Snoopy

[Snoopy87]

• Rework AhbInterconnect (Decode + arbitrer)

• Add AHB tests for arbitrer/decoder/interconnect

• Add PhaseInitReg (Init all register uninitiated )

• You can add comments of a component in the rtf file generated

[jijingg]

Add PhaseInitReg (Init all register uninitiated )

hi @Snoopy87 i have question ,dose this will automatic init all register uninitiated? or just a manually option

[Snoopy87]

Add PhaseInitReg (Init all register uninitiated )

hi @Snoopy87 i have question ,dose this will automatic init all register uninitiated? or just a manually option

It is a manually option. In the SpinalConfig you can add transfomration phases
SpinalConfig( mode = VHDL, transformationPhases = ArrayBuffer(new PhaseInitReg()) ).generate(newTopLevel)

[jijingg]

@Snoopy87, manually option is ok . if you can add report log of uninitiated register it 'will be great .
because for some team, they prefer to remove reset on data-path for purpose of saving area(there is no functional risk), but on control path must with reset logic, if spinal give report log of uninitiated, the designer can find out which control path without uninitiated.

[Snoopy87]

@Snoopy87, manually option is ok . if you can add report log of uninitiated register it 'will be great .
because for some team, they prefer to remove reset on data-path for purpose of saving area(there is no functional risk), but on control path must with reset logic, if spinal give report log of uninitiated, the designer can find out which control path without uninitiated.

I have added messages in the PhaseInitReg

[typingArtist]

I very much dislike this code, both in its old and in its new form. AMBA 3 AHB-Lite specification, Section 3.5.2 states that

Although masters are not permitted to terminate a burst request early, slaves must be designed to work correctly if the burst is not completed.

This means that decoding the last information is quite worthless and leading to wrong conclusions in unlocking from a burst. IMHO, the only condition for unlocking is checking for HMASTLOCK, and whether HTRANS is either IDLE or NONSEQ. I am aware of the implications on timing, but relying on last is improper.

[typingArtist]

isLast doesn’t indicate properly a burst of type INCR, i.e. a burst of undefined length. In its current form, it takes a burst of type INCR just the same like the type SINGLE which is quite the opposite of INCR.

Edit: Also, according to Section 3.5.2 of the AMBA 3 AHB-Lite specification, an ERROR response doesn’t implicitly terminate a burst. A master is allowed to continue a burst, and a slave must be prepared for this situation.

However, see my other note on using last() in the AhbLite3Arbiter for why I see this method useless in the first place.