Assertion failure during compilation to assembly

[seanmkauffman]

llc fourierf.ll -mtriple avr-none -o fourierf.s

llc: /home/seanmk/workspace/llvm/lib/Target/AVR/AVRISelLowering.cpp:1137: virtual llvm::SDValue llvm::AVRTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &, SmallVectorImpl<llvm::SDValue> &) const: Assertion `Mask && "Missing call preserved mask for calling convention"' failed.
#0 0x1ffb4ae llvm::sys::PrintStackTrace(llvm::raw_ostream&) /home/seanmk/workspace/llvm/lib/Support/Unix/Signals.inc:424:15
#1 0x1ffc509 PrintStackTraceSignalHandler(void*) /home/seanmk/workspace/llvm/lib/Support/Unix/Signals.inc:483:1
#2 0x1fff5a4 SignalHandler(int) /home/seanmk/workspace/llvm/lib/Support/Unix/Signals.inc:199:60
#3 0x7fb68080a250 __restore_rt (/lib64/libpthread.so.0+0x11250)
#4 0x7fb67f80d795 gsignal (/lib64/libc.so.6+0x38795)
#5 0x7fb67f80ec18 abort (/lib64/libc.so.6+0x39c18)
#6 0x7fb67f806812 (/lib64/libc.so.6+0x31812)
#7 0x7fb67f8068c2 (/lib64/libc.so.6+0x318c2)
#8 0xbdf3cb llvm::AVRTargetLowering::LowerCall(llvm::TargetLowering::CallLoweringInfo&, llvm::SmallVectorImpl<llvm::SDValue>&) const /home/seanmk/workspace/llvm/lib/Target/AVR/AVRISelLowering.cpp:1137:3
#9 0x146f9d8 llvm::TargetLowering::LowerCallTo(llvm::TargetLowering::CallLoweringInfo&) const /home/seanmk/workspace/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp:7309:15
#10 0x1471307 llvm::SelectionDAGBuilder::lowerInvokable(llvm::TargetLowering::CallLoweringInfo&, llvm::MachineBasicBlock*) /home/seanmk/workspace/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp:5484:3
#11 0x1455a0f llvm::SelectionDAGBuilder::LowerCallTo(llvm::ImmutableCallSite, llvm::SDValue, bool, llvm::MachineBasicBlock*) /home/seanmk/workspace/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp:5550:39
#12 0x1443740 llvm::SelectionDAGBuilder::visitCall(llvm::CallInst const&) /home/seanmk/workspace/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp:6086:1
#13 0x143ad2e llvm::SelectionDAGBuilder::visit(unsigned int, llvm::User const&) /home/seanmk/workspace/llvm/include/llvm/IR/Instruction.def:165:1
#14 0x143a223 llvm::SelectionDAGBuilder::visit(llvm::Instruction const&) /home/seanmk/workspace/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp:967:3
#15 0x149f95c llvm::SelectionDAGISel::SelectBasicBlock(llvm::ilist_iterator<llvm::Instruction const>, llvm::ilist_iterator<llvm::Instruction const>, bool&) /home/seanmk/workspace/llvm/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp:601:5
#16 0x149f3e1 llvm::SelectionDAGISel::SelectAllBasicBlocks(llvm::Function const&) /home/seanmk/workspace/llvm/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp:1328:5
#17 0x149d287 llvm::SelectionDAGISel::runOnMachineFunction(llvm::MachineFunction&) /home/seanmk/workspace/llvm/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp:464:33
#18 0xbd041b (anonymous namespace)::AVRDAGToDAGISel::runOnMachineFunction(llvm::MachineFunction&) /home/seanmk/workspace/llvm/lib/Target/AVR/AVRISelDAGToDAG.cpp:70:14
#19 0x170d2ee llvm::MachineFunctionPass::runOnFunction(llvm::Function&) /home/seanmk/workspace/llvm/lib/CodeGen/MachineFunctionPass.cpp:40:10
#20 0x1f31100 llvm::FPPassManager::runOnFunction(llvm::Function&) /home/seanmk/workspace/llvm/lib/IR/LegacyPassManager.cpp:1538:23
#21 0x1f31408 llvm::FPPassManager::runOnModule(llvm::Module&) /home/seanmk/workspace/llvm/lib/IR/LegacyPassManager.cpp:1558:16
#22 0x1f31adc (anonymous namespace)::MPPassManager::runOnModule(llvm::Module&) /home/seanmk/workspace/llvm/lib/IR/LegacyPassManager.cpp:1616:23
#23 0x1f316be llvm::legacy::PassManagerImpl::run(llvm::Module&) /home/seanmk/workspace/llvm/lib/IR/LegacyPassManager.cpp:1723:16
#24 0x1f320a1 llvm::legacy::PassManager::run(llvm::Module&) /home/seanmk/workspace/llvm/lib/IR/LegacyPassManager.cpp:1756:10
#25 0x751cb0 compileModule(char**, llvm::LLVMContext&) /home/seanmk/workspace/llvm/tools/llc/llc.cpp:351:3
#26 0x750bf2 main /home/seanmk/workspace/llvm/tools/llc/llc.cpp:200:22
#27 0x7fb67f7f9db5 __libc_start_main (/lib64/libc.so.6+0x24db5)
#28 0x74e3c9 _start (/home/seanmk/workspace/avr-work-newer/bin/llc+0x74e3c9)
Stack dump:
0.  Program arguments: /home/seanmk/workspace/avr-work-newer/bin/llc -mtriple=avr-none fourierf.ll -o fourierf.s 
1.  Running pass 'Function Pass Manager' on module 'fourierf.ll'.
2.  Running pass 'AVR DAG->DAG Instruction Selection' on function '@fft_float'

fourierf.ll:

; ModuleID = 'fourierf.c'
target datalayout = "e-p:16:8:8-i8:8:8-i16:8:8-i32:8:8-i64:8:8-f32:8:8-f64:8:8-n8"
target triple = "avr-none-none"

%struct.__file = type { i8*, i8, i8, i16, i16, i16 (i8, %struct.__file*)*, i16 (%struct.__file*)*, i8* }

@.str = private unnamed_addr constant [7 x i8] c"RealIn\00", align 1
@.str1 = private unnamed_addr constant [8 x i8] c"RealOut\00", align 1
@.str2 = private unnamed_addr constant [8 x i8] c"ImagOut\00", align 1
@__iob = external global [0 x %struct.__file*]
@.str3 = private unnamed_addr constant [35 x i8] c"Error in fft_float():  %s == NULL\0A\00", align 1

; Function Attrs: nounwind
define void @fft_float(i16 %NumSamples, i16 %InverseTransform, float* readonly %RealIn, float* readonly %ImagIn, float* %RealOut, float* %ImagOut) #0 {
entry:
  %tobool = icmp ne i16 %InverseTransform, 0
  %. = select i1 %tobool, double 0xC01921FB54442D18, double 0x401921FB54442D18
  %0 = bitcast float* %RealIn to i8*
  tail call fastcc void @CheckPointer(i8* %0, i8* getelementptr inbounds ([7 x i8], [7 x i8]* @.str, i16 0, i16 0))
  %1 = bitcast float* %RealOut to i8*
  tail call fastcc void @CheckPointer(i8* %1, i8* getelementptr inbounds ([8 x i8], [8 x i8]* @.str1, i16 0, i16 0))
  %2 = bitcast float* %ImagOut to i8*
  tail call fastcc void @CheckPointer(i8* %2, i8* getelementptr inbounds ([8 x i8], [8 x i8]* @.str2, i16 0, i16 0))
  %call = tail call i16 @NumberOfBitsNeeded(i16 %NumSamples) #4
  %cmp195 = icmp eq i16 %NumSamples, 0
  br i1 %cmp195, label %for.end94, label %for.body.lr.ph

for.body.lr.ph:                                   ; preds = %entry
  %cmp3 = icmp eq float* %ImagIn, null
  br label %for.body

for.cond7.preheader:                              ; preds = %cond.end
  %cmp8191 = icmp ult i16 %NumSamples, 2
  br i1 %cmp8191, label %for.end94, label %for.body10

for.body:                                         ; preds = %for.body.lr.ph, %cond.end
  %i.0196 = phi i16 [ 0, %for.body.lr.ph ], [ %inc, %cond.end ]
  %call1 = tail call i16 @ReverseBits(i16 %i.0196, i16 %call) #4
  %arrayidx = getelementptr inbounds float, float* %RealIn, i16 %i.0196
  %3 = load float, float* %arrayidx, align 4, !tbaa !1
  %arrayidx2 = getelementptr inbounds float, float* %RealOut, i16 %call1
  store float %3, float* %arrayidx2, align 4, !tbaa !1
  br i1 %cmp3, label %cond.end, label %cond.false

cond.false:                                       ; preds = %for.body
  %arrayidx4 = getelementptr inbounds float, float* %ImagIn, i16 %i.0196
  %4 = load float, float* %arrayidx4, align 4, !tbaa !1
  %conv = fpext float %4 to double
  br label %cond.end

cond.end:                                         ; preds = %for.body, %cond.false
  %cond = phi double [ %conv, %cond.false ], [ 0.000000e+00, %for.body ]
  %conv5 = fptrunc double %cond to float
  %arrayidx6 = getelementptr inbounds float, float* %ImagOut, i16 %call1
  store float %conv5, float* %arrayidx6, align 4, !tbaa !1
  %inc = add nuw i16 %i.0196, 1
  %cmp = icmp ult i16 %inc, %NumSamples
  br i1 %cmp, label %for.body, label %for.cond7.preheader

for.body10:                                       ; preds = %for.cond7.preheader, %for.end92
  %BlockSize.0193 = phi i16 [ %shl, %for.end92 ], [ 2, %for.cond7.preheader ]
  %BlockEnd.0192 = phi i16 [ %BlockSize.0193, %for.end92 ], [ 1, %for.cond7.preheader ]
  %conv11 = uitofp i16 %BlockSize.0193 to double
  %div = fdiv double %., %conv11
  %mul = fmul double %div, -2.000000e+00
  %call12 = tail call double @sin(double %mul) #5
  %sub13 = fsub double -0.000000e+00, %div
  %call14 = tail call double @sin(double %sub13) #5
  %call16 = tail call double @cos(double %mul) #5
  %cos = tail call double @cos(double %div)
  %mul19 = fmul double %cos, 2.000000e+00
  %cmp29182 = icmp eq i16 %BlockEnd.0192, 0
  br label %for.cond28.preheader

for.cond28.preheader:                             ; preds = %for.body10, %for.inc90
  %i.1190 = phi i16 [ 0, %for.body10 ], [ %add91, %for.inc90 ]
  br i1 %cmp29182, label %for.inc90, label %for.body31

for.body31:                                       ; preds = %for.cond28.preheader, %for.body31
  %j.0188 = phi i16 [ %inc87, %for.body31 ], [ %i.1190, %for.cond28.preheader ]
  %ai.sroa.4.0187 = phi double [ %sub44, %for.body31 ], [ %call14, %for.cond28.preheader ]
  %ai.sroa.8.0186 = phi double [ %ai.sroa.4.0187, %for.body31 ], [ %call12, %for.cond28.preheader ]
  %n.0185 = phi i16 [ %inc88, %for.body31 ], [ 0, %for.cond28.preheader ]
  %ar.sroa.4.0184 = phi double [ %sub35, %for.body31 ], [ %cos, %for.cond28.preheader ]
  %ar.sroa.8.0183 = phi double [ %ar.sroa.4.0184, %for.body31 ], [ %call16, %for.cond28.preheader ]
  %mul33 = fmul double %mul19, %ar.sroa.4.0184
  %sub35 = fsub double %mul33, %ar.sroa.8.0183
  %mul42 = fmul double %mul19, %ai.sroa.4.0187
  %sub44 = fsub double %mul42, %ai.sroa.8.0186
  %add = add i16 %j.0188, %BlockEnd.0192
  %arrayidx51 = getelementptr inbounds float, float* %RealOut, i16 %add
  %5 = load float, float* %arrayidx51, align 4, !tbaa !1
  %conv52 = fpext float %5 to double
  %mul53 = fmul double %sub35, %conv52
  %arrayidx55 = getelementptr inbounds float, float* %ImagOut, i16 %add
  %6 = load float, float* %arrayidx55, align 4, !tbaa !1
  %conv56 = fpext float %6 to double
  %mul57 = fmul double %sub44, %conv56
  %sub58 = fsub double %mul53, %mul57
  %mul62 = fmul double %sub35, %conv56
  %mul66 = fmul double %sub44, %conv52
  %add67 = fadd double %mul66, %mul62
  %arrayidx68 = getelementptr inbounds float, float* %RealOut, i16 %j.0188
  %7 = load float, float* %arrayidx68, align 4, !tbaa !1
  %conv69 = fpext float %7 to double
  %sub70 = fsub double %conv69, %sub58
  %conv71 = fptrunc double %sub70 to float
  store float %conv71, float* %arrayidx51, align 4, !tbaa !1
  %arrayidx73 = getelementptr inbounds float, float* %ImagOut, i16 %j.0188
  %8 = load float, float* %arrayidx73, align 4, !tbaa !1
  %conv74 = fpext float %8 to double
  %sub75 = fsub double %conv74, %add67
  %conv76 = fptrunc double %sub75 to float
  store float %conv76, float* %arrayidx55, align 4, !tbaa !1
  %9 = load float, float* %arrayidx68, align 4, !tbaa !1
  %conv79 = fpext float %9 to double
  %add80 = fadd double %sub58, %conv79
  %conv81 = fptrunc double %add80 to float
  store float %conv81, float* %arrayidx68, align 4, !tbaa !1
  %10 = load float, float* %arrayidx73, align 4, !tbaa !1
  %conv83 = fpext float %10 to double
  %add84 = fadd double %add67, %conv83
  %conv85 = fptrunc double %add84 to float
  store float %conv85, float* %arrayidx73, align 4, !tbaa !1
  %inc87 = add i16 %j.0188, 1
  %inc88 = add nuw i16 %n.0185, 1
  %cmp29 = icmp ult i16 %inc88, %BlockEnd.0192
  br i1 %cmp29, label %for.body31, label %for.inc90

for.inc90:                                        ; preds = %for.body31, %for.cond28.preheader
  %add91 = add i16 %i.1190, %BlockSize.0193
  %cmp21 = icmp ult i16 %add91, %NumSamples
  br i1 %cmp21, label %for.cond28.preheader, label %for.end92

for.end92:                                        ; preds = %for.inc90
  %shl = shl i16 %BlockSize.0193, 1
  %cmp8 = icmp ugt i16 %shl, %NumSamples
  br i1 %cmp8, label %for.end94, label %for.body10

for.end94:                                        ; preds = %for.end92, %entry, %for.cond7.preheader
  br i1 %tobool, label %if.then96, label %if.end113

if.then96:                                        ; preds = %for.end94
  %conv97 = uitofp i16 %NumSamples to double
  %cmp99180 = icmp eq i16 %NumSamples, 0
  br i1 %cmp99180, label %if.end113, label %for.body101

for.body101:                                      ; preds = %if.then96, %for.body101
  %i.2181 = phi i16 [ %inc111, %for.body101 ], [ 0, %if.then96 ]
  %arrayidx102 = getelementptr inbounds float, float* %RealOut, i16 %i.2181
  %11 = load float, float* %arrayidx102, align 4, !tbaa !1
  %conv103 = fpext float %11 to double
  %div104 = fdiv double %conv103, %conv97
  %conv105 = fptrunc double %div104 to float
  store float %conv105, float* %arrayidx102, align 4, !tbaa !1
  %arrayidx106 = getelementptr inbounds float, float* %ImagOut, i16 %i.2181
  %12 = load float, float* %arrayidx106, align 4, !tbaa !1
  %conv107 = fpext float %12 to double
  %div108 = fdiv double %conv107, %conv97
  %conv109 = fptrunc double %div108 to float
  store float %conv109, float* %arrayidx106, align 4, !tbaa !1
  %inc111 = add nuw i16 %i.2181, 1
  %cmp99 = icmp ult i16 %inc111, %NumSamples
  br i1 %cmp99, label %for.body101, label %if.end113

if.end113:                                        ; preds = %for.body101, %if.then96, %for.end94
  ret void
}

; Function Attrs: nounwind
define internal fastcc void @CheckPointer(i8* readnone %p, i8* %name) #0 {
entry:
  %cmp = icmp eq i8* %p, null
  br i1 %cmp, label %if.then, label %if.end

if.then:                                          ; preds = %entry
  %0 = load %struct.__file*, %struct.__file** getelementptr inbounds ([0 x %struct.__file*], [0 x %struct.__file*]* @__iob, i16 0, i16 2), align 2, !tbaa !5
  %call = tail call i16 (%struct.__file*, i8*, ...)* @fprintf(%struct.__file* %0, i8* getelementptr inbounds ([35 x i8], [35 x i8]* @.str3, i16 0, i16 0), i8* %name) #4
  tail call void @exit(i16 1) #6
  unreachable

if.end:                                           ; preds = %entry
  ret void
}

declare i16 @NumberOfBitsNeeded(i16) #1

declare i16 @ReverseBits(i16, i16) #1

; Function Attrs: nounwind readnone
declare double @sin(double) #2

; Function Attrs: nounwind readnone
declare double @cos(double) #2

; Function Attrs: nounwind
declare i16 @fprintf(%struct.__file* nocapture, i8* nocapture readonly, ...) #0

; Function Attrs: noreturn
declare void @exit(i16) #3

attributes #0 = { nounwind "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
attributes #1 = { "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
attributes #2 = { nounwind readnone "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
attributes #3 = { noreturn "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
attributes #4 = { nounwind }
attributes #5 = { nounwind readnone }
attributes #6 = { noreturn nounwind }

!llvm.ident = !{!0}

!0 = !{!"clang version 3.7.0 (https://github.com/avr-llvm/clang.git 00a0b3863ac817790d9b344da4fc499b71140a7a) (llvm/llvm.git 70f966eb9eb5e2cae5d4f62ae213ae035677efee)"}
!1 = !{!2, !2, i64 0}
!2 = !{!"float", !3, i64 0}
!3 = !{!"omnipotent char", !4, i64 0}
!4 = !{!"Simple C/C++ TBAA"}
!5 = !{!6, !6, i64 0}
!6 = !{!"any pointer", !3, i64 0}

[dylanmckay]

This was caused because the AVRRegisterInfo::getCallPreservedMask function was not actually overriding the virtual function it should have been, because the signature had been changed and AVRRegisterInfo::getCallPreservedMask was not qualified with the override keyword, so the problem happened silently.

I have audited all classes in the backend, and added missing override keywords where they were needed.

[dylanmckay]

Note that when compiling your test case, compilation fails with another error: "instruction selection failed".

This is presumably because the MCU targeted (by default, avr2) does not support the operations used (such as multiplication/division, I'm not sure which).

If you select one of the higher end MCUs (such as avrxmega7, by using llc fourierf.ll -march=avr -mcpu=avrxmega7), instruction selection completes, but another error is hit - #37.

[dylanmckay]

I have opened up an issue for the instruction selection error - #76.

[seanmkauffman]

Thanks for the help. FYI, avr-llvm will not compile right now due to an error in the clang branch. I filed an issue. avr-llvm/clang#3

[dylanmckay]

Now that #37 is fixed, I can run llc on fourierf.ll and it succeeds.