1 /*****************************************************************************/
5 /* Optimizer subroutines */
9 /* (C) 2001-2009 Ullrich von Bassewitz */
10 /* Roemerstrasse 52 */
11 /* D-70794 Filderstadt */
12 /* EMail: uz@cc65.org */
15 /* This software is provided 'as-is', without any expressed or implied */
16 /* warranty. In no event will the authors be held liable for any damages */
17 /* arising from the use of this software. */
19 /* Permission is granted to anyone to use this software for any purpose, */
20 /* including commercial applications, and to alter it and redistribute it */
21 /* freely, subject to the following restrictions: */
23 /* 1. The origin of this software must not be misrepresented; you must not */
24 /* claim that you wrote the original software. If you use this software */
25 /* in a product, an acknowledgment in the product documentation would be */
26 /* appreciated but is not required. */
27 /* 2. Altered source versions must be plainly marked as such, and must not */
28 /* be misrepresented as being the original software. */
29 /* 3. This notice may not be removed or altered from any source */
32 /*****************************************************************************/
55 #include "coptptrload.h"
59 #include "coptstore.h"
68 /*****************************************************************************/
70 /*****************************************************************************/
85 /*****************************************************************************/
87 /*****************************************************************************/
91 static unsigned OptShift1 (CodeSeg* S)
92 /* A call to the shlaxN routine may get replaced by one or more asl insns
93 * if the value of X is not used later. If X is used later, but it is zero
94 * on entry and it's a shift by one, it may get replaced by:
103 unsigned Changes = 0;
105 /* Generate register info */
108 /* Walk over the entries */
110 while (I < CS_GetEntryCount (S)) {
117 CodeEntry* E = CS_GetEntry (S, I);
119 /* Check for the sequence */
120 if (E->OPC == OP65_JSR &&
121 (strncmp (E->Arg, "shlax", 5) == 0 ||
122 strncmp (E->Arg, "aslax", 5) == 0) &&
123 strlen (E->Arg) == 6 &&
124 IsDigit (E->Arg[5])) {
126 if (!RegXUsed (S, I+1)) {
128 /* Insert shift insns */
129 unsigned Count = E->Arg[5] - '0';
131 X = NewCodeEntry (OP65_ASL, AM65_ACC, "a", 0, E->LI);
132 CS_InsertEntry (S, X, I+1);
135 /* Delete the call to shlax */
138 /* Remember, we had changes */
141 } else if (E->RI->In.RegX == 0 &&
143 (N = CS_GetNextEntry (S, I)) != 0) {
146 X = NewCodeEntry (OP65_ASL, AM65_ACC, "a", 0, E->LI);
147 CS_InsertEntry (S, X, I+1);
150 L = CS_GenLabel (S, N);
151 X = NewCodeEntry (OP65_BCC, AM65_BRA, L->Name, L, E->LI);
152 CS_InsertEntry (S, X, I+2);
155 X = NewCodeEntry (OP65_INX, AM65_IMP, 0, 0, E->LI);
156 CS_InsertEntry (S, X, I+3);
158 /* Delete the call to shlax */
161 /* Remember, we had changes */
172 /* Free the register info */
175 /* Return the number of changes made */
181 static unsigned OptShift2 (CodeSeg* S)
182 /* A call to the shraxN routine may get replaced by one or more lsr insns
183 * if the value of X is zero.
186 unsigned Changes = 0;
189 /* Generate register info */
192 /* Walk over the entries */
194 while (I < CS_GetEntryCount (S)) {
197 CodeEntry* E = CS_GetEntry (S, I);
199 /* Check for the sequence */
200 if (E->OPC == OP65_JSR &&
201 strncmp (E->Arg, "shrax", 5) == 0 &&
202 strlen (E->Arg) == 6 &&
203 IsDigit (E->Arg[5]) &&
204 E->RI->In.RegX == 0) {
206 /* Insert shift insns */
207 unsigned Count = E->Arg[5] - '0';
209 CodeEntry* X = NewCodeEntry (OP65_LSR, AM65_ACC, "a", 0, E->LI);
210 CS_InsertEntry (S, X, I+1);
213 /* Delete the call to shlax */
216 /* Remember, we had changes */
226 /* Free the register info */
229 /* Return the number of changes made */
235 static unsigned GetShiftType (const char* Sub)
236 /* Helper function for OptShift3 */
239 if (strcmp (Sub+1, "slax1") == 0) {
241 } else if (strcmp (Sub+1, "srax1") == 0) {
244 } else if (*Sub == 's') {
245 if (strcmp (Sub+1, "hlax1") == 0) {
247 } else if (strcmp (Sub+1, "hrax1") == 0) {
256 static unsigned OptShift3 (CodeSeg* S)
257 /* Search for the sequence
261 * jsr aslax1/asrax1/shlax1/shrax1
274 * or similar, provided that a/x is not used later
277 unsigned Changes = 0;
279 /* Walk over the entries */
281 while (I < CS_GetEntryCount (S)) {
287 L[0] = CS_GetEntry (S, I);
289 /* Check for the sequence */
290 if (L[0]->OPC == OP65_LDA &&
291 (L[0]->AM == AM65_ABS || L[0]->AM == AM65_ZP) &&
292 CS_GetEntries (S, L+1, I+1, 4) &&
293 !CS_RangeHasLabel (S, I+1, 4) &&
294 L[1]->OPC == OP65_LDX &&
295 (L[1]->AM == AM65_ABS || L[1]->AM == AM65_ZP) &&
296 L[2]->OPC == OP65_JSR &&
297 (ShiftType = GetShiftType (L[2]->Arg)) != SHIFT_NONE&&
298 L[3]->OPC == OP65_STA &&
299 (L[3]->AM == AM65_ABS || L[3]->AM == AM65_ZP) &&
300 L[4]->OPC == OP65_STX &&
301 (L[4]->AM == AM65_ABS || L[4]->AM == AM65_ZP) &&
302 !RegAXUsed (S, I+5)) {
306 /* Handle the four shift types differently */
310 X = NewCodeEntry (OP65_LDA, L[1]->AM, L[1]->Arg, 0, L[1]->LI);
311 CS_InsertEntry (S, X, I+5);
312 X = NewCodeEntry (OP65_CMP, AM65_IMM, "$80", 0, L[2]->LI);
313 CS_InsertEntry (S, X, I+6);
314 X = NewCodeEntry (OP65_ROR, AM65_ACC, "a", 0, L[2]->LI);
315 CS_InsertEntry (S, X, I+7);
316 X = NewCodeEntry (OP65_STA, L[4]->AM, L[4]->Arg, 0, L[4]->LI);
317 CS_InsertEntry (S, X, I+8);
318 X = NewCodeEntry (OP65_LDA, L[0]->AM, L[0]->Arg, 0, L[0]->LI);
319 CS_InsertEntry (S, X, I+9);
320 X = NewCodeEntry (OP65_ROR, AM65_ACC, "a", 0, L[2]->LI);
321 CS_InsertEntry (S, X, I+10);
322 X = NewCodeEntry (OP65_STA, L[3]->AM, L[3]->Arg, 0, L[3]->LI);
323 CS_InsertEntry (S, X, I+11);
324 CS_DelEntries (S, I, 5);
328 X = NewCodeEntry (OP65_LDA, L[1]->AM, L[1]->Arg, 0, L[1]->LI);
329 CS_InsertEntry (S, X, I+5);
330 X = NewCodeEntry (OP65_LSR, AM65_ACC, "a", 0, L[2]->LI);
331 CS_InsertEntry (S, X, I+6);
332 X = NewCodeEntry (OP65_STA, L[4]->AM, L[4]->Arg, 0, L[4]->LI);
333 CS_InsertEntry (S, X, I+7);
334 X = NewCodeEntry (OP65_LDA, L[0]->AM, L[0]->Arg, 0, L[0]->LI);
335 CS_InsertEntry (S, X, I+8);
336 X = NewCodeEntry (OP65_ROR, AM65_ACC, "a", 0, L[2]->LI);
337 CS_InsertEntry (S, X, I+9);
338 X = NewCodeEntry (OP65_STA, L[3]->AM, L[3]->Arg, 0, L[3]->LI);
339 CS_InsertEntry (S, X, I+10);
340 CS_DelEntries (S, I, 5);
345 /* These two are identical */
346 X = NewCodeEntry (OP65_ASL, AM65_ACC, "a", 0, L[2]->LI);
347 CS_InsertEntry (S, X, I+1);
348 X = NewCodeEntry (OP65_STA, L[3]->AM, L[3]->Arg, 0, L[3]->LI);
349 CS_InsertEntry (S, X, I+2);
350 X = NewCodeEntry (OP65_LDA, L[1]->AM, L[1]->Arg, 0, L[1]->LI);
351 CS_InsertEntry (S, X, I+3);
352 X = NewCodeEntry (OP65_ROL, AM65_ACC, "a", 0, L[2]->LI);
353 CS_InsertEntry (S, X, I+4);
354 X = NewCodeEntry (OP65_STA, L[4]->AM, L[4]->Arg, 0, L[4]->LI);
355 CS_InsertEntry (S, X, I+5);
356 CS_DelEntries (S, I+6, 4);
361 /* Remember, we had changes */
371 /* Return the number of changes made */
377 static unsigned OptShift4 (CodeSeg* S)
378 /* Inline the shift subroutines. */
380 unsigned Changes = 0;
382 /* Walk over the entries */
384 while (I < CS_GetEntryCount (S)) {
390 CodeEntry* E = CS_GetEntry (S, I);
392 /* Check for a call to one of the shift routine */
393 if (E->OPC == OP65_JSR &&
394 (strncmp (E->Arg, "shlax", 5) == 0 ||
395 strncmp (E->Arg, "aslax", 5) == 0) &&
396 strlen (E->Arg) == 6 &&
397 IsDigit (E->Arg[5])) {
399 /* Get number of shifts */
400 unsigned ShiftCount = (E->Arg[5] - '0');
407 * (repeat ShiftCount-1 times)
410 * which makes 4 + 3 * ShiftCount bytes, compared to the original
411 * 3 bytes for the subroutine call. However, in most cases, the
412 * final load of the X register gets merged with some other insn
413 * and replaces a txa, so for a shift count of 1, we get a factor
414 * of 200, which matches nicely the CodeSizeFactor enabled with -Oi
416 if (ShiftCount > 1 || S->CodeSizeFactor > 200) {
417 unsigned Size = 4 + 3 * ShiftCount;
418 if ((Size * 100 / 3) > S->CodeSizeFactor) {
424 /* Inline the code. Insertion point is behind the subroutine call */
428 X = NewCodeEntry (OP65_STX, AM65_ZP, "tmp1", 0, E->LI);
429 CS_InsertEntry (S, X, IP++);
431 while (ShiftCount--) {
433 X = NewCodeEntry (OP65_ASL, AM65_ACC, "a", 0, E->LI);
434 CS_InsertEntry (S, X, IP++);
437 X = NewCodeEntry (OP65_ROL, AM65_ZP, "tmp1", 0, E->LI);
438 CS_InsertEntry (S, X, IP++);
442 X = NewCodeEntry (OP65_LDX, AM65_ZP, "tmp1", 0, E->LI);
443 CS_InsertEntry (S, X, IP++);
445 /* Remove the subroutine call */
448 /* Remember, we had changes */
458 /* Return the number of changes made */
464 /*****************************************************************************/
466 /*****************************************************************************/
470 static unsigned OptLoad1 (CodeSeg* S)
471 /* Search for a call to ldaxysp where X is not used later and replace it by
472 * a load of just the A register.
476 unsigned Changes = 0;
478 /* Generate register info */
481 /* Walk over the entries */
483 while (I < CS_GetEntryCount (S)) {
488 E = CS_GetEntry (S, I);
490 /* Check for the sequence */
491 if (CE_IsCallTo (E, "ldaxysp") &&
492 RegValIsKnown (E->RI->In.RegY) &&
493 !RegXUsed (S, I+1)) {
497 /* Reload the Y register */
498 const char* Arg = MakeHexArg (E->RI->In.RegY - 1);
499 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
500 CS_InsertEntry (S, X, I+1);
502 /* Load from stack */
503 X = NewCodeEntry (OP65_LDA, AM65_ZP_INDY, "sp", 0, E->LI);
504 CS_InsertEntry (S, X, I+2);
506 /* Now remove the call to the subroutine */
509 /* Remember, we had changes */
519 /* Free the register info */
522 /* Return the number of changes made */
528 /*****************************************************************************/
529 /* Optimize stores through pointers */
530 /*****************************************************************************/
534 static unsigned OptPtrStore1Sub (CodeSeg* S, unsigned I, CodeEntry** const L)
535 /* Check if this is one of the allowed suboperation for OptPtrStore1 */
537 /* Check for a label attached to the entry */
538 if (CE_HasLabel (L[0])) {
542 /* Check for single insn sub ops */
543 if (L[0]->OPC == OP65_AND ||
544 L[0]->OPC == OP65_EOR ||
545 L[0]->OPC == OP65_ORA ||
546 (L[0]->OPC == OP65_JSR && strncmp (L[0]->Arg, "shlax", 5) == 0) ||
547 (L[0]->OPC == OP65_JSR && strncmp (L[0]->Arg, "shrax", 5) == 0)) {
552 } else if (L[0]->OPC == OP65_CLC &&
553 (L[1] = CS_GetNextEntry (S, I)) != 0 &&
554 L[1]->OPC == OP65_ADC &&
555 !CE_HasLabel (L[1])) {
557 } else if (L[0]->OPC == OP65_SEC &&
558 (L[1] = CS_GetNextEntry (S, I)) != 0 &&
559 L[1]->OPC == OP65_SBC &&
560 !CE_HasLabel (L[1])) {
572 static unsigned OptPtrStore1 (CodeSeg* S)
573 /* Search for the sequence:
593 * In case a/x is loaded from the register bank before the pushax, we can even
594 * use the register bank instead of ptr1.
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 (CE_IsCallTo (L[0], "pushax") &&
622 CS_GetEntries (S, L+1, I+1, 3) &&
623 L[1]->OPC == OP65_LDY &&
624 CE_IsConstImm (L[1]) &&
625 !CE_HasLabel (L[1]) &&
626 CE_IsCallTo (L[2], "ldauidx") &&
627 !CE_HasLabel (L[2]) &&
628 (K = OptPtrStore1Sub (S, I+3, L+3)) > 0 &&
629 CS_GetEntries (S, L+3+K, I+3+K, 2) &&
630 L[3+K]->OPC == OP65_LDY &&
631 CE_IsConstImm (L[3+K]) &&
632 !CE_HasLabel (L[3+K]) &&
633 CE_IsCallTo (L[4+K], "staspidx") &&
634 !CE_HasLabel (L[4+K])) {
637 const char* RegBank = 0;
638 const char* ZPLoc = "ptr1";
642 /* Get the preceeding two instructions and check them. We check
649 P[0] = CS_GetEntry (S, I-2);
650 P[1] = CS_GetEntry (S, I-1);
651 if (P[0]->OPC == OP65_LDA &&
652 P[0]->AM == AM65_ZP &&
653 P[1]->OPC == OP65_LDX &&
654 P[1]->AM == AM65_ZP &&
655 !CE_HasLabel (P[1]) &&
656 strncmp (P[0]->Arg, "regbank+", 8) == 0) {
658 unsigned Len = strlen (P[0]->Arg);
660 if (strncmp (P[0]->Arg, P[1]->Arg, Len) == 0 &&
661 P[1]->Arg[Len+0] == '+' &&
662 P[1]->Arg[Len+1] == '1' &&
663 P[1]->Arg[Len+2] == '\0') {
665 /* Ok, found. Use the name of the register bank */
666 RegBank = ZPLoc = P[0]->Arg;
671 /* Insert the load via the zp pointer */
672 X = NewCodeEntry (OP65_LDX, AM65_IMM, "$00", 0, L[3]->LI);
673 CS_InsertEntry (S, X, I+3);
674 X = NewCodeEntry (OP65_LDA, AM65_ZP_INDY, ZPLoc, 0, L[2]->LI);
675 CS_InsertEntry (S, X, I+4);
677 /* Insert the store through the zp pointer */
678 X = NewCodeEntry (OP65_STA, AM65_ZP_INDY, ZPLoc, 0, L[3]->LI);
679 CS_InsertEntry (S, X, I+6+K);
681 /* Delete the old code */
682 CS_DelEntry (S, I+7+K); /* jsr spaspidx */
683 CS_DelEntry (S, I+2); /* jsr ldauidx */
685 /* Create and insert the stores into the zp pointer if needed */
687 X = NewCodeEntry (OP65_STA, AM65_ZP, "ptr1", 0, L[0]->LI);
688 CS_InsertEntry (S, X, I+1);
689 X = NewCodeEntry (OP65_STX, AM65_ZP, "ptr1+1", 0, L[0]->LI);
690 CS_InsertEntry (S, X, I+2);
693 /* Delete more old code. Do it here to keep a label attached to
696 CS_DelEntry (S, I); /* jsr pushax */
698 /* Remember, we had changes */
708 /* Return the number of changes made */
714 static unsigned OptPtrStore2 (CodeSeg* S)
715 /* Search for the sequence:
737 unsigned Changes = 0;
739 /* Walk over the entries */
741 while (I < CS_GetEntryCount (S)) {
747 L[0] = CS_GetEntry (S, I);
749 /* Check for the sequence */
750 if (L[0]->OPC == OP65_LDA &&
751 L[0]->AM == AM65_IMM &&
752 CS_GetEntries (S, L+1, I+1, 10) &&
753 L[1]->OPC == OP65_LDX &&
754 L[1]->AM == AM65_IMM &&
755 L[2]->OPC == OP65_CLC &&
756 L[3]->OPC == OP65_ADC &&
757 (L[3]->AM == AM65_ABS || L[3]->AM == AM65_ZP) &&
758 (L[4]->OPC == OP65_BCC || L[4]->OPC == OP65_JCC) &&
760 L[4]->JumpTo->Owner == L[6] &&
761 L[5]->OPC == OP65_INX &&
762 CE_IsCallTo (L[6], "pushax") &&
763 L[7]->OPC == OP65_LDX &&
764 L[8]->OPC == OP65_LDA &&
765 L[9]->OPC == OP65_LDY &&
766 CE_IsKnownImm (L[9], 0) &&
767 CE_IsCallTo (L[10], "staspidx") &&
768 !CS_RangeHasLabel (S, I+1, 5) &&
769 !CS_RangeHasLabel (S, I+7, 4) &&
770 /* Check the label last because this is quite costly */
771 (Len = strlen (L[0]->Arg)) > 3 &&
772 L[0]->Arg[0] == '<' &&
773 L[0]->Arg[1] == '(' &&
774 strlen (L[1]->Arg) == Len &&
775 L[1]->Arg[0] == '>' &&
776 memcmp (L[0]->Arg+1, L[1]->Arg+1, Len-1) == 0) {
781 /* We will create all the new stuff behind the current one so
782 * we keep the line references.
784 X = NewCodeEntry (OP65_LDY, L[3]->AM, L[3]->Arg, 0, L[0]->LI);
785 CS_InsertEntry (S, X, I+11);
787 X = NewCodeEntry (OP65_LDX, L[7]->AM, L[7]->Arg, 0, L[7]->LI);
788 CS_InsertEntry (S, X, I+12);
790 X = NewCodeEntry (OP65_LDA, L[8]->AM, L[8]->Arg, 0, L[8]->LI);
791 CS_InsertEntry (S, X, I+13);
793 Label = memcpy (xmalloc (Len-2), L[0]->Arg+2, Len-3);
795 X = NewCodeEntry (OP65_STA, AM65_ABSY, Label, 0, L[10]->LI);
796 CS_InsertEntry (S, X, I+14);
799 /* Remove the old code */
800 CS_DelEntries (S, I, 11);
802 /* Remember, we had changes */
812 /* Return the number of changes made */
818 /*****************************************************************************/
819 /* Decouple operations */
820 /*****************************************************************************/
824 static unsigned OptDecouple (CodeSeg* S)
825 /* Decouple operations, that is, do the following replacements:
839 * Provided that the register values are known of course.
842 unsigned Changes = 0;
845 /* Generate register info for the following step */
848 /* Walk over the entries */
850 while (I < CS_GetEntryCount (S)) {
854 /* Get next entry and it's input register values */
855 CodeEntry* E = CS_GetEntry (S, I);
856 const RegContents* In = &E->RI->In;
858 /* Assume we have no replacement */
861 /* Check the instruction */
865 if (RegValIsKnown (In->RegA)) {
866 Arg = MakeHexArg ((In->RegA - 1) & 0xFF);
867 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
872 if (RegValIsKnown (In->RegX)) {
873 Arg = MakeHexArg ((In->RegX - 1) & 0xFF);
874 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
879 if (RegValIsKnown (In->RegY)) {
880 Arg = MakeHexArg ((In->RegY - 1) & 0xFF);
881 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
886 if (RegValIsKnown (In->RegA)) {
887 Arg = MakeHexArg ((In->RegA + 1) & 0xFF);
888 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
893 if (RegValIsKnown (In->RegX)) {
894 Arg = MakeHexArg ((In->RegX + 1) & 0xFF);
895 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
900 if (RegValIsKnown (In->RegY)) {
901 Arg = MakeHexArg ((In->RegY + 1) & 0xFF);
902 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
907 if (E->AM == AM65_ZP) {
908 switch (GetKnownReg (E->Use & REG_ZP, In)) {
910 Arg = MakeHexArg (In->Tmp1);
911 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
915 Arg = MakeHexArg (In->Ptr1Lo);
916 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
920 Arg = MakeHexArg (In->Ptr1Hi);
921 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
925 Arg = MakeHexArg (In->SRegLo);
926 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
930 Arg = MakeHexArg (In->SRegHi);
931 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
938 if (E->AM == AM65_ZP) {
939 switch (GetKnownReg (E->Use & REG_ZP, In)) {
941 Arg = MakeHexArg (In->Tmp1);
942 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
946 Arg = MakeHexArg (In->Ptr1Lo);
947 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
951 Arg = MakeHexArg (In->Ptr1Hi);
952 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
956 Arg = MakeHexArg (In->SRegLo);
957 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
961 Arg = MakeHexArg (In->SRegHi);
962 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
969 if (E->AM == AM65_ZP) {
970 switch (GetKnownReg (E->Use, In)) {
972 Arg = MakeHexArg (In->Tmp1);
973 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
977 Arg = MakeHexArg (In->Ptr1Lo);
978 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
982 Arg = MakeHexArg (In->Ptr1Hi);
983 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
987 Arg = MakeHexArg (In->SRegLo);
988 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
992 Arg = MakeHexArg (In->SRegHi);
993 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
1000 if (E->RI->In.RegA >= 0) {
1001 Arg = MakeHexArg (In->RegA);
1002 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, E->LI);
1007 if (E->RI->In.RegA >= 0) {
1008 Arg = MakeHexArg (In->RegA);
1009 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
1014 if (E->RI->In.RegX >= 0) {
1015 Arg = MakeHexArg (In->RegX);
1016 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
1021 if (E->RI->In.RegY >= 0) {
1022 Arg = MakeHexArg (In->RegY);
1023 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, E->LI);
1028 /* Avoid gcc warnings */
1033 /* Insert the replacement if we have one */
1035 CS_InsertEntry (S, X, I+1);
1045 /* Free register info */
1048 /* Return the number of changes made */
1054 /*****************************************************************************/
1055 /* struct OptFunc */
1056 /*****************************************************************************/
1060 typedef struct OptFunc OptFunc;
1062 unsigned (*Func) (CodeSeg*); /* Optimizer function */
1063 const char* Name; /* Name of the function/group */
1064 unsigned CodeSizeFactor; /* Code size factor for this opt func */
1065 unsigned long TotalRuns; /* Total number of runs */
1066 unsigned long LastRuns; /* Last number of runs */
1067 unsigned long TotalChanges; /* Total number of changes */
1068 unsigned long LastChanges; /* Last number of changes */
1069 char Disabled; /* True if function disabled */
1074 /*****************************************************************************/
1076 /*****************************************************************************/
1080 /* A list of all the function descriptions */
1081 static OptFunc DOpt65C02BitOps = { Opt65C02BitOps, "Opt65C02BitOps", 66, 0, 0, 0, 0, 0 };
1082 static OptFunc DOpt65C02Ind = { Opt65C02Ind, "Opt65C02Ind", 100, 0, 0, 0, 0, 0 };
1083 static OptFunc DOpt65C02Stores = { Opt65C02Stores, "Opt65C02Stores", 100, 0, 0, 0, 0, 0 };
1084 static OptFunc DOptAdd1 = { OptAdd1, "OptAdd1", 125, 0, 0, 0, 0, 0 };
1085 static OptFunc DOptAdd2 = { OptAdd2, "OptAdd2", 200, 0, 0, 0, 0, 0 };
1086 static OptFunc DOptAdd3 = { OptAdd3, "OptAdd3", 65, 0, 0, 0, 0, 0 };
1087 static OptFunc DOptAdd4 = { OptAdd4, "OptAdd4", 90, 0, 0, 0, 0, 0 };
1088 static OptFunc DOptAdd5 = { OptAdd5, "OptAdd5", 100, 0, 0, 0, 0, 0 };
1089 static OptFunc DOptAdd6 = { OptAdd6, "OptAdd6", 40, 0, 0, 0, 0, 0 };
1090 static OptFunc DOptBoolTrans = { OptBoolTrans, "OptBoolTrans", 100, 0, 0, 0, 0, 0 };
1091 static OptFunc DOptBranchDist = { OptBranchDist, "OptBranchDist", 0, 0, 0, 0, 0, 0 };
1092 static OptFunc DOptCmp1 = { OptCmp1, "OptCmp1", 42, 0, 0, 0, 0, 0 };
1093 static OptFunc DOptCmp2 = { OptCmp2, "OptCmp2", 85, 0, 0, 0, 0, 0 };
1094 static OptFunc DOptCmp3 = { OptCmp3, "OptCmp3", 75, 0, 0, 0, 0, 0 };
1095 static OptFunc DOptCmp4 = { OptCmp4, "OptCmp4", 75, 0, 0, 0, 0, 0 };
1096 static OptFunc DOptCmp5 = { OptCmp5, "OptCmp5", 100, 0, 0, 0, 0, 0 };
1097 static OptFunc DOptCmp6 = { OptCmp6, "OptCmp6", 100, 0, 0, 0, 0, 0 };
1098 static OptFunc DOptCmp7 = { OptCmp7, "OptCmp7", 85, 0, 0, 0, 0, 0 };
1099 static OptFunc DOptCmp8 = { OptCmp8, "OptCmp8", 50, 0, 0, 0, 0, 0 };
1100 static OptFunc DOptCmp9 = { OptCmp9, "OptCmp9", 85, 0, 0, 0, 0, 0 };
1101 static OptFunc DOptCondBranches1= { OptCondBranches1,"OptCondBranches1", 80, 0, 0, 0, 0, 0 };
1102 static OptFunc DOptCondBranches2= { OptCondBranches2,"OptCondBranches2", 0, 0, 0, 0, 0, 0 };
1103 static OptFunc DOptDeadCode = { OptDeadCode, "OptDeadCode", 100, 0, 0, 0, 0, 0 };
1104 static OptFunc DOptDeadJumps = { OptDeadJumps, "OptDeadJumps", 100, 0, 0, 0, 0, 0 };
1105 static OptFunc DOptDecouple = { OptDecouple, "OptDecouple", 100, 0, 0, 0, 0, 0 };
1106 static OptFunc DOptDupLoads = { OptDupLoads, "OptDupLoads", 0, 0, 0, 0, 0, 0 };
1107 static OptFunc DOptJumpCascades = { OptJumpCascades, "OptJumpCascades", 100, 0, 0, 0, 0, 0 };
1108 static OptFunc DOptJumpTarget1 = { OptJumpTarget1, "OptJumpTarget1", 100, 0, 0, 0, 0, 0 };
1109 static OptFunc DOptJumpTarget2 = { OptJumpTarget2, "OptJumpTarget2", 100, 0, 0, 0, 0, 0 };
1110 static OptFunc DOptLoad1 = { OptLoad1, "OptLoad1", 100, 0, 0, 0, 0, 0 };
1111 static OptFunc DOptRTS = { OptRTS, "OptRTS", 100, 0, 0, 0, 0, 0 };
1112 static OptFunc DOptRTSJumps1 = { OptRTSJumps1, "OptRTSJumps1", 100, 0, 0, 0, 0, 0 };
1113 static OptFunc DOptRTSJumps2 = { OptRTSJumps2, "OptRTSJumps2", 100, 0, 0, 0, 0, 0 };
1114 static OptFunc DOptNegA1 = { OptNegA1, "OptNegA1", 100, 0, 0, 0, 0, 0 };
1115 static OptFunc DOptNegA2 = { OptNegA2, "OptNegA2", 100, 0, 0, 0, 0, 0 };
1116 static OptFunc DOptNegAX1 = { OptNegAX1, "OptNegAX1", 100, 0, 0, 0, 0, 0 };
1117 static OptFunc DOptNegAX2 = { OptNegAX2, "OptNegAX2", 100, 0, 0, 0, 0, 0 };
1118 static OptFunc DOptNegAX3 = { OptNegAX3, "OptNegAX3", 100, 0, 0, 0, 0, 0 };
1119 static OptFunc DOptNegAX4 = { OptNegAX4, "OptNegAX4", 100, 0, 0, 0, 0, 0 };
1120 static OptFunc DOptPrecalc = { OptPrecalc, "OptPrecalc", 100, 0, 0, 0, 0, 0 };
1121 static OptFunc DOptPtrLoad1 = { OptPtrLoad1, "OptPtrLoad1", 100, 0, 0, 0, 0, 0 };
1122 static OptFunc DOptPtrLoad2 = { OptPtrLoad2, "OptPtrLoad2", 100, 0, 0, 0, 0, 0 };
1123 static OptFunc DOptPtrLoad3 = { OptPtrLoad3, "OptPtrLoad3", 100, 0, 0, 0, 0, 0 };
1124 static OptFunc DOptPtrLoad4 = { OptPtrLoad4, "OptPtrLoad4", 100, 0, 0, 0, 0, 0 };
1125 static OptFunc DOptPtrLoad5 = { OptPtrLoad5, "OptPtrLoad5", 50, 0, 0, 0, 0, 0 };
1126 static OptFunc DOptPtrLoad6 = { OptPtrLoad6, "OptPtrLoad6", 60, 0, 0, 0, 0, 0 };
1127 static OptFunc DOptPtrLoad7 = { OptPtrLoad7, "OptPtrLoad7", 140, 0, 0, 0, 0, 0 };
1128 static OptFunc DOptPtrLoad11 = { OptPtrLoad11, "OptPtrLoad11", 92, 0, 0, 0, 0, 0 };
1129 static OptFunc DOptPtrLoad12 = { OptPtrLoad12, "OptPtrLoad12", 50, 0, 0, 0, 0, 0 };
1130 static OptFunc DOptPtrLoad13 = { OptPtrLoad13, "OptPtrLoad13", 65, 0, 0, 0, 0, 0 };
1131 static OptFunc DOptPtrLoad14 = { OptPtrLoad14, "OptPtrLoad14", 108, 0, 0, 0, 0, 0 };
1132 static OptFunc DOptPtrLoad15 = { OptPtrLoad15, "OptPtrLoad15", 86, 0, 0, 0, 0, 0 };
1133 static OptFunc DOptPtrLoad16 = { OptPtrLoad16, "OptPtrLoad16", 100, 0, 0, 0, 0, 0 };
1134 static OptFunc DOptPtrLoad17 = { OptPtrLoad17, "OptPtrLoad17", 190, 0, 0, 0, 0, 0 };
1135 static OptFunc DOptPtrStore1 = { OptPtrStore1, "OptPtrStore1", 100, 0, 0, 0, 0, 0 };
1136 static OptFunc DOptPtrStore2 = { OptPtrStore2, "OptPtrStore2", 40, 0, 0, 0, 0, 0 };
1137 static OptFunc DOptPush1 = { OptPush1, "OptPush1", 65, 0, 0, 0, 0, 0 };
1138 static OptFunc DOptPush2 = { OptPush2, "OptPush2", 50, 0, 0, 0, 0, 0 };
1139 static OptFunc DOptPushPop = { OptPushPop, "OptPushPop", 0, 0, 0, 0, 0, 0 };
1140 static OptFunc DOptShift1 = { OptShift1, "OptShift1", 100, 0, 0, 0, 0, 0 };
1141 static OptFunc DOptShift2 = { OptShift2, "OptShift2", 100, 0, 0, 0, 0, 0 };
1142 static OptFunc DOptShift3 = { OptShift3, "OptShift3", 110, 0, 0, 0, 0, 0 };
1143 static OptFunc DOptShift4 = { OptShift4, "OptShift4", 200, 0, 0, 0, 0, 0 };
1144 static OptFunc DOptSize1 = { OptSize1, "OptSize1", 100, 0, 0, 0, 0, 0 };
1145 static OptFunc DOptSize2 = { OptSize2, "OptSize2", 100, 0, 0, 0, 0, 0 };
1146 static OptFunc DOptStackOps = { OptStackOps, "OptStackOps", 100, 0, 0, 0, 0, 0 };
1147 static OptFunc DOptStore1 = { OptStore1, "OptStore1", 70, 0, 0, 0, 0, 0 };
1148 static OptFunc DOptStore2 = { OptStore2, "OptStore2", 220, 0, 0, 0, 0, 0 };
1149 static OptFunc DOptStore3 = { OptStore3, "OptStore3", 120, 0, 0, 0, 0, 0 };
1150 static OptFunc DOptStore4 = { OptStore4, "OptStore4", 50, 0, 0, 0, 0, 0 };
1151 static OptFunc DOptStore5 = { OptStore5, "OptStore5", 100, 0, 0, 0, 0, 0 };
1152 static OptFunc DOptStoreLoad = { OptStoreLoad, "OptStoreLoad", 0, 0, 0, 0, 0, 0 };
1153 static OptFunc DOptSub1 = { OptSub1, "OptSub1", 100, 0, 0, 0, 0, 0 };
1154 static OptFunc DOptSub2 = { OptSub2, "OptSub2", 100, 0, 0, 0, 0, 0 };
1155 static OptFunc DOptSub3 = { OptSub3, "OptSub3", 100, 0, 0, 0, 0, 0 };
1156 static OptFunc DOptTest1 = { OptTest1, "OptTest1", 100, 0, 0, 0, 0, 0 };
1157 static OptFunc DOptTransfers1 = { OptTransfers1, "OptTransfers1", 0, 0, 0, 0, 0, 0 };
1158 static OptFunc DOptTransfers2 = { OptTransfers2, "OptTransfers2", 60, 0, 0, 0, 0, 0 };
1159 static OptFunc DOptTransfers3 = { OptTransfers3, "OptTransfers3", 65, 0, 0, 0, 0, 0 };
1160 static OptFunc DOptTransfers4 = { OptTransfers4, "OptTransfers4", 65, 0, 0, 0, 0, 0 };
1161 static OptFunc DOptUnusedLoads = { OptUnusedLoads, "OptUnusedLoads", 0, 0, 0, 0, 0, 0 };
1162 static OptFunc DOptUnusedStores = { OptUnusedStores, "OptUnusedStores", 0, 0, 0, 0, 0, 0 };
1165 /* Table containing all the steps in alphabetical order */
1166 static OptFunc* OptFuncs[] = {
1250 #define OPTFUNC_COUNT (sizeof(OptFuncs) / sizeof(OptFuncs[0]))
1254 static int CmpOptStep (const void* Key, const void* Func)
1255 /* Compare function for bsearch */
1257 return strcmp (Key, (*(const OptFunc**)Func)->Name);
1262 static OptFunc* FindOptFunc (const char* Name)
1263 /* Find an optimizer step by name in the table and return a pointer. Return
1264 * NULL if no such step is found.
1267 /* Search for the function in the list */
1268 OptFunc** O = bsearch (Name, OptFuncs, OPTFUNC_COUNT, sizeof (OptFuncs[0]), CmpOptStep);
1274 static OptFunc* GetOptFunc (const char* Name)
1275 /* Find an optimizer step by name in the table and return a pointer. Print an
1276 * error and call AbEnd if not found.
1279 /* Search for the function in the list */
1280 OptFunc* F = FindOptFunc (Name);
1283 AbEnd ("Optimization step `%s' not found", Name);
1290 void DisableOpt (const char* Name)
1291 /* Disable the optimization with the given name */
1293 if (strcmp (Name, "any") == 0) {
1295 for (I = 0; I < OPTFUNC_COUNT; ++I) {
1296 OptFuncs[I]->Disabled = 1;
1299 GetOptFunc(Name)->Disabled = 1;
1305 void EnableOpt (const char* Name)
1306 /* Enable the optimization with the given name */
1308 if (strcmp (Name, "any") == 0) {
1310 for (I = 0; I < OPTFUNC_COUNT; ++I) {
1311 OptFuncs[I]->Disabled = 0;
1314 GetOptFunc(Name)->Disabled = 0;
1320 void ListOptSteps (FILE* F)
1321 /* List all optimization steps */
1324 for (I = 0; I < OPTFUNC_COUNT; ++I) {
1325 fprintf (F, "%s\n", OptFuncs[I]->Name);
1331 static void ReadOptStats (const char* Name)
1332 /* Read the optimizer statistics file */
1337 /* Try to open the file */
1338 FILE* F = fopen (Name, "r");
1340 /* Ignore the error */
1344 /* Read and parse the lines */
1346 while (fgets (Buf, sizeof (Buf), F) != 0) {
1354 unsigned long TotalRuns;
1355 unsigned long TotalChanges;
1360 /* Remove trailing white space including the line terminator */
1363 while (Len > 0 && IsSpace (B[Len-1])) {
1368 /* Remove leading whitespace */
1369 while (IsSpace (*B)) {
1373 /* Check for empty and comment lines */
1374 if (*B == '\0' || *B == ';' || *B == '#') {
1378 /* Parse the line */
1379 if (sscanf (B, "%31s %lu %*u %lu %*u", Name, &TotalRuns, &TotalChanges) != 3) {
1384 /* Search for the optimizer step. */
1385 Func = FindOptFunc (Name);
1391 /* Found the step, set the fields */
1392 Func->TotalRuns = TotalRuns;
1393 Func->TotalChanges = TotalChanges;
1397 /* Close the file, ignore errors here. */
1403 static void WriteOptStats (const char* Name)
1404 /* Write the optimizer statistics file */
1408 /* Try to open the file */
1409 FILE* F = fopen (Name, "w");
1411 /* Ignore the error */
1415 /* Write a header */
1417 "; Optimizer Total Last Total Last\n"
1418 "; Step Runs Runs Chg Chg\n");
1421 /* Write the data */
1422 for (I = 0; I < OPTFUNC_COUNT; ++I) {
1423 const OptFunc* O = OptFuncs[I];
1425 "%-20s %10lu %10lu %10lu %10lu\n",
1433 /* Close the file, ignore errors here. */
1439 static unsigned RunOptFunc (CodeSeg* S, OptFunc* F, unsigned Max)
1440 /* Run one optimizer function Max times or until there are no more changes */
1442 unsigned Changes, C;
1444 /* Don't run the function if it is disabled or if it is prohibited by the
1447 if (F->Disabled || F->CodeSizeFactor > S->CodeSizeFactor) {
1451 /* Run this until there are no more changes */
1455 /* Run the function */
1462 F->TotalChanges += C;
1463 F->LastChanges += C;
1465 } while (--Max && C > 0);
1467 /* Return the number of changes */
1473 static unsigned RunOptGroup1 (CodeSeg* S)
1474 /* Run the first group of optimization steps. These steps translate known
1475 * patterns emitted by the code generator into more optimal patterns. Order
1476 * of the steps is important, because some of the steps done earlier cover
1477 * the same patterns as later steps as subpatterns.
1480 unsigned Changes = 0;
1482 Changes += RunOptFunc (S, &DOptPtrStore1, 1);
1483 Changes += RunOptFunc (S, &DOptPtrStore2, 1);
1484 Changes += RunOptFunc (S, &DOptAdd3, 1); /* Before OptPtrLoad5! */
1485 Changes += RunOptFunc (S, &DOptPtrLoad1, 1);
1486 Changes += RunOptFunc (S, &DOptPtrLoad2, 1);
1487 Changes += RunOptFunc (S, &DOptPtrLoad3, 1);
1488 Changes += RunOptFunc (S, &DOptPtrLoad4, 1);
1489 Changes += RunOptFunc (S, &DOptPtrLoad5, 1);
1490 Changes += RunOptFunc (S, &DOptPtrLoad6, 1);
1491 Changes += RunOptFunc (S, &DOptPtrLoad7, 1);
1492 Changes += RunOptFunc (S, &DOptPtrLoad11, 1);
1493 Changes += RunOptFunc (S, &DOptPtrLoad12, 1);
1494 Changes += RunOptFunc (S, &DOptPtrLoad13, 1);
1495 Changes += RunOptFunc (S, &DOptPtrLoad14, 1);
1496 Changes += RunOptFunc (S, &DOptPtrLoad15, 1);
1497 Changes += RunOptFunc (S, &DOptPtrLoad16, 1);
1498 Changes += RunOptFunc (S, &DOptPtrLoad17, 1);
1499 Changes += RunOptFunc (S, &DOptNegAX1, 1);
1500 Changes += RunOptFunc (S, &DOptNegAX2, 1);
1501 Changes += RunOptFunc (S, &DOptNegAX3, 1);
1502 Changes += RunOptFunc (S, &DOptNegAX4, 1);
1503 Changes += RunOptFunc (S, &DOptAdd1, 1);
1504 Changes += RunOptFunc (S, &DOptAdd2, 1);
1505 Changes += RunOptFunc (S, &DOptAdd4, 1);
1506 Changes += RunOptFunc (S, &DOptStore4, 1);
1507 Changes += RunOptFunc (S, &DOptStore5, 1);
1508 Changes += RunOptFunc (S, &DOptShift1, 1);
1509 Changes += RunOptFunc (S, &DOptShift2, 1);
1510 Changes += RunOptFunc (S, &DOptShift3, 1);
1511 Changes += RunOptFunc (S, &DOptShift4, 1);
1512 Changes += RunOptFunc (S, &DOptStore1, 1);
1513 Changes += RunOptFunc (S, &DOptStore2, 5);
1514 Changes += RunOptFunc (S, &DOptStore3, 5);
1516 /* Return the number of changes */
1522 static unsigned RunOptGroup2 (CodeSeg* S)
1523 /* Run one group of optimization steps. This step involves just decoupling
1524 * instructions by replacing them by instructions that do not depend on
1525 * previous instructions. This makes it easier to find instructions that
1529 unsigned Changes = 0;
1531 Changes += RunOptFunc (S, &DOptDecouple, 1);
1533 /* Return the number of changes */
1539 static unsigned RunOptGroup3 (CodeSeg* S)
1540 /* Run one group of optimization steps. These steps depend on each other,
1541 * that means that one step may allow another step to do additional work,
1542 * so we will repeat the steps as long as we see any changes.
1545 unsigned Changes, C;
1551 C += RunOptFunc (S, &DOptNegA1, 1);
1552 C += RunOptFunc (S, &DOptNegA2, 1);
1553 C += RunOptFunc (S, &DOptSub1, 1);
1554 C += RunOptFunc (S, &DOptSub2, 1);
1555 C += RunOptFunc (S, &DOptSub3, 1);
1556 C += RunOptFunc (S, &DOptAdd5, 1);
1557 C += RunOptFunc (S, &DOptAdd6, 1);
1558 C += RunOptFunc (S, &DOptStackOps, 1);
1559 C += RunOptFunc (S, &DOptJumpCascades, 1);
1560 C += RunOptFunc (S, &DOptDeadJumps, 1);
1561 C += RunOptFunc (S, &DOptRTS, 1);
1562 C += RunOptFunc (S, &DOptDeadCode, 1);
1563 C += RunOptFunc (S, &DOptBoolTrans, 1);
1564 C += RunOptFunc (S, &DOptJumpTarget1, 1);
1565 C += RunOptFunc (S, &DOptJumpTarget2, 1);
1566 C += RunOptFunc (S, &DOptCondBranches1, 1);
1567 C += RunOptFunc (S, &DOptCondBranches2, 1);
1568 C += RunOptFunc (S, &DOptRTSJumps1, 1);
1569 C += RunOptFunc (S, &DOptCmp1, 1);
1570 C += RunOptFunc (S, &DOptCmp2, 1);
1571 C += RunOptFunc (S, &DOptCmp3, 1);
1572 C += RunOptFunc (S, &DOptCmp4, 1);
1573 C += RunOptFunc (S, &DOptCmp5, 1);
1574 C += RunOptFunc (S, &DOptCmp6, 1);
1575 C += RunOptFunc (S, &DOptCmp7, 1);
1576 C += RunOptFunc (S, &DOptCmp8, 1);
1577 C += RunOptFunc (S, &DOptCmp9, 1);
1578 C += RunOptFunc (S, &DOptTest1, 1);
1579 C += RunOptFunc (S, &DOptLoad1, 1);
1580 C += RunOptFunc (S, &DOptUnusedLoads, 1);
1581 C += RunOptFunc (S, &DOptUnusedStores, 1);
1582 C += RunOptFunc (S, &DOptDupLoads, 1);
1583 C += RunOptFunc (S, &DOptStoreLoad, 1);
1584 C += RunOptFunc (S, &DOptTransfers1, 1);
1585 C += RunOptFunc (S, &DOptTransfers3, 1);
1586 C += RunOptFunc (S, &DOptTransfers4, 1);
1587 C += RunOptFunc (S, &DOptStore1, 1);
1588 C += RunOptFunc (S, &DOptStore5, 1);
1589 C += RunOptFunc (S, &DOptPushPop, 1);
1590 C += RunOptFunc (S, &DOptPrecalc, 1);
1596 /* Return the number of changes */
1602 static unsigned RunOptGroup4 (CodeSeg* S)
1603 /* 65C02 specific optimizations. */
1605 unsigned Changes = 0;
1607 if (CPUIsets[CPU] & CPU_ISET_65SC02) {
1608 Changes += RunOptFunc (S, &DOpt65C02BitOps, 1);
1609 Changes += RunOptFunc (S, &DOpt65C02Ind, 1);
1610 Changes += RunOptFunc (S, &DOpt65C02Stores, 1);
1612 /* The 65C02 replacement codes do often make the use of a register
1613 * value unnecessary, so if we have changes, run another load
1616 Changes += RunOptFunc (S, &DOptUnusedLoads, 1);
1620 /* Return the number of changes */
1626 static unsigned RunOptGroup5 (CodeSeg* S)
1627 /* Run another round of pattern replacements. These are done late, since there
1628 * may be better replacements before.
1631 unsigned Changes = 0;
1633 Changes += RunOptFunc (S, &DOptPush1, 1);
1634 Changes += RunOptFunc (S, &DOptPush2, 1);
1635 /* Repeat some of the other optimizations now */
1636 Changes += RunOptFunc (S, &DOptUnusedLoads, 1);
1637 Changes += RunOptFunc (S, &DOptTransfers2, 1);
1639 /* Return the number of changes */
1645 static unsigned RunOptGroup6 (CodeSeg* S)
1646 /* The last group of optimization steps. Adjust branches, do size optimizations.
1649 unsigned Changes = 0;
1652 /* Optimize for size, that is replace operations by shorter ones, even
1653 * if this does hinder further optimizations (no problem since we're
1656 C = RunOptFunc (S, &DOptSize1, 1);
1659 /* Run some optimization passes again, since the size optimizations
1660 * may have opened new oportunities.
1662 Changes += RunOptFunc (S, &DOptUnusedLoads, 1);
1663 Changes += RunOptFunc (S, &DOptUnusedStores, 1);
1664 Changes += RunOptFunc (S, &DOptJumpTarget1, 5);
1665 Changes += RunOptFunc (S, &DOptStore5, 1);
1668 C = RunOptFunc (S, &DOptSize2, 1);
1671 /* Run some optimization passes again, since the size optimizations
1672 * may have opened new oportunities.
1674 Changes += RunOptFunc (S, &DOptUnusedLoads, 1);
1675 Changes += RunOptFunc (S, &DOptJumpTarget1, 5);
1676 Changes += RunOptFunc (S, &DOptStore5, 1);
1677 Changes += RunOptFunc (S, &DOptTransfers3, 1);
1680 /* Adjust branch distances */
1681 Changes += RunOptFunc (S, &DOptBranchDist, 3);
1683 /* Replace conditional branches to RTS. If we had changes, we must run dead
1684 * code elimination again, since the change may have introduced dead code.
1686 C = RunOptFunc (S, &DOptRTSJumps2, 1);
1689 Changes += RunOptFunc (S, &DOptDeadCode, 1);
1692 /* Return the number of changes */
1698 void RunOpt (CodeSeg* S)
1699 /* Run the optimizer */
1701 const char* StatFileName;
1703 /* If we shouldn't run the optimizer, bail out */
1708 /* Check if we are requested to write optimizer statistics */
1709 StatFileName = getenv ("CC65_OPTSTATS");
1711 ReadOptStats (StatFileName);
1714 /* Print the name of the function we are working on */
1716 Print (stdout, 1, "Running optimizer for function `%s'\n", S->Func->Name);
1718 Print (stdout, 1, "Running optimizer for global code segment\n");
1721 /* Run groups of optimizations */
1729 /* Write statistics */
1731 WriteOptStats (StatFileName);