/* */
/* */
/* */
-/* (C) 2001-2004 Ullrich von Bassewitz */
-/* Römerstrasse 52 */
+/* (C) 2001-2009 Ullrich von Bassewitz */
+/* Roemerstrasse 52 */
/* D-70794 Filderstadt */
/* EMail: uz@cc65.org */
/* */
+/*****************************************************************************/
+/* Load tracking data */
+/*****************************************************************************/
+
+
+
+/* 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 */
+
+/* Structure that tells us how to load the lhs values */
+typedef struct LoadRegInfo LoadRegInfo;
+struct LoadRegInfo {
+ 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 */
+};
+
+/* Now combined for both registers */
+typedef struct LoadInfo LoadInfo;
+struct LoadInfo {
+ LoadRegInfo A; /* Info for A register */
+ LoadRegInfo X; /* Info for X register */
+ LoadRegInfo Y; /* Info for Y register */
+};
+
+
+
/*****************************************************************************/
/* Data */
/*****************************************************************************/
-/* Structure that holds the needed data */
+/* Flags for the functions */
+typedef enum {
+ OP_NONE = 0x00, /* Nothing special */
+ 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_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 */
typedef struct StackOpData StackOpData;
+
+/* Structure that describes an optimizer subfunction for a specific op */
+typedef unsigned (*OptFunc) (StackOpData* D);
+typedef struct OptFuncDesc OptFuncDesc;
+struct OptFuncDesc {
+ const char* Name; /* Name of the replaced runtime function */
+ OptFunc Func; /* Function pointer */
+ unsigned UnusedRegs; /* Regs that must not be used later */
+ OP_FLAGS Flags; /* Flags */
+};
+
+/* Structure that holds the needed data */
struct StackOpData {
- CodeSeg* Code; /* Pointer to code segment */
- unsigned Flags; /* Flags to remember things */
- unsigned PushIndex; /* Index of call to pushax in codeseg */
- unsigned OpIndex; /* Index of actual operation */
- CodeEntry* PrevEntry; /* Entry before the call to pushax */
- CodeEntry* PushEntry; /* Pointer to entry with call to pushax */
- CodeEntry* OpEntry; /* Pointer to entry with op */
- CodeEntry* NextEntry; /* Entry after the op */
- const char* ZPLo; /* Lo byte of zero page loc to use */
- const char* ZPHi; /* Hi byte of zero page loc to use */
- unsigned IP; /* Insertion point used by some routines */
+ CodeSeg* Code; /* Pointer to code segment */
+ unsigned Flags; /* Flags to remember things */
+
+ /* Pointer to optimizer subfunction description */
+ const OptFuncDesc* OptFunc;
+
+ /* ZP register usage inside the sequence */
+ unsigned UsedRegs;
+
+ /* Register load information for lhs and rhs */
+ LoadInfo Lhs;
+ LoadInfo Rhs;
+
+ /* Several indices of insns in the code segment */
+ int PushIndex; /* Index of call to pushax in codeseg */
+ int OpIndex; /* Index of actual operation */
+
+ /* Pointers to insns in the code segment */
+ CodeEntry* PrevEntry; /* Entry before the call to pushax */
+ CodeEntry* PushEntry; /* Pointer to entry with call to pushax */
+ CodeEntry* OpEntry; /* Pointer to entry with op */
+ CodeEntry* NextEntry; /* Entry after the op */
+
+ const char* ZPLo; /* Lo byte of zero page loc to use */
+ const char* ZPHi; /* Hi byte of zero page loc to use */
+ unsigned IP; /* Insertion point used by some routines */
};
-/* Flags returned by DirectOp */
-#define OP_DIRECT 0x01 /* Direct op may be used */
-#define OP_RELOAD_Y 0x02 /* Must reload index register Y */
+
+
+/*****************************************************************************/
+/* Load tracking code */
+/*****************************************************************************/
+
+
+
+static void ClearLoadRegInfo (LoadRegInfo* RI)
+/* Clear a LoadRegInfo struct */
+{
+ RI->LoadIndex = -1;
+}
+
+
+
+static void FinalizeLoadRegInfo (LoadRegInfo* RI, CodeSeg* S)
+/* Prepare a LoadRegInfo struct for use */
+{
+ /* Get the entry */
+ if (RI->LoadIndex >= 0) {
+ RI->LoadEntry = CS_GetEntry (S, RI->LoadIndex);
+ } else {
+ RI->LoadEntry = 0;
+ }
+ RI->Flags = LI_NONE;
+}
+
+
+
+static void ClearLoadInfo (LoadInfo* LI)
+/* Clear a LoadInfo struct */
+{
+ ClearLoadRegInfo (&LI->A);
+ ClearLoadRegInfo (&LI->X);
+ ClearLoadRegInfo (&LI->Y);
+}
+
+
+
+static void AdjustLoadRegInfo (LoadRegInfo* RI, int DelIndex, int Change)
+/* Adjust a load register info struct after deleting or inserting an entry
+ * with a given index
+ */
+{
+ CHECK (abs (Change) == 1);
+ if (Change < 0) {
+ /* Deletion */
+ if (DelIndex < RI->LoadIndex) {
+ --RI->LoadIndex;
+ } else if (DelIndex == RI->LoadIndex) {
+ /* Has been removed */
+ RI->LoadIndex = -1;
+ RI->LoadEntry = 0;
+ }
+ } else {
+ /* Insertion */
+ if (DelIndex <= RI->LoadIndex) {
+ ++RI->LoadIndex;
+ }
+ }
+}
+
+
+
+static void FinalizeLoadInfo (LoadInfo* LI, CodeSeg* S)
+/* Prepare a LoadInfo struct for use */
+{
+ /* Get the entries */
+ FinalizeLoadRegInfo (&LI->A, S);
+ FinalizeLoadRegInfo (&LI->X, S);
+ FinalizeLoadRegInfo (&LI->Y, S);
+}
+
+
+
+static void AdjustLoadInfo (LoadInfo* LI, int DelIndex, 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);
+}
+
+
+
+static void TrackLoads (LoadInfo* LI, CodeEntry* E, int I)
+/* Track loads for a code entry */
+{
+ if (E->Info & OF_LOAD) {
+ if (E->Chg & REG_A) {
+ LI->A.LoadIndex = I;
+ }
+ if (E->Chg & REG_X) {
+ LI->X.LoadIndex = I;
+ }
+ if (E->Chg & REG_Y) {
+ LI->Y.LoadIndex = I;
+ }
+ } else if (E->Info & OF_XFR) {
+ 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;
+ }
+ } else if (CE_IsCallTo (E, "ldaxysp")) {
+ /* Both registers set, Y changed */
+ LI->A.LoadIndex = I;
+ LI->X.LoadIndex = I;
+ LI->Y.LoadIndex = -1;
+ } else {
+ if (E->Chg & REG_A) {
+ LI->A.LoadIndex = -1;
+ }
+ if (E->Chg & REG_X) {
+ LI->X.LoadIndex = -1;
+ }
+ if (E->Chg & REG_Y) {
+ LI->Y.LoadIndex = -1;
+ }
+ }
+}
-static unsigned AdjustStackOffset (CodeSeg* S, unsigned Start, unsigned Stop,
- unsigned Offs)
-/* Adjust the offset for all stack accesses in the range Start to Stop, both
- * inclusive. The function returns the number of instructions that have been
- * inserted.
+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.
*/
{
- /* Number of inserted instructions */
- unsigned Inserted = 0;
-
/* Walk over all entries */
- unsigned I = Start;
- while (I <= Stop) {
+ int I = D->PushIndex + 1;
+ while (I < D->OpIndex) {
- CodeEntry* E = CS_GetEntry (S, I);
+ CodeEntry* E = CS_GetEntry (D->Code, I);
int NeedCorrection = 0;
if ((E->Use & REG_SP) != 0) {
if (NeedCorrection) {
- CodeEntry* P;
-
- /* If the Y register value is needed later, we have to reload the
- * register after changing it.
- */
- int NeedY = RegYUsed (S, I+1);
- unsigned YVal = E->RI->In.RegY;
-
/* Get the code entry before this one. If it's a LDY, adjust the
* value.
*/
- P = CS_GetPrevEntry (S, I);
- if (P && P->OPC == OP65_LDY && CE_KnownImm (P)) {
+ 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);
+ /* 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 (YVal - Offs);
- CodeEntry* X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, E->LI);
- CS_InsertEntry (S, X, I);
+ /* 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 */
- ++Inserted;
- ++Stop;
-
- /* Be sure to skip the stack access for the next round */
- ++I;
+ ++D->OpIndex;
}
/* If we need the value of Y later, be sure to reload it */
- if (NeedY) {
- const char* Arg = MakeHexArg (YVal);
+ 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 (S, X, I+1);
+ CS_InsertEntry (D->Code, X, I+1);
/* One more inserted entries */
- ++Inserted;
- ++Stop;
+ ++D->OpIndex;
- /* Skip this instruction int the next round */
+ /* Skip this instruction in the next round */
++I;
}
}
/* Next entry */
++I;
}
-
- /* Return the number of inserted entries */
- return Inserted;
}
-static void InsertEntry (StackOpData* D, CodeEntry* E, unsigned Index)
+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.
*/
/* Insert the entry into the code segment */
CS_InsertEntry (D->Code, E, Index);
+ /* Adjust register loads if necessary */
+ AdjustLoadInfo (&D->Lhs, Index, 1);
+ AdjustLoadInfo (&D->Rhs, Index, 1);
+
/* Adjust the indices if necessary */
if (D->PushEntry && Index <= D->PushIndex) {
++D->PushIndex;
-static void DelEntry (StackOpData* D, unsigned Index)
+static void DelEntry (StackOpData* D, int Index)
/* Delete an entry. Depending on Index, D->PushIndex and D->OpIndex will be
* adjusted by this function, and PushEntry/OpEntry may get invalidated.
*/
/* Delete the entry from the code segment */
CS_DelEntry (D->Code, Index);
- /* Adjust the indices if necessary */
+ /* Adjust register loads if necessary */
+ AdjustLoadInfo (&D->Lhs, Index, -1);
+ AdjustLoadInfo (&D->Rhs, Index, -1);
+
+ /* Adjust the other indices if necessary */
if (Index < D->PushIndex) {
--D->PushIndex;
} else if (Index == D->PushIndex) {
-static void CheckDirectOp (StackOpData* D)
+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.
*/
{
- /* We need the entry before the push */
- CodeEntry* E;
- CHECK ((E = D->PrevEntry) != 0);
+ /* Get the load entry */
+ CodeEntry* E = LI->LoadEntry;
+ if (E == 0) {
+ /* No load insn */
+ return;
+ }
- if (E->OPC == OP65_LDA) {
- if (E->AM == AM65_IMM || E->AM == AM65_ZP || E->AM == AM65_ABS) {
+ /* 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 */
- D->Flags |= OP_DIRECT;
- } else if (E->AM == AM65_ZP_INDY && RegValIsKnown (E->RI->In.RegY) &&
+ 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
* these locations may change between the push and the actual
* operation.
*/
- D->Flags |= (OP_DIRECT | OP_RELOAD_Y);
+ LI->Offs = (unsigned char) E->RI->In.RegY;
+ LI->Flags |= (LI_DIRECT | LI_RELOAD_Y);
}
}
}
+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.
+ */
+{
+ /* 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);
+}
+
+
+
static void ReplacePushByStore (StackOpData* D)
/* Replace the call to the push subroutine by a store into the zero page
* location (actually, the push is not replaced, because we need it for
{
CodeEntry* X;
- /* Store the value into the zeropage instead of pushing it */
- X = NewCodeEntry (OP65_STX, AM65_ZP, D->ZPHi, 0, D->PushEntry->LI);
- InsertEntry (D, X, D->PushIndex+1);
- if ((D->Flags & OP_DIRECT) == 0) {
+ /* 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);
+ }
+ 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);
}
{
CodeEntry* X;
- if ((D->Flags & OP_DIRECT) != 0) {
+ if ((D->Lhs.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->Flags & OP_RELOAD_Y) != 0) {
- const char* Arg = MakeHexArg (D->PrevEntry->RI->In.RegY);
+ if ((D->Lhs.A.Flags & LI_RELOAD_Y) == 0) {
+
+ /* opc ... */
+ CodeEntry* LoadA = D->Lhs.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);
X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, D->OpEntry->LI);
InsertEntry (D, X, D->IP++);
+
+ /* opc (sp),y */
+ X = NewCodeEntry (OPC, AM65_ZP_INDY, "sp", 0, D->OpEntry->LI);
+ InsertEntry (D, X, D->IP++);
+
}
- X = NewCodeEntry (OPC, D->PrevEntry->AM, D->PrevEntry->Arg, 0, D->OpEntry->LI);
+
+ /* In both cases, we can remove the load */
+ D->Lhs.A.Flags |= LI_REMOVE;
+
} else {
- /* Op with temp storage */
- X = NewCodeEntry (OPC, AM65_ZP, D->ZPLo, 0, D->OpEntry->LI);
+
+ /* Op with temp storage */
+ X = NewCodeEntry (OPC, AM65_ZP, D->ZPLo, 0, D->OpEntry->LI);
+ InsertEntry (D, X, D->IP++);
+
}
- InsertEntry (D, X, D->IP++);
}
{
CodeEntry* X;
- /* High byte is unknown */
- X = NewCodeEntry (OP65_STA, AM65_ZP, D->ZPLo, 0, D->OpEntry->LI);
+ /* 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++);
- X = NewCodeEntry (OPC, AM65_ZP, D->ZPHi, 0, D->OpEntry->LI);
- InsertEntry (D, X, D->IP++);
+
+ if ((D->Lhs.X.Flags & LI_DIRECT) != 0) {
+
+ if ((D->Lhs.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);
+ InsertEntry (D, X, D->IP++);
+
+ } else {
+
+ /* ldy #const */
+ const char* Arg = MakeHexArg (D->Lhs.X.Offs);
+ X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, D->OpEntry->LI);
+ InsertEntry (D, X, D->IP++);
+
+ /* opc (sp),y */
+ X = NewCodeEntry (OPC, AM65_ZP_INDY, "sp", 0, D->OpEntry->LI);
+ InsertEntry (D, X, D->IP++);
+ }
+
+ /* In both cases, we can remove the load */
+ D->Lhs.X.Flags |= LI_REMOVE;
+
+ } else {
+ /* opc zphi */
+ X = NewCodeEntry (OPC, AM65_ZP, D->ZPHi, 0, D->OpEntry->LI);
+ InsertEntry (D, X, D->IP++);
+ }
+
+ /* tax */
X = NewCodeEntry (OP65_TAX, AM65_IMP, 0, 0, D->OpEntry->LI);
InsertEntry (D, X, D->IP++);
- X = NewCodeEntry (OP65_LDA, AM65_ZP, D->ZPLo, 0, D->OpEntry->LI);
+
+ /* pla */
+ X = NewCodeEntry (OP65_PLA, AM65_IMP, 0, 0, D->OpEntry->LI);
InsertEntry (D, X, D->IP++);
}
-static void RemovePushAndOp (StackOpData* D)
-/* Remove the call to pushax and the call to the operator subroutine */
+static void RemoveRegLoads (StackOpData* D, LoadInfo* LI)
+/* Remove register load insns */
{
+ /* 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->X.LoadIndex >= 0 && (LI->X.Flags & LI_REMOVE)) {
+ DelEntry (D, LI->X.LoadIndex);
+ }
+}
+
+
+
+static void RemoveRemainders (StackOpData* D)
+/* Remove the code that is unnecessary after translation of the sequence */
+{
+ /* Remove the register loads for lhs and rhs */
+ RemoveRegLoads (D, &D->Lhs);
+ RemoveRegLoads (D, &D->Rhs);
+
+ /* Remove the push and the operator routine */
DelEntry (D, D->OpIndex);
DelEntry (D, D->PushIndex);
}
static int IsRegVar (StackOpData* D)
-/* If the value pushed is that of a register variable, replace ZPLo and ZPHi
- * in the given StackOpData struct by the register variables and return true.
- * Otherwise leave D untouched and return false.
+/* If the value pushed is that of a zeropage variable, replace ZPLo and ZPHi
+ * in the given StackOpData struct by the variable and return true. Otherwise
+ * leave D untouched and return false.
*/
{
- CodeEntry* P;
-
- if (D->PushIndex >= 2 &&
- (P = D->PrevEntry) != 0 &&
- P->OPC == OP65_LDX &&
- P->AM == AM65_ZP &&
- strncmp (P->Arg, "regbank+", 7) == 0 &&
- IsDigit (P->Arg[8]) &&
- (P = CS_GetEntry (D->Code, D->PushIndex-2)) != 0 &&
- P->OPC == OP65_LDA &&
- P->AM == AM65_ZP &&
- strncmp (P->Arg, "regbank+", 7) == 0 &&
- IsDigit (P->Arg[8])) {
- /* Ok, it loads the register variable */
- D->ZPHi = D->PrevEntry->Arg;
- D->ZPLo = P->Arg;
- return 1;
- } else {
+ CodeEntry* LoadA = D->Lhs.A.LoadEntry;
+ CodeEntry* LoadX = D->Lhs.X.LoadEntry;
+ unsigned Len;
+
+ /* Must have both load insns */
+ if (LoadA == 0 || LoadX == 0) {
return 0;
}
+
+ /* Must be loads from zp */
+ if (LoadA->AM != AM65_ZP || LoadX->AM != AM65_ZP) {
+ return 0;
+ }
+
+ /* Must be the same zp loc with high byte in X */
+ Len = strlen (LoadA->Arg);
+ if (strncmp (LoadA->Arg, LoadX->Arg, Len) != 0 ||
+ strcmp (LoadX->Arg + Len, "+1") != 0) {
+ return 0;
+ }
+
+ /* Use the zero page location directly */
+ D->ZPLo = LoadA->Arg;
+ D->ZPHi = LoadX->Arg;
+ return 1;
}
*/
if (D->OpEntry->RI->In.RegA != 0) {
+ /* lda #$00 */
+ X = NewCodeEntry (OP65_LDA, AM65_IMM, "$00", 0, D->OpEntry->LI);
+ InsertEntry (D, X, D->OpIndex+1);
+
/* The value of A is known */
if (D->OpEntry->RI->In.RegA <= 0x81) {
/* Loop using the sign bit */
- X = NewCodeEntry (OP65_LDA, AM65_IMM, "$00", 0, D->OpEntry->LI);
- InsertEntry (D, X, D->OpIndex+1);
+ /* ldy #count-1 */
Arg = MakeHexArg (D->OpEntry->RI->In.RegA - 1);
X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, D->OpEntry->LI);
InsertEntry (D, X, D->OpIndex+2);
+ /* L: sta (zp),y */
X = NewCodeEntry (OP65_STA, AM65_ZP_INDY, D->ZPLo, 0, D->OpEntry->LI);
InsertEntry (D, X, D->OpIndex+3);
L = CS_GenLabel (D->Code, X);
+ /* dey */
X = NewCodeEntry (OP65_DEY, AM65_IMP, 0, 0, D->OpEntry->LI);
InsertEntry (D, X, D->OpIndex+4);
+ /* bpl L */
X = NewCodeEntry (OP65_BPL, AM65_BRA, L->Name, L, D->OpEntry->LI);
InsertEntry (D, X, D->OpIndex+5);
} else {
/* Loop using an explicit compare */
- X = NewCodeEntry (OP65_LDA, AM65_IMM, "$00", 0, D->OpEntry->LI);
- InsertEntry (D, X, D->OpIndex+1);
+ /* ldy #$00 */
X = NewCodeEntry (OP65_LDY, AM65_IMM, "$00", 0, D->OpEntry->LI);
InsertEntry (D, X, D->OpIndex+2);
+ /* L: sta (zp),y */
X = NewCodeEntry (OP65_STA, AM65_ZP_INDY, D->ZPLo, 0, D->OpEntry->LI);
InsertEntry (D, X, D->OpIndex+3);
L = CS_GenLabel (D->Code, X);
+ /* iny */
X = NewCodeEntry (OP65_INY, AM65_IMP, 0, 0, D->OpEntry->LI);
InsertEntry (D, X, D->OpIndex+4);
+ /* cpy #count */
Arg = MakeHexArg (D->OpEntry->RI->In.RegA);
X = NewCodeEntry (OP65_CPY, AM65_IMM, Arg, 0, D->OpEntry->LI);
InsertEntry (D, X, D->OpIndex+5);
- X = NewCodeEntry (OP65_BPL, AM65_BRA, L->Name, L, D->OpEntry->LI);
+ /* bne L */
+ X = NewCodeEntry (OP65_BNE, AM65_BRA, L->Name, L, D->OpEntry->LI);
InsertEntry (D, X, D->OpIndex+6);
}
}
/* Remove the push and the call to the __bzero function */
- RemovePushAndOp (D);
+ RemoveRemainders (D);
/* We changed the sequence */
return 1;
InsertEntry (D, X, D->OpIndex+1);
/* Remove the push and the call to the staspidx function */
- RemovePushAndOp (D);
+ RemoveRemainders (D);
/* We changed the sequence */
return 1;
}
/* Inline the store */
+
+ /* sta (zp),y */
X = NewCodeEntry (OP65_STA, AM65_ZP_INDY, D->ZPLo, 0, D->OpEntry->LI);
InsertEntry (D, X, D->OpIndex+1);
+
if (RegValIsKnown (D->OpEntry->RI->In.RegY)) {
/* Value of Y is known */
- const char* Arg = MakeHexArg (D->OpEntry->RI->In.RegY + 1);
+ const char* Arg = MakeHexArg (D->OpEntry->RI->In.RegY + 1);
X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, D->OpEntry->LI);
} else {
X = NewCodeEntry (OP65_INY, AM65_IMP, 0, 0, D->OpEntry->LI);
}
InsertEntry (D, X, D->OpIndex+2);
+
if (RegValIsKnown (D->OpEntry->RI->In.RegX)) {
- /* Value of X is known */
- const char* Arg = MakeHexArg (D->OpEntry->RI->In.RegX);
+ /* Value of X is known */
+ const char* Arg = MakeHexArg (D->OpEntry->RI->In.RegX);
X = NewCodeEntry (OP65_LDA, AM65_IMM, Arg, 0, D->OpEntry->LI);
} else {
- /* Value unknown */
- X = NewCodeEntry (OP65_TXA, AM65_IMP, 0, 0, D->OpEntry->LI);
+ /* Value unknown */
+ X = NewCodeEntry (OP65_TXA, AM65_IMP, 0, 0, D->OpEntry->LI);
}
InsertEntry (D, X, D->OpIndex+3);
+
+ /* sta (zp),y */
X = NewCodeEntry (OP65_STA, AM65_ZP_INDY, D->ZPLo, 0, D->OpEntry->LI);
InsertEntry (D, X, D->OpIndex+4);
- /* Remove the push and the call to the staspidx function */
- RemovePushAndOp (D);
+ /* If we remove staxspidx, we must restore the Y register to what the
+ * function would return.
+ */
+ X = NewCodeEntry (OP65_LDY, AM65_IMM, "$00", 0, D->OpEntry->LI);
+ InsertEntry (D, X, D->OpIndex+5);
+
+ /* Remove the push and the call to the staxspidx function */
+ RemoveRemainders (D);
/* We changed the sequence */
return 1;
/* Optimize the tosaddax sequence if possible */
{
CodeEntry* X;
-
+ CodeEntry* N;
/* We need the entry behind the add */
CHECK (D->NextEntry != 0);
- /* 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.
+ /* Check if the X register is known and zero when the add is done, and
+ * if the add is followed by
+ *
+ * ldy #$00
+ * jsr ldauidx ; or ldaidx
+ *
+ * If this is true, the addition does actually add an offset to a pointer
+ * before it is dereferenced. Since both subroutines take an offset in Y,
+ * we can pass the offset (instead of #$00) and remove the addition
+ * alltogether.
*/
- CheckDirectOp (D);
+ if (D->OpEntry->RI->In.RegX == 0 &&
+ D->NextEntry->OPC == OP65_LDY &&
+ CE_IsKnownImm (D->NextEntry, 0) &&
+ !CE_HasLabel (D->NextEntry) &&
+ (N = CS_GetNextEntry (D->Code, D->OpIndex + 1)) != 0 &&
+ (CE_IsCallTo (N, "ldauidx") ||
+ CE_IsCallTo (N, "ldaidx"))) {
- /* Store the value into the zeropage instead of pushing it */
- ReplacePushByStore (D);
+ int Signed = (strcmp (N->Arg, "ldaidx") == 0);
- /* Inline the add */
- D->IP = D->OpIndex+1;
- X = NewCodeEntry (OP65_CLC, AM65_IMP, 0, 0, D->OpEntry->LI);
- InsertEntry (D, X, D->IP++);
+ /* Store the value into the zeropage instead of pushing it */
+ ReplacePushByStore (D);
- /* Low byte */
- AddOpLow (D, OP65_ADC);
+ /* 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
+ * insn.
+ */
+ X = NewCodeEntry (OP65_TAY, AM65_IMP, 0, 0, D->NextEntry->LI);
+ DelEntry (D, D->OpIndex + 1);
+ InsertEntry (D, X, D->OpIndex + 1);
+
+ /* Replace the call to ldaidx/ldauidx. Since X is already zero, and
+ * the ptr is in the zero page location, we just need to load from
+ * the pointer, and fix X in case of ldaidx.
+ */
+ X = NewCodeEntry (OP65_LDA, AM65_ZP_INDY, D->ZPLo, 0, N->LI);
+ DelEntry (D, D->OpIndex + 2);
+ InsertEntry (D, X, D->OpIndex + 2);
+ if (Signed) {
+
+ CodeLabel* L;
+
+ /* Add sign extension - N is unused now */
+ N = CS_GetNextEntry (D->Code, D->OpIndex + 2);
+ CHECK (N != 0);
+ L = CS_GenLabel (D->Code, N);
+
+ X = NewCodeEntry (OP65_BPL, AM65_BRA, L->Name, L, X->LI);
+ InsertEntry (D, X, D->OpIndex + 3);
+
+ X = NewCodeEntry (OP65_DEX, AM65_IMP, 0, 0, X->LI);
+ InsertEntry (D, X, D->OpIndex + 4);
+ }
- /* 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);
- X = NewCodeEntry (OP65_BCC, AM65_BRA, L->Name, L, D->OpEntry->LI);
- InsertEntry (D, X, D->IP++);
- X = NewCodeEntry (OP65_INX, AM65_IMP, 0, 0, D->OpEntry->LI);
- InsertEntry (D, X, D->IP++);
- } else if (D->OpEntry->RI->In.RegX == 0) {
- /* The high byte is that of the first operand plus carry */
- CodeLabel* L;
- if (RegValIsKnown (D->PushEntry->RI->In.RegX)) {
- /* Value of first op high byte is known */
- 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);
- }
- InsertEntry (D, X, D->IP++);
- L = CS_GenLabel (D->Code, D->NextEntry);
- X = NewCodeEntry (OP65_BCC, AM65_BRA, L->Name, L, D->OpEntry->LI);
- InsertEntry (D, X, D->IP++);
- X = NewCodeEntry (OP65_INX, AM65_IMP, 0, 0, D->OpEntry->LI);
- InsertEntry (D, X, D->IP++);
} else {
- /* High byte is unknown */
- AddOpHigh (D, OP65_ADC);
+
+ /* Store the value into the zeropage instead of pushing it */
+ ReplacePushByStore (D);
+
+ /* Inline the add */
+ D->IP = D->OpIndex+1;
+
+ /* clc */
+ X = NewCodeEntry (OP65_CLC, AM65_IMP, 0, 0, D->OpEntry->LI);
+ InsertEntry (D, X, D->IP++);
+
+ /* Low byte */
+ AddOpLow (D, OP65_ADC);
+
+ /* 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);
+ X = NewCodeEntry (OP65_BCC, AM65_BRA, L->Name, L, D->OpEntry->LI);
+ InsertEntry (D, X, D->IP++);
+ X = NewCodeEntry (OP65_INX, AM65_IMP, 0, 0, D->OpEntry->LI);
+ InsertEntry (D, X, D->IP++);
+ } else if (D->OpEntry->RI->In.RegX == 0) {
+ /* The high byte is that of the first operand plus carry */
+ CodeLabel* L;
+ if (RegValIsKnown (D->PushEntry->RI->In.RegX)) {
+ /* Value of first op high byte is known */
+ 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);
+ }
+ InsertEntry (D, X, D->IP++);
+
+ /* bcc label */
+ L = CS_GenLabel (D->Code, D->NextEntry);
+ 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 {
+ /* High byte is unknown */
+ AddOpHigh (D, OP65_ADC);
+ }
}
/* Remove the push and the call to the tosaddax function */
- RemovePushAndOp (D);
+ RemoveRemainders (D);
/* We changed the sequence */
return 1;
{
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);
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++);
+ InsertEntry (D, X, D->IP++);
} else {
/* High byte is unknown */
AddOpHigh (D, OP65_AND);
}
/* Remove the push and the call to the tosandax function */
- RemovePushAndOp (D);
+ RemoveRemainders (D);
+
+ /* We changed the sequence */
+ return 1;
+}
+
+
+
+static unsigned Opt_tosgeax (StackOpData* D)
+/* Optimize the tosgeax sequence if possible. */
+{
+ 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 & 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) {
+
+ /* cmp ... */
+ CodeEntry* LoadA = D->Rhs.A.LoadEntry;
+ X = NewCodeEntry (OP65_CMP, LoadA->AM, LoadA->Arg, 0, D->OpEntry->LI);
+ InsertEntry (D, X, D->IP++);
+
+ } else {
+
+ /* 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++);
+
+ /* cmp (sp),y */
+ X = NewCodeEntry (OP65_CMP, 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;
+
+ /* txa */
+ X = NewCodeEntry (OP65_TXA, AM65_IMP, 0, 0, D->OpEntry->LI);
+ InsertEntry (D, X, D->IP++);
+
+ /* Must be true because of OP_RHS_LOAD */
+ CHECK ((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) {
+
+ /* sbc ... */
+ CodeEntry* LoadX = D->Rhs.X.LoadEntry;
+ X = NewCodeEntry (OP65_SBC, LoadX->AM, LoadX->Arg, 0, D->OpEntry->LI);
+ InsertEntry (D, X, D->IP++);
+
+ } else {
+
+ /* 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++);
+
+ /* 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;
+
+ /* 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 if possible. */
+{
+ 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 & 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) {
+
+ /* cmp ... */
+ CodeEntry* LoadA = D->Rhs.A.LoadEntry;
+ X = NewCodeEntry (OP65_CMP, LoadA->AM, LoadA->Arg, 0, D->OpEntry->LI);
+ InsertEntry (D, X, D->IP++);
+
+ } else {
+
+ /* 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++);
+
+ /* cmp (sp),y */
+ X = NewCodeEntry (OP65_CMP, 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;
+
+ /* txa */
+ X = NewCodeEntry (OP65_TXA, AM65_IMP, 0, 0, D->OpEntry->LI);
+ InsertEntry (D, X, D->IP++);
+
+ /* Must be true because of OP_RHS_LOAD */
+ CHECK ((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) {
+
+ /* sbc ... */
+ CodeEntry* LoadX = D->Rhs.X.LoadEntry;
+ X = NewCodeEntry (OP65_SBC, LoadX->AM, LoadX->Arg, 0, D->OpEntry->LI);
+ InsertEntry (D, X, D->IP++);
+
+ } else {
+
+ /* 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++);
+
+ /* 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;
+
+ /* 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;
{
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);
/* High byte */
if (RegValIsKnown (D->PushEntry->RI->In.RegX) &&
RegValIsKnown (D->OpEntry->RI->In.RegX)) {
- /* Both values known, precalculate the result */
+ /* 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 */
+ /* High byte is unknown */
AddOpHigh (D, OP65_ORA);
}
/* Remove the push and the call to the tosorax function */
- RemovePushAndOp (D);
+ RemoveRemainders (D);
+
+ /* We changed the sequence */
+ return 1;
+}
+
+
+
+static unsigned Opt_tossubax (StackOpData* D)
+/* Optimize the tossubax sequence if possible. Note: subtraction is not
+ * commutative!
+ */
+{
+ CodeEntry* X;
+
+
+ /* Inline the sbc */
+ D->IP = D->OpIndex+1;
+
+ /* sec */
+ X = NewCodeEntry (OP65_SEC, AM65_IMP, 0, 0, D->OpEntry->LI);
+ InsertEntry (D, X, D->IP++);
+
+ /* Must be true because of OP_RHS_LOAD */
+ CHECK ((D->Rhs.A.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) {
+
+ /* sbc ... */
+ CodeEntry* LoadA = D->Rhs.A.LoadEntry;
+ X = NewCodeEntry (OP65_SBC, LoadA->AM, LoadA->Arg, 0, D->OpEntry->LI);
+ InsertEntry (D, X, D->IP++);
+
+ } else {
+
+ /* 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++);
+
+ /* 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;
+
+ /* 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++);
+
+ /* Must be true because of OP_RHS_LOAD */
+ CHECK ((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) {
+
+ /* sbc ... */
+ CodeEntry* LoadX = D->Rhs.X.LoadEntry;
+ X = NewCodeEntry (OP65_SBC, LoadX->AM, LoadX->Arg, 0, D->OpEntry->LI);
+ InsertEntry (D, X, D->IP++);
+
+ } else {
+
+ /* 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++);
+
+ /* 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;
+
+ /* 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++);
+
+ /* Remove the push and the call to the tossubax function */
+ RemoveRemainders (D);
+
+ /* We changed the sequence */
+ return 1;
+}
+
+
+
+static unsigned Opt_tosugeax (StackOpData* D)
+/* Optimize the tosugeax sequence if possible. */
+{
+ CodeEntry* X;
+
+
+ /* Inline the sbc */
+ D->IP = D->OpIndex+1;
+
+ /* Must be true because of OP_RHS_LOAD */
+ CHECK ((D->Rhs.A.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) {
+
+ /* cmp ... */
+ CodeEntry* LoadA = D->Rhs.A.LoadEntry;
+ X = NewCodeEntry (OP65_CMP, LoadA->AM, LoadA->Arg, 0, D->OpEntry->LI);
+ InsertEntry (D, X, D->IP++);
+
+ } else {
+
+ /* 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++);
+
+ /* cmp (sp),y */
+ X = NewCodeEntry (OP65_CMP, 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;
+
+ /* txa */
+ X = NewCodeEntry (OP65_TXA, AM65_IMP, 0, 0, D->OpEntry->LI);
+ InsertEntry (D, X, D->IP++);
+
+ /* Must be true because of OP_RHS_LOAD */
+ CHECK ((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) {
+
+ /* sbc ... */
+ CodeEntry* LoadX = D->Rhs.X.LoadEntry;
+ X = NewCodeEntry (OP65_SBC, LoadX->AM, LoadX->Arg, 0, D->OpEntry->LI);
+ InsertEntry (D, X, D->IP++);
+
+ } else {
+
+ /* 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++);
+
+ /* 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;
+
+ /* 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 tosugeax function */
+ RemoveRemainders (D);
/* We changed the sequence */
return 1;
{
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);
}
/* Remove the push and the call to the tosandax function */
- RemovePushAndOp (D);
+ RemoveRemainders (D);
/* We changed the sequence */
return 1;
/*****************************************************************************/
-/* Code */
+/* Code */
/*****************************************************************************/
-/* Flags for the functions */
-typedef enum {
- STOP_NONE = 0x00, /* Nothing special */
- STOP_A_UNUSED = 0x01, /* Call only if a unused later */
- STOP_A_KNOWN = 0x02, /* Call only if A is known */
- STOP_X_ZERO = 0x04 /* Call only if X is zero */
-} STOP_FLAGS;
-
-
-typedef unsigned (*OptFunc) (StackOpData* D);
-typedef struct OptFuncDesc OptFuncDesc;
-struct OptFuncDesc {
- const char* Name; /* Name of the replaced runtime function */
- OptFunc Func; /* Function pointer */
- STOP_FLAGS Flags; /* Flags */
-};
-
static const OptFuncDesc FuncTable[] = {
- { "__bzero", Opt___bzero, STOP_X_ZERO },
- { "staspidx", Opt_staspidx, STOP_NONE },
- { "staxspidx", Opt_staxspidx, STOP_A_UNUSED },
- { "tosaddax", Opt_tosaddax, STOP_NONE },
- { "tosandax", Opt_tosandax, STOP_NONE },
- { "tosorax", Opt_tosorax, STOP_NONE },
- { "tosxorax", Opt_tosxorax, STOP_NONE },
+ { "__bzero", Opt___bzero, REG_NONE, OP_X_ZERO | OP_A_KNOWN },
+ { "staspidx", Opt_staspidx, REG_NONE, OP_NONE },
+ { "staxspidx", Opt_staxspidx, REG_AX, OP_NONE },
+ { "tosaddax", Opt_tosaddax, REG_NONE, OP_NONE },
+ { "tosandax", Opt_tosandax, REG_NONE, OP_NONE },
+ { "tosgeax", Opt_tosgeax, REG_NONE, OP_RHS_LOAD_DIRECT },
+ { "tosltax", Opt_tosltax, REG_NONE, OP_RHS_LOAD_DIRECT },
+ { "tosorax", Opt_tosorax, REG_NONE, OP_NONE },
+ { "tossubax", Opt_tossubax, REG_NONE, OP_RHS_LOAD_DIRECT },
+ { "tosugeax", Opt_tosugeax, REG_NONE, OP_RHS_LOAD_DIRECT },
+ { "tosxorax", Opt_tosxorax, REG_NONE, OP_NONE },
};
#define FUNC_COUNT (sizeof(FuncTable) / sizeof(FuncTable[0]))
*/
{
static const char* Tab[] = {
+ "aslax1",
+ "aslax2",
+ "asrax1",
+ "asrax2",
+ "bnegax",
"ldaxidx",
"ldaxysp",
"negax",
+ "shlax1",
+ "shlax2",
+ "shrax1",
+ "shrax2",
};
void* R = bsearch (Name,
+static void ResetStackOpData (StackOpData* Data)
+/* Reset the given data structure */
+{
+ Data->OptFunc = 0;
+ Data->UsedRegs = REG_NONE;
+
+ ClearLoadInfo (&Data->Lhs);
+ ClearLoadInfo (&Data->Rhs);
+
+ Data->PushIndex = -1;
+ Data->OpIndex = -1;
+}
+
+
+
+static int PreCondOk (StackOpData* D)
+/* Check if the preconditions for a call to the optimizer subfunction are
+ * satisfied. As a side effect, this function will also choose the zero page
+ * register to use.
+ */
+{
+ /* Check the flags */
+ unsigned UnusedRegs = D->OptFunc->UnusedRegs;
+ if (UnusedRegs != REG_NONE &&
+ (GetRegInfo (D->Code, D->OpIndex+1, UnusedRegs) & UnusedRegs) != 0) {
+ /* Cannot optimize */
+ return 0;
+ }
+ if ((D->OptFunc->Flags & OP_A_KNOWN) != 0 &&
+ RegValIsUnknown (D->OpEntry->RI->In.RegA)) {
+ /* Cannot optimize */
+ return 0;
+ }
+ if ((D->OptFunc->Flags & OP_X_ZERO) != 0 &&
+ D->OpEntry->RI->In.RegX != 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;
+ }
+ }
+ }
+
+ /* Determine the zero page locations to use */
+ if ((D->UsedRegs & REG_PTR1) == REG_NONE) {
+ D->ZPLo = "ptr1";
+ D->ZPHi = "ptr1+1";
+ } else if ((D->UsedRegs & REG_SREG) == REG_NONE) {
+ D->ZPLo = "sreg";
+ D->ZPHi = "sreg+1";
+ } else if ((D->UsedRegs & REG_PTR2) == REG_NONE) {
+ D->ZPLo = "ptr2";
+ D->ZPHi = "ptr2+1";
+ } else {
+ /* No registers available */
+ return 0;
+ }
+
+ /* Determine if we have a basic block */
+ return CS_IsBasicBlock (D->Code, D->PushIndex, D->OpIndex);
+}
+
+
+
/*****************************************************************************/
-/* Code */
+/* Code */
/*****************************************************************************/
unsigned OptStackOps (CodeSeg* S)
/* Optimize operations that take operands via the stack */
{
- unsigned Changes = 0; /* Number of changes in one run */
- int InSeq = 0; /* Inside a sequence */
- unsigned Push = 0; /* Index of pushax */
- unsigned UsedRegs = 0; /* Zeropage registers used in sequence */
- unsigned I;
+ unsigned Changes = 0; /* Number of changes in one run */
+ StackOpData Data;
+ unsigned I;
+
+ enum {
+ Initialize,
+ Search,
+ FoundPush,
+ FoundOp
+ } State = Initialize;
/* Generate register info */
CS_GenRegInfo (S);
+ /* Remember the code segment in the info struct */
+ Data.Code = S;
+
/* Look for a call to pushax followed by a call to some other function
* that takes it's first argument on the stack, and the second argument
* in the primary register.
I = 0;
while (I < CS_GetEntryCount (S)) {
- /* Get the next entry */
- CodeEntry* E = CS_GetEntry (S, I);
+ /* Get the next entry */
+ CodeEntry* E = CS_GetEntry (S, I);
- /* Handling depends if we're inside a sequence or not */
- if (InSeq) {
+ /* Actions depend on state */
+ switch (State) {
- /* If we are using the stack, and we don't have "indirect Y"
- * addressing mode, or the value of Y is unknown, or less than
- * two, we cannot cope with this piece of code. Having an unknown
- * value of Y means that we cannot correct the stack offset, while
- * having an offset less than two means that the code works with
- * the value on stack which is to be removed.
- */
- if ((E->Use & REG_SP) != 0 &&
- (E->AM != AM65_ZP_INDY || RegValIsUnknown (E->RI->In.RegY) ||
- E->RI->In.RegY < 2)) {
-
- /* All this stuff is not allowed in a sequence */
- InSeq = 0;
-
- } else if (E->OPC == OP65_JSR) {
-
- /* Subroutine call: Check if this is one of our functions */
- const OptFuncDesc* F = FindFunc (E->Arg);
- if (F) {
-
- StackOpData Data;
- int PreCondOk = 1;
-
- /* Check the flags */
- if ((F->Flags & STOP_A_UNUSED) != 0 && RegAUsed (S, I+1)) {
- /* Cannot optimize */
- PreCondOk = 0;
- } else if ((F->Flags & STOP_A_KNOWN) != 0 && RegValIsUnknown (E->RI->In.RegA)) {
- /* Cannot optimize */
- PreCondOk = 0;
- } else if ((F->Flags & STOP_X_ZERO) != 0 && E->RI->In.RegX != 0) {
- /* Cannot optimize */
- PreCondOk = 0;
- }
-
- /* Determine the zero page locations to use */
- if (PreCondOk) {
- UsedRegs |= GetRegInfo (S, I+1, REG_SREG | REG_PTR1 | REG_PTR2);
- if ((UsedRegs & REG_SREG) == REG_NONE) {
- /* SREG is available */
- Data.ZPLo = "sreg";
- Data.ZPHi = "sreg+1";
- } else if ((UsedRegs & REG_PTR1) == REG_NONE) {
- Data.ZPLo = "ptr1";
- Data.ZPHi = "ptr1+1";
- } else if ((UsedRegs & REG_PTR2) == REG_NONE) {
- Data.ZPLo = "ptr2";
- Data.ZPHi = "ptr2+1";
- } else {
- /* No registers available */
- PreCondOk = 0;
- }
- }
-
- /* Determine if we have a basic block */
- if (PreCondOk) {
- PreCondOk = CS_IsBasicBlock (S, Push, I);
- }
+ case Initialize:
+ ResetStackOpData (&Data);
+ State = Search;
+ /* FALLTHROUGH */
- /* If preconditions are ok, call the optimizer function */
- if (PreCondOk) {
-
- /* Adjust stack offsets */
- Data.OpIndex = I + AdjustStackOffset (S, Push, I, 2);
-
- /* Prepare the remainder of the data structure */
- Data.Code = S;
- Data.Flags = 0;
- Data.PushIndex = Push;
- Data.PrevEntry = CS_GetPrevEntry (S, Data.PushIndex);
- Data.PushEntry = CS_GetEntry (S, Data.PushIndex);
- Data.OpEntry = E;
- Data.NextEntry = CS_GetNextEntry (S, Data.OpIndex);
-
- /* Call the optimizer function */
- Changes += F->Func (&Data);
-
- /* Regenerate register info */
- CS_GenRegInfo (S);
- }
-
- /* End of sequence */
- InSeq = 0;
-
- } else if (strcmp (E->Arg, "pushax") == 0) {
- /* Restart the sequence */
- Push = I;
- UsedRegs = REG_NONE;
- } else if (HarmlessCall (E->Arg)) {
- /* Track zeropage register usage */
- UsedRegs |= (E->Use | E->Chg);
+ case Search:
+ /* While searching, track register load insns, so we can tell
+ * what is in a register once pushax is encountered.
+ */
+ if (CE_IsCallTo (E, "pushax")) {
+ Data.PushIndex = I;
+ State = FoundPush;
} else {
- /* A call to an unkown subroutine ends the sequence */
- InSeq = 0;
- }
-
- } else {
- /* Other stuff: Track zeropage register usage */
- UsedRegs |= (E->Use | E->Chg);
- }
+ /* Track load insns */
+ TrackLoads (&Data.Lhs, E, I);
+ }
+ break;
+
+ case FoundPush:
+ /* We' found a pushax before. Search for a stack op that may
+ * follow and in the meantime, track zeropage usage and check
+ * for code that will disable us from translating the sequence.
+ */
+ if (E->OPC == OP65_JSR) {
+
+ /* Subroutine call: Check if this is one of the functions,
+ * we're going to replace.
+ */
+ Data.OptFunc = FindFunc (E->Arg);
+ if (Data.OptFunc) {
+ /* Remember the op index and go on */
+ Data.OpIndex = I;
+ Data.OpEntry = E;
+ State = FoundOp;
+ break;
+ } else if (!HarmlessCall (E->Arg)) {
+ /* A call to an unkown subroutine: We need to start
+ * over after the last pushax. Note: This will also
+ * happen if we encounter a call to pushax!
+ */
+ I = Data.PushIndex;
+ State = Initialize;
+ break;
+ } else {
+ /* Track register usage */
+ Data.UsedRegs |= (E->Use | E->Chg);
+ TrackLoads (&Data.Rhs, E, I);
+ }
- } else if (CE_IsCallTo (E, "pushax")) {
+ } else if (E->Info & OF_STORE && (E->Chg & REG_ZP) == 0) {
+
+ /* Too dangerous - there may be a change of a variable
+ * within the sequence.
+ */
+ I = Data.PushIndex;
+ State = Initialize;
+ break;
+
+ } else if ((E->Use & REG_SP) != 0 &&
+ (E->AM != AM65_ZP_INDY ||
+ RegValIsUnknown (E->RI->In.RegY) ||
+ E->RI->In.RegY < 2)) {
+
+ /* If we are using the stack, and we don't have "indirect Y"
+ * addressing mode, or the value of Y is unknown, or less
+ * than two, we cannot cope with this piece of code. Having
+ * an unknown value of Y means that we cannot correct the
+ * stack offset, while having an offset less than two means
+ * that the code works with the value on stack which is to
+ * be removed.
+ */
+ I = Data.PushIndex;
+ State = Initialize;
+ break;
- /* This starts a sequence */
- Push = I;
- UsedRegs = REG_NONE;
- InSeq = 1;
+ } else {
+ /* Other stuff: Track register usage */
+ Data.UsedRegs |= (E->Use | E->Chg);
+ TrackLoads (&Data.Rhs, E, I);
+ }
+ break;
+
+ case FoundOp:
+ /* Track zero page location usage beyond this point */
+ Data.UsedRegs |= GetRegInfo (S, I, REG_SREG | REG_PTR1 | REG_PTR2);
+
+ /* Finalize the load info */
+ FinalizeLoadInfo (&Data.Lhs, S);
+ FinalizeLoadInfo (&Data.Rhs, S);
+
+ /* Set flags for direct operations */
+ CheckDirectOp (&Data);
+
+ /* If the Lhs loads do load from zeropage, we have to include
+ * them into UsedRegs registers used. The Rhs loads have already
+ * been tracked.
+ */
+ if (Data.Lhs.A.LoadEntry && Data.Lhs.A.LoadEntry->AM == AM65_ZP) {
+ Data.UsedRegs |= Data.Lhs.A.LoadEntry->Use;
+ }
+ if (Data.Lhs.X.LoadEntry && Data.Lhs.X.LoadEntry->AM == AM65_ZP) {
+ Data.UsedRegs |= Data.Lhs.X.LoadEntry->Use;
+ }
+
+ /* Check the preconditions. If they aren't ok, reset the insn
+ * pointer to the pushax and start over. We will loose part of
+ * load tracking but at least a/x has probably lost between
+ * pushax and here and will be tracked again when restarting.
+ */
+ if (!PreCondOk (&Data)) {
+ I = Data.PushIndex;
+ State = Initialize;
+ 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);
+
+ /* Call the optimizer function */
+ Changes += Data.OptFunc->Func (&Data);
+
+ /* Regenerate register info */
+ CS_GenRegInfo (S);
+
+ /* Done */
+ State = Initialize;
+ break;
}
projects / cc65 / blobdiff