, $cr1" %}
+  ins_encode %{
+    // TODO: PPC port $archOpcode(ppc64Opcode_compound);
+    __ string_indexof($result$$Register,
+                      $haystack$$Register, $haycnt$$Register,
+                      $needle$$Register, NULL, $needlecnt$$Register, 0,  // needlecnt not constant.
+                      $tmp1$$Register, $tmp2$$Register, $tmp3$$Register, $tmp4$$Register, StrIntrinsicNode::UL);
+  %}
+  ins_pipe(pipe_class_compare);
+%}
+
+// char[] to byte[] compression
+instruct string_compress(rarg1RegP src, rarg2RegP dst, iRegIsrc len, iRegIdst result, iRegLdst tmp1,
+                         iRegLdst tmp2, iRegLdst tmp3, iRegLdst tmp4, iRegLdst tmp5, regCTR ctr, flagsRegCR0 cr0) %{
+  match(Set result (StrCompressedCopy src (Binary dst len)));
+  effect(TEMP_DEF result, TEMP tmp1, TEMP tmp2, TEMP tmp3, TEMP tmp4, TEMP tmp5,
+         USE_KILL src, USE_KILL dst, KILL ctr, KILL cr0);
+  ins_cost(300);
+  format %{ "String Compress $src,$dst,$len -> $result \t// KILL $tmp1, $tmp2, $tmp3, $tmp4, $tmp5" %}
+  ins_encode %{
+    // TODO: PPC port $archOpcode(ppc64Opcode_compound);
+    Label Lskip, Ldone;
+    __ li($result$$Register, 0);
+    __ string_compress_16($src$$Register, $dst$$Register, $len$$Register, $tmp1$$Register,
+                          $tmp2$$Register, $tmp3$$Register, $tmp4$$Register, $tmp5$$Register, Ldone);
+    __ rldicl_($tmp1$$Register, $len$$Register, 0, 64-3); // Remaining characters.
+    __ beq(CCR0, Lskip);
+    __ string_compress($src$$Register, $dst$$Register, $tmp1$$Register, $tmp2$$Register, Ldone);
+    __ bind(Lskip);
+    __ mr($result$$Register, $len$$Register);
+    __ bind(Ldone);
+  %}
+  ins_pipe(pipe_class_default);
+%}
+
+// byte[] to char[] inflation
+instruct string_inflate(Universe dummy, rarg1RegP src, rarg2RegP dst, iRegIsrc len, iRegLdst tmp1,
+                        iRegLdst tmp2, iRegLdst tmp3, iRegLdst tmp4, iRegLdst tmp5, regCTR ctr, flagsRegCR0 cr0) %{
+  match(Set dummy (StrInflatedCopy src (Binary dst len)));
+  effect(TEMP tmp1, TEMP tmp2, TEMP tmp3, TEMP tmp4, TEMP tmp5, USE_KILL src, USE_KILL dst, KILL ctr, KILL cr0);
+  ins_cost(300);
+  format %{ "String Inflate $src,$dst,$len \t// KILL $tmp1, $tmp2, $tmp3, $tmp4, $tmp5" %}
+  ins_encode %{
+    // TODO: PPC port $archOpcode(ppc64Opcode_compound);
+    Label Ldone;
+    __ string_inflate_16($src$$Register, $dst$$Register, $len$$Register, $tmp1$$Register,
+                         $tmp2$$Register, $tmp3$$Register, $tmp4$$Register, $tmp5$$Register);
+    __ rldicl_($tmp1$$Register, $len$$Register, 0, 64-3); // Remaining characters.
+    __ beq(CCR0, Ldone);
+    __ string_inflate($src$$Register, $dst$$Register, $tmp1$$Register, $tmp2$$Register);
+    __ bind(Ldone);
+  %}
+  ins_pipe(pipe_class_default);
+%}
+
+// StringCoding.java intrinsics
+instruct has_negatives(rarg1RegP ary1, iRegIsrc len, iRegIdst result, iRegLdst tmp1, iRegLdst tmp2,
+                       regCTR ctr, flagsRegCR0 cr0)
+%{
+  match(Set result (HasNegatives ary1 len));
+  effect(TEMP_DEF result, USE_KILL ary1, TEMP tmp1, TEMP tmp2, KILL ctr, KILL cr0);
+  ins_cost(300);
+  format %{ "has negatives byte[] $ary1,$len -> $result \t// KILL $tmp1, $tmp2" %}
+  ins_encode %{
+    // TODO: PPC port $archOpcode(ppc64Opcode_compound);
+    __ has_negatives($ary1$$Register, $len$$Register, $result$$Register,
+                     $tmp1$$Register, $tmp2$$Register);
+  %}
+  ins_pipe(pipe_class_default);
+%}
+
+// encode char[] to byte[] in ISO_8859_1
+instruct encode_iso_array(rarg1RegP src, rarg2RegP dst, iRegIsrc len, iRegIdst result, iRegLdst tmp1,
+                          iRegLdst tmp2, iRegLdst tmp3, iRegLdst tmp4, iRegLdst tmp5, regCTR ctr, flagsRegCR0 cr0) %{
+  match(Set result (EncodeISOArray src (Binary dst len)));
+  effect(TEMP_DEF result, TEMP tmp1, TEMP tmp2, TEMP tmp3, TEMP tmp4, TEMP tmp5,
+         USE_KILL src, USE_KILL dst, KILL ctr, KILL cr0);
+  ins_cost(300);
+  format %{ "Encode array $src,$dst,$len -> $result \t// KILL $tmp1, $tmp2, $tmp3, $tmp4, $tmp5" %}
+  ins_encode %{
+    // TODO: PPC port $archOpcode(ppc64Opcode_compound);
+    Label Lslow, Lfailure1, Lfailure2, Ldone;
+    __ string_compress_16($src$$Register, $dst$$Register, $len$$Register, $tmp1$$Register,
+                          $tmp2$$Register, $tmp3$$Register, $tmp4$$Register, $tmp5$$Register, Lfailure1);
+    __ rldicl_($result$$Register, $len$$Register, 0, 64-3); // Remaining characters.
+    __ beq(CCR0, Ldone);
+    __ bind(Lslow);
+    __ string_compress($src$$Register, $dst$$Register, $result$$Register, $tmp2$$Register, Lfailure2);
+    __ li($result$$Register, 0);
+    __ b(Ldone);
+
+    __ bind(Lfailure1);
+    __ mr($result$$Register, $len$$Register);
+    __ mfctr($tmp1$$Register);
+    __ rldimi_($result$$Register, $tmp1$$Register, 3, 0); // Remaining characters.
+    __ beq(CCR0, Ldone);
+    __ b(Lslow);
+
+    __ bind(Lfailure2);
+    __ mfctr($result$$Register); // Remaining characters.
+
+    __ bind(Ldone);
+    __ subf($result$$Register, $result$$Register, $len$$Register);
+  %}
+  ins_pipe(pipe_class_default);
+%}
+
+
 // String_IndexOf for needle of length 1.
 //
 // Match needle into immediate operands: no loadConP node needed. Saves one
@@ -11060,11 +11638,11 @@
     if (java_lang_String::has_coder_field()) {
       // New compact strings byte array strings
 #ifdef VM_LITTLE_ENDIAN
-      chr = (((jchar)needle_values->element_value(1).as_byte()) << 8) |
-              (jchar)needle_values->element_value(0).as_byte();
+    chr = (((jchar)(unsigned char)needle_values->element_value(1).as_byte()) << 8) |
+           ((jchar)(unsigned char)needle_values->element_value(0).as_byte());
 #else
-      chr = (((jchar)needle_values->element_value(0).as_byte()) << 8) |
-              (jchar)needle_values->element_value(1).as_byte();
+    chr = (((jchar)(unsigned char)needle_values->element_value(0).as_byte()) << 8) |
+           ((jchar)(unsigned char)needle_values->element_value(1).as_byte());
 #endif
     } else {
       // Old char array strings
@@ -11115,11 +11693,11 @@
     if (java_lang_String::has_coder_field()) {
       // New compact strings byte array strings
 #ifdef VM_LITTLE_ENDIAN
-      chr = (((jchar)needle_values->element_value(1).as_byte()) << 8) |
-              (jchar)needle_values->element_value(0).as_byte();
+    chr = (((jchar)(unsigned char)needle_values->element_value(1).as_byte()) << 8) |
+           ((jchar)(unsigned char)needle_values->element_value(0).as_byte());
 #else
-      chr = (((jchar)needle_values->element_value(0).as_byte()) << 8) |
-              (jchar)needle_values->element_value(1).as_byte();
+    chr = (((jchar)(unsigned char)needle_values->element_value(0).as_byte()) << 8) |
+           ((jchar)(unsigned char)needle_values->element_value(1).as_byte());
 #endif
     } else {
       // Old char array strings
@@ -11321,6 +11899,20 @@
   %}
 %}
 
+instruct minI_reg_reg_isel(iRegIdst dst, iRegIsrc src1, iRegIsrc src2, flagsRegCR0 cr0) %{
+  match(Set dst (MinI src1 src2));
+  effect(KILL cr0);
+  predicate(VM_Version::has_isel());
+  ins_cost(DEFAULT_COST*2);
+
+  ins_encode %{
+    // TODO: PPC port $archOpcode(ppc64Opcode_compound);
+    __ cmpw(CCR0, $src1$$Register, $src2$$Register);
+    __ isel($dst$$Register, CCR0, Assembler::less, /*invert*/false, $src1$$Register, $src2$$Register);
+  %}
+  ins_pipe(pipe_class_default);
+%}
+
 instruct maxI_reg_reg_Ex(iRegIdst dst, iRegIsrc src1, iRegIsrc src2) %{
   match(Set dst (MaxI src1 src2));
   ins_cost(DEFAULT_COST*6);
@@ -11341,6 +11933,20 @@
   %}
 %}
 
+instruct maxI_reg_reg_isel(iRegIdst dst, iRegIsrc src1, iRegIsrc src2, flagsRegCR0 cr0) %{
+  match(Set dst (MaxI src1 src2));
+  effect(KILL cr0);
+  predicate(VM_Version::has_isel());
+  ins_cost(DEFAULT_COST*2);
+
+  ins_encode %{
+    // TODO: PPC port $archOpcode(ppc64Opcode_compound);
+    __ cmpw(CCR0, $src1$$Register, $src2$$Register);
+    __ isel($dst$$Register, CCR0, Assembler::greater, /*invert*/false, $src1$$Register, $src2$$Register);
+  %}
+  ins_pipe(pipe_class_default);
+%}
+
 //---------- Population Count Instructions ------------------------------------
 
 // Popcnt for Power7.
--- a/hotspot/src/cpu/ppc/vm/stubGenerator_ppc.cpp	Fri Feb 19 20:45:26 2016 +0300
+++ b/hotspot/src/cpu/ppc/vm/stubGenerator_ppc.cpp	Fri Feb 19 11:09:59 2016 +0100
@@ -2609,9 +2609,7 @@
    *   R5_ARG3    - int   length (of buffer)
    *
    * scratch:
-   *   R6_ARG4    - crc table address
-   *   R7_ARG5    - tmp1
-   *   R8_ARG6    - tmp2
+   *   R2, R6-R12
    *
    * Ouput:
    *   R3_RET     - int   crc result
@@ -2623,22 +2621,25 @@
     address start = __ function_entry();  // Remember stub start address (is rtn value).
 
     // arguments to kernel_crc32:
-    Register       crc     = R3_ARG1;  // Current checksum, preset by caller or result from previous call.
-    Register       data    = R4_ARG2;  // source byte array
-    Register       dataLen = R5_ARG3;  // #bytes to process
-    Register       table   = R6_ARG4;  // crc table address
-
-    Register       t0      = R9;       // work reg for kernel* emitters
-    Register       t1      = R10;      // work reg for kernel* emitters
-    Register       t2      = R11;      // work reg for kernel* emitters
-    Register       t3      = R12;      // work reg for kernel* emitters
+    const Register crc     = R3_ARG1;  // Current checksum, preset by caller or result from previous call.
+    const Register data    = R4_ARG2;  // source byte array
+    const Register dataLen = R5_ARG3;  // #bytes to process
+    const Register table   = R6_ARG4;  // crc table address
+
+    const Register t0      = R2;
+    const Register t1      = R7;
+    const Register t2      = R8;
+    const Register t3      = R9;
+    const Register tc0     = R10;
+    const Register tc1     = R11;
+    const Register tc2     = R12;
 
     BLOCK_COMMENT("Stub body {");
     assert_different_registers(crc, data, dataLen, table);
 
     StubRoutines::ppc64::generate_load_crc_table_addr(_masm, table);
 
-    __ kernel_crc32_1byte(crc, data, dataLen, table, t0, t1, t2, t3);
+    __ kernel_crc32_1word(crc, data, dataLen, table, t0, t1, t2, t3, tc0, tc1, tc2, table);
 
     BLOCK_COMMENT("return");
     __ mr_if_needed(R3_RET, crc);      // Updated crc is function result. No copying required (R3_ARG1 == R3_RET).
--- a/hotspot/src/cpu/ppc/vm/vm_version_ppc.cpp	Fri Feb 19 20:45:26 2016 +0300
+++ b/hotspot/src/cpu/ppc/vm/vm_version_ppc.cpp	Fri Feb 19 11:09:59 2016 +0100
@@ -1,6 +1,6 @@
 /*
- * Copyright (c) 1997, 2015, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2012, 2015 SAP SE. All rights reserved.
+ * Copyright (c) 1997, 2016, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2012, 2016 SAP SE. 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
@@ -53,7 +53,7 @@
 
   // If PowerArchitecturePPC64 hasn't been specified explicitly determine from features.
   if (FLAG_IS_DEFAULT(PowerArchitecturePPC64)) {
-    if (VM_Version::has_tcheck() && VM_Version::has_lqarx()) {
+    if (VM_Version::has_lqarx()) {
       FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 8);
     } else if (VM_Version::has_popcntw()) {
       FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 7);
@@ -68,8 +68,7 @@
 
   bool PowerArchitecturePPC64_ok = false;
   switch (PowerArchitecturePPC64) {
-    case 8: if (!VM_Version::has_tcheck() ) break;
-            if (!VM_Version::has_lqarx()  ) break;
+    case 8: if (!VM_Version::has_lqarx()  ) break;
     case 7: if (!VM_Version::has_popcntw()) break;
     case 6: if (!VM_Version::has_cmpb()   ) break;
     case 5: if (!VM_Version::has_popcntb()) break;
@@ -80,7 +79,7 @@
             UINTX_FORMAT " on this machine", PowerArchitecturePPC64);
 
   // Power 8: Configure Data Stream Control Register.
-  if (PowerArchitecturePPC64 >= 8) {
+  if (has_mfdscr()) {
     config_dscr();
   }
 
@@ -112,7 +111,7 @@
   // Create and print feature-string.
   char buf[(num_features+1) * 16]; // Max 16 chars per feature.
   jio_snprintf(buf, sizeof(buf),
-               "ppc64%s%s%s%s%s%s%s%s%s%s%s%s",
+               "ppc64%s%s%s%s%s%s%s%s%s%s%s%s%s",
                (has_fsqrt()   ? " fsqrt"   : ""),
                (has_isel()    ? " isel"    : ""),
                (has_lxarxeh() ? " lxarxeh" : ""),
@@ -125,7 +124,8 @@
                (has_lqarx()   ? " lqarx"   : ""),
                (has_vcipher() ? " vcipher" : ""),
                (has_vpmsumb() ? " vpmsumb" : ""),
-               (has_tcheck()  ? " tcheck"  : "")
+               (has_tcheck()  ? " tcheck"  : ""),
+               (has_mfdscr()  ? " mfdscr"  : "")
                // Make sure number of %s matches num_features!
               );
   _features_string = os::strdup(buf);
@@ -610,6 +610,7 @@
   a->vcipher(VR0, VR1, VR2);                   // code[10] -> vcipher
   a->vpmsumb(VR0, VR1, VR2);                   // code[11] -> vpmsumb
   a->tcheck(0);                                // code[12] -> tcheck
+  a->mfdscr(R0);                               // code[13] -> mfdscr
   a->blr();
 
   // Emit function to set one cache line to zero. Emit function descriptor and get pointer to it.
@@ -657,6 +658,7 @@
   if (code[feature_cntr++]) features |= vcipher_m;
   if (code[feature_cntr++]) features |= vpmsumb_m;
   if (code[feature_cntr++]) features |= tcheck_m;
+  if (code[feature_cntr++]) features |= mfdscr_m;
 
   // Print the detection code.
   if (PrintAssembly) {
@@ -670,8 +672,6 @@
 
 // Power 8: Configure Data Stream Control Register.
 void VM_Version::config_dscr() {
-  assert(has_tcheck(), "Only execute on Power 8 or later!");
-
   // 7 InstWords for each call (function descriptor + blr instruction).
   const int code_size = (2+2*7)*BytesPerInstWord;
 
--- a/hotspot/src/cpu/ppc/vm/vm_version_ppc.hpp	Fri Feb 19 20:45:26 2016 +0300
+++ b/hotspot/src/cpu/ppc/vm/vm_version_ppc.hpp	Fri Feb 19 11:09:59 2016 +0100
@@ -1,6 +1,6 @@
 /*
- * Copyright (c) 1997, 2015, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2012, 2015 SAP SE. All rights reserved.
+ * Copyright (c) 1997, 2016, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2012, 2016 SAP SE. 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
@@ -45,6 +45,7 @@
     vcipher,
     vpmsumb,
     tcheck,
+    mfdscr,
     num_features // last entry to count features
   };
   enum Feature_Flag_Set {
@@ -62,6 +63,7 @@
     vcipher_m             = (1 << vcipher),
     vpmsumb_m             = (1 << vpmsumb),
     tcheck_m              = (1 << tcheck ),
+    mfdscr_m              = (1 << mfdscr ),
     all_features_m        = (unsigned long)-1
   };
 
@@ -94,6 +96,7 @@
   static bool has_vcipher() { return (_features & vcipher_m) != 0; }
   static bool has_vpmsumb() { return (_features & vpmsumb_m) != 0; }
   static bool has_tcheck()  { return (_features & tcheck_m) != 0; }
+  static bool has_mfdscr()  { return (_features & mfdscr_m) != 0; }
 
   // Assembler testing
   static void allow_all();
--- a/hotspot/test/compiler/intrinsics/string/TestStringIntrinsics2.java	Fri Feb 19 20:45:26 2016 +0300
+++ b/hotspot/test/compiler/intrinsics/string/TestStringIntrinsics2.java	Fri Feb 19 11:09:59 2016 +0100
@@ -494,6 +494,29 @@
         return s.indexOf("1");
     }
 
+    static String text1UTF16 = "A" + "\u05d0" + "\u05d1" + "B";
+
+    @Test(role = Role.TEST_ENTRY)
+    public static void test_indexOf_immUTF16() {
+        assertEquals(      3, indexOf_imm1Latin1_needle(text1UTF16), "test_indexOf_immUTF16");
+        assertEquals(      1, indexOf_imm1UTF16_needle(text1UTF16), "test_indexOf_immUTF16");
+        assertEquals(      1, indexOf_immUTF16_needle(text1UTF16), "test_indexOf_immUTF16");
+    }
+
+    @Test(role = Role.TEST_HELPER, compileAt = 4, warmup = 1, warmupArgs = { "A" + "\u05d0" + "\u05d1" + "B" })
+    static int indexOf_imm1Latin1_needle(String s) {
+        return s.indexOf("B");
+    }
+
+    @Test(role = Role.TEST_HELPER, compileAt = 4, warmup = 1, warmupArgs = { "A" + "\u05d0" + "\u05d1" + "B" })
+    static int indexOf_imm1UTF16_needle(String s) {
+        return s.indexOf("\u05d0");
+    }
+
+    @Test(role = Role.TEST_HELPER, compileAt = 4, warmup = 1, warmupArgs = { "A" + "\u05d0" + "\u05d1" + "B" })
+    static int indexOf_immUTF16_needle(String s) {
+        return s.indexOf("\u05d0" + "\u05d1");
+    }
 
     @Test(role = Role.TEST_HELPER, compileAt = 4, warmup = 1, warmupArgs = { "abc", "abcd" })
     public static int asmStringCompareTo(String a, String b) {