Implementation of the Control Unit

[thegabriele97]

• Basic working implementation of it (with J and JAL and all basic OP required to be implemented in the base DLX)
• The CALL/RET op should be executed only if there are no hazard (the hazard table is all at 0). If it's not, we should add some stall until the hazard table is all at 0. Maybe a signal is needed from the DECODE stage (and from the HazardTable unit) that tells us if there is at least a 1 in the table
• When a JUMP/BRANCH occurs, a fetch cycle must be skipped (i.e. the IR enable should go off). I think it can be accomplished by setting the bit IF_EN (or the one that enables the IR) to 0 for each jump/branch instruction

3. Acoomplished by setting a bit of the IR to 0, and when it's 0 IR <- NOP so a NOP is executed after a jump

[giovannipollo]

The CALL/RET op should be executed only if there are no hazard (the hazard table is all at 0). If it's not, we should add some stall until the hazard table is all at 0. Maybe a signal is needed from the DECODE stage (and from the HazardTable unit) that tells us if there is at least a 1 in the table

The control if the Table is all 0 is ready to be tested. However I was not able to test it, because the waveform does not show up (did not understand why).

EDIT (by @thegabriele97 ): problem solved

[thegabriele97]

The CALL/RET op should be executed only if there are no hazard (the hazard table is all at 0). If it's not, we should add some stall until the hazard table is all at 0. Maybe a signal is needed from the DECODE stage (and from the HazardTable unit) that tells us if there is at least a 1 in the table

The control if the Table is all 0 is ready to be tested. However I was not able to test it, because the waveform does not show up (did not understand why).

EDIT (by @thegabriele97 ): problem solved

@giop98 is the signal you made (the Hazard table is all at 0) refereed to the second point of this issue?

[MatteoBattilana]

CALL and RET implementation is totally missing both from the VHD and compiler part.
I think that the CALL/RET signal must be put in an AND together the all zero signal. We have this signal both at DLX and Decode unit level, maybe we can propagate into the datapath to inhibit pipeline and put NOP.

[thegabriele97]

CALL and RET implementation is totally missing both from the VHD and compiler part.
I think that the CALL/RET signal must be put in an AND together the all zero signal. We have this signal both at DLX and Decode unit level, maybe we can propagate into the datapath to inhibit pipeline and put NOP.

Yes but I was thinking about looking only for R8..R31 about "busy registers" in order to insert the stall after a CALL/RET. This is because R0..R7 are in the GLOBALS so they are always the same and they are always available so there is no need to stall if they are the only busy registers

[thegabriele97]

CALL and RET implementation is totally missing both from the VHD and compiler part.
I think that the CALL/RET signal must be put in an AND together the all zero signal. We have this signal both at DLX and Decode unit level, maybe we can propagate into the datapath to inhibit pipeline and put NOP.

Yes however about the other things, i think we should do something similar to the HAZARD_SIG, so stalling the pipeline when ALL_ZEROS & CALL/RET are both at 1

[giovannipollo]

@giop98 is the signal you made (the Hazard table is all at 0) refereed to the second point of this issue?

Yes, when I did the signal it was referring to the second point. But now, as both of you said, we may need to change something.

[giovannipollo]

Yes but I was thinking about looking only for R8..R31 about "busy registers" in order to insert the stall after a CALL/RET. This is because R0..R7 are in the GLOBALS so they are always the same and they are always available so there is no need to stall if they are the only busy registers

I think this is a nice idea.

Indeed, taking about PUSH/POP, I'm not sure if we have to do something (if I'm not wrong everything is already managed inside the windowed RF).
In case we should modify something, an idea, in order to check if we have free space, would be to check if, in the next window, we have no valid data (so value of the registers at 0).