16099227452

Committed 06 Jul 2025 12:45PM UTC coverage: 92.335% (+0.05%) from 92.284%

Build # 16099227452

Build Type

push

github

Committed by

nickg

Commit Message

Handle underscores in Verilog decimal literals

Fixes #1230

Run Details

18 of 20 new or added lines in 1 file covered. (90.0%)

598 existing lines in 16 files now uncovered.

71069 of 76969 relevant lines covered (92.33%)

564781.41 hits per line

Source File
Press 'n' to go to next uncovered line, 'b' for previous

94.07

/src/rt/model.c

//
//  Copyright (C) 2011-2024  Nick Gasson
//
//  This program is free software: you can redistribute it and/or modify
//  it under the terms of the GNU General Public License as published by
//  the Free Software Foundation, either version 3 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, see <http://www.gnu.org/licenses/>.
//

#include "util.h"
#include "array.h"
#include "common.h"
#include "debug.h"
#include "hash.h"
#include "jit/jit-exits.h"
#include "jit/jit.h"
#include "lib.h"
#include "option.h"
#include "psl/psl-node.h"
#include "rt/assert.h"
#include "rt/copy.h"
#include "rt/heap.h"
#include "rt/model.h"
#include "rt/random.h"
#include "rt/structs.h"
#include "thread.h"
#include "tree.h"
#include "type.h"
#include "vlog/vlog-node.h"
#include "vlog/vlog-util.h"

#include <assert.h>
#include <inttypes.h>
#include <stdlib.h>
#include <string.h>

typedef struct _rt_callback rt_callback_t;
typedef struct _memblock memblock_t;

typedef struct _rt_callback {
   rt_event_fn_t  fn;
   void          *user;
   rt_callback_t *next;
} rt_callback_t;

typedef enum {
   EVENT_TIMEOUT,
   EVENT_DRIVER,
   EVENT_PROCESS,
} event_kind_t;

#define MEMBLOCK_ALIGN   64
#define MEMBLOCK_PAGE_SZ 0x800000
#define TRIGGER_TAB_SIZE 64

#if ASAN_ENABLED
#define MEMBLOCK_REDZONE 16
#else
#define MEMBLOCK_REDZONE 0
#endif

typedef struct _memblock {
   memblock_t *chain;
   size_t      alloc;
   size_t      limit;
   uint8_t     data[];
} memblock_t;

STATIC_ASSERT(sizeof(memblock_t) <= MEMBLOCK_ALIGN);

typedef struct {
   waveform_t    *free_waveforms;
   tlab_t        *tlab;
   rt_wakeable_t *active_obj;
   rt_scope_t    *active_scope;
} __attribute__((aligned(64))) model_thread_t;

typedef void (*defer_fn_t)(rt_model_t *, void *);

typedef struct {
   defer_fn_t  fn;
   void       *arg;
} defer_task_t;

typedef struct {
   defer_task_t *tasks;
   unsigned      count;
   unsigned      max;
} deferq_t;

typedef struct _rt_model {
   tree_t             top;
   hash_t            *scopes;
   rt_scope_t        *root;
   mspace_t          *mspace;
   jit_t             *jit;
   rt_nexus_t        *nexuses;
   rt_nexus_t       **nexus_tail;
   delta_cycle_t      stop_delta;
   int                iteration;
   uint64_t           now;
   bool               can_create_delta;
   bool               next_is_delta;
   bool               force_stop;
   unsigned           n_signals;
   heap_t            *eventq_heap;
   ihash_t           *res_memo;
   rt_watch_t        *watches;
   deferq_t           procq;
   deferq_t           delta_procq;
   deferq_t           driverq;
   deferq_t           delta_driverq;
   deferq_t           postponedq;
   deferq_t           implicitq;
   heap_t            *driving_heap;
   heap_t            *effective_heap;
   rt_callback_t     *global_cbs[RT_LAST_EVENT];
   cover_data_t      *cover;
   nvc_rusage_t       ready_rusage;
   nvc_lock_t         memlock;
   memblock_t        *memblocks;
   model_thread_t    *threads[MAX_THREADS];
   signal_list_t      eventsigs;
   bool               shuffle;
   bool               liveness;
   rt_trigger_t      *triggertab[TRIGGER_TAB_SIZE];
} rt_model_t;

#define FMT_VALUES_SZ   128
#define NEXUS_INDEX_MIN 8
#define TRACE_SIGNALS   1
#define WAVEFORM_CHUNK  256
#define PENDING_MIN     4
#define MAX_RANK        UINT8_MAX

#define TRACE(...) do {                                 \
      if (unlikely(__trace_on))                         \
         __model_trace(get_model(), __VA_ARGS__);       \
   } while (0)

#define MODEL_ENTRY(m)                                                  \
   rt_model_t *__save __attribute__((unused, cleanup(__model_exit)));   \
   __model_entry(m, &__save);                                           \

#if USE_EMUTLS
static rt_model_t *__model = NULL;
#else
static __thread rt_model_t *__model = NULL;
#endif

static bool __trace_on = false;

static void *source_value(rt_nexus_t *nexus, rt_source_t *src);
static void free_value(rt_nexus_t *n, rt_value_t v);
static rt_nexus_t *clone_nexus(rt_model_t *m, rt_nexus_t *old, int offset);
static void put_driving(rt_model_t *m, rt_nexus_t *n, const void *value);
static void put_effective(rt_model_t *m, rt_nexus_t *n, const void *value);
static void update_implicit_signal(rt_model_t *m, rt_implicit_t *imp);
static bool run_trigger(rt_model_t *m, rt_trigger_t *t);
static void reset_scope(rt_model_t *m, rt_scope_t *s);
static void async_run_process(rt_model_t *m, void *arg);
static void async_update_property(rt_model_t *m, void *arg);
static void async_update_driver(rt_model_t *m, void *arg);
static void async_fast_driver(rt_model_t *m, void *arg);
static void async_fast_all_drivers(rt_model_t *m, void *arg);
static void async_pseudo_source(rt_model_t *m, void *arg);
static void async_transfer_signal(rt_model_t *m, void *arg);
static void async_update_implicit_signal(rt_model_t *m, void *arg);

static int fmt_time_r(char *buf, size_t len, int64_t t, const char *sep)
{
   static const struct {
      int64_t time;
      const char *unit;
   } units[] = {
      { INT64_C(1), "fs" },
      { INT64_C(1000), "ps" },
      { INT64_C(1000000), "ns" },
      { INT64_C(1000000000), "us" },
      { INT64_C(1000000000000), "ms" },
      { 0, NULL }
   };

   int u = 0;
   while (units[u + 1].unit && (t % units[u + 1].time == 0))
      ++u;

   return checked_sprintf(buf, len, "%"PRIi64"%s%s",
                          t / units[u].time, sep, units[u].unit);
}

__attribute__((format(printf, 2, 3)))
static void __model_trace(rt_model_t *m, const char *fmt, ...)
{
   va_list ap;
   va_start(ap, fmt);

   static nvc_lock_t lock = 0;
   {
      SCOPED_LOCK(lock);

      if (m->iteration < 0)
         fprintf(stderr, "TRACE (init): ");
      else {
         char buf[64];
         fmt_time_r(buf, sizeof(buf), m->now, "");
         fprintf(stderr, "TRACE %s+%d: ", buf, m->iteration);
      }
      vfprintf(stderr, fmt, ap);
      fprintf(stderr, "\n");
      fflush(stderr);
   }

   va_end(ap);
}

static const char *trace_time(uint64_t value)
{
   static __thread char buf[2][32];
   static __thread int which = 0;

   which ^= 1;
   fmt_time_r(buf[which], 32, value, "");
   return buf[which];
}

static const char *trace_states(bit_mask_t *mask)
{
   static __thread text_buf_t *tb = NULL;

   if (tb == NULL)
      tb = tb_new();

   tb_rewind(tb);
   tb_append(tb, '{');

   size_t bit = -1;
   while (mask_iter(mask, &bit))
      tb_printf(tb, "%s%zd", tb_len(tb) > 1 ? "," : "", bit);

   tb_append(tb, '}');

   return tb_get(tb);
}

static const char *trace_nexus(rt_nexus_t *n)
{
   static __thread text_buf_t *tb = NULL;

   if (tb == NULL)
      tb = tb_new();

   tb_rewind(tb);

   if (is_signal_scope(n->signal->parent))
      tb_printf(tb, "%s.", istr(n->signal->parent->name));

   tb_istr(tb, tree_ident(n->signal->where));

   if (n->width * n->size < n->signal->shared.size)
      tb_printf(tb, "[%d:%d]", n->offset, n->offset + n->width - 1);

   return tb_get(tb);
}

static void model_diag_cb(diag_t *d, void *arg)
{
   rt_model_t *m = arg;

   if (m->iteration < 0)
      diag_printf(d, "(init): ");
   else  {
      char tmbuf[64];
      fmt_time_r(tmbuf, sizeof(tmbuf), m->now, "");

      diag_printf(d, "%s+%d: ", tmbuf, m->iteration);
   }
}

static void __model_entry(rt_model_t *m, rt_model_t **save)
{
   if (__model == NULL)
      diag_add_hint_fn(model_diag_cb, m);

   *save = __model;
   __model = m;
}

static void __model_exit(rt_model_t **save)
{
   __model = *save;
   *save = NULL;

   if (__model == NULL)
      diag_remove_hint_fn(model_diag_cb);
}

static char *fmt_values_r(const void *values, size_t len, char *buf, size_t max)
{
   char *p = buf;
   const uint8_t *vptr = values;

   for (unsigned i = 0; i < len; i++) {
      if (buf + max - p <= 5) {
         checked_sprintf(p, buf + max - p, "...");
         break;
      }
      else
         p += checked_sprintf(p, buf + max - p, "%02x", *vptr++);
   }

   return buf;
}

static const char *fmt_nexus(rt_nexus_t *n, const void *values)
{
   static char buf[FMT_VALUES_SZ*2 + 2];
   return fmt_values_r(values, n->size * n->width, buf, sizeof(buf));
}

static const char *fmt_values(const void *values, uint32_t len)
{
   static char buf[FMT_VALUES_SZ*2 + 2];
   return fmt_values_r(values, len, buf, sizeof(buf));
}

static const char *fmt_jit_value(jit_scalar_t value, bool scalar, uint32_t len)
{
   static char buf[FMT_VALUES_SZ*2 + 2];
   if (scalar) {
      checked_sprintf(buf, sizeof(buf), "%"PRIx64, value.integer);
      return buf;
   }
   else
      return fmt_values_r(value.pointer, len, buf, sizeof(buf));
}

static model_thread_t *model_thread(rt_model_t *m)
{
#if RT_MULTITHREADED
   const int my_id = thread_id();

   if (unlikely(m->threads[my_id] == NULL))
      return (m->threads[my_id] = xcalloc(sizeof(model_thread_t)));

   return m->threads[my_id];
#else
   assert(thread_id() == 0);
   return m->threads[0];
#endif
}

__attribute__((cold, noinline))
static void deferq_grow(deferq_t *dq)
{
   dq->max = MAX(dq->max * 2, 64);
   dq->tasks = xrealloc_array(dq->tasks, dq->max, sizeof(defer_task_t));
}

static inline void deferq_do(deferq_t *dq, defer_fn_t fn, void *arg)
{
   if (unlikely(dq->count == dq->max))
      deferq_grow(dq);

   dq->tasks[dq->count++] = (defer_task_t){ fn, arg };
}

static void deferq_scan(deferq_t *dq, scan_fn_t fn, void *arg)
{
   for (int i = 0; i < dq->count; i++)
      (*fn)(dq->tasks[i].fn, dq->tasks[i].arg, arg);
}

static void deferq_shuffle(deferq_t *dq)
{
   int cur = dq->count;
   while (cur > 0) {
      const int swap = get_random() % cur--;
      const defer_task_t tmp = dq->tasks[cur];
      dq->tasks[cur] = dq->tasks[swap];
      dq->tasks[swap] = tmp;
   }
}

static void deferq_run(rt_model_t *m, deferq_t *dq)
{
   const defer_task_t *tasks = dq->tasks;
   const int count = dq->count;

   int i = 0;
   for (; i < count - 1; i++) {
      // Prefetch ahead the next task argument to avoid cache misses
      // when we execute it
      prefetch_read(tasks[i + 1].arg);
      (*tasks[i].fn)(m, tasks[i].arg);
   }
   for (; i < count; i++)
      (*tasks[i].fn)(m, tasks[i].arg);

   assert(dq->tasks == tasks);
   assert(dq->count == count);

   dq->count = 0;
}

static void *static_alloc(rt_model_t *m, size_t size)
{
   const int total_bytes = ALIGN_UP(size + MEMBLOCK_REDZONE, MEMBLOCK_ALIGN);

   RT_LOCK(m->memlock);

   memblock_t *mb = m->memblocks;

   if (mb == NULL || mb->alloc + total_bytes > mb->limit) {
      const size_t pagesz =
         MAX(MEMBLOCK_PAGE_SZ, total_bytes + 2 * MEMBLOCK_ALIGN);

      mb = map_huge_pages(MEMBLOCK_ALIGN, pagesz);
      mb->alloc = MEMBLOCK_ALIGN;
      mb->limit = pagesz - MEMBLOCK_ALIGN;   // Allow overreading in intrinsics

      ASAN_POISON(mb->data, pagesz - sizeof(memblock_t));

      m->memblocks = mb;
   }

   assert((mb->alloc & (MEMBLOCK_ALIGN - 1)) == 0);

   void *ptr = (void *)mb + mb->alloc;
   mb->alloc += total_bytes;

   ASAN_UNPOISON(ptr, size);
   return ptr;
}

static void global_event(rt_model_t *m, rt_event_t kind)
{
   rt_callback_t *list = m->global_cbs[kind];
   m->global_cbs[kind] = NULL;

   for (rt_callback_t *it = list, *tmp; it; it = tmp) {
      tmp = it->next;
      (*it->fn)(m, it->user);
      free(it);
   }
}

static void restore_scopes(rt_model_t *m, tree_t block, rt_scope_t *parent)
{
   rt_scope_t *s = create_scope(m, block, parent);

   const int nstmts = tree_stmts(block);
   for (int i = 0; i < nstmts; i++) {
      tree_t t = tree_stmt(block, i);
      if (tree_kind(t) == T_BLOCK)
         restore_scopes(m, t, s);
   }
}

rt_model_t *model_new(jit_t *jit, cover_data_t *cover)
{
   rt_model_t *m = xcalloc(sizeof(rt_model_t));
   m->scopes      = hash_new(256);
   m->mspace      = jit_get_mspace(jit);
   m->jit         = jit;
   m->nexus_tail  = &(m->nexuses);
   m->iteration   = -1;
   m->eventq_heap = heap_new(512);
   m->res_memo    = ihash_new(128);
   m->cover       = cover;

   m->driving_heap   = heap_new(64);
   m->effective_heap = heap_new(64);

   m->can_create_delta = true;

   m->threads[thread_id()] = static_alloc(m, sizeof(model_thread_t));

   __trace_on = opt_get_int(OPT_RT_TRACE);

   return m;
}

rt_model_t *get_model(void)
{
   assert(__model != NULL);
#ifdef USE_EMUTLS
   assert(thread_id() == 0);
#endif
   return __model;
}

rt_model_t *get_model_or_null(void)
{
   return __model;
}

static rt_wakeable_t *get_active_wakeable(void)
{
   return __model ? model_thread(__model)->active_obj : NULL;
}

rt_proc_t *get_active_proc(void)
{
   rt_wakeable_t *obj = get_active_wakeable();
   if (obj == NULL)
      return NULL;

   assert(obj->kind == W_PROC);
   return container_of(obj, rt_proc_t, wakeable);
}

rt_scope_t *get_active_scope(rt_model_t *m)
{
   return model_thread(m)->active_scope;
}

static void free_waveform(rt_model_t *m, waveform_t *w)
{
   model_thread_t *thread = model_thread(m);
   w->next = thread->free_waveforms;
   thread->free_waveforms = w;
}

static void cleanup_nexus(rt_model_t *m, rt_nexus_t *n)
{
   for (rt_source_t *s = &(n->sources); s; s = s->chain_input) {
      if (s->tag != SOURCE_PORT)
         continue;

      rt_conv_func_t *cf = s->u.port.conv_func;
      if (cf == NULL)
         continue;
      else if (cf->inputs != NULL && cf->inputs != cf->tail) {
         free(cf->inputs);
         s->u.port.conv_func->inputs = NULL;
      }
   }

   if (n->pending != NULL && pointer_tag(n->pending) == 0)
      free(n->pending);
}

static void cleanup_signal(rt_model_t *m, rt_signal_t *s)
{
   rt_nexus_t *n = &(s->nexus), *tmp;
   for (int i = 0; i < s->n_nexus; i++, n = tmp) {
      tmp = n->chain;
      cleanup_nexus(m, n);
   }

   free(s->index);
}

static void cleanup_scope(rt_model_t *m, rt_scope_t *scope)
{
   for (int i = 0; i < scope->procs.count; i++) {
      rt_proc_t *p = scope->procs.items[i];
      mptr_free(m->mspace, &(p->privdata));
      tlab_release(p->tlab);
      free(p);
   }
   ACLEAR(scope->procs);

   for (int i = 0; i < scope->signals.count; i++)
      cleanup_signal(m, scope->signals.items[i]);
   ACLEAR(scope->signals);

   for (int i = 0; i < scope->aliases.count; i++)
      free(scope->aliases.items[i]);
   ACLEAR(scope->aliases);

   for (int i = 0; i < scope->properties.count; i++) {
      rt_prop_t *p = scope->properties.items[i];
      mask_free(&p->state);
      mask_free(&p->newstate);
      mptr_free(m->mspace, &(p->privdata));
      free(p);
   }
   ACLEAR(scope->properties);

   for (int i = 0; i < scope->children.count; i++)
      cleanup_scope(m, scope->children.items[i]);
   ACLEAR(scope->children);

   mptr_free(m->mspace, &(scope->privdata));
   free(scope);
}

void model_free(rt_model_t *m)
{
   if (opt_get_int(OPT_RT_STATS)) {
      nvc_rusage_t ru;
      nvc_rusage(&ru);

      unsigned mem = 0;
      for (memblock_t *mb = m->memblocks; mb; mb = mb->chain)
         mem += mb->alloc;

      notef("setup:%ums run:%ums user:%ums sys:%ums maxrss:%ukB static:%ukB",
            m->ready_rusage.ms, ru.ms, ru.user, ru.sys, ru.rss, mem / 1024);
   }

   while (heap_size(m->eventq_heap) > 0) {
      void *e = heap_extract_min(m->eventq_heap);
      if (pointer_tag(e) == EVENT_TIMEOUT)
         free(untag_pointer(e, rt_callback_t));
   }

   if (m->root != NULL)
      cleanup_scope(m, m->root);

   for (int i = 0; i < MAX_THREADS; i++) {
      model_thread_t *thread = m->threads[i];
      if (thread != NULL)
         tlab_release(thread->tlab);
   }

   free(m->procq.tasks);
   free(m->delta_procq.tasks);
   free(m->postponedq.tasks);
   free(m->implicitq.tasks);
   free(m->driverq.tasks);
   free(m->delta_driverq.tasks);

   for (rt_watch_t *it = m->watches, *tmp; it; it = tmp) {
      tmp = it->chain_all;
      free(it);
   }

   for (int i = 0; i < RT_LAST_EVENT; i++) {
      for (rt_callback_t *it = m->global_cbs[i], *tmp; it; it = tmp) {
         tmp = it->next;
         free(it);
      }
   }

   for (memblock_t *mb = m->memblocks, *tmp; mb; mb = tmp) {
      tmp = mb->chain;
      nvc_munmap(mb, mb->limit + MEMBLOCK_ALIGN);
   }

   heap_free(m->effective_heap);
   heap_free(m->driving_heap);
   heap_free(m->eventq_heap);
   hash_free(m->scopes);
   ihash_free(m->res_memo);
   ACLEAR(m->eventsigs);
   free(m);
}

bool is_signal_scope(rt_scope_t *s)
{
   return s->kind == SCOPE_RECORD || s->kind == SCOPE_ARRAY;
}

rt_signal_t *find_signal(rt_scope_t *scope, tree_t decl)
{
   for (int i = 0; i < scope->signals.count; i++) {
      if (scope->signals.items[i]->where == decl)
         return scope->signals.items[i];
   }

   for (int i = 0; i < scope->aliases.count; i++) {
      if (scope->aliases.items[i]->where == decl)
         return scope->aliases.items[i]->signal;
   }

   return NULL;
}

rt_proc_t *find_proc(rt_scope_t *scope, tree_t proc)
{
   for (int i = 0; i < scope->procs.count; i++) {
      if (scope->procs.items[i]->where == proc)
         return scope->procs.items[i];
   }

   return NULL;
}

rt_watch_t *find_watch(rt_nexus_t *n, sig_event_fn_t fn)
{
   if (n->pending == NULL)
      return NULL;
   else if (pointer_tag(n->pending) == 1) {
      rt_wakeable_t *obj = untag_pointer(n->pending, rt_wakeable_t);
      if (obj->kind == W_WATCH) {
         rt_watch_t *w = container_of(obj, rt_watch_t, wakeable);
         if (w->fn == fn)
            return w;
      }

      return NULL;
   }
   else {
      rt_pending_t *p = untag_pointer(n->pending, rt_pending_t);

      for (int i = 0; i < p->count; i++) {
         rt_wakeable_t *obj = untag_pointer(p->wake[i], rt_wakeable_t);
         if (obj->kind == W_WATCH) {
            rt_watch_t *w = container_of(obj, rt_watch_t, wakeable);
            if (w->fn == fn)
               return w;
         }
      }

      return NULL;
   }
}

rt_scope_t *create_scope(rt_model_t *m, tree_t block, rt_scope_t *parent)
{
   if (parent == NULL) {
      assert(m->top == NULL);
      assert(tree_kind(block) == T_ELAB);

      m->top = block;

      m->root = xcalloc(sizeof(rt_scope_t));
      m->root->kind     = SCOPE_ROOT;
      m->root->where    = block;
      m->root->privdata = MPTR_INVALID;
      m->root->name     = lib_name(lib_work());

      if (tree_stmts(block) > 0)
         restore_scopes(m, tree_stmt(block, 0), m->root);

      return m->root;
   }
   else {
      rt_scope_t *s = xcalloc(sizeof(rt_scope_t));
      s->where    = block;
      s->kind     = SCOPE_INSTANCE;
      s->privdata = mptr_new(m->mspace, "block privdata");
      s->parent   = parent;
      s->name     = ident_prefix(parent->name, tree_ident(block), '.');

      APUSH(parent->children, s);

      hash_put(m->scopes, block, s);

      MODEL_ENTRY(m);

      TRACE("initialise scope %s", istr(s->name));

      model_thread_t *thread = model_thread(m);
      thread->active_scope = s;

      jit_handle_t handle = jit_lazy_compile(m->jit, s->name);
      if (handle == JIT_HANDLE_INVALID)
         fatal_trace("failed to compile %s", istr(s->name));

      jit_scalar_t result, context = { .pointer = NULL };
      jit_scalar_t p2 = { .integer = 0 };

      if (s->parent->kind != SCOPE_ROOT)
         context.pointer = *mptr_get(s->parent->privdata);

      tlab_t tlab = jit_null_tlab(m->jit);

      if (jit_fastcall(m->jit, handle, &result, context, p2, &tlab))
         *mptr_get(s->privdata) = result.pointer;
      else
         m->force_stop = true;

      assert(thread->active_scope == s);
      thread->active_scope = NULL;
      return s;
   }
}

rt_scope_t *find_scope(rt_model_t *m, tree_t container)
{
   return hash_get(m->scopes, container);
}

rt_scope_t *root_scope(rt_model_t *m)
{
   return m->root;
}

rt_scope_t *child_scope(rt_scope_t *scope, tree_t decl)
{
   for (int i = 0; i < scope->children.count; i++) {
      rt_scope_t *s = scope->children.items[i];
      if (s->where == decl)
         return s;
   }

   return NULL;
}

rt_scope_t *child_scope_at(rt_scope_t *scope, int index)
{
   return AGET(scope->children, index);
}

const void *signal_value(rt_signal_t *s)
{
   return s->shared.data;
}

const void *signal_last_value(rt_signal_t *s)
{
   return s->shared.data + s->shared.size;
}

uint8_t signal_size(rt_signal_t *s)
{
   return s->nexus.size;
}

uint32_t signal_width(rt_signal_t *s)
{
   return s->shared.size / s->nexus.size;
}

size_t signal_expand(rt_signal_t *s, uint64_t *buf, size_t max)
{
   const size_t total = s->shared.size / s->nexus.size;

#define SIGNAL_READ_EXPAND_U64(type) do {                               \
      const type *sp = (type *)s->shared.data;                          \
      for (int i = 0; i < max && i < total; i++)                        \
         buf[i] = sp[i];                                                \
   } while (0)

   FOR_ALL_SIZES(s->nexus.size, SIGNAL_READ_EXPAND_U64);

   return total;
}

static inline void set_pending(rt_wakeable_t *wake)
{
   assert(!wake->pending);
   assert(!wake->delayed);
   wake->pending = true;
}

static void deltaq_insert_proc(rt_model_t *m, uint64_t delta, rt_proc_t *proc)
{
   if (delta == 0) {
      set_pending(&proc->wakeable);
      deferq_do(&m->delta_procq, async_run_process, proc);
      m->next_is_delta = true;
   }
   else {
      assert(!proc->wakeable.delayed);
      proc->wakeable.delayed = true;

      void *e = tag_pointer(proc, EVENT_PROCESS);
      heap_insert(m->eventq_heap, m->now + delta, e);
   }
}

static void deltaq_insert_driver(rt_model_t *m, uint64_t delta,
                                 rt_source_t *source)
{
   if (delta == 0) {
      deferq_do(&m->delta_driverq, async_update_driver, source);
      m->next_is_delta = true;
   }
   else {
      void *e = tag_pointer(source, EVENT_DRIVER);
      heap_insert(m->eventq_heap, m->now + delta, e);
   }
}

static void deltaq_insert_pseudo_source(rt_model_t *m, rt_source_t *src)
{
   deferq_do(&m->delta_driverq, async_pseudo_source, src);
   m->next_is_delta = true;
}

static void reset_process(rt_model_t *m, rt_proc_t *proc)
{
   TRACE("reset process %s", istr(proc->name));

   assert(proc->tlab == NULL);
   assert(model_thread(m)->tlab == NULL);   // Not used during reset

   model_thread_t *thread = model_thread(m);
   thread->active_obj = &(proc->wakeable);
   thread->active_scope = proc->scope;

   jit_scalar_t context = {
      .pointer = *mptr_get(proc->scope->privdata)
   };
   jit_scalar_t state = { .pointer = NULL };
   jit_scalar_t result;

   tlab_t tlab = jit_null_tlab(m->jit);

   if (jit_fastcall(m->jit, proc->handle, &result, state, context, &tlab))
      *mptr_get(proc->privdata) = result.pointer;
   else
      m->force_stop = true;

   thread->active_obj = NULL;
   thread->active_scope = NULL;

   // Schedule the process to run immediately
   deltaq_insert_proc(m, 0, proc);
}

static void reset_property(rt_model_t *m, rt_prop_t *prop)
{
   TRACE("reset property %s", istr(prop->name));

   assert(model_thread(m)->tlab == NULL);   // Not used during reset

   model_thread_t *thread = model_thread(m);
   thread->active_obj = &(prop->wakeable);
   thread->active_scope = prop->scope;

   tlab_t tlab = jit_null_tlab(m->jit);

   jit_scalar_t args[] = {
      { .pointer = NULL },
      { .pointer = *mptr_get(prop->scope->privdata) },
      { .integer = -1 },
   }, results[2];

   if (jit_vfastcall(m->jit, prop->handle, args, ARRAY_LEN(args),
                     results, ARRAY_LEN(results), &tlab))
      *mptr_get(prop->privdata) = results[0].pointer;
   else
      m->force_stop = true;

   TRACE("needs %"PRIi64" state bits", results[1].integer);

   mask_init(&prop->state, results[1].integer);
   mask_init(&prop->newstate, results[1].integer);

   mask_set(&prop->state, 0);
   mask_set(&prop->state, results[1].integer - 1);   // Update prev() variables

   thread->active_obj = NULL;
   thread->active_scope = NULL;
}

static void run_process(rt_model_t *m, rt_proc_t *proc)
{
   TRACE("run %sprocess %s", *mptr_get(proc->privdata) ? "" :  "stateless ",
         istr(proc->name));

   rt_wakeable_t *obj = &(proc->wakeable);

   if (obj->trigger != NULL && !run_trigger(m, obj->trigger))
      return;   // Filtered

   model_thread_t *thread = model_thread(m);
   assert(thread->tlab != NULL);
   assert(thread->tlab->alloc == 0);

   thread->active_obj = obj;
   thread->active_scope = proc->scope;

   // Stateless processes have NULL privdata so pass a dummy pointer
   // value in so it can be distinguished from a reset
   jit_scalar_t state = {
      .pointer = *mptr_get(proc->privdata) ?: (void *)-1
   };

   jit_scalar_t result;
   jit_scalar_t context = {
      .pointer = *mptr_get(proc->scope->privdata)
   };

   if (!jit_fastcall(m->jit, proc->handle, &result, state, context,
                     proc->tlab ?: thread->tlab))
      m->force_stop = true;

   if (proc->tlab != NULL && result.pointer == NULL) {
      tlab_release(proc->tlab);
      proc->tlab = NULL;
   }
   else if (proc->tlab == NULL && result.pointer != NULL) {
      TRACE("claiming TLAB for private use (used %u/%u)",
            thread->tlab->alloc, thread->tlab->limit);
      proc->tlab = thread->tlab;
      thread->tlab = tlab_acquire(m->mspace);
   }
   else if (proc->tlab == NULL)
      tlab_reset(thread->tlab);   // All data inside the TLAB is dead

   thread->active_obj = NULL;
   thread->active_scope = NULL;
}

static void reset_scope(rt_model_t *m, rt_scope_t *s)
{
   for (int i = 0; i < s->children.count; i++)
      reset_scope(m, s->children.items[i]);

   for (int i = 0; i < s->procs.count; i++)
      reset_process(m, s->procs.items[i]);

   for (int i = 0; i < s->properties.count; i++)
      reset_property(m, s->properties.items[i]);
}

static res_memo_t *memo_resolution_fn(rt_model_t *m, rt_signal_t *signal,
                                      ffi_closure_t closure, int32_t nlits,
                                      res_flags_t flags)
{
   // Optimise some common resolution functions by memoising them

   res_memo_t *memo = ihash_get(m->res_memo, closure.handle);
   if (memo != NULL)
      return memo;

   memo = static_alloc(m, sizeof(res_memo_t));
   memo->closure = closure;
   memo->flags   = flags;

   ihash_put(m->res_memo, memo->closure.handle, memo);

   if (nlits == 0 || nlits > 16)
      return memo;

   const vhdl_severity_t old_severity = set_exit_severity(SEVERITY_NOTE);

   jit_set_silent(m->jit, true);

   // Memoise the function for all two value cases

   for (int i = 0; i < nlits; i++) {
      for (int j = 0; j < nlits; j++) {
         int8_t args[2] = { i, j };
         jit_scalar_t result;
         if (jit_try_call(m->jit, memo->closure.handle, &result,
                          memo->closure.context, args, 2)) {
            assert(result.integer < nlits && result.integer >= 0);
            memo->tab2[i][j] = result.integer;
         }
      }
   }

   // Memoise the function for all single value cases and determine if the
   // function behaves like the identity function

   bool identity = true;
   for (int i = 0; i < nlits; i++) {
      int8_t args[1] = { i };
      jit_scalar_t result;
      if (jit_try_call(m->jit, memo->closure.handle, &result,
                       memo->closure.context, args, 1)) {
         memo->tab1[i] = result.integer;
         identity = identity && (memo->tab1[i] == i);
      }
   }

   if (model_exit_status(m) == 0) {
      memo->flags |= R_MEMO;
      if (identity)
         memo->flags |= R_IDENT;
   }

   TRACE("memoised resolution function %s for type %s",
         istr(jit_get_name(m->jit, closure.handle)),
         type_pp(tree_type(signal->where)));

   jit_set_silent(m->jit, false);
   jit_reset_exit_status(m->jit);

   set_exit_severity(old_severity);

   return memo;
}

static inline void *nexus_effective(rt_nexus_t *n)
{
   return n->signal->shared.data + n->offset;
}

static inline void *nexus_last_value(rt_nexus_t *n)
{
   return n->signal->shared.data + n->offset + n->signal->shared.size;
}

static inline void *nexus_driving(rt_nexus_t *n)
{
   assert(n->flags & NET_F_EFFECTIVE);
   return n->signal->shared.data + n->offset + 2*n->signal->shared.size;
}

static inline void *nexus_initial(rt_nexus_t *n)
{
   assert(n->flags & NET_F_HAS_INITIAL);
   return n->signal->shared.data + n->offset + 2*n->signal->shared.size;
}

static rt_value_t alloc_value(rt_model_t *m, rt_nexus_t *n)
{
   rt_value_t result = {};

   const size_t valuesz = n->size * n->width;
   if (valuesz > sizeof(rt_value_t)) {
      if (n->free_value != NULL) {
         result.ext = n->free_value;
         n->free_value = *(void **)result.ext;
      }
      else
         result.ext = static_alloc(m, valuesz);
   }

   return result;
}

static void free_value(rt_nexus_t *n, rt_value_t v)
{
   const size_t valuesz = n->width * n->size;
   if (valuesz > sizeof(rt_value_t)) {
      *(void **)v.ext = n->free_value;
      n->free_value = v.ext;
   }
}

static inline uint8_t *value_ptr(rt_nexus_t *n, rt_value_t *v)
{
   const size_t valuesz = n->width * n->size;
   if (valuesz <= sizeof(rt_value_t))
      return v->bytes;
   else
      return v->ext;
}

static void copy_value_ptr(rt_nexus_t *n, rt_value_t *v, const void *p)
{
   const size_t valuesz = n->width * n->size;
   if (valuesz <= sizeof(rt_value_t)) {
#if ASAN_ENABLED
      memcpy(v->bytes, p, valuesz);
#else
      v->qword = unaligned_load(p, uint64_t);
#endif
   }
   else
      memcpy(v->ext, p, valuesz);
}

static inline bool cmp_values(rt_nexus_t *n, rt_value_t a, rt_value_t b)
{
   const size_t valuesz = n->width * n->size;
   if (valuesz <= sizeof(rt_value_t))
      return a.qword == b.qword;
   else
      return cmp_bytes(a.ext, b.ext, valuesz);
}

static inline bool is_pseudo_source(source_kind_t kind)
{
   return kind == SOURCE_FORCING || kind == SOURCE_DEPOSIT
      || kind == SOURCE_IMPLICIT;
}

static void check_multiple_sources(rt_nexus_t *n, source_kind_t kind)
{
   if (n->signal->resolution != NULL || is_pseudo_source(kind))
      return;

   diag_t *d;
   if (is_signal_scope(n->signal->parent)) {
      rt_scope_t *root = n->signal->parent;
      for (; is_signal_scope(root->parent); root = root->parent);

      d = diag_new(DIAG_FATAL, tree_loc(root->where));
      diag_printf(d, "element %s of signal %s has multiple sources",
                  istr(tree_ident(n->signal->where)),
                  istr(tree_ident(root->where)));
      diag_hint(d, tree_loc(n->signal->where), "element %s declared here",
                istr(tree_ident(n->signal->where)));
      diag_hint(d, tree_loc(root->where), "composite signal %s declared with "
                "unresolved type %s", istr(tree_ident(root->where)),
                type_pp(tree_type(root->where)));
   }
   else {
      d = diag_new(DIAG_FATAL, tree_loc(n->signal->where));
      diag_printf(d, "unresolved signal %s has multiple sources",
                  istr(tree_ident(n->signal->where)));
      diag_hint(d, tree_loc(n->signal->where), "signal %s declared with "
                "unresolved type %s", istr(tree_ident(n->signal->where)),
                type_pp(tree_type(n->signal->where)));
   }

   if (n->sources.tag == SOURCE_DRIVER) {
      const rt_proc_t *p = n->sources.u.driver.proc;
      diag_hint(d, tree_loc(p->where), "driven by process %s", istr(p->name));
   }
   else if (n->sources.tag == SOURCE_PORT) {
      const rt_signal_t *s = n->sources.u.port.input->signal;
      tree_t where = s->where;
      if (is_signal_scope(s->parent)) {
         for (rt_scope_t *it = s->parent; is_signal_scope(it); it = it->parent)
            where = it->where;
      }

      if (tree_kind(where) == T_PORT_DECL)
         diag_hint(d, tree_loc(where), "connected to %s port %s",
                   port_mode_str(tree_subkind(where)), istr(tree_ident(where)));
      else
         diag_hint(d, tree_loc(where), "connected to signal %s",
                   istr(tree_ident(where)));
   }

   if (kind == SOURCE_DRIVER) {
      const rt_proc_t *p = get_active_proc();
      diag_hint(d, tree_loc(p->where), "driven by process %s", istr(p->name));
   }

   diag_emit(d);
   jit_abort_with_status(EXIT_FAILURE);
}

static rt_source_t *add_source(rt_model_t *m, rt_nexus_t *n, source_kind_t kind)
{
   rt_source_t *src = NULL;
   if (n->n_sources == 0)
      src = &(n->sources);
   else {
      check_multiple_sources(n, kind);

      rt_source_t **p;
      for (p = &(n->sources.chain_input); *p; p = &((*p)->chain_input))
         ;
      *p = src = static_alloc(m, sizeof(rt_source_t));
   }

   // The only interesting values of n_sources are 0, 1, and 2
   if (n->n_sources < UINT8_MAX)
      n->n_sources++;

   if (n->n_sources > 1)
      n->flags &= ~NET_F_FAST_DRIVER;

   src->chain_input  = NULL;
   src->chain_output = NULL;
   src->tag          = kind;
   src->disconnected = 0;
   src->fastqueued   = 0;
   src->sigqueued    = 0;
   src->pseudoqueued = 0;

   switch (kind) {
   case SOURCE_DRIVER:
      {
         src->u.driver.proc  = NULL;
         src->u.driver.nexus = n;

         waveform_t *w0 = &(src->u.driver.waveforms);
         w0->when  = TIME_HIGH;
         w0->next  = NULL;
      }
      break;

   case SOURCE_PORT:
      src->u.port.conv_func = NULL;
      src->u.port.input     = NULL;
      src->u.port.output    = n;
      break;

   case SOURCE_DEPOSIT:
   case SOURCE_FORCING:
   case SOURCE_IMPLICIT:
      src->u.pseudo.nexus = n;
      src->u.pseudo.value = alloc_value(m, n);
      break;
   }

   return src;
}

static inline int map_index(rt_index_t *index, unsigned offset)
{
   if (likely(index->how >= 0))
      return offset >> index->how;
   else
      return offset / -(index->how);
}

static inline int unmap_index(rt_index_t *index, unsigned key)
{
   if (likely(index->how >= 0))
      return key << index->how;
   else
      return key * -(index->how);
}

static inline bool index_valid(rt_index_t *index, unsigned offset)
{
   if (likely(index->how >= 0))
      return (offset >> index->how) << index->how == offset;
   else
      return offset % -(index->how) == 0;
}

static void build_index(rt_signal_t *signal)
{
   const unsigned signal_w = signal->shared.size / signal->nexus.size;

   int shift = INT_MAX, gcd = 0;
   rt_nexus_t *n = &(signal->nexus);
   for (int i = 0, offset = 0; i < signal->n_nexus;
        i++, offset += n->width, n = n->chain) {
      if (offset > 0) {
         const int tzc = __builtin_ctz(offset);
         shift = MIN(shift, tzc);
      }

      // Compute greatest common divisor
      for (int b = offset; b > 0;) {
         int temp = b;
         b = gcd % b;
         gcd = temp;
      }
   }

   const int how = gcd > 1 && gcd > (1 << shift) && gcd > 1 ? -gcd : shift;
   const int count =
      how < 0 ? (signal_w - how - 1) / -how : (signal_w >> shift) + 1;

   TRACE("create index for signal %s how=%d count=%d",
         istr(tree_ident(signal->where)), how, count);

   rt_index_t *index = xcalloc_flex(sizeof(rt_index_t), count,
                                    sizeof(rt_nexus_t *));
   index->how = how;

   n = &(signal->nexus);
   for (int i = 0, offset = 0; i < signal->n_nexus;
        i++, offset += n->width, n = n->chain) {
      index->nexus[map_index(index, offset)] = n;
   }

   free(signal->index);
   signal->index = index;
}

static void update_index(rt_signal_t *s, rt_nexus_t *n)
{
   const unsigned offset = n->offset / n->size;

   if (!index_valid(s->index, offset)) {
      TRACE("rebuild index for %s offset=%d how=%d",
            istr(tree_ident(s->where)), offset, s->index->how);
      build_index(s);
      assert(s->index->nexus[map_index(s->index, offset)] == n);
   }
   else {
      const int elt = map_index(s->index, offset);
      assert(s->index->nexus[elt] == NULL);
      s->index->nexus[elt] = n;
   }
}

static rt_nexus_t *lookup_index(rt_signal_t *s, int *offset)
{
   if (likely(offset == 0 || s->index == NULL))
      return &(s->nexus);
   else if (!index_valid(s->index, *offset)) {
      TRACE("invalid index for %s offset=%d how=%d", istr(tree_ident(s->where)),
            *offset, s->index->how);
      free(s->index);
      s->index = NULL;
      return &(s->nexus);
   }
   else {
      const int key = map_index(s->index, *offset);
      for (int k = key; k >= 0; k--) {
         rt_nexus_t *n = s->index->nexus[k];
         if (n != NULL) {
            *offset = unmap_index(s->index, key - k);
            return n;
         }
      }
      return &(s->nexus);
   }
}

static waveform_t *alloc_waveform(rt_model_t *m)
{
   model_thread_t *thread = model_thread(m);

   if (thread->free_waveforms == NULL) {
      // Ensure waveforms are always within one cache line
      STATIC_ASSERT(sizeof(waveform_t) <= 32);
      char *mem = static_alloc(m, WAVEFORM_CHUNK * 32);
      for (int i = 1; i < WAVEFORM_CHUNK; i++)
         free_waveform(m, (waveform_t *)(mem + i*32));

      return (waveform_t *)mem;
   }
   else {
      waveform_t *w = thread->free_waveforms;
      thread->free_waveforms = w->next;
      prefetch_write(w->next);
      w->next = NULL;
      return w;
   }
}

static void add_conversion_input(rt_model_t *m, rt_conv_func_t *cf,
                                 rt_nexus_t *in)
{
   if (cf->ninputs == cf->maxinputs) {
      const size_t per_block = MEMBLOCK_ALIGN / sizeof(conv_input_t);
      cf->maxinputs = ALIGN_UP(MAX(4, cf->maxinputs * 2), per_block);

      if (cf->inputs == cf->tail) {
         void *new = xmalloc_array(cf->maxinputs, sizeof(conv_input_t));
         memcpy(new, cf->inputs, cf->ninputs * sizeof(conv_input_t));
         cf->inputs = new;
      }
      else
         cf->inputs = xrealloc_array(cf->inputs, cf->maxinputs,
                                     sizeof(conv_input_t));
   }

   cf->inputs[cf->ninputs++] = (conv_input_t){
      .nexus  = in,
      .result = alloc_value(m, in),
   };
}

static rt_value_t *find_conversion_input(rt_conv_func_t *cf, rt_nexus_t *n)
{
   for (int i = 0; i < cf->ninputs; i++) {
      if (cf->inputs[i].nexus == n)
         return &(cf->inputs[i].result);
   }

   return NULL;
}

static void split_value(rt_nexus_t *nexus, rt_value_t *v_new,
                        rt_value_t *v_old, int offset)
{
   const int split = offset * nexus->size;
   const int oldsz = (offset + nexus->width) * nexus->size;
   const int newsz = nexus->width * nexus->size;

   if (split > sizeof(rt_value_t) && newsz > sizeof(rt_value_t)) {
      // Split the external memory with no copying
      v_new->ext = (char *)v_old->ext + split;
   }
   else if (newsz > sizeof(rt_value_t)) {
      // Wasting up to eight bytes at the start of the the old waveform
      char *ext = v_old->ext;
      v_old->qword = *(uint64_t *)ext;
      v_new->ext = ext + split;
   }
   else if (split > sizeof(rt_value_t)) {
      // Wasting up to eight bytes at the end of the the old waveform
      memcpy(v_new->bytes, v_old->ext + split, newsz);
   }
   else if (oldsz > sizeof(rt_value_t)) {
      // The memory backing this waveform is lost now but this can only
      // happen a bounded number of times as nexuses only ever shrink
      char *ext = v_old->ext;
      memcpy(v_new->bytes, ext + split, newsz);
      v_old->qword = *(uint64_t *)ext;
   }
   else {
      // This trick with shifting probably only works on little-endian
      // systems
      v_new->qword = v_old->qword >> (split * 8);
   }
}

static void clone_source(rt_model_t *m, rt_nexus_t *nexus,
                         rt_source_t *old, int offset)
{
   rt_source_t *new = add_source(m, nexus, old->tag);

   switch (old->tag) {
   case SOURCE_PORT:
      {
         new->u.port.input = old->u.port.input;

         if (old->u.port.conv_func != NULL)
            new->u.port.conv_func = old->u.port.conv_func;
         else {
            if (old->u.port.input->width == offset)
               new->u.port.input = old->u.port.input->chain;  // Cycle breaking
            else {
               RT_LOCK(old->u.port.input->signal->lock);
               rt_nexus_t *n = clone_nexus(m, old->u.port.input, offset);
               new->u.port.input = n;
            }
            assert(new->u.port.input->width == nexus->width);
         }
      }
      break;

   case SOURCE_DRIVER:
      {
         new->u.driver.proc = old->u.driver.proc;

         // Current transaction
         waveform_t *w_new = &(new->u.driver.waveforms);
         waveform_t *w_old = &(old->u.driver.waveforms);
         w_new->when = w_old->when;
         w_new->next = NULL;

         split_value(nexus, &w_new->value, &w_old->value, offset);

         // Pending fast driver update
         if ((nexus->flags & NET_F_FAST_DRIVER) && old->fastqueued) {
            rt_nexus_t *n0 = &(nexus->signal->nexus);
            if (!n0->sources.sigqueued)
               deferq_do(&m->delta_driverq, async_fast_driver, new);
            new->fastqueued = 1;
         }

         new->was_active = old->was_active;

         // Future transactions
         for (w_old = w_old->next; w_old; w_old = w_old->next) {
            w_new = (w_new->next = alloc_waveform(m));
            w_new->when = w_old->when;
            w_new->next = NULL;

            split_value(nexus, &w_new->value, &w_old->value, offset);

            assert(w_old->when >= m->now);
            deltaq_insert_driver(m, w_new->when - m->now, new);
         }
      }
      break;

   case SOURCE_FORCING:
   case SOURCE_DEPOSIT:
      split_value(nexus, &(new->u.pseudo.value), &(old->u.pseudo.value),
                  offset);
      assert(!old->pseudoqueued);   // TODO
      break;

   case SOURCE_IMPLICIT:
      break;
   }
}

static rt_nexus_t *clone_nexus(rt_model_t *m, rt_nexus_t *old, int offset)
{
   assert(offset < old->width);

   rt_signal_t *signal = old->signal;
   MULTITHREADED_ONLY(assert_lock_held(&signal->lock));
   signal->n_nexus++;

   if (signal->n_nexus == 2 && (old->flags & NET_F_FAST_DRIVER))
      signal->shared.flags |= NET_F_FAST_DRIVER;

   rt_nexus_t *new = static_alloc(m, sizeof(rt_nexus_t));
   new->width        = old->width - offset;
   new->size         = old->size;
   new->signal       = signal;
   new->offset       = old->offset + offset * old->size;
   new->chain        = old->chain;
   new->flags        = old->flags;
   new->active_delta = old->active_delta;
   new->event_delta  = old->event_delta;
   new->last_event   = old->last_event;
   new->rank         = old->rank;

   old->chain = new;
   old->width = offset;

   if (old->pending == NULL)
      new->pending = NULL;
   else if (pointer_tag(old->pending) == 1)
      new->pending = old->pending;
   else {
      rt_pending_t *old_p = untag_pointer(old->pending, rt_pending_t);
      rt_pending_t *new_p = xmalloc_flex(sizeof(rt_pending_t), old_p->count,
                                         sizeof(rt_wakeable_t *));

      new_p->count = new_p->max = old_p->count;

      for (int i = 0; i < old_p->count; i++)
         new_p->wake[i] = old_p->wake[i];

      new->pending = tag_pointer(new_p, 0);
   }

   if (new->chain == NULL)
      m->nexus_tail = &(new->chain);

   if (old->n_sources > 0) {
      for (rt_source_t *it = &(old->sources); it; it = it->chain_input)
         clone_source(m, new, it, offset);
   }

   for (rt_source_t *old_o = old->outputs; old_o; old_o = old_o->chain_output) {
      assert(old_o->tag == SOURCE_PORT || old_o->tag == SOURCE_IMPLICIT);

      if (old_o->tag == SOURCE_PORT && old_o->u.port.conv_func != NULL) {
         new->outputs = old_o;
         add_conversion_input(m, old_o->u.port.conv_func, new);
      }
      else {
         rt_nexus_t *out_n;
         if (old_o->tag == SOURCE_IMPLICIT)
            out_n = old_o->u.port.output;
         else if (old_o->u.port.output->width == offset)
            out_n = old_o->u.port.output->chain;   // Cycle breaking
         else {
            RT_LOCK(old_o->u.port.output->signal->lock);
            out_n = clone_nexus(m, old_o->u.port.output, offset);
         }

         for (rt_source_t *s = &(out_n->sources); s; s = s->chain_input) {
            if (s->tag != old_o->tag)
               continue;
            else if (s->u.port.input == new || s->u.port.input == old) {
               s->u.port.input = new;
               s->chain_output = new->outputs;
               new->outputs = s;
               break;
            }
         }
      }
   }

   if (signal->index == NULL && signal->n_nexus >= NEXUS_INDEX_MIN)
      build_index(signal);
   else if (signal->index != NULL)
      update_index(signal, new);

   return new;
}

static rt_nexus_t *split_nexus_slow(rt_model_t *m, rt_signal_t *s,
                                    int offset, int count)
{
   rt_nexus_t *result = NULL;
   for (rt_nexus_t *it = lookup_index(s, &offset); count > 0; it = it->chain) {
      if (offset >= it->width) {
         offset -= it->width;
         continue;
      }
      else if (offset > 0) {
         clone_nexus(m, it, offset);
         offset = 0;
         continue;
      }
      else {
         if (it->width > count)
            clone_nexus(m, it, count);

         count -= it->width;

         if (result == NULL)
            result = it;
      }
   }

   return result;
}

static inline rt_nexus_t *split_nexus(rt_model_t *m, rt_signal_t *s,
                                      int offset, int count)
{
   MULTITHREADED_ONLY(assert_lock_held(&s->lock));

   rt_nexus_t *n0 = &(s->nexus);
   if (likely(offset == 0 && n0->width == count))
      return n0;
   else if (offset == 0 && count == s->shared.size / n0->size)
      return n0;

   return split_nexus_slow(m, s, offset, count);
}

static void setup_signal(rt_model_t *m, rt_signal_t *s, tree_t where,
                         unsigned count, unsigned size, sig_flags_t flags,
                         unsigned offset)
{
   rt_scope_t *parent = model_thread(m)->active_scope;

   s->where   = where;
   s->n_nexus = 1;
   s->offset  = offset;
   s->parent  = parent;

   s->shared.flags = flags;
   s->shared.size  = count * size;

   APUSH(parent->signals, s);

   s->nexus.width        = count;
   s->nexus.size         = size;
   s->nexus.n_sources    = 0;
   s->nexus.offset       = 0;
   s->nexus.flags        = flags | NET_F_FAST_DRIVER | NET_F_HAS_INITIAL;
   s->nexus.signal       = s;
   s->nexus.pending      = NULL;
   s->nexus.active_delta = DELTA_CYCLE_MAX;
   s->nexus.event_delta  = DELTA_CYCLE_MAX;
   s->nexus.last_event   = TIME_HIGH;

   *m->nexus_tail = &(s->nexus);
   m->nexus_tail = &(s->nexus.chain);

   m->n_signals++;
}

static void copy_sub_signal_sources(rt_scope_t *scope, void *buf, int stride)
{
   assert(is_signal_scope(scope));

   for (int i = 0; i < scope->signals.count; i++) {
      rt_signal_t *s = scope->signals.items[i];
      rt_nexus_t *n = &(s->nexus);
      for (unsigned i = 0; i < s->n_nexus; i++) {
         unsigned o = 0;
         for (rt_source_t *src = &(n->sources); src; src = src->chain_input) {
            const void *data = source_value(n, src);
            if (data == NULL)
               continue;

            memcpy(buf + s->offset + (o++ * stride), data, n->size * n->width);
         }
      }
   }

   for (int i = 0; i < scope->children.count; i++)
      copy_sub_signal_sources(scope->children.items[i], buf, stride);
}

static void convert_driving(rt_conv_func_t *cf)
{
   rt_model_t *m = get_model();

   if (cf->effective.handle == JIT_HANDLE_INVALID) {
      // Ensure effective value is only updated once per cycle
      if (cf->when == m->now && cf->iteration == m->iteration)
         return;

      cf->when = m->now;
      cf->iteration = m->iteration;
   }

   TRACE("call driving conversion function %s",
         istr(jit_get_name(m->jit, cf->driving.handle)));

   model_thread_t *thread = model_thread(m);

   jit_scalar_t context = { .pointer = cf->driving.context };
   jit_scalar_t arg = { .pointer = cf }, result;
   if (!jit_fastcall(m->jit, cf->driving.handle, &result, context, arg,
                     thread->tlab))
      m->force_stop = true;

   tlab_reset(thread->tlab);   // No allocations can be live past here
}

static void convert_effective(rt_conv_func_t *cf)
{
   rt_model_t *m = get_model();

   // Ensure effective value is only updated once per cycle
   if (cf->when == m->now && cf->iteration == m->iteration)
      return;

   cf->when = m->now;
   cf->iteration = m->iteration;

   TRACE("call effective conversion function %s",
         istr(jit_get_name(m->jit, cf->effective.handle)));

   model_thread_t *thread = model_thread(m);

   jit_scalar_t context = { .pointer = cf->effective.context };
   jit_scalar_t arg = { .pointer = cf }, result;
   if (!jit_fastcall(m->jit, cf->effective.handle, &result, context, arg,
                     thread->tlab))
      m->force_stop = true;

   tlab_reset(thread->tlab);   // No allocations can be live past here
}

static void *source_value(rt_nexus_t *nexus, rt_source_t *src)
{
   switch (src->tag) {
   case SOURCE_DRIVER:
      if (unlikely(src->disconnected))
         return NULL;
      else
         return value_ptr(nexus, &(src->u.driver.waveforms.value));

   case SOURCE_PORT:
      if (likely(src->u.port.conv_func == NULL)) {
         if (src->u.port.input->flags & NET_F_EFFECTIVE)
            return nexus_driving(src->u.port.input);
         else
            return nexus_effective(src->u.port.input);
      }
      else {
         convert_driving(src->u.port.conv_func);
         return value_ptr(nexus, &src->u.port.conv_result);
      }

   case SOURCE_FORCING:
   case SOURCE_DEPOSIT:
      assert(src->disconnected);
      return NULL;

   case SOURCE_IMPLICIT:
      return NULL;
   }

   return NULL;
}

static void call_resolution(rt_model_t *m, rt_nexus_t *n, res_memo_t *r,
                            int nonnull, rt_source_t *s0)
{
   if ((n->flags & NET_F_R_IDENT) && nonnull == 1) {
      // Resolution function behaves like identity for a single driver
      put_driving(m, n, source_value(n, s0));
   }
   else if ((r->flags & R_MEMO) && nonnull == 1) {
      // Resolution function has been memoised so do a table lookup

      model_thread_t *thread = model_thread(m);
      assert(thread->tlab != NULL);

      void *resolved = tlab_alloc(thread->tlab, n->width * n->size);
      char *p0 = source_value(n, s0);

      for (int j = 0; j < n->width; j++) {
         const int index = ((uint8_t *)p0)[j];
         ((int8_t *)resolved)[j] = r->tab1[index];
      }

      put_driving(m, n, resolved);
      tlab_reset(thread->tlab);   // No allocations can be live past here
   }
   else if ((r->flags & R_MEMO) && nonnull == 2) {
      // Resolution function has been memoised so do a table lookup

      model_thread_t *thread = model_thread(m);
      assert(thread->tlab != NULL);

      void *resolved = tlab_alloc(thread->tlab, n->width * n->size);

      char *p0 = source_value(n, s0), *p1 = NULL;
      for (rt_source_t *s1 = s0->chain_input;
           s1 && (p1 = source_value(n, s1)) == NULL;
           s1 = s1->chain_input)
         ;

      for (int j = 0; j < n->width; j++)
         ((int8_t *)resolved)[j] = r->tab2[(int)p0[j]][(int)p1[j]];

      put_driving(m, n, resolved);
      tlab_reset(thread->tlab);   // No allocations can be live past here
   }
   else if (r->flags & R_COMPOSITE) {
      // Call resolution function of composite type

      rt_scope_t *scope = n->signal->parent, *rscope = scope;
      while (is_signal_scope(scope->parent)) {
         scope = scope->parent;
         if (scope->flags & SCOPE_F_RESOLVED)
            rscope = scope;
      }

      TRACE("resolved composite signal needs %d bytes", scope->size);

      model_thread_t *thread = model_thread(m);
      assert(thread->tlab != NULL);

      uint8_t *inputs = tlab_alloc(thread->tlab, nonnull * scope->size);
      copy_sub_signal_sources(scope, inputs, scope->size);

      jit_scalar_t result;
      if (jit_try_call(m->jit, r->closure.handle, &result,
                       r->closure.context, inputs, nonnull))
         put_driving(m, n, result.pointer + n->signal->offset
                     + n->offset - rscope->offset);
      else
         m->force_stop = true;

      tlab_reset(thread->tlab);   // No allocations can be live past here
   }
   else {
      model_thread_t *thread = model_thread(m);
      assert(thread->tlab != NULL);

      void *resolved = tlab_alloc(thread->tlab, n->width * n->size);

      for (int j = 0; j < n->width; j++) {
#define CALL_RESOLUTION_FN(type) do {                                   \
            type vals[nonnull];                                         \
            unsigned o = 0;                                             \
            for (rt_source_t *s = s0; s; s = s->chain_input) {          \
               const void *data = source_value(n, s);                   \
               if (data != NULL)                                        \
                  vals[o++] = ((const type *)data)[j];                  \
            }                                                           \
            assert(o == nonnull);                                       \
            type *p = (type *)resolved;                                 \
            jit_scalar_t result;                                        \
            if (!jit_try_call(m->jit, r->closure.handle, &result,       \
                              r->closure.context, vals, nonnull))       \
               m->force_stop = true;                                    \
            p[j] = result.integer;                                      \
         } while (0)

         FOR_ALL_SIZES(n->size, CALL_RESOLUTION_FN);
      }

      put_driving(m, n, resolved);
      tlab_reset(thread->tlab);   // No allocations can be live past here
   }
}

static rt_source_t *get_pseudo_source(rt_model_t *m, rt_nexus_t *n,
                                      source_kind_t kind)
{
   assert(is_pseudo_source(kind));

   if (n->n_sources > 0) {
      for (rt_source_t *s = &(n->sources); s; s = s->chain_input) {
         if (s->tag == kind)
            return s;
      }
   }

   return add_source(m, n, kind);
}

__attribute__((cold, noinline))
static void schedule_implicit_update(rt_model_t *m, rt_nexus_t *n)
{
   rt_implicit_t *imp = container_of(n->signal, rt_implicit_t, signal);

   if (!imp->wakeable.pending) {
      deferq_do(&m->implicitq, async_update_implicit_signal, imp);
      set_pending(&imp->wakeable);
   }
}

static void calculate_driving_value(rt_model_t *m, rt_nexus_t *n)
{
   // Algorithm for driving values is in LRM 08 section 14.7.3.2

   // If S has no source, then the driving value of S is given by the
   // default value associated with S
   if (n->n_sources == 0) {
      put_driving(m, n, nexus_initial(n));
      return;
   }

   res_memo_t *r = n->signal->resolution;

   int nonnull = 0;
   rt_source_t *s0 = NULL;
   for (rt_source_t *s = &(n->sources); s; s = s->chain_input) {
      if (s->disconnected)
         continue;
      else if (s->tag == SOURCE_FORCING) {
         // If S is driving-value forced, the driving value of S is
         // unchanged from its previous value; no further steps are
         // required.
         put_driving(m, n, value_ptr(n, &(s->u.pseudo.value)));
         return;
      }
      else if (s->tag == SOURCE_DEPOSIT) {
         // If a driving-value deposit is scheduled for S or for a
         // signal of which S is a subelement, the driving value of S is
         // the driving deposit value for S or the element of the
         // driving deposit value for the signal of which S is a
         // subelement, as appropriate.
         s->disconnected = 1;
         put_driving(m, n, value_ptr(n, &(s->u.pseudo.value)));
         return;
      }
      else if (unlikely(s->tag == SOURCE_IMPLICIT)) {
         // At least one of the inputs is active so schedule an update
         // to the value of an implicit 'TRANSACTION or 'QUIET signal
         schedule_implicit_update(m, n);
         return;
      }
      else if (s0 == NULL)
         s0 = s;
      nonnull++;
   }

   if (unlikely(s0 == NULL)) {
      // If S is of signal kind register and all the sources of S have
      // values determined by the null transaction, then the driving
      // value of S is unchanged from its previous value.
      if (n->signal->shared.flags & SIG_F_REGISTER)
         put_driving(m, n, nexus_effective(n));
      else if (r == NULL || is_pseudo_source(n->sources.tag))
         put_driving(m, n, nexus_initial(n));
      else
         call_resolution(m, n, r, nonnull, s0);
   }
   else if (r == NULL) {
      switch (s0->tag) {
      case SOURCE_DRIVER:
         // If S has one source that is a driver and S is not a resolved
         // signal, then the driving value of S is the current value of
         // that driver.
         assert(!s0->disconnected);
         put_driving(m, n, value_ptr(n, &(s0->u.driver.waveforms.value)));
         break;

      case SOURCE_PORT:
         // If S has one source that is a port and S is not a resolved
         // signal, then the driving value of S is the driving value of
         // the formal part of the association element that associates S
         // with that port
         if (likely(s0->u.port.conv_func == NULL)) {
            if (s0->u.port.input->flags & NET_F_EFFECTIVE)
               put_driving(m, n, nexus_driving(s0->u.port.input));
            else
               put_driving(m, n, nexus_effective(s0->u.port.input));
         }
         else {
            convert_driving(s0->u.port.conv_func);
            put_driving(m, n, value_ptr(n, &(s0->u.port.conv_result)));
         }
         break;

      default:
         break;
      }
   }
   else {
      // Otherwise, the driving value of S is obtained by executing the
      // resolution function associated with S
      call_resolution(m, n, r, nonnull, s0);
   }
}

static void calculate_effective_value(rt_model_t *m, rt_nexus_t *n)
{
   // Algorithm for effective values is in LRM 08 section 14.7.7.3

   // If S is a connected port of mode in or inout, then the effective
   // value of S is the same as the effective value of the actual part
   // of the association element that associates an actual with S
   if (n->flags & NET_F_INOUT) {
      for (rt_source_t *s = n->outputs; s; s = s->chain_output) {
         if (s->tag == SOURCE_PORT) {
            if (likely(s->u.port.conv_func == NULL))
               put_effective(m, n, nexus_effective(s->u.port.output));
            else {
               rt_value_t *v = find_conversion_input(s->u.port.conv_func, n);
               assert(v != NULL);

               convert_effective(s->u.port.conv_func);
               put_effective(m, n, value_ptr(n, v));
            }
            return;
         }
      }
   }

   // If S is a signal declared by a signal declaration, a port of mode
   // out or buffer, or an unconnected port of mode inout, then the
   // effective value of S is the same as the driving value of S.
   if (n->flags & NET_F_EFFECTIVE)
      put_effective(m, n, nexus_driving(n));

   // If S is an unconnected port of mode in, the effective value of S
   // is given by the default value associated with S.
}

static void calculate_initial_value(rt_model_t *m, rt_nexus_t *n)
{
   calculate_driving_value(m, n);

   if (n->flags & NET_F_EFFECTIVE) {
      // Driving and effective values must be calculated separately
      assert(n->flags & NET_F_PENDING);
   }
   else {
      // Effective value is always the same as the driving value
      memcpy(nexus_last_value(n), nexus_effective(n), n->size * n->width);
   }
}

static int nexus_rank(rt_nexus_t *n)
{
   if (n->rank > 0)
      return n->rank;   // Already calculated
   else if (n->n_sources > 0) {
      int rank = 0;
      for (rt_source_t *s = &(n->sources); s; s = s->chain_input) {
         if (s->tag != SOURCE_PORT)
            continue;
         else if (s->u.port.conv_func != NULL) {
            rt_conv_func_t *cf = s->u.port.conv_func;
            for (int i = 0; i < cf->ninputs; i++)
               rank = MAX(rank, nexus_rank(cf->inputs[i].nexus) + 1);
         }
         else
            rank = MAX(rank, nexus_rank(s->u.port.input) + 1);
      }
      return (n->rank = rank);
   }
   else
      return 0;
}

cover_data_t *get_coverage(rt_model_t *m)
{
   return m->cover;
}

#if TRACE_SIGNALS
static void dump_one_signal(rt_model_t *m, rt_scope_t *scope, rt_signal_t *s,
                            tree_t alias)
{
   rt_nexus_t *n = &(s->nexus);

   LOCAL_TEXT_BUF tb = tb_new();
   if (is_signal_scope(scope))
      tb_printf(tb, "%s.", istr(scope->name));
   tb_cat(tb, istr(tree_ident(alias ?: s->where)));
   if (alias != NULL)
      tb_append(tb, '*');

   for (int nth = 0; nth < s->n_nexus; nth++, n = n->chain) {
      int n_outputs = 0;
      for (rt_source_t *s = n->outputs; s != NULL; s = s->chain_output)
         n_outputs++;

      const void *driving = NULL;
      if (n->flags & NET_F_EFFECTIVE)
         driving = nexus_driving(n);

      fprintf(stderr, "%-20s %-5d %-4d %-7d %-7d %-4d ",
              nth == 0 ? tb_get(tb) : "+",
              n->width, n->size, n->n_sources, n_outputs, n->rank);

      if (n->event_delta == m->iteration && n->last_event == m->now)
         fprintf(stderr, "%s -> ", fmt_nexus(n, nexus_last_value(n)));

      fputs(fmt_nexus(n, nexus_effective(n)), stderr);

      if (driving != NULL)
         fprintf(stderr, " (%s)", fmt_nexus(n, driving));

      fputs("\n", stderr);
   }
}

static void dump_signals(rt_model_t *m, rt_scope_t *scope)
{
   if (scope->signals.count == 0 && scope->children.count == 0)
      return;

   if (!is_signal_scope(scope)) {
      const char *sname = istr(scope->name);
      fprintf(stderr, "== %s ", sname);
      for (int pad = 74 - strlen(sname); pad > 0; pad--)
         fputc('=', stderr);
      fputc('\n', stderr);

      fprintf(stderr, "%-20s %5s %4s %7s %7s %4s %s\n",
              "Signal", "Width", "Size", "Sources", "Outputs", "Rank", "Value");
   }

   for (int i = 0; i < scope->signals.count; i++)
      dump_one_signal(m, scope, scope->signals.items[i], NULL);

   for (int i = 0; i < scope->aliases.count; i++) {
      rt_alias_t *a = scope->aliases.items[i];
      dump_one_signal(m, scope, a->signal, a->where);
   }

   for (int i = 0; i < scope->children.count; i++) {
      rt_scope_t *c = scope->children.items[i];
      if (is_signal_scope(c))
         dump_signals(m, c);
   }

   for (int i = 0; i < scope->children.count; i++) {
      rt_scope_t *c = scope->children.items[i];
      if (!is_signal_scope(c))
         dump_signals(m, c);
   }
}
#endif   // TRACE_SIGNALS

static text_buf_t *signal_full_name(rt_signal_t *s)
{
   text_buf_t *tb = tb_new();
   if (is_signal_scope(s->parent))
      tb_printf(tb, "%s.", istr(s->parent->name));
   tb_cat(tb, istr(tree_ident(s->where)));
   return tb;
}

static void check_undriven_std_logic(rt_nexus_t *n)
{
   // Print a warning if any STD_LOGIC signal has multiple sources one
   // of which is an undriven port with initial value 'U'. The resolved
   // value will then always be 'U' which often confuses users.

   if (n->n_sources < 2 || !(n->signal->shared.flags & SIG_F_STD_LOGIC))
      return;

   rt_signal_t *undriven = NULL;
   for (rt_source_t *s = &(n->sources); s; s = s->chain_input) {
      if (s->tag == SOURCE_PORT && s->u.port.conv_func == NULL) {
         rt_nexus_t *input = s->u.port.input;
         if (input->n_sources == 0) {
            const unsigned char *init = nexus_effective(input), *p = init;
            for (; *p == 0 && p < init + input->width; p++);

            if (p == init + input->width)
               undriven = s->u.port.input->signal;
         }
      }
   }

   if (undriven == NULL)
      return;

   LOCAL_TEXT_BUF sig_name = signal_full_name(n->signal);
   LOCAL_TEXT_BUF port_name = signal_full_name(undriven);

   const loc_t *sig_loc = tree_loc(n->signal->where);
   rt_scope_t *sig_scope = n->signal->parent;
   for (; is_signal_scope(sig_scope); sig_scope = sig_scope->parent)
      sig_loc = tree_loc(sig_scope->where);

   const loc_t *port_loc = tree_loc(undriven->where);
   rt_scope_t *port_scope = undriven->parent;
   for (; is_signal_scope(port_scope); port_scope = port_scope->parent)
      port_loc = tree_loc(port_scope->where);

   diag_t *d = diag_new(DIAG_WARN, sig_loc);
   diag_printf(d, "%ssignal %s has %d sources including port %s which has "
               "initial value 'U' and no driver in instance %s",
               n->signal->n_nexus > 1 ? "sub-element of " : "",
               tb_get(sig_name), n->n_sources,
               tb_get(port_name), istr(tree_ident(port_scope->where)));
   diag_hint(d, sig_loc, "signal %s declared here", tb_get(sig_name));
   diag_hint(d, port_loc, "sourced by port %s which always contributes 'U'",
             tb_get(port_name));
   diag_hint(d, NULL, "the resolved value will always be 'U' which was almost "
             "certainly not intended");
   diag_emit(d);

   // Prevent multiple warnings for the same signal
   n->signal->shared.flags &= ~SIG_F_STD_LOGIC;
}

static void create_processes(rt_model_t *m, rt_scope_t *s)
{
   for (int i = 0; i < s->children.count; i++) {
      if (s->children.items[i]->kind == SCOPE_INSTANCE)
         create_processes(m, s->children.items[i]);
   }

   if (s->kind != SCOPE_INSTANCE)
      return;

   tree_t hier = tree_decl(s->where, 0);
   assert(tree_kind(hier) == T_HIER);

   LOCAL_TEXT_BUF tb = tb_new();
   get_path_name(s, tb);

   ident_t path = ident_new(tb_get(tb));
   ident_t sym_prefix = tree_ident2(hier);

   const int nstmts = tree_stmts(s->where);
   for (int i = 0; i < nstmts; i++) {
      tree_t t = tree_stmt(s->where, i);
      switch (tree_kind(t)) {
      case T_VERILOG:
         {
            ident_t name = tree_ident(t);
            ident_t sym = ident_prefix(sym_prefix, name, '.');

            rt_proc_t *p = xcalloc(sizeof(rt_proc_t));
            p->where     = t;
            p->name      = ident_prefix(path, ident_downcase(name), ':');
            p->handle    = jit_lazy_compile(m->jit, sym);
            p->scope     = s;
            p->privdata  = mptr_new(m->mspace, "process privdata");

            p->wakeable.kind      = W_PROC;
            p->wakeable.pending   = false;
            p->wakeable.postponed = false;
            p->wakeable.delayed   = false;

            APUSH(s->procs, p);
         }
         break;

      case T_PROCESS:
         {
            ident_t name = tree_ident(t);
            ident_t sym = ident_prefix(sym_prefix, name, '.');

            rt_proc_t *p = xcalloc(sizeof(rt_proc_t));
            p->where     = t;
            p->name      = ident_prefix(path, ident_downcase(name), ':');
            p->handle    = jit_lazy_compile(m->jit, sym);
            p->scope     = s;
            p->privdata  = mptr_new(m->mspace, "process privdata");

            p->wakeable.kind      = W_PROC;
            p->wakeable.pending   = false;
            p->wakeable.postponed = !!(tree_flags(t) & TREE_F_POSTPONED);
            p->wakeable.delayed   = false;

            APUSH(s->procs, p);
         }
         break;

      case T_PSL_DIRECT:
         {
            psl_node_t psl = tree_psl(t);

            const psl_kind_t kind = psl_kind(psl);
            if (kind != P_ASSERT && kind != P_COVER)
               continue;

            ident_t name = tree_ident(t);
            ident_t sym = ident_prefix(s->name, name, '.');

            rt_prop_t *p = xcalloc(sizeof(rt_prop_t));
            p->where    = tree_psl(t);
            p->handle   = jit_lazy_compile(m->jit, sym);
            p->scope    = s;
            p->name     = sym;
            p->privdata = mptr_new(m->mspace, "property privdata");

            p->wakeable.kind      = W_PROPERTY;
            p->wakeable.pending   = false;
            p->wakeable.postponed = false;
            p->wakeable.delayed   = false;

            APUSH(s->properties, p);
         }
         break;

      default:
         break;
      }
   }
}

void model_reset(rt_model_t *m)
{
   MODEL_ENTRY(m);

   // Re-read options as these may have changed
   m->stop_delta = opt_get_int(OPT_STOP_DELTA);
   m->shuffle    = opt_get_int(OPT_SHUFFLE_PROCS);

   __trace_on = opt_get_int(OPT_RT_TRACE);

   create_processes(m, m->root);

   nvc_rusage(&m->ready_rusage);

   // Initialisation is described in LRM 93 section 12.6.4

   reset_scope(m, m->root);

   if (m->force_stop)
      return;   // Error in intialisation

#if TRACE_SIGNALS > 0
   if (__trace_on)
      dump_signals(m, m->root);
#endif

   TRACE("calculate initial signal values");

   model_thread_t *thread = model_thread(m);
   thread->tlab = tlab_acquire(m->mspace);

   // The signals in the model are updated as follows in an order such
   // that if a given signal R depends upon the current value of another
   // signal S, then the current value of S is updated prior to the
   // updating of the current value of R.

   for (rt_nexus_t *n = m->nexuses; n != NULL; n = n->chain) {
      // The initial value of each driver is the default value of the signal
      if (n->n_sources > 0) {
         for (rt_source_t *s = &(n->sources); s; s = s->chain_input) {
            if (s->tag == SOURCE_DRIVER)
               copy_value_ptr(n, &(s->u.driver.waveforms.value),
                              nexus_effective(n));
         }
      }

      const int rank = nexus_rank(n);
      if (rank > MAX_RANK)
         fatal_at(tree_loc(n->signal->where), "signal rank %d is greater "
                  "than the maximum supported %d", rank, MAX_RANK);
      else if (rank > 0 || n->n_sources > 1)
         heap_insert(m->driving_heap, rank, n);
      else {
         calculate_initial_value(m, n);
         check_undriven_std_logic(n);
      }
   }

   while (heap_size(m->driving_heap) > 0) {
      rt_nexus_t *n = heap_extract_min(m->driving_heap);
      calculate_initial_value(m, n);
      check_undriven_std_logic(n);
   }

   // Update effective values after all initial driving values calculated
   while (heap_size(m->effective_heap) > 0) {
      rt_nexus_t *n = heap_extract_min(m->effective_heap);
      n->flags &= ~NET_F_PENDING;

      calculate_effective_value(m, n);
   }

   tlab_reset(thread->tlab);   // No allocations can be live past here

   global_event(m, RT_END_OF_INITIALISATION);
}

static void update_property(rt_model_t *m, rt_prop_t *prop)
{
   TRACE("update property %s state %s", istr(prop->name),
         trace_states(&prop->state));

   rt_wakeable_t *obj = &(prop->wakeable);

   if (obj->trigger != NULL && !run_trigger(m, obj->trigger))
      return;   // Filtered

   model_thread_t *thread = model_thread(m);
   assert(thread->tlab != NULL);

   thread->active_obj = obj;
   thread->active_scope = prop->scope;

   jit_scalar_t args[] = {
      { .pointer = *mptr_get(prop->privdata) ?: (void *)-1 },
      { .pointer = *mptr_get(prop->scope->privdata) },
      { .integer = -1 },
   };

   mask_clearall(&prop->newstate);
   prop->strong = false;

   size_t bit = -1;
   while (mask_iter(&prop->state, &bit)) {
      args[2].integer = bit;

      if (!jit_vfastcall(m->jit, prop->handle, args, ARRAY_LEN(args),
                         NULL, 0, thread->tlab))
         m->force_stop = true;
   }

   tlab_reset(thread->tlab);   // No allocations can be live past here

   thread->active_obj = NULL;
   thread->active_scope = NULL;

   TRACE("new state %s%s", trace_states(&prop->newstate),
         prop->strong ? " strong" : "");

   mask_copy(&prop->state, &prop->newstate);

   m->liveness |= prop->strong;
}

static void sched_event(rt_model_t *m, rt_nexus_t *n, rt_wakeable_t *obj)
{
   if (n->pending == NULL)
      n->pending = tag_pointer(obj, 1);
   else if (pointer_tag(n->pending) == 1) {
      rt_wakeable_t *cur = untag_pointer(n->pending, rt_wakeable_t);
      if (cur == obj)
         return;

      rt_pending_t *p = xmalloc_flex(sizeof(rt_pending_t), PENDING_MIN,
                                     sizeof(rt_wakeable_t *));
      p->max = PENDING_MIN;
      p->count = 2;
      p->wake[0] = cur;
      p->wake[1] = obj;

      n->pending = tag_pointer(p, 0);
   }
   else {
      rt_pending_t *p = untag_pointer(n->pending, rt_pending_t);

      for (int i = 0; i < p->count; i++) {
         if (p->wake[i] == NULL || p->wake[i] == obj) {
            p->wake[i] = obj;
            return;
         }
      }

      if (p->count == p->max) {
         p->max = MAX(PENDING_MIN, p->max * 2);
         p = xrealloc_flex(p, sizeof(rt_pending_t), p->max,
                           sizeof(rt_wakeable_t *));
         n->pending = tag_pointer(p, 0);
      }

      p->wake[p->count++] = obj;
   }
}

static void clear_event(rt_model_t *m, rt_nexus_t *n, rt_wakeable_t *obj)
{
   if (pointer_tag(n->pending) == 1) {
      rt_wakeable_t *wake = untag_pointer(n->pending, rt_wakeable_t);
      if (wake == obj)
         n->pending = NULL;
   }
   else if (n->pending != NULL) {
      rt_pending_t *p = untag_pointer(n->pending, rt_pending_t);
      for (int i = 0; i < p->count; i++) {
         if (p->wake[i] == obj) {
            p->wake[i] = NULL;
            return;
         }
      }
   }
}

static rt_source_t *find_driver(rt_nexus_t *nexus, rt_proc_t *proc)
{
   // Try to find this process in the list of existing drivers
   for (rt_source_t *d = &(nexus->sources); d; d = d->chain_input) {
      if (d->tag == SOURCE_DRIVER && d->u.driver.proc == proc)
         return d;
   }

   return NULL;
}

static inline bool insert_transaction(rt_model_t *m, rt_nexus_t *nexus,
                                      rt_source_t *source, waveform_t *w,
                                      uint64_t when, uint64_t reject)
{
   waveform_t *last = &(source->u.driver.waveforms);
   waveform_t *it   = last->next;
   while (it != NULL && it->when < when) {
      // If the current transaction is within the pulse rejection interval
      // and the value is different to that of the new transaction then
      // delete the current transaction
      assert(it->when >= m->now);
      if (it->when >= when - reject
          && !cmp_values(nexus, it->value, w->value)) {
         waveform_t *next = it->next;
         last->next = next;
         free_value(nexus, it->value);
         free_waveform(m, it);
         it = next;
      }
      else {
         last = it;
         it = it->next;
      }
   }
   last->next = w;

   // Delete all transactions later than this
   // We could remove this transaction from the deltaq as well but the
   // overhead of doing so is probably higher than the cost of waking
   // up for the empty event
   bool already_scheduled = false;
   for (waveform_t *next; it != NULL; it = next) {
      next = it->next;
      already_scheduled |= (it->when == when);
      free_value(nexus, it->value);
      free_waveform(m, it);
   }

   return already_scheduled;
}

static void sched_driver(rt_model_t *m, rt_nexus_t *n, uint64_t after,
                         uint64_t reject, const void *value, rt_proc_t *proc)
{
   if (after == 0 && (n->flags & NET_F_FAST_DRIVER)) {
      rt_source_t *d = &(n->sources);
      assert(n->n_sources == 1);

      waveform_t *w = &d->u.driver.waveforms;
      w->when = m->now;
      assert(w->next == NULL);

      rt_signal_t *signal = n->signal;
      rt_source_t *d0 = &(signal->nexus.sources);

      if (d->fastqueued)
         assert(m->next_is_delta);
      else if ((signal->shared.flags & NET_F_FAST_DRIVER) && d0->sigqueued) {
         assert(m->next_is_delta);
         d->fastqueued = 1;
      }
      else if (cmp_bytes(value, value_ptr(n, &w->value), n->width * n->size)) {
         m->next_is_delta = true;
         d->was_active = (n->active_delta == m->iteration);
         n->active_delta = m->iteration + 1;
         return;
      }
      else if (signal->shared.flags & NET_F_FAST_DRIVER) {
         deferq_do(&m->delta_driverq, async_fast_all_drivers, signal);
         m->next_is_delta = true;
         d0->sigqueued = 1;
         d->fastqueued = 1;
      }
      else {
         deferq_do(&m->delta_driverq, async_fast_driver, d);
         m->next_is_delta = true;
         d->fastqueued = 1;
      }

      copy_value_ptr(n, &w->value, value);
   }
   else {
      rt_source_t *d = find_driver(n, proc);
      assert(d != NULL);

      if ((n->flags & NET_F_FAST_DRIVER) && d->fastqueued) {
         // A fast update to this driver is already scheduled
         waveform_t *w0 = alloc_waveform(m);
         w0->when  = m->now;
         w0->next  = NULL;
         w0->value = alloc_value(m, n);

         const uint8_t *prev = value_ptr(n, &(d->u.driver.waveforms.value));
         copy_value_ptr(n, &w0->value, prev);

         assert(d->u.driver.waveforms.next == NULL);
         d->u.driver.waveforms.next = w0;
      }

      n->flags &= ~NET_F_FAST_DRIVER;

      waveform_t *w = alloc_waveform(m);
      w->when  = m->now + after;
      w->next  = NULL;
      w->value = alloc_value(m, n);

      copy_value_ptr(n, &w->value, value);

      if (!insert_transaction(m, n, d, w, w->when, reject))
         deltaq_insert_driver(m, after, d);
   }
}

static void sched_disconnect(rt_model_t *m, rt_nexus_t *nexus, uint64_t after,
                             uint64_t reject, rt_proc_t *proc)
{
   rt_source_t *d = find_driver(nexus, proc);
   assert(d != NULL);

   const uint64_t when = m->now + after;

   // Need update_driver to clear disconnected flag
   nexus->flags &= ~NET_F_FAST_DRIVER;

   waveform_t *w = alloc_waveform(m);
   w->when = -when;   // Use sign bit to represent null
   w->next = NULL;
   w->value.qword = 0;

   if (!insert_transaction(m, nexus, d, w, when, reject))
      deltaq_insert_driver(m, after, d);
}

static void async_watch_callback(rt_model_t *m, void *arg)
{
   rt_watch_t *w = arg;

   assert(w->wakeable.pending);
   w->wakeable.pending = false;

   if (w->wakeable.zombie)
      free(w);
   else
      (*w->fn)(m->now, w->signals[0], w, w->user_data);
}

static void async_timeout_callback(rt_model_t *m, void *arg)
{
   rt_callback_t *cb = arg;
   (*cb->fn)(m, cb->user);
   free(cb);
}

static void async_update_implicit_signal(rt_model_t *m, void *arg)
{
   rt_implicit_t *imp = arg;

   assert(imp->wakeable.pending);
   imp->wakeable.pending = false;

   update_implicit_signal(m, imp);
}

static void async_run_process(rt_model_t *m, void *arg)
{
   rt_proc_t *proc = arg;

   assert(proc->wakeable.pending);
   proc->wakeable.pending = false;

   run_process(m, proc);
}

static void async_update_property(rt_model_t *m, void *arg)
{
   rt_prop_t *prop = arg;

   assert(prop->wakeable.pending);
   prop->wakeable.pending = false;

   update_property(m, prop);
}

static bool heap_delete_proc_cb(uint64_t key, void *value, void *search)
{
   if (pointer_tag(value) != EVENT_PROCESS)
      return false;

   return untag_pointer(value, rt_proc_t) == search;
}

static bool run_trigger(rt_model_t *m, rt_trigger_t *t)
{
   if (t->when == m->now && t->iteration == m->iteration)
      return t->result.integer != 0;   // Cached

   switch (t->kind) {
   case FUNC_TRIGGER:
      {
         tlab_t tlab = jit_null_tlab(m->jit);
         if (!jit_vfastcall(m->jit, t->handle, t->args, t->nargs,
                            &t->result, 1, &tlab))
            m->force_stop = true;

         TRACE("run trigger %p %s ==> %"PRIi64, t,
               istr(jit_get_name(m->jit, t->handle)), t->result.integer);
      }
      break;

   case OR_TRIGGER:
      {
         rt_trigger_t *left = t->args[0].pointer;
         rt_trigger_t *right = t->args[1].pointer;
         t->result.integer = run_trigger(m, left) || run_trigger(m, right);

         TRACE("or trigger %p ==> %"PRIi64, t, t->result.integer);
      }
      break;

   case CMP_TRIGGER:
      {
         rt_signal_t *s = t->args[0].pointer;
         uint32_t offset = t->args[1].integer;
         int64_t right = t->args[2].integer;

#define COMPARE_SCALAR(type) do {                                       \
            const type *data = (type *)s->shared.data;                  \
            t->result.integer = (data[offset] == right);                \
      } while (0)

         FOR_ALL_SIZES(s->nexus.size, COMPARE_SCALAR);

         TRACE("cmp trigger %p ==> %"PRIi64, t, t->result.integer);
      }
      break;
   }

   t->when = m->now;
   t->iteration = m->iteration;

   return t->result.integer != 0;
}

static void wakeup_one(rt_model_t *m, rt_wakeable_t *obj)
{
   if (obj->pending)
      return;   // Already scheduled

   deferq_t *dq = obj->postponed ? &m->postponedq : &m->procq;

   switch (obj->kind) {
   case W_PROC:
      {
         rt_proc_t *proc = container_of(obj, rt_proc_t, wakeable);
         TRACE("wakeup %sprocess %s", obj->postponed ? "postponed " : "",
               istr(proc->name));
         deferq_do(dq, async_run_process, proc);

         if (proc->wakeable.delayed) {
            // This process was already scheduled to run at a later
            // time so we need to delete it from the simulation queue
            heap_delete(m->eventq_heap, heap_delete_proc_cb, proc);
            proc->wakeable.delayed = false;
         }
      }
      break;

   case W_PROPERTY:
      {
         rt_prop_t *prop = container_of(obj, rt_prop_t, wakeable);
         TRACE("wakeup property %s", istr(prop->name));
         deferq_do(dq, async_update_property, prop);
      }
      break;

   case W_IMPLICIT:
      {
         rt_implicit_t *imp = container_of(obj, rt_implicit_t, wakeable);
         TRACE("wakeup implicit signal %s closure %s",
               istr(tree_ident(imp->signal.where)),
               istr(jit_get_name(m->jit, imp->closure.handle)));
         deferq_do(&m->implicitq, async_update_implicit_signal, imp);
      }
      break;

   case W_WATCH:
      {
         rt_watch_t *w = container_of(obj, rt_watch_t, wakeable);
         TRACE("wakeup %svalue change callback %p %s",
               obj->postponed ? "postponed " : "", w,
               debug_symbol_name(w->fn));

         assert(!w->wakeable.zombie);
         deferq_do(dq, async_watch_callback, w);
      }
      break;

   case W_TRANSFER:
      {
         rt_transfer_t *t = container_of(obj, rt_transfer_t, wakeable);
         TRACE("wakeup signal transfer for %s",
               istr(tree_ident(t->target->signal->where)));
         deferq_do(dq, async_transfer_signal, t);
      }
      break;
   }

   set_pending(obj);
}

static void notify_event(rt_model_t *m, rt_nexus_t *n)
{
   // Must only be called once per cycle
   assert(n->last_event != m->now || n->event_delta != m->iteration);

   n->last_event = m->now;
   n->event_delta = m->iteration;

   if (n->flags & NET_F_CACHE_EVENT)
      n->signal->shared.flags |= SIG_F_EVENT_FLAG;

   if (pointer_tag(n->pending) == 1) {
      rt_wakeable_t *wake = untag_pointer(n->pending, rt_wakeable_t);
      wakeup_one(m, wake);
   }
   else if (n->pending != NULL) {
      rt_pending_t *p = untag_pointer(n->pending, rt_pending_t);
      for (int i = 0; i < p->count; i++) {
         if (p->wake[i] != NULL)
            wakeup_one(m, p->wake[i]);
      }
   }
}

static void put_effective(rt_model_t *m, rt_nexus_t *n, const void *value)
{
   TRACE("update %s effective value %s", trace_nexus(n), fmt_nexus(n, value));

   unsigned char *eff = nexus_effective(n);
   unsigned char *last = nexus_last_value(n);

   const size_t valuesz = n->size * n->width;

   if (!cmp_bytes(eff, value, valuesz)) {
      copy2(last, eff, value, valuesz);
      notify_event(m, n);
   }
}

static void enqueue_effective(rt_model_t *m, rt_nexus_t *n)
{
   if (n->flags & NET_F_PENDING)
      return;

   n->flags |= NET_F_PENDING;
   heap_insert(m->effective_heap, MAX_RANK - n->rank, n);
}

static void update_effective(rt_model_t *m, rt_nexus_t *n)
{
   n->active_delta = m->iteration;
   n->flags &= ~NET_F_PENDING;

   calculate_effective_value(m, n);

   if (n->n_sources > 0) {
      for (rt_source_t *s = &(n->sources); s; s = s->chain_input) {
         if (s->tag != SOURCE_PORT)
            continue;
         else if (s->u.port.conv_func != NULL) {
            rt_conv_func_t *cf = s->u.port.conv_func;
            for (int i = 0; i < cf->ninputs; i++) {
               if (cf->inputs[i].nexus->flags & NET_F_INOUT)
                  enqueue_effective(m, cf->inputs[i].nexus);
            }
         }
         else if (s->u.port.input->flags & NET_F_INOUT)
            enqueue_effective(m, s->u.port.input);
      }
   }
}

static void put_driving(rt_model_t *m, rt_nexus_t *n, const void *value)
{
   if (n->flags & NET_F_EFFECTIVE) {
      TRACE("update %s driving value %s", trace_nexus(n), fmt_nexus(n, value));

      memcpy(nexus_driving(n), value, n->size * n->width);

      assert(!(n->flags & NET_F_PENDING));
      n->flags |= NET_F_PENDING;
      n->flags &= ~NET_F_HAS_INITIAL;
      heap_insert(m->effective_heap, MAX_RANK - n->rank, n);
   }
   else
      put_effective(m, n, value);
}

static void defer_driving_update(rt_model_t *m, rt_nexus_t *n)
{
   if (n->flags & NET_F_PENDING)
      return;

   TRACE("defer %s driving value update", trace_nexus(n));
   heap_insert(m->driving_heap, n->rank, n);
   n->flags |= NET_F_PENDING;
}

static void update_driving(rt_model_t *m, rt_nexus_t *n, bool safe)
{
   if (n->n_sources == 1 || safe) {
      n->active_delta = m->iteration;
      n->flags &= ~NET_F_PENDING;

      calculate_driving_value(m, n);

      // Update outputs if the effective value must be calculated
      // separately or there was an event on this signal
      const bool update_outputs = !!(n->flags & NET_F_EFFECTIVE)
         || (n->event_delta == m->iteration && n->last_event == m->now);

      if (update_outputs) {
         for (rt_source_t *o = n->outputs; o; o = o->chain_output) {
            switch (o->tag) {
            case SOURCE_PORT:
               if (o->u.port.conv_func != NULL)
                  defer_driving_update(m, o->u.port.output);
               else
                  update_driving(m, o->u.port.output, false);
               break;
            case SOURCE_IMPLICIT:
               update_driving(m, o->u.pseudo.nexus , false);
               break;
            default:
               should_not_reach_here();
            }
         }
      }
   }
   else
      defer_driving_update(m, n);
}

static void update_driver(rt_model_t *m, rt_nexus_t *n, rt_source_t *source)
{
   waveform_t *w_now  = &(source->u.driver.waveforms);
   waveform_t *w_next = w_now->next;

   if (likely(w_next != NULL && w_next->when == m->now)) {
      free_value(n, w_now->value);
      *w_now = *w_next;
      free_waveform(m, w_next);
      source->disconnected = 0;
      update_driving(m, n, false);
   }
   else if (unlikely(w_next != NULL && w_next->when == -m->now)) {
      // Disconnect source due to null transaction
      *w_now = *w_next;
      free_waveform(m, w_next);
      source->disconnected = 1;
      update_driving(m, n, false);
   }
}

static void fast_update_driver(rt_model_t *m, rt_nexus_t *nexus)
{
   rt_source_t *src = &(nexus->sources);

   if (likely(nexus->flags & NET_F_FAST_DRIVER)) {
      // Preconditions for fast driver updates
      assert(nexus->n_sources == 1);
      assert(src->tag == SOURCE_DRIVER);
      assert(src->u.driver.waveforms.next == NULL);

      update_driving(m, nexus, false);
   }
   else
      update_driver(m, nexus, src);

   assert(src->fastqueued);
   src->fastqueued = 0;
}

static void fast_update_all_drivers(rt_model_t *m, rt_signal_t *signal)
{
   assert(signal->shared.flags & NET_F_FAST_DRIVER);

   rt_nexus_t *n = &(signal->nexus);
   assert(n->sources.sigqueued);
   n->sources.sigqueued = 0;

   int count = 0;
   for (int i = 0; i < signal->n_nexus; i++, n = n->chain) {
      if (n->sources.fastqueued) {
         fast_update_driver(m, n);
         count++;
      }
   }

   if (count < signal->n_nexus >> 1) {
      // Unlikely to be worth the iteration cost
      signal->shared.flags &= ~NET_F_FAST_DRIVER;
   }
}

static void async_update_driver(rt_model_t *m, void *arg)
{
   rt_source_t *src = arg;
   update_driver(m, src->u.driver.nexus, src);
}

static void async_fast_driver(rt_model_t *m, void *arg)
{
   rt_source_t *src = arg;
   fast_update_driver(m, src->u.driver.nexus);
}

static void async_fast_all_drivers(rt_model_t *m, void *arg)
{
   rt_signal_t *signal = arg;
   fast_update_all_drivers(m, signal);
}

static void async_pseudo_source(rt_model_t *m, void *arg)
{
   rt_source_t *src = arg;
   assert(src->tag == SOURCE_FORCING || src->tag == SOURCE_DEPOSIT);

   update_driving(m, src->u.pseudo.nexus, false);

   assert(src->pseudoqueued);
   src->pseudoqueued = 0;
}

static void async_transfer_signal(rt_model_t *m, void *arg)
{
   rt_transfer_t *t = arg;

   assert(t->wakeable.pending);
   t->wakeable.pending = false;

   rt_nexus_t *n = t->target;
   char *vptr = nexus_effective(t->source);
   for (int count = t->count; count > 0; n = n->chain) {
      count -= n->width;
      assert(count >= 0);

      sched_driver(m, n, t->after, t->reject, vptr, t->proc);
      vptr += n->width * n->size;
   }
}

static void update_implicit_signal(rt_model_t *m, rt_implicit_t *imp)
{
   model_thread_t *thread = model_thread(m);
   assert(thread->active_obj == NULL);
   thread->active_obj = &(imp->wakeable);

   jit_scalar_t result;
   if (!jit_try_call(m->jit, imp->closure.handle, &result,
                     imp->closure.context, imp->signal.shared.data[0]))
      m->force_stop = true;

   thread->active_obj = NULL;

   TRACE("implicit signal %s new value %"PRIi64,
         istr(tree_ident(imp->signal.where)), result.integer);

   assert(imp->signal.n_nexus == 1);
   rt_nexus_t *n0 = &(imp->signal.nexus);

   n0->active_delta = m->iteration;

   if (n0->n_sources > 0 && n0->sources.tag == SOURCE_DRIVER) {
      if (!result.integer) {
         // Update driver for 'STABLE and 'QUIET
         // TODO: this should happen inside the callback
         waveform_t *w = alloc_waveform(m);
         w->when  = m->now + imp->delay;
         w->next  = NULL;
         w->value = alloc_value(m, n0);

         w->value.bytes[0] = 1;   // Boolean TRUE

         if (!insert_transaction(m, n0, &(n0->sources), w, w->when, imp->delay))
            deltaq_insert_driver(m, imp->delay, &(n0->sources));

         put_effective(m, n0, &result.integer);
      }
      else if (n0->sources.u.driver.waveforms.next == NULL)
         put_effective(m, n0, &result.integer);
   }
   else
      put_effective(m, n0, &result.integer);
}

static void iteration_limit_proc_cb(void *fn, void *arg, void *extra)
{
   diag_t *d = extra;
   rt_proc_t *proc = NULL;

   if (fn == async_run_process)
      proc = arg;
   else if (fn == async_transfer_signal) {
      rt_transfer_t *t = arg;
      proc = t->proc;
   }

   if (proc == NULL)
      return;

   const loc_t *loc = tree_loc(proc->where);
   diag_hint(d, loc, "process %s is active", istr(proc->name));
}

static void iteration_limit_driver_cb(void *fn, void *arg, void *extra)
{
   diag_t *d = extra;
   tree_t decl = NULL;

   if (fn == async_update_driver || fn == async_fast_driver) {
      rt_source_t *src = arg;
      if (src->tag == SOURCE_DRIVER)
         decl = src->u.driver.nexus->signal->where;
   }
   else if (fn == async_fast_all_drivers) {
      rt_signal_t *s = arg;
      decl = s->where;
   }

   if (decl == NULL)
      return;

   diag_hint(d, tree_loc(decl), "driver for %s %s is active",
             tree_kind(decl) == T_PORT_DECL ? "port" : "signal",
             istr(tree_ident(decl)));
}

static void reached_iteration_limit(rt_model_t *m)
{
   diag_t *d = diag_new(DIAG_FATAL, NULL);

   diag_printf(d, "limit of %d delta cycles reached", m->stop_delta);

   deferq_scan(&m->delta_procq, iteration_limit_proc_cb, d);
   deferq_scan(&m->delta_driverq, iteration_limit_driver_cb, d);

   diag_hint(d, NULL, "you can increase this limit with $bold$--stop-delta$$");
   diag_emit(d);

   m->force_stop = true;
}

static void sync_event_cache(rt_model_t *m)
{
   for (int i = 0; i < m->eventsigs.count; i++) {
      rt_signal_t *s = m->eventsigs.items[i];
      assert(s->shared.flags & SIG_F_CACHE_EVENT);

      const bool event = s->nexus.last_event == m->now
         && s->nexus.event_delta == m->iteration;

      TRACE("sync event flag %d for %s", event, istr(tree_ident(s->where)));

      if (event)
         assert(s->shared.flags & SIG_F_EVENT_FLAG);   // Set by notify_event
      else
         s->shared.flags &= ~SIG_F_EVENT_FLAG;
   }
}

static void swap_deferq(deferq_t *a, deferq_t *b)
{
   deferq_t tmp = *a;
   *a = *b;
   *b = tmp;
}

static void model_cycle(rt_model_t *m)
{
   // Simulation cycle is described in LRM 93 section 12.6.4

   const bool is_delta_cycle = m->next_is_delta;
   m->next_is_delta = false;

   if (is_delta_cycle)
      m->iteration = m->iteration + 1;
   else {
      m->now = heap_min_key(m->eventq_heap);
      m->iteration = 0;
   }

   TRACE("begin cycle");

#if TRACE_DELTAQ > 0
   if (__trace_on)
      deltaq_dump(m);
#endif

   swap_deferq(&m->procq, &m->delta_procq);
   swap_deferq(&m->driverq, &m->delta_driverq);

   if (m->iteration == 0)
      global_event(m, RT_NEXT_TIME_STEP);

   global_event(m, RT_NEXT_CYCLE);

   if (!is_delta_cycle) {
      for (;;) {
         void *e = heap_extract_min(m->eventq_heap);
         switch (pointer_tag(e)) {
         case EVENT_PROCESS:
            {
               rt_proc_t *proc = untag_pointer(e, rt_proc_t);
               assert(proc->wakeable.delayed);
               proc->wakeable.delayed = false;
               set_pending(&proc->wakeable);
               deferq_do(&m->procq, async_run_process, proc);
            }
            break;
         case EVENT_DRIVER:
            {
               rt_source_t *source = untag_pointer(e, rt_source_t);
               deferq_do(&m->driverq, async_update_driver, source);
            }
            break;
         case EVENT_TIMEOUT:
            {
               rt_callback_t *cb = untag_pointer(e, rt_callback_t);
               deferq_do(&m->driverq, async_timeout_callback, cb);
            }
            break;
         }

         if (heap_size(m->eventq_heap) == 0)
            break;
         else if (heap_min_key(m->eventq_heap) > m->now)
            break;
      }
   }

   deferq_run(m, &m->driverq);

   while (heap_size(m->driving_heap) > 0) {
      rt_nexus_t *n = heap_extract_min(m->driving_heap);
      update_driving(m, n, true);
   }

   while (heap_size(m->effective_heap) > 0) {
      rt_nexus_t *n = heap_extract_min(m->effective_heap);
      update_effective(m, n);
   }

   sync_event_cache(m);

   // Update implicit signals
   deferq_run(m, &m->implicitq);

#if TRACE_SIGNALS > 0
   if (__trace_on)
      dump_signals(m, m->root);
#endif

   if (m->shuffle)
      deferq_shuffle(&m->procq);

   // Run all non-postponed processes and event callbacks
   deferq_run(m, &m->procq);

   global_event(m, RT_END_OF_PROCESSES);

   if (!m->next_is_delta)
      global_event(m, RT_LAST_KNOWN_DELTA_CYCLE);

   if (!m->next_is_delta) {
      m->can_create_delta = false;

      // Run all postponed processes and event callbacks
      deferq_run(m, &m->postponedq);

      global_event(m, RT_END_TIME_STEP);

      m->can_create_delta = true;
   }
   else if (m->stop_delta > 0 && m->iteration == m->stop_delta)
      reached_iteration_limit(m);
}

static bool should_stop_now(rt_model_t *m, uint64_t stop_time)
{
   if (m->force_stop) {
      // Make sure we print the interrupted message if this was the
      // result of an interrupt
      jit_check_interrupt(m->jit);
      return true;
   }
   else if (m->next_is_delta)
      return false;
   else if (heap_size(m->eventq_heap) == 0)
      return true;
   else
      return heap_min_key(m->eventq_heap) > stop_time;
}

static void check_liveness_properties(rt_model_t *m, rt_scope_t *s)
{
   model_thread_t *thread = model_thread(m);

   for (int i = 0; i < s->properties.count; i++) {
      rt_prop_t *p = s->properties.items[i];
      if (p->strong) {
         TRACE("property %s in strong state", istr(p->name));

         // Passing an invalid state triggers the assertion failure
         jit_scalar_t args[] = {
            { .pointer = *mptr_get(p->privdata) ?: (void *)-1 },
            { .pointer = *mptr_get(p->scope->privdata) },
            { .integer = INT_MAX },
         };
         jit_vfastcall(m->jit, p->handle, args, ARRAY_LEN(args),
                       NULL, 0, thread->tlab);
      }
   }

   for (int i = 0; i < s->children.count; i++)
      check_liveness_properties(m, s->children.items[i]);
}

void model_run(rt_model_t *m, uint64_t stop_time)
{
   MODEL_ENTRY(m);

   if (m->force_stop)
      return;   // Was error during intialisation

   global_event(m, RT_START_OF_SIMULATION);

   while (!should_stop_now(m, stop_time))
      model_cycle(m);

   global_event(m, RT_END_OF_SIMULATION);

   if (m->liveness)
      check_liveness_properties(m, m->root);
}

bool model_step(rt_model_t *m)
{
   MODEL_ENTRY(m);

   if (!m->force_stop)
      model_cycle(m);

   return should_stop_now(m, TIME_HIGH);
}

static inline void check_postponed(int64_t after, rt_proc_t *proc)
{
   if (unlikely(proc->wakeable.postponed && (after == 0)))
      fatal("postponed process %s cannot cause a delta cycle",
            istr(proc->name));
}

static inline void check_reject_limit(rt_signal_t *s, uint64_t after,
                                      uint64_t reject)
{
   if (unlikely(reject > after))
      jit_msg(NULL, DIAG_FATAL, "signal %s pulse reject limit %s is greater "
              "than delay %s", istr(tree_ident(s->where)),
              trace_time(reject), trace_time(after));
}

static inline void check_delay(int64_t delay)
{
   if (unlikely(delay < 0)) {
      char buf[32];
      fmt_time_r(buf, sizeof(buf), delay, " ");
      jit_msg(NULL, DIAG_FATAL, "illegal negative delay %s", buf);
   }
}

void force_signal(rt_model_t *m, rt_signal_t *s, const void *values,
                  int offset, size_t count)
{
   RT_LOCK(s->lock);

   TRACE("force signal %s+%d to %s", istr(tree_ident(s->where)), offset,
         fmt_values(values, count));

   assert(m->can_create_delta);

   rt_nexus_t *n = split_nexus(m, s, offset, count);
   const char *vptr = values;
   for (; count > 0; n = n->chain) {
      count -= n->width;
      assert(count >= 0);

      n->flags |= NET_F_FORCED;

      rt_source_t *src = get_pseudo_source(m, n, SOURCE_FORCING);
      copy_value_ptr(n, &(src->u.pseudo.value), vptr);
      src->disconnected = 0;

      if (!src->pseudoqueued) {
         deltaq_insert_pseudo_source(m, src);
         src->pseudoqueued = 1;
      }

      vptr += n->width * n->size;
   }
}

void release_signal(rt_model_t *m, rt_signal_t *s, int offset, size_t count)
{
   RT_LOCK(s->lock);

   TRACE("release signal %s+%d", istr(tree_ident(s->where)), offset);

   assert(m->can_create_delta);

   rt_nexus_t *n = split_nexus(m, s, offset, count);
   for (; count > 0; n = n->chain) {
      count -= n->width;
      assert(count >= 0);

      n->flags &= ~NET_F_FORCED;

      rt_source_t *src = get_pseudo_source(m, n, SOURCE_FORCING);
      src->disconnected = 1;

      if (!src->pseudoqueued) {
         deltaq_insert_pseudo_source(m, src);
         src->pseudoqueued = 1;
      }
   }
}

void deposit_signal(rt_model_t *m, rt_signal_t *s, const void *values,
                    int offset, size_t count)
{
   RT_LOCK(s->lock);

   TRACE("deposit signal %s+%d to %s", istr(tree_ident(s->where)),
         offset, fmt_values(values, count));

   assert(m->can_create_delta);

   rt_nexus_t *n = split_nexus(m, s, offset, count);
   const char *vptr = values;
   for (; count > 0; n = n->chain) {
      count -= n->width;
      assert(count >= 0);

      rt_source_t *src = get_pseudo_source(m, n, SOURCE_DEPOSIT);
      copy_value_ptr(n, &(src->u.pseudo.value), vptr);
      src->disconnected = 0;

      if (!src->pseudoqueued) {
         deltaq_insert_pseudo_source(m, src);
         src->pseudoqueued = 1;
      }

      vptr += n->width * n->size;
   }
}

bool model_can_create_delta(rt_model_t *m)
{
   return m->can_create_delta;
}

int64_t model_now(rt_model_t *m, unsigned *deltas)
{
   if (deltas != NULL)
      *deltas = MAX(m->iteration, 0);

   return m->now;
}

int64_t model_next_time(rt_model_t *m)
{
   if (heap_size(m->eventq_heap) == 0)
      return TIME_HIGH;
   else
      return heap_min_key(m->eventq_heap);
}

void model_stop(rt_model_t *m)
{
   relaxed_store(&m->force_stop, true);
}

void model_set_global_cb(rt_model_t *m, rt_event_t event, rt_event_fn_t fn,
                         void *user)
{
   assert(event < RT_LAST_EVENT);

   // Add to end of list so callbacks are called in registration order
   rt_callback_t **p = &(m->global_cbs[event]);
   for (; *p; p = &(*p)->next);

   rt_callback_t *cb = xcalloc(sizeof(rt_callback_t));
   cb->next = NULL;
   cb->fn   = fn;
   cb->user = user;

   *p = cb;
}

void model_set_timeout_cb(rt_model_t *m, uint64_t when, rt_event_fn_t fn,
                          void *user)
{
   rt_callback_t *cb = xcalloc(sizeof(rt_callback_t));
   cb->next = NULL;
   cb->fn   = fn;
   cb->user = user;

   assert(when > m->now);   // TODO: delta timeouts?

   void *e = tag_pointer(cb, EVENT_TIMEOUT);
   heap_insert(m->eventq_heap, when, e);
}

rt_watch_t *watch_new(rt_model_t *m, sig_event_fn_t fn, void *user,
                      watch_kind_t kind, unsigned slots)
{
   rt_watch_t *w = xcalloc_flex(sizeof(rt_watch_t), slots,
                                sizeof(rt_signal_t *));
   w->fn        = fn;
   w->chain_all = m->watches;
   w->user_data = user;
   w->num_slots = slots;

   w->wakeable.kind      = W_WATCH;
   w->wakeable.postponed = (kind == WATCH_POSTPONED);
   w->wakeable.pending   = false;
   w->wakeable.delayed   = false;

   m->watches = w;

   return w;
}

void watch_free(rt_model_t *m, rt_watch_t *w)
{
   assert(!w->wakeable.zombie);

   for (int i = 0; i < w->next_slot; i++) {
      rt_nexus_t *n = &(w->signals[i]->nexus);
      for (int j = 0; j < w->signals[i]->n_nexus; j++, n = n->chain)
         clear_event(m, n, &(w->wakeable));
   }

   rt_watch_t **last = &m->watches;
   for (rt_watch_t *it = *last; it;
        last = &(it->chain_all), it = it->chain_all) {
      if (it == w) {
         *last = it->chain_all;
         if (w->wakeable.pending)
            w->wakeable.zombie = true;   // Will be freed in callback
         else
            free(w);
         return;
      }
   }

   should_not_reach_here();
}

rt_watch_t *model_set_event_cb(rt_model_t *m, rt_signal_t *s, rt_watch_t *w)
{
   assert(!w->wakeable.zombie);
   assert(w->next_slot < w->num_slots);

   w->signals[w->next_slot++] = s;

   rt_nexus_t *n = &(s->nexus);
   for (int i = 0; i < s->n_nexus; i++, n = n->chain)
      sched_event(m, n, &(w->wakeable));

   return w;
}

static void handle_interrupt_cb(jit_t *j, void *ctx)
{
   rt_proc_t *proc = get_active_proc();

   if (proc != NULL)
      jit_msg(NULL, DIAG_FATAL, "interrupted in process %s", istr(proc->name));
   else {
      diag_t *d = diag_new(DIAG_FATAL, NULL);
      diag_printf(d, "interrupted");
      diag_emit(d);
   }
}

void model_interrupt(rt_model_t *m)
{
   model_stop(m);
   jit_interrupt(m->jit, handle_interrupt_cb, m);
}

int model_exit_status(rt_model_t *m)
{
   int status;
   if (jit_exit_status(m->jit, &status))
      return status;
   else if (m->stop_delta > 0 && m->iteration == m->stop_delta)
      return EXIT_FAILURE;
   else
      return get_vhdl_assert_exit_status();
}

static bool nexus_active(rt_model_t *m, rt_nexus_t *nexus)
{
   if (nexus->n_sources > 0) {
      for (rt_source_t *s = &(nexus->sources); s; s = s->chain_input) {
         if (s->tag == SOURCE_PORT) {
            rt_conv_func_t *cf = s->u.port.conv_func;
            if (cf == NULL) {
               RT_LOCK(s->u.port.input->signal->lock);
               if (nexus_active(m, s->u.port.input))
                  return true;
            }
            else {
               for (int i = 0; i < cf->ninputs; i++) {
                  if (nexus_active(m, cf->inputs[i].nexus))
                     return true;
               }
            }
         }
         else if (s->tag == SOURCE_DRIVER
                  && s->u.driver.waveforms.when == m->now) {
            if (nexus->active_delta == m->iteration)
               return true;
            else if (nexus->active_delta == m->iteration + 1 && s->was_active)
               return true;
         }
      }
   }

   return false;
}

static uint64_t nexus_last_active(rt_model_t *m, rt_nexus_t *nexus)
{
   int64_t last = TIME_HIGH;

   if (nexus->n_sources > 0) {
      for (rt_source_t *s = &(nexus->sources); s; s = s->chain_input) {
          if (s->tag == SOURCE_PORT) {
            rt_conv_func_t *cf = s->u.port.conv_func;
            if (cf == NULL) {
               RT_LOCK(s->u.port.input->signal->lock);
               last = MIN(last, nexus_last_active(m, s->u.port.input));
            }
            else {
               for (int i = 0; i < cf->ninputs; i++) {
                  RT_LOCK(cf->inputs[i]->signal->lock);
                  last = MIN(last, nexus_last_active(m, cf->inputs[i].nexus));
               }
            }
         }
         else if (s->tag == SOURCE_DRIVER
                  && s->u.driver.waveforms.when <= m->now)
            last = MIN(last, m->now - s->u.driver.waveforms.when);
      }
   }

   return last;
}

void get_forcing_value(rt_signal_t *s, uint8_t *value)
{
   uint8_t *p = value;
   rt_nexus_t *n = &(s->nexus);
   for (int i = 0; i < s->n_nexus; i++) {
      assert(n->n_sources > 0);
      rt_source_t *s = NULL;
      for (s = &(n->sources); s; s = s->chain_input) {
         if (s->tag == SOURCE_FORCING)
            break;
      }
      assert(s != NULL);

      memcpy(p, s->u.pseudo.value.bytes, n->width * n->size);
      p += n->width * n->size;
   }
   assert(p == value + s->shared.size);
}

int32_t *get_cover_counter(rt_model_t *m, int32_t tag, int count)
{
   assert(tag >= 0);
   assert(m->cover != NULL);
   return jit_get_cover_mem(m->jit, tag + count) + tag;
}

static rt_trigger_t *new_trigger(rt_model_t *m, trigger_kind_t kind,
                                 uint64_t hash, jit_handle_t handle,
                                 unsigned nargs, const jit_scalar_t *args)
{
   rt_trigger_t **bucket = &(m->triggertab[hash % TRIGGER_TAB_SIZE]);

   for (rt_trigger_t *exist = *bucket; exist; exist = exist->chain) {
      bool hit = exist->handle == handle
         && exist->nargs == nargs
         && exist->kind == kind;

      for (int i = 0; hit && i < nargs; i++)
         hit &= (exist->args[i].integer == args[i].integer);

      if (hit)
         return exist;
   }

   const size_t argsz = nargs * sizeof(jit_scalar_t);

   rt_trigger_t *t = static_alloc(m, sizeof(rt_trigger_t) + argsz);
   t->handle = handle;
   t->nargs  = nargs;
   t->when   = TIME_HIGH;
   t->kind   = kind;
   t->chain  = *bucket;
   memcpy(t->args, args, argsz);

   return (*bucket = t);
}

void call_with_model(rt_model_t *m, void (*cb)(void *), void *arg)
{
   MODEL_ENTRY(m);
   (*cb)(arg);
}

void get_instance_name(rt_scope_t *s, text_buf_t *tb)
{
   if (s->kind == SCOPE_ROOT)
      return;

   tree_t hier = tree_decl(s->where, 0);
   assert(tree_kind(hier) == T_HIER);

   switch (tree_subkind(hier)) {
   case T_ARCH:
      {
         tree_t unit = tree_ref(hier);

         get_instance_name(s->parent, tb);

         if (s->parent->kind != SCOPE_ROOT) {
            tree_t hier2 = tree_decl(s->parent->where, 0);
            assert(tree_kind(hier2) == T_HIER);

            if (tree_subkind(hier2) != T_COMPONENT) {
               tb_append(tb, ':');
               tb_istr(tb, tree_ident(s->where));
            }

            tb_append(tb, '@');
         }
         else
            tb_append(tb, ':');

         const char *arch = strchr(istr(tree_ident(unit)), '-') + 1;
         tb_printf(tb, "%s(%s)", istr(tree_ident2(unit)), arch);
      }
      break;

   case T_BLOCK:
   case T_FOR_GENERATE:
   case T_IF_GENERATE:
   case T_CASE_GENERATE:
      get_instance_name(s->parent, tb);
      tb_append(tb, ':');
      tb_printf(tb, "%s", istr(tree_ident(s->where)));
      break;

   case T_COMPONENT:
      get_instance_name(s->parent, tb);
      tb_printf(tb, ":%s", istr(tree_ident(s->where)));
      break;

   default:
      should_not_reach_here();
   }

   tb_downcase(tb);
}

void get_path_name(rt_scope_t *s, text_buf_t *tb)
{
   if (s->kind == SCOPE_ROOT)
      return;

   get_path_name(s->parent, tb);

   if (s->parent->kind != SCOPE_ROOT) {
      tree_t hier = tree_decl(s->parent->where, 0);
      assert(tree_kind(hier) == T_HIER);

      if (tree_subkind(hier) == T_COMPONENT)
         return;   // Skip implicit block for components
   }

   tb_append(tb, ':');
   tb_istr(tb, tree_ident(s->where));
   tb_downcase(tb);
}

////////////////////////////////////////////////////////////////////////////////
// Entry points from compiled code

sig_shared_t *x_init_signal(int64_t count, uint32_t size, jit_scalar_t value,
                            bool scalar, sig_flags_t flags, tree_t where,
                            int32_t offset)
{
   TRACE("init signal %s count=%"PRIi64" size=%d value=%s flags=%x offset=%d",
         istr(tree_ident(where)), count, size,
         fmt_jit_value(value, scalar, size * count), flags, offset);

   rt_model_t *m = get_model();

   if (count > INT32_MAX)
      jit_msg(tree_loc(where), DIAG_FATAL, "signal %s has %"PRIi64
              " sub-elements which is greater than the maximum supported %d",
              istr(tree_ident(where)), count, INT32_MAX);

   const size_t datasz = MAX(3 * count * size, 8);
   rt_signal_t *s = static_alloc(m, sizeof(rt_signal_t) + datasz);
   setup_signal(m, s, where, count, size, flags, offset);

   // The driving value area is also used to save the default value
   void *driving = s->shared.data + 2*s->shared.size;

   if (scalar) {
#define COPY_SCALAR(type) do {                  \
         type *pi = (type *)s->shared.data;     \
         type *pd = (type *)driving;            \
         for (int i = 0; i < count; i++)        \
            pi[i] = pd[i] = value.integer;      \
      } while (0)

      FOR_ALL_SIZES(size, COPY_SCALAR);
   }
   else {
      memcpy(s->shared.data, value.pointer, s->shared.size);
      memcpy(driving, value.pointer, s->shared.size);
   }

   return &(s->shared);
}

void x_drive_signal(sig_shared_t *ss, uint32_t offset, int32_t count)
{
   rt_signal_t *s = container_of(ss, rt_signal_t, shared);
   RT_LOCK(s->lock);

   TRACE("drive signal %s+%d count=%d", istr(tree_ident(s->where)),
         offset, count);

   rt_model_t *m = get_model();
   rt_proc_t *proc = get_active_proc();
   rt_nexus_t *n = split_nexus(m, s, offset, count);
   for (; count > 0; n = n->chain) {
      rt_source_t *s;
      for (s = &(n->sources); s; s = s->chain_input) {
         if (s->tag == SOURCE_DRIVER && s->u.driver.proc == proc)
            break;
      }

      if (s == NULL) {
         s = add_source(m, n, SOURCE_DRIVER);
         s->u.driver.waveforms.value = alloc_value(m, n);
         s->u.driver.proc = proc;
      }

      count -= n->width;
      assert(count >= 0);
   }
}

void x_sched_process(int64_t delay)
{
   rt_proc_t *proc = get_active_proc();
   if (proc == NULL)
      return;    // May be called during constant folding

   TRACE("schedule process %s delay=%s", istr(proc->name), trace_time(delay));

   check_delay(delay);
   deltaq_insert_proc(get_model(), delay, proc);
}

void x_sched_waveform_s(sig_shared_t *ss, uint32_t offset, uint64_t scalar,
                        int64_t after, int64_t reject)
{
   rt_signal_t *s = container_of(ss, rt_signal_t, shared);
   RT_LOCK(s->lock);

   TRACE("_sched_waveform_s %s+%d value=%"PRIi64" after=%s reject=%s",
         istr(tree_ident(s->where)), offset, scalar, trace_time(after),
         trace_time(reject));

   rt_proc_t *proc = get_active_proc();

   check_delay(after);
   check_postponed(after, proc);
   check_reject_limit(s, after, reject);

   rt_model_t *m = get_model();
   rt_nexus_t *n = split_nexus(m, s, offset, 1);

   sched_driver(m, n, after, reject, &scalar, proc);
}

void x_sched_waveform(sig_shared_t *ss, uint32_t offset, void *values,
                      int32_t count, int64_t after, int64_t reject)
{
   rt_signal_t *s = container_of(ss, rt_signal_t, shared);
   RT_LOCK(s->lock);

   TRACE("_sched_waveform %s+%d value=%s count=%d after=%s reject=%s",
         istr(tree_ident(s->where)), offset,
         fmt_values(values, count * s->nexus.size),
         count, trace_time(after), trace_time(reject));

   rt_proc_t *proc = get_active_proc();

   check_delay(after);
   check_postponed(after, proc);
   check_reject_limit(s, after, reject);

   rt_model_t *m = get_model();
   rt_nexus_t *n = split_nexus(m, s, offset, count);
   char *vptr = values;
   for (; count > 0; n = n->chain) {
      count -= n->width;
      assert(count >= 0);

      sched_driver(m, n, after, reject, vptr, proc);
      vptr += n->width * n->size;
   }
}

void x_transfer_signal(sig_shared_t *target_ss, uint32_t toffset,
                       sig_shared_t *source_ss, uint32_t soffset,
                       int32_t count, int64_t after, int64_t reject)
{
   rt_signal_t *target = container_of(target_ss, rt_signal_t, shared);
   rt_signal_t *source = container_of(source_ss, rt_signal_t, shared);

   TRACE("transfer signal %s+%d to %s+%d count=%d",
         istr(tree_ident(source->where)), soffset,
         istr(tree_ident(target->where)), toffset, count);

   rt_proc_t *proc = get_active_proc();

   check_delay(after);
   check_postponed(after, proc);
   check_reject_limit(target, after, reject);

   rt_model_t *m = get_model();

   rt_transfer_t *t = static_alloc(m, sizeof(rt_transfer_t));
   t->proc   = proc;
   t->target = split_nexus(m, target, toffset, count);
   t->source = split_nexus(m, source, soffset, count);
   t->count  = count;
   t->after  = after;
   t->reject = reject;

   t->wakeable.kind      = W_TRANSFER;
   t->wakeable.postponed = false;
   t->wakeable.pending   = false;
   t->wakeable.delayed   = false;

   for (rt_nexus_t *n = t->source; count > 0; n = n->chain) {
      sched_event(m, n, &(t->wakeable));

      if (!t->wakeable.pending) {
         // Schedule initial update immediately
         deferq_do(&m->delta_procq, async_transfer_signal, t);
         t->wakeable.pending = true;
      }

      count -= n->width;
      assert(count >= 0);
   }
}

int32_t x_test_net_event(sig_shared_t *ss, uint32_t offset, int32_t count)
{
   rt_signal_t *s = container_of(ss, rt_signal_t, shared);
   RT_LOCK(s->lock);

   TRACE("_test_net_event %s offset=%d count=%d",
         istr(tree_ident(s->where)), offset, count);

   int32_t result = 0;
   rt_model_t *m = get_model();
   rt_nexus_t *n = split_nexus(m, s, offset, count);
   for (; count > 0; n = n->chain) {
      if (n->last_event == m->now && n->event_delta == m->iteration) {
         result = 1;
         break;
      }

      count -= n->width;
      assert(count >= 0);
   }

   if (ss->size == s->nexus.size) {
      assert(!(ss->flags & SIG_F_CACHE_EVENT));   // Should have taken fast-path
      ss->flags |= SIG_F_CACHE_EVENT | (result ? SIG_F_EVENT_FLAG : 0);
      s->nexus.flags |= NET_F_CACHE_EVENT;
      APUSH(m->eventsigs, s);
   }

   return result;
}

int32_t x_test_net_active(sig_shared_t *ss, uint32_t offset, int32_t count)
{
   rt_signal_t *s = container_of(ss, rt_signal_t, shared);
   RT_LOCK(s->lock);

   TRACE("_test_net_active %s offset=%d count=%d",
         istr(tree_ident(s->where)), offset, count);

   rt_model_t *m = get_model();
   rt_nexus_t *n = split_nexus(m, s, offset, count);
   for (; count > 0; n = n->chain) {
      if (nexus_active(m, n))
         return 1;

      count -= n->width;
      assert(count >= 0);
   }

   return 0;
}

void x_sched_event(sig_shared_t *ss, uint32_t offset, int32_t count)
{
   rt_signal_t *s = container_of(ss, rt_signal_t, shared);
   RT_LOCK(s->lock);

   TRACE("_sched_event %s+%d count=%d", istr(tree_ident(s->where)),
         offset, count);

   rt_wakeable_t *obj = get_active_wakeable();

   rt_model_t *m = get_model();
   rt_nexus_t *n = split_nexus(m, s, offset, count);
   for (; count > 0; n = n->chain) {
      sched_event(m, n, obj);

      count -= n->width;
      assert(count >= 0);
   }
}

void x_clear_event(sig_shared_t *ss, uint32_t offset, int32_t count)
{
   rt_signal_t *s = container_of(ss, rt_signal_t, shared);
   RT_LOCK(s->lock);

   TRACE("clear event %s+%d count=%d",
         istr(tree_ident(s->where)), offset, count);

   rt_model_t *m = get_model();
   rt_proc_t *proc = get_active_proc();
   rt_nexus_t *n = split_nexus(m, s, offset, count);
   for (; count > 0; n = n->chain) {
      clear_event(m, n, &(proc->wakeable));

      count -= n->width;
      assert(count >= 0);
   }
}

void x_enter_state(int32_t state, bool strong)
{
   rt_wakeable_t *obj = get_active_wakeable();
   assert(obj->kind == W_PROPERTY);

   rt_prop_t *prop = container_of(obj, rt_prop_t, wakeable);
   mask_set(&prop->newstate, state);
   prop->strong |= strong;
}

void x_alias_signal(sig_shared_t *ss, tree_t where)
{
   rt_signal_t *s = container_of(ss, rt_signal_t, shared);
   RT_LOCK(s->lock);

   TRACE("alias signal %s to %s", istr(tree_ident(s->where)),
         istr(tree_ident(where)));

   rt_alias_t *a = xcalloc(sizeof(rt_alias_t));
   a->where  = where;
   a->signal = s;

   model_thread_t *thread = model_thread(get_model());
   APUSH(thread->active_scope->aliases, a);
}

int64_t x_last_event(sig_shared_t *ss, uint32_t offset, int32_t count)
{
   rt_signal_t *s = container_of(ss, rt_signal_t, shared);
   RT_LOCK(s->lock);

   TRACE("_last_event %s offset=%d count=%d",
         istr(tree_ident(s->where)), offset, count);

   int64_t last = TIME_HIGH;

   rt_model_t *m = get_model();
   rt_nexus_t *n = split_nexus(m, s, offset, count);
   for (; count > 0; n = n->chain) {
      if (n->last_event <= m->now)
         last = MIN(last, m->now - n->last_event);

      count -= n->width;
      assert(count >= 0);
   }

   return last;
}

int64_t x_last_active(sig_shared_t *ss, uint32_t offset, int32_t count)
{
   rt_signal_t *s = container_of(ss, rt_signal_t, shared);
   RT_LOCK(s->lock);

   TRACE("_last_active %s offset=%d count=%d",
         istr(tree_ident(s->where)), offset, count);

   int64_t last = TIME_HIGH;

   rt_model_t *m = get_model();
   rt_nexus_t *n = split_nexus(m, s, offset, count);
   for (; count > 0; n = n->chain) {
      last = MIN(last, nexus_last_active(m, n));

      count -= n->width;
      assert(count >= 0);
   }

   return last;
}

void x_map_signal(sig_shared_t *src_ss, uint32_t src_offset,
                  sig_shared_t *dst_ss, uint32_t dst_offset, uint32_t count)
{
   rt_signal_t *src_s = container_of(src_ss, rt_signal_t, shared);
   RT_LOCK(src_s->lock);

   rt_signal_t *dst_s = container_of(dst_ss, rt_signal_t, shared);
   RT_LOCK(dst_s->lock);

   TRACE("map signal %s+%d to %s+%d count %d",
         istr(tree_ident(src_s->where)), src_offset,
         istr(tree_ident(dst_s->where)), dst_offset, count);

   assert(src_s != dst_s);

   rt_model_t *m = get_model();

   rt_nexus_t *src_n = split_nexus(m, src_s, src_offset, count);
   rt_nexus_t *dst_n = split_nexus(m, dst_s, dst_offset, count);

   while (count > 0) {
      if (src_n->width > dst_n->width)
         clone_nexus(m, src_n, dst_n->width);
      else if (src_n->width < dst_n->width)
         clone_nexus(m, dst_n, src_n->width);

      assert(src_n->width == dst_n->width);
      assert(src_n->size == dst_n->size);

      // Effective value updates must propagate through ports
      src_n->flags |= (dst_n->flags & NET_F_EFFECTIVE);
      dst_n->flags |= (src_n->flags & NET_F_EFFECTIVE);

      rt_source_t *port = add_source(m, dst_n, SOURCE_PORT);
      port->u.port.input = src_n;

      port->chain_output = src_n->outputs;
      src_n->outputs = port;

      count -= src_n->width;
      assert(count >= 0);

      src_n = src_n->chain;
      dst_n = dst_n->chain;
   }
}

void x_map_const(sig_shared_t *ss, uint32_t offset,
                 const uint8_t *values, uint32_t count)
{
   rt_signal_t *s = container_of(ss, rt_signal_t, shared);
   RT_LOCK(s->lock);

   TRACE("map const %s to %s+%d count %d", fmt_values(values, count),
         istr(tree_ident(s->where)), offset, count);

   rt_model_t *m = get_model();
   rt_nexus_t *n = split_nexus(m, s, offset, count);
   for (; count > 0; n = n->chain) {
      const size_t valuesz = n->size * n->width;
      memcpy(nexus_effective(n), values, valuesz);
      memcpy(nexus_initial(n), values, valuesz);

      n->flags |= NET_F_HAS_INITIAL;
      values += valuesz;

      count -= n->width;
      assert(count >= 0);
   }
}

void x_map_implicit(sig_shared_t *src_ss, uint32_t src_offset,
                    sig_shared_t *dst_ss, uint32_t dst_offset,
                    uint32_t count)
{
   rt_signal_t *src_s = container_of(src_ss, rt_signal_t, shared);
   RT_LOCK(src_s->lock);

   rt_signal_t *dst_s = container_of(dst_ss, rt_signal_t, shared);
   RT_LOCK(dst_s->lock);

   TRACE("map implicit signal %s+%d to %s+%d count %d",
         istr(tree_ident(src_s->where)), src_offset,
         istr(tree_ident(dst_s->where)), dst_offset, count);

   assert(src_s != dst_s);
   assert(dst_offset == 0);

   rt_model_t *m = get_model();
   rt_nexus_t *n = split_nexus(m, src_s, src_offset, count);
   for (; count > 0; n = n->chain) {
      count -= n->width;
      assert(count >= 0);

      rt_source_t *src = add_source(m, &(dst_s->nexus), SOURCE_IMPLICIT);
      src->u.port.input = n;

      src->chain_output = n->outputs;
      n->outputs = src;

      n->flags |= NET_F_EFFECTIVE;   // Update outputs when active
      n->flags &= ~NET_F_FAST_DRIVER;
   }
}

void x_push_scope(tree_t where, int32_t size, rt_scope_kind_t kind)
{
   TRACE("push scope %s size=%d kind=%d", istr(tree_ident(where)), size, kind);

   rt_model_t *m = get_model();
   model_thread_t *thread = model_thread(m);

   ident_t name;
   if (thread->active_scope && thread->active_scope->kind == SCOPE_ARRAY)
      name = ident_sprintf("%s(%d)", istr(thread->active_scope->name),
                           thread->active_scope->children.count);
   else if (thread->active_scope && thread->active_scope->kind == SCOPE_RECORD)
      name = ident_prefix(thread->active_scope->name, tree_ident(where), '.');
   else
      name = tree_ident(where);

   rt_scope_t *s = xcalloc(sizeof(rt_scope_t));
   s->where    = where;
   s->name     = name;
   s->kind     = kind;
   s->parent   = thread->active_scope;
   s->size     = size;
   s->privdata = MPTR_INVALID;

   if (kind != SCOPE_PACKAGE) {
      type_t type = tree_type(where);
      assert(type_is_composite(type));
      if (type_kind(type) == T_SUBTYPE && type_has_resolution(type))
         s->flags |= SCOPE_F_RESOLVED;
   }

   thread->active_scope = s;
}

void x_pop_scope(void)
{
   rt_model_t *m = get_model();
   model_thread_t *thread = model_thread(m);

   rt_scope_t *pop = thread->active_scope, *old = pop->parent;

   TRACE("pop scope %s", istr(tree_ident(pop->where)));

   int offset = INT_MAX;
   for (int i = 0; i < pop->children.count; i++)
      offset = MIN(offset, pop->children.items[i]->offset);
   for (int i = 0; i < pop->signals.count; i++)
      offset = MIN(offset, pop->signals.items[i]->offset);
   pop->offset = offset;

   thread->active_scope = old;

   if (pop->kind == SCOPE_PACKAGE)
      pop->parent = m->root;   // Always attach packages to root scope

   APUSH(pop->parent->children, pop);
}

bool x_driving(sig_shared_t *ss, uint32_t offset, int32_t count)
{
   rt_signal_t *s = container_of(ss, rt_signal_t, shared);
   RT_LOCK(s->lock);

   TRACE("_driving %s offset=%d count=%d",
         istr(tree_ident(s->where)), offset, count);

   int ntotal = 0, ndriving = 0;
   bool found = false;
   rt_model_t *m = get_model();
   rt_proc_t *proc = get_active_proc();
   rt_nexus_t *n = split_nexus(m, s, offset, count);
   for (; count > 0; n = n->chain) {
      if (n->n_sources > 0) {
         rt_source_t *src = find_driver(n, proc);
         if (src != NULL) {
            if (!src->disconnected) ndriving++;
            found = true;
         }
      }

      ntotal++;
      count -= n->width;
      assert(count >= 0);
   }

   if (!found)
      jit_msg(NULL, DIAG_FATAL, "process %s does not contain a driver for %s",
              istr(proc->name), istr(tree_ident(s->where)));

   return ntotal == ndriving;
}

void *x_driving_value(sig_shared_t *ss, uint32_t offset, int32_t count)
{
   rt_signal_t *s = container_of(ss, rt_signal_t, shared);
   RT_LOCK(s->lock);

   TRACE("driving value %s offset=%d count=%d", istr(tree_ident(s->where)),
         offset, count);

   rt_model_t *m = get_model();
   rt_nexus_t *n = split_nexus(m, s, offset, count);

   rt_proc_t *proc = get_active_proc();
   if (proc == NULL) {   // Called in output conversion
      if (n->flags & NET_F_EFFECTIVE)
         return nexus_driving(n);
      else
         return nexus_effective(n);
   }

   void *result = tlab_alloc(model_thread(m)->tlab, s->shared.size);

   uint8_t *p = result;
   for (; count > 0; n = n->chain) {
      rt_source_t *src = find_driver(n, proc);
      if (src == NULL)
         jit_msg(NULL, DIAG_FATAL, "process %s does not contain a driver "
                 "for %s", istr(proc->name), istr(tree_ident(s->where)));

      const uint8_t *driving;
      if (n->flags & NET_F_FAST_DRIVER)
         driving = nexus_effective(n);
      else
         driving = value_ptr(n, &(src->u.driver.waveforms.value));

      memcpy(p, driving, n->width * n->size);
      p += n->width * n->size;

      count -= n->width;
      assert(count >= 0);
   }

   return result;
}

sig_shared_t *x_implicit_signal(uint32_t count, uint32_t size, tree_t where,
                                implicit_kind_t kind, ffi_closure_t *closure,
                                int64_t delay)
{
   TRACE("implicit signal %s count=%d size=%d kind=%d",
         istr(tree_ident(where)), count, size, kind);

   rt_model_t *m = get_model();

   const size_t datasz = MAX(2 * count * size, 8);
   rt_implicit_t *imp = static_alloc(m, sizeof(rt_implicit_t) + datasz);
   setup_signal(m, &(imp->signal), where, count, size, SIG_F_IMPLICIT, 0);

   imp->closure = *closure;
   imp->delay = delay;
   imp->wakeable.kind = W_IMPLICIT;

   deferq_do(&m->implicitq, async_update_implicit_signal, imp);
   set_pending(&(imp->wakeable));

   if (kind == IMPLICIT_STABLE || kind == IMPLICIT_QUIET) {
      add_source(m, &(imp->signal.nexus), SOURCE_DRIVER);
      imp->signal.shared.data[0] = 1;    // X'STABLE initally true
   }

   return &(imp->signal.shared);
}

void x_disconnect(sig_shared_t *ss, uint32_t offset, int32_t count,
                  int64_t after, int64_t reject)
{
   rt_signal_t *s = container_of(ss, rt_signal_t, shared);

   TRACE("_disconnect %s+%d len=%d after=%s reject=%s",
         istr(tree_ident(s->where)), offset, count, trace_time(after),
         trace_time(reject));

   rt_proc_t *proc = get_active_proc();

   check_postponed(after, proc);
   check_reject_limit(s, after, reject);

   rt_model_t *m = get_model();
   rt_nexus_t *n = split_nexus(m, s, offset, count);
   for (; count > 0; n = n->chain) {
      count -= n->width;
      assert(count >= 0);

      sched_disconnect(m, n, after, reject, proc);
   }
}

void x_force(sig_shared_t *ss, uint32_t offset, int32_t count, void *values)
{
   rt_signal_t *s = container_of(ss, rt_signal_t, shared);

   TRACE("force signal %s+%d value=%s count=%d", istr(tree_ident(s->where)),
         offset, fmt_values(values, count), count);

   rt_proc_t *proc = get_active_proc();
   rt_model_t *m = get_model();

   check_postponed(0, proc);

   force_signal(m, s, values, offset, count);
}

void x_release(sig_shared_t *ss, uint32_t offset, int32_t count)
{
   rt_signal_t *s = container_of(ss, rt_signal_t, shared);

   TRACE("release signal %s+%d count=%d", istr(tree_ident(s->where)),
         offset, count);

   rt_proc_t *proc = get_active_proc();
   rt_model_t *m = get_model();

   check_postponed(0, proc);

   release_signal(m, s, offset, count);
}

void x_deposit_signal(sig_shared_t *ss, uint32_t offset, int32_t count,
                      void *values)
{
   rt_signal_t *s = container_of(ss, rt_signal_t, shared);

   TRACE("deposit signal %s+%d value=%s count=%d", istr(tree_ident(s->where)),
         offset, fmt_values(values, count * s->nexus.size), count);

   rt_proc_t *proc = get_active_proc();
   rt_model_t *m = get_model();

   check_postponed(0, proc);

   deposit_signal(m, s, values, offset, count);
}

void x_put_conversion(rt_conv_func_t *cf, sig_shared_t *ss, uint32_t offset,
                      int32_t count, void *values)
{
   rt_signal_t *s = container_of(ss, rt_signal_t, shared);

   TRACE("put conversion %s+%d value=%s count=%d", istr(tree_ident(s->where)),
         offset, fmt_values(values, count * s->nexus.size), count);

   rt_model_t *m = get_model();
   rt_nexus_t *n = split_nexus(m, s, offset, count);
   for (; count > 0; n = n->chain) {
      count -= n->width;
      assert(count >= 0);

      rt_source_t *s = &(n->sources);
      for (; s; s = s->chain_input) {
         if (s->tag == SOURCE_PORT && s->u.port.conv_func == cf)
            break;
      }

      rt_value_t *result;
      if (s != NULL)
         result = &(s->u.port.conv_result);
      else {
         assert(n->flags & NET_F_EFFECTIVE);
         result = find_conversion_input(cf, n);
         assert(result != NULL);
      }

      copy_value_ptr(n, result, values);

      values += n->width * n->size;
   }
}

void x_resolve_signal(sig_shared_t *ss, jit_handle_t handle, void *context,
                      int32_t nlits, int32_t flags)
{
   rt_signal_t *s = container_of(ss, rt_signal_t, shared);

   TRACE("resolve signal %s", istr(tree_ident(s->where)));

   ffi_closure_t closure = {
      .handle = handle,
      .context = context
   };

   rt_model_t *m = get_model();
   s->resolution = memo_resolution_fn(m, s, closure, nlits, flags);

   // Copy R_IDENT into the nexus flags to avoid rt_resolve_nexus_fast
   // having to dereference the resolution pointer in the common case
   if (s->resolution->flags & R_IDENT) {
      s->shared.flags |= NET_F_R_IDENT;

      rt_nexus_t *n = &(s->nexus);
      for (int i = 0; i < s->n_nexus; i++, n = n->chain)
         n->flags |= NET_F_R_IDENT;
   }
}

void x_process_init(jit_handle_t handle, tree_t where)
{
   rt_model_t *m = get_model();
   ident_t name = jit_get_name(m->jit, handle);

   TRACE("init process %s", istr(name));

   rt_scope_t *s = model_thread(m)->active_scope;
   assert(s != NULL);
   assert(s->kind == SCOPE_INSTANCE);

   rt_proc_t *p = xcalloc(sizeof(rt_proc_t));
   p->where     = where;
   p->name      = name;
   p->handle    = handle;
   p->scope     = s;
   p->privdata  = mptr_new(m->mspace, "process privdata");

   p->wakeable.kind      = W_PROC;
   p->wakeable.pending   = false;
   p->wakeable.postponed = false;
   p->wakeable.delayed   = false;

   APUSH(s->procs, p);
}

void *x_function_trigger(jit_handle_t handle, unsigned nargs,
                         const jit_scalar_t *args)
{
   rt_model_t *m = get_model();

   uint64_t hash = mix_bits_32(handle);
   for (int i = 0; i < nargs; i++)
      hash ^= mix_bits_64(args[i].integer);

   TRACE("function trigger %s nargs=%u hash=%"PRIx64,
         istr(jit_get_name(m->jit, handle)), nargs, hash);

   return new_trigger(m, FUNC_TRIGGER, hash, handle, nargs, args);
}

void *x_or_trigger(void *left, void *right)
{
   rt_model_t *m = get_model();

   uint64_t hash = mix_bits_64(left) ^ mix_bits_64(right);

   TRACE("or trigger %p %p hash=%"PRIx64, left, right, hash);

   const jit_scalar_t args[] = {
      { .pointer = left < right ? left : right },
      { .pointer = left < right ? right : left }
   };

   return new_trigger(m, OR_TRIGGER, hash, JIT_HANDLE_INVALID, 2, args);
}

void *x_cmp_trigger(sig_shared_t *ss, uint32_t offset, int64_t right)
{
   rt_model_t *m = get_model();
   rt_signal_t *s = container_of(ss, rt_signal_t, shared);

   uint64_t hash = mix_bits_64(s) ^ mix_bits_32(offset) ^ mix_bits_64(right);

   TRACE("cmp trigger %s+%d right=%"PRIi64" hash=%"PRIx64,
         istr(tree_ident(s->where)), offset, right, hash);

   const jit_scalar_t args[] = {
      { .pointer = s },
      { .integer = offset },
      { .integer = right }
   };
   return new_trigger(m, CMP_TRIGGER, hash, JIT_HANDLE_INVALID, 3, args);
}

void x_add_trigger(void *ptr)
{
   TRACE("add trigger %p", ptr);

   rt_wakeable_t *obj = get_active_wakeable();
   assert(obj->trigger == NULL);

   obj->trigger = ptr;
}

void *x_port_conversion(const ffi_closure_t *driving,
                        const ffi_closure_t *effective)
{
   rt_model_t *m = get_model();

   TRACE("port conversion %s context %p",
         istr(jit_get_name(m->jit, driving->handle)), driving->context);

   if (effective->handle != JIT_HANDLE_INVALID)
      TRACE("effective value conversion %s context %p",
            istr(jit_get_name(m->jit, effective->handle)), effective->context);

   const size_t tail_bytes = ALIGN_UP(sizeof(rt_conv_func_t), MEMBLOCK_ALIGN)
      - sizeof(rt_conv_func_t);
   const int tail_max_inputs = tail_bytes / sizeof(conv_input_t);
   assert(tail_max_inputs > 0);

   const size_t total_bytes =
      sizeof(rt_conv_func_t) + tail_max_inputs * sizeof(conv_input_t);

   rt_conv_func_t *cf = static_alloc(m, total_bytes);
   cf->driving   = *driving;
   cf->effective = *effective;
   cf->ninputs   = 0;
   cf->maxinputs = tail_max_inputs;
   cf->outputs   = NULL;
   cf->inputs    = cf->tail;
   cf->when      = TIME_HIGH;
   cf->iteration = UINT_MAX;

   return cf;
}

void x_convert_in(void *ptr, sig_shared_t *ss, uint32_t offset, int32_t count)
{
   rt_signal_t *s = container_of(ss, rt_signal_t, shared);

   TRACE("convert in %p %s+%d count=%d", ptr, istr(tree_ident(s->where)),
         offset, count);

   rt_conv_func_t *cf = ptr;
   rt_model_t *m = get_model();

   rt_nexus_t *n = split_nexus(m, s, offset, count);
   for (; count > 0; n = n->chain) {
      count -= n->width;
      assert(count >= 0);

      add_conversion_input(m, cf, n);

      rt_source_t **p = &(n->outputs);
      for (; *p != NULL && *p != cf->outputs; p = &((*p)->chain_output));
      *p = cf->outputs;
   }
}

void x_convert_out(void *ptr, sig_shared_t *ss, uint32_t offset, int32_t count)
{
   rt_signal_t *s = container_of(ss, rt_signal_t, shared);

   TRACE("convert out %p %s+%d count=%d", ptr, istr(tree_ident(s->where)),
         offset, count);

   rt_conv_func_t *cf = ptr;
   rt_model_t *m = get_model();

   assert(cf->ninputs == 0);    // Add outputs first

   rt_nexus_t *n = split_nexus(m, s, offset, count);
   for (; count > 0; n = n->chain) {
      count -= n->width;
      assert(count >= 0);

      rt_source_t *src = add_source(m, n, SOURCE_PORT);
      src->u.port.conv_func   = cf;
      src->u.port.conv_result = alloc_value(m, n);

      src->chain_output = cf->outputs;
      cf->outputs = src;
   }
}

void x_instance_name(attr_kind_t kind, text_buf_t *tb)
{
   rt_model_t *m = get_model();
   rt_scope_t *s = get_active_scope(m);

   switch (kind) {
   case ATTR_INSTANCE_NAME:
      get_instance_name(s, tb);
      break;
   case ATTR_PATH_NAME:
      get_path_name(s, tb);
      break;
   default:
      should_not_reach_here();
   }
}

nickg / nvc / 16099227452

Source File Press 'n' to go to next uncovered line, 'b' for previous

Source File
Press 'n' to go to next uncovered line, 'b' for previous