[llvm-commits] Deterministic finite automaton based packetizer for VLIW architectures

Anshuman Dasgupta adasgupt at codeaurora.org
Mon Nov 28 07:13:47 PST 2011


Did anyone get a chance to take a look at the patch? Is it okay to commit?


Qualcomm Innovation Center, Inc is a member of Code Aurora Forum

On 11/18/2011 10:51 PM, Anshuman Dasgupta wrote:
> Hi,
> I'm attaching a patch that adds support for a deterministic finite 
> automaton (DFA) based packetizer for VLIW architectures. Specifically, 
> it automatically generates a DFA from a VLIW target's Schedule.td 
> file. In a VLIW machine, an instruction can typically be dispatched to 
> one or many function units. The DFA determines whether there exists a 
> legal mapping of instructions to functional unit assignments in a 
> packet. This DFA can then be queried by a backend packetization pass 
> to determine which instructions can be grouped into a VLIW packet.
> This patch contains the machine-independent code that adds a component 
> to TableGen. The component autogenerates the DFA and implements the 
> API that can be used to query the DFA during packetization. This can 
> be used by any VLIW target to packetize its instructions. After the 
> Hexagon backend is committed, I will post another Hexagon-specific 
> patch that uses this mechanism to construct packets in the Hexagon 
> backend. I ran the llvm test suite and this patch did not cause any 
> regressions.
> I would appreciate reviews and comments.
> Thanks
> -Anshu
> --
> Qualcomm Innovation Center, Inc is a member of Code Aurora Forum
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20111128/9feb80b7/attachment.html>

More information about the llvm-commits mailing list