src/share/vm/adlc/adlparse.cpp

Tue, 24 Dec 2013 11:48:39 -0800

author
mikael
date
Tue, 24 Dec 2013 11:48:39 -0800
changeset 6198
55fb97c4c58d
parent 5791
c9ccd7b85f20
child 6503
a9becfeecd1b
permissions
-rw-r--r--

8029233: Update copyright year to match last edit in jdk8 hotspot repository for 2013
Summary: Copyright year updated for files modified during 2013
Reviewed-by: twisti, iveresov

     1 /*
     2  * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved.
     3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
     4  *
     5  * This code is free software; you can redistribute it and/or modify it
     6  * under the terms of the GNU General Public License version 2 only, as
     7  * published by the Free Software Foundation.
     8  *
     9  * This code is distributed in the hope that it will be useful, but WITHOUT
    10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
    11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
    12  * version 2 for more details (a copy is included in the LICENSE file that
    13  * accompanied this code).
    14  *
    15  * You should have received a copy of the GNU General Public License version
    16  * 2 along with this work; if not, write to the Free Software Foundation,
    17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
    18  *
    19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
    20  * or visit www.oracle.com if you need additional information or have any
    21  * questions.
    22  *
    23  */
    25 // ADLPARSE.CPP - Architecture Description Language Parser
    26 // Authors: Chris Vick and Mike Paleczny
    27 #include "adlc.hpp"
    29 //----------------------------ADLParser----------------------------------------
    30 // Create a new ADL parser
    31 ADLParser::ADLParser(FileBuff& buffer, ArchDesc& archDesc)
    32   : _buf(buffer), _AD(archDesc),
    33     _globalNames(archDesc.globalNames()) {
    34   _AD._syntax_errs = _AD._semantic_errs = 0; // No errors so far this file
    35   _AD._warnings    = 0;                      // No warnings either
    36   _curline         = _ptr = NULL;            // No pointers into buffer yet
    38   _preproc_depth = 0;
    39   _preproc_not_taken = 0;
    41   // Delimit command-line definitions from in-file definitions:
    42   _AD._preproc_list.add_signal();
    43 }
    45 //------------------------------~ADLParser-------------------------------------
    46 // Delete an ADL parser.
    47 ADLParser::~ADLParser() {
    48   if (!_AD._quiet_mode)
    49     fprintf(stderr,"---------------------------- Errors and Warnings ----------------------------\n");
    50 #ifndef ASSERT
    51   fprintf(stderr, "**************************************************************\n");
    52   fprintf(stderr, "***** WARNING: ASSERT is undefined, assertions disabled. *****\n");
    53   fprintf(stderr, "**************************************************************\n");
    54 #endif
    55   if( _AD._syntax_errs + _AD._semantic_errs + _AD._warnings == 0 ) {
    56     if (!_AD._quiet_mode)
    57       fprintf(stderr,"No errors or warnings to report from phase-1 parse.\n" );
    58   }
    59   else {
    60     if( _AD._syntax_errs ) {      // Any syntax errors?
    61       fprintf(stderr,"%s:  Found %d syntax error", _buf._fp->_name, _AD._syntax_errs);
    62       if( _AD._syntax_errs > 1 ) fprintf(stderr,"s.\n\n");
    63       else fprintf(stderr,".\n\n");
    64     }
    65     if( _AD._semantic_errs ) {    // Any semantic errors?
    66       fprintf(stderr,"%s:  Found %d semantic error", _buf._fp->_name, _AD._semantic_errs);
    67       if( _AD._semantic_errs > 1 ) fprintf(stderr,"s.\n\n");
    68       else fprintf(stderr,".\n\n");
    69     }
    70     if( _AD._warnings ) {         // Any warnings?
    71       fprintf(stderr,"%s:  Found %d warning", _buf._fp->_name, _AD._warnings);
    72       if( _AD._warnings > 1 ) fprintf(stderr,"s.\n\n");
    73       else fprintf(stderr,".\n\n");
    74     }
    75   }
    76   if (!_AD._quiet_mode)
    77     fprintf(stderr,"-----------------------------------------------------------------------------\n");
    78   _AD._TotalLines += linenum()-1;     // -1 for overshoot in "nextline" routine
    80   // Write out information we have stored
    81   // // UNIXism == fsync(stderr);
    82 }
    84 //------------------------------parse------------------------------------------
    85 // Each top-level keyword should appear as the first non-whitespace on a line.
    86 //
    87 void ADLParser::parse() {
    88   char *ident;
    90   // Iterate over the lines in the file buffer parsing Level 1 objects
    91   for( next_line(); _curline != NULL; next_line()) {
    92     _ptr = _curline;             // Reset ptr to start of new line
    93     skipws();                    // Skip any leading whitespace
    94     ident = get_ident();         // Get first token
    95     if (ident == NULL) {         // Empty line
    96       continue;                  // Get the next line
    97     }
    98          if (!strcmp(ident, "instruct"))   instr_parse();
    99     else if (!strcmp(ident, "operand"))    oper_parse();
   100     else if (!strcmp(ident, "opclass"))    opclass_parse();
   101     else if (!strcmp(ident, "ins_attrib")) ins_attr_parse();
   102     else if (!strcmp(ident, "op_attrib"))  op_attr_parse();
   103     else if (!strcmp(ident, "source"))     source_parse();
   104     else if (!strcmp(ident, "source_hpp")) source_hpp_parse();
   105     else if (!strcmp(ident, "register"))   reg_parse();
   106     else if (!strcmp(ident, "frame"))      frame_parse();
   107     else if (!strcmp(ident, "encode"))     encode_parse();
   108     else if (!strcmp(ident, "pipeline"))   pipe_parse();
   109     else if (!strcmp(ident, "definitions")) definitions_parse();
   110     else if (!strcmp(ident, "peephole"))   peep_parse();
   111     else if (!strcmp(ident, "#line"))      preproc_line();
   112     else if (!strcmp(ident, "#define"))    preproc_define();
   113     else if (!strcmp(ident, "#undef"))     preproc_undef();
   114     else {
   115       parse_err(SYNERR, "expected one of - instruct, operand, ins_attrib, op_attrib, source, register, pipeline, encode\n     Found %s",ident);
   116     }
   117   }
   118   // Add reg_class spill_regs after parsing.
   119   RegisterForm *regBlock = _AD.get_registers();
   120   if (regBlock == NULL) {
   121     parse_err(SEMERR, "Did not declare 'register' definitions");
   122   }
   123   regBlock->addSpillRegClass();
   125   // Done with parsing, check consistency.
   127   if (_preproc_depth != 0) {
   128     parse_err(SYNERR, "End of file inside #ifdef");
   129   }
   131   // AttributeForms ins_cost and op_cost must be defined for default behaviour
   132   if (_globalNames[AttributeForm::_ins_cost] == NULL) {
   133     parse_err(SEMERR, "Did not declare 'ins_cost' attribute");
   134   }
   135   if (_globalNames[AttributeForm::_op_cost] == NULL) {
   136     parse_err(SEMERR, "Did not declare 'op_cost' attribute");
   137   }
   138 }
   140 // ******************** Private Level 1 Parse Functions ********************
   141 //------------------------------instr_parse------------------------------------
   142 // Parse the contents of an instruction definition, build the InstructForm to
   143 // represent that instruction, and add it to the InstructForm list.
   144 void ADLParser::instr_parse(void) {
   145   char          *ident;
   146   InstructForm  *instr;
   147   MatchRule     *rule;
   148   int            match_rules_cnt = 0;
   150   // First get the name of the instruction
   151   if( (ident = get_unique_ident(_globalNames,"instruction")) == NULL )
   152     return;
   153   instr = new InstructForm(ident); // Create new instruction form
   154   instr->_linenum = linenum();
   155   _globalNames.Insert(ident, instr); // Add name to the name table
   156   // Debugging Stuff
   157   if (_AD._adl_debug > 1)
   158     fprintf(stderr,"Parsing Instruction Form %s\n", ident);
   160   // Then get the operands
   161   skipws();
   162   if (_curchar != '(') {
   163     parse_err(SYNERR, "missing '(' in instruct definition\n");
   164   }
   165   // Parse the operand list
   166   else get_oplist(instr->_parameters, instr->_localNames);
   167   skipws();                        // Skip leading whitespace
   168   // Check for block delimiter
   169   if ( (_curchar != '%')
   170        || ( next_char(),  (_curchar != '{')) ) {
   171     parse_err(SYNERR, "missing '%%{' in instruction definition\n");
   172     return;
   173   }
   174   next_char();                     // Maintain the invariant
   175   do {
   176     ident = get_ident();           // Grab next identifier
   177     if (ident == NULL) {
   178       parse_err(SYNERR, "keyword identifier expected at %c\n", _curchar);
   179       continue;
   180     }
   181     if      (!strcmp(ident, "predicate")) instr->_predicate = pred_parse();
   182     else if      (!strcmp(ident, "match")) {
   183       // Allow one instruction have several match rules.
   184       rule = instr->_matrule;
   185       if (rule == NULL) {
   186         // This is first match rule encountered
   187         rule = match_parse(instr->_localNames);
   188         if (rule) {
   189           instr->_matrule = rule;
   190           // Special case the treatment of Control instructions.
   191           if( instr->is_ideal_control() ) {
   192             // Control instructions return a special result, 'Universe'
   193             rule->_result = "Universe";
   194           }
   195           // Check for commutative operations with tree operands.
   196           matchrule_clone_and_swap(rule, instr->_ident, match_rules_cnt);
   197         }
   198       } else {
   199         // Find the end of the match rule list
   200         while (rule->_next != NULL)
   201           rule = rule->_next;
   202         // Add the new match rule to the list
   203         rule->_next = match_parse(instr->_localNames);
   204         if (rule->_next) {
   205           rule = rule->_next;
   206           if( instr->is_ideal_control() ) {
   207             parse_err(SYNERR, "unique match rule expected for %s\n", rule->_name);
   208             return;
   209           }
   210           assert(match_rules_cnt < 100," too many match rule clones");
   211           char* buf = (char*) malloc(strlen(instr->_ident) + 4);
   212           sprintf(buf, "%s_%d", instr->_ident, match_rules_cnt++);
   213           rule->_result = buf;
   214           // Check for commutative operations with tree operands.
   215           matchrule_clone_and_swap(rule, instr->_ident, match_rules_cnt);
   216         }
   217       }
   218     }
   219     else if (!strcmp(ident, "encode"))  {
   220       parse_err(SYNERR, "Instructions specify ins_encode, not encode\n");
   221     }
   222     else if (!strcmp(ident, "ins_encode"))     ins_encode_parse(*instr);
   223     else if (!strcmp(ident, "opcode"))         instr->_opcode    = opcode_parse(instr);
   224     else if (!strcmp(ident, "size"))           instr->_size      = size_parse(instr);
   225     else if (!strcmp(ident, "effect"))         effect_parse(instr);
   226     else if (!strcmp(ident, "expand"))         instr->_exprule   = expand_parse(instr);
   227     else if (!strcmp(ident, "rewrite"))        instr->_rewrule   = rewrite_parse();
   228     else if (!strcmp(ident, "constraint")) {
   229       parse_err(SYNERR, "Instructions do not specify a constraint\n");
   230     }
   231     else if (!strcmp(ident, "construct")) {
   232       parse_err(SYNERR, "Instructions do not specify a construct\n");
   233     }
   234     else if (!strcmp(ident, "format"))         instr->_format    = format_parse();
   235     else if (!strcmp(ident, "interface")) {
   236       parse_err(SYNERR, "Instructions do not specify an interface\n");
   237     }
   238     else if (!strcmp(ident, "ins_pipe"))        ins_pipe_parse(*instr);
   239     else {  // Done with staticly defined parts of instruction definition
   240       // Check identifier to see if it is the name of an attribute
   241       const Form    *form = _globalNames[ident];
   242       AttributeForm *attr = form ? form->is_attribute() : NULL;
   243       if( attr && (attr->_atype == INS_ATTR) ) {
   244         // Insert the new attribute into the linked list.
   245         Attribute *temp = attr_parse(ident);
   246         temp->_next = instr->_attribs;
   247         instr->_attribs = temp;
   248       } else {
   249         parse_err(SYNERR, "expected one of:\n predicate, match, encode, or the name of an instruction attribute at %s\n", ident);
   250       }
   251     }
   252     skipws();
   253   } while(_curchar != '%');
   254   next_char();
   255   if (_curchar != '}') {
   256     parse_err(SYNERR, "missing '%%}' in instruction definition\n");
   257     return;
   258   }
   259   // Check for "Set" form of chain rule
   260   adjust_set_rule(instr);
   261   if (_AD._pipeline ) {
   262     if( instr->expands() ) {
   263       if( instr->_ins_pipe )
   264         parse_err(WARN, "ins_pipe and expand rule both specified for instruction \"%s\"; ins_pipe will be unused\n", instr->_ident);
   265     } else {
   266       if( !instr->_ins_pipe )
   267         parse_err(WARN, "No ins_pipe specified for instruction \"%s\"\n", instr->_ident);
   268     }
   269   }
   270   // Add instruction to tail of instruction list
   271   _AD.addForm(instr);
   273   // Create instruction form for each additional match rule
   274   rule = instr->_matrule;
   275   if (rule != NULL) {
   276     rule = rule->_next;
   277     while (rule != NULL) {
   278       ident = (char*)rule->_result;
   279       InstructForm *clone = new InstructForm(ident, instr, rule); // Create new instruction form
   280       _globalNames.Insert(ident, clone); // Add name to the name table
   281       // Debugging Stuff
   282       if (_AD._adl_debug > 1)
   283         fprintf(stderr,"Parsing Instruction Form %s\n", ident);
   284       // Check for "Set" form of chain rule
   285       adjust_set_rule(clone);
   286       // Add instruction to tail of instruction list
   287       _AD.addForm(clone);
   288       rule = rule->_next;
   289       clone->_matrule->_next = NULL; // One match rule per clone
   290     }
   291   }
   292 }
   294 //------------------------------matchrule_clone_and_swap-----------------------
   295 // Check for commutative operations with subtree operands,
   296 // create clones and swap operands.
   297 void ADLParser::matchrule_clone_and_swap(MatchRule* rule, const char* instr_ident, int& match_rules_cnt) {
   298   // Check for commutative operations with tree operands.
   299   int count = 0;
   300   rule->count_commutative_op(count);
   301   if (count > 0) {
   302     // Clone match rule and swap commutative operation's operands.
   303     rule->matchrule_swap_commutative_op(instr_ident, count, match_rules_cnt);
   304   }
   305 }
   307 //------------------------------adjust_set_rule--------------------------------
   308 // Check for "Set" form of chain rule
   309 void ADLParser::adjust_set_rule(InstructForm *instr) {
   310   if (instr->_matrule == NULL || instr->_matrule->_rChild == NULL) return;
   311   const char *rch = instr->_matrule->_rChild->_opType;
   312   const Form *frm = _globalNames[rch];
   313   if( (! strcmp(instr->_matrule->_opType,"Set")) &&
   314       frm && frm->is_operand() && (! frm->ideal_only()) ) {
   315     // Previous implementation, which missed leaP*, but worked for loadCon*
   316     unsigned    position = 0;
   317     const char *result   = NULL;
   318     const char *name     = NULL;
   319     const char *optype   = NULL;
   320     MatchNode  *right    = instr->_matrule->_rChild;
   321     if (right->base_operand(position, _globalNames, result, name, optype)) {
   322       position = 1;
   323       const char *result2  = NULL;
   324       const char *name2    = NULL;
   325       const char *optype2  = NULL;
   326       // Can not have additional base operands in right side of match!
   327       if ( ! right->base_operand( position, _globalNames, result2, name2, optype2) ) {
   328         if (instr->_predicate != NULL)
   329           parse_err(SYNERR, "ADLC does not support instruction chain rules with predicates");
   330         // Chain from input  _ideal_operand_type_,
   331         // Needed for shared roots of match-trees
   332         ChainList *lst = (ChainList *)_AD._chainRules[optype];
   333         if (lst == NULL) {
   334           lst = new ChainList();
   335           _AD._chainRules.Insert(optype, lst);
   336         }
   337         if (!lst->search(instr->_matrule->_lChild->_opType)) {
   338           const char *cost = instr->cost();
   339           if (cost == NULL) {
   340             cost = ((AttributeForm*)_globalNames[AttributeForm::_ins_cost])->_attrdef;
   341           }
   342           // The ADLC does not support chaining from the ideal operand type
   343           // of a predicated user-defined operand
   344           if( frm->is_operand() == NULL || frm->is_operand()->_predicate == NULL ) {
   345             lst->insert(instr->_matrule->_lChild->_opType,cost,instr->_ident);
   346           }
   347         }
   348         // Chain from input  _user_defined_operand_type_,
   349         lst = (ChainList *)_AD._chainRules[result];
   350         if (lst == NULL) {
   351           lst = new ChainList();
   352           _AD._chainRules.Insert(result, lst);
   353         }
   354         if (!lst->search(instr->_matrule->_lChild->_opType)) {
   355           const char *cost = instr->cost();
   356           if (cost == NULL) {
   357             cost = ((AttributeForm*)_globalNames[AttributeForm::_ins_cost])->_attrdef;
   358           }
   359           // It is safe to chain from the top-level user-defined operand even
   360           // if it has a predicate, since the predicate is checked before
   361           // the user-defined type is available.
   362           lst->insert(instr->_matrule->_lChild->_opType,cost,instr->_ident);
   363         }
   364       } else {
   365         // May have instruction chain rule if root of right-tree is an ideal
   366         OperandForm *rightOp = _globalNames[right->_opType]->is_operand();
   367         if( rightOp ) {
   368           const Form *rightRoot = _globalNames[rightOp->_matrule->_opType];
   369           if( rightRoot && rightRoot->ideal_only() ) {
   370             const char *chain_op = NULL;
   371             if( rightRoot->is_instruction() )
   372               chain_op = rightOp->_ident;
   373             if( chain_op ) {
   374               // Look-up the operation in chain rule table
   375               ChainList *lst = (ChainList *)_AD._chainRules[chain_op];
   376               if (lst == NULL) {
   377                 lst = new ChainList();
   378                 _AD._chainRules.Insert(chain_op, lst);
   379               }
   380               // if (!lst->search(instr->_matrule->_lChild->_opType)) {
   381               const char *cost = instr->cost();
   382               if (cost == NULL) {
   383                 cost = ((AttributeForm*)_globalNames[AttributeForm::_ins_cost])->_attrdef;
   384               }
   385               // This chains from a top-level operand whose predicate, if any,
   386               // has been checked.
   387               lst->insert(instr->_matrule->_lChild->_opType,cost,instr->_ident);
   388               // }
   389             }
   390           }
   391         }
   392       } // end chain rule from right-tree's ideal root
   393     }
   394   }
   395 }
   398 //------------------------------oper_parse-------------------------------------
   399 void ADLParser::oper_parse(void) {
   400   char          *ident;
   401   OperandForm   *oper;
   402   AttributeForm *attr;
   403   MatchRule     *rule;
   405   // First get the name of the operand
   406   skipws();
   407   if( (ident = get_unique_ident(_globalNames,"operand")) == NULL )
   408     return;
   409   oper = new OperandForm(ident);        // Create new operand form
   410   oper->_linenum = linenum();
   411   _globalNames.Insert(ident, oper); // Add name to the name table
   413   // Debugging Stuff
   414   if (_AD._adl_debug > 1) fprintf(stderr,"Parsing Operand Form %s\n", ident);
   416   // Get the component operands
   417   skipws();
   418   if (_curchar != '(') {
   419     parse_err(SYNERR, "missing '(' in operand definition\n");
   420     return;
   421   }
   422   else get_oplist(oper->_parameters, oper->_localNames); // Parse the component operand list
   423   skipws();
   424   // Check for block delimiter
   425   if ((_curchar != '%') || (*(_ptr+1) != '{')) { // If not open block
   426     parse_err(SYNERR, "missing '%%{' in operand definition\n");
   427     return;
   428   }
   429   next_char(); next_char();        // Skip over "%{" symbol
   430   do {
   431     ident = get_ident();           // Grab next identifier
   432     if (ident == NULL) {
   433       parse_err(SYNERR, "keyword identifier expected at %c\n", _curchar);
   434       continue;
   435     }
   436     if      (!strcmp(ident, "predicate")) oper->_predicate = pred_parse();
   437     else if (!strcmp(ident, "match"))     {
   438       // Find the end of the match rule list
   439       rule = oper->_matrule;
   440       if (rule) {
   441         while (rule->_next) rule = rule->_next;
   442         // Add the new match rule to the list
   443         rule->_next = match_parse(oper->_localNames);
   444         if (rule->_next) {
   445           rule->_next->_result = oper->_ident;
   446         }
   447       }
   448       else {
   449         // This is first match rule encountered
   450         oper->_matrule = match_parse(oper->_localNames);
   451         if (oper->_matrule) {
   452           oper->_matrule->_result = oper->_ident;
   453         }
   454       }
   455     }
   456     else if (!strcmp(ident, "encode"))    oper->_interface = interface_parse();
   457     else if (!strcmp(ident, "ins_encode")) {
   458       parse_err(SYNERR, "Operands specify 'encode', not 'ins_encode'\n");
   459     }
   460     else if (!strcmp(ident, "opcode"))    {
   461       parse_err(SYNERR, "Operands do not specify an opcode\n");
   462     }
   463     else if (!strcmp(ident, "effect"))    {
   464       parse_err(SYNERR, "Operands do not specify an effect\n");
   465     }
   466     else if (!strcmp(ident, "expand"))    {
   467       parse_err(SYNERR, "Operands do not specify an expand\n");
   468     }
   469     else if (!strcmp(ident, "rewrite"))   {
   470       parse_err(SYNERR, "Operands do not specify a rewrite\n");
   471     }
   472     else if (!strcmp(ident, "constraint"))oper->_constraint= constraint_parse();
   473     else if (!strcmp(ident, "construct")) oper->_construct = construct_parse();
   474     else if (!strcmp(ident, "format"))    oper->_format    = format_parse();
   475     else if (!strcmp(ident, "interface")) oper->_interface = interface_parse();
   476     // Check identifier to see if it is the name of an attribute
   477     else if (((attr = _globalNames[ident]->is_attribute()) != NULL) &&
   478              (attr->_atype == OP_ATTR))   oper->_attribs   = attr_parse(ident);
   479     else {
   480       parse_err(SYNERR, "expected one of - constraint, predicate, match, encode, format, construct, or the name of a defined operand attribute at %s\n", ident);
   481     }
   482     skipws();
   483   } while(_curchar != '%');
   484   next_char();
   485   if (_curchar != '}') {
   486     parse_err(SYNERR, "missing '%%}' in operand definition\n");
   487     return;
   488   }
   489   // Add operand to tail of operand list
   490   _AD.addForm(oper);
   491 }
   493 //------------------------------opclass_parse----------------------------------
   494 // Operand Classes are a block with a comma delimited list of operand names
   495 void ADLParser::opclass_parse(void) {
   496   char          *ident;
   497   OpClassForm   *opc;
   498   OperandForm   *opForm;
   500   // First get the name of the operand class
   501   skipws();
   502   if( (ident = get_unique_ident(_globalNames,"opclass")) == NULL )
   503     return;
   504   opc = new OpClassForm(ident);             // Create new operand class form
   505   _globalNames.Insert(ident, opc);  // Add name to the name table
   507   // Debugging Stuff
   508   if (_AD._adl_debug > 1)
   509     fprintf(stderr,"Parsing Operand Class Form %s\n", ident);
   511   // Get the list of operands
   512   skipws();
   513   if (_curchar != '(') {
   514     parse_err(SYNERR, "missing '(' in operand definition\n");
   515     return;
   516   }
   517   do {
   518     next_char();                            // Skip past open paren or comma
   519     ident = get_ident();                    // Grab next identifier
   520     if (ident == NULL) {
   521       parse_err(SYNERR, "keyword identifier expected at %c\n", _curchar);
   522       continue;
   523     }
   524     // Check identifier to see if it is the name of an operand
   525     const Form *form = _globalNames[ident];
   526     opForm     = form ? form->is_operand() : NULL;
   527     if ( opForm ) {
   528       opc->_oplst.addName(ident);           // Add operand to opclass list
   529       opForm->_classes.addName(opc->_ident);// Add opclass to operand list
   530     }
   531     else {
   532       parse_err(SYNERR, "expected name of a defined operand at %s\n", ident);
   533     }
   534     skipws();                               // skip trailing whitespace
   535   } while (_curchar == ',');                // Check for the comma
   536   // Check for closing ')'
   537   if (_curchar != ')') {
   538     parse_err(SYNERR, "missing ')' or ',' in opclass definition\n");
   539     return;
   540   }
   541   next_char();                              // Consume the ')'
   542   skipws();
   543   // Check for closing ';'
   544   if (_curchar != ';') {
   545     parse_err(SYNERR, "missing ';' in opclass definition\n");
   546     return;
   547   }
   548   next_char();                             // Consume the ';'
   549   // Add operand to tail of operand list
   550   _AD.addForm(opc);
   551 }
   553 //------------------------------ins_attr_parse---------------------------------
   554 void ADLParser::ins_attr_parse(void) {
   555   char          *ident;
   556   char          *aexpr;
   557   AttributeForm *attrib;
   559   // get name for the instruction attribute
   560   skipws();                      // Skip leading whitespace
   561   if( (ident = get_unique_ident(_globalNames,"inst_attrib")) == NULL )
   562     return;
   563   // Debugging Stuff
   564   if (_AD._adl_debug > 1) fprintf(stderr,"Parsing Ins_Attribute Form %s\n", ident);
   566   // Get default value of the instruction attribute
   567   skipws();                      // Skip whitespace
   568   if ((aexpr = get_paren_expr("attribute default expression string")) == NULL) {
   569     parse_err(SYNERR, "missing '(' in ins_attrib definition\n");
   570     return;
   571   }
   572   // Debug Stuff
   573   if (_AD._adl_debug > 1) fprintf(stderr,"Attribute Expression: %s\n", aexpr);
   575   // Check for terminator
   576   if (_curchar != ';') {
   577     parse_err(SYNERR, "missing ';' in ins_attrib definition\n");
   578     return;
   579   }
   580   next_char();                    // Advance past the ';'
   582   // Construct the attribute, record global name, and store in ArchDesc
   583   attrib = new AttributeForm(ident, INS_ATTR, aexpr);
   584   _globalNames.Insert(ident, attrib);  // Add name to the name table
   585   _AD.addForm(attrib);
   586 }
   588 //------------------------------op_attr_parse----------------------------------
   589 void ADLParser::op_attr_parse(void) {
   590   char          *ident;
   591   char          *aexpr;
   592   AttributeForm *attrib;
   594   // get name for the operand attribute
   595   skipws();                      // Skip leading whitespace
   596   if( (ident = get_unique_ident(_globalNames,"op_attrib")) == NULL )
   597     return;
   598   // Debugging Stuff
   599   if (_AD._adl_debug > 1) fprintf(stderr,"Parsing Op_Attribute Form %s\n", ident);
   601   // Get default value of the instruction attribute
   602   skipws();                      // Skip whitespace
   603   if ((aexpr = get_paren_expr("attribute default expression string")) == NULL) {
   604     parse_err(SYNERR, "missing '(' in op_attrib definition\n");
   605     return;
   606   }
   607   // Debug Stuff
   608   if (_AD._adl_debug > 1) fprintf(stderr,"Attribute Expression: %s\n", aexpr);
   610   // Check for terminator
   611   if (_curchar != ';') {
   612     parse_err(SYNERR, "missing ';' in op_attrib definition\n");
   613     return;
   614   }
   615   next_char();                    // Advance past the ';'
   617   // Construct the attribute, record global name, and store in ArchDesc
   618   attrib = new AttributeForm(ident, OP_ATTR, aexpr);
   619   _globalNames.Insert(ident, attrib);
   620   _AD.addForm(attrib);
   621 }
   623 //------------------------------definitions_parse-----------------------------------
   624 void ADLParser::definitions_parse(void) {
   625   skipws();                       // Skip leading whitespace
   626   if (_curchar == '%' && *(_ptr+1) == '{') {
   627     next_char(); next_char();     // Skip "%{"
   628     skipws();
   629     while (_curchar != '%' && *(_ptr+1) != '}') {
   630       // Process each definition until finding closing string "%}"
   631       char *token = get_ident();
   632       if (token == NULL) {
   633         parse_err(SYNERR, "missing identifier inside definitions block.\n");
   634         return;
   635       }
   636       if (strcmp(token,"int_def")==0)     { int_def_parse(); }
   637       // if (strcmp(token,"str_def")==0)   { str_def_parse(); }
   638       skipws();
   639     }
   640   }
   641   else {
   642     parse_err(SYNERR, "Missing %%{ ... %%} block after definitions keyword.\n");
   643     return;
   644   }
   645 }
   647 //------------------------------int_def_parse----------------------------------
   648 // Parse Example:
   649 // int_def    MEMORY_REF_COST      (         200,  DEFAULT_COST * 2);
   650 // <keyword>  <name>               ( <int_value>,   <description>  );
   651 //
   652 void ADLParser::int_def_parse(void) {
   653   char *name        = NULL;         // Name of definition
   654   char *value       = NULL;         // its value,
   655   int   int_value   = -1;           // positive values only
   656   char *description = NULL;         // textual description
   658   // Get definition name
   659   skipws();                      // Skip whitespace
   660   name = get_ident();
   661   if (name == NULL) {
   662     parse_err(SYNERR, "missing definition name after int_def\n");
   663     return;
   664   }
   666   // Check for value of int_def dname( integer_value [, string_expression ] )
   667   skipws();
   668   if (_curchar == '(') {
   670     // Parse the integer value.
   671     next_char();
   672     value = get_ident();
   673     if (value == NULL) {
   674       parse_err(SYNERR, "missing value in int_def\n");
   675       return;
   676     }
   677     if( !is_int_token(value, int_value) ) {
   678       parse_err(SYNERR, "value in int_def is not recognized as integer\n");
   679       return;
   680     }
   681     skipws();
   683     // Check for description
   684     if (_curchar == ',') {
   685       next_char();   // skip ','
   687       description = get_expr("int_def description", ")");
   688       if (description == NULL) {
   689         parse_err(SYNERR, "invalid or missing description in int_def\n");
   690         return;
   691       }
   692       trim(description);
   693     }
   695     if (_curchar != ')') {
   696       parse_err(SYNERR, "missing ')' in register definition statement\n");
   697       return;
   698     }
   699     next_char();
   700   }
   702   // Check for closing ';'
   703   skipws();
   704   if (_curchar != ';') {
   705     parse_err(SYNERR, "missing ';' after int_def\n");
   706     return;
   707   }
   708   next_char();                   // move past ';'
   710   // Debug Stuff
   711   if (_AD._adl_debug > 1) {
   712     fprintf(stderr,"int_def: %s ( %s, %s )\n", name,
   713             (value), (description ? description : ""));
   714   }
   716   // Record new definition.
   717   Expr *expr     = new Expr(name, description, int_value, int_value);
   718   const Expr *old_expr = _AD.globalDefs().define(name, expr);
   719   if (old_expr != NULL) {
   720     parse_err(SYNERR, "Duplicate definition\n");
   721     return;
   722   }
   724   return;
   725 }
   728 //------------------------------source_parse-----------------------------------
   729 void ADLParser::source_parse(void) {
   730   SourceForm *source;             // Encode class for instruction/operand
   731   char   *rule = NULL;            // String representation of encode rule
   733   skipws();                       // Skip leading whitespace
   734   if ( (rule = find_cpp_block("source block")) == NULL ) {
   735     parse_err(SYNERR, "incorrect or missing block for 'source'.\n");
   736     return;
   737   }
   738   // Debug Stuff
   739   if (_AD._adl_debug > 1) fprintf(stderr,"Source Form: %s\n", rule);
   741   source = new SourceForm(rule);    // Build new Source object
   742   _AD.addForm(source);
   743   // skipws();
   744 }
   746 //------------------------------source_hpp_parse-------------------------------
   747 // Parse a source_hpp %{ ... %} block.
   748 // The code gets stuck into the ad_<arch>.hpp file.
   749 // If the source_hpp block appears before the register block in the AD
   750 // file, it goes up at the very top of the ad_<arch>.hpp file, so that
   751 // it can be used by register encodings, etc.  Otherwise, it goes towards
   752 // the bottom, where it's useful as a global definition to *.cpp files.
   753 void ADLParser::source_hpp_parse(void) {
   754   char   *rule = NULL;            // String representation of encode rule
   756   skipws();                       // Skip leading whitespace
   757   if ( (rule = find_cpp_block("source_hpp block")) == NULL ) {
   758     parse_err(SYNERR, "incorrect or missing block for 'source_hpp'.\n");
   759     return;
   760   }
   761   // Debug Stuff
   762   if (_AD._adl_debug > 1) fprintf(stderr,"Header Form: %s\n", rule);
   764   if (_AD.get_registers() == NULL) {
   765     // Very early in the file, before reg_defs, we collect pre-headers.
   766     PreHeaderForm* pre_header = new PreHeaderForm(rule);
   767     _AD.addForm(pre_header);
   768   } else {
   769     // Normally, we collect header info, placed at the bottom of the hpp file.
   770     HeaderForm* header = new HeaderForm(rule);
   771     _AD.addForm(header);
   772   }
   773 }
   775 //------------------------------reg_parse--------------------------------------
   776 void ADLParser::reg_parse(void) {
   777   RegisterForm *regBlock = _AD.get_registers(); // Information about registers encoding
   778   if (regBlock == NULL) {
   779     // Create the RegisterForm for the architecture description.
   780     regBlock = new RegisterForm();    // Build new Source object
   781     _AD.addForm(regBlock);
   782   }
   784   skipws();                       // Skip leading whitespace
   785   if (_curchar == '%' && *(_ptr+1) == '{') {
   786     next_char(); next_char();     // Skip "%{"
   787     skipws();
   788     while (_curchar != '%' && *(_ptr+1) != '}') {
   789       char *token = get_ident();
   790       if (token == NULL) {
   791         parse_err(SYNERR, "missing identifier inside register block.\n");
   792         return;
   793       }
   794       if (strcmp(token,"reg_def")==0)          { reg_def_parse(); }
   795       else if (strcmp(token,"reg_class")==0)   { reg_class_parse(); }
   796       else if (strcmp(token,"alloc_class")==0) { alloc_class_parse(); }
   797       else if (strcmp(token,"#define")==0)     { preproc_define(); }
   798       else { parse_err(SYNERR, "bad token %s inside register block.\n", token); break; }
   799       skipws();
   800     }
   801   }
   802   else {
   803     parse_err(SYNERR, "Missing %c{ ... %c} block after register keyword.\n",'%','%');
   804     return;
   805   }
   806 }
   808 //------------------------------encode_parse-----------------------------------
   809 void ADLParser::encode_parse(void) {
   810   EncodeForm *encBlock;         // Information about instruction/operand encoding
   812   _AD.getForm(&encBlock);
   813   if ( encBlock == NULL) {
   814     // Create the EncodeForm for the architecture description.
   815     encBlock = new EncodeForm();    // Build new Source object
   816     _AD.addForm(encBlock);
   817   }
   819   skipws();                       // Skip leading whitespace
   820   if (_curchar == '%' && *(_ptr+1) == '{') {
   821     next_char(); next_char();     // Skip "%{"
   822     skipws();
   823     while (_curchar != '%' && *(_ptr+1) != '}') {
   824       char *token = get_ident();
   825       if (token == NULL) {
   826             parse_err(SYNERR, "missing identifier inside encoding block.\n");
   827             return;
   828       }
   829       if (strcmp(token,"enc_class")==0)   { enc_class_parse(); }
   830       skipws();
   831     }
   832   }
   833   else {
   834     parse_err(SYNERR, "Missing %c{ ... %c} block after encode keyword.\n",'%','%');
   835     return;
   836   }
   837 }
   839 //------------------------------enc_class_parse--------------------------------
   840 void ADLParser::enc_class_parse(void) {
   841   char       *ec_name;           // Name of encoding class being defined
   843   // Get encoding class name
   844   skipws();                      // Skip whitespace
   845   ec_name = get_ident();
   846   if (ec_name == NULL) {
   847     parse_err(SYNERR, "missing encoding class name after encode.\n");
   848     return;
   849   }
   851   EncClass  *encoding = _AD._encode->add_EncClass(ec_name);
   852   encoding->_linenum = linenum();
   854   skipws();                      // Skip leading whitespace
   855   // Check for optional parameter list
   856   if (_curchar == '(') {
   857     do {
   858       char *pType = NULL;        // parameter type
   859       char *pName = NULL;        // parameter name
   861       next_char();               // skip open paren & comma characters
   862       skipws();
   863       if (_curchar == ')') break;
   865       // Get parameter type
   866       pType = get_ident();
   867       if (pType == NULL) {
   868         parse_err(SYNERR, "parameter type expected at %c\n", _curchar);
   869         return;
   870       }
   872       skipws();
   873       // Get parameter name
   874       pName = get_ident();
   875       if (pName == NULL) {
   876         parse_err(SYNERR, "parameter name expected at %c\n", _curchar);
   877         return;
   878       }
   880       // Record parameter type and name
   881       encoding->add_parameter( pType, pName );
   883       skipws();
   884     } while(_curchar == ',');
   886     if (_curchar != ')') parse_err(SYNERR, "missing ')'\n");
   887     else {
   888       next_char();                  // Skip ')'
   889     }
   890   } // Done with parameter list
   892   skipws();
   893   // Check for block starting delimiters
   894   if ((_curchar != '%') || (*(_ptr+1) != '{')) { // If not open block
   895     parse_err(SYNERR, "missing '%c{' in enc_class definition\n", '%');
   896     return;
   897   }
   898   next_char();                      // Skip '%'
   899   next_char();                      // Skip '{'
   901   enc_class_parse_block(encoding, ec_name);
   902 }
   905 void ADLParser::enc_class_parse_block(EncClass* encoding, char* ec_name) {
   906   skipws_no_preproc();              // Skip leading whitespace
   907   // Prepend location descriptor, for debugging; cf. ADLParser::find_cpp_block
   908   if (_AD._adlocation_debug) {
   909     encoding->add_code(get_line_string());
   910   }
   912   // Collect the parts of the encode description
   913   // (1) strings that are passed through to output
   914   // (2) replacement/substitution variable, preceeded by a '$'
   915   while ( (_curchar != '%') && (*(_ptr+1) != '}') ) {
   917     // (1)
   918     // Check if there is a string to pass through to output
   919     char *start = _ptr;       // Record start of the next string
   920     while ((_curchar != '$') && ((_curchar != '%') || (*(_ptr+1) != '}')) ) {
   921       // If at the start of a comment, skip past it
   922       if( (_curchar == '/') && ((*(_ptr+1) == '/') || (*(_ptr+1) == '*')) ) {
   923         skipws_no_preproc();
   924       } else {
   925         // ELSE advance to the next character, or start of the next line
   926         next_char_or_line();
   927       }
   928     }
   929     // If a string was found, terminate it and record in EncClass
   930     if ( start != _ptr ) {
   931       *_ptr  = '\0';          // Terminate the string
   932       encoding->add_code(start);
   933     }
   935     // (2)
   936     // If we are at a replacement variable,
   937     // copy it and record in EncClass
   938     if (_curchar == '$') {
   939       // Found replacement Variable
   940       char* rep_var = get_rep_var_ident_dup();
   941       // Add flag to _strings list indicating we should check _rep_vars
   942       encoding->add_rep_var(rep_var);
   943     }
   944   } // end while part of format description
   945   next_char();                      // Skip '%'
   946   next_char();                      // Skip '}'
   948   skipws();
   950   if (_AD._adlocation_debug) {
   951     encoding->add_code(end_line_marker());
   952   }
   954   // Debug Stuff
   955   if (_AD._adl_debug > 1) fprintf(stderr,"EncodingClass Form: %s\n", ec_name);
   956 }
   958 //------------------------------frame_parse-----------------------------------
   959 void ADLParser::frame_parse(void) {
   960   FrameForm  *frame;              // Information about stack-frame layout
   961   char       *desc = NULL;        // String representation of frame
   963   skipws();                       // Skip leading whitespace
   965   frame = new FrameForm();        // Build new Frame object
   966   // Check for open block sequence
   967   skipws();                       // Skip leading whitespace
   968   if (_curchar == '%' && *(_ptr+1) == '{') {
   969     next_char(); next_char();     // Skip "%{"
   970     skipws();
   971     while (_curchar != '%' && *(_ptr+1) != '}') {
   972       char *token = get_ident();
   973       if (token == NULL) {
   974             parse_err(SYNERR, "missing identifier inside frame block.\n");
   975             return;
   976       }
   977       if (strcmp(token,"stack_direction")==0) {
   978         stack_dir_parse(frame);
   979       }
   980       if (strcmp(token,"sync_stack_slots")==0) {
   981         sync_stack_slots_parse(frame);
   982       }
   983       if (strcmp(token,"frame_pointer")==0) {
   984         frame_pointer_parse(frame, false);
   985       }
   986       if (strcmp(token,"interpreter_frame_pointer")==0) {
   987         interpreter_frame_pointer_parse(frame, false);
   988       }
   989       if (strcmp(token,"inline_cache_reg")==0) {
   990         inline_cache_parse(frame, false);
   991       }
   992       if (strcmp(token,"compiler_method_oop_reg")==0) {
   993         parse_err(WARN, "Using obsolete Token, compiler_method_oop_reg");
   994         skipws();
   995       }
   996       if (strcmp(token,"interpreter_method_oop_reg")==0) {
   997         interpreter_method_oop_parse(frame, false);
   998       }
   999       if (strcmp(token,"cisc_spilling_operand_name")==0) {
  1000         cisc_spilling_operand_name_parse(frame, false);
  1002       if (strcmp(token,"stack_alignment")==0) {
  1003         stack_alignment_parse(frame);
  1005       if (strcmp(token,"return_addr")==0) {
  1006         return_addr_parse(frame, false);
  1008       if (strcmp(token,"in_preserve_stack_slots")==0) {
  1009         preserve_stack_parse(frame);
  1011       if (strcmp(token,"out_preserve_stack_slots")==0) {
  1012         parse_err(WARN, "Using obsolete token, out_preserve_stack_slots");
  1013         skipws();
  1015       if (strcmp(token,"varargs_C_out_slots_killed")==0) {
  1016         frame->_varargs_C_out_slots_killed = parse_one_arg("varargs C out slots killed");
  1018       if (strcmp(token,"calling_convention")==0) {
  1019         frame->_calling_convention = calling_convention_parse();
  1021       if (strcmp(token,"return_value")==0) {
  1022         frame->_return_value = return_value_parse();
  1024       if (strcmp(token,"c_frame_pointer")==0) {
  1025         frame_pointer_parse(frame, true);
  1027       if (strcmp(token,"c_return_addr")==0) {
  1028         return_addr_parse(frame, true);
  1030       if (strcmp(token,"c_calling_convention")==0) {
  1031         frame->_c_calling_convention = calling_convention_parse();
  1033       if (strcmp(token,"c_return_value")==0) {
  1034         frame->_c_return_value = return_value_parse();
  1037       skipws();
  1040   else {
  1041     parse_err(SYNERR, "Missing %c{ ... %c} block after encode keyword.\n",'%','%');
  1042     return;
  1044   // All Java versions are required, native versions are optional
  1045   if(frame->_frame_pointer == NULL) {
  1046     parse_err(SYNERR, "missing frame pointer definition in frame section.\n");
  1047     return;
  1049   // !!!!! !!!!!
  1050   // if(frame->_interpreter_frame_ptr_reg == NULL) {
  1051   //   parse_err(SYNERR, "missing interpreter frame pointer definition in frame section.\n");
  1052   //   return;
  1053   // }
  1054   if(frame->_alignment == NULL) {
  1055     parse_err(SYNERR, "missing alignment definition in frame section.\n");
  1056     return;
  1058   if(frame->_return_addr == NULL) {
  1059     parse_err(SYNERR, "missing return address location in frame section.\n");
  1060     return;
  1062   if(frame->_in_preserve_slots == NULL) {
  1063     parse_err(SYNERR, "missing stack slot preservation definition in frame section.\n");
  1064     return;
  1066   if(frame->_varargs_C_out_slots_killed == NULL) {
  1067     parse_err(SYNERR, "missing varargs C out slots killed definition in frame section.\n");
  1068     return;
  1070   if(frame->_calling_convention == NULL) {
  1071     parse_err(SYNERR, "missing calling convention definition in frame section.\n");
  1072     return;
  1074   if(frame->_return_value == NULL) {
  1075     parse_err(SYNERR, "missing return value definition in frame section.\n");
  1076     return;
  1078   // Fill natives in identically with the Java versions if not present.
  1079   if(frame->_c_frame_pointer == NULL) {
  1080     frame->_c_frame_pointer = frame->_frame_pointer;
  1082   if(frame->_c_return_addr == NULL) {
  1083     frame->_c_return_addr = frame->_return_addr;
  1084     frame->_c_return_addr_loc = frame->_return_addr_loc;
  1086   if(frame->_c_calling_convention == NULL) {
  1087     frame->_c_calling_convention = frame->_calling_convention;
  1089   if(frame->_c_return_value == NULL) {
  1090     frame->_c_return_value = frame->_return_value;
  1093   // Debug Stuff
  1094   if (_AD._adl_debug > 1) fprintf(stderr,"Frame Form: %s\n", desc);
  1096   // Create the EncodeForm for the architecture description.
  1097   _AD.addForm(frame);
  1098   // skipws();
  1101 //------------------------------stack_dir_parse--------------------------------
  1102 void ADLParser::stack_dir_parse(FrameForm *frame) {
  1103   char *direction = parse_one_arg("stack direction entry");
  1104   if (strcmp(direction, "TOWARDS_LOW") == 0) {
  1105     frame->_direction = false;
  1107   else if (strcmp(direction, "TOWARDS_HIGH") == 0) {
  1108     frame->_direction = true;
  1110   else {
  1111     parse_err(SYNERR, "invalid value inside stack direction entry.\n");
  1112     return;
  1116 //------------------------------sync_stack_slots_parse-------------------------
  1117 void ADLParser::sync_stack_slots_parse(FrameForm *frame) {
  1118     // Assign value into frame form
  1119     frame->_sync_stack_slots = parse_one_arg("sync stack slots entry");
  1122 //------------------------------frame_pointer_parse----------------------------
  1123 void ADLParser::frame_pointer_parse(FrameForm *frame, bool native) {
  1124   char *frame_pointer = parse_one_arg("frame pointer entry");
  1125   // Assign value into frame form
  1126   if (native) { frame->_c_frame_pointer = frame_pointer; }
  1127   else        { frame->_frame_pointer   = frame_pointer; }
  1130 //------------------------------interpreter_frame_pointer_parse----------------------------
  1131 void ADLParser::interpreter_frame_pointer_parse(FrameForm *frame, bool native) {
  1132   frame->_interpreter_frame_pointer_reg = parse_one_arg("interpreter frame pointer entry");
  1135 //------------------------------inline_cache_parse-----------------------------
  1136 void ADLParser::inline_cache_parse(FrameForm *frame, bool native) {
  1137   frame->_inline_cache_reg = parse_one_arg("inline cache reg entry");
  1140 //------------------------------interpreter_method_oop_parse------------------
  1141 void ADLParser::interpreter_method_oop_parse(FrameForm *frame, bool native) {
  1142   frame->_interpreter_method_oop_reg = parse_one_arg("method oop reg entry");
  1145 //------------------------------cisc_spilling_operand_parse---------------------
  1146 void ADLParser::cisc_spilling_operand_name_parse(FrameForm *frame, bool native) {
  1147   frame->_cisc_spilling_operand_name = parse_one_arg("cisc spilling operand name");
  1150 //------------------------------stack_alignment_parse--------------------------
  1151 void ADLParser::stack_alignment_parse(FrameForm *frame) {
  1152   char *alignment = parse_one_arg("stack alignment entry");
  1153   // Assign value into frame
  1154   frame->_alignment   = alignment;
  1157 //------------------------------parse_one_arg-------------------------------
  1158 char *ADLParser::parse_one_arg(const char *description) {
  1159   char *token = NULL;
  1160   if(_curchar == '(') {
  1161     next_char();
  1162     skipws();
  1163     token = get_expr(description, ")");
  1164     if (token == NULL) {
  1165       parse_err(SYNERR, "missing value inside %s.\n", description);
  1166       return NULL;
  1168     next_char();           // skip the close paren
  1169     if(_curchar != ';') {  // check for semi-colon
  1170       parse_err(SYNERR, "missing %c in.\n", ';', description);
  1171       return NULL;
  1173     next_char();           // skip the semi-colon
  1175   else {
  1176     parse_err(SYNERR, "Missing %c in.\n", '(', description);
  1177     return NULL;
  1180   trim(token);
  1181   return token;
  1184 //------------------------------return_addr_parse------------------------------
  1185 void ADLParser::return_addr_parse(FrameForm *frame, bool native) {
  1186   bool in_register  = true;
  1187   if(_curchar == '(') {
  1188     next_char();
  1189     skipws();
  1190     char *token = get_ident();
  1191     if (token == NULL) {
  1192       parse_err(SYNERR, "missing value inside return address entry.\n");
  1193       return;
  1195     // check for valid values for stack/register
  1196     if (strcmp(token, "REG") == 0) {
  1197       in_register = true;
  1199     else if (strcmp(token, "STACK") == 0) {
  1200       in_register = false;
  1202     else {
  1203       parse_err(SYNERR, "invalid value inside return_address entry.\n");
  1204       return;
  1206     if (native) { frame->_c_return_addr_loc = in_register; }
  1207     else        { frame->_return_addr_loc   = in_register; }
  1209     // Parse expression that specifies register or stack position
  1210     skipws();
  1211     char *token2 = get_expr("return address entry", ")");
  1212     if (token2 == NULL) {
  1213       parse_err(SYNERR, "missing value inside return address entry.\n");
  1214       return;
  1216     next_char();           // skip the close paren
  1217     if (native) { frame->_c_return_addr = token2; }
  1218     else        { frame->_return_addr   = token2; }
  1220     if(_curchar != ';') {  // check for semi-colon
  1221       parse_err(SYNERR, "missing %c in return address entry.\n", ';');
  1222       return;
  1224     next_char();           // skip the semi-colon
  1226   else {
  1227     parse_err(SYNERR, "Missing %c in return_address entry.\n", '(');
  1231 //------------------------------preserve_stack_parse---------------------------
  1232 void ADLParser::preserve_stack_parse(FrameForm *frame) {
  1233   if(_curchar == '(') {
  1234     char *token = get_paren_expr("preserve_stack_slots");
  1235     frame->_in_preserve_slots   = token;
  1237     if(_curchar != ';') {  // check for semi-colon
  1238       parse_err(SYNERR, "missing %c in preserve stack slot entry.\n", ';');
  1239       return;
  1241     next_char();           // skip the semi-colon
  1243   else {
  1244     parse_err(SYNERR, "Missing %c in preserve stack slot entry.\n", '(');
  1248 //------------------------------calling_convention_parse-----------------------
  1249 char *ADLParser::calling_convention_parse() {
  1250   char   *desc = NULL;          // String representation of calling_convention
  1252   skipws();                     // Skip leading whitespace
  1253   if ( (desc = find_cpp_block("calling convention block")) == NULL ) {
  1254     parse_err(SYNERR, "incorrect or missing block for 'calling_convention'.\n");
  1256   return desc;
  1259 //------------------------------return_value_parse-----------------------------
  1260 char *ADLParser::return_value_parse() {
  1261   char   *desc = NULL;          // String representation of calling_convention
  1263   skipws();                     // Skip leading whitespace
  1264   if ( (desc = find_cpp_block("return value block")) == NULL ) {
  1265     parse_err(SYNERR, "incorrect or missing block for 'return_value'.\n");
  1267   return desc;
  1270 //------------------------------ins_pipe_parse---------------------------------
  1271 void ADLParser::ins_pipe_parse(InstructForm &instr) {
  1272   char * ident;
  1274   skipws();
  1275   if ( _curchar != '(' ) {       // Check for delimiter
  1276     parse_err(SYNERR, "missing \"(\" in ins_pipe definition\n");
  1277     return;
  1280   next_char();
  1281   ident = get_ident();           // Grab next identifier
  1283   if (ident == NULL) {
  1284     parse_err(SYNERR, "keyword identifier expected at %c\n", _curchar);
  1285     return;
  1288   skipws();
  1289   if ( _curchar != ')' ) {       // Check for delimiter
  1290     parse_err(SYNERR, "missing \")\" in ins_pipe definition\n");
  1291     return;
  1294   next_char();                   // skip the close paren
  1295   if(_curchar != ';') {          // check for semi-colon
  1296     parse_err(SYNERR, "missing %c in return value entry.\n", ';');
  1297     return;
  1299   next_char();                   // skip the semi-colon
  1301   // Check ident for validity
  1302   if (_AD._pipeline && !_AD._pipeline->_classlist.search(ident)) {
  1303     parse_err(SYNERR, "\"%s\" is not a valid pipeline class\n", ident);
  1304     return;
  1307   // Add this instruction to the list in the pipeline class
  1308   _AD._pipeline->_classdict[ident]->is_pipeclass()->_instructs.addName(instr._ident);
  1310   // Set the name of the pipeline class in the instruction
  1311   instr._ins_pipe = ident;
  1312   return;
  1315 //------------------------------pipe_parse-------------------------------------
  1316 void ADLParser::pipe_parse(void) {
  1317   PipelineForm *pipeline;         // Encode class for instruction/operand
  1318   char * ident;
  1320   pipeline = new PipelineForm();  // Build new Source object
  1321   _AD.addForm(pipeline);
  1323   skipws();                       // Skip leading whitespace
  1324   // Check for block delimiter
  1325   if ( (_curchar != '%')
  1326        || ( next_char(),  (_curchar != '{')) ) {
  1327     parse_err(SYNERR, "missing '%%{' in pipeline definition\n");
  1328     return;
  1330   next_char();                     // Maintain the invariant
  1331   do {
  1332     ident = get_ident();           // Grab next identifier
  1333     if (ident == NULL) {
  1334       parse_err(SYNERR, "keyword identifier expected at %c\n", _curchar);
  1335       continue;
  1337     if      (!strcmp(ident, "resources" )) resource_parse(*pipeline);
  1338     else if (!strcmp(ident, "pipe_desc" )) pipe_desc_parse(*pipeline);
  1339     else if (!strcmp(ident, "pipe_class")) pipe_class_parse(*pipeline);
  1340     else if (!strcmp(ident, "define")) {
  1341       skipws();
  1342       if ( (_curchar != '%')
  1343            || ( next_char(),  (_curchar != '{')) ) {
  1344         parse_err(SYNERR, "expected '%%{'\n");
  1345         return;
  1347       next_char(); skipws();
  1349       char *node_class = get_ident();
  1350       if (node_class == NULL) {
  1351         parse_err(SYNERR, "expected identifier, found \"%c\"\n", _curchar);
  1352         return;
  1355       skipws();
  1356       if (_curchar != ',' && _curchar != '=') {
  1357         parse_err(SYNERR, "expected `=`, found '%c'\n", _curchar);
  1358         break;
  1360       next_char(); skipws();
  1362       char *pipe_class = get_ident();
  1363       if (pipe_class == NULL) {
  1364         parse_err(SYNERR, "expected identifier, found \"%c\"\n", _curchar);
  1365         return;
  1367       if (_curchar != ';' ) {
  1368         parse_err(SYNERR, "expected `;`, found '%c'\n", _curchar);
  1369         break;
  1371       next_char();              // Skip over semi-colon
  1373       skipws();
  1374       if ( (_curchar != '%')
  1375            || ( next_char(),  (_curchar != '}')) ) {
  1376         parse_err(SYNERR, "expected '%%}', found \"%c\"\n", _curchar);
  1378       next_char();
  1380       // Check ident for validity
  1381       if (_AD._pipeline && !_AD._pipeline->_classlist.search(pipe_class)) {
  1382         parse_err(SYNERR, "\"%s\" is not a valid pipeline class\n", pipe_class);
  1383         return;
  1386       // Add this machine node to the list in the pipeline class
  1387       _AD._pipeline->_classdict[pipe_class]->is_pipeclass()->_instructs.addName(node_class);
  1389       MachNodeForm *machnode = new MachNodeForm(node_class); // Create new machnode form
  1390       machnode->_machnode_pipe = pipe_class;
  1392       _AD.addForm(machnode);
  1394     else if (!strcmp(ident, "attributes")) {
  1395       bool vsi_seen = false;
  1397       skipws();
  1398       if ( (_curchar != '%')
  1399            || ( next_char(),  (_curchar != '{')) ) {
  1400         parse_err(SYNERR, "expected '%%{'\n");
  1401         return;
  1403       next_char(); skipws();
  1405       while (_curchar != '%') {
  1406         ident = get_ident();
  1407         if (ident == NULL)
  1408           break;
  1410         if (!strcmp(ident, "variable_size_instructions")) {
  1411           skipws();
  1412           if (_curchar == ';') {
  1413             next_char(); skipws();
  1416           pipeline->_variableSizeInstrs = true;
  1417           vsi_seen = true;
  1418           continue;
  1421         if (!strcmp(ident, "fixed_size_instructions")) {
  1422           skipws();
  1423           if (_curchar == ';') {
  1424             next_char(); skipws();
  1427           pipeline->_variableSizeInstrs = false;
  1428           vsi_seen = true;
  1429           continue;
  1432         if (!strcmp(ident, "branch_has_delay_slot")) {
  1433           skipws();
  1434           if (_curchar == ';') {
  1435             next_char(); skipws();
  1438           pipeline->_branchHasDelaySlot = true;
  1439           continue;
  1442         if (!strcmp(ident, "max_instructions_per_bundle")) {
  1443           skipws();
  1444           if (_curchar != '=') {
  1445             parse_err(SYNERR, "expected `=`\n");
  1446             break;
  1449           next_char(); skipws();
  1450           pipeline->_maxInstrsPerBundle = get_int();
  1451           skipws();
  1453           if (_curchar == ';') {
  1454             next_char(); skipws();
  1457           continue;
  1460         if (!strcmp(ident, "max_bundles_per_cycle")) {
  1461           skipws();
  1462           if (_curchar != '=') {
  1463             parse_err(SYNERR, "expected `=`\n");
  1464             break;
  1467           next_char(); skipws();
  1468           pipeline->_maxBundlesPerCycle = get_int();
  1469           skipws();
  1471           if (_curchar == ';') {
  1472             next_char(); skipws();
  1475           continue;
  1478         if (!strcmp(ident, "instruction_unit_size")) {
  1479           skipws();
  1480           if (_curchar != '=') {
  1481             parse_err(SYNERR, "expected `=`, found '%c'\n", _curchar);
  1482             break;
  1485           next_char(); skipws();
  1486           pipeline->_instrUnitSize = get_int();
  1487           skipws();
  1489           if (_curchar == ';') {
  1490             next_char(); skipws();
  1493           continue;
  1496         if (!strcmp(ident, "bundle_unit_size")) {
  1497           skipws();
  1498           if (_curchar != '=') {
  1499             parse_err(SYNERR, "expected `=`, found '%c'\n", _curchar);
  1500             break;
  1503           next_char(); skipws();
  1504           pipeline->_bundleUnitSize = get_int();
  1505           skipws();
  1507           if (_curchar == ';') {
  1508             next_char(); skipws();
  1511           continue;
  1514         if (!strcmp(ident, "instruction_fetch_unit_size")) {
  1515           skipws();
  1516           if (_curchar != '=') {
  1517             parse_err(SYNERR, "expected `=`, found '%c'\n", _curchar);
  1518             break;
  1521           next_char(); skipws();
  1522           pipeline->_instrFetchUnitSize = get_int();
  1523           skipws();
  1525           if (_curchar == ';') {
  1526             next_char(); skipws();
  1529           continue;
  1532         if (!strcmp(ident, "instruction_fetch_units")) {
  1533           skipws();
  1534           if (_curchar != '=') {
  1535             parse_err(SYNERR, "expected `=`, found '%c'\n", _curchar);
  1536             break;
  1539           next_char(); skipws();
  1540           pipeline->_instrFetchUnits = get_int();
  1541           skipws();
  1543           if (_curchar == ';') {
  1544             next_char(); skipws();
  1547           continue;
  1550         if (!strcmp(ident, "nops")) {
  1551           skipws();
  1552           if (_curchar != '(') {
  1553             parse_err(SYNERR, "expected `(`, found '%c'\n", _curchar);
  1554             break;
  1557           next_char(); skipws();
  1559           while (_curchar != ')') {
  1560             ident = get_ident();
  1561             if (ident == NULL) {
  1562               parse_err(SYNERR, "expected identifier for nop instruction, found '%c'\n", _curchar);
  1563               break;
  1566             pipeline->_noplist.addName(ident);
  1567             pipeline->_nopcnt++;
  1568             skipws();
  1570             if (_curchar == ',') {
  1571               next_char(); skipws();
  1575           next_char(); skipws();
  1577           if (_curchar == ';') {
  1578             next_char(); skipws();
  1581           continue;
  1584         parse_err(SYNERR, "unknown specifier \"%s\"\n", ident);
  1587       if ( (_curchar != '%')
  1588            || ( next_char(),  (_curchar != '}')) ) {
  1589         parse_err(SYNERR, "expected '%%}', found \"%c\"\n", _curchar);
  1591       next_char(); skipws();
  1593       if (pipeline->_maxInstrsPerBundle == 0)
  1594         parse_err(SYNERR, "\"max_instructions_per_bundle\" unspecified\n");
  1595       if (pipeline->_instrUnitSize == 0 && pipeline->_bundleUnitSize == 0)
  1596         parse_err(SYNERR, "\"instruction_unit_size\" and \"bundle_unit_size\" unspecified\n");
  1597       if (pipeline->_instrFetchUnitSize == 0)
  1598         parse_err(SYNERR, "\"instruction_fetch_unit_size\" unspecified\n");
  1599       if (pipeline->_instrFetchUnits == 0)
  1600         parse_err(SYNERR, "\"instruction_fetch_units\" unspecified\n");
  1601       if (!vsi_seen)
  1602         parse_err(SYNERR, "\"variable_size_instruction\" or \"fixed_size_instruction\" unspecified\n");
  1604     else {  // Done with staticly defined parts of instruction definition
  1605       parse_err(SYNERR, "expected one of \"resources\", \"pipe_desc\", \"pipe_class\", found \"%s\"\n", ident);
  1606       return;
  1608     skipws();
  1609     if (_curchar == ';')
  1610       skipws();
  1611   } while(_curchar != '%');
  1613   next_char();
  1614   if (_curchar != '}') {
  1615     parse_err(SYNERR, "missing \"%%}\" in pipeline definition\n");
  1616     return;
  1619   next_char();
  1622 //------------------------------resource_parse----------------------------
  1623 void ADLParser::resource_parse(PipelineForm &pipeline) {
  1624   ResourceForm *resource;
  1625   char * ident;
  1626   char * expr;
  1627   unsigned mask;
  1628   pipeline._rescount = 0;
  1630   skipws();                       // Skip leading whitespace
  1632   if (_curchar != '(') {
  1633     parse_err(SYNERR, "missing \"(\" in resource definition\n");
  1634     return;
  1637   do {
  1638     next_char();                   // Skip "(" or ","
  1639     ident = get_ident();           // Grab next identifier
  1641     if (_AD._adl_debug > 1) {
  1642       if (ident != NULL) {
  1643         fprintf(stderr, "resource_parse: identifier: %s\n", ident);
  1647     if (ident == NULL) {
  1648       parse_err(SYNERR, "keyword identifier expected at \"%c\"\n", _curchar);
  1649       return;
  1651     skipws();
  1653     if (_curchar != '=') {
  1654       mask = (1 << pipeline._rescount++);
  1656     else {
  1657       next_char(); skipws();
  1658       expr = get_ident();          // Grab next identifier
  1659       if (expr == NULL) {
  1660         parse_err(SYNERR, "keyword identifier expected at \"%c\"\n", _curchar);
  1661         return;
  1663       resource = (ResourceForm *) pipeline._resdict[expr];
  1664       if (resource == NULL) {
  1665         parse_err(SYNERR, "resource \"%s\" is not defined\n", expr);
  1666         return;
  1668       mask = resource->mask();
  1670       skipws();
  1671       while (_curchar == '|') {
  1672         next_char(); skipws();
  1674         expr = get_ident();          // Grab next identifier
  1675         if (expr == NULL) {
  1676           parse_err(SYNERR, "keyword identifier expected at \"%c\"\n", _curchar);
  1677           return;
  1680         resource = (ResourceForm *) pipeline._resdict[expr];   // Look up the value
  1681         if (resource == NULL) {
  1682           parse_err(SYNERR, "resource \"%s\" is not defined\n", expr);
  1683           return;
  1686         mask |= resource->mask();
  1687         skipws();
  1691     resource = new ResourceForm(mask);
  1693     pipeline._resdict.Insert(ident, resource);
  1694     pipeline._reslist.addName(ident);
  1695   } while (_curchar == ',');
  1697   if (_curchar != ')') {
  1698       parse_err(SYNERR, "\")\" expected at \"%c\"\n", _curchar);
  1699       return;
  1702   next_char();                 // Skip ")"
  1703   if (_curchar == ';')
  1704     next_char();               // Skip ";"
  1707 //------------------------------resource_parse----------------------------
  1708 void ADLParser::pipe_desc_parse(PipelineForm &pipeline) {
  1709   char * ident;
  1711   skipws();                       // Skip leading whitespace
  1713   if (_curchar != '(') {
  1714     parse_err(SYNERR, "missing \"(\" in pipe_desc definition\n");
  1715     return;
  1718   do {
  1719     next_char();                   // Skip "(" or ","
  1720     ident = get_ident();           // Grab next identifier
  1721     if (ident == NULL) {
  1722       parse_err(SYNERR, "keyword identifier expected at \"%c\"\n", _curchar);
  1723       return;
  1726     // Add the name to the list
  1727     pipeline._stages.addName(ident);
  1728     pipeline._stagecnt++;
  1730     skipws();
  1731   } while (_curchar == ',');
  1733   if (_curchar != ')') {
  1734       parse_err(SYNERR, "\")\" expected at \"%c\"\n", _curchar);
  1735       return;
  1738   next_char();                     // Skip ")"
  1739   if (_curchar == ';')
  1740     next_char();                   // Skip ";"
  1743 //------------------------------pipe_class_parse--------------------------
  1744 void ADLParser::pipe_class_parse(PipelineForm &pipeline) {
  1745   PipeClassForm *pipe_class;
  1746   char * ident;
  1747   char * stage;
  1748   char * read_or_write;
  1749   int is_write;
  1750   int is_read;
  1751   OperandForm  *oper;
  1753   skipws();                       // Skip leading whitespace
  1755   ident = get_ident();            // Grab next identifier
  1757   if (ident == NULL) {
  1758     parse_err(SYNERR, "keyword identifier expected at \"%c\"\n", _curchar);
  1759     return;
  1762   // Create a record for the pipe_class
  1763   pipe_class = new PipeClassForm(ident, ++pipeline._classcnt);
  1764   pipeline._classdict.Insert(ident, pipe_class);
  1765   pipeline._classlist.addName(ident);
  1767   // Then get the operands
  1768   skipws();
  1769   if (_curchar != '(') {
  1770     parse_err(SYNERR, "missing \"(\" in pipe_class definition\n");
  1772   // Parse the operand list
  1773   else get_oplist(pipe_class->_parameters, pipe_class->_localNames);
  1774   skipws();                        // Skip leading whitespace
  1775   // Check for block delimiter
  1776   if ( (_curchar != '%')
  1777        || ( next_char(),  (_curchar != '{')) ) {
  1778     parse_err(SYNERR, "missing \"%%{\" in pipe_class definition\n");
  1779     return;
  1781   next_char();
  1783   do {
  1784     ident = get_ident();           // Grab next identifier
  1785     if (ident == NULL) {
  1786       parse_err(SYNERR, "keyword identifier expected at \"%c\"\n", _curchar);
  1787       continue;
  1789     skipws();
  1791     if (!strcmp(ident, "fixed_latency")) {
  1792       skipws();
  1793       if (_curchar != '(') {
  1794         parse_err(SYNERR, "missing \"(\" in latency definition\n");
  1795         return;
  1797       next_char(); skipws();
  1798       if( !isdigit(_curchar) ) {
  1799         parse_err(SYNERR, "number expected for \"%c\" in latency definition\n", _curchar);
  1800         return;
  1802       int fixed_latency = get_int();
  1803       skipws();
  1804       if (_curchar != ')') {
  1805         parse_err(SYNERR, "missing \")\" in latency definition\n");
  1806         return;
  1808       next_char(); skipws();
  1809       if (_curchar != ';') {
  1810         parse_err(SYNERR, "missing \";\" in latency definition\n");
  1811         return;
  1814       pipe_class->setFixedLatency(fixed_latency);
  1815       next_char(); skipws();
  1816       continue;
  1819     if (!strcmp(ident, "zero_instructions") ||
  1820         !strcmp(ident, "no_instructions")) {
  1821       skipws();
  1822       if (_curchar != ';') {
  1823         parse_err(SYNERR, "missing \";\" in latency definition\n");
  1824         return;
  1827       pipe_class->setInstructionCount(0);
  1828       next_char(); skipws();
  1829       continue;
  1832     if (!strcmp(ident, "one_instruction_with_delay_slot") ||
  1833         !strcmp(ident, "single_instruction_with_delay_slot")) {
  1834       skipws();
  1835       if (_curchar != ';') {
  1836         parse_err(SYNERR, "missing \";\" in latency definition\n");
  1837         return;
  1840       pipe_class->setInstructionCount(1);
  1841       pipe_class->setBranchDelay(true);
  1842       next_char(); skipws();
  1843       continue;
  1846     if (!strcmp(ident, "one_instruction") ||
  1847         !strcmp(ident, "single_instruction")) {
  1848       skipws();
  1849       if (_curchar != ';') {
  1850         parse_err(SYNERR, "missing \";\" in latency definition\n");
  1851         return;
  1854       pipe_class->setInstructionCount(1);
  1855       next_char(); skipws();
  1856       continue;
  1859     if (!strcmp(ident, "instructions_in_first_bundle") ||
  1860         !strcmp(ident, "instruction_count")) {
  1861       skipws();
  1863       int number_of_instructions = 1;
  1865       if (_curchar != '(') {
  1866         parse_err(SYNERR, "\"(\" expected at \"%c\"\n", _curchar);
  1867         continue;
  1870       next_char(); skipws();
  1871       number_of_instructions = get_int();
  1873       skipws();
  1874       if (_curchar != ')') {
  1875         parse_err(SYNERR, "\")\" expected at \"%c\"\n", _curchar);
  1876         continue;
  1879       next_char(); skipws();
  1880       if (_curchar != ';') {
  1881         parse_err(SYNERR, "missing \";\" in latency definition\n");
  1882         return;
  1885       pipe_class->setInstructionCount(number_of_instructions);
  1886       next_char(); skipws();
  1887       continue;
  1890     if (!strcmp(ident, "multiple_bundles")) {
  1891       skipws();
  1892       if (_curchar != ';') {
  1893         parse_err(SYNERR, "missing \";\" after multiple bundles\n");
  1894         return;
  1897       pipe_class->setMultipleBundles(true);
  1898       next_char(); skipws();
  1899       continue;
  1902     if (!strcmp(ident, "has_delay_slot")) {
  1903       skipws();
  1904       if (_curchar != ';') {
  1905         parse_err(SYNERR, "missing \";\" after \"has_delay_slot\"\n");
  1906         return;
  1909       pipe_class->setBranchDelay(true);
  1910       next_char(); skipws();
  1911       continue;
  1914     if (!strcmp(ident, "force_serialization")) {
  1915       skipws();
  1916       if (_curchar != ';') {
  1917         parse_err(SYNERR, "missing \";\" after \"force_serialization\"\n");
  1918         return;
  1921       pipe_class->setForceSerialization(true);
  1922       next_char(); skipws();
  1923       continue;
  1926     if (!strcmp(ident, "may_have_no_code")) {
  1927       skipws();
  1928       if (_curchar != ';') {
  1929         parse_err(SYNERR, "missing \";\" after \"may_have_no_code\"\n");
  1930         return;
  1933       pipe_class->setMayHaveNoCode(true);
  1934       next_char(); skipws();
  1935       continue;
  1938     const Form *parm = pipe_class->_localNames[ident];
  1939     if (parm != NULL) {
  1940       oper = parm->is_operand();
  1941       if (oper == NULL && !parm->is_opclass()) {
  1942         parse_err(SYNERR, "operand name expected at %s\n", ident);
  1943         continue;
  1946       if (_curchar != ':') {
  1947         parse_err(SYNERR, "\":\" expected at \"%c\"\n", _curchar);
  1948         continue;
  1950       next_char(); skipws();
  1951       stage = get_ident();
  1952       if (stage == NULL) {
  1953         parse_err(SYNERR, "pipeline stage identifier expected at \"%c\"\n", _curchar);
  1954         continue;
  1957       skipws();
  1958       if (_curchar != '(') {
  1959         parse_err(SYNERR, "\"(\" expected at \"%c\"\n", _curchar);
  1960         continue;
  1963       next_char();
  1964       read_or_write = get_ident();
  1965       if (read_or_write == NULL) {
  1966         parse_err(SYNERR, "\"read\" or \"write\" expected at \"%c\"\n", _curchar);
  1967         continue;
  1970       is_read  = strcmp(read_or_write, "read")   == 0;
  1971       is_write = strcmp(read_or_write, "write")  == 0;
  1972       if (!is_read && !is_write) {
  1973         parse_err(SYNERR, "\"read\" or \"write\" expected at \"%c\"\n", _curchar);
  1974         continue;
  1977       skipws();
  1978       if (_curchar != ')') {
  1979         parse_err(SYNERR, "\")\" expected at \"%c\"\n", _curchar);
  1980         continue;
  1983       next_char(); skipws();
  1984       int more_instrs = 0;
  1985       if (_curchar == '+') {
  1986           next_char(); skipws();
  1987           if (_curchar < '0' || _curchar > '9') {
  1988             parse_err(SYNERR, "<number> expected at \"%c\"\n", _curchar);
  1989             continue;
  1991           while (_curchar >= '0' && _curchar <= '9') {
  1992             more_instrs *= 10;
  1993             more_instrs += _curchar - '0';
  1994             next_char();
  1996           skipws();
  1999       PipeClassOperandForm *pipe_operand = new PipeClassOperandForm(stage, is_write, more_instrs);
  2000       pipe_class->_localUsage.Insert(ident, pipe_operand);
  2002       if (_curchar == '%')
  2003           continue;
  2005       if (_curchar != ';') {
  2006         parse_err(SYNERR, "\";\" expected at \"%c\"\n", _curchar);
  2007         continue;
  2009       next_char(); skipws();
  2010       continue;
  2013     // Scan for Resource Specifier
  2014     const Form *res = pipeline._resdict[ident];
  2015     if (res != NULL) {
  2016       int cyclecnt = 1;
  2017       if (_curchar != ':') {
  2018         parse_err(SYNERR, "\":\" expected at \"%c\"\n", _curchar);
  2019         continue;
  2021       next_char(); skipws();
  2022       stage = get_ident();
  2023       if (stage == NULL) {
  2024         parse_err(SYNERR, "pipeline stage identifier expected at \"%c\"\n", _curchar);
  2025         continue;
  2028       skipws();
  2029       if (_curchar == '(') {
  2030         next_char();
  2031         cyclecnt = get_int();
  2033         skipws();
  2034         if (_curchar != ')') {
  2035           parse_err(SYNERR, "\")\" expected at \"%c\"\n", _curchar);
  2036           continue;
  2039         next_char(); skipws();
  2042       PipeClassResourceForm *resource = new PipeClassResourceForm(ident, stage, cyclecnt);
  2043       int stagenum = pipeline._stages.index(stage);
  2044       if (pipeline._maxcycleused < (stagenum+cyclecnt))
  2045         pipeline._maxcycleused = (stagenum+cyclecnt);
  2046       pipe_class->_resUsage.addForm(resource);
  2048       if (_curchar == '%')
  2049           continue;
  2051       if (_curchar != ';') {
  2052         parse_err(SYNERR, "\";\" expected at \"%c\"\n", _curchar);
  2053         continue;
  2055       next_char(); skipws();
  2056       continue;
  2059     parse_err(SYNERR, "resource expected at \"%s\"\n", ident);
  2060     return;
  2061   } while(_curchar != '%');
  2063   next_char();
  2064   if (_curchar != '}') {
  2065     parse_err(SYNERR, "missing \"%%}\" in pipe_class definition\n");
  2066     return;
  2069   next_char();
  2072 //------------------------------peep_parse-------------------------------------
  2073 void ADLParser::peep_parse(void) {
  2074   Peephole  *peep;                // Pointer to current peephole rule form
  2075   char      *desc = NULL;         // String representation of rule
  2077   skipws();                       // Skip leading whitespace
  2079   peep = new Peephole();          // Build new Peephole object
  2080   // Check for open block sequence
  2081   skipws();                       // Skip leading whitespace
  2082   if (_curchar == '%' && *(_ptr+1) == '{') {
  2083     next_char(); next_char();     // Skip "%{"
  2084     skipws();
  2085     while (_curchar != '%' && *(_ptr+1) != '}') {
  2086       char *token = get_ident();
  2087       if (token == NULL) {
  2088         parse_err(SYNERR, "missing identifier inside peephole rule.\n");
  2089         return;
  2091       // check for legal subsections of peephole rule
  2092       if (strcmp(token,"peepmatch")==0) {
  2093         peep_match_parse(*peep); }
  2094       else if (strcmp(token,"peepconstraint")==0) {
  2095         peep_constraint_parse(*peep); }
  2096       else if (strcmp(token,"peepreplace")==0) {
  2097         peep_replace_parse(*peep); }
  2098       else {
  2099         parse_err(SYNERR, "expected peepmatch, peepconstraint, or peepreplace for identifier %s.\n", token);
  2101       skipws();
  2104   else {
  2105     parse_err(SYNERR, "Missing %%{ ... %%} block after peephole keyword.\n");
  2106     return;
  2108   next_char();                    // Skip past '%'
  2109   next_char();                    // Skip past '}'
  2112 // ******************** Private Level 2 Parse Functions ********************
  2113 //------------------------------constraint_parse------------------------------
  2114 Constraint *ADLParser::constraint_parse(void) {
  2115   char *func;
  2116   char *arg;
  2118   // Check for constraint expression
  2119   skipws();
  2120   if (_curchar != '(') {
  2121     parse_err(SYNERR, "missing constraint expression, (...)\n");
  2122     return NULL;
  2124   next_char();                    // Skip past '('
  2126   // Get constraint function
  2127   skipws();
  2128   func = get_ident();
  2129   if (func == NULL) {
  2130     parse_err(SYNERR, "missing function in constraint expression.\n");
  2131     return NULL;
  2133   if (strcmp(func,"ALLOC_IN_RC")==0
  2134       || strcmp(func,"IS_R_CLASS")==0) {
  2135     // Check for '(' before argument
  2136     skipws();
  2137     if (_curchar != '(') {
  2138       parse_err(SYNERR, "missing '(' for constraint function's argument.\n");
  2139       return NULL;
  2141     next_char();
  2143     // Get it's argument
  2144     skipws();
  2145     arg = get_ident();
  2146     if (arg == NULL) {
  2147       parse_err(SYNERR, "missing argument for constraint function %s\n",func);
  2148       return NULL;
  2150     // Check for ')' after argument
  2151     skipws();
  2152     if (_curchar != ')') {
  2153       parse_err(SYNERR, "missing ')' after constraint function argument %s\n",arg);
  2154       return NULL;
  2156     next_char();
  2157   } else {
  2158     parse_err(SYNERR, "Invalid constraint function %s\n",func);
  2159     return NULL;
  2162   // Check for closing paren and ';'
  2163   skipws();
  2164   if (_curchar != ')') {
  2165     parse_err(SYNERR, "Missing ')' for constraint function %s\n",func);
  2166     return NULL;
  2168   next_char();
  2169   skipws();
  2170   if (_curchar != ';') {
  2171     parse_err(SYNERR, "Missing ';' after constraint.\n");
  2172     return NULL;
  2174   next_char();
  2176   // Create new "Constraint"
  2177   Constraint *constraint = new Constraint(func,arg);
  2178   return constraint;
  2181 //------------------------------constr_parse-----------------------------------
  2182 ConstructRule *ADLParser::construct_parse(void) {
  2183   return NULL;
  2187 //------------------------------reg_def_parse----------------------------------
  2188 void ADLParser::reg_def_parse(void) {
  2189   char *rname;                   // Name of register being defined
  2191   // Get register name
  2192   skipws();                      // Skip whitespace
  2193   rname = get_ident();
  2194   if (rname == NULL) {
  2195     parse_err(SYNERR, "missing register name after reg_def\n");
  2196     return;
  2199   // Check for definition of register calling convention (save on call, ...),
  2200   // register save type, and register encoding value.
  2201   skipws();
  2202   char *callconv  = NULL;
  2203   char *c_conv    = NULL;
  2204   char *idealtype = NULL;
  2205   char *encoding  = NULL;
  2206   char *concrete = NULL;
  2207   if (_curchar == '(') {
  2208     next_char();
  2209     callconv = get_ident();
  2210     // Parse the internal calling convention, must be NS, SOC, SOE, or AS.
  2211     if (callconv == NULL) {
  2212       parse_err(SYNERR, "missing register calling convention value\n");
  2213       return;
  2215     if(strcmp(callconv, "SOC") && strcmp(callconv,"SOE") &&
  2216        strcmp(callconv, "NS") && strcmp(callconv, "AS")) {
  2217       parse_err(SYNERR, "invalid value for register calling convention\n");
  2219     skipws();
  2220     if (_curchar != ',') {
  2221       parse_err(SYNERR, "missing comma in register definition statement\n");
  2222       return;
  2224     next_char();
  2226     // Parse the native calling convention, must be NS, SOC, SOE, AS
  2227     c_conv = get_ident();
  2228     if (c_conv == NULL) {
  2229       parse_err(SYNERR, "missing register native calling convention value\n");
  2230       return;
  2232     if(strcmp(c_conv, "SOC") && strcmp(c_conv,"SOE") &&
  2233        strcmp(c_conv, "NS") && strcmp(c_conv, "AS")) {
  2234       parse_err(SYNERR, "invalid value for register calling convention\n");
  2236     skipws();
  2237     if (_curchar != ',') {
  2238       parse_err(SYNERR, "missing comma in register definition statement\n");
  2239       return;
  2241     next_char();
  2242     skipws();
  2244     // Parse the ideal save type
  2245     idealtype = get_ident();
  2246     if (idealtype == NULL) {
  2247       parse_err(SYNERR, "missing register save type value\n");
  2248       return;
  2250     skipws();
  2251     if (_curchar != ',') {
  2252       parse_err(SYNERR, "missing comma in register definition statement\n");
  2253       return;
  2255     next_char();
  2256     skipws();
  2258     // Parse the encoding value
  2259     encoding = get_expr("encoding", ",");
  2260     if (encoding == NULL) {
  2261       parse_err(SYNERR, "missing register encoding value\n");
  2262       return;
  2264     trim(encoding);
  2265     if (_curchar != ',') {
  2266       parse_err(SYNERR, "missing comma in register definition statement\n");
  2267       return;
  2269     next_char();
  2270     skipws();
  2271     // Parse the concrete name type
  2272     // concrete = get_ident();
  2273     concrete = get_expr("concrete", ")");
  2274     if (concrete == NULL) {
  2275       parse_err(SYNERR, "missing vm register name value\n");
  2276       return;
  2279     if (_curchar != ')') {
  2280       parse_err(SYNERR, "missing ')' in register definition statement\n");
  2281       return;
  2283     next_char();
  2286   // Check for closing ';'
  2287   skipws();
  2288   if (_curchar != ';') {
  2289     parse_err(SYNERR, "missing ';' after reg_def\n");
  2290     return;
  2292   next_char();                   // move past ';'
  2294   // Debug Stuff
  2295   if (_AD._adl_debug > 1) {
  2296     fprintf(stderr,"Register Definition: %s ( %s, %s %s )\n", rname,
  2297             (callconv ? callconv : ""), (c_conv ? c_conv : ""), concrete);
  2300   // Record new register definition.
  2301   _AD._register->addRegDef(rname, callconv, c_conv, idealtype, encoding, concrete);
  2302   return;
  2305 //------------------------------reg_class_parse--------------------------------
  2306 void ADLParser::reg_class_parse(void) {
  2307   char *cname;                    // Name of register class being defined
  2309   // Get register class name
  2310   skipws();                       // Skip leading whitespace
  2311   cname = get_ident();
  2312   if (cname == NULL) {
  2313     parse_err(SYNERR, "missing register class name after 'reg_class'\n");
  2314     return;
  2316   // Debug Stuff
  2317   if (_AD._adl_debug >1) fprintf(stderr,"Register Class: %s\n", cname);
  2319   RegClass *reg_class = _AD._register->addRegClass(cname);
  2321   // Collect registers in class
  2322   skipws();
  2323   if (_curchar == '(') {
  2324     next_char();                  // Skip '('
  2325     skipws();
  2326     while (_curchar != ')') {
  2327       char *rname = get_ident();
  2328       if (rname==NULL) {
  2329         parse_err(SYNERR, "missing identifier inside reg_class list.\n");
  2330         return;
  2332       RegDef *regDef = _AD._register->getRegDef(rname);
  2333       if (!regDef) {
  2334         parse_err(SEMERR, "unknown identifier %s inside reg_class list.\n", rname);
  2335       } else {
  2336         reg_class->addReg(regDef); // add regDef to regClass
  2339       // Check for ',' and position to next token.
  2340       skipws();
  2341       if (_curchar == ',') {
  2342         next_char();              // Skip trailing ','
  2343         skipws();
  2346     next_char();                  // Skip closing ')'
  2347   } else if (_curchar == '%') {
  2348     char *code = find_cpp_block("reg class");
  2349     if (code == NULL) {
  2350       parse_err(SYNERR, "missing code declaration for reg class.\n");
  2351       return;
  2353     reg_class->_user_defined = code;
  2354     return;
  2357   // Check for terminating ';'
  2358   skipws();
  2359   if (_curchar != ';') {
  2360     parse_err(SYNERR, "missing ';' at end of reg_class definition.\n");
  2361     return;
  2363   next_char();                    // Skip trailing ';'
  2365   // Check RegClass size, must be <= 32 registers in class.
  2367   return;
  2370 //------------------------------alloc_class_parse------------------------------
  2371 void ADLParser::alloc_class_parse(void) {
  2372   char *name;                     // Name of allocation class being defined
  2374   // Get allocation class name
  2375   skipws();                       // Skip leading whitespace
  2376   name = get_ident();
  2377   if (name == NULL) {
  2378     parse_err(SYNERR, "missing allocation class name after 'reg_class'\n");
  2379     return;
  2381   // Debug Stuff
  2382   if (_AD._adl_debug >1) fprintf(stderr,"Allocation Class: %s\n", name);
  2384   AllocClass *alloc_class = _AD._register->addAllocClass(name);
  2386   // Collect registers in class
  2387   skipws();
  2388   if (_curchar == '(') {
  2389     next_char();                  // Skip '('
  2390     skipws();
  2391     while (_curchar != ')') {
  2392       char *rname = get_ident();
  2393       if (rname==NULL) {
  2394         parse_err(SYNERR, "missing identifier inside reg_class list.\n");
  2395         return;
  2397       // Check if name is a RegDef
  2398       RegDef *regDef = _AD._register->getRegDef(rname);
  2399       if (regDef) {
  2400         alloc_class->addReg(regDef);   // add regDef to allocClass
  2401       } else {
  2403         // name must be a RegDef or a RegClass
  2404         parse_err(SYNERR, "name %s should be a previously defined reg_def.\n", rname);
  2405         return;
  2408       // Check for ',' and position to next token.
  2409       skipws();
  2410       if (_curchar == ',') {
  2411         next_char();              // Skip trailing ','
  2412         skipws();
  2415     next_char();                  // Skip closing ')'
  2418   // Check for terminating ';'
  2419   skipws();
  2420   if (_curchar != ';') {
  2421     parse_err(SYNERR, "missing ';' at end of reg_class definition.\n");
  2422     return;
  2424   next_char();                    // Skip trailing ';'
  2426   return;
  2429 //------------------------------peep_match_child_parse-------------------------
  2430 InstructForm *ADLParser::peep_match_child_parse(PeepMatch &match, int parent, int &position, int input){
  2431   char      *token  = NULL;
  2432   int        lparen = 0;          // keep track of parenthesis nesting depth
  2433   int        rparen = 0;          // position of instruction at this depth
  2434   InstructForm *inst_seen  = NULL;
  2436   // Walk the match tree,
  2437   // Record <parent, position, instruction name, input position>
  2438   while ( lparen >= rparen ) {
  2439     skipws();
  2440     // Left paren signals start of an input, collect with recursive call
  2441     if (_curchar == '(') {
  2442       ++lparen;
  2443       next_char();
  2444       ( void ) peep_match_child_parse(match, parent, position, rparen);
  2446     // Right paren signals end of an input, may be more
  2447     else if (_curchar == ')') {
  2448       ++rparen;
  2449       if( rparen == lparen ) { // IF rparen matches an lparen I've seen
  2450         next_char();           //    move past ')'
  2451       } else {                 // ELSE leave ')' for parent
  2452         assert( rparen == lparen + 1, "Should only see one extra ')'");
  2453         // if an instruction was not specified for this paren-pair
  2454         if( ! inst_seen ) {   // record signal entry
  2455           match.add_instruction( parent, position, NameList::_signal, input );
  2456           ++position;
  2458         // ++input;   // TEMPORARY
  2459         return inst_seen;
  2462     // if no parens, then check for instruction name
  2463     // This instruction is the parent of a sub-tree
  2464     else if ((token = get_ident_dup()) != NULL) {
  2465       const Form *form = _AD._globalNames[token];
  2466       if (form) {
  2467         InstructForm *inst = form->is_instruction();
  2468         // Record the first instruction at this level
  2469         if( inst_seen == NULL ) {
  2470           inst_seen = inst;
  2472         if (inst) {
  2473           match.add_instruction( parent, position, token, input );
  2474           parent = position;
  2475           ++position;
  2476         } else {
  2477           parse_err(SYNERR, "instruction name expected at identifier %s.\n",
  2478                     token);
  2479           return inst_seen;
  2482       else {
  2483         parse_err(SYNERR, "missing identifier in peepmatch rule.\n");
  2484         return NULL;
  2487     else {
  2488       parse_err(SYNERR, "missing identifier in peepmatch rule.\n");
  2489       return NULL;
  2492   } // end while
  2494   assert( false, "ShouldNotReachHere();");
  2495   return NULL;
  2498 //------------------------------peep_match_parse-------------------------------
  2499 // Syntax for a peepmatch rule
  2500 //
  2501 // peepmatch ( root_instr_name [(instruction subtree)] [,(instruction subtree)]* );
  2502 //
  2503 void ADLParser::peep_match_parse(Peephole &peep) {
  2505   skipws();
  2506   // Check the structure of the rule
  2507   // Check for open paren
  2508   if (_curchar != '(') {
  2509     parse_err(SYNERR, "missing '(' at start of peepmatch rule.\n");
  2510     return;
  2512   next_char();   // skip '('
  2514   // Construct PeepMatch and parse the peepmatch rule.
  2515   PeepMatch *match = new PeepMatch(_ptr);
  2516   int  parent   = -1;                   // parent of root
  2517   int  position = 0;                    // zero-based positions
  2518   int  input    = 0;                    // input position in parent's operands
  2519   InstructForm *root= peep_match_child_parse( *match, parent, position, input);
  2520   if( root == NULL ) {
  2521     parse_err(SYNERR, "missing instruction-name at start of peepmatch.\n");
  2522     return;
  2525   if( _curchar != ')' ) {
  2526     parse_err(SYNERR, "missing ')' at end of peepmatch.\n");
  2527     return;
  2529   next_char();   // skip ')'
  2531   // Check for closing semicolon
  2532   skipws();
  2533   if( _curchar != ';' ) {
  2534     parse_err(SYNERR, "missing ';' at end of peepmatch.\n");
  2535     return;
  2537   next_char();   // skip ';'
  2539   // Store match into peep, and store peep into instruction
  2540   peep.add_match(match);
  2541   root->append_peephole(&peep);
  2544 //------------------------------peep_constraint_parse--------------------------
  2545 // Syntax for a peepconstraint rule
  2546 // A parenthesized list of relations between operands in peepmatch subtree
  2547 //
  2548 // peepconstraint %{
  2549 // (instruction_number.operand_name
  2550 //     relational_op
  2551 //  instruction_number.operand_name OR register_name
  2552 //  [, ...] );
  2553 //
  2554 // // instruction numbers are zero-based using topological order in peepmatch
  2555 //
  2556 void ADLParser::peep_constraint_parse(Peephole &peep) {
  2558   skipws();
  2559   // Check the structure of the rule
  2560   // Check for open paren
  2561   if (_curchar != '(') {
  2562     parse_err(SYNERR, "missing '(' at start of peepconstraint rule.\n");
  2563     return;
  2565   else {
  2566     next_char();                  // Skip '('
  2569   // Check for a constraint
  2570   skipws();
  2571   while( _curchar != ')' ) {
  2572     // Get information on the left instruction and its operand
  2573     // left-instructions's number
  2574     int left_inst = get_int();
  2575     // Left-instruction's operand
  2576     skipws();
  2577     if( _curchar != '.' ) {
  2578       parse_err(SYNERR, "missing '.' in peepconstraint after instruction number.\n");
  2579       return;
  2581     next_char();                  // Skip '.'
  2582     char *left_op = get_ident_dup();
  2584     skipws();
  2585     // Collect relational operator
  2586     char *relation = get_relation_dup();
  2588     skipws();
  2589     // Get information on the right instruction and its operand
  2590     int right_inst;        // Right-instructions's number
  2591     if( isdigit(_curchar) ) {
  2592       right_inst = get_int();
  2593       // Right-instruction's operand
  2594       skipws();
  2595       if( _curchar != '.' ) {
  2596         parse_err(SYNERR, "missing '.' in peepconstraint after instruction number.\n");
  2597         return;
  2599       next_char();              // Skip '.'
  2600     } else {
  2601       right_inst = -1;          // Flag as being a register constraint
  2604     char *right_op = get_ident_dup();
  2606     // Construct the next PeepConstraint
  2607     PeepConstraint *constraint = new PeepConstraint( left_inst, left_op,
  2608                                                      relation,
  2609                                                      right_inst, right_op );
  2610     // And append it to the list for this peephole rule
  2611     peep.append_constraint( constraint );
  2613     // Check for another constraint, or end of rule
  2614     skipws();
  2615     if( _curchar == ',' ) {
  2616       next_char();                // Skip ','
  2617       skipws();
  2619     else if( _curchar != ')' ) {
  2620       parse_err(SYNERR, "expected ',' or ')' after peephole constraint.\n");
  2621       return;
  2623   } // end while( processing constraints )
  2624   next_char();                    // Skip ')'
  2626   // Check for terminating ';'
  2627   skipws();
  2628   if (_curchar != ';') {
  2629     parse_err(SYNERR, "missing ';' at end of peepconstraint.\n");
  2630     return;
  2632   next_char();                    // Skip trailing ';'
  2636 //------------------------------peep_replace_parse-----------------------------
  2637 // Syntax for a peepreplace rule
  2638 // root instruction name followed by a
  2639 // parenthesized list of whitespace separated instruction.operand specifiers
  2640 //
  2641 // peepreplace ( instr_name  ( [instruction_number.operand_name]* ) );
  2642 //
  2643 //
  2644 void ADLParser::peep_replace_parse(Peephole &peep) {
  2645   int          lparen = 0;        // keep track of parenthesis nesting depth
  2646   int          rparen = 0;        // keep track of parenthesis nesting depth
  2647   int          icount = 0;        // count of instructions in rule for naming
  2648   char        *str    = NULL;
  2649   char        *token  = NULL;
  2651   skipws();
  2652   // Check for open paren
  2653   if (_curchar != '(') {
  2654     parse_err(SYNERR, "missing '(' at start of peepreplace rule.\n");
  2655     return;
  2657   else {
  2658     lparen++;
  2659     next_char();
  2662   // Check for root instruction
  2663   char       *inst = get_ident_dup();
  2664   const Form *form = _AD._globalNames[inst];
  2665   if( form == NULL || form->is_instruction() == NULL ) {
  2666     parse_err(SYNERR, "Instruction name expected at start of peepreplace.\n");
  2667     return;
  2670   // Store string representation of rule into replace
  2671   PeepReplace *replace = new PeepReplace(str);
  2672   replace->add_instruction( inst );
  2674   skipws();
  2675   // Start of root's operand-list
  2676   if (_curchar != '(') {
  2677     parse_err(SYNERR, "missing '(' at peepreplace root's operand-list.\n");
  2678     return;
  2680   else {
  2681     lparen++;
  2682     next_char();
  2685   skipws();
  2686   // Get the list of operands
  2687   while( _curchar != ')' ) {
  2688     // Get information on an instruction and its operand
  2689     // instructions's number
  2690     int   inst_num = get_int();
  2691     // Left-instruction's operand
  2692     skipws();
  2693     if( _curchar != '.' ) {
  2694       parse_err(SYNERR, "missing '.' in peepreplace after instruction number.\n");
  2695       return;
  2697     next_char();                  // Skip '.'
  2698     char *inst_op = get_ident_dup();
  2699     if( inst_op == NULL ) {
  2700       parse_err(SYNERR, "missing operand identifier in peepreplace.\n");
  2701       return;
  2704     // Record this operand's position in peepmatch
  2705     replace->add_operand( inst_num, inst_op );
  2706     skipws();
  2709   // Check for the end of operands list
  2710   skipws();
  2711   assert( _curchar == ')', "While loop should have advanced to ')'.");
  2712   next_char();  // Skip ')'
  2714   skipws();
  2715   // Check for end of peepreplace
  2716   if( _curchar != ')' ) {
  2717     parse_err(SYNERR, "missing ')' at end of peepmatch.\n");
  2718     parse_err(SYNERR, "Support one replacement instruction.\n");
  2719     return;
  2721   next_char(); // Skip ')'
  2723   // Check for closing semicolon
  2724   skipws();
  2725   if( _curchar != ';' ) {
  2726     parse_err(SYNERR, "missing ';' at end of peepreplace.\n");
  2727     return;
  2729   next_char();   // skip ';'
  2731   // Store replace into peep
  2732   peep.add_replace( replace );
  2735 //------------------------------pred_parse-------------------------------------
  2736 Predicate *ADLParser::pred_parse(void) {
  2737   Predicate *predicate;           // Predicate class for operand
  2738   char      *rule = NULL;         // String representation of predicate
  2740   skipws();                       // Skip leading whitespace
  2741   int line = linenum();
  2742   if ( (rule = get_paren_expr("pred expression", true)) == NULL ) {
  2743     parse_err(SYNERR, "incorrect or missing expression for 'predicate'\n");
  2744     return NULL;
  2746   // Debug Stuff
  2747   if (_AD._adl_debug > 1) fprintf(stderr,"Predicate: %s\n", rule);
  2748   if (_curchar != ';') {
  2749     parse_err(SYNERR, "missing ';' in predicate definition\n");
  2750     return NULL;
  2752   next_char();                     // Point after the terminator
  2754   predicate = new Predicate(rule); // Build new predicate object
  2755   skipws();
  2756   return predicate;
  2760 //------------------------------ins_encode_parse_block-------------------------
  2761 // Parse the block form of ins_encode.  See ins_encode_parse for more details
  2762 void ADLParser::ins_encode_parse_block(InstructForm& inst) {
  2763   // Create a new encoding name based on the name of the instruction
  2764   // definition, which should be unique.
  2765   const char* prefix = "__ins_encode_";
  2766   char* ec_name = (char*) malloc(strlen(inst._ident) + strlen(prefix) + 1);
  2767   sprintf(ec_name, "%s%s", prefix, inst._ident);
  2769   assert(_AD._encode->encClass(ec_name) == NULL, "shouldn't already exist");
  2770   EncClass* encoding = _AD._encode->add_EncClass(ec_name);
  2771   encoding->_linenum = linenum();
  2773   // synthesize the arguments list for the enc_class from the
  2774   // arguments to the instruct definition.
  2775   const char* param = NULL;
  2776   inst._parameters.reset();
  2777   while ((param = inst._parameters.iter()) != NULL) {
  2778     OperandForm* opForm = (OperandForm*) inst._localNames[param];
  2779     encoding->add_parameter(opForm->_ident, param);
  2782   // Define a MacroAssembler instance for use by the encoding.  The
  2783   // name is chosen to match the __ idiom used for assembly in other
  2784   // parts of hotspot and assumes the existence of the standard
  2785   // #define __ _masm.
  2786   encoding->add_code("    MacroAssembler _masm(&cbuf);\n");
  2788   // Parse the following %{ }% block
  2789   ins_encode_parse_block_impl(inst, encoding, ec_name);
  2791   // Build an encoding rule which invokes the encoding rule we just
  2792   // created, passing all arguments that we received.
  2793   InsEncode*   encrule = new InsEncode(); // Encode class for instruction
  2794   NameAndList* params  = encrule->add_encode(ec_name);
  2795   inst._parameters.reset();
  2796   while ((param = inst._parameters.iter()) != NULL) {
  2797     params->add_entry(param);
  2800   // Check for duplicate ins_encode sections after parsing the block
  2801   // so that parsing can continue and find any other errors.
  2802   if (inst._insencode != NULL) {
  2803     parse_err(SYNERR, "Multiple ins_encode sections defined\n");
  2804     return;
  2807   // Set encode class of this instruction.
  2808   inst._insencode = encrule;
  2812 void ADLParser::ins_encode_parse_block_impl(InstructForm& inst, EncClass* encoding, char* ec_name) {
  2813   skipws_no_preproc();              // Skip leading whitespace
  2814   // Prepend location descriptor, for debugging; cf. ADLParser::find_cpp_block
  2815   if (_AD._adlocation_debug) {
  2816     encoding->add_code(get_line_string());
  2819   // Collect the parts of the encode description
  2820   // (1) strings that are passed through to output
  2821   // (2) replacement/substitution variable, preceeded by a '$'
  2822   while ((_curchar != '%') && (*(_ptr+1) != '}')) {
  2824     // (1)
  2825     // Check if there is a string to pass through to output
  2826     char *start = _ptr;       // Record start of the next string
  2827     while ((_curchar != '$') && ((_curchar != '%') || (*(_ptr+1) != '}')) ) {
  2828       // If at the start of a comment, skip past it
  2829       if( (_curchar == '/') && ((*(_ptr+1) == '/') || (*(_ptr+1) == '*')) ) {
  2830         skipws_no_preproc();
  2831       } else {
  2832         // ELSE advance to the next character, or start of the next line
  2833         next_char_or_line();
  2836     // If a string was found, terminate it and record in EncClass
  2837     if (start != _ptr) {
  2838       *_ptr = '\0';          // Terminate the string
  2839       encoding->add_code(start);
  2842     // (2)
  2843     // If we are at a replacement variable,
  2844     // copy it and record in EncClass
  2845     if (_curchar == '$') {
  2846       // Found replacement Variable
  2847       char* rep_var = get_rep_var_ident_dup();
  2849       // Add flag to _strings list indicating we should check _rep_vars
  2850       encoding->add_rep_var(rep_var);
  2852       skipws();
  2854       // Check if this instruct is a MachConstantNode.
  2855       if (strcmp(rep_var, "constanttablebase") == 0) {
  2856         // This instruct is a MachConstantNode.
  2857         inst.set_is_mach_constant(true);
  2859         if (_curchar == '(')  {
  2860           parse_err(SYNERR, "constanttablebase in instruct %s cannot have an argument (only constantaddress and constantoffset)", ec_name);
  2861           return;
  2864       else if ((strcmp(rep_var, "constantaddress")   == 0) ||
  2865                (strcmp(rep_var, "constantoffset")    == 0)) {
  2866         // This instruct is a MachConstantNode.
  2867         inst.set_is_mach_constant(true);
  2869         // If the constant keyword has an argument, parse it.
  2870         if (_curchar == '(')  constant_parse(inst);
  2873   } // end while part of format description
  2874   next_char();                      // Skip '%'
  2875   next_char();                      // Skip '}'
  2877   skipws();
  2879   if (_AD._adlocation_debug) {
  2880     encoding->add_code(end_line_marker());
  2883   // Debug Stuff
  2884   if (_AD._adl_debug > 1)  fprintf(stderr, "EncodingClass Form: %s\n", ec_name);
  2888 //------------------------------ins_encode_parse-------------------------------
  2889 // Encode rules have the form
  2890 //   ins_encode( encode_class_name(parameter_list), ... );
  2891 //
  2892 // The "encode_class_name" must be defined in the encode section
  2893 // The parameter list contains $names that are locals.
  2894 //
  2895 // Alternatively it can be written like this:
  2896 //
  2897 //   ins_encode %{
  2898 //      ... // body
  2899 //   %}
  2900 //
  2901 // which synthesizes a new encoding class taking the same arguments as
  2902 // the InstructForm, and automatically prefixes the definition with:
  2903 //
  2904 //    MacroAssembler masm(&cbuf);\n");
  2905 //
  2906 //  making it more compact to take advantage of the MacroAssembler and
  2907 //  placing the assembly closer to it's use by instructions.
  2908 void ADLParser::ins_encode_parse(InstructForm& inst) {
  2910   // Parse encode class name
  2911   skipws();                        // Skip whitespace
  2912   if (_curchar != '(') {
  2913     // Check for ins_encode %{ form
  2914     if ((_curchar == '%') && (*(_ptr+1) == '{')) {
  2915       next_char();                      // Skip '%'
  2916       next_char();                      // Skip '{'
  2918       // Parse the block form of ins_encode
  2919       ins_encode_parse_block(inst);
  2920       return;
  2923     parse_err(SYNERR, "missing '%%{' or '(' in ins_encode definition\n");
  2924     return;
  2926   next_char();                     // move past '('
  2927   skipws();
  2929   InsEncode *encrule  = new InsEncode(); // Encode class for instruction
  2930   encrule->_linenum = linenum();
  2931   char      *ec_name  = NULL;      // String representation of encode rule
  2932   // identifier is optional.
  2933   while (_curchar != ')') {
  2934     ec_name = get_ident();
  2935     if (ec_name == NULL) {
  2936       parse_err(SYNERR, "Invalid encode class name after 'ins_encode('.\n");
  2937       return;
  2939     // Check that encoding is defined in the encode section
  2940     EncClass *encode_class = _AD._encode->encClass(ec_name);
  2941     if (encode_class == NULL) {
  2942       // Like to defer checking these till later...
  2943       // parse_err(WARN, "Using an undefined encode class '%s' in 'ins_encode'.\n", ec_name);
  2946     // Get list for encode method's parameters
  2947     NameAndList *params = encrule->add_encode(ec_name);
  2949     // Parse the parameters to this encode method.
  2950     skipws();
  2951     if ( _curchar == '(' ) {
  2952       next_char();                 // move past '(' for parameters
  2954       // Parse the encode method's parameters
  2955       while (_curchar != ')') {
  2956         char *param = get_ident_or_literal_constant("encoding operand");
  2957         if ( param != NULL ) {
  2958           // Found a parameter:
  2959           // Check it is a local name, add it to the list, then check for more
  2960           // New: allow hex constants as parameters to an encode method.
  2961           // New: allow parenthesized expressions as parameters.
  2962           // New: allow "primary", "secondary", "tertiary" as parameters.
  2963           // New: allow user-defined register name as parameter
  2964           if ( (inst._localNames[param] == NULL) &&
  2965                !ADLParser::is_literal_constant(param) &&
  2966                (Opcode::as_opcode_type(param) == Opcode::NOT_AN_OPCODE) &&
  2967                ((_AD._register == NULL ) || (_AD._register->getRegDef(param) == NULL)) ) {
  2968             parse_err(SYNERR, "Using non-locally defined parameter %s for encoding %s.\n", param, ec_name);
  2969             return;
  2971           params->add_entry(param);
  2973           skipws();
  2974           if (_curchar == ',' ) {
  2975             // More parameters to come
  2976             next_char();           // move past ',' between parameters
  2977             skipws();              // Skip to next parameter
  2979           else if (_curchar == ')') {
  2980             // Done with parameter list
  2982           else {
  2983             // Only ',' or ')' are valid after a parameter name
  2984             parse_err(SYNERR, "expected ',' or ')' after parameter %s.\n",
  2985                       ec_name);
  2986             return;
  2989         } else {
  2990           skipws();
  2991           // Did not find a parameter
  2992           if (_curchar == ',') {
  2993             parse_err(SYNERR, "Expected encode parameter before ',' in encoding %s.\n", ec_name);
  2994             return;
  2996           if (_curchar != ')') {
  2997             parse_err(SYNERR, "Expected ')' after encode parameters.\n");
  2998             return;
  3001       } // WHILE loop collecting parameters
  3002       next_char();                   // move past ')' at end of parameters
  3003     } // done with parameter list for encoding
  3005     // Check for ',' or ')' after encoding
  3006     skipws();                      // move to character after parameters
  3007     if ( _curchar == ',' ) {
  3008       // Found a ','
  3009       next_char();                 // move past ',' between encode methods
  3010       skipws();
  3012     else if ( _curchar != ')' ) {
  3013       // If not a ',' then only a ')' is allowed
  3014       parse_err(SYNERR, "Expected ')' after encoding %s.\n", ec_name);
  3015       return;
  3018     // Check for ',' separating parameters
  3019     // if ( _curchar != ',' && _curchar != ')' ) {
  3020     //   parse_err(SYNERR, "expected ',' or ')' after encode method inside ins_encode.\n");
  3021     //   return NULL;
  3022     // }
  3024   } // done parsing ins_encode methods and their parameters
  3025   if (_curchar != ')') {
  3026     parse_err(SYNERR, "Missing ')' at end of ins_encode description.\n");
  3027     return;
  3029   next_char();                     // move past ')'
  3030   skipws();                        // Skip leading whitespace
  3032   if ( _curchar != ';' ) {
  3033     parse_err(SYNERR, "Missing ';' at end of ins_encode.\n");
  3034     return;
  3036   next_char();                     // move past ';'
  3037   skipws();                        // be friendly to oper_parse()
  3039   // Check for duplicate ins_encode sections after parsing the block
  3040   // so that parsing can continue and find any other errors.
  3041   if (inst._insencode != NULL) {
  3042     parse_err(SYNERR, "Multiple ins_encode sections defined\n");
  3043     return;
  3046   // Debug Stuff
  3047   if (_AD._adl_debug > 1) fprintf(stderr,"Instruction Encode: %s\n", ec_name);
  3049   // Set encode class of this instruction.
  3050   inst._insencode = encrule;
  3054 //------------------------------constant_parse---------------------------------
  3055 // Parse a constant expression.
  3056 void ADLParser::constant_parse(InstructForm& inst) {
  3057   // Create a new encoding name based on the name of the instruction
  3058   // definition, which should be unique.
  3059   const char* prefix = "__constant_";
  3060   char* ec_name = (char*) malloc(strlen(inst._ident) + strlen(prefix) + 1);
  3061   sprintf(ec_name, "%s%s", prefix, inst._ident);
  3063   assert(_AD._encode->encClass(ec_name) == NULL, "shouldn't already exist");
  3064   EncClass* encoding = _AD._encode->add_EncClass(ec_name);
  3065   encoding->_linenum = linenum();
  3067   // synthesize the arguments list for the enc_class from the
  3068   // arguments to the instruct definition.
  3069   const char* param = NULL;
  3070   inst._parameters.reset();
  3071   while ((param = inst._parameters.iter()) != NULL) {
  3072     OperandForm* opForm = (OperandForm*) inst._localNames[param];
  3073     encoding->add_parameter(opForm->_ident, param);
  3076   // Parse the following ( ) expression.
  3077   constant_parse_expression(encoding, ec_name);
  3079   // Build an encoding rule which invokes the encoding rule we just
  3080   // created, passing all arguments that we received.
  3081   InsEncode*   encrule = new InsEncode(); // Encode class for instruction
  3082   NameAndList* params  = encrule->add_encode(ec_name);
  3083   inst._parameters.reset();
  3084   while ((param = inst._parameters.iter()) != NULL) {
  3085     params->add_entry(param);
  3088   // Set encode class of this instruction.
  3089   inst._constant = encrule;
  3093 //------------------------------constant_parse_expression----------------------
  3094 void ADLParser::constant_parse_expression(EncClass* encoding, char* ec_name) {
  3095   skipws();
  3097   // Prepend location descriptor, for debugging; cf. ADLParser::find_cpp_block
  3098   if (_AD._adlocation_debug) {
  3099     encoding->add_code(get_line_string());
  3102   // Start code line.
  3103   encoding->add_code("    _constant = C->constant_table().add");
  3105   // Parse everything in ( ) expression.
  3106   encoding->add_code("(this, ");
  3107   next_char();  // Skip '('
  3108   int parens_depth = 1;
  3110   // Collect the parts of the constant expression.
  3111   // (1) strings that are passed through to output
  3112   // (2) replacement/substitution variable, preceeded by a '$'
  3113   while (parens_depth > 0) {
  3114     if (_curchar == '(') {
  3115       parens_depth++;
  3116       encoding->add_code("(");
  3117       next_char();
  3119     else if (_curchar == ')') {
  3120       parens_depth--;
  3121       if (parens_depth > 0)
  3122         encoding->add_code(")");
  3123       next_char();
  3125     else {
  3126       // (1)
  3127       // Check if there is a string to pass through to output
  3128       char *start = _ptr;  // Record start of the next string
  3129       while ((_curchar != '$') && (_curchar != '(') && (_curchar != ')')) {
  3130         next_char();
  3132       // If a string was found, terminate it and record in EncClass
  3133       if (start != _ptr) {
  3134         *_ptr = '\0';  // Terminate the string
  3135         encoding->add_code(start);
  3138       // (2)
  3139       // If we are at a replacement variable, copy it and record in EncClass.
  3140       if (_curchar == '$') {
  3141         // Found replacement Variable
  3142         char* rep_var = get_rep_var_ident_dup();
  3143         encoding->add_rep_var(rep_var);
  3148   // Finish code line.
  3149   encoding->add_code(");");
  3151   if (_AD._adlocation_debug) {
  3152     encoding->add_code(end_line_marker());
  3155   // Debug Stuff
  3156   if (_AD._adl_debug > 1)  fprintf(stderr, "EncodingClass Form: %s\n", ec_name);
  3160 //------------------------------size_parse-----------------------------------
  3161 // Parse a 'size(<expr>)' attribute which specifies the size of the
  3162 // emitted instructions in bytes. <expr> can be a C++ expression,
  3163 // e.g. a constant.
  3164 char* ADLParser::size_parse(InstructForm *instr) {
  3165   char* sizeOfInstr = NULL;
  3167   // Get value of the instruction's size
  3168   skipws();
  3170   // Parse size
  3171   sizeOfInstr = get_paren_expr("size expression");
  3172   if (sizeOfInstr == NULL) {
  3173      parse_err(SYNERR, "size of opcode expected at %c\n", _curchar);
  3174      return NULL;
  3177   skipws();
  3179   // Check for terminator
  3180   if (_curchar != ';') {
  3181     parse_err(SYNERR, "missing ';' in ins_attrib definition\n");
  3182     return NULL;
  3184   next_char();                     // Advance past the ';'
  3185   skipws();                        // necessary for instr_parse()
  3187   // Debug Stuff
  3188   if (_AD._adl_debug > 1) {
  3189     if (sizeOfInstr != NULL) {
  3190       fprintf(stderr,"size of opcode: %s\n", sizeOfInstr);
  3194   return sizeOfInstr;
  3198 //------------------------------opcode_parse-----------------------------------
  3199 Opcode * ADLParser::opcode_parse(InstructForm *instr) {
  3200   char *primary   = NULL;
  3201   char *secondary = NULL;
  3202   char *tertiary  = NULL;
  3204   char   *val    = NULL;
  3205   Opcode *opcode = NULL;
  3207   // Get value of the instruction's opcode
  3208   skipws();
  3209   if (_curchar != '(') {         // Check for parenthesized operand list
  3210     parse_err(SYNERR, "missing '(' in expand instruction declaration\n");
  3211     return NULL;
  3213   next_char();                   // skip open paren
  3214   skipws();
  3215   if (_curchar != ')') {
  3216     // Parse primary, secondary, and tertiary opcodes, if provided.
  3217     if ( ((primary = get_ident_or_literal_constant("primary opcode")) == NULL) ) {
  3218         parse_err(SYNERR, "primary hex opcode expected at %c\n", _curchar);
  3219         return NULL;
  3221     skipws();
  3222     if (_curchar == ',') {
  3223       next_char();
  3224       skipws();
  3225       // Parse secondary opcode
  3226       if ( ((secondary = get_ident_or_literal_constant("secondary opcode")) == NULL) ) {
  3227         parse_err(SYNERR, "secondary hex opcode expected at %c\n", _curchar);
  3228         return NULL;
  3230       skipws();
  3231       if (_curchar == ',') {
  3232         next_char();
  3233         skipws();
  3234         // Parse tertiary opcode
  3235         if ( ((tertiary = get_ident_or_literal_constant("tertiary opcode")) == NULL) ) {
  3236           parse_err(SYNERR,"tertiary hex opcode expected at %c\n", _curchar);
  3237           return NULL;
  3239         skipws();
  3242     skipws();
  3243     if (_curchar != ')') {
  3244       parse_err(SYNERR, "Missing ')' in opcode description\n");
  3245       return NULL;
  3248   next_char();                     // Skip ')'
  3249   skipws();
  3250   // Check for terminator
  3251   if (_curchar != ';') {
  3252     parse_err(SYNERR, "missing ';' in ins_attrib definition\n");
  3253     return NULL;
  3255   next_char();                     // Advance past the ';'
  3256   skipws();                        // necessary for instr_parse()
  3258   // Debug Stuff
  3259   if (_AD._adl_debug > 1) {
  3260     if (primary   != NULL) fprintf(stderr,"primary   opcode: %s\n", primary);
  3261     if (secondary != NULL) fprintf(stderr,"secondary opcode: %s\n", secondary);
  3262     if (tertiary  != NULL) fprintf(stderr,"tertiary  opcode: %s\n", tertiary);
  3265   // Generate new object and return
  3266   opcode = new Opcode(primary, secondary, tertiary);
  3267   return opcode;
  3271 //------------------------------interface_parse--------------------------------
  3272 Interface *ADLParser::interface_parse(void) {
  3273   char *iface_name  = NULL;      // Name of interface class being used
  3274   char *iface_code  = NULL;      // Describe components of this class
  3276   // Get interface class name
  3277   skipws();                       // Skip whitespace
  3278   if (_curchar != '(') {
  3279     parse_err(SYNERR, "Missing '(' at start of interface description.\n");
  3280     return NULL;
  3282   next_char();                    // move past '('
  3283   skipws();
  3284   iface_name = get_ident();
  3285   if (iface_name == NULL) {
  3286     parse_err(SYNERR, "missing interface name after 'interface'.\n");
  3287     return NULL;
  3289   skipws();
  3290   if (_curchar != ')') {
  3291     parse_err(SYNERR, "Missing ')' after name of interface.\n");
  3292     return NULL;
  3294   next_char();                    // move past ')'
  3296   // Get details of the interface,
  3297   // for the type of interface indicated by iface_name.
  3298   Interface *inter = NULL;
  3299   skipws();
  3300   if ( _curchar != ';' ) {
  3301     if ( strcmp(iface_name,"MEMORY_INTER") == 0 ) {
  3302       inter = mem_interface_parse();
  3304     else if ( strcmp(iface_name,"COND_INTER") == 0 ) {
  3305       inter = cond_interface_parse();
  3307     // The parse routines consume the "%}"
  3309     // Check for probable extra ';' after defining block.
  3310     if ( _curchar == ';' ) {
  3311       parse_err(SYNERR, "Extra ';' after defining interface block.\n");
  3312       next_char();                // Skip ';'
  3313       return NULL;
  3315   } else {
  3316     next_char();                  // move past ';'
  3318     // Create appropriate interface object
  3319     if ( strcmp(iface_name,"REG_INTER") == 0 ) {
  3320       inter = new RegInterface();
  3322     else if ( strcmp(iface_name,"CONST_INTER") == 0 ) {
  3323       inter = new ConstInterface();
  3326   skipws();                       // be friendly to oper_parse()
  3327   // Debug Stuff
  3328   if (_AD._adl_debug > 1) fprintf(stderr,"Interface Form: %s\n", iface_name);
  3330   // Create appropriate interface object and return.
  3331   return inter;
  3335 //------------------------------mem_interface_parse----------------------------
  3336 Interface *ADLParser::mem_interface_parse(void) {
  3337   // Fields for MemInterface
  3338   char *base        = NULL;
  3339   char *index       = NULL;
  3340   char *scale       = NULL;
  3341   char *disp        = NULL;
  3343   if (_curchar != '%') {
  3344     parse_err(SYNERR, "Missing '%%{' for 'interface' block.\n");
  3345     return NULL;
  3347   next_char();                  // Skip '%'
  3348   if (_curchar != '{') {
  3349     parse_err(SYNERR, "Missing '%%{' for 'interface' block.\n");
  3350     return NULL;
  3352   next_char();                  // Skip '{'
  3353   skipws();
  3354   do {
  3355     char *field = get_ident();
  3356     if (field == NULL) {
  3357       parse_err(SYNERR, "Expected keyword, base|index|scale|disp,  or '%%}' ending interface.\n");
  3358       return NULL;
  3360     if ( strcmp(field,"base") == 0 ) {
  3361       base  = interface_field_parse();
  3363     else if ( strcmp(field,"index") == 0 ) {
  3364       index = interface_field_parse();
  3366     else if ( strcmp(field,"scale") == 0 ) {
  3367       scale = interface_field_parse();
  3369     else if ( strcmp(field,"disp") == 0 ) {
  3370       disp  = interface_field_parse();
  3372     else {
  3373       parse_err(SYNERR, "Expected keyword, base|index|scale|disp,  or '%%}' ending interface.\n");
  3374       return NULL;
  3376   } while( _curchar != '%' );
  3377   next_char();                  // Skip '%'
  3378   if ( _curchar != '}' ) {
  3379     parse_err(SYNERR, "Missing '%%}' for 'interface' block.\n");
  3380     return NULL;
  3382   next_char();                  // Skip '}'
  3384   // Construct desired object and return
  3385   Interface *inter = new MemInterface(base, index, scale, disp);
  3386   return inter;
  3390 //------------------------------cond_interface_parse---------------------------
  3391 Interface *ADLParser::cond_interface_parse(void) {
  3392   char *equal;
  3393   char *not_equal;
  3394   char *less;
  3395   char *greater_equal;
  3396   char *less_equal;
  3397   char *greater;
  3398   char *overflow;
  3399   char *no_overflow;
  3400   const char *equal_format = "eq";
  3401   const char *not_equal_format = "ne";
  3402   const char *less_format = "lt";
  3403   const char *greater_equal_format = "ge";
  3404   const char *less_equal_format = "le";
  3405   const char *greater_format = "gt";
  3406   const char *overflow_format = "o";
  3407   const char *no_overflow_format = "no";
  3409   if (_curchar != '%') {
  3410     parse_err(SYNERR, "Missing '%%{' for 'cond_interface' block.\n");
  3411     return NULL;
  3413   next_char();                  // Skip '%'
  3414   if (_curchar != '{') {
  3415     parse_err(SYNERR, "Missing '%%{' for 'cond_interface' block.\n");
  3416     return NULL;
  3418   next_char();                  // Skip '{'
  3419   skipws();
  3420   do {
  3421     char *field = get_ident();
  3422     if (field == NULL) {
  3423       parse_err(SYNERR, "Expected keyword, base|index|scale|disp,  or '%%}' ending interface.\n");
  3424       return NULL;
  3426     if ( strcmp(field,"equal") == 0 ) {
  3427       equal  = interface_field_parse(&equal_format);
  3429     else if ( strcmp(field,"not_equal") == 0 ) {
  3430       not_equal = interface_field_parse(&not_equal_format);
  3432     else if ( strcmp(field,"less") == 0 ) {
  3433       less = interface_field_parse(&less_format);
  3435     else if ( strcmp(field,"greater_equal") == 0 ) {
  3436       greater_equal  = interface_field_parse(&greater_equal_format);
  3438     else if ( strcmp(field,"less_equal") == 0 ) {
  3439       less_equal = interface_field_parse(&less_equal_format);
  3441     else if ( strcmp(field,"greater") == 0 ) {
  3442       greater = interface_field_parse(&greater_format);
  3444     else if ( strcmp(field,"overflow") == 0 ) {
  3445       overflow = interface_field_parse(&overflow_format);
  3447     else if ( strcmp(field,"no_overflow") == 0 ) {
  3448       no_overflow = interface_field_parse(&no_overflow_format);
  3450     else {
  3451       parse_err(SYNERR, "Expected keyword, base|index|scale|disp,  or '%%}' ending interface.\n");
  3452       return NULL;
  3454   } while( _curchar != '%' );
  3455   next_char();                  // Skip '%'
  3456   if ( _curchar != '}' ) {
  3457     parse_err(SYNERR, "Missing '%%}' for 'interface' block.\n");
  3458     return NULL;
  3460   next_char();                  // Skip '}'
  3462   // Construct desired object and return
  3463   Interface *inter = new CondInterface(equal,         equal_format,
  3464                                        not_equal,     not_equal_format,
  3465                                        less,          less_format,
  3466                                        greater_equal, greater_equal_format,
  3467                                        less_equal,    less_equal_format,
  3468                                        greater,       greater_format,
  3469                                        overflow,      overflow_format,
  3470                                        no_overflow,   no_overflow_format);
  3471   return inter;
  3475 //------------------------------interface_field_parse--------------------------
  3476 char *ADLParser::interface_field_parse(const char ** format) {
  3477   char *iface_field = NULL;
  3479   // Get interface field
  3480   skipws();                      // Skip whitespace
  3481   if (_curchar != '(') {
  3482     parse_err(SYNERR, "Missing '(' at start of interface field.\n");
  3483     return NULL;
  3485   next_char();                   // move past '('
  3486   skipws();
  3487   if ( _curchar != '0' && _curchar != '$' ) {
  3488     parse_err(SYNERR, "missing or invalid interface field contents.\n");
  3489     return NULL;
  3491   iface_field = get_rep_var_ident();
  3492   if (iface_field == NULL) {
  3493     parse_err(SYNERR, "missing or invalid interface field contents.\n");
  3494     return NULL;
  3496   skipws();
  3497   if (format != NULL && _curchar == ',') {
  3498     next_char();
  3499     skipws();
  3500     if (_curchar != '"') {
  3501       parse_err(SYNERR, "Missing '\"' in field format .\n");
  3502       return NULL;
  3504     next_char();
  3505     char *start = _ptr;       // Record start of the next string
  3506     while ((_curchar != '"') && (_curchar != '%') && (_curchar != '\n')) {
  3507       if (_curchar == '\\')  next_char();  // superquote
  3508       if (_curchar == '\n')  parse_err(SYNERR, "newline in string");  // unimplemented!
  3509       next_char();
  3511     if (_curchar != '"') {
  3512       parse_err(SYNERR, "Missing '\"' at end of field format .\n");
  3513       return NULL;
  3515     // If a string was found, terminate it and record in FormatRule
  3516     if ( start != _ptr ) {
  3517       *_ptr  = '\0';          // Terminate the string
  3518       *format = start;
  3520     next_char();
  3521     skipws();
  3523   if (_curchar != ')') {
  3524     parse_err(SYNERR, "Missing ')' after interface field.\n");
  3525     return NULL;
  3527   next_char();                   // move past ')'
  3528   skipws();
  3529   if ( _curchar != ';' ) {
  3530     parse_err(SYNERR, "Missing ';' at end of interface field.\n");
  3531     return NULL;
  3533   next_char();                    // move past ';'
  3534   skipws();                       // be friendly to interface_parse()
  3536   return iface_field;
  3540 //------------------------------match_parse------------------------------------
  3541 MatchRule *ADLParser::match_parse(FormDict &operands) {
  3542   MatchRule *match;               // Match Rule class for instruction/operand
  3543   char      *cnstr = NULL;        // Code for constructor
  3544   int        depth = 0;           // Counter for matching parentheses
  3545   int        numleaves = 0;       // Counter for number of leaves in rule
  3547   // Parse the match rule tree
  3548   MatchNode *mnode = matchNode_parse(operands, depth, numleaves, true);
  3550   // Either there is a block with a constructor, or a ';' here
  3551   skipws();                       // Skip whitespace
  3552   if ( _curchar == ';' ) {        // Semicolon is valid terminator
  3553     cnstr = NULL;                 // no constructor for this form
  3554     next_char();                  // Move past the ';', replaced with '\0'
  3556   else if ((cnstr = find_cpp_block("match constructor")) == NULL ) {
  3557     parse_err(SYNERR, "invalid construction of match rule\n"
  3558               "Missing ';' or invalid '%%{' and '%%}' constructor\n");
  3559     return NULL;                  // No MatchRule to return
  3561   if (_AD._adl_debug > 1)
  3562     if (cnstr) fprintf(stderr,"Match Constructor: %s\n", cnstr);
  3563   // Build new MatchRule object
  3564   match = new MatchRule(_AD, mnode, depth, cnstr, numleaves);
  3565   skipws();                       // Skip any trailing whitespace
  3566   return match;                   // Return MatchRule object
  3569 //------------------------------format_parse-----------------------------------
  3570 FormatRule* ADLParser::format_parse(void) {
  3571   char       *desc   = NULL;
  3572   FormatRule *format = (new FormatRule(desc));
  3574   // Without expression form, MUST have a code block;
  3575   skipws();                       // Skip whitespace
  3576   if ( _curchar == ';' ) {        // Semicolon is valid terminator
  3577     desc  = NULL;                 // no constructor for this form
  3578     next_char();                  // Move past the ';', replaced with '\0'
  3580   else if ( _curchar == '%' && *(_ptr+1) == '{') {
  3581     next_char();                  // Move past the '%'
  3582     next_char();                  // Move past the '{'
  3584     skipws();
  3585     if (_curchar == '$') {
  3586       char* ident = get_rep_var_ident();
  3587       if (strcmp(ident, "$$template") == 0) return template_parse();
  3588       parse_err(SYNERR, "Unknown \"%s\" directive in format", ident);
  3589       return NULL;
  3591     // Check for the opening '"' inside the format description
  3592     if ( _curchar == '"' ) {
  3593       next_char();              // Move past the initial '"'
  3594       if( _curchar == '"' ) {   // Handle empty format string case
  3595         *_ptr = '\0';           // Terminate empty string
  3596         format->_strings.addName(_ptr);
  3599       // Collect the parts of the format description
  3600       // (1) strings that are passed through to tty->print
  3601       // (2) replacement/substitution variable, preceeded by a '$'
  3602       // (3) multi-token ANSIY C style strings
  3603       while ( true ) {
  3604         if ( _curchar == '%' || _curchar == '\n' ) {
  3605           if ( _curchar != '"' ) {
  3606             parse_err(SYNERR, "missing '\"' at end of format block");
  3607             return NULL;
  3611         // (1)
  3612         // Check if there is a string to pass through to output
  3613         char *start = _ptr;       // Record start of the next string
  3614         while ((_curchar != '$') && (_curchar != '"') && (_curchar != '%') && (_curchar != '\n')) {
  3615           if (_curchar == '\\') {
  3616             next_char();  // superquote
  3617             if ((_curchar == '$') || (_curchar == '%'))
  3618               // hack to avoid % escapes and warnings about undefined \ escapes
  3619               *(_ptr-1) = _curchar;
  3621           if (_curchar == '\n')  parse_err(SYNERR, "newline in string");  // unimplemented!
  3622           next_char();
  3624         // If a string was found, terminate it and record in FormatRule
  3625         if ( start != _ptr ) {
  3626           *_ptr  = '\0';          // Terminate the string
  3627           format->_strings.addName(start);
  3630         // (2)
  3631         // If we are at a replacement variable,
  3632         // copy it and record in FormatRule
  3633         if ( _curchar == '$' ) {
  3634           next_char();          // Move past the '$'
  3635           char* rep_var = get_ident(); // Nil terminate the variable name
  3636           rep_var = strdup(rep_var);// Copy the string
  3637           *_ptr   = _curchar;     // and replace Nil with original character
  3638           format->_rep_vars.addName(rep_var);
  3639           // Add flag to _strings list indicating we should check _rep_vars
  3640           format->_strings.addName(NameList::_signal);
  3643         // (3)
  3644         // Allow very long strings to be broken up,
  3645         // using the ANSI C syntax "foo\n" <newline> "bar"
  3646         if ( _curchar == '"') {
  3647           next_char();           // Move past the '"'
  3648           skipws();              // Skip white space before next string token
  3649           if ( _curchar != '"') {
  3650             break;
  3651           } else {
  3652             // Found one.  Skip both " and the whitespace in between.
  3653             next_char();
  3656       } // end while part of format description
  3658       // Check for closing '"' and '%}' in format description
  3659       skipws();                   // Move to closing '%}'
  3660       if ( _curchar != '%' ) {
  3661         parse_err(SYNERR, "non-blank characters between closing '\"' and '%%' in format");
  3662         return NULL;
  3664     } // Done with format description inside
  3666     skipws();
  3667     // Past format description, at '%'
  3668     if ( _curchar != '%' || *(_ptr+1) != '}' ) {
  3669       parse_err(SYNERR, "missing '%%}' at end of format block");
  3670       return NULL;
  3672     next_char();                  // Move past the '%'
  3673     next_char();                  // Move past the '}'
  3675   else {  // parameter list alone must terminate with a ';'
  3676     parse_err(SYNERR, "missing ';' after Format expression");
  3677     return NULL;
  3679   // Debug Stuff
  3680   if (_AD._adl_debug > 1) fprintf(stderr,"Format Rule: %s\n", desc);
  3682   skipws();
  3683   return format;
  3687 //------------------------------template_parse-----------------------------------
  3688 FormatRule* ADLParser::template_parse(void) {
  3689   char       *desc   = NULL;
  3690   FormatRule *format = (new FormatRule(desc));
  3692   skipws();
  3693   while ( (_curchar != '%') && (*(_ptr+1) != '}') ) {
  3695     // (1)
  3696     // Check if there is a string to pass through to output
  3698       char *start = _ptr;       // Record start of the next string
  3699       while ((_curchar != '$') && ((_curchar != '%') || (*(_ptr+1) != '}')) ) {
  3700         // If at the start of a comment, skip past it
  3701         if( (_curchar == '/') && ((*(_ptr+1) == '/') || (*(_ptr+1) == '*')) ) {
  3702           skipws_no_preproc();
  3703         } else {
  3704           // ELSE advance to the next character, or start of the next line
  3705           next_char_or_line();
  3708       // If a string was found, terminate it and record in EncClass
  3709       if ( start != _ptr ) {
  3710         *_ptr  = '\0';          // Terminate the string
  3711         // Add flag to _strings list indicating we should check _rep_vars
  3712         format->_strings.addName(NameList::_signal2);
  3713         format->_strings.addName(start);
  3717     // (2)
  3718     // If we are at a replacement variable,
  3719     // copy it and record in EncClass
  3720     if ( _curchar == '$' ) {
  3721       // Found replacement Variable
  3722       char *rep_var = get_rep_var_ident_dup();
  3723       if (strcmp(rep_var, "$emit") == 0) {
  3724         // switch to normal format parsing
  3725         next_char();
  3726         next_char();
  3727         skipws();
  3728         // Check for the opening '"' inside the format description
  3729         if ( _curchar == '"' ) {
  3730           next_char();              // Move past the initial '"'
  3731           if( _curchar == '"' ) {   // Handle empty format string case
  3732             *_ptr = '\0';           // Terminate empty string
  3733             format->_strings.addName(_ptr);
  3736           // Collect the parts of the format description
  3737           // (1) strings that are passed through to tty->print
  3738           // (2) replacement/substitution variable, preceeded by a '$'
  3739           // (3) multi-token ANSIY C style strings
  3740           while ( true ) {
  3741             if ( _curchar == '%' || _curchar == '\n' ) {
  3742               parse_err(SYNERR, "missing '\"' at end of format block");
  3743               return NULL;
  3746             // (1)
  3747             // Check if there is a string to pass through to output
  3748             char *start = _ptr;       // Record start of the next string
  3749             while ((_curchar != '$') && (_curchar != '"') && (_curchar != '%') && (_curchar != '\n')) {
  3750               if (_curchar == '\\')  next_char();  // superquote
  3751               if (_curchar == '\n')  parse_err(SYNERR, "newline in string");  // unimplemented!
  3752               next_char();
  3754             // If a string was found, terminate it and record in FormatRule
  3755             if ( start != _ptr ) {
  3756               *_ptr  = '\0';          // Terminate the string
  3757               format->_strings.addName(start);
  3760             // (2)
  3761             // If we are at a replacement variable,
  3762             // copy it and record in FormatRule
  3763             if ( _curchar == '$' ) {
  3764               next_char();          // Move past the '$'
  3765               char* next_rep_var = get_ident(); // Nil terminate the variable name
  3766               next_rep_var = strdup(next_rep_var);// Copy the string
  3767               *_ptr   = _curchar;     // and replace Nil with original character
  3768               format->_rep_vars.addName(next_rep_var);
  3769               // Add flag to _strings list indicating we should check _rep_vars
  3770               format->_strings.addName(NameList::_signal);
  3773             // (3)
  3774             // Allow very long strings to be broken up,
  3775             // using the ANSI C syntax "foo\n" <newline> "bar"
  3776             if ( _curchar == '"') {
  3777               next_char();           // Move past the '"'
  3778               skipws();              // Skip white space before next string token
  3779               if ( _curchar != '"') {
  3780                 break;
  3781               } else {
  3782                 // Found one.  Skip both " and the whitespace in between.
  3783                 next_char();
  3786           } // end while part of format description
  3788       } else {
  3789         // Add flag to _strings list indicating we should check _rep_vars
  3790         format->_rep_vars.addName(rep_var);
  3791         // Add flag to _strings list indicating we should check _rep_vars
  3792         format->_strings.addName(NameList::_signal3);
  3794     } // end while part of format description
  3797   skipws();
  3798   // Past format description, at '%'
  3799   if ( _curchar != '%' || *(_ptr+1) != '}' ) {
  3800     parse_err(SYNERR, "missing '%%}' at end of format block");
  3801     return NULL;
  3803   next_char();                  // Move past the '%'
  3804   next_char();                  // Move past the '}'
  3806   // Debug Stuff
  3807   if (_AD._adl_debug > 1) fprintf(stderr,"Format Rule: %s\n", desc);
  3809   skipws();
  3810   return format;
  3814 //------------------------------effect_parse-----------------------------------
  3815 void ADLParser::effect_parse(InstructForm *instr) {
  3816   char* desc   = NULL;
  3818   skipws();                      // Skip whitespace
  3819   if (_curchar != '(') {
  3820     parse_err(SYNERR, "missing '(' in effect definition\n");
  3821     return;
  3823   // Get list of effect-operand pairs and insert into dictionary
  3824   else get_effectlist(instr->_effects, instr->_localNames, instr->_has_call);
  3826   // Debug Stuff
  3827   if (_AD._adl_debug > 1) fprintf(stderr,"Effect description: %s\n", desc);
  3828   if (_curchar != ';') {
  3829     parse_err(SYNERR, "missing ';' in Effect definition\n");
  3831   next_char();                  // Skip ';'
  3835 //------------------------------expand_parse-----------------------------------
  3836 ExpandRule* ADLParser::expand_parse(InstructForm *instr) {
  3837   char         *ident, *ident2;
  3838   OperandForm  *oper;
  3839   InstructForm *ins;
  3840   NameAndList  *instr_and_operands = NULL;
  3841   ExpandRule   *exp = new ExpandRule();
  3843   // Expand is a block containing an ordered list of instructions, each of
  3844   // which has an ordered list of operands.
  3845   // Check for block delimiter
  3846   skipws();                        // Skip leading whitespace
  3847   if ((_curchar != '%')
  3848       || (next_char(), (_curchar != '{')) ) { // If not open block
  3849     parse_err(SYNERR, "missing '%%{' in expand definition\n");
  3850     return(NULL);
  3852   next_char();                     // Maintain the invariant
  3853   do {
  3854     ident = get_ident();           // Grab next identifier
  3855     if (ident == NULL) {
  3856       parse_err(SYNERR, "identifier expected at %c\n", _curchar);
  3857       continue;
  3858     }                              // Check that you have a valid instruction
  3859     const Form *form = _globalNames[ident];
  3860     ins = form ? form->is_instruction() : NULL;
  3861     if (ins == NULL) {
  3862       // This is a new operand
  3863       oper = form ? form->is_operand() : NULL;
  3864       if (oper == NULL) {
  3865         parse_err(SYNERR, "instruction/operand name expected at %s\n", ident);
  3866         continue;
  3868       // Throw the operand on the _newopers list
  3869       skipws();
  3870       ident = get_unique_ident(instr->_localNames,"Operand");
  3871       if (ident == NULL) {
  3872         parse_err(SYNERR, "identifier expected at %c\n", _curchar);
  3873         continue;
  3875       exp->_newopers.addName(ident);
  3876       // Add new operand to LocalNames
  3877       instr->_localNames.Insert(ident, oper);
  3878       // Grab any constructor code and save as a string
  3879       char *c = NULL;
  3880       skipws();
  3881       if (_curchar == '%') { // Need a constructor for the operand
  3882         c = find_cpp_block("Operand Constructor");
  3883         if (c == NULL) {
  3884           parse_err(SYNERR, "Invalid code block for operand constructor\n", _curchar);
  3885           continue;
  3887         // Add constructor to _newopconst Dict
  3888         exp->_newopconst.Insert(ident, c);
  3890       else if (_curchar != ';') { // If no constructor, need a ;
  3891         parse_err(SYNERR, "Missing ; in expand rule operand declaration\n");
  3892         continue;
  3894       else next_char(); // Skip the ;
  3895       skipws();
  3897     else {
  3898       // Add instruction to list
  3899       instr_and_operands = new NameAndList(ident);
  3900       // Grab operands, build nameList of them, and then put into dictionary
  3901       skipws();
  3902       if (_curchar != '(') {         // Check for parenthesized operand list
  3903         parse_err(SYNERR, "missing '(' in expand instruction declaration\n");
  3904         continue;
  3906       do {
  3907         next_char();                 // skip open paren & comma characters
  3908         skipws();
  3909         if (_curchar == ')') break;
  3910         ident2 = get_ident();
  3911         skipws();
  3912         if (ident2 == NULL) {
  3913           parse_err(SYNERR, "identifier expected at %c\n", _curchar);
  3914           continue;
  3915         }                            // Check that you have a valid operand
  3916         const Form *form2 = instr->_localNames[ident2];
  3917         if (!form2) {
  3918           parse_err(SYNERR, "operand name expected at %s\n", ident2);
  3919           continue;
  3921         oper = form2->is_operand();
  3922         if (oper == NULL && !form2->is_opclass()) {
  3923           parse_err(SYNERR, "operand name expected at %s\n", ident2);
  3924           continue;
  3925         }                            // Add operand to list
  3926         instr_and_operands->add_entry(ident2);
  3927       } while(_curchar == ',');
  3928       if (_curchar != ')') {
  3929         parse_err(SYNERR, "missing ')'in expand instruction declaration\n");
  3930         continue;
  3932       next_char();
  3933       if (_curchar != ';') {
  3934         parse_err(SYNERR, "missing ';'in expand instruction declaration\n");
  3935         continue;
  3937       next_char();
  3939       // Record both instruction name and its operand list
  3940       exp->add_instruction(instr_and_operands);
  3942       skipws();
  3945   } while(_curchar != '%');
  3946   next_char();
  3947   if (_curchar != '}') {
  3948     parse_err(SYNERR, "missing '%%}' in expand rule definition\n");
  3949     return(NULL);
  3951   next_char();
  3953   // Debug Stuff
  3954   if (_AD._adl_debug > 1) fprintf(stderr,"Expand Rule:\n");
  3956   skipws();
  3957   return (exp);
  3960 //------------------------------rewrite_parse----------------------------------
  3961 RewriteRule* ADLParser::rewrite_parse(void) {
  3962   char* params = NULL;
  3963   char* desc   = NULL;
  3966   // This feature targeted for second generation description language.
  3968   skipws();                      // Skip whitespace
  3969   // Get parameters for rewrite
  3970   if ((params = get_paren_expr("rewrite parameters")) == NULL) {
  3971     parse_err(SYNERR, "missing '(' in rewrite rule\n");
  3972     return NULL;
  3974   // Debug Stuff
  3975   if (_AD._adl_debug > 1) fprintf(stderr,"Rewrite parameters: %s\n", params);
  3977   // For now, grab entire block;
  3978   skipws();
  3979   if ( (desc = find_cpp_block("rewrite block")) == NULL ) {
  3980     parse_err(SYNERR, "incorrect or missing block for 'rewrite'.\n");
  3981     return NULL;
  3983   // Debug Stuff
  3984   if (_AD._adl_debug > 1) fprintf(stderr,"Rewrite Rule: %s\n", desc);
  3986   skipws();
  3987   return (new RewriteRule(params,desc));
  3990 //------------------------------attr_parse-------------------------------------
  3991 Attribute *ADLParser::attr_parse(char* ident) {
  3992   Attribute *attrib;              // Attribute class
  3993   char      *cost = NULL;         // String representation of cost attribute
  3995   skipws();                       // Skip leading whitespace
  3996   if ( (cost = get_paren_expr("attribute")) == NULL ) {
  3997     parse_err(SYNERR, "incorrect or missing expression for 'attribute'\n");
  3998     return NULL;
  4000   // Debug Stuff
  4001   if (_AD._adl_debug > 1) fprintf(stderr,"Attribute: %s\n", cost);
  4002   if (_curchar != ';') {
  4003     parse_err(SYNERR, "missing ';' in attribute definition\n");
  4004     return NULL;
  4006   next_char();                   // Point after the terminator
  4008   skipws();
  4009   attrib = new Attribute(ident,cost,INS_ATTR); // Build new predicate object
  4010   return attrib;
  4014 //------------------------------matchNode_parse--------------------------------
  4015 MatchNode *ADLParser::matchNode_parse(FormDict &operands, int &depth, int &numleaves, bool atroot) {
  4016   // Count depth of parenthesis nesting for both left and right children
  4017   int   lParens = depth;
  4018   int   rParens = depth;
  4020   // MatchNode objects for left, right, and root of subtree.
  4021   MatchNode *lChild = NULL;
  4022   MatchNode *rChild = NULL;
  4023   char      *token;               // Identifier which may be opcode or operand
  4025   // Match expression starts with a '('
  4026   if (cur_char() != '(')
  4027     return NULL;
  4029   next_char();                    // advance past '('
  4031   // Parse the opcode
  4032   token = get_ident();            // Get identifier, opcode
  4033   if (token == NULL) {
  4034     parse_err(SYNERR, "missing opcode in match expression\n");
  4035     return NULL;
  4038   // Take note if we see one of a few special operations - those that are
  4039   // treated differently on different architectures in the sense that on
  4040   // one architecture there is a match rule and on another there isn't (so
  4041   // a call will eventually be generated).
  4043   for (int i = _last_machine_leaf + 1; i < _last_opcode; i++) {
  4044     if (strcmp(token, NodeClassNames[i]) == 0) {
  4045       _AD.has_match_rule(i, true);
  4049   // Lookup the root value in the operands dict to perform substitution
  4050   const char  *result    = NULL;  // Result type will be filled in later
  4051   const char  *name      = token; // local name associated with this node
  4052   const char  *operation = token; // remember valid operation for later
  4053   const Form  *form      = operands[token];
  4054   OpClassForm *opcForm = form ? form->is_opclass() : NULL;
  4055   if (opcForm != NULL) {
  4056     // If this token is an entry in the local names table, record its type
  4057     if (!opcForm->ideal_only()) {
  4058       operation = opcForm->_ident;
  4059       result = operation;         // Operands result in their own type
  4061     // Otherwise it is an ideal type, and so, has no local name
  4062     else                        name = NULL;
  4065   // Parse the operands
  4066   skipws();
  4067   if (cur_char() != ')') {
  4069     // Parse the left child
  4070     if (strcmp(operation,"Set"))
  4071       lChild = matchChild_parse(operands, lParens, numleaves, false);
  4072     else
  4073       lChild = matchChild_parse(operands, lParens, numleaves, true);
  4075     skipws();
  4076     if (cur_char() != ')' ) {
  4077       if(strcmp(operation, "Set"))
  4078         rChild = matchChild_parse(operands,rParens,numleaves,false);
  4079       else
  4080         rChild = matchChild_parse(operands,rParens,numleaves,true);
  4084   // Check for required ')'
  4085   skipws();
  4086   if (cur_char() != ')') {
  4087     parse_err(SYNERR, "missing ')' in match expression\n");
  4088     return NULL;
  4090   next_char();                    // skip the ')'
  4092   MatchNode* mroot = new MatchNode(_AD,result,name,operation,lChild,rChild);
  4094   // If not the root, reduce this subtree to an internal operand
  4095   if (!atroot) {
  4096     mroot->build_internalop();
  4098   // depth is greater of left and right paths.
  4099   depth = (lParens > rParens) ? lParens : rParens;
  4101   return mroot;
  4105 //------------------------------matchChild_parse-------------------------------
  4106 MatchNode *ADLParser::matchChild_parse(FormDict &operands, int &parens, int &numleaves, bool atroot) {
  4107   MatchNode  *child  = NULL;
  4108   const char *result = NULL;
  4109   const char *token  = NULL;
  4110   const char *opType = NULL;
  4112   if (cur_char() == '(') {         // child is an operation
  4113     ++parens;
  4114     child = matchNode_parse(operands, parens, numleaves, atroot);
  4116   else {                           // child is an operand
  4117     token = get_ident();
  4118     const Form  *form    = operands[token];
  4119     OpClassForm *opcForm = form ? form->is_opclass() : NULL;
  4120     if (opcForm != NULL) {
  4121       opType = opcForm->_ident;
  4122       result = opcForm->_ident;    // an operand's result matches its type
  4123     } else {
  4124       parse_err(SYNERR, "undefined operand %s in match rule\n", token);
  4125       return NULL;
  4128     if (opType == NULL) {
  4129       parse_err(SYNERR, "missing type for argument '%s'\n", token);
  4132     child = new MatchNode(_AD, result, token, opType);
  4133     ++numleaves;
  4136   return child;
  4141 // ******************** Private Utility Functions *************************
  4144 char* ADLParser::find_cpp_block(const char* description) {
  4145   char *next;                     // Pointer for finding block delimiters
  4146   char* cppBlock = NULL;          // Beginning of C++ code block
  4148   if (_curchar == '%') {          // Encoding is a C++ expression
  4149     next_char();
  4150     if (_curchar != '{') {
  4151       parse_err(SYNERR, "missing '{' in %s \n", description);
  4152       return NULL;
  4154     next_char();                  // Skip block delimiter
  4155     skipws_no_preproc();          // Skip leading whitespace
  4156     cppBlock = _ptr;              // Point to start of expression
  4157     int line = linenum();
  4158     next = _ptr + 1;
  4159     while(((_curchar != '%') || (*next != '}')) && (_curchar != '\0')) {
  4160       next_char_or_line();
  4161       next = _ptr+1;              // Maintain the next pointer
  4162     }                             // Grab string
  4163     if (_curchar == '\0') {
  4164       parse_err(SYNERR, "invalid termination of %s \n", description);
  4165       return NULL;
  4167     *_ptr = '\0';                 // Terminate string
  4168     _ptr += 2;                    // Skip block delimiter
  4169     _curchar = *_ptr;             // Maintain invariant
  4171     // Prepend location descriptor, for debugging.
  4172     if (_AD._adlocation_debug) {
  4173       char* location = get_line_string(line);
  4174       char* end_loc  = end_line_marker();
  4175       char* result = (char *)malloc(strlen(location) + strlen(cppBlock) + strlen(end_loc) + 1);
  4176       strcpy(result, location);
  4177       strcat(result, cppBlock);
  4178       strcat(result, end_loc);
  4179       cppBlock = result;
  4180       free(location);
  4184   return cppBlock;
  4187 // Move to the closing token of the expression we are currently at,
  4188 // as defined by stop_chars.  Match parens and quotes.
  4189 char* ADLParser::get_expr(const char *desc, const char *stop_chars) {
  4190   char* expr = NULL;
  4191   int   paren = 0;
  4193   expr = _ptr;
  4194   while (paren > 0 || !strchr(stop_chars, _curchar)) {
  4195     if (_curchar == '(') {        // Down level of nesting
  4196       paren++;                    // Bump the parenthesis counter
  4197       next_char();                // maintain the invariant
  4199     else if (_curchar == ')') {   // Up one level of nesting
  4200       if (paren == 0) {
  4201         // Paren underflow:  We didn't encounter the required stop-char.
  4202         parse_err(SYNERR, "too many )'s, did not find %s after %s\n",
  4203                   stop_chars, desc);
  4204         return NULL;
  4206       paren--;                    // Drop the parenthesis counter
  4207       next_char();                // Maintain the invariant
  4209     else if (_curchar == '"' || _curchar == '\'') {
  4210       int qchar = _curchar;
  4211       while (true) {
  4212         next_char();
  4213         if (_curchar == qchar) { next_char(); break; }
  4214         if (_curchar == '\\')  next_char();  // superquote
  4215         if (_curchar == '\n' || _curchar == '\0') {
  4216           parse_err(SYNERR, "newline in string in %s\n", desc);
  4217           return NULL;
  4221     else if (_curchar == '%' && (_ptr[1] == '{' || _ptr[1] == '}')) {
  4222       // Make sure we do not stray into the next ADLC-level form.
  4223       parse_err(SYNERR, "unexpected %%%c in %s\n", _ptr[1], desc);
  4224       return NULL;
  4226     else if (_curchar == '\0') {
  4227       parse_err(SYNERR, "unexpected EOF in %s\n", desc);
  4228       return NULL;
  4230     else {
  4231       // Always walk over whitespace, comments, preprocessor directives, etc.
  4232       char* pre_skip_ptr = _ptr;
  4233       skipws();
  4234       // If the parser declined to make progress on whitespace,
  4235       // skip the next character, which is therefore NOT whitespace.
  4236       if (pre_skip_ptr == _ptr) {
  4237         next_char();
  4238       } else if (pre_skip_ptr+strlen(pre_skip_ptr) != _ptr+strlen(_ptr)) {
  4239         parse_err(SYNERR, "unimplemented: preprocessor must not elide subexpression in %s", desc);
  4244   assert(strchr(stop_chars, _curchar), "non-null return must be at stop-char");
  4245   *_ptr = '\0';               // Replace ')' or other stop-char with '\0'
  4246   return expr;
  4249 // Helper function around get_expr
  4250 // Sets _curchar to '(' so that get_paren_expr will search for a matching ')'
  4251 char *ADLParser::get_paren_expr(const char *description, bool include_location) {
  4252   int line = linenum();
  4253   if (_curchar != '(')            // Escape if not valid starting position
  4254     return NULL;
  4255   next_char();                    // Skip the required initial paren.
  4256   char *token2 = get_expr(description, ")");
  4257   if (_curchar == ')')
  4258     next_char();                  // Skip required final paren.
  4259   int junk = 0;
  4260   if (include_location && _AD._adlocation_debug && !is_int_token(token2, junk)) {
  4261     // Prepend location descriptor, for debugging.
  4262     char* location = get_line_string(line);
  4263     char* end_loc  = end_line_marker();
  4264     char* result = (char *)malloc(strlen(location) + strlen(token2) + strlen(end_loc) + 1);
  4265     strcpy(result, location);
  4266     strcat(result, token2);
  4267     strcat(result, end_loc);
  4268     token2 = result;
  4269     free(location);
  4271   return token2;
  4274 //------------------------------get_ident_common-------------------------------
  4275 // Looks for an identifier in the buffer, and turns it into a null terminated
  4276 // string(still inside the file buffer).  Returns a pointer to the string or
  4277 // NULL if some other token is found instead.
  4278 char *ADLParser::get_ident_common(bool do_preproc) {
  4279   register char c;
  4280   char *start;                    // Pointer to start of token
  4281   char *end;                      // Pointer to end of token
  4283   if( _curline == NULL )          // Return NULL at EOF.
  4284     return NULL;
  4286   skipws_common(do_preproc);      // Skip whitespace before identifier
  4287   start = end = _ptr;             // Start points at first character
  4288   end--;                          // unwind end by one to prepare for loop
  4289   do {
  4290     end++;                        // Increment end pointer
  4291     c = *end;                     // Grab character to test
  4292   } while ( ((c >= 'a') && (c <= 'z')) || ((c >= 'A') && (c <= 'Z'))
  4293             || ((c >= '0') && (c <= '9'))
  4294             || ((c == '_')) || ((c == ':')) || ((c == '#')) );
  4295   if (start == end) {             // We popped out on the first try
  4296     // It can occur that `start' contains the rest of the input file.
  4297     // In this case the output should be truncated.
  4298     if (strlen(start) > 24) {
  4299       char buf[32];
  4300       strncpy(buf, start, 20);
  4301       buf[20] = '\0';
  4302       strcat(buf, "[...]");
  4303       parse_err(SYNERR, "Identifier expected, but found '%s'.", buf);
  4304     } else {
  4305       parse_err(SYNERR, "Identifier expected, but found '%s'.", start);
  4307     start = NULL;
  4309   else {
  4310     _curchar = c;                 // Save the first character of next token
  4311     *end = '\0';                  // NULL terminate the string in place
  4313   _ptr = end;                     // Reset _ptr to point to next char after token
  4315   // Make sure we do not try to use #defined identifiers.  If start is
  4316   // NULL an error was already reported.
  4317   if (do_preproc && start != NULL) {
  4318     const char* def = _AD.get_preproc_def(start);
  4319     if (def != NULL && strcmp(def, start)) {
  4320       const char* def1 = def;
  4321       const char* def2 = _AD.get_preproc_def(def1);
  4322       // implement up to 2 levels of #define
  4323       if (def2 != NULL && strcmp(def2, def1)) {
  4324         def = def2;
  4325         const char* def3 = _AD.get_preproc_def(def2);
  4326         if (def3 != NULL && strcmp(def3, def2) && strcmp(def3, def1)) {
  4327           parse_err(SYNERR, "unimplemented: using %s defined as %s => %s => %s",
  4328                     start, def1, def2, def3);
  4331       start = strdup(def);
  4335   return start;                   // Pointer to token in filebuf
  4338 //------------------------------get_ident_dup----------------------------------
  4339 // Looks for an identifier in the buffer, and returns a duplicate
  4340 // or NULL if some other token is found instead.
  4341 char *ADLParser::get_ident_dup(void) {
  4342   char *ident = get_ident();
  4344   // Duplicate an identifier before returning and restore string.
  4345   if( ident != NULL ) {
  4346     ident = strdup(ident);  // Copy the string
  4347     *_ptr   = _curchar;         // and replace Nil with original character
  4350   return ident;
  4353 //----------------------get_ident_or_literal_constant--------------------------
  4354 // Looks for an identifier in the buffer, or a parenthesized expression.
  4355 char *ADLParser::get_ident_or_literal_constant(const char* description) {
  4356   char* param = NULL;
  4357   skipws();
  4358   if (_curchar == '(') {
  4359     // Grab a constant expression.
  4360     param = get_paren_expr(description);
  4361     if (param[0] != '(') {
  4362       char* buf = (char*) malloc(strlen(param) + 3);
  4363       sprintf(buf, "(%s)", param);
  4364       param = buf;
  4366     assert(is_literal_constant(param),
  4367            "expr must be recognizable as a constant");
  4368   } else {
  4369     param = get_ident();
  4371   return param;
  4374 //------------------------------get_rep_var_ident-----------------------------
  4375 // Do NOT duplicate,
  4376 // Leave nil terminator in buffer
  4377 // Preserve initial '$'(s) in string
  4378 char *ADLParser::get_rep_var_ident(void) {
  4379   // Remember starting point
  4380   char *rep_var = _ptr;
  4382   // Check for replacement variable indicator '$' and pass if present
  4383   if ( _curchar == '$' ) {
  4384     next_char();
  4386   // Check for a subfield indicator, a second '$', and pass if present
  4387   if ( _curchar == '$' ) {
  4388     next_char();
  4391   // Check for a control indicator, a third '$':
  4392   if ( _curchar == '$' ) {
  4393     next_char();
  4396   // Check for more than three '$'s in sequence, SYNERR
  4397   if( _curchar == '$' ) {
  4398     parse_err(SYNERR, "Replacement variables and field specifiers can not start with '$$$$'");
  4399     next_char();
  4400     return NULL;
  4403   // Nil terminate the variable name following the '$'
  4404   char *rep_var_name = get_ident();
  4405   assert( rep_var_name != NULL,
  4406           "Missing identifier after replacement variable indicator '$'");
  4408   return rep_var;
  4413 //------------------------------get_rep_var_ident_dup-------------------------
  4414 // Return the next replacement variable identifier, skipping first '$'
  4415 // given a pointer into a line of the buffer.
  4416 // Null terminates string, still inside the file buffer,
  4417 // Returns a pointer to a copy of the string, or NULL on failure
  4418 char *ADLParser::get_rep_var_ident_dup(void) {
  4419   if( _curchar != '$' ) return NULL;
  4421   next_char();                // Move past the '$'
  4422   char *rep_var = _ptr;       // Remember starting point
  4424   // Check for a subfield indicator, a second '$':
  4425   if ( _curchar == '$' ) {
  4426     next_char();
  4429   // Check for a control indicator, a third '$':
  4430   if ( _curchar == '$' ) {
  4431     next_char();
  4434   // Check for more than three '$'s in sequence, SYNERR
  4435   if( _curchar == '$' ) {
  4436     parse_err(SYNERR, "Replacement variables and field specifiers can not start with '$$$$'");
  4437     next_char();
  4438     return NULL;
  4441   // Nil terminate the variable name following the '$'
  4442   char *rep_var_name = get_ident();
  4443   assert( rep_var_name != NULL,
  4444           "Missing identifier after replacement variable indicator '$'");
  4445   rep_var = strdup(rep_var);  // Copy the string
  4446   *_ptr   = _curchar;         // and replace Nil with original character
  4448   return rep_var;
  4452 //------------------------------get_unique_ident------------------------------
  4453 // Looks for an identifier in the buffer, terminates it with a NULL,
  4454 // and checks that it is unique
  4455 char *ADLParser::get_unique_ident(FormDict& dict, const char* nameDescription){
  4456   char* ident = get_ident();
  4458   if (ident == NULL) {
  4459     parse_err(SYNERR, "missing %s identifier at %c\n", nameDescription, _curchar);
  4461   else {
  4462     if (dict[ident] != NULL) {
  4463       parse_err(SYNERR, "duplicate name %s for %s\n", ident, nameDescription);
  4464       ident = NULL;
  4468   return ident;
  4472 //------------------------------get_int----------------------------------------
  4473 // Looks for a character string integer in the buffer, and turns it into an int
  4474 // invokes a parse_err if the next token is not an integer.
  4475 // This routine does not leave the integer null-terminated.
  4476 int ADLParser::get_int(void) {
  4477   register char c;
  4478   char         *start;            // Pointer to start of token
  4479   char         *end;              // Pointer to end of token
  4480   int           result;           // Storage for integer result
  4482   if( _curline == NULL )          // Return NULL at EOF.
  4483     return 0;
  4485   skipws();                       // Skip whitespace before identifier
  4486   start = end = _ptr;             // Start points at first character
  4487   c = *end;                       // Grab character to test
  4488   while ((c >= '0') && (c <= '9')
  4489          || ((c == '-') && (end == start))) {
  4490     end++;                        // Increment end pointer
  4491     c = *end;                     // Grab character to test
  4493   if (start == end) {             // We popped out on the first try
  4494     parse_err(SYNERR, "integer expected at %c\n", c);
  4495     result = 0;
  4497   else {
  4498     _curchar = c;                 // Save the first character of next token
  4499     *end = '\0';                  // NULL terminate the string in place
  4500     result = atoi(start);         // Convert the string to an integer
  4501     *end = _curchar;              // Restore buffer to original condition
  4504   // Reset _ptr to next char after token
  4505   _ptr = end;
  4507   return result;                   // integer
  4511 //------------------------------get_relation_dup------------------------------
  4512 // Looks for a relational operator in the buffer
  4513 // invokes a parse_err if the next token is not a relation
  4514 // This routine creates a duplicate of the string in the buffer.
  4515 char *ADLParser::get_relation_dup(void) {
  4516   char         *result = NULL;    // relational operator being returned
  4518   if( _curline == NULL )          // Return NULL at EOF.
  4519     return  NULL;
  4521   skipws();                       // Skip whitespace before relation
  4522   char *start = _ptr;             // Store start of relational operator
  4523   char first  = *_ptr;            // the first character
  4524   if( (first == '=') || (first == '!') || (first == '<') || (first == '>') ) {
  4525     next_char();
  4526     char second = *_ptr;          // the second character
  4527     if( (second == '=') ) {
  4528       next_char();
  4529       char tmp  = *_ptr;
  4530       *_ptr = '\0';               // NULL terminate
  4531       result = strdup(start);     // Duplicate the string
  4532       *_ptr = tmp;                // restore buffer
  4533     } else {
  4534       parse_err(SYNERR, "relational operator expected at %s\n", _ptr);
  4536   } else {
  4537     parse_err(SYNERR, "relational operator expected at %s\n", _ptr);
  4540   return result;
  4545 //------------------------------get_oplist-------------------------------------
  4546 // Looks for identifier pairs where first must be the name of an operand, and
  4547 // second must be a name unique in the scope of this instruction.  Stores the
  4548 // names with a pointer to the OpClassForm of their type in a local name table.
  4549 void ADLParser::get_oplist(NameList &parameters, FormDict &operands) {
  4550   OpClassForm *opclass = NULL;
  4551   char        *ident   = NULL;
  4553   do {
  4554     next_char();             // skip open paren & comma characters
  4555     skipws();
  4556     if (_curchar == ')') break;
  4558     // Get operand type, and check it against global name table
  4559     ident = get_ident();
  4560     if (ident == NULL) {
  4561       parse_err(SYNERR, "optype identifier expected at %c\n", _curchar);
  4562       return;
  4564     else {
  4565       const Form  *form = _globalNames[ident];
  4566       if( form == NULL ) {
  4567         parse_err(SYNERR, "undefined operand type %s\n", ident);
  4568         return;
  4571       // Check for valid operand type
  4572       OpClassForm *opc  = form->is_opclass();
  4573       OperandForm *oper = form->is_operand();
  4574       if((oper == NULL) && (opc == NULL)) {
  4575         parse_err(SYNERR, "identifier %s not operand type\n", ident);
  4576         return;
  4578       opclass = opc;
  4580     // Debugging Stuff
  4581     if (_AD._adl_debug > 1) fprintf(stderr, "\tOperand Type: %s\t", ident);
  4583     // Get name of operand and add it to local name table
  4584     if( (ident = get_unique_ident(operands, "operand")) == NULL) {
  4585       return;
  4587     // Parameter names must not be global names.
  4588     if( _globalNames[ident] != NULL ) {
  4589          parse_err(SYNERR, "Reuse of global name %s as operand.\n",ident);
  4590          return;
  4592     operands.Insert(ident, opclass);
  4593     parameters.addName(ident);
  4595     // Debugging Stuff
  4596     if (_AD._adl_debug > 1) fprintf(stderr, "\tOperand Name: %s\n", ident);
  4597     skipws();
  4598   } while(_curchar == ',');
  4600   if (_curchar != ')') parse_err(SYNERR, "missing ')'\n");
  4601   else {
  4602     next_char();  // set current character position past the close paren
  4607 //------------------------------get_effectlist---------------------------------
  4608 // Looks for identifier pairs where first must be the name of a pre-defined,
  4609 // effect, and the second must be the name of an operand defined in the
  4610 // operand list of this instruction.  Stores the names with a pointer to the
  4611 // effect form in a local effects table.
  4612 void ADLParser::get_effectlist(FormDict &effects, FormDict &operands, bool& has_call) {
  4613   OperandForm *opForm;
  4614   Effect      *eForm;
  4615   char        *ident;
  4617   do {
  4618     next_char();             // skip open paren & comma characters
  4619     skipws();
  4620     if (_curchar == ')') break;
  4622     // Get effect type, and check it against global name table
  4623     ident = get_ident();
  4624     if (ident == NULL) {
  4625       parse_err(SYNERR, "effect type identifier expected at %c\n", _curchar);
  4626       return;
  4628     else {
  4629       // Check for valid effect type
  4630       const Form *form = _globalNames[ident];
  4631       if( form == NULL ) {
  4632         parse_err(SYNERR, "undefined effect type %s\n", ident);
  4633         return;
  4635       else {
  4636         if( (eForm = form->is_effect()) == NULL) {
  4637           parse_err(SYNERR, "identifier %s not effect type\n", ident);
  4638           return;
  4642       // Debugging Stuff
  4643     if (_AD._adl_debug > 1) fprintf(stderr, "\tEffect Type: %s\t", ident);
  4644     skipws();
  4645     if (eForm->is(Component::CALL)) {
  4646       if (_AD._adl_debug > 1) fprintf(stderr, "\n");
  4647       has_call = true;
  4648     } else {
  4649       // Get name of operand and check that it is in the local name table
  4650       if( (ident = get_unique_ident(effects, "effect")) == NULL) {
  4651         parse_err(SYNERR, "missing operand identifier in effect list\n");
  4652         return;
  4654       const Form *form = operands[ident];
  4655       opForm = form ? form->is_operand() : NULL;
  4656       if( opForm == NULL ) {
  4657         if( form && form->is_opclass() ) {
  4658           const char* cname = form->is_opclass()->_ident;
  4659           parse_err(SYNERR, "operand classes are illegal in effect lists (found %s %s)\n", cname, ident);
  4660         } else {
  4661           parse_err(SYNERR, "undefined operand %s in effect list\n", ident);
  4663         return;
  4665       // Add the pair to the effects table
  4666       effects.Insert(ident, eForm);
  4667       // Debugging Stuff
  4668       if (_AD._adl_debug > 1) fprintf(stderr, "\tOperand Name: %s\n", ident);
  4670     skipws();
  4671   } while(_curchar == ',');
  4673   if (_curchar != ')') parse_err(SYNERR, "missing ')'\n");
  4674   else {
  4675     next_char();  // set current character position past the close paren
  4680 //-------------------------------preproc_line----------------------------------
  4681 // A "#line" keyword has been seen, so parse the rest of the line.
  4682 void ADLParser::preproc_line(void) {
  4683   int line = get_int();
  4684   skipws_no_preproc();
  4685   const char* file = NULL;
  4686   if (_curchar == '"') {
  4687     next_char();              // Move past the initial '"'
  4688     file = _ptr;
  4689     while (true) {
  4690       if (_curchar == '\n') {
  4691         parse_err(SYNERR, "missing '\"' at end of #line directive");
  4692         return;
  4694       if (_curchar == '"') {
  4695         *_ptr  = '\0';          // Terminate the string
  4696         next_char();
  4697         skipws_no_preproc();
  4698         break;
  4700       next_char();
  4703   ensure_end_of_line();
  4704   if (file != NULL)
  4705     _AD._ADL_file._name = file;
  4706   _buf.set_linenum(line);
  4709 //------------------------------preproc_define---------------------------------
  4710 // A "#define" keyword has been seen, so parse the rest of the line.
  4711 void ADLParser::preproc_define(void) {
  4712   char* flag = get_ident_no_preproc();
  4713   skipws_no_preproc();
  4714   // only #define x y is supported for now
  4715   char* def = get_ident_no_preproc();
  4716   _AD.set_preproc_def(flag, def);
  4717   skipws_no_preproc();
  4718   if (_curchar != '\n') {
  4719     parse_err(SYNERR, "non-identifier in preprocessor definition\n");
  4723 //------------------------------preproc_undef----------------------------------
  4724 // An "#undef" keyword has been seen, so parse the rest of the line.
  4725 void ADLParser::preproc_undef(void) {
  4726   char* flag = get_ident_no_preproc();
  4727   skipws_no_preproc();
  4728   ensure_end_of_line();
  4729   _AD.set_preproc_def(flag, NULL);
  4734 //------------------------------parse_err--------------------------------------
  4735 // Issue a parser error message, and skip to the end of the current line
  4736 void ADLParser::parse_err(int flag, const char *fmt, ...) {
  4737   va_list args;
  4739   va_start(args, fmt);
  4740   if (flag == 1)
  4741     _AD._syntax_errs += _AD.emit_msg(0, flag, linenum(), fmt, args);
  4742   else if (flag == 2)
  4743     _AD._semantic_errs += _AD.emit_msg(0, flag, linenum(), fmt, args);
  4744   else
  4745     _AD._warnings += _AD.emit_msg(0, flag, linenum(), fmt, args);
  4747   int error_char = _curchar;
  4748   char* error_ptr = _ptr+1;
  4749   for(;*_ptr != '\n'; _ptr++) ; // Skip to the end of the current line
  4750   _curchar = '\n';
  4751   va_end(args);
  4752   _AD._no_output = 1;
  4754   if (flag == 1) {
  4755     char* error_tail = strchr(error_ptr, '\n');
  4756     char tem = *error_ptr;
  4757     error_ptr[-1] = '\0';
  4758     char* error_head = error_ptr-1;
  4759     while (error_head > _curline && *error_head)  --error_head;
  4760     if (error_tail)  *error_tail = '\0';
  4761     fprintf(stderr, "Error Context:  %s>>>%c<<<%s\n",
  4762             error_head, error_char, error_ptr);
  4763     if (error_tail)  *error_tail = '\n';
  4764     error_ptr[-1] = tem;
  4768 //---------------------------ensure_start_of_line------------------------------
  4769 // A preprocessor directive has been encountered.  Be sure it has fallen at
  4770 // the beginning of a line, or else report an error.
  4771 void ADLParser::ensure_start_of_line(void) {
  4772   if (_curchar == '\n') { next_line(); return; }
  4773   assert( _ptr >= _curline && _ptr < _curline+strlen(_curline),
  4774           "Must be able to find which line we are in" );
  4776   for (char *s = _curline; s < _ptr; s++) {
  4777     if (*s > ' ') {
  4778       parse_err(SYNERR, "'%c' must be at beginning of line\n", _curchar);
  4779       break;
  4784 //---------------------------ensure_end_of_line--------------------------------
  4785 // A preprocessor directive has been parsed.  Be sure there is no trailing
  4786 // garbage at the end of this line.  Set the scan point to the beginning of
  4787 // the next line.
  4788 void ADLParser::ensure_end_of_line(void) {
  4789   skipws_no_preproc();
  4790   if (_curchar != '\n' && _curchar != '\0') {
  4791     parse_err(SYNERR, "garbage char '%c' at end of line\n", _curchar);
  4792   } else {
  4793     next_char_or_line();
  4797 //---------------------------handle_preproc------------------------------------
  4798 // The '#' character introducing a preprocessor directive has been found.
  4799 // Parse the whole directive name (e.g., #define, #endif) and take appropriate
  4800 // action.  If we are in an "untaken" span of text, simply keep track of
  4801 // #ifdef nesting structure, so we can find out when to start taking text
  4802 // again.  (In this state, we "sort of support" C's #if directives, enough
  4803 // to disregard their associated #else and #endif lines.)  If we are in a
  4804 // "taken" span of text, there are two cases:  "#define" and "#undef"
  4805 // directives are preserved and passed up to the caller, which eventually
  4806 // passes control to the top-level parser loop, which handles #define and
  4807 // #undef directly.  (This prevents these directives from occurring in
  4808 // arbitrary positions in the AD file--we require better structure than C.)
  4809 // In the other case, and #ifdef, #ifndef, #else, or #endif is silently
  4810 // processed as whitespace, with the "taken" state of the text correctly
  4811 // updated.  This routine returns "false" exactly in the case of a "taken"
  4812 // #define or #undef, which tells the caller that a preprocessor token
  4813 // has appeared which must be handled explicitly by the parse loop.
  4814 bool ADLParser::handle_preproc_token() {
  4815   assert(*_ptr == '#', "must be at start of preproc");
  4816   ensure_start_of_line();
  4817   next_char();
  4818   skipws_no_preproc();
  4819   char* start_ident = _ptr;
  4820   char* ident = (_curchar == '\n') ? NULL : get_ident_no_preproc();
  4821   if (ident == NULL) {
  4822     parse_err(SYNERR, "expected preprocessor command, got end of line\n");
  4823   } else if (!strcmp(ident, "ifdef") ||
  4824              !strcmp(ident, "ifndef")) {
  4825     char* flag = get_ident_no_preproc();
  4826     ensure_end_of_line();
  4827     // Test the identifier only if we are already in taken code:
  4828     bool flag_def  = preproc_taken() && (_AD.get_preproc_def(flag) != NULL);
  4829     bool now_taken = !strcmp(ident, "ifdef") ? flag_def : !flag_def;
  4830     begin_if_def(now_taken);
  4831   } else if (!strcmp(ident, "if")) {
  4832     if (preproc_taken())
  4833       parse_err(SYNERR, "unimplemented: #%s %s", ident, _ptr+1);
  4834     next_line();
  4835     // Intelligently skip this nested C preprocessor directive:
  4836     begin_if_def(true);
  4837   } else if (!strcmp(ident, "else")) {
  4838     ensure_end_of_line();
  4839     invert_if_def();
  4840   } else if (!strcmp(ident, "endif")) {
  4841     ensure_end_of_line();
  4842     end_if_def();
  4843   } else if (preproc_taken()) {
  4844     // pass this token up to the main parser as "#define" or "#undef"
  4845     _ptr = start_ident;
  4846     _curchar = *--_ptr;
  4847     if( _curchar != '#' ) {
  4848       parse_err(SYNERR, "no space allowed after # in #define or #undef");
  4849       assert(_curchar == '#', "no space allowed after # in #define or #undef");
  4851     return false;
  4853   return true;
  4856 //---------------------------skipws_common-------------------------------------
  4857 // Skip whitespace, including comments and newlines, while keeping an accurate
  4858 // line count.
  4859 // Maybe handle certain preprocessor constructs: #ifdef, #ifndef, #else, #endif
  4860 void ADLParser::skipws_common(bool do_preproc) {
  4861   char *start = _ptr;
  4862   char *next = _ptr + 1;
  4864   if (*_ptr == '\0') {
  4865     // Check for string terminator
  4866     if (_curchar > ' ')  return;
  4867     if (_curchar == '\n') {
  4868       if (!do_preproc)  return;            // let caller handle the newline
  4869       next_line();
  4870       _ptr = _curline; next = _ptr + 1;
  4872     else if (_curchar == '#' ||
  4873         (_curchar == '/' && (*next == '/' || *next == '*'))) {
  4874       parse_err(SYNERR, "unimplemented: comment token in a funny place");
  4877   while(_curline != NULL) {                // Check for end of file
  4878     if (*_ptr == '\n') {                   // keep proper track of new lines
  4879       if (!do_preproc)  break;             // let caller handle the newline
  4880       next_line();
  4881       _ptr = _curline; next = _ptr + 1;
  4883     else if ((*_ptr == '/') && (*next == '/'))      // C++ comment
  4884       do { _ptr++; next++; } while(*_ptr != '\n');  // So go to end of line
  4885     else if ((*_ptr == '/') && (*next == '*')) {    // C comment
  4886       _ptr++; next++;
  4887       do {
  4888         _ptr++; next++;
  4889         if (*_ptr == '\n') {               // keep proper track of new lines
  4890           next_line();                     // skip newlines within comments
  4891           if (_curline == NULL) {          // check for end of file
  4892             parse_err(SYNERR, "end-of-file detected inside comment\n");
  4893             break;
  4895           _ptr = _curline; next = _ptr + 1;
  4897       } while(!((*_ptr == '*') && (*next == '/'))); // Go to end of comment
  4898       _ptr = ++next; next++;               // increment _ptr past comment end
  4900     else if (do_preproc && *_ptr == '#') {
  4901       // Note that this calls skipws_common(false) recursively!
  4902       bool preproc_handled = handle_preproc_token();
  4903       if (!preproc_handled) {
  4904         if (preproc_taken()) {
  4905           return;  // short circuit
  4907         ++_ptr;    // skip the preprocessor character
  4909       next = _ptr+1;
  4910     } else if(*_ptr > ' ' && !(do_preproc && !preproc_taken())) {
  4911       break;
  4913     else if (*_ptr == '"' || *_ptr == '\'') {
  4914       assert(do_preproc, "only skip strings if doing preproc");
  4915       // skip untaken quoted string
  4916       int qchar = *_ptr;
  4917       while (true) {
  4918         ++_ptr;
  4919         if (*_ptr == qchar) { ++_ptr; break; }
  4920         if (*_ptr == '\\')  ++_ptr;
  4921         if (*_ptr == '\n' || *_ptr == '\0') {
  4922           parse_err(SYNERR, "newline in string");
  4923           break;
  4926       next = _ptr + 1;
  4928     else { ++_ptr; ++next; }
  4930   if( _curline != NULL )            // at end of file _curchar isn't valid
  4931     _curchar = *_ptr;               // reset _curchar to maintain invariant
  4934 //---------------------------cur_char-----------------------------------------
  4935 char ADLParser::cur_char() {
  4936   return (_curchar);
  4939 //---------------------------next_char-----------------------------------------
  4940 void ADLParser::next_char() {
  4941   if (_curchar == '\n')  parse_err(WARN, "must call next_line!");
  4942   _curchar = *++_ptr;
  4943   // if ( _curchar == '\n' ) {
  4944   //   next_line();
  4945   // }
  4948 //---------------------------next_char_or_line---------------------------------
  4949 void ADLParser::next_char_or_line() {
  4950   if ( _curchar != '\n' ) {
  4951     _curchar = *++_ptr;
  4952   } else {
  4953     next_line();
  4954     _ptr = _curline;
  4955     _curchar = *_ptr;  // maintain invariant
  4959 //---------------------------next_line-----------------------------------------
  4960 void ADLParser::next_line() {
  4961   _curline = _buf.get_line();
  4962   _curchar = ' ';
  4965 //------------------------get_line_string--------------------------------------
  4966 // Prepended location descriptor, for debugging.
  4967 // Must return a malloced string (that can be freed if desired).
  4968 char* ADLParser::get_line_string(int linenum) {
  4969   const char* file = _AD._ADL_file._name;
  4970   int         line = linenum ? linenum : this->linenum();
  4971   char* location = (char *)malloc(strlen(file) + 100);
  4972   sprintf(location, "\n#line %d \"%s\"\n", line, file);
  4973   return location;
  4976 //-------------------------is_literal_constant---------------------------------
  4977 bool ADLParser::is_literal_constant(const char *param) {
  4978   if (param[0] == 0)     return false;  // null string
  4979   if (param[0] == '(')   return true;   // parenthesized expression
  4980   if (param[0] == '0' && (param[1] == 'x' || param[1] == 'X')) {
  4981     // Make sure it's a hex constant.
  4982     int i = 2;
  4983     do {
  4984       if( !ADLParser::is_hex_digit(*(param+i)) )  return false;
  4985       ++i;
  4986     } while( *(param+i) != 0 );
  4987     return true;
  4989   return false;
  4992 //---------------------------is_hex_digit--------------------------------------
  4993 bool ADLParser::is_hex_digit(char digit) {
  4994   return ((digit >= '0') && (digit <= '9'))
  4995        ||((digit >= 'a') && (digit <= 'f'))
  4996        ||((digit >= 'A') && (digit <= 'F'));
  4999 //---------------------------is_int_token--------------------------------------
  5000 bool ADLParser::is_int_token(const char* token, int& intval) {
  5001   const char* cp = token;
  5002   while (*cp != '\0' && *cp <= ' ')  cp++;
  5003   if (*cp == '-')  cp++;
  5004   int ndigit = 0;
  5005   while (*cp >= '0' && *cp <= '9')  { cp++; ndigit++; }
  5006   while (*cp != '\0' && *cp <= ' ')  cp++;
  5007   if (ndigit == 0 || *cp != '\0') {
  5008     return false;
  5010   intval = atoi(token);
  5011   return true;
  5014 static const char* skip_expr_ws(const char* str) {
  5015   const char * cp = str;
  5016   while (cp[0]) {
  5017     if (cp[0] <= ' ') {
  5018       ++cp;
  5019     } else if (cp[0] == '#') {
  5020       ++cp;
  5021       while (cp[0] == ' ')  ++cp;
  5022       assert(0 == strncmp(cp, "line", 4), "must be a #line directive");
  5023       const char* eol = strchr(cp, '\n');
  5024       assert(eol != NULL, "must find end of line");
  5025       if (eol == NULL)  eol = cp + strlen(cp);
  5026       cp = eol;
  5027     } else {
  5028       break;
  5031   return cp;
  5034 //-----------------------equivalent_expressions--------------------------------
  5035 bool ADLParser::equivalent_expressions(const char* str1, const char* str2) {
  5036   if (str1 == str2)
  5037     return true;
  5038   else if (str1 == NULL || str2 == NULL)
  5039     return false;
  5040   const char* cp1 = str1;
  5041   const char* cp2 = str2;
  5042   char in_quote = '\0';
  5043   while (cp1[0] && cp2[0]) {
  5044     if (!in_quote) {
  5045       // skip spaces and/or cpp directives
  5046       const char* cp1a = skip_expr_ws(cp1);
  5047       const char* cp2a = skip_expr_ws(cp2);
  5048       if (cp1a > cp1 && cp2a > cp2) {
  5049         cp1 = cp1a; cp2 = cp2a;
  5050         continue;
  5052       if (cp1a > cp1 || cp2a > cp2)  break; // fail
  5054     // match one non-space char
  5055     if (cp1[0] != cp2[0])  break; // fail
  5056     char ch = cp1[0];
  5057     cp1++; cp2++;
  5058     // watch for quotes
  5059     if (in_quote && ch == '\\') {
  5060       if (cp1[0] != cp2[0])  break; // fail
  5061       if (!cp1[0])  break;
  5062       cp1++; cp2++;
  5064     if (in_quote && ch == in_quote) {
  5065       in_quote = '\0';
  5066     } else if (!in_quote && (ch == '"' || ch == '\'')) {
  5067       in_quote = ch;
  5070   return (!cp1[0] && !cp2[0]);
  5074 //-------------------------------trim------------------------------------------
  5075 void ADLParser::trim(char* &token) {
  5076   while (*token <= ' ')  token++;
  5077   char* end = token + strlen(token);
  5078   while (end > token && *(end-1) <= ' ')  --end;
  5079   *end = '\0';

mercurial