aboutsummaryrefslogtreecommitdiffstats
path: root/arch/alpha/lib/ev6-copy_user.S
diff options
context:
space:
mode:
Diffstat (limited to 'arch/alpha/lib/ev6-copy_user.S')
-rw-r--r--arch/alpha/lib/ev6-copy_user.S104
1 files changed, 45 insertions, 59 deletions
diff --git a/arch/alpha/lib/ev6-copy_user.S b/arch/alpha/lib/ev6-copy_user.S
index be720b518af9..35e6710d0700 100644
--- a/arch/alpha/lib/ev6-copy_user.S
+++ b/arch/alpha/lib/ev6-copy_user.S
@@ -12,21 +12,6 @@
* only _after_ a successful copy). There is also some rather minor
* exception setup stuff..
*
- * NOTE! This is not directly C-callable, because the calling semantics are
- * different:
- *
- * Inputs:
- * length in $0
- * destination address in $6
- * source address in $7
- * return address in $28
- *
- * Outputs:
- * bytes left to copy in $0
- *
- * Clobbers:
- * $1,$2,$3,$4,$5,$6,$7
- *
* Much of the information about 21264 scheduling/coding comes from:
* Compiler Writer's Guide for the Alpha 21264
* abbreviated as 'CWG' in other comments here
@@ -60,10 +45,11 @@
# Pipeline info: Slotting & Comments
__copy_user:
.prologue 0
- subq $0, 32, $1 # .. E .. .. : Is this going to be a small copy?
+ andq $18, $18, $0
+ subq $18, 32, $1 # .. E .. .. : Is this going to be a small copy?
beq $0, $zerolength # U .. .. .. : U L U L
- and $6,7,$3 # .. .. .. E : is leading dest misalignment
+ and $16,7,$3 # .. .. .. E : is leading dest misalignment
ble $1, $onebyteloop # .. .. U .. : 1st branch : small amount of data
beq $3, $destaligned # .. U .. .. : 2nd (one cycle fetcher stall)
subq $3, 8, $3 # E .. .. .. : L U U L : trip counter
@@ -73,17 +59,17 @@ __copy_user:
* We know we have at least one trip through this loop
*/
$aligndest:
- EXI( ldbu $1,0($7) ) # .. .. .. L : Keep loads separate from stores
- addq $6,1,$6 # .. .. E .. : Section 3.8 in the CWG
+ EXI( ldbu $1,0($17) ) # .. .. .. L : Keep loads separate from stores
+ addq $16,1,$16 # .. .. E .. : Section 3.8 in the CWG
addq $3,1,$3 # .. E .. .. :
nop # E .. .. .. : U L U L
/*
- * the -1 is to compensate for the inc($6) done in a previous quadpack
+ * the -1 is to compensate for the inc($16) done in a previous quadpack
* which allows us zero dependencies within either quadpack in the loop
*/
- EXO( stb $1,-1($6) ) # .. .. .. L :
- addq $7,1,$7 # .. .. E .. : Section 3.8 in the CWG
+ EXO( stb $1,-1($16) ) # .. .. .. L :
+ addq $17,1,$17 # .. .. E .. : Section 3.8 in the CWG
subq $0,1,$0 # .. E .. .. :
bne $3, $aligndest # U .. .. .. : U L U L
@@ -92,29 +78,29 @@ $aligndest:
* If we arrived via branch, we have a minimum of 32 bytes
*/
$destaligned:
- and $7,7,$1 # .. .. .. E : Check _current_ source alignment
+ and $17,7,$1 # .. .. .. E : Check _current_ source alignment
bic $0,7,$4 # .. .. E .. : number bytes as a quadword loop
- EXI( ldq_u $3,0($7) ) # .. L .. .. : Forward fetch for fallthrough code
+ EXI( ldq_u $3,0($17) ) # .. L .. .. : Forward fetch for fallthrough code
beq $1,$quadaligned # U .. .. .. : U L U L
/*
- * In the worst case, we've just executed an ldq_u here from 0($7)
+ * In the worst case, we've just executed an ldq_u here from 0($17)
* and we'll repeat it once if we take the branch
*/
/* Misaligned quadword loop - not unrolled. Leave it that way. */
$misquad:
- EXI( ldq_u $2,8($7) ) # .. .. .. L :
+ EXI( ldq_u $2,8($17) ) # .. .. .. L :
subq $4,8,$4 # .. .. E .. :
- extql $3,$7,$3 # .. U .. .. :
- extqh $2,$7,$1 # U .. .. .. : U U L L
+ extql $3,$17,$3 # .. U .. .. :
+ extqh $2,$17,$1 # U .. .. .. : U U L L
bis $3,$1,$1 # .. .. .. E :
- EXO( stq $1,0($6) ) # .. .. L .. :
- addq $7,8,$7 # .. E .. .. :
+ EXO( stq $1,0($16) ) # .. .. L .. :
+ addq $17,8,$17 # .. E .. .. :
subq $0,8,$0 # E .. .. .. : U L L U
- addq $6,8,$6 # .. .. .. E :
+ addq $16,8,$16 # .. .. .. E :
bis $2,$2,$3 # .. .. E .. :
nop # .. E .. .. :
bne $4,$misquad # U .. .. .. : U L U L
@@ -125,8 +111,8 @@ $misquad:
beq $0,$zerolength # U .. .. .. : U L U L
/* We know we have at least one trip through the byte loop */
- EXI ( ldbu $2,0($7) ) # .. .. .. L : No loads in the same quad
- addq $6,1,$6 # .. .. E .. : as the store (Section 3.8 in CWG)
+ EXI ( ldbu $2,0($17) ) # .. .. .. L : No loads in the same quad
+ addq $16,1,$16 # .. .. E .. : as the store (Section 3.8 in CWG)
nop # .. E .. .. :
br $31, $dirtyentry # L0 .. .. .. : L U U L
/* Do the trailing byte loop load, then hop into the store part of the loop */
@@ -136,8 +122,8 @@ $misquad:
* Based upon the usage context, it's worth the effort to unroll this loop
* $0 - number of bytes to be moved
* $4 - number of bytes to move as quadwords
- * $6 is current destination address
- * $7 is current source address
+ * $16 is current destination address
+ * $17 is current source address
*/
$quadaligned:
subq $4, 32, $2 # .. .. .. E : do not unroll for small stuff
@@ -155,29 +141,29 @@ $quadaligned:
* instruction memory hint instruction).
*/
$unroll4:
- EXI( ldq $1,0($7) ) # .. .. .. L
- EXI( ldq $2,8($7) ) # .. .. L ..
+ EXI( ldq $1,0($17) ) # .. .. .. L
+ EXI( ldq $2,8($17) ) # .. .. L ..
subq $4,32,$4 # .. E .. ..
nop # E .. .. .. : U U L L
- addq $7,16,$7 # .. .. .. E
- EXO( stq $1,0($6) ) # .. .. L ..
- EXO( stq $2,8($6) ) # .. L .. ..
+ addq $17,16,$17 # .. .. .. E
+ EXO( stq $1,0($16) ) # .. .. L ..
+ EXO( stq $2,8($16) ) # .. L .. ..
subq $0,16,$0 # E .. .. .. : U L L U
- addq $6,16,$6 # .. .. .. E
- EXI( ldq $1,0($7) ) # .. .. L ..
- EXI( ldq $2,8($7) ) # .. L .. ..
+ addq $16,16,$16 # .. .. .. E
+ EXI( ldq $1,0($17) ) # .. .. L ..
+ EXI( ldq $2,8($17) ) # .. L .. ..
subq $4, 32, $3 # E .. .. .. : U U L L : is there enough for another trip?
- EXO( stq $1,0($6) ) # .. .. .. L
- EXO( stq $2,8($6) ) # .. .. L ..
+ EXO( stq $1,0($16) ) # .. .. .. L
+ EXO( stq $2,8($16) ) # .. .. L ..
subq $0,16,$0 # .. E .. ..
- addq $7,16,$7 # E .. .. .. : U L L U
+ addq $17,16,$17 # E .. .. .. : U L L U
nop # .. .. .. E
nop # .. .. E ..
- addq $6,16,$6 # .. E .. ..
+ addq $16,16,$16 # .. E .. ..
bgt $3,$unroll4 # U .. .. .. : U L U L
nop
@@ -186,14 +172,14 @@ $unroll4:
beq $4, $noquads
$onequad:
- EXI( ldq $1,0($7) )
+ EXI( ldq $1,0($17) )
subq $4,8,$4
- addq $7,8,$7
+ addq $17,8,$17
nop
- EXO( stq $1,0($6) )
+ EXO( stq $1,0($16) )
subq $0,8,$0
- addq $6,8,$6
+ addq $16,8,$16
bne $4,$onequad
$noquads:
@@ -207,23 +193,23 @@ $noquads:
* There's no point in doing a lot of complex alignment calculations to try to
* to quadword stuff for a small amount of data.
* $0 - remaining number of bytes left to copy
- * $6 - current dest addr
- * $7 - current source addr
+ * $16 - current dest addr
+ * $17 - current source addr
*/
$onebyteloop:
- EXI ( ldbu $2,0($7) ) # .. .. .. L : No loads in the same quad
- addq $6,1,$6 # .. .. E .. : as the store (Section 3.8 in CWG)
+ EXI ( ldbu $2,0($17) ) # .. .. .. L : No loads in the same quad
+ addq $16,1,$16 # .. .. E .. : as the store (Section 3.8 in CWG)
nop # .. E .. .. :
nop # E .. .. .. : U L U L
$dirtyentry:
/*
- * the -1 is to compensate for the inc($6) done in a previous quadpack
+ * the -1 is to compensate for the inc($16) done in a previous quadpack
* which allows us zero dependencies within either quadpack in the loop
*/
- EXO ( stb $2,-1($6) ) # .. .. .. L :
- addq $7,1,$7 # .. .. E .. : quadpack as the load
+ EXO ( stb $2,-1($16) ) # .. .. .. L :
+ addq $17,1,$17 # .. .. E .. : quadpack as the load
subq $0,1,$0 # .. E .. .. : change count _after_ copy
bgt $0,$onebyteloop # U .. .. .. : U L U L
@@ -233,7 +219,7 @@ $exitout: # Destination for exception recovery(?)
nop # .. .. .. E
nop # .. .. E ..
nop # .. E .. ..
- ret $31,($28),1 # L0 .. .. .. : L U L U
+ ret $31,($26),1 # L0 .. .. .. : L U L U
.end __copy_user
EXPORT_SYMBOL(__copy_user)