|
|
@@ -24,7 +24,7 @@
|
|
|
|
|
|
{$define FPC_SYSTEM_HAS_MOVE}
|
|
|
|
|
|
-procedure Move(var sou{}rce;var dest;count:longint);assembler;
|
|
|
+procedure Move(var source;var dest;count:longint);assembler;
|
|
|
asm
|
|
|
{ count <= 0 ? }
|
|
|
cmpwi cr0,r5,0
|
|
|
@@ -32,19 +32,19 @@ asm
|
|
|
sub r30,r4,r3
|
|
|
{ carry := boolean(dest-source < count) = boolean(overlap) }
|
|
|
subc r30,r30,r5
|
|
|
-
|
|
|
+
|
|
|
{ count < 11 ? (to decide whether we will move dwords or bytes }
|
|
|
cmpwi cr1,r5,11
|
|
|
-
|
|
|
+
|
|
|
{ if overlap, then r30 := -1 else r30 := 0 }
|
|
|
subfe r30,r30,r30
|
|
|
-
|
|
|
+
|
|
|
{ count < 39 ? (32 + max. alignment (7) }
|
|
|
cmpwi cr7,r5,39
|
|
|
-
|
|
|
+
|
|
|
{ if count <= 0, stop }
|
|
|
ble cr0,LMoveDone
|
|
|
-
|
|
|
+
|
|
|
{ if overlap, then r29 := count else r29 := 0 }
|
|
|
and r29,r5,r30
|
|
|
{ if overlap, then point source and dest to the end }
|
|
|
@@ -85,12 +85,17 @@ LMove4ByteAlignLoop:
|
|
|
{ if 11 <= count < 39, copy using dwords }
|
|
|
blt cr7,LMoveDWords
|
|
|
|
|
|
+ { multiply the update count with 4 }
|
|
|
+ slwi r30,r30,2
|
|
|
+
|
|
|
beq cr0,L8BytesAligned
|
|
|
-
|
|
|
+
|
|
|
{ count >= 39 -> align to 8 byte boundary and then use the FPU }
|
|
|
{ since we're already at 4 byte alignment, use dword store }
|
|
|
- lwzux r29,r3,r30
|
|
|
- stwux r29,r4,r30
|
|
|
+ lwz r29,0(r3)
|
|
|
+ add r3,r3,r30,
|
|
|
+ stw r29,0(r4)
|
|
|
+ add r4,r4,r30,
|
|
|
L8BytesAligned:
|
|
|
{ count div 32 ( >= 1, since count was >=39 }
|
|
|
srwi r29,r5,5
|
|
|
@@ -99,13 +104,13 @@ L8BytesAligned:
|
|
|
{ to decide if we will do some dword stores afterwards or not }
|
|
|
cmpwi cr1,r5,11
|
|
|
mtctr r29
|
|
|
-
|
|
|
+
|
|
|
{ r29 := count div 4, will be moved to ctr when copying dwords }
|
|
|
srwi r29,r5,2
|
|
|
-
|
|
|
+
|
|
|
{ adjust the update count: it will now be 8 or -8 depending on overlap }
|
|
|
- slwi r30,r30,3
|
|
|
-
|
|
|
+ slwi r30,r30,1
|
|
|
+
|
|
|
{ adjust source and dest pointers: because of the above loop, dest is now }
|
|
|
{ aligned to 8 bytes. So if we substract r30 we will still have an 8 bytes }
|
|
|
{ aligned address) }
|
|
|
@@ -127,8 +132,8 @@ LMove32ByteLoop:
|
|
|
beq cr0,LMoveDone
|
|
|
|
|
|
{ make r30 again -1 or 1, but first adjust source/dest pointers }
|
|
|
- add r3,r3,r30
|
|
|
- add r4,r4,r30
|
|
|
+ add r3,r3,r30
|
|
|
+ add r4,r4,r30
|
|
|
srawi r30,r30,3
|
|
|
sub r3,r3,r30
|
|
|
sub r4,r4,r30
|
|
|
@@ -143,8 +148,8 @@ LMoveDWords:
|
|
|
andi. r5,r5,3
|
|
|
{ r30 * 4 }
|
|
|
slwi r30,r30,2
|
|
|
- sub r3,r3,r30
|
|
|
- sub r4,r4,r30
|
|
|
+ sub r3,r3,r30
|
|
|
+ sub r4,r4,r30
|
|
|
|
|
|
LMoveDWordsLoop:
|
|
|
lwzux r29,r3,r30
|
|
|
@@ -153,8 +158,8 @@ LMoveDWordsLoop:
|
|
|
|
|
|
beq cr0,LMoveDone
|
|
|
{ make r30 again -1 or 1 }
|
|
|
- add r3,r3,r30
|
|
|
- add r4,r4,r30
|
|
|
+ add r3,r3,r30
|
|
|
+ add r4,r4,r30
|
|
|
srawi r30,r30,2
|
|
|
sub r3,r3,r30
|
|
|
sub r4,r4,r30
|
|
|
@@ -433,7 +438,7 @@ asm
|
|
|
subme r3,r3,r3 { if r3 >= r4 then r3' := 0 else r3' := -1 }
|
|
|
and r3,r29,r3 { if r3 >= r4 then r3' := 0 else r3' := r3-r30 }
|
|
|
add r3,r3,r30 { if r3 >= r4 then r3' := r30 else r3' := r3 }
|
|
|
-
|
|
|
+
|
|
|
cmpli r3,0
|
|
|
{ put length in ctr }
|
|
|
mtctr r3
|
|
|
@@ -448,7 +453,10 @@ end ['r3','r4','r5','r29','r30','cr0','ctr'];
|
|
|
|
|
|
{
|
|
|
$Log$
|
|
|
- Revision 1.3 2001-03-02 13:24:10 jonas
|
|
|
+ Revision 1.4 2001-03-03 13:53:36 jonas
|
|
|
+ * fixed small bug in move
|
|
|
+
|
|
|
+ Revision 1.3 2001/03/02 13:24:10 jonas
|
|
|
+ new, complete implementation of move procedure (including support for
|
|
|
overlapping regions)
|
|
|
|
Jonas Maebe