PROJECT : 9th OCT

Agenda : Trace selection and Trace collection

Global table : table which hold the info about the traces.

1. Start of Trace.
- destination of backward jump.
- exits from fragments already present.
- all function calls destination

2. End of Trace.

- two possibilities
- trace is a loop
- no of insn > Threshold
- encountering start of trace
- begining of fucntion definition
- no of insn > Threshold
- ret insn is encountered

Now, We just take a look at how the control flow is gonna take place.

Interpretation Starts.

- from begining of code.
- a branch tranfer insn is encountered.

- forward jump
- change the fetch addr for interpreter.

- backward jump
- its a special case.
- exect'n of collected linear code.
- making an entry in global table.
- check count, perform reqd. operation.
- function calls
- exect'n of collected linear code.
- entry made in global table.
- if already present, increment the count.
- and check for threshold.

Once you find trace crossed the threshold :

- start collecting the trace
- start of trace already defined
- end of trace alreay defined

Contents of trace can be :

- forward jumps inside fragments
- backward jumps inside fragments
- jumps to outside the fragment
- instructions - calls to functions

Actions taken for these above mentioned cases :
- forward jump inside fragments
- care taken during relocation.

- backward jump inside fragments
- care taken during relocation.

- jump outside fragments
- control transferred to epilogue.
- an entry is present for desti'n of exits.
as all exits from fragments are probabale
candidates for trace.
- incrementaing the count in the global table.
- control finally transferred to epilogue.

- function calls
- should be replaced by exit stubs.
- if this exit stub crosses the threshold ,
it is optimised and linked into fragment cache.

Some important considerations :
- Every fragment will have a unique fragment id.
- Also in the global table, there will be a field for this _frag_id.
- The use of this fragment id :

- If two or more fragments have same exit destinations
from the fragment, we willtwo separate entries for
that in the gloabal table.

- If sometime later, an exit is made through the
exit stub, the corrseponding entry in the global
table will be bumped.

- This is to help, the linking of fragments, e.g. if
siometimes later the exit path's counter execeeds
the threshold, we can optimise it and link it to the
existing fragments.