reference, declarationdefinition
definition → references, declarations, derived classes, virtual overrides
reference to multiple definitions → definitions
unreferenced
    1
    2
    3
    4
    5
    6
    7
    8
    9
   10
   11
   12
   13
   14
   15
   16
   17
   18
   19
   20
   21
   22
   23
   24
   25
   26
   27
   28
   29
   30
   31
   32
   33
   34
   35
   36
   37
   38
   39
   40
   41
# RUN: not llc -mtriple=aarch64-- -run-pass=legalizer %s -o - 2>&1 | FileCheck %s
# REQUIRES: asserts

# This is to demonstrate what kind of bugs we're missing w/o some kind
# of validation for LegalizerInfo: G_INTTOPTR could only be legal /
# could be legalized if its destination operand has a pointer type and
# its source - a scalar type of an appropriate size. This test meets
# the requirements for type index 0 (the pointer) and LLT-size
# requirements for type index 1 (64 bits for AArch64), but has a
# non-scalar (vector) type for type index 1. The Legalizer is expected
# to fail on it with an appropriate error message. Prior to
# LegalizerInfo::verify AArch64 legalizer had a subtle bug in its
# definition that caused it to accept the following MIR as legal.
# Namely, it checked that type index 0 is either s64 or p0 and
# implicitly declared any type for type index 1 as legal (as soon as
# its size is 64 bits). As LegalizerInfo::verify asserts on such a
# definition due to type index 1 not being covered by a specific
# action (not just `unsupportedIf`) it forces to review the definition
# and fix the mistake: check that type index 0 is p0 and type index 1
# is s64.

# CHECK: Bad machine code: operand types must be all-vector or all-scalar
# CHECK: LLVM ERROR: Found 1 machine code errors.

---
name:            broken
alignment:       4
tracksRegLiveness: true
registers:
  - { id: 0, class: _ }
  - { id: 1, class: _ }
body:             |
  bb.1:
    liveins: $d0

    %0:_(<4 x s16>) = COPY $d0
    %1:_(p0) = G_INTTOPTR %0(<4 x s16>)
    $x0 = COPY %1(p0)
    RET_ReallyLR implicit $x0

...