[llvm-bugs] [Bug 34584] New: [AVX512] eliminate compare of bool mask for masked store

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=34584

            Bug ID: 34584
           Summary: [AVX512] eliminate compare of bool mask for masked
                    store
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: All
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: Backend: X86
          Assignee: unassignedbugs at nondot.org
          Reporter: spatel+llvm at rotateright.com
                CC: llvm-bugs at lists.llvm.org

As discussed in D37446:
https://reviews.llvm.org/D37446

If we have a bool vector mask for a masked store, we shouldn't need to generate
a compare to set the sign bits of that mask.

declare void @llvm.masked.store.v4f32.p0v4f32(<4 x float>, <4 x float>*, i32,
<4 x i1>)
declare void @llvm.masked.store.v16f32.p0v16f32(<16 x float>, <16 x float>*,
i32, <16 x i1>)

define void @generic_masked_stores(<4 x float> %x, <4 x float>* %ptr, <4 x
float> %y, <4 x i32> %mask) {
  %trunc = icmp slt <4 x i32> %mask, zeroinitializer
  call void @llvm.masked.store.v4f32.p0v4f32(<4 x float> %x, <4 x float>* %ptr,
i32 1, <4 x i1> %trunc)
  ret void
}

define void @generic_masked_stores_16(<16 x float> %x, <16 x float>* %ptr, <16
x float> %y, <16 x i32> %mask) {
  %trunc = icmp slt <16 x i32> %mask, zeroinitializer
  call void @llvm.masked.store.v16f32.p0v16f32(<16 x float> %x, <16 x float>*
%ptr, i32 1, <16 x i1> %trunc)
  ret void
}

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

$ ./llc -o - maskedstore_truncmask.ll -mattr=avx512f,avx512vl,avx512bw
_generic_masked_stores:               
        vpxor   %xmm1, %xmm1, %xmm1
        vpcmpgtd        %xmm2, %xmm1, %k1
        vmovups %xmm0, (%rdi) {%k1}
        retq
_generic_masked_stores_16:
        vpxor   %xmm1, %xmm1, %xmm1
        vpcmpgtd        %zmm2, %zmm1, %k1
        vmovups %zmm0, (%rdi) {%k1}
        vzeroupper
        retq

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

A better option would be:
   vpmovd2m %xmm2, %k1
   vmovups %xmm0, (%rdi) {%k1}

Shorter, but not a universal solution for AVX512:
   vmaskmovps %xmm0, %xmm2, (%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/20170912/e10bb2c8/attachment.html>