[LLVMbugs] [Bug 22685] making zeros the hard way: scalar load into zero vector (AVX)

[bugzilla-daemon at llvm.org]

https://llvm.org/bugs/show_bug.cgi?id=22685

Sanjay Patel <spatel+llvm at rotateright.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|---                         |FIXED

--- Comment #5 from Sanjay Patel <spatel+llvm at rotateright.com> ---
http://reviews.llvm.org/rL233704
http://reviews.llvm.org/rL233724
http://reviews.llvm.org/rL233931
http://reviews.llvm.org/rL233941

After those, I think we can declare victory. The AVX2 codegen for the test
cases in comment 4 is ideal:

_mov_v8f32:                             ## @mov_v8f32
    .cfi_startproc
## BB#0:
    vmovss    (%rdi), %xmm1           ## xmm1 = mem[0],zero,zero,zero
    vaddps    %ymm0, %ymm1, %ymm0
    retq
    .cfi_endproc

    .globl    _mov_v4f64
    .align    4, 0x90
_mov_v4f64:                             ## @mov_v4f64
    .cfi_startproc
## BB#0:
    vmovsd    (%rdi), %xmm1           ## xmm1 = mem[0],zero
    vaddpd    %ymm0, %ymm1, %ymm0
    retq
    .cfi_endproc

    .globl    _mov_v4i64
    .align    4, 0x90
_mov_v4i64:                             ## @mov_v4i64
    .cfi_startproc
## BB#0:
    vmovq    (%rdi), %xmm1           ## xmm1 = mem[0],zero
    vpaddq    %ymm0, %ymm1, %ymm0
    retq
    .cfi_endproc

    .globl    _mov_v8i32
    .align    4, 0x90
_mov_v8i32:                             ## @mov_v8i32
    .cfi_startproc
## BB#0:
    vmovd    (%rdi), %xmm1           ## xmm1 = mem[0],zero,zero,zero
    vpaddd    %ymm0, %ymm1, %ymm0
    retq
    .cfi_endproc

    .globl    _mov_v16i16
    .align    4, 0x90
_mov_v16i16:                            ## @mov_v16i16
    .cfi_startproc
## BB#0:
    movzwl    (%rdi), %eax
    vmovd    %eax, %xmm0
    vpaddw    %ymm0, %ymm1, %ymm0
    retq
    .cfi_endproc

    .globl    _mov_v32i8
    .align    4, 0x90
_mov_v32i8:                             ## @mov_v32i8
    .cfi_startproc
## BB#0:
    movzbl    (%rdi), %eax
    vmovd    %eax, %xmm1
    vpaddb    %ymm0, %ymm1, %ymm0
    retq

-------------------------------------------------------------------------------

There's still something screwy with AVX1 and the integer cases, but this will
be a separate bug:

    vmovd    (%rdi), %xmm1           ## xmm1 = mem[0],zero,zero,zero
    vpaddd    %xmm0, %xmm1, %xmm1
    vextractf128    $1, %ymm0, %xmm0
    vinsertf128    $1, %xmm0, %ymm1, %ymm0
    retq

-- 
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/20150402/c7c2c63f/attachment.html>