[llvm] 9273f86 - [RISCV] Prevent performCombineVMergeAndVOps from creating cycles in the DAG.

[Craig Topper via llvm-commits]

Author: Craig Topper
Date: 2022-09-30T20:01:45-07:00
New Revision: 9273f860c089fd94a4f929ff24da7b331d70f64d

URL: https://github.com/llvm/llvm-project/commit/9273f860c089fd94a4f929ff24da7b331d70f64d
DIFF: https://github.com/llvm/llvm-project/commit/9273f860c089fd94a4f929ff24da7b331d70f64d.diff

LOG: [RISCV] Prevent performCombineVMergeAndVOps from creating cycles in the DAG.

If True has a Chain result, the other operands of the vmerge may
depend on it through that Chain. We need to ensure it isn't a
predecessor of those operands.

Reviewed By: fakepaper56

Differential Revision: https://reviews.llvm.org/D134980

Added: 
    

Modified: 
    llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp
    llvm/test/CodeGen/RISCV/rvv/rvv-peephole-vmerge-vops.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp b/llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp
index 2a6bdb2fb752..be1656518119 100644
--- a/llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp
@@ -2667,6 +2667,20 @@ bool RISCVDAGToDAGISel::performCombineVMergeAndVOps(SDNode *N, bool IsTA) {
   bool HasChainOp =
       True.getOperand(True.getNumOperands() - 1).getValueType() == MVT::Other;
 
+  if (HasChainOp) {
+    // Avoid creating cycles in the DAG. We must ensure that none of the other
+    // operands depend on True through it's Chain.
+    SmallVector<const SDNode *, 4> LoopWorklist;
+    SmallPtrSet<const SDNode *, 16> Visited;
+    LoopWorklist.push_back(False.getNode());
+    LoopWorklist.push_back(Mask.getNode());
+    LoopWorklist.push_back(VL.getNode());
+    if (SDNode *Glued = N->getGluedNode())
+      LoopWorklist.push_back(Glued);
+    if (SDNode::hasPredecessorHelper(True.getNode(), Visited, LoopWorklist))
+      return false;
+  }
+
   // Need True has same VL with N.
   unsigned TrueVLIndex = True.getNumOperands() - HasChainOp - 2;
   SDValue TrueVL = True.getOperand(TrueVLIndex);

diff  --git a/llvm/test/CodeGen/RISCV/rvv/rvv-peephole-vmerge-vops.ll b/llvm/test/CodeGen/RISCV/rvv/rvv-peephole-vmerge-vops.ll
index 97678dd731f5..9530bedd8d99 100644
--- a/llvm/test/CodeGen/RISCV/rvv/rvv-peephole-vmerge-vops.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/rvv-peephole-vmerge-vops.ll
@@ -843,3 +843,40 @@ define <vscale x 2 x i32> @vpselect_trunc(<vscale x 2 x i32> %passthru, <vscale
   %b = call <vscale x 2 x i32> @llvm.vp.select.nxv2i32(<vscale x 2 x i1> %m, <vscale x 2 x i32> %a, <vscale x 2 x i32> %passthru, i32 %vl)
   ret <vscale x 2 x i32> %b
 }
+
+; Folding this would create a loop in the DAG becuase the chain from the VLE is
+; used by the vssubu.
+define void @test_dag_loop() {
+; CHECK-LABEL: test_dag_loop:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vsetivli zero, 1, e16, m8, ta, mu
+; CHECK-NEXT:    vle16.v v8, (zero)
+; CHECK-NEXT:    vsetvli a0, zero, e8, m4, ta, mu
+; CHECK-NEXT:    vmclr.m v0
+; CHECK-NEXT:    vmv.v.i v16, 0
+; CHECK-NEXT:    vsetivli zero, 0, e8, m4, tu, mu
+; CHECK-NEXT:    vmv4r.v v20, v16
+; CHECK-NEXT:    vssubu.vx v20, v16, zero, v0.t
+; CHECK-NEXT:    vsetvli zero, zero, e8, m4, ta, mu
+; CHECK-NEXT:    vmseq.vv v0, v20, v16
+; CHECK-NEXT:    vsetvli a0, zero, e16, m8, ta, mu
+; CHECK-NEXT:    vmv.v.i v16, 0
+; CHECK-NEXT:    vsetivli zero, 1, e16, m8, tu, mu
+; CHECK-NEXT:    vmerge.vvm v16, v16, v8, v0
+; CHECK-NEXT:    vsetivli zero, 0, e16, m8, ta, mu
+; CHECK-NEXT:    vse16.v v16, (zero)
+; CHECK-NEXT:    ret
+entry:
+  %0 = call <vscale x 32 x i16> @llvm.riscv.vle.nxv32i16.i64(<vscale x 32 x i16> undef, <vscale x 32 x i16>* null, i64 1)
+  %1 = tail call <vscale x 32 x i8> @llvm.riscv.vssubu.mask.nxv32i8.i8.i64(<vscale x 32 x i8> zeroinitializer, <vscale x 32 x i8> zeroinitializer, i8 0, <vscale x 32 x i1> zeroinitializer, i64 0, i64 0)
+  %2 = tail call <vscale x 32 x i1> @llvm.riscv.vmseq.nxv32i8.nxv32i8.i64(<vscale x 32 x i8> %1, <vscale x 32 x i8> zeroinitializer, i64 0)
+  %3 = tail call <vscale x 32 x i16> @llvm.riscv.vmerge.nxv32i16.nxv32i16.i64(<vscale x 32 x i16> zeroinitializer, <vscale x 32 x i16> zeroinitializer, <vscale x 32 x i16> %0, <vscale x 32 x i1> %2, i64 1)
+  call void @llvm.riscv.vse.nxv32i16.i64(<vscale x 32 x i16> %3, <vscale x 32 x i16>* null, i64 0)
+  ret void
+}
+
+declare <vscale x 32 x i16> @llvm.riscv.vle.nxv32i16.i64(<vscale x 32 x i16>, <vscale x 32 x i16>* nocapture, i64)
+declare <vscale x 32 x i8> @llvm.riscv.vssubu.mask.nxv32i8.i8.i64(<vscale x 32 x i8>, <vscale x 32 x i8>, i8, <vscale x 32 x i1>, i64, i64 immarg)
+declare <vscale x 32 x i1> @llvm.riscv.vmseq.nxv32i8.nxv32i8.i64(<vscale x 32 x i8>, <vscale x 32 x i8>, i64)
+declare <vscale x 32 x i16> @llvm.riscv.vmerge.nxv32i16.nxv32i16.i64(<vscale x 32 x i16>, <vscale x 32 x i16>, <vscale x 32 x i16>, <vscale x 32 x i1>, i64)
+declare void @llvm.riscv.vse.nxv32i16.i64(<vscale x 32 x i16>, <vscale x 32 x i16>* nocapture, i64)