[LLVMdev] CodeGen instructions and patterns

Jim Grosbach grosbach at apple.com
Fri Feb 24 14:35:42 CST 2012


(readding cc llvmdev)

On Feb 24, 2012, at 11:46 AM,    محمد ﻋﻤﺮ  ﺩﻫﻠﻮﻯ <omerbeg at gmail.com> wrote:

> 
> > otherwise how are the machine instructions being accessed/matched for instruction selection ?
> >
> 
> Have a look at the code in SelectionDAGISel.
> 
> 
> I am looking at the ARM backend.
> In specific, the instruction selection. 
> In Select(N), I only see ad-hoc matches on the basis of the opcode in the provided SDNode of the DAG, and then getMachineNode() is being used to generate the machine instruction.
> Am I understanding this correctly ?    
> 

That's just the target-specific part. Also look in lib/CodeGen/SelectionDAG.

You may find the debug output from the compiler useful. Try running llc with the "-debug-only=isel" command line option. That will give lots of information about how the selector is going about figuring out which patterns to use.

-Jim

> Consider the following multiply add insstruction, given in ARMInstrInfo.td 
> 
> def MLA  : AsMul1I32<0b0000001, (outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm, GPR:$Ra),
>                     IIC_iMAC32, "mla", "\t$Rd, $Rn, $Rm, $Ra",
>                    [(set GPR:$Rd, (add (mul GPR:$Rn, GPR:$Rm), GPR:$Ra))]>
> 
> How do I get this instruction and it corresponding pattern in the selector ?
> 
> -Omer 
> 
> 
> 
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.cs.uiuc.edu/pipermail/llvmdev/attachments/20120224/ab5c00a6/attachment.html 


More information about the LLVMdev mailing list