[LLVMbugs] [Bug 13928] New: Suboptimal instruction ordering from split vector operations

[bugzilla-daemon at llvm.org]

http://llvm.org/bugs/show_bug.cgi?id=13928

             Bug #: 13928
           Summary: Suboptimal instruction ordering from split vector
                    operations
           Product: new-bugs
           Version: 3.1
          Platform: PC
        OS/Version: Linux
            Status: NEW
          Severity: normal
          Priority: P
         Component: new bugs
        AssignedTo: unassignedbugs at nondot.org
        ReportedBy: mike.gist at morganstanley.com
                CC: llvmbugs at cs.uiuc.edu
    Classification: Unclassified


Created attachment 9264
  --> http://llvm.org/bugs/attachment.cgi?id=9264
Example IR

After splitting vectors into smaller types, the resulting instructions aren't
reordered, sometimes producing suboptimal performance.

The IR and asm (x86_64) for a simple example are attached.

I would expect the loop body to look something more like:

movapd  (%rax), %xmm0
addpd   (%rcx), %xmm0
movapd  %xmm0, (%rsi)
...

-- 
Configure bugmail: http://llvm.org/bugs/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.