* Irregexp: Move from a byte-oriented bytecode format to a 32-bit oriented

bytecode format.  This provides a nice speedup on Intel and probably an
even better one on ARM.  Also removes the 256-register limitation on the
interpreter.
Review URL: http://codereview.chromium.org/18363

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@1111 ce2b1a6d-e550-0410-aec6-3dcde31c8c00

src/bytecodes-irregexp.h

@@ -31,50 +31,59 @@
 
 namespace v8 { namespace internal {
 
+
+static const int BYTECODE_MASK = 0xff;
+static const unsigned int MAX_FIRST_ARG = 0xffffffu;
+static const int BYTECODE_SHIFT = 8;
+
 #define BYTECODE_ITERATOR(V)                                                   \
-V(BREAK,              0, 1) /* break                                        */ \
-V(PUSH_CP,            1, 5) /* push_cp offset32                             */ \
-V(PUSH_BT,            2, 5) /* push_bt addr32                               */ \
-V(PUSH_REGISTER,      3, 2) /* push_register register_index                 */ \
-V(SET_REGISTER_TO_CP, 4, 6) /* set_register_to_cp register_index offset32   */ \
-V(SET_CP_TO_REGISTER, 5, 2) /* set_cp_to_registger register_index           */ \
-V(SET_REGISTER_TO_SP, 6, 2) /* set_register_to_sp register_index            */ \
-V(SET_SP_TO_REGISTER, 7, 2) /* set_sp_to_registger register_index           */ \
-V(SET_REGISTER,       8, 6) /* set_register register_index value32          */ \
-V(ADVANCE_REGISTER,   9, 6) /* advance_register register_index value32      */ \
-V(POP_CP,            10, 1) /* pop_cp                                       */ \
-V(POP_BT,            11, 1) /* pop_bt                                       */ \
-V(POP_REGISTER,      12, 2) /* pop_register register_index                  */ \
-V(FAIL,              13, 1) /* fail                                         */ \
-V(SUCCEED,           14, 1) /* succeed                                      */ \
-V(ADVANCE_CP,        15, 5) /* advance_cp offset32                          */ \
-V(GOTO,              16, 5) /* goto addr32                                  */ \
-V(LOAD_CURRENT_CHAR, 17, 9) /* load offset32 addr32                         */ \
-V(LOAD_CURRENT_CHAR_UNCHECKED, 18, 5) /* load offset32                      */ \
-V(LOAD_2_CURRENT_CHARS, 19, 9) /* load offset32 addr32                      */ \
-V(LOAD_2_CURRENT_CHARS_UNCHECKED, 20, 5) /* load offset32                   */ \
-V(LOAD_4_CURRENT_CHARS, 21, 9) /* load offset32 addr32                      */ \
-V(LOAD_4_CURRENT_CHARS_UNCHECKED, 22, 5) /* load offset32                   */ \
-V(CHECK_CHAR,        23, 9) /* check_char uint32 addr32                     */ \
-V(CHECK_NOT_CHAR,    24, 9) /* check_not_char uint32 addr32                 */ \
-V(AND_CHECK_CHAR,    25, 13) /* and_check_char uint32 uint32 addr32         */ \
-V(AND_CHECK_NOT_CHAR, 26, 13) /* and_check_not_char uint32 uint32 addr32    */ \
-V(MINUS_AND_CHECK_NOT_CHAR, 27, 11) /* minus_and_check_not_char uc16 uc16...*/ \
-V(CHECK_LT,          28, 7) /* check_lt uc16 addr32                         */ \
-V(CHECK_GT,          29, 7) /* check_gr uc16 addr32                         */ \
-V(CHECK_NOT_BACK_REF, 30, 6) /* check_not_back_ref capture_idx addr32       */ \
-V(CHECK_NOT_BACK_REF_NO_CASE, 31, 6) /* check_not_back_ref_no_case captu... */ \
-V(CHECK_NOT_REGS_EQUAL, 32, 7) /* check_not_regs_equal reg1 reg2 addr32     */ \
-V(LOOKUP_MAP1,       33, 11) /* l_map1 start16 bit_map_addr32 addr32        */ \
-V(LOOKUP_MAP2,       34, 99) /* l_map2 start16 half_nibble_map_addr32*      */ \
-V(LOOKUP_MAP8,       35, 99) /* l_map8 start16 byte_map addr32*             */ \
-V(LOOKUP_HI_MAP8,    36, 99) /* l_himap8 start8 byte_map_addr32 addr32*     */ \
-V(CHECK_REGISTER_LT, 37, 8) /* check_reg_lt register_index value16 addr32   */ \
-V(CHECK_REGISTER_GE, 38, 8) /* check_reg_ge register_index value16 addr32   */ \
-V(CHECK_REGISTER_EQ_POS, 39, 6) /* check_register_eq_pos index addr32       */ \
-V(CHECK_AT_START,    40, 5) /* check_at_start addr32                        */ \
-V(CHECK_NOT_AT_START, 41, 5) /* check_not_at_start addr32                   */ \
-V(CHECK_GREEDY,      42, 5) /* check_greedy addr32                          */
+V(BREAK,              0, 4)   /* bc8                                        */ \
+V(PUSH_CP,            1, 4)   /* bc8 pad24                                  */ \
+V(PUSH_BT,            2, 8)   /* bc8 pad24 offset32                         */ \
+V(PUSH_REGISTER,      3, 4)   /* bc8 reg_idx24                              */ \
+V(SET_REGISTER_TO_CP, 4, 8)   /* bc8 reg_idx24 offset32                     */ \
+V(SET_CP_TO_REGISTER, 5, 4)   /* bc8 reg_idx24                              */ \
+V(SET_REGISTER_TO_SP, 6, 4)   /* bc8 reg_idx24                              */ \
+V(SET_SP_TO_REGISTER, 7, 4)   /* bc8 reg_idx24                              */ \
+V(SET_REGISTER,       8, 8)   /* bc8 reg_idx24 value32                      */ \
+V(ADVANCE_REGISTER,   9, 8)   /* bc8 reg_idx24 value32                      */ \
+V(POP_CP,            10, 4)   /* bc8 pad24                                  */ \
+V(POP_BT,            11, 4)   /* bc8 pad24                                  */ \
+V(POP_REGISTER,      12, 4)   /* bc8 reg_idx24                              */ \
+V(FAIL,              13, 4)   /* bc8 pad24                                  */ \
+V(SUCCEED,           14, 4)   /* bc8 pad24                                  */ \
+V(ADVANCE_CP,        15, 4)   /* bc8 offset24                               */ \
+V(GOTO,              16, 8)   /* bc8 pad24 addr32                           */ \
+V(LOAD_CURRENT_CHAR, 17, 8)   /* bc8 offset24 addr32                        */ \
+V(LOAD_CURRENT_CHAR_UNCHECKED, 18, 4) /* bc8 offset24                       */ \
+V(LOAD_2_CURRENT_CHARS, 19, 8) /* bc8 offset24 addr32                       */ \
+V(LOAD_2_CURRENT_CHARS_UNCHECKED, 20, 4) /* bc8 offset24                    */ \
+V(LOAD_4_CURRENT_CHARS, 21, 8) /* bc8 offset24 addr32                       */ \
+V(LOAD_4_CURRENT_CHARS_UNCHECKED, 22, 4) /* bc8 offset24                    */ \
+V(CHECK_4_CHARS,     23, 12)  /* bc8 pad24 uint32 addr32                    */ \
+V(CHECK_CHAR,        24, 8)   /* bc8 pad8 uint16 addr32                     */ \
+V(CHECK_NOT_4_CHARS, 25, 12)  /* bc8 pad24 uint32 addr32                    */ \
+V(CHECK_NOT_CHAR,    26, 8)   /* bc8 pad8 uint16 addr32                     */ \
+V(AND_CHECK_4_CHARS, 27, 16)  /* bc8 pad24 uint32 uint32 addr32             */ \
+V(AND_CHECK_CHAR,    28, 12)  /* bc8 pad8 uint16 uint32 addr32              */ \
+V(AND_CHECK_NOT_4_CHARS, 29, 16) /* bc8 pad24 uint32 uint32 addr32          */ \
+V(AND_CHECK_NOT_CHAR, 30, 12) /* bc8 pad8 uint16 uint32 addr32              */ \
+V(MINUS_AND_CHECK_NOT_CHAR, 31, 12) /* bc8 pad8 uc16 uc16 addr32            */ \
+V(CHECK_LT,          32, 8)   /* bc8 pad8 uc16 addr32                       */ \
+V(CHECK_GT,          33, 8)   /* bc8 pad8 uc16 addr32                       */ \
+V(CHECK_NOT_BACK_REF, 34, 8)  /* bc8 reg_idx24 addr32                       */ \
+V(CHECK_NOT_BACK_REF_NO_CASE, 35, 8) /* bc8 reg_idx24 addr32                */ \
+V(CHECK_NOT_REGS_EQUAL, 36, 12) /* bc8 regidx24 reg_idx32 addr32            */ \
+V(LOOKUP_MAP1,       37, 12)  /* bc8 pad8 start16 bit_map_addr32 addr32     */ \
+V(LOOKUP_MAP2,       38, 96)  /* bc8 pad8 start16 half_nibble_map_addr32*   */ \
+V(LOOKUP_MAP8,       39, 96)  /* bc8 pad8  start16 byte_map addr32*         */ \
+V(LOOKUP_HI_MAP8,    40, 96)  /* bc8 start24 byte_map_addr32 addr32*        */ \
+V(CHECK_REGISTER_LT, 41, 12)  /* bc8 reg_idx24 value32 addr32               */ \
+V(CHECK_REGISTER_GE, 42, 12)  /* bc8 reg_idx24 value32 addr32               */ \
+V(CHECK_REGISTER_EQ_POS, 43, 8) /* bc8 reg_idx24 addr32                     */ \
+V(CHECK_AT_START,    44, 8)   /* bc8 pad24 addr32                           */ \
+V(CHECK_NOT_AT_START, 45, 8)  /* bc8 pad24 addr32                           */ \
+V(CHECK_GREEDY,      46, 8)   /* bc8 pad24 addr32                           */
 
 #define DECLARE_BYTECODES(name, code, length) \
   static const int BC_##name = code;

src/regexp-macro-assembler-irregexp-inl.h

@@ -36,12 +36,14 @@
 namespace v8 { namespace internal {
 
 
-void RegExpMacroAssemblerIrregexp::Emit(uint32_t byte) {
+void RegExpMacroAssemblerIrregexp::Emit(uint32_t byte, uint32_t twenty_four_bits) {
+  uint32_t word = ((twenty_four_bits << BYTECODE_SHIFT) | byte);
   ASSERT(pc_ <= buffer_.length());
-  if (pc_ == buffer_.length()) {
+  if (pc_  + 3 >= buffer_.length()) {
     Expand();
   }
-  buffer_[pc_++] = byte;
+  Store32(buffer_.start() + pc_, word);
+  pc_ += 4;
 }
 
 

src/regexp-macro-assembler-irregexp.cc

@@ -84,8 +84,7 @@ void RegExpMacroAssemblerIrregexp::EmitOrLink(Label* l) {
 
 
 void RegExpMacroAssemblerIrregexp::PopRegister(int register_index) {
-  Emit(BC_POP_REGISTER);
-  Emit(register_index);
+  Emit(BC_POP_REGISTER, register_index);
 }
 
 
@@ -93,16 +92,14 @@ void RegExpMacroAssemblerIrregexp::PushRegister(
     int register_index,
     StackCheckFlag check_stack_limit) {
   ASSERT(register_index >= 0);
-  Emit(BC_PUSH_REGISTER);
-  Emit(register_index);
+  Emit(BC_PUSH_REGISTER, register_index);
 }
 
 
 void RegExpMacroAssemblerIrregexp::WriteCurrentPositionToRegister(
     int register_index, int cp_offset) {
   ASSERT(register_index >= 0);
-  Emit(BC_SET_REGISTER_TO_CP);
-  Emit(register_index);
+  Emit(BC_SET_REGISTER_TO_CP, register_index);
   Emit32(cp_offset);  // Current position offset.
 }
 
@@ -115,90 +112,83 @@ void RegExpMacroAssemblerIrregexp::ClearRegister(int reg) {
 void RegExpMacroAssemblerIrregexp::ReadCurrentPositionFromRegister(
     int register_index) {
   ASSERT(register_index >= 0);
-  Emit(BC_SET_CP_TO_REGISTER);
-  Emit(register_index);
+  Emit(BC_SET_CP_TO_REGISTER, register_index);
 }
 
 
 void RegExpMacroAssemblerIrregexp::WriteStackPointerToRegister(
     int register_index) {
   ASSERT(register_index >= 0);
-  Emit(BC_SET_REGISTER_TO_SP);
-  Emit(register_index);
+  Emit(BC_SET_REGISTER_TO_SP, register_index);
 }
 
 
 void RegExpMacroAssemblerIrregexp::ReadStackPointerFromRegister(
     int register_index) {
   ASSERT(register_index >= 0);
-  Emit(BC_SET_SP_TO_REGISTER);
-  Emit(register_index);
+  Emit(BC_SET_SP_TO_REGISTER, register_index);
 }
 
 
 void RegExpMacroAssemblerIrregexp::SetRegister(int register_index, int to) {
   ASSERT(register_index >= 0);
-  Emit(BC_SET_REGISTER);
-  Emit(register_index);
+  Emit(BC_SET_REGISTER, register_index);
   Emit32(to);
 }
 
 
 void RegExpMacroAssemblerIrregexp::AdvanceRegister(int register_index, int by) {
   ASSERT(register_index >= 0);
-  Emit(BC_ADVANCE_REGISTER);
-  Emit(register_index);
+  Emit(BC_ADVANCE_REGISTER, register_index);
   Emit32(by);
 }
 
 
 void RegExpMacroAssemblerIrregexp::PopCurrentPosition() {
-  Emit(BC_POP_CP);
+  Emit(BC_POP_CP, 0);
 }
 
 
 void RegExpMacroAssemblerIrregexp::PushCurrentPosition() {
-  Emit(BC_PUSH_CP);
-  Emit32(0);  // Current position offset.
+  Emit(BC_PUSH_CP, 0);
 }
 
 
 void RegExpMacroAssemblerIrregexp::Backtrack() {
-  Emit(BC_POP_BT);
+  Emit(BC_POP_BT, 0);
 }
 
 
 void RegExpMacroAssemblerIrregexp::GoTo(Label* l) {
-  Emit(BC_GOTO);
+  Emit(BC_GOTO, 0);
   EmitOrLink(l);
 }
 
 
 void RegExpMacroAssemblerIrregexp::PushBacktrack(Label* l) {
-  Emit(BC_PUSH_BT);
+  Emit(BC_PUSH_BT, 0);
   EmitOrLink(l);
 }
 
 
 void RegExpMacroAssemblerIrregexp::Succeed() {
-  Emit(BC_SUCCEED);
+  Emit(BC_SUCCEED, 0);
 }
 
 
 void RegExpMacroAssemblerIrregexp::Fail() {
-  Emit(BC_FAIL);
+  Emit(BC_FAIL, 0);
 }
 
 
 void RegExpMacroAssemblerIrregexp::AdvanceCurrentPosition(int by) {
-  Emit(BC_ADVANCE_CP);
-  Emit32(by);
+  Emit(BC_ADVANCE_CP, by);
 }
 
 
 void RegExpMacroAssemblerIrregexp::CheckGreedyLoop(
       Label* on_tos_equals_current_position) {
-  Emit(BC_CHECK_GREEDY);
+  Emit(BC_CHECK_GREEDY, 0);
   EmitOrLink(on_tos_equals_current_position);
 }
 
@@ -227,51 +217,56 @@ void RegExpMacroAssemblerIrregexp::LoadCurrentCharacter(int cp_offset,
       bytecode = BC_LOAD_CURRENT_CHAR_UNCHECKED;
     }
   }
-  Emit(bytecode);
-  Emit32(cp_offset);
+  Emit(bytecode, cp_offset);
   if (check_bounds) EmitOrLink(on_failure);
 }
 
 
 void RegExpMacroAssemblerIrregexp::CheckCharacterLT(uc16 limit,
                                                     Label* on_less) {
-  Emit(BC_CHECK_LT);
-  Emit16(limit);
+  Emit(BC_CHECK_LT, limit);
   EmitOrLink(on_less);
 }
 
 
 void RegExpMacroAssemblerIrregexp::CheckCharacterGT(uc16 limit,
                                                     Label* on_greater) {
-  Emit(BC_CHECK_GT);
-  Emit16(limit);
+  Emit(BC_CHECK_GT, limit);
   EmitOrLink(on_greater);
 }
 
 
 void RegExpMacroAssemblerIrregexp::CheckCharacter(uint32_t c, Label* on_equal) {
-  Emit(BC_CHECK_CHAR);
-  Emit32(c);
+  if (c > MAX_FIRST_ARG) {
+    Emit(BC_CHECK_4_CHARS, 0);
+    Emit32(c);
+  } else {
+    Emit(BC_CHECK_CHAR, c);
+  }
   EmitOrLink(on_equal);
 }
 
 
 void RegExpMacroAssemblerIrregexp::CheckAtStart(Label* on_at_start) {
-  Emit(BC_CHECK_AT_START);
+  Emit(BC_CHECK_AT_START, 0);
   EmitOrLink(on_at_start);
 }
 
 
 void RegExpMacroAssemblerIrregexp::CheckNotAtStart(Label* on_not_at_start) {
-  Emit(BC_CHECK_NOT_AT_START);
+  Emit(BC_CHECK_NOT_AT_START, 0);
   EmitOrLink(on_not_at_start);
 }
 
 
 void RegExpMacroAssemblerIrregexp::CheckNotCharacter(uint32_t c,
                                                      Label* on_not_equal) {
-  Emit(BC_CHECK_NOT_CHAR);
-  Emit32(c);
+  if (c > MAX_FIRST_ARG) {
+    Emit(BC_CHECK_NOT_4_CHARS, 0);
+    Emit32(c);
+  } else {
+    Emit(BC_CHECK_NOT_CHAR, c);
+  }
   EmitOrLink(on_not_equal);
 }
 
@@ -280,8 +275,12 @@ void RegExpMacroAssemblerIrregexp::CheckCharacterAfterAnd(
     uint32_t c,
     uint32_t mask,
     Label* on_equal) {
-  Emit(BC_AND_CHECK_CHAR);
-  Emit32(c);
+  if (c > MAX_FIRST_ARG) {
+    Emit(BC_AND_CHECK_4_CHARS, 0);
+    Emit32(c);
+  } else {
+    Emit(BC_AND_CHECK_CHAR, c);
+  }
   Emit32(mask);
   EmitOrLink(on_equal);
 }
@@ -291,8 +290,12 @@ void RegExpMacroAssemblerIrregexp::CheckNotCharacterAfterAnd(
     uint32_t c,
     uint32_t mask,
     Label* on_not_equal) {
-  Emit(BC_AND_CHECK_NOT_CHAR);
-  Emit32(c);
+  if (c > MAX_FIRST_ARG) {
+    Emit(BC_AND_CHECK_NOT_4_CHARS, 0);
+    Emit32(c);
+  } else {
+    Emit(BC_AND_CHECK_NOT_CHAR, c);
+  }
   Emit32(mask);
   EmitOrLink(on_not_equal);
 }
@@ -303,8 +306,7 @@ void RegExpMacroAssemblerIrregexp::CheckNotCharacterAfterMinusAnd(
     uc16 minus,
     uc16 mask,
     Label* on_not_equal) {
-  Emit(BC_MINUS_AND_CHECK_NOT_CHAR);
-  Emit16(c);
+  Emit(BC_MINUS_AND_CHECK_NOT_CHAR, c);
   Emit16(minus);
   Emit16(mask);
   EmitOrLink(on_not_equal);
@@ -313,8 +315,7 @@ void RegExpMacroAssemblerIrregexp::CheckNotCharacterAfterMinusAnd(
 
 void RegExpMacroAssemblerIrregexp::CheckNotBackReference(int start_reg,
                                                          Label* on_not_equal) {
-  Emit(BC_CHECK_NOT_BACK_REF);
-  Emit(start_reg);
+  Emit(BC_CHECK_NOT_BACK_REF, start_reg);
   EmitOrLink(on_not_equal);
 }
 
@@ -322,8 +323,7 @@ void RegExpMacroAssemblerIrregexp::CheckNotBackReference(int start_reg,
 void RegExpMacroAssemblerIrregexp::CheckNotBackReferenceIgnoreCase(
     int start_reg,
     Label* on_not_equal) {
-  Emit(BC_CHECK_NOT_BACK_REF_NO_CASE);
-  Emit(start_reg);
+  Emit(BC_CHECK_NOT_BACK_REF_NO_CASE, start_reg);
   EmitOrLink(on_not_equal);
 }
 
@@ -331,9 +331,8 @@ void RegExpMacroAssemblerIrregexp::CheckNotBackReferenceIgnoreCase(
 void RegExpMacroAssemblerIrregexp::CheckNotRegistersEqual(int reg1,
                                                           int reg2,
                                                           Label* on_not_equal) {
-  Emit(BC_CHECK_NOT_REGS_EQUAL);
-  Emit(reg1);
-  Emit(reg2);
+  Emit(BC_CHECK_NOT_REGS_EQUAL, reg1);
+  Emit32(reg2);
   EmitOrLink(on_not_equal);
 }
 
@@ -378,15 +377,12 @@ void RegExpMacroAssemblerIrregexp::CheckCharacters(
   // load below.
   for (int i = str.length() - 1; i >= 0; i--) {
     if (check_end_of_string && i == str.length() - 1) {
-      Emit(BC_LOAD_CURRENT_CHAR);
-      Emit32(cp_offset + i);
+      Emit(BC_LOAD_CURRENT_CHAR, cp_offset + i);
       EmitOrLink(on_failure);
     } else {
-      Emit(BC_LOAD_CURRENT_CHAR_UNCHECKED);
-      Emit32(cp_offset + i);
+      Emit(BC_LOAD_CURRENT_CHAR_UNCHECKED, cp_offset + i);
     }
-    Emit(BC_CHECK_NOT_CHAR);
-    Emit32(str[i]);
+    Emit(BC_CHECK_NOT_CHAR, str[i]);
     EmitOrLink(on_failure);
   }
 }
@@ -396,9 +392,8 @@ void RegExpMacroAssemblerIrregexp::IfRegisterLT(int register_index,
                                                 int comparand,
                                                 Label* on_less_than) {
   ASSERT(comparand >= 0 && comparand <= 65535);
-  Emit(BC_CHECK_REGISTER_LT);
-  Emit(register_index);
-  Emit16(comparand);
+  Emit(BC_CHECK_REGISTER_LT, register_index);
+  Emit32(comparand);
   EmitOrLink(on_less_than);
 }
 
@@ -407,24 +402,22 @@ void RegExpMacroAssemblerIrregexp::IfRegisterGE(int register_index,
                                                 int comparand,
                                                 Label* on_greater_or_equal) {
   ASSERT(comparand >= 0 && comparand <= 65535);
-  Emit(BC_CHECK_REGISTER_GE);
-  Emit(register_index);
-  Emit16(comparand);
+  Emit(BC_CHECK_REGISTER_GE, register_index);
+  Emit32(comparand);
   EmitOrLink(on_greater_or_equal);
 }
 
 
 void RegExpMacroAssemblerIrregexp::IfRegisterEqPos(int register_index,
                                                    Label* on_eq) {
-  Emit(BC_CHECK_REGISTER_EQ_POS);
-  Emit(register_index);
+  Emit(BC_CHECK_REGISTER_EQ_POS, register_index);
   EmitOrLink(on_eq);
 }
 
 
 Handle<Object> RegExpMacroAssemblerIrregexp::GetCode(Handle<String> source) {
   Bind(&backtrack_);
-  Emit(BC_POP_BT);
+  Emit(BC_POP_BT, 0);
   Handle<ByteArray> array = Factory::NewByteArray(length());
   Copy(array->GetDataStartAddress());
   return array;

src/regexp-macro-assembler-irregexp.h

@@ -120,7 +120,7 @@ class RegExpMacroAssemblerIrregexp: public RegExpMacroAssembler {
   // Code and bitmap emission.
   inline void Emit32(uint32_t x);
   inline void Emit16(uint32_t x);
-  inline void Emit(uint32_t x);
+  inline void Emit(uint32_t bc, uint32_t arg);
   // Bytecode buffer.
   int length();
   void Copy(Address a);