1 /*****************************************************************************/
5 /* Optimizer subroutines */
9 /* (C) 2001-2012, 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 /*****************************************************************************/
44 #include "debugflag.h"
60 #include "coptptrload.h"
61 #include "coptptrstore.h"
63 #include "coptshift.h"
66 #include "coptstore.h"
75 /*****************************************************************************/
77 /*****************************************************************************/
81 static unsigned OptLoad1 (CodeSeg* S)
82 /* Search for a call to ldaxysp where X is not used later and replace it by
83 * a load of just the A register.
89 /* Walk over the entries */
91 while (I < CS_GetEntryCount (S)) {
96 E = CS_GetEntry (S, I);
98 /* Check for the sequence */
99 if (CE_IsCallTo (E, "ldaxysp") &&
100 RegValIsKnown (E->RI->In.RegY) &&
101 !RegXUsed (S, I+1)) {
105 /* Reload the Y register */
106 const char* Arg = MakeHexArg (E->RI->In.RegY - 1);
107 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
108 CS_InsertEntry (S, X, I+1);
110 /* Load from stack */
111 X = NewCodeEntry (OP65_LDA, AM65_ZP_INDY, "sp", 0, E->LI);
112 CS_InsertEntry (S, X, I+2);
114 /* Now remove the call to the subroutine */
117 /* Remember, we had changes */
127 /* Return the number of changes made */
133 static unsigned OptLoad2 (CodeSeg* S)
134 /* Replace calls to ldaxysp by inline code */
137 unsigned Changes = 0;
139 /* Walk over the entries */
141 while (I < CS_GetEntryCount (S)) {
146 L[0] = CS_GetEntry (S, I);
148 /* Check for the sequence */
149 if (CE_IsCallTo (L[0], "ldaxysp")) {
153 /* Followed by sta abs/stx abs? */
154 if (CS_GetEntries (S, L+1, I+1, 2) &&
155 L[1]->OPC == OP65_STA &&
156 L[2]->OPC == OP65_STX &&
159 strcmp (L[1]->Arg, L[2]->Arg) != 0) &&
160 !CS_RangeHasLabel (S, I+1, 2) &&
161 !RegXUsed (S, I+3)) {
163 /* A/X are stored into memory somewhere and X is not used
168 X = NewCodeEntry (OP65_LDA, AM65_ZP_INDY, "sp", 0, L[0]->LI);
169 CS_InsertEntry (S, X, I+3);
172 X = NewCodeEntry (OP65_STA, L[2]->AM, L[2]->Arg, 0, L[2]->LI);
173 CS_InsertEntry (S, X, I+4);
176 X = NewCodeEntry (OP65_DEY, AM65_IMP, 0, 0, L[0]->LI);
177 CS_InsertEntry (S, X, I+5);
180 X = NewCodeEntry (OP65_LDA, AM65_ZP_INDY, "sp", 0, L[0]->LI);
181 CS_InsertEntry (S, X, I+6);
184 X = NewCodeEntry (OP65_STA, L[1]->AM, L[1]->Arg, 0, L[1]->LI);
185 CS_InsertEntry (S, X, I+7);
187 /* Now remove the call to the subroutine and the sta/stx */
188 CS_DelEntries (S, I, 3);
192 /* Standard replacement */
195 X = NewCodeEntry (OP65_LDA, AM65_ZP_INDY, "sp", 0, L[0]->LI);
196 CS_InsertEntry (S, X, I+1);
199 X = NewCodeEntry (OP65_TAX, AM65_IMP, 0, 0, L[0]->LI);
200 CS_InsertEntry (S, X, I+2);
203 X = NewCodeEntry (OP65_DEY, AM65_IMP, 0, 0, L[0]->LI);
204 CS_InsertEntry (S, X, I+3);
207 X = NewCodeEntry (OP65_LDA, AM65_ZP_INDY, "sp", 0, L[0]->LI);
208 CS_InsertEntry (S, X, I+4);
210 /* Now remove the call to the subroutine */
215 /* Remember, we had changes */
224 /* Return the number of changes made */
230 static unsigned OptLoad3 (CodeSeg* S)
231 /* Remove repeated loads from one and the same memory location */
233 unsigned Changes = 0;
236 /* Walk over the entries */
238 while (I < CS_GetEntryCount (S)) {
241 CodeEntry* E = CS_GetEntry (S, I);
243 /* Forget a preceeding load if we have a label */
244 if (Load && CE_HasLabel (E)) {
248 /* Check if this insn is a load */
249 if (E->Info & OF_LOAD) {
251 /* If we had a preceeding load that is identical, remove this one.
252 * If it is not identical, or we didn't have one, remember it.
255 E->OPC == Load->OPC &&
257 ((E->Arg == 0 && Load->Arg == 0) ||
258 strcmp (E->Arg, Load->Arg) == 0)) {
260 /* Now remove the call to the subroutine */
263 /* Remember, we had changes */
275 } else if ((E->Info & OF_CMP) == 0 && (E->Info & OF_CBRA) == 0) {
276 /* Forget the first load on occurance of any insn we don't like */
284 /* Return the number of changes made */
290 /*****************************************************************************/
291 /* Decouple operations */
292 /*****************************************************************************/
296 static unsigned OptDecouple (CodeSeg* S)
297 /* Decouple operations, that is, do the following replacements:
311 * Provided that the register values are known of course.
314 unsigned Changes = 0;
317 /* Walk over the entries */
319 while (I < CS_GetEntryCount (S)) {
323 /* Get next entry and it's input register values */
324 CodeEntry* E = CS_GetEntry (S, I);
325 const RegContents* In = &E->RI->In;
327 /* Assume we have no replacement */
330 /* Check the instruction */
334 if (RegValIsKnown (In->RegA)) {
335 Arg = MakeHexArg ((In->RegA - 1) & 0xFF);
336 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
341 if (RegValIsKnown (In->RegX)) {
342 Arg = MakeHexArg ((In->RegX - 1) & 0xFF);
343 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
348 if (RegValIsKnown (In->RegY)) {
349 Arg = MakeHexArg ((In->RegY - 1) & 0xFF);
350 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
355 if (RegValIsKnown (In->RegA)) {
356 Arg = MakeHexArg ((In->RegA + 1) & 0xFF);
357 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
362 if (RegValIsKnown (In->RegX)) {
363 Arg = MakeHexArg ((In->RegX + 1) & 0xFF);
364 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
369 if (RegValIsKnown (In->RegY)) {
370 Arg = MakeHexArg ((In->RegY + 1) & 0xFF);
371 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
376 if (E->AM == AM65_ZP) {
377 switch (GetKnownReg (E->Use & REG_ZP, In)) {
379 Arg = MakeHexArg (In->Tmp1);
380 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
384 Arg = MakeHexArg (In->Ptr1Lo);
385 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
389 Arg = MakeHexArg (In->Ptr1Hi);
390 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
394 Arg = MakeHexArg (In->SRegLo);
395 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
399 Arg = MakeHexArg (In->SRegHi);
400 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
407 if (E->AM == AM65_ZP) {
408 switch (GetKnownReg (E->Use & REG_ZP, In)) {
410 Arg = MakeHexArg (In->Tmp1);
411 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
415 Arg = MakeHexArg (In->Ptr1Lo);
416 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
420 Arg = MakeHexArg (In->Ptr1Hi);
421 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
425 Arg = MakeHexArg (In->SRegLo);
426 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
430 Arg = MakeHexArg (In->SRegHi);
431 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
438 if (E->AM == AM65_ZP) {
439 switch (GetKnownReg (E->Use, In)) {
441 Arg = MakeHexArg (In->Tmp1);
442 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
446 Arg = MakeHexArg (In->Ptr1Lo);
447 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
451 Arg = MakeHexArg (In->Ptr1Hi);
452 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
456 Arg = MakeHexArg (In->SRegLo);
457 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
461 Arg = MakeHexArg (In->SRegHi);
462 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
469 if (E->RI->In.RegA >= 0) {
470 Arg = MakeHexArg (In->RegA);
471 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
476 if (E->RI->In.RegA >= 0) {
477 Arg = MakeHexArg (In->RegA);
478 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
483 if (E->RI->In.RegX >= 0) {
484 Arg = MakeHexArg (In->RegX);
485 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
490 if (E->RI->In.RegY >= 0) {
491 Arg = MakeHexArg (In->RegY);
492 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
497 /* Avoid gcc warnings */
502 /* Insert the replacement if we have one */
504 CS_InsertEntry (S, X, I+1);
514 /* Return the number of changes made */
520 /*****************************************************************************/
521 /* Optimize stack pointer ops */
522 /*****************************************************************************/
526 static unsigned IsDecSP (const CodeEntry* E)
527 /* Check if this is an insn that decrements the stack pointer. If so, return
528 * the decrement. If not, return zero.
529 * The function expects E to be a subroutine call.
532 if (strncmp (E->Arg, "decsp", 5) == 0) {
533 if (E->Arg[5] >= '1' && E->Arg[5] <= '8') {
534 return (E->Arg[5] - '0');
536 } else if (strcmp (E->Arg, "subysp") == 0 && RegValIsKnown (E->RI->In.RegY)) {
537 return E->RI->In.RegY;
540 /* If we come here, it's not a decsp op */
546 static unsigned OptStackPtrOps (CodeSeg* S)
547 /* Merge adjacent calls to decsp into one. NOTE: This function won't merge all
551 unsigned Changes = 0;
554 /* Walk over the entries */
556 while (I < CS_GetEntryCount (S)) {
562 /* Get the next entry */
563 const CodeEntry* E = CS_GetEntry (S, I);
565 /* Check for decspn or subysp */
566 if (E->OPC == OP65_JSR &&
567 (Dec1 = IsDecSP (E)) > 0 &&
568 (N = CS_GetNextEntry (S, I)) != 0 &&
569 (Dec2 = IsDecSP (N)) > 0 &&
570 (Dec1 += Dec2) <= 255 &&
576 /* We can combine the two */
578 /* Insert a call to decsp */
579 xsprintf (Buf, sizeof (Buf), "decsp%u", Dec1);
580 X = NewCodeEntry (OP65_JSR, AM65_ABS, Buf, 0, N->LI);
581 CS_InsertEntry (S, X, I+2);
583 /* Insert a call to subysp */
584 const char* Arg = MakeHexArg (Dec1);
585 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, N->LI);
586 CS_InsertEntry (S, X, I+2);
587 X = NewCodeEntry (OP65_JSR, AM65_ABS, "subysp", 0, N->LI);
588 CS_InsertEntry (S, X, I+3);
591 /* Delete the old code */
592 CS_DelEntries (S, I, 2);
594 /* Regenerate register info */
597 /* Remember we had changes */
608 /* Return the number of changes made */
614 /*****************************************************************************/
616 /*****************************************************************************/
620 typedef struct OptFunc OptFunc;
622 unsigned (*Func) (CodeSeg*); /* Optimizer function */
623 const char* Name; /* Name of the function/group */
624 unsigned CodeSizeFactor; /* Code size factor for this opt func */
625 unsigned long TotalRuns; /* Total number of runs */
626 unsigned long LastRuns; /* Last number of runs */
627 unsigned long TotalChanges; /* Total number of changes */
628 unsigned long LastChanges; /* Last number of changes */
629 char Disabled; /* True if function disabled */
634 /*****************************************************************************/
636 /*****************************************************************************/
640 /* A list of all the function descriptions */
641 static OptFunc DOpt65C02BitOps = { Opt65C02BitOps, "Opt65C02BitOps", 66, 0, 0, 0, 0, 0 };
642 static OptFunc DOpt65C02Ind = { Opt65C02Ind, "Opt65C02Ind", 100, 0, 0, 0, 0, 0 };
643 static OptFunc DOpt65C02Stores = { Opt65C02Stores, "Opt65C02Stores", 100, 0, 0, 0, 0, 0 };
644 static OptFunc DOptAdd1 = { OptAdd1, "OptAdd1", 125, 0, 0, 0, 0, 0 };
645 static OptFunc DOptAdd2 = { OptAdd2, "OptAdd2", 200, 0, 0, 0, 0, 0 };
646 static OptFunc DOptAdd3 = { OptAdd3, "OptAdd3", 65, 0, 0, 0, 0, 0 };
647 static OptFunc DOptAdd4 = { OptAdd4, "OptAdd4", 90, 0, 0, 0, 0, 0 };
648 static OptFunc DOptAdd5 = { OptAdd5, "OptAdd5", 100, 0, 0, 0, 0, 0 };
649 static OptFunc DOptAdd6 = { OptAdd6, "OptAdd6", 40, 0, 0, 0, 0, 0 };
650 static OptFunc DOptBNegA1 = { OptBNegA1, "OptBNegA1", 100, 0, 0, 0, 0, 0 };
651 static OptFunc DOptBNegA2 = { OptBNegA2, "OptBNegA2", 100, 0, 0, 0, 0, 0 };
652 static OptFunc DOptBNegAX1 = { OptBNegAX1, "OptBNegAX1", 100, 0, 0, 0, 0, 0 };
653 static OptFunc DOptBNegAX2 = { OptBNegAX2, "OptBNegAX2", 100, 0, 0, 0, 0, 0 };
654 static OptFunc DOptBNegAX3 = { OptBNegAX3, "OptBNegAX3", 100, 0, 0, 0, 0, 0 };
655 static OptFunc DOptBNegAX4 = { OptBNegAX4, "OptBNegAX4", 100, 0, 0, 0, 0, 0 };
656 static OptFunc DOptBoolTrans = { OptBoolTrans, "OptBoolTrans", 100, 0, 0, 0, 0, 0 };
657 static OptFunc DOptBranchDist = { OptBranchDist, "OptBranchDist", 0, 0, 0, 0, 0, 0 };
658 static OptFunc DOptCmp1 = { OptCmp1, "OptCmp1", 42, 0, 0, 0, 0, 0 };
659 static OptFunc DOptCmp2 = { OptCmp2, "OptCmp2", 85, 0, 0, 0, 0, 0 };
660 static OptFunc DOptCmp3 = { OptCmp3, "OptCmp3", 75, 0, 0, 0, 0, 0 };
661 static OptFunc DOptCmp4 = { OptCmp4, "OptCmp4", 75, 0, 0, 0, 0, 0 };
662 static OptFunc DOptCmp5 = { OptCmp5, "OptCmp5", 100, 0, 0, 0, 0, 0 };
663 static OptFunc DOptCmp6 = { OptCmp6, "OptCmp6", 100, 0, 0, 0, 0, 0 };
664 static OptFunc DOptCmp7 = { OptCmp7, "OptCmp7", 85, 0, 0, 0, 0, 0 };
665 static OptFunc DOptCmp8 = { OptCmp8, "OptCmp8", 50, 0, 0, 0, 0, 0 };
666 static OptFunc DOptCmp9 = { OptCmp9, "OptCmp9", 85, 0, 0, 0, 0, 0 };
667 static OptFunc DOptComplAX1 = { OptComplAX1, "OptComplAX1", 65, 0, 0, 0, 0, 0 };
668 static OptFunc DOptCondBranches1= { OptCondBranches1,"OptCondBranches1", 80, 0, 0, 0, 0, 0 };
669 static OptFunc DOptCondBranches2= { OptCondBranches2,"OptCondBranches2", 0, 0, 0, 0, 0, 0 };
670 static OptFunc DOptDeadCode = { OptDeadCode, "OptDeadCode", 100, 0, 0, 0, 0, 0 };
671 static OptFunc DOptDeadJumps = { OptDeadJumps, "OptDeadJumps", 100, 0, 0, 0, 0, 0 };
672 static OptFunc DOptDecouple = { OptDecouple, "OptDecouple", 100, 0, 0, 0, 0, 0 };
673 static OptFunc DOptDupLoads = { OptDupLoads, "OptDupLoads", 0, 0, 0, 0, 0, 0 };
674 static OptFunc DOptIndLoads1 = { OptIndLoads1, "OptIndLoads1", 0, 0, 0, 0, 0, 0 };
675 static OptFunc DOptIndLoads2 = { OptIndLoads2, "OptIndLoads2", 0, 0, 0, 0, 0, 0 };
676 static OptFunc DOptJumpCascades = { OptJumpCascades, "OptJumpCascades", 100, 0, 0, 0, 0, 0 };
677 static OptFunc DOptJumpTarget1 = { OptJumpTarget1, "OptJumpTarget1", 100, 0, 0, 0, 0, 0 };
678 static OptFunc DOptJumpTarget2 = { OptJumpTarget2, "OptJumpTarget2", 100, 0, 0, 0, 0, 0 };
679 static OptFunc DOptJumpTarget3 = { OptJumpTarget3, "OptJumpTarget3", 100, 0, 0, 0, 0, 0 };
680 static OptFunc DOptLoad1 = { OptLoad1, "OptLoad1", 100, 0, 0, 0, 0, 0 };
681 static OptFunc DOptLoad2 = { OptLoad2, "OptLoad2", 200, 0, 0, 0, 0, 0 };
682 static OptFunc DOptLoad3 = { OptLoad3, "OptLoad3", 0, 0, 0, 0, 0, 0 };
683 static OptFunc DOptNegAX1 = { OptNegAX1, "OptNegAX1", 165, 0, 0, 0, 0, 0 };
684 static OptFunc DOptNegAX2 = { OptNegAX2, "OptNegAX2", 200, 0, 0, 0, 0, 0 };
685 static OptFunc DOptRTS = { OptRTS, "OptRTS", 100, 0, 0, 0, 0, 0 };
686 static OptFunc DOptRTSJumps1 = { OptRTSJumps1, "OptRTSJumps1", 100, 0, 0, 0, 0, 0 };
687 static OptFunc DOptRTSJumps2 = { OptRTSJumps2, "OptRTSJumps2", 100, 0, 0, 0, 0, 0 };
688 static OptFunc DOptPrecalc = { OptPrecalc, "OptPrecalc", 100, 0, 0, 0, 0, 0 };
689 static OptFunc DOptPtrLoad1 = { OptPtrLoad1, "OptPtrLoad1", 100, 0, 0, 0, 0, 0 };
690 static OptFunc DOptPtrLoad2 = { OptPtrLoad2, "OptPtrLoad2", 100, 0, 0, 0, 0, 0 };
691 static OptFunc DOptPtrLoad3 = { OptPtrLoad3, "OptPtrLoad3", 100, 0, 0, 0, 0, 0 };
692 static OptFunc DOptPtrLoad4 = { OptPtrLoad4, "OptPtrLoad4", 100, 0, 0, 0, 0, 0 };
693 static OptFunc DOptPtrLoad5 = { OptPtrLoad5, "OptPtrLoad5", 50, 0, 0, 0, 0, 0 };
694 static OptFunc DOptPtrLoad6 = { OptPtrLoad6, "OptPtrLoad6", 60, 0, 0, 0, 0, 0 };
695 static OptFunc DOptPtrLoad7 = { OptPtrLoad7, "OptPtrLoad7", 140, 0, 0, 0, 0, 0 };
696 static OptFunc DOptPtrLoad11 = { OptPtrLoad11, "OptPtrLoad11", 92, 0, 0, 0, 0, 0 };
697 static OptFunc DOptPtrLoad12 = { OptPtrLoad12, "OptPtrLoad12", 50, 0, 0, 0, 0, 0 };
698 static OptFunc DOptPtrLoad13 = { OptPtrLoad13, "OptPtrLoad13", 65, 0, 0, 0, 0, 0 };
699 static OptFunc DOptPtrLoad14 = { OptPtrLoad14, "OptPtrLoad14", 108, 0, 0, 0, 0, 0 };
700 static OptFunc DOptPtrLoad15 = { OptPtrLoad15, "OptPtrLoad15", 86, 0, 0, 0, 0, 0 };
701 static OptFunc DOptPtrLoad16 = { OptPtrLoad16, "OptPtrLoad16", 100, 0, 0, 0, 0, 0 };
702 static OptFunc DOptPtrLoad17 = { OptPtrLoad17, "OptPtrLoad17", 190, 0, 0, 0, 0, 0 };
703 static OptFunc DOptPtrStore1 = { OptPtrStore1, "OptPtrStore1", 65, 0, 0, 0, 0, 0 };
704 static OptFunc DOptPtrStore2 = { OptPtrStore2, "OptPtrStore2", 65, 0, 0, 0, 0, 0 };
705 static OptFunc DOptPtrStore3 = { OptPtrStore3, "OptPtrStore3", 100, 0, 0, 0, 0, 0 };
706 static OptFunc DOptPush1 = { OptPush1, "OptPush1", 65, 0, 0, 0, 0, 0 };
707 static OptFunc DOptPush2 = { OptPush2, "OptPush2", 50, 0, 0, 0, 0, 0 };
708 static OptFunc DOptPushPop = { OptPushPop, "OptPushPop", 0, 0, 0, 0, 0, 0 };
709 static OptFunc DOptShift1 = { OptShift1, "OptShift1", 100, 0, 0, 0, 0, 0 };
710 static OptFunc DOptShift2 = { OptShift2, "OptShift2", 100, 0, 0, 0, 0, 0 };
711 static OptFunc DOptShift3 = { OptShift3, "OptShift3", 17, 0, 0, 0, 0, 0 };
712 static OptFunc DOptShift4 = { OptShift4, "OptShift4", 100, 0, 0, 0, 0, 0 };
713 static OptFunc DOptShift5 = { OptShift5, "OptShift5", 110, 0, 0, 0, 0, 0 };
714 static OptFunc DOptShift6 = { OptShift6, "OptShift6", 200, 0, 0, 0, 0, 0 };
715 static OptFunc DOptSize1 = { OptSize1, "OptSize1", 100, 0, 0, 0, 0, 0 };
716 static OptFunc DOptSize2 = { OptSize2, "OptSize2", 100, 0, 0, 0, 0, 0 };
717 static OptFunc DOptStackOps = { OptStackOps, "OptStackOps", 100, 0, 0, 0, 0, 0 };
718 static OptFunc DOptStackPtrOps = { OptStackPtrOps, "OptStackPtrOps", 50, 0, 0, 0, 0, 0 };
719 static OptFunc DOptStore1 = { OptStore1, "OptStore1", 70, 0, 0, 0, 0, 0 };
720 static OptFunc DOptStore2 = { OptStore2, "OptStore2", 115, 0, 0, 0, 0, 0 };
721 static OptFunc DOptStore3 = { OptStore3, "OptStore3", 120, 0, 0, 0, 0, 0 };
722 static OptFunc DOptStore4 = { OptStore4, "OptStore4", 50, 0, 0, 0, 0, 0 };
723 static OptFunc DOptStore5 = { OptStore5, "OptStore5", 100, 0, 0, 0, 0, 0 };
724 static OptFunc DOptStoreLoad = { OptStoreLoad, "OptStoreLoad", 0, 0, 0, 0, 0, 0 };
725 static OptFunc DOptSub1 = { OptSub1, "OptSub1", 100, 0, 0, 0, 0, 0 };
726 static OptFunc DOptSub2 = { OptSub2, "OptSub2", 100, 0, 0, 0, 0, 0 };
727 static OptFunc DOptSub3 = { OptSub3, "OptSub3", 100, 0, 0, 0, 0, 0 };
728 static OptFunc DOptTest1 = { OptTest1, "OptTest1", 65, 0, 0, 0, 0, 0 };
729 static OptFunc DOptTest2 = { OptTest2, "OptTest2", 50, 0, 0, 0, 0, 0 };
730 static OptFunc DOptTransfers1 = { OptTransfers1, "OptTransfers1", 0, 0, 0, 0, 0, 0 };
731 static OptFunc DOptTransfers2 = { OptTransfers2, "OptTransfers2", 60, 0, 0, 0, 0, 0 };
732 static OptFunc DOptTransfers3 = { OptTransfers3, "OptTransfers3", 65, 0, 0, 0, 0, 0 };
733 static OptFunc DOptTransfers4 = { OptTransfers4, "OptTransfers4", 65, 0, 0, 0, 0, 0 };
734 static OptFunc DOptUnusedLoads = { OptUnusedLoads, "OptUnusedLoads", 0, 0, 0, 0, 0, 0 };
735 static OptFunc DOptUnusedStores = { OptUnusedStores, "OptUnusedStores", 0, 0, 0, 0, 0, 0 };
738 /* Table containing all the steps in alphabetical order */
739 static OptFunc* OptFuncs[] = {
836 #define OPTFUNC_COUNT (sizeof(OptFuncs) / sizeof(OptFuncs[0]))
840 static int CmpOptStep (const void* Key, const void* Func)
841 /* Compare function for bsearch */
843 return strcmp (Key, (*(const OptFunc**)Func)->Name);
848 static OptFunc* FindOptFunc (const char* Name)
849 /* Find an optimizer step by name in the table and return a pointer. Return
850 * NULL if no such step is found.
853 /* Search for the function in the list */
854 OptFunc** O = bsearch (Name, OptFuncs, OPTFUNC_COUNT, sizeof (OptFuncs[0]), CmpOptStep);
860 static OptFunc* GetOptFunc (const char* Name)
861 /* Find an optimizer step by name in the table and return a pointer. Print an
862 * error and call AbEnd if not found.
865 /* Search for the function in the list */
866 OptFunc* F = FindOptFunc (Name);
869 AbEnd ("Optimization step `%s' not found", Name);
876 void DisableOpt (const char* Name)
877 /* Disable the optimization with the given name */
879 if (strcmp (Name, "any") == 0) {
881 for (I = 0; I < OPTFUNC_COUNT; ++I) {
882 OptFuncs[I]->Disabled = 1;
885 GetOptFunc(Name)->Disabled = 1;
891 void EnableOpt (const char* Name)
892 /* Enable the optimization with the given name */
894 if (strcmp (Name, "any") == 0) {
896 for (I = 0; I < OPTFUNC_COUNT; ++I) {
897 OptFuncs[I]->Disabled = 0;
900 GetOptFunc(Name)->Disabled = 0;
906 void ListOptSteps (FILE* F)
907 /* List all optimization steps */
910 for (I = 0; I < OPTFUNC_COUNT; ++I) {
911 fprintf (F, "%s\n", OptFuncs[I]->Name);
917 static void ReadOptStats (const char* Name)
918 /* Read the optimizer statistics file */
923 /* Try to open the file */
924 FILE* F = fopen (Name, "r");
926 /* Ignore the error */
930 /* Read and parse the lines */
932 while (fgets (Buf, sizeof (Buf), F) != 0) {
940 unsigned long TotalRuns;
941 unsigned long TotalChanges;
946 /* Remove trailing white space including the line terminator */
949 while (Len > 0 && IsSpace (B[Len-1])) {
954 /* Remove leading whitespace */
955 while (IsSpace (*B)) {
959 /* Check for empty and comment lines */
960 if (*B == '\0' || *B == ';' || *B == '#') {
965 if (sscanf (B, "%31s %lu %*u %lu %*u", Name, &TotalRuns, &TotalChanges) != 3) {
970 /* Search for the optimizer step. */
971 Func = FindOptFunc (Name);
977 /* Found the step, set the fields */
978 Func->TotalRuns = TotalRuns;
979 Func->TotalChanges = TotalChanges;
983 /* Close the file, ignore errors here. */
989 static void WriteOptStats (const char* Name)
990 /* Write the optimizer statistics file */
994 /* Try to open the file */
995 FILE* F = fopen (Name, "w");
997 /* Ignore the error */
1001 /* Write a header */
1003 "; Optimizer Total Last Total Last\n"
1004 "; Step Runs Runs Chg Chg\n");
1007 /* Write the data */
1008 for (I = 0; I < OPTFUNC_COUNT; ++I) {
1009 const OptFunc* O = OptFuncs[I];
1011 "%-20s %10lu %10lu %10lu %10lu\n",
1019 /* Close the file, ignore errors here. */
1025 static void OpenDebugFile (const CodeSeg* S)
1026 /* Open the debug file for the given segment if the flag is on */
1028 if (DebugOptOutput) {
1029 StrBuf Name = AUTO_STRBUF_INITIALIZER;
1031 SB_CopyStr (&Name, S->Func->Name);
1033 SB_CopyStr (&Name, "global");
1035 SB_AppendStr (&Name, ".opt");
1036 SB_Terminate (&Name);
1037 OpenDebugOutputFile (SB_GetConstBuf (&Name));
1044 static void WriteDebugOutput (CodeSeg* S, const char* Step)
1045 /* Write a separator line into the debug file if the flag is on */
1047 if (DebugOptOutput) {
1048 /* Output a separator */
1049 WriteOutput ("=========================================================================\n");
1051 /* Output a header line */
1053 /* Initial output */
1054 WriteOutput ("Initial code for function `%s':\n",
1055 S->Func? S->Func->Name : "<global>");
1057 WriteOutput ("Code after applying `%s':\n", Step);
1060 /* Output the code segment */
1067 static unsigned RunOptFunc (CodeSeg* S, OptFunc* F, unsigned Max)
1068 /* Run one optimizer function Max times or until there are no more changes */
1070 unsigned Changes, C;
1072 /* Don't run the function if it is disabled or if it is prohibited by the
1075 if (F->Disabled || F->CodeSizeFactor > S->CodeSizeFactor) {
1079 /* Run this until there are no more changes */
1083 /* Run the function */
1090 F->TotalChanges += C;
1091 F->LastChanges += C;
1093 /* If we had changes, output stuff and regenerate register info */
1096 printf ("Applied %s: %u changes\n", F->Name, C);
1098 WriteDebugOutput (S, F->Name);
1102 } while (--Max && C > 0);
1104 /* Return the number of changes */
1110 static unsigned RunOptGroup1 (CodeSeg* S)
1111 /* Run the first group of optimization steps. These steps translate known
1112 * patterns emitted by the code generator into more optimal patterns. Order
1113 * of the steps is important, because some of the steps done earlier cover
1114 * the same patterns as later steps as subpatterns.
1117 unsigned Changes = 0;
1119 Changes += RunOptFunc (S, &DOptStackPtrOps, 5);
1120 Changes += RunOptFunc (S, &DOptPtrStore1, 1);
1121 Changes += RunOptFunc (S, &DOptPtrStore2, 1);
1122 Changes += RunOptFunc (S, &DOptPtrStore3, 1);
1123 Changes += RunOptFunc (S, &DOptAdd3, 1); /* Before OptPtrLoad5! */
1124 Changes += RunOptFunc (S, &DOptPtrLoad1, 1);
1125 Changes += RunOptFunc (S, &DOptPtrLoad2, 1);
1126 Changes += RunOptFunc (S, &DOptPtrLoad3, 1);
1127 Changes += RunOptFunc (S, &DOptPtrLoad4, 1);
1128 Changes += RunOptFunc (S, &DOptPtrLoad5, 1);
1129 Changes += RunOptFunc (S, &DOptPtrLoad6, 1);
1130 Changes += RunOptFunc (S, &DOptPtrLoad7, 1);
1131 Changes += RunOptFunc (S, &DOptPtrLoad11, 1);
1132 Changes += RunOptFunc (S, &DOptPtrLoad12, 1);
1133 Changes += RunOptFunc (S, &DOptPtrLoad13, 1);
1134 Changes += RunOptFunc (S, &DOptPtrLoad14, 1);
1135 Changes += RunOptFunc (S, &DOptPtrLoad15, 1);
1136 Changes += RunOptFunc (S, &DOptPtrLoad16, 1);
1137 Changes += RunOptFunc (S, &DOptPtrLoad17, 1);
1138 Changes += RunOptFunc (S, &DOptBNegAX1, 1);
1139 Changes += RunOptFunc (S, &DOptBNegAX2, 1);
1140 Changes += RunOptFunc (S, &DOptBNegAX3, 1);
1141 Changes += RunOptFunc (S, &DOptBNegAX4, 1);
1142 Changes += RunOptFunc (S, &DOptAdd1, 1);
1143 Changes += RunOptFunc (S, &DOptAdd2, 1);
1144 Changes += RunOptFunc (S, &DOptAdd4, 1);
1145 Changes += RunOptFunc (S, &DOptAdd5, 1);
1146 Changes += RunOptFunc (S, &DOptAdd6, 1);
1147 Changes += RunOptFunc (S, &DOptSub1, 1);
1148 Changes += RunOptFunc (S, &DOptSub3, 1);
1149 Changes += RunOptFunc (S, &DOptStore4, 1);
1150 Changes += RunOptFunc (S, &DOptStore5, 1);
1151 Changes += RunOptFunc (S, &DOptShift1, 1);
1152 Changes += RunOptFunc (S, &DOptShift2, 1);
1153 Changes += RunOptFunc (S, &DOptShift5, 1);
1154 Changes += RunOptFunc (S, &DOptShift6, 1);
1155 Changes += RunOptFunc (S, &DOptStore1, 1);
1156 Changes += RunOptFunc (S, &DOptStore2, 5);
1157 Changes += RunOptFunc (S, &DOptStore3, 5);
1159 /* Return the number of changes */
1165 static unsigned RunOptGroup2 (CodeSeg* S)
1166 /* Run one group of optimization steps. This step involves just decoupling
1167 * instructions by replacing them by instructions that do not depend on
1168 * previous instructions. This makes it easier to find instructions that
1172 unsigned Changes = 0;
1174 Changes += RunOptFunc (S, &DOptDecouple, 1);
1176 /* Return the number of changes */
1182 static unsigned RunOptGroup3 (CodeSeg* S)
1183 /* Run one group of optimization steps. These steps depend on each other,
1184 * that means that one step may allow another step to do additional work,
1185 * so we will repeat the steps as long as we see any changes.
1188 unsigned Changes, C;
1194 C += RunOptFunc (S, &DOptBNegA1, 1);
1195 C += RunOptFunc (S, &DOptBNegA2, 1);
1196 C += RunOptFunc (S, &DOptNegAX1, 1);
1197 C += RunOptFunc (S, &DOptNegAX2, 1);
1198 C += RunOptFunc (S, &DOptStackOps, 3);
1199 C += RunOptFunc (S, &DOptShift1, 1);
1200 C += RunOptFunc (S, &DOptShift4, 1);
1201 C += RunOptFunc (S, &DOptComplAX1, 1);
1202 C += RunOptFunc (S, &DOptSub1, 1);
1203 C += RunOptFunc (S, &DOptSub2, 1);
1204 C += RunOptFunc (S, &DOptSub3, 1);
1205 C += RunOptFunc (S, &DOptAdd5, 1);
1206 C += RunOptFunc (S, &DOptAdd6, 1);
1207 C += RunOptFunc (S, &DOptJumpCascades, 1);
1208 C += RunOptFunc (S, &DOptDeadJumps, 1);
1209 C += RunOptFunc (S, &DOptRTS, 1);
1210 C += RunOptFunc (S, &DOptDeadCode, 1);
1211 C += RunOptFunc (S, &DOptBoolTrans, 1);
1212 C += RunOptFunc (S, &DOptJumpTarget1, 1);
1213 C += RunOptFunc (S, &DOptJumpTarget2, 1);
1214 C += RunOptFunc (S, &DOptCondBranches1, 1);
1215 C += RunOptFunc (S, &DOptCondBranches2, 1);
1216 C += RunOptFunc (S, &DOptRTSJumps1, 1);
1217 C += RunOptFunc (S, &DOptCmp1, 1);
1218 C += RunOptFunc (S, &DOptCmp2, 1);
1219 C += RunOptFunc (S, &DOptCmp8, 1); /* Must run before OptCmp3 */
1220 C += RunOptFunc (S, &DOptCmp3, 1);
1221 C += RunOptFunc (S, &DOptCmp4, 1);
1222 C += RunOptFunc (S, &DOptCmp5, 1);
1223 C += RunOptFunc (S, &DOptCmp6, 1);
1224 C += RunOptFunc (S, &DOptCmp7, 1);
1225 C += RunOptFunc (S, &DOptCmp9, 1);
1226 C += RunOptFunc (S, &DOptTest1, 1);
1227 C += RunOptFunc (S, &DOptLoad1, 1);
1228 C += RunOptFunc (S, &DOptJumpTarget3, 1); /* After OptCondBranches2 */
1229 C += RunOptFunc (S, &DOptUnusedLoads, 1);
1230 C += RunOptFunc (S, &DOptUnusedStores, 1);
1231 C += RunOptFunc (S, &DOptDupLoads, 1);
1232 C += RunOptFunc (S, &DOptStoreLoad, 1);
1233 C += RunOptFunc (S, &DOptTransfers1, 1);
1234 C += RunOptFunc (S, &DOptTransfers3, 1);
1235 C += RunOptFunc (S, &DOptTransfers4, 1);
1236 C += RunOptFunc (S, &DOptStore1, 1);
1237 C += RunOptFunc (S, &DOptStore5, 1);
1238 C += RunOptFunc (S, &DOptPushPop, 1);
1239 C += RunOptFunc (S, &DOptPrecalc, 1);
1245 /* Return the number of changes */
1251 static unsigned RunOptGroup4 (CodeSeg* S)
1252 /* Run another round of pattern replacements. These are done late, since there
1253 * may be better replacements before.
1256 unsigned Changes = 0;
1258 /* Repeat some of the steps here */
1259 Changes += RunOptFunc (S, &DOptShift3, 1);
1260 Changes += RunOptFunc (S, &DOptPush1, 1);
1261 Changes += RunOptFunc (S, &DOptPush2, 1);
1262 Changes += RunOptFunc (S, &DOptUnusedLoads, 1);
1263 Changes += RunOptFunc (S, &DOptTest2, 1);
1264 Changes += RunOptFunc (S, &DOptTransfers2, 1);
1265 Changes += RunOptFunc (S, &DOptLoad2, 1);
1266 Changes += RunOptFunc (S, &DOptLoad3, 1);
1267 Changes += RunOptFunc (S, &DOptDupLoads, 1);
1269 /* Return the number of changes */
1275 static unsigned RunOptGroup5 (CodeSeg* S)
1276 /* 65C02 specific optimizations. */
1278 unsigned Changes = 0;
1280 if (CPUIsets[CPU] & CPU_ISET_65SC02) {
1281 Changes += RunOptFunc (S, &DOpt65C02BitOps, 1);
1282 Changes += RunOptFunc (S, &DOpt65C02Ind, 1);
1283 Changes += RunOptFunc (S, &DOpt65C02Stores, 1);
1285 /* The 65C02 replacement codes do often make the use of a register
1286 * value unnecessary, so if we have changes, run another load
1289 Changes += RunOptFunc (S, &DOptUnusedLoads, 1);
1293 /* Return the number of changes */
1299 static unsigned RunOptGroup6 (CodeSeg* S)
1300 /* This one is quite special. It tries to replace "lda (sp),y" by "lda (sp,x)".
1301 * The latter is ony cycle slower, but if we're able to remove the necessary
1302 * load of the Y register, because X is zero anyway, we gain 1 cycle and
1303 * shorten the code by one (transfer) or two bytes (load). So what we do is
1304 * to replace the insns, remove unused loads, and then change back all insns
1305 * where Y is still zero (meaning that the load has not been removed).
1308 unsigned Changes = 0;
1310 /* This group will only run for a standard 6502, because the 65C02 has a
1311 * better addressing mode that covers this case.
1313 if ((CPUIsets[CPU] & CPU_ISET_65SC02) == 0) {
1314 Changes += RunOptFunc (S, &DOptIndLoads1, 1);
1315 Changes += RunOptFunc (S, &DOptUnusedLoads, 1);
1316 Changes += RunOptFunc (S, &DOptIndLoads2, 1);
1319 /* Return the number of changes */
1325 static unsigned RunOptGroup7 (CodeSeg* S)
1326 /* The last group of optimization steps. Adjust branches, do size optimizations.
1329 unsigned Changes = 0;
1332 /* Optimize for size, that is replace operations by shorter ones, even
1333 * if this does hinder further optimizations (no problem since we're
1336 C = RunOptFunc (S, &DOptSize1, 1);
1339 /* Run some optimization passes again, since the size optimizations
1340 * may have opened new oportunities.
1342 Changes += RunOptFunc (S, &DOptUnusedLoads, 1);
1343 Changes += RunOptFunc (S, &DOptUnusedStores, 1);
1344 Changes += RunOptFunc (S, &DOptJumpTarget1, 5);
1345 Changes += RunOptFunc (S, &DOptStore5, 1);
1348 C = RunOptFunc (S, &DOptSize2, 1);
1351 /* Run some optimization passes again, since the size optimizations
1352 * may have opened new oportunities.
1354 Changes += RunOptFunc (S, &DOptUnusedLoads, 1);
1355 Changes += RunOptFunc (S, &DOptJumpTarget1, 5);
1356 Changes += RunOptFunc (S, &DOptStore5, 1);
1357 Changes += RunOptFunc (S, &DOptTransfers1, 1);
1358 Changes += RunOptFunc (S, &DOptTransfers3, 1);
1361 /* Adjust branch distances */
1362 Changes += RunOptFunc (S, &DOptBranchDist, 3);
1364 /* Replace conditional branches to RTS. If we had changes, we must run dead
1365 * code elimination again, since the change may have introduced dead code.
1367 C = RunOptFunc (S, &DOptRTSJumps2, 1);
1370 Changes += RunOptFunc (S, &DOptDeadCode, 1);
1373 /* Return the number of changes */
1379 void RunOpt (CodeSeg* S)
1380 /* Run the optimizer */
1382 const char* StatFileName;
1384 /* If we shouldn't run the optimizer, bail out */
1389 /* Check if we are requested to write optimizer statistics */
1390 StatFileName = getenv ("CC65_OPTSTATS");
1392 ReadOptStats (StatFileName);
1395 /* Print the name of the function we are working on */
1397 Print (stdout, 1, "Running optimizer for function `%s'\n", S->Func->Name);
1399 Print (stdout, 1, "Running optimizer for global code segment\n");
1402 /* If requested, open an output file */
1404 WriteDebugOutput (S, 0);
1406 /* Generate register info for all instructions */
1409 /* Run groups of optimizations */
1418 /* Free register info */
1421 /* Close output file if necessary */
1422 if (DebugOptOutput) {
1426 /* Write statistics */
1428 WriteOptStats (StatFileName);