The assemble optimization for ARM (with G729_ARM option) has issues

```
What steps will reproduce the problem?
1. Compile with G729_ARM option, use the codec to communicate with other phone.
2. The program will crash while decoding first packet, I found it stops at
--Word16 div_s(Word16 var1, Word16 var2), e.g, the var1 and var2 always is less
than 0. The call path is: Post_Filter() --> agc() --> goes to line "s =
L_deposit_l(div_s(gain_out,gain_in));", stop at div_s(gain_out, gain_in). As I
print the two value, found they are less than 0, it causes program exit.
3. If I compile it without G729_ARM option, it is ok, no problem.

What is the expected output? What do you see instead?
--> Compare difference of the compilation option, it illustrates that the
Assemble optimization for ARM has issues.

What version of the product are you using? On what operating system?
---> I'm testing on HTC handset, it is ARM V7a.

Please provide any additional information below.

```

Original issue reported on code.google.com by `perty....@gmail.com` on 9 Sep 2011 at 3:19