llvm dev - Jan 2016 - Question on Instruction Combining

Hello,

In InstructionCombining.cpp, under AddReachableCodeToWorklist
<http://llvm.org/docs/doxygen/html/InstructionCombining_8cpp.html#a65beadbfc094f4040b353518b98d0c07>,
DCE and ConstFold are performed but MadeIRChange is not set to true (lines
2912 and 2924). Is there some underlying assumption that supports this?

Intuitively not setting MadeIRChanges to true at these points could allow
Instruction Combining to exit early while there is still work to do.

Thank you,
Sain
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20160127/a5796113/attachment.html>

At first blush, DCE can reduce the number of uses which might encourage
InstCombine to further canonicalize which would indicate that we should set
the flag.  However, that would only matter if the users of the instruction
made it to the worklist in the first place which doesn't appear to be a
guarantee.

Do you have any code which is not fully canonicalized by InstCombine?

On Wed, Jan 27, 2016 at 12:35 PM, Sain-Zee Ueng via llvm-dev <
llvm-dev at lists.llvm.org> wrote:
> Hello,
>
> In InstructionCombining.cpp, under AddReachableCodeToWorklist
>
<http://llvm.org/docs/doxygen/html/InstructionCombining_8cpp.html#a65beadbfc094f4040b353518b98d0c07>,
> DCE and ConstFold are performed but MadeIRChange is not set to true (lines
> 2912 and 2924). Is there some underlying assumption that supports this?
>
> Intuitively not setting MadeIRChanges to true at these points could allow
> Instruction Combining to exit early while there is still work to do.
>
> Thank you,
> Sain
>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20160127/3799588c/attachment.html>