1 /*****************************************************************************/
5 /* Code segment structure */
9 /* (C) 2001 Ullrich von Bassewitz */
11 /* D-70597 Stuttgart */
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 /*****************************************************************************/
58 /*****************************************************************************/
59 /* Helper functions */
60 /*****************************************************************************/
64 static void MoveLabelsToPool (CodeSeg* S, CodeEntry* E)
65 /* Move the labels of the code entry E to the label pool of the code segment */
67 unsigned LabelCount = GetCodeLabelCount (E);
68 while (LabelCount--) {
69 CodeLabel* L = GetCodeLabel (E, LabelCount);
72 CollAppend (&S->Labels, L);
74 CollDeleteAll (&E->Labels);
79 static CodeLabel* FindCodeLabel (CodeSeg* S, const char* Name, unsigned Hash)
80 /* Find the label with the given name. Return the label or NULL if not found */
82 /* Get the first hash chain entry */
83 CodeLabel* L = S->LabelHash[Hash];
87 if (strcmp (Name, L->Name) == 0) {
98 static CodeLabel* NewCodeSegLabel (CodeSeg* S, const char* Name, unsigned Hash)
99 /* Create a new label and insert it into the label hash table */
101 /* Not found - create a new one */
102 CodeLabel* L = NewCodeLabel (Name, Hash);
104 /* Enter the label into the hash table */
105 L->Next = S->LabelHash[L->Hash];
106 S->LabelHash[L->Hash] = L;
108 /* Return the new label */
114 static void RemoveLabelFromHash (CodeSeg* S, CodeLabel* L)
115 /* Remove the given code label from the hash list */
117 /* Get the first entry in the hash chain */
118 CodeLabel* List = S->LabelHash[L->Hash];
121 /* First, remove the label from the hash chain */
123 /* First entry in hash chain */
124 S->LabelHash[L->Hash] = L->Next;
126 /* Must search through the chain */
127 while (List->Next != L) {
128 /* If we've reached the end of the chain, something is *really* wrong */
129 CHECK (List->Next != 0);
133 /* The next entry is the one, we have been searching for */
134 List->Next = L->Next;
140 /*****************************************************************************/
141 /* Functions for parsing instructions */
142 /*****************************************************************************/
146 static const char* SkipSpace (const char* S)
147 /* Skip white space and return an updated pointer */
149 while (IsSpace (*S)) {
157 static const char* ReadToken (const char* L, const char* Term,
158 char* Buf, unsigned BufSize)
159 /* Read the next token into Buf, return the updated line pointer. The
160 * token is terminated by one of the characters given in term.
163 /* Read/copy the token */
165 unsigned ParenCount = 0;
166 while (*L && (ParenCount > 0 || strchr (Term, *L) == 0)) {
172 } else if (*L == '(') {
178 /* Terminate the buffer contents */
181 /* Return the updated line pointer */
187 static CodeEntry* ParseInsn (CodeSeg* S, const char* L)
188 /* Parse an instruction nnd generate a code entry from it. If the line contains
189 * errors, output an error message and return NULL.
190 * For simplicity, we don't accept the broad range of input a "real" assembler
191 * does. The instruction and the argument are expected to be separated by
192 * white space, for example.
197 am_t AM = 0; /* Initialize to keep gcc silent */
204 L = ReadToken (L, " \t", Mnemo, sizeof (Mnemo));
206 /* Try to find the opcode description for the mnemonic */
207 OPC = FindOpcode (Mnemo);
209 /* If we didn't find the opcode, print an error and bail out */
211 Error ("ASM code error: %s is not a valid mnemonic", Mnemo);
215 /* Skip separator white space */
218 /* Get the addressing mode */
229 StrCopy (Arg, sizeof (Arg), L+1);
235 L = ReadToken (L+1, ",)", Arg, sizeof (Arg));
237 /* Check for errors */
239 Error ("ASM code error: syntax error");
243 /* Check the different indirect modes */
245 /* Expect zp x indirect */
247 if (toupper (*L) != 'X') {
248 Error ("ASM code error: `X' expected");
253 Error ("ASM code error: `)' expected");
258 Error ("ASM code error: syntax error");
262 } else if (*L == ')') {
263 /* zp indirect or zp indirect, y */
267 if (toupper (*L) != 'Y') {
268 Error ("ASM code error: `Y' expected");
273 Error ("ASM code error: syntax error");
277 } else if (*L == '\0') {
280 Error ("ASM code error: syntax error");
296 /* Absolute, maybe indexed */
297 L = ReadToken (L, ",", Arg, sizeof (Arg));
299 /* Assume absolute */
301 } else if (*L == ',') {
305 Error ("ASM code error: syntax error");
312 } else if (Reg == 'Y') {
315 Error ("ASM code error: syntax error");
319 Error ("ASM code error: syntax error");
328 /* If the instruction is a branch, check for the label and generate it
329 * if it does not exist. Ignore anything but local labels here.
332 if ((OPC->Info & OF_BRA) != 0 && Arg[0] == 'L') {
336 /* Addressing mode must be alsobute or something is really wrong */
337 CHECK (AM == AM_ABS);
339 /* Addressing mode is a branch/jump */
342 /* Generate the hash over the label, then search for the label */
343 Hash = HashStr (Arg) % CS_LABEL_HASH_SIZE;
344 Label = FindCodeLabel (S, Arg, Hash);
346 /* If we don't have the label, it's a forward ref - create it */
348 /* Generate a new label */
349 Label = NewCodeSegLabel (S, Arg, Hash);
353 /* We do now have the addressing mode in AM. Allocate a new CodeEntry
354 * structure and initialize it.
356 E = NewCodeEntry (OPC, AM, Arg, Label);
358 /* Return the new code entry */
364 /*****************************************************************************/
366 /*****************************************************************************/
370 CodeSeg* NewCodeSeg (const char* SegName, SymEntry* Func)
371 /* Create a new code segment, initialize and return it */
375 /* Allocate memory */
376 CodeSeg* S = xmalloc (sizeof (CodeSeg));
378 /* Initialize the fields */
379 S->SegName = xstrdup (SegName);
381 InitCollection (&S->Entries);
382 InitCollection (&S->Labels);
383 for (I = 0; I < sizeof(S->LabelHash) / sizeof(S->LabelHash[0]); ++I) {
387 /* If we have a function given, get the return type of the function.
388 * Assume ANY return type besides void will use the A and X registers.
390 if (S->Func && !IsTypeVoid (GetFuncReturn (Func->Type))) {
391 S->ExitRegs = REG_AX;
393 S->ExitRegs = REG_NONE;
396 /* Return the new struct */
402 void AddCodeEntry (CodeSeg* S, const char* Format, va_list ap)
403 /* Add a line to the given code segment */
409 /* Format the line */
411 xvsprintf (Buf, sizeof (Buf), Format, ap);
413 /* Skip whitespace */
416 /* Check which type of instruction we have */
417 E = 0; /* Assume no insn created */
421 /* Empty line, just ignore it */
425 /* Comment or hint, ignore it for now */
429 /* Control instruction */
430 ReadToken (L, " \t", Token, sizeof (Token));
431 Error ("ASM code error: Pseudo instruction `%s' not supported", Token);
435 E = ParseInsn (S, L);
439 /* If we have a code entry, transfer the labels and insert it */
442 /* Transfer the labels if we have any */
444 unsigned LabelCount = CollCount (&S->Labels);
445 for (I = 0; I < LabelCount; ++I) {
448 CodeLabel* L = CollAt (&S->Labels, I);
450 /* Attach it to the entry */
451 AttachCodeLabel (E, L);
454 /* Delete the transfered labels */
455 CollDeleteAll (&S->Labels);
457 /* Add the entry to the list of code entries in this segment */
458 CollAppend (&S->Entries, E);
465 void InsertCodeEntry (CodeSeg* S, struct CodeEntry* E, unsigned Index)
466 /* Insert the code entry at the index given. Following code entries will be
467 * moved to slots with higher indices.
470 /* Insert the entry into the collection */
471 CollInsert (&S->Entries, E, Index);
476 void DelCodeEntry (CodeSeg* S, unsigned Index)
477 /* Delete an entry from the code segment. This includes moving any associated
478 * labels, removing references to labels and even removing the referenced labels
479 * if the reference count drops to zero.
482 /* Get the code entry for the given index */
483 CodeEntry* E = GetCodeEntry (S, Index);
485 /* If the entry has a labels, we have to move this label to the next insn.
486 * If there is no next insn, move the label into the code segement label
487 * pool. The operation is further complicated by the fact that the next
488 * insn may already have a label. In that case change all reference to
489 * this label and delete the label instead of moving it.
491 unsigned Count = GetCodeLabelCount (E);
494 /* The instruction has labels attached. Check if there is a next
497 if (Index == GetCodeEntryCount (S)-1) {
499 /* No next instruction, move to the codeseg label pool */
500 MoveLabelsToPool (S, E);
504 /* There is a next insn, get it */
505 CodeEntry* N = GetCodeEntry (S, Index+1);
507 /* Move labels to the next entry */
508 MoveCodeLabels (S, E, N);
513 /* If this insn references a label, remove the reference. And, if the
514 * the reference count for this label drops to zero, remove this label.
517 /* Remove the reference */
518 RemoveCodeLabelRef (S, E);
521 /* Delete the pointer to the insn */
522 CollDelete (&S->Entries, Index);
524 /* Delete the instruction itself */
530 struct CodeEntry* GetCodeEntry (CodeSeg* S, unsigned Index)
531 /* Get an entry from the given code segment */
533 return CollAt (&S->Entries, Index);
538 struct CodeEntry* GetNextCodeEntry (CodeSeg* S, unsigned Index)
539 /* Get the code entry following the one with the index Index. If there is no
540 * following code entry, return NULL.
543 if (Index >= CollCount (&S->Entries)-1) {
544 /* This is the last entry */
547 /* Code entries left */
548 return CollAt (&S->Entries, Index+1);
554 unsigned GetCodeEntryIndex (CodeSeg* S, struct CodeEntry* E)
555 /* Return the index of a code entry */
557 int Index = CollIndex (&S->Entries, E);
564 void AddCodeLabel (CodeSeg* S, const char* Name)
565 /* Add a code label for the next instruction to follow */
567 /* Calculate the hash from the name */
568 unsigned Hash = HashStr (Name) % CS_LABEL_HASH_SIZE;
570 /* Try to find the code label if it does already exist */
571 CodeLabel* L = FindCodeLabel (S, Name, Hash);
573 /* Did we find it? */
575 /* We found it - be sure it does not already have an owner */
576 CHECK (L->Owner == 0);
578 /* Not found - create a new one */
579 L = NewCodeSegLabel (S, Name, Hash);
582 /* Safety. This call is quite costly, but safety is better */
583 if (CollIndex (&S->Labels, L) >= 0) {
584 Internal ("AddCodeLabel: Label `%s' already defined", Name);
587 /* We do now have a valid label. Remember it for later */
588 CollAppend (&S->Labels, L);
593 CodeLabel* GenCodeLabel (CodeSeg* S, struct CodeEntry* E)
594 /* If the code entry E does already have a label, return it. Otherwise
595 * create a new label, attach it to E and return it.
600 if (CodeEntryHasLabel (E)) {
602 /* Get the label from this entry */
603 L = GetCodeLabel (E, 0);
608 const char* Name = LocalLabelName (GetLocalLabel ());
610 /* Generate the hash over the name */
611 unsigned Hash = HashStr (Name) % CS_LABEL_HASH_SIZE;
613 /* Create a new label */
614 L = NewCodeSegLabel (S, Name, Hash);
616 /* Attach this label to the code entry */
617 AttachCodeLabel (E, L);
621 /* Return the label */
627 void DelCodeLabel (CodeSeg* S, CodeLabel* L)
628 /* Remove references from this label and delete it. */
632 /* First, remove the label from the hash chain */
633 RemoveLabelFromHash (S, L);
635 /* Remove references from insns jumping to this label */
636 Count = CollCount (&L->JumpFrom);
637 for (I = 0; I < Count; ++I) {
638 /* Get the insn referencing this label */
639 CodeEntry* E = CollAt (&L->JumpFrom, I);
640 /* Remove the reference */
643 CollDeleteAll (&L->JumpFrom);
645 /* Remove the reference to the owning instruction if it has one. The
646 * function may be called for a label without an owner when deleting
647 * unfinished parts of the code. This is unfortunate since it allows
648 * errors to slip through.
651 CollDeleteItem (&L->Owner->Labels, L);
654 /* All references removed, delete the label itself */
660 void MergeCodeLabels (CodeSeg* S)
661 /* Merge code labels. That means: For each instruction, remove all labels but
662 * one and adjust references accordingly.
667 /* Walk over all code entries */
668 unsigned EntryCount = GetCodeEntryCount (S);
669 for (I = 0; I < EntryCount; ++I) {
674 /* Get a pointer to the next entry */
675 CodeEntry* E = GetCodeEntry (S, I);
677 /* If this entry has zero labels, continue with the next one */
678 unsigned LabelCount = GetCodeLabelCount (E);
679 if (LabelCount == 0) {
683 /* We have at least one label. Use the first one as reference label. */
684 RefLab = GetCodeLabel (E, 0);
686 /* Walk through the remaining labels and change references to these
687 * labels to a reference to the one and only label. Delete the labels
688 * that are no longer used. To increase performance, walk backwards
691 for (J = LabelCount-1; J >= 1; --J) {
693 /* Get the next label */
694 CodeLabel* L = GetCodeLabel (E, J);
696 /* Move all references from this label to the reference label */
697 MoveLabelRefs (L, RefLab);
699 /* Remove the label completely. */
703 /* The reference label is the only remaining label. Check if there
704 * are any references to this label, and delete it if this is not
707 if (CollCount (&RefLab->JumpFrom) == 0) {
708 /* Delete the label */
709 DelCodeLabel (S, RefLab);
716 void MoveCodeLabels (CodeSeg* S, struct CodeEntry* Old, struct CodeEntry* New)
717 /* Move all labels from Old to New. The routine will move the labels itself
718 * if New does not have any labels, and move references if there is at least
719 * a label for new. If references are moved, the old label is deleted
723 /* Get the number of labels to move */
724 unsigned OldLabelCount = GetCodeLabelCount (Old);
726 /* Does the new entry have itself a label? */
727 if (CodeEntryHasLabel (New)) {
729 /* The new entry does already have a label - move references */
730 CodeLabel* NewLabel = GetCodeLabel (New, 0);
731 while (OldLabelCount--) {
733 /* Get the next label */
734 CodeLabel* OldLabel = GetCodeLabel (Old, OldLabelCount);
736 /* Move references */
737 MoveLabelRefs (OldLabel, NewLabel);
739 /* Delete the label */
740 DelCodeLabel (S, OldLabel);
746 /* The new entry does not have a label, just move them */
747 while (OldLabelCount--) {
749 /* Move the label to the new entry */
750 MoveCodeLabel (GetCodeLabel (Old, OldLabelCount), New);
759 void RemoveCodeLabelRef (CodeSeg* S, struct CodeEntry* E)
760 /* Remove the reference between E and the label it jumps to. The reference
761 * will be removed on both sides and E->JumpTo will be 0 after that. If
762 * the reference was the only one for the label, the label will get
766 /* Get a pointer to the label and make sure it exists */
767 CodeLabel* L = E->JumpTo;
770 /* Delete the entry from the label */
771 CollDeleteItem (&L->JumpFrom, E);
773 /* The entry jumps no longer to L */
776 /* If there are no more references, delete the label */
777 if (CollCount (&L->JumpFrom) == 0) {
784 void MoveCodeLabelRef (CodeSeg* S, struct CodeEntry* E, CodeLabel* L)
785 /* Change the reference of E to L instead of the current one. If this
786 * was the only reference to the old label, the old label will get
790 /* Get the old label */
791 CodeLabel* OldLabel = E->JumpTo;
793 /* Be sure that code entry references a label */
794 PRECONDITION (OldLabel != 0);
796 /* Remove the reference to our label */
797 RemoveCodeLabelRef (S, E);
799 /* Use the new label */
805 void AddCodeSegHint (CodeSeg* S, unsigned Hint)
806 /* Add a hint for the preceeding instruction */
810 /* Get the number of entries in this segment */
811 unsigned EntryCount = GetCodeEntryCount (S);
813 /* Must have at least one entry */
814 CHECK (EntryCount > 0);
816 /* Get the last entry */
817 E = GetCodeEntry (S, EntryCount-1);
825 void DelCodeSegAfter (CodeSeg* S, unsigned Last)
826 /* Delete all entries including the given one */
828 /* Get the number of entries in this segment */
829 unsigned Count = GetCodeEntryCount (S);
831 /* First pass: Delete all references to labels. If the reference count
832 * for a label drops to zero, delete it.
837 /* Get the next entry */
838 CodeEntry* E = GetCodeEntry (S, C);
840 /* Check if this entry has a label reference */
842 /* Remove the reference to the label */
843 RemoveCodeLabelRef (S, E);
848 /* Second pass: Delete the instructions. If a label attached to an
849 * instruction still has references, it must be references from outside
850 * the deleted area. Don't delete the label in this case, just make it
851 * ownerless and move it to the label pool.
856 /* Get the next entry */
857 CodeEntry* E = GetCodeEntry (S, C);
859 /* Check if this entry has a label attached */
860 if (CodeEntryHasLabel (E)) {
861 /* Move the labels to the pool and clear the owner pointer */
862 MoveLabelsToPool (S, E);
865 /* Delete the pointer to the entry */
866 CollDelete (&S->Entries, C);
868 /* Delete the entry itself */
875 void OutputCodeSeg (const CodeSeg* S, FILE* F)
876 /* Output the code segment data to a file */
880 /* Get the number of entries in this segment */
881 unsigned Count = GetCodeEntryCount (S);
883 /* If the code segment is empty, bail out here */
888 /* Output the segment directive */
889 fprintf (F, ".segment\t\"%s\"\n\n", S->SegName);
891 /* If this is a segment for a function, enter a function */
893 fprintf (F, ".proc\t_%s\n\n", S->Func->Name);
896 /* Output all entries */
897 for (I = 0; I < Count; ++I) {
901 OutputCodeEntry (CollConstAt (&S->Entries, I), F);
903 /* Print usage info */
904 Use = GetRegInfo ((CodeSeg*) S, I+1);
907 (Use & REG_A)? 'A' : '_',
908 (Use & REG_X)? 'X' : '_',
909 (Use & REG_Y)? 'Y' : '_');
912 /* If this is a segment for a function, leave the function */
914 fprintf (F, "\n.endproc\n\n");
920 unsigned GetCodeEntryCount (const CodeSeg* S)
921 /* Return the number of entries for the given code segment */
923 return CollCount (&S->Entries);