Use stack instead of AA, BB, CC and DD variables

The motivation is to hope for increase in speed where we avoid
getting and setting these variables once for every 64-byte block.

Author: amn

MD5.wat.m4

@@ -27,17 +27,17 @@ define(`I', `(i32.xor (local.get $$2) (i32.or (local.get $$1) (i32.xor (local.ge
 	(func (export "update") (param $start i32) (param $n_bytes i32) (param $context i32)
 		(local $end i32)
 		(local $A i32) (local $B i32) (local $C i32) (local $D i32)
-		(local $AA i32) (local $BB i32) (local $CC i32) (local $DD i32)
 		(local.set $A (i32.load offset=0 (local.get $context)))
 		(local.set $B (i32.load offset=4 (local.get $context)))
 		(local.set $C (i32.load offset=8 (local.get $context)))
 		(local.set $D (i32.load offset=12 (local.get $context)))
 		(local.set $end (i32.add (local.get $start) (local.get $n_bytes)))
 		(loop $process_block
-			(local.set $AA (local.get $A))
-			(local.tee $BB (local.get $B))
-			(local.set $CC (local.get $C))
-			(local.set $DD (local.get $D))
+			local.get $C
+			local.get $D
+			local.get $A
+			local.get $B
+			local.get $B
 			R(A, B, C, D, 0, 0xd76aa478, 7, `F')
 			R(D, A, B, C, 1, 0xe8c7b756, 12, `F')
 			R(C, D, A, B, 2, 0x242070db, 17, `F')
@@ -105,10 +105,10 @@ define(`I', `(i32.xor (local.get $$2) (i32.or (local.get $$1) (i32.xor (local.ge
 			R(D, A, B, C, 11, 0xbd3af235, 10, `I')
 			R(C, D, A, B, 2, 0x2ad7d2bb, 15, `I')
 			R(B, C, D, A, 9, 0xeb86d391, 21, `I')
-			(local.set $A (i32.add (local.get $A) (local.get $AA)))
-			(local.set $B (i32.add (local.get $BB)))
-			(local.set $C (i32.add (local.get $C) (local.get $CC)))
-			(local.set $D (i32.add (local.get $D) (local.get $DD)))
+			(local.set $B (i32.add))
+			(local.set $A (i32.add (local.get $A)))
+			(local.set $D (i32.add (local.get $D)))
+			(local.set $C (i32.add (local.get $C)))
 			(local.set $start (i32.add (local.get $start) (i32.const 64)))
 			(br_if 0 (i32.ne (local.get $start) (local.get $end))))
 		(i32.store offset=0 (local.get $context) (local.get $A))