1 /*****************************************************************************/
5 /* Code segment structure */
9 /* (C) 2001-2002 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 /*****************************************************************************/
60 /*****************************************************************************/
61 /* Helper functions */
62 /*****************************************************************************/
66 static void CS_PrintFunctionHeader (const CodeSeg* S, FILE* F)
67 /* Print a comment with the function signature to the given file */
69 /* Get the associated function */
70 const SymEntry* Func = S->Func;
72 /* If this is a global code segment, do nothing */
75 "; ---------------------------------------------------------------\n"
77 PrintFuncSig (F, Func->Name, Func->Type);
80 "; ---------------------------------------------------------------\n"
87 static void CS_MoveLabelsToEntry (CodeSeg* S, CodeEntry* E)
88 /* Move all labels from the label pool to the given entry and remove them
92 /* Transfer the labels if we have any */
94 unsigned LabelCount = CollCount (&S->Labels);
95 for (I = 0; I < LabelCount; ++I) {
98 CodeLabel* L = CollAt (&S->Labels, I);
100 /* Attach it to the entry */
101 CE_AttachLabel (E, L);
104 /* Delete the transfered labels */
105 CollDeleteAll (&S->Labels);
110 static void CS_MoveLabelsToPool (CodeSeg* S, CodeEntry* E)
111 /* Move the labels of the code entry E to the label pool of the code segment */
113 unsigned LabelCount = CE_GetLabelCount (E);
114 while (LabelCount--) {
115 CodeLabel* L = CE_GetLabel (E, LabelCount);
117 CollAppend (&S->Labels, L);
119 CollDeleteAll (&E->Labels);
124 static CodeLabel* CS_FindLabel (CodeSeg* S, const char* Name, unsigned Hash)
125 /* Find the label with the given name. Return the label or NULL if not found */
127 /* Get the first hash chain entry */
128 CodeLabel* L = S->LabelHash[Hash];
130 /* Search the list */
132 if (strcmp (Name, L->Name) == 0) {
143 static CodeLabel* CS_NewCodeLabel (CodeSeg* S, const char* Name, unsigned Hash)
144 /* Create a new label and insert it into the label hash table */
146 /* Create a new label */
147 CodeLabel* L = NewCodeLabel (Name, Hash);
149 /* Enter the label into the hash table */
150 L->Next = S->LabelHash[L->Hash];
151 S->LabelHash[L->Hash] = L;
153 /* Return the new label */
159 static void CS_RemoveLabelFromHash (CodeSeg* S, CodeLabel* L)
160 /* Remove the given code label from the hash list */
162 /* Get the first entry in the hash chain */
163 CodeLabel* List = S->LabelHash[L->Hash];
166 /* First, remove the label from the hash chain */
168 /* First entry in hash chain */
169 S->LabelHash[L->Hash] = L->Next;
171 /* Must search through the chain */
172 while (List->Next != L) {
173 /* If we've reached the end of the chain, something is *really* wrong */
174 CHECK (List->Next != 0);
178 /* The next entry is the one, we have been searching for */
179 List->Next = L->Next;
185 static CodeLabel* CS_AddLabelInternal (CodeSeg* S, const char* Name,
186 void (*ErrorFunc) (const char*, ...))
187 /* Add a code label for the next instruction to follow */
189 /* Calculate the hash from the name */
190 unsigned Hash = HashStr (Name) % CS_LABEL_HASH_SIZE;
192 /* Try to find the code label if it does already exist */
193 CodeLabel* L = CS_FindLabel (S, Name, Hash);
195 /* Did we find it? */
197 /* We found it - be sure it does not already have an owner */
199 ErrorFunc ("ASM label `%s' is already defined", Name);
202 /* Not found - create a new one */
203 L = CS_NewCodeLabel (S, Name, Hash);
206 /* Safety. This call is quite costly, but safety is better */
207 if (CollIndex (&S->Labels, L) >= 0) {
208 ErrorFunc ("ASM label `%s' is already defined", Name);
211 /* We do now have a valid label. Remember it for later */
212 CollAppend (&S->Labels, L);
214 /* Return the label */
220 /*****************************************************************************/
221 /* Functions for parsing instructions */
222 /*****************************************************************************/
226 static const char* SkipSpace (const char* S)
227 /* Skip white space and return an updated pointer */
229 while (IsSpace (*S)) {
237 static const char* ReadToken (const char* L, const char* Term,
238 char* Buf, unsigned BufSize)
239 /* Read the next token into Buf, return the updated line pointer. The
240 * token is terminated by one of the characters given in term.
243 /* Read/copy the token */
245 unsigned ParenCount = 0;
246 while (*L && (ParenCount > 0 || strchr (Term, *L) == 0)) {
249 } else if (I == BufSize-1) {
250 /* Cannot store this character, this is an input error (maybe
251 * identifier too long or similar).
253 Error ("ASM code error: syntax error");
258 } else if (*L == '(') {
264 /* Terminate the buffer contents */
267 /* Return the updated line pointer */
273 static CodeEntry* ParseInsn (CodeSeg* S, LineInfo* LI, const char* L)
274 /* Parse an instruction nnd generate a code entry from it. If the line contains
275 * errors, output an error message and return NULL.
276 * For simplicity, we don't accept the broad range of input a "real" assembler
277 * does. The instruction and the argument are expected to be separated by
278 * white space, for example.
281 char Mnemo[IDENTSIZE+10];
283 am_t AM = 0; /* Initialize to keep gcc silent */
284 char Arg[IDENTSIZE+10];
289 /* Read the first token and skip white space after it */
290 L = SkipSpace (ReadToken (L, " \t:", Mnemo, sizeof (Mnemo)));
292 /* Check if we have a label */
295 /* Skip the colon and following white space */
299 CS_AddLabelInternal (S, Mnemo, Error);
301 /* If we have reached end of line, bail out, otherwise a mnemonic
308 L = SkipSpace (ReadToken (L, " \t", Mnemo, sizeof (Mnemo)));
311 /* Try to find the opcode description for the mnemonic */
312 OPC = FindOP65 (Mnemo);
314 /* If we didn't find the opcode, print an error and bail out */
316 Error ("ASM code error: %s is not a valid mnemonic", Mnemo);
320 /* Get the addressing mode */
331 StrCopy (Arg, sizeof (Arg), L+1);
337 L = ReadToken (L+1, ",)", Arg, sizeof (Arg));
339 /* Check for errors */
341 Error ("ASM code error: syntax error");
345 /* Check the different indirect modes */
347 /* Expect zp x indirect */
349 if (toupper (*L) != 'X') {
350 Error ("ASM code error: `X' expected");
355 Error ("ASM code error: `)' expected");
360 Error ("ASM code error: syntax error");
364 } else if (*L == ')') {
365 /* zp indirect or zp indirect, y */
369 if (toupper (*L) != 'Y') {
370 Error ("ASM code error: `Y' expected");
375 Error ("ASM code error: syntax error");
379 } else if (*L == '\0') {
382 Error ("ASM code error: syntax error");
398 /* Absolute, maybe indexed */
399 L = ReadToken (L, ",", Arg, sizeof (Arg));
401 /* Absolute, zeropage or branch */
402 if ((OPC->Info & OF_BRA) != 0) {
405 } else if (GetZPInfo(Arg) != 0) {
410 } else if (*L == ',') {
414 Error ("ASM code error: syntax error");
420 if (GetZPInfo(Arg) != 0) {
425 } else if (Reg == 'Y') {
428 Error ("ASM code error: syntax error");
432 Error ("ASM code error: syntax error");
441 /* If the instruction is a branch, check for the label and generate it
442 * if it does not exist. This may lead to unused labels (if the label
443 * is actually an external one) which are removed by the CS_MergeLabels
447 if (AM == AM65_BRA) {
449 /* Generate the hash over the label, then search for the label */
450 unsigned Hash = HashStr (Arg) % CS_LABEL_HASH_SIZE;
451 Label = CS_FindLabel (S, Arg, Hash);
453 /* If we don't have the label, it's a forward ref - create it */
455 /* Generate a new label */
456 Label = CS_NewCodeLabel (S, Arg, Hash);
460 /* We do now have the addressing mode in AM. Allocate a new CodeEntry
461 * structure and initialize it.
463 E = NewCodeEntry (OPC->OPC, AM, Arg, Label, LI);
465 /* Return the new code entry */
471 /*****************************************************************************/
473 /*****************************************************************************/
477 CodeSeg* NewCodeSeg (const char* SegName, SymEntry* Func)
478 /* Create a new code segment, initialize and return it */
483 /* Allocate memory */
484 CodeSeg* S = xmalloc (sizeof (CodeSeg));
486 /* Initialize the fields */
487 S->SegName = xstrdup (SegName);
489 InitCollection (&S->Entries);
490 InitCollection (&S->Labels);
491 for (I = 0; I < sizeof(S->LabelHash) / sizeof(S->LabelHash[0]); ++I) {
495 /* If we have a function given, get the return type of the function.
496 * Assume ANY return type besides void will use the A and X registers.
498 if (S->Func && !IsTypeVoid ((RetType = GetFuncReturn (Func->Type)))) {
499 if (SizeOf (RetType) == SizeOf (type_long)) {
500 S->ExitRegs = REG_EAX;
502 S->ExitRegs = REG_AX;
505 S->ExitRegs = REG_NONE;
508 /* Return the new struct */
514 void CS_AddEntry (CodeSeg* S, struct CodeEntry* E)
515 /* Add an entry to the given code segment */
517 /* Transfer the labels if we have any */
518 CS_MoveLabelsToEntry (S, E);
520 /* Add the entry to the list of code entries in this segment */
521 CollAppend (&S->Entries, E);
526 void CS_AddVLine (CodeSeg* S, LineInfo* LI, const char* Format, va_list ap)
527 /* Add a line to the given code segment */
531 char Token[IDENTSIZE+10];
533 /* Format the line */
535 xvsprintf (Buf, sizeof (Buf), Format, ap);
537 /* Skip whitespace */
540 /* Check which type of instruction we have */
541 E = 0; /* Assume no insn created */
545 /* Empty line, just ignore it */
549 /* Comment or hint, ignore it for now */
553 /* Control instruction */
554 ReadToken (L, " \t", Token, sizeof (Token));
555 Error ("ASM code error: Pseudo instruction `%s' not supported", Token);
559 E = ParseInsn (S, LI, L);
563 /* If we have a code entry, transfer the labels and insert it */
571 void CS_AddLine (CodeSeg* S, LineInfo* LI, const char* Format, ...)
572 /* Add a line to the given code segment */
575 va_start (ap, Format);
576 CS_AddVLine (S, LI, Format, ap);
582 void CS_InsertEntry (CodeSeg* S, struct CodeEntry* E, unsigned Index)
583 /* Insert the code entry at the index given. Following code entries will be
584 * moved to slots with higher indices.
587 /* Insert the entry into the collection */
588 CollInsert (&S->Entries, E, Index);
593 void CS_DelEntry (CodeSeg* S, unsigned Index)
594 /* Delete an entry from the code segment. This includes moving any associated
595 * labels, removing references to labels and even removing the referenced labels
596 * if the reference count drops to zero.
597 * Note: Labels are moved forward if possible, that is, they are moved to the
598 * next insn (not the preceeding one).
601 /* Get the code entry for the given index */
602 CodeEntry* E = CS_GetEntry (S, Index);
604 /* If the entry has a labels, we have to move this label to the next insn.
605 * If there is no next insn, move the label into the code segement label
606 * pool. The operation is further complicated by the fact that the next
607 * insn may already have a label. In that case change all reference to
608 * this label and delete the label instead of moving it.
610 unsigned Count = CE_GetLabelCount (E);
613 /* The instruction has labels attached. Check if there is a next
616 if (Index == CS_GetEntryCount (S)-1) {
618 /* No next instruction, move to the codeseg label pool */
619 CS_MoveLabelsToPool (S, E);
623 /* There is a next insn, get it */
624 CodeEntry* N = CS_GetEntry (S, Index+1);
626 /* Move labels to the next entry */
627 CS_MoveLabels (S, E, N);
632 /* If this insn references a label, remove the reference. And, if the
633 * the reference count for this label drops to zero, remove this label.
636 /* Remove the reference */
637 CS_RemoveLabelRef (S, E);
640 /* Delete the pointer to the insn */
641 CollDelete (&S->Entries, Index);
643 /* Delete the instruction itself */
649 void CS_DelEntries (CodeSeg* S, unsigned Start, unsigned Count)
650 /* Delete a range of code entries. This includes removing references to labels,
651 * labels attached to the entries and so on.
654 /* Start deleting the entries from the rear, because this involves less
658 CS_DelEntry (S, Start + Count);
664 void CS_MoveEntries (CodeSeg* S, unsigned Start, unsigned Count, unsigned NewPos)
665 /* Move a range of entries from one position to another. Start is the index
666 * of the first entry to move, Count is the number of entries and NewPos is
667 * the index of the target entry. The entry with the index Start will later
668 * have the index NewPos. All entries with indices NewPos and above are
669 * moved to higher indices. If the code block is moved to the end of the
670 * current code, and if pending labels exist, these labels will get attached
671 * to the first instruction of the moved block (the first one after the
675 /* If NewPos is at the end of the code segment, move any labels from the
676 * label pool to the first instruction of the moved range.
678 if (NewPos == CS_GetEntryCount (S)) {
679 CS_MoveLabelsToEntry (S, CS_GetEntry (S, Start));
682 /* Move the code block to the destination */
683 CollMoveMultiple (&S->Entries, Start, Count, NewPos);
688 struct CodeEntry* CS_GetPrevEntry (CodeSeg* S, unsigned Index)
689 /* Get the code entry preceeding the one with the index Index. If there is no
690 * preceeding code entry, return NULL.
694 /* This is the first entry */
697 /* Previous entry available */
698 return CollAtUnchecked (&S->Entries, Index-1);
704 struct CodeEntry* CS_GetNextEntry (CodeSeg* S, unsigned Index)
705 /* Get the code entry following the one with the index Index. If there is no
706 * following code entry, return NULL.
709 if (Index >= CollCount (&S->Entries)-1) {
710 /* This is the last entry */
713 /* Code entries left */
714 return CollAtUnchecked (&S->Entries, Index+1);
720 int CS_GetEntries (CodeSeg* S, struct CodeEntry** List,
721 unsigned Start, unsigned Count)
722 /* Get Count code entries into List starting at index start. Return true if
723 * we got the lines, return false if not enough lines were available.
726 /* Check if enough entries are available */
727 if (Start + Count > CollCount (&S->Entries)) {
731 /* Copy the entries */
733 *List++ = CollAtUnchecked (&S->Entries, Start++);
736 /* We have the entries */
742 unsigned CS_GetEntryIndex (CodeSeg* S, struct CodeEntry* E)
743 /* Return the index of a code entry */
745 int Index = CollIndex (&S->Entries, E);
752 int CS_RangeHasLabel (CodeSeg* S, unsigned Start, unsigned Count)
753 /* Return true if any of the code entries in the given range has a label
754 * attached. If the code segment does not span the given range, check the
755 * possible span instead.
758 unsigned EntryCount = CS_GetEntryCount(S);
760 /* Adjust count. We expect at least Start to be valid. */
761 CHECK (Start < EntryCount);
762 if (Start + Count > EntryCount) {
763 Count = EntryCount - Start;
766 /* Check each entry. Since we have validated the index above, we may
767 * use the unchecked access function in the loop which is faster.
770 const CodeEntry* E = CollAtUnchecked (&S->Entries, Start++);
771 if (CE_HasLabel (E)) {
776 /* No label in the complete range */
782 CodeLabel* CS_AddLabel (CodeSeg* S, const char* Name)
783 /* Add a code label for the next instruction to follow */
785 return CS_AddLabelInternal (S, Name, Internal);
790 CodeLabel* CS_GenLabel (CodeSeg* S, struct CodeEntry* E)
791 /* If the code entry E does already have a label, return it. Otherwise
792 * create a new label, attach it to E and return it.
797 if (CE_HasLabel (E)) {
799 /* Get the label from this entry */
800 L = CE_GetLabel (E, 0);
805 const char* Name = LocalLabelName (GetLocalLabel ());
807 /* Generate the hash over the name */
808 unsigned Hash = HashStr (Name) % CS_LABEL_HASH_SIZE;
810 /* Create a new label */
811 L = CS_NewCodeLabel (S, Name, Hash);
813 /* Attach this label to the code entry */
814 CE_AttachLabel (E, L);
818 /* Return the label */
824 void CS_DelLabel (CodeSeg* S, CodeLabel* L)
825 /* Remove references from this label and delete it. */
829 /* First, remove the label from the hash chain */
830 CS_RemoveLabelFromHash (S, L);
832 /* Remove references from insns jumping to this label */
833 Count = CollCount (&L->JumpFrom);
834 for (I = 0; I < Count; ++I) {
835 /* Get the insn referencing this label */
836 CodeEntry* E = CollAt (&L->JumpFrom, I);
837 /* Remove the reference */
840 CollDeleteAll (&L->JumpFrom);
842 /* Remove the reference to the owning instruction if it has one. The
843 * function may be called for a label without an owner when deleting
844 * unfinished parts of the code. This is unfortunate since it allows
845 * errors to slip through.
848 CollDeleteItem (&L->Owner->Labels, L);
851 /* All references removed, delete the label itself */
857 void CS_MergeLabels (CodeSeg* S)
858 /* Merge code labels. That means: For each instruction, remove all labels but
859 * one and adjust references accordingly.
865 /* First, remove all labels from the label symbol table that don't have an
866 * owner (this means that they are actually external labels but we didn't
867 * know that previously since they may have also been forward references).
869 for (I = 0; I < CS_LABEL_HASH_SIZE; ++I) {
871 /* Get the first label in this hash chain */
872 CodeLabel** L = &S->LabelHash[I];
874 if ((*L)->Owner == 0) {
876 /* The label does not have an owner, remove it from the chain */
880 /* Cleanup any entries jumping to this label */
881 for (J = 0; J < CL_GetRefCount (X); ++J) {
882 /* Get the entry referencing this label */
883 CodeEntry* E = CL_GetRef (X, J);
884 /* And remove the reference */
888 /* Print some debugging output */
890 printf ("Removing unused global label `%s'", X->Name);
893 /* And free the label */
896 /* Label is owned, point to next code label pointer */
902 /* Walk over all code entries */
903 for (I = 0; I < CS_GetEntryCount (S); ++I) {
908 /* Get a pointer to the next entry */
909 CodeEntry* E = CS_GetEntry (S, I);
911 /* If this entry has zero labels, continue with the next one */
912 unsigned LabelCount = CE_GetLabelCount (E);
913 if (LabelCount == 0) {
917 /* We have at least one label. Use the first one as reference label. */
918 RefLab = CE_GetLabel (E, 0);
920 /* Walk through the remaining labels and change references to these
921 * labels to a reference to the one and only label. Delete the labels
922 * that are no longer used. To increase performance, walk backwards
925 for (J = LabelCount-1; J >= 1; --J) {
927 /* Get the next label */
928 CodeLabel* L = CE_GetLabel (E, J);
930 /* Move all references from this label to the reference label */
931 CL_MoveRefs (L, RefLab);
933 /* Remove the label completely. */
937 /* The reference label is the only remaining label. Check if there
938 * are any references to this label, and delete it if this is not
941 if (CollCount (&RefLab->JumpFrom) == 0) {
942 /* Delete the label */
943 CS_DelLabel (S, RefLab);
950 void CS_MoveLabels (CodeSeg* S, struct CodeEntry* Old, struct CodeEntry* New)
951 /* Move all labels from Old to New. The routine will move the labels itself
952 * if New does not have any labels, and move references if there is at least
953 * a label for new. If references are moved, the old label is deleted
957 /* Get the number of labels to move */
958 unsigned OldLabelCount = CE_GetLabelCount (Old);
960 /* Does the new entry have itself a label? */
961 if (CE_HasLabel (New)) {
963 /* The new entry does already have a label - move references */
964 CodeLabel* NewLabel = CE_GetLabel (New, 0);
965 while (OldLabelCount--) {
967 /* Get the next label */
968 CodeLabel* OldLabel = CE_GetLabel (Old, OldLabelCount);
970 /* Move references */
971 CL_MoveRefs (OldLabel, NewLabel);
973 /* Delete the label */
974 CS_DelLabel (S, OldLabel);
980 /* The new entry does not have a label, just move them */
981 while (OldLabelCount--) {
983 /* Move the label to the new entry */
984 CE_MoveLabel (CE_GetLabel (Old, OldLabelCount), New);
993 void CS_RemoveLabelRef (CodeSeg* S, struct CodeEntry* E)
994 /* Remove the reference between E and the label it jumps to. The reference
995 * will be removed on both sides and E->JumpTo will be 0 after that. If
996 * the reference was the only one for the label, the label will get
1000 /* Get a pointer to the label and make sure it exists */
1001 CodeLabel* L = E->JumpTo;
1004 /* Delete the entry from the label */
1005 CollDeleteItem (&L->JumpFrom, E);
1007 /* The entry jumps no longer to L */
1010 /* If there are no more references, delete the label */
1011 if (CollCount (&L->JumpFrom) == 0) {
1018 void CS_MoveLabelRef (CodeSeg* S, struct CodeEntry* E, CodeLabel* L)
1019 /* Change the reference of E to L instead of the current one. If this
1020 * was the only reference to the old label, the old label will get
1024 /* Get the old label */
1025 CodeLabel* OldLabel = E->JumpTo;
1027 /* Be sure that code entry references a label */
1028 PRECONDITION (OldLabel != 0);
1030 /* Remove the reference to our label */
1031 CS_RemoveLabelRef (S, E);
1033 /* Use the new label */
1039 void CS_DelCodeAfter (CodeSeg* S, unsigned Last)
1040 /* Delete all entries including the given one */
1042 /* Get the number of entries in this segment */
1043 unsigned Count = CS_GetEntryCount (S);
1045 /* First pass: Delete all references to labels. If the reference count
1046 * for a label drops to zero, delete it.
1049 while (Last < C--) {
1051 /* Get the next entry */
1052 CodeEntry* E = CS_GetEntry (S, C);
1054 /* Check if this entry has a label reference */
1056 /* If the label is a label in the label pool and this is the last
1057 * reference to the label, remove the label from the pool.
1059 CodeLabel* L = E->JumpTo;
1060 int Index = CollIndex (&S->Labels, L);
1061 if (Index >= 0 && CollCount (&L->JumpFrom) == 1) {
1062 /* Delete it from the pool */
1063 CollDelete (&S->Labels, Index);
1066 /* Remove the reference to the label */
1067 CS_RemoveLabelRef (S, E);
1072 /* Second pass: Delete the instructions. If a label attached to an
1073 * instruction still has references, it must be references from outside
1074 * the deleted area. Don't delete the label in this case, just make it
1075 * ownerless and move it to the label pool.
1078 while (Last < C--) {
1080 /* Get the next entry */
1081 CodeEntry* E = CS_GetEntry (S, C);
1083 /* Check if this entry has a label attached */
1084 if (CE_HasLabel (E)) {
1085 /* Move the labels to the pool and clear the owner pointer */
1086 CS_MoveLabelsToPool (S, E);
1089 /* Delete the pointer to the entry */
1090 CollDelete (&S->Entries, C);
1092 /* Delete the entry itself */
1099 void CS_ResetMarks (CodeSeg* S, unsigned First, unsigned Last)
1100 /* Remove all user marks from the entries in the given range */
1102 while (First <= Last) {
1103 CE_ResetMark (CS_GetEntry (S, First++));
1109 int CS_IsBasicBlock (CodeSeg* S, unsigned First, unsigned Last)
1110 /* Check if the given code segment range is a basic block. That is, check if
1111 * First is the only entrance and Last is the only exit. This means that no
1112 * jump/branch inside the block may jump to an insn below First or after(!)
1113 * Last, and that no insn may jump into this block from the outside.
1118 /* Don't accept invalid ranges */
1119 CHECK (First <= Last);
1121 /* First pass: Walk over the range and remove all marks from the entries */
1122 CS_ResetMarks (S, First, Last);
1124 /* Second pass: Walk over the range checking all labels. Note: There may be
1125 * label on the first insn which is ok.
1130 /* Get the next entry */
1131 CodeEntry* E = CS_GetEntry (S, I);
1133 /* Check if this entry has one or more labels, if so, check which
1134 * entries jump to this label.
1136 unsigned LabelCount = CE_GetLabelCount (E);
1137 unsigned LabelIndex;
1138 for (LabelIndex = 0; LabelIndex < LabelCount; ++LabelIndex) {
1140 /* Get this label */
1141 CodeLabel* L = CE_GetLabel (E, LabelIndex);
1143 /* Walk over all entries that jump to this label. Check for each
1144 * of the entries if it is out of the range.
1146 unsigned RefCount = CL_GetRefCount (L);
1148 for (RefIndex = 0; RefIndex < RefCount; ++RefIndex) {
1150 /* Get the code entry that jumps here */
1151 CodeEntry* Ref = CL_GetRef (L, RefIndex);
1153 /* Walk over out complete range and check if we find the
1154 * refering entry. This is cheaper than using CS_GetEntryIndex,
1155 * because CS_GetEntryIndex will search the complete code
1156 * segment and not just our range.
1159 for (J = First; J <= Last; ++J) {
1160 if (Ref == CS_GetEntry (S, J)) {
1165 /* We did not find the entry. This means that the jump to
1166 * out code segment entry E came from outside the range,
1167 * which in turn means that the given range is not a basic
1170 CS_ResetMarks (S, First, Last);
1174 /* If we come here, we found the entry. Mark it, so we know
1175 * that the branch to the label is in range.
1185 /* Third pass: Walk again over the range and check all branches. If we
1186 * find a branch that is not marked, its target is not inside the range
1187 * (since we checked all the labels in the range before).
1192 /* Get the next entry */
1193 CodeEntry* E = CS_GetEntry (S, I);
1195 /* Check if this is a branch and if so, if it has a mark */
1196 if (E->Info & (OF_UBRA | OF_CBRA)) {
1197 if (!CE_HasMark (E)) {
1198 /* No mark means not a basic block. Before bailing out, be sure
1199 * to remove the marks from the remaining entries.
1201 CS_ResetMarks (S, I+1, Last);
1205 /* Remove the mark */
1213 /* Done - this is a basic block */
1219 void CS_OutputPrologue (const CodeSeg* S, FILE* F)
1220 /* If the given code segment is a code segment for a function, output the
1221 * assembler prologue into the file. That is: Output a comment header, switch
1222 * to the correct segment and enter the local function scope. If the code
1223 * segment is global, do nothing.
1226 /* Get the function associated with the code segment */
1227 SymEntry* Func = S->Func;
1229 /* If the code segment is associated with a function, print a function
1230 * header and enter a local scope. Be sure to switch to the correct
1231 * segment before outputing the function label.
1234 CS_PrintFunctionHeader (S, F);
1235 fprintf (F, ".segment\t\"%s\"\n\n.proc\t_%s\n\n", S->SegName, Func->Name);
1242 void CS_OutputEpilogue (const CodeSeg* S, FILE* F)
1243 /* If the given code segment is a code segment for a function, output the
1244 * assembler epilogue into the file. That is: Close the local function scope.
1248 fprintf (F, "\n.endproc\n\n");
1254 void CS_Output (const CodeSeg* S, FILE* F)
1255 /* Output the code segment data to a file */
1260 /* Get the number of entries in this segment */
1261 unsigned Count = CS_GetEntryCount (S);
1263 /* If the code segment is empty, bail out here */
1268 /* Output the segment directive */
1269 fprintf (F, ".segment\t\"%s\"\n\n", S->SegName);
1271 /* Output all entries, prepended by the line information if it has changed */
1273 for (I = 0; I < Count; ++I) {
1274 /* Get the next entry */
1275 const CodeEntry* E = CollConstAt (&S->Entries, I);
1276 /* Check if the line info has changed. If so, output the source line
1277 * if the option is enabled and output debug line info if the debug
1278 * option is enabled.
1281 /* Line info has changed, remember the new line info */
1284 /* Add the source line as a comment */
1286 fprintf (F, ";\n; %s\n;\n", LI->Line);
1289 /* Add line debug info */
1291 fprintf (F, "\t.dbg\tline, \"%s\", %u\n",
1292 GetInputName (LI), GetInputLine (LI));
1295 /* Output the code */
1299 /* If debug info is enabled, terminate the last line number information */
1301 fprintf (F, "\t.dbg\tline\n");
1307 void CS_FreeRegInfo (CodeSeg* S)
1308 /* Free register infos for all instructions */
1311 for (I = 0; I < CS_GetEntryCount (S); ++I) {
1312 CE_FreeRegInfo (CS_GetEntry(S, I));
1318 void CS_GenRegInfo (CodeSeg* S)
1319 /* Generate register infos for all instructions */
1322 RegContents Regs; /* Initial register contents */
1323 RegContents* CurrentRegs; /* Current register contents */
1324 int WasJump; /* True if last insn was a jump */
1325 int Done; /* All runs done flag */
1327 /* Be sure to delete all register infos */
1330 /* We may need two runs to get back references right */
1333 /* Assume we're done after this run */
1336 /* On entry, the register contents are unknown */
1337 RC_Invalidate (&Regs);
1338 CurrentRegs = &Regs;
1340 /* Walk over all insns and note just the changes from one insn to the
1344 for (I = 0; I < CS_GetEntryCount (S); ++I) {
1348 /* Get the next instruction */
1349 CodeEntry* E = CollAtUnchecked (&S->Entries, I);
1351 /* If the instruction has a label, we need some special handling */
1352 unsigned LabelCount = CE_GetLabelCount (E);
1353 if (LabelCount > 0) {
1355 /* Loop over all entry points that jump here. If these entry
1356 * points already have register info, check if all values are
1357 * known and identical. If all values are identical, and the
1358 * preceeding instruction was not an unconditional branch, check
1359 * if the register value on exit of the preceeding instruction
1360 * is also identical. If all these values are identical, the
1361 * value of a register is known, otherwise it is unknown.
1363 CodeLabel* Label = CE_GetLabel (E, 0);
1366 /* Preceeding insn was an unconditional branch */
1367 CodeEntry* J = CL_GetRef(Label, 0);
1371 RC_Invalidate (&Regs);
1375 Regs = *CurrentRegs;
1379 while (Entry < CL_GetRefCount (Label)) {
1380 /* Get this entry */
1381 CodeEntry* J = CL_GetRef (Label, Entry);
1383 /* No register info for this entry. This means that the
1384 * instruction that jumps here is at higher addresses and
1385 * the jump is a backward jump. We need a second run to
1386 * get the register info right in this case. Until then,
1387 * assume unknown register contents.
1390 RC_Invalidate (&Regs);
1393 if (J->RI->Out2.RegA != Regs.RegA) {
1396 if (J->RI->Out2.RegX != Regs.RegX) {
1399 if (J->RI->Out2.RegY != Regs.RegY) {
1402 if (J->RI->Out2.SRegLo != Regs.SRegLo) {
1405 if (J->RI->Out2.SRegHi != Regs.SRegHi) {
1411 /* Use this register info */
1412 CurrentRegs = &Regs;
1416 /* Generate register info for this instruction */
1417 CE_GenRegInfo (E, CurrentRegs);
1419 /* Remember for the next insn if this insn was an uncondition branch */
1420 WasJump = (E->Info & OF_UBRA) != 0;
1422 /* Output registers for this insn are input for the next */
1423 CurrentRegs = &E->RI->Out;
1425 /* If this insn is a branch on zero flag, we may have more info on
1426 * register contents for one of both flow directions, but only if
1427 * there is a previous instruction.
1429 if ((E->Info & OF_ZBRA) != 0 && (P = CS_GetPrevEntry (S, I)) != 0) {
1431 /* Get the branch condition */
1432 bc_t BC = GetBranchCond (E->OPC);
1434 /* Check the previous instruction */
1446 /* A is zero in one execution flow direction */
1448 E->RI->Out2.RegA = 0;
1450 E->RI->Out.RegA = 0;
1455 /* If this is an immidiate compare, the A register has
1456 * the value of the compare later.
1458 if (CE_KnownImm (P)) {
1460 E->RI->Out2.RegA = (unsigned char)P->Num;
1462 E->RI->Out.RegA = (unsigned char)P->Num;
1468 /* If this is an immidiate compare, the X register has
1469 * the value of the compare later.
1471 if (CE_KnownImm (P)) {
1473 E->RI->Out2.RegX = (unsigned char)P->Num;
1475 E->RI->Out.RegX = (unsigned char)P->Num;
1481 /* If this is an immidiate compare, the Y register has
1482 * the value of the compare later.
1484 if (CE_KnownImm (P)) {
1486 E->RI->Out2.RegY = (unsigned char)P->Num;
1488 E->RI->Out.RegY = (unsigned char)P->Num;
1497 /* X is zero in one execution flow direction */
1499 E->RI->Out2.RegX = 0;
1501 E->RI->Out.RegX = 0;
1509 /* X is zero in one execution flow direction */
1511 E->RI->Out2.RegY = 0;
1513 E->RI->Out.RegY = 0;
1519 /* If the branch is a beq, both A and X are zero at the
1520 * branch target, otherwise they are zero at the next
1524 E->RI->Out2.RegA = E->RI->Out2.RegX = 0;
1526 E->RI->Out.RegA = E->RI->Out.RegX = 0;
1532 /* If the branch is a beq, both A and Y are zero at the
1533 * branch target, otherwise they are zero at the next
1537 E->RI->Out2.RegA = E->RI->Out2.RegY = 0;
1539 E->RI->Out.RegA = E->RI->Out.RegY = 0;