Now almost as fast as the C code.

Rev: src/nettle/sparc/aes.asm:1.5

diff --git a/sparc/aes.asm b/sparc/aes.asm
index 8860acb039f31066831b6de4aa78265424c7909b..300295524923e84a6754fbe86d2b57bf69a10003 100644 (file)
--- a/sparc/aes.asm
+++ b/sparc/aes.asm
@@ -59,7 +59,8 @@ key_addition_8to32:
        retl
        nop
 
-       ! FIXME:         Inline, unroll?
+! key_addition32(const uint32_t *txt, const uint32_t *keys, uint32_t *out)
+
        .size   key_addition_8to32,.LLfe1-key_addition_8to32
        .align 4
        .type   key_addition32,#function
@@ -67,27 +68,25 @@ key_addition_8to32:
 key_addition32:
        ! Use %g2 and %g3 as temporaries, %o3 as counter
        mov     0, %o3
+       ! Decrement out, so we can increment the counter earlier in the loop
+       sub     %o2, 4, %o2
 .LL26:
-       ! Get *txt++
-       ld      [%o0], %g2
-       add     %o0, 4, %o0
-       ! Get *keys++
-       ld      [%o1], %g3
-       add     %o1, 4, %o1
+       ! Get txt[i]
+       ld      [%o0+%o3], %g2
+       ! Get keys[i]
+       ld      [%o1+%o3], %g3
+       cmp     %o3, 12
+       add     %o3, 4, %o3
+
        xor     %g2, %g3, %g3
-       st      %g3, [%o2]
-       ! Incrementing %o2 in the delay slot
-       add     %o3, 1, %o3
 
-       ! FIXME:         Unroll or inline?
-       cmp     %o3, 3
        bleu    .LL26
-       add     %o2, 4, %o2
+       st      %g3, [%o2+%o3]
        
        retl
        nop
 
-!      ! And three more times
+
 !      
 !      mov     %o0, %o4
 !      mov     0, %o3