1 /*****************************************************************************/
5 /* Code segment structure */
9 /* (C) 2001-2006, Ullrich von Bassewitz */
10 /* Römerstrasse 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 /*****************************************************************************/
42 #include "debugflag.h"
62 /*****************************************************************************/
63 /* Helper functions */
64 /*****************************************************************************/
68 static void CS_PrintFunctionHeader (const CodeSeg* S, FILE* F)
69 /* Print a comment with the function signature to the given file */
71 /* Get the associated function */
72 const SymEntry* Func = S->Func;
74 /* If this is a global code segment, do nothing */
77 "; ---------------------------------------------------------------\n"
79 PrintFuncSig (F, Func->Name, Func->Type);
82 "; ---------------------------------------------------------------\n"
89 static void CS_MoveLabelsToEntry (CodeSeg* S, CodeEntry* E)
90 /* Move all labels from the label pool to the given entry and remove them
94 /* Transfer the labels if we have any */
96 unsigned LabelCount = CollCount (&S->Labels);
97 for (I = 0; I < LabelCount; ++I) {
100 CodeLabel* L = CollAt (&S->Labels, I);
102 /* Attach it to the entry */
103 CE_AttachLabel (E, L);
106 /* Delete the transfered labels */
107 CollDeleteAll (&S->Labels);
112 static void CS_MoveLabelsToPool (CodeSeg* S, CodeEntry* E)
113 /* Move the labels of the code entry E to the label pool of the code segment */
115 unsigned LabelCount = CE_GetLabelCount (E);
116 while (LabelCount--) {
117 CodeLabel* L = CE_GetLabel (E, LabelCount);
119 CollAppend (&S->Labels, L);
121 CollDeleteAll (&E->Labels);
126 static CodeLabel* CS_FindLabel (CodeSeg* S, const char* Name, unsigned Hash)
127 /* Find the label with the given name. Return the label or NULL if not found */
129 /* Get the first hash chain entry */
130 CodeLabel* L = S->LabelHash[Hash];
132 /* Search the list */
134 if (strcmp (Name, L->Name) == 0) {
145 static CodeLabel* CS_NewCodeLabel (CodeSeg* S, const char* Name, unsigned Hash)
146 /* Create a new label and insert it into the label hash table */
148 /* Create a new label */
149 CodeLabel* L = NewCodeLabel (Name, Hash);
151 /* Enter the label into the hash table */
152 L->Next = S->LabelHash[L->Hash];
153 S->LabelHash[L->Hash] = L;
155 /* Return the new label */
161 static void CS_RemoveLabelFromHash (CodeSeg* S, CodeLabel* L)
162 /* Remove the given code label from the hash list */
164 /* Get the first entry in the hash chain */
165 CodeLabel* List = S->LabelHash[L->Hash];
168 /* First, remove the label from the hash chain */
170 /* First entry in hash chain */
171 S->LabelHash[L->Hash] = L->Next;
173 /* Must search through the chain */
174 while (List->Next != L) {
175 /* If we've reached the end of the chain, something is *really* wrong */
176 CHECK (List->Next != 0);
180 /* The next entry is the one, we have been searching for */
181 List->Next = L->Next;
187 /*****************************************************************************/
188 /* Functions for parsing instructions */
189 /*****************************************************************************/
193 static const char* SkipSpace (const char* S)
194 /* Skip white space and return an updated pointer */
196 while (IsSpace (*S)) {
204 static const char* ReadToken (const char* L, const char* Term,
205 char* Buf, unsigned BufSize)
206 /* Read the next token into Buf, return the updated line pointer. The
207 * token is terminated by one of the characters given in term.
210 /* Read/copy the token */
212 unsigned ParenCount = 0;
213 while (*L && (ParenCount > 0 || strchr (Term, *L) == 0)) {
216 } else if (I == BufSize-1) {
217 /* Cannot store this character, this is an input error (maybe
218 * identifier too long or similar).
220 Error ("ASM code error: syntax error");
225 } else if (*L == '(') {
231 /* Terminate the buffer contents */
234 /* Return the updated line pointer */
240 static CodeEntry* ParseInsn (CodeSeg* S, LineInfo* LI, const char* L)
241 /* Parse an instruction nnd generate a code entry from it. If the line contains
242 * errors, output an error message and return NULL.
243 * For simplicity, we don't accept the broad range of input a "real" assembler
244 * does. The instruction and the argument are expected to be separated by
245 * white space, for example.
248 char Mnemo[IDENTSIZE+10];
250 am_t AM = 0; /* Initialize to keep gcc silent */
251 char Arg[IDENTSIZE+10];
256 /* Read the first token and skip white space after it */
257 L = SkipSpace (ReadToken (L, " \t:", Mnemo, sizeof (Mnemo)));
259 /* Check if we have a label */
262 /* Skip the colon and following white space */
266 CS_AddLabel (S, Mnemo);
268 /* If we have reached end of line, bail out, otherwise a mnemonic
275 L = SkipSpace (ReadToken (L, " \t", Mnemo, sizeof (Mnemo)));
278 /* Try to find the opcode description for the mnemonic */
279 OPC = FindOP65 (Mnemo);
281 /* If we didn't find the opcode, print an error and bail out */
283 Error ("ASM code error: %s is not a valid mnemonic", Mnemo);
287 /* Get the addressing mode */
292 /* Implicit or accu */
293 if (OPC->Info & OF_NOIMP) {
302 StrCopy (Arg, sizeof (Arg), L+1);
308 L = ReadToken (L+1, ",)", Arg, sizeof (Arg));
310 /* Check for errors */
312 Error ("ASM code error: syntax error");
316 /* Check the different indirect modes */
318 /* Expect zp x indirect */
320 if (toupper (*L) != 'X') {
321 Error ("ASM code error: `X' expected");
326 Error ("ASM code error: `)' expected");
331 Error ("ASM code error: syntax error");
335 } else if (*L == ')') {
336 /* zp indirect or zp indirect, y */
340 if (toupper (*L) != 'Y') {
341 Error ("ASM code error: `Y' expected");
346 Error ("ASM code error: syntax error");
350 } else if (*L == '\0') {
353 Error ("ASM code error: syntax error");
369 /* Absolute, maybe indexed */
370 L = ReadToken (L, ",", Arg, sizeof (Arg));
372 /* Absolute, zeropage or branch */
373 if ((OPC->Info & OF_BRA) != 0) {
376 } else if (GetZPInfo(Arg) != 0) {
381 } else if (*L == ',') {
385 Error ("ASM code error: syntax error");
391 if (GetZPInfo(Arg) != 0) {
396 } else if (Reg == 'Y') {
399 Error ("ASM code error: syntax error");
403 Error ("ASM code error: syntax error");
412 /* If the instruction is a branch, check for the label and generate it
413 * if it does not exist. This may lead to unused labels (if the label
414 * is actually an external one) which are removed by the CS_MergeLabels
418 if (AM == AM65_BRA) {
420 /* Generate the hash over the label, then search for the label */
421 unsigned Hash = HashStr (Arg) % CS_LABEL_HASH_SIZE;
422 Label = CS_FindLabel (S, Arg, Hash);
424 /* If we don't have the label, it's a forward ref - create it */
426 /* Generate a new label */
427 Label = CS_NewCodeLabel (S, Arg, Hash);
431 /* We do now have the addressing mode in AM. Allocate a new CodeEntry
432 * structure and initialize it.
434 E = NewCodeEntry (OPC->OPC, AM, Arg, Label, LI);
436 /* Return the new code entry */
442 /*****************************************************************************/
444 /*****************************************************************************/
448 CodeSeg* NewCodeSeg (const char* SegName, SymEntry* Func)
449 /* Create a new code segment, initialize and return it */
454 /* Allocate memory */
455 CodeSeg* S = xmalloc (sizeof (CodeSeg));
457 /* Initialize the fields */
458 S->SegName = xstrdup (SegName);
460 InitCollection (&S->Entries);
461 InitCollection (&S->Labels);
462 for (I = 0; I < sizeof(S->LabelHash) / sizeof(S->LabelHash[0]); ++I) {
466 /* If we have a function given, get the return type of the function.
467 * Assume ANY return type besides void will use the A and X registers.
469 if (S->Func && !IsTypeVoid ((RetType = GetFuncReturn (Func->Type)))) {
470 if (SizeOf (RetType) == SizeOf (type_long)) {
471 S->ExitRegs = REG_EAX;
473 S->ExitRegs = REG_AX;
476 S->ExitRegs = REG_NONE;
479 /* Copy the global optimization settings */
480 S->Optimize = (unsigned char) IS_Get (&Optimize);
481 S->CodeSizeFactor = (unsigned) IS_Get (&CodeSizeFactor);
483 /* Return the new struct */
489 void CS_AddEntry (CodeSeg* S, struct CodeEntry* E)
490 /* Add an entry to the given code segment */
492 /* Transfer the labels if we have any */
493 CS_MoveLabelsToEntry (S, E);
495 /* Add the entry to the list of code entries in this segment */
496 CollAppend (&S->Entries, E);
501 void CS_AddVLine (CodeSeg* S, LineInfo* LI, const char* Format, va_list ap)
502 /* Add a line to the given code segment */
506 char Token[IDENTSIZE+10];
508 /* Format the line */
509 StrBuf Buf = STATIC_STRBUF_INITIALIZER;
510 SB_VPrintf (&Buf, Format, ap);
512 /* Skip whitespace */
513 L = SkipSpace (SB_GetConstBuf (&Buf));
515 /* Check which type of instruction we have */
516 E = 0; /* Assume no insn created */
520 /* Empty line, just ignore it */
524 /* Comment or hint, ignore it for now */
528 /* Control instruction */
529 ReadToken (L, " \t", Token, sizeof (Token));
530 Error ("ASM code error: Pseudo instruction `%s' not supported", Token);
534 E = ParseInsn (S, LI, L);
538 /* If we have a code entry, transfer the labels and insert it */
543 /* Cleanup the string buffer */
549 void CS_AddLine (CodeSeg* S, LineInfo* LI, const char* Format, ...)
550 /* Add a line to the given code segment */
553 va_start (ap, Format);
554 CS_AddVLine (S, LI, Format, ap);
560 void CS_InsertEntry (CodeSeg* S, struct CodeEntry* E, unsigned Index)
561 /* Insert the code entry at the index given. Following code entries will be
562 * moved to slots with higher indices.
565 /* Insert the entry into the collection */
566 CollInsert (&S->Entries, E, Index);
571 void CS_DelEntry (CodeSeg* S, unsigned Index)
572 /* Delete an entry from the code segment. This includes moving any associated
573 * labels, removing references to labels and even removing the referenced labels
574 * if the reference count drops to zero.
575 * Note: Labels are moved forward if possible, that is, they are moved to the
576 * next insn (not the preceeding one).
579 /* Get the code entry for the given index */
580 CodeEntry* E = CS_GetEntry (S, Index);
582 /* If the entry has a labels, we have to move this label to the next insn.
583 * If there is no next insn, move the label into the code segement label
584 * pool. The operation is further complicated by the fact that the next
585 * insn may already have a label. In that case change all reference to
586 * this label and delete the label instead of moving it.
588 unsigned Count = CE_GetLabelCount (E);
591 /* The instruction has labels attached. Check if there is a next
594 if (Index == CS_GetEntryCount (S)-1) {
596 /* No next instruction, move to the codeseg label pool */
597 CS_MoveLabelsToPool (S, E);
601 /* There is a next insn, get it */
602 CodeEntry* N = CS_GetEntry (S, Index+1);
604 /* Move labels to the next entry */
605 CS_MoveLabels (S, E, N);
610 /* If this insn references a label, remove the reference. And, if the
611 * the reference count for this label drops to zero, remove this label.
614 /* Remove the reference */
615 CS_RemoveLabelRef (S, E);
618 /* Delete the pointer to the insn */
619 CollDelete (&S->Entries, Index);
621 /* Delete the instruction itself */
627 void CS_DelEntries (CodeSeg* S, unsigned Start, unsigned Count)
628 /* Delete a range of code entries. This includes removing references to labels,
629 * labels attached to the entries and so on.
632 /* Start deleting the entries from the rear, because this involves less
636 CS_DelEntry (S, Start + Count);
642 void CS_MoveEntries (CodeSeg* S, unsigned Start, unsigned Count, unsigned NewPos)
643 /* Move a range of entries from one position to another. Start is the index
644 * of the first entry to move, Count is the number of entries and NewPos is
645 * the index of the target entry. The entry with the index Start will later
646 * have the index NewPos. All entries with indices NewPos and above are
647 * moved to higher indices. If the code block is moved to the end of the
648 * current code, and if pending labels exist, these labels will get attached
649 * to the first instruction of the moved block (the first one after the
653 /* If NewPos is at the end of the code segment, move any labels from the
654 * label pool to the first instruction of the moved range.
656 if (NewPos == CS_GetEntryCount (S)) {
657 CS_MoveLabelsToEntry (S, CS_GetEntry (S, Start));
660 /* Move the code block to the destination */
661 CollMoveMultiple (&S->Entries, Start, Count, NewPos);
666 struct CodeEntry* CS_GetPrevEntry (CodeSeg* S, unsigned Index)
667 /* Get the code entry preceeding the one with the index Index. If there is no
668 * preceeding code entry, return NULL.
672 /* This is the first entry */
675 /* Previous entry available */
676 return CollAtUnchecked (&S->Entries, Index-1);
682 struct CodeEntry* CS_GetNextEntry (CodeSeg* S, unsigned Index)
683 /* Get the code entry following the one with the index Index. If there is no
684 * following code entry, return NULL.
687 if (Index >= CollCount (&S->Entries)-1) {
688 /* This is the last entry */
691 /* Code entries left */
692 return CollAtUnchecked (&S->Entries, Index+1);
698 int CS_GetEntries (CodeSeg* S, struct CodeEntry** List,
699 unsigned Start, unsigned Count)
700 /* Get Count code entries into List starting at index start. Return true if
701 * we got the lines, return false if not enough lines were available.
704 /* Check if enough entries are available */
705 if (Start + Count > CollCount (&S->Entries)) {
709 /* Copy the entries */
711 *List++ = CollAtUnchecked (&S->Entries, Start++);
714 /* We have the entries */
720 unsigned CS_GetEntryIndex (CodeSeg* S, struct CodeEntry* E)
721 /* Return the index of a code entry */
723 int Index = CollIndex (&S->Entries, E);
730 int CS_RangeHasLabel (CodeSeg* S, unsigned Start, unsigned Count)
731 /* Return true if any of the code entries in the given range has a label
732 * attached. If the code segment does not span the given range, check the
733 * possible span instead.
736 unsigned EntryCount = CS_GetEntryCount(S);
738 /* Adjust count. We expect at least Start to be valid. */
739 CHECK (Start < EntryCount);
740 if (Start + Count > EntryCount) {
741 Count = EntryCount - Start;
744 /* Check each entry. Since we have validated the index above, we may
745 * use the unchecked access function in the loop which is faster.
748 const CodeEntry* E = CollAtUnchecked (&S->Entries, Start++);
749 if (CE_HasLabel (E)) {
754 /* No label in the complete range */
760 CodeLabel* CS_AddLabel (CodeSeg* S, const char* Name)
761 /* Add a code label for the next instruction to follow */
763 /* Calculate the hash from the name */
764 unsigned Hash = HashStr (Name) % CS_LABEL_HASH_SIZE;
766 /* Try to find the code label if it does already exist */
767 CodeLabel* L = CS_FindLabel (S, Name, Hash);
769 /* Did we find it? */
771 /* We found it - be sure it does not already have an owner */
773 Error ("ASM label `%s' is already defined", Name);
777 /* Not found - create a new one */
778 L = CS_NewCodeLabel (S, Name, Hash);
781 /* Safety. This call is quite costly, but safety is better */
782 if (CollIndex (&S->Labels, L) >= 0) {
783 Error ("ASM label `%s' is already defined", Name);
787 /* We do now have a valid label. Remember it for later */
788 CollAppend (&S->Labels, L);
790 /* Return the label */
796 CodeLabel* CS_GenLabel (CodeSeg* S, struct CodeEntry* E)
797 /* If the code entry E does already have a label, return it. Otherwise
798 * create a new label, attach it to E and return it.
803 if (CE_HasLabel (E)) {
805 /* Get the label from this entry */
806 L = CE_GetLabel (E, 0);
811 const char* Name = LocalLabelName (GetLocalLabel ());
813 /* Generate the hash over the name */
814 unsigned Hash = HashStr (Name) % CS_LABEL_HASH_SIZE;
816 /* Create a new label */
817 L = CS_NewCodeLabel (S, Name, Hash);
819 /* Attach this label to the code entry */
820 CE_AttachLabel (E, L);
824 /* Return the label */
830 void CS_DelLabel (CodeSeg* S, CodeLabel* L)
831 /* Remove references from this label and delete it. */
835 /* First, remove the label from the hash chain */
836 CS_RemoveLabelFromHash (S, L);
838 /* Remove references from insns jumping to this label */
839 Count = CollCount (&L->JumpFrom);
840 for (I = 0; I < Count; ++I) {
841 /* Get the insn referencing this label */
842 CodeEntry* E = CollAt (&L->JumpFrom, I);
843 /* Remove the reference */
846 CollDeleteAll (&L->JumpFrom);
848 /* Remove the reference to the owning instruction if it has one. The
849 * function may be called for a label without an owner when deleting
850 * unfinished parts of the code. This is unfortunate since it allows
851 * errors to slip through.
854 CollDeleteItem (&L->Owner->Labels, L);
857 /* All references removed, delete the label itself */
863 void CS_MergeLabels (CodeSeg* S)
864 /* Merge code labels. That means: For each instruction, remove all labels but
865 * one and adjust references accordingly.
871 /* First, remove all labels from the label symbol table that don't have an
872 * owner (this means that they are actually external labels but we didn't
873 * know that previously since they may have also been forward references).
875 for (I = 0; I < CS_LABEL_HASH_SIZE; ++I) {
877 /* Get the first label in this hash chain */
878 CodeLabel** L = &S->LabelHash[I];
880 if ((*L)->Owner == 0) {
882 /* The label does not have an owner, remove it from the chain */
886 /* Cleanup any entries jumping to this label */
887 for (J = 0; J < CL_GetRefCount (X); ++J) {
888 /* Get the entry referencing this label */
889 CodeEntry* E = CL_GetRef (X, J);
890 /* And remove the reference. Do NOT call CE_ClearJumpTo
891 * here, because this will also clear the label name,
892 * which is not what we want.
897 /* Print some debugging output */
899 printf ("Removing unused global label `%s'", X->Name);
902 /* And free the label */
905 /* Label is owned, point to next code label pointer */
911 /* Walk over all code entries */
912 for (I = 0; I < CS_GetEntryCount (S); ++I) {
917 /* Get a pointer to the next entry */
918 CodeEntry* E = CS_GetEntry (S, I);
920 /* If this entry has zero labels, continue with the next one */
921 unsigned LabelCount = CE_GetLabelCount (E);
922 if (LabelCount == 0) {
926 /* We have at least one label. Use the first one as reference label. */
927 RefLab = CE_GetLabel (E, 0);
929 /* Walk through the remaining labels and change references to these
930 * labels to a reference to the one and only label. Delete the labels
931 * that are no longer used. To increase performance, walk backwards
934 for (J = LabelCount-1; J >= 1; --J) {
936 /* Get the next label */
937 CodeLabel* L = CE_GetLabel (E, J);
939 /* Move all references from this label to the reference label */
940 CL_MoveRefs (L, RefLab);
942 /* Remove the label completely. */
946 /* The reference label is the only remaining label. Check if there
947 * are any references to this label, and delete it if this is not
950 if (CollCount (&RefLab->JumpFrom) == 0) {
951 /* Delete the label */
952 CS_DelLabel (S, RefLab);
959 void CS_MoveLabels (CodeSeg* S, struct CodeEntry* Old, struct CodeEntry* New)
960 /* Move all labels from Old to New. The routine will move the labels itself
961 * if New does not have any labels, and move references if there is at least
962 * a label for new. If references are moved, the old label is deleted
966 /* Get the number of labels to move */
967 unsigned OldLabelCount = CE_GetLabelCount (Old);
969 /* Does the new entry have itself a label? */
970 if (CE_HasLabel (New)) {
972 /* The new entry does already have a label - move references */
973 CodeLabel* NewLabel = CE_GetLabel (New, 0);
974 while (OldLabelCount--) {
976 /* Get the next label */
977 CodeLabel* OldLabel = CE_GetLabel (Old, OldLabelCount);
979 /* Move references */
980 CL_MoveRefs (OldLabel, NewLabel);
982 /* Delete the label */
983 CS_DelLabel (S, OldLabel);
989 /* The new entry does not have a label, just move them */
990 while (OldLabelCount--) {
992 /* Move the label to the new entry */
993 CE_MoveLabel (CE_GetLabel (Old, OldLabelCount), New);
1002 void CS_RemoveLabelRef (CodeSeg* S, struct CodeEntry* E)
1003 /* Remove the reference between E and the label it jumps to. The reference
1004 * will be removed on both sides and E->JumpTo will be 0 after that. If
1005 * the reference was the only one for the label, the label will get
1009 /* Get a pointer to the label and make sure it exists */
1010 CodeLabel* L = E->JumpTo;
1013 /* Delete the entry from the label */
1014 CollDeleteItem (&L->JumpFrom, E);
1016 /* The entry jumps no longer to L */
1019 /* If there are no more references, delete the label */
1020 if (CollCount (&L->JumpFrom) == 0) {
1027 void CS_MoveLabelRef (CodeSeg* S, struct CodeEntry* E, CodeLabel* L)
1028 /* Change the reference of E to L instead of the current one. If this
1029 * was the only reference to the old label, the old label will get
1033 /* Get the old label */
1034 CodeLabel* OldLabel = E->JumpTo;
1036 /* Be sure that code entry references a label */
1037 PRECONDITION (OldLabel != 0);
1039 /* Remove the reference to our label */
1040 CS_RemoveLabelRef (S, E);
1042 /* Use the new label */
1048 void CS_DelCodeAfter (CodeSeg* S, unsigned Last)
1049 /* Delete all entries including the given one */
1051 /* Get the number of entries in this segment */
1052 unsigned Count = CS_GetEntryCount (S);
1054 /* First pass: Delete all references to labels. If the reference count
1055 * for a label drops to zero, delete it.
1058 while (Last < C--) {
1060 /* Get the next entry */
1061 CodeEntry* E = CS_GetEntry (S, C);
1063 /* Check if this entry has a label reference */
1065 /* If the label is a label in the label pool and this is the last
1066 * reference to the label, remove the label from the pool.
1068 CodeLabel* L = E->JumpTo;
1069 int Index = CollIndex (&S->Labels, L);
1070 if (Index >= 0 && CollCount (&L->JumpFrom) == 1) {
1071 /* Delete it from the pool */
1072 CollDelete (&S->Labels, Index);
1075 /* Remove the reference to the label */
1076 CS_RemoveLabelRef (S, E);
1081 /* Second pass: Delete the instructions. If a label attached to an
1082 * instruction still has references, it must be references from outside
1083 * the deleted area. Don't delete the label in this case, just make it
1084 * ownerless and move it to the label pool.
1087 while (Last < C--) {
1089 /* Get the next entry */
1090 CodeEntry* E = CS_GetEntry (S, C);
1092 /* Check if this entry has a label attached */
1093 if (CE_HasLabel (E)) {
1094 /* Move the labels to the pool and clear the owner pointer */
1095 CS_MoveLabelsToPool (S, E);
1098 /* Delete the pointer to the entry */
1099 CollDelete (&S->Entries, C);
1101 /* Delete the entry itself */
1108 void CS_ResetMarks (CodeSeg* S, unsigned First, unsigned Last)
1109 /* Remove all user marks from the entries in the given range */
1111 while (First <= Last) {
1112 CE_ResetMark (CS_GetEntry (S, First++));
1118 int CS_IsBasicBlock (CodeSeg* S, unsigned First, unsigned Last)
1119 /* Check if the given code segment range is a basic block. That is, check if
1120 * First is the only entrance and Last is the only exit. This means that no
1121 * jump/branch inside the block may jump to an insn below First or after(!)
1122 * Last, and that no insn may jump into this block from the outside.
1127 /* Don't accept invalid ranges */
1128 CHECK (First <= Last);
1130 /* First pass: Walk over the range and remove all marks from the entries */
1131 CS_ResetMarks (S, First, Last);
1133 /* Second pass: Walk over the range checking all labels. Note: There may be
1134 * label on the first insn which is ok.
1139 /* Get the next entry */
1140 CodeEntry* E = CS_GetEntry (S, I);
1142 /* Check if this entry has one or more labels, if so, check which
1143 * entries jump to this label.
1145 unsigned LabelCount = CE_GetLabelCount (E);
1146 unsigned LabelIndex;
1147 for (LabelIndex = 0; LabelIndex < LabelCount; ++LabelIndex) {
1149 /* Get this label */
1150 CodeLabel* L = CE_GetLabel (E, LabelIndex);
1152 /* Walk over all entries that jump to this label. Check for each
1153 * of the entries if it is out of the range.
1155 unsigned RefCount = CL_GetRefCount (L);
1157 for (RefIndex = 0; RefIndex < RefCount; ++RefIndex) {
1159 /* Get the code entry that jumps here */
1160 CodeEntry* Ref = CL_GetRef (L, RefIndex);
1162 /* Walk over out complete range and check if we find the
1163 * refering entry. This is cheaper than using CS_GetEntryIndex,
1164 * because CS_GetEntryIndex will search the complete code
1165 * segment and not just our range.
1168 for (J = First; J <= Last; ++J) {
1169 if (Ref == CS_GetEntry (S, J)) {
1174 /* We did not find the entry. This means that the jump to
1175 * out code segment entry E came from outside the range,
1176 * which in turn means that the given range is not a basic
1179 CS_ResetMarks (S, First, Last);
1183 /* If we come here, we found the entry. Mark it, so we know
1184 * that the branch to the label is in range.
1194 /* Third pass: Walk again over the range and check all branches. If we
1195 * find a branch that is not marked, its target is not inside the range
1196 * (since we checked all the labels in the range before).
1201 /* Get the next entry */
1202 CodeEntry* E = CS_GetEntry (S, I);
1204 /* Check if this is a branch and if so, if it has a mark */
1205 if (E->Info & (OF_UBRA | OF_CBRA)) {
1206 if (!CE_HasMark (E)) {
1207 /* No mark means not a basic block. Before bailing out, be sure
1208 * to remove the marks from the remaining entries.
1210 CS_ResetMarks (S, I+1, Last);
1214 /* Remove the mark */
1222 /* Done - this is a basic block */
1228 void CS_OutputPrologue (const CodeSeg* S, FILE* F)
1229 /* If the given code segment is a code segment for a function, output the
1230 * assembler prologue into the file. That is: Output a comment header, switch
1231 * to the correct segment and enter the local function scope. If the code
1232 * segment is global, do nothing.
1235 /* Get the function associated with the code segment */
1236 SymEntry* Func = S->Func;
1238 /* If the code segment is associated with a function, print a function
1239 * header and enter a local scope. Be sure to switch to the correct
1240 * segment before outputing the function label.
1243 /* Get the function descriptor */
1244 const FuncDesc* D = GetFuncDesc (Func->Type);
1245 CS_PrintFunctionHeader (S, F);
1246 fprintf (F, ".segment\t\"%s\"\n\n.proc\t_%s", S->SegName, Func->Name);
1247 if (D->Flags & FD_NEAR) {
1248 fputs (": near", F);
1249 } else if (D->Flags & FD_FAR) {
1259 void CS_OutputEpilogue (const CodeSeg* S, FILE* F)
1260 /* If the given code segment is a code segment for a function, output the
1261 * assembler epilogue into the file. That is: Close the local function scope.
1265 fputs ("\n.endproc\n\n", F);
1271 void CS_Output (CodeSeg* S, FILE* F)
1272 /* Output the code segment data to a file */
1277 /* Get the number of entries in this segment */
1278 unsigned Count = CS_GetEntryCount (S);
1280 /* If the code segment is empty, bail out here */
1285 /* Generate register info */
1288 /* Output the segment directive */
1289 fprintf (F, ".segment\t\"%s\"\n\n", S->SegName);
1291 /* Output all entries, prepended by the line information if it has changed */
1293 for (I = 0; I < Count; ++I) {
1294 /* Get the next entry */
1295 const CodeEntry* E = CollConstAt (&S->Entries, I);
1296 /* Check if the line info has changed. If so, output the source line
1297 * if the option is enabled and output debug line info if the debug
1298 * option is enabled.
1301 /* Line info has changed, remember the new line info */
1304 /* Add the source line as a comment. Beware: When line continuation
1305 * was used, the line may contain newlines.
1308 const char* L = LI->Line;
1321 /* Add line debug info */
1323 fprintf (F, "\t.dbg\tline, \"%s\", %u\n",
1324 GetInputName (LI), GetInputLine (LI));
1327 /* Output the code */
1331 /* If debug info is enabled, terminate the last line number information */
1333 fputs ("\t.dbg\tline\n", F);
1336 /* Free register info */
1342 void CS_FreeRegInfo (CodeSeg* S)
1343 /* Free register infos for all instructions */
1346 for (I = 0; I < CS_GetEntryCount (S); ++I) {
1347 CE_FreeRegInfo (CS_GetEntry(S, I));
1353 void CS_GenRegInfo (CodeSeg* S)
1354 /* Generate register infos for all instructions */
1357 RegContents Regs; /* Initial register contents */
1358 RegContents* CurrentRegs; /* Current register contents */
1359 int WasJump; /* True if last insn was a jump */
1360 int Done; /* All runs done flag */
1362 /* Be sure to delete all register infos */
1365 /* We may need two runs to get back references right */
1368 /* Assume we're done after this run */
1371 /* On entry, the register contents are unknown */
1372 RC_Invalidate (&Regs);
1373 CurrentRegs = &Regs;
1375 /* Walk over all insns and note just the changes from one insn to the
1379 for (I = 0; I < CS_GetEntryCount (S); ++I) {
1383 /* Get the next instruction */
1384 CodeEntry* E = CollAtUnchecked (&S->Entries, I);
1386 /* If the instruction has a label, we need some special handling */
1387 unsigned LabelCount = CE_GetLabelCount (E);
1388 if (LabelCount > 0) {
1390 /* Loop over all entry points that jump here. If these entry
1391 * points already have register info, check if all values are
1392 * known and identical. If all values are identical, and the
1393 * preceeding instruction was not an unconditional branch, check
1394 * if the register value on exit of the preceeding instruction
1395 * is also identical. If all these values are identical, the
1396 * value of a register is known, otherwise it is unknown.
1398 CodeLabel* Label = CE_GetLabel (E, 0);
1401 /* Preceeding insn was an unconditional branch */
1402 CodeEntry* J = CL_GetRef(Label, 0);
1406 RC_Invalidate (&Regs);
1410 Regs = *CurrentRegs;
1414 while (Entry < CL_GetRefCount (Label)) {
1415 /* Get this entry */
1416 CodeEntry* J = CL_GetRef (Label, Entry);
1418 /* No register info for this entry. This means that the
1419 * instruction that jumps here is at higher addresses and
1420 * the jump is a backward jump. We need a second run to
1421 * get the register info right in this case. Until then,
1422 * assume unknown register contents.
1425 RC_Invalidate (&Regs);
1428 if (J->RI->Out2.RegA != Regs.RegA) {
1429 Regs.RegA = UNKNOWN_REGVAL;
1431 if (J->RI->Out2.RegX != Regs.RegX) {
1432 Regs.RegX = UNKNOWN_REGVAL;
1434 if (J->RI->Out2.RegY != Regs.RegY) {
1435 Regs.RegY = UNKNOWN_REGVAL;
1437 if (J->RI->Out2.SRegLo != Regs.SRegLo) {
1438 Regs.SRegLo = UNKNOWN_REGVAL;
1440 if (J->RI->Out2.SRegHi != Regs.SRegHi) {
1441 Regs.SRegHi = UNKNOWN_REGVAL;
1443 if (J->RI->Out2.Tmp1 != Regs.Tmp1) {
1444 Regs.Tmp1 = UNKNOWN_REGVAL;
1449 /* Use this register info */
1450 CurrentRegs = &Regs;
1454 /* Generate register info for this instruction */
1455 CE_GenRegInfo (E, CurrentRegs);
1457 /* Remember for the next insn if this insn was an uncondition branch */
1458 WasJump = (E->Info & OF_UBRA) != 0;
1460 /* Output registers for this insn are input for the next */
1461 CurrentRegs = &E->RI->Out;
1463 /* If this insn is a branch on zero flag, we may have more info on
1464 * register contents for one of both flow directions, but only if
1465 * there is a previous instruction.
1467 if ((E->Info & OF_ZBRA) != 0 && (P = CS_GetPrevEntry (S, I)) != 0) {
1469 /* Get the branch condition */
1470 bc_t BC = GetBranchCond (E->OPC);
1472 /* Check the previous instruction */
1484 /* A is zero in one execution flow direction */
1486 E->RI->Out2.RegA = 0;
1488 E->RI->Out.RegA = 0;
1493 /* If this is an immidiate compare, the A register has
1494 * the value of the compare later.
1496 if (CE_IsConstImm (P)) {
1498 E->RI->Out2.RegA = (unsigned char)P->Num;
1500 E->RI->Out.RegA = (unsigned char)P->Num;
1506 /* If this is an immidiate compare, the X register has
1507 * the value of the compare later.
1509 if (CE_IsConstImm (P)) {
1511 E->RI->Out2.RegX = (unsigned char)P->Num;
1513 E->RI->Out.RegX = (unsigned char)P->Num;
1519 /* If this is an immidiate compare, the Y register has
1520 * the value of the compare later.
1522 if (CE_IsConstImm (P)) {
1524 E->RI->Out2.RegY = (unsigned char)P->Num;
1526 E->RI->Out.RegY = (unsigned char)P->Num;
1535 /* X is zero in one execution flow direction */
1537 E->RI->Out2.RegX = 0;
1539 E->RI->Out.RegX = 0;
1547 /* X is zero in one execution flow direction */
1549 E->RI->Out2.RegY = 0;
1551 E->RI->Out.RegY = 0;
1557 /* If the branch is a beq, both A and X are zero at the
1558 * branch target, otherwise they are zero at the next
1562 E->RI->Out2.RegA = E->RI->Out2.RegX = 0;
1564 E->RI->Out.RegA = E->RI->Out.RegX = 0;
1570 /* If the branch is a beq, both A and Y are zero at the
1571 * branch target, otherwise they are zero at the next
1575 E->RI->Out2.RegA = E->RI->Out2.RegY = 0;
1577 E->RI->Out.RegA = E->RI->Out.RegY = 0;