From 4389252c58b783ce5b62b1942ea4b85eed556fce Mon Sep 17 00:00:00 2001
From: Fangrui Song <i@maskray.me>
Date: Fri, 22 Sep 2023 00:14:31 -0700
Subject: [PATCH] Revert "[DAG] getNode() - remove oneuse limit from (zext
 (trunc (assertzext x))) -> (assertzext x) fold"

This reverts commit 05926a5a557878aa233ac8431b3acddf54422e58.

Caused AArch64 crash

 #12 0x00007f09eec09181 skipExtensionForVectorMULL(llvm::SDNode*, llvm::SelectionDAG&)
 #13 0x00007f09eec08289 llvm::AArch64TargetLowering::LowerMUL(llvm::SDValue, llvm::SelectionDAG&) const
 #14 0x00007f09eec1a3fd llvm::AArch64TargetLowering::LowerOperation(llvm::SDValue, llvm::SelectionDAG&) const
 #15 0x00007f09dc8586a7 (anonymous namespace)::VectorLegalizer::LowerOperationWrapper(llvm::SDNode*, llvm::SmallVectorImpl<llvm::SDValue>&)
---
 llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp          | 2 +-
 llvm/lib/Target/X86/X86ISelLoweringCall.cpp             | 3 +--
 llvm/test/CodeGen/AArch64/setcc_knownbits.ll            | 2 ++
 llvm/test/CodeGen/RISCV/rvv/fold-vp-fadd-and-vp-fmul.ll | 6 +++---
 4 files changed, 7 insertions(+), 6 deletions(-)

diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
index f12db53c7f08..c15f056551b9 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -5700,7 +5700,7 @@ SDValue SelectionDAG::getNode(unsigned Opcode, const SDLoc &DL, EVT VT,
     if (OpOpcode == ISD::TRUNCATE) {
       SDValue OpOp = N1.getOperand(0);
       if (OpOp.getValueType() == VT) {
-        if (OpOp.getOpcode() == ISD::AssertZext) {
+        if (OpOp.getOpcode() == ISD::AssertZext && N1->hasOneUse()) {
           APInt HiBits = APInt::getBitsSetFrom(VT.getScalarSizeInBits(),
                                                N1.getScalarValueSizeInBits());
           if (MaskedValueIsZero(OpOp, HiBits)) {
diff --git a/llvm/lib/Target/X86/X86ISelLoweringCall.cpp b/llvm/lib/Target/X86/X86ISelLoweringCall.cpp
index c47ddae072b4..754d2042105e 100644
--- a/llvm/lib/Target/X86/X86ISelLoweringCall.cpp
+++ b/llvm/lib/Target/X86/X86ISelLoweringCall.cpp
@@ -2645,8 +2645,7 @@ bool MatchingStackOffset(SDValue Arg, unsigned Offset, ISD::ArgFlagsTy Flags,
   for (;;) {
     // Look through nodes that don't alter the bits of the incoming value.
     unsigned Op = Arg.getOpcode();
-    if (Op == ISD::ZERO_EXTEND || Op == ISD::ANY_EXTEND || Op == ISD::BITCAST ||
-        Op == ISD::AssertZext) {
+    if (Op == ISD::ZERO_EXTEND || Op == ISD::ANY_EXTEND || Op == ISD::BITCAST) {
       Arg = Arg.getOperand(0);
       continue;
     }
diff --git a/llvm/test/CodeGen/AArch64/setcc_knownbits.ll b/llvm/test/CodeGen/AArch64/setcc_knownbits.ll
index af5c1586a4c6..aa62a7aa176c 100644
--- a/llvm/test/CodeGen/AArch64/setcc_knownbits.ll
+++ b/llvm/test/CodeGen/AArch64/setcc_knownbits.ll
@@ -4,6 +4,8 @@
 define i1 @load_bv_v4i8(i1 zeroext %a) {
 ; CHECK-LABEL: load_bv_v4i8:
 ; CHECK:       // %bb.0:
+; CHECK-NEXT:    cmp w0, #0
+; CHECK-NEXT:    cset w0, ne
 ; CHECK-NEXT:    ret
   %b = zext i1 %a to i32
   %c = icmp eq i32 %b, 1
diff --git a/llvm/test/CodeGen/RISCV/rvv/fold-vp-fadd-and-vp-fmul.ll b/llvm/test/CodeGen/RISCV/rvv/fold-vp-fadd-and-vp-fmul.ll
index f773de3b518c..1eccaaa26154 100644
--- a/llvm/test/CodeGen/RISCV/rvv/fold-vp-fadd-and-vp-fmul.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/fold-vp-fadd-and-vp-fmul.ll
@@ -62,9 +62,9 @@ define <vscale x 1 x double> @fma_reassociate(<vscale x 1 x double> %a, <vscale
 ; CHECK-LABEL: fma_reassociate:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vsetvli zero, a0, e64, m1, ta, ma
-; CHECK-NEXT:    vfmadd.vv v11, v10, v12, v0.t
-; CHECK-NEXT:    vfmadd.vv v9, v8, v11, v0.t
-; CHECK-NEXT:    vmv.v.v v8, v9
+; CHECK-NEXT:    vfmadd.vv v9, v8, v12, v0.t
+; CHECK-NEXT:    vfmadd.vv v11, v10, v9, v0.t
+; CHECK-NEXT:    vmv.v.v v8, v11
 ; CHECK-NEXT:    ret
   %1 = call fast <vscale x 1 x double> @llvm.vp.fmul.nxv1f64(<vscale x 1 x double> %a, <vscale x 1 x double> %b, <vscale x 1 x i1> %m, i32 %vl)
   %2 = call fast <vscale x 1 x double> @llvm.vp.fmul.nxv1f64(<vscale x 1 x double> %c, <vscale x 1 x double> %d, <vscale x 1 x i1> %m, i32 %vl)