Skip to content

Commit

Permalink
[AArch64][GlobalISel] Constrain reg operands in selectBrJT
Browse files Browse the repository at this point in the history
This was causing a machine verifier failure on the test suite.

Make sure that we don't end up with a weird register class here.

Failure for reference:

*** Bad machine code: Illegal virtual register for instruction ***
- function:    check_constrain
- basic block: %bb.1  (0x7f8b70839f80)
- instruction: early-clobber %6:gpr64, early-clobber %7:gpr64sp =
  JumpTableDest32 %5:gpr64, %1:gpr64sp, %jump-table.0
- operand 3:   %1:gpr64sp
Expected a GPR64 register, but got a GPR64sp register

Differential Revision: https://reviews.llvm.org/D77349
  • Loading branch information
Jessica Paquette committed Apr 3, 2020
1 parent fe8ac0f commit 71947ed
Show file tree
Hide file tree
Showing 2 changed files with 65 additions and 5 deletions.
9 changes: 4 additions & 5 deletions llvm/lib/Target/AArch64/AArch64InstructionSelector.cpp
Expand Up @@ -2766,14 +2766,13 @@ bool AArch64InstructionSelector::selectBrJT(MachineInstr &I,

Register TargetReg = MRI.createVirtualRegister(&AArch64::GPR64RegClass);
Register ScratchReg = MRI.createVirtualRegister(&AArch64::GPR64spRegClass);
MIB.buildInstr(AArch64::JumpTableDest32, {TargetReg, ScratchReg},
{JTAddr, Index})
.addJumpTableIndex(JTI);

auto JumpTableInst = MIB.buildInstr(AArch64::JumpTableDest32,
{TargetReg, ScratchReg}, {JTAddr, Index})
.addJumpTableIndex(JTI);
// Build the indirect branch.
MIB.buildInstr(AArch64::BR, {}, {TargetReg});
I.eraseFromParent();
return true;
return constrainSelectedInstRegOperands(*JumpTableInst, TII, TRI, RBI);
}

bool AArch64InstructionSelector::selectJumpTable(
Expand Down
@@ -0,0 +1,61 @@
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -mtriple=aarch64-unknown-unknown -o - -verify-machineinstrs -run-pass=instruction-select %s | FileCheck %s

# When we select the G_ZEXTLOAD, the SUBREG_TO_REG will initially land on a
# gpr64sp register.
#
# This caused a test failure when selecting the G_BRJT, because it was not being
# constrained. This test checks that the G_BRJT is actually being constrained.
# As a result, the SUBREG_TO_REG should end up on a gpr64common.

...
---
name: check_constrain
legalized: true
regBankSelected: true
tracksRegLiveness: true
jumpTable:
kind: block-address
entries:
- id: 0
blocks: [ '%bb.4' ]
body: |
; CHECK-LABEL: name: check_constrain
; CHECK: bb.0:
; CHECK: successors: %bb.3(0x40000000), %bb.1(0x40000000)
; CHECK: [[DEF:%[0-9]+]]:gpr64common = IMPLICIT_DEF
; CHECK: [[LDRBBui:%[0-9]+]]:gpr32 = LDRBBui [[DEF]], 0 :: (load 1)
; CHECK: [[SUBREG_TO_REG:%[0-9]+]]:gpr64common = SUBREG_TO_REG 0, [[LDRBBui]], %subreg.sub_32
; CHECK: $xzr = SUBSXri [[SUBREG_TO_REG]], 8, 0, implicit-def $nzcv
; CHECK: Bcc 8, %bb.3, implicit $nzcv
; CHECK: bb.1:
; CHECK: successors: %bb.2(0x40000000), %bb.3(0x40000000)
; CHECK: [[MOVaddrJT:%[0-9]+]]:gpr64 = MOVaddrJT target-flags(aarch64-page) %jump-table.0, target-flags(aarch64-pageoff, aarch64-nc) %jump-table.0
; CHECK: early-clobber %6:gpr64, early-clobber %7:gpr64sp = JumpTableDest32 [[MOVaddrJT]], [[SUBREG_TO_REG]], %jump-table.0
; CHECK: BR %6
; CHECK: bb.2:
; CHECK: successors: %bb.3(0x80000000)
; CHECK: B %bb.3
; CHECK: bb.3:
; CHECK: RET_ReallyLR
bb.1:
%1:gpr(p0) = G_IMPLICIT_DEF
%5:gpr(s64) = G_ZEXTLOAD %1(p0) :: (load 1)
%7:gpr(s64) = G_CONSTANT i64 8
%16:gpr(s32) = G_ICMP intpred(ugt), %5(s64), %7
%8:gpr(s1) = G_TRUNC %16(s32)
G_BRCOND %8(s1), %bb.4
bb.2:
successors: %bb.3, %bb.4
%9:gpr(p0) = G_JUMP_TABLE %jump-table.0
G_BRJT %9(p0), %jump-table.0, %5(s64)
bb.3:
G_BR %bb.4
bb.4:
RET_ReallyLR
...

0 comments on commit 71947ed

Please sign in to comment.