[cfe-dev] Question on "phi-node-folding-threshold"

Joan Lluch via cfe-dev cfe-dev at lists.llvm.org
Wed Sep 18 22:05:48 PDT 2019


Hi Craig,

This worked. Thank you very much.

Is there a way to specify this by default for a particular target?. I implemented a LLVM backend and added all the required code so that Clang recognises it as a target, but I’m unsure about how to deal with this particularly. Ideally, Clang should use ‘phi-node-folding-threshold=1’ automatically when my target is specified. Any suggestions are appreciated. 

Thanks again.

John


> On 18 Sep 2019, at 08:34, Craig Topper <craig.topper at gmail.com> wrote:
> 
> You can set it on the command line by doing "clang -mllvm -phi-node-folding-threshold=1”  The -mllvm tells clang to pass the next argument through to the command line system used by IR passes.
> 
> ~Craig
> 
> 
>> On Tue, Sep 17, 2019 at 11:05 PM Joan Lluch via cfe-dev <cfe-dev at lists.llvm.org> wrote:
>> Hi All,
>> 
>> For a custom backend I’m developing, where jumps are not particularly expensive, I want to reduce the phi-node-folding-threshold attribute value from the default 2 to just 1, as it produces better code in my case.
>> 
>> I first tried to set the attribute as a clang command line option, but it does not seem to be available. ( I tried “clang  -phi-node-folding-threshold=1” but this returns an “unknown argument” error)  
>> 
>> However, it works as intended if I replace the default value by hard coding in it in SimplifyCFG.cpp and recompiling clang
>> 
>>  My question is, Is there a way to set that attribute to a different value for a specific target?. Or at least to set it on the command line?
>> 
>> Thanks in advance
>> 
>> John
>> _______________________________________________
>> cfe-dev mailing list
>> cfe-dev at lists.llvm.org
>> https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20190919/ad63171f/attachment.html>


More information about the cfe-dev mailing list