1 /*****************************************************************************/
5 /* Optimizer subroutines */
9 /* (C) 2001-2009 Ullrich von Bassewitz */
10 /* Roemerstrasse 52 */
11 /* D-70794 Filderstadt */
12 /* EMail: uz@cc65.org */
15 /* This software is provided 'as-is', without any expressed or implied */
16 /* warranty. In no event will the authors be held liable for any damages */
17 /* arising from the use of this software. */
19 /* Permission is granted to anyone to use this software for any purpose, */
20 /* including commercial applications, and to alter it and redistribute it */
21 /* freely, subject to the following restrictions: */
23 /* 1. The origin of this software must not be misrepresented; you must not */
24 /* claim that you wrote the original software. If you use this software */
25 /* in a product, an acknowledgment in the product documentation would be */
26 /* appreciated but is not required. */
27 /* 2. Altered source versions must be plainly marked as such, and must not */
28 /* be misrepresented as being the original software. */
29 /* 3. This notice may not be removed or altered from any source */
32 /*****************************************************************************/
55 #include "coptptrload.h"
59 #include "coptstore.h"
68 /*****************************************************************************/
70 /*****************************************************************************/
85 /*****************************************************************************/
87 /*****************************************************************************/
91 static unsigned OptShift1 (CodeSeg* S)
92 /* A call to the shlaxN routine may get replaced by one or more asl insns
93 * if the value of X is not used later.
98 /* Walk over the entries */
100 while (I < CS_GetEntryCount (S)) {
103 CodeEntry* E = CS_GetEntry (S, I);
105 /* Check for the sequence */
106 if (E->OPC == OP65_JSR &&
107 (strncmp (E->Arg, "shlax", 5) == 0 ||
108 strncmp (E->Arg, "aslax", 5) == 0) &&
109 strlen (E->Arg) == 6 &&
110 IsDigit (E->Arg[5]) &&
111 !RegXUsed (S, I+1)) {
113 /* Insert shift insns */
114 unsigned Count = E->Arg[5] - '0';
116 CodeEntry* X = NewCodeEntry (OP65_ASL, AM65_ACC, "a", 0, E->LI);
117 CS_InsertEntry (S, X, I+1);
120 /* Delete the call to shlax */
123 /* Remember, we had changes */
133 /* Return the number of changes made */
139 static unsigned OptShift2 (CodeSeg* S)
140 /* A call to the shraxN routine may get replaced by one or more lsr insns
141 * if the value of X is zero.
144 unsigned Changes = 0;
147 /* Generate register info */
150 /* Walk over the entries */
152 while (I < CS_GetEntryCount (S)) {
155 CodeEntry* E = CS_GetEntry (S, I);
157 /* Check for the sequence */
158 if (E->OPC == OP65_JSR &&
159 strncmp (E->Arg, "shrax", 5) == 0 &&
160 strlen (E->Arg) == 6 &&
161 IsDigit (E->Arg[5]) &&
162 E->RI->In.RegX == 0) {
164 /* Insert shift insns */
165 unsigned Count = E->Arg[5] - '0';
167 CodeEntry* X = NewCodeEntry (OP65_LSR, AM65_ACC, "a", 0, E->LI);
168 CS_InsertEntry (S, X, I+1);
171 /* Delete the call to shlax */
174 /* Remember, we had changes */
184 /* Free the register info */
187 /* Return the number of changes made */
193 static unsigned GetShiftType (const char* Sub)
194 /* Helper function for OptShift3 */
197 if (strcmp (Sub+1, "slax1") == 0) {
199 } else if (strcmp (Sub+1, "srax1") == 0) {
202 } else if (*Sub == 's') {
203 if (strcmp (Sub+1, "hlax1") == 0) {
205 } else if (strcmp (Sub+1, "hrax1") == 0) {
214 static unsigned OptShift3 (CodeSeg* S)
215 /* Search for the sequence
219 * jsr aslax1/asrax1/shlax1/shrax1
232 * or similar, provided that a/x is not used later
235 unsigned Changes = 0;
237 /* Walk over the entries */
239 while (I < CS_GetEntryCount (S)) {
245 L[0] = CS_GetEntry (S, I);
247 /* Check for the sequence */
248 if (L[0]->OPC == OP65_LDA &&
249 (L[0]->AM == AM65_ABS || L[0]->AM == AM65_ZP) &&
250 CS_GetEntries (S, L+1, I+1, 4) &&
251 !CS_RangeHasLabel (S, I+1, 4) &&
252 L[1]->OPC == OP65_LDX &&
253 (L[1]->AM == AM65_ABS || L[1]->AM == AM65_ZP) &&
254 L[2]->OPC == OP65_JSR &&
255 (ShiftType = GetShiftType (L[2]->Arg)) != SHIFT_NONE&&
256 L[3]->OPC == OP65_STA &&
257 (L[3]->AM == AM65_ABS || L[3]->AM == AM65_ZP) &&
258 L[4]->OPC == OP65_STX &&
259 (L[4]->AM == AM65_ABS || L[4]->AM == AM65_ZP) &&
260 !RegAXUsed (S, I+5)) {
264 /* Handle the four shift types differently */
268 X = NewCodeEntry (OP65_LDA, L[1]->AM, L[1]->Arg, 0, L[1]->LI);
269 CS_InsertEntry (S, X, I+5);
270 X = NewCodeEntry (OP65_CMP, AM65_IMM, "$80", 0, L[2]->LI);
271 CS_InsertEntry (S, X, I+6);
272 X = NewCodeEntry (OP65_ROR, AM65_ACC, "a", 0, L[2]->LI);
273 CS_InsertEntry (S, X, I+7);
274 X = NewCodeEntry (OP65_STA, L[4]->AM, L[4]->Arg, 0, L[4]->LI);
275 CS_InsertEntry (S, X, I+8);
276 X = NewCodeEntry (OP65_LDA, L[0]->AM, L[0]->Arg, 0, L[0]->LI);
277 CS_InsertEntry (S, X, I+9);
278 X = NewCodeEntry (OP65_ROR, AM65_ACC, "a", 0, L[2]->LI);
279 CS_InsertEntry (S, X, I+10);
280 X = NewCodeEntry (OP65_STA, L[3]->AM, L[3]->Arg, 0, L[3]->LI);
281 CS_InsertEntry (S, X, I+11);
282 CS_DelEntries (S, I, 5);
286 X = NewCodeEntry (OP65_LDA, L[1]->AM, L[1]->Arg, 0, L[1]->LI);
287 CS_InsertEntry (S, X, I+5);
288 X = NewCodeEntry (OP65_LSR, AM65_ACC, "a", 0, L[2]->LI);
289 CS_InsertEntry (S, X, I+6);
290 X = NewCodeEntry (OP65_STA, L[4]->AM, L[4]->Arg, 0, L[4]->LI);
291 CS_InsertEntry (S, X, I+7);
292 X = NewCodeEntry (OP65_LDA, L[0]->AM, L[0]->Arg, 0, L[0]->LI);
293 CS_InsertEntry (S, X, I+8);
294 X = NewCodeEntry (OP65_ROR, AM65_ACC, "a", 0, L[2]->LI);
295 CS_InsertEntry (S, X, I+9);
296 X = NewCodeEntry (OP65_STA, L[3]->AM, L[3]->Arg, 0, L[3]->LI);
297 CS_InsertEntry (S, X, I+10);
298 CS_DelEntries (S, I, 5);
303 /* These two are identical */
304 X = NewCodeEntry (OP65_ASL, AM65_ACC, "a", 0, L[2]->LI);
305 CS_InsertEntry (S, X, I+1);
306 X = NewCodeEntry (OP65_STA, L[3]->AM, L[3]->Arg, 0, L[3]->LI);
307 CS_InsertEntry (S, X, I+2);
308 X = NewCodeEntry (OP65_LDA, L[1]->AM, L[1]->Arg, 0, L[1]->LI);
309 CS_InsertEntry (S, X, I+3);
310 X = NewCodeEntry (OP65_ROL, AM65_ACC, "a", 0, L[2]->LI);
311 CS_InsertEntry (S, X, I+4);
312 X = NewCodeEntry (OP65_STA, L[4]->AM, L[4]->Arg, 0, L[4]->LI);
313 CS_InsertEntry (S, X, I+5);
314 CS_DelEntries (S, I+6, 4);
319 /* Remember, we had changes */
329 /* Return the number of changes made */
335 /*****************************************************************************/
337 /*****************************************************************************/
341 static unsigned OptLoad1 (CodeSeg* S)
342 /* Search for a call to ldaxysp where X is not used later and replace it by
343 * a load of just the A register.
347 unsigned Changes = 0;
349 /* Generate register info */
352 /* Walk over the entries */
354 while (I < CS_GetEntryCount (S)) {
359 E = CS_GetEntry (S, I);
361 /* Check for the sequence */
362 if (CE_IsCallTo (E, "ldaxysp") &&
363 RegValIsKnown (E->RI->In.RegY) &&
364 !RegXUsed (S, I+1)) {
368 /* Reload the Y register */
369 const char* Arg = MakeHexArg (E->RI->In.RegY - 1);
370 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
371 CS_InsertEntry (S, X, I+1);
373 /* Load from stack */
374 X = NewCodeEntry (OP65_LDA, AM65_ZP_INDY, "sp", 0, E->LI);
375 CS_InsertEntry (S, X, I+2);
377 /* Now remove the call to the subroutine */
380 /* Remember, we had changes */
390 /* Free the register info */
393 /* Return the number of changes made */
399 /*****************************************************************************/
400 /* Optimize stores through pointers */
401 /*****************************************************************************/
405 static unsigned OptPtrStore1Sub (CodeSeg* S, unsigned I, CodeEntry** const L)
406 /* Check if this is one of the allowed suboperation for OptPtrStore1 */
408 /* Check for a label attached to the entry */
409 if (CE_HasLabel (L[0])) {
413 /* Check for single insn sub ops */
414 if (L[0]->OPC == OP65_AND ||
415 L[0]->OPC == OP65_EOR ||
416 L[0]->OPC == OP65_ORA ||
417 (L[0]->OPC == OP65_JSR && strncmp (L[0]->Arg, "shlax", 5) == 0) ||
418 (L[0]->OPC == OP65_JSR && strncmp (L[0]->Arg, "shrax", 5) == 0)) {
423 } else if (L[0]->OPC == OP65_CLC &&
424 (L[1] = CS_GetNextEntry (S, I)) != 0 &&
425 L[1]->OPC == OP65_ADC &&
426 !CE_HasLabel (L[1])) {
428 } else if (L[0]->OPC == OP65_SEC &&
429 (L[1] = CS_GetNextEntry (S, I)) != 0 &&
430 L[1]->OPC == OP65_SBC &&
431 !CE_HasLabel (L[1])) {
443 static unsigned OptPtrStore1 (CodeSeg* S)
444 /* Search for the sequence:
464 * In case a/x is loaded from the register bank before the pushax, we can even
465 * use the register bank instead of ptr1.
479 unsigned Changes = 0;
481 /* Walk over the entries */
483 while (I < CS_GetEntryCount (S)) {
489 L[0] = CS_GetEntry (S, I);
491 /* Check for the sequence */
492 if (CE_IsCallTo (L[0], "pushax") &&
493 CS_GetEntries (S, L+1, I+1, 3) &&
494 L[1]->OPC == OP65_LDY &&
495 CE_IsConstImm (L[1]) &&
496 !CE_HasLabel (L[1]) &&
497 CE_IsCallTo (L[2], "ldauidx") &&
498 !CE_HasLabel (L[2]) &&
499 (K = OptPtrStore1Sub (S, I+3, L+3)) > 0 &&
500 CS_GetEntries (S, L+3+K, I+3+K, 2) &&
501 L[3+K]->OPC == OP65_LDY &&
502 CE_IsConstImm (L[3+K]) &&
503 !CE_HasLabel (L[3+K]) &&
504 CE_IsCallTo (L[4+K], "staspidx") &&
505 !CE_HasLabel (L[4+K])) {
508 const char* RegBank = 0;
509 const char* ZPLoc = "ptr1";
513 /* Get the preceeding two instructions and check them. We check
520 P[0] = CS_GetEntry (S, I-2);
521 P[1] = CS_GetEntry (S, I-1);
522 if (P[0]->OPC == OP65_LDA &&
523 P[0]->AM == AM65_ZP &&
524 P[1]->OPC == OP65_LDX &&
525 P[1]->AM == AM65_ZP &&
526 !CE_HasLabel (P[1]) &&
527 strncmp (P[0]->Arg, "regbank+", 8) == 0) {
529 unsigned Len = strlen (P[0]->Arg);
531 if (strncmp (P[0]->Arg, P[1]->Arg, Len) == 0 &&
532 P[1]->Arg[Len+0] == '+' &&
533 P[1]->Arg[Len+1] == '1' &&
534 P[1]->Arg[Len+2] == '\0') {
536 /* Ok, found. Use the name of the register bank */
537 RegBank = ZPLoc = P[0]->Arg;
542 /* Insert the load via the zp pointer */
543 X = NewCodeEntry (OP65_LDX, AM65_IMM, "$00", 0, L[3]->LI);
544 CS_InsertEntry (S, X, I+3);
545 X = NewCodeEntry (OP65_LDA, AM65_ZP_INDY, ZPLoc, 0, L[2]->LI);
546 CS_InsertEntry (S, X, I+4);
548 /* Insert the store through the zp pointer */
549 X = NewCodeEntry (OP65_STA, AM65_ZP_INDY, ZPLoc, 0, L[3]->LI);
550 CS_InsertEntry (S, X, I+6+K);
552 /* Delete the old code */
553 CS_DelEntry (S, I+7+K); /* jsr spaspidx */
554 CS_DelEntry (S, I+2); /* jsr ldauidx */
556 /* Create and insert the stores into the zp pointer if needed */
558 X = NewCodeEntry (OP65_STA, AM65_ZP, "ptr1", 0, L[0]->LI);
559 CS_InsertEntry (S, X, I+1);
560 X = NewCodeEntry (OP65_STX, AM65_ZP, "ptr1+1", 0, L[0]->LI);
561 CS_InsertEntry (S, X, I+2);
564 /* Delete more old code. Do it here to keep a label attached to
567 CS_DelEntry (S, I); /* jsr pushax */
569 /* Remember, we had changes */
579 /* Return the number of changes made */
585 static unsigned OptPtrStore2 (CodeSeg* S)
586 /* Search for the sequence:
608 unsigned Changes = 0;
610 /* Walk over the entries */
612 while (I < CS_GetEntryCount (S)) {
618 L[0] = CS_GetEntry (S, I);
620 /* Check for the sequence */
621 if (L[0]->OPC == OP65_LDA &&
622 L[0]->AM == AM65_IMM &&
623 CS_GetEntries (S, L+1, I+1, 10) &&
624 L[1]->OPC == OP65_LDX &&
625 L[1]->AM == AM65_IMM &&
626 L[2]->OPC == OP65_CLC &&
627 L[3]->OPC == OP65_ADC &&
628 (L[3]->AM == AM65_ABS || L[3]->AM == AM65_ZP) &&
629 (L[4]->OPC == OP65_BCC || L[4]->OPC == OP65_JCC) &&
631 L[4]->JumpTo->Owner == L[6] &&
632 L[5]->OPC == OP65_INX &&
633 CE_IsCallTo (L[6], "pushax") &&
634 L[7]->OPC == OP65_LDX &&
635 L[8]->OPC == OP65_LDA &&
636 L[9]->OPC == OP65_LDY &&
637 CE_IsKnownImm (L[9], 0) &&
638 CE_IsCallTo (L[10], "staspidx") &&
639 !CS_RangeHasLabel (S, I+1, 5) &&
640 !CS_RangeHasLabel (S, I+7, 4) &&
641 /* Check the label last because this is quite costly */
642 (Len = strlen (L[0]->Arg)) > 3 &&
643 L[0]->Arg[0] == '<' &&
644 L[0]->Arg[1] == '(' &&
645 strlen (L[1]->Arg) == Len &&
646 L[1]->Arg[0] == '>' &&
647 memcmp (L[0]->Arg+1, L[1]->Arg+1, Len-1) == 0) {
652 /* We will create all the new stuff behind the current one so
653 * we keep the line references.
655 X = NewCodeEntry (OP65_LDY, L[3]->AM, L[3]->Arg, 0, L[0]->LI);
656 CS_InsertEntry (S, X, I+11);
658 X = NewCodeEntry (OP65_LDX, L[7]->AM, L[7]->Arg, 0, L[7]->LI);
659 CS_InsertEntry (S, X, I+12);
661 X = NewCodeEntry (OP65_LDA, L[8]->AM, L[8]->Arg, 0, L[8]->LI);
662 CS_InsertEntry (S, X, I+13);
664 Label = memcpy (xmalloc (Len-2), L[0]->Arg+2, Len-3);
666 X = NewCodeEntry (OP65_STA, AM65_ABSY, Label, 0, L[10]->LI);
667 CS_InsertEntry (S, X, I+14);
670 /* Remove the old code */
671 CS_DelEntries (S, I, 11);
673 /* Remember, we had changes */
683 /* Return the number of changes made */
689 /*****************************************************************************/
690 /* Decouple operations */
691 /*****************************************************************************/
695 static unsigned OptDecouple (CodeSeg* S)
696 /* Decouple operations, that is, do the following replacements:
710 * Provided that the register values are known of course.
713 unsigned Changes = 0;
716 /* Generate register info for the following step */
719 /* Walk over the entries */
721 while (I < CS_GetEntryCount (S)) {
725 /* Get next entry and it's input register values */
726 CodeEntry* E = CS_GetEntry (S, I);
727 const RegContents* In = &E->RI->In;
729 /* Assume we have no replacement */
732 /* Check the instruction */
736 if (RegValIsKnown (In->RegA)) {
737 Arg = MakeHexArg ((In->RegA - 1) & 0xFF);
738 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
743 if (RegValIsKnown (In->RegX)) {
744 Arg = MakeHexArg ((In->RegX - 1) & 0xFF);
745 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
750 if (RegValIsKnown (In->RegY)) {
751 Arg = MakeHexArg ((In->RegY - 1) & 0xFF);
752 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
757 if (RegValIsKnown (In->RegA)) {
758 Arg = MakeHexArg ((In->RegA + 1) & 0xFF);
759 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
764 if (RegValIsKnown (In->RegX)) {
765 Arg = MakeHexArg ((In->RegX + 1) & 0xFF);
766 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
771 if (RegValIsKnown (In->RegY)) {
772 Arg = MakeHexArg ((In->RegY + 1) & 0xFF);
773 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
778 if (E->AM == AM65_ZP) {
779 switch (GetKnownReg (E->Use & REG_ZP, In)) {
781 Arg = MakeHexArg (In->Tmp1);
782 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
786 Arg = MakeHexArg (In->Ptr1Lo);
787 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
791 Arg = MakeHexArg (In->Ptr1Hi);
792 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
796 Arg = MakeHexArg (In->SRegLo);
797 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
801 Arg = MakeHexArg (In->SRegHi);
802 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
809 if (E->AM == AM65_ZP) {
810 switch (GetKnownReg (E->Use & REG_ZP, In)) {
812 Arg = MakeHexArg (In->Tmp1);
813 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
817 Arg = MakeHexArg (In->Ptr1Lo);
818 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
822 Arg = MakeHexArg (In->Ptr1Hi);
823 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
827 Arg = MakeHexArg (In->SRegLo);
828 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
832 Arg = MakeHexArg (In->SRegHi);
833 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
840 if (E->AM == AM65_ZP) {
841 switch (GetKnownReg (E->Use, In)) {
843 Arg = MakeHexArg (In->Tmp1);
844 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
848 Arg = MakeHexArg (In->Ptr1Lo);
849 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
853 Arg = MakeHexArg (In->Ptr1Hi);
854 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
858 Arg = MakeHexArg (In->SRegLo);
859 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
863 Arg = MakeHexArg (In->SRegHi);
864 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
871 if (E->RI->In.RegA >= 0) {
872 Arg = MakeHexArg (In->RegA);
873 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
878 if (E->RI->In.RegA >= 0) {
879 Arg = MakeHexArg (In->RegA);
880 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
885 if (E->RI->In.RegX >= 0) {
886 Arg = MakeHexArg (In->RegX);
887 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
892 if (E->RI->In.RegY >= 0) {
893 Arg = MakeHexArg (In->RegY);
894 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
899 /* Avoid gcc warnings */
904 /* Insert the replacement if we have one */
906 CS_InsertEntry (S, X, I+1);
916 /* Free register info */
919 /* Return the number of changes made */
925 /*****************************************************************************/
927 /*****************************************************************************/
931 typedef struct OptFunc OptFunc;
933 unsigned (*Func) (CodeSeg*); /* Optimizer function */
934 const char* Name; /* Name of the function/group */
935 unsigned CodeSizeFactor; /* Code size factor for this opt func */
936 unsigned long TotalRuns; /* Total number of runs */
937 unsigned long LastRuns; /* Last number of runs */
938 unsigned long TotalChanges; /* Total number of changes */
939 unsigned long LastChanges; /* Last number of changes */
940 char Disabled; /* True if function disabled */
945 /*****************************************************************************/
947 /*****************************************************************************/
951 /* A list of all the function descriptions */
952 static OptFunc DOpt65C02BitOps = { Opt65C02BitOps, "Opt65C02BitOps", 66, 0, 0, 0, 0, 0 };
953 static OptFunc DOpt65C02Ind = { Opt65C02Ind, "Opt65C02Ind", 100, 0, 0, 0, 0, 0 };
954 static OptFunc DOpt65C02Stores = { Opt65C02Stores, "Opt65C02Stores", 100, 0, 0, 0, 0, 0 };
955 static OptFunc DOptAdd1 = { OptAdd1, "OptAdd1", 125, 0, 0, 0, 0, 0 };
956 static OptFunc DOptAdd2 = { OptAdd2, "OptAdd2", 200, 0, 0, 0, 0, 0 };
957 static OptFunc DOptAdd3 = { OptAdd3, "OptAdd3", 65, 0, 0, 0, 0, 0 };
958 static OptFunc DOptAdd4 = { OptAdd4, "OptAdd4", 90, 0, 0, 0, 0, 0 };
959 static OptFunc DOptAdd5 = { OptAdd5, "OptAdd5", 100, 0, 0, 0, 0, 0 };
960 static OptFunc DOptAdd6 = { OptAdd6, "OptAdd6", 40, 0, 0, 0, 0, 0 };
961 static OptFunc DOptBoolTrans = { OptBoolTrans, "OptBoolTrans", 100, 0, 0, 0, 0, 0 };
962 static OptFunc DOptBranchDist = { OptBranchDist, "OptBranchDist", 0, 0, 0, 0, 0, 0 };
963 static OptFunc DOptCmp1 = { OptCmp1, "OptCmp1", 42, 0, 0, 0, 0, 0 };
964 static OptFunc DOptCmp2 = { OptCmp2, "OptCmp2", 85, 0, 0, 0, 0, 0 };
965 static OptFunc DOptCmp3 = { OptCmp3, "OptCmp3", 75, 0, 0, 0, 0, 0 };
966 static OptFunc DOptCmp4 = { OptCmp4, "OptCmp4", 75, 0, 0, 0, 0, 0 };
967 static OptFunc DOptCmp5 = { OptCmp5, "OptCmp5", 100, 0, 0, 0, 0, 0 };
968 static OptFunc DOptCmp6 = { OptCmp6, "OptCmp6", 100, 0, 0, 0, 0, 0 };
969 static OptFunc DOptCmp7 = { OptCmp7, "OptCmp7", 85, 0, 0, 0, 0, 0 };
970 static OptFunc DOptCmp8 = { OptCmp8, "OptCmp8", 50, 0, 0, 0, 0, 0 };
971 static OptFunc DOptCondBranches = { OptCondBranches, "OptCondBranches", 80, 0, 0, 0, 0, 0 };
972 static OptFunc DOptDeadCode = { OptDeadCode, "OptDeadCode", 100, 0, 0, 0, 0, 0 };
973 static OptFunc DOptDeadJumps = { OptDeadJumps, "OptDeadJumps", 100, 0, 0, 0, 0, 0 };
974 static OptFunc DOptDecouple = { OptDecouple, "OptDecouple", 100, 0, 0, 0, 0, 0 };
975 static OptFunc DOptDupLoads = { OptDupLoads, "OptDupLoads", 0, 0, 0, 0, 0, 0 };
976 static OptFunc DOptJumpCascades = { OptJumpCascades, "OptJumpCascades", 100, 0, 0, 0, 0, 0 };
977 static OptFunc DOptJumpTarget = { OptJumpTarget, "OptJumpTarget", 100, 0, 0, 0, 0, 0 };
978 static OptFunc DOptLoad1 = { OptLoad1, "OptLoad1", 100, 0, 0, 0, 0, 0 };
979 static OptFunc DOptRTS = { OptRTS, "OptRTS", 100, 0, 0, 0, 0, 0 };
980 static OptFunc DOptRTSJumps1 = { OptRTSJumps1, "OptRTSJumps1", 100, 0, 0, 0, 0, 0 };
981 static OptFunc DOptRTSJumps2 = { OptRTSJumps2, "OptRTSJumps2", 100, 0, 0, 0, 0, 0 };
982 static OptFunc DOptNegA1 = { OptNegA1, "OptNegA1", 100, 0, 0, 0, 0, 0 };
983 static OptFunc DOptNegA2 = { OptNegA2, "OptNegA2", 100, 0, 0, 0, 0, 0 };
984 static OptFunc DOptNegAX1 = { OptNegAX1, "OptNegAX1", 100, 0, 0, 0, 0, 0 };
985 static OptFunc DOptNegAX2 = { OptNegAX2, "OptNegAX2", 100, 0, 0, 0, 0, 0 };
986 static OptFunc DOptNegAX3 = { OptNegAX3, "OptNegAX3", 100, 0, 0, 0, 0, 0 };
987 static OptFunc DOptNegAX4 = { OptNegAX4, "OptNegAX4", 100, 0, 0, 0, 0, 0 };
988 static OptFunc DOptPrecalc = { OptPrecalc, "OptPrecalc", 100, 0, 0, 0, 0, 0 };
989 static OptFunc DOptPtrLoad1 = { OptPtrLoad1, "OptPtrLoad1", 100, 0, 0, 0, 0, 0 };
990 static OptFunc DOptPtrLoad2 = { OptPtrLoad2, "OptPtrLoad2", 100, 0, 0, 0, 0, 0 };
991 static OptFunc DOptPtrLoad3 = { OptPtrLoad3, "OptPtrLoad3", 100, 0, 0, 0, 0, 0 };
992 static OptFunc DOptPtrLoad4 = { OptPtrLoad4, "OptPtrLoad4", 100, 0, 0, 0, 0, 0 };
993 static OptFunc DOptPtrLoad5 = { OptPtrLoad5, "OptPtrLoad5", 92, 0, 0, 0, 0, 0 };
994 static OptFunc DOptPtrLoad6 = { OptPtrLoad6, "OptPtrLoad6", 50, 0, 0, 0, 0, 0 };
995 static OptFunc DOptPtrLoad7 = { OptPtrLoad7, "OptPtrLoad7", 65, 0, 0, 0, 0, 0 };
996 static OptFunc DOptPtrLoad8 = { OptPtrLoad8, "OptPtrLoad8", 108, 0, 0, 0, 0, 0 };
997 static OptFunc DOptPtrLoad9 = { OptPtrLoad9, "OptPtrLoad9", 86, 0, 0, 0, 0, 0 };
998 static OptFunc DOptPtrLoad10 = { OptPtrLoad10, "OptPtrLoad10", 100, 0, 0, 0, 0, 0 };
999 static OptFunc DOptPtrStore1 = { OptPtrStore1, "OptPtrStore1", 100, 0, 0, 0, 0, 0 };
1000 static OptFunc DOptPtrStore2 = { OptPtrStore2, "OptPtrStore2", 40, 0, 0, 0, 0, 0 };
1001 static OptFunc DOptPush1 = { OptPush1, "OptPush1", 65, 0, 0, 0, 0, 0 };
1002 static OptFunc DOptPush2 = { OptPush2, "OptPush2", 50, 0, 0, 0, 0, 0 };
1003 static OptFunc DOptPushPop = { OptPushPop, "OptPushPop", 0, 0, 0, 0, 0, 0 };
1004 static OptFunc DOptShift1 = { OptShift1, "OptShift1", 100, 0, 0, 0, 0, 0 };
1005 static OptFunc DOptShift2 = { OptShift2, "OptShift2", 100, 0, 0, 0, 0, 0 };
1006 static OptFunc DOptShift3 = { OptShift3, "OptShift3", 110, 0, 0, 0, 0, 0 };
1007 static OptFunc DOptSize1 = { OptSize1, "OptSize1", 100, 0, 0, 0, 0, 0 };
1008 static OptFunc DOptSize2 = { OptSize2, "OptSize2", 100, 0, 0, 0, 0, 0 };
1009 static OptFunc DOptStackOps = { OptStackOps, "OptStackOps", 100, 0, 0, 0, 0, 0 };
1010 static OptFunc DOptStore1 = { OptStore1, "OptStore1", 70, 0, 0, 0, 0, 0 };
1011 static OptFunc DOptStore2 = { OptStore2, "OptStore2", 220, 0, 0, 0, 0, 0 };
1012 static OptFunc DOptStore3 = { OptStore3, "OptStore3", 120, 0, 0, 0, 0, 0 };
1013 static OptFunc DOptStore4 = { OptStore4, "OptStore4", 50, 0, 0, 0, 0, 0 };
1014 static OptFunc DOptStore5 = { OptStore5, "OptStore5", 100, 0, 0, 0, 0, 0 };
1015 static OptFunc DOptStoreLoad = { OptStoreLoad, "OptStoreLoad", 0, 0, 0, 0, 0, 0 };
1016 static OptFunc DOptSub1 = { OptSub1, "OptSub1", 100, 0, 0, 0, 0, 0 };
1017 static OptFunc DOptSub2 = { OptSub2, "OptSub2", 100, 0, 0, 0, 0, 0 };
1018 static OptFunc DOptSub3 = { OptSub3, "OptSub3", 100, 0, 0, 0, 0, 0 };
1019 static OptFunc DOptTest1 = { OptTest1, "OptTest1", 100, 0, 0, 0, 0, 0 };
1020 static OptFunc DOptTransfers1 = { OptTransfers1, "OptTransfers1", 0, 0, 0, 0, 0, 0 };
1021 static OptFunc DOptTransfers2 = { OptTransfers2, "OptTransfers2", 60, 0, 0, 0, 0, 0 };
1022 static OptFunc DOptTransfers3 = { OptTransfers3, "OptTransfers3", 65, 0, 0, 0, 0, 0 };
1023 static OptFunc DOptTransfers4 = { OptTransfers4, "OptTransfers4", 65, 0, 0, 0, 0, 0 };
1024 static OptFunc DOptUnusedLoads = { OptUnusedLoads, "OptUnusedLoads", 0, 0, 0, 0, 0, 0 };
1025 static OptFunc DOptUnusedStores = { OptUnusedStores, "OptUnusedStores", 0, 0, 0, 0, 0, 0 };
1028 /* Table containing all the steps in alphabetical order */
1029 static OptFunc* OptFuncs[] = {
1105 #define OPTFUNC_COUNT (sizeof(OptFuncs) / sizeof(OptFuncs[0]))
1109 static int CmpOptStep (const void* Key, const void* Func)
1110 /* Compare function for bsearch */
1112 return strcmp (Key, (*(const OptFunc**)Func)->Name);
1117 static OptFunc* FindOptFunc (const char* Name)
1118 /* Find an optimizer step by name in the table and return a pointer. Return
1119 * NULL if no such step is found.
1122 /* Search for the function in the list */
1123 OptFunc** O = bsearch (Name, OptFuncs, OPTFUNC_COUNT, sizeof (OptFuncs[0]), CmpOptStep);
1129 static OptFunc* GetOptFunc (const char* Name)
1130 /* Find an optimizer step by name in the table and return a pointer. Print an
1131 * error and call AbEnd if not found.
1134 /* Search for the function in the list */
1135 OptFunc* F = FindOptFunc (Name);
1138 AbEnd ("Optimization step `%s' not found", Name);
1145 void DisableOpt (const char* Name)
1146 /* Disable the optimization with the given name */
1148 if (strcmp (Name, "any") == 0) {
1150 for (I = 0; I < OPTFUNC_COUNT; ++I) {
1151 OptFuncs[I]->Disabled = 1;
1154 GetOptFunc(Name)->Disabled = 1;
1160 void EnableOpt (const char* Name)
1161 /* Enable the optimization with the given name */
1163 if (strcmp (Name, "any") == 0) {
1165 for (I = 0; I < OPTFUNC_COUNT; ++I) {
1166 OptFuncs[I]->Disabled = 0;
1169 GetOptFunc(Name)->Disabled = 0;
1175 void ListOptSteps (FILE* F)
1176 /* List all optimization steps */
1179 for (I = 0; I < OPTFUNC_COUNT; ++I) {
1180 fprintf (F, "%s\n", OptFuncs[I]->Name);
1186 static void ReadOptStats (const char* Name)
1187 /* Read the optimizer statistics file */
1192 /* Try to open the file */
1193 FILE* F = fopen (Name, "r");
1195 /* Ignore the error */
1199 /* Read and parse the lines */
1201 while (fgets (Buf, sizeof (Buf), F) != 0) {
1209 unsigned long TotalRuns;
1210 unsigned long TotalChanges;
1215 /* Remove trailing white space including the line terminator */
1218 while (Len > 0 && IsSpace (B[Len-1])) {
1223 /* Remove leading whitespace */
1224 while (IsSpace (*B)) {
1228 /* Check for empty and comment lines */
1229 if (*B == '\0' || *B == ';' || *B == '#') {
1233 /* Parse the line */
1234 if (sscanf (B, "%31s %lu %*u %lu %*u", Name, &TotalRuns, &TotalChanges) != 3) {
1239 /* Search for the optimizer step. */
1240 Func = FindOptFunc (Name);
1246 /* Found the step, set the fields */
1247 Func->TotalRuns = TotalRuns;
1248 Func->TotalChanges = TotalChanges;
1252 /* Close the file, ignore errors here. */
1258 static void WriteOptStats (const char* Name)
1259 /* Write the optimizer statistics file */
1263 /* Try to open the file */
1264 FILE* F = fopen (Name, "w");
1266 /* Ignore the error */
1270 /* Write a header */
1272 "; Optimizer Total Last Total Last\n"
1273 "; Step Runs Runs Chg Chg\n");
1276 /* Write the data */
1277 for (I = 0; I < OPTFUNC_COUNT; ++I) {
1278 const OptFunc* O = OptFuncs[I];
1280 "%-20s %10lu %10lu %10lu %10lu\n",
1288 /* Close the file, ignore errors here. */
1294 static unsigned RunOptFunc (CodeSeg* S, OptFunc* F, unsigned Max)
1295 /* Run one optimizer function Max times or until there are no more changes */
1297 unsigned Changes, C;
1299 /* Don't run the function if it is disabled or if it is prohibited by the
1302 if (F->Disabled || F->CodeSizeFactor > S->CodeSizeFactor) {
1306 /* Run this until there are no more changes */
1310 /* Run the function */
1317 F->TotalChanges += C;
1318 F->LastChanges += C;
1320 } while (--Max && C > 0);
1322 /* Return the number of changes */
1328 static unsigned RunOptGroup1 (CodeSeg* S)
1329 /* Run the first group of optimization steps. These steps translate known
1330 * patterns emitted by the code generator into more optimal patterns. Order
1331 * of the steps is important, because some of the steps done earlier cover
1332 * the same patterns as later steps as subpatterns.
1335 unsigned Changes = 0;
1337 Changes += RunOptFunc (S, &DOptPtrStore1, 1);
1338 Changes += RunOptFunc (S, &DOptPtrStore2, 1);
1339 Changes += RunOptFunc (S, &DOptAdd3, 1); /* Before OptPtrLoad5! */
1340 Changes += RunOptFunc (S, &DOptPtrLoad1, 1);
1341 Changes += RunOptFunc (S, &DOptPtrLoad2, 1);
1342 Changes += RunOptFunc (S, &DOptPtrLoad3, 1);
1343 Changes += RunOptFunc (S, &DOptPtrLoad4, 1);
1344 Changes += RunOptFunc (S, &DOptPtrLoad5, 1);
1345 Changes += RunOptFunc (S, &DOptPtrLoad6, 1);
1346 Changes += RunOptFunc (S, &DOptPtrLoad7, 1);
1347 Changes += RunOptFunc (S, &DOptPtrLoad8, 1);
1348 Changes += RunOptFunc (S, &DOptPtrLoad9, 1);
1349 Changes += RunOptFunc (S, &DOptNegAX1, 1);
1350 Changes += RunOptFunc (S, &DOptNegAX2, 1);
1351 Changes += RunOptFunc (S, &DOptNegAX3, 1);
1352 Changes += RunOptFunc (S, &DOptNegAX4, 1);
1353 Changes += RunOptFunc (S, &DOptAdd1, 1);
1354 Changes += RunOptFunc (S, &DOptAdd2, 1);
1355 Changes += RunOptFunc (S, &DOptAdd4, 1);
1356 Changes += RunOptFunc (S, &DOptStore4, 1);
1357 Changes += RunOptFunc (S, &DOptStore5, 1);
1358 Changes += RunOptFunc (S, &DOptShift1, 1);
1359 Changes += RunOptFunc (S, &DOptShift2, 1);
1360 Changes += RunOptFunc (S, &DOptShift3, 1);
1361 Changes += RunOptFunc (S, &DOptStore1, 1);
1362 Changes += RunOptFunc (S, &DOptStore2, 5);
1363 Changes += RunOptFunc (S, &DOptStore3, 5);
1365 /* Return the number of changes */
1371 static unsigned RunOptGroup2 (CodeSeg* S)
1372 /* Run one group of optimization steps. This step involves just decoupling
1373 * instructions by replacing them by instructions that do not depend on
1374 * previous instructions. This makes it easier to find instructions that
1378 unsigned Changes = 0;
1380 Changes += RunOptFunc (S, &DOptDecouple, 1);
1382 /* Return the number of changes */
1388 static unsigned RunOptGroup3 (CodeSeg* S)
1389 /* Run one group of optimization steps. These steps depend on each other,
1390 * that means that one step may allow another step to do additional work,
1391 * so we will repeat the steps as long as we see any changes.
1394 unsigned Changes, C;
1400 C += RunOptFunc (S, &DOptPtrLoad10, 1);
1401 C += RunOptFunc (S, &DOptNegA1, 1);
1402 C += RunOptFunc (S, &DOptNegA2, 1);
1403 C += RunOptFunc (S, &DOptSub1, 1);
1404 C += RunOptFunc (S, &DOptSub2, 1);
1405 C += RunOptFunc (S, &DOptSub3, 1);
1406 C += RunOptFunc (S, &DOptAdd5, 1);
1407 C += RunOptFunc (S, &DOptAdd6, 1);
1408 C += RunOptFunc (S, &DOptStackOps, 1);
1409 C += RunOptFunc (S, &DOptJumpCascades, 1);
1410 C += RunOptFunc (S, &DOptDeadJumps, 1);
1411 C += RunOptFunc (S, &DOptRTS, 1);
1412 C += RunOptFunc (S, &DOptDeadCode, 1);
1413 C += RunOptFunc (S, &DOptJumpTarget, 1);
1414 C += RunOptFunc (S, &DOptCondBranches, 1);
1415 C += RunOptFunc (S, &DOptRTSJumps1, 1);
1416 C += RunOptFunc (S, &DOptBoolTrans, 1);
1417 C += RunOptFunc (S, &DOptCmp1, 1);
1418 C += RunOptFunc (S, &DOptCmp2, 1);
1419 C += RunOptFunc (S, &DOptCmp3, 1);
1420 C += RunOptFunc (S, &DOptCmp4, 1);
1421 C += RunOptFunc (S, &DOptCmp5, 1);
1422 C += RunOptFunc (S, &DOptCmp6, 1);
1423 C += RunOptFunc (S, &DOptCmp7, 1);
1424 C += RunOptFunc (S, &DOptCmp8, 1);
1425 C += RunOptFunc (S, &DOptTest1, 1);
1426 C += RunOptFunc (S, &DOptLoad1, 1);
1427 C += RunOptFunc (S, &DOptUnusedLoads, 1);
1428 C += RunOptFunc (S, &DOptUnusedStores, 1);
1429 C += RunOptFunc (S, &DOptDupLoads, 1);
1430 C += RunOptFunc (S, &DOptStoreLoad, 1);
1431 C += RunOptFunc (S, &DOptTransfers1, 1);
1432 C += RunOptFunc (S, &DOptTransfers3, 1);
1433 C += RunOptFunc (S, &DOptTransfers4, 1);
1434 C += RunOptFunc (S, &DOptStore1, 1);
1435 C += RunOptFunc (S, &DOptStore5, 1);
1436 C += RunOptFunc (S, &DOptPushPop, 1);
1437 C += RunOptFunc (S, &DOptPrecalc, 1);
1443 /* Return the number of changes */
1449 static unsigned RunOptGroup4 (CodeSeg* S)
1450 /* 65C02 specific optimizations. */
1452 unsigned Changes = 0;
1454 if (CPUIsets[CPU] & CPU_ISET_65SC02) {
1455 Changes += RunOptFunc (S, &DOpt65C02BitOps, 1);
1456 Changes += RunOptFunc (S, &DOpt65C02Ind, 1);
1457 Changes += RunOptFunc (S, &DOpt65C02Stores, 1);
1459 /* The 65C02 replacement codes do often make the use of a register
1460 * value unnecessary, so if we have changes, run another load
1463 Changes += RunOptFunc (S, &DOptUnusedLoads, 1);
1467 /* Return the number of changes */
1473 static unsigned RunOptGroup5 (CodeSeg* S)
1474 /* Run another round of pattern replacements. These are done late, since there
1475 * may be better replacements before.
1478 unsigned Changes = 0;
1480 Changes += RunOptFunc (S, &DOptPush1, 1);
1481 Changes += RunOptFunc (S, &DOptPush2, 1);
1482 Changes += RunOptFunc (S, &DOptUnusedLoads, 1);
1483 Changes += RunOptFunc (S, &DOptTransfers2, 1);
1485 /* Return the number of changes */
1491 static unsigned RunOptGroup6 (CodeSeg* S)
1492 /* The last group of optimization steps. Adjust branches, do size optimizations.
1495 unsigned Changes = 0;
1498 /* Optimize for size, that is replace operations by shorter ones, even
1499 * if this does hinder further optimizations (no problem since we're
1502 C = RunOptFunc (S, &DOptSize1, 1);
1505 /* Run some optimization passes again, since the size optimizations
1506 * may have opened new oportunities.
1508 Changes += RunOptFunc (S, &DOptUnusedLoads, 1);
1509 Changes += RunOptFunc (S, &DOptJumpTarget, 5);
1510 Changes += RunOptFunc (S, &DOptStore5, 1);
1513 C = RunOptFunc (S, &DOptSize2, 1);
1516 /* Run some optimization passes again, since the size optimizations
1517 * may have opened new oportunities.
1519 Changes += RunOptFunc (S, &DOptUnusedLoads, 1);
1520 Changes += RunOptFunc (S, &DOptJumpTarget, 5);
1521 Changes += RunOptFunc (S, &DOptStore5, 1);
1524 /* Adjust branch distances */
1525 Changes += RunOptFunc (S, &DOptBranchDist, 3);
1527 /* Replace conditional branches to RTS. If we had changes, we must run dead
1528 * code elimination again, since the change may have introduced dead code.
1530 C = RunOptFunc (S, &DOptRTSJumps2, 1);
1533 Changes += RunOptFunc (S, &DOptDeadCode, 1);
1536 /* Return the number of changes */
1542 void RunOpt (CodeSeg* S)
1543 /* Run the optimizer */
1545 const char* StatFileName;
1547 /* If we shouldn't run the optimizer, bail out */
1552 /* Check if we are requested to write optimizer statistics */
1553 StatFileName = getenv ("CC65_OPTSTATS");
1555 ReadOptStats (StatFileName);
1558 /* Print the name of the function we are working on */
1560 Print (stdout, 1, "Running optimizer for function `%s'\n", S->Func->Name);
1562 Print (stdout, 1, "Running optimizer for global code segment\n");
1565 /* Run groups of optimizations */
1573 /* Write statistics */
1575 WriteOptStats (StatFileName);