/* */
/* */
/* */
-/* (C) 2001 Ullrich von Bassewitz */
+/* (C) 2001-2002 Ullrich von Bassewitz */
/* Wacholderweg 14 */
/* D-70597 Stuttgart */
/* EMail: uz@cc65.org */
+#include <stdlib.h>
#include <string.h>
/* common */
#include "abend.h"
+#include "chartype.h"
#include "print.h"
+#include "xmalloc.h"
+#include "xsprintf.h"
/* cc65 */
#include "asmlabel.h"
#include "codeent.h"
#include "codeinfo.h"
+#include "coptadd.h"
+#include "coptc02.h"
+#include "coptcmp.h"
#include "coptind.h"
+#include "coptneg.h"
+#include "coptpush.h"
+#include "coptstop.h"
+#include "coptsub.h"
+#include "copttest.h"
+#include "cpu.h"
#include "error.h"
#include "global.h"
#include "codeopt.h"
/*****************************************************************************/
-/* Data */
+/* Optimize shifts */
/*****************************************************************************/
-/* Defines for the conditions in a compare */
-typedef enum {
- CMP_INV = -1,
- CMP_EQ,
- CMP_NE,
- CMP_GT,
- CMP_GE,
- CMP_LT,
- CMP_LE,
- CMP_UGT,
- CMP_UGE,
- CMP_ULT,
- CMP_ULE
-} cmp_t;
-
-/* Table with the compare suffixes */
-static const char CmpSuffixTab [][4] = {
- "eq", "ne", "gt", "ge", "lt", "le", "ugt", "uge", "ult", "ule"
-};
-
-/* Table used to invert a condition, indexed by condition */
-static const unsigned char CmpInvertTab [] = {
- CMP_NE, CMP_EQ,
- CMP_LE, CMP_LT, CMP_GE, CMP_GT,
- CMP_ULE, CMP_ULT, CMP_UGE, CMP_UGT
-};
+static unsigned OptShift1 (CodeSeg* S)
+/* A call to the shlaxN routine may get replaced by one or more asl insns
+ * if the value of X is not used later.
+ */
+{
+ unsigned Changes = 0;
-/* Table to show which compares are signed (use the N flag) */
-static const char CmpSignedTab [] = {
- 0, 0, 1, 1, 1, 1, 0, 0, 0, 0
-};
+ /* Walk over the entries */
+ unsigned I = 0;
+ while (I < CS_GetEntryCount (S)) {
+ /* Get next entry */
+ CodeEntry* E = CS_GetEntry (S, I);
+ /* Check for the sequence */
+ if (E->OPC == OP65_JSR &&
+ (strncmp (E->Arg, "shlax", 5) == 0 ||
+ strncmp (E->Arg, "aslax", 5) == 0) &&
+ strlen (E->Arg) == 6 &&
+ IsDigit (E->Arg[5]) &&
+ !RegXUsed (S, I+1)) {
+
+ /* Insert shift insns */
+ unsigned Count = E->Arg[5] - '0';
+ while (Count--) {
+ CodeEntry* X = NewCodeEntry (OP65_ASL, AM65_ACC, "a", 0, E->LI);
+ CS_InsertEntry (S, X, I+1);
+ }
-/*****************************************************************************/
-/* Helper functions */
-/*****************************************************************************/
+ /* Delete the call to shlax */
+ CS_DelEntry (S, I);
+ /* Remember, we had changes */
+ ++Changes;
+ }
-static cmp_t FindCmpCond (const char* Code, unsigned CodeLen)
-/* Search for a compare condition by the given code using the given length */
-{
- unsigned I;
+ /* Next entry */
+ ++I;
- /* Linear search */
- for (I = 0; I < sizeof (CmpSuffixTab) / sizeof (CmpSuffixTab [0]); ++I) {
- if (strncmp (Code, CmpSuffixTab [I], CodeLen) == 0) {
- /* Found */
- return I;
- }
}
- /* Not found */
- return CMP_INV;
+ /* Return the number of changes made */
+ return Changes;
}
-static cmp_t FindBoolCmpCond (const char* Name)
-/* Map a condition suffix to a code. Return the code or CMP_INV on failure */
+static unsigned OptShift2 (CodeSeg* S)
+/* A call to the shraxN routine may get replaced by one or more lsr insns
+ * if the value of X is not used later.
+ */
{
- /* Check for the correct subroutine name */
- if (strncmp (Name, "bool", 4) == 0) {
- /* Name is ok, search for the code in the table */
- return FindCmpCond (Name+4, strlen(Name)-4);
- } else {
- /* Not found */
- return CMP_INV;
- }
-}
+ unsigned Changes = 0;
+
+ /* Walk over the entries */
+ unsigned I = 0;
+ while (I < CS_GetEntryCount (S)) {
+ /* Get next entry */
+ CodeEntry* E = CS_GetEntry (S, I);
+ /* Check for the sequence */
+ if (E->OPC == OP65_JSR &&
+ strncmp (E->Arg, "shrax", 5) == 0 &&
+ strlen (E->Arg) == 6 &&
+ IsDigit (E->Arg[5]) &&
+ !RegXUsed (S, I+1)) {
+
+ /* Insert shift insns */
+ unsigned Count = E->Arg[5] - '0';
+ while (Count--) {
+ CodeEntry* X = NewCodeEntry (OP65_LSR, AM65_ACC, "a", 0, E->LI);
+ CS_InsertEntry (S, X, I+1);
+ }
-static int FindTosCmpCond (const char* Name)
-/* Check if this is a call to one of the TOS compare functions (tosgtax).
- * Return the condition code or CMP_INV on failure.
- */
-{
- unsigned Len = strlen (Name);
+ /* Delete the call to shlax */
+ CS_DelEntry (S, I);
+
+ /* Remember, we had changes */
+ ++Changes;
+
+ }
+
+ /* Next entry */
+ ++I;
- /* Check for the correct subroutine name */
- if (strncmp (Name, "tos", 3) == 0 && strcmp (Name+Len-2, "ax") == 0) {
- /* Name is ok, search for the code in the table */
- return FindCmpCond (Name+3, Len-3-2);
- } else {
- /* Not found */
- return CMP_INV;
}
+
+ /* Return the number of changes made */
+ return Changes;
}
-static int IsUnsignedCmp (int Code)
-/* Check if this is an unsigned compare */
-{
- CHECK (Code >= 0);
- return CmpSignedTab [Code] == 0;
-}
+/*****************************************************************************/
+/* Optimize stores through pointers */
+/*****************************************************************************/
-static const char* MakeBoolCmp (cmp_t Cond)
-/* Create the name of a bool transformer subroutine for the given code. The
- * result is placed into a static buffer, so beware!
- */
+static unsigned OptPtrStore1Sub (CodeSeg* S, unsigned I, CodeEntry** const L)
+/* Check if this is one of the allowed suboperation for OptPtrStore1 */
{
- static char Buffer[20];
- CHECK (Cond != CMP_INV);
- sprintf (Buffer, "bool%s", CmpSuffixTab[Cond]);
- return Buffer;
-}
+ /* Check for a label attached to the entry */
+ if (CE_HasLabel (L[0])) {
+ return 0;
+ }
+ /* Check for single insn sub ops */
+ if (L[0]->OPC == OP65_AND ||
+ L[0]->OPC == OP65_EOR ||
+ L[0]->OPC == OP65_ORA ||
+ (L[0]->OPC == OP65_JSR && strncmp (L[0]->Arg, "shlax", 5) == 0) ||
+ (L[0]->OPC == OP65_JSR && strncmp (L[0]->Arg, "shrax", 5) == 0)) {
+
+ /* One insn */
+ return 1;
+
+ } else if (L[0]->OPC == OP65_CLC &&
+ (L[1] = CS_GetNextEntry (S, I)) != 0 &&
+ L[1]->OPC == OP65_ADC &&
+ !CE_HasLabel (L[1])) {
+ return 2;
+ } else if (L[0]->OPC == OP65_SEC &&
+ (L[1] = CS_GetNextEntry (S, I)) != 0 &&
+ L[1]->OPC == OP65_SBC &&
+ !CE_HasLabel (L[1])) {
+ return 2;
+ }
-static const char* MakeTosCmp (cmp_t Cond)
-/* Create the name of a tos compare subroutine for the given code. The
- * result is placed into a static buffer, so beware!
- */
-{
- static char Buffer[20];
- CHECK (Cond != CMP_INV);
- sprintf (Buffer, "tos%sax", CmpSuffixTab[Cond]);
- return Buffer;
+
+ /* Not found */
+ return 0;
}
-static int IsBitOp (const CodeEntry* E)
-/* Check if E is one of the bit operations (and, or, eor) */
+static unsigned OptPtrStore1 (CodeSeg* S)
+/* Search for the sequence:
+ *
+ * jsr pushax
+ * ldy xxx
+ * jsr ldauidx
+ * subop
+ * ldy yyy
+ * jsr staspidx
+ *
+ * and replace it by:
+ *
+ * sta ptr1
+ * stx ptr1+1
+ * ldy xxx
+ * ldx #$00
+ * lda (ptr1),y
+ * subop
+ * ldy yyy
+ * sta (ptr1),y
+ */
{
- return (E->OPC == OPC_AND || E->OPC == OPC_ORA || E->OPC == OPC_EOR);
-}
+ unsigned Changes = 0;
+ /* Walk over the entries */
+ unsigned I = 0;
+ while (I < CS_GetEntryCount (S)) {
+ unsigned K;
+ CodeEntry* L[10];
-static int IsCmpToZero (const CodeEntry* E)
-/* Check if the given instrcuction is a compare to zero instruction */
-{
- return (E->OPC == OPC_CMP &&
- E->AM == AM_IMM &&
- (E->Flags & CEF_NUMARG) != 0 &&
- E->Num == 0);
-}
+ /* Get next entry */
+ L[0] = CS_GetEntry (S, I);
+ /* Check for the sequence */
+ if (CE_IsCall (L[0], "pushax") &&
+ CS_GetEntries (S, L+1, I+1, 3) &&
+ L[1]->OPC == OP65_LDY &&
+ CE_KnownImm (L[1]) &&
+ !CE_HasLabel (L[1]) &&
+ CE_IsCall (L[2], "ldauidx") &&
+ !CE_HasLabel (L[2]) &&
+ (K = OptPtrStore1Sub (S, I+3, L+3)) > 0 &&
+ CS_GetEntries (S, L+3+K, I+3+K, 2) &&
+ L[3+K]->OPC == OP65_LDY &&
+ CE_KnownImm (L[3+K]) &&
+ !CE_HasLabel (L[3+K]) &&
+ CE_IsCall (L[4+K], "staspidx") &&
+ !CE_HasLabel (L[4+K])) {
+ CodeEntry* X;
-static int IsSpLoad (const CodeEntry* E)
-/* Return true if this is the load of A from the stack */
-{
- return E->OPC == OPC_LDA && E->AM == AM_ZP_INDY && strcmp (E->Arg, "sp") == 0;
-}
+ /* Create and insert the stores */
+ X = NewCodeEntry (OP65_STA, AM65_ZP, "ptr1", 0, L[0]->LI);
+ CS_InsertEntry (S, X, I+1);
+ X = NewCodeEntry (OP65_STX, AM65_ZP, "ptr1+1", 0, L[0]->LI);
+ CS_InsertEntry (S, X, I+2);
+ /* Delete the call to pushax */
+ CS_DelEntry (S, I);
-static int IsLocalLoad16 (CodeSeg* S, unsigned Index,
- CodeEntry** L, unsigned Count)
-/* Check if a 16 bit load of a local variable follows:
- *
- * ldy #$xx
- * lda (sp),y
- * tax
- * dey
- * lda (sp),y
- *
- * If so, read Count entries following the first ldy into L and return true
- * if this is possible. Otherwise return false.
- */
-{
- /* Be sure we read enough entries for the check */
- CHECK (Count >= 5);
-
- /* Read the first entry */
- L[0] = GetCodeEntry (S, Index);
-
- /* Check for the sequence */
- return (L[0]->OPC == OPC_LDY &&
- L[0]->AM == AM_IMM &&
- (L[0]->Flags & CEF_NUMARG) != 0 &&
- GetCodeEntries (S, L+1, Index+1, Count-1) &&
- IsSpLoad (L[1]) &&
- !CodeEntryHasLabel (L[1]) &&
- L[2]->OPC == OPC_TAX &&
- !CodeEntryHasLabel (L[2]) &&
- L[3]->OPC == OPC_DEY &&
- !CodeEntryHasLabel (L[3]) &&
- IsSpLoad (L[4]) &&
- !CodeEntryHasLabel (L[4]));
-}
+ /* Delete the call to ldauidx */
+ CS_DelEntry (S, I+3);
+ /* Insert the load from ptr1 */
+ X = NewCodeEntry (OP65_LDX, AM65_IMM, "$00", 0, L[3]->LI);
+ CS_InsertEntry (S, X, I+3);
+ X = NewCodeEntry (OP65_LDA, AM65_ZP_INDY, "ptr1", 0, L[2]->LI);
+ CS_InsertEntry (S, X, I+4);
+ /* Insert the store through ptr1 */
+ X = NewCodeEntry (OP65_STA, AM65_ZP_INDY, "ptr1", 0, L[3]->LI);
+ CS_InsertEntry (S, X, I+6+K);
-static int IsImmCmp16 (CodeSeg* S, CodeEntry** L)
-/* Check if the instructions at L are an immidiate compare of a/x:
- *
- *
- */
-{
- return (L[0]->OPC == OPC_CPX &&
- L[0]->AM == AM_IMM &&
- (L[0]->Flags & CEF_NUMARG) != 0 &&
- !CodeEntryHasLabel (L[0]) &&
- (L[1]->OPC == OPC_JNE || L[1]->OPC == OPC_BNE) &&
- L[1]->JumpTo != 0 &&
- !CodeEntryHasLabel (L[1]) &&
- L[2]->OPC == OPC_CMP &&
- L[2]->AM == AM_IMM &&
- (L[2]->Flags & CEF_NUMARG) != 0 &&
- (L[3]->Info & OF_ZBRA) != 0 &&
- L[3]->JumpTo != 0 &&
- (L[1]->JumpTo->Owner == L[3] || L[1]->JumpTo == L[3]->JumpTo));
-}
+ /* Delete the call to staspidx */
+ CS_DelEntry (S, I+7+K);
+ /* Remember, we had changes */
+ ++Changes;
+ }
-/*****************************************************************************/
-/* Remove calls to the bool transformer subroutines */
-/*****************************************************************************/
+ /* Next entry */
+ ++I;
+ }
+
+ /* Return the number of changes made */
+ return Changes;
+}
-static unsigned OptBoolTransforms (CodeSeg* S)
-/* Try to remove the call to boolean transformer routines where the call is
- * not really needed.
+
+static unsigned OptPtrStore2 (CodeSeg* S)
+/* Search for the sequence:
+ *
+ * jsr pushax
+ * lda xxx
+ * ldy yyy
+ * jsr staspidx
+ *
+ * and replace it by:
+ *
+ * sta ptr1
+ * stx ptr1+1
+ * lda xxx
+ * ldy yyy
+ * sta (ptr1),y
*/
{
unsigned Changes = 0;
/* Walk over the entries */
unsigned I = 0;
- while (I < GetCodeEntryCount (S)) {
+ while (I < CS_GetEntryCount (S)) {
- CodeEntry* N;
- cmp_t Cond;
+ CodeEntry* L[4];
/* Get next entry */
- CodeEntry* E = GetCodeEntry (S, I);
+ L[0] = CS_GetEntry (S, I);
- /* Check for a boolean transformer */
- if (E->OPC == OPC_JSR &&
- (Cond = FindBoolCmpCond (E->Arg)) != CMP_INV &&
- (N = GetNextCodeEntry (S, I)) != 0 &&
- (N->Info & OF_ZBRA) != 0) {
+ /* Check for the sequence */
+ if (CE_IsCall (L[0], "pushax") &&
+ CS_GetEntries (S, L+1, I+1, 3) &&
+ L[1]->OPC == OP65_LDA &&
+ !CE_HasLabel (L[1]) &&
+ L[2]->OPC == OP65_LDY &&
+ !CE_HasLabel (L[2]) &&
+ CE_IsCall (L[3], "staspidx") &&
+ !CE_HasLabel (L[3])) {
CodeEntry* X;
- CodeLabel* L;
- /* Make the boolean transformer unnecessary by changing the
- * the conditional jump to evaluate the condition flags that
- * are set after the compare directly. Note: jeq jumps if
- * the condition is not met, jne jumps if the condition is met.
- * Invert the code if we jump on condition not met.
- */
- if (GetBranchCond (N->OPC) == BC_EQ) {
- /* Jumps if condition false, invert condition */
- Cond = CmpInvertTab [Cond];
- }
-
- /* Check if we can replace the code by something better */
- switch (Cond) {
-
- case CMP_EQ:
- ReplaceOPC (N, OPC_JEQ);
- break;
-
- case CMP_NE:
- ReplaceOPC (N, OPC_JNE);
- break;
-
- case CMP_GT:
- /* Replace by
- * beq @L
- * jpl Target
- * @L: ...
- */
- if ((X = GetNextCodeEntry (S, I+1)) == 0) {
- /* No such entry */
- goto NextEntry;
- }
- L = GenCodeLabel (S, X);
- X = NewCodeEntry (OPC_BEQ, AM_BRA, L->Name, L);
- InsertCodeEntry (S, X, I+1);
- ReplaceOPC (N, OPC_JPL);
- break;
-
- case CMP_GE:
- ReplaceOPC (N, OPC_JPL);
- break;
-
- case CMP_LT:
- ReplaceOPC (N, OPC_JMI);
- break;
-
- case CMP_LE:
- /* Replace by
- * jmi Target
- * jeq Target
- */
- ReplaceOPC (N, OPC_JMI);
- L = N->JumpTo;
- X = NewCodeEntry (OPC_JEQ, AM_BRA, L->Name, L);
- InsertCodeEntry (S, X, I+2);
- break;
-
- case CMP_UGT:
- /* Replace by
- * beq @L
- * jcs Target
- * @L: ...
- */
- if ((X = GetNextCodeEntry (S, I+1)) == 0) {
- /* No such entry */
- goto NextEntry;
- }
- L = GenCodeLabel (S, X);
- X = NewCodeEntry (OPC_BEQ, AM_BRA, L->Name, L);
- InsertCodeEntry (S, X, I+1);
- ReplaceOPC (N, OPC_JCS);
- break;
-
- case CMP_UGE:
- ReplaceOPC (N, OPC_JCS);
- break;
-
- case CMP_ULT:
- ReplaceOPC (N, OPC_JCC);
- break;
-
- case CMP_ULE:
- /* Replace by
- * jcc Target
- * jeq Target
- */
- ReplaceOPC (N, OPC_JCC);
- L = N->JumpTo;
- X = NewCodeEntry (OPC_JEQ, AM_BRA, L->Name, L);
- InsertCodeEntry (S, X, I+2);
- break;
-
- default:
- Internal ("Unknown jump condition: %d", Cond);
+ /* Create and insert the stores */
+ X = NewCodeEntry (OP65_STA, AM65_ZP, "ptr1", 0, L[0]->LI);
+ CS_InsertEntry (S, X, I+1);
- }
+ X = NewCodeEntry (OP65_STX, AM65_ZP, "ptr1+1", 0, L[0]->LI);
+ CS_InsertEntry (S, X, I+2);
+
+ /* Delete the call to pushax */
+ CS_DelEntry (S, I);
- /* Remove the call to the bool transformer */
- DelCodeEntry (S, I);
+ /* Insert the store through ptr1 */
+ X = NewCodeEntry (OP65_STA, AM65_ZP_INDY, "ptr1", 0, L[3]->LI);
+ CS_InsertEntry (S, X, I+4);
+
+ /* Delete the call to staspidx */
+ CS_DelEntry (S, I+5);
/* Remember, we had changes */
++Changes;
}
-NextEntry:
/* Next entry */
++I;
/*****************************************************************************/
-/* Optimize subtractions */
+/* Optimize loads through pointers */
/*****************************************************************************/
-static unsigned OptSub1 (CodeSeg* S)
-/* Search for the sequence
+static unsigned OptPtrLoad1 (CodeSeg* S)
+/* Search for the sequence:
*
- * sbc ...
- * bcs L
- * dex
- * L:
+ * tax
+ * dey
+ * lda (sp),y # May be any destination
+ * ldy ...
+ * jsr ldauidx
+ *
+ * and replace it by:
*
- * and remove the handling of the high byte if X is not used later.
+ * sta ptr1+1
+ * dey
+ * lda (sp),y
+ * sta ptr1
+ * ldy ...
+ * ldx #$00
+ * lda (ptr1),y
*/
{
unsigned Changes = 0;
/* Walk over the entries */
unsigned I = 0;
- while (I < GetCodeEntryCount (S)) {
+ while (I < CS_GetEntryCount (S)) {
- CodeEntry* L[3];
+ CodeEntry* L[5];
/* Get next entry */
- CodeEntry* E = GetCodeEntry (S, I);
+ L[0] = CS_GetEntry (S, I);
/* Check for the sequence */
- if (E->OPC == OPC_SBC &&
- GetCodeEntries (S, L, I+1, 3) &&
- (L[0]->OPC == OPC_BCS || L[0]->OPC == OPC_JCS) &&
- L[0]->JumpTo != 0 &&
- !CodeEntryHasLabel (L[0]) &&
- L[1]->OPC == OPC_DEX &&
- !CodeEntryHasLabel (L[1]) &&
- L[0]->JumpTo->Owner == L[2] &&
- !RegXUsed (S, I+3)) {
-
- /* Remove the bcs/dex */
- DelCodeEntries (S, I+1, 2);
+ if (L[0]->OPC == OP65_TAX &&
+ CS_GetEntries (S, L+1, I+1, 4) &&
+ L[1]->OPC == OP65_DEY &&
+ !CE_HasLabel (L[1]) &&
+ L[2]->OPC == OP65_LDA &&
+ !CE_HasLabel (L[2]) &&
+ L[3]->OPC == OP65_LDY &&
+ !CE_HasLabel (L[3]) &&
+ CE_IsCall (L[4], "ldauidx") &&
+ !CE_HasLabel (L[4])) {
+
+ CodeEntry* X;
+
+ /* Store the high byte and remove the TAX instead */
+ X = NewCodeEntry (OP65_STA, AM65_ZP, "ptr1+1", 0, L[0]->LI);
+ CS_InsertEntry (S, X, I+1);
+ CS_DelEntry (S, I);
+
+ /* Store the low byte */
+ X = NewCodeEntry (OP65_STA, AM65_ZP, "ptr1", 0, L[2]->LI);
+ CS_InsertEntry (S, X, I+3);
+
+ /* Delete the call to ldauidx */
+ CS_DelEntry (S, I+5);
+
+ /* Load high and low byte */
+ X = NewCodeEntry (OP65_LDX, AM65_IMM, "$00", 0, L[3]->LI);
+ CS_InsertEntry (S, X, I+5);
+ X = NewCodeEntry (OP65_LDA, AM65_ZP_INDY, "ptr1", 0, L[3]->LI);
+ CS_InsertEntry (S, X, I+6);
/* Remember, we had changes */
++Changes;
-static unsigned OptSub2 (CodeSeg* S)
-/* Search for the sequence
+static unsigned OptPtrLoad2 (CodeSeg* S)
+/* Search for the sequence:
*
- * lda xx
- * sec
- * sta tmp1
- * lda yy
- * sbc tmp1
- * sta yy
+ * clc
+ * adc xxx
+ * tay
+ * txa
+ * adc yyy
+ * tax
+ * tya
+ * ldy
+ * jsr ldauidx
*
- * and replace it by
+ * and replace it by:
*
- * sec
- * lda yy
- * sbc xx
- * sta yy
+ * clc
+ * adc xxx
+ * sta ptr1
+ * txa
+ * adc yyy
+ * sta ptr1+1
+ * ldy
+ * ldx #$00
+ * lda (ptr1),y
*/
{
unsigned Changes = 0;
/* Walk over the entries */
unsigned I = 0;
- while (I < GetCodeEntryCount (S)) {
+ while (I < CS_GetEntryCount (S)) {
- CodeEntry* L[5];
+ CodeEntry* L[9];
/* Get next entry */
- CodeEntry* E = GetCodeEntry (S, I);
+ L[0] = CS_GetEntry (S, I);
/* Check for the sequence */
- if (E->OPC == OPC_LDA &&
- GetCodeEntries (S, L, I+1, 5) &&
- L[0]->OPC == OPC_SEC &&
- !CodeEntryHasLabel (L[0]) &&
- L[1]->OPC == OPC_STA &&
- strcmp (L[1]->Arg, "tmp1") == 0 &&
- !CodeEntryHasLabel (L[1]) &&
- L[2]->OPC == OPC_LDA &&
- !CodeEntryHasLabel (L[2]) &&
- L[3]->OPC == OPC_SBC &&
- strcmp (L[3]->Arg, "tmp1") == 0 &&
- !CodeEntryHasLabel (L[3]) &&
- L[4]->OPC == OPC_STA &&
- strcmp (L[4]->Arg, L[2]->Arg) == 0 &&
- !CodeEntryHasLabel (L[4])) {
-
- /* Remove the store to tmp1 */
- DelCodeEntry (S, I+2);
-
- /* Remove the subtraction */
- DelCodeEntry (S, I+3);
-
- /* Move the lda to the position of the subtraction and change the
- * op to SBC.
+ if (L[0]->OPC == OP65_CLC &&
+ CS_GetEntries (S, L+1, I+1, 8) &&
+ L[1]->OPC == OP65_ADC &&
+ !CE_HasLabel (L[1]) &&
+ L[2]->OPC == OP65_TAY &&
+ !CE_HasLabel (L[2]) &&
+ L[3]->OPC == OP65_TXA &&
+ !CE_HasLabel (L[3]) &&
+ L[4]->OPC == OP65_ADC &&
+ !CE_HasLabel (L[4]) &&
+ L[5]->OPC == OP65_TAX &&
+ !CE_HasLabel (L[5]) &&
+ L[6]->OPC == OP65_TYA &&
+ !CE_HasLabel (L[6]) &&
+ L[7]->OPC == OP65_LDY &&
+ !CE_HasLabel (L[7]) &&
+ CE_IsCall (L[8], "ldauidx") &&
+ !CE_HasLabel (L[8])) {
+
+ CodeEntry* X;
+ CodeEntry* P;
+
+ /* Store the low byte and remove the TAY instead */
+ X = NewCodeEntry (OP65_STA, AM65_ZP, "ptr1", 0, L[1]->LI);
+ CS_InsertEntry (S, X, I+2);
+ CS_DelEntry (S, I+3);
+
+ /* Store the high byte */
+ X = NewCodeEntry (OP65_STA, AM65_ZP, "ptr1+1", 0, L[4]->LI);
+ CS_InsertEntry (S, X, I+5);
+
+ /* If the instruction before the adc is a ldx, replace the
+ * txa by and lda with the same location of the ldx.
*/
- MoveCodeEntry (S, I, I+3);
- ReplaceOPC (E, OPC_SBC);
+ if ((P = CS_GetPrevEntry (S, I)) != 0 &&
+ P->OPC == OP65_LDX &&
+ !CE_HasLabel (P)) {
+
+ X = NewCodeEntry (OP65_LDA, P->AM, P->Arg, 0, P->LI);
+ CS_InsertEntry (S, X, I+4);
+ CS_DelEntry (S, I+3);
+ }
+
+ /* Delete more transfer insns */
+ CS_DelEntry (S, I+7);
+ CS_DelEntry (S, I+6);
+
+ /* Delete the call to ldauidx */
+ CS_DelEntry (S, I+7);
+
+ /* Load high and low byte */
+ X = NewCodeEntry (OP65_LDX, AM65_IMM, "$00", 0, L[7]->LI);
+ CS_InsertEntry (S, X, I+7);
+ X = NewCodeEntry (OP65_LDA, AM65_ZP_INDY, "ptr1", 0, L[7]->LI);
+ CS_InsertEntry (S, X, I+8);
/* Remember, we had changes */
++Changes;
-/*****************************************************************************/
-/* Optimize additions */
-/*****************************************************************************/
-
-
-
-static unsigned OptAdd1 (CodeSeg* S)
-/* Search for the sequence
+static unsigned OptPtrLoad3 (CodeSeg* S)
+/* Search for the sequence:
*
- * adc ...
- * bcc L
- * inx
- * L:
+ * adc xxx
+ * pha
+ * txa
+ * iny
+ * adc yyy
+ * tax
+ * pla
+ * ldy
+ * jsr ldauidx
+ *
+ * and replace it by:
*
- * and remove the handling of the high byte if X is not used later.
+ * adc xxx
+ * sta ptr1
+ * txa
+ * iny
+ * adc yyy
+ * sta ptr1+1
+ * ldy
+ * ldx #$00
+ * lda (ptr1),y
*/
{
unsigned Changes = 0;
/* Walk over the entries */
unsigned I = 0;
- while (I < GetCodeEntryCount (S)) {
+ while (I < CS_GetEntryCount (S)) {
- CodeEntry* L[3];
+ CodeEntry* L[9];
/* Get next entry */
- CodeEntry* E = GetCodeEntry (S, I);
+ L[0] = CS_GetEntry (S, I);
/* Check for the sequence */
- if (E->OPC == OPC_ADC &&
- GetCodeEntries (S, L, I+1, 3) &&
- (L[0]->OPC == OPC_BCC || L[0]->OPC == OPC_JCC) &&
- L[0]->JumpTo != 0 &&
- !CodeEntryHasLabel (L[0]) &&
- L[1]->OPC == OPC_INX &&
- !CodeEntryHasLabel (L[1]) &&
- L[0]->JumpTo->Owner == L[2] &&
- !RegXUsed (S, I+3)) {
-
- /* Remove the bcs/dex */
- DelCodeEntries (S, I+1, 2);
+ if (L[0]->OPC == OP65_ADC &&
+ CS_GetEntries (S, L+1, I+1, 8) &&
+ L[1]->OPC == OP65_PHA &&
+ !CE_HasLabel (L[1]) &&
+ L[2]->OPC == OP65_TXA &&
+ !CE_HasLabel (L[2]) &&
+ L[3]->OPC == OP65_INY &&
+ !CE_HasLabel (L[3]) &&
+ L[4]->OPC == OP65_ADC &&
+ !CE_HasLabel (L[4]) &&
+ L[5]->OPC == OP65_TAX &&
+ !CE_HasLabel (L[5]) &&
+ L[6]->OPC == OP65_PLA &&
+ !CE_HasLabel (L[6]) &&
+ L[7]->OPC == OP65_LDY &&
+ !CE_HasLabel (L[7]) &&
+ CE_IsCall (L[8], "ldauidx") &&
+ !CE_HasLabel (L[8])) {
+
+ CodeEntry* X;
+
+ /* Store the low byte and remove the PHA instead */
+ X = NewCodeEntry (OP65_STA, AM65_ZP, "ptr1", 0, L[0]->LI);
+ CS_InsertEntry (S, X, I+1);
+ CS_DelEntry (S, I+2);
+
+ /* Store the high byte */
+ X = NewCodeEntry (OP65_STA, AM65_ZP, "ptr1+1", 0, L[4]->LI);
+ CS_InsertEntry (S, X, I+5);
+
+ /* Delete more transfer and PLA insns */
+ CS_DelEntry (S, I+7);
+ CS_DelEntry (S, I+6);
+
+ /* Delete the call to ldauidx */
+ CS_DelEntry (S, I+7);
+
+ /* Load high and low byte */
+ X = NewCodeEntry (OP65_LDX, AM65_IMM, "$00", 0, L[6]->LI);
+ CS_InsertEntry (S, X, I+7);
+ X = NewCodeEntry (OP65_LDA, AM65_ZP_INDY, "ptr1", 0, L[6]->LI);
+ CS_InsertEntry (S, X, I+8);
/* Remember, we had changes */
++Changes;
-/*****************************************************************************/
-/* Optimizations for compares */
-/*****************************************************************************/
-
-
-
-static unsigned OptCmp1 (CodeSeg* S)
-/* Search for the sequence
+static unsigned OptPtrLoad4 (CodeSeg* S)
+/* Search for the sequence:
*
- * stx xx
- * stx tmp1
- * ora tmp1
+ * lda #<(label+0)
+ * ldx #>(label+0)
+ * clc
+ * adc xxx
+ * bcc L
+ * inx
+ * L: ldy #$00
+ * jsr ldauidx
*
- * and replace it by
+ * and replace it by:
*
- * stx xx
- * ora xx
+ * ldy xxx
+ * ldx #$00
+ * lda label,y
*/
{
unsigned Changes = 0;
/* Walk over the entries */
unsigned I = 0;
- while (I < GetCodeEntryCount (S)) {
+ while (I < CS_GetEntryCount (S)) {
- CodeEntry* L[2];
+ CodeEntry* L[8];
+ unsigned Len;
/* Get next entry */
- CodeEntry* E = GetCodeEntry (S, I);
+ L[0] = CS_GetEntry (S, I);
/* Check for the sequence */
- if (E->OPC == OPC_STX &&
- GetCodeEntries (S, L, I+1, 2) &&
- L[0]->OPC == OPC_STX &&
- strcmp (L[0]->Arg, "tmp1") == 0 &&
- !CodeEntryHasLabel (L[0]) &&
- L[1]->OPC == OPC_ORA &&
- strcmp (L[1]->Arg, "tmp1") == 0 &&
- !CodeEntryHasLabel (L[1])) {
+ if (L[0]->OPC == OP65_LDA &&
+ L[0]->AM == AM65_IMM &&
+ CS_GetEntries (S, L+1, I+1, 7) &&
+ L[1]->OPC == OP65_LDX &&
+ L[1]->AM == AM65_IMM &&
+ !CE_HasLabel (L[1]) &&
+ L[2]->OPC == OP65_CLC &&
+ !CE_HasLabel (L[2]) &&
+ L[3]->OPC == OP65_ADC &&
+ (L[3]->AM == AM65_ABS || L[3]->AM == AM65_ZP) &&
+ !CE_HasLabel (L[3]) &&
+ (L[4]->OPC == OP65_BCC || L[4]->OPC == OP65_JCC) &&
+ L[4]->JumpTo != 0 &&
+ L[4]->JumpTo->Owner == L[6] &&
+ !CE_HasLabel (L[4]) &&
+ L[5]->OPC == OP65_INX &&
+ !CE_HasLabel (L[5]) &&
+ L[6]->OPC == OP65_LDY &&
+ CE_KnownImm (L[6]) &&
+ L[6]->Num == 0 &&
+ CE_IsCall (L[7], "ldauidx") &&
+ !CE_HasLabel (L[7]) &&
+ /* Check the label last because this is quite costly */
+ (Len = strlen (L[0]->Arg)) > 3 &&
+ L[0]->Arg[0] == '<' &&
+ L[0]->Arg[1] == '(' &&
+ strlen (L[1]->Arg) == Len &&
+ L[1]->Arg[0] == '>' &&
+ memcmp (L[0]->Arg+1, L[1]->Arg+1, Len-1) == 0) {
- /* Remove the remaining instructions */
- DelCodeEntries (S, I+1, 2);
+ CodeEntry* X;
+ char* Label;
+
+ /* We will create all the new stuff behind the current one so
+ * we keep the line references.
+ */
+ X = NewCodeEntry (OP65_LDY, L[3]->AM, L[3]->Arg, 0, L[0]->LI);
+ CS_InsertEntry (S, X, I+8);
- /* Insert the ora instead */
- InsertCodeEntry (S, NewCodeEntry (OPC_ORA, E->AM, E->Arg, 0), I+1);
+ X = NewCodeEntry (OP65_LDX, AM65_IMM, "$00", 0, L[0]->LI);
+ CS_InsertEntry (S, X, I+9);
+
+ Label = memcpy (xmalloc (Len-2), L[0]->Arg+2, Len-3);
+ Label[Len-3] = '\0';
+ X = NewCodeEntry (OP65_LDA, AM65_ABSY, Label, 0, L[0]->LI);
+ CS_InsertEntry (S, X, I+10);
+ xfree (Label);
+
+ /* Remove the old code */
+ CS_DelEntries (S, I, 8);
/* Remember, we had changes */
++Changes;
-static unsigned OptCmp2 (CodeSeg* S)
-/* Search for
+static unsigned OptPtrLoad5 (CodeSeg* S)
+/* Search for the sequence:
*
- * lda/and/ora/eor ...
- * cmp #$00
- * jeq/jne
+ * lda #<(label+0)
+ * ldx #>(label+0)
+ * ldy #$xx
+ * clc
+ * adc (sp),y
+ * bcc L
+ * inx
+ * L: ldy #$00
+ * jsr ldauidx
+ *
+ * and replace it by:
*
- * and remove the cmp.
+ * ldy #$xx
+ * lda (sp),y
+ * tay
+ * ldx #$00
+ * lda label,y
*/
{
unsigned Changes = 0;
/* Walk over the entries */
unsigned I = 0;
- while (I < GetCodeEntryCount (S)) {
+ while (I < CS_GetEntryCount (S)) {
- CodeEntry* L[2];
+ CodeEntry* L[9];
+ unsigned Len;
/* Get next entry */
- CodeEntry* E = GetCodeEntry (S, I);
+ L[0] = CS_GetEntry (S, I);
/* Check for the sequence */
- if ((E->OPC == OPC_LDA || IsBitOp (E)) &&
- GetCodeEntries (S, L, I+1, 2) &&
- IsCmpToZero (L[0]) &&
- !CodeEntryHasLabel (L[0]) &&
- (L[1]->Info & OF_FBRA) != 0 &&
- !CodeEntryHasLabel (L[1])) {
+ if (L[0]->OPC == OP65_LDA &&
+ L[0]->AM == AM65_IMM &&
+ CS_GetEntries (S, L+1, I+1, 8) &&
+ L[1]->OPC == OP65_LDX &&
+ L[1]->AM == AM65_IMM &&
+ !CE_HasLabel (L[1]) &&
+ L[2]->OPC == OP65_LDY &&
+ CE_KnownImm (L[2]) &&
+ !CE_HasLabel (L[2]) &&
+ L[3]->OPC == OP65_CLC &&
+ !CE_HasLabel (L[3]) &&
+ L[4]->OPC == OP65_ADC &&
+ L[4]->AM == AM65_ZP_INDY &&
+ !CE_HasLabel (L[4]) &&
+ (L[5]->OPC == OP65_BCC || L[5]->OPC == OP65_JCC) &&
+ L[5]->JumpTo != 0 &&
+ L[5]->JumpTo->Owner == L[7] &&
+ !CE_HasLabel (L[5]) &&
+ L[6]->OPC == OP65_INX &&
+ !CE_HasLabel (L[6]) &&
+ L[7]->OPC == OP65_LDY &&
+ CE_KnownImm (L[7]) &&
+ L[7]->Num == 0 &&
+ CE_IsCall (L[8], "ldauidx") &&
+ !CE_HasLabel (L[8]) &&
+ /* Check the label last because this is quite costly */
+ (Len = strlen (L[0]->Arg)) > 3 &&
+ L[0]->Arg[0] == '<' &&
+ L[0]->Arg[1] == '(' &&
+ strlen (L[1]->Arg) == Len &&
+ L[1]->Arg[0] == '>' &&
+ memcmp (L[0]->Arg+1, L[1]->Arg+1, Len-1) == 0) {
+
+ CodeEntry* X;
+ char* Label;
+
+ /* Add the lda */
+ X = NewCodeEntry (OP65_LDA, AM65_ZP_INDY, L[4]->Arg, 0, L[0]->LI);
+ CS_InsertEntry (S, X, I+3);
+
+ /* Add the tay */
+ X = NewCodeEntry (OP65_TAY, AM65_IMP, 0, 0, L[0]->LI);
+ CS_InsertEntry (S, X, I+4);
+
+ /* Add the ldx */
+ X = NewCodeEntry (OP65_LDX, AM65_IMM, "$00", 0, L[0]->LI);
+ CS_InsertEntry (S, X, I+5);
+
+ /* Add the lda */
+ Label = memcpy (xmalloc (Len-2), L[0]->Arg+2, Len-3);
+ Label[Len-3] = '\0';
+ X = NewCodeEntry (OP65_LDA, AM65_ABSY, Label, 0, L[0]->LI);
+ CS_InsertEntry (S, X, I+6);
+ xfree (Label);
- /* Remove the compare */
- DelCodeEntry (S, I+1);
+ /* Remove the old code */
+ CS_DelEntries (S, I, 2);
+ CS_DelEntries (S, I+5, 6);
/* Remember, we had changes */
++Changes;
-static unsigned OptCmp3 (CodeSeg* S)
-/* Search for
+static unsigned OptPtrLoad6 (CodeSeg* S)
+/* Search for the sequence
*
- * lda x
- * ldx y
- * cpx #a
- * bne L1
- * cmp #b
- * jne/jeq L2
+ * ldy ...
+ * jsr ldauidx
*
- * If a is zero, we may remove the compare. If a and b are both zero, we may
- * replace it by the sequence
+ * and replace it by:
*
- * lda x
- * ora x+1
- * jne/jeq ...
+ * ldy ...
+ * stx ptr1+1
+ * sta ptr1
+ * ldx #$00
+ * lda (ptr1),y
*
- * L1 may be either the label at the branch instruction, or the target label
- * of this instruction.
+ * This step must be execute *after* OptPtrLoad1!
*/
{
unsigned Changes = 0;
/* Walk over the entries */
unsigned I = 0;
- while (I < GetCodeEntryCount (S)) {
+ while (I < CS_GetEntryCount (S)) {
- CodeEntry* L[5];
+ CodeEntry* L[2];
/* Get next entry */
- CodeEntry* E = GetCodeEntry (S, I);
+ L[0] = CS_GetEntry (S, I);
/* Check for the sequence */
- if (E->OPC == OPC_LDA &&
- GetCodeEntries (S, L, I+1, 5) &&
- L[0]->OPC == OPC_LDX &&
- !CodeEntryHasLabel (L[0]) &&
- IsImmCmp16 (S, L+1)) {
-
- if (L[1]->Num == 0 && L[3]->Num == 0) {
- /* The value is zero, we may use the simple code version. */
- ReplaceOPC (L[0], OPC_ORA);
- DelCodeEntries (S, I+2, 3);
- } else {
- /* Move the lda instruction after the first branch. This will
- * improve speed, since the load is delayed after the first
- * test.
- */
- MoveCodeEntry (S, I, I+4);
-
- /* We will replace the ldx/cpx by lda/cmp */
- ReplaceOPC (L[0], OPC_LDA);
- ReplaceOPC (L[1], OPC_CMP);
+ if (L[0]->OPC == OP65_LDY &&
+ CS_GetEntries (S, L+1, I+1, 1) &&
+ CE_IsCall (L[1], "ldauidx") &&
+ !CE_HasLabel (L[1])) {
- }
+ CodeEntry* X;
+
+ /* Store the high byte */
+ X = NewCodeEntry (OP65_STA, AM65_ZP, "ptr1", 0, L[0]->LI);
+ CS_InsertEntry (S, X, I+1);
+
+ /* Store the low byte */
+ X = NewCodeEntry (OP65_STX, AM65_ZP, "ptr1+1", 0, L[0]->LI);
+ CS_InsertEntry (S, X, I+2);
+
+ /* Delete the call to ldauidx */
+ CS_DelEntry (S, I+3);
+ /* Load the high and low byte */
+ X = NewCodeEntry (OP65_LDX, AM65_IMM, "$00", 0, L[0]->LI);
+ CS_InsertEntry (S, X, I+3);
+ X = NewCodeEntry (OP65_LDA, AM65_ZP_INDY, "ptr1", 0, L[0]->LI);
+ CS_InsertEntry (S, X, I+4);
+
+ /* Remember, we had changes */
++Changes;
+
}
/* Next entry */
-static unsigned OptCmp4 (CodeSeg* S)
-/* Optimize compares of local variables:
+/*****************************************************************************/
+/* Decouple operations */
+/*****************************************************************************/
+
+
+
+static unsigned OptDecouple (CodeSeg* S)
+/* Decouple operations, that is, do the following replacements:
*
- * ldy #o
- * lda (sp),y
- * tax
- * dey
- * lda (sp),y
- * cpx #a
- * bne L1
- * cmp #b
- * jne/jeq L2
+ * dex -> ldx #imm
+ * inx -> ldx #imm
+ * dey -> ldy #imm
+ * iny -> ldy #imm
+ * tax -> ldx #imm
+ * txa -> lda #imm
+ * tay -> ldy #imm
+ * tya -> lda #imm
+ * lda sreg -> lda #imm
+ * ldx sreg -> ldx #imm
+ * ldy sreg -> ldy #imm
+ *
+ * Provided that the register values are known of course.
*/
{
unsigned Changes = 0;
+ unsigned I;
+
+ /* Generate register info for the following step */
+ CS_GenRegInfo (S);
/* Walk over the entries */
- unsigned I = 0;
- while (I < GetCodeEntryCount (S)) {
+ I = 0;
+ while (I < CS_GetEntryCount (S)) {
- CodeEntry* L[9];
+ char Buf [16];
- /* Check for the sequence */
- if (IsLocalLoad16 (S, I, L, 9) && IsImmCmp16 (S, L+5)) {
-
- if (L[5]->Num == 0 && L[7]->Num == 0) {
-
- /* The value is zero, we may use the simple code version:
- * ldy #o
- * lda (sp),y
- * dey
- * ora (sp),y
- * jne/jeq ...
- */
- ReplaceOPC (L[4], OPC_ORA);
- DelCodeEntries (S, I+5, 3); /* cpx/bne/cmp */
- DelCodeEntry (S, I+2); /* tax */
-
- } else {
-
- /* Change the code to just use the A register. Move the load
- * of the low byte after the first branch if possible:
- *
- * ldy #o
- * lda (sp),y
- * cmp #a
- * bne L1
- * dey
- * lda (sp),y
- * cmp #b
- * jne/jeq ...
- */
- DelCodeEntry (S, I+2); /* tax */
- ReplaceOPC (L[5], OPC_CMP); /* cpx -> cmp */
- MoveCodeEntry (S, I+4, I+2); /* cmp */
- MoveCodeEntry (S, I+5, I+3); /* bne */
+ /* Get next entry */
+ CodeEntry* E = CS_GetEntry (S, I);
+
+ /* Assume we have no replacement */
+ CodeEntry* X = 0;
+
+ /* Check the instruction */
+ switch (E->OPC) {
+
+ case OP65_DEX:
+ if (E->RI->In.RegX >= 0) {
+ xsprintf (Buf, sizeof (Buf), "$%02X", (E->RI->In.RegX - 1) & 0xFF);
+ X = NewCodeEntry (OP65_LDX, AM65_IMM, Buf, 0, E->LI);
+ }
+ break;
+
+ case OP65_DEY:
+ if (E->RI->In.RegY >= 0) {
+ xsprintf (Buf, sizeof (Buf), "$%02X", (E->RI->In.RegY - 1) & 0xFF);
+ X = NewCodeEntry (OP65_LDY, AM65_IMM, Buf, 0, E->LI);
+ }
+ break;
+
+ case OP65_INX:
+ if (E->RI->In.RegX >= 0) {
+ xsprintf (Buf, sizeof (Buf), "$%02X", (E->RI->In.RegX + 1) & 0xFF);
+ X = NewCodeEntry (OP65_LDX, AM65_IMM, Buf, 0, E->LI);
+ }
+ break;
+
+ case OP65_INY:
+ if (E->RI->In.RegY >= 0) {
+ xsprintf (Buf, sizeof (Buf), "$%02X", (E->RI->In.RegY + 1) & 0xFF);
+ X = NewCodeEntry (OP65_LDY, AM65_IMM, Buf, 0, E->LI);
+ }
+ break;
+
+ case OP65_LDA:
+ if (E->AM == AM65_ZP) {
+ if ((E->Use & REG_SREG_LO) != 0 && E->RI->In.SRegLo >= 0) {
+ xsprintf (Buf, sizeof (Buf), "$%02X", E->RI->In.SRegLo);
+ X = NewCodeEntry (OP65_LDA, AM65_IMM, Buf, 0, E->LI);
+ } else if ((E->Use & REG_SREG_HI) != 0 && E->RI->In.SRegHi >= 0) {
+ xsprintf (Buf, sizeof (Buf), "$%02X", E->RI->In.SRegHi);
+ X = NewCodeEntry (OP65_LDA, AM65_IMM, Buf, 0, E->LI);
+ }
+ }
+ break;
+
+ case OP65_LDX:
+ if (E->AM == AM65_ZP) {
+ if ((E->Use & REG_SREG_LO) != 0 && E->RI->In.SRegLo >= 0) {
+ xsprintf (Buf, sizeof (Buf), "$%02X", E->RI->In.SRegLo);
+ X = NewCodeEntry (OP65_LDX, AM65_IMM, Buf, 0, E->LI);
+ } else if ((E->Use & REG_SREG_HI) != 0 && E->RI->In.SRegHi >= 0) {
+ xsprintf (Buf, sizeof (Buf), "$%02X", E->RI->In.SRegHi);
+ X = NewCodeEntry (OP65_LDX, AM65_IMM, Buf, 0, E->LI);
+ }
+ }
+ break;
+
+ case OP65_LDY:
+ if (E->AM == AM65_ZP) {
+ if ((E->Use & REG_SREG_LO) != 0 && E->RI->In.SRegLo >= 0) {
+ xsprintf (Buf, sizeof (Buf), "$%02X", E->RI->In.SRegLo);
+ X = NewCodeEntry (OP65_LDY, AM65_IMM, Buf, 0, E->LI);
+ } else if ((E->Use & REG_SREG_HI) != 0 && E->RI->In.SRegHi >= 0) {
+ xsprintf (Buf, sizeof (Buf), "$%02X", E->RI->In.SRegHi);
+ X = NewCodeEntry (OP65_LDY, AM65_IMM, Buf, 0, E->LI);
+ }
+ }
+ break;
+
+ case OP65_TAX:
+ if (E->RI->In.RegA >= 0) {
+ xsprintf (Buf, sizeof (Buf), "$%02X", E->RI->In.RegA);
+ X = NewCodeEntry (OP65_LDX, AM65_IMM, Buf, 0, E->LI);
+ }
+ break;
+
+ case OP65_TAY:
+ if (E->RI->In.RegA >= 0) {
+ xsprintf (Buf, sizeof (Buf), "$%02X", E->RI->In.RegA);
+ X = NewCodeEntry (OP65_LDY, AM65_IMM, Buf, 0, E->LI);
+ }
+ break;
+
+ case OP65_TXA:
+ if (E->RI->In.RegX >= 0) {
+ xsprintf (Buf, sizeof (Buf), "$%02X", E->RI->In.RegX);
+ X = NewCodeEntry (OP65_LDA, AM65_IMM, Buf, 0, E->LI);
+ }
+ break;
+
+ case OP65_TYA:
+ if (E->RI->In.RegY >= 0) {
+ xsprintf (Buf, sizeof (Buf), "$%02X", E->RI->In.RegY);
+ X = NewCodeEntry (OP65_LDA, AM65_IMM, Buf, 0, E->LI);
+ }
+ break;
+
+ default:
+ /* Avoid gcc warnings */
+ break;
- }
+ }
+ /* Insert the replacement if we have one */
+ if (X) {
+ CS_InsertEntry (S, X, I+1);
+ CS_DelEntry (S, I);
++Changes;
}
}
+ /* Free register info */
+ CS_FreeRegInfo (S);
+
/* Return the number of changes made */
return Changes;
}
/*****************************************************************************/
-/* nega optimizations */
+/* Size optimization */
/*****************************************************************************/
-static unsigned OptNegA1 (CodeSeg* S)
-/* Check for
- *
- * ldx #$00
- * lda ..
- * jsr bnega
- *
- * Remove the ldx if the lda does not use it.
+#if 0
+static unsigned OptSize1 (CodeSeg* S)
+/* Do size optimization by calling special subroutines that preload registers.
+ * This routine does not work standalone, it needs a following register load
+ * removal pass.
*/
{
+ static const char* Func = {
+ "stax0sp", /* staxysp, y = 0 */
+ "addeq0sp",
+ "ldax0sp", /* ldaxysp, y = 1 */
+ "ldeax0sp", /* ldeaxysp, y = 3 */
+ "push0", /* pushax, a = 0, x = 0 */
+ "pusha0", /* pushax, x = 0 */
+ "pushaFF", /* pushax, x = ff */
+ "pusha0sp", /* pushaysp, y = 0 */
+ "tosadda0", /* tosaddax, x = 0 */
+ "tosanda0", /* tosandax, x = 0 */
+ "tosdiva0", /* tosdivax, x = 0 */
+ "toseqa0", /* toseqax, x = 0 */
+ "tosgea0", /* tosgeax, x = 0 */
+ "tosgta0", /* tosgtax, x = 0 */
+ "tosadd0ax", /* tosaddeax, sreg = 0 */
+ "laddeqa", /* laddeq, sreg = 0, x = 0 */
+ "laddeq1", /* laddeq, sreg = 0, x = 0, a = 1 */
+ "laddeq0sp", /* laddeqysp, y = 0 */
+ "tosand0ax", /* tosandeax, sreg = 0 */
+ "ldaxi", /* ldaxidx, y = 1 */
+ "ldeaxi", /* ldeaxidx, y = 3 */
+ "ldeax0sp", /* ldeaxysp, y = 3 */
+ "tosdiv0ax", /* tosdiveax, sreg = 0 */
+ "toslea0", /* tosleax, x = 0 */
+ "tosmod0ax", /* tosmodeax, sreg = 0 */
+ "tosmul0ax", /* tosmuleax, sreg = 0 */
+ "tosumul0ax", /* tosumuleax, sreg = 0 */
+ "tosor0ax", /* tosoreax, sreg = 0 */
+ "push0ax", /* pusheax, sreg = 0 */
+ "tosrsub0ax", /* tosrsubeax, sreg = 0 */
+ "tosshl0ax", /* tosshleax, sreg = 0 */
+ "tosasl0ax", /* tosasleax, sreg = 0 */
+ "tosshr0ax", /* tosshreax, sreg = 0 */
+ "tosasr0ax", /* tosasreax, sreg = 0 */
+ "tossub0ax", /* tossubeax, sreg = 0 */
+ "lsubeqa", /* lsubeq, sreg = 0, x = 0 */
+ "lsubeq1", /* lsubeq, sreg = 0, x = 0, a = 1 */
+ "lsubeq0sp", /* lsubeqysp, y = 0 */
+ "toslta0", /* tosltax, x = 0 */
+ "tosudiv0ax", /* tosudiveax, sreg = 0 */
+ "tosumod0ax", /* tosumodeax, sreg = 0 */
+ "tosxor0ax", /* tosxoreax, sreg = 0 */
+ "tosmoda0", /* tosmodax, x = 0 */
+ "tosmula0", /* tosmulax, x = 0 */
+ "tosumula0", /* tosumulax, x = 0 */
+ "tosnea0", /* tosneax, x = 0 */
+ "tosora0", /* tosorax, x = 0 */
+ "push1", /* pushax, x = 0, a = 1 */
+ "push2", /* pushax, x = 0, a = 2 */
+ "push3", /* pushax, x = 0, a = 3 */
+ "push4", /* pushax, x = 0, a = 4 */
+ "push5", /* pushax, x = 0, a = 5 */
+ "push6", /* pushax, x = 0, a = 6 */
+ "push7", /* pushax, x = 0, a = 7 */
+ "pushc0", /* pusha, a = 0 */
+ "pushc1", /* pusha, a = 1 */
+ "pushc2", /* pusha, a = 2 */
+ "tosrsuba0", /* tosrsubax, x = 0 */
+ "tosshla0", /* tosshlax, x = 0 */
+ "tosasla0", /* tosaslax, x = 0 */
+ "tosshra0", /* tosshrax, x = 0 */
+ "tosasra0", /* tosasrax, x = 0 */
+ "steax0sp", /* steaxsp, y = 0 */
+ "tossuba0", /* tossubax, x = 0 */
+ "subeq0sp", /* subeqysp, y = 0 */
+ "tosudiva0", /* tosudivax, x = 0 */
+ "tosugea0", /* tosugeax, x = 0 */
+ "tosugta0", /* tosugtax, x = 0 */
+ "tosulea0", /* tosuleax, x = 0 */
+ "tosulta0", /* tosultax, x = 0 */
+ "tosumoda0", /* tosumodax, x = 0 */
+ "tosxora0", /* tosxorax, x = 0 */
+ };
+
unsigned Changes = 0;
+ unsigned I;
- /* Walk over the entries */
- unsigned I = 0;
- while (I < GetCodeEntryCount (S)) {
+ /* Generate register info for the following step */
+ CS_GenRegInfo (S);
- CodeEntry* L[2];
+ /* Walk over the entries */
+ I = 0;
+ while (I < CS_GetEntryCount (S)) {
/* Get next entry */
- CodeEntry* E = GetCodeEntry (S, I);
-
- /* Check for a ldx */
- if (E->OPC == OPC_LDX &&
- E->AM == AM_IMM &&
- (E->Flags & CEF_NUMARG) != 0 &&
- E->Num == 0 &&
- GetCodeEntries (S, L, I+1, 2) &&
- L[0]->OPC == OPC_LDA &&
- (L[0]->Use & REG_X) == 0 &&
- L[1]->OPC == OPC_JSR &&
- strcmp (L[1]->Arg, "bnega") == 0) {
-
- /* Remove the ldx instruction */
- DelCodeEntry (S, I);
+ CodeEntry* E = CS_GetEntry (S, I);
- /* Remember, we had changes */
- ++Changes;
-
- }
/* Next entry */
++I;
}
+ /* Free register info */
+ CS_FreeRegInfo (S);
+
/* Return the number of changes made */
return Changes;
}
+#endif
-static unsigned OptNegA2 (CodeSeg* S)
-/* Check for
- *
- * lda ..
- * jsr bnega
- * jeq/jne ..
- *
- * Adjust the conditional branch and remove the call to the subroutine.
+static unsigned OptSize2 (CodeSeg* S)
+/* Do size optimization by using shorter code sequences, even if this
+ * introduces relations between instructions. This step must be one of the
+ * last steps, because it makes further work much more difficult.
*/
{
unsigned Changes = 0;
+ unsigned I;
+
+ /* Generate register info for the following step */
+ CS_GenRegInfo (S);
/* Walk over the entries */
- unsigned I = 0;
- while (I < GetCodeEntryCount (S)) {
+ I = 0;
+ while (I < CS_GetEntryCount (S)) {
- CodeEntry* L[2];
/* Get next entry */
- CodeEntry* E = GetCodeEntry (S, I);
-
- /* Check for the sequence */
- if (E->OPC == OPC_LDA &&
- GetCodeEntries (S, L, I+1, 2) &&
- L[0]->OPC == OPC_JSR &&
- strcmp (L[0]->Arg, "bnega") == 0 &&
- !CodeEntryHasLabel (L[0]) &&
- (L[1]->Info & OF_ZBRA) != 0) {
+ CodeEntry* E = CS_GetEntry (S, I);
+
+ /* Assume we have no replacement */
+ CodeEntry* X = 0;
+
+ /* Check the instruction */
+ switch (E->OPC) {
+
+ case OP65_LDA:
+ if (CE_KnownImm (E)) {
+ short Val = (short) E->Num;
+ if (Val == E->RI->In.RegX) {
+ X = NewCodeEntry (OP65_TXA, AM65_IMP, 0, 0, E->LI);
+ } else if (Val == E->RI->In.RegY) {
+ X = NewCodeEntry (OP65_TYA, AM65_IMP, 0, 0, E->LI);
+ } else if (E->RI->In.RegA >= 0 && CPU >= CPU_65C02) {
+ if (Val == ((E->RI->In.RegA - 1) & 0xFF)) {
+ X = NewCodeEntry (OP65_DEA, AM65_IMP, 0, 0, E->LI);
+ } else if (Val == ((E->RI->In.RegA + 1) & 0xFF)) {
+ X = NewCodeEntry (OP65_INA, AM65_IMP, 0, 0, E->LI);
+ }
+ }
+ }
+ break;
+
+ case OP65_LDX:
+ if (CE_KnownImm (E)) {
+ short Val = (short) E->Num;
+ if (E->RI->In.RegX >= 0 && Val == ((E->RI->In.RegX - 1) & 0xFF)) {
+ X = NewCodeEntry (OP65_DEX, AM65_IMP, 0, 0, E->LI);
+ } else if (E->RI->In.RegX >= 0 && Val == ((E->RI->In.RegX + 1) & 0xFF)) {
+ X = NewCodeEntry (OP65_INX, AM65_IMP, 0, 0, E->LI);
+ } else if (Val == E->RI->In.RegA) {
+ X = NewCodeEntry (OP65_TAX, AM65_IMP, 0, 0, E->LI);
+ }
+ }
+ break;
+
+ case OP65_LDY:
+ if (CE_KnownImm (E)) {
+ short Val = (short) E->Num;
+ if (E->RI->In.RegY >= 0 && Val == ((E->RI->In.RegY - 1) & 0xFF)) {
+ X = NewCodeEntry (OP65_DEY, AM65_IMP, 0, 0, E->LI);
+ } else if (E->RI->In.RegY >= 0 && Val == ((E->RI->In.RegY + 1) & 0xFF)) {
+ X = NewCodeEntry (OP65_INY, AM65_IMP, 0, 0, E->LI);
+ } else if (Val == E->RI->In.RegA) {
+ X = NewCodeEntry (OP65_TAY, AM65_IMP, 0, 0, E->LI);
+ }
+ }
+ break;
- /* Invert the branch */
- ReplaceOPC (L[1], GetInverseBranch (L[1]->OPC));
+ default:
+ /* Avoid gcc warnings */
+ break;
- /* Delete the subroutine call */
- DelCodeEntry (S, I+1);
+ }
- /* Remember, we had changes */
+ /* Insert the replacement if we have one */
+ if (X) {
+ CS_InsertEntry (S, X, I+1);
+ CS_DelEntry (S, I);
++Changes;
-
}
/* Next entry */
}
+ /* Free register info */
+ CS_FreeRegInfo (S);
+
/* Return the number of changes made */
return Changes;
}
/*****************************************************************************/
-/* negax optimizations */
+/* struct OptFunc */
/*****************************************************************************/
-static unsigned OptNegAX1 (CodeSeg* S)
-/* Search for the sequence:
- *
- * lda (xx),y
- * tax
- * dey
- * lda (xx),y
- * jsr bnegax
- * jne/jeq ...
- *
- * and replace it by
- *
- * lda (xx),y
- * dey
- * ora (xx),y
- * jeq/jne ...
+typedef struct OptFunc OptFunc;
+struct OptFunc {
+ unsigned (*Func) (CodeSeg*); /* Optimizer function */
+ const char* Name; /* Name of the function/group */
+ unsigned CodeSizeFactor; /* Code size factor for this opt func */
+ unsigned long TotalRuns; /* Total number of runs */
+ unsigned long LastRuns; /* Last number of runs */
+ unsigned long TotalChanges; /* Total number of changes */
+ unsigned long LastChanges; /* Last number of changes */
+ char Disabled; /* True if function disabled */
+};
+
+
+
+/*****************************************************************************/
+/* Code */
+/*****************************************************************************/
+
+
+
+/* Macro that builds a function description */
+#define OptFuncEntry(func) static OptFuncDesc D##func = { func, #func, 0 }
+
+/* A list of all the function descriptions */
+static OptFunc DOpt65C02Ind = { Opt65C02Ind, "Opt65C02Ind", 100, 0, 0, 0, 0, 0 };
+static OptFunc DOptAdd1 = { OptAdd1, "OptAdd1", 60, 0, 0, 0, 0, 0 };
+static OptFunc DOptAdd2 = { OptAdd2, "OptAdd2", 200, 0, 0, 0, 0, 0 };
+static OptFunc DOptAdd3 = { OptAdd3, "OptAdd3", 40, 0, 0, 0, 0, 0 };
+static OptFunc DOptBoolTrans = { OptBoolTrans, "OptBoolTrans", 100, 0, 0, 0, 0, 0 };
+static OptFunc DOptBranchDist = { OptBranchDist, "OptBranchDist", 0, 0, 0, 0, 0, 0 };
+static OptFunc DOptCmp1 = { OptCmp1, "OptCmp1", 85, 0, 0, 0, 0, 0 };
+static OptFunc DOptCmp2 = { OptCmp2, "OptCmp2", 75, 0, 0, 0, 0, 0 };
+static OptFunc DOptCmp3 = { OptCmp3, "OptCmp3", 75, 0, 0, 0, 0, 0 };
+static OptFunc DOptCmp4 = { OptCmp4, "OptCmp4", 100, 0, 0, 0, 0, 0 };
+static OptFunc DOptCmp5 = { OptCmp5, "OptCmp5", 100, 0, 0, 0, 0, 0 };
+static OptFunc DOptCmp6 = { OptCmp6, "OptCmp6", 85, 0, 0, 0, 0, 0 };
+static OptFunc DOptCmp7 = { OptCmp7, "OptCmp7", 50, 0, 0, 0, 0, 0 };
+static OptFunc DOptCondBranches = { OptCondBranches, "OptCondBranches", 80, 0, 0, 0, 0, 0 };
+static OptFunc DOptDeadCode = { OptDeadCode, "OptDeadCode", 100, 0, 0, 0, 0, 0 };
+static OptFunc DOptDeadJumps = { OptDeadJumps, "OptDeadJumps", 100, 0, 0, 0, 0, 0 };
+static OptFunc DOptDecouple = { OptDecouple, "OptDecouple", 100, 0, 0, 0, 0, 0 };
+static OptFunc DOptDupLoads = { OptDupLoads, "OptDupLoads", 0, 0, 0, 0, 0, 0 };
+static OptFunc DOptJumpCascades = { OptJumpCascades, "OptJumpCascades", 100, 0, 0, 0, 0, 0 };
+static OptFunc DOptJumpTarget = { OptJumpTarget, "OptJumpTarget", 100, 0, 0, 0, 0, 0 };
+static OptFunc DOptRTS = { OptRTS, "OptRTS", 100, 0, 0, 0, 0, 0 };
+static OptFunc DOptRTSJumps1 = { OptRTSJumps1, "OptRTSJumps1", 100, 0, 0, 0, 0, 0 };
+static OptFunc DOptRTSJumps2 = { OptRTSJumps2, "OptRTSJumps2", 100, 0, 0, 0, 0, 0 };
+static OptFunc DOptNegA1 = { OptNegA1, "OptNegA1", 100, 0, 0, 0, 0, 0 };
+static OptFunc DOptNegA2 = { OptNegA2, "OptNegA2", 100, 0, 0, 0, 0, 0 };
+static OptFunc DOptNegAX1 = { OptNegAX1, "OptNegAX1", 100, 0, 0, 0, 0, 0 };
+static OptFunc DOptNegAX2 = { OptNegAX2, "OptNegAX2", 100, 0, 0, 0, 0, 0 };
+static OptFunc DOptNegAX3 = { OptNegAX3, "OptNegAX3", 100, 0, 0, 0, 0, 0 };
+static OptFunc DOptNegAX4 = { OptNegAX4, "OptNegAX4", 100, 0, 0, 0, 0, 0 };
+static OptFunc DOptPtrLoad1 = { OptPtrLoad1, "OptPtrLoad1", 100, 0, 0, 0, 0, 0 };
+static OptFunc DOptPtrLoad2 = { OptPtrLoad2, "OptPtrLoad2", 100, 0, 0, 0, 0, 0 };
+static OptFunc DOptPtrLoad3 = { OptPtrLoad3, "OptPtrLoad3", 100, 0, 0, 0, 0, 0 };
+static OptFunc DOptPtrLoad4 = { OptPtrLoad4, "OptPtrLoad4", 100, 0, 0, 0, 0, 0 };
+static OptFunc DOptPtrLoad5 = { OptPtrLoad5, "OptPtrLoad5", 100, 0, 0, 0, 0, 0 };
+static OptFunc DOptPtrLoad6 = { OptPtrLoad6, "OptPtrLoad6", 100, 0, 0, 0, 0, 0 };
+static OptFunc DOptPtrStore1 = { OptPtrStore1, "OptPtrStore1", 100, 0, 0, 0, 0, 0 };
+static OptFunc DOptPtrStore2 = { OptPtrStore2, "OptPtrStore2", 100, 0, 0, 0, 0, 0 };
+static OptFunc DOptPush1 = { OptPush1, "OptPush1", 65, 0, 0, 0, 0, 0 };
+static OptFunc DOptShift1 = { OptShift1, "OptShift1", 100, 0, 0, 0, 0, 0 };
+static OptFunc DOptShift2 = { OptShift2, "OptShift2", 100, 0, 0, 0, 0, 0 };
+/*static OptFunc DOptSize1 = { OptSize1, "OptSize1", 100, 0, 0, 0, 0, 0 };*/
+static OptFunc DOptSize2 = { OptSize2, "OptSize2", 100, 0, 0, 0, 0, 0 };
+static OptFunc DOptStackOps = { OptStackOps, "OptStackOps", 100, 0, 0, 0, 0, 0 };
+static OptFunc DOptStoreLoad = { OptStoreLoad, "OptStoreLoad", 0, 0, 0, 0, 0, 0 };
+static OptFunc DOptSub1 = { OptSub1, "OptSub1", 100, 0, 0, 0, 0, 0 };
+static OptFunc DOptSub2 = { OptSub2, "OptSub2", 100, 0, 0, 0, 0, 0 };
+static OptFunc DOptTest1 = { OptTest1, "OptTest1", 100, 0, 0, 0, 0, 0 };
+static OptFunc DOptTransfers = { OptTransfers, "OptTransfers", 0, 0, 0, 0, 0, 0 };
+static OptFunc DOptUnusedLoads = { OptUnusedLoads, "OptUnusedLoads", 0, 0, 0, 0, 0, 0 };
+static OptFunc DOptUnusedStores = { OptUnusedStores, "OptUnusedStores", 0, 0, 0, 0, 0, 0 };
+
+
+/* Table containing all the steps in alphabetical order */
+static OptFunc* OptFuncs[] = {
+ &DOpt65C02Ind,
+ &DOptAdd1,
+ &DOptAdd2,
+ &DOptAdd3,
+ &DOptBoolTrans,
+ &DOptBranchDist,
+ &DOptCmp1,
+ &DOptCmp2,
+ &DOptCmp3,
+ &DOptCmp4,
+ &DOptCmp5,
+ &DOptCmp6,
+ &DOptCmp7,
+ &DOptCondBranches,
+ &DOptDeadCode,
+ &DOptDeadJumps,
+ &DOptDecouple,
+ &DOptDupLoads,
+ &DOptJumpCascades,
+ &DOptJumpTarget,
+ &DOptNegA1,
+ &DOptNegA2,
+ &DOptNegAX1,
+ &DOptNegAX2,
+ &DOptNegAX3,
+ &DOptNegAX4,
+ &DOptPtrLoad1,
+ &DOptPtrLoad2,
+ &DOptPtrLoad3,
+ &DOptPtrLoad4,
+ &DOptPtrLoad5,
+ &DOptPtrLoad6,
+ &DOptPtrStore1,
+ &DOptPtrStore2,
+ &DOptPush1,
+ &DOptRTS,
+ &DOptRTSJumps1,
+ &DOptRTSJumps2,
+ &DOptShift1,
+ &DOptShift2,
+ /*&DOptSize1,*/
+ &DOptSize2,
+ &DOptStackOps,
+ &DOptStoreLoad,
+ &DOptSub1,
+ &DOptSub2,
+ &DOptTest1,
+ &DOptTransfers,
+ &DOptUnusedLoads,
+ &DOptUnusedStores,
+};
+#define OPTFUNC_COUNT (sizeof(OptFuncs) / sizeof(OptFuncs[0]))
+
+
+
+static int CmpOptStep (const void* Key, const void* Func)
+/* Compare function for bsearch */
+{
+ return strcmp (Key, (*(const OptFunc**)Func)->Name);
+}
+
+
+
+static OptFunc* FindOptFunc (const char* Name)
+/* Find an optimizer step by name in the table and return a pointer. Return
+ * NULL if no such step is found.
*/
{
- unsigned Changes = 0;
+ /* Search for the function in the list */
+ OptFunc** O = bsearch (Name, OptFuncs, OPTFUNC_COUNT, sizeof (OptFuncs[0]), CmpOptStep);
+ return O? *O : 0;
+}
- /* Walk over the entries */
- unsigned I = 0;
- while (I < GetCodeEntryCount (S)) {
- CodeEntry* L[5];
- /* Get next entry */
- CodeEntry* E = GetCodeEntry (S, I);
+static OptFunc* GetOptFunc (const char* Name)
+/* Find an optimizer step by name in the table and return a pointer. Print an
+ * error and call AbEnd if not found.
+ */
+{
+ /* Search for the function in the list */
+ OptFunc* F = FindOptFunc (Name);
+ if (F == 0) {
+ /* Not found */
+ AbEnd ("Optimization step `%s' not found", Name);
+ }
+ return F;
+}
- /* Check for the sequence */
- if (E->OPC == OPC_LDA &&
- E->AM == AM_ZP_INDY &&
- GetCodeEntries (S, L, I+1, 5) &&
- L[0]->OPC == OPC_TAX &&
- L[1]->OPC == OPC_DEY &&
- L[2]->OPC == OPC_LDA &&
- L[2]->AM == AM_ZP_INDY &&
- strcmp (L[2]->Arg, E->Arg) == 0 &&
- !CodeEntryHasLabel (L[2]) &&
- L[3]->OPC == OPC_JSR &&
- strcmp (L[3]->Arg, "bnegax") == 0 &&
- !CodeEntryHasLabel (L[3]) &&
- (L[4]->Info & OF_ZBRA) != 0) {
-
- /* lda --> ora */
- ReplaceOPC (L[2], OPC_ORA);
-
- /* Invert the branch */
- ReplaceOPC (L[4], GetInverseBranch (L[4]->OPC));
-
- /* Delete the entries no longer needed. Beware: Deleting entries
- * will change the indices.
- */
- DelCodeEntry (S, I+4); /* jsr bnegax */
- DelCodeEntry (S, I+1); /* tax */
- /* Remember, we had changes */
- ++Changes;
- }
+void DisableOpt (const char* Name)
+/* Disable the optimization with the given name */
+{
+ if (strcmp (Name, "any") == 0) {
+ unsigned I;
+ for (I = 0; I < OPTFUNC_COUNT; ++I) {
+ OptFuncs[I]->Disabled = 1;
+ }
+ } else {
+ GetOptFunc(Name)->Disabled = 1;
+ }
+}
+
- /* Next entry */
- ++I;
+void EnableOpt (const char* Name)
+/* Enable the optimization with the given name */
+{
+ if (strcmp (Name, "any") == 0) {
+ unsigned I;
+ for (I = 0; I < OPTFUNC_COUNT; ++I) {
+ OptFuncs[I]->Disabled = 0;
+ }
+ } else {
+ GetOptFunc(Name)->Disabled = 0;
}
+}
- /* Return the number of changes made */
- return Changes;
+
+
+void ListOptSteps (FILE* F)
+/* List all optimization steps */
+{
+ unsigned I;
+ for (I = 0; I < OPTFUNC_COUNT; ++I) {
+ fprintf (F, "%s\n", OptFuncs[I]->Name);
+ }
}
-static unsigned OptNegAX2 (CodeSeg* S)
-/* Search for the sequence:
- *
- * lda xx
- * ldx yy
- * jsr bnegax
- * jne/jeq ...
- *
- * and replace it by
- *
- * lda xx
- * ora xx+1
- * jeq/jne ...
- */
+static void ReadOptStats (const char* Name)
+/* Read the optimizer statistics file */
{
- unsigned Changes = 0;
+ char Buf [256];
+ unsigned Lines;
+
+ /* Try to open the file */
+ FILE* F = fopen (Name, "r");
+ if (F == 0) {
+ /* Ignore the error */
+ return;
+ }
- /* Walk over the entries */
- unsigned I = 0;
- while (I < GetCodeEntryCount (S)) {
+ /* Read and parse the lines */
+ Lines = 0;
+ while (fgets (Buf, sizeof (Buf), F) != 0) {
- CodeEntry* L[3];
+ char* B;
+ unsigned Len;
+ OptFunc* Func;
- /* Get next entry */
- CodeEntry* E = GetCodeEntry (S, I);
+ /* Fields */
+ char Name[32];
+ unsigned long TotalRuns;
+ unsigned long TotalChanges;
- /* Check for the sequence */
- if (E->OPC == OPC_LDA &&
- GetCodeEntries (S, L, I+1, 3) &&
- L[0]->OPC == OPC_LDX &&
- !CodeEntryHasLabel (L[0]) &&
- L[1]->OPC == OPC_JSR &&
- strcmp (L[1]->Arg, "bnegax") == 0 &&
- !CodeEntryHasLabel (L[1]) &&
- (L[2]->Info & OF_ZBRA) != 0) {
+ /* Count lines */
+ ++Lines;
- /* ldx --> ora */
- ReplaceOPC (L[0], OPC_ORA);
+ /* Remove trailing white space including the line terminator */
+ B = Buf;
+ Len = strlen (B);
+ while (Len > 0 && IsSpace (B[Len-1])) {
+ --Len;
+ }
+ B[Len] = '\0';
- /* Invert the branch */
- ReplaceOPC (L[2], GetInverseBranch (L[2]->OPC));
+ /* Remove leading whitespace */
+ while (IsSpace (*B)) {
+ ++B;
+ }
- /* Delete the subroutine call */
- DelCodeEntry (S, I+2);
+ /* Check for empty and comment lines */
+ if (*B == '\0' || *B == ';' || *B == '#') {
+ continue;
+ }
- /* Remember, we had changes */
- ++Changes;
+ /* Parse the line */
+ if (sscanf (B, "%31s %lu %*u %lu %*u", Name, &TotalRuns, &TotalChanges) != 3) {
+ /* Syntax error */
+ continue;
+ }
+ /* Search for the optimizer step. */
+ Func = FindOptFunc (Name);
+ if (Func == 0) {
+ /* Not found */
+ continue;
}
- /* Next entry */
- ++I;
+ /* Found the step, set the fields */
+ Func->TotalRuns = TotalRuns;
+ Func->TotalChanges = TotalChanges;
}
- /* Return the number of changes made */
- return Changes;
+ /* Close the file, ignore errors here. */
+ fclose (F);
}
-static unsigned OptNegAX3 (CodeSeg* S)
-/* Search for the sequence:
- *
- * jsr _xxx
- * jsr bnega(x)
- * jeq/jne ...
- *
- * and replace it by:
- *
- * jsr _xxx
- * <boolean test>
- * jne/jeq ...
- */
+static void WriteOptStats (const char* Name)
+/* Write the optimizer statistics file */
{
- unsigned Changes = 0;
+ unsigned I;
- /* Walk over the entries */
- unsigned I = 0;
- while (I < GetCodeEntryCount (S)) {
+ /* Try to open the file */
+ FILE* F = fopen (Name, "w");
+ if (F == 0) {
+ /* Ignore the error */
+ return;
+ }
- CodeEntry* L[2];
+ /* Write the file */
+ for (I = 0; I < OPTFUNC_COUNT; ++I) {
+ const OptFunc* O = OptFuncs[I];
+ fprintf (F,
+ "%-20s %6lu %6lu %6lu %6lu\n",
+ O->Name,
+ O->TotalRuns,
+ O->LastRuns,
+ O->TotalChanges,
+ O->LastChanges);
+ }
- /* Get next entry */
- CodeEntry* E = GetCodeEntry (S, I);
+ /* Close the file, ignore errors here. */
+ fclose (F);
+}
- /* Check for the sequence */
- if (E->OPC == OPC_JSR &&
- E->Arg[0] == '_' &&
- GetCodeEntries (S, L, I+1, 2) &&
- L[0]->OPC == OPC_JSR &&
- strncmp (L[0]->Arg,"bnega",5) == 0 &&
- !CodeEntryHasLabel (L[0]) &&
- (L[1]->Info & OF_ZBRA) != 0) {
-
- /* Check if we're calling bnega or bnegax */
- int ByteSized = (strcmp (L[0]->Arg, "bnega") == 0);
-
- /* Delete the subroutine call */
- DelCodeEntry (S, I+1);
-
- /* Insert apropriate test code */
- if (ByteSized) {
- /* Test bytes */
- InsertCodeEntry (S, NewCodeEntry (OPC_TAX, AM_IMP, 0, 0), I+1);
- } else {
- /* Test words */
- InsertCodeEntry (S, NewCodeEntry (OPC_STX, AM_ZP, "tmp1", 0), I+1);
- InsertCodeEntry (S, NewCodeEntry (OPC_ORA, AM_ZP, "tmp1", 0), I+2);
- }
- /* Invert the branch */
- ReplaceOPC (L[1], GetInverseBranch (L[1]->OPC));
- /* Remember, we had changes */
- ++Changes;
+static unsigned RunOptFunc (CodeSeg* S, OptFunc* F, unsigned Max)
+/* Run one optimizer function Max times or until there are no more changes */
+{
+ unsigned Changes, C;
- }
+ /* Don't run the function if it is disabled or if it is prohibited by the
+ * code size factor
+ */
+ if (F->Disabled || CodeSizeFactor < F->CodeSizeFactor) {
+ return 0;
+ }
- /* Next entry */
- ++I;
+ /* Run this until there are no more changes */
+ Changes = 0;
+ do {
- }
+ /* Run the function */
+ C = F->Func (S);
+ Changes += C;
- /* Return the number of changes made */
+ /* Do statistics */
+ ++F->TotalRuns;
+ ++F->LastRuns;
+ F->TotalChanges += C;
+ F->LastChanges += C;
+
+ } while (--Max && C > 0);
+
+ /* Return the number of changes */
return Changes;
}
-/*****************************************************************************/
-/* Code */
-/*****************************************************************************/
+static unsigned RunOptGroup1 (CodeSeg* S)
+/* Run the first group of optimization steps. These steps translate known
+ * patterns emitted by the code generator into more optimal patterns. Order
+ * of the steps is important, because some of the steps done earlier cover
+ * the same patterns as later steps as subpatterns.
+ */
+{
+ unsigned Changes = 0;
+ Changes += RunOptFunc (S, &DOptPtrStore1, 1);
+ Changes += RunOptFunc (S, &DOptPtrStore2, 1);
+ Changes += RunOptFunc (S, &DOptPtrLoad1, 1);
+ Changes += RunOptFunc (S, &DOptPtrLoad2, 1);
+ Changes += RunOptFunc (S, &DOptPtrLoad3, 1);
+ Changes += RunOptFunc (S, &DOptPtrLoad4, 1);
+ Changes += RunOptFunc (S, &DOptPtrLoad5, 1);
+ Changes += RunOptFunc (S, &DOptNegAX1, 1);
+ Changes += RunOptFunc (S, &DOptNegAX2, 1);
+ Changes += RunOptFunc (S, &DOptNegAX3, 1);
+ Changes += RunOptFunc (S, &DOptNegAX4, 1);
+ Changes += RunOptFunc (S, &DOptAdd1, 1);
+ Changes += RunOptFunc (S, &DOptAdd2, 1);
+ Changes += RunOptFunc (S, &DOptShift1, 1);
+ Changes += RunOptFunc (S, &DOptShift2, 1);
+
+ /* Return the number of changes */
+ return Changes;
+}
-/* Table with all the optimization functions */
-typedef struct OptFunc OptFunc;
-struct OptFunc {
- unsigned (*Func) (CodeSeg*);/* Optimizer function */
- const char* Name; /* Name of optimizer step */
- char Disabled; /* True if pass disabled */
-};
+static unsigned RunOptGroup2 (CodeSeg* S)
+/* Run one group of optimization steps. This step involves just decoupling
+ * instructions by replacing them by instructions that do not depend on
+ * previous instructions. This makes it easier to find instructions that
+ * aren't used.
+ */
+{
+ unsigned Changes = 0;
+ Changes += RunOptFunc (S, &DOptDecouple, 1);
-/* Table with optimizer steps - are called in this order */
-static OptFunc OptFuncs [] = {
- /* Optimize subtractions */
- { OptSub1, "OptSub1", 0 },
- { OptSub2, "OptSub2", 0 },
- /* Optimize additions */
- { OptAdd1, "OptAdd1", 0 },
- /* Optimize jump cascades */
- { OptJumpCascades, "OptJumpCascades", 0 },
- /* Remove dead jumps */
- { OptDeadJumps, "OptDeadJumps", 0 },
- /* Change jsr/rts to jmp */
- { OptRTS, "OptRTS", 0 },
- /* Remove dead code */
- { OptDeadCode, "OptDeadCode", 0 },
- /* Optimize jump targets */
- { OptJumpTarget, "OptJumpTarget", 0 },
- /* Optimize conditional branches */
- { OptCondBranches, "OptCondBranches", 0 },
- /* Replace jumps to RTS by RTS */
- { OptRTSJumps, "OptRTSJumps", 0 },
- /* Remove calls to the bool transformer subroutines */
- { OptBoolTransforms, "OptBoolTransforms", 0 },
- /* Optimize calls to nega */
- { OptNegA1, "OptNegA1", 0 },
- { OptNegA2, "OptNegA2", 0 },
- /* Optimize calls to negax */
- { OptNegAX1, "OptNegAX1", 0 },
- { OptNegAX2, "OptNegAX2", 0 },
- { OptNegAX3, "OptNegAX3", 0 },
- /* Optimize compares */
- { OptCmp1, "OptCmp1", 0 },
- { OptCmp2, "OptCmp2", 0 },
- { OptCmp3, "OptCmp3", 0 },
- { OptCmp4, "OptCmp4", 0 },
- /* Remove unused loads */
- { OptUnusedLoads, "OptUnusedLoads", 0 },
- /* Optimize branch distance */
- { OptBranchDist, "OptBranchDist", 0 },
-};
+ /* Return the number of changes */
+ return Changes;
+}
-static OptFunc* FindOptStep (const char* Name)
-/* Find an optimizer step by name in the table and return a pointer. Print an
- * error and call AbEnd if not found.
+static unsigned RunOptGroup3 (CodeSeg* S)
+/* Run one group of optimization steps. These steps depend on each other,
+ * that means that one step may allow another step to do additional work,
+ * so we will repeat the steps as long as we see any changes.
*/
{
- unsigned I;
+ unsigned Changes, C;
- /* Run all optimization steps */
- for (I = 0; I < sizeof(OptFuncs)/sizeof(OptFuncs[0]); ++I) {
- if (strcmp (OptFuncs[I].Name, Name) == 0) {
- /* Found */
- return OptFuncs+I;
+ Changes = 0;
+ do {
+ C = 0;
+
+ C += RunOptFunc (S, &DOptPtrLoad6, 1);
+ C += RunOptFunc (S, &DOptNegA1, 1);
+ C += RunOptFunc (S, &DOptNegA2, 1);
+ C += RunOptFunc (S, &DOptSub1, 1);
+ C += RunOptFunc (S, &DOptSub2, 1);
+ C += RunOptFunc (S, &DOptAdd3, 1);
+ C += RunOptFunc (S, &DOptStackOps, 1);
+ C += RunOptFunc (S, &DOptJumpCascades, 1);
+ C += RunOptFunc (S, &DOptDeadJumps, 1);
+ C += RunOptFunc (S, &DOptRTS, 1);
+ C += RunOptFunc (S, &DOptDeadCode, 1);
+ C += RunOptFunc (S, &DOptJumpTarget, 1);
+ C += RunOptFunc (S, &DOptCondBranches, 1);
+ C += RunOptFunc (S, &DOptRTSJumps1, 1);
+ C += RunOptFunc (S, &DOptBoolTrans, 1);
+ C += RunOptFunc (S, &DOptCmp1, 1);
+ C += RunOptFunc (S, &DOptCmp2, 1);
+ C += RunOptFunc (S, &DOptCmp3, 1);
+ C += RunOptFunc (S, &DOptCmp4, 1);
+ C += RunOptFunc (S, &DOptCmp5, 1);
+ C += RunOptFunc (S, &DOptCmp6, 1);
+ C += RunOptFunc (S, &DOptCmp7, 1);
+ C += RunOptFunc (S, &DOptTest1, 1);
+ C += RunOptFunc (S, &DOptUnusedLoads, 1);
+ C += RunOptFunc (S, &DOptUnusedStores, 1);
+ C += RunOptFunc (S, &DOptDupLoads, 1);
+ C += RunOptFunc (S, &DOptStoreLoad, 1);
+ C += RunOptFunc (S, &DOptTransfers, 1);
+
+ Changes += C;
+
+ } while (C);
+
+ /* Return the number of changes */
+ return Changes;
+}
+
+
+
+static unsigned RunOptGroup4 (CodeSeg* S)
+/* 65C02 specific optimizations. */
+{
+ unsigned Changes = 0;
+
+ if (CPU >= CPU_65C02) {
+ /* Replace (zp),y by (zp) if Y is zero. If we have changes, run register
+ * load optimization again, since loads of Y may have become unnecessary.
+ */
+ unsigned C = RunOptFunc (S, &DOpt65C02Ind, 1);
+ Changes += C;
+ if (C) {
+ Changes += RunOptFunc (S, &DOptUnusedLoads, 1);
}
}
- /* Not found */
- AbEnd ("Optimization step `%s' not found", Name);
- return 0;
+ /* Return the number of changes */
+ return Changes;
}
-void DisableOpt (const char* Name)
-/* Disable the optimization with the given name */
+static unsigned RunOptGroup5 (CodeSeg* S)
+/* Run another round of pattern replacements. These are done late, since there
+ * may be better replacements before.
+ */
{
- OptFunc* F = FindOptStep (Name);
- F->Disabled = 1;
+ unsigned Changes = 0;
+
+ Changes += RunOptFunc (S, &DOptPush1, 1);
+
+ /* Return the number of changes */
+ return Changes;
}
-void EnableOpt (const char* Name)
-/* Enable the optimization with the given name */
+static unsigned RunOptGroup6 (CodeSeg* S)
+/* The last group of optimization steps. Adjust branches, do size optimizations.
+ */
{
- OptFunc* F = FindOptStep (Name);
- F->Disabled = 0;
+ unsigned Changes = 0;
+ unsigned C;
+
+ /* Optimize for size, that is replace operations by shorter ones, even
+ * if this does hinder further optimizations (no problem since we're
+ * done soon).
+ */
+ Changes += RunOptFunc (S, &DOptSize2, 1);
+
+ /* Run the jump target optimization again, since the size optimization
+ * above may have opened new oportunities.
+ */
+ Changes += RunOptFunc (S, &DOptJumpTarget, 5);
+
+ /* Adjust branch distances */
+ Changes += RunOptFunc (S, &DOptBranchDist, 3);
+
+ /* Replace conditional branches to RTS. If we had changes, we must run dead
+ * code elimination again, since the change may have introduced dead code.
+ */
+ C = RunOptFunc (S, &DOptRTSJumps2, 1);
+ Changes += C;
+ if (C) {
+ Changes += RunOptFunc (S, &DOptDeadCode, 1);
+ }
+
+ /* Return the number of changes */
+ return Changes;
}
void RunOpt (CodeSeg* S)
/* Run the optimizer */
{
- unsigned I;
- unsigned Pass = 0;
- unsigned OptChanges;
+ const char* StatFileName;
/* If we shouldn't run the optimizer, bail out */
if (!Optimize) {
- return;
+ return;
+ }
+
+ /* Check if we are requested to write optimizer statistics */
+ StatFileName = getenv ("CC65_OPTSTATS");
+ if (StatFileName) {
+ ReadOptStats (StatFileName);
}
/* Print the name of the function we are working on */
Print (stdout, 1, "Running optimizer for global code segment\n");
}
- /* Repeat all steps until there are no more changes */
- do {
- /* Reset the number of changes */
- OptChanges = 0;
-
- /* Keep the user hapy */
- Print (stdout, 1, " Optimizer pass %u:\n", ++Pass);
-
- /* Run all optimization steps */
- for (I = 0; I < sizeof(OptFuncs)/sizeof(OptFuncs[0]); ++I) {
-
- /* Print the name of the following optimizer step */
- Print (stdout, 1, " %s:%*s", OptFuncs[I].Name,
- (int) (30-strlen(OptFuncs[I].Name)), "");
-
- /* Check if the step is disabled */
- if (OptFuncs[I].Disabled) {
- Print (stdout, 1, "Disabled\n");
- } else {
- unsigned Changes = OptFuncs[I].Func (S);
- OptChanges += Changes;
- Print (stdout, 1, "%u Changes\n", Changes);
- }
- }
-
- } while (OptChanges > 0);
+ /* Run groups of optimizations */
+ RunOptGroup1 (S);
+ RunOptGroup2 (S);
+ RunOptGroup3 (S);
+ RunOptGroup4 (S);
+ RunOptGroup5 (S);
+ RunOptGroup6 (S);
+
+ /* Write statistics */
+ if (StatFileName) {
+ WriteOptStats (StatFileName);
+ }
}