X-Git-Url: https://git.sur5r.net/?a=blobdiff_plain;f=src%2Fcc65%2Fcoptstop.c;h=85824b64d148f6bced31f0e8338c263584f16270;hb=c6ee5a841878ccdbf9ab85d3dafdd088648b93ba;hp=f59d22d1abd3e3c2a27dfd0fa4ceb8679f3b70af;hpb=08d6d96264434cce1e437b92093820e46914edf7;p=cc65 diff --git a/src/cc65/coptstop.c b/src/cc65/coptstop.c index f59d22d1a..85824b64d 100644 --- a/src/cc65/coptstop.c +++ b/src/cc65/coptstop.c @@ -42,6 +42,7 @@ #include "codeent.h" #include "codeinfo.h" #include "coptstop.h" +#include "error.h" @@ -52,18 +53,23 @@ /* LoadRegInfo flags set by DirectOp */ -#define LI_NONE 0x00 -#define LI_DIRECT 0x01 /* Direct op may be used */ -#define LI_RELOAD_Y 0x02 /* Reload index register Y */ -#define LI_REMOVE 0x04 /* Load may be removed */ +typedef enum { + LI_NONE = 0x00, + LI_DIRECT = 0x01, /* Direct op may be used */ + LI_RELOAD_Y = 0x02, /* Reload index register Y */ + LI_REMOVE = 0x04, /* Load may be removed */ + LI_DUP_LOAD = 0x08, /* Duplicate load */ +} LI_FLAGS; /* Structure that tells us how to load the lhs values */ typedef struct LoadRegInfo LoadRegInfo; struct LoadRegInfo { + LI_FLAGS Flags; /* Tells us how to load */ int LoadIndex; /* Index of load insn, -1 if invalid */ CodeEntry* LoadEntry; /* The actual entry, 0 if invalid */ - unsigned char Flags; /* Tells us how to load */ - unsigned char Offs; /* Stack offset if data is on stack */ + int XferIndex; /* Index of transfer insn */ + CodeEntry* XferEntry; /* The actual transfer entry */ + int Offs; /* Stack offset if data is on stack */ }; /* Now combined for both registers */ @@ -88,7 +94,9 @@ typedef enum { OP_A_KNOWN = 0x01, /* Value of A must be known */ OP_X_ZERO = 0x02, /* X must be zero */ OP_LHS_LOAD = 0x04, /* Must have load insns for LHS */ - OP_RHS_LOAD = 0x08, /* Must have load insns for RHS */ + OP_LHS_LOAD_DIRECT = 0x0C, /* Must have direct load insn for LHS */ + OP_RHS_LOAD = 0x10, /* Must have load insns for RHS */ + OP_RHS_LOAD_DIRECT = 0x30, /* Must have direct load insn for RHS */ } OP_FLAGS; /* Structure forward decl */ @@ -145,7 +153,10 @@ struct StackOpData { static void ClearLoadRegInfo (LoadRegInfo* RI) /* Clear a LoadRegInfo struct */ { + RI->Flags = LI_NONE; RI->LoadIndex = -1; + RI->XferIndex = -1; + RI->Offs = 0; } @@ -153,13 +164,17 @@ static void ClearLoadRegInfo (LoadRegInfo* RI) static void FinalizeLoadRegInfo (LoadRegInfo* RI, CodeSeg* S) /* Prepare a LoadRegInfo struct for use */ { - /* Get the entry */ + /* Get the entries */ if (RI->LoadIndex >= 0) { RI->LoadEntry = CS_GetEntry (S, RI->LoadIndex); } else { RI->LoadEntry = 0; } - RI->Flags = LI_NONE; + if (RI->XferIndex >= 0) { + RI->XferEntry = CS_GetEntry (S, RI->XferIndex); + } else { + RI->XferEntry = 0; + } } @@ -174,7 +189,7 @@ static void ClearLoadInfo (LoadInfo* LI) -static void AdjustLoadRegInfo (LoadRegInfo* RI, int DelIndex, int Change) +static void AdjustLoadRegInfo (LoadRegInfo* RI, int Index, int Change) /* Adjust a load register info struct after deleting or inserting an entry * with a given index */ @@ -182,18 +197,28 @@ static void AdjustLoadRegInfo (LoadRegInfo* RI, int DelIndex, int Change) CHECK (abs (Change) == 1); if (Change < 0) { /* Deletion */ - if (DelIndex < RI->LoadIndex) { + if (Index < RI->LoadIndex) { --RI->LoadIndex; - } else if (DelIndex == RI->LoadIndex) { + } else if (Index == RI->LoadIndex) { /* Has been removed */ RI->LoadIndex = -1; RI->LoadEntry = 0; } + if (Index < RI->XferIndex) { + --RI->XferIndex; + } else if (Index == RI->XferIndex) { + /* Has been removed */ + RI->XferIndex = -1; + RI->XferEntry = 0; + } } else { /* Insertion */ - if (DelIndex <= RI->LoadIndex) { + if (Index <= RI->LoadIndex) { ++RI->LoadIndex; } + if (Index <= RI->XferIndex) { + ++RI->XferIndex; + } } } @@ -210,12 +235,12 @@ static void FinalizeLoadInfo (LoadInfo* LI, CodeSeg* S) -static void AdjustLoadInfo (LoadInfo* LI, int DelIndex, int Change) +static void AdjustLoadInfo (LoadInfo* LI, int Index, int Change) /* Adjust a load info struct after deleting entry with a given index */ { - AdjustLoadRegInfo (&LI->A, DelIndex, Change); - AdjustLoadRegInfo (&LI->X, DelIndex, Change); - AdjustLoadRegInfo (&LI->Y, DelIndex, Change); + AdjustLoadRegInfo (&LI->A, Index, Change); + AdjustLoadRegInfo (&LI->X, Index, Change); + AdjustLoadRegInfo (&LI->Y, Index, Change); } @@ -224,37 +249,114 @@ static void TrackLoads (LoadInfo* LI, CodeEntry* E, int I) /* Track loads for a code entry */ { if (E->Info & OF_LOAD) { + + LoadRegInfo* RI = 0; + + /* Determine, which register was loaded */ if (E->Chg & REG_A) { - LI->A.LoadIndex = I; + RI = &LI->A; + } else if (E->Chg & REG_X) { + RI = &LI->X; + } else if (E->Chg & REG_Y) { + RI = &LI->Y; } - if (E->Chg & REG_X) { - LI->X.LoadIndex = I; + CHECK (RI != 0); + + /* If we had a load or xfer op before, this is a duplicate load which + * can cause problems if it encountered between the pushax and the op, + * so remember it. + */ + if (RI->LoadIndex >= 0 || RI->XferIndex >= 0) { + RI->Flags |= LI_DUP_LOAD; } - if (E->Chg & REG_Y) { - LI->Y.LoadIndex = I; + + /* Remember the load */ + RI->LoadIndex = I; + RI->XferIndex = -1; + + /* Set load flags */ + RI->Flags &= ~(LI_DIRECT | LI_RELOAD_Y); + if (E->AM == AM65_IMM || E->AM == AM65_ZP || E->AM == AM65_ABS) { + /* These insns are all ok and replaceable */ + RI->Flags |= LI_DIRECT; + } else if (E->AM == AM65_ZP_INDY && + RegValIsKnown (E->RI->In.RegY) && + strcmp (E->Arg, "sp") == 0) { + /* A load from the stack with known offset is also ok, but in this + * case we must reload the index register later. Please note that + * a load indirect via other zero page locations is not ok, since + * these locations may change between the push and the actual + * operation. + */ + RI->Offs = (unsigned char) E->RI->In.RegY; + RI->Flags |= (LI_DIRECT | LI_RELOAD_Y); } + + } else if (E->Info & OF_XFR) { + + /* Determine source and target of the transfer and handle the TSX insn */ + LoadRegInfo* Src; + LoadRegInfo* Tgt; switch (E->OPC) { - case OP65_TAX: LI->X.LoadIndex = LI->A.LoadIndex; break; - case OP65_TAY: LI->Y.LoadIndex = LI->A.LoadIndex; break; - case OP65_TXA: LI->A.LoadIndex = LI->X.LoadIndex; break; - case OP65_TYA: LI->A.LoadIndex = LI->Y.LoadIndex; break; - default: break; + case OP65_TAX: Src = &LI->A; Tgt = &LI->X; break; + case OP65_TAY: Src = &LI->A; Tgt = &LI->Y; break; + case OP65_TXA: Src = &LI->X; Tgt = &LI->A; break; + case OP65_TYA: Src = &LI->Y; Tgt = &LI->A; break; + case OP65_TSX: ClearLoadRegInfo (&LI->X); return; + case OP65_TXS: return; + default: Internal ("Unknown XFR insn in TrackLoads"); } - } else if (CE_IsCallTo (E, "ldaxysp")) { + + /* If we had a load or xfer op before, this is a duplicate load which + * can cause problems if it encountered between the pushax and the op, + * so remember it. + */ + if (Tgt->LoadIndex >= 0 || Tgt->XferIndex >= 0) { + Tgt->Flags |= LI_DUP_LOAD; + } + + /* Transfer the data */ + Tgt->LoadIndex = Src->LoadIndex; + Tgt->XferIndex = I; + Tgt->Offs = Src->Offs; + Tgt->Flags &= ~(LI_DIRECT | LI_RELOAD_Y); + Tgt->Flags |= Src->Flags & (LI_DIRECT | LI_RELOAD_Y); + + } else if (CE_IsCallTo (E, "ldaxysp") && RegValIsKnown (E->RI->In.RegY)) { + + /* If we had a load or xfer op before, this is a duplicate load which + * can cause problems if it encountered between the pushax and the op, + * so remember it for both registers involved. + */ + if (LI->A.LoadIndex >= 0 || LI->A.XferIndex >= 0) { + LI->A.Flags |= LI_DUP_LOAD; + } + if (LI->X.LoadIndex >= 0 || LI->X.XferIndex >= 0) { + LI->X.Flags |= LI_DUP_LOAD; + } + /* Both registers set, Y changed */ LI->A.LoadIndex = I; + LI->A.XferIndex = -1; + LI->A.Flags |= (LI_DIRECT | LI_RELOAD_Y); + LI->A.Offs = (unsigned char) E->RI->In.RegY - 1; + LI->X.LoadIndex = I; - LI->Y.LoadIndex = -1; + LI->X.XferIndex = -1; + LI->X.Flags |= (LI_DIRECT | LI_RELOAD_Y); + LI->X.Offs = (unsigned char) E->RI->In.RegY; + + ClearLoadRegInfo (&LI->Y); } else { if (E->Chg & REG_A) { - LI->A.LoadIndex = -1; + ClearLoadRegInfo (&LI->A); } if (E->Chg & REG_X) { - LI->X.LoadIndex = -1; + ClearLoadRegInfo (&LI->X); } if (E->Chg & REG_Y) { - LI->Y.LoadIndex = -1; + ClearLoadRegInfo (&LI->Y); } } } @@ -262,83 +364,11 @@ static void TrackLoads (LoadInfo* LI, CodeEntry* E, int I) /*****************************************************************************/ -/* Helpers */ +/* Helpers */ /*****************************************************************************/ -static void AdjustStackOffset (StackOpData* D, unsigned Offs) -/* Adjust the offset for all stack accesses in the range PushIndex to OpIndex. - * OpIndex is adjusted according to the insertions. - */ -{ - /* Walk over all entries */ - int I = D->PushIndex + 1; - while (I < D->OpIndex) { - - CodeEntry* E = CS_GetEntry (D->Code, I); - - int NeedCorrection = 0; - if ((E->Use & REG_SP) != 0) { - - /* Check for some things that should not happen */ - CHECK (E->AM == AM65_ZP_INDY || E->RI->In.RegY >= (short) Offs); - CHECK (strcmp (E->Arg, "sp") == 0); - - /* We need to correct this one */ - NeedCorrection = 1; - - } else if (CE_IsCallTo (E, "ldaxysp")) { - - /* We need to correct this one */ - NeedCorrection = 1; - - } - - if (NeedCorrection) { - - /* Get the code entry before this one. If it's a LDY, adjust the - * value. - */ - CodeEntry* P = CS_GetPrevEntry (D->Code, I); - if (P && P->OPC == OP65_LDY && CE_IsConstImm (P)) { - - /* The Y load is just before the stack access, adjust it */ - CE_SetNumArg (P, P->Num - Offs); - - } else { - - /* Insert a new load instruction before the stack access */ - const char* Arg = MakeHexArg (E->RI->In.RegY - Offs); - CodeEntry* X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI); - CS_InsertEntry (D->Code, X, I++); - - /* One more inserted entries */ - ++D->OpIndex; - - } - - /* If we need the value of Y later, be sure to reload it */ - if (RegYUsed (D->Code, I+1)) { - const char* Arg = MakeHexArg (E->RI->In.RegY); - CodeEntry* X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI); - CS_InsertEntry (D->Code, X, I+1); - - /* One more inserted entries */ - ++D->OpIndex; - - /* Skip this instruction in the next round */ - ++I; - } - } - - /* Next entry */ - ++I; - } -} - - - static void InsertEntry (StackOpData* D, CodeEntry* E, int Index) /* Insert a new entry. Depending on Index, D->PushIndex and D->OpIndex will * be adjusted by this function. @@ -389,64 +419,96 @@ static void DelEntry (StackOpData* D, int Index) -static void CheckOneDirectOp (LoadRegInfo* LI, unsigned char Offs) -/* Check if the given entry is a lda instruction with an addressing mode - * that allows us to replace it by another operation (like ora). If so, we may - * use this location for the or and must not save the value in the zero - * page location. +static void AdjustStackOffset (StackOpData* D, unsigned Offs) +/* Adjust the offset for all stack accesses in the range PushIndex to OpIndex. + * OpIndex is adjusted according to the insertions. */ { - /* Get the load entry */ - CodeEntry* E = LI->LoadEntry; - if (E == 0) { - /* No load insn */ - return; - } + /* Walk over all entries */ + int I = D->PushIndex + 1; + while (I < D->OpIndex) { + + CodeEntry* E = CS_GetEntry (D->Code, I); + + int NeedCorrection = 0; + if ((E->Use & REG_SP) != 0) { + + /* Check for some things that should not happen */ + CHECK (E->AM == AM65_ZP_INDY || E->RI->In.RegY >= (short) Offs); + CHECK (strcmp (E->Arg, "sp") == 0); + + /* We need to correct this one */ + NeedCorrection = 1; + + } else if (CE_IsCallTo (E, "ldaxysp")) { + + /* We need to correct this one */ + NeedCorrection = 1; - /* Check the load entry */ - if (E) { - /* Must check the call first since addressing mode is ABS, so second - * "if" will catch otherwise. - */ - if (CE_IsCallTo (E, "ldaxysp")) { - /* Same as single loads from stack. Since we must distinguish - * between A and X here, the necessary offset is passed to the - * function as a parameter. - */ - LI->Offs = (unsigned char) E->RI->In.RegY - Offs; - LI->Flags |= (LI_DIRECT | LI_RELOAD_Y); - } else if (E->AM == AM65_IMM || E->AM == AM65_ZP || E->AM == AM65_ABS) { - /* These insns are all ok and replaceable */ - LI->Flags |= LI_DIRECT; - } else if (E->AM == AM65_ZP_INDY && - RegValIsKnown (E->RI->In.RegY) && - strcmp (E->Arg, "sp") == 0) { - /* A load from the stack with known offset is also ok, but in this - * case we must reload the index register later. Please note that - * a load indirect via other zero page locations is not ok, since - * these locations may change between the push and the actual - * operation. - */ - LI->Offs = (unsigned char) E->RI->In.RegY; - LI->Flags |= (LI_DIRECT | LI_RELOAD_Y); } + + if (NeedCorrection) { + + /* Get the code entry before this one. If it's a LDY, adjust the + * value. + */ + CodeEntry* P = CS_GetPrevEntry (D->Code, I); + if (P && P->OPC == OP65_LDY && CE_IsConstImm (P)) { + + /* The Y load is just before the stack access, adjust it */ + CE_SetNumArg (P, P->Num - Offs); + + } else { + + /* Insert a new load instruction before the stack access */ + const char* Arg = MakeHexArg (E->RI->In.RegY - Offs); + CodeEntry* X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI); + InsertEntry (D, X, I++); + + } + + /* If we need the value of Y later, be sure to reload it */ + if (RegYUsed (D->Code, I+1)) { + const char* Arg = MakeHexArg (E->RI->In.RegY); + CodeEntry* X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI); + InsertEntry (D, X, I+1); + + /* Skip this instruction in the next round */ + ++I; + } + } + + /* Next entry */ + ++I; + } + + /* If we have rhs load insns that load from stack, we'll have to adjust + * the offsets for these also. + */ + if (D->Rhs.A.Flags & LI_RELOAD_Y) { + D->Rhs.A.Offs -= Offs; + } + if (D->Rhs.X.Flags & LI_RELOAD_Y) { + D->Rhs.X.Offs -= Offs; } } -static void CheckDirectOp (StackOpData* D) -/* Check if the given entry is a lda instruction with an addressing mode - * that allows us to replace it by another operation (like ora). If so, we may - * use this location for the or and must not save the value in the zero - * page location. - */ +static void AddStoreA (StackOpData* D) +/* Add a store to zero page after the push insn */ { - /* Check flags for all load instructions */ - CheckOneDirectOp (&D->Lhs.A, 1); - CheckOneDirectOp (&D->Lhs.X, 0); - CheckOneDirectOp (&D->Rhs.A, 1); - CheckOneDirectOp (&D->Rhs.X, 0); + CodeEntry* X = NewCodeEntry (OP65_STA, AM65_ZP, D->ZPLo, 0, D->PushEntry->LI); + InsertEntry (D, X, D->PushIndex+1); +} + + + +static void AddStoreX (StackOpData* D) +/* Add a store to zero page after the push insn */ +{ + CodeEntry* X = NewCodeEntry (OP65_STX, AM65_ZP, D->ZPHi, 0, D->PushEntry->LI); + InsertEntry (D, X, D->PushIndex+1); } @@ -458,24 +520,20 @@ static void ReplacePushByStore (StackOpData* D) * end of each routine). */ { - CodeEntry* X; - /* Store the value into the zeropage instead of pushing it. Check high * byte first so that the store is later in A/X order. */ if ((D->Lhs.X.Flags & LI_DIRECT) == 0) { - X = NewCodeEntry (OP65_STX, AM65_ZP, D->ZPHi, 0, D->PushEntry->LI); - InsertEntry (D, X, D->PushIndex+1); + AddStoreX (D); } if ((D->Lhs.A.Flags & LI_DIRECT) == 0) { - X = NewCodeEntry (OP65_STA, AM65_ZP, D->ZPLo, 0, D->PushEntry->LI); - InsertEntry (D, X, D->PushIndex+1); + AddStoreA (D); } } -static void AddOpLow (StackOpData* D, opc_t OPC) +static void AddOpLow (StackOpData* D, opc_t OPC, LoadInfo* LI) /* Add an op for the low byte of an operator. This function honours the * OP_DIRECT and OP_RELOAD_Y flags and generates the necessary instructions. * All code is inserted at the current insertion point. @@ -483,21 +541,21 @@ static void AddOpLow (StackOpData* D, opc_t OPC) { CodeEntry* X; - if ((D->Lhs.A.Flags & LI_DIRECT) != 0) { + if ((LI->A.Flags & LI_DIRECT) != 0) { /* Op with a variable location. If the location is on the stack, we * need to reload the Y register. */ - if ((D->Lhs.A.Flags & LI_RELOAD_Y) == 0) { + if ((LI->A.Flags & LI_RELOAD_Y) == 0) { /* opc ... */ - CodeEntry* LoadA = D->Lhs.A.LoadEntry; + CodeEntry* LoadA = LI->A.LoadEntry; X = NewCodeEntry (OPC, LoadA->AM, LoadA->Arg, 0, D->OpEntry->LI); InsertEntry (D, X, D->IP++); } else { /* ldy #offs */ - const char* Arg = MakeHexArg (D->Lhs.A.Offs); + const char* Arg = MakeHexArg (LI->A.Offs); X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, D->OpEntry->LI); InsertEntry (D, X, D->IP++); @@ -508,7 +566,7 @@ static void AddOpLow (StackOpData* D, opc_t OPC) } /* In both cases, we can remove the load */ - D->Lhs.A.Flags |= LI_REMOVE; + LI->A.Flags |= LI_REMOVE; } else { @@ -521,7 +579,7 @@ static void AddOpLow (StackOpData* D, opc_t OPC) -static void AddOpHigh (StackOpData* D, opc_t OPC) +static void AddOpHigh (StackOpData* D, opc_t OPC, LoadInfo* LI, int KeepResult) /* Add an op for the high byte of an operator. Special cases (constant values * or similar) have to be checked separately, the function covers only the * generic case. Code is inserted at the insertion point. @@ -529,27 +587,29 @@ static void AddOpHigh (StackOpData* D, opc_t OPC) { CodeEntry* X; - /* pha */ - X = NewCodeEntry (OP65_PHA, AM65_IMP, 0, 0, D->OpEntry->LI); - InsertEntry (D, X, D->IP++); + if (KeepResult) { + /* pha */ + X = NewCodeEntry (OP65_PHA, AM65_IMP, 0, 0, D->OpEntry->LI); + InsertEntry (D, X, D->IP++); + } /* txa */ X = NewCodeEntry (OP65_TXA, AM65_IMP, 0, 0, D->OpEntry->LI); InsertEntry (D, X, D->IP++); - if ((D->Lhs.X.Flags & LI_DIRECT) != 0) { + if ((LI->X.Flags & LI_DIRECT) != 0) { - if ((D->Lhs.X.Flags & LI_RELOAD_Y) == 0) { + if ((LI->X.Flags & LI_RELOAD_Y) == 0) { /* opc xxx */ - CodeEntry* LoadX = D->Lhs.X.LoadEntry; - X = NewCodeEntry (OPC, LoadX->AM, LoadX->Arg, 0, D->OpEntry->LI); + CodeEntry* LoadX = LI->X.LoadEntry; + X = NewCodeEntry (OPC, LoadX->AM, LoadX->Arg, 0, D->OpEntry->LI); InsertEntry (D, X, D->IP++); } else { /* ldy #const */ - const char* Arg = MakeHexArg (D->Lhs.X.Offs); + const char* Arg = MakeHexArg (LI->X.Offs); X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, D->OpEntry->LI); InsertEntry (D, X, D->IP++); @@ -559,7 +619,7 @@ static void AddOpHigh (StackOpData* D, opc_t OPC) } /* In both cases, we can remove the load */ - D->Lhs.X.Flags |= LI_REMOVE; + LI->X.Flags |= LI_REMOVE; } else { /* opc zphi */ @@ -567,13 +627,15 @@ static void AddOpHigh (StackOpData* D, opc_t OPC) InsertEntry (D, X, D->IP++); } - /* tax */ - X = NewCodeEntry (OP65_TAX, AM65_IMP, 0, 0, D->OpEntry->LI); - InsertEntry (D, X, D->IP++); + if (KeepResult) { + /* tax */ + X = NewCodeEntry (OP65_TAX, AM65_IMP, 0, 0, D->OpEntry->LI); + InsertEntry (D, X, D->IP++); - /* pla */ - X = NewCodeEntry (OP65_PLA, AM65_IMP, 0, 0, D->OpEntry->LI); - InsertEntry (D, X, D->IP++); + /* pla */ + X = NewCodeEntry (OP65_PLA, AM65_IMP, 0, 0, D->OpEntry->LI); + InsertEntry (D, X, D->IP++); + } } @@ -584,11 +646,21 @@ static void RemoveRegLoads (StackOpData* D, LoadInfo* LI) /* Both registers may be loaded with one insn, but DelEntry will in this * case clear the other one. */ - if (LI->A.LoadIndex >= 0 && (LI->A.Flags & LI_REMOVE)) { - DelEntry (D, LI->A.LoadIndex); + if (LI->A.Flags & LI_REMOVE) { + if (LI->A.LoadIndex >= 0) { + DelEntry (D, LI->A.LoadIndex); + } + if (LI->A.XferIndex >= 0) { + DelEntry (D, LI->A.XferIndex); + } } - if (LI->X.LoadIndex >= 0 && (LI->X.Flags & LI_REMOVE)) { - DelEntry (D, LI->X.LoadIndex); + if (LI->X.Flags & LI_REMOVE) { + if (LI->X.LoadIndex >= 0) { + DelEntry (D, LI->X.LoadIndex); + } + if (LI->X.XferIndex >= 0) { + DelEntry (D, LI->X.XferIndex); + } } } @@ -649,8 +721,119 @@ static int IsRegVar (StackOpData* D) +static unsigned Opt_toseqax_tosneax (StackOpData* D, const char* BoolTransformer) +/* Optimize the toseqax and tosneax sequences. */ +{ + CodeEntry* X; + CodeLabel* L; + + /* Create a call to the boolean transformer function and a label for this + * insn. This is needed for all variants. Other insns are inserted *before* + * the call. + */ + X = NewCodeEntry (OP65_JSR, AM65_ABS, BoolTransformer, 0, D->OpEntry->LI); + InsertEntry (D, X, D->OpIndex + 1); + L = CS_GenLabel (D->Code, X); + + /* If the lhs is direct (but not stack relative), encode compares with lhs + * effectively reverting the order (which doesn't matter for ==). + */ + if ((D->Lhs.A.Flags & (LI_DIRECT | LI_RELOAD_Y)) == LI_DIRECT && + (D->Lhs.X.Flags & (LI_DIRECT | LI_RELOAD_Y)) == LI_DIRECT) { + + CodeEntry* LoadX = D->Lhs.X.LoadEntry; + CodeEntry* LoadA = D->Lhs.A.LoadEntry; + + D->IP = D->OpIndex+1; + + /* cpx */ + X = NewCodeEntry (OP65_CPX, LoadX->AM, LoadX->Arg, 0, D->OpEntry->LI); + InsertEntry (D, X, D->IP++); + + /* bne L */ + X = NewCodeEntry (OP65_BNE, AM65_BRA, L->Name, L, D->OpEntry->LI); + InsertEntry (D, X, D->IP++); + + /* cmp */ + X = NewCodeEntry (OP65_CMP, LoadA->AM, LoadA->Arg, 0, D->OpEntry->LI); + InsertEntry (D, X, D->IP++); + + /* Lhs load entries can be removed */ + D->Lhs.X.Flags |= LI_REMOVE; + D->Lhs.A.Flags |= LI_REMOVE; + + } else if ((D->Rhs.A.Flags & (LI_DIRECT | LI_RELOAD_Y)) == LI_DIRECT && + (D->Rhs.X.Flags & (LI_DIRECT | LI_RELOAD_Y)) == LI_DIRECT) { + + CodeEntry* LoadX = D->Rhs.X.LoadEntry; + CodeEntry* LoadA = D->Rhs.A.LoadEntry; + + D->IP = D->OpIndex+1; + + /* cpx */ + X = NewCodeEntry (OP65_CPX, LoadX->AM, LoadX->Arg, 0, D->OpEntry->LI); + InsertEntry (D, X, D->IP++); + + /* bne L */ + X = NewCodeEntry (OP65_BNE, AM65_BRA, L->Name, L, D->OpEntry->LI); + InsertEntry (D, X, D->IP++); + + /* cmp */ + X = NewCodeEntry (OP65_CMP, LoadA->AM, LoadA->Arg, 0, D->OpEntry->LI); + InsertEntry (D, X, D->IP++); + + /* Rhs load entries can be removed */ + D->Rhs.X.Flags |= LI_REMOVE; + D->Rhs.A.Flags |= LI_REMOVE; + + } else if ((D->Rhs.A.Flags & LI_DIRECT) != 0 && + (D->Rhs.X.Flags & LI_DIRECT) != 0) { + + D->IP = D->OpIndex+1; + + /* Add operand for low byte */ + AddOpLow (D, OP65_CMP, &D->Rhs); + + /* bne L */ + X = NewCodeEntry (OP65_BNE, AM65_BRA, L->Name, L, D->OpEntry->LI); + InsertEntry (D, X, D->IP++); + + /* Add operand for high byte */ + AddOpHigh (D, OP65_CMP, &D->Rhs, 0); + + } else { + + /* Save lhs into zeropage, then compare */ + AddStoreX (D); + AddStoreA (D); + + D->IP = D->OpIndex+1; + + /* cpx */ + X = NewCodeEntry (OP65_CPX, AM65_ZP, D->ZPHi, 0, D->OpEntry->LI); + InsertEntry (D, X, D->IP++); + + /* bne L */ + X = NewCodeEntry (OP65_BNE, AM65_BRA, L->Name, L, D->OpEntry->LI); + InsertEntry (D, X, D->IP++); + + /* cmp */ + X = NewCodeEntry (OP65_CMP, AM65_ZP, D->ZPLo, 0, D->OpEntry->LI); + InsertEntry (D, X, D->IP++); + + } + + /* Remove the push and the call to the tosgeax function */ + RemoveRemainders (D); + + /* We changed the sequence */ + return 1; +} + + + static unsigned Opt___bzero (StackOpData* D) -/* Optimize the __bzero sequence if possible */ +/* Optimize the __bzero sequence */ { CodeEntry* X; const char* Arg; @@ -659,7 +842,8 @@ static unsigned Opt___bzero (StackOpData* D) /* Check if we're using a register variable */ if (!IsRegVar (D)) { /* Store the value into the zeropage instead of pushing it */ - ReplacePushByStore (D); + AddStoreX (D); + AddStoreA (D); } /* If the return value of __bzero is used, we have to add code to reload @@ -743,14 +927,15 @@ static unsigned Opt___bzero (StackOpData* D) static unsigned Opt_staspidx (StackOpData* D) -/* Optimize the staspidx sequence if possible */ +/* Optimize the staspidx sequence */ { CodeEntry* X; /* Check if we're using a register variable */ if (!IsRegVar (D)) { /* Store the value into the zeropage instead of pushing it */ - ReplacePushByStore (D); + AddStoreX (D); + AddStoreA (D); } /* Replace the store subroutine call by a direct op */ @@ -767,14 +952,15 @@ static unsigned Opt_staspidx (StackOpData* D) static unsigned Opt_staxspidx (StackOpData* D) -/* Optimize the staxspidx sequence if possible */ +/* Optimize the staxspidx sequence */ { CodeEntry* X; /* Check if we're using a register variable */ if (!IsRegVar (D)) { /* Store the value into the zeropage instead of pushing it */ - ReplacePushByStore (D); + AddStoreX (D); + AddStoreA (D); } /* Inline the store */ @@ -822,7 +1008,7 @@ static unsigned Opt_staxspidx (StackOpData* D) static unsigned Opt_tosaddax (StackOpData* D) -/* Optimize the tosaddax sequence if possible */ +/* Optimize the tosaddax sequence */ { CodeEntry* X; CodeEntry* N; @@ -852,7 +1038,8 @@ static unsigned Opt_tosaddax (StackOpData* D) int Signed = (strcmp (N->Arg, "ldaidx") == 0); /* Store the value into the zeropage instead of pushing it */ - ReplacePushByStore (D); + AddStoreX (D); + AddStoreA (D); /* Replace the ldy by a tay. Be sure to create the new entry before * deleting the ldy, since we will reference the line info from this @@ -887,13 +1074,6 @@ static unsigned Opt_tosaddax (StackOpData* D) } else { - /* Check the entry before the push. If it's a lda instruction with an - * addressing mode that allows us to replace it, we may use this - * location for the op and must not save the value in the zero page - * location. - */ - CheckDirectOp (D); - /* Store the value into the zeropage instead of pushing it */ ReplacePushByStore (D); @@ -905,17 +1085,26 @@ static unsigned Opt_tosaddax (StackOpData* D) InsertEntry (D, X, D->IP++); /* Low byte */ - AddOpLow (D, OP65_ADC); + AddOpLow (D, OP65_ADC, &D->Lhs); /* High byte */ if (D->PushEntry->RI->In.RegX == 0) { + /* The high byte is the value in X plus the carry */ CodeLabel* L = CS_GenLabel (D->Code, D->NextEntry); + + /* bcc L */ X = NewCodeEntry (OP65_BCC, AM65_BRA, L->Name, L, D->OpEntry->LI); InsertEntry (D, X, D->IP++); + + /* inx */ X = NewCodeEntry (OP65_INX, AM65_IMP, 0, 0, D->OpEntry->LI); InsertEntry (D, X, D->IP++); - } else if (D->OpEntry->RI->In.RegX == 0) { + + } else if (D->OpEntry->RI->In.RegX == 0 && + (RegValIsKnown (D->PushEntry->RI->In.RegX) || + (D->Lhs.X.Flags & LI_RELOAD_Y) == 0)) { + /* The high byte is that of the first operand plus carry */ CodeLabel* L; if (RegValIsKnown (D->PushEntry->RI->In.RegX)) { @@ -923,8 +1112,15 @@ static unsigned Opt_tosaddax (StackOpData* D) const char* Arg = MakeHexArg (D->PushEntry->RI->In.RegX); X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, D->OpEntry->LI); } else { - /* Value of first op high byte is unknown */ - X = NewCodeEntry (OP65_LDX, AM65_ZP, D->ZPHi, 0, D->OpEntry->LI); + /* Value of first op high byte is unknown. Load from ZP or + * original storage. + */ + if (D->Lhs.X.Flags & LI_DIRECT) { + CodeEntry* LoadX = D->Lhs.X.LoadEntry; + X = NewCodeEntry (OP65_LDX, LoadX->AM, LoadX->Arg, 0, D->OpEntry->LI); + } else { + X = NewCodeEntry (OP65_LDX, AM65_ZP, D->ZPHi, 0, D->OpEntry->LI); + } } InsertEntry (D, X, D->IP++); @@ -938,7 +1134,7 @@ static unsigned Opt_tosaddax (StackOpData* D) InsertEntry (D, X, D->IP++); } else { /* High byte is unknown */ - AddOpHigh (D, OP65_ADC); + AddOpHigh (D, OP65_ADC, &D->Lhs, 1); } } @@ -952,33 +1148,17 @@ static unsigned Opt_tosaddax (StackOpData* D) static unsigned Opt_tosandax (StackOpData* D) -/* Optimize the tosandax sequence if possible */ +/* Optimize the tosandax sequence */ { - CodeEntry* X; - - /* Check the entry before the push. If it's a lda instruction with an - * addressing mode that allows us to replace it, we may use this - * location for the op and must not save the value in the zero page - * location. - */ - CheckDirectOp (D); - /* Store the value into the zeropage instead of pushing it */ ReplacePushByStore (D); /* Inline the and, low byte */ D->IP = D->OpIndex + 1; - AddOpLow (D, OP65_AND); + AddOpLow (D, OP65_AND, &D->Lhs); /* High byte */ - if (D->PushEntry->RI->In.RegX == 0 || D->OpEntry->RI->In.RegX == 0) { - /* The high byte is zero */ - X = NewCodeEntry (OP65_LDX, AM65_IMM, "$00", 0, D->OpEntry->LI); - InsertEntry (D, X, D->IP++); - } else { - /* High byte is unknown */ - AddOpHigh (D, OP65_AND); - } + AddOpHigh (D, OP65_AND, &D->Lhs, 1); /* Remove the push and the call to the tosandax function */ RemoveRemainders (D); @@ -989,37 +1169,141 @@ static unsigned Opt_tosandax (StackOpData* D) -static unsigned Opt_tosorax (StackOpData* D) -/* Optimize the tosorax sequence if possible */ +static unsigned Opt_toseqax (StackOpData* D) +/* Optimize the toseqax sequence */ +{ + return Opt_toseqax_tosneax (D, "booleq"); +} + + + +static unsigned Opt_tosgeax (StackOpData* D) +/* Optimize the tosgeax sequence */ { CodeEntry* X; + CodeLabel* L; + + /* Inline the sbc */ + D->IP = D->OpIndex+1; + + /* Must be true because of OP_RHS_LOAD */ + CHECK ((D->Rhs.A.Flags & D->Rhs.X.Flags & LI_DIRECT) != 0); + + /* Add code for low operand */ + AddOpLow (D, OP65_CMP, &D->Rhs); + + /* Add code for high operand */ + AddOpHigh (D, OP65_SBC, &D->Rhs, 0); + + /* eor #$80 */ + X = NewCodeEntry (OP65_EOR, AM65_IMM, "$80", 0, D->OpEntry->LI); + InsertEntry (D, X, D->IP++); + + /* asl a */ + X = NewCodeEntry (OP65_ASL, AM65_ACC, "a", 0, D->OpEntry->LI); + InsertEntry (D, X, D->IP++); + L = CS_GenLabel (D->Code, X); + + /* Insert a bvs L before the eor insn */ + X = NewCodeEntry (OP65_BVS, AM65_BRA, L->Name, L, D->OpEntry->LI); + InsertEntry (D, X, D->IP - 2); + ++D->IP; + + /* lda #$00 */ + X = NewCodeEntry (OP65_LDA, AM65_IMM, "$00", 0, D->OpEntry->LI); + InsertEntry (D, X, D->IP++); + + /* ldx #$00 */ + X = NewCodeEntry (OP65_LDX, AM65_IMM, "$00", 0, D->OpEntry->LI); + InsertEntry (D, X, D->IP++); + + /* rol a */ + X = NewCodeEntry (OP65_ROL, AM65_ACC, "a", 0, D->OpEntry->LI); + InsertEntry (D, X, D->IP++); + + /* Remove the push and the call to the tosgeax function */ + RemoveRemainders (D); + + /* We changed the sequence */ + return 1; +} + + + +static unsigned Opt_tosltax (StackOpData* D) +/* Optimize the tosltax sequence */ +{ + CodeEntry* X; + CodeLabel* L; - /* Check the entry before the push. If it's a lda instruction with an - * addressing mode that allows us to replace it, we may use this - * location for the op and must not save the value in the zero page - * location. - */ - CheckDirectOp (D); + /* Inline the sbc */ + D->IP = D->OpIndex+1; + + /* Must be true because of OP_RHS_LOAD */ + CHECK ((D->Rhs.A.Flags & D->Rhs.X.Flags & LI_DIRECT) != 0); + + /* Add code for low operand */ + AddOpLow (D, OP65_CMP, &D->Rhs); + + /* Add code for high operand */ + AddOpHigh (D, OP65_SBC, &D->Rhs, 0); + + /* eor #$80 */ + X = NewCodeEntry (OP65_EOR, AM65_IMM, "$80", 0, D->OpEntry->LI); + InsertEntry (D, X, D->IP++); + + /* asl a */ + X = NewCodeEntry (OP65_ASL, AM65_ACC, "a", 0, D->OpEntry->LI); + InsertEntry (D, X, D->IP++); + L = CS_GenLabel (D->Code, X); + + /* Insert a bvc L before the eor insn */ + X = NewCodeEntry (OP65_BVC, AM65_BRA, L->Name, L, D->OpEntry->LI); + InsertEntry (D, X, D->IP - 2); + ++D->IP; + + /* lda #$00 */ + X = NewCodeEntry (OP65_LDA, AM65_IMM, "$00", 0, D->OpEntry->LI); + InsertEntry (D, X, D->IP++); + + /* ldx #$00 */ + X = NewCodeEntry (OP65_LDX, AM65_IMM, "$00", 0, D->OpEntry->LI); + InsertEntry (D, X, D->IP++); + + /* rol a */ + X = NewCodeEntry (OP65_ROL, AM65_ACC, "a", 0, D->OpEntry->LI); + InsertEntry (D, X, D->IP++); + + /* Remove the push and the call to the tosltax function */ + RemoveRemainders (D); + + /* We changed the sequence */ + return 1; +} + + + +static unsigned Opt_tosneax (StackOpData* D) +/* Optimize the tosneax sequence */ +{ + return Opt_toseqax_tosneax (D, "boolne"); +} + + + +static unsigned Opt_tosorax (StackOpData* D) +/* Optimize the tosorax sequence */ +{ /* Store the value into the zeropage instead of pushing it */ ReplacePushByStore (D); /* Inline the or, low byte */ D->IP = D->OpIndex + 1; - AddOpLow (D, OP65_ORA); + AddOpLow (D, OP65_ORA, &D->Lhs); /* High byte */ - if (RegValIsKnown (D->PushEntry->RI->In.RegX) && - RegValIsKnown (D->OpEntry->RI->In.RegX)) { - /* Both values known, precalculate the result */ - unsigned char Result = D->PushEntry->RI->In.RegX | D->OpEntry->RI->In.RegX; - const char* Arg = MakeHexArg (Result); - X = NewCodeEntry (OP65_LDX, AM65_IMM, Arg, 0, D->OpEntry->LI); - InsertEntry (D, X, D->IP++); - } else if (D->PushEntry->RI->In.RegX != 0) { - /* High byte is unknown */ - AddOpHigh (D, OP65_ORA); - } + AddOpHigh (D, OP65_ORA, &D->Lhs, 1); /* Remove the push and the call to the tosorax function */ RemoveRemainders (D); @@ -1031,18 +1315,10 @@ static unsigned Opt_tosorax (StackOpData* D) static unsigned Opt_tossubax (StackOpData* D) -/* Optimize the tossubax sequence if possible. Note: subtraction is not - * commutative! - */ +/* Optimize the tossubax sequence. Note: subtraction is not commutative! */ { CodeEntry* X; - /* Check the load entry before the push. If it's a lda instruction with an - * addressing mode that allows us to replace it, we may use this - * location for the op and must not save the value in the zero page - * location. - */ - CheckDirectOp (D); /* Inline the sbc */ D->IP = D->OpIndex+1; @@ -1052,74 +1328,171 @@ static unsigned Opt_tossubax (StackOpData* D) InsertEntry (D, X, D->IP++); /* Must be true because of OP_RHS_LOAD */ - CHECK ((D->Rhs.A.Flags & LI_DIRECT) != 0); + CHECK ((D->Rhs.A.Flags & D->Rhs.X.Flags & LI_DIRECT) != 0); - /* If the location is on the stack, we need to reload the Y register. */ - if ((D->Rhs.A.Flags & LI_RELOAD_Y) == 0) { + /* Add code for low operand */ + AddOpLow (D, OP65_SBC, &D->Rhs); - /* sbc ... */ - CodeEntry* LoadA = D->Rhs.A.LoadEntry; - X = NewCodeEntry (OP65_SBC, LoadA->AM, LoadA->Arg, 0, D->OpEntry->LI); - InsertEntry (D, X, D->IP++); + /* Add code for high operand */ + AddOpHigh (D, OP65_SBC, &D->Rhs, 1); - } else { + /* Remove the push and the call to the tossubax function */ + RemoveRemainders (D); - /* ldy #offs */ - const char* Arg = MakeHexArg (D->Rhs.A.Offs); - X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, D->OpEntry->LI); - InsertEntry (D, X, D->IP++); + /* We changed the sequence */ + return 1; +} + + + +static unsigned Opt_tosugeax (StackOpData* D) +/* Optimize the tosugeax sequence */ +{ + CodeEntry* X; - /* sbc (sp),y */ - X = NewCodeEntry (OP65_SBC, AM65_ZP_INDY, "sp", 0, D->OpEntry->LI); - InsertEntry (D, X, D->IP++); - } - /* In both cases, we can remove the load */ - D->Rhs.A.Flags |= LI_REMOVE; + /* Inline the sbc */ + D->IP = D->OpIndex+1; + + /* Must be true because of OP_RHS_LOAD */ + CHECK ((D->Rhs.A.Flags & D->Rhs.X.Flags & LI_DIRECT) != 0); - /* pha */ - X = NewCodeEntry (OP65_PHA, AM65_IMP, 0, 0, D->OpEntry->LI); + /* Add code for low operand */ + AddOpLow (D, OP65_CMP, &D->Rhs); + + /* Add code for high operand */ + AddOpHigh (D, OP65_SBC, &D->Rhs, 0); + + /* lda #$00 */ + X = NewCodeEntry (OP65_LDA, AM65_IMM, "$00", 0, D->OpEntry->LI); InsertEntry (D, X, D->IP++); - /* txa */ - X = NewCodeEntry (OP65_TXA, AM65_IMP, 0, 0, D->OpEntry->LI); + /* ldx #$00 */ + X = NewCodeEntry (OP65_LDX, AM65_IMM, "$00", 0, D->OpEntry->LI); InsertEntry (D, X, D->IP++); + /* rol a */ + X = NewCodeEntry (OP65_ROL, AM65_ACC, "a", 0, D->OpEntry->LI); + InsertEntry (D, X, D->IP++); + + /* Remove the push and the call to the tosugeax function */ + RemoveRemainders (D); + + /* We changed the sequence */ + return 1; +} + + + +static unsigned Opt_tosugtax (StackOpData* D) +/* Optimize the tosugtax sequence */ +{ + CodeEntry* X; + + + /* Inline the sbc */ + D->IP = D->OpIndex+1; + /* Must be true because of OP_RHS_LOAD */ - CHECK ((D->Rhs.X.Flags & LI_DIRECT) != 0); + CHECK ((D->Rhs.A.Flags & D->Rhs.X.Flags & LI_DIRECT) != 0); - /* If the location is on the stack, we need to reload the Y register. */ - if ((D->Rhs.X.Flags & LI_RELOAD_Y) == 0) { + /* sec */ + X = NewCodeEntry (OP65_SEC, AM65_IMP, 0, 0, D->OpEntry->LI); + InsertEntry (D, X, D->IP++); - /* sbc ... */ - CodeEntry* LoadX = D->Rhs.X.LoadEntry; - X = NewCodeEntry (OP65_SBC, LoadX->AM, LoadX->Arg, 0, D->OpEntry->LI); - InsertEntry (D, X, D->IP++); + /* Add code for low operand */ + AddOpLow (D, OP65_SBC, &D->Rhs); - } else { + /* We need the zero flag, so remember the immediate result */ + X = NewCodeEntry (OP65_STA, AM65_ZP, "tmp1", 0, D->OpEntry->LI); + InsertEntry (D, X, D->IP++); - /* ldy #offs */ - const char* Arg = MakeHexArg (D->Rhs.X.Offs); - X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, D->OpEntry->LI); - InsertEntry (D, X, D->IP++); + /* Add code for high operand */ + AddOpHigh (D, OP65_SBC, &D->Rhs, 0); + + /* Set Z flag */ + X = NewCodeEntry (OP65_ORA, AM65_ZP, "tmp1", 0, D->OpEntry->LI); + InsertEntry (D, X, D->IP++); + + /* Transform to boolean */ + X = NewCodeEntry (OP65_JSR, AM65_ABS, "boolugt", 0, D->OpEntry->LI); + InsertEntry (D, X, D->IP++); + + /* Remove the push and the call to the operator function */ + RemoveRemainders (D); + + /* We changed the sequence */ + return 1; +} + + + +static unsigned Opt_tosuleax (StackOpData* D) +/* Optimize the tosuleax sequence */ +{ + CodeEntry* X; - /* sbc (sp),y */ - X = NewCodeEntry (OP65_SBC, AM65_ZP_INDY, "sp", 0, D->OpEntry->LI); - InsertEntry (D, X, D->IP++); - } - /* In both cases, we can remove the load */ - D->Rhs.X.Flags |= LI_REMOVE; + /* Inline the sbc */ + D->IP = D->OpIndex+1; + + /* Must be true because of OP_RHS_LOAD */ + CHECK ((D->Rhs.A.Flags & D->Rhs.X.Flags & LI_DIRECT) != 0); - /* tax */ - X = NewCodeEntry (OP65_TAX, AM65_IMP, 0, 0, D->OpEntry->LI); + /* sec */ + X = NewCodeEntry (OP65_SEC, AM65_IMP, 0, 0, D->OpEntry->LI); InsertEntry (D, X, D->IP++); - /* pla */ - X = NewCodeEntry (OP65_PLA, AM65_IMP, 0, 0, D->OpEntry->LI); + /* Add code for low operand */ + AddOpLow (D, OP65_SBC, &D->Rhs); + + /* We need the zero flag, so remember the immediate result */ + X = NewCodeEntry (OP65_STA, AM65_ZP, "tmp1", 0, D->OpEntry->LI); InsertEntry (D, X, D->IP++); - /* Remove the push and the call to the tossubax function */ + /* Add code for high operand */ + AddOpHigh (D, OP65_SBC, &D->Rhs, 0); + + /* Set Z flag */ + X = NewCodeEntry (OP65_ORA, AM65_ZP, "tmp1", 0, D->OpEntry->LI); + InsertEntry (D, X, D->IP++); + + /* Transform to boolean */ + X = NewCodeEntry (OP65_JSR, AM65_ABS, "boolule", 0, D->OpEntry->LI); + InsertEntry (D, X, D->IP++); + + /* Remove the push and the call to the operator function */ + RemoveRemainders (D); + + /* We changed the sequence */ + return 1; +} + + + +static unsigned Opt_tosultax (StackOpData* D) +/* Optimize the tosultax sequence */ +{ + CodeEntry* X; + + + /* Inline the sbc */ + D->IP = D->OpIndex+1; + + /* Must be true because of OP_RHS_LOAD */ + CHECK ((D->Rhs.A.Flags & D->Rhs.X.Flags & LI_DIRECT) != 0); + + /* Add code for low operand */ + AddOpLow (D, OP65_CMP, &D->Rhs); + + /* Add code for high operand */ + AddOpHigh (D, OP65_SBC, &D->Rhs, 0); + + /* Transform to boolean */ + X = NewCodeEntry (OP65_JSR, AM65_ABS, "boolult", 0, D->OpEntry->LI); + InsertEntry (D, X, D->IP++); + + /* Remove the push and the call to the operator function */ RemoveRemainders (D); /* We changed the sequence */ @@ -1129,23 +1502,17 @@ static unsigned Opt_tossubax (StackOpData* D) static unsigned Opt_tosxorax (StackOpData* D) -/* Optimize the tosxorax sequence if possible */ +/* Optimize the tosxorax sequence */ { CodeEntry* X; - /* Check the entry before the push. If it's a lda instruction with an - * addressing mode that allows us to replace it, we may use this - * location for the op and must not save the value in the zero page - * location. - */ - CheckDirectOp (D); /* Store the value into the zeropage instead of pushing it */ ReplacePushByStore (D); /* Inline the xor, low byte */ D->IP = D->OpIndex + 1; - AddOpLow (D, OP65_EOR); + AddOpLow (D, OP65_EOR, &D->Lhs); /* High byte */ if (RegValIsKnown (D->PushEntry->RI->In.RegX) && @@ -1156,7 +1523,7 @@ static unsigned Opt_tosxorax (StackOpData* D) InsertEntry (D, X, D->IP++); } else if (D->PushEntry->RI->In.RegX != 0) { /* High byte is unknown */ - AddOpHigh (D, OP65_EOR); + AddOpHigh (D, OP65_EOR, &D->Lhs, 1); } /* Remove the push and the call to the tosandax function */ @@ -1180,10 +1547,16 @@ static const OptFuncDesc FuncTable[] = { { "staxspidx", Opt_staxspidx, REG_AX, OP_NONE }, { "tosaddax", Opt_tosaddax, REG_NONE, OP_NONE }, { "tosandax", Opt_tosandax, REG_NONE, OP_NONE }, + { "toseqax", Opt_toseqax, REG_NONE, OP_NONE }, + { "tosgeax", Opt_tosgeax, REG_NONE, OP_RHS_LOAD_DIRECT }, + { "tosltax", Opt_tosltax, REG_NONE, OP_RHS_LOAD_DIRECT }, + { "tosneax", Opt_tosneax, REG_NONE, OP_NONE }, { "tosorax", Opt_tosorax, REG_NONE, OP_NONE }, -#if 1 - { "tossubax", Opt_tossubax, REG_NONE, OP_NONE }, -#endif + { "tossubax", Opt_tossubax, REG_NONE, OP_RHS_LOAD_DIRECT }, + { "tosugeax", Opt_tosugeax, REG_NONE, OP_RHS_LOAD_DIRECT }, + { "tosugtax", Opt_tosugtax, REG_NONE, OP_RHS_LOAD_DIRECT }, + { "tosuleax", Opt_tosuleax, REG_NONE, OP_RHS_LOAD_DIRECT }, + { "tosultax", Opt_tosultax, REG_NONE, OP_RHS_LOAD_DIRECT }, { "tosxorax", Opt_tosxorax, REG_NONE, OP_NONE }, }; #define FUNC_COUNT (sizeof(FuncTable) / sizeof(FuncTable[0])) @@ -1224,16 +1597,42 @@ static int HarmlessCall (const char* Name) static const char* Tab[] = { "aslax1", "aslax2", + "aslax3", + "aslax4", "asrax1", "asrax2", + "asrax3", + "asrax4", "bnegax", + "decax1", + "decax2", + "decax3", + "decax4", + "decax5", + "decax6", + "decax7", + "decax8", + "decaxy", + "incax1", + "incax2", + "incax3", + "incax4", + "incax5", + "incax6", + "incax7", + "incax8", + "incaxy", "ldaxidx", "ldaxysp", "negax", "shlax1", "shlax2", + "shlax3", + "shlax4", "shrax1", "shrax2", + "shrax3", + "shrax4", }; void* R = bsearch (Name, @@ -1284,13 +1683,29 @@ static int PreCondOk (StackOpData* D) /* Cannot optimize */ return 0; } - if ((D->OptFunc->Flags & OP_LHS_LOAD) != 0 && - (D->Lhs.A.LoadIndex < 0 || D->Lhs.X.LoadIndex < 0)) { - /* Cannot optimize */ - return 0; + if ((D->OptFunc->Flags & OP_LHS_LOAD) != 0) { + if (D->Lhs.A.LoadIndex < 0 || D->Lhs.X.LoadIndex < 0) { + /* Cannot optimize */ + return 0; + } else if ((D->OptFunc->Flags & OP_LHS_LOAD_DIRECT) != 0) { + if ((D->Lhs.A.Flags & D->Lhs.X.Flags & LI_DIRECT) == 0) { + /* Cannot optimize */ + return 0; + } + } + } + if ((D->OptFunc->Flags & OP_RHS_LOAD) != 0) { + if (D->Rhs.A.LoadIndex < 0 || D->Rhs.X.LoadIndex < 0) { + /* Cannot optimize */ + return 0; + } else if ((D->OptFunc->Flags & OP_RHS_LOAD_DIRECT) != 0) { + if ((D->Rhs.A.Flags & D->Rhs.X.Flags & LI_DIRECT) == 0) { + /* Cannot optimize */ + return 0; + } + } } - if ((D->OptFunc->Flags & OP_RHS_LOAD) != 0 && - (D->Rhs.A.LoadIndex < 0 || D->Rhs.X.LoadIndex < 0)) { + if ((D->Rhs.A.Flags | D->Rhs.X.Flags) & LI_DUP_LOAD) { /* Cannot optimize */ return 0; } @@ -1302,9 +1717,6 @@ static int PreCondOk (StackOpData* D) } else if ((D->UsedRegs & REG_SREG) == REG_NONE) { D->ZPLo = "sreg"; D->ZPHi = "sreg+1"; - } else if ((D->UsedRegs & REG_PTR1) == REG_NONE) { - D->ZPLo = "ptr1"; - D->ZPHi = "ptr1+1"; } else if ((D->UsedRegs & REG_PTR2) == REG_NONE) { D->ZPLo = "ptr2"; D->ZPHi = "ptr2+1"; @@ -1330,7 +1742,8 @@ unsigned OptStackOps (CodeSeg* S) { unsigned Changes = 0; /* Number of changes in one run */ StackOpData Data; - unsigned I; + int I; + int OldEntryCount; /* Old number of entries */ enum { Initialize, @@ -1340,9 +1753,6 @@ unsigned OptStackOps (CodeSeg* S) } State = Initialize; - /* Generate register info */ - CS_GenRegInfo (S); - /* Remember the code segment in the info struct */ Data.Code = S; @@ -1362,7 +1772,7 @@ unsigned OptStackOps (CodeSeg* S) * intermediate code for zero page use. */ I = 0; - while (I < CS_GetEntryCount (S)) { + while (I < (int)CS_GetEntryCount (S)) { /* Get the next entry */ CodeEntry* E = CS_GetEntry (S, I); @@ -1372,12 +1782,17 @@ unsigned OptStackOps (CodeSeg* S) case Initialize: ResetStackOpData (&Data); + State = Search; /* FALLTHROUGH */ case Search: /* While searching, track register load insns, so we can tell * what is in a register once pushax is encountered. */ + if (CE_HasLabel (E)) { + /* Currently we don't track across branches */ + ClearLoadInfo (&Data.Lhs); + } if (CE_IsCallTo (E, "pushax")) { Data.PushIndex = I; State = FoundPush; @@ -1392,6 +1807,10 @@ unsigned OptStackOps (CodeSeg* S) * follow and in the meantime, track zeropage usage and check * for code that will disable us from translating the sequence. */ + if (CE_HasLabel (E)) { + /* Currently we don't track across branches */ + ClearLoadInfo (&Data.Rhs); + } if (E->OPC == OP65_JSR) { /* Subroutine call: Check if this is one of the functions, @@ -1418,7 +1837,7 @@ unsigned OptStackOps (CodeSeg* S) TrackLoads (&Data.Rhs, E, I); } - } else if (E->Info & OF_STORE) { + } else if (E->Info & OF_STORE && (E->Chg & REG_ZP) == 0) { /* Too dangerous - there may be a change of a variable * within the sequence. @@ -1481,20 +1900,18 @@ unsigned OptStackOps (CodeSeg* S) break; } - /* Adjust stack offsets to account for the upcoming removal */ - AdjustStackOffset (&Data, 2); - - /* Regenerate register info, since AdjustStackOffset changed - * the code - */ - CS_GenRegInfo (S); - /* Prepare the remainder of the data structure. */ Data.PrevEntry = CS_GetPrevEntry (S, Data.PushIndex); Data.PushEntry = CS_GetEntry (S, Data.PushIndex); Data.OpEntry = CS_GetEntry (S, Data.OpIndex); Data.NextEntry = CS_GetNextEntry (S, Data.OpIndex); + /* Remember the current number of code lines */ + OldEntryCount = CS_GetEntryCount (S); + + /* Adjust stack offsets to account for the upcoming removal */ + AdjustStackOffset (&Data, 2); + /* Regenerate register info, since AdjustStackOffset changed * the code */ @@ -1503,12 +1920,17 @@ unsigned OptStackOps (CodeSeg* S) /* Call the optimizer function */ Changes += Data.OptFunc->Func (&Data); + /* Since the function may have added or deleted entries, + * correct the index. + */ + I += CS_GetEntryCount (S) - OldEntryCount; + /* Regenerate register info */ CS_GenRegInfo (S); /* Done */ State = Initialize; - break; + continue; } @@ -1517,9 +1939,6 @@ unsigned OptStackOps (CodeSeg* S) } - /* Free the register info */ - CS_FreeRegInfo (S); - /* Return the number of changes made */ return Changes; }