[SelectionDAG] Check CALLSEQ_BEGIN nodes in DelayForLiveRegs

[samparker]

Summary

This is an alternative patch for D29492, the bug was originally reported in https://bugs.llvm.org/show_bug.cgi?id=30911.

The issue arises when multiple CALLSEQ_BEGIN nodes are unscheduled as the last node to be unscheduled will gain access to the CallResource register. But when a node is being picked, only CALLSEQ_END nodes are checked against the CallResource and have their chains evaluated. This then means that other CALLSEQ_BEGIN nodes can be scheduled before the existing call sequence has been finalised. This patch adds a check against the FrameSetup nodes in DelayForLiveRegs to prevent this from happening.

[grosser]

Thanks Sam, you are fast! This patch fixes the crash for me and all test cases still pass. Very nice!

I am not an expert in this area, so I will let others review the patch. Thanks again for helping make the AOSP buildbot green!

lib/CodeGen/SelectionDAG/ScheduleDAGRRList.cpp

1309 ↗

(On Diff #93645)

There seems to be something wrong with the opening braces here. I had to fix them before this compiled.

[atrick]

I don't remember (or never understood) how all this code works. But the fix looks reasonable.

[MatzeB]

This needs a test!

[samparker]

Added the test already provided in D29492. Also fixed the typo.

[grosser]

Very nice. I think this is good to go.

llvm/trunk/lib/CodeGen/SelectionDAG/ScheduleDAGRRList.cpp

if (Node->getOpcode() == ISD::INLINEASM) {
       continue;
     }
 
     if (!Node->isMachineOpcode())
       continue;
     // If we're in the middle of scheduling a call, don't begin scheduling
     // another call. Also, don't allow any physical registers to be live across
     // the call.
-    if (Node->getMachineOpcode() == (unsigned)TII->getCallFrameDestroyOpcode()) {
+    if ((Node->getMachineOpcode() == TII->getCallFrameDestroyOpcode()) ||
+        (Node->getMachineOpcode() == TII->getCallFrameSetupOpcode())) {
       // Check the special calling-sequence resource.
       unsigned CallResource = TRI->getNumRegs();
       if (LiveRegDefs[CallResource]) {
         SDNode *Gen = LiveRegGens[CallResource]->getNode();
         while (SDNode *Glued = Gen->getGluedNode())
           Gen = Glued;
         if (!IsChainDependent(Gen, Node, 0, TII) &&
             RegAdded.insert(CallResource).second)

llvm/trunk/test/CodeGen/ARM/unschedule-first-call.ll

+; RUN: llc < %s
+; PR30911
+
+target datalayout = "e-m:e-p:32:32-i64:64-v128:64:128-a:0:32-n32-S64"
+target triple = "armv6kz--linux-gnueabihf"
+
+; Function Attrs: nounwind
+define void @dradbg(i32, i32, float*, float*, float*, float*, float*) #0 {
+  br i1 undef, label %.critedge, label %8
+
+.critedge:                                        ; preds = %7
+  %.mux2 = select i1 undef, i1 undef, i1 true
+  br label %8
+
+; <label>:8:                                      ; preds = %.critedge, %7
+  %9 = getelementptr float, float* %3, i64 undef
+  %10 = ptrtoint float* %9 to i32
+  %11 = icmp ule i32 %10, undef
+  %12 = getelementptr float, float* %5, i64 undef
+  %13 = call { i64, i1 } @llvm.sadd.with.overflow.i64(i64 undef, i64 undef)
+  %14 = extractvalue { i64, i1 } %13, 0
+  %15 = call { i64, i1 } @llvm.sadd.with.overflow.i64(i64 %14, i64 1)
+  %16 = extractvalue { i64, i1 } %15, 0
+  %17 = icmp slt i64 1, %16
+  %18 = select i1 %17, i64 1, i64 %16
+  %19 = sext i32 %1 to i64
+  %20 = call { i64, i1 } @llvm.smul.with.overflow.i64(i64 %18, i64 %19)
+  %21 = extractvalue { i64, i1 } %20, 0
+  %22 = call { i64, i1 } @llvm.sadd.with.overflow.i64(i64 %21, i64 0)
+  %23 = extractvalue { i64, i1 } %22, 0
+  %24 = call { i64, i1 } @llvm.smul.with.overflow.i64(i64 %23, i64 undef)
+  %25 = extractvalue { i64, i1 } %24, 0
+  %26 = call { i64, i1 } @llvm.sadd.with.overflow.i64(i64 %25, i64 0)
+  %27 = extractvalue { i64, i1 } %26, 0
+  %28 = getelementptr float, float* %3, i64 %27
+  %29 = ptrtoint float* %12 to i32
+  %30 = ptrtoint float* %28 to i32
+  %31 = icmp ule i32 %29, %30
+  %32 = or i1 %11, %31
+  %33 = and i1 false, %32
+  %34 = call { i64, i1 } @llvm.sadd.with.overflow.i64(i64 0, i64 undef)
+  %35 = extractvalue { i64, i1 } %34, 0
+  %36 = call { i64, i1 } @llvm.sadd.with.overflow.i64(i64 %35, i64 1)
+  %37 = extractvalue { i64, i1 } %36, 0
+  %38 = icmp slt i64 1, %37
+  %39 = select i1 %38, i64 1, i64 %37
+  %40 = sext i32 %1 to i64
+  %41 = call { i64, i1 } @llvm.smul.with.overflow.i64(i64 %39, i64 %40)
+  %42 = extractvalue { i64, i1 } %41, 0
+  %43 = call { i64, i1 } @llvm.sadd.with.overflow.i64(i64 %42, i64 0)
+  %44 = extractvalue { i64, i1 } %43, 0
+  %45 = sext i32 %0 to i64
+  %46 = call { i64, i1 } @llvm.smul.with.overflow.i64(i64 %44, i64 %45)
+  %47 = extractvalue { i64, i1 } %46, 0
+  %48 = call { i64, i1 } @llvm.sadd.with.overflow.i64(i64 %47, i64 0)
+  %49 = extractvalue { i64, i1 } %48, 0
+  %50 = getelementptr float, float* %5, i64 %49
+  %51 = ptrtoint float* %50 to i32
+  %52 = icmp ule i32 undef, %51
+  %53 = getelementptr float, float* %4, i64 undef
+  %54 = ptrtoint float* %53 to i32
+  %55 = icmp ule i32 undef, %54
+  %56 = or i1 %52, %55
+  %57 = and i1 %33, %56
+  %58 = getelementptr float, float* %2, i64 undef
+  %59 = ptrtoint float* %58 to i32
+  %60 = icmp ule i32 %59, undef
+  %61 = select i1 undef, i64 undef, i64 0
+  %62 = call { i64, i1 } @llvm.smul.with.overflow.i64(i64 %61, i64 undef)
+  %63 = extractvalue { i64, i1 } %62, 0
+  %64 = call { i64, i1 } @llvm.ssub.with.overflow.i64(i64 undef, i64 1)
+  %65 = extractvalue { i64, i1 } %64, 0
+  %66 = call { i64, i1 } @llvm.sadd.with.overflow.i64(i64 %63, i64 %65)
+  %67 = extractvalue { i64, i1 } %66, 0
+  %68 = sext i32 %0 to i64
+  %69 = call { i64, i1 } @llvm.smul.with.overflow.i64(i64 %67, i64 %68)
+  %70 = extractvalue { i64, i1 } %69, 0
+  %71 = call { i64, i1 } @llvm.sadd.with.overflow.i64(i64 %70, i64 0)
+  %72 = extractvalue { i64, i1 } %71, 0
+  %73 = getelementptr float, float* %5, i64 %72
+  %74 = ptrtoint float* %73 to i32
+  %75 = icmp ule i32 %74, undef
+  %76 = or i1 %60, %75
+  %77 = and i1 %57, %76
+  %78 = getelementptr float, float* %6, i64 undef
+  %79 = ptrtoint float* %78 to i32
+  %80 = icmp ule i32 %79, undef
+  %81 = getelementptr float, float* %5, i64 undef
+  %82 = ptrtoint float* %81 to i32
+  %83 = icmp ule i32 %82, undef
+  %84 = or i1 %80, %83
+  %85 = and i1 %77, %84
+  %86 = and i1 %85, undef
+  %87 = and i1 %86, undef
+  %88 = and i1 %87, undef
+  %89 = and i1 %88, undef
+  %90 = and i1 %89, undef
+  %91 = and i1 %90, undef
+  %92 = and i1 %91, undef
+  %93 = and i1 %92, undef
+  %94 = and i1 %93, undef
+  %95 = and i1 %94, undef
+  br i1 %95, label %97, label %96
+
+; <label>:96:                                     ; preds = %8
+  br i1 undef, label %.critedge122, label %.critedge110
+
+.critedge122:                                     ; preds = %.critedge122, %96
+  br i1 false, label %.critedge122, label %.critedge110
+
+.critedge110:                                     ; preds = %.critedge219, %97, %.critedge122, %96
+  ret void
+
+; <label>:97:                                     ; preds = %8
+  br i1 undef, label %.critedge219, label %.critedge110
+
+.critedge219:                                     ; preds = %.critedge219, %97
+  %.pr287 = phi i1 [ undef, %.critedge219 ], [ true, %97 ]
+  br i1 %.pr287, label %.critedge219, label %.critedge110
+}
+
+; Function Attrs: nounwind readnone
+declare { i64, i1 } @llvm.smul.with.overflow.i64(i64, i64) #1
+
+; Function Attrs: nounwind readnone
+declare { i64, i1 } @llvm.sadd.with.overflow.i64(i64, i64) #1
+
+; Function Attrs: nounwind readnone
+declare { i64, i1 } @llvm.ssub.with.overflow.i64(i64, i64) #1
+
+attributes #0 = { nounwind "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "polly-optimized" "stack-protector-buffer-size"="8" "target-cpu"="arm1176jzf-s" "target-features"="+dsp,+strict-align,+vfp2" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #1 = { nounwind readnone }
+
+!llvm.ident = !{!0}
+
+!0 = !{!"clang version 4.0.0 (trunk 285923) (llvm/trunk 285921)"}