[llvm-bugs] [Bug 48760] New: Suboptimal cmov generation

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=48760

            Bug ID: 48760
           Summary: Suboptimal cmov generation
           Product: new-bugs
           Version: trunk
          Hardware: PC
                OS: All
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: new bugs
          Assignee: unassignedbugs at nondot.org
          Reporter: henrik at gramner.com
                CC: htmldeveloper at gmail.com, llvm-bugs at lists.llvm.org

LLVM prefers to generate cmov instructions with as many flag dependencies as
possible, which is suboptimal. It should be doing the opposite.

For example, both of those functions are semantically identical:

unsigned a(unsigned x) { return x >  3 ? x : 0; }
unsigned b(unsigned x) { return x >= 4 ? x : 0; }

They both generate identical code:

    xor    eax, eax
    cmp    edi, 3
    cmova  eax, edi
    ret

The better code would be:

    xor    eax, eax
    cmp    edi, 4
    cmovae eax, edi
    ret

cmovae has a dependency on CF whereas cmova has dependencies on both CF and ZF.
Many (most?) x86 CPUs will execute cmov instructions with a single flag
dependency in a single µop, but splits them into two µops if there are multiple
flag dependencies.

-- 
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/20210115/5ef81e89/attachment.html>