bslathi19/cc65
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

Unified the "store via pointer" optimization routines.

Browse Source 
git-svn-id: svn://svn.cc65.org/cc65/trunk@5719 b7a2c559-68d2-44c3-8de9-860c34a00d81
This commit is contained in:
uz
2012-06-16 15:46:45 +00:00
parent 7c9b27c048
commit c1d7ed26e8
4 changed files with 317 additions and 449 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
src/cc65/codeopt.c
View File

@@ -1201,10 +1201,8 @@ static OptFunc DOptPtrLoad15 = { OptPtrLoad15, "OptPtrLoad15", 86, 0,
static OptFunc DOptPtrLoad16 = { OptPtrLoad16, "OptPtrLoad16", 100, 0, 0, 0, 0, 0 }; static OptFunc DOptPtrLoad16 = { OptPtrLoad16, "OptPtrLoad16", 100, 0, 0, 0, 0, 0 };
static OptFunc DOptPtrLoad17 = { OptPtrLoad17, "OptPtrLoad17", 190, 0, 0, 0, 0, 0 }; static OptFunc DOptPtrLoad17 = { OptPtrLoad17, "OptPtrLoad17", 190, 0, 0, 0, 0, 0 };
static OptFunc DOptPtrStore1 = { OptPtrStore1, "OptPtrStore1", 65, 0, 0, 0, 0, 0 }; static OptFunc DOptPtrStore1 = { OptPtrStore1, "OptPtrStore1", 65, 0, 0, 0, 0, 0 };
static OptFunc DOptPtrStore2 = { OptPtrStore2, "OptPtrStore2", 50, 0, 0, 0, 0, 0 }; static OptFunc DOptPtrStore2 = { OptPtrStore2, "OptPtrStore2", 65, 0, 0, 0, 0, 0 };
static OptFunc DOptPtrStore3 = { OptPtrStore3, "OptPtrStore3", 50, 0, 0, 0, 0, 0 }; static OptFunc DOptPtrStore3 = { OptPtrStore3, "OptPtrStore3", 100, 0, 0, 0, 0, 0 };
static OptFunc DOptPtrStore4 = { OptPtrStore4, "OptPtrStore4", 65, 0, 0, 0, 0, 0 };
static OptFunc DOptPtrStore5 = { OptPtrStore5, "OptPtrStore5", 100, 0, 0, 0, 0, 0 };
static OptFunc DOptPush1 = { OptPush1, "OptPush1", 65, 0, 0, 0, 0, 0 }; static OptFunc DOptPush1 = { OptPush1, "OptPush1", 65, 0, 0, 0, 0, 0 };
static OptFunc DOptPush2 = { OptPush2, "OptPush2", 50, 0, 0, 0, 0, 0 }; static OptFunc DOptPush2 = { OptPush2, "OptPush2", 50, 0, 0, 0, 0, 0 };
static OptFunc DOptPushPop = { OptPushPop, "OptPushPop", 0, 0, 0, 0, 0, 0 }; static OptFunc DOptPushPop = { OptPushPop, "OptPushPop", 0, 0, 0, 0, 0, 0 };
@@ -1297,8 +1295,6 @@ static OptFunc* OptFuncs[] = {
&DOptPtrStore1, &DOptPtrStore1,
&DOptPtrStore2, &DOptPtrStore2,
&DOptPtrStore3, &DOptPtrStore3,
&DOptPtrStore4,
&DOptPtrStore5,
&DOptPush1, &DOptPush1,
&DOptPush2, &DOptPush2,
&DOptPushPop, &DOptPushPop,
@@ -1569,11 +1565,10 @@ static unsigned RunOptGroup1 (CodeSeg* S)
unsigned Changes = 0; unsigned Changes = 0;
Changes += RunOptFunc (S, &DOptStackPtrOps, 5); Changes += RunOptFunc (S, &DOptStackPtrOps, 5);
Changes += RunOptFunc (S, &DOptPtrLoad13, 1); /* ### */
Changes += RunOptFunc (S, &DOptPtrStore1, 1); Changes += RunOptFunc (S, &DOptPtrStore1, 1);
Changes += RunOptFunc (S, &DOptPtrStore2, 1); Changes += RunOptFunc (S, &DOptPtrStore2, 1);
Changes += RunOptFunc (S, &DOptPtrStore3, 1); Changes += RunOptFunc (S, &DOptPtrStore3, 1);
Changes += RunOptFunc (S, &DOptPtrStore4, 1);
Changes += RunOptFunc (S, &DOptPtrStore5, 1);
Changes += RunOptFunc (S, &DOptAdd3, 1); /* Before OptPtrLoad5! */ Changes += RunOptFunc (S, &DOptAdd3, 1); /* Before OptPtrLoad5! */
Changes += RunOptFunc (S, &DOptPtrLoad1, 1); Changes += RunOptFunc (S, &DOptPtrLoad1, 1);
Changes += RunOptFunc (S, &DOptPtrLoad2, 1); Changes += RunOptFunc (S, &DOptPtrLoad2, 1);

4
src/cc65/coptind.c
View File

@@ -1950,7 +1950,8 @@ unsigned OptPushPop (CodeSeg* S)
case FoundPop: case FoundPop:
/* We're at the instruction after the PLA. /* We're at the instruction after the PLA.
* Check for the following conditions: * Check for the following conditions:
* - If this instruction is a store of A, does not have a * - If this instruction is a store of A that doesn't use
* another register, if the instruction does not have a
* label, and A is not used later, we may replace the PHA * label, and A is not used later, we may replace the PHA
* by the store and remove pla if several other conditions * by the store and remove pla if several other conditions
* are met. * are met.
@@ -1959,6 +1960,7 @@ unsigned OptPushPop (CodeSeg* S)
* between push and pop, we may remove PHA and PLA. * between push and pop, we may remove PHA and PLA.
*/ */
if (E->OPC == OP65_STA && if (E->OPC == OP65_STA &&
(E->AM == AM65_ABS || E->AM == AM65_ZP) &&
!CE_HasLabel (E) && !CE_HasLabel (E) &&
!RegAUsed (S, I+1) && !RegAUsed (S, I+1) &&
!MemAccess (S, Push+1, Pop-1, E)) { !MemAccess (S, Push+1, Pop-1, E)) {

530
src/cc65/coptptrstore.c
View File

@@ -92,14 +92,16 @@ static unsigned OptPtrStore1Sub (CodeSeg* S, unsigned I, CodeEntry** const L)
static const char* LoadAXZP (CodeSeg* S, unsigned I) static const char* LoadAXZP (CodeSeg* S, unsigned I)
/* If the two instructions at S/I are a load of A/X from a two byte zero byte /* If the two instructions preceeding S/I are a load of A/X from a two byte
* location, return the name of the zero page location. Otherwise return NULL. * zero byte location, return the name of the zero page location. Otherwise
* return NULL.
*/ */
{ {
CodeEntry* L[2]; CodeEntry* L[2];
unsigned Len; unsigned Len;
if (CS_GetEntries (S, L, I, 2) && if (I >= 2 &&
CS_GetEntries (S, L, I-2, 2) &&
L[0]->OPC == OP65_LDA && L[0]->OPC == OP65_LDA &&
L[0]->AM == AM65_ZP && L[0]->AM == AM65_ZP &&
L[1]->OPC == OP65_LDX && L[1]->OPC == OP65_LDX &&
@@ -124,35 +126,28 @@ static const char* LoadAXZP (CodeSeg* S, unsigned I)
static const char* LoadAXImm (CodeSeg* S, unsigned I) static const char* LoadAXImm (CodeSeg* S, unsigned I)
/* If the two instructions at S/I are a load of A/X of a constant value or a /* If the instructions preceeding S/I are a load of A/X of a constant value
* wqord sized address label, return the address of the location as a string. * or a word sized address label, return the address of the location as a
* string.
* Beware: In case of a numeric value, the result is returned in static * Beware: In case of a numeric value, the result is returned in static
* storage which is overwritten with each call. * storage which is overwritten with each call.
*/ */
{ {
static StrBuf Buf = STATIC_STRBUF_INITIALIZER; static StrBuf Buf = STATIC_STRBUF_INITIALIZER;
CodeEntry* L[2]; CodeEntry* L[2];
CodeEntry* ALoad;
CodeEntry* XLoad;
unsigned Len; unsigned Len;
if (CS_GetEntries (S, L, I, 2) && /* Fetch entry at I and check if A/X is known */
((L[0]->OPC == OP65_LDA && L[1]->OPC == OP65_LDX) || L[0] = CS_GetEntry (S, I);
(L[0]->OPC == OP65_LDX && L[1]->OPC == OP65_LDA)) && if (L[0] != 0 &&
L[0]->AM == AM65_IMM && RegValIsKnown (L[0]->RI->In.RegA) &&
L[1]->AM == AM65_IMM && RegValIsKnown (L[0]->RI->In.RegX)) {
!CE_HasLabel (L[1])) {
/* Immediate load of A/X */
if (CE_HasNumArg (L[0]) && CE_HasNumArg (L[1])) {
/* Numeric argument - get low and high byte */ /* Numeric argument - get low and high byte */
unsigned Hi, Lo; unsigned Lo = (L[0]->RI->In.RegA & 0xFF);
if (L[0]->OPC == OP65_LDA) { unsigned Hi = (L[0]->RI->In.RegX & 0xFF);
Lo = (L[0]->Num & 0xFF);
Hi = (L[1]->Num & 0xFF);
} else {
Lo = (L[1]->Num & 0xFF);
Hi = (L[0]->Num & 0xFF);
}
/* Format into buffer */ /* Format into buffer */
SB_Printf (&Buf, "$%04X", Lo | (Hi << 8)); SB_Printf (&Buf, "$%04X", Lo | (Hi << 8));
@@ -160,31 +155,58 @@ static const char* LoadAXImm (CodeSeg* S, unsigned I)
/* Return the address as a string */ /* Return the address as a string */
return SB_GetConstBuf (&Buf); return SB_GetConstBuf (&Buf);
} else if ((Len = strlen (L[0]->Arg)) > 3 && }
L[0]->Arg[0] == '<' &&
L[0]->Arg[1] == '(' && /* Search back for the two instructions loading A and X. Abort
strlen (L[1]->Arg) == Len && * the search if the registers are changed in any other way or
L[1]->Arg[0] == '>' && * if a label is reached while we don't have both loads.
memcmp (L[0]->Arg+1, L[1]->Arg+1, Len-1) == 0) { */
ALoad = 0;
XLoad = 0;
while (I-- > 0) {
/* Get next entry */
CodeEntry* E = CS_GetEntry (S, I);
/* Check for the loads of A and X */
if (ALoad == 0 && E->OPC == OP65_LDA && E->AM == AM65_IMM) {
ALoad = E;
} else if (E->Chg & REG_A) {
/* A is changed before we get the load */
return 0;
} else if (XLoad == 0 && E->OPC == OP65_LDX && E->AM == AM65_IMM) {
XLoad = E;
} else if (E->Chg & REG_X) {
/* X is changed before we get the load */
return 0;
}
if (ALoad != 0 && XLoad != 0) {
/* We have both */
break;
}
/* If we have a label, before both are found, bail out */
if (CE_HasLabel (E)) {
return 0;
}
}
/* Check for a load of a label address */
if ((Len = strlen (ALoad->Arg)) > 3 &&
ALoad->Arg[0] == '<' &&
ALoad->Arg[1] == '(' &&
strlen (XLoad->Arg) == Len &&
XLoad->Arg[0] == '>' &&
memcmp (ALoad->Arg+1, XLoad->Arg+1, Len-1) == 0) {
/* Load of an address label */ /* Load of an address label */
SB_CopyBuf (&Buf, L[0]->Arg + 2, Len - 3); SB_CopyBuf (&Buf, ALoad->Arg + 2, Len - 3);
SB_Terminate (&Buf); SB_Terminate (&Buf);
return SB_GetConstBuf (&Buf); return SB_GetConstBuf (&Buf);
}
} else {
/* Not found */ /* Not found */
return 0; return 0;
}
} else {
/* Not found */
return 0;
}
} }
@@ -222,6 +244,13 @@ unsigned OptPtrStore1 (CodeSeg* S)
* ldy xxx * ldy xxx
* ldx #$00 * ldx #$00
* lda yyy * lda yyy
* sta (zp),y
*
* or by
*
* ldy xxx
* ldx #$00
* lda yyy
* sta label,y * sta label,y
* *
* or by * or by
@@ -231,20 +260,17 @@ unsigned OptPtrStore1 (CodeSeg* S)
* lda yyy * lda yyy
* sta $xxxx,y * sta $xxxx,y
* *
* or by * depending on the code preceeding the sequence above.
*
* ldy xxx
* ldx #$00
* lda yyy
* sta (zp),y
*
* depending on the two instructions preceeding the sequence above.
*/ */
{ {
unsigned Changes = 0; unsigned Changes = 0;
unsigned I;
/* Generate register info */
CS_GenRegInfo (S);
/* Walk over the entries */ /* Walk over the entries */
unsigned I = 0; I = 0;
while (I < CS_GetEntryCount (S)) { while (I < CS_GetEntryCount (S)) {
CodeEntry* L[9]; CodeEntry* L[9];
@@ -258,7 +284,9 @@ unsigned OptPtrStore1 (CodeSeg* S)
L[1]->OPC == OP65_ADC && L[1]->OPC == OP65_ADC &&
(L[1]->AM == AM65_ABS || (L[1]->AM == AM65_ABS ||
L[1]->AM == AM65_ZP || L[1]->AM == AM65_ZP ||
L[1]->AM == AM65_IMM) && L[1]->AM == AM65_IMM ||
(L[1]->AM == AM65_ZP_INDY &&
RegValIsKnown (L[1]->RI->In.RegY))) &&
(L[2]->OPC == OP65_BCC || L[2]->OPC == OP65_JCC) && (L[2]->OPC == OP65_BCC || L[2]->OPC == OP65_JCC) &&
L[2]->JumpTo != 0 && L[2]->JumpTo != 0 &&
L[2]->JumpTo->Owner == L[4] && L[2]->JumpTo->Owner == L[4] &&
@@ -278,30 +306,21 @@ unsigned OptPtrStore1 (CodeSeg* S)
/* Track the insertion point */ /* Track the insertion point */
unsigned IP = I + 9; unsigned IP = I + 9;
if ((Loc = LoadAXZP (S, I)) != 0) {
unsigned DeleteStart = I;
unsigned DeleteCount = 9;
if (I >= 2) {
if ((Loc = LoadAXZP (S, I-2)) != 0) {
/* If the sequence is preceeded by a load of a ZP value, /* If the sequence is preceeded by a load of a ZP value,
* we can use this ZP value as a pointer using ZP * we can use this ZP value as a pointer using ZP
* indirect Y addressing. * indirect Y addressing.
*/ */
AM = AM65_ZP_INDY; AM = AM65_ZP_INDY;
DeleteStart -= 2; } else if ((Loc = LoadAXImm (S, I)) != 0) {
DeleteCount += 2;
} else if ((Loc = LoadAXImm (S, I-2)) != 0) {
/* If the sequence is preceeded by a load of an immediate /* If the sequence is preceeded by a load of an immediate
* value, we can use this absolute value as an address * value, we can use this absolute value as an address
* using absolute indexed Y addressing. * using absolute indexed Y addressing.
*/ */
AM = AM65_ABSY; AM = AM65_ABSY;
DeleteStart -= 2;
DeleteCount += 2;
}
} }
/* If we don't have a zero page location, we use ptr1 with zp /* If we don't have a store location, we use ptr1 with zp
* indirect Y addressing. We must store the value in A/X into * indirect Y addressing. We must store the value in A/X into
* ptr1 in this case. * ptr1 in this case.
*/ */
@@ -319,8 +338,21 @@ unsigned OptPtrStore1 (CodeSeg* S)
} }
/* If the index is loaded from (zp),y, we cannot do that directly.
* Note: In this case, the Y register will contain the correct
* value after removing the old code, so we don't need to load
* it here.
*/
if (L[1]->AM == AM65_ZP_INDY) {
X = NewCodeEntry (OP65_LDA, L[1]->AM, L[1]->Arg, 0, L[1]->LI);
CS_InsertEntry (S, X, IP++);
X = NewCodeEntry (OP65_TAY, AM65_IMP, 0, 0, L[1]->LI);
CS_InsertEntry (S, X, IP++);
} else {
X = NewCodeEntry (OP65_LDY, L[1]->AM, L[1]->Arg, 0, L[1]->LI); X = NewCodeEntry (OP65_LDY, L[1]->AM, L[1]->Arg, 0, L[1]->LI);
CS_InsertEntry (S, X, IP++); CS_InsertEntry (S, X, IP++);
}
X = NewCodeEntry (OP65_LDX, L[5]->AM, L[5]->Arg, 0, L[5]->LI); X = NewCodeEntry (OP65_LDX, L[5]->AM, L[5]->Arg, 0, L[5]->LI);
CS_InsertEntry (S, X, IP++); CS_InsertEntry (S, X, IP++);
@@ -332,9 +364,9 @@ unsigned OptPtrStore1 (CodeSeg* S)
CS_InsertEntry (S, X, IP++); CS_InsertEntry (S, X, IP++);
/* Remove the old code */ /* Remove the old code */
CS_DelEntries (S, DeleteStart, DeleteCount); CS_DelEntries (S, I, 9);
/* Skip most of the generated replacement */ /* Skip most of the generated replacement code */
I += 3; I += 3;
/* Remember, we had changes */ /* Remember, we had changes */
@@ -347,6 +379,9 @@ unsigned OptPtrStore1 (CodeSeg* S)
} }
/* Free the register info */
CS_FreeRegInfo (S);
/* Return the number of changes made */ /* Return the number of changes made */
return Changes; return Changes;
} }
@@ -354,245 +389,6 @@ unsigned OptPtrStore1 (CodeSeg* S)
unsigned OptPtrStore2 (CodeSeg* S) unsigned OptPtrStore2 (CodeSeg* S)
/* Search for the sequence:
*
* lda #<(label+0)
* ldx #>(label+0)
* ldy aaa
* clc
* adc (sp),y
* bcc L
* inx
* L: jsr pushax
* ldx #$00
* lda yyy
* ldy #$00
* jsr staspidx
*
* and replace it by:
*
* ldy aaa
* ldx #$00
* lda (sp),y
* tay
* lda yyy
* sta label,y
*/
{
unsigned Changes = 0;
/* Walk over the entries */
unsigned I = 0;
while (I < CS_GetEntryCount (S)) {
CodeEntry* L[12];
unsigned Len;
/* Get next entry */
L[0] = CS_GetEntry (S, I);
/* Check for the sequence */
if (L[0]->OPC == OP65_LDA &&
L[0]->AM == AM65_IMM &&
CS_GetEntries (S, L+1, I+1, 11) &&
L[1]->OPC == OP65_LDX &&
L[1]->AM == AM65_IMM &&
L[2]->OPC == OP65_LDY &&
L[3]->OPC == OP65_CLC &&
L[4]->OPC == OP65_ADC &&
L[4]->AM == AM65_ZP_INDY &&
(L[5]->OPC == OP65_BCC || L[5]->OPC == OP65_JCC) &&
L[5]->JumpTo != 0 &&
L[5]->JumpTo->Owner == L[7] &&
L[6]->OPC == OP65_INX &&
CE_IsCallTo (L[7], "pushax") &&
L[8]->OPC == OP65_LDX &&
L[9]->OPC == OP65_LDA &&
L[10]->OPC == OP65_LDY &&
CE_IsKnownImm (L[10], 0) &&
CE_IsCallTo (L[11], "staspidx") &&
!CS_RangeHasLabel (S, I+1, 6) &&
!CS_RangeHasLabel (S, I+8, 4) &&
/* 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;
/* We will create all the new stuff behind the current one so
* we keep the line references.
*/
X = NewCodeEntry (OP65_LDY, L[2]->AM, L[2]->Arg, 0, L[2]->LI);
CS_InsertEntry (S, X, I+12);
X = NewCodeEntry (OP65_LDX, L[8]->AM, L[8]->Arg, 0, L[8]->LI);
CS_InsertEntry (S, X, I+13);
X = NewCodeEntry (OP65_LDA, L[4]->AM, L[4]->Arg, 0, L[4]->LI);
CS_InsertEntry (S, X, I+14);
X = NewCodeEntry (OP65_TAY, AM65_IMP, 0, 0, L[4]->LI);
CS_InsertEntry (S, X, I+15);
X = NewCodeEntry (OP65_LDA, L[9]->AM, L[9]->Arg, 0, L[9]->LI);
CS_InsertEntry (S, X, I+16);
Label = memcpy (xmalloc (Len-2), L[0]->Arg+2, Len-3);
Label[Len-3] = '\0';
X = NewCodeEntry (OP65_STA, AM65_ABSY, Label, 0, L[11]->LI);
CS_InsertEntry (S, X, I+17);
xfree (Label);
/* Remove the old code */
CS_DelEntries (S, I, 12);
/* Remember, we had changes */
++Changes;
}
/* Next entry */
++I;
}
/* Return the number of changes made */
return Changes;
}
unsigned OptPtrStore3 (CodeSeg* S)
/* Search for the sequence:
*
* lda #<(label+0)
* ldx #>(label+0)
* ldy aaa
* clc
* adc (sp),y
* bcc L
* inx
* L: jsr pushax
* ldy #bbb
* ldx #$00
* lda (sp),y
* ldy #$00
* jsr staspidx
*
* and replace it by:
*
* ldy aaa
* lda (sp),y
* tax
* ldy #bbb-2
* lda (sp),y
* sta label,x
* ldx #$00
*/
{
unsigned Changes = 0;
/* Walk over the entries */
unsigned I = 0;
while (I < CS_GetEntryCount (S)) {
CodeEntry* L[13];
unsigned Len;
/* Get next entry */
L[0] = CS_GetEntry (S, I);
/* Check for the sequence */
if (L[0]->OPC == OP65_LDA &&
L[0]->AM == AM65_IMM &&
CS_GetEntries (S, L+1, I+1, 12) &&
L[1]->OPC == OP65_LDX &&
L[1]->AM == AM65_IMM &&
L[2]->OPC == OP65_LDY &&
L[3]->OPC == OP65_CLC &&
L[4]->OPC == OP65_ADC &&
L[4]->AM == AM65_ZP_INDY &&
(L[5]->OPC == OP65_BCC || L[5]->OPC == OP65_JCC) &&
L[5]->JumpTo != 0 &&
L[5]->JumpTo->Owner == L[7] &&
L[6]->OPC == OP65_INX &&
CE_IsCallTo (L[7], "pushax") &&
L[8]->OPC == OP65_LDY &&
CE_IsConstImm (L[8]) &&
L[8]->Num >= 2 &&
L[9]->OPC == OP65_LDX &&
L[10]->OPC == OP65_LDA &&
L[10]->AM == AM65_ZP_INDY &&
L[11]->OPC == OP65_LDY &&
CE_IsKnownImm (L[11], 0) &&
CE_IsCallTo (L[12], "staspidx") &&
!CS_RangeHasLabel (S, I+1, 6) &&
!CS_RangeHasLabel (S, I+8, 5) &&
/* 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;
const char* Arg;
char* Label;
/* We will create all the new stuff behind the current one so
* we keep the line references.
*/
X = NewCodeEntry (OP65_LDY, L[2]->AM, L[2]->Arg, 0, L[2]->LI);
CS_InsertEntry (S, X, I+13);
X = NewCodeEntry (OP65_LDA, L[4]->AM, L[4]->Arg, 0, L[4]->LI);
CS_InsertEntry (S, X, I+14);
X = NewCodeEntry (OP65_TAX, AM65_IMP, 0, 0, L[4]->LI);
CS_InsertEntry (S, X, I+15);
Arg = MakeHexArg (L[8]->Num - 2);
X = NewCodeEntry (OP65_LDY, L[8]->AM, Arg, 0, L[8]->LI);
CS_InsertEntry (S, X, I+16);
X = NewCodeEntry (OP65_LDA, L[10]->AM, L[10]->Arg, 0, L[10]->LI);
CS_InsertEntry (S, X, I+17);
Label = memcpy (xmalloc (Len-2), L[0]->Arg+2, Len-3);
Label[Len-3] = '\0';
X = NewCodeEntry (OP65_STA, AM65_ABSX, Label, 0, L[12]->LI);
CS_InsertEntry (S, X, I+18);
xfree (Label);
X = NewCodeEntry (OP65_LDX, L[9]->AM, L[9]->Arg, 0, L[9]->LI);
CS_InsertEntry (S, X, I+19);
/* Remove the old code */
CS_DelEntries (S, I, 13);
/* Remember, we had changes */
++Changes;
}
/* Next entry */
++I;
}
/* Return the number of changes made */
return Changes;
}
unsigned OptPtrStore4 (CodeSeg* S)
/* Search for the sequence: /* Search for the sequence:
* *
* clc * clc
@@ -616,14 +412,41 @@ unsigned OptPtrStore4 (CodeSeg* S)
* ldy xxx * ldy xxx
* sta (ptr1),y * sta (ptr1),y
* *
* In case a/x is loaded from the register bank before the clc, we can even * or by
* use the register bank instead of ptr1. *
* ldy yyy-2
* ldx #$00
* lda (sp),y
* ldy xxx
* sta (zp),y
*
* or by
*
* ldy yyy-2
* ldx #$00
* lda (sp),y
* ldy xxx
* sta label,y
*
* or by
*
* ldy yyy-2
* ldx #$00
* lda (sp),y
* ldy xxx
* sta $xxxx,y
*
* depending on the code preceeding the sequence above.
*/ */
{ {
unsigned Changes = 0; unsigned Changes = 0;
unsigned I;
/* Generate register info */
CS_GenRegInfo (S);
/* Walk over the entries */ /* Walk over the entries */
unsigned I = 0; I = 0;
while (I < CS_GetEntryCount (S)) { while (I < CS_GetEntryCount (S)) {
CodeEntry* L[10]; CodeEntry* L[10];
@@ -637,7 +460,9 @@ unsigned OptPtrStore4 (CodeSeg* S)
L[1]->OPC == OP65_ADC && L[1]->OPC == OP65_ADC &&
(L[1]->AM == AM65_ABS || (L[1]->AM == AM65_ABS ||
L[1]->AM == AM65_ZP || L[1]->AM == AM65_ZP ||
L[1]->AM == AM65_IMM) && L[1]->AM == AM65_IMM ||
(L[1]->AM == AM65_ZP_INDY &&
RegValIsKnown (L[1]->RI->In.RegY))) &&
(L[2]->OPC == OP65_BCC || L[2]->OPC == OP65_JCC) && (L[2]->OPC == OP65_BCC || L[2]->OPC == OP65_JCC) &&
L[2]->JumpTo != 0 && L[2]->JumpTo != 0 &&
L[2]->JumpTo->Owner == L[4] && L[2]->JumpTo->Owner == L[4] &&
@@ -664,24 +489,21 @@ unsigned OptPtrStore4 (CodeSeg* S)
/* Track the insertion point */ /* Track the insertion point */
unsigned IP = I + 10; unsigned IP = I + 10;
if ((Loc = LoadAXZP (S, I)) != 0) {
if (I >= 2) {
if ((Loc = LoadAXZP (S, I-2)) != 0) {
/* If the sequence is preceeded by a load of a ZP value, /* If the sequence is preceeded by a load of a ZP value,
* we can use this ZP value as a pointer using ZP * we can use this ZP value as a pointer using ZP
* indirect Y addressing. * indirect Y addressing.
*/ */
AM = AM65_ZP_INDY; AM = AM65_ZP_INDY;
} else if ((Loc = LoadAXImm (S, I-2)) != 0) { } else if ((Loc = LoadAXImm (S, I)) != 0) {
/* If the sequence is preceeded by a load of an immediate /* If the sequence is preceeded by a load of an immediate
* value, we can use this absolute value as an address * value, we can use this absolute value as an address
* using absolute indexed Y addressing. * using absolute indexed Y addressing.
*/ */
AM = AM65_ABSY; AM = AM65_ABSY;
} }
}
/* If we don't have a zero page location, we use ptr1 with zp /* If we don't have a store location, we use ptr1 with zp
* indirect Y addressing. We must store the value in A/X into * indirect Y addressing. We must store the value in A/X into
* ptr1 in this case. * ptr1 in this case.
*/ */
@@ -699,6 +521,24 @@ unsigned OptPtrStore4 (CodeSeg* S)
} }
/* Generate four different replacements depending on the addressing
* mode of the store and from where the index is loaded:
*
* 1. If the index is not loaded ZP indirect Y, we can use Y for
* the store index.
*
* 2. If the index is loaded ZP indirect Y and we store absolute
* indexed, we need Y to load the index and will therefore
* use X as index for the store. The disadvantage is that we
* need to reload X later.
*
* 3. If the index is loaded ZP indirect Y and we store ZP indirect
* Y, we must use Y for load and store and must therefore save
* the A register when loading Y the second time.
*/
if (L[1]->AM != AM65_ZP_INDY) {
/* Case 1 */
Arg = MakeHexArg (L[5]->Num - 2); Arg = MakeHexArg (L[5]->Num - 2);
X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, L[5]->LI); X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, L[5]->LI);
CS_InsertEntry (S, X, IP++); CS_InsertEntry (S, X, IP++);
@@ -715,9 +555,68 @@ unsigned OptPtrStore4 (CodeSeg* S)
X = NewCodeEntry (OP65_STA, AM, Loc, 0, L[9]->LI); X = NewCodeEntry (OP65_STA, AM, Loc, 0, L[9]->LI);
CS_InsertEntry (S, X, IP++); CS_InsertEntry (S, X, IP++);
} else if (AM == AM65_ABSY) {
/* Case 2 */
X = NewCodeEntry (OP65_LDA, L[1]->AM, L[1]->Arg, 0, L[1]->LI);
CS_InsertEntry (S, X, IP++);
X = NewCodeEntry (OP65_TAX, AM65_IMP, 0, 0, L[1]->LI);
CS_InsertEntry (S, X, IP++);
Arg = MakeHexArg (L[5]->Num - 2);
X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, L[5]->LI);
CS_InsertEntry (S, X, IP++);
X = NewCodeEntry (OP65_LDA, L[7]->AM, L[7]->Arg, 0, L[7]->LI);
CS_InsertEntry (S, X, IP++);
X = NewCodeEntry (OP65_STA, AM65_ABSX, Loc, 0, L[9]->LI);
CS_InsertEntry (S, X, IP++);
X = NewCodeEntry (OP65_LDX, L[6]->AM, L[6]->Arg, 0, L[6]->LI);
CS_InsertEntry (S, X, IP++);
} else {
/* Case 3 */
Arg = MakeHexArg (L[5]->Num - 2);
X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, L[5]->LI);
CS_InsertEntry (S, X, IP++);
X = NewCodeEntry (OP65_LDX, L[6]->AM, L[6]->Arg, 0, L[6]->LI);
CS_InsertEntry (S, X, IP++);
X = NewCodeEntry (OP65_LDA, L[7]->AM, L[7]->Arg, 0, L[7]->LI);
CS_InsertEntry (S, X, IP++);
X = NewCodeEntry (OP65_PHA, AM65_IMP, 0, 0, L[6]->LI);
CS_InsertEntry (S, X, IP++);
Arg = MakeHexArg (L[1]->RI->In.RegY);
X = NewCodeEntry (OP65_LDY, AM65_IMM, Arg, 0, L[1]->LI);
CS_InsertEntry (S, X, IP++);
X = NewCodeEntry (OP65_LDA, L[1]->AM, L[1]->Arg, 0, L[1]->LI);
CS_InsertEntry (S, X, IP++);
X = NewCodeEntry (OP65_TAY, AM65_IMP, 0, 0, L[1]->LI);
CS_InsertEntry (S, X, IP++);
X = NewCodeEntry (OP65_PLA, AM65_IMP, 0, 0, L[6]->LI);
CS_InsertEntry (S, X, IP++);
X = NewCodeEntry (OP65_STA, AM, Loc, 0, L[9]->LI);
CS_InsertEntry (S, X, IP++);
}
/* Remove the old code */ /* Remove the old code */
CS_DelEntries (S, I, 10); CS_DelEntries (S, I, 10);
/* Skip most of the generated replacement code */
I += 4;
/* Remember, we had changes */ /* Remember, we had changes */
++Changes; ++Changes;
@@ -728,13 +627,16 @@ unsigned OptPtrStore4 (CodeSeg* S)
} }
/* Free the register info */
CS_FreeRegInfo (S);
/* Return the number of changes made */ /* Return the number of changes made */
return Changes; return Changes;
} }
unsigned OptPtrStore5 (CodeSeg* S) unsigned OptPtrStore3 (CodeSeg* S)
/* Search for the sequence: /* Search for the sequence:
* *
* jsr pushax * jsr pushax

97
src/cc65/coptptrstore.h
View File

@@ -76,6 +76,13 @@ unsigned OptPtrStore1 (CodeSeg* S);
* ldy xxx * ldy xxx
* ldx #$00 * ldx #$00
* lda yyy * lda yyy
* sta (zp),y
*
* or by
*
* ldy xxx
* ldx #$00
* lda yyy
* sta label,y * sta label,y
* *
* or by * or by
@@ -85,71 +92,10 @@ unsigned OptPtrStore1 (CodeSeg* S);
* lda yyy * lda yyy
* sta $xxxx,y * sta $xxxx,y
* *
* or by
*
* ldy xxx
* ldx #$00
* lda yyy
* sta (zp),y
*
* depending on the two instructions preceeding the sequence above. * depending on the two instructions preceeding the sequence above.
*/ */
unsigned OptPtrStore2 (CodeSeg* S); unsigned OptPtrStore2 (CodeSeg* S);
/* Search for the sequence:
*
* lda #<(label+0)
* ldx #>(label+0)
* ldy aaa
* clc
* adc (sp),y
* bcc L
* inx
* L: jsr pushax
* ldx #$00
* lda yyy
* ldy #$00
* jsr staspidx
*
* and replace it by:
*
* ldy aaa
* ldx #$00
* lda (sp),y
* tay
* lda yyy
* sta label,y
*/
unsigned OptPtrStore3 (CodeSeg* S);
/* Search for the sequence:
*
* lda #<(label+0)
* ldx #>(label+0)
* ldy aaa
* clc
* adc (sp),y
* bcc L
* inx
* L: jsr pushax
* ldy #bbb
* ldx #$00
* lda (sp),y
* ldy #$00
* jsr staspidx
*
* and replace it by:
*
* ldy aaa
* lda (sp),y
* tax
* ldy #bbb-2
* lda (sp),y
* sta label,x
* ldx #$00
*/
unsigned OptPtrStore4 (CodeSeg* S);
/* Search for the sequence: /* Search for the sequence:
* *
* clc * clc
@@ -173,11 +119,34 @@ unsigned OptPtrStore4 (CodeSeg* S);
* ldy xxx * ldy xxx
* sta (ptr1),y * sta (ptr1),y
* *
* In case a/x is loaded from the register bank before the clc, we can even * or by
* use the register bank instead of ptr1. *
* ldy yyy-2
* ldx #$00
* lda (sp),y
* ldy xxx
* sta (zp),y
*
* or by
*
* ldy yyy-2
* ldx #$00
* lda (sp),y
* ldy xxx
* sta label,y
*
* or by
*
* ldy yyy-2
* ldx #$00
* lda (sp),y
* ldy xxx
* sta $xxxx,y
*
* depending on the code preceeding the sequence above.
*/ */
unsigned OptPtrStore5 (CodeSeg* S); unsigned OptPtrStore3 (CodeSeg* S);
/* Search for the sequence: /* Search for the sequence:
* *
* jsr pushax * jsr pushax