8141132: JEP 254: Compact Strings

Adopt a more space-efficient internal representation for strings.

Co-authored-by: Brent Christian <brent.christian@oracle.com>
Co-authored-by: Vivek Deshpande <vivek.r.deshpande@intel.com>
Co-authored-by: Charlie Hunt <charlie.hunt@oracle.com>
Co-authored-by: Vladimir Kozlov <vladimir.kozlov@oracle.com>
Co-authored-by: Roger Riggs <roger.riggs@oracle.com>
Co-authored-by: Xueming Shen <xueming.shen@oracle.com>
Co-authored-by: Aleksey Shipilev <aleksey.shipilev@oracle.com>
Co-authored-by: Sandhya Viswanathan <sandhya.viswanathan@intel.com>
Reviewed-by: alanb, bdelsart, coleenp, iklam, jiangli, jrose, kevinw, naoto, pliden, roland, smarks, twisti

Author: 9 people

hotspot/agent/src/share/classes/sun/jvm/hotspot/oops/OopUtilities.java

@@ -40,8 +40,7 @@ public class OopUtilities implements /* imports */ JVMTIThreadState {
   // FIXME: access should be synchronized and cleared when VM is
   // resumed
   // String fields
-  private static IntField offsetField;
-  private static IntField countField;
+  private static ByteField coderField;
   private static OopField valueField;
   // ThreadGroup fields
   private static OopField threadGroupParentField;
@@ -96,20 +95,30 @@ public static String charArrayToString(TypeArray charArray) {
     if (charArray == null) {
       return null;
     }
-    return charArrayToString(charArray, 0, (int) charArray.getLength());
+    int length = (int)charArray.getLength();
+    StringBuffer buf = new StringBuffer(length);
+    for (int i = 0; i < length; i++) {
+      buf.append(charArray.getCharAt(i));
+    }
+    return buf.toString();
   }
 
-  public static String charArrayToString(TypeArray charArray, int offset, int length) {
-    if (charArray == null) {
+  public static String byteArrayToString(TypeArray byteArray, byte coder) {
+    if (byteArray == null) {
       return null;
     }
-    final int limit = offset + length;
-    if (Assert.ASSERTS_ENABLED) {
-      Assert.that(offset >= 0 && limit <= charArray.getLength(), "out of bounds");
-    }
+    int length = (int)byteArray.getLength() >> coder;
     StringBuffer buf = new StringBuffer(length);
-    for (int i = offset; i < limit; i++) {
-      buf.append(charArray.getCharAt(i));
+    if (coder == 0) {
+      // Latin1 encoded
+      for (int i = 0; i < length; i++) {
+        buf.append((char)(byteArray.getByteAt(i) & 0xff));
+      }
+    } else {
+      // UTF16 encoded
+      for (int i = 0; i < length; i++) {
+        buf.append(byteArray.getCharAt(i));
+      }
     }
     return buf.toString();
   }
@@ -141,21 +150,14 @@ public static String escapeString(String s) {
   }
 
   public static String stringOopToString(Oop stringOop) {
-    if (offsetField == null) {
-      InstanceKlass k = (InstanceKlass) stringOop.getKlass();
-      offsetField = (IntField) k.findField("offset", "I");   // optional
-      countField  = (IntField) k.findField("count",  "I");   // optional
-      valueField  = (OopField) k.findField("value",  "[C");
-      if (Assert.ASSERTS_ENABLED) {
-         Assert.that(valueField != null, "Field \'value\' of java.lang.String not found");
-      }
-    }
-    if (offsetField != null && countField != null) {
-      return charArrayToString((TypeArray) valueField.getValue(stringOop),
-                               offsetField.getValue(stringOop),
-                               countField.getValue(stringOop));
+    InstanceKlass k = (InstanceKlass) stringOop.getKlass();
+    coderField  = (ByteField) k.findField("coder", "B");
+    valueField  = (OopField) k.findField("value",  "[B");
+    if (Assert.ASSERTS_ENABLED) {
+       Assert.that(coderField != null, "Field \'coder\' of java.lang.String not found");
+       Assert.that(valueField != null, "Field \'value\' of java.lang.String not found");
     }
-    return  charArrayToString((TypeArray) valueField.getValue(stringOop));
+    return byteArrayToString((TypeArray) valueField.getValue(stringOop), coderField.getValue(stringOop));
   }
 
   public static String stringOopToEscapedString(Oop stringOop) {

hotspot/agent/src/share/classes/sun/jvm/hotspot/tools/HeapSummary.java

@@ -268,8 +268,8 @@ class StringStat implements StringTable.StringVisitor {
             VM vm = VM.getVM();
             SystemDictionary sysDict = vm.getSystemDictionary();
             InstanceKlass strKlass = sysDict.getStringKlass();
-            // String has a field named 'value' of type 'char[]'.
-            stringValueField = (OopField) strKlass.findField("value", "[C");
+            // String has a field named 'value' of type 'byte[]'.
+            stringValueField = (OopField) strKlass.findField("value", "[B");
          }
 
          private long stringSize(Instance instance) {

hotspot/agent/src/share/classes/sun/jvm/hotspot/utilities/Hashtable.java

@@ -61,9 +61,8 @@ protected static long hashSymbol(byte[] buf) {
     long h = 0;
     int s = 0;
     int len = buf.length;
-    // Emulate the unsigned int in java_lang_String::hash_code
     while (len-- > 0) {
-      h = 31*h + (0xFFFFFFFFL & buf[s]);
+      h = 31*h + (0xFFL & buf[s]);
       s++;
     }
     return h & 0xFFFFFFFFL;

hotspot/src/cpu/aarch64/vm/aarch64.ad

@@ -14150,6 +14150,7 @@ instruct partialSubtypeCheckVsZero(iRegP_R4 sub, iRegP_R0 super, iRegP_R2 temp,
 instruct string_compare(iRegP_R1 str1, iRegI_R2 cnt1, iRegP_R3 str2, iRegI_R4 cnt2,
                         iRegI_R0 result, iRegP_R10 tmp1, rFlagsReg cr)
 %{
+  predicate(!CompactStrings);
   match(Set result (StrComp (Binary str1 cnt1) (Binary str2 cnt2)));
   effect(KILL tmp1, USE_KILL str1, USE_KILL str2, USE_KILL cnt1, USE_KILL cnt2, KILL cr);
 
@@ -14165,6 +14166,7 @@ instruct string_compare(iRegP_R1 str1, iRegI_R2 cnt1, iRegP_R3 str2, iRegI_R4 cn
 instruct string_indexof(iRegP_R1 str1, iRegI_R4 cnt1, iRegP_R3 str2, iRegI_R2 cnt2,
        iRegI_R0 result, iRegI tmp1, iRegI tmp2, iRegI tmp3, iRegI tmp4, rFlagsReg cr)
 %{
+  predicate(!CompactStrings);
   match(Set result (StrIndexOf (Binary str1 cnt1) (Binary str2 cnt2)));
   effect(USE_KILL str1, USE_KILL str2, USE_KILL cnt1, USE_KILL cnt2,
          TEMP tmp1, TEMP tmp2, TEMP tmp3, TEMP tmp4, KILL cr);
@@ -14184,6 +14186,7 @@ instruct string_indexof_con(iRegP_R1 str1, iRegI_R4 cnt1, iRegP_R3 str2,
                  immI_le_4 int_cnt2, iRegI_R0 result, iRegI tmp1, iRegI tmp2,
                  iRegI tmp3, iRegI tmp4, rFlagsReg cr)
 %{
+  predicate(!CompactStrings);
   match(Set result (StrIndexOf (Binary str1 cnt1) (Binary str2 int_cnt2)));
   effect(USE_KILL str1, USE_KILL str2, USE_KILL cnt1,
          TEMP tmp1, TEMP tmp2, TEMP tmp3, TEMP tmp4, KILL cr);
@@ -14203,6 +14206,7 @@ instruct string_indexof_con(iRegP_R1 str1, iRegI_R4 cnt1, iRegP_R3 str2,
 instruct string_equals(iRegP_R1 str1, iRegP_R3 str2, iRegI_R4 cnt,
                         iRegI_R0 result, iRegP_R10 tmp, rFlagsReg cr)
 %{
+  predicate(!CompactStrings);
   match(Set result (StrEquals (Binary str1 str2) cnt));
   effect(KILL tmp, USE_KILL str1, USE_KILL str2, USE_KILL cnt, KILL cr);
 
@@ -14218,6 +14222,7 @@ instruct string_equals(iRegP_R1 str1, iRegP_R3 str2, iRegI_R4 cnt,
 instruct array_equals(iRegP_R1 ary1, iRegP_R2 ary2, iRegI_R0 result,
                       iRegP_R10 tmp, rFlagsReg cr)
 %{
+  predicate(((AryEqNode*)n)->encoding() == StrIntrinsicNode::UU);
   match(Set result (AryEq ary1 ary2));
   effect(KILL tmp, USE_KILL ary1, USE_KILL ary2, KILL cr);
 

hotspot/src/cpu/aarch64/vm/c1_LIRAssembler_aarch64.cpp

@@ -483,15 +483,6 @@ int LIR_Assembler::emit_deopt_handler() {
   return offset;
 }
 
-
-// This is the fast version of java.lang.String.compare; it has not
-// OSR-entry and therefore, we generate a slow version for OSR's
-void LIR_Assembler::emit_string_compare(LIR_Opr arg0, LIR_Opr arg1, LIR_Opr dst, CodeEmitInfo* info)  {
-  __ mov(r2, (address)__FUNCTION__);
-  __ call_Unimplemented();
-}
-
-
 void LIR_Assembler::add_debug_info_for_branch(address adr, CodeEmitInfo* info) {
   _masm->code_section()->relocate(adr, relocInfo::poll_type);
   int pc_offset = code_offset();

hotspot/src/cpu/aarch64/vm/globals_aarch64.hpp

@@ -79,6 +79,9 @@ define_pd_global(uintx, CMSYoungGenPerWorker, 64*M);  // default max size of CMS
 
 define_pd_global(uintx, TypeProfileLevel, 111);
 
+// No performance work done here yet.
+define_pd_global(bool, CompactStrings, false);
+
 // avoid biased locking while we are bootstrapping the aarch64 build
 define_pd_global(bool, UseBiasedLocking, false);
 

hotspot/src/cpu/ppc/vm/globals_ppc.hpp

@@ -72,6 +72,9 @@ define_pd_global(size_t, CMSYoungGenPerWorker, 16*M);  // Default max size of CM
 
 define_pd_global(uintx, TypeProfileLevel, 111);
 
+// No performance work done here yet.
+define_pd_global(bool, CompactStrings, false);
+
 // Platform dependent flag handling: flags only defined on this platform.
 #define ARCH_FLAGS(develop, product, diagnostic, experimental, notproduct, range, constraint)  \
                                                                             \

hotspot/src/cpu/ppc/vm/ppc.ad

@@ -2054,11 +2054,11 @@ const bool Matcher::match_rule_supported(int opcode) {
     return (UsePopCountInstruction && VM_Version::has_popcntw());
 
   case Op_StrComp:
-    return SpecialStringCompareTo;
+    return SpecialStringCompareTo && !CompactStrings;
   case Op_StrEquals:
-    return SpecialStringEquals;
+    return SpecialStringEquals && !CompactStrings;
   case Op_StrIndexOf:
-    return SpecialStringIndexOf;
+    return SpecialStringIndexOf && !CompactStrings;
   }
 
   return true;  // Per default match rules are supported.
@@ -11077,7 +11077,7 @@ instruct string_indexOf_imm1_char(iRegIdst result, iRegPsrc haystack, iRegIsrc h
                                   immP needleImm, immL offsetImm, immI_1 needlecntImm,
                                   iRegIdst tmp1, iRegIdst tmp2,
                                   flagsRegCR0 cr0, flagsRegCR1 cr1) %{
-  predicate(SpecialStringIndexOf);  // type check implicit by parameter type, See Matcher::match_rule_supported
+  predicate(SpecialStringIndexOf && !CompactStrings);  // type check implicit by parameter type, See Matcher::match_rule_supported
   match(Set result (StrIndexOf (Binary haystack haycnt) (Binary (AddP needleImm offsetImm) needlecntImm)));
 
   effect(TEMP_DEF result, TEMP tmp1, TEMP tmp2, KILL cr0, KILL cr1);
@@ -11120,7 +11120,7 @@ instruct string_indexOf_imm1(iRegIdst result, iRegPsrc haystack, iRegIsrc haycnt
   effect(USE_KILL needle, /* TDEF needle, */ TEMP_DEF result,
          TEMP tmp1, TEMP tmp2);
   // Required for EA: check if it is still a type_array.
-  predicate(SpecialStringIndexOf && n->in(3)->in(1)->bottom_type()->is_aryptr()->const_oop() &&
+  predicate(SpecialStringIndexOf && !CompactStrings && 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);
 
@@ -11167,7 +11167,7 @@ instruct string_indexOf_imm(iRegIdst result, iRegPsrc haystack, rscratch1RegI ha
   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);
   // Required for EA: check if it is still a type_array.
-  predicate(SpecialStringIndexOf && n->in(3)->in(1)->bottom_type()->is_aryptr()->const_oop() &&
+  predicate(SpecialStringIndexOf && !CompactStrings && 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);
 
@@ -11200,7 +11200,7 @@ instruct string_indexOf(iRegIdst result, iRegPsrc haystack, rscratch1RegI haycnt
   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);
-  predicate(SpecialStringIndexOf);  // See Matcher::match_rule_supported.
+  predicate(SpecialStringIndexOf && !CompactStrings);  // See Matcher::match_rule_supported.
   ins_cost(300);
 
   ins_alignment(8); // 'compute_padding()' gets called, up to this number-1 nops will get inserted.
@@ -11224,7 +11224,7 @@ instruct string_equals_imm(iRegPsrc str1, iRegPsrc str2, uimmI15 cntImm, iRegIds
   match(Set result (StrEquals (Binary str1 str2) cntImm));
   effect(TEMP_DEF result, TEMP tmp1, TEMP tmp2,
          KILL cr0, KILL cr6, KILL ctr);
-  predicate(SpecialStringEquals);  // See Matcher::match_rule_supported.
+  predicate(SpecialStringEquals && !CompactStrings);  // See Matcher::match_rule_supported.
   ins_cost(250);
 
   ins_alignment(8); // 'compute_padding()' gets called, up to this number-1 nops will get inserted.
@@ -11247,7 +11247,7 @@ instruct string_equals(iRegPsrc str1, iRegPsrc str2, iRegIsrc cnt, iRegIdst resu
   match(Set result (StrEquals (Binary str1 str2) cnt));
   effect(TEMP_DEF result, TEMP tmp1, TEMP tmp2, TEMP tmp3, TEMP tmp4, TEMP tmp5,
          KILL cr0, KILL cr1, KILL cr6, KILL ctr);
-  predicate(SpecialStringEquals);  // See Matcher::match_rule_supported.
+  predicate(SpecialStringEquals && !CompactStrings);  // See Matcher::match_rule_supported.
   ins_cost(300);
 
   ins_alignment(8); // 'compute_padding()' gets called, up to this number-1 nops will get inserted.
@@ -11267,6 +11267,7 @@ instruct string_equals(iRegPsrc str1, iRegPsrc str2, iRegIsrc cnt, iRegIdst resu
 // Use dst register classes if register gets killed, as it is the case for TEMP operands!
 instruct string_compare(rarg1RegP str1, rarg2RegP str2, rarg3RegI cnt1, rarg4RegI cnt2, iRegIdst result,
                         iRegPdst tmp, flagsRegCR0 cr0, regCTR ctr) %{
+  predicate(!CompactStrings);
   match(Set result (StrComp (Binary str1 cnt1) (Binary str2 cnt2)));
   effect(USE_KILL cnt1, USE_KILL cnt2, USE_KILL str1, USE_KILL str2, TEMP_DEF result, TEMP tmp, KILL cr0, KILL ctr);
   ins_cost(300);

hotspot/src/cpu/sparc/vm/assembler_sparc.hpp

@@ -124,6 +124,8 @@ class Assembler : public AbstractAssembler  {
     impdep1_op3  = 0x36,
     aes3_op3     = 0x36,
     sha_op3      = 0x36,
+    bmask_op3    = 0x36,
+    bshuffle_op3   = 0x36,
     alignaddr_op3  = 0x36,
     faligndata_op3 = 0x36,
     flog3_op3    = 0x36,
@@ -194,6 +196,7 @@ class Assembler : public AbstractAssembler  {
     fnegd_opf          = 0x06,
 
     alignaddr_opf      = 0x18,
+    bmask_opf          = 0x19,
 
     fadds_opf          = 0x41,
     faddd_opf          = 0x42,
@@ -204,6 +207,7 @@ class Assembler : public AbstractAssembler  {
 
     fmuls_opf          = 0x49,
     fmuld_opf          = 0x4a,
+    bshuffle_opf       = 0x4c,
     fdivs_opf          = 0x4d,
     fdivd_opf          = 0x4e,
 
@@ -1226,6 +1230,9 @@ class Assembler : public AbstractAssembler  {
 
   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)); }
 
+  void bmask( Register s1, Register s2, Register d ) { vis2_only(); emit_int32( op(arith_op) | rd(d) | op3(bmask_op3) | rs1(s1) | opf(bmask_opf) | rs2(s2)); }
+  void bshuffle( FloatRegister s1, FloatRegister s2, FloatRegister d ) { vis2_only(); emit_int32( op(arith_op) | fd(d, FloatRegisterImpl::D) | op3(bshuffle_op3) | fs1(s1, FloatRegisterImpl::D) | opf(bshuffle_opf) | fs2(s2, FloatRegisterImpl::D)); }
+
   // 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)); }