[llvm-bugs] [Bug 50384] New: SCEV/IndVars has quadratic compile time complexity

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=50384

            Bug ID: 50384
           Summary: SCEV/IndVars has quadratic compile time complexity
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: Transformation Utilities
          Assignee: unassignedbugs at nondot.org
          Reporter: lebedev.ri at gmail.com
                CC: llvm-bugs at lists.llvm.org

Created attachment 24863
  --> https://bugs.llvm.org/attachment.cgi?id=24863&action=edit
n=32

I have not fully assessed what is going on, but -indvars, clearly,
scales non-linearly with the IR size.

https://www.wolframalpha.com/input/?i=fit+%7B32%2C+0.2114%7D%2C%7B64%2C+0.4191%7D%2C%7B128%2C+1.3613%7D%2C%7B256%2C+4.7540%7D%2C%7B512%2C+17.7136%7D%2C%7B1024%2C+74.2987%7D

Where x is LoopMicroOpBufferSize in X86ScheduleZnver3.td
(aka the desired unrolled loop size), and y is the time -indvars takes.

As it stands right now, i've put myself into a corner,
and with -march=znver3 it is simply unfeasible
to compile e.g. vanilla test-suite, it just takes too long.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20210518/42b3c5dc/attachment.html>