[llvm-bugs] [Bug 35977] New: Vector shift misoptimized with AVX512

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=35977

            Bug ID: 35977
           Summary: Vector shift misoptimized with AVX512
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: All
            Status: NEW
          Severity: normal
          Priority: P
         Component: Backend: X86
          Assignee: unassignedbugs at nondot.org
          Reporter: benny.kra at gmail.com
                CC: llvm-bugs at lists.llvm.org, llvm-dev at redking.me.uk

Created attachment 19688
  --> https://bugs.llvm.org/attachment.cgi?id=19688&action=edit
IR test case

The attached testcase (reduced from llvmpipe) is erroneously optimized away
when AVX512 is enabled.

$ llc -mcpu=skylake < t.ll
test:                                   # @test
        .cfi_startproc
# %bb.0:                                # %entry
        vmovq   (%rdi), %xmm0           # xmm0 = mem[0],zero
        vpxor   %xmm1, %xmm1, %xmm1
        vpunpcklbw      %xmm0, %xmm1, %xmm0 # xmm0 =
xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
        vmovdqa %xmm0, (%rsi)
        retq

$ llc -mcpu=skylake-avx512 < t.ll
test:                                   # @test
        vxorps  %xmm0, %xmm0, %xmm0
        vmovaps %xmm0, (%rsi)
        retq

This seems to have started with r321085, but seems like an unrelated latent bug
in shuffle lowering.

-- 
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/20180116/b733aef9/attachment.html>