1 /*****************************************************************************/
5 /* Optimize operations that take operands via the stack */
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 /*****************************************************************************/
48 /*****************************************************************************/
49 /* Load tracking data */
50 /*****************************************************************************/
54 /* LoadRegInfo flags set by DirectOp */
57 LI_DIRECT = 0x01, /* Direct op may be used */
58 LI_RELOAD_Y = 0x02, /* Reload index register Y */
59 LI_REMOVE = 0x04, /* Load may be removed */
62 /* Structure that tells us how to load the lhs values */
63 typedef struct LoadRegInfo LoadRegInfo;
65 int LoadIndex; /* Index of load insn, -1 if invalid */
66 CodeEntry* LoadEntry; /* The actual entry, 0 if invalid */
67 LI_FLAGS Flags; /* Tells us how to load */
68 unsigned char Offs; /* Stack offset if data is on stack */
71 /* Now combined for both registers */
72 typedef struct LoadInfo LoadInfo;
74 LoadRegInfo A; /* Info for A register */
75 LoadRegInfo X; /* Info for X register */
76 LoadRegInfo Y; /* Info for Y register */
81 /*****************************************************************************/
83 /*****************************************************************************/
87 /* Flags for the functions */
89 OP_NONE = 0x00, /* Nothing special */
90 OP_A_KNOWN = 0x01, /* Value of A must be known */
91 OP_X_ZERO = 0x02, /* X must be zero */
92 OP_LHS_LOAD = 0x04, /* Must have load insns for LHS */
93 OP_LHS_LOAD_DIRECT = 0x0C, /* Must have direct load insn for LHS */
94 OP_RHS_LOAD = 0x10, /* Must have load insns for RHS */
95 OP_RHS_LOAD_DIRECT = 0x30, /* Must have direct load insn for RHS */
98 /* Structure forward decl */
99 typedef struct StackOpData StackOpData;
101 /* Structure that describes an optimizer subfunction for a specific op */
102 typedef unsigned (*OptFunc) (StackOpData* D);
103 typedef struct OptFuncDesc OptFuncDesc;
105 const char* Name; /* Name of the replaced runtime function */
106 OptFunc Func; /* Function pointer */
107 unsigned UnusedRegs; /* Regs that must not be used later */
108 OP_FLAGS Flags; /* Flags */
111 /* Structure that holds the needed data */
113 CodeSeg* Code; /* Pointer to code segment */
114 unsigned Flags; /* Flags to remember things */
116 /* Pointer to optimizer subfunction description */
117 const OptFuncDesc* OptFunc;
119 /* ZP register usage inside the sequence */
122 /* Register load information for lhs and rhs */
126 /* Several indices of insns in the code segment */
127 int PushIndex; /* Index of call to pushax in codeseg */
128 int OpIndex; /* Index of actual operation */
130 /* Pointers to insns in the code segment */
131 CodeEntry* PrevEntry; /* Entry before the call to pushax */
132 CodeEntry* PushEntry; /* Pointer to entry with call to pushax */
133 CodeEntry* OpEntry; /* Pointer to entry with op */
134 CodeEntry* NextEntry; /* Entry after the op */
136 const char* ZPLo; /* Lo byte of zero page loc to use */
137 const char* ZPHi; /* Hi byte of zero page loc to use */
138 unsigned IP; /* Insertion point used by some routines */
143 /*****************************************************************************/
144 /* Load tracking code */
145 /*****************************************************************************/
149 static void ClearLoadRegInfo (LoadRegInfo* RI)
150 /* Clear a LoadRegInfo struct */
157 static void FinalizeLoadRegInfo (LoadRegInfo* RI, CodeSeg* S)
158 /* Prepare a LoadRegInfo struct for use */
161 if (RI->LoadIndex >= 0) {
162 RI->LoadEntry = CS_GetEntry (S, RI->LoadIndex);
171 static void ClearLoadInfo (LoadInfo* LI)
172 /* Clear a LoadInfo struct */
174 ClearLoadRegInfo (&LI->A);
175 ClearLoadRegInfo (&LI->X);
176 ClearLoadRegInfo (&LI->Y);
181 static void AdjustLoadRegInfo (LoadRegInfo* RI, int DelIndex, int Change)
182 /* Adjust a load register info struct after deleting or inserting an entry
186 CHECK (abs (Change) == 1);
189 if (DelIndex < RI->LoadIndex) {
191 } else if (DelIndex == RI->LoadIndex) {
192 /* Has been removed */
198 if (DelIndex <= RI->LoadIndex) {
206 static void FinalizeLoadInfo (LoadInfo* LI, CodeSeg* S)
207 /* Prepare a LoadInfo struct for use */
209 /* Get the entries */
210 FinalizeLoadRegInfo (&LI->A, S);
211 FinalizeLoadRegInfo (&LI->X, S);
212 FinalizeLoadRegInfo (&LI->Y, S);
217 static void AdjustLoadInfo (LoadInfo* LI, int DelIndex, int Change)
218 /* Adjust a load info struct after deleting entry with a given index */
220 AdjustLoadRegInfo (&LI->A, DelIndex, Change);
221 AdjustLoadRegInfo (&LI->X, DelIndex, Change);
222 AdjustLoadRegInfo (&LI->Y, DelIndex, Change);
227 static void TrackLoads (LoadInfo* LI, CodeEntry* E, int I)
228 /* Track loads for a code entry */
230 if (E->Info & OF_LOAD) {
231 if (E->Chg & REG_A) {
234 if (E->Chg & REG_X) {
237 if (E->Chg & REG_Y) {
240 } else if (E->Info & OF_XFR) {
242 case OP65_TAX: LI->X.LoadIndex = LI->A.LoadIndex; break;
243 case OP65_TAY: LI->Y.LoadIndex = LI->A.LoadIndex; break;
244 case OP65_TXA: LI->A.LoadIndex = LI->X.LoadIndex; break;
245 case OP65_TYA: LI->A.LoadIndex = LI->Y.LoadIndex; break;
248 } else if (CE_IsCallTo (E, "ldaxysp")) {
249 /* Both registers set, Y changed */
252 LI->Y.LoadIndex = -1;
254 if (E->Chg & REG_A) {
255 LI->A.LoadIndex = -1;
257 if (E->Chg & REG_X) {
258 LI->X.LoadIndex = -1;
260 if (E->Chg & REG_Y) {
261 LI->Y.LoadIndex = -1;
268 /*****************************************************************************/
270 /*****************************************************************************/
274 static void AdjustStackOffset (StackOpData* D, unsigned Offs)
275 /* Adjust the offset for all stack accesses in the range PushIndex to OpIndex.
276 * OpIndex is adjusted according to the insertions.
279 /* Walk over all entries */
280 int I = D->PushIndex + 1;
281 while (I < D->OpIndex) {
283 CodeEntry* E = CS_GetEntry (D->Code, I);
285 int NeedCorrection = 0;
286 if ((E->Use & REG_SP) != 0) {
288 /* Check for some things that should not happen */
289 CHECK (E->AM == AM65_ZP_INDY || E->RI->In.RegY >= (short) Offs);
290 CHECK (strcmp (E->Arg, "sp") == 0);
292 /* We need to correct this one */
295 } else if (CE_IsCallTo (E, "ldaxysp")) {
297 /* We need to correct this one */
302 if (NeedCorrection) {
304 /* Get the code entry before this one. If it's a LDY, adjust the
307 CodeEntry* P = CS_GetPrevEntry (D->Code, I);
308 if (P && P->OPC == OP65_LDY && CE_IsConstImm (P)) {
310 /* The Y load is just before the stack access, adjust it */
311 CE_SetNumArg (P, P->Num - Offs);
315 /* Insert a new load instruction before the stack access */
316 const char* Arg = MakeHexArg (E->RI->In.RegY - Offs);
317 CodeEntry* X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
318 CS_InsertEntry (D->Code, X, I++);
320 /* One more inserted entries */
325 /* If we need the value of Y later, be sure to reload it */
326 if (RegYUsed (D->Code, I+1)) {
327 const char* Arg = MakeHexArg (E->RI->In.RegY);
328 CodeEntry* X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
329 CS_InsertEntry (D->Code, X, I+1);
331 /* One more inserted entries */
334 /* Skip this instruction in the next round */
343 /* If we have rhs load insns that load from stack, we'll have to adjust
344 * the offsets for these also.
346 if (D->Rhs.A.Flags & LI_RELOAD_Y) {
347 D->Rhs.A.Offs -= Offs;
349 if (D->Rhs.X.Flags & LI_RELOAD_Y) {
350 D->Rhs.X.Offs -= Offs;
356 static void InsertEntry (StackOpData* D, CodeEntry* E, int Index)
357 /* Insert a new entry. Depending on Index, D->PushIndex and D->OpIndex will
358 * be adjusted by this function.
361 /* Insert the entry into the code segment */
362 CS_InsertEntry (D->Code, E, Index);
364 /* Adjust register loads if necessary */
365 AdjustLoadInfo (&D->Lhs, Index, 1);
366 AdjustLoadInfo (&D->Rhs, Index, 1);
368 /* Adjust the indices if necessary */
369 if (D->PushEntry && Index <= D->PushIndex) {
372 if (D->OpEntry && Index <= D->OpIndex) {
379 static void DelEntry (StackOpData* D, int Index)
380 /* Delete an entry. Depending on Index, D->PushIndex and D->OpIndex will be
381 * adjusted by this function, and PushEntry/OpEntry may get invalidated.
384 /* Delete the entry from the code segment */
385 CS_DelEntry (D->Code, Index);
387 /* Adjust register loads if necessary */
388 AdjustLoadInfo (&D->Lhs, Index, -1);
389 AdjustLoadInfo (&D->Rhs, Index, -1);
391 /* Adjust the other indices if necessary */
392 if (Index < D->PushIndex) {
394 } else if (Index == D->PushIndex) {
397 if (Index < D->OpIndex) {
399 } else if (Index == D->OpIndex) {
406 static void CheckOneDirectOp (LoadRegInfo* LI, unsigned char Offs)
407 /* Check if the given entry is a lda instruction with an addressing mode
408 * that allows us to replace it by another operation (like ora). If so, we may
409 * use this location for the or and must not save the value in the zero
413 /* Get the load entry */
414 CodeEntry* E = LI->LoadEntry;
420 /* Check the load entry */
422 /* Must check the call first since addressing mode is ABS, so second
423 * "if" will catch otherwise.
425 if (CE_IsCallTo (E, "ldaxysp")) {
426 /* Same as single loads from stack. Since we must distinguish
427 * between A and X here, the necessary offset is passed to the
428 * function as a parameter.
430 LI->Offs = (unsigned char) E->RI->In.RegY - Offs;
431 LI->Flags |= (LI_DIRECT | LI_RELOAD_Y);
432 } else if (E->AM == AM65_IMM || E->AM == AM65_ZP || E->AM == AM65_ABS) {
433 /* These insns are all ok and replaceable */
434 LI->Flags |= LI_DIRECT;
435 } else if (E->AM == AM65_ZP_INDY &&
436 RegValIsKnown (E->RI->In.RegY) &&
437 strcmp (E->Arg, "sp") == 0) {
438 /* A load from the stack with known offset is also ok, but in this
439 * case we must reload the index register later. Please note that
440 * a load indirect via other zero page locations is not ok, since
441 * these locations may change between the push and the actual
444 LI->Offs = (unsigned char) E->RI->In.RegY;
445 LI->Flags |= (LI_DIRECT | LI_RELOAD_Y);
452 static void CheckDirectOp (StackOpData* D)
453 /* Check if the given entry is a lda instruction with an addressing mode
454 * that allows us to replace it by another operation (like ora). If so, we may
455 * use this location for the or and must not save the value in the zero
459 /* Check flags for all load instructions */
460 CheckOneDirectOp (&D->Lhs.A, 1);
461 CheckOneDirectOp (&D->Lhs.X, 0);
462 CheckOneDirectOp (&D->Rhs.A, 1);
463 CheckOneDirectOp (&D->Rhs.X, 0);
468 static void AddStoreA (StackOpData* D)
469 /* Add a store to zero page after the push insn */
471 CodeEntry* X = NewCodeEntry (OP65_STA, AM65_ZP, D->ZPLo, 0, D->PushEntry->LI);
472 InsertEntry (D, X, D->PushIndex+1);
477 static void AddStoreX (StackOpData* D)
478 /* Add a store to zero page after the push insn */
480 CodeEntry* X = NewCodeEntry (OP65_STX, AM65_ZP, D->ZPHi, 0, D->PushEntry->LI);
481 InsertEntry (D, X, D->PushIndex+1);
486 static void ReplacePushByStore (StackOpData* D)
487 /* Replace the call to the push subroutine by a store into the zero page
488 * location (actually, the push is not replaced, because we need it for
489 * later, but the name is still ok since the push will get removed at the
490 * end of each routine).
493 /* Store the value into the zeropage instead of pushing it. Check high
494 * byte first so that the store is later in A/X order.
496 if ((D->Lhs.X.Flags & LI_DIRECT) == 0) {
499 if ((D->Lhs.A.Flags & LI_DIRECT) == 0) {
506 static void AddOpLow (StackOpData* D, opc_t OPC)
507 /* Add an op for the low byte of an operator. This function honours the
508 * OP_DIRECT and OP_RELOAD_Y flags and generates the necessary instructions.
509 * All code is inserted at the current insertion point.
514 if ((D->Lhs.A.Flags & LI_DIRECT) != 0) {
515 /* Op with a variable location. If the location is on the stack, we
516 * need to reload the Y register.
518 if ((D->Lhs.A.Flags & LI_RELOAD_Y) == 0) {
521 CodeEntry* LoadA = D->Lhs.A.LoadEntry;
522 X = NewCodeEntry (OPC, LoadA->AM, LoadA->Arg, 0, D->OpEntry->LI);
523 InsertEntry (D, X, D->IP++);
528 const char* Arg = MakeHexArg (D->Lhs.A.Offs);
529 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, D->OpEntry->LI);
530 InsertEntry (D, X, D->IP++);
533 X = NewCodeEntry (OPC, AM65_ZP_INDY, "sp", 0, D->OpEntry->LI);
534 InsertEntry (D, X, D->IP++);
538 /* In both cases, we can remove the load */
539 D->Lhs.A.Flags |= LI_REMOVE;
543 /* Op with temp storage */
544 X = NewCodeEntry (OPC, AM65_ZP, D->ZPLo, 0, D->OpEntry->LI);
545 InsertEntry (D, X, D->IP++);
552 static void AddOpHigh (StackOpData* D, opc_t OPC)
553 /* Add an op for the high byte of an operator. Special cases (constant values
554 * or similar) have to be checked separately, the function covers only the
555 * generic case. Code is inserted at the insertion point.
561 X = NewCodeEntry (OP65_PHA, AM65_IMP, 0, 0, D->OpEntry->LI);
562 InsertEntry (D, X, D->IP++);
565 X = NewCodeEntry (OP65_TXA, AM65_IMP, 0, 0, D->OpEntry->LI);
566 InsertEntry (D, X, D->IP++);
568 if ((D->Lhs.X.Flags & LI_DIRECT) != 0) {
570 if ((D->Lhs.X.Flags & LI_RELOAD_Y) == 0) {
573 CodeEntry* LoadX = D->Lhs.X.LoadEntry;
574 X = NewCodeEntry (OPC, LoadX->AM, LoadX->Arg, 0, D->OpEntry->LI);
575 InsertEntry (D, X, D->IP++);
580 const char* Arg = MakeHexArg (D->Lhs.X.Offs);
581 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, D->OpEntry->LI);
582 InsertEntry (D, X, D->IP++);
585 X = NewCodeEntry (OPC, AM65_ZP_INDY, "sp", 0, D->OpEntry->LI);
586 InsertEntry (D, X, D->IP++);
589 /* In both cases, we can remove the load */
590 D->Lhs.X.Flags |= LI_REMOVE;
594 X = NewCodeEntry (OPC, AM65_ZP, D->ZPHi, 0, D->OpEntry->LI);
595 InsertEntry (D, X, D->IP++);
599 X = NewCodeEntry (OP65_TAX, AM65_IMP, 0, 0, D->OpEntry->LI);
600 InsertEntry (D, X, D->IP++);
603 X = NewCodeEntry (OP65_PLA, AM65_IMP, 0, 0, D->OpEntry->LI);
604 InsertEntry (D, X, D->IP++);
609 static void RemoveRegLoads (StackOpData* D, LoadInfo* LI)
610 /* Remove register load insns */
612 /* Both registers may be loaded with one insn, but DelEntry will in this
613 * case clear the other one.
615 if (LI->A.LoadIndex >= 0 && (LI->A.Flags & LI_REMOVE)) {
616 DelEntry (D, LI->A.LoadIndex);
618 if (LI->X.LoadIndex >= 0 && (LI->X.Flags & LI_REMOVE)) {
619 DelEntry (D, LI->X.LoadIndex);
625 static void RemoveRemainders (StackOpData* D)
626 /* Remove the code that is unnecessary after translation of the sequence */
628 /* Remove the register loads for lhs and rhs */
629 RemoveRegLoads (D, &D->Lhs);
630 RemoveRegLoads (D, &D->Rhs);
632 /* Remove the push and the operator routine */
633 DelEntry (D, D->OpIndex);
634 DelEntry (D, D->PushIndex);
639 static int IsRegVar (StackOpData* D)
640 /* If the value pushed is that of a zeropage variable, replace ZPLo and ZPHi
641 * in the given StackOpData struct by the variable and return true. Otherwise
642 * leave D untouched and return false.
645 CodeEntry* LoadA = D->Lhs.A.LoadEntry;
646 CodeEntry* LoadX = D->Lhs.X.LoadEntry;
649 /* Must have both load insns */
650 if (LoadA == 0 || LoadX == 0) {
654 /* Must be loads from zp */
655 if (LoadA->AM != AM65_ZP || LoadX->AM != AM65_ZP) {
659 /* Must be the same zp loc with high byte in X */
660 Len = strlen (LoadA->Arg);
661 if (strncmp (LoadA->Arg, LoadX->Arg, Len) != 0 ||
662 strcmp (LoadX->Arg + Len, "+1") != 0) {
666 /* Use the zero page location directly */
667 D->ZPLo = LoadA->Arg;
668 D->ZPHi = LoadX->Arg;
674 /*****************************************************************************/
675 /* Actual optimization functions */
676 /*****************************************************************************/
680 static unsigned Opt_toseqax_tosneax (StackOpData* D, const char* BoolTransformer)
681 /* Optimize the toseqax and tosneax sequences. */
686 /* Create a call to the boolean transformer function and a label for this
687 * insn. This is needed for all variants. Other insns are inserted *before*
690 X = NewCodeEntry (OP65_JSR, AM65_ABS, BoolTransformer, 0, D->OpEntry->LI);
691 InsertEntry (D, X, D->OpIndex + 1);
692 L = CS_GenLabel (D->Code, X);
694 /* If the lhs is direct (but not stack relative), encode compares with lhs
695 * effectively reverting the order (which doesn't matter for ==).
697 if ((D->Lhs.A.Flags & (LI_DIRECT | LI_RELOAD_Y)) == LI_DIRECT &&
698 (D->Lhs.X.Flags & (LI_DIRECT | LI_RELOAD_Y)) == LI_DIRECT) {
700 CodeEntry* LoadX = D->Lhs.X.LoadEntry;
701 CodeEntry* LoadA = D->Lhs.A.LoadEntry;
703 D->IP = D->OpIndex+1;
706 X = NewCodeEntry (OP65_CPX, LoadX->AM, LoadX->Arg, 0, D->OpEntry->LI);
707 InsertEntry (D, X, D->IP++);
710 X = NewCodeEntry (OP65_BNE, AM65_BRA, L->Name, L, D->OpEntry->LI);
711 InsertEntry (D, X, D->IP++);
714 X = NewCodeEntry (OP65_CMP, LoadA->AM, LoadA->Arg, 0, D->OpEntry->LI);
715 InsertEntry (D, X, D->IP++);
717 /* Lhs load entries can be removed */
718 D->Lhs.X.Flags |= LI_REMOVE;
719 D->Lhs.A.Flags |= LI_REMOVE;
721 } else if ((D->Rhs.A.Flags & (LI_DIRECT | LI_RELOAD_Y)) == LI_DIRECT &&
722 (D->Rhs.X.Flags & (LI_DIRECT | LI_RELOAD_Y)) == LI_DIRECT) {
724 CodeEntry* LoadX = D->Rhs.X.LoadEntry;
725 CodeEntry* LoadA = D->Rhs.A.LoadEntry;
727 D->IP = D->OpIndex+1;
730 X = NewCodeEntry (OP65_CPX, LoadX->AM, LoadX->Arg, 0, D->OpEntry->LI);
731 InsertEntry (D, X, D->IP++);
734 X = NewCodeEntry (OP65_BNE, AM65_BRA, L->Name, L, D->OpEntry->LI);
735 InsertEntry (D, X, D->IP++);
738 X = NewCodeEntry (OP65_CMP, LoadA->AM, LoadA->Arg, 0, D->OpEntry->LI);
739 InsertEntry (D, X, D->IP++);
741 /* Rhs load entries can be removed */
742 D->Rhs.X.Flags |= LI_REMOVE;
743 D->Rhs.A.Flags |= LI_REMOVE;
745 } else if ((D->Rhs.A.Flags & LI_DIRECT) != 0 &&
746 (D->Rhs.X.Flags & LI_DIRECT) != 0) {
748 D->IP = D->OpIndex+1;
750 /* If the location is on the stack, we need to reload the Y register. */
751 if ((D->Rhs.A.Flags & LI_RELOAD_Y) == 0) {
754 CodeEntry* LoadA = D->Rhs.A.LoadEntry;
755 X = NewCodeEntry (OP65_CMP, LoadA->AM, LoadA->Arg, 0, D->OpEntry->LI);
756 InsertEntry (D, X, D->IP++);
761 const char* Arg = MakeHexArg (D->Rhs.A.Offs);
762 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, D->OpEntry->LI);
763 InsertEntry (D, X, D->IP++);
766 X = NewCodeEntry (OP65_CMP, AM65_ZP_INDY, "sp", 0, D->OpEntry->LI);
767 InsertEntry (D, X, D->IP++);
770 /* In both cases, we can remove the load */
771 D->Rhs.A.Flags |= LI_REMOVE;
774 X = NewCodeEntry (OP65_BNE, AM65_BRA, L->Name, L, D->OpEntry->LI);
775 InsertEntry (D, X, D->IP++);
778 X = NewCodeEntry (OP65_TXA, AM65_IMP, 0, 0, D->OpEntry->LI);
779 InsertEntry (D, X, D->IP++);
781 /* If the location is on the stack, we need to reload the Y register. */
782 if ((D->Rhs.X.Flags & LI_RELOAD_Y) == 0) {
785 CodeEntry* LoadX = D->Rhs.X.LoadEntry;
786 X = NewCodeEntry (OP65_CMP, LoadX->AM, LoadX->Arg, 0, D->OpEntry->LI);
787 InsertEntry (D, X, D->IP++);
792 const char* Arg = MakeHexArg (D->Rhs.X.Offs);
793 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, D->OpEntry->LI);
794 InsertEntry (D, X, D->IP++);
797 X = NewCodeEntry (OP65_CMP, AM65_ZP_INDY, "sp", 0, D->OpEntry->LI);
798 InsertEntry (D, X, D->IP++);
801 /* In both cases, we can remove the load */
802 D->Rhs.X.Flags |= LI_REMOVE;
806 /* Save lhs into zeropage, then compare */
810 D->IP = D->OpIndex+1;
813 X = NewCodeEntry (OP65_CPX, AM65_ZP, D->ZPHi, 0, D->OpEntry->LI);
814 InsertEntry (D, X, D->IP++);
817 X = NewCodeEntry (OP65_BNE, AM65_BRA, L->Name, L, D->OpEntry->LI);
818 InsertEntry (D, X, D->IP++);
821 X = NewCodeEntry (OP65_CMP, AM65_ZP, D->ZPLo, 0, D->OpEntry->LI);
822 InsertEntry (D, X, D->IP++);
826 /* Remove the push and the call to the tosgeax function */
827 RemoveRemainders (D);
829 /* We changed the sequence */
835 static unsigned Opt___bzero (StackOpData* D)
836 /* Optimize the __bzero sequence */
842 /* Check if we're using a register variable */
844 /* Store the value into the zeropage instead of pushing it */
849 /* If the return value of __bzero is used, we have to add code to reload
850 * a/x from the pointer variable.
852 if (RegAXUsed (D->Code, D->OpIndex+1)) {
853 X = NewCodeEntry (OP65_LDA, AM65_ZP, D->ZPLo, 0, D->OpEntry->LI);
854 InsertEntry (D, X, D->OpIndex+1);
855 X = NewCodeEntry (OP65_LDX, AM65_ZP, D->ZPHi, 0, D->OpEntry->LI);
856 InsertEntry (D, X, D->OpIndex+2);
859 /* X is always zero, A contains the size of the data area to zero.
860 * Note: A may be zero, in which case the operation is null op.
862 if (D->OpEntry->RI->In.RegA != 0) {
865 X = NewCodeEntry (OP65_LDA, AM65_IMM, "$00", 0, D->OpEntry->LI);
866 InsertEntry (D, X, D->OpIndex+1);
868 /* The value of A is known */
869 if (D->OpEntry->RI->In.RegA <= 0x81) {
871 /* Loop using the sign bit */
874 Arg = MakeHexArg (D->OpEntry->RI->In.RegA - 1);
875 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, D->OpEntry->LI);
876 InsertEntry (D, X, D->OpIndex+2);
879 X = NewCodeEntry (OP65_STA, AM65_ZP_INDY, D->ZPLo, 0, D->OpEntry->LI);
880 InsertEntry (D, X, D->OpIndex+3);
881 L = CS_GenLabel (D->Code, X);
884 X = NewCodeEntry (OP65_DEY, AM65_IMP, 0, 0, D->OpEntry->LI);
885 InsertEntry (D, X, D->OpIndex+4);
888 X = NewCodeEntry (OP65_BPL, AM65_BRA, L->Name, L, D->OpEntry->LI);
889 InsertEntry (D, X, D->OpIndex+5);
893 /* Loop using an explicit compare */
896 X = NewCodeEntry (OP65_LDY, AM65_IMM, "$00", 0, D->OpEntry->LI);
897 InsertEntry (D, X, D->OpIndex+2);
900 X = NewCodeEntry (OP65_STA, AM65_ZP_INDY, D->ZPLo, 0, D->OpEntry->LI);
901 InsertEntry (D, X, D->OpIndex+3);
902 L = CS_GenLabel (D->Code, X);
905 X = NewCodeEntry (OP65_INY, AM65_IMP, 0, 0, D->OpEntry->LI);
906 InsertEntry (D, X, D->OpIndex+4);
909 Arg = MakeHexArg (D->OpEntry->RI->In.RegA);
910 X = NewCodeEntry (OP65_CPY, AM65_IMM, Arg, 0, D->OpEntry->LI);
911 InsertEntry (D, X, D->OpIndex+5);
914 X = NewCodeEntry (OP65_BNE, AM65_BRA, L->Name, L, D->OpEntry->LI);
915 InsertEntry (D, X, D->OpIndex+6);
920 /* Remove the push and the call to the __bzero function */
921 RemoveRemainders (D);
923 /* We changed the sequence */
929 static unsigned Opt_staspidx (StackOpData* D)
930 /* Optimize the staspidx sequence */
934 /* Check if we're using a register variable */
936 /* Store the value into the zeropage instead of pushing it */
941 /* Replace the store subroutine call by a direct op */
942 X = NewCodeEntry (OP65_STA, AM65_ZP_INDY, D->ZPLo, 0, D->OpEntry->LI);
943 InsertEntry (D, X, D->OpIndex+1);
945 /* Remove the push and the call to the staspidx function */
946 RemoveRemainders (D);
948 /* We changed the sequence */
954 static unsigned Opt_staxspidx (StackOpData* D)
955 /* Optimize the staxspidx sequence */
959 /* Check if we're using a register variable */
961 /* Store the value into the zeropage instead of pushing it */
966 /* Inline the store */
969 X = NewCodeEntry (OP65_STA, AM65_ZP_INDY, D->ZPLo, 0, D->OpEntry->LI);
970 InsertEntry (D, X, D->OpIndex+1);
972 if (RegValIsKnown (D->OpEntry->RI->In.RegY)) {
973 /* Value of Y is known */
974 const char* Arg = MakeHexArg (D->OpEntry->RI->In.RegY + 1);
975 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, D->OpEntry->LI);
977 X = NewCodeEntry (OP65_INY, AM65_IMP, 0, 0, D->OpEntry->LI);
979 InsertEntry (D, X, D->OpIndex+2);
981 if (RegValIsKnown (D->OpEntry->RI->In.RegX)) {
982 /* Value of X is known */
983 const char* Arg = MakeHexArg (D->OpEntry->RI->In.RegX);
984 X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, D->OpEntry->LI);
987 X = NewCodeEntry (OP65_TXA, AM65_IMP, 0, 0, D->OpEntry->LI);
989 InsertEntry (D, X, D->OpIndex+3);
992 X = NewCodeEntry (OP65_STA, AM65_ZP_INDY, D->ZPLo, 0, D->OpEntry->LI);
993 InsertEntry (D, X, D->OpIndex+4);
995 /* If we remove staxspidx, we must restore the Y register to what the
996 * function would return.
998 X = NewCodeEntry (OP65_LDY, AM65_IMM, "$00", 0, D->OpEntry->LI);
999 InsertEntry (D, X, D->OpIndex+5);
1001 /* Remove the push and the call to the staxspidx function */
1002 RemoveRemainders (D);
1004 /* We changed the sequence */
1010 static unsigned Opt_tosaddax (StackOpData* D)
1011 /* Optimize the tosaddax sequence */
1016 /* We need the entry behind the add */
1017 CHECK (D->NextEntry != 0);
1019 /* Check if the X register is known and zero when the add is done, and
1020 * if the add is followed by
1023 * jsr ldauidx ; or ldaidx
1025 * If this is true, the addition does actually add an offset to a pointer
1026 * before it is dereferenced. Since both subroutines take an offset in Y,
1027 * we can pass the offset (instead of #$00) and remove the addition
1030 if (D->OpEntry->RI->In.RegX == 0 &&
1031 D->NextEntry->OPC == OP65_LDY &&
1032 CE_IsKnownImm (D->NextEntry, 0) &&
1033 !CE_HasLabel (D->NextEntry) &&
1034 (N = CS_GetNextEntry (D->Code, D->OpIndex + 1)) != 0 &&
1035 (CE_IsCallTo (N, "ldauidx") ||
1036 CE_IsCallTo (N, "ldaidx"))) {
1038 int Signed = (strcmp (N->Arg, "ldaidx") == 0);
1040 /* Store the value into the zeropage instead of pushing it */
1044 /* Replace the ldy by a tay. Be sure to create the new entry before
1045 * deleting the ldy, since we will reference the line info from this
1048 X = NewCodeEntry (OP65_TAY, AM65_IMP, 0, 0, D->NextEntry->LI);
1049 DelEntry (D, D->OpIndex + 1);
1050 InsertEntry (D, X, D->OpIndex + 1);
1052 /* Replace the call to ldaidx/ldauidx. Since X is already zero, and
1053 * the ptr is in the zero page location, we just need to load from
1054 * the pointer, and fix X in case of ldaidx.
1056 X = NewCodeEntry (OP65_LDA, AM65_ZP_INDY, D->ZPLo, 0, N->LI);
1057 DelEntry (D, D->OpIndex + 2);
1058 InsertEntry (D, X, D->OpIndex + 2);
1063 /* Add sign extension - N is unused now */
1064 N = CS_GetNextEntry (D->Code, D->OpIndex + 2);
1066 L = CS_GenLabel (D->Code, N);
1068 X = NewCodeEntry (OP65_BPL, AM65_BRA, L->Name, L, X->LI);
1069 InsertEntry (D, X, D->OpIndex + 3);
1071 X = NewCodeEntry (OP65_DEX, AM65_IMP, 0, 0, X->LI);
1072 InsertEntry (D, X, D->OpIndex + 4);
1077 /* Store the value into the zeropage instead of pushing it */
1078 ReplacePushByStore (D);
1080 /* Inline the add */
1081 D->IP = D->OpIndex+1;
1084 X = NewCodeEntry (OP65_CLC, AM65_IMP, 0, 0, D->OpEntry->LI);
1085 InsertEntry (D, X, D->IP++);
1088 AddOpLow (D, OP65_ADC);
1091 if (D->PushEntry->RI->In.RegX == 0) {
1092 /* The high byte is the value in X plus the carry */
1093 CodeLabel* L = CS_GenLabel (D->Code, D->NextEntry);
1094 X = NewCodeEntry (OP65_BCC, AM65_BRA, L->Name, L, D->OpEntry->LI);
1095 InsertEntry (D, X, D->IP++);
1096 X = NewCodeEntry (OP65_INX, AM65_IMP, 0, 0, D->OpEntry->LI);
1097 InsertEntry (D, X, D->IP++);
1098 } else if (D->OpEntry->RI->In.RegX == 0) {
1099 /* The high byte is that of the first operand plus carry */
1101 if (RegValIsKnown (D->PushEntry->RI->In.RegX)) {
1102 /* Value of first op high byte is known */
1103 const char* Arg = MakeHexArg (D->PushEntry->RI->In.RegX);
1104 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, D->OpEntry->LI);
1106 /* Value of first op high byte is unknown */
1107 X = NewCodeEntry (OP65_LDX, AM65_ZP, D->ZPHi, 0, D->OpEntry->LI);
1109 InsertEntry (D, X, D->IP++);
1112 L = CS_GenLabel (D->Code, D->NextEntry);
1113 X = NewCodeEntry (OP65_BCC, AM65_BRA, L->Name, L, D->OpEntry->LI);
1114 InsertEntry (D, X, D->IP++);
1117 X = NewCodeEntry (OP65_INX, AM65_IMP, 0, 0, D->OpEntry->LI);
1118 InsertEntry (D, X, D->IP++);
1120 /* High byte is unknown */
1121 AddOpHigh (D, OP65_ADC);
1125 /* Remove the push and the call to the tosaddax function */
1126 RemoveRemainders (D);
1128 /* We changed the sequence */
1134 static unsigned Opt_tosandax (StackOpData* D)
1135 /* Optimize the tosandax sequence */
1139 /* Store the value into the zeropage instead of pushing it */
1140 ReplacePushByStore (D);
1142 /* Inline the and, low byte */
1143 D->IP = D->OpIndex + 1;
1144 AddOpLow (D, OP65_AND);
1147 if (D->PushEntry->RI->In.RegX == 0 || D->OpEntry->RI->In.RegX == 0) {
1148 /* The high byte is zero */
1149 X = NewCodeEntry (OP65_LDX, AM65_IMM, "$00", 0, D->OpEntry->LI);
1150 InsertEntry (D, X, D->IP++);
1152 /* High byte is unknown */
1153 AddOpHigh (D, OP65_AND);
1156 /* Remove the push and the call to the tosandax function */
1157 RemoveRemainders (D);
1159 /* We changed the sequence */
1165 static unsigned Opt_toseqax (StackOpData* D)
1166 /* Optimize the toseqax sequence */
1168 return Opt_toseqax_tosneax (D, "booleq");
1173 static unsigned Opt_tosgeax (StackOpData* D)
1174 /* Optimize the tosgeax sequence */
1179 /* Inline the sbc */
1180 D->IP = D->OpIndex+1;
1182 /* Must be true because of OP_RHS_LOAD */
1183 CHECK ((D->Rhs.A.Flags & LI_DIRECT) != 0);
1185 /* If the location is on the stack, we need to reload the Y register. */
1186 if ((D->Rhs.A.Flags & LI_RELOAD_Y) == 0) {
1189 CodeEntry* LoadA = D->Rhs.A.LoadEntry;
1190 X = NewCodeEntry (OP65_CMP, LoadA->AM, LoadA->Arg, 0, D->OpEntry->LI);
1191 InsertEntry (D, X, D->IP++);
1196 const char* Arg = MakeHexArg (D->Rhs.A.Offs);
1197 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, D->OpEntry->LI);
1198 InsertEntry (D, X, D->IP++);
1201 X = NewCodeEntry (OP65_CMP, AM65_ZP_INDY, "sp", 0, D->OpEntry->LI);
1202 InsertEntry (D, X, D->IP++);
1205 /* In both cases, we can remove the load */
1206 D->Rhs.A.Flags |= LI_REMOVE;
1209 X = NewCodeEntry (OP65_TXA, AM65_IMP, 0, 0, D->OpEntry->LI);
1210 InsertEntry (D, X, D->IP++);
1212 /* Must be true because of OP_RHS_LOAD */
1213 CHECK ((D->Rhs.X.Flags & LI_DIRECT) != 0);
1215 /* If the location is on the stack, we need to reload the Y register. */
1216 if ((D->Rhs.X.Flags & LI_RELOAD_Y) == 0) {
1219 CodeEntry* LoadX = D->Rhs.X.LoadEntry;
1220 X = NewCodeEntry (OP65_SBC, LoadX->AM, LoadX->Arg, 0, D->OpEntry->LI);
1221 InsertEntry (D, X, D->IP++);
1226 const char* Arg = MakeHexArg (D->Rhs.X.Offs);
1227 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, D->OpEntry->LI);
1228 InsertEntry (D, X, D->IP++);
1231 X = NewCodeEntry (OP65_SBC, AM65_ZP_INDY, "sp", 0, D->OpEntry->LI);
1232 InsertEntry (D, X, D->IP++);
1235 /* In both cases, we can remove the load */
1236 D->Rhs.X.Flags |= LI_REMOVE;
1239 X = NewCodeEntry (OP65_EOR, AM65_IMM, "$80", 0, D->OpEntry->LI);
1240 InsertEntry (D, X, D->IP++);
1243 X = NewCodeEntry (OP65_ASL, AM65_ACC, "a", 0, D->OpEntry->LI);
1244 InsertEntry (D, X, D->IP++);
1245 L = CS_GenLabel (D->Code, X);
1247 /* Insert a bvs L before the eor insn */
1248 X = NewCodeEntry (OP65_BVS, AM65_BRA, L->Name, L, D->OpEntry->LI);
1249 InsertEntry (D, X, D->IP - 2);
1253 X = NewCodeEntry (OP65_LDA, AM65_IMM, "$00", 0, D->OpEntry->LI);
1254 InsertEntry (D, X, D->IP++);
1257 X = NewCodeEntry (OP65_LDX, AM65_IMM, "$00", 0, D->OpEntry->LI);
1258 InsertEntry (D, X, D->IP++);
1261 X = NewCodeEntry (OP65_ROL, AM65_ACC, "a", 0, D->OpEntry->LI);
1262 InsertEntry (D, X, D->IP++);
1264 /* Remove the push and the call to the tosgeax function */
1265 RemoveRemainders (D);
1267 /* We changed the sequence */
1273 static unsigned Opt_tosltax (StackOpData* D)
1274 /* Optimize the tosltax sequence */
1280 /* Inline the sbc */
1281 D->IP = D->OpIndex+1;
1283 /* Must be true because of OP_RHS_LOAD */
1284 CHECK ((D->Rhs.A.Flags & LI_DIRECT) != 0);
1286 /* If the location is on the stack, we need to reload the Y register. */
1287 if ((D->Rhs.A.Flags & LI_RELOAD_Y) == 0) {
1290 CodeEntry* LoadA = D->Rhs.A.LoadEntry;
1291 X = NewCodeEntry (OP65_CMP, LoadA->AM, LoadA->Arg, 0, D->OpEntry->LI);
1292 InsertEntry (D, X, D->IP++);
1297 const char* Arg = MakeHexArg (D->Rhs.A.Offs);
1298 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, D->OpEntry->LI);
1299 InsertEntry (D, X, D->IP++);
1302 X = NewCodeEntry (OP65_CMP, AM65_ZP_INDY, "sp", 0, D->OpEntry->LI);
1303 InsertEntry (D, X, D->IP++);
1306 /* In both cases, we can remove the load */
1307 D->Rhs.A.Flags |= LI_REMOVE;
1310 X = NewCodeEntry (OP65_TXA, AM65_IMP, 0, 0, D->OpEntry->LI);
1311 InsertEntry (D, X, D->IP++);
1313 /* Must be true because of OP_RHS_LOAD */
1314 CHECK ((D->Rhs.X.Flags & LI_DIRECT) != 0);
1316 /* If the location is on the stack, we need to reload the Y register. */
1317 if ((D->Rhs.X.Flags & LI_RELOAD_Y) == 0) {
1320 CodeEntry* LoadX = D->Rhs.X.LoadEntry;
1321 X = NewCodeEntry (OP65_SBC, LoadX->AM, LoadX->Arg, 0, D->OpEntry->LI);
1322 InsertEntry (D, X, D->IP++);
1327 const char* Arg = MakeHexArg (D->Rhs.X.Offs);
1328 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, D->OpEntry->LI);
1329 InsertEntry (D, X, D->IP++);
1332 X = NewCodeEntry (OP65_SBC, AM65_ZP_INDY, "sp", 0, D->OpEntry->LI);
1333 InsertEntry (D, X, D->IP++);
1336 /* In both cases, we can remove the load */
1337 D->Rhs.X.Flags |= LI_REMOVE;
1340 X = NewCodeEntry (OP65_EOR, AM65_IMM, "$80", 0, D->OpEntry->LI);
1341 InsertEntry (D, X, D->IP++);
1344 X = NewCodeEntry (OP65_ASL, AM65_ACC, "a", 0, D->OpEntry->LI);
1345 InsertEntry (D, X, D->IP++);
1346 L = CS_GenLabel (D->Code, X);
1348 /* Insert a bvc L before the eor insn */
1349 X = NewCodeEntry (OP65_BVC, AM65_BRA, L->Name, L, D->OpEntry->LI);
1350 InsertEntry (D, X, D->IP - 2);
1354 X = NewCodeEntry (OP65_LDA, AM65_IMM, "$00", 0, D->OpEntry->LI);
1355 InsertEntry (D, X, D->IP++);
1358 X = NewCodeEntry (OP65_LDX, AM65_IMM, "$00", 0, D->OpEntry->LI);
1359 InsertEntry (D, X, D->IP++);
1362 X = NewCodeEntry (OP65_ROL, AM65_ACC, "a", 0, D->OpEntry->LI);
1363 InsertEntry (D, X, D->IP++);
1365 /* Remove the push and the call to the tosltax function */
1366 RemoveRemainders (D);
1368 /* We changed the sequence */
1374 static unsigned Opt_tosneax (StackOpData* D)
1375 /* Optimize the tosneax sequence */
1377 return Opt_toseqax_tosneax (D, "boolne");
1382 static unsigned Opt_tosorax (StackOpData* D)
1383 /* Optimize the tosorax sequence */
1387 /* Store the value into the zeropage instead of pushing it */
1388 ReplacePushByStore (D);
1390 /* Inline the or, low byte */
1391 D->IP = D->OpIndex + 1;
1392 AddOpLow (D, OP65_ORA);
1395 if (RegValIsKnown (D->PushEntry->RI->In.RegX) &&
1396 RegValIsKnown (D->OpEntry->RI->In.RegX)) {
1397 /* Both values known, precalculate the result */
1398 unsigned char Result = D->PushEntry->RI->In.RegX | D->OpEntry->RI->In.RegX;
1399 const char* Arg = MakeHexArg (Result);
1400 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, D->OpEntry->LI);
1401 InsertEntry (D, X, D->IP++);
1402 } else if (D->PushEntry->RI->In.RegX != 0) {
1403 /* High byte is unknown */
1404 AddOpHigh (D, OP65_ORA);
1407 /* Remove the push and the call to the tosorax function */
1408 RemoveRemainders (D);
1410 /* We changed the sequence */
1416 static unsigned Opt_tossubax (StackOpData* D)
1417 /* Optimize the tossubax sequence. Note: subtraction is not commutative! */
1422 /* Inline the sbc */
1423 D->IP = D->OpIndex+1;
1426 X = NewCodeEntry (OP65_SEC, AM65_IMP, 0, 0, D->OpEntry->LI);
1427 InsertEntry (D, X, D->IP++);
1429 /* Must be true because of OP_RHS_LOAD */
1430 CHECK ((D->Rhs.A.Flags & LI_DIRECT) != 0);
1432 /* If the location is on the stack, we need to reload the Y register. */
1433 if ((D->Rhs.A.Flags & LI_RELOAD_Y) == 0) {
1436 CodeEntry* LoadA = D->Rhs.A.LoadEntry;
1437 X = NewCodeEntry (OP65_SBC, LoadA->AM, LoadA->Arg, 0, D->OpEntry->LI);
1438 InsertEntry (D, X, D->IP++);
1443 const char* Arg = MakeHexArg (D->Rhs.A.Offs);
1444 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, D->OpEntry->LI);
1445 InsertEntry (D, X, D->IP++);
1448 X = NewCodeEntry (OP65_SBC, AM65_ZP_INDY, "sp", 0, D->OpEntry->LI);
1449 InsertEntry (D, X, D->IP++);
1452 /* In both cases, we can remove the load */
1453 D->Rhs.A.Flags |= LI_REMOVE;
1456 X = NewCodeEntry (OP65_PHA, AM65_IMP, 0, 0, D->OpEntry->LI);
1457 InsertEntry (D, X, D->IP++);
1460 X = NewCodeEntry (OP65_TXA, AM65_IMP, 0, 0, D->OpEntry->LI);
1461 InsertEntry (D, X, D->IP++);
1463 /* Must be true because of OP_RHS_LOAD */
1464 CHECK ((D->Rhs.X.Flags & LI_DIRECT) != 0);
1466 /* If the location is on the stack, we need to reload the Y register. */
1467 if ((D->Rhs.X.Flags & LI_RELOAD_Y) == 0) {
1470 CodeEntry* LoadX = D->Rhs.X.LoadEntry;
1471 X = NewCodeEntry (OP65_SBC, LoadX->AM, LoadX->Arg, 0, D->OpEntry->LI);
1472 InsertEntry (D, X, D->IP++);
1477 const char* Arg = MakeHexArg (D->Rhs.X.Offs);
1478 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, D->OpEntry->LI);
1479 InsertEntry (D, X, D->IP++);
1482 X = NewCodeEntry (OP65_SBC, AM65_ZP_INDY, "sp", 0, D->OpEntry->LI);
1483 InsertEntry (D, X, D->IP++);
1486 /* In both cases, we can remove the load */
1487 D->Rhs.X.Flags |= LI_REMOVE;
1490 X = NewCodeEntry (OP65_TAX, AM65_IMP, 0, 0, D->OpEntry->LI);
1491 InsertEntry (D, X, D->IP++);
1494 X = NewCodeEntry (OP65_PLA, AM65_IMP, 0, 0, D->OpEntry->LI);
1495 InsertEntry (D, X, D->IP++);
1497 /* Remove the push and the call to the tossubax function */
1498 RemoveRemainders (D);
1500 /* We changed the sequence */
1506 static unsigned Opt_tosugeax (StackOpData* D)
1507 /* Optimize the tosugeax sequence */
1512 /* Inline the sbc */
1513 D->IP = D->OpIndex+1;
1515 /* Must be true because of OP_RHS_LOAD */
1516 CHECK ((D->Rhs.A.Flags & LI_DIRECT) != 0);
1518 /* If the location is on the stack, we need to reload the Y register. */
1519 if ((D->Rhs.A.Flags & LI_RELOAD_Y) == 0) {
1522 CodeEntry* LoadA = D->Rhs.A.LoadEntry;
1523 X = NewCodeEntry (OP65_CMP, LoadA->AM, LoadA->Arg, 0, D->OpEntry->LI);
1524 InsertEntry (D, X, D->IP++);
1529 const char* Arg = MakeHexArg (D->Rhs.A.Offs);
1530 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, D->OpEntry->LI);
1531 InsertEntry (D, X, D->IP++);
1534 X = NewCodeEntry (OP65_CMP, AM65_ZP_INDY, "sp", 0, D->OpEntry->LI);
1535 InsertEntry (D, X, D->IP++);
1538 /* In both cases, we can remove the load */
1539 D->Rhs.A.Flags |= LI_REMOVE;
1542 X = NewCodeEntry (OP65_TXA, AM65_IMP, 0, 0, D->OpEntry->LI);
1543 InsertEntry (D, X, D->IP++);
1545 /* Must be true because of OP_RHS_LOAD */
1546 CHECK ((D->Rhs.X.Flags & LI_DIRECT) != 0);
1548 /* If the location is on the stack, we need to reload the Y register. */
1549 if ((D->Rhs.X.Flags & LI_RELOAD_Y) == 0) {
1552 CodeEntry* LoadX = D->Rhs.X.LoadEntry;
1553 X = NewCodeEntry (OP65_SBC, LoadX->AM, LoadX->Arg, 0, D->OpEntry->LI);
1554 InsertEntry (D, X, D->IP++);
1559 const char* Arg = MakeHexArg (D->Rhs.X.Offs);
1560 X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, D->OpEntry->LI);
1561 InsertEntry (D, X, D->IP++);
1564 X = NewCodeEntry (OP65_SBC, AM65_ZP_INDY, "sp", 0, D->OpEntry->LI);
1565 InsertEntry (D, X, D->IP++);
1568 /* In both cases, we can remove the load */
1569 D->Rhs.X.Flags |= LI_REMOVE;
1572 X = NewCodeEntry (OP65_LDA, AM65_IMM, "$00", 0, D->OpEntry->LI);
1573 InsertEntry (D, X, D->IP++);
1576 X = NewCodeEntry (OP65_LDX, AM65_IMM, "$00", 0, D->OpEntry->LI);
1577 InsertEntry (D, X, D->IP++);
1580 X = NewCodeEntry (OP65_ROL, AM65_ACC, "a", 0, D->OpEntry->LI);
1581 InsertEntry (D, X, D->IP++);
1583 /* Remove the push and the call to the tosugeax function */
1584 RemoveRemainders (D);
1586 /* We changed the sequence */
1592 static unsigned Opt_tosxorax (StackOpData* D)
1593 /* Optimize the tosxorax sequence */
1598 /* Store the value into the zeropage instead of pushing it */
1599 ReplacePushByStore (D);
1601 /* Inline the xor, low byte */
1602 D->IP = D->OpIndex + 1;
1603 AddOpLow (D, OP65_EOR);
1606 if (RegValIsKnown (D->PushEntry->RI->In.RegX) &&
1607 RegValIsKnown (D->OpEntry->RI->In.RegX)) {
1608 /* Both values known, precalculate the result */
1609 const char* Arg = MakeHexArg (D->PushEntry->RI->In.RegX ^ D->OpEntry->RI->In.RegX);
1610 X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, D->OpEntry->LI);
1611 InsertEntry (D, X, D->IP++);
1612 } else if (D->PushEntry->RI->In.RegX != 0) {
1613 /* High byte is unknown */
1614 AddOpHigh (D, OP65_EOR);
1617 /* Remove the push and the call to the tosandax function */
1618 RemoveRemainders (D);
1620 /* We changed the sequence */
1626 /*****************************************************************************/
1628 /*****************************************************************************/
1632 static const OptFuncDesc FuncTable[] = {
1633 { "__bzero", Opt___bzero, REG_NONE, OP_X_ZERO | OP_A_KNOWN },
1634 { "staspidx", Opt_staspidx, REG_NONE, OP_NONE },
1635 { "staxspidx", Opt_staxspidx, REG_AX, OP_NONE },
1636 { "tosaddax", Opt_tosaddax, REG_NONE, OP_NONE },
1637 { "tosandax", Opt_tosandax, REG_NONE, OP_NONE },
1638 { "toseqax", Opt_toseqax, REG_NONE, OP_NONE },
1639 { "tosgeax", Opt_tosgeax, REG_NONE, OP_RHS_LOAD_DIRECT },
1640 { "tosltax", Opt_tosltax, REG_NONE, OP_RHS_LOAD_DIRECT },
1641 { "tosneax", Opt_tosneax, REG_NONE, OP_NONE },
1642 { "tosorax", Opt_tosorax, REG_NONE, OP_NONE },
1643 { "tossubax", Opt_tossubax, REG_NONE, OP_RHS_LOAD_DIRECT },
1644 { "tosugeax", Opt_tosugeax, REG_NONE, OP_RHS_LOAD_DIRECT },
1645 { "tosxorax", Opt_tosxorax, REG_NONE, OP_NONE },
1647 #define FUNC_COUNT (sizeof(FuncTable) / sizeof(FuncTable[0]))
1651 static int CmpFunc (const void* Key, const void* Func)
1652 /* Compare function for bsearch */
1654 return strcmp (Key, ((const OptFuncDesc*) Func)->Name);
1659 static const OptFuncDesc* FindFunc (const char* Name)
1660 /* Find the function with the given name. Return a pointer to the table entry
1661 * or NULL if the function was not found.
1664 return bsearch (Name, FuncTable, FUNC_COUNT, sizeof(OptFuncDesc), CmpFunc);
1669 static int CmpHarmless (const void* Key, const void* Entry)
1670 /* Compare function for bsearch */
1672 return strcmp (Key, *(const char**)Entry);
1677 static int HarmlessCall (const char* Name)
1678 /* Check if this is a call to a harmless subroutine that will not interrupt
1679 * the pushax/op sequence when encountered.
1682 static const char* Tab[] = {
1705 void* R = bsearch (Name,
1707 sizeof (Tab) / sizeof (Tab[0]),
1715 static void ResetStackOpData (StackOpData* Data)
1716 /* Reset the given data structure */
1719 Data->UsedRegs = REG_NONE;
1721 ClearLoadInfo (&Data->Lhs);
1722 ClearLoadInfo (&Data->Rhs);
1724 Data->PushIndex = -1;
1730 static int PreCondOk (StackOpData* D)
1731 /* Check if the preconditions for a call to the optimizer subfunction are
1732 * satisfied. As a side effect, this function will also choose the zero page
1736 /* Check the flags */
1737 unsigned UnusedRegs = D->OptFunc->UnusedRegs;
1738 if (UnusedRegs != REG_NONE &&
1739 (GetRegInfo (D->Code, D->OpIndex+1, UnusedRegs) & UnusedRegs) != 0) {
1740 /* Cannot optimize */
1743 if ((D->OptFunc->Flags & OP_A_KNOWN) != 0 &&
1744 RegValIsUnknown (D->OpEntry->RI->In.RegA)) {
1745 /* Cannot optimize */
1748 if ((D->OptFunc->Flags & OP_X_ZERO) != 0 &&
1749 D->OpEntry->RI->In.RegX != 0) {
1750 /* Cannot optimize */
1753 if ((D->OptFunc->Flags & OP_LHS_LOAD) != 0) {
1754 if (D->Lhs.A.LoadIndex < 0 || D->Lhs.X.LoadIndex < 0) {
1755 /* Cannot optimize */
1757 } else if ((D->OptFunc->Flags & OP_LHS_LOAD_DIRECT) != 0) {
1758 if ((D->Lhs.A.Flags & D->Lhs.X.Flags & LI_DIRECT) == 0) {
1759 /* Cannot optimize */
1764 if ((D->OptFunc->Flags & OP_RHS_LOAD) != 0) {
1765 if (D->Rhs.A.LoadIndex < 0 || D->Rhs.X.LoadIndex < 0) {
1766 /* Cannot optimize */
1768 } else if ((D->OptFunc->Flags & OP_RHS_LOAD_DIRECT) != 0) {
1769 if ((D->Rhs.A.Flags & D->Rhs.X.Flags & LI_DIRECT) == 0) {
1770 /* Cannot optimize */
1776 /* Determine the zero page locations to use */
1777 if ((D->UsedRegs & REG_PTR1) == REG_NONE) {
1780 } else if ((D->UsedRegs & REG_SREG) == REG_NONE) {
1783 } else if ((D->UsedRegs & REG_PTR2) == REG_NONE) {
1787 /* No registers available */
1791 /* Determine if we have a basic block */
1792 return CS_IsBasicBlock (D->Code, D->PushIndex, D->OpIndex);
1797 /*****************************************************************************/
1799 /*****************************************************************************/
1803 unsigned OptStackOps (CodeSeg* S)
1804 /* Optimize operations that take operands via the stack */
1806 unsigned Changes = 0; /* Number of changes in one run */
1815 } State = Initialize;
1818 /* Generate register info */
1821 /* Remember the code segment in the info struct */
1824 /* Look for a call to pushax followed by a call to some other function
1825 * that takes it's first argument on the stack, and the second argument
1826 * in the primary register.
1827 * It depends on the code between the two if we can handle/transform the
1828 * sequence, so check this code for the following list of things:
1830 * - the range must be a basic block (one entry, one exit)
1831 * - there may not be accesses to local variables with unknown
1832 * offsets (because we have to adjust these offsets).
1833 * - no subroutine calls
1836 * Since we need a zero page register later, do also check the
1837 * intermediate code for zero page use.
1840 while (I < CS_GetEntryCount (S)) {
1842 /* Get the next entry */
1843 CodeEntry* E = CS_GetEntry (S, I);
1845 /* Actions depend on state */
1849 ResetStackOpData (&Data);
1854 /* While searching, track register load insns, so we can tell
1855 * what is in a register once pushax is encountered.
1857 if (CE_IsCallTo (E, "pushax")) {
1861 /* Track load insns */
1862 TrackLoads (&Data.Lhs, E, I);
1867 /* We' found a pushax before. Search for a stack op that may
1868 * follow and in the meantime, track zeropage usage and check
1869 * for code that will disable us from translating the sequence.
1871 if (E->OPC == OP65_JSR) {
1873 /* Subroutine call: Check if this is one of the functions,
1874 * we're going to replace.
1876 Data.OptFunc = FindFunc (E->Arg);
1878 /* Remember the op index and go on */
1883 } else if (!HarmlessCall (E->Arg)) {
1884 /* A call to an unkown subroutine: We need to start
1885 * over after the last pushax. Note: This will also
1886 * happen if we encounter a call to pushax!
1892 /* Track register usage */
1893 Data.UsedRegs |= (E->Use | E->Chg);
1894 TrackLoads (&Data.Rhs, E, I);
1897 } else if (E->Info & OF_STORE && (E->Chg & REG_ZP) == 0) {
1899 /* Too dangerous - there may be a change of a variable
1900 * within the sequence.
1906 } else if ((E->Use & REG_SP) != 0 &&
1907 (E->AM != AM65_ZP_INDY ||
1908 RegValIsUnknown (E->RI->In.RegY) ||
1909 E->RI->In.RegY < 2)) {
1911 /* If we are using the stack, and we don't have "indirect Y"
1912 * addressing mode, or the value of Y is unknown, or less
1913 * than two, we cannot cope with this piece of code. Having
1914 * an unknown value of Y means that we cannot correct the
1915 * stack offset, while having an offset less than two means
1916 * that the code works with the value on stack which is to
1924 /* Other stuff: Track register usage */
1925 Data.UsedRegs |= (E->Use | E->Chg);
1926 TrackLoads (&Data.Rhs, E, I);
1931 /* Track zero page location usage beyond this point */
1932 Data.UsedRegs |= GetRegInfo (S, I, REG_SREG | REG_PTR1 | REG_PTR2);
1934 /* Finalize the load info */
1935 FinalizeLoadInfo (&Data.Lhs, S);
1936 FinalizeLoadInfo (&Data.Rhs, S);
1938 /* Set flags for direct operations */
1939 CheckDirectOp (&Data);
1941 /* If the Lhs loads do load from zeropage, we have to include
1942 * them into UsedRegs registers used. The Rhs loads have already
1945 if (Data.Lhs.A.LoadEntry && Data.Lhs.A.LoadEntry->AM == AM65_ZP) {
1946 Data.UsedRegs |= Data.Lhs.A.LoadEntry->Use;
1948 if (Data.Lhs.X.LoadEntry && Data.Lhs.X.LoadEntry->AM == AM65_ZP) {
1949 Data.UsedRegs |= Data.Lhs.X.LoadEntry->Use;
1952 /* Check the preconditions. If they aren't ok, reset the insn
1953 * pointer to the pushax and start over. We will loose part of
1954 * load tracking but at least a/x has probably lost between
1955 * pushax and here and will be tracked again when restarting.
1957 if (!PreCondOk (&Data)) {
1963 /* Adjust stack offsets to account for the upcoming removal */
1964 AdjustStackOffset (&Data, 2);
1966 /* Regenerate register info, since AdjustStackOffset changed
1971 /* Prepare the remainder of the data structure. */
1972 Data.PrevEntry = CS_GetPrevEntry (S, Data.PushIndex);
1973 Data.PushEntry = CS_GetEntry (S, Data.PushIndex);
1974 Data.OpEntry = CS_GetEntry (S, Data.OpIndex);
1975 Data.NextEntry = CS_GetNextEntry (S, Data.OpIndex);
1977 /* Call the optimizer function */
1978 Changes += Data.OptFunc->Func (&Data);
1980 /* Regenerate register info */
1994 /* Free the register info */
1997 /* Return the number of changes made */