[AMDGPU] Avoid DAG combining assert with fneg(fadd(A,0))

fneg combining attempts to turn it into fadd(fneg(A), fneg(0)), but
creating the new fadd folds to just fneg(A). When A has multiple uses,
this confuses it and you get an assert. Fixed.

Differential Revision: https://reviews.llvm.org/D60633

Change-Id: I0ddc9b7286abe78edc0cd8d734fdeb05ff09821c

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@358640 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/AMDGPU/AMDGPUISelLowering.cpp b/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
index 623d9817f45f6e45ad94929742ae87b67c5c67c0..08affa8f0574ff1cee1daf05f85e32d1d0706c92 100644 (file)
--- a/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
+++ b/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
@@ -3692,6 +3692,8 @@ SDValue AMDGPUTargetLowering::performFNegCombine(SDNode *N,
       RHS = RHS.getOperand(0);
 
     SDValue Res = DAG.getNode(ISD::FADD, SL, VT, LHS, RHS, N0->getFlags());
+    if (Res.getOpcode() != ISD::FADD)
+      return SDValue(); // Op got folded away.
     if (!N0.hasOneUse())
       DAG.ReplaceAllUsesWith(N0, DAG.getNode(ISD::FNEG, SL, VT, Res));
     return Res;
@@ -3711,6 +3713,8 @@ SDValue AMDGPUTargetLowering::performFNegCombine(SDNode *N,
       RHS = DAG.getNode(ISD::FNEG, SL, VT, RHS);
 
     SDValue Res = DAG.getNode(Opc, SL, VT, LHS, RHS, N0->getFlags());
+    if (Res.getOpcode() != Opc)
+      return SDValue(); // Op got folded away.
     if (!N0.hasOneUse())
       DAG.ReplaceAllUsesWith(N0, DAG.getNode(ISD::FNEG, SL, VT, Res));
     return Res;
@@ -3738,6 +3742,8 @@ SDValue AMDGPUTargetLowering::performFNegCombine(SDNode *N,
       RHS = RHS.getOperand(0);
 
     SDValue Res = DAG.getNode(Opc, SL, VT, LHS, MHS, RHS);
+    if (Res.getOpcode() != Opc)
+      return SDValue(); // Op got folded away.
     if (!N0.hasOneUse())
       DAG.ReplaceAllUsesWith(N0, DAG.getNode(ISD::FNEG, SL, VT, Res));
     return Res;
@@ -3766,6 +3772,8 @@ SDValue AMDGPUTargetLowering::performFNegCombine(SDNode *N,
     unsigned Opposite = inverseMinMax(Opc);
 
     SDValue Res = DAG.getNode(Opposite, SL, VT, NegLHS, NegRHS, N0->getFlags());
+    if (Res.getOpcode() != Opposite)
+      return SDValue(); // Op got folded away.
     if (!N0.hasOneUse())
       DAG.ReplaceAllUsesWith(N0, DAG.getNode(ISD::FNEG, SL, VT, Res));
     return Res;
@@ -3776,6 +3784,8 @@ SDValue AMDGPUTargetLowering::performFNegCombine(SDNode *N,
       Ops[I] = DAG.getNode(ISD::FNEG, SL, VT, N0->getOperand(I), N0->getFlags());
 
     SDValue Res = DAG.getNode(AMDGPUISD::FMED3, SL, VT, Ops, N0->getFlags());
+    if (Res.getOpcode() != AMDGPUISD::FMED3)
+      return SDValue(); // Op got folded away.
     if (!N0.hasOneUse())
       DAG.ReplaceAllUsesWith(N0, DAG.getNode(ISD::FNEG, SL, VT, Res));
     return Res;

diff --git a/test/CodeGen/AMDGPU/fneg-combines.ll b/test/CodeGen/AMDGPU/fneg-combines.ll
index d705f3194371232aec827df809af9e407458096d..d2edfa566c8cbf742ce5bae2baf7bfd65319b456 100644 (file)
--- a/test/CodeGen/AMDGPU/fneg-combines.ll
+++ b/test/CodeGen/AMDGPU/fneg-combines.ll
@@ -214,6 +214,28 @@ define amdgpu_kernel void @v_fneg_add_multi_use_fneg_x_f32(float addrspace(1)* %
   ret void
 }
 
+; This one asserted with -enable-no-signed-zeros-fp-math
+; GCN-LABEL: {{^}}fneg_fadd_0:
+; GCN-SAFE-DAG: v_mad_f32 [[A:v[0-9]+]],
+; GCN-SAFE-DAG: v_xor_b32_e32 [[B:v[0-9]+]], 0x80000000
+; GCN-SAFE-DAG: v_cmp_ngt_f32_e32 {{.*}}, [[A]]
+; GCN-NSZ-DAG: v_mac_f32_e32 [[C:v[0-9]+]],
+; GCN-NSZ-DAG: v_cmp_nlt_f32_e64 {{.*}}, -[[C]]
+
+define amdgpu_ps float @fneg_fadd_0(float inreg %tmp2, float inreg %tmp6, <4 x i32> %arg) local_unnamed_addr #0 {
+.entry:
+  %tmp7 = fdiv float 1.000000e+00, %tmp6
+  %tmp8 = fmul float 0.000000e+00, %tmp7
+  %tmp9 = fmul reassoc nnan arcp contract float 0.000000e+00, %tmp8
+  %.i188 = fadd float %tmp9, 0.000000e+00
+  %tmp10 = fcmp uge float %.i188, %tmp2
+  %tmp11 = fsub float -0.000000e+00, %.i188
+  %.i092 = select i1 %tmp10, float %tmp2, float %tmp11
+  %tmp12 = fcmp ule float %.i092, 0.000000e+00
+  %.i198 = select i1 %tmp12, float 0.000000e+00, float 0x7FF8000000000000
+  ret float %.i198
+}
+
 ; --------------------------------------------------------------------------------
 ; fmul tests
 ; --------------------------------------------------------------------------------