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 */
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_KnownImm (L[9]) &&
639 CE_IsCallTo (L[10], "staspidx") &&
640 !CS_RangeHasLabel (S, I+1, 5) &&
641 !CS_RangeHasLabel (S, I+7, 4) &&
642 /* Check the label last because this is quite costly */
643 (Len = strlen (L[0]->Arg)) > 3 &&
644 L[0]->Arg[0] == '<' &&
645 L[0]->Arg[1] == '(' &&
646 strlen (L[1]->Arg) == Len &&
647 L[1]->Arg[0] == '>' &&
648 memcmp (L[0]->Arg+1, L[1]->Arg+1, Len-1) == 0) {
653 /* We will create all the new stuff behind the current one so
654 * we keep the line references.
656 X = NewCodeEntry (OP65_LDY, L[3]->AM, L[3]->Arg, 0, L[0]->LI);
657 CS_InsertEntry (S, X, I+11);
659 X = NewCodeEntry (OP65_LDX, L[7]->AM, L[7]->Arg, 0, L[7]->LI);
660 CS_InsertEntry (S, X, I+12);
662 X = NewCodeEntry (OP65_LDA, L[8]->AM, L[8]->Arg, 0, L[8]->LI);
663 CS_InsertEntry (S, X, I+13);
665 Label = memcpy (xmalloc (Len-2), L[0]->Arg+2, Len-3);
667 X = NewCodeEntry (OP65_STA, AM65_ABSY, Label, 0, L[10]->LI);
668 CS_InsertEntry (S, X, I+14);
671 /* Remove the old code */
672 CS_DelEntries (S, I, 11);
674 /* Remember, we had changes */
684 /* Return the number of changes made */
690 /*****************************************************************************/
691 /* Optimize loads through pointers */
692 /*****************************************************************************/
696 static unsigned OptPtrLoad1 (CodeSeg* S)
697 /* Search for the sequence:
722 unsigned Changes = 0;
724 /* Walk over the entries */
726 while (I < CS_GetEntryCount (S)) {
731 L[0] = CS_GetEntry (S, I);
733 /* Check for the sequence */
734 if (L[0]->OPC == OP65_CLC &&
735 CS_GetEntries (S, L+1, I+1, 8) &&
736 L[1]->OPC == OP65_ADC &&
737 L[2]->OPC == OP65_TAY &&
738 L[3]->OPC == OP65_TXA &&
739 L[4]->OPC == OP65_ADC &&
740 L[5]->OPC == OP65_TAX &&
741 L[6]->OPC == OP65_TYA &&
742 L[7]->OPC == OP65_LDY &&
743 CE_IsCallTo (L[8], "ldauidx") &&
744 !CS_RangeHasLabel (S, I+1, 8)) {
749 /* Track the insertion point */
753 X = NewCodeEntry (OP65_STA, AM65_ZP, "ptr1", 0, L[2]->LI);
754 CS_InsertEntry (S, X, IP++);
756 /* If the instruction before the clc is a ldx, replace the
757 * txa by an lda with the same location of the ldx. Otherwise
758 * transfer the value in X to A.
760 if ((P = CS_GetPrevEntry (S, I)) != 0 &&
761 P->OPC == OP65_LDX &&
763 X = NewCodeEntry (OP65_LDA, P->AM, P->Arg, 0, P->LI);
765 X = NewCodeEntry (OP65_TXA, AM65_IMP, 0, 0, L[3]->LI);
767 CS_InsertEntry (S, X, IP++);
770 X = NewCodeEntry (OP65_ADC, L[4]->AM, L[4]->Arg, 0, L[4]->LI);
771 CS_InsertEntry (S, X, IP++);
774 X = NewCodeEntry (OP65_STA, AM65_ZP, "ptr1+1", 0, L[5]->LI);
775 CS_InsertEntry (S, X, IP++);
778 X = NewCodeEntry (OP65_LDY, L[7]->AM, L[7]->Arg, 0, L[7]->LI);
779 CS_InsertEntry (S, X, IP++);
782 X = NewCodeEntry (OP65_LDX, AM65_IMM, "$00", 0, L[8]->LI);
783 CS_InsertEntry (S, X, IP++);
786 X = NewCodeEntry (OP65_LDA, AM65_ZP_INDY, "ptr1", 0, L[8]->LI);
787 CS_InsertEntry (S, X, IP++);
789 /* Remove the old instructions */
790 CS_DelEntries (S, IP, 7);
792 /* Remember, we had changes */
802 /* Return the number of changes made */
808 static unsigned OptPtrLoad2 (CodeSeg* S)
809 /* Search for the sequence:
834 unsigned Changes = 0;
836 /* Walk over the entries */
838 while (I < CS_GetEntryCount (S)) {
843 L[0] = CS_GetEntry (S, I);
845 /* Check for the sequence */
846 if (L[0]->OPC == OP65_ADC &&
847 CS_GetEntries (S, L+1, I+1, 8) &&
848 L[1]->OPC == OP65_PHA &&
849 L[2]->OPC == OP65_TXA &&
850 L[3]->OPC == OP65_INY &&
851 L[4]->OPC == OP65_ADC &&
852 L[5]->OPC == OP65_TAX &&
853 L[6]->OPC == OP65_PLA &&
854 L[7]->OPC == OP65_LDY &&
855 CE_IsCallTo (L[8], "ldauidx") &&
856 !CS_RangeHasLabel (S, I+1, 8)) {
860 /* Store the low byte and remove the PHA instead */
861 X = NewCodeEntry (OP65_STA, AM65_ZP, "ptr1", 0, L[0]->LI);
862 CS_InsertEntry (S, X, I+1);
864 /* Store the high byte */
865 X = NewCodeEntry (OP65_STA, AM65_ZP, "ptr1+1", 0, L[4]->LI);
866 CS_InsertEntry (S, X, I+6);
868 /* Load high and low byte */
869 X = NewCodeEntry (OP65_LDX, AM65_IMM, "$00", 0, L[6]->LI);
870 CS_InsertEntry (S, X, I+10);
871 X = NewCodeEntry (OP65_LDA, AM65_ZP_INDY, "ptr1", 0, L[6]->LI);
872 CS_InsertEntry (S, X, I+11);
874 /* Delete the old code */
875 CS_DelEntry (S, I+12); /* jsr ldauidx */
876 CS_DelEntry (S, I+8); /* pla */
877 CS_DelEntry (S, I+7); /* tax */
878 CS_DelEntry (S, I+2); /* pha */
880 /* Remember, we had changes */
890 /* Return the number of changes made */
896 static unsigned OptPtrLoad3 (CodeSeg* S)
897 /* Search for the sequence:
915 unsigned Changes = 0;
917 /* Walk over the entries */
919 while (I < CS_GetEntryCount (S)) {
925 L[0] = CS_GetEntry (S, I);
927 /* Check for the sequence */
928 if (L[0]->OPC == OP65_LDA &&
929 L[0]->AM == AM65_IMM &&
930 CS_GetEntries (S, L+1, I+1, 7) &&
931 L[1]->OPC == OP65_LDX &&
932 L[1]->AM == AM65_IMM &&
933 L[2]->OPC == OP65_CLC &&
934 L[3]->OPC == OP65_ADC &&
935 (L[3]->AM == AM65_ABS || L[3]->AM == AM65_ZP) &&
936 (L[4]->OPC == OP65_BCC || L[4]->OPC == OP65_JCC) &&
938 L[4]->JumpTo->Owner == L[6] &&
939 L[5]->OPC == OP65_INX &&
940 L[6]->OPC == OP65_LDY &&
941 CE_KnownImm (L[6]) &&
943 CE_IsCallTo (L[7], "ldauidx") &&
944 !CS_RangeHasLabel (S, I+1, 5) &&
945 !CE_HasLabel (L[7]) &&
946 /* Check the label last because this is quite costly */
947 (Len = strlen (L[0]->Arg)) > 3 &&
948 L[0]->Arg[0] == '<' &&
949 L[0]->Arg[1] == '(' &&
950 strlen (L[1]->Arg) == Len &&
951 L[1]->Arg[0] == '>' &&
952 memcmp (L[0]->Arg+1, L[1]->Arg+1, Len-1) == 0) {
957 /* We will create all the new stuff behind the current one so
958 * we keep the line references.
960 X = NewCodeEntry (OP65_LDY, L[3]->AM, L[3]->Arg, 0, L[0]->LI);
961 CS_InsertEntry (S, X, I+8);
963 X = NewCodeEntry (OP65_LDX, AM65_IMM, "$00", 0, L[0]->LI);
964 CS_InsertEntry (S, X, I+9);
966 Label = memcpy (xmalloc (Len-2), L[0]->Arg+2, Len-3);
968 X = NewCodeEntry (OP65_LDA, AM65_ABSY, Label, 0, L[0]->LI);
969 CS_InsertEntry (S, X, I+10);
972 /* Remove the old code */
973 CS_DelEntries (S, I, 8);
975 /* Remember, we had changes */
985 /* Return the number of changes made */
991 static unsigned OptPtrLoad4 (CodeSeg* S)
992 /* Search for the sequence:
1004 * and replace it by:
1013 unsigned Changes = 0;
1015 /* Walk over the entries */
1017 while (I < CS_GetEntryCount (S)) {
1022 /* Get next entry */
1023 L[0] = CS_GetEntry (S, I);
1025 /* Check for the sequence */
1026 if (L[0]->OPC == OP65_LDA &&
1027 L[0]->AM == AM65_IMM &&
1028 CS_GetEntries (S, L+1, I+1, 8) &&
1029 L[1]->OPC == OP65_LDX &&
1030 L[1]->AM == AM65_IMM &&
1031 !CE_HasLabel (L[1]) &&
1032 L[2]->OPC == OP65_LDY &&
1033 CE_KnownImm (L[2]) &&
1034 !CE_HasLabel (L[2]) &&
1035 L[3]->OPC == OP65_CLC &&
1036 !CE_HasLabel (L[3]) &&
1037 L[4]->OPC == OP65_ADC &&
1038 L[4]->AM == AM65_ZP_INDY &&
1039 !CE_HasLabel (L[4]) &&
1040 (L[5]->OPC == OP65_BCC || L[5]->OPC == OP65_JCC) &&
1041 L[5]->JumpTo != 0 &&
1042 L[5]->JumpTo->Owner == L[7] &&
1043 !CE_HasLabel (L[5]) &&
1044 L[6]->OPC == OP65_INX &&
1045 !CE_HasLabel (L[6]) &&
1046 L[7]->OPC == OP65_LDY &&
1047 CE_KnownImm (L[7]) &&
1049 CE_IsCallTo (L[8], "ldauidx") &&
1050 !CE_HasLabel (L[8]) &&
1051 /* Check the label last because this is quite costly */
1052 (Len = strlen (L[0]->Arg)) > 3 &&
1053 L[0]->Arg[0] == '<' &&
1054 L[0]->Arg[1] == '(' &&
1055 strlen (L[1]->Arg) == Len &&
1056 L[1]->Arg[0] == '>' &&
1057 memcmp (L[0]->Arg+1, L[1]->Arg+1, Len-1) == 0) {
1063 X = NewCodeEntry (OP65_LDA, AM65_ZP_INDY, L[4]->Arg, 0, L[0]->LI);
1064 CS_InsertEntry (S, X, I+3);
1067 X = NewCodeEntry (OP65_TAY, AM65_IMP, 0, 0, L[0]->LI);
1068 CS_InsertEntry (S, X, I+4);
1071 X = NewCodeEntry (OP65_LDX, AM65_IMM, "$00", 0, L[0]->LI);
1072 CS_InsertEntry (S, X, I+5);
1075 Label = memcpy (xmalloc (Len-2), L[0]->Arg+2, Len-3);
1076 Label[Len-3] = '\0';
1077 X = NewCodeEntry (OP65_LDA, AM65_ABSY, Label, 0, L[0]->LI);
1078 CS_InsertEntry (S, X, I+6);
1081 /* Remove the old code */
1082 CS_DelEntries (S, I, 2);
1083 CS_DelEntries (S, I+5, 6);
1085 /* Remember, we had changes */
1095 /* Return the number of changes made */
1101 static unsigned OptPtrLoad5 (CodeSeg* S)
1102 /* Search for the sequence:
1119 * and replace it by:
1127 * L1: tay <- only if flags are used
1129 * This function must execute before OptPtrLoad5!
1133 unsigned Changes = 0;
1135 /* Walk over the entries */
1137 while (I < CS_GetEntryCount (S)) {
1142 /* Get next entry */
1143 L[0] = CS_GetEntry (S, I);
1145 /* Check for the sequence */
1146 if (L[0]->OPC == OP65_LDA &&
1147 L[0]->AM == AM65_ZP &&
1148 strncmp (L[0]->Arg, "regbank+", 8) == 0 &&
1149 (Len = strlen (L[0]->Arg)) > 0 &&
1150 CS_GetEntries (S, L+1, I+1, 14) &&
1151 !CS_RangeHasLabel (S, I+1, 7) &&
1152 !CS_RangeHasLabel (S, I+9, 5) &&
1153 L[1]->OPC == OP65_LDX &&
1154 L[1]->AM == AM65_ZP &&
1155 strncmp (L[1]->Arg, L[0]->Arg, Len) == 0 &&
1156 strcmp (L[1]->Arg+Len, "+1") == 0 &&
1157 L[2]->OPC == OP65_STA &&
1158 L[2]->AM == AM65_ZP &&
1159 strcmp (L[2]->Arg, "regsave") == 0 &&
1160 L[3]->OPC == OP65_STX &&
1161 L[3]->AM == AM65_ZP &&
1162 strcmp (L[3]->Arg, "regsave+1") == 0 &&
1163 L[4]->OPC == OP65_CLC &&
1164 L[5]->OPC == OP65_ADC &&
1165 CE_KnownImm (L[5]) &&
1167 L[6]->OPC == OP65_BCC &&
1168 L[6]->JumpTo != 0 &&
1169 L[6]->JumpTo->Owner == L[8] &&
1170 L[7]->OPC == OP65_INX &&
1171 L[8]->OPC == OP65_STA &&
1172 L[8]->AM == AM65_ZP &&
1173 strcmp (L[8]->Arg, L[0]->Arg) == 0 &&
1174 L[9]->OPC == OP65_STX &&
1175 L[9]->AM == AM65_ZP &&
1176 strcmp (L[9]->Arg, L[1]->Arg) == 0 &&
1177 L[10]->OPC == OP65_LDA &&
1178 L[10]->AM == AM65_ZP &&
1179 strcmp (L[10]->Arg, "regsave") == 0 &&
1180 L[11]->OPC == OP65_LDX &&
1181 L[11]->AM == AM65_ZP &&
1182 strcmp (L[11]->Arg, "regsave+1") == 0 &&
1183 L[12]->OPC == OP65_LDY &&
1184 CE_KnownImm (L[12]) &&
1185 CE_IsCallTo (L[13], "ldauidx")) {
1190 /* Check if the instruction following the sequence uses the flags
1191 * set by the load. If so, insert a test of the value in the
1194 if (CE_UseLoadFlags (L[14])) {
1195 X = NewCodeEntry (OP65_TAY, AM65_IMP, 0, 0, L[13]->LI);
1196 CS_InsertEntry (S, X, I+14);
1199 /* Attach a label to L[14]. This may be either the just inserted
1200 * instruction, or the one following the sequence.
1202 Label = CS_GenLabel (S, L[14]);
1205 X = NewCodeEntry (OP65_LDY, AM65_IMM, L[12]->Arg, 0, L[12]->LI);
1206 CS_InsertEntry (S, X, I+14);
1209 X = NewCodeEntry (OP65_LDX, AM65_IMM, "$00", 0, L[13]->LI);
1210 CS_InsertEntry (S, X, I+15);
1212 /* lda (regbank+n),y */
1213 X = NewCodeEntry (OP65_LDA, AM65_ZP_INDY, L[0]->Arg, 0, L[13]->LI);
1214 CS_InsertEntry (S, X, I+16);
1217 X = NewCodeEntry (OP65_INC, AM65_ZP, L[0]->Arg, 0, L[5]->LI);
1218 CS_InsertEntry (S, X, I+17);
1221 X = NewCodeEntry (OP65_BNE, AM65_BRA, Label->Name, Label, L[6]->LI);
1222 CS_InsertEntry (S, X, I+18);
1224 /* inc regbank+n+1 */
1225 X = NewCodeEntry (OP65_INC, AM65_ZP, L[1]->Arg, 0, L[7]->LI);
1226 CS_InsertEntry (S, X, I+19);
1228 /* Delete the old code */
1229 CS_DelEntries (S, I, 14);
1231 /* Remember, we had changes */
1241 /* Return the number of changes made */
1247 static unsigned OptPtrLoad6 (CodeSeg* S)
1248 /* Search for the sequence:
1255 * and replace it by:
1262 unsigned Changes = 0;
1264 /* Walk over the entries */
1266 while (I < CS_GetEntryCount (S)) {
1271 /* Get next entry */
1272 L[0] = CS_GetEntry (S, I);
1274 /* Check for the sequence */
1275 if (L[0]->OPC == OP65_LDA && L[0]->AM == AM65_ZP &&
1276 CS_GetEntries (S, L+1, I+1, 3) &&
1277 !CS_RangeHasLabel (S, I+1, 3) &&
1278 L[1]->OPC == OP65_LDX && L[1]->AM == AM65_ZP &&
1279 (Len = strlen (L[0]->Arg)) > 0 &&
1280 strncmp (L[0]->Arg, L[1]->Arg, Len) == 0 &&
1281 strcmp (L[1]->Arg + Len, "+1") == 0 &&
1282 L[2]->OPC == OP65_LDY &&
1283 CE_IsCallTo (L[3], "ldauidx")) {
1288 X = NewCodeEntry (OP65_LDX, AM65_IMM, "$00", 0, L[3]->LI);
1289 CS_InsertEntry (S, X, I+3);
1292 X = NewCodeEntry (OP65_LDA, AM65_ZP_INDY, L[0]->Arg, 0, L[3]->LI);
1293 CS_InsertEntry (S, X, I+4);
1295 /* Remove the old code */
1296 CS_DelEntry (S, I+5);
1297 CS_DelEntries (S, I, 2);
1299 /* Remember, we had changes */
1309 /* Return the number of changes made */
1315 static unsigned OptPtrLoad7 (CodeSeg* S)
1316 /* Search for the sequence:
1323 * and replace it by:
1332 unsigned Changes = 0;
1334 /* Walk over the entries */
1336 while (I < CS_GetEntryCount (S)) {
1341 /* Get next entry */
1342 L[0] = CS_GetEntry (S, I);
1344 /* Check for the sequence */
1345 if (L[0]->OPC == OP65_LDA && L[0]->AM == AM65_ZP &&
1346 CS_GetEntries (S, L+1, I+1, 3) &&
1347 !CS_RangeHasLabel (S, I+1, 3) &&
1348 L[1]->OPC == OP65_LDX && L[1]->AM == AM65_ZP &&
1349 (Len = strlen (L[0]->Arg)) > 0 &&
1350 strncmp (L[0]->Arg, L[1]->Arg, Len) == 0 &&
1351 strcmp (L[1]->Arg + Len, "+1") == 0 &&
1352 L[2]->OPC == OP65_LDY &&
1353 CE_IsCallTo (L[3], "ldaxidx")) {
1358 X = NewCodeEntry (OP65_LDA, AM65_ZP_INDY, L[0]->Arg, 0, L[3]->LI);
1359 CS_InsertEntry (S, X, I+4);
1362 X = NewCodeEntry (OP65_TAX, AM65_IMP, 0, 0, L[3]->LI);
1363 CS_InsertEntry (S, X, I+5);
1366 X = NewCodeEntry (OP65_DEY, AM65_IMP, 0, 0, L[3]->LI);
1367 CS_InsertEntry (S, X, I+6);
1370 X = NewCodeEntry (OP65_LDA, AM65_ZP_INDY, L[0]->Arg, 0, L[3]->LI);
1371 CS_InsertEntry (S, X, I+7);
1373 /* Remove the old code */
1374 CS_DelEntry (S, I+3);
1375 CS_DelEntries (S, I, 2);
1377 /* Remember, we had changes */
1387 /* Return the number of changes made */
1393 static unsigned OptPtrLoad8 (CodeSeg* S)
1394 /* Search for the sequence
1399 * and replace it by:
1407 * This step must be executed *after* OptPtrLoad1!
1410 unsigned Changes = 0;
1412 /* Walk over the entries */
1414 while (I < CS_GetEntryCount (S)) {
1418 /* Get next entry */
1419 L[0] = CS_GetEntry (S, I);
1421 /* Check for the sequence */
1422 if (L[0]->OPC == OP65_LDY &&
1423 CS_GetEntries (S, L+1, I+1, 1) &&
1424 CE_IsCallTo (L[1], "ldauidx") &&
1425 !CE_HasLabel (L[1])) {
1429 /* Store the high byte */
1430 X = NewCodeEntry (OP65_STA, AM65_ZP, "ptr1", 0, L[0]->LI);
1431 CS_InsertEntry (S, X, I+1);
1433 /* Store the low byte */
1434 X = NewCodeEntry (OP65_STX, AM65_ZP, "ptr1+1", 0, L[0]->LI);
1435 CS_InsertEntry (S, X, I+2);
1437 /* Delete the call to ldauidx */
1438 CS_DelEntry (S, I+3);
1440 /* Load the high and low byte */
1441 X = NewCodeEntry (OP65_LDX, AM65_IMM, "$00", 0, L[0]->LI);
1442 CS_InsertEntry (S, X, I+3);
1443 X = NewCodeEntry (OP65_LDA, AM65_ZP_INDY, "ptr1", 0, L[0]->LI);
1444 CS_InsertEntry (S, X, I+4);
1446 /* Remember, we had changes */
1456 /* Return the number of changes made */
1462 /*****************************************************************************/
1463 /* Decouple operations */
1464 /*****************************************************************************/
1468 static unsigned OptDecouple (CodeSeg* S)
1469 /* Decouple operations, that is, do the following replacements:
1479 * lda zp -> lda #imm
1480 * ldx zp -> ldx #imm
1481 * ldy zp -> ldy #imm
1483 * Provided that the register values are known of course.
1486 unsigned Changes = 0;
1489 /* Generate register info for the following step */
1492 /* Walk over the entries */
1494 while (I < CS_GetEntryCount (S)) {
1498 /* Get next entry and it's input register values */
1499 CodeEntry* E = CS_GetEntry (S, I);
1500 const RegContents* In = &E->RI->In;
1502 /* Assume we have no replacement */
1505 /* Check the instruction */
1509 if (RegValIsKnown (In->RegA)) {
1510 Arg = MakeHexArg ((In->RegA - 1) & 0xFF);
1511 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
1516 if (RegValIsKnown (In->RegX)) {
1517 Arg = MakeHexArg ((In->RegX - 1) & 0xFF);
1518 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
1523 if (RegValIsKnown (In->RegY)) {
1524 Arg = MakeHexArg ((In->RegY - 1) & 0xFF);
1525 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
1530 if (RegValIsKnown (In->RegA)) {
1531 Arg = MakeHexArg ((In->RegA + 1) & 0xFF);
1532 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
1537 if (RegValIsKnown (In->RegX)) {
1538 Arg = MakeHexArg ((In->RegX + 1) & 0xFF);
1539 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
1544 if (RegValIsKnown (In->RegY)) {
1545 Arg = MakeHexArg ((In->RegY + 1) & 0xFF);
1546 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
1551 if (E->AM == AM65_ZP) {
1552 switch (GetKnownReg (E->Use & REG_ZP, In)) {
1554 Arg = MakeHexArg (In->Tmp1);
1555 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
1559 Arg = MakeHexArg (In->Ptr1Lo);
1560 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
1564 Arg = MakeHexArg (In->Ptr1Hi);
1565 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
1569 Arg = MakeHexArg (In->SRegLo);
1570 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
1574 Arg = MakeHexArg (In->SRegHi);
1575 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
1582 if (E->AM == AM65_ZP) {
1583 switch (GetKnownReg (E->Use & REG_ZP, In)) {
1585 Arg = MakeHexArg (In->Tmp1);
1586 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
1590 Arg = MakeHexArg (In->Ptr1Lo);
1591 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
1595 Arg = MakeHexArg (In->Ptr1Hi);
1596 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
1600 Arg = MakeHexArg (In->SRegLo);
1601 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
1605 Arg = MakeHexArg (In->SRegHi);
1606 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
1613 if (E->AM == AM65_ZP) {
1614 switch (GetKnownReg (E->Use, In)) {
1616 Arg = MakeHexArg (In->Tmp1);
1617 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
1621 Arg = MakeHexArg (In->Ptr1Lo);
1622 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
1626 Arg = MakeHexArg (In->Ptr1Hi);
1627 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
1631 Arg = MakeHexArg (In->SRegLo);
1632 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
1636 Arg = MakeHexArg (In->SRegHi);
1637 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
1644 if (E->RI->In.RegA >= 0) {
1645 Arg = MakeHexArg (In->RegA);
1646 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
1651 if (E->RI->In.RegA >= 0) {
1652 Arg = MakeHexArg (In->RegA);
1653 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
1658 if (E->RI->In.RegX >= 0) {
1659 Arg = MakeHexArg (In->RegX);
1660 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
1665 if (E->RI->In.RegY >= 0) {
1666 Arg = MakeHexArg (In->RegY);
1667 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
1672 /* Avoid gcc warnings */
1677 /* Insert the replacement if we have one */
1679 CS_InsertEntry (S, X, I+1);
1689 /* Free register info */
1692 /* Return the number of changes made */
1698 /*****************************************************************************/
1699 /* struct OptFunc */
1700 /*****************************************************************************/
1704 typedef struct OptFunc OptFunc;
1706 unsigned (*Func) (CodeSeg*); /* Optimizer function */
1707 const char* Name; /* Name of the function/group */
1708 unsigned CodeSizeFactor; /* Code size factor for this opt func */
1709 unsigned long TotalRuns; /* Total number of runs */
1710 unsigned long LastRuns; /* Last number of runs */
1711 unsigned long TotalChanges; /* Total number of changes */
1712 unsigned long LastChanges; /* Last number of changes */
1713 char Disabled; /* True if function disabled */
1718 /*****************************************************************************/
1720 /*****************************************************************************/
1724 /* A list of all the function descriptions */
1725 static OptFunc DOpt65C02BitOps = { Opt65C02BitOps, "Opt65C02BitOps", 66, 0, 0, 0, 0, 0 };
1726 static OptFunc DOpt65C02Ind = { Opt65C02Ind, "Opt65C02Ind", 100, 0, 0, 0, 0, 0 };
1727 static OptFunc DOpt65C02Stores = { Opt65C02Stores, "Opt65C02Stores", 100, 0, 0, 0, 0, 0 };
1728 static OptFunc DOptAdd1 = { OptAdd1, "OptAdd1", 125, 0, 0, 0, 0, 0 };
1729 static OptFunc DOptAdd2 = { OptAdd2, "OptAdd2", 200, 0, 0, 0, 0, 0 };
1730 static OptFunc DOptAdd3 = { OptAdd3, "OptAdd3", 90, 0, 0, 0, 0, 0 };
1731 static OptFunc DOptAdd4 = { OptAdd4, "OptAdd4", 100, 0, 0, 0, 0, 0 };
1732 static OptFunc DOptAdd5 = { OptAdd5, "OptAdd5", 40, 0, 0, 0, 0, 0 };
1733 static OptFunc DOptBoolTrans = { OptBoolTrans, "OptBoolTrans", 100, 0, 0, 0, 0, 0 };
1734 static OptFunc DOptBranchDist = { OptBranchDist, "OptBranchDist", 0, 0, 0, 0, 0, 0 };
1735 static OptFunc DOptCmp1 = { OptCmp1, "OptCmp1", 42, 0, 0, 0, 0, 0 };
1736 static OptFunc DOptCmp2 = { OptCmp2, "OptCmp2", 85, 0, 0, 0, 0, 0 };
1737 static OptFunc DOptCmp3 = { OptCmp3, "OptCmp3", 75, 0, 0, 0, 0, 0 };
1738 static OptFunc DOptCmp4 = { OptCmp4, "OptCmp4", 75, 0, 0, 0, 0, 0 };
1739 static OptFunc DOptCmp5 = { OptCmp5, "OptCmp5", 100, 0, 0, 0, 0, 0 };
1740 static OptFunc DOptCmp6 = { OptCmp6, "OptCmp6", 100, 0, 0, 0, 0, 0 };
1741 static OptFunc DOptCmp7 = { OptCmp7, "OptCmp7", 85, 0, 0, 0, 0, 0 };
1742 static OptFunc DOptCmp8 = { OptCmp8, "OptCmp8", 50, 0, 0, 0, 0, 0 };
1743 static OptFunc DOptCondBranches = { OptCondBranches, "OptCondBranches", 80, 0, 0, 0, 0, 0 };
1744 static OptFunc DOptDeadCode = { OptDeadCode, "OptDeadCode", 100, 0, 0, 0, 0, 0 };
1745 static OptFunc DOptDeadJumps = { OptDeadJumps, "OptDeadJumps", 100, 0, 0, 0, 0, 0 };
1746 static OptFunc DOptDecouple = { OptDecouple, "OptDecouple", 100, 0, 0, 0, 0, 0 };
1747 static OptFunc DOptDupLoads = { OptDupLoads, "OptDupLoads", 0, 0, 0, 0, 0, 0 };
1748 static OptFunc DOptJumpCascades = { OptJumpCascades, "OptJumpCascades", 100, 0, 0, 0, 0, 0 };
1749 static OptFunc DOptJumpTarget = { OptJumpTarget, "OptJumpTarget", 100, 0, 0, 0, 0, 0 };
1750 static OptFunc DOptLoad1 = { OptLoad1, "OptLoad1", 100, 0, 0, 0, 0, 0 };
1751 static OptFunc DOptRTS = { OptRTS, "OptRTS", 100, 0, 0, 0, 0, 0 };
1752 static OptFunc DOptRTSJumps1 = { OptRTSJumps1, "OptRTSJumps1", 100, 0, 0, 0, 0, 0 };
1753 static OptFunc DOptRTSJumps2 = { OptRTSJumps2, "OptRTSJumps2", 100, 0, 0, 0, 0, 0 };
1754 static OptFunc DOptNegA1 = { OptNegA1, "OptNegA1", 100, 0, 0, 0, 0, 0 };
1755 static OptFunc DOptNegA2 = { OptNegA2, "OptNegA2", 100, 0, 0, 0, 0, 0 };
1756 static OptFunc DOptNegAX1 = { OptNegAX1, "OptNegAX1", 100, 0, 0, 0, 0, 0 };
1757 static OptFunc DOptNegAX2 = { OptNegAX2, "OptNegAX2", 100, 0, 0, 0, 0, 0 };
1758 static OptFunc DOptNegAX3 = { OptNegAX3, "OptNegAX3", 100, 0, 0, 0, 0, 0 };
1759 static OptFunc DOptNegAX4 = { OptNegAX4, "OptNegAX4", 100, 0, 0, 0, 0, 0 };
1760 static OptFunc DOptPrecalc = { OptPrecalc, "OptPrecalc", 100, 0, 0, 0, 0, 0 };
1761 static OptFunc DOptPtrLoad1 = { OptPtrLoad1, "OptPtrLoad1", 100, 0, 0, 0, 0, 0 };
1762 static OptFunc DOptPtrLoad2 = { OptPtrLoad2, "OptPtrLoad2", 100, 0, 0, 0, 0, 0 };
1763 static OptFunc DOptPtrLoad3 = { OptPtrLoad3, "OptPtrLoad3", 100, 0, 0, 0, 0, 0 };
1764 static OptFunc DOptPtrLoad4 = { OptPtrLoad4, "OptPtrLoad4", 100, 0, 0, 0, 0, 0 };
1765 static OptFunc DOptPtrLoad5 = { OptPtrLoad5, "OptPtrLoad5", 50, 0, 0, 0, 0, 0 };
1766 static OptFunc DOptPtrLoad6 = { OptPtrLoad6, "OptPtrLoad6", 65, 0, 0, 0, 0, 0 };
1767 static OptFunc DOptPtrLoad7 = { OptPtrLoad7, "OptPtrLoad7", 86, 0, 0, 0, 0, 0 };
1768 static OptFunc DOptPtrLoad8 = { OptPtrLoad8, "OptPtrLoad8", 100, 0, 0, 0, 0, 0 };
1769 static OptFunc DOptPtrStore1 = { OptPtrStore1, "OptPtrStore1", 100, 0, 0, 0, 0, 0 };
1770 static OptFunc DOptPtrStore2 = { OptPtrStore2, "OptPtrStore2", 40, 0, 0, 0, 0, 0 };
1771 static OptFunc DOptPush1 = { OptPush1, "OptPush1", 65, 0, 0, 0, 0, 0 };
1772 static OptFunc DOptPush2 = { OptPush2, "OptPush2", 50, 0, 0, 0, 0, 0 };
1773 static OptFunc DOptPushPop = { OptPushPop, "OptPushPop", 0, 0, 0, 0, 0, 0 };
1774 static OptFunc DOptShift1 = { OptShift1, "OptShift1", 100, 0, 0, 0, 0, 0 };
1775 static OptFunc DOptShift2 = { OptShift2, "OptShift2", 100, 0, 0, 0, 0, 0 };
1776 static OptFunc DOptShift3 = { OptShift3, "OptShift3", 110, 0, 0, 0, 0, 0 };
1777 static OptFunc DOptSize1 = { OptSize1, "OptSize1", 100, 0, 0, 0, 0, 0 };
1778 static OptFunc DOptSize2 = { OptSize2, "OptSize2", 100, 0, 0, 0, 0, 0 };
1779 static OptFunc DOptStackOps = { OptStackOps, "OptStackOps", 100, 0, 0, 0, 0, 0 };
1780 static OptFunc DOptStore1 = { OptStore1, "OptStore1", 70, 0, 0, 0, 0, 0 };
1781 static OptFunc DOptStore2 = { OptStore2, "OptStore2", 220, 0, 0, 0, 0, 0 };
1782 static OptFunc DOptStore3 = { OptStore3, "OptStore3", 120, 0, 0, 0, 0, 0 };
1783 static OptFunc DOptStore4 = { OptStore4, "OptStore4", 50, 0, 0, 0, 0, 0 };
1784 static OptFunc DOptStoreLoad = { OptStoreLoad, "OptStoreLoad", 0, 0, 0, 0, 0, 0 };
1785 static OptFunc DOptSub1 = { OptSub1, "OptSub1", 100, 0, 0, 0, 0, 0 };
1786 static OptFunc DOptSub2 = { OptSub2, "OptSub2", 100, 0, 0, 0, 0, 0 };
1787 static OptFunc DOptTest1 = { OptTest1, "OptTest1", 100, 0, 0, 0, 0, 0 };
1788 static OptFunc DOptTransfers1 = { OptTransfers1, "OptTransfers1", 0, 0, 0, 0, 0, 0 };
1789 static OptFunc DOptTransfers2 = { OptTransfers2, "OptTransfers2", 60, 0, 0, 0, 0, 0 };
1790 static OptFunc DOptUnusedLoads = { OptUnusedLoads, "OptUnusedLoads", 0, 0, 0, 0, 0, 0 };
1791 static OptFunc DOptUnusedStores = { OptUnusedStores, "OptUnusedStores", 0, 0, 0, 0, 0, 0 };
1794 /* Table containing all the steps in alphabetical order */
1795 static OptFunc* OptFuncs[] = {
1864 #define OPTFUNC_COUNT (sizeof(OptFuncs) / sizeof(OptFuncs[0]))
1868 static int CmpOptStep (const void* Key, const void* Func)
1869 /* Compare function for bsearch */
1871 return strcmp (Key, (*(const OptFunc**)Func)->Name);
1876 static OptFunc* FindOptFunc (const char* Name)
1877 /* Find an optimizer step by name in the table and return a pointer. Return
1878 * NULL if no such step is found.
1881 /* Search for the function in the list */
1882 OptFunc** O = bsearch (Name, OptFuncs, OPTFUNC_COUNT, sizeof (OptFuncs[0]), CmpOptStep);
1888 static OptFunc* GetOptFunc (const char* Name)
1889 /* Find an optimizer step by name in the table and return a pointer. Print an
1890 * error and call AbEnd if not found.
1893 /* Search for the function in the list */
1894 OptFunc* F = FindOptFunc (Name);
1897 AbEnd ("Optimization step `%s' not found", Name);
1904 void DisableOpt (const char* Name)
1905 /* Disable the optimization with the given name */
1907 if (strcmp (Name, "any") == 0) {
1909 for (I = 0; I < OPTFUNC_COUNT; ++I) {
1910 OptFuncs[I]->Disabled = 1;
1913 GetOptFunc(Name)->Disabled = 1;
1919 void EnableOpt (const char* Name)
1920 /* Enable the optimization with the given name */
1922 if (strcmp (Name, "any") == 0) {
1924 for (I = 0; I < OPTFUNC_COUNT; ++I) {
1925 OptFuncs[I]->Disabled = 0;
1928 GetOptFunc(Name)->Disabled = 0;
1934 void ListOptSteps (FILE* F)
1935 /* List all optimization steps */
1938 for (I = 0; I < OPTFUNC_COUNT; ++I) {
1939 fprintf (F, "%s\n", OptFuncs[I]->Name);
1945 static void ReadOptStats (const char* Name)
1946 /* Read the optimizer statistics file */
1951 /* Try to open the file */
1952 FILE* F = fopen (Name, "r");
1954 /* Ignore the error */
1958 /* Read and parse the lines */
1960 while (fgets (Buf, sizeof (Buf), F) != 0) {
1968 unsigned long TotalRuns;
1969 unsigned long TotalChanges;
1974 /* Remove trailing white space including the line terminator */
1977 while (Len > 0 && IsSpace (B[Len-1])) {
1982 /* Remove leading whitespace */
1983 while (IsSpace (*B)) {
1987 /* Check for empty and comment lines */
1988 if (*B == '\0' || *B == ';' || *B == '#') {
1992 /* Parse the line */
1993 if (sscanf (B, "%31s %lu %*u %lu %*u", Name, &TotalRuns, &TotalChanges) != 3) {
1998 /* Search for the optimizer step. */
1999 Func = FindOptFunc (Name);
2005 /* Found the step, set the fields */
2006 Func->TotalRuns = TotalRuns;
2007 Func->TotalChanges = TotalChanges;
2011 /* Close the file, ignore errors here. */
2017 static void WriteOptStats (const char* Name)
2018 /* Write the optimizer statistics file */
2022 /* Try to open the file */
2023 FILE* F = fopen (Name, "w");
2025 /* Ignore the error */
2029 /* Write a header */
2031 "; Optimizer Total Last Total Last\n"
2032 "; Step Runs Runs Chg Chg\n");
2035 /* Write the data */
2036 for (I = 0; I < OPTFUNC_COUNT; ++I) {
2037 const OptFunc* O = OptFuncs[I];
2039 "%-20s %10lu %10lu %10lu %10lu\n",
2047 /* Close the file, ignore errors here. */
2053 static unsigned RunOptFunc (CodeSeg* S, OptFunc* F, unsigned Max)
2054 /* Run one optimizer function Max times or until there are no more changes */
2056 unsigned Changes, C;
2058 /* Don't run the function if it is disabled or if it is prohibited by the
2061 if (F->Disabled || F->CodeSizeFactor > S->CodeSizeFactor) {
2065 /* Run this until there are no more changes */
2069 /* Run the function */
2076 F->TotalChanges += C;
2077 F->LastChanges += C;
2079 } while (--Max && C > 0);
2081 /* Return the number of changes */
2087 static unsigned RunOptGroup1 (CodeSeg* S)
2088 /* Run the first group of optimization steps. These steps translate known
2089 * patterns emitted by the code generator into more optimal patterns. Order
2090 * of the steps is important, because some of the steps done earlier cover
2091 * the same patterns as later steps as subpatterns.
2094 unsigned Changes = 0;
2096 Changes += RunOptFunc (S, &DOptPtrStore1, 1);
2097 Changes += RunOptFunc (S, &DOptPtrStore2, 1);
2098 Changes += RunOptFunc (S, &DOptPtrLoad1, 1);
2099 Changes += RunOptFunc (S, &DOptPtrLoad2, 1);
2100 Changes += RunOptFunc (S, &DOptPtrLoad3, 1);
2101 Changes += RunOptFunc (S, &DOptPtrLoad4, 1);
2102 Changes += RunOptFunc (S, &DOptPtrLoad5, 1);
2103 Changes += RunOptFunc (S, &DOptPtrLoad6, 1);
2104 Changes += RunOptFunc (S, &DOptPtrLoad7, 1);
2105 Changes += RunOptFunc (S, &DOptNegAX1, 1);
2106 Changes += RunOptFunc (S, &DOptNegAX2, 1);
2107 Changes += RunOptFunc (S, &DOptNegAX3, 1);
2108 Changes += RunOptFunc (S, &DOptNegAX4, 1);
2109 Changes += RunOptFunc (S, &DOptAdd1, 1);
2110 Changes += RunOptFunc (S, &DOptAdd2, 1);
2111 Changes += RunOptFunc (S, &DOptAdd3, 1);
2112 Changes += RunOptFunc (S, &DOptStore4, 1);
2113 Changes += RunOptFunc (S, &DOptShift1, 1);
2114 Changes += RunOptFunc (S, &DOptShift2, 1);
2115 Changes += RunOptFunc (S, &DOptShift3, 1);
2116 Changes += RunOptFunc (S, &DOptStore1, 1);
2117 Changes += RunOptFunc (S, &DOptStore2, 5);
2118 Changes += RunOptFunc (S, &DOptStore3, 5);
2120 /* Return the number of changes */
2126 static unsigned RunOptGroup2 (CodeSeg* S)
2127 /* Run one group of optimization steps. This step involves just decoupling
2128 * instructions by replacing them by instructions that do not depend on
2129 * previous instructions. This makes it easier to find instructions that
2133 unsigned Changes = 0;
2135 Changes += RunOptFunc (S, &DOptDecouple, 1);
2137 /* Return the number of changes */
2143 static unsigned RunOptGroup3 (CodeSeg* S)
2144 /* Run one group of optimization steps. These steps depend on each other,
2145 * that means that one step may allow another step to do additional work,
2146 * so we will repeat the steps as long as we see any changes.
2149 unsigned Changes, C;
2155 C += RunOptFunc (S, &DOptPtrLoad8, 1);
2156 C += RunOptFunc (S, &DOptNegA1, 1);
2157 C += RunOptFunc (S, &DOptNegA2, 1);
2158 C += RunOptFunc (S, &DOptSub1, 1);
2159 C += RunOptFunc (S, &DOptSub2, 1);
2160 C += RunOptFunc (S, &DOptAdd4, 1);
2161 C += RunOptFunc (S, &DOptAdd5, 1);
2162 C += RunOptFunc (S, &DOptStackOps, 1);
2163 C += RunOptFunc (S, &DOptJumpCascades, 1);
2164 C += RunOptFunc (S, &DOptDeadJumps, 1);
2165 C += RunOptFunc (S, &DOptRTS, 1);
2166 C += RunOptFunc (S, &DOptDeadCode, 1);
2167 C += RunOptFunc (S, &DOptJumpTarget, 1);
2168 C += RunOptFunc (S, &DOptCondBranches, 1);
2169 C += RunOptFunc (S, &DOptRTSJumps1, 1);
2170 C += RunOptFunc (S, &DOptBoolTrans, 1);
2171 C += RunOptFunc (S, &DOptCmp1, 1);
2172 C += RunOptFunc (S, &DOptCmp2, 1);
2173 C += RunOptFunc (S, &DOptCmp3, 1);
2174 C += RunOptFunc (S, &DOptCmp4, 1);
2175 C += RunOptFunc (S, &DOptCmp5, 1);
2176 C += RunOptFunc (S, &DOptCmp6, 1);
2177 C += RunOptFunc (S, &DOptCmp7, 1);
2178 C += RunOptFunc (S, &DOptCmp8, 1);
2179 C += RunOptFunc (S, &DOptTest1, 1);
2180 C += RunOptFunc (S, &DOptLoad1, 1);
2181 C += RunOptFunc (S, &DOptUnusedLoads, 1);
2182 C += RunOptFunc (S, &DOptUnusedStores, 1);
2183 C += RunOptFunc (S, &DOptDupLoads, 1);
2184 C += RunOptFunc (S, &DOptStoreLoad, 1);
2185 C += RunOptFunc (S, &DOptTransfers1, 1);
2186 C += RunOptFunc (S, &DOptPushPop, 1);
2187 C += RunOptFunc (S, &DOptPrecalc, 1);
2193 /* Return the number of changes */
2199 static unsigned RunOptGroup4 (CodeSeg* S)
2200 /* 65C02 specific optimizations. */
2202 unsigned Changes = 0;
2204 if (CPUIsets[CPU] & CPU_ISET_65SC02) {
2205 Changes += RunOptFunc (S, &DOpt65C02BitOps, 1);
2206 Changes += RunOptFunc (S, &DOpt65C02Ind, 1);
2207 Changes += RunOptFunc (S, &DOpt65C02Stores, 1);
2209 /* The 65C02 replacement codes do often make the use of a register
2210 * value unnecessary, so if we have changes, run another load
2213 Changes += RunOptFunc (S, &DOptUnusedLoads, 1);
2217 /* Return the number of changes */
2223 static unsigned RunOptGroup5 (CodeSeg* S)
2224 /* Run another round of pattern replacements. These are done late, since there
2225 * may be better replacements before.
2228 unsigned Changes = 0;
2230 Changes += RunOptFunc (S, &DOptPush1, 1);
2231 Changes += RunOptFunc (S, &DOptPush2, 1);
2232 Changes += RunOptFunc (S, &DOptUnusedLoads, 1);
2233 Changes += RunOptFunc (S, &DOptTransfers2, 1);
2235 /* Return the number of changes */
2241 static unsigned RunOptGroup6 (CodeSeg* S)
2242 /* The last group of optimization steps. Adjust branches, do size optimizations.
2245 unsigned Changes = 0;
2248 if (S->CodeSizeFactor <= 100) {
2249 /* Optimize for size, that is replace operations by shorter ones, even
2250 * if this does hinder further optimizations (no problem since we're
2253 C = RunOptFunc (S, &DOptSize1, 1);
2256 /* Run some optimization passes again, since the size optimizations
2257 * may have opened new oportunities.
2259 Changes += RunOptFunc (S, &DOptUnusedLoads, 1);
2260 Changes += RunOptFunc (S, &DOptJumpTarget, 5);
2263 C = RunOptFunc (S, &DOptSize2, 1);
2266 /* Run some optimization passes again, since the size optimizations
2267 * may have opened new oportunities.
2269 Changes += RunOptFunc (S, &DOptUnusedLoads, 1);
2270 Changes += RunOptFunc (S, &DOptJumpTarget, 5);
2273 /* Adjust branch distances */
2274 Changes += RunOptFunc (S, &DOptBranchDist, 3);
2276 /* Replace conditional branches to RTS. If we had changes, we must run dead
2277 * code elimination again, since the change may have introduced dead code.
2279 C = RunOptFunc (S, &DOptRTSJumps2, 1);
2282 Changes += RunOptFunc (S, &DOptDeadCode, 1);
2285 /* Return the number of changes */
2291 void RunOpt (CodeSeg* S)
2292 /* Run the optimizer */
2294 const char* StatFileName;
2296 /* If we shouldn't run the optimizer, bail out */
2301 /* Check if we are requested to write optimizer statistics */
2302 StatFileName = getenv ("CC65_OPTSTATS");
2304 ReadOptStats (StatFileName);
2307 /* Print the name of the function we are working on */
2309 Print (stdout, 1, "Running optimizer for function `%s'\n", S->Func->Name);
2311 Print (stdout, 1, "Running optimizer for global code segment\n");
2314 /* Run groups of optimizations */
2322 /* Write statistics */
2324 WriteOptStats (StatFileName);