| 1 |
48 |
robfinch |
// ============================================================================
|
| 2 |
|
|
// __
|
| 3 |
|
|
// \\__/ o\ (C) 2017-2018 Robert Finch, Waterloo
|
| 4 |
|
|
// \ __ / All rights reserved.
|
| 5 |
|
|
// \/_// robfinch<remove>@finitron.ca
|
| 6 |
|
|
// ||
|
| 7 |
|
|
//
|
| 8 |
|
|
// CC64 - 'C' derived language compiler
|
| 9 |
|
|
// - 64 bit CPU
|
| 10 |
|
|
//
|
| 11 |
|
|
// This source file is free software: you can redistribute it and/or modify
|
| 12 |
|
|
// it under the terms of the GNU Lesser General Public License as published
|
| 13 |
|
|
// by the Free Software Foundation, either version 3 of the License, or
|
| 14 |
|
|
// (at your option) any later version.
|
| 15 |
|
|
//
|
| 16 |
|
|
// This source file is distributed in the hope that it will be useful,
|
| 17 |
|
|
// but WITHOUT ANY WARRANTY; without even the implied warranty of
|
| 18 |
|
|
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
| 19 |
|
|
// GNU General Public License for more details.
|
| 20 |
|
|
//
|
| 21 |
|
|
// You should have received a copy of the GNU General Public License
|
| 22 |
|
|
// along with this program. If not, see <http://www.gnu.org/licenses/>.
|
| 23 |
|
|
//
|
| 24 |
|
|
// ============================================================================
|
| 25 |
|
|
//
|
| 26 |
|
|
#include "stdafx.h"
|
| 27 |
|
|
|
| 28 |
|
|
void CSETable::Assign(CSETable *t)
|
| 29 |
|
|
{
|
| 30 |
|
|
memcpy(this, t, sizeof(CSETable));
|
| 31 |
|
|
}
|
| 32 |
|
|
|
| 33 |
|
|
static int CSECmp(const void *a, const void *b)
|
| 34 |
|
|
{
|
| 35 |
|
|
CSE *csp1, *csp2;
|
| 36 |
|
|
int aa, bb;
|
| 37 |
|
|
|
| 38 |
|
|
csp1 = (CSE *)a;
|
| 39 |
|
|
csp2 = (CSE *)b;
|
| 40 |
|
|
aa = csp1->OptimizationDesireability();
|
| 41 |
|
|
bb = csp2->OptimizationDesireability();
|
| 42 |
|
|
if (aa < bb)
|
| 43 |
|
|
return (1);
|
| 44 |
|
|
else if (aa == bb)
|
| 45 |
|
|
return (0);
|
| 46 |
|
|
else
|
| 47 |
|
|
return (-1);
|
| 48 |
|
|
}
|
| 49 |
|
|
|
| 50 |
|
|
|
| 51 |
|
|
void CSETable::Sort(int(*cmp)(const void *a, const void *b))
|
| 52 |
|
|
{
|
| 53 |
|
|
qsort(table, (size_t)csendx, sizeof(CSE), cmp);
|
| 54 |
|
|
}
|
| 55 |
|
|
|
| 56 |
|
|
// InsertNodeIntoCSEList will enter a reference to an expression node into the
|
| 57 |
|
|
// common expression table. duse is a flag indicating whether or not
|
| 58 |
|
|
// this reference will be dereferenced.
|
| 59 |
|
|
|
| 60 |
|
|
CSE *CSETable::InsertNode(ENODE *node, int duse)
|
| 61 |
|
|
{
|
| 62 |
|
|
CSE *csp;
|
| 63 |
|
|
|
| 64 |
|
|
if ((csp = Search(node)) == nullptr) { /* add to tree */
|
| 65 |
|
|
if (csendx > 499)
|
| 66 |
|
|
throw new C64PException(ERR_CSETABLE, 0x01);
|
| 67 |
|
|
csp = &table[csendx];
|
| 68 |
|
|
csendx++;
|
| 69 |
|
|
if (loop_active > 1) {
|
| 70 |
|
|
csp->uses = (loop_active - 1) * 5;
|
| 71 |
|
|
csp->duses = (duse != 0) * ((loop_active - 1) * 5);
|
| 72 |
|
|
}
|
| 73 |
|
|
else {
|
| 74 |
|
|
csp->uses = 1;
|
| 75 |
|
|
csp->duses = (duse != 0);
|
| 76 |
|
|
}
|
| 77 |
|
|
csp->exp = node->Clone();
|
| 78 |
|
|
csp->voidf = 0;
|
| 79 |
|
|
csp->reg = 0;
|
| 80 |
|
|
csp->isfp = csp->exp->IsFloatType();
|
| 81 |
|
|
return (csp);
|
| 82 |
|
|
}
|
| 83 |
|
|
if (loop_active < 2) {
|
| 84 |
|
|
(csp->uses)++;
|
| 85 |
|
|
if (duse)
|
| 86 |
|
|
(csp->duses)++;
|
| 87 |
|
|
}
|
| 88 |
|
|
else {
|
| 89 |
|
|
(csp->uses) += ((loop_active - 1) * 5);
|
| 90 |
|
|
if (duse)
|
| 91 |
|
|
(csp->duses) += ((loop_active - 1) * 5);
|
| 92 |
|
|
}
|
| 93 |
|
|
return (csp);
|
| 94 |
|
|
}
|
| 95 |
|
|
|
| 96 |
|
|
//
|
| 97 |
|
|
// SearchCSEList will search the common expression table for an entry
|
| 98 |
|
|
// that matches the node passed and return a pointer to it.
|
| 99 |
|
|
//
|
| 100 |
|
|
CSE *CSETable::Search(ENODE *node)
|
| 101 |
|
|
{
|
| 102 |
|
|
int cnt;
|
| 103 |
|
|
|
| 104 |
|
|
for (cnt = 0; cnt < csendx; cnt++) {
|
| 105 |
|
|
if (ENODE::IsEqual(node, table[cnt].exp))
|
| 106 |
|
|
return (&table[cnt]);
|
| 107 |
|
|
}
|
| 108 |
|
|
return ((CSE *)nullptr);
|
| 109 |
|
|
}
|
| 110 |
|
|
|
| 111 |
|
|
// voidauto2 searches the entire CSE table for auto dereferenced node which
|
| 112 |
|
|
// point to the passed node. There might be more than one LValue that matches.
|
| 113 |
|
|
// voidauto will void an auto dereference node which points to
|
| 114 |
|
|
// the same auto constant as node.
|
| 115 |
|
|
//
|
| 116 |
|
|
int CSETable::voidauto2(ENODE *node)
|
| 117 |
|
|
{
|
| 118 |
|
|
int uses;
|
| 119 |
|
|
bool voided;
|
| 120 |
|
|
int cnt;
|
| 121 |
|
|
|
| 122 |
|
|
uses = 0;
|
| 123 |
|
|
voided = false;
|
| 124 |
|
|
for (cnt = 0; cnt < csendx; cnt++) {
|
| 125 |
|
|
if (IsLValue(table[cnt].exp) && ENODE::IsEqual(node, table[cnt].exp->p[0])) {
|
| 126 |
|
|
table[cnt].voidf = 1;
|
| 127 |
|
|
voided = true;
|
| 128 |
|
|
uses += table[cnt].uses;
|
| 129 |
|
|
}
|
| 130 |
|
|
}
|
| 131 |
|
|
return (voided ? uses : -1);
|
| 132 |
|
|
}
|
| 133 |
|
|
|
| 134 |
|
|
// Make multiple passes over the CSE table in order to use
|
| 135 |
|
|
// up all temporary registers. Allocates on the progressively
|
| 136 |
|
|
// less desirable.
|
| 137 |
|
|
|
| 138 |
|
|
int CSETable::AllocateGPRegisters()
|
| 139 |
|
|
{
|
| 140 |
|
|
CSE *csp;
|
| 141 |
|
|
bool alloc;
|
| 142 |
|
|
int pass;
|
| 143 |
|
|
int reg;
|
| 144 |
|
|
|
| 145 |
|
|
reg = regFirstRegvar;
|
| 146 |
|
|
for (pass = 0; pass < 4; pass++) {
|
| 147 |
|
|
for (csp = First(); csp; csp = Next()) {
|
| 148 |
|
|
if (csp->OptimizationDesireability() != 0) {
|
| 149 |
|
|
if (!csp->voidf && csp->reg == -1) {
|
| 150 |
|
|
if (csp->exp->etype != bt_vector && !csp->isfp) {
|
| 151 |
|
|
switch (pass)
|
| 152 |
|
|
{
|
| 153 |
|
|
case 0:
|
| 154 |
|
|
case 1:
|
| 155 |
|
|
case 2: alloc = (csp->OptimizationDesireability() >= 4) && reg < regLastRegvar; break;
|
| 156 |
|
|
case 3: alloc = (csp->OptimizationDesireability() >= 4) && reg < regLastRegvar; break;
|
| 157 |
|
|
}
|
| 158 |
|
|
if (alloc)
|
| 159 |
|
|
csp->reg = reg++;
|
| 160 |
|
|
else
|
| 161 |
|
|
csp->reg = -1;
|
| 162 |
|
|
}
|
| 163 |
|
|
}
|
| 164 |
|
|
}
|
| 165 |
|
|
}
|
| 166 |
|
|
}
|
| 167 |
|
|
return (reg);
|
| 168 |
|
|
}
|
| 169 |
|
|
|
| 170 |
|
|
int CSETable::AllocateFPRegisters()
|
| 171 |
|
|
{
|
| 172 |
|
|
CSE *csp;
|
| 173 |
|
|
bool alloc;
|
| 174 |
|
|
int pass;
|
| 175 |
|
|
int reg;
|
| 176 |
|
|
|
| 177 |
|
|
reg = regFirstRegvar;
|
| 178 |
|
|
for (pass = 0; pass < 4; pass++) {
|
| 179 |
|
|
for (csp = First(); csp; csp = Next()) {
|
| 180 |
|
|
if (csp->OptimizationDesireability() != 0) {
|
| 181 |
|
|
if (!csp->voidf && csp->reg == -1) {
|
| 182 |
|
|
if (csp->isfp) {
|
| 183 |
|
|
switch (pass)
|
| 184 |
|
|
{
|
| 185 |
|
|
case 0:
|
| 186 |
|
|
case 1:
|
| 187 |
|
|
case 2: alloc = (csp->OptimizationDesireability() >= 4) && reg < regLastRegvar; break;
|
| 188 |
|
|
case 3: alloc = (csp->OptimizationDesireability() >= 4) && reg < regLastRegvar; break;
|
| 189 |
|
|
// if(( csp->duses > csp->uses / (8 << nn)) && reg < regLastRegvar ) // <- address register assignments
|
| 190 |
|
|
}
|
| 191 |
|
|
if (alloc)
|
| 192 |
|
|
csp->reg = reg++;
|
| 193 |
|
|
else
|
| 194 |
|
|
csp->reg = -1;
|
| 195 |
|
|
}
|
| 196 |
|
|
}
|
| 197 |
|
|
}
|
| 198 |
|
|
}
|
| 199 |
|
|
}
|
| 200 |
|
|
return (reg);
|
| 201 |
|
|
}
|
| 202 |
|
|
|
| 203 |
|
|
int CSETable::AllocateVectorRegisters()
|
| 204 |
|
|
{
|
| 205 |
|
|
int nn, vreg;
|
| 206 |
|
|
CSE *csp;
|
| 207 |
|
|
bool alloc;
|
| 208 |
|
|
|
| 209 |
|
|
vreg = regFirstRegvar;
|
| 210 |
|
|
for (nn = 0; nn < 4; nn++) {
|
| 211 |
|
|
for (csp = First(); csp; csp = Next()) {
|
| 212 |
|
|
if (csp->exp) {
|
| 213 |
|
|
if (csp->exp->etype == bt_vector && csp->reg == -1 && vreg < regLastRegvar) {
|
| 214 |
|
|
switch (nn) {
|
| 215 |
|
|
case 0:
|
| 216 |
|
|
case 1:
|
| 217 |
|
|
case 2: alloc = (csp->OptimizationDesireability() >= 4 - nn)
|
| 218 |
|
|
&& (csp->duses > csp->uses / (8 << nn));
|
| 219 |
|
|
break;
|
| 220 |
|
|
case 3: alloc = (!csp->voidf) && (csp->uses > 3);
|
| 221 |
|
|
break;
|
| 222 |
|
|
}
|
| 223 |
|
|
if (alloc)
|
| 224 |
|
|
csp->reg = vreg++;
|
| 225 |
|
|
else
|
| 226 |
|
|
csp->reg = -1;
|
| 227 |
|
|
}
|
| 228 |
|
|
}
|
| 229 |
|
|
}
|
| 230 |
|
|
}
|
| 231 |
|
|
return (vreg);
|
| 232 |
|
|
}
|
| 233 |
|
|
|
| 234 |
|
|
void CSETable::InitializeTempRegs()
|
| 235 |
|
|
{
|
| 236 |
|
|
Operand *ap, *ap2, *ap3;
|
| 237 |
|
|
CSE *csp;
|
| 238 |
|
|
ENODE *exptr;
|
| 239 |
|
|
int size;
|
| 240 |
|
|
|
| 241 |
|
|
for (csp = First(); csp; csp = Next()) {
|
| 242 |
|
|
if (csp->reg != -1)
|
| 243 |
|
|
{ // see if preload needed
|
| 244 |
|
|
exptr = csp->exp;
|
| 245 |
|
|
if (1 || !IsLValue(exptr) || (exptr->p[0]->i > 0))
|
| 246 |
|
|
{
|
| 247 |
|
|
initstack();
|
| 248 |
|
|
{
|
| 249 |
|
|
ap = GenerateExpression(exptr, F_REG | F_IMMED | F_MEM | F_FPREG, sizeOfWord);
|
| 250 |
|
|
ap2 = csp->isfp ? makefpreg(csp->reg) : makereg(csp->reg);
|
| 251 |
|
|
if (csp->isfp)
|
| 252 |
|
|
ap2->type = ap->type;
|
| 253 |
|
|
ap2->isPtr = ap->isPtr;
|
| 254 |
|
|
if (ap->mode == am_imm) {
|
| 255 |
|
|
if (ap2->mode == am_fpreg) {
|
| 256 |
|
|
ap3 = GetTempRegister();
|
| 257 |
|
|
GenLdi(ap3, ap);
|
| 258 |
|
|
GenerateDiadic(op_mov, 0, ap2, ap3);
|
| 259 |
|
|
ReleaseTempReg(ap3);
|
| 260 |
|
|
}
|
| 261 |
|
|
else
|
| 262 |
|
|
GenLdi(ap2, ap);
|
| 263 |
|
|
}
|
| 264 |
|
|
else if (ap->mode == am_reg)
|
| 265 |
|
|
GenerateDiadic(op_mov, 0, ap2, ap);
|
| 266 |
|
|
else {
|
| 267 |
|
|
size = GetNaturalSize(exptr);
|
| 268 |
|
|
ap->isUnsigned = exptr->isUnsigned;
|
| 269 |
|
|
GenLoad(ap2, ap, size, size);
|
| 270 |
|
|
}
|
| 271 |
|
|
}
|
| 272 |
|
|
ReleaseTempReg(ap);
|
| 273 |
|
|
}
|
| 274 |
|
|
}
|
| 275 |
|
|
}
|
| 276 |
|
|
|
| 277 |
|
|
}
|
| 278 |
|
|
|
| 279 |
|
|
void CSETable::GenerateRegMask(CSE *csp, uint64_t *mask, uint64_t *rmask)
|
| 280 |
|
|
{
|
| 281 |
|
|
if (csp->reg != -1)
|
| 282 |
|
|
{
|
| 283 |
|
|
*rmask = *rmask | (1LL << (63 - csp->reg));
|
| 284 |
|
|
*mask = *mask | (1LL << csp->reg);
|
| 285 |
|
|
}
|
| 286 |
|
|
}
|
| 287 |
|
|
|
| 288 |
|
|
// ----------------------------------------------------------------------------
|
| 289 |
|
|
// AllocateRegisterVars will allocate registers for the expressions that have
|
| 290 |
|
|
// a high enough desirability.
|
| 291 |
|
|
// ----------------------------------------------------------------------------
|
| 292 |
|
|
|
| 293 |
|
|
int CSETable::AllocateRegisterVars()
|
| 294 |
|
|
{
|
| 295 |
|
|
CSE *csp;
|
| 296 |
|
|
uint64_t mask, rmask;
|
| 297 |
|
|
uint64_t fpmask, fprmask;
|
| 298 |
|
|
uint64_t vmask, vrmask;
|
| 299 |
|
|
|
| 300 |
|
|
mask = 0;
|
| 301 |
|
|
rmask = 0;
|
| 302 |
|
|
fpmask = 0;
|
| 303 |
|
|
fprmask = 0;
|
| 304 |
|
|
vmask = 0;
|
| 305 |
|
|
vrmask = 0;
|
| 306 |
|
|
|
| 307 |
|
|
// Sort the CSE table according to desirability of allocating
|
| 308 |
|
|
// a register.
|
| 309 |
|
|
if (pass == 1)
|
| 310 |
|
|
Sort(CSECmp);
|
| 311 |
|
|
|
| 312 |
|
|
// Initialize to no allocated registers
|
| 313 |
|
|
for (csp = First(); csp; csp = Next())
|
| 314 |
|
|
csp->reg = -1;
|
| 315 |
|
|
|
| 316 |
|
|
AllocateGPRegisters();
|
| 317 |
|
|
AllocateFPRegisters();
|
| 318 |
|
|
AllocateVectorRegisters();
|
| 319 |
|
|
|
| 320 |
|
|
// Generate bit masks of allocated registers
|
| 321 |
|
|
for (csp = First(); csp; csp = Next()) {
|
| 322 |
|
|
if (csp->exp) {
|
| 323 |
|
|
if (csp->exp->IsFloatType())
|
| 324 |
|
|
GenerateRegMask(csp, &fpmask, &fprmask);
|
| 325 |
|
|
else if (csp->exp->etype == bt_vector)
|
| 326 |
|
|
GenerateRegMask(csp, &vrmask, &vmask);
|
| 327 |
|
|
else
|
| 328 |
|
|
GenerateRegMask(csp, &mask, &rmask);
|
| 329 |
|
|
}
|
| 330 |
|
|
else
|
| 331 |
|
|
GenerateRegMask(csp, &mask, &rmask);
|
| 332 |
|
|
}
|
| 333 |
|
|
|
| 334 |
|
|
Dump();
|
| 335 |
|
|
|
| 336 |
|
|
// Push temporaries on the stack.
|
| 337 |
|
|
SaveRegisterVars(mask, rmask);
|
| 338 |
|
|
SaveFPRegisterVars(fpmask, fprmask);
|
| 339 |
|
|
|
| 340 |
|
|
save_mask = mask;
|
| 341 |
|
|
fpsave_mask = fpmask;
|
| 342 |
|
|
|
| 343 |
|
|
InitializeTempRegs();
|
| 344 |
|
|
return (popcnt(mask));
|
| 345 |
|
|
}
|
| 346 |
|
|
|
| 347 |
|
|
/*
|
| 348 |
|
|
* opt1 is the externally callable optimization routine. it will
|
| 349 |
|
|
* collect and allocate common subexpressions and substitute the
|
| 350 |
|
|
* tempref for all occurrances of the expression within the block.
|
| 351 |
|
|
*/
|
| 352 |
|
|
int CSETable::Optimize(Statement *block)
|
| 353 |
|
|
{
|
| 354 |
|
|
int nn;
|
| 355 |
|
|
|
| 356 |
|
|
//csendx = 0;
|
| 357 |
|
|
dfs.printf("<CSETable__Optimize>");
|
| 358 |
|
|
nn = 0;
|
| 359 |
|
|
if (pass == 1) {
|
| 360 |
|
|
if (currentFn->csetbl == nullptr) {
|
| 361 |
|
|
currentFn->csetbl = new CSETable;
|
| 362 |
|
|
}
|
| 363 |
|
|
Clear();
|
| 364 |
|
|
}
|
| 365 |
|
|
else if (pass == 2) {
|
| 366 |
|
|
//Assign(currentFn->csetbl);
|
| 367 |
|
|
}
|
| 368 |
|
|
dfs.printf("Pass:%d ", pass);
|
| 369 |
|
|
if (opt_noregs == FALSE) {
|
| 370 |
|
|
if (pass == 1)
|
| 371 |
|
|
block->scan(); /* collect expressions */
|
| 372 |
|
|
nn = AllocateRegisterVars();
|
| 373 |
|
|
if (pass == 2)
|
| 374 |
|
|
block->repcse(); /* replace allocated expressions */
|
| 375 |
|
|
}
|
| 376 |
|
|
if (pass == 1)
|
| 377 |
|
|
;// currentFn->csetbl->Assign(pCSETable);
|
| 378 |
|
|
else if (pass == 2) {
|
| 379 |
|
|
if (currentFn->csetbl && !currentFn->IsInline) {
|
| 380 |
|
|
delete currentFn->csetbl;
|
| 381 |
|
|
currentFn->csetbl = nullptr;
|
| 382 |
|
|
}
|
| 383 |
|
|
}
|
| 384 |
|
|
dfs.printf("</CSETable__Optimize>\n");
|
| 385 |
|
|
return (nn);
|
| 386 |
|
|
}
|
| 387 |
|
|
|
| 388 |
|
|
// Immediate constants have low priority.
|
| 389 |
|
|
// Even though their use might be high, they are given a low priority.
|
| 390 |
|
|
|
| 391 |
|
|
void CSETable::Dump()
|
| 392 |
|
|
{
|
| 393 |
|
|
int nn;
|
| 394 |
|
|
CSE *csp;
|
| 395 |
|
|
|
| 396 |
|
|
dfs.printf("<CSETable>For %s\n", (char *)currentFn->sym->name->c_str());
|
| 397 |
|
|
dfs.printf(
|
| 398 |
|
|
"*The expression must be used three or more times before it will be allocated\n"
|
| 399 |
|
|
"to a register.\n");
|
| 400 |
|
|
dfs.printf("N OD Uses DUses Void Reg Sym\n");
|
| 401 |
|
|
for (nn = 0; nn < csendx; nn++) {
|
| 402 |
|
|
csp = &table[nn];
|
| 403 |
|
|
dfs.printf("%d: ", nn);
|
| 404 |
|
|
dfs.printf("%d ", csp->OptimizationDesireability());
|
| 405 |
|
|
dfs.printf("%d ", csp->uses);
|
| 406 |
|
|
dfs.printf("%d ", csp->duses);
|
| 407 |
|
|
dfs.printf("%d ", (int)csp->voidf);
|
| 408 |
|
|
dfs.printf("%d ", csp->reg);
|
| 409 |
|
|
if (csp->exp && csp->exp->sym)
|
| 410 |
|
|
dfs.printf("%s ", (char *)csp->exp->sym->name->c_str());
|
| 411 |
|
|
if (csp->exp && csp->exp->sp)
|
| 412 |
|
|
dfs.printf("%s ", (char *)((std::string *)(csp->exp->sp))->c_str());
|
| 413 |
|
|
dfs.printf("\n");
|
| 414 |
|
|
}
|
| 415 |
|
|
dfs.printf("</CSETable>\n");
|
| 416 |
|
|
}
|
| 417 |
|
|
|
