[LLVMbugs] [Bug 17236] missed opportunity to use vector FMA instruction

bugzilla-daemon at llvm.org bugzilla-daemon at llvm.org
Thu Apr 10 14:06:36 PDT 2014


Sanjay Patel <sanjay3.0.0.0 at gmail.com> changed:

           What    |Removed                     |Added
             Status|NEW                         |RESOLVED
                 CC|                            |sanjay3.0.0.0 at gmail.com
         Resolution|---                         |FIXED

--- Comment #2 from Sanjay Patel <sanjay3.0.0.0 at gmail.com> ---
$ ./clang -v
clang version 3.5.0 (trunk 205798) (llvm/trunk 205792)
Target: x86_64-apple-darwin13.1.0
Thread model: posix

The SLP vectorizer is now able to generate packed FMA instructions for this
test case. It's not using 32-byte wide operations, but that problem is
addressed by bug 17170.

_foo:                                   ## @foo
## BB#0:                                ## %entry
    vmovupd    (%rdi), %xmm0
    vfmadd213pd    %xmm0, %xmm0, %xmm0
    vmovupd    %xmm0, (%rdi)
    vmovupd    16(%rdi), %xmm0
    vfmadd213pd    %xmm0, %xmm0, %xmm0
    vmovupd    %xmm0, 16(%rdi)
    vmovupd    32(%rdi), %xmm0
    vfmadd213pd    %xmm0, %xmm0, %xmm0
    vmovupd    %xmm0, 32(%rdi)
    vmovupd    48(%rdi), %xmm0
    vfmadd213pd    %xmm0, %xmm0, %xmm0
    vmovupd    %xmm0, 48(%rdi)

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/20140410/16bca653/attachment.html>

More information about the llvm-bugs mailing list