[llvm-bugs] [Bug 25373] New: -fomit-frame-pointer introduces spills

[via llvm-bugs]

https://llvm.org/bugs/show_bug.cgi?id=25373

            Bug ID: 25373
           Summary: -fomit-frame-pointer introduces spills
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: All
            Status: NEW
          Severity: normal
          Priority: P
         Component: Register Allocator
          Assignee: unassignedbugs at nondot.org
          Reporter: james.molloy at arm.com
                CC: llvm-bugs at lists.llvm.org
    Classification: Unclassified

Created attachment 15200
  --> https://llvm.org/bugs/attachment.cgi?id=15200&action=edit
Reproducer

The attached IR compiles with no spills or reloads when the frame pointer is
enabled. However if the frame pointer is disabled, 

$ llc -disable-fp-elim=true 2.ll -o - | grep -c Reload
0

$ llc -disable-fp-elim=false 2.ll -o - | grep -c Reload
2

This was reduced from a benchmark where our spill decisions were suboptimal.
Enabling FP omission (increasing the register bank by one) caused 10% more
spills. Even worse, reducing the register bank by one also caused 10% more
spills, but reducing it one further so there were only 10 registers reduced the
spill count again:

  R0..R12: N+10% spills
  R0..R11: N spills
  R0..R10: N+10% spills
  R0..R9:  N spills

So actually we do just as good a job with 10 regs as we do with 12 regs, and
better than the job we do with 13 regs!

I hope the reduced testcase actually points to the same underlying problem.

-- 
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/20151102/fe4c1feb/attachment.html>