X-Git-Url: https://git.sur5r.net/?a=blobdiff_plain;f=src%2Fcc65%2Fcodeseg.c;h=1c7c0b5f6d89ee6878e1c867ad51091f9a4f2707;hb=a84080febe4bffb7382c763e756d185c3e812646;hp=6565d2b8fc496544c2a9471dff668e0eb973c776;hpb=4c19a5b69938f653aa8718880cdc5e390a7f0ff8;p=cc65 diff --git a/src/cc65/codeseg.c b/src/cc65/codeseg.c index 6565d2b8f..1c7c0b5f6 100644 --- a/src/cc65/codeseg.c +++ b/src/cc65/codeseg.c @@ -6,10 +6,10 @@ /* */ /* */ /* */ -/* (C) 2001 Ullrich von Bassewitz */ -/* Wacholderweg 14 */ -/* D-70597 Stuttgart */ -/* EMail: uz@cc65.org */ +/* (C) 2001-2006, Ullrich von Bassewitz */ +/* Römerstrasse 52 */ +/* D-70794 Filderstadt */ +/* EMail: uz@cc65.org */ /* */ /* */ /* This software is provided 'as-is', without any expressed or implied */ @@ -39,19 +39,23 @@ /* common */ #include "chartype.h" #include "check.h" +#include "debugflag.h" +#include "global.h" #include "hashstr.h" +#include "strbuf.h" #include "strutil.h" #include "xmalloc.h" -#include "xsprintf.h" /* cc65 */ #include "asmlabel.h" #include "codeent.h" #include "codeinfo.h" +#include "codeseg.h" #include "datatype.h" #include "error.h" +#include "global.h" +#include "ident.h" #include "symentry.h" -#include "codeseg.h" @@ -61,13 +65,56 @@ -static void MoveLabelsToPool (CodeSeg* S, CodeEntry* E) +static void CS_PrintFunctionHeader (const CodeSeg* S, FILE* F) +/* Print a comment with the function signature to the given file */ +{ + /* Get the associated function */ + const SymEntry* Func = S->Func; + + /* If this is a global code segment, do nothing */ + if (Func) { + fprintf (F, + "; ---------------------------------------------------------------\n" + "; "); + PrintFuncSig (F, Func->Name, Func->Type); + fprintf (F, + "\n" + "; ---------------------------------------------------------------\n" + "\n"); + } +} + + + +static void CS_MoveLabelsToEntry (CodeSeg* S, CodeEntry* E) +/* Move all labels from the label pool to the given entry and remove them + * from the pool. + */ +{ + /* Transfer the labels if we have any */ + unsigned I; + unsigned LabelCount = CollCount (&S->Labels); + for (I = 0; I < LabelCount; ++I) { + + /* Get the label */ + CodeLabel* L = CollAt (&S->Labels, I); + + /* Attach it to the entry */ + CE_AttachLabel (E, L); + } + + /* Delete the transfered labels */ + CollDeleteAll (&S->Labels); +} + + + +static void CS_MoveLabelsToPool (CodeSeg* S, CodeEntry* E) /* Move the labels of the code entry E to the label pool of the code segment */ { - unsigned LabelCount = GetCodeLabelCount (E); + unsigned LabelCount = CE_GetLabelCount (E); while (LabelCount--) { - CodeLabel* L = GetCodeLabel (E, LabelCount); - L->Flags &= ~LF_DEF; + CodeLabel* L = CE_GetLabel (E, LabelCount); L->Owner = 0; CollAppend (&S->Labels, L); } @@ -76,7 +123,7 @@ static void MoveLabelsToPool (CodeSeg* S, CodeEntry* E) -static CodeLabel* FindCodeLabel (CodeSeg* S, const char* Name, unsigned Hash) +static CodeLabel* CS_FindLabel (CodeSeg* S, const char* Name, unsigned Hash) /* Find the label with the given name. Return the label or NULL if not found */ { /* Get the first hash chain entry */ @@ -95,10 +142,10 @@ static CodeLabel* FindCodeLabel (CodeSeg* S, const char* Name, unsigned Hash) -static CodeLabel* NewCodeSegLabel (CodeSeg* S, const char* Name, unsigned Hash) +static CodeLabel* CS_NewCodeLabel (CodeSeg* S, const char* Name, unsigned Hash) /* Create a new label and insert it into the label hash table */ { - /* Not found - create a new one */ + /* Create a new label */ CodeLabel* L = NewCodeLabel (Name, Hash); /* Enter the label into the hash table */ @@ -111,7 +158,7 @@ static CodeLabel* NewCodeSegLabel (CodeSeg* S, const char* Name, unsigned Hash) -static void RemoveLabelFromHash (CodeSeg* S, CodeLabel* L) +static void CS_RemoveLabelFromHash (CodeSeg* S, CodeLabel* L) /* Remove the given code label from the hash list */ { /* Get the first entry in the hash chain */ @@ -155,7 +202,7 @@ static const char* SkipSpace (const char* S) static const char* ReadToken (const char* L, const char* Term, - char* Buf, unsigned BufSize) + char* Buf, unsigned BufSize) /* Read the next token into Buf, return the updated line pointer. The * token is terminated by one of the characters given in term. */ @@ -165,8 +212,14 @@ static const char* ReadToken (const char* L, const char* Term, unsigned ParenCount = 0; while (*L && (ParenCount > 0 || strchr (Term, *L) == 0)) { if (I < BufSize-1) { - Buf[I++] = *L; + Buf[I] = *L; + } else if (I == BufSize-1) { + /* Cannot store this character, this is an input error (maybe + * identifier too long or similar). + */ + Error ("ASM code error: syntax error"); } + ++I; if (*L == ')') { --ParenCount; } else if (*L == '(') { @@ -184,7 +237,7 @@ static const char* ReadToken (const char* L, const char* Term, -static CodeEntry* ParseInsn (CodeSeg* S, const char* L) +static CodeEntry* ParseInsn (CodeSeg* S, LineInfo* LI, const char* L) /* Parse an instruction nnd generate a code entry from it. If the line contains * errors, output an error message and return NULL. * For simplicity, we don't accept the broad range of input a "real" assembler @@ -192,19 +245,38 @@ static CodeEntry* ParseInsn (CodeSeg* S, const char* L) * white space, for example. */ { - char Mnemo[16]; - const OPCDesc* OPC; - am_t AM = 0; /* Initialize to keep gcc silent */ - char Arg[64]; - char Reg; + char Mnemo[IDENTSIZE+10]; + const OPCDesc* OPC; + am_t AM = 0; /* Initialize to keep gcc silent */ + char Arg[IDENTSIZE+10]; + char Reg; CodeEntry* E; - CodeLabel* Label; + CodeLabel* Label; + + /* Read the first token and skip white space after it */ + L = SkipSpace (ReadToken (L, " \t:", Mnemo, sizeof (Mnemo))); + + /* Check if we have a label */ + if (*L == ':') { + + /* Skip the colon and following white space */ + L = SkipSpace (L+1); + + /* Add the label */ + CS_AddLabel (S, Mnemo); - /* Mnemonic */ - L = ReadToken (L, " \t", Mnemo, sizeof (Mnemo)); + /* If we have reached end of line, bail out, otherwise a mnemonic + * may follow. + */ + if (*L == '\0') { + return 0; + } + + L = SkipSpace (ReadToken (L, " \t", Mnemo, sizeof (Mnemo))); + } /* Try to find the opcode description for the mnemonic */ - OPC = FindOpcode (Mnemo); + OPC = FindOP65 (Mnemo); /* If we didn't find the opcode, print an error and bail out */ if (OPC == 0) { @@ -212,22 +284,23 @@ static CodeEntry* ParseInsn (CodeSeg* S, const char* L) return 0; } - /* Skip separator white space */ - L = SkipSpace (L); - /* Get the addressing mode */ Arg[0] = '\0'; switch (*L) { case '\0': - /* Implicit */ - AM = AM_IMP; + /* Implicit or accu */ + if (OPC->Info & OF_NOIMP) { + AM = AM65_ACC; + } else { + AM = AM65_IMP; + } break; case '#': /* Immidiate */ StrCopy (Arg, sizeof (Arg), L+1); - AM = AM_IMM; + AM = AM65_IMM; break; case '(': @@ -235,7 +308,7 @@ static CodeEntry* ParseInsn (CodeSeg* S, const char* L) L = ReadToken (L+1, ",)", Arg, sizeof (Arg)); /* Check for errors */ - if (*L == '\0') { + if (*L == '\0') { Error ("ASM code error: syntax error"); return 0; } @@ -254,11 +327,11 @@ static CodeEntry* ParseInsn (CodeSeg* S, const char* L) return 0; } L = SkipSpace (L+1); - if (*L != '\0') { + if (*L != '\0') { Error ("ASM code error: syntax error"); return 0; } - AM = AM_ZPX_IND; + AM = AM65_ZPX_IND; } else if (*L == ')') { /* zp indirect or zp indirect, y */ L = SkipSpace (L+1); @@ -266,19 +339,19 @@ static CodeEntry* ParseInsn (CodeSeg* S, const char* L) L = SkipSpace (L+1); if (toupper (*L) != 'Y') { Error ("ASM code error: `Y' expected"); - return 0; + return 0; } L = SkipSpace (L+1); if (*L != '\0') { Error ("ASM code error: syntax error"); return 0; } - AM = AM_ZP_INDY; + AM = AM65_ZP_INDY; } else if (*L == '\0') { - AM = AM_ZP_IND; + AM = AM65_ZP_IND; } else { Error ("ASM code error: syntax error"); - return 0; + return 0; } } break; @@ -287,7 +360,7 @@ static CodeEntry* ParseInsn (CodeSeg* S, const char* L) case 'A': /* Accumulator? */ if (L[1] == '\0') { - AM = AM_ACC; + AM = AM65_ACC; break; } /* FALLTHROUGH */ @@ -296,8 +369,15 @@ static CodeEntry* ParseInsn (CodeSeg* S, const char* L) /* Absolute, maybe indexed */ L = ReadToken (L, ",", Arg, sizeof (Arg)); if (*L == '\0') { - /* Assume absolute */ - AM = AM_ABS; + /* Absolute, zeropage or branch */ + if ((OPC->Info & OF_BRA) != 0) { + /* Branch */ + AM = AM65_BRA; + } else if (GetZPInfo(Arg) != 0) { + AM = AM65_ZP; + } else { + AM = AM65_ABS; + } } else if (*L == ',') { /* Indexed */ L = SkipSpace (L+1); @@ -308,9 +388,13 @@ static CodeEntry* ParseInsn (CodeSeg* S, const char* L) Reg = toupper (*L); L = SkipSpace (L+1); if (Reg == 'X') { - AM = AM_ABSX; + if (GetZPInfo(Arg) != 0) { + AM = AM65_ZPX; + } else { + AM = AM65_ABSX; + } } else if (Reg == 'Y') { - AM = AM_ABSY; + AM = AM65_ABSY; } else { Error ("ASM code error: syntax error"); return 0; @@ -326,34 +410,28 @@ static CodeEntry* ParseInsn (CodeSeg* S, const char* L) } /* If the instruction is a branch, check for the label and generate it - * if it does not exist. Ignore anything but local labels here. + * if it does not exist. This may lead to unused labels (if the label + * is actually an external one) which are removed by the CS_MergeLabels + * function later. */ Label = 0; - if ((OPC->Info & OF_BRA) != 0 && Arg[0] == 'L') { - - unsigned Hash; - - /* Addressing mode must be alsobute or something is really wrong */ - CHECK (AM == AM_ABS); - - /* Addressing mode is a branch/jump */ - AM = AM_BRA; + if (AM == AM65_BRA) { /* Generate the hash over the label, then search for the label */ - Hash = HashStr (Arg) % CS_LABEL_HASH_SIZE; - Label = FindCodeLabel (S, Arg, Hash); + unsigned Hash = HashStr (Arg) % CS_LABEL_HASH_SIZE; + Label = CS_FindLabel (S, Arg, Hash); /* If we don't have the label, it's a forward ref - create it */ if (Label == 0) { /* Generate a new label */ - Label = NewCodeSegLabel (S, Arg, Hash); + Label = CS_NewCodeLabel (S, Arg, Hash); } } /* We do now have the addressing mode in AM. Allocate a new CodeEntry * structure and initialize it. */ - E = NewCodeEntry (OPC->OPC, AM, Arg, Label); + E = NewCodeEntry (OPC->OPC, AM, Arg, Label, LI); /* Return the new code entry */ return E; @@ -362,7 +440,7 @@ static CodeEntry* ParseInsn (CodeSeg* S, const char* L) /*****************************************************************************/ -/* Code */ +/* Code */ /*****************************************************************************/ @@ -371,6 +449,7 @@ CodeSeg* NewCodeSeg (const char* SegName, SymEntry* Func) /* Create a new code segment, initialize and return it */ { unsigned I; + const Type* RetType; /* Allocate memory */ CodeSeg* S = xmalloc (sizeof (CodeSeg)); @@ -387,34 +466,54 @@ CodeSeg* NewCodeSeg (const char* SegName, SymEntry* Func) /* If we have a function given, get the return type of the function. * Assume ANY return type besides void will use the A and X registers. */ - if (S->Func && !IsTypeVoid (GetFuncReturn (Func->Type))) { - S->ExitRegs = REG_AX; + if (S->Func && !IsTypeVoid ((RetType = GetFuncReturn (Func->Type)))) { + if (SizeOf (RetType) == SizeOf (type_long)) { + S->ExitRegs = REG_EAX; + } else { + S->ExitRegs = REG_AX; + } } else { S->ExitRegs = REG_NONE; } + /* Copy the global optimization settings */ + S->Optimize = (unsigned char) IS_Get (&Optimize); + S->CodeSizeFactor = (unsigned) IS_Get (&CodeSizeFactor); + /* Return the new struct */ return S; } -void AddCodeEntry (CodeSeg* S, const char* Format, va_list ap) +void CS_AddEntry (CodeSeg* S, struct CodeEntry* E) +/* Add an entry to the given code segment */ +{ + /* Transfer the labels if we have any */ + CS_MoveLabelsToEntry (S, E); + + /* Add the entry to the list of code entries in this segment */ + CollAppend (&S->Entries, E); +} + + + +void CS_AddVLine (CodeSeg* S, LineInfo* LI, const char* Format, va_list ap) /* Add a line to the given code segment */ { const char* L; CodeEntry* E; - char Token[64]; + char Token[IDENTSIZE+10]; /* Format the line */ - char Buf [256]; - xvsprintf (Buf, sizeof (Buf), Format, ap); + StrBuf Buf = STATIC_STRBUF_INITIALIZER; + SB_VPrintf (&Buf, Format, ap); /* Skip whitespace */ - L = SkipSpace (Buf); + L = SkipSpace (SB_GetConstBuf (&Buf)); /* Check which type of instruction we have */ - E = 0; /* Assume no insn created */ + E = 0; /* Assume no insn created */ switch (*L) { case '\0': @@ -432,37 +531,33 @@ void AddCodeEntry (CodeSeg* S, const char* Format, va_list ap) break; default: - E = ParseInsn (S, L); + E = ParseInsn (S, LI, L); break; } /* If we have a code entry, transfer the labels and insert it */ if (E) { + CS_AddEntry (S, E); + } - /* Transfer the labels if we have any */ - unsigned I; - unsigned LabelCount = CollCount (&S->Labels); - for (I = 0; I < LabelCount; ++I) { - - /* Get the label */ - CodeLabel* L = CollAt (&S->Labels, I); - - /* Attach it to the entry */ - AttachCodeLabel (E, L); - } + /* Cleanup the string buffer */ + DoneStrBuf (&Buf); +} - /* Delete the transfered labels */ - CollDeleteAll (&S->Labels); - /* Add the entry to the list of code entries in this segment */ - CollAppend (&S->Entries, E); - } +void CS_AddLine (CodeSeg* S, LineInfo* LI, const char* Format, ...) +/* Add a line to the given code segment */ +{ + va_list ap; + va_start (ap, Format); + CS_AddVLine (S, LI, Format, ap); + va_end (ap); } -void InsertCodeEntry (CodeSeg* S, struct CodeEntry* E, unsigned Index) +void CS_InsertEntry (CodeSeg* S, struct CodeEntry* E, unsigned Index) /* Insert the code entry at the index given. Following code entries will be * moved to slots with higher indices. */ @@ -473,14 +568,16 @@ void InsertCodeEntry (CodeSeg* S, struct CodeEntry* E, unsigned Index) -void DelCodeEntry (CodeSeg* S, unsigned Index) +void CS_DelEntry (CodeSeg* S, unsigned Index) /* Delete an entry from the code segment. This includes moving any associated * labels, removing references to labels and even removing the referenced labels * if the reference count drops to zero. + * Note: Labels are moved forward if possible, that is, they are moved to the + * next insn (not the preceeding one). */ { /* Get the code entry for the given index */ - CodeEntry* E = GetCodeEntry (S, Index); + CodeEntry* E = CS_GetEntry (S, Index); /* If the entry has a labels, we have to move this label to the next insn. * If there is no next insn, move the label into the code segement label @@ -488,24 +585,24 @@ void DelCodeEntry (CodeSeg* S, unsigned Index) * insn may already have a label. In that case change all reference to * this label and delete the label instead of moving it. */ - unsigned Count = GetCodeLabelCount (E); + unsigned Count = CE_GetLabelCount (E); if (Count > 0) { /* The instruction has labels attached. Check if there is a next * instruction. */ - if (Index == GetCodeEntryCount (S)-1) { + if (Index == CS_GetEntryCount (S)-1) { /* No next instruction, move to the codeseg label pool */ - MoveLabelsToPool (S, E); + CS_MoveLabelsToPool (S, E); } else { /* There is a next insn, get it */ - CodeEntry* N = GetCodeEntry (S, Index+1); + CodeEntry* N = CS_GetEntry (S, Index+1); /* Move labels to the next entry */ - MoveCodeLabels (S, E, N); + CS_MoveLabels (S, E, N); } } @@ -515,7 +612,7 @@ void DelCodeEntry (CodeSeg* S, unsigned Index) */ if (E->JumpTo) { /* Remove the reference */ - RemoveCodeLabelRef (S, E); + CS_RemoveLabelRef (S, E); } /* Delete the pointer to the insn */ @@ -527,7 +624,7 @@ void DelCodeEntry (CodeSeg* S, unsigned Index) -void DelCodeEntries (CodeSeg* S, unsigned Start, unsigned Count) +void CS_DelEntries (CodeSeg* S, unsigned Start, unsigned Count) /* Delete a range of code entries. This includes removing references to labels, * labels attached to the entries and so on. */ @@ -536,51 +633,70 @@ void DelCodeEntries (CodeSeg* S, unsigned Start, unsigned Count) * memory moving. */ while (Count--) { - DelCodeEntry (S, Start + Count); + CS_DelEntry (S, Start + Count); } } -void MoveCodeEntry (CodeSeg* S, unsigned OldPos, unsigned NewPos) -/* Move an entry from one position to another. OldPos is the current position - * of the entry, NewPos is the new position of the entry. +void CS_MoveEntries (CodeSeg* S, unsigned Start, unsigned Count, unsigned NewPos) +/* Move a range of entries from one position to another. Start is the index + * of the first entry to move, Count is the number of entries and NewPos is + * the index of the target entry. The entry with the index Start will later + * have the index NewPos. All entries with indices NewPos and above are + * moved to higher indices. If the code block is moved to the end of the + * current code, and if pending labels exist, these labels will get attached + * to the first instruction of the moved block (the first one after the + * current code end) */ { - /* Get the code entry and remove it from the collection */ - CodeEntry* E = GetCodeEntry (S, OldPos); - CollDelete (&S->Entries, OldPos); - - /* Correct NewPos if needed */ - if (NewPos >= OldPos) { - /* Position has changed with removal */ - --NewPos; + /* If NewPos is at the end of the code segment, move any labels from the + * label pool to the first instruction of the moved range. + */ + if (NewPos == CS_GetEntryCount (S)) { + CS_MoveLabelsToEntry (S, CS_GetEntry (S, Start)); } - /* Now insert it at the new position */ - CollInsert (&S->Entries, E, NewPos); + /* Move the code block to the destination */ + CollMoveMultiple (&S->Entries, Start, Count, NewPos); +} + + + +struct CodeEntry* CS_GetPrevEntry (CodeSeg* S, unsigned Index) +/* Get the code entry preceeding the one with the index Index. If there is no + * preceeding code entry, return NULL. + */ +{ + if (Index == 0) { + /* This is the first entry */ + return 0; + } else { + /* Previous entry available */ + return CollAtUnchecked (&S->Entries, Index-1); + } } -struct CodeEntry* GetNextCodeEntry (CodeSeg* S, unsigned Index) +struct CodeEntry* CS_GetNextEntry (CodeSeg* S, unsigned Index) /* Get the code entry following the one with the index Index. If there is no * following code entry, return NULL. */ { if (Index >= CollCount (&S->Entries)-1) { - /* This is the last entry */ - return 0; + /* This is the last entry */ + return 0; } else { - /* Code entries left */ + /* Code entries left */ return CollAtUnchecked (&S->Entries, Index+1); } } -int GetCodeEntries (CodeSeg* S, struct CodeEntry** List, - unsigned Start, unsigned Count) +int CS_GetEntries (CodeSeg* S, struct CodeEntry** List, + unsigned Start, unsigned Count) /* Get Count code entries into List starting at index start. Return true if * we got the lines, return false if not enough lines were available. */ @@ -601,7 +717,7 @@ int GetCodeEntries (CodeSeg* S, struct CodeEntry** List, -unsigned GetCodeEntryIndex (CodeSeg* S, struct CodeEntry* E) +unsigned CS_GetEntryIndex (CodeSeg* S, struct CodeEntry* E) /* Return the index of a code entry */ { int Index = CollIndex (&S->Entries, E); @@ -611,46 +727,83 @@ unsigned GetCodeEntryIndex (CodeSeg* S, struct CodeEntry* E) -void AddCodeLabel (CodeSeg* S, const char* Name) +int CS_RangeHasLabel (CodeSeg* S, unsigned Start, unsigned Count) +/* Return true if any of the code entries in the given range has a label + * attached. If the code segment does not span the given range, check the + * possible span instead. + */ +{ + unsigned EntryCount = CS_GetEntryCount(S); + + /* Adjust count. We expect at least Start to be valid. */ + CHECK (Start < EntryCount); + if (Start + Count > EntryCount) { + Count = EntryCount - Start; + } + + /* Check each entry. Since we have validated the index above, we may + * use the unchecked access function in the loop which is faster. + */ + while (Count--) { + const CodeEntry* E = CollAtUnchecked (&S->Entries, Start++); + if (CE_HasLabel (E)) { + return 1; + } + } + + /* No label in the complete range */ + return 0; +} + + + +CodeLabel* CS_AddLabel (CodeSeg* S, const char* Name) /* Add a code label for the next instruction to follow */ { /* Calculate the hash from the name */ unsigned Hash = HashStr (Name) % CS_LABEL_HASH_SIZE; /* Try to find the code label if it does already exist */ - CodeLabel* L = FindCodeLabel (S, Name, Hash); + CodeLabel* L = CS_FindLabel (S, Name, Hash); /* Did we find it? */ if (L) { /* We found it - be sure it does not already have an owner */ - CHECK (L->Owner == 0); + if (L->Owner) { + Error ("ASM label `%s' is already defined", Name); + return L; + } } else { /* Not found - create a new one */ - L = NewCodeSegLabel (S, Name, Hash); + L = CS_NewCodeLabel (S, Name, Hash); } /* Safety. This call is quite costly, but safety is better */ if (CollIndex (&S->Labels, L) >= 0) { - Internal ("AddCodeLabel: Label `%s' already defined", Name); + Error ("ASM label `%s' is already defined", Name); + return L; } /* We do now have a valid label. Remember it for later */ CollAppend (&S->Labels, L); + + /* Return the label */ + return L; } -CodeLabel* GenCodeLabel (CodeSeg* S, struct CodeEntry* E) +CodeLabel* CS_GenLabel (CodeSeg* S, struct CodeEntry* E) /* If the code entry E does already have a label, return it. Otherwise * create a new label, attach it to E and return it. */ { CodeLabel* L; - if (CodeEntryHasLabel (E)) { + if (CE_HasLabel (E)) { /* Get the label from this entry */ - L = GetCodeLabel (E, 0); + L = CE_GetLabel (E, 0); } else { @@ -661,10 +814,10 @@ CodeLabel* GenCodeLabel (CodeSeg* S, struct CodeEntry* E) unsigned Hash = HashStr (Name) % CS_LABEL_HASH_SIZE; /* Create a new label */ - L = NewCodeSegLabel (S, Name, Hash); + L = CS_NewCodeLabel (S, Name, Hash); /* Attach this label to the code entry */ - AttachCodeLabel (E, L); + CE_AttachLabel (E, L); } @@ -674,13 +827,13 @@ CodeLabel* GenCodeLabel (CodeSeg* S, struct CodeEntry* E) -void DelCodeLabel (CodeSeg* S, CodeLabel* L) +void CS_DelLabel (CodeSeg* S, CodeLabel* L) /* Remove references from this label and delete it. */ { unsigned Count, I; /* First, remove the label from the hash chain */ - RemoveLabelFromHash (S, L); + CS_RemoveLabelFromHash (S, L); /* Remove references from insns jumping to this label */ Count = CollCount (&L->JumpFrom); @@ -688,7 +841,7 @@ void DelCodeLabel (CodeSeg* S, CodeLabel* L) /* Get the insn referencing this label */ CodeEntry* E = CollAt (&L->JumpFrom, I); /* Remove the reference */ - E->JumpTo = 0; + CE_ClearJumpTo (E); } CollDeleteAll (&L->JumpFrom); @@ -698,7 +851,7 @@ void DelCodeLabel (CodeSeg* S, CodeLabel* L) * errors to slip through. */ if (L->Owner) { - CollDeleteItem (&L->Owner->Labels, L); + CollDeleteItem (&L->Owner->Labels, L); } /* All references removed, delete the label itself */ @@ -707,63 +860,103 @@ void DelCodeLabel (CodeSeg* S, CodeLabel* L) -void MergeCodeLabels (CodeSeg* S) +void CS_MergeLabels (CodeSeg* S) /* Merge code labels. That means: For each instruction, remove all labels but * one and adjust references accordingly. */ { unsigned I; + unsigned J; + + /* First, remove all labels from the label symbol table that don't have an + * owner (this means that they are actually external labels but we didn't + * know that previously since they may have also been forward references). + */ + for (I = 0; I < CS_LABEL_HASH_SIZE; ++I) { + + /* Get the first label in this hash chain */ + CodeLabel** L = &S->LabelHash[I]; + while (*L) { + if ((*L)->Owner == 0) { + + /* The label does not have an owner, remove it from the chain */ + CodeLabel* X = *L; + *L = X->Next; + + /* Cleanup any entries jumping to this label */ + for (J = 0; J < CL_GetRefCount (X); ++J) { + /* Get the entry referencing this label */ + CodeEntry* E = CL_GetRef (X, J); + /* And remove the reference. Do NOT call CE_ClearJumpTo + * here, because this will also clear the label name, + * which is not what we want. + */ + E->JumpTo = 0; + } + + /* Print some debugging output */ + if (Debug) { + printf ("Removing unused global label `%s'", X->Name); + } + + /* And free the label */ + FreeCodeLabel (X); + } else { + /* Label is owned, point to next code label pointer */ + L = &((*L)->Next); + } + } + } /* Walk over all code entries */ - unsigned EntryCount = GetCodeEntryCount (S); - for (I = 0; I < EntryCount; ++I) { + for (I = 0; I < CS_GetEntryCount (S); ++I) { CodeLabel* RefLab; - unsigned J; + unsigned J; - /* Get a pointer to the next entry */ - CodeEntry* E = GetCodeEntry (S, I); + /* Get a pointer to the next entry */ + CodeEntry* E = CS_GetEntry (S, I); /* If this entry has zero labels, continue with the next one */ - unsigned LabelCount = GetCodeLabelCount (E); - if (LabelCount == 0) { - continue; - } + unsigned LabelCount = CE_GetLabelCount (E); + if (LabelCount == 0) { + continue; + } - /* We have at least one label. Use the first one as reference label. */ - RefLab = GetCodeLabel (E, 0); + /* We have at least one label. Use the first one as reference label. */ + RefLab = CE_GetLabel (E, 0); - /* Walk through the remaining labels and change references to these - * labels to a reference to the one and only label. Delete the labels - * that are no longer used. To increase performance, walk backwards - * through the list. - */ + /* Walk through the remaining labels and change references to these + * labels to a reference to the one and only label. Delete the labels + * that are no longer used. To increase performance, walk backwards + * through the list. + */ for (J = LabelCount-1; J >= 1; --J) { - /* Get the next label */ - CodeLabel* L = GetCodeLabel (E, J); + /* Get the next label */ + CodeLabel* L = CE_GetLabel (E, J); - /* Move all references from this label to the reference label */ - MoveLabelRefs (L, RefLab); + /* Move all references from this label to the reference label */ + CL_MoveRefs (L, RefLab); /* Remove the label completely. */ - DelCodeLabel (S, L); + CS_DelLabel (S, L); } - /* The reference label is the only remaining label. Check if there - * are any references to this label, and delete it if this is not - * the case. - */ + /* The reference label is the only remaining label. Check if there + * are any references to this label, and delete it if this is not + * the case. + */ if (CollCount (&RefLab->JumpFrom) == 0) { /* Delete the label */ - DelCodeLabel (S, RefLab); + CS_DelLabel (S, RefLab); } } } -void MoveCodeLabels (CodeSeg* S, struct CodeEntry* Old, struct CodeEntry* New) +void CS_MoveLabels (CodeSeg* S, struct CodeEntry* Old, struct CodeEntry* New) /* Move all labels from Old to New. The routine will move the labels itself * if New does not have any labels, and move references if there is at least * a label for new. If references are moved, the old label is deleted @@ -771,23 +964,23 @@ void MoveCodeLabels (CodeSeg* S, struct CodeEntry* Old, struct CodeEntry* New) */ { /* Get the number of labels to move */ - unsigned OldLabelCount = GetCodeLabelCount (Old); + unsigned OldLabelCount = CE_GetLabelCount (Old); /* Does the new entry have itself a label? */ - if (CodeEntryHasLabel (New)) { + if (CE_HasLabel (New)) { /* The new entry does already have a label - move references */ - CodeLabel* NewLabel = GetCodeLabel (New, 0); + CodeLabel* NewLabel = CE_GetLabel (New, 0); while (OldLabelCount--) { /* Get the next label */ - CodeLabel* OldLabel = GetCodeLabel (Old, OldLabelCount); + CodeLabel* OldLabel = CE_GetLabel (Old, OldLabelCount); /* Move references */ - MoveLabelRefs (OldLabel, NewLabel); + CL_MoveRefs (OldLabel, NewLabel); /* Delete the label */ - DelCodeLabel (S, OldLabel); + CS_DelLabel (S, OldLabel); } @@ -797,7 +990,7 @@ void MoveCodeLabels (CodeSeg* S, struct CodeEntry* Old, struct CodeEntry* New) while (OldLabelCount--) { /* Move the label to the new entry */ - MoveCodeLabel (GetCodeLabel (Old, OldLabelCount), New); + CE_MoveLabel (CE_GetLabel (Old, OldLabelCount), New); } @@ -806,7 +999,7 @@ void MoveCodeLabels (CodeSeg* S, struct CodeEntry* Old, struct CodeEntry* New) -void RemoveCodeLabelRef (CodeSeg* S, struct CodeEntry* E) +void CS_RemoveLabelRef (CodeSeg* S, struct CodeEntry* E) /* Remove the reference between E and the label it jumps to. The reference * will be removed on both sides and E->JumpTo will be 0 after that. If * the reference was the only one for the label, the label will get @@ -821,17 +1014,17 @@ void RemoveCodeLabelRef (CodeSeg* S, struct CodeEntry* E) CollDeleteItem (&L->JumpFrom, E); /* The entry jumps no longer to L */ - E->JumpTo = 0; + CE_ClearJumpTo (E); /* If there are no more references, delete the label */ if (CollCount (&L->JumpFrom) == 0) { - DelCodeLabel (S, L); + CS_DelLabel (S, L); } } -void MoveCodeLabelRef (CodeSeg* S, struct CodeEntry* E, CodeLabel* L) +void CS_MoveLabelRef (CodeSeg* S, struct CodeEntry* E, CodeLabel* L) /* Change the reference of E to L instead of the current one. If this * was the only reference to the old label, the old label will get * deleted. @@ -844,39 +1037,19 @@ void MoveCodeLabelRef (CodeSeg* S, struct CodeEntry* E, CodeLabel* L) PRECONDITION (OldLabel != 0); /* Remove the reference to our label */ - RemoveCodeLabelRef (S, E); + CS_RemoveLabelRef (S, E); /* Use the new label */ - AddLabelRef (L, E); + CL_AddRef (L, E); } -void AddCodeSegHint (CodeSeg* S, unsigned Hint) -/* Add a hint for the preceeding instruction */ -{ - CodeEntry* E; - - /* Get the number of entries in this segment */ - unsigned EntryCount = GetCodeEntryCount (S); - - /* Must have at least one entry */ - CHECK (EntryCount > 0); - - /* Get the last entry */ - E = GetCodeEntry (S, EntryCount-1); - - /* Add the hint */ - E->Hints |= Hint; -} - - - -void DelCodeSegAfter (CodeSeg* S, unsigned Last) +void CS_DelCodeAfter (CodeSeg* S, unsigned Last) /* Delete all entries including the given one */ { /* Get the number of entries in this segment */ - unsigned Count = GetCodeEntryCount (S); + unsigned Count = CS_GetEntryCount (S); /* First pass: Delete all references to labels. If the reference count * for a label drops to zero, delete it. @@ -884,23 +1057,23 @@ void DelCodeSegAfter (CodeSeg* S, unsigned Last) unsigned C = Count; while (Last < C--) { - /* Get the next entry */ - CodeEntry* E = GetCodeEntry (S, C); - - /* Check if this entry has a label reference */ - if (E->JumpTo) { - /* If the label is a label in the label pool and this is the last - * reference to the label, remove the label from the pool. - */ - CodeLabel* L = E->JumpTo; - int Index = CollIndex (&S->Labels, L); - if (Index >= 0 && CollCount (&L->JumpFrom) == 1) { - /* Delete it from the pool */ - CollDelete (&S->Labels, Index); - } + /* Get the next entry */ + CodeEntry* E = CS_GetEntry (S, C); + + /* Check if this entry has a label reference */ + if (E->JumpTo) { + /* If the label is a label in the label pool and this is the last + * reference to the label, remove the label from the pool. + */ + CodeLabel* L = E->JumpTo; + int Index = CollIndex (&S->Labels, L); + if (Index >= 0 && CollCount (&L->JumpFrom) == 1) { + /* Delete it from the pool */ + CollDelete (&S->Labels, Index); + } /* Remove the reference to the label */ - RemoveCodeLabelRef (S, E); + CS_RemoveLabelRef (S, E); } } @@ -914,12 +1087,12 @@ void DelCodeSegAfter (CodeSeg* S, unsigned Last) while (Last < C--) { /* Get the next entry */ - CodeEntry* E = GetCodeEntry (S, C); + CodeEntry* E = CS_GetEntry (S, C); /* Check if this entry has a label attached */ - if (CodeEntryHasLabel (E)) { + if (CE_HasLabel (E)) { /* Move the labels to the pool and clear the owner pointer */ - MoveLabelsToPool (S, E); + CS_MoveLabelsToPool (S, E); } /* Delete the pointer to the entry */ @@ -932,52 +1105,488 @@ void DelCodeSegAfter (CodeSeg* S, unsigned Last) -void OutputCodeSeg (const CodeSeg* S, FILE* F) +void CS_ResetMarks (CodeSeg* S, unsigned First, unsigned Last) +/* Remove all user marks from the entries in the given range */ +{ + while (First <= Last) { + CE_ResetMark (CS_GetEntry (S, First++)); + } +} + + + +int CS_IsBasicBlock (CodeSeg* S, unsigned First, unsigned Last) +/* Check if the given code segment range is a basic block. That is, check if + * First is the only entrance and Last is the only exit. This means that no + * jump/branch inside the block may jump to an insn below First or after(!) + * Last, and that no insn may jump into this block from the outside. + */ +{ + unsigned I; + + /* Don't accept invalid ranges */ + CHECK (First <= Last); + + /* First pass: Walk over the range and remove all marks from the entries */ + CS_ResetMarks (S, First, Last); + + /* Second pass: Walk over the range checking all labels. Note: There may be + * label on the first insn which is ok. + */ + I = First + 1; + while (I <= Last) { + + /* Get the next entry */ + CodeEntry* E = CS_GetEntry (S, I); + + /* Check if this entry has one or more labels, if so, check which + * entries jump to this label. + */ + unsigned LabelCount = CE_GetLabelCount (E); + unsigned LabelIndex; + for (LabelIndex = 0; LabelIndex < LabelCount; ++LabelIndex) { + + /* Get this label */ + CodeLabel* L = CE_GetLabel (E, LabelIndex); + + /* Walk over all entries that jump to this label. Check for each + * of the entries if it is out of the range. + */ + unsigned RefCount = CL_GetRefCount (L); + unsigned RefIndex; + for (RefIndex = 0; RefIndex < RefCount; ++RefIndex) { + + /* Get the code entry that jumps here */ + CodeEntry* Ref = CL_GetRef (L, RefIndex); + + /* Walk over out complete range and check if we find the + * refering entry. This is cheaper than using CS_GetEntryIndex, + * because CS_GetEntryIndex will search the complete code + * segment and not just our range. + */ + unsigned J; + for (J = First; J <= Last; ++J) { + if (Ref == CS_GetEntry (S, J)) { + break; + } + } + if (J > Last) { + /* We did not find the entry. This means that the jump to + * out code segment entry E came from outside the range, + * which in turn means that the given range is not a basic + * block. + */ + CS_ResetMarks (S, First, Last); + return 0; + } + + /* If we come here, we found the entry. Mark it, so we know + * that the branch to the label is in range. + */ + CE_SetMark (Ref); + } + } + + /* Next entry */ + ++I; + } + + /* Third pass: Walk again over the range and check all branches. If we + * find a branch that is not marked, its target is not inside the range + * (since we checked all the labels in the range before). + */ + I = First; + while (I <= Last) { + + /* Get the next entry */ + CodeEntry* E = CS_GetEntry (S, I); + + /* Check if this is a branch and if so, if it has a mark */ + if (E->Info & (OF_UBRA | OF_CBRA)) { + if (!CE_HasMark (E)) { + /* No mark means not a basic block. Before bailing out, be sure + * to remove the marks from the remaining entries. + */ + CS_ResetMarks (S, I+1, Last); + return 0; + } + + /* Remove the mark */ + CE_ResetMark (E); + } + + /* Next entry */ + ++I; + } + + /* Done - this is a basic block */ + return 1; +} + + + +void CS_OutputPrologue (const CodeSeg* S, FILE* F) +/* If the given code segment is a code segment for a function, output the + * assembler prologue into the file. That is: Output a comment header, switch + * to the correct segment and enter the local function scope. If the code + * segment is global, do nothing. + */ +{ + /* Get the function associated with the code segment */ + SymEntry* Func = S->Func; + + /* If the code segment is associated with a function, print a function + * header and enter a local scope. Be sure to switch to the correct + * segment before outputing the function label. + */ + if (Func) { + /* Get the function descriptor */ + const FuncDesc* D = GetFuncDesc (Func->Type); + CS_PrintFunctionHeader (S, F); + fprintf (F, ".segment\t\"%s\"\n\n.proc\t_%s", S->SegName, Func->Name); + if (D->Flags & FD_NEAR) { + fputs (": near", F); + } else if (D->Flags & FD_FAR) { + fputs (": far", F); + } + fputs ("\n\n", F); + } + +} + + + +void CS_OutputEpilogue (const CodeSeg* S, FILE* F) +/* If the given code segment is a code segment for a function, output the + * assembler epilogue into the file. That is: Close the local function scope. + */ +{ + if (S->Func) { + fputs ("\n.endproc\n\n", F); + } +} + + + +void CS_Output (CodeSeg* S, FILE* F) /* Output the code segment data to a file */ { unsigned I; + const LineInfo* LI; /* Get the number of entries in this segment */ - unsigned Count = GetCodeEntryCount (S); + unsigned Count = CS_GetEntryCount (S); /* If the code segment is empty, bail out here */ if (Count == 0) { return; } + /* Generate register info */ + CS_GenRegInfo (S); + /* Output the segment directive */ fprintf (F, ".segment\t\"%s\"\n\n", S->SegName); - /* If this is a segment for a function, enter a function */ - if (S->Func) { - fprintf (F, ".proc\t_%s\n\n", S->Func->Name); + /* Output all entries, prepended by the line information if it has changed */ + LI = 0; + for (I = 0; I < Count; ++I) { + /* Get the next entry */ + const CodeEntry* E = CollConstAt (&S->Entries, I); + /* Check if the line info has changed. If so, output the source line + * if the option is enabled and output debug line info if the debug + * option is enabled. + */ + if (E->LI != LI) { + /* Line info has changed, remember the new line info */ + LI = E->LI; + + /* Add the source line as a comment. Beware: When line continuation + * was used, the line may contain newlines. + */ + if (AddSource) { + const char* L = LI->Line; + fputs (";\n; ", F); + while (*L) { + if (*L == '\n') { + fputs ("\n; ", F); + } else { + fputc (*L, F); + } + ++L; + } + fputs ("\n;\n", F); + } + + /* Add line debug info */ + if (DebugInfo) { + fprintf (F, "\t.dbg\tline, \"%s\", %u\n", + GetInputName (LI), GetInputLine (LI)); + } + } + /* Output the code */ + CE_Output (E, F); } - /* Output all entries */ - for (I = 0; I < Count; ++I) { + /* If debug info is enabled, terminate the last line number information */ + if (DebugInfo) { + fputs ("\t.dbg\tline\n", F); + } - unsigned char Use; + /* Free register info */ + CS_FreeRegInfo (S); +} - OutputCodeEntry (CollConstAt (&S->Entries, I), F); - /* Print usage info */ - Use = GetRegInfo ((CodeSeg*) S, I+1); - fprintf (F, - " Use: %c%c%c\n", - (Use & REG_A)? 'A' : '_', - (Use & REG_X)? 'X' : '_', - (Use & REG_Y)? 'Y' : '_'); - } - /* If this is a segment for a function, leave the function */ - if (S->Func) { - fprintf (F, "\n.endproc\n\n"); +void CS_FreeRegInfo (CodeSeg* S) +/* Free register infos for all instructions */ +{ + unsigned I; + for (I = 0; I < CS_GetEntryCount (S); ++I) { + CE_FreeRegInfo (CS_GetEntry(S, I)); } } +void CS_GenRegInfo (CodeSeg* S) +/* Generate register infos for all instructions */ +{ + unsigned I; + RegContents Regs; /* Initial register contents */ + RegContents* CurrentRegs; /* Current register contents */ + int WasJump; /* True if last insn was a jump */ + int Done; /* All runs done flag */ + + /* Be sure to delete all register infos */ + CS_FreeRegInfo (S); + + /* We may need two runs to get back references right */ + do { + + /* Assume we're done after this run */ + Done = 1; + + /* On entry, the register contents are unknown */ + RC_Invalidate (&Regs); + CurrentRegs = &Regs; + /* Walk over all insns and note just the changes from one insn to the + * next one. + */ + WasJump = 0; + for (I = 0; I < CS_GetEntryCount (S); ++I) { + + CodeEntry* P; + + /* Get the next instruction */ + CodeEntry* E = CollAtUnchecked (&S->Entries, I); + + /* If the instruction has a label, we need some special handling */ + unsigned LabelCount = CE_GetLabelCount (E); + if (LabelCount > 0) { + + /* Loop over all entry points that jump here. If these entry + * points already have register info, check if all values are + * known and identical. If all values are identical, and the + * preceeding instruction was not an unconditional branch, check + * if the register value on exit of the preceeding instruction + * is also identical. If all these values are identical, the + * value of a register is known, otherwise it is unknown. + */ + CodeLabel* Label = CE_GetLabel (E, 0); + unsigned Entry; + if (WasJump) { + /* Preceeding insn was an unconditional branch */ + CodeEntry* J = CL_GetRef(Label, 0); + if (J->RI) { + Regs = J->RI->Out2; + } else { + RC_Invalidate (&Regs); + } + Entry = 1; + } else { + Regs = *CurrentRegs; + Entry = 0; + } + + while (Entry < CL_GetRefCount (Label)) { + /* Get this entry */ + CodeEntry* J = CL_GetRef (Label, Entry); + if (J->RI == 0) { + /* No register info for this entry. This means that the + * instruction that jumps here is at higher addresses and + * the jump is a backward jump. We need a second run to + * get the register info right in this case. Until then, + * assume unknown register contents. + */ + Done = 0; + RC_Invalidate (&Regs); + break; + } + if (J->RI->Out2.RegA != Regs.RegA) { + Regs.RegA = UNKNOWN_REGVAL; + } + if (J->RI->Out2.RegX != Regs.RegX) { + Regs.RegX = UNKNOWN_REGVAL; + } + if (J->RI->Out2.RegY != Regs.RegY) { + Regs.RegY = UNKNOWN_REGVAL; + } + if (J->RI->Out2.SRegLo != Regs.SRegLo) { + Regs.SRegLo = UNKNOWN_REGVAL; + } + if (J->RI->Out2.SRegHi != Regs.SRegHi) { + Regs.SRegHi = UNKNOWN_REGVAL; + } + if (J->RI->Out2.Tmp1 != Regs.Tmp1) { + Regs.Tmp1 = UNKNOWN_REGVAL; + } + ++Entry; + } + + /* Use this register info */ + CurrentRegs = &Regs; + + } + + /* Generate register info for this instruction */ + CE_GenRegInfo (E, CurrentRegs); + + /* Remember for the next insn if this insn was an uncondition branch */ + WasJump = (E->Info & OF_UBRA) != 0; + + /* Output registers for this insn are input for the next */ + CurrentRegs = &E->RI->Out; + + /* If this insn is a branch on zero flag, we may have more info on + * register contents for one of both flow directions, but only if + * there is a previous instruction. + */ + if ((E->Info & OF_ZBRA) != 0 && (P = CS_GetPrevEntry (S, I)) != 0) { + + /* Get the branch condition */ + bc_t BC = GetBranchCond (E->OPC); + + /* Check the previous instruction */ + switch (P->OPC) { + + case OP65_ADC: + case OP65_AND: + case OP65_DEA: + case OP65_EOR: + case OP65_INA: + case OP65_LDA: + case OP65_ORA: + case OP65_PLA: + case OP65_SBC: + /* A is zero in one execution flow direction */ + if (BC == BC_EQ) { + E->RI->Out2.RegA = 0; + } else { + E->RI->Out.RegA = 0; + } + break; + + case OP65_CMP: + /* If this is an immidiate compare, the A register has + * the value of the compare later. + */ + if (CE_IsConstImm (P)) { + if (BC == BC_EQ) { + E->RI->Out2.RegA = (unsigned char)P->Num; + } else { + E->RI->Out.RegA = (unsigned char)P->Num; + } + } + break; + + case OP65_CPX: + /* If this is an immidiate compare, the X register has + * the value of the compare later. + */ + if (CE_IsConstImm (P)) { + if (BC == BC_EQ) { + E->RI->Out2.RegX = (unsigned char)P->Num; + } else { + E->RI->Out.RegX = (unsigned char)P->Num; + } + } + break; + + case OP65_CPY: + /* If this is an immidiate compare, the Y register has + * the value of the compare later. + */ + if (CE_IsConstImm (P)) { + if (BC == BC_EQ) { + E->RI->Out2.RegY = (unsigned char)P->Num; + } else { + E->RI->Out.RegY = (unsigned char)P->Num; + } + } + break; + + case OP65_DEX: + case OP65_INX: + case OP65_LDX: + case OP65_PLX: + /* X is zero in one execution flow direction */ + if (BC == BC_EQ) { + E->RI->Out2.RegX = 0; + } else { + E->RI->Out.RegX = 0; + } + break; + + case OP65_DEY: + case OP65_INY: + case OP65_LDY: + case OP65_PLY: + /* X is zero in one execution flow direction */ + if (BC == BC_EQ) { + E->RI->Out2.RegY = 0; + } else { + E->RI->Out.RegY = 0; + } + break; + + case OP65_TAX: + case OP65_TXA: + /* If the branch is a beq, both A and X are zero at the + * branch target, otherwise they are zero at the next + * insn. + */ + if (BC == BC_EQ) { + E->RI->Out2.RegA = E->RI->Out2.RegX = 0; + } else { + E->RI->Out.RegA = E->RI->Out.RegX = 0; + } + break; + + case OP65_TAY: + case OP65_TYA: + /* If the branch is a beq, both A and Y are zero at the + * branch target, otherwise they are zero at the next + * insn. + */ + if (BC == BC_EQ) { + E->RI->Out2.RegA = E->RI->Out2.RegY = 0; + } else { + E->RI->Out.RegA = E->RI->Out.RegY = 0; + } + break; + + default: + break; + + } + } + } + } while (!Done); + +}