[Speculation] SaveCommit units don't handle NoCmp signal correctly

[shundroid]

Background (from Haoran's thesis):

• When the speculator receives real data before it speculates, the speculator sends a NoCmp signal to SaveCommit units through SCSaveCtrl channel.
• When SaveCommit units receive a NoCmp signal, it should perform resend. That is, SaveCommit units should send out the oldest data token and drop it from the internal buffer.
  • Note that SaveCommit units also perform resend when the speculator finds out its oldest prediction is wrong (imagine that Save units also need to resend non-spec tokens in this case).

In Haoran's implementation, NoCmp signal seems to be handled incorrectly and causes a bug that SaveCommit units cannot accept any control signals after NoCmp signal is sent to them.

In detail as follows:

1. NoCmpEn signal is sent to the save-commit unit. That is, pValidArray(1) = '1' and controlInArray(0) = "100".
2. Then, NoCmpEn becomes 1 and other something-En signals are 0 (en_proc process in the save_commit_control arch)
3. readyArray(1) becomes 0 because NoCmpEn is not handled in the ready_proc process
4. It never accepts the NoCmpEn signal and stalls permanently

[shundroid]

Apart from the thesis, NoCmpEn signal is defined separately from ResendEn signal and NoCmpEn signal is not handled correctly in the current implementation. I guess that this NoCmpEn signal is probably old and unnecessary. I removed NoCmpEn signal and replaced it with ResendEn signal and it worked.

Fixed in #172