Straight-line code in DeferredTaggedToI

Straight-line code in DeferredTaggedToI

Code is shorter because the multiple branches to deoptimization are
short form branches to a long form jump.

Also: use 1-byte form of push reg instead of 2 byte form in several
places.

Example: this deferred code is 20% shorter.

0xf4b270bf  1087  8179ffa1a03bf5 cmp [ecx+0xff],0xf53ba0a1    ;; object: 0xf53ba0a1 <Map>
0xf4b270c6  1094  0f85d0300400   jnz 0xf4b6a19c              ;; deoptimization bailout 38
0xf4b270cc  1100  f20f104103     movsd xmm0,[ecx+0x3]
0xf4b270d1  1105  f20f2cc8       cvttsd2si ecx,xmm0
0xf4b270d5  1109  f20f2ac9       cvtsi2sd xmm1,ecx
0xf4b270d9  1113  660f2ec1       ucomisd xmm0,xmm1
0xf4b270dd  1117  0f85b9300400   jnz 0xf4b6a19c              ;; deoptimization bailout 38
0xf4b270e3  1123  0f8ab3300400   jpe 0xf4b6a19c              ;; deoptimization bailout 38
0xf4b270e9  1129  e987feffff     jmp 757  (0xf4b26f75)
0xf4b270ee  1134

1134 = 1087 = 47

0xf4b8749c  1052  8179ffa1a041f5 cmp [ecx+0xff],0xf541a0a1    ;; object: 0xf541a0a1 <Map>
0xf4b874a3  1059  7519           jnz 1086  (0xf4b874be)
0xf4b874a5  1061  f20f104103     movsd xmm0,[ecx+0x3]
0xf4b874aa  1066  f20f2cc8       cvttsd2si ecx,xmm0
0xf4b874ae  1070  f20f2ac9       cvtsi2sd xmm1,ecx
0xf4b874b2  1074  660f2ec1       ucomisd xmm0,xmm1
0xf4b874b6  1078  7506           jnz 1086  (0xf4b874be)
0xf4b874b8  1080  0f8bb0feffff   jpo 750  (0xf4b8736e)
0xf4b874be  1086  e9d92c0400     jmp 0xf4bca19c              ;; deoptimization bailout 38
0xf4b874c3  1091

1091 - 1052 = 39

[fschneider, 2011-07-05 10:04:23]

LGTM. The same also works on x64. Do you want me to land this for you?