[llvm-bugs] [Bug 37397] New: [GlobalISel][AArch64] i1 value in aggregate not zero-extend prior to store

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=37397

            Bug ID: 37397
           Summary: [GlobalISel][AArch64] i1 value in aggregate not
                    zero-extend prior to store
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: Backend: AArch64
          Assignee: unassignedbugs at nondot.org
          Reporter: tstellar at redhat.com
                CC: efriedma at codeaurora.org, jistone at redhat.com,
                    llvm-bugs at lists.llvm.org
            Blocks: 36649

This testcase comes from  https://github.com/rust-lang/rust/issues/50516 and
shows that GlobalISel does not zero-extend an i1 value to 8-bits prior to
writing it to memory:

target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
target triple = "aarch64-unknown-linux-gnu"
define void @i1_store() {
start:
  %tmp0 = alloca { i32, i1 }, align 4
  %tmp1 = call { i32, i1 } @foo()
  store { i32, i1 } %tmp1, { i32, i1 }* %tmp0, align 4
  ret void
}
declare { i32, i1 } @foo()

llc -O0 gives us:

        sub     sp, sp, #32             // =32
        str     x30, [sp, #16]          // 8-byte Folded Spill
        .cfi_def_cfa_offset 32
        .cfi_offset w30, -16
        bl      foo
                                        // implicit-def: $x8
        mov     w9, w0
        bfxil   x8, x9, #0, #32
        mov     w9, w1
        bfi     x8, x9, #32, #1
        str     x8, [sp, #8]


i1 values are stored in memory as 8-bit values, but it is expected that value
will be either 0x0 or 0x1.  In this case, since Bits 33-39 of x8 remain
uninitialized, when this value is loaded there will be random values in the
high 7-bits.

Fast-ISel seems to handle this test case correctly:


        sub     sp, sp, #32             // =32
        str     x30, [sp, #16]          // 8-byte Folded Spill
        .cfi_def_cfa_offset 32
        .cfi_offset w30, -16
        bl      foo
        str     w0, [sp, #8]
        and     w0, w1, #0x1
        strb    w0, [sp, #12]
        ldr     x30, [sp, #16]          // 8-byte Folded Reload
        add     sp, sp, #32             // =32
        ret


Referenced Bugs:

https://bugs.llvm.org/show_bug.cgi?id=36649
[Bug 36649] [meta] 6.0.1 Release Blockers
-- 
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/20180509/b73631be/attachment.html>