1
0
mirror of https://github.com/cc65/cc65.git synced 2024-11-19 06:31:31 +00:00

Another optimization for array stores.

git-svn-id: svn://svn.cc65.org/cc65/trunk@5715 b7a2c559-68d2-44c3-8de9-860c34a00d81
This commit is contained in:
uz 2012-06-12 18:11:53 +00:00
parent 383157e743
commit 9142a300c1
3 changed files with 161 additions and 24 deletions

View File

@ -1204,7 +1204,8 @@ static OptFunc DOptPtrStore1 = { OptPtrStore1, "OptPtrStore1", 40, 0,
static OptFunc DOptPtrStore2 = { OptPtrStore2, "OptPtrStore2", 50, 0, 0, 0, 0, 0 };
static OptFunc DOptPtrStore3 = { OptPtrStore3, "OptPtrStore3", 50, 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 DOptPtrStore5 = { OptPtrStore5, "OptPtrStore5", 65, 0, 0, 0, 0, 0 };
static OptFunc DOptPtrStore6 = { OptPtrStore6, "OptPtrStore6", 100, 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 DOptPushPop = { OptPushPop, "OptPushPop", 0, 0, 0, 0, 0, 0 };
@ -1299,6 +1300,7 @@ static OptFunc* OptFuncs[] = {
&DOptPtrStore3,
&DOptPtrStore4,
&DOptPtrStore5,
&DOptPtrStore6,
&DOptPush1,
&DOptPush2,
&DOptPushPop,
@ -1574,6 +1576,7 @@ static unsigned RunOptGroup1 (CodeSeg* S)
Changes += RunOptFunc (S, &DOptPtrStore3, 1);
Changes += RunOptFunc (S, &DOptPtrStore4, 1);
Changes += RunOptFunc (S, &DOptPtrStore5, 1);
Changes += RunOptFunc (S, &DOptPtrStore6, 1);
Changes += RunOptFunc (S, &DOptAdd3, 1); /* Before OptPtrLoad5! */
Changes += RunOptFunc (S, &DOptPtrLoad1, 1);
Changes += RunOptFunc (S, &DOptPtrLoad2, 1);

View File

@ -491,6 +491,9 @@ unsigned OptPtrStore4 (CodeSeg* S)
* ldx #$00
* lda yyy
* sta (ptr1),y
*
* In case a/x is loaded from the register bank before the clc, we can even
* use the register bank instead of ptr1.
*/
{
unsigned Changes = 0;
@ -526,7 +529,7 @@ unsigned OptPtrStore4 (CodeSeg* S)
CodeEntry* X;
const char* ZPLoc;
/* Track the insertion point */
unsigned IP = I + 9;
@ -567,7 +570,7 @@ unsigned OptPtrStore4 (CodeSeg* S)
}
/* Next entry */
++I;
++I;
}
@ -578,6 +581,129 @@ unsigned OptPtrStore4 (CodeSeg* S)
unsigned OptPtrStore5 (CodeSeg* S)
/* Search for the sequence:
*
* clc
* adc xxx
* bcc L
* inx
* L: jsr pushax
* ldy yyy
* ldx #$00
* lda (sp),y
* ldy #$00
* jsr staspidx
*
* and replace it by:
*
* sta ptr1
* stx ptr1+1
* ldy yyy-2
* ldx #$00
* lda (sp),y
* ldy xxx
* sta (ptr1),y
*
* In case a/x is loaded from the register bank before the clc, we can even
* use the register bank instead of ptr1.
*/
{
unsigned Changes = 0;
/* Walk over the entries */
unsigned I = 0;
while (I < CS_GetEntryCount (S)) {
CodeEntry* L[10];
/* Get next entry */
L[0] = CS_GetEntry (S, I);
/* Check for the sequence */
if (L[0]->OPC == OP65_CLC &&
CS_GetEntries (S, L+1, I+1, 9) &&
L[1]->OPC == OP65_ADC &&
(L[1]->AM == AM65_ABS ||
L[1]->AM == AM65_ZP ||
L[1]->AM == AM65_IMM) &&
(L[2]->OPC == OP65_BCC || L[2]->OPC == OP65_JCC) &&
L[2]->JumpTo != 0 &&
L[2]->JumpTo->Owner == L[4] &&
L[3]->OPC == OP65_INX &&
CE_IsCallTo (L[4], "pushax") &&
L[5]->OPC == OP65_LDY &&
CE_IsConstImm (L[5]) &&
L[6]->OPC == OP65_LDX &&
L[7]->OPC == OP65_LDA &&
L[7]->AM == AM65_ZP_INDY &&
strcmp (L[7]->Arg, "sp") == 0 &&
L[8]->OPC == OP65_LDY &&
(L[8]->AM == AM65_ABS ||
L[8]->AM == AM65_ZP ||
L[8]->AM == AM65_IMM) &&
CE_IsCallTo (L[9], "staspidx") &&
!CS_RangeHasLabel (S, I+1, 3) &&
!CS_RangeHasLabel (S, I+5, 5)) {
CodeEntry* X;
const char* Arg;
const char* ZPLoc;
/* Track the insertion point */
unsigned IP = I + 10;
/* If the sequence is preceeded by a load of a ZP value, we can
* use this ZP value as a pointer.
*/
if (I < 2 || (ZPLoc = ZPLoadAX (S, I-2)) == 0) {
ZPLoc = "ptr1";
/* Must use ptr1 */
X = NewCodeEntry (OP65_STA, AM65_ZP, "ptr1", 0, L[9]->LI);
CS_InsertEntry (S, X, IP++);
X = NewCodeEntry (OP65_STX, AM65_ZP, "ptr1+1", 0, L[9]->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_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_LDY, L[1]->AM, L[1]->Arg, 0, L[1]->LI);
CS_InsertEntry (S, X, IP++);
X = NewCodeEntry (OP65_STA, AM65_ZP_INDY, ZPLoc, 0, L[9]->LI);
CS_InsertEntry (S, X, IP++);
/* Remove the old code */
CS_DelEntries (S, I, 10);
/* Remember, we had changes */
++Changes;
}
/* Next entry */
++I;
}
/* Return the number of changes made */
return Changes;
}
unsigned OptPtrStore6 (CodeSeg* S)
/* Search for the sequence:
*
* jsr pushax
@ -601,16 +727,6 @@ unsigned OptPtrStore5 (CodeSeg* S)
* In case a/x is loaded from the register bank before the pushax, we can even
* use the register bank instead of ptr1.
*
*
* jsr pushax
* ldy xxx
* jsr ldauidx
* ldx #$00
* lda (zp),y
* subop
* ldy yyy
* sta (zp),y
* jsr staspidx
*/
{
unsigned Changes = 0;

View File

@ -153,6 +153,34 @@ unsigned OptPtrStore4 (CodeSeg* S);
*/
unsigned OptPtrStore5 (CodeSeg* S);
/* Search for the sequence:
*
* clc
* adc xxx
* bcc L
* inx
* L: jsr pushax
* ldy yyy
* ldx #$00
* lda (sp),y
* ldy #$00
* jsr staspidx
*
* and replace it by:
*
* sta ptr1
* stx ptr1+1
* ldy yyy-2
* ldx #$00
* lda (sp),y
* ldy xxx
* sta (ptr1),y
*
* In case a/x is loaded from the register bank before the clc, we can even
* use the register bank instead of ptr1.
*/
unsigned OptPtrStore6 (CodeSeg* S);
/* Search for the sequence:
*
* jsr pushax
@ -176,16 +204,6 @@ unsigned OptPtrStore5 (CodeSeg* S);
* In case a/x is loaded from the register bank before the pushax, we can even
* use the register bank instead of ptr1.
*
*
* jsr pushax
* ldy xxx
* jsr ldauidx
* ldx #$00
* lda (zp),y
* subop
* ldy yyy
* sta (zp),y
* jsr staspidx
*/
@ -195,4 +213,4 @@ unsigned OptPtrStore5 (CodeSeg* S);