llvm dev - May 2007 - [LLVMdev] Instruction Scheduling in LLVM

Hello,

I am working with the SelectionDAG/ScheduleDAG framework to implement a
variation of the List scheduling Algorithm in LLVM.
I was trying to understand the existing List scheduler implementation in LLVM. I
have a doubt about the SUnits structure which contain flagged nodes together.
The instructions within a Sunit are scheduled as a single unit. My understanding
is that the nodes in the original DAG fall into exactly one of the Sunits. So, I
was wondering if the instructions (SDNodes) within a SUnit need to be scheduled
seperately? For example, it may be necessary to insert a No-op between the
instructions (within one SUnit) in certain cases. Or is the nodes which are
flagged together (apart from the main node) are psuedo ops which don't need
to be actually emitted?
A general overview will also be helpful. 
(The documentation does not provide much details and the source is too
complicated to understand at times).

Thank you
-Kaushik
 
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20070501/94d2c626/attachment.html>

Some isel nodes produce MVT::Flag value because it has to scheduled  
right before its use (one use per flag value). An example of this is  
a compare and a conditional branch (only because we do not model  
condition code as a register right now).

The scheduler is honoring the request by treating a chain of flagged  
nodes as a single scheduling unit.

Evan

On May 1, 2007, at 6:19 PM, Kaushik Kumar wrote:
> Hello,
>
> I am working with the SelectionDAG/ScheduleDAG framework to  
> implement a variation of the List scheduling Algorithm in LLVM.
> I was trying to understand the existing List scheduler  
> implementation in LLVM. I have a doubt about the SUnits structure  
> which contain flagged nodes together. The instructions within a  
> Sunit are scheduled as a single unit. My understanding is that the  
> nodes in the original DAG fall into exactly one of the Sunits. So,  
> I was wondering if the instructions (SDNodes) within a SUnit need  
> to be scheduled seperately? For example, it may be necessary to  
> insert a No-op between the instructions (within one SUnit) in  
> certain cases. Or is the nodes which are flagged together (apart  
> from the main node) are psuedo ops which don't need to be actually  
> emitted?
> A general overview will also be helpful.
> (The documentation does not provide much details and the source is  
> too complicated to understand at times).
>
> Thank you
> -Kaushik
>
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20070502/7c7e7d98/attachment.html>