void *ph_malloc(size_t nbytes)
{ unsigned char *p;
#ifdef DEBUG
util_progress();
#endif
nbytes += sizeof(unsigned) * 2;
nbytes &= ~(sizeof(unsigned) - 1);
if (nbytes >= heap->nleft)
ph_newheap(nbytes);
p = heap->p;
heap->p += nbytes;
heap->nleft -= nbytes;
*(unsigned *)p = nbytes - sizeof(unsigned);
p += sizeof(unsigned);
return p;
}
void optfunc()
{
#if !HTOD
block *b;
int iter; // iteration count
clock_t starttime;
cmes ("optfunc()\n");
dbg_optprint("optfunc\n");
#ifdef DEBUG
if (debugb)
{
dbg_printf("................Before optimization.........\n");
WRfunc();
}
#endif
iter = 0;
if (localgot)
{ // Initialize with:
// localgot = OPgot;
elem *e = el_long(TYnptr, 0);
e->Eoper = OPgot;
e = el_bin(OPeq, TYnptr, el_var(localgot), e);
startblock->Belem = el_combine(e, startblock->Belem);
}
// Each pass through the loop can reduce only one level of comma expression.
// The infinite loop check needs to take this into account.
int iterationLimit = 0;
for (b = startblock; b; b = b->Bnext)
{
if (!b->Belem)
continue;
int d = el_countCommas(b->Belem);
if (d > iterationLimit)
iterationLimit = d;
}
// Some functions can take enormous amounts of time to optimize.
// We try to put a lid on it.
starttime = clock();
do
{
//printf("iter = %d\n", iter);
#if TX86
if (++iter > 200)
{ assert(iter < iterationLimit); // infinite loop check
break;
}
#else
L1:
#endif
#if MARS
util_progress();
#else
file_progress();
#endif
//printf("optelem\n");
/* canonicalize the trees */
for (b = startblock; b; b = b->Bnext)
if (b->Belem)
{
#if DEBUG
if(debuge)
{
dbg_printf("before\n");
elem_print(b->Belem);
//el_check(b->Belem);
}
#endif
b->Belem = doptelem(b->Belem,bc_goal[b->BC] | GOALagain);
#if DEBUG
if(0 && debugf)
{
dbg_printf("after\n");
elem_print(b->Belem);
}
#endif
}
//printf("blockopt\n");
#if TX86
if (mfoptim & MFdc)
blockopt(0); // do block optimization
out_regcand(&globsym); // recompute register candidates
changes = 0; /* no changes yet */
if (mfoptim & MFcnp)
constprop(); /* make relationals unsigned */
if (mfoptim & (MFli | MFliv))
#else
if (config.optimized && (mfoptim & MFdc))
blockopt(0); // do block optimization
dbg_optprint("blockopt\n");
out_regcand(); // recompute register candidates
changes = 0; /* no changes yet */
dbg_optprint("constprop\n");
if (config.optimized && (mfoptim & MFcnp))
constprop(); /* make relationals unsigned */
dbg_optprint("loopopt\n");
if (config.optimized && (mfoptim & (MFli | MFliv)))
#endif
loopopt(); /* remove loop invariants and */
/* induction vars */
/* do loop rotation */
else
for (b = startblock; b; b = b->Bnext)
b->Bweight = 1;
dbg_optprint("boolopt\n");
#if TX86
if (mfoptim & MFcnp)
boolopt(); // optimize boolean values
if (changes && mfoptim & MFloop && (clock() - starttime) < 30 * CLOCKS_PER_SEC)
continue;
if (mfoptim & MFcnp)
constprop(); /* constant propagation */
if (mfoptim & MFcp)
copyprop(); /* do copy propagation */
/* Floating point constants and string literals need to be
* replaced with loads from variables in read-only data.
* This can result in localgot getting needed.
*/
symbol *localgotsave = localgot;
for (b = startblock; b; b = b->Bnext)
{
if (b->Belem)
{
b->Belem = doptelem(b->Belem,bc_goal[b->BC] | GOALstruct);
if (b->Belem)
b->Belem = el_convert(b->Belem);
}
}
if (localgot != localgotsave)
{ /* Looks like we did need localgot, initialize with:
* localgot = OPgot;
*/
elem *e = el_long(TYnptr, 0);
e->Eoper = OPgot;
e = el_bin(OPeq, TYnptr, el_var(localgot), e);
startblock->Belem = el_combine(e, startblock->Belem);
}
/* localize() is after localgot, otherwise we wind up with
* more than one OPgot in a function, which mucks up OSX
* code generation which assumes at most one (localgotoffset).
*/
if (mfoptim & MFlocal)
localize(); // improve expression locality
if (mfoptim & MFda)
rmdeadass(); /* remove dead assignments */
cmes2 ("changes = %d\n", changes);
if (!(changes && mfoptim & MFloop && (clock() - starttime) < 30 * CLOCKS_PER_SEC))
break;
} while (1);
cmes2("%d iterations\n",iter);
if (mfoptim & MFdc)
blockopt(1); // do block optimization
#else
if (config.optimized && (mfoptim & MFcnp))
boolopt(); // optimize boolean values
util_progress();
if (changes)
goto L1;
dbg_optprint("constprop\n");
if (config.optimized && (mfoptim & MFcnp))
constprop(); /* constant propagation */
dbg_optprint("copyprop\n");
if (config.optimized && (mfoptim & MFcp))
copyprop(); /* do copy propagation */
dbg_optprint("localize\n");
if (config.optimized && (mfoptim & MFlocal))
localize(); // improve expression locality
dbg_optprint("rmdeadass\n");
if (config.optimized && (mfoptim & MFda))
rmdeadass(); /* remove dead assignments */
cmes2 ("changes = %d\n", changes);
iter++;
assert (iter < 80); /* infinite loop check */
} while (changes && (config.optimized) && (mfoptim & MFloop));
