1 /*****************************************************************************/
5 /* Optimizer subroutines */
9 /* (C) 2001-2003 Ullrich von Bassewitz */
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 /*****************************************************************************/
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_KnownImm (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_KnownImm (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 */
555 CS_DelEntry (S, I); /* jsr pushax */
557 /* Create and insert the stores into the zp pointer if needed */
559 X = NewCodeEntry (OP65_STA, AM65_ZP, "ptr1", 0, L[0]->LI);
560 CS_InsertEntry (S, X, I);
561 X = NewCodeEntry (OP65_STX, AM65_ZP, "ptr1+1", 0, L[0]->LI);
562 CS_InsertEntry (S, X, I+1);
565 /* Remember, we had changes */
575 /* Return the number of changes made */
581 static unsigned OptPtrStore2 (CodeSeg* S)
582 /* Search for the sequence:
604 unsigned Changes = 0;
606 /* Walk over the entries */
608 while (I < CS_GetEntryCount (S)) {
614 L[0] = CS_GetEntry (S, I);
616 /* Check for the sequence */
617 if (L[0]->OPC == OP65_LDA &&
618 L[0]->AM == AM65_IMM &&
619 CS_GetEntries (S, L+1, I+1, 10) &&
620 L[1]->OPC == OP65_LDX &&
621 L[1]->AM == AM65_IMM &&
622 L[2]->OPC == OP65_CLC &&
623 L[3]->OPC == OP65_ADC &&
624 (L[3]->AM == AM65_ABS || L[3]->AM == AM65_ZP) &&
625 (L[4]->OPC == OP65_BCC || L[4]->OPC == OP65_JCC) &&
627 L[4]->JumpTo->Owner == L[6] &&
628 L[5]->OPC == OP65_INX &&
629 CE_IsCallTo (L[6], "pushax") &&
630 L[7]->OPC == OP65_LDX &&
631 L[8]->OPC == OP65_LDA &&
632 L[9]->OPC == OP65_LDY &&
633 CE_KnownImm (L[9]) &&
635 CE_IsCallTo (L[10], "staspidx") &&
636 !CS_RangeHasLabel (S, I+1, 5) &&
637 !CS_RangeHasLabel (S, I+7, 4) &&
638 /* Check the label last because this is quite costly */
639 (Len = strlen (L[0]->Arg)) > 3 &&
640 L[0]->Arg[0] == '<' &&
641 L[0]->Arg[1] == '(' &&
642 strlen (L[1]->Arg) == Len &&
643 L[1]->Arg[0] == '>' &&
644 memcmp (L[0]->Arg+1, L[1]->Arg+1, Len-1) == 0) {
649 /* We will create all the new stuff behind the current one so
650 * we keep the line references.
652 X = NewCodeEntry (OP65_LDY, L[3]->AM, L[3]->Arg, 0, L[0]->LI);
653 CS_InsertEntry (S, X, I+11);
655 X = NewCodeEntry (OP65_LDX, L[7]->AM, L[7]->Arg, 0, L[7]->LI);
656 CS_InsertEntry (S, X, I+12);
658 X = NewCodeEntry (OP65_LDA, L[8]->AM, L[8]->Arg, 0, L[8]->LI);
659 CS_InsertEntry (S, X, I+13);
661 Label = memcpy (xmalloc (Len-2), L[0]->Arg+2, Len-3);
663 X = NewCodeEntry (OP65_STA, AM65_ABSY, Label, 0, L[10]->LI);
664 CS_InsertEntry (S, X, I+14);
667 /* Remove the old code */
668 CS_DelEntries (S, I, 11);
670 /* Remember, we had changes */
680 /* Return the number of changes made */
686 /*****************************************************************************/
687 /* Optimize loads through pointers */
688 /*****************************************************************************/
692 static unsigned OptPtrLoad1 (CodeSeg* S)
693 /* Search for the sequence:
718 unsigned Changes = 0;
720 /* Walk over the entries */
722 while (I < CS_GetEntryCount (S)) {
727 L[0] = CS_GetEntry (S, I);
729 /* Check for the sequence */
730 if (L[0]->OPC == OP65_CLC &&
731 CS_GetEntries (S, L+1, I+1, 8) &&
732 L[1]->OPC == OP65_ADC &&
733 L[2]->OPC == OP65_TAY &&
734 L[3]->OPC == OP65_TXA &&
735 L[4]->OPC == OP65_ADC &&
736 L[5]->OPC == OP65_TAX &&
737 L[6]->OPC == OP65_TYA &&
738 L[7]->OPC == OP65_LDY &&
739 CE_IsCallTo (L[8], "ldauidx") &&
740 !CS_RangeHasLabel (S, I+1, 8)) {
745 /* Track the insertion point */
749 X = NewCodeEntry (OP65_STA, AM65_ZP, "ptr1", 0, L[2]->LI);
750 CS_InsertEntry (S, X, IP++);
752 /* If the instruction before the clc is a ldx, replace the
753 * txa by an lda with the same location of the ldx. Otherwise
754 * transfer the value in X to A.
756 if ((P = CS_GetPrevEntry (S, I)) != 0 &&
757 P->OPC == OP65_LDX &&
759 X = NewCodeEntry (OP65_LDA, P->AM, P->Arg, 0, P->LI);
761 X = NewCodeEntry (OP65_TXA, AM65_IMP, 0, 0, L[3]->LI);
763 CS_InsertEntry (S, X, IP++);
766 X = NewCodeEntry (OP65_ADC, L[4]->AM, L[4]->Arg, 0, L[4]->LI);
767 CS_InsertEntry (S, X, IP++);
770 X = NewCodeEntry (OP65_STA, AM65_ZP, "ptr1+1", 0, L[5]->LI);
771 CS_InsertEntry (S, X, IP++);
774 X = NewCodeEntry (OP65_LDY, L[7]->AM, L[7]->Arg, 0, L[7]->LI);
775 CS_InsertEntry (S, X, IP++);
778 X = NewCodeEntry (OP65_LDX, AM65_IMM, "$00", 0, L[8]->LI);
779 CS_InsertEntry (S, X, IP++);
782 X = NewCodeEntry (OP65_LDA, AM65_ZP_INDY, "ptr1", 0, L[8]->LI);
783 CS_InsertEntry (S, X, IP++);
785 /* Remove the old instructions */
786 CS_DelEntries (S, IP, 7);
788 /* Remember, we had changes */
798 /* Return the number of changes made */
804 static unsigned OptPtrLoad2 (CodeSeg* S)
805 /* Search for the sequence:
830 unsigned Changes = 0;
832 /* Walk over the entries */
834 while (I < CS_GetEntryCount (S)) {
839 L[0] = CS_GetEntry (S, I);
841 /* Check for the sequence */
842 if (L[0]->OPC == OP65_ADC &&
843 CS_GetEntries (S, L+1, I+1, 8) &&
844 L[1]->OPC == OP65_PHA &&
845 L[2]->OPC == OP65_TXA &&
846 L[3]->OPC == OP65_INY &&
847 L[4]->OPC == OP65_ADC &&
848 L[5]->OPC == OP65_TAX &&
849 L[6]->OPC == OP65_PLA &&
850 L[7]->OPC == OP65_LDY &&
851 CE_IsCallTo (L[8], "ldauidx") &&
852 !CS_RangeHasLabel (S, I+1, 8)) {
856 /* Store the low byte and remove the PHA instead */
857 X = NewCodeEntry (OP65_STA, AM65_ZP, "ptr1", 0, L[0]->LI);
858 CS_InsertEntry (S, X, I+1);
860 /* Store the high byte */
861 X = NewCodeEntry (OP65_STA, AM65_ZP, "ptr1+1", 0, L[4]->LI);
862 CS_InsertEntry (S, X, I+6);
864 /* Load high and low byte */
865 X = NewCodeEntry (OP65_LDX, AM65_IMM, "$00", 0, L[6]->LI);
866 CS_InsertEntry (S, X, I+10);
867 X = NewCodeEntry (OP65_LDA, AM65_ZP_INDY, "ptr1", 0, L[6]->LI);
868 CS_InsertEntry (S, X, I+11);
870 /* Delete the old code */
871 CS_DelEntry (S, I+12); /* jsr ldauidx */
872 CS_DelEntry (S, I+8); /* pla */
873 CS_DelEntry (S, I+7); /* tax */
874 CS_DelEntry (S, I+2); /* pha */
876 /* Remember, we had changes */
886 /* Return the number of changes made */
892 static unsigned OptPtrLoad3 (CodeSeg* S)
893 /* Search for the sequence:
911 unsigned Changes = 0;
913 /* Walk over the entries */
915 while (I < CS_GetEntryCount (S)) {
921 L[0] = CS_GetEntry (S, I);
923 /* Check for the sequence */
924 if (L[0]->OPC == OP65_LDA &&
925 L[0]->AM == AM65_IMM &&
926 CS_GetEntries (S, L+1, I+1, 7) &&
927 L[1]->OPC == OP65_LDX &&
928 L[1]->AM == AM65_IMM &&
929 L[2]->OPC == OP65_CLC &&
930 L[3]->OPC == OP65_ADC &&
931 (L[3]->AM == AM65_ABS || L[3]->AM == AM65_ZP) &&
932 (L[4]->OPC == OP65_BCC || L[4]->OPC == OP65_JCC) &&
934 L[4]->JumpTo->Owner == L[6] &&
935 L[5]->OPC == OP65_INX &&
936 L[6]->OPC == OP65_LDY &&
937 CE_KnownImm (L[6]) &&
939 CE_IsCallTo (L[7], "ldauidx") &&
940 !CS_RangeHasLabel (S, I+1, 5) &&
941 !CE_HasLabel (L[7]) &&
942 /* Check the label last because this is quite costly */
943 (Len = strlen (L[0]->Arg)) > 3 &&
944 L[0]->Arg[0] == '<' &&
945 L[0]->Arg[1] == '(' &&
946 strlen (L[1]->Arg) == Len &&
947 L[1]->Arg[0] == '>' &&
948 memcmp (L[0]->Arg+1, L[1]->Arg+1, Len-1) == 0) {
953 /* We will create all the new stuff behind the current one so
954 * we keep the line references.
956 X = NewCodeEntry (OP65_LDY, L[3]->AM, L[3]->Arg, 0, L[0]->LI);
957 CS_InsertEntry (S, X, I+8);
959 X = NewCodeEntry (OP65_LDX, AM65_IMM, "$00", 0, L[0]->LI);
960 CS_InsertEntry (S, X, I+9);
962 Label = memcpy (xmalloc (Len-2), L[0]->Arg+2, Len-3);
964 X = NewCodeEntry (OP65_LDA, AM65_ABSY, Label, 0, L[0]->LI);
965 CS_InsertEntry (S, X, I+10);
968 /* Remove the old code */
969 CS_DelEntries (S, I, 8);
971 /* Remember, we had changes */
981 /* Return the number of changes made */
987 static unsigned OptPtrLoad4 (CodeSeg* S)
988 /* Search for the sequence:
1000 * and replace it by:
1009 unsigned Changes = 0;
1011 /* Walk over the entries */
1013 while (I < CS_GetEntryCount (S)) {
1018 /* Get next entry */
1019 L[0] = CS_GetEntry (S, I);
1021 /* Check for the sequence */
1022 if (L[0]->OPC == OP65_LDA &&
1023 L[0]->AM == AM65_IMM &&
1024 CS_GetEntries (S, L+1, I+1, 8) &&
1025 L[1]->OPC == OP65_LDX &&
1026 L[1]->AM == AM65_IMM &&
1027 !CE_HasLabel (L[1]) &&
1028 L[2]->OPC == OP65_LDY &&
1029 CE_KnownImm (L[2]) &&
1030 !CE_HasLabel (L[2]) &&
1031 L[3]->OPC == OP65_CLC &&
1032 !CE_HasLabel (L[3]) &&
1033 L[4]->OPC == OP65_ADC &&
1034 L[4]->AM == AM65_ZP_INDY &&
1035 !CE_HasLabel (L[4]) &&
1036 (L[5]->OPC == OP65_BCC || L[5]->OPC == OP65_JCC) &&
1037 L[5]->JumpTo != 0 &&
1038 L[5]->JumpTo->Owner == L[7] &&
1039 !CE_HasLabel (L[5]) &&
1040 L[6]->OPC == OP65_INX &&
1041 !CE_HasLabel (L[6]) &&
1042 L[7]->OPC == OP65_LDY &&
1043 CE_KnownImm (L[7]) &&
1045 CE_IsCallTo (L[8], "ldauidx") &&
1046 !CE_HasLabel (L[8]) &&
1047 /* Check the label last because this is quite costly */
1048 (Len = strlen (L[0]->Arg)) > 3 &&
1049 L[0]->Arg[0] == '<' &&
1050 L[0]->Arg[1] == '(' &&
1051 strlen (L[1]->Arg) == Len &&
1052 L[1]->Arg[0] == '>' &&
1053 memcmp (L[0]->Arg+1, L[1]->Arg+1, Len-1) == 0) {
1059 X = NewCodeEntry (OP65_LDA, AM65_ZP_INDY, L[4]->Arg, 0, L[0]->LI);
1060 CS_InsertEntry (S, X, I+3);
1063 X = NewCodeEntry (OP65_TAY, AM65_IMP, 0, 0, L[0]->LI);
1064 CS_InsertEntry (S, X, I+4);
1067 X = NewCodeEntry (OP65_LDX, AM65_IMM, "$00", 0, L[0]->LI);
1068 CS_InsertEntry (S, X, I+5);
1071 Label = memcpy (xmalloc (Len-2), L[0]->Arg+2, Len-3);
1072 Label[Len-3] = '\0';
1073 X = NewCodeEntry (OP65_LDA, AM65_ABSY, Label, 0, L[0]->LI);
1074 CS_InsertEntry (S, X, I+6);
1077 /* Remove the old code */
1078 CS_DelEntries (S, I, 2);
1079 CS_DelEntries (S, I+5, 6);
1081 /* Remember, we had changes */
1091 /* Return the number of changes made */
1097 static unsigned OptPtrLoad5 (CodeSeg* S)
1098 /* Search for the sequence:
1105 * and replace it by:
1112 unsigned Changes = 0;
1114 /* Walk over the entries */
1116 while (I < CS_GetEntryCount (S)) {
1121 /* Get next entry */
1122 L[0] = CS_GetEntry (S, I);
1124 /* Check for the sequence */
1125 if (L[0]->OPC == OP65_LDA && L[0]->AM == AM65_ZP &&
1126 CS_GetEntries (S, L+1, I+1, 3) &&
1127 !CS_RangeHasLabel (S, I+1, 3) &&
1128 L[1]->OPC == OP65_LDX && L[1]->AM == AM65_ZP &&
1129 (Len = strlen (L[0]->Arg)) > 0 &&
1130 strncmp (L[0]->Arg, L[1]->Arg, Len) == 0 &&
1131 strcmp (L[1]->Arg + Len, "+1") == 0 &&
1132 L[2]->OPC == OP65_LDY &&
1133 CE_IsCallTo (L[3], "ldauidx")) {
1138 X = NewCodeEntry (OP65_LDX, AM65_IMM, "$00", 0, L[3]->LI);
1139 CS_InsertEntry (S, X, I+3);
1142 X = NewCodeEntry (OP65_LDA, AM65_ZP_INDY, L[0]->Arg, 0, L[3]->LI);
1143 CS_InsertEntry (S, X, I+4);
1145 /* Remove the old code */
1146 CS_DelEntry (S, I+5);
1147 CS_DelEntries (S, I, 2);
1149 /* Remember, we had changes */
1159 /* Return the number of changes made */
1165 static unsigned OptPtrLoad6 (CodeSeg* S)
1166 /* Search for the sequence:
1173 * and replace it by:
1182 unsigned Changes = 0;
1184 /* Walk over the entries */
1186 while (I < CS_GetEntryCount (S)) {
1191 /* Get next entry */
1192 L[0] = CS_GetEntry (S, I);
1194 /* Check for the sequence */
1195 if (L[0]->OPC == OP65_LDA && L[0]->AM == AM65_ZP &&
1196 CS_GetEntries (S, L+1, I+1, 3) &&
1197 !CS_RangeHasLabel (S, I+1, 3) &&
1198 L[1]->OPC == OP65_LDX && L[1]->AM == AM65_ZP &&
1199 (Len = strlen (L[0]->Arg)) > 0 &&
1200 strncmp (L[0]->Arg, L[1]->Arg, Len) == 0 &&
1201 strcmp (L[1]->Arg + Len, "+1") == 0 &&
1202 L[2]->OPC == OP65_LDY &&
1203 CE_IsCallTo (L[3], "ldaxidx")) {
1208 X = NewCodeEntry (OP65_LDA, AM65_ZP_INDY, L[0]->Arg, 0, L[3]->LI);
1209 CS_InsertEntry (S, X, I+4);
1212 X = NewCodeEntry (OP65_TAX, AM65_IMP, 0, 0, L[3]->LI);
1213 CS_InsertEntry (S, X, I+5);
1216 X = NewCodeEntry (OP65_DEY, AM65_IMP, 0, 0, L[3]->LI);
1217 CS_InsertEntry (S, X, I+6);
1220 X = NewCodeEntry (OP65_LDA, AM65_ZP_INDY, L[0]->Arg, 0, L[3]->LI);
1221 CS_InsertEntry (S, X, I+7);
1223 /* Remove the old code */
1224 CS_DelEntry (S, I+3);
1225 CS_DelEntries (S, I, 2);
1227 /* Remember, we had changes */
1237 /* Return the number of changes made */
1243 static unsigned OptPtrLoad7 (CodeSeg* S)
1244 /* Search for the sequence
1249 * and replace it by:
1257 * This step must be execute *after* OptPtrLoad1!
1260 unsigned Changes = 0;
1262 /* Walk over the entries */
1264 while (I < CS_GetEntryCount (S)) {
1268 /* Get next entry */
1269 L[0] = CS_GetEntry (S, I);
1271 /* Check for the sequence */
1272 if (L[0]->OPC == OP65_LDY &&
1273 CS_GetEntries (S, L+1, I+1, 1) &&
1274 CE_IsCallTo (L[1], "ldauidx") &&
1275 !CE_HasLabel (L[1])) {
1279 /* Store the high byte */
1280 X = NewCodeEntry (OP65_STA, AM65_ZP, "ptr1", 0, L[0]->LI);
1281 CS_InsertEntry (S, X, I+1);
1283 /* Store the low byte */
1284 X = NewCodeEntry (OP65_STX, AM65_ZP, "ptr1+1", 0, L[0]->LI);
1285 CS_InsertEntry (S, X, I+2);
1287 /* Delete the call to ldauidx */
1288 CS_DelEntry (S, I+3);
1290 /* Load the high and low byte */
1291 X = NewCodeEntry (OP65_LDX, AM65_IMM, "$00", 0, L[0]->LI);
1292 CS_InsertEntry (S, X, I+3);
1293 X = NewCodeEntry (OP65_LDA, AM65_ZP_INDY, "ptr1", 0, L[0]->LI);
1294 CS_InsertEntry (S, X, I+4);
1296 /* Remember, we had changes */
1306 /* Return the number of changes made */
1312 /*****************************************************************************/
1313 /* Decouple operations */
1314 /*****************************************************************************/
1318 static unsigned OptDecouple (CodeSeg* S)
1319 /* Decouple operations, that is, do the following replacements:
1329 * lda zp -> lda #imm
1330 * ldx zp -> ldx #imm
1331 * ldy zp -> ldy #imm
1333 * Provided that the register values are known of course.
1336 unsigned Changes = 0;
1339 /* Generate register info for the following step */
1342 /* Walk over the entries */
1344 while (I < CS_GetEntryCount (S)) {
1348 /* Get next entry and it's input register values */
1349 CodeEntry* E = CS_GetEntry (S, I);
1350 const RegContents* In = &E->RI->In;
1352 /* Assume we have no replacement */
1355 /* Check the instruction */
1359 if (RegValIsKnown (In->RegA)) {
1360 Arg = MakeHexArg ((In->RegA - 1) & 0xFF);
1361 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
1366 if (RegValIsKnown (In->RegX)) {
1367 Arg = MakeHexArg ((In->RegX - 1) & 0xFF);
1368 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
1373 if (RegValIsKnown (In->RegY)) {
1374 Arg = MakeHexArg ((In->RegY - 1) & 0xFF);
1375 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
1380 if (RegValIsKnown (In->RegA)) {
1381 Arg = MakeHexArg ((In->RegA + 1) & 0xFF);
1382 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
1387 if (RegValIsKnown (In->RegX)) {
1388 Arg = MakeHexArg ((In->RegX + 1) & 0xFF);
1389 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
1394 if (RegValIsKnown (In->RegY)) {
1395 Arg = MakeHexArg ((In->RegY + 1) & 0xFF);
1396 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
1401 if (E->AM == AM65_ZP) {
1402 switch (GetKnownReg (E->Use & REG_ZP, In)) {
1404 Arg = MakeHexArg (In->Tmp1);
1405 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
1409 Arg = MakeHexArg (In->Ptr1Lo);
1410 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
1414 Arg = MakeHexArg (In->Ptr1Hi);
1415 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
1419 Arg = MakeHexArg (In->SRegLo);
1420 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
1424 Arg = MakeHexArg (In->SRegHi);
1425 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
1432 if (E->AM == AM65_ZP) {
1433 switch (GetKnownReg (E->Use & REG_ZP, In)) {
1435 Arg = MakeHexArg (In->Tmp1);
1436 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
1440 Arg = MakeHexArg (In->Ptr1Lo);
1441 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
1445 Arg = MakeHexArg (In->Ptr1Hi);
1446 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
1450 Arg = MakeHexArg (In->SRegLo);
1451 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
1455 Arg = MakeHexArg (In->SRegHi);
1456 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
1463 if (E->AM == AM65_ZP) {
1464 switch (GetKnownReg (E->Use, In)) {
1466 Arg = MakeHexArg (In->Tmp1);
1467 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
1471 Arg = MakeHexArg (In->Ptr1Lo);
1472 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
1476 Arg = MakeHexArg (In->Ptr1Hi);
1477 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
1481 Arg = MakeHexArg (In->SRegLo);
1482 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
1486 Arg = MakeHexArg (In->SRegHi);
1487 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
1494 if (E->RI->In.RegA >= 0) {
1495 Arg = MakeHexArg (In->RegA);
1496 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
1501 if (E->RI->In.RegA >= 0) {
1502 Arg = MakeHexArg (In->RegA);
1503 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
1508 if (E->RI->In.RegX >= 0) {
1509 Arg = MakeHexArg (In->RegX);
1510 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
1515 if (E->RI->In.RegY >= 0) {
1516 Arg = MakeHexArg (In->RegY);
1517 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
1522 /* Avoid gcc warnings */
1527 /* Insert the replacement if we have one */
1529 CS_InsertEntry (S, X, I+1);
1539 /* Free register info */
1542 /* Return the number of changes made */
1548 /*****************************************************************************/
1549 /* struct OptFunc */
1550 /*****************************************************************************/
1554 typedef struct OptFunc OptFunc;
1556 unsigned (*Func) (CodeSeg*); /* Optimizer function */
1557 const char* Name; /* Name of the function/group */
1558 unsigned CodeSizeFactor; /* Code size factor for this opt func */
1559 unsigned long TotalRuns; /* Total number of runs */
1560 unsigned long LastRuns; /* Last number of runs */
1561 unsigned long TotalChanges; /* Total number of changes */
1562 unsigned long LastChanges; /* Last number of changes */
1563 char Disabled; /* True if function disabled */
1568 /*****************************************************************************/
1570 /*****************************************************************************/
1574 /* A list of all the function descriptions */
1575 static OptFunc DOpt65C02BitOps = { Opt65C02BitOps, "Opt65C02BitOps", 66, 0, 0, 0, 0, 0 };
1576 static OptFunc DOpt65C02Ind = { Opt65C02Ind, "Opt65C02Ind", 100, 0, 0, 0, 0, 0 };
1577 static OptFunc DOpt65C02Stores = { Opt65C02Stores, "Opt65C02Stores", 100, 0, 0, 0, 0, 0 };
1578 static OptFunc DOptAdd1 = { OptAdd1, "OptAdd1", 125, 0, 0, 0, 0, 0 };
1579 static OptFunc DOptAdd2 = { OptAdd2, "OptAdd2", 200, 0, 0, 0, 0, 0 };
1580 static OptFunc DOptAdd3 = { OptAdd3, "OptAdd3", 90, 0, 0, 0, 0, 0 };
1581 static OptFunc DOptAdd4 = { OptAdd4, "OptAdd4", 100, 0, 0, 0, 0, 0 };
1582 static OptFunc DOptAdd5 = { OptAdd5, "OptAdd5", 40, 0, 0, 0, 0, 0 };
1583 static OptFunc DOptBoolTrans = { OptBoolTrans, "OptBoolTrans", 100, 0, 0, 0, 0, 0 };
1584 static OptFunc DOptBranchDist = { OptBranchDist, "OptBranchDist", 0, 0, 0, 0, 0, 0 };
1585 static OptFunc DOptCmp1 = { OptCmp1, "OptCmp1", 85, 0, 0, 0, 0, 0 };
1586 static OptFunc DOptCmp2 = { OptCmp2, "OptCmp2", 75, 0, 0, 0, 0, 0 };
1587 static OptFunc DOptCmp3 = { OptCmp3, "OptCmp3", 75, 0, 0, 0, 0, 0 };
1588 static OptFunc DOptCmp4 = { OptCmp4, "OptCmp4", 100, 0, 0, 0, 0, 0 };
1589 static OptFunc DOptCmp5 = { OptCmp5, "OptCmp5", 100, 0, 0, 0, 0, 0 };
1590 static OptFunc DOptCmp6 = { OptCmp6, "OptCmp6", 85, 0, 0, 0, 0, 0 };
1591 static OptFunc DOptCmp7 = { OptCmp7, "OptCmp7", 50, 0, 0, 0, 0, 0 };
1592 static OptFunc DOptCondBranches = { OptCondBranches, "OptCondBranches", 80, 0, 0, 0, 0, 0 };
1593 static OptFunc DOptDeadCode = { OptDeadCode, "OptDeadCode", 100, 0, 0, 0, 0, 0 };
1594 static OptFunc DOptDeadJumps = { OptDeadJumps, "OptDeadJumps", 100, 0, 0, 0, 0, 0 };
1595 static OptFunc DOptDecouple = { OptDecouple, "OptDecouple", 100, 0, 0, 0, 0, 0 };
1596 static OptFunc DOptDupLoads = { OptDupLoads, "OptDupLoads", 0, 0, 0, 0, 0, 0 };
1597 static OptFunc DOptJumpCascades = { OptJumpCascades, "OptJumpCascades", 100, 0, 0, 0, 0, 0 };
1598 static OptFunc DOptJumpTarget = { OptJumpTarget, "OptJumpTarget", 100, 0, 0, 0, 0, 0 };
1599 static OptFunc DOptLoad1 = { OptLoad1, "OptLoad1", 100, 0, 0, 0, 0, 0 };
1600 static OptFunc DOptRTS = { OptRTS, "OptRTS", 100, 0, 0, 0, 0, 0 };
1601 static OptFunc DOptRTSJumps1 = { OptRTSJumps1, "OptRTSJumps1", 100, 0, 0, 0, 0, 0 };
1602 static OptFunc DOptRTSJumps2 = { OptRTSJumps2, "OptRTSJumps2", 100, 0, 0, 0, 0, 0 };
1603 static OptFunc DOptNegA1 = { OptNegA1, "OptNegA1", 100, 0, 0, 0, 0, 0 };
1604 static OptFunc DOptNegA2 = { OptNegA2, "OptNegA2", 100, 0, 0, 0, 0, 0 };
1605 static OptFunc DOptNegAX1 = { OptNegAX1, "OptNegAX1", 100, 0, 0, 0, 0, 0 };
1606 static OptFunc DOptNegAX2 = { OptNegAX2, "OptNegAX2", 100, 0, 0, 0, 0, 0 };
1607 static OptFunc DOptNegAX3 = { OptNegAX3, "OptNegAX3", 100, 0, 0, 0, 0, 0 };
1608 static OptFunc DOptNegAX4 = { OptNegAX4, "OptNegAX4", 100, 0, 0, 0, 0, 0 };
1609 static OptFunc DOptPrecalc = { OptPrecalc, "OptPrecalc", 100, 0, 0, 0, 0, 0 };
1610 static OptFunc DOptPtrLoad1 = { OptPtrLoad1, "OptPtrLoad1", 100, 0, 0, 0, 0, 0 };
1611 static OptFunc DOptPtrLoad2 = { OptPtrLoad2, "OptPtrLoad2", 100, 0, 0, 0, 0, 0 };
1612 static OptFunc DOptPtrLoad3 = { OptPtrLoad3, "OptPtrLoad3", 100, 0, 0, 0, 0, 0 };
1613 static OptFunc DOptPtrLoad4 = { OptPtrLoad4, "OptPtrLoad4", 100, 0, 0, 0, 0, 0 };
1614 static OptFunc DOptPtrLoad5 = { OptPtrLoad5, "OptPtrLoad5", 65, 0, 0, 0, 0, 0 };
1615 static OptFunc DOptPtrLoad6 = { OptPtrLoad6, "OptPtrLoad6", 86, 0, 0, 0, 0, 0 };
1616 static OptFunc DOptPtrLoad7 = { OptPtrLoad7, "OptPtrLoad7", 100, 0, 0, 0, 0, 0 };
1617 static OptFunc DOptPtrStore1 = { OptPtrStore1, "OptPtrStore1", 100, 0, 0, 0, 0, 0 };
1618 static OptFunc DOptPtrStore2 = { OptPtrStore2, "OptPtrStore2", 40, 0, 0, 0, 0, 0 };
1619 static OptFunc DOptPush1 = { OptPush1, "OptPush1", 65, 0, 0, 0, 0, 0 };
1620 static OptFunc DOptPush2 = { OptPush2, "OptPush2", 50, 0, 0, 0, 0, 0 };
1621 static OptFunc DOptPushPop = { OptPushPop, "OptPushPop", 0, 0, 0, 0, 0, 0 };
1622 static OptFunc DOptShift1 = { OptShift1, "OptShift1", 100, 0, 0, 0, 0, 0 };
1623 static OptFunc DOptShift2 = { OptShift2, "OptShift2", 100, 0, 0, 0, 0, 0 };
1624 static OptFunc DOptShift3 = { OptShift3, "OptShift3", 110, 0, 0, 0, 0, 0 };
1625 static OptFunc DOptSize1 = { OptSize1, "OptSize1", 100, 0, 0, 0, 0, 0 };
1626 static OptFunc DOptSize2 = { OptSize2, "OptSize2", 100, 0, 0, 0, 0, 0 };
1627 static OptFunc DOptStackOps = { OptStackOps, "OptStackOps", 100, 0, 0, 0, 0, 0 };
1628 static OptFunc DOptStore1 = { OptStore1, "OptStore1", 70, 0, 0, 0, 0, 0 };
1629 static OptFunc DOptStore2 = { OptStore2, "OptStore2", 220, 0, 0, 0, 0, 0 };
1630 static OptFunc DOptStore3 = { OptStore3, "OptStore3", 120, 0, 0, 0, 0, 0 };
1631 static OptFunc DOptStore4 = { OptStore4, "OptStore4", 50, 0, 0, 0, 0, 0 };
1632 static OptFunc DOptStoreLoad = { OptStoreLoad, "OptStoreLoad", 0, 0, 0, 0, 0, 0 };
1633 static OptFunc DOptSub1 = { OptSub1, "OptSub1", 100, 0, 0, 0, 0, 0 };
1634 static OptFunc DOptSub2 = { OptSub2, "OptSub2", 100, 0, 0, 0, 0, 0 };
1635 static OptFunc DOptTest1 = { OptTest1, "OptTest1", 100, 0, 0, 0, 0, 0 };
1636 static OptFunc DOptTransfers1 = { OptTransfers1, "OptTransfers1", 0, 0, 0, 0, 0, 0 };
1637 static OptFunc DOptTransfers2 = { OptTransfers2, "OptTransfers2", 60, 0, 0, 0, 0, 0 };
1638 static OptFunc DOptUnusedLoads = { OptUnusedLoads, "OptUnusedLoads", 0, 0, 0, 0, 0, 0 };
1639 static OptFunc DOptUnusedStores = { OptUnusedStores, "OptUnusedStores", 0, 0, 0, 0, 0, 0 };
1642 /* Table containing all the steps in alphabetical order */
1643 static OptFunc* OptFuncs[] = {
1710 #define OPTFUNC_COUNT (sizeof(OptFuncs) / sizeof(OptFuncs[0]))
1714 static int CmpOptStep (const void* Key, const void* Func)
1715 /* Compare function for bsearch */
1717 return strcmp (Key, (*(const OptFunc**)Func)->Name);
1722 static OptFunc* FindOptFunc (const char* Name)
1723 /* Find an optimizer step by name in the table and return a pointer. Return
1724 * NULL if no such step is found.
1727 /* Search for the function in the list */
1728 OptFunc** O = bsearch (Name, OptFuncs, OPTFUNC_COUNT, sizeof (OptFuncs[0]), CmpOptStep);
1734 static OptFunc* GetOptFunc (const char* Name)
1735 /* Find an optimizer step by name in the table and return a pointer. Print an
1736 * error and call AbEnd if not found.
1739 /* Search for the function in the list */
1740 OptFunc* F = FindOptFunc (Name);
1743 AbEnd ("Optimization step `%s' not found", Name);
1750 void DisableOpt (const char* Name)
1751 /* Disable the optimization with the given name */
1753 if (strcmp (Name, "any") == 0) {
1755 for (I = 0; I < OPTFUNC_COUNT; ++I) {
1756 OptFuncs[I]->Disabled = 1;
1759 GetOptFunc(Name)->Disabled = 1;
1765 void EnableOpt (const char* Name)
1766 /* Enable the optimization with the given name */
1768 if (strcmp (Name, "any") == 0) {
1770 for (I = 0; I < OPTFUNC_COUNT; ++I) {
1771 OptFuncs[I]->Disabled = 0;
1774 GetOptFunc(Name)->Disabled = 0;
1780 void ListOptSteps (FILE* F)
1781 /* List all optimization steps */
1784 for (I = 0; I < OPTFUNC_COUNT; ++I) {
1785 fprintf (F, "%s\n", OptFuncs[I]->Name);
1791 static void ReadOptStats (const char* Name)
1792 /* Read the optimizer statistics file */
1797 /* Try to open the file */
1798 FILE* F = fopen (Name, "r");
1800 /* Ignore the error */
1804 /* Read and parse the lines */
1806 while (fgets (Buf, sizeof (Buf), F) != 0) {
1814 unsigned long TotalRuns;
1815 unsigned long TotalChanges;
1820 /* Remove trailing white space including the line terminator */
1823 while (Len > 0 && IsSpace (B[Len-1])) {
1828 /* Remove leading whitespace */
1829 while (IsSpace (*B)) {
1833 /* Check for empty and comment lines */
1834 if (*B == '\0' || *B == ';' || *B == '#') {
1838 /* Parse the line */
1839 if (sscanf (B, "%31s %lu %*u %lu %*u", Name, &TotalRuns, &TotalChanges) != 3) {
1844 /* Search for the optimizer step. */
1845 Func = FindOptFunc (Name);
1851 /* Found the step, set the fields */
1852 Func->TotalRuns = TotalRuns;
1853 Func->TotalChanges = TotalChanges;
1857 /* Close the file, ignore errors here. */
1863 static void WriteOptStats (const char* Name)
1864 /* Write the optimizer statistics file */
1868 /* Try to open the file */
1869 FILE* F = fopen (Name, "w");
1871 /* Ignore the error */
1875 /* Write a header */
1877 "; Optimizer Total Last Total Last\n"
1878 "; Step Runs Runs Chg Chg\n");
1881 /* Write the data */
1882 for (I = 0; I < OPTFUNC_COUNT; ++I) {
1883 const OptFunc* O = OptFuncs[I];
1885 "%-20s %10lu %10lu %10lu %10lu\n",
1893 /* Close the file, ignore errors here. */
1899 static unsigned RunOptFunc (CodeSeg* S, OptFunc* F, unsigned Max)
1900 /* Run one optimizer function Max times or until there are no more changes */
1902 unsigned Changes, C;
1904 /* Don't run the function if it is disabled or if it is prohibited by the
1907 if (F->Disabled || F->CodeSizeFactor > CodeSizeFactor) {
1911 /* Run this until there are no more changes */
1915 /* Run the function */
1922 F->TotalChanges += C;
1923 F->LastChanges += C;
1925 } while (--Max && C > 0);
1927 /* Return the number of changes */
1933 static unsigned RunOptGroup1 (CodeSeg* S)
1934 /* Run the first group of optimization steps. These steps translate known
1935 * patterns emitted by the code generator into more optimal patterns. Order
1936 * of the steps is important, because some of the steps done earlier cover
1937 * the same patterns as later steps as subpatterns.
1940 unsigned Changes = 0;
1942 Changes += RunOptFunc (S, &DOptPtrStore1, 1);
1943 Changes += RunOptFunc (S, &DOptPtrStore2, 1);
1944 Changes += RunOptFunc (S, &DOptPtrLoad1, 1);
1945 Changes += RunOptFunc (S, &DOptPtrLoad2, 1);
1946 Changes += RunOptFunc (S, &DOptPtrLoad3, 1);
1947 Changes += RunOptFunc (S, &DOptPtrLoad4, 1);
1948 Changes += RunOptFunc (S, &DOptPtrLoad5, 1);
1949 Changes += RunOptFunc (S, &DOptPtrLoad6, 1);
1950 Changes += RunOptFunc (S, &DOptNegAX1, 1);
1951 Changes += RunOptFunc (S, &DOptNegAX2, 1);
1952 Changes += RunOptFunc (S, &DOptNegAX3, 1);
1953 Changes += RunOptFunc (S, &DOptNegAX4, 1);
1954 Changes += RunOptFunc (S, &DOptAdd1, 1);
1955 Changes += RunOptFunc (S, &DOptAdd2, 1);
1956 Changes += RunOptFunc (S, &DOptAdd3, 1);
1957 Changes += RunOptFunc (S, &DOptStore4, 1);
1958 Changes += RunOptFunc (S, &DOptShift1, 1);
1959 Changes += RunOptFunc (S, &DOptShift2, 1);
1960 Changes += RunOptFunc (S, &DOptShift3, 1);
1961 Changes += RunOptFunc (S, &DOptStore1, 1);
1962 Changes += RunOptFunc (S, &DOptStore2, 5);
1963 Changes += RunOptFunc (S, &DOptStore3, 5);
1965 /* Return the number of changes */
1971 static unsigned RunOptGroup2 (CodeSeg* S)
1972 /* Run one group of optimization steps. This step involves just decoupling
1973 * instructions by replacing them by instructions that do not depend on
1974 * previous instructions. This makes it easier to find instructions that
1978 unsigned Changes = 0;
1980 Changes += RunOptFunc (S, &DOptDecouple, 1);
1982 /* Return the number of changes */
1988 static unsigned RunOptGroup3 (CodeSeg* S)
1989 /* Run one group of optimization steps. These steps depend on each other,
1990 * that means that one step may allow another step to do additional work,
1991 * so we will repeat the steps as long as we see any changes.
1994 unsigned Changes, C;
2000 C += RunOptFunc (S, &DOptPtrLoad7, 1);
2001 C += RunOptFunc (S, &DOptNegA1, 1);
2002 C += RunOptFunc (S, &DOptNegA2, 1);
2003 C += RunOptFunc (S, &DOptSub1, 1);
2004 C += RunOptFunc (S, &DOptSub2, 1);
2005 C += RunOptFunc (S, &DOptAdd4, 1);
2006 C += RunOptFunc (S, &DOptAdd5, 1);
2007 C += RunOptFunc (S, &DOptStackOps, 1);
2008 C += RunOptFunc (S, &DOptJumpCascades, 1);
2009 C += RunOptFunc (S, &DOptDeadJumps, 1);
2010 C += RunOptFunc (S, &DOptRTS, 1);
2011 C += RunOptFunc (S, &DOptDeadCode, 1);
2012 C += RunOptFunc (S, &DOptJumpTarget, 1);
2013 C += RunOptFunc (S, &DOptCondBranches, 1);
2014 C += RunOptFunc (S, &DOptRTSJumps1, 1);
2015 C += RunOptFunc (S, &DOptBoolTrans, 1);
2016 C += RunOptFunc (S, &DOptCmp1, 1);
2017 C += RunOptFunc (S, &DOptCmp2, 1);
2018 C += RunOptFunc (S, &DOptCmp3, 1);
2019 C += RunOptFunc (S, &DOptCmp4, 1);
2020 C += RunOptFunc (S, &DOptCmp5, 1);
2021 C += RunOptFunc (S, &DOptCmp6, 1);
2022 C += RunOptFunc (S, &DOptCmp7, 1);
2023 C += RunOptFunc (S, &DOptTest1, 1);
2024 C += RunOptFunc (S, &DOptLoad1, 1);
2025 C += RunOptFunc (S, &DOptUnusedLoads, 1);
2026 C += RunOptFunc (S, &DOptUnusedStores, 1);
2027 C += RunOptFunc (S, &DOptDupLoads, 1);
2028 C += RunOptFunc (S, &DOptStoreLoad, 1);
2029 C += RunOptFunc (S, &DOptTransfers1, 1);
2030 C += RunOptFunc (S, &DOptPushPop, 1);
2031 C += RunOptFunc (S, &DOptPrecalc, 1);
2037 /* Return the number of changes */
2043 static unsigned RunOptGroup4 (CodeSeg* S)
2044 /* 65C02 specific optimizations. */
2046 unsigned Changes = 0;
2048 if (CPUIsets[CPU] & CPU_ISET_65SC02) {
2049 Changes += RunOptFunc (S, &DOpt65C02BitOps, 1);
2050 Changes += RunOptFunc (S, &DOpt65C02Ind, 1);
2051 Changes += RunOptFunc (S, &DOpt65C02Stores, 1);
2053 /* The 65C02 replacement codes do often make the use of a register
2054 * value unnecessary, so if we have changes, run another load
2057 Changes += RunOptFunc (S, &DOptUnusedLoads, 1);
2061 /* Return the number of changes */
2067 static unsigned RunOptGroup5 (CodeSeg* S)
2068 /* Run another round of pattern replacements. These are done late, since there
2069 * may be better replacements before.
2072 unsigned Changes = 0;
2074 Changes += RunOptFunc (S, &DOptPush1, 1);
2075 Changes += RunOptFunc (S, &DOptPush2, 1);
2076 Changes += RunOptFunc (S, &DOptUnusedLoads, 1);
2077 Changes += RunOptFunc (S, &DOptTransfers2, 1);
2079 /* Return the number of changes */
2085 static unsigned RunOptGroup6 (CodeSeg* S)
2086 /* The last group of optimization steps. Adjust branches, do size optimizations.
2089 unsigned Changes = 0;
2092 if (CodeSizeFactor <= 100) {
2093 /* Optimize for size, that is replace operations by shorter ones, even
2094 * if this does hinder further optimizations (no problem since we're
2097 C = RunOptFunc (S, &DOptSize1, 1);
2100 /* Run some optimization passes again, since the size optimizations
2101 * may have opened new oportunities.
2103 Changes += RunOptFunc (S, &DOptUnusedLoads, 1);
2104 Changes += RunOptFunc (S, &DOptJumpTarget, 5);
2107 C = RunOptFunc (S, &DOptSize2, 1);
2110 /* Run some optimization passes again, since the size optimizations
2111 * may have opened new oportunities.
2113 Changes += RunOptFunc (S, &DOptUnusedLoads, 1);
2114 Changes += RunOptFunc (S, &DOptJumpTarget, 5);
2117 /* Adjust branch distances */
2118 Changes += RunOptFunc (S, &DOptBranchDist, 3);
2120 /* Replace conditional branches to RTS. If we had changes, we must run dead
2121 * code elimination again, since the change may have introduced dead code.
2123 C = RunOptFunc (S, &DOptRTSJumps2, 1);
2126 Changes += RunOptFunc (S, &DOptDeadCode, 1);
2129 /* Return the number of changes */
2135 void RunOpt (CodeSeg* S)
2136 /* Run the optimizer */
2138 const char* StatFileName;
2140 /* If we shouldn't run the optimizer, bail out */
2145 /* Check if we are requested to write optimizer statistics */
2146 StatFileName = getenv ("CC65_OPTSTATS");
2148 ReadOptStats (StatFileName);
2151 /* Print the name of the function we are working on */
2153 Print (stdout, 1, "Running optimizer for function `%s'\n", S->Func->Name);
2155 Print (stdout, 1, "Running optimizer for global code segment\n");
2158 /* Run groups of optimizations */
2166 /* Write statistics */
2168 WriteOptStats (StatFileName);