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
|
# RUN: not --crash llc -mtriple=aarch64-- -run-pass=legalizer %s -o - 2>&1 | FileCheck %s
# 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
...
|
