[AVR] Fix a backend bug that left extraneous operands after expansion

This patch fixes a bug in the AVR FRMIDX expansion logic.

The expansion would leave a leftover operand from the original FRMIDX,
but now attached to a MOVWRdRr instruction. The MOVWRdRr instruction
did not expect this operand and so LLVM rejected the machine
instruction.

This would trigger an assertion:

    Assertion failed: ((isImpReg || Op.isRegMask() || MCID->isVariadic() ||
                        OpNo < MCID->getNumOperands() || isMetaDataOp) &&
                        "Trying to add an operand to a machine instr that is already done!"),
    function addOperand, file llvm/lib/CodeGen/MachineInstr.cpp

Tim fixed this so that now the FRMIDX is expanded correctly into
a well-formed MOVWRdRr.

Patch by Tim Neumann

llvm-svn: 346117
This commit is contained in:
Dylan McKay 2018-11-05 05:49:04 +00:00
parent 1e255e7a7b
commit 4c5a5c8db6
2 changed files with 49 additions and 0 deletions

View File

@ -152,6 +152,7 @@ void AVRRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
if (MI.getOpcode() == AVR::FRMIDX) {
MI.setDesc(TII.get(AVR::MOVWRdRr));
MI.getOperand(FIOperandNum).ChangeToRegister(AVR::R29R28, false);
MI.RemoveOperand(2);
assert(Offset > 0 && "Invalid offset");

View File

@ -0,0 +1,48 @@
; RUN: llc < %s -march=avr | FileCheck %s
; The avr-rust bug can be found here:
; https://github.com/avr-rust/rust/issues/112
;
; In this test, the codegen stage generates a FRMIDX
; instruction. Later in the pipeline, the frame index
; gets expanded into a 16-bit MOVWRdRr instruction.
;
; There was a bug in the FRMIDX->MOVWRdRr expansion logic
; that could leave the MOVW instruction with an extraneous
; operand, left over from the original FRMIDX.
;
; This would trigger an assertion:
;
; Assertion failed: ((isImpReg || Op.isRegMask() || MCID->isVariadic() ||
; OpNo < MCID->getNumOperands() || isMetaDataOp) &&
; "Trying to add an operand to a machine instr that is already done!"),
; function addOperand, file llvm/lib/CodeGen/MachineInstr.cpp
;
; The logic has since been fixed.
; CHECK-LABEL: "core::str::slice_error_fail"
define void @"core::str::slice_error_fail"(i16 %arg) personality i32 (...) addrspace(1)* @rust_eh_personality {
start:
%char_range = alloca { i16, i16 }, align 1
br i1 undef, label %"<core::option::Option<T>>::unwrap.exit.thread", label %bb11.i.i
"<core::option::Option<T>>::unwrap.exit.thread":
br label %"core::char::methods::<impl char>::len_utf8.exit"
bb11.i.i:
%tmp = bitcast { i16, i16 }* %char_range to i8*
%tmp1 = icmp ult i32 undef, 65536
%..i = select i1 %tmp1, i16 3, i16 4
br label %"core::char::methods::<impl char>::len_utf8.exit"
"core::char::methods::<impl char>::len_utf8.exit":
%tmp2 = phi i8* [ %tmp, %bb11.i.i ], [ undef, %"<core::option::Option<T>>::unwrap.exit.thread" ]
%_0.0.i12 = phi i16 [ %..i, %bb11.i.i ], [ 1, %"<core::option::Option<T>>::unwrap.exit.thread" ]
%tmp3 = add i16 %_0.0.i12, %arg
store i16 %tmp3, i16* undef, align 1
store i8* %tmp2, i8** undef, align 1
unreachable
}
declare i32 @rust_eh_personality(...) addrspace(1)