[llvm-bugs] [Bug 36796] New: most llvm.experimental.vector.reduce intrinsics fail for most vector types on AArch64 only

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=36796

            Bug ID: 36796
           Summary: most llvm.experimental.vector.reduce intrinsics fail
                    for most vector types on AArch64 only
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: All
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: Backend: AArch64
          Assignee: unassignedbugs at nondot.org
          Reporter: gonzalobg88 at gmail.com
                CC: llvm-bugs at lists.llvm.org

Most of the llvm.experimental.vector.reduce intrinsics fail to lower on AArch64
only -
 they generate correct code in other targets, including the ARM target.

Example:

declare i8 @llvm.experimental.vector.reduce.mul.i8.v8i8(<8 x i8> %a)
define i8 
    @foo(<8 x i8>* noalias nocapture 
         readonly dereferenceable(16) %val) 
unnamed_addr {
  %a = load <8 x i8>, <8 x i8>* %val, align 16
  %b = tail call 
      i8 @llvm.experimental.vector.reduce.mul.i8.v8i8(<8 x i8> %a)
  ret i8 %b
}

makes the AArch64 backend segfault:

#0 0x0000000001b707da (/opt/compiler-explorer/clang-5.0.0/bin/llc+0x1b707da)
#1 0x0000000001b6e89e (/opt/compiler-explorer/clang-5.0.0/bin/llc+0x1b6e89e)
#2 0x0000000001b6e9ec (/opt/compiler-explorer/clang-5.0.0/bin/llc+0x1b6e9ec)
#3 0x00007f71aeed5390 __restore_rt
(/lib/x86_64-linux-gnu/libpthread.so.0+0x11390)
#4 0x0000000001a91844 (/opt/compiler-explorer/clang-5.0.0/bin/llc+0x1a91844)
#5 0x0000000001ae743c (/opt/compiler-explorer/clang-5.0.0/bin/llc+0x1ae743c)
Stack dump:
0.      Program arguments: /opt/compiler-explorer/clang-5.0.0/bin/llc -o
/tmp/compiler-explorer-compiler118219-63-a628my.hwd0a/output.s
-x86-asm-syntax=intel -O3 -march=aarch64 <source> 
1.      Running pass 'Function Pass Manager' on module '<source>'.
2.      Running pass 'AArch64 Instruction Selection' on function '@foo'
/compiler-explorer/c-preload/compiler-wrapper: line 21:  8596 Segmentation
fault      (core dumped) "$@"
Compiler returned: 139

while replacing mul with other reduction intrinsics some times work, sometimes
error with failed to select instruction, and some times segfault. These errors
are not exclusive of i8 x 8, but also apply to all vector types that I have
tried.

-- 
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/20180319/9365f31d/attachment.html>