[llvm-bugs] [Bug 38287] New: llvm.experimental.vector.reduce.{fmin, fmax} broken on Mips

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=38287

            Bug ID: 38287
           Summary: llvm.experimental.vector.reduce.{fmin,fmax} broken on
                    Mips
           Product: libraries
           Version: trunk
          Hardware: Other
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: P
         Component: Common Code Generator Code
          Assignee: unassignedbugs at nondot.org
          Reporter: dragan.mladjenovic at rt-rk.com
                CC: llvm-bugs at lists.llvm.org

Created attachment 20596
  --> https://bugs.llvm.org/attachment.cgi?id=20596&action=edit
Reproducer

Running the attached reproducer on trunk after rL332358 results in error:

LLVM ERROR: Cannot select: 0x55877f48eda0: v2i64 = setcc 0x55877f48ed38,
0x55877f48f218, setlt:ch
  0x55877f48ed38: v2f64 = bitcast 0x55877f48f350
    0x55877f48f350: v2i64 = insert_vector_elt 0x55877f48f280, 0x55877f48ec68,
Constant:i32<1>
      0x55877f48f280: v2i64 = insert_vector_elt 0x55877f48ecd0, 0x55877f48eb98,
Constant:i32<0>
        0x55877f48ecd0: v2i64 = BUILD_VECTOR Constant:i64<0>, Constant:i64<0>
          0x55877f48f010: i64 = Constant<0>
          0x55877f48f010: i64 = Constant<0>
        0x55877f48eb98: i64,ch = CopyFromReg 0x55877f429508, Register:i64 %0
          0x55877f48eb30: i64 = Register %0
        0x55877f48ef40: i32 = Constant<0>
      0x55877f48ec68: i64,ch = CopyFromReg 0x55877f429508, Register:i64 %1
        0x55877f48ec00: i64 = Register %1
      0x55877f48f2e8: i32 = Constant<1>
  0x55877f48f218: v2f64 = bitcast 0x55877f48efa8
    0x55877f48efa8: v2i64 = BUILD_VECTOR 0x55877f48ec68, 0x55877f48ec68
      0x55877f48ec68: i64,ch = CopyFromReg 0x55877f429508, Register:i64 %1
        0x55877f48ec00: i64 = Register %1
      0x55877f48ec68: i64,ch = CopyFromReg 0x55877f429508, Register:i64 %1
        0x55877f48ec00: i64 = Register %1

Before the rL332358 the 'setcc setolt' wasn't replaced with 'setcc setlt' which
Mips backend can't select on v2f64 types. 

Possibly a second issue is that reduction expansion produces same 'fcmp fast
olt' even when intrinsic is not used in fastmath context.


*** IR Dump After Expand reduction intrinsics ***
; Function Attrs: uwtable
define double @test(<2 x double>) unnamed_addr #0 {
start:
  %rdx.shuf = shufflevector <2 x double> %0, <2 x double> undef, <2 x i32> <i32
1, i32 undef>
  %rdx.minmax.cmp = fcmp fast olt <2 x double> %0, %rdx.shuf
  %rdx.minmax.select = select <2 x i1> %rdx.minmax.cmp, <2 x double> %0, <2 x
double> %rdx.shuf
  %1 = extractelement <2 x double> %rdx.minmax.select, i32 0
  ret double %1
}

-- 
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/20180724/06f81025/attachment-0001.html>