llvm dev - Mar 2013 - [LLVMdev] Question about simple constant propagation pass

Hi,

I just started working with llvm, I am trying to test and improve the
constant propagation pass (-constprop),

I have two main questions,

1. I wrote my own test file, which has something like
   int a = 1
   int b = a + 2;
   int c = b + 3;
   print c

    However, non of the instruction was killed according to -stat. I think
it might due to the fact the llvm instructions generated was a bit
convoluted, but overall, I am not sure why this simple test case has no
instruction killed? Shouldn't c be just 6 and show at least two
instructions killed?

2. The description states that this pass is very simple. Can anyone point
me to a few directions that might be interesting to improve it? I am not
sure about why is it labeled "simple" ?

Thanks

Yunming
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20130327/b2f5a295/attachment.html>

Hi Yunming,

On 28/03/13 03:15, yunming zhang wrote:
> Hi,
>
> I just started working with llvm, I am trying to test and improve the
constant
> propagation pass (-constprop),
>
> I have two main questions,
>
> 1. I wrote my own test file, which has something like
>     int a = 1
>     int b = a + 2;
>     int c = b + 3;
>     print c
>
>      However, non of the instruction was killed according to -stat. I think
it
> might due to the fact the llvm instructions generated was a bit convoluted,
but
> overall, I am not sure why this simple test case has no instruction killed?
> Shouldn't c be just 6 and show at least two instructions killed?
optimization passes usually assume you have run at least -mem2reg (or -sroa)
first and won't do much if you haven't.

Ciao, Duncan.
>
> 2. The description states that this pass is very simple. Can anyone point
me to
> a few directions that might be interesting to improve it? I am not sure
about
> why is it labeled "simple" ?
>
> Thanks
>
> Yunming
>
>
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>

Thanks a lot Duncan!!

It worked like magic,

Yunming


On Thu, Mar 28, 2013 at 7:11 AM, Duncan Sands <baldrick at free.fr> wrote:
> Hi Yunming,
>
>
> On 28/03/13 03:15, yunming zhang wrote:
>
>> Hi,
>>
>> I just started working with llvm, I am trying to test and improve the
>> constant
>> propagation pass (-constprop),
>>
>> I have two main questions,
>>
>> 1. I wrote my own test file, which has something like
>>     int a = 1
>>     int b = a + 2;
>>     int c = b + 3;
>>     print c
>>
>>      However, non of the instruction was killed according to -stat. I
>> think it
>> might due to the fact the llvm instructions generated was a bit
>> convoluted, but
>> overall, I am not sure why this simple test case has no instruction
>> killed?
>> Shouldn't c be just 6 and show at least two instructions killed?
>>
>
> optimization passes usually assume you have run at least -mem2reg (or
> -sroa)
> first and won't do much if you haven't.
>
> Ciao, Duncan.
>
>
>> 2. The description states that this pass is very simple. Can anyone
point
>> me to
>> a few directions that might be interesting to improve it? I am not sure
>> about
>> why is it labeled "simple" ?
>>
>> Thanks
>>
>> Yunming
>>
>>
>> ______________________________**_________________
>> LLVM Developers mailing list
>> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
>>
http://lists.cs.uiuc.edu/**mailman/listinfo/llvmdev<http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev>
>>
>>
> ______________________________**_________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
>
http://lists.cs.uiuc.edu/**mailman/listinfo/llvmdev<http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20130328/6d4a0db8/attachment.html>