Initialize assignments of reals

[iverilog.git] / elab_lval.cc

/*
 * Copyright (c) 2000-2006 Stephen Williams (steve@icarus.com)
 *
 *    This source code is free software; you can redistribute it
 *    and/or modify it in source code form under the terms of the GNU
 *    General Public License as published by the Free Software
 *    Foundation; either version 2 of the License, or (at your option)
 *    any later version.
 *
 *    This program is distributed in the hope that it will be useful,
 *    but WITHOUT ANY WARRANTY; without even the implied warranty of
 *    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 *    GNU General Public License for more details.
 *
 *    You should have received a copy of the GNU General Public License
 *    along with this program; if not, write to the Free Software
 *    Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA
 */
#ifdef HAVE_CVS_IDENT
#ident "$Id: elab_lval.cc,v 1.44 2007/06/02 03:42:12 steve Exp $"
#endif

# include "config.h"

# include  "PExpr.h"
# include  "netlist.h"
# include  "netmisc.h"
# include  "compiler.h"
# include  <iostream>
# include  "ivl_assert.h"

/*
 * These methods generate a NetAssign_ object for the l-value of the
 * assignment. This is common code for the = and <= statements.
 *
 * What gets generated depends on the structure of the l-value. If the
 * l-value is a simple name (i.e., foo <= <value>) the the NetAssign_
 * is created the width of the foo reg and connected to all the
 * bits.
 *
 * If there is a part select (i.e., foo[3:1] <= <value>) the NetAssign_
 * is made only as wide as it needs to be (3 bits in this example) and
 * connected to the correct bits of foo. A constant bit select is a
 * special case of the part select.
 *
 * If the bit-select is non-constant (i.e., foo[<expr>] = <value>) the
 * NetAssign_ is made wide enough to connect to all the bits of foo,
 * then the mux expression is elaborated and attached to the
 * NetAssign_ node as a b_mux value. The target must interpret the
 * presence of a bmux value as taking a single bit and assigning it to
 * the bit selected by the bmux expression.
 *
 * If the l-value expression is non-trivial, but can be fully
 * evaluated at compile time (meaning any bit selects are constant)
 * then elaboration will make a single NetAssign_ that connects to a
 * synthetic reg that in turn connects to all the proper pins of the
 * l-value.
 *
 * This last case can turn up in statements like: {a, b[1]} = c;
 * rather then create a NetAssign_ for each item in the concatenation,
 * elaboration makes a single NetAssign_ and connects it up properly.
 */


/*
 * The default interpretation of an l-value to a procedural assignment
 * is to try to make a net elaboration, and see if the result is
 * suitable for assignment.
 */
NetAssign_* PExpr::elaborate_lval(Design*des,
                                  NetScope*scope,
                                  bool is_force) const
{
      NetNet*ll = 0;
      if (ll == 0) {
            cerr << get_line() << ": Assignment l-value too complex."
                 << endl;
            return 0;
      }

      NetAssign_*lv = new NetAssign_(ll);
      return lv;
}

/*
 * Concatenation expressions can appear as l-values. Handle them here.
 *
 * If adjacent l-values in the concatenation are not bit selects, then
 * merge them into a single NetAssign_ object. This can happen is code
 * like ``{ ...a, b, ...}''. As long as "a" and "b" do not have bit
 * selects (or the bit selects are constant) we can merge the
 * NetAssign_ objects.
 *
 * Be careful to get the bit order right. In the expression ``{a, b}''
 * a is the MSB and b the LSB. Connect the LSB to the low pins of the
 * NetAssign_ object.
 */
NetAssign_* PEConcat::elaborate_lval(Design*des,
                                     NetScope*scope,
                                     bool is_force) const
{
      if (repeat_) {
            cerr << get_line() << ": error: Repeat concatenations make "
                  "no sense in l-value expressions. I refuse." << endl;
            des->errors += 1;
            return 0;
      }

      NetAssign_*res = 0;

      for (unsigned idx = 0 ;  idx < parms_.count() ;  idx += 1) {

            if (parms_[idx] == 0) {
                  cerr << get_line() << ": error: Empty expressions "
                       << "not allowed in concatenations." << endl;
                  des->errors += 1;
                  continue;
            }

            NetAssign_*tmp = parms_[idx]->elaborate_lval(des, scope, is_force);

              /* If the l-value doesn't elaborate, the error was
                 already detected and printed. We just skip it and let
                 the compiler catch more errors. */
            if (tmp == 0)
                  continue;

            assert(tmp);


              /* Link the new l-value to the previous one. */

            NetAssign_*last = tmp;
            while (last->more)
                  last = last->more;

            last->more = res;
            res = tmp;
      }

      return res;
}

/*
 * Handle the ident as an l-value. This includes bit and part selects
 * of that ident.
 */
NetAssign_* PEIdent::elaborate_lval(Design*des,
                                    NetScope*scope,
                                    bool is_force) const
{
      NetNet*       reg = 0;
      const NetExpr*par = 0;
      NetEvent*     eve = 0;

      symbol_search(des, scope, path_, reg, par, eve);
      if (reg == 0) {
            cerr << get_line() << ": error: Could not find variable ``"
                 << path_ << "'' in ``" << scope_path(scope) <<
                  "''" << endl;

            des->errors += 1;
            return 0;
      }

      ivl_assert(*this, reg);

      const name_component_t&name_tail = path_.back();

      index_component_t::ctype_t use_sel = index_component_t::SEL_NONE;
      if (!name_tail.index.empty())
            use_sel = name_tail.index.back().sel;

        // This is the special case that the l-value is an entire
        // memory. This is, in fact, an error.
      if (reg->array_dimensions() > 0 && name_tail.index.empty()) {
            cerr << get_line() << ": error: Cannot assign to array "
                 << path_ << ". Did you forget a word index?" << endl;
            des->errors += 1;
            return 0;
      }

      if (reg->array_dimensions() > 0)
            return elaborate_lval_net_word_(des, scope, reg);

      if (use_sel == index_component_t::SEL_PART) {
            NetAssign_*lv = new NetAssign_(reg);
            elaborate_lval_net_part_(des, scope, lv);
            return lv;
      }

      if (use_sel == index_component_t::SEL_IDX_UP) {
            NetAssign_*lv = new NetAssign_(reg);
            elaborate_lval_net_idx_up_(des, scope, lv);
            return lv;
      }

      if (use_sel == index_component_t::SEL_IDX_DO) {
            NetAssign_*lv = new NetAssign_(reg);
            elaborate_lval_net_idx_do_(des, scope, lv);
            return lv;
      }

        /* Get the signal referenced by the identifier, and make sure
           it is a register. Wires are not allows in this context,
           unless this is the l-value of a force. */
      if ((reg->type() != NetNet::REG) && !is_force) {
            cerr << get_line() << ": error: " << path_ <<
                  " is not a valid l-value in " << scope_path(scope) <<
                  "." << endl;
            cerr << reg->get_line() << ":      : " << path_ <<
                  " is declared here as " << reg->type() << "." << endl;
            des->errors += 1;
            return 0;
      }

      long msb, lsb;
      NetExpr*mux;

      if (use_sel == index_component_t::SEL_BIT) {

            const index_component_t&index_tail = name_tail.index.back();
            ivl_assert(*this, index_tail.msb != 0);
            ivl_assert(*this, index_tail.lsb == 0);

              /* If there is only a single select expression, it is a
                 bit select. Evaluate the constant value and treat it
                 as a part select with a bit width of 1. If the
                 expression it not constant, then return the
                 expression as a mux. */

            NetExpr*index_expr = elab_and_eval(des, scope, index_tail.msb, -1);

            if (NetEConst*index_con = dynamic_cast<NetEConst*> (index_expr)) {
                  msb = index_con->value().as_long();
                  lsb = index_con->value().as_long();
                  mux = 0;

            } else {
                  msb = 0;
                  lsb = 0;
                  mux = index_expr;
            }

      } else {

              /* No select expressions, so presume a part select the
                 width of the register. */

            msb = reg->msb();
            lsb = reg->lsb();
            mux = 0;
      }


      NetAssign_*lv;
      if (mux) {

              /* If there is a non-constant bit select, make a
                 NetAssign_ to the target reg and attach a
                 bmux to select the target bit. */
            lv = new NetAssign_(reg);

              /* Correct the mux for the range of the vector. */
            if (reg->msb() < reg->lsb())
                  mux = make_sub_expr(reg->lsb(), mux);
            else if (reg->lsb() != 0)
                  mux = make_add_expr(mux, - reg->lsb());

            lv->set_part(mux, 1);

      } else if (msb == reg->msb() && lsb == reg->lsb()) {

              /* No bit select, and part select covers the entire
                 vector. Simplest case. */
            lv = new NetAssign_(reg);

      } else {

              /* If the bit/part select is constant, then make the
                 NetAssign_ only as wide as it needs to be and connect
                 only to the selected bits of the reg. */
            unsigned loff = reg->sb_to_idx(lsb);
            unsigned moff = reg->sb_to_idx(msb);
            unsigned wid = moff - loff + 1;

            if (moff < loff) {
                  cerr << get_line() << ": error: part select "
                       << reg->name() << "[" << msb<<":"<<lsb<<"]"
                       << " is reversed." << endl;
                  des->errors += 1;
                  return 0;
            }

              /* If the part select extends beyond the extreme of the
                 variable, then report an error. Note that loff is
                 converted to normalized form so is relative the
                 variable pins. */

            if ((wid + loff) > reg->vector_width()) {
                  cerr << get_line() << ": error: bit/part select "
                       << reg->name() << "[" << msb<<":"<<lsb<<"]"
                       << " is out of range." << endl;
                  des->errors += 1;
                  return 0;
            }

            lv = new NetAssign_(reg);
            lv->set_part(new NetEConst(verinum(loff)), wid);
      }


      return lv;
}

NetAssign_* PEIdent::elaborate_lval_net_word_(Design*des,
                                              NetScope*scope,
                                              NetNet*reg) const
{
      const name_component_t&name_tail = path_.back();
      ivl_assert(*this, !name_tail.index.empty());

      const index_component_t&index_head = name_tail.index.front();
      ivl_assert(*this, index_head.sel == index_component_t::SEL_BIT);
      ivl_assert(*this, index_head.msb != 0);
      ivl_assert(*this, index_head.lsb == 0);

      NetExpr*word = elab_and_eval(des, scope, index_head.msb, -1);

        // If there is a non-zero base to the memory, then build an
        // expression to calculate the canonical address.
      if (long base = reg->array_first()) {

            word = make_add_expr(word, 0-base);
            if (NetExpr*tmp = word->eval_tree()) {
                  word = tmp;
            }
      }

      NetAssign_*lv = new NetAssign_(reg);
      lv->set_word(word);

      if (debug_elaborate)
            cerr << get_line() << ": debug: Set array word=" << *word << endl;

        // Test for the case that the index is a constant, and is out
        // of bounds. The "word" expression is the word index already
        // converted to canonical address, so this just needs to check
        // that the address is not too big.
      if (NetEConst*word_const = dynamic_cast<NetEConst*>(word)) {
            verinum word_val = word_const->value();
            long index = word_val.as_long();
            if (index < 0 || index >= reg->array_count()) {
                  cerr << get_line() << ": warning: Constant array index "
                       << (index + reg->array_first())
                       << " is out of range for array "
                       << reg->name() << "." << endl;
            }
      }

        /* An array word may also have part selects applied to them. */

      index_component_t::ctype_t use_sel = index_component_t::SEL_NONE;
      if (name_tail.index.size() > 1)
            use_sel = name_tail.index.back().sel;

      if (use_sel == index_component_t::SEL_PART)
            elaborate_lval_net_part_(des, scope, lv);

      if (use_sel == index_component_t::SEL_IDX_UP)
            elaborate_lval_net_idx_up_(des, scope, lv);

      if (use_sel == index_component_t::SEL_IDX_DO)
            elaborate_lval_net_idx_do_(des, scope, lv);

      return lv;
}

bool PEIdent::elaborate_lval_net_part_(Design*des,
                                       NetScope*scope,
                                       NetAssign_*lv) const
{
      long msb, lsb;
      bool flag = calculate_parts_(des, scope, msb, lsb);
      if (!flag)
            return false;

      NetNet*reg = lv->sig();
      assert(reg);

      if (msb == reg->msb() && lsb == reg->lsb()) {

              /* No bit select, and part select covers the entire
                 vector. Simplest case. */

      } else {

              /* If the bit/part select is constant, then make the
                 NetAssign_ only as wide as it needs to be and connect
                 only to the selected bits of the reg. */
            unsigned loff = reg->sb_to_idx(lsb);
            unsigned moff = reg->sb_to_idx(msb);
            unsigned wid = moff - loff + 1;

            if (moff < loff) {
                  cerr << get_line() << ": error: part select "
                       << reg->name() << "[" << msb<<":"<<lsb<<"]"
                       << " is reversed." << endl;
                  des->errors += 1;
                  return false;
            }

              /* If the part select extends beyond the extreme of the
                 variable, then report an error. Note that loff is
                 converted to normalized form so is relative the
                 variable pins. */

            if ((wid + loff) > reg->vector_width()) {
                  cerr << get_line() << ": error: bit/part select "
                       << reg->name() << "[" << msb<<":"<<lsb<<"]"
                       << " is out of range." << endl;
                  des->errors += 1;
                  return false;
            }

            lv->set_part(new NetEConst(verinum(loff)), wid);
      }

      return true;
}

bool PEIdent::elaborate_lval_net_idx_up_(Design*des,
                                         NetScope*scope,
                                         NetAssign_*lv) const
{
      const name_component_t&name_tail = path_.back();;
      ivl_assert(*this, !name_tail.index.empty());

      const index_component_t&index_tail = name_tail.index.back();
      ivl_assert(*this, index_tail.msb != 0);
      ivl_assert(*this, index_tail.lsb != 0);

      NetNet*reg = lv->sig();
      assert(reg);

      if (reg->type() != NetNet::REG) {
            cerr << get_line() << ": error: " << path_ <<
                  " is not a reg/integer/time in " << scope_path(scope) <<
                  "." << endl;
            cerr << reg->get_line() << ":      : " << path_ <<
                  " is declared here as " << reg->type() << "." << endl;
            des->errors += 1;
            return false;
      }

      unsigned long wid;
      calculate_up_do_width_(des, scope, wid);

      NetExpr*base = elab_and_eval(des, scope, index_tail.msb, -1);

        /* Correct the mux for the range of the vector. */
      if (reg->msb() < reg->lsb())
            base = make_sub_expr(reg->lsb(), base);
      else if (reg->lsb() != 0)
            base = make_add_expr(base, - reg->lsb());

      if (debug_elaborate)
            cerr << get_line() << ": debug: Set part select width="
                 << wid << ", base=" << *base << endl;

      lv->set_part(base, wid);

      return true;
}

bool PEIdent::elaborate_lval_net_idx_do_(Design*des,
                                         NetScope*scope,
                                         NetAssign_*lv) const
{
      const name_component_t&name_tail = path_.back();;
      ivl_assert(*this, !name_tail.index.empty());

      const index_component_t&index_tail = name_tail.index.back();
      ivl_assert(*this, index_tail.msb != 0);
      ivl_assert(*this, index_tail.lsb != 0);

      cerr << get_line() << ": internal error: don't know how to "
            "deal with SEL_IDX_DO in lval?" << endl;
      des->errors += 1;
      return false;
}

NetAssign_* PENumber::elaborate_lval(Design*des, NetScope*, bool) const
{
      cerr << get_line() << ": error: Constant values not allowed "
           << "in l-value expressions." << endl;
      des->errors += 1;
      return 0;
}

/*
 * $Log: elab_lval.cc,v $
 * Revision 1.44  2007/06/02 03:42:12  steve
 *  Properly evaluate scope path expressions.
 *
 * Revision 1.43  2007/05/24 04:07:11  steve
 *  Rework the heirarchical identifier parse syntax and pform
 *  to handle more general combinations of heirarch and bit selects.
 *
 * Revision 1.42  2007/03/14 05:06:49  steve
 *  Replace some asserts with ivl_asserts.
 *
 * Revision 1.41  2007/03/05 05:59:10  steve
 *  Handle processes within generate loops.
 *
 * Revision 1.40  2007/02/27 05:14:38  steve
 *  Detect and warn about lval array index out fo bounds.
 *
 * Revision 1.39  2007/02/01 05:25:26  steve
 *  Error message better reflects more general reality.
 *
 * Revision 1.38  2007/01/16 05:44:15  steve
 *  Major rework of array handling. Memories are replaced with the
 *  more general concept of arrays. The NetMemory and NetEMemory
 *  classes are removed from the ivl core program, and the IVL_LPM_RAM
 *  lpm type is removed from the ivl_target API.
 *
 * Revision 1.37  2006/11/04 06:19:25  steve
 *  Remove last bits of relax_width methods, and use test_width
 *  to calculate the width of an r-value expression that may
 *  contain unsized numbers.
 *
 * Revision 1.36  2006/06/02 04:48:50  steve
 *  Make elaborate_expr methods aware of the width that the context
 *  requires of it. In the process, fix sizing of the width of unary
 *  minus is context determined sizes.
 *
 * Revision 1.35  2006/04/16 00:54:04  steve
 *  Cleanup lval part select handling.
 *
 * Revision 1.34  2006/04/16 00:15:43  steve
 *  Fix part selects in l-values.
 *
 * Revision 1.33  2006/02/02 02:43:57  steve
 *  Allow part selects of memory words in l-values.
 *
 * Revision 1.32  2005/07/11 16:56:50  steve
 *  Remove NetVariable and ivl_variable_t structures.
 *
 * Revision 1.31  2004/12/29 23:55:43  steve
 *  Unify elaboration of l-values for all proceedural assignments,
 *  including assing, cassign and force.
 *
 *  Generate NetConcat devices for gate outputs that feed into a
 *  vector results. Use this to hande gate arrays. Also let gate
 *  arrays handle vectors of gates when the outputs allow for it.
 *
 * Revision 1.30  2004/12/11 02:31:25  steve
 *  Rework of internals to carry vectors through nexus instead
 *  of single bits. Make the ivl, tgt-vvp and vvp initial changes
 *  down this path.
 *
 * Revision 1.29  2004/10/04 01:10:52  steve
 *  Clean up spurious trailing white space.
 *
 * Revision 1.28  2004/08/28 14:59:44  steve
 *  More detailed error message about bad variable.
 *
 * Revision 1.27  2003/09/19 03:30:05  steve
 *  Fix name search in elab_lval.
 */