alpha: fix alignment problem in csum_ipv6_magic()

[powerpc.git] / arch / alpha / lib / ev6-csum_ipv6_magic.S

diff --git a/arch/alpha/lib/ev6-csum_ipv6_magic.S b/arch/alpha/lib/ev6-csum_ipv6_magic.S
index de1948a..fc0bc39 100644 (file)
--- a/arch/alpha/lib/ev6-csum_ipv6_magic.S
+++ b/arch/alpha/lib/ev6-csum_ipv6_magic.S
@@ -46,6 +46,10 @@
  * add the 3 low ushorts together, generating a uint
  * a final add of the 2 lower ushorts
  * truncating the result.
+ *
+ * Misalignment handling added by Ivan Kokshaysky <ink@jurassic.park.msu.ru>
+ * The cost is 16 instructions (~8 cycles), including two extra loads which
+ * may cause additional delay in rare cases (load-load replay traps).
  */
 
        .globl csum_ipv6_magic
@@ -55,25 +59,45 @@
 csum_ipv6_magic:
        .prologue 0
 
-       ldq     $0,0($16)       # L : Latency: 3
+       ldq_u   $0,0($16)       # L : Latency: 3
        inslh   $18,7,$4        # U : 0000000000AABBCC
-       ldq     $1,8($16)       # L : Latency: 3
+       ldq_u   $1,8($16)       # L : Latency: 3
        sll     $19,8,$7        # U : U L U L : 0x00000000 00aabb00
 
+       and     $16,7,$6        # E : src misalignment
+       ldq_u   $5,15($16)      # L : Latency: 3
        zapnot  $20,15,$20      # U : zero extend incoming csum
-       ldq     $2,0($17)       # L : Latency: 3
-       sll     $19,24,$19      # U : U L L U : 0x000000aa bb000000
+       ldq_u   $2,0($17)       # L : U L U L : Latency: 3
+
+       extql   $0,$6,$0        # U :
+       extqh   $1,$6,$22       # U :
+       ldq_u   $3,8($17)       # L : Latency: 3
+       sll     $19,24,$19      # U : U U L U : 0x000000aa bb000000
+
+       cmoveq  $6,$31,$22      # E : src aligned?
+       ldq_u   $23,15($17)     # L : Latency: 3
        inswl   $18,3,$18       # U : 000000CCDD000000
+       addl    $19,$7,$19      # E : U L U L : <sign bits>bbaabb00
 
-       ldq     $3,8($17)       # L : Latency: 3
-       bis     $18,$4,$18      # E : 000000CCDDAABBCC
-       addl    $19,$7,$19      # E : <sign bits>bbaabb00
-       nop                     # E : U L U L
+       or      $0,$22,$0       # E : 1st src word complete
+       extql   $1,$6,$1        # U :
+       or      $18,$4,$18      # E : 000000CCDDAABBCC
+       extqh   $5,$6,$5        # U : L U L U
 
+       and     $17,7,$6        # E : dst misalignment
+       extql   $2,$6,$2        # U :
+       or      $1,$5,$1        # E : 2nd src word complete
+       extqh   $3,$6,$22       # U : L U L U :
+
+       cmoveq  $6,$31,$22      # E : dst aligned?
+       extql   $3,$6,$3        # U :
        addq    $20,$0,$20      # E : begin summing the words
+       extqh   $23,$6,$23      # U : L U L U :
+
        srl     $18,16,$4       # U : 0000000000CCDDAA
+       or      $2,$22,$2       # E : 1st dst word complete
        zap     $19,0x3,$19     # U : <sign bits>bbaa0000
-       nop                     # E : L U U L
+       or      $3,$23,$3       # E : U L U L : 2nd dst word complete
 
        cmpult  $20,$0,$0       # E :
        addq    $20,$1,$20      # E :