diff --git a/llvm/lib/Analysis/ConstantFolding.cpp b/llvm/lib/Analysis/ConstantFolding.cpp
--- a/llvm/lib/Analysis/ConstantFolding.cpp
+++ b/llvm/lib/Analysis/ConstantFolding.cpp
@@ -1459,6 +1459,10 @@
   case Intrinsic::convert_from_fp16:
   case Intrinsic::convert_to_fp16:
   case Intrinsic::bitreverse:
+  case Intrinsic::amdgcn_cubeid:
+  case Intrinsic::amdgcn_cubema:
+  case Intrinsic::amdgcn_cubesc:
+  case Intrinsic::amdgcn_cubetc:
   case Intrinsic::amdgcn_fmul_legacy:
   case Intrinsic::x86_sse_cvtss2si:
   case Intrinsic::x86_sse_cvtss2si64:
@@ -2303,6 +2307,53 @@
   return nullptr;
 }
 
+static float ConstantFoldAMDGCNCubeIntrinsic(Intrinsic::ID IntrinsicID, float S0, float S1, float S2) {
+  float ID, MA, SC, TC;
+  if (fabs(S2) >= fabs(S0) && fabs(S2) >= fabs(S1)) {
+    if (S2 < 0) {
+      ID = 5.0;
+      SC = -S0;
+    } else {
+      ID = 4.0;
+      SC = S0;
+    }
+    MA = 2.0 * S2;
+    TC = -S1;
+  } else if (fabs(S1) >= fabs(S0)) {
+    if (S1 < 0) {
+      ID = 3.0;
+      TC = -S2;
+    } else {
+      ID = 2.0;
+      TC = S2;
+    }
+    MA = 2.0 * S1;
+    SC = S0;
+  } else {
+    if (S0 < 0) {
+      ID = 1.0;
+      SC = S2;
+    } else {
+      ID = 0.0;
+      SC = -S2;
+    }
+    MA = 2.0 * S0;
+    TC = -S1;
+  }
+  switch (IntrinsicID) {
+  default:
+    llvm_unreachable("unhandled amdgcn cube intrinsic");
+  case Intrinsic::amdgcn_cubeid:
+    return ID;
+  case Intrinsic::amdgcn_cubema:
+    return MA;
+  case Intrinsic::amdgcn_cubesc:
+    return SC;
+  case Intrinsic::amdgcn_cubetc:
+    return TC;
+  }
+}
+
 static Constant *ConstantFoldScalarCall3(StringRef Name,
                                          Intrinsic::ID IntrinsicID,
                                          Type *Ty,
@@ -2314,15 +2365,27 @@
   if (const auto *Op1 = dyn_cast<ConstantFP>(Operands[0])) {
     if (const auto *Op2 = dyn_cast<ConstantFP>(Operands[1])) {
       if (const auto *Op3 = dyn_cast<ConstantFP>(Operands[2])) {
+        APFloat V1 = Op1->getValueAPF();
+        APFloat V2 = Op2->getValueAPF();
+        APFloat V3 = Op3->getValueAPF();
         switch (IntrinsicID) {
         default: break;
         case Intrinsic::fma:
         case Intrinsic::fmuladd: {
-          APFloat V = Op1->getValueAPF();
-          V.fusedMultiplyAdd(Op2->getValueAPF(), Op3->getValueAPF(),
-                             APFloat::rmNearestTiesToEven);
+          APFloat V(V1);
+          V.fusedMultiplyAdd(V2, V3, APFloat::rmNearestTiesToEven);
           return ConstantFP::get(Ty->getContext(), V);
         }
+        case Intrinsic::amdgcn_cubeid:
+        case Intrinsic::amdgcn_cubema:
+        case Intrinsic::amdgcn_cubesc:
+        case Intrinsic::amdgcn_cubetc: {
+          float V = ConstantFoldAMDGCNCubeIntrinsic(IntrinsicID,
+                                                    V1.convertToFloat(),
+                                                    V2.convertToFloat(),
+                                                    V3.convertToFloat());
+          return ConstantFP::get(Ty, V);
+        }
         }
       }
     }
diff --git a/llvm/test/Analysis/ConstantFolding/AMDGPU/cubeid.ll b/llvm/test/Analysis/ConstantFolding/AMDGPU/cubeid.ll
new file mode 100644
--- /dev/null
+++ b/llvm/test/Analysis/ConstantFolding/AMDGPU/cubeid.ll
@@ -0,0 +1,155 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -instsimplify -S | FileCheck %s
+
+declare float @llvm.amdgcn.cubeid(float, float, float)
+
+define void @test(float* %p) {
+; CHECK-LABEL: @test(
+; CHECK-NEXT:    store volatile float 4.000000e+00, float* [[P:%.*]]
+; CHECK-NEXT:    store volatile float 2.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 2.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 0.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 0.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 5.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 2.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 5.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 2.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 0.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 0.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 0.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 0.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 5.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 5.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 0.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 0.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 2.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 2.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 1.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 1.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 5.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 2.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 5.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 2.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 1.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 1.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 1.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 1.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 5.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 5.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 1.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 1.000000e+00, float* [[P]]
+; CHECK-NEXT:    ret void
+;
+  %p3p4p5 = call float @llvm.amdgcn.cubeid(float +3.0, float +4.0, float +5.0)
+  store volatile float %p3p4p5, float* %p
+  %p3p5p4 = call float @llvm.amdgcn.cubeid(float +3.0, float +5.0, float +4.0)
+  store volatile float %p3p5p4, float* %p
+  %p4p3p5 = call float @llvm.amdgcn.cubeid(float +4.0, float +3.0, float +5.0)
+  store volatile float %p4p3p5, float* %p
+  %p4p5p3 = call float @llvm.amdgcn.cubeid(float +4.0, float +5.0, float +3.0)
+  store volatile float %p4p5p3, float* %p
+  %p5p3p4 = call float @llvm.amdgcn.cubeid(float +5.0, float +3.0, float +4.0)
+  store volatile float %p5p3p4, float* %p
+  %p5p4p3 = call float @llvm.amdgcn.cubeid(float +5.0, float +4.0, float +3.0)
+  store volatile float %p5p4p3, float* %p
+  %p3p4n5 = call float @llvm.amdgcn.cubeid(float +3.0, float +4.0, float -5.0)
+  store volatile float %p3p4n5, float* %p
+  %p3p5n4 = call float @llvm.amdgcn.cubeid(float +3.0, float +5.0, float -4.0)
+  store volatile float %p3p5n4, float* %p
+  %p4p3n5 = call float @llvm.amdgcn.cubeid(float +4.0, float +3.0, float -5.0)
+  store volatile float %p4p3n5, float* %p
+  %p4p5n3 = call float @llvm.amdgcn.cubeid(float +4.0, float +5.0, float -3.0)
+  store volatile float %p4p5n3, float* %p
+  %p5p3n4 = call float @llvm.amdgcn.cubeid(float +5.0, float +3.0, float -4.0)
+  store volatile float %p5p3n4, float* %p
+  %p5p4n3 = call float @llvm.amdgcn.cubeid(float +5.0, float +4.0, float -3.0)
+  store volatile float %p5p4n3, float* %p
+  %p3n4p5 = call float @llvm.amdgcn.cubeid(float +3.0, float -4.0, float +5.0)
+  store volatile float %p3n4p5, float* %p
+  %p3n5p4 = call float @llvm.amdgcn.cubeid(float +3.0, float -5.0, float +4.0)
+  store volatile float %p3n5p4, float* %p
+  %p4n3p5 = call float @llvm.amdgcn.cubeid(float +4.0, float -3.0, float +5.0)
+  store volatile float %p4n3p5, float* %p
+  %p4n5p3 = call float @llvm.amdgcn.cubeid(float +4.0, float -5.0, float +3.0)
+  store volatile float %p4n5p3, float* %p
+  %p5n3p4 = call float @llvm.amdgcn.cubeid(float +5.0, float -3.0, float +4.0)
+  store volatile float %p5n3p4, float* %p
+  %p5n4p3 = call float @llvm.amdgcn.cubeid(float +5.0, float -4.0, float +3.0)
+  store volatile float %p5n4p3, float* %p
+  %p3n4n5 = call float @llvm.amdgcn.cubeid(float +3.0, float -4.0, float -5.0)
+  store volatile float %p3n4n5, float* %p
+  %p3n5n4 = call float @llvm.amdgcn.cubeid(float +3.0, float -5.0, float -4.0)
+  store volatile float %p3n5n4, float* %p
+  %p4n3n5 = call float @llvm.amdgcn.cubeid(float +4.0, float -3.0, float -5.0)
+  store volatile float %p4n3n5, float* %p
+  %p4n5n3 = call float @llvm.amdgcn.cubeid(float +4.0, float -5.0, float -3.0)
+  store volatile float %p4n5n3, float* %p
+  %p5n3n4 = call float @llvm.amdgcn.cubeid(float +5.0, float -3.0, float -4.0)
+  store volatile float %p5n3n4, float* %p
+  %p5n4n3 = call float @llvm.amdgcn.cubeid(float +5.0, float -4.0, float -3.0)
+  store volatile float %p5n4n3, float* %p
+  %n3p4p5 = call float @llvm.amdgcn.cubeid(float -3.0, float +4.0, float +5.0)
+  store volatile float %n3p4p5, float* %p
+  %n3p5p4 = call float @llvm.amdgcn.cubeid(float -3.0, float +5.0, float +4.0)
+  store volatile float %n3p5p4, float* %p
+  %n4p3p5 = call float @llvm.amdgcn.cubeid(float -4.0, float +3.0, float +5.0)
+  store volatile float %n4p3p5, float* %p
+  %n4p5p3 = call float @llvm.amdgcn.cubeid(float -4.0, float +5.0, float +3.0)
+  store volatile float %n4p5p3, float* %p
+  %n5p3p4 = call float @llvm.amdgcn.cubeid(float -5.0, float +3.0, float +4.0)
+  store volatile float %n5p3p4, float* %p
+  %n5p4p3 = call float @llvm.amdgcn.cubeid(float -5.0, float +4.0, float +3.0)
+  store volatile float %n5p4p3, float* %p
+  %n3p4n5 = call float @llvm.amdgcn.cubeid(float -3.0, float +4.0, float -5.0)
+  store volatile float %n3p4n5, float* %p
+  %n3p5n4 = call float @llvm.amdgcn.cubeid(float -3.0, float +5.0, float -4.0)
+  store volatile float %n3p5n4, float* %p
+  %n4p3n5 = call float @llvm.amdgcn.cubeid(float -4.0, float +3.0, float -5.0)
+  store volatile float %n4p3n5, float* %p
+  %n4p5n3 = call float @llvm.amdgcn.cubeid(float -4.0, float +5.0, float -3.0)
+  store volatile float %n4p5n3, float* %p
+  %n5p3n4 = call float @llvm.amdgcn.cubeid(float -5.0, float +3.0, float -4.0)
+  store volatile float %n5p3n4, float* %p
+  %n5p4n3 = call float @llvm.amdgcn.cubeid(float -5.0, float +4.0, float -3.0)
+  store volatile float %n5p4n3, float* %p
+  %n3n4p5 = call float @llvm.amdgcn.cubeid(float -3.0, float -4.0, float +5.0)
+  store volatile float %n3n4p5, float* %p
+  %n3n5p4 = call float @llvm.amdgcn.cubeid(float -3.0, float -5.0, float +4.0)
+  store volatile float %n3n5p4, float* %p
+  %n4n3p5 = call float @llvm.amdgcn.cubeid(float -4.0, float -3.0, float +5.0)
+  store volatile float %n4n3p5, float* %p
+  %n4n5p3 = call float @llvm.amdgcn.cubeid(float -4.0, float -5.0, float +3.0)
+  store volatile float %n4n5p3, float* %p
+  %n5n3p4 = call float @llvm.amdgcn.cubeid(float -5.0, float -3.0, float +4.0)
+  store volatile float %n5n3p4, float* %p
+  %n5n4p3 = call float @llvm.amdgcn.cubeid(float -5.0, float -4.0, float +3.0)
+  store volatile float %n5n4p3, float* %p
+  %n3n4n5 = call float @llvm.amdgcn.cubeid(float -3.0, float -4.0, float -5.0)
+  store volatile float %n3n4n5, float* %p
+  %n3n5n4 = call float @llvm.amdgcn.cubeid(float -3.0, float -5.0, float -4.0)
+  store volatile float %n3n5n4, float* %p
+  %n4n3n5 = call float @llvm.amdgcn.cubeid(float -4.0, float -3.0, float -5.0)
+  store volatile float %n4n3n5, float* %p
+  %n4n5n3 = call float @llvm.amdgcn.cubeid(float -4.0, float -5.0, float -3.0)
+  store volatile float %n4n5n3, float* %p
+  %n5n3n4 = call float @llvm.amdgcn.cubeid(float -5.0, float -3.0, float -4.0)
+  store volatile float %n5n3n4, float* %p
+  %n5n4n3 = call float @llvm.amdgcn.cubeid(float -5.0, float -4.0, float -3.0)
+  store volatile float %n5n4n3, float* %p
+  ret void
+}
diff --git a/llvm/test/Analysis/ConstantFolding/AMDGPU/cubema.ll b/llvm/test/Analysis/ConstantFolding/AMDGPU/cubema.ll
new file mode 100644
--- /dev/null
+++ b/llvm/test/Analysis/ConstantFolding/AMDGPU/cubema.ll
@@ -0,0 +1,155 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -instsimplify -S | FileCheck %s
+
+declare float @llvm.amdgcn.cubema(float, float, float)
+
+define void @test(float* %p) {
+; CHECK-LABEL: @test(
+; CHECK-NEXT:    store volatile float 1.000000e+01, float* [[P:%.*]]
+; CHECK-NEXT:    store volatile float 1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float 1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float 1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float 1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float 1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float -1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float 1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float -1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float 1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float 1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float 1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float 1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float -1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float 1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float -1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float 1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float 1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float -1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float -1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float -1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float -1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float 1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float 1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float 1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float 1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float 1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float 1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float -1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float -1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float -1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float 1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float -1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float 1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float -1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float -1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float 1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float -1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float 1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float -1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float -1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float -1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float -1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float -1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float -1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float -1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float -1.000000e+01, float* [[P]]
+; CHECK-NEXT:    store volatile float -1.000000e+01, float* [[P]]
+; CHECK-NEXT:    ret void
+;
+  %p3p4p5 = call float @llvm.amdgcn.cubema(float +3.0, float +4.0, float +5.0)
+  store volatile float %p3p4p5, float* %p
+  %p3p5p4 = call float @llvm.amdgcn.cubema(float +3.0, float +5.0, float +4.0)
+  store volatile float %p3p5p4, float* %p
+  %p4p3p5 = call float @llvm.amdgcn.cubema(float +4.0, float +3.0, float +5.0)
+  store volatile float %p4p3p5, float* %p
+  %p4p5p3 = call float @llvm.amdgcn.cubema(float +4.0, float +5.0, float +3.0)
+  store volatile float %p4p5p3, float* %p
+  %p5p3p4 = call float @llvm.amdgcn.cubema(float +5.0, float +3.0, float +4.0)
+  store volatile float %p5p3p4, float* %p
+  %p5p4p3 = call float @llvm.amdgcn.cubema(float +5.0, float +4.0, float +3.0)
+  store volatile float %p5p4p3, float* %p
+  %p3p4n5 = call float @llvm.amdgcn.cubema(float +3.0, float +4.0, float -5.0)
+  store volatile float %p3p4n5, float* %p
+  %p3p5n4 = call float @llvm.amdgcn.cubema(float +3.0, float +5.0, float -4.0)
+  store volatile float %p3p5n4, float* %p
+  %p4p3n5 = call float @llvm.amdgcn.cubema(float +4.0, float +3.0, float -5.0)
+  store volatile float %p4p3n5, float* %p
+  %p4p5n3 = call float @llvm.amdgcn.cubema(float +4.0, float +5.0, float -3.0)
+  store volatile float %p4p5n3, float* %p
+  %p5p3n4 = call float @llvm.amdgcn.cubema(float +5.0, float +3.0, float -4.0)
+  store volatile float %p5p3n4, float* %p
+  %p5p4n3 = call float @llvm.amdgcn.cubema(float +5.0, float +4.0, float -3.0)
+  store volatile float %p5p4n3, float* %p
+  %p3n4p5 = call float @llvm.amdgcn.cubema(float +3.0, float -4.0, float +5.0)
+  store volatile float %p3n4p5, float* %p
+  %p3n5p4 = call float @llvm.amdgcn.cubema(float +3.0, float -5.0, float +4.0)
+  store volatile float %p3n5p4, float* %p
+  %p4n3p5 = call float @llvm.amdgcn.cubema(float +4.0, float -3.0, float +5.0)
+  store volatile float %p4n3p5, float* %p
+  %p4n5p3 = call float @llvm.amdgcn.cubema(float +4.0, float -5.0, float +3.0)
+  store volatile float %p4n5p3, float* %p
+  %p5n3p4 = call float @llvm.amdgcn.cubema(float +5.0, float -3.0, float +4.0)
+  store volatile float %p5n3p4, float* %p
+  %p5n4p3 = call float @llvm.amdgcn.cubema(float +5.0, float -4.0, float +3.0)
+  store volatile float %p5n4p3, float* %p
+  %p3n4n5 = call float @llvm.amdgcn.cubema(float +3.0, float -4.0, float -5.0)
+  store volatile float %p3n4n5, float* %p
+  %p3n5n4 = call float @llvm.amdgcn.cubema(float +3.0, float -5.0, float -4.0)
+  store volatile float %p3n5n4, float* %p
+  %p4n3n5 = call float @llvm.amdgcn.cubema(float +4.0, float -3.0, float -5.0)
+  store volatile float %p4n3n5, float* %p
+  %p4n5n3 = call float @llvm.amdgcn.cubema(float +4.0, float -5.0, float -3.0)
+  store volatile float %p4n5n3, float* %p
+  %p5n3n4 = call float @llvm.amdgcn.cubema(float +5.0, float -3.0, float -4.0)
+  store volatile float %p5n3n4, float* %p
+  %p5n4n3 = call float @llvm.amdgcn.cubema(float +5.0, float -4.0, float -3.0)
+  store volatile float %p5n4n3, float* %p
+  %n3p4p5 = call float @llvm.amdgcn.cubema(float -3.0, float +4.0, float +5.0)
+  store volatile float %n3p4p5, float* %p
+  %n3p5p4 = call float @llvm.amdgcn.cubema(float -3.0, float +5.0, float +4.0)
+  store volatile float %n3p5p4, float* %p
+  %n4p3p5 = call float @llvm.amdgcn.cubema(float -4.0, float +3.0, float +5.0)
+  store volatile float %n4p3p5, float* %p
+  %n4p5p3 = call float @llvm.amdgcn.cubema(float -4.0, float +5.0, float +3.0)
+  store volatile float %n4p5p3, float* %p
+  %n5p3p4 = call float @llvm.amdgcn.cubema(float -5.0, float +3.0, float +4.0)
+  store volatile float %n5p3p4, float* %p
+  %n5p4p3 = call float @llvm.amdgcn.cubema(float -5.0, float +4.0, float +3.0)
+  store volatile float %n5p4p3, float* %p
+  %n3p4n5 = call float @llvm.amdgcn.cubema(float -3.0, float +4.0, float -5.0)
+  store volatile float %n3p4n5, float* %p
+  %n3p5n4 = call float @llvm.amdgcn.cubema(float -3.0, float +5.0, float -4.0)
+  store volatile float %n3p5n4, float* %p
+  %n4p3n5 = call float @llvm.amdgcn.cubema(float -4.0, float +3.0, float -5.0)
+  store volatile float %n4p3n5, float* %p
+  %n4p5n3 = call float @llvm.amdgcn.cubema(float -4.0, float +5.0, float -3.0)
+  store volatile float %n4p5n3, float* %p
+  %n5p3n4 = call float @llvm.amdgcn.cubema(float -5.0, float +3.0, float -4.0)
+  store volatile float %n5p3n4, float* %p
+  %n5p4n3 = call float @llvm.amdgcn.cubema(float -5.0, float +4.0, float -3.0)
+  store volatile float %n5p4n3, float* %p
+  %n3n4p5 = call float @llvm.amdgcn.cubema(float -3.0, float -4.0, float +5.0)
+  store volatile float %n3n4p5, float* %p
+  %n3n5p4 = call float @llvm.amdgcn.cubema(float -3.0, float -5.0, float +4.0)
+  store volatile float %n3n5p4, float* %p
+  %n4n3p5 = call float @llvm.amdgcn.cubema(float -4.0, float -3.0, float +5.0)
+  store volatile float %n4n3p5, float* %p
+  %n4n5p3 = call float @llvm.amdgcn.cubema(float -4.0, float -5.0, float +3.0)
+  store volatile float %n4n5p3, float* %p
+  %n5n3p4 = call float @llvm.amdgcn.cubema(float -5.0, float -3.0, float +4.0)
+  store volatile float %n5n3p4, float* %p
+  %n5n4p3 = call float @llvm.amdgcn.cubema(float -5.0, float -4.0, float +3.0)
+  store volatile float %n5n4p3, float* %p
+  %n3n4n5 = call float @llvm.amdgcn.cubema(float -3.0, float -4.0, float -5.0)
+  store volatile float %n3n4n5, float* %p
+  %n3n5n4 = call float @llvm.amdgcn.cubema(float -3.0, float -5.0, float -4.0)
+  store volatile float %n3n5n4, float* %p
+  %n4n3n5 = call float @llvm.amdgcn.cubema(float -4.0, float -3.0, float -5.0)
+  store volatile float %n4n3n5, float* %p
+  %n4n5n3 = call float @llvm.amdgcn.cubema(float -4.0, float -5.0, float -3.0)
+  store volatile float %n4n5n3, float* %p
+  %n5n3n4 = call float @llvm.amdgcn.cubema(float -5.0, float -3.0, float -4.0)
+  store volatile float %n5n3n4, float* %p
+  %n5n4n3 = call float @llvm.amdgcn.cubema(float -5.0, float -4.0, float -3.0)
+  store volatile float %n5n4n3, float* %p
+  ret void
+}
diff --git a/llvm/test/Analysis/ConstantFolding/AMDGPU/cubesc.ll b/llvm/test/Analysis/ConstantFolding/AMDGPU/cubesc.ll
new file mode 100644
--- /dev/null
+++ b/llvm/test/Analysis/ConstantFolding/AMDGPU/cubesc.ll
@@ -0,0 +1,155 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -instsimplify -S | FileCheck %s
+
+declare float @llvm.amdgcn.cubesc(float, float, float)
+
+define void @test(float* %p) {
+; CHECK-LABEL: @test(
+; CHECK-NEXT:    store volatile float 3.000000e+00, float* [[P:%.*]]
+; CHECK-NEXT:    store volatile float 3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -3.000000e+00, float* [[P]]
+; CHECK-NEXT:    ret void
+;
+  %p3p4p5 = call float @llvm.amdgcn.cubesc(float +3.0, float +4.0, float +5.0)
+  store volatile float %p3p4p5, float* %p
+  %p3p5p4 = call float @llvm.amdgcn.cubesc(float +3.0, float +5.0, float +4.0)
+  store volatile float %p3p5p4, float* %p
+  %p4p3p5 = call float @llvm.amdgcn.cubesc(float +4.0, float +3.0, float +5.0)
+  store volatile float %p4p3p5, float* %p
+  %p4p5p3 = call float @llvm.amdgcn.cubesc(float +4.0, float +5.0, float +3.0)
+  store volatile float %p4p5p3, float* %p
+  %p5p3p4 = call float @llvm.amdgcn.cubesc(float +5.0, float +3.0, float +4.0)
+  store volatile float %p5p3p4, float* %p
+  %p5p4p3 = call float @llvm.amdgcn.cubesc(float +5.0, float +4.0, float +3.0)
+  store volatile float %p5p4p3, float* %p
+  %p3p4n5 = call float @llvm.amdgcn.cubesc(float +3.0, float +4.0, float -5.0)
+  store volatile float %p3p4n5, float* %p
+  %p3p5n4 = call float @llvm.amdgcn.cubesc(float +3.0, float +5.0, float -4.0)
+  store volatile float %p3p5n4, float* %p
+  %p4p3n5 = call float @llvm.amdgcn.cubesc(float +4.0, float +3.0, float -5.0)
+  store volatile float %p4p3n5, float* %p
+  %p4p5n3 = call float @llvm.amdgcn.cubesc(float +4.0, float +5.0, float -3.0)
+  store volatile float %p4p5n3, float* %p
+  %p5p3n4 = call float @llvm.amdgcn.cubesc(float +5.0, float +3.0, float -4.0)
+  store volatile float %p5p3n4, float* %p
+  %p5p4n3 = call float @llvm.amdgcn.cubesc(float +5.0, float +4.0, float -3.0)
+  store volatile float %p5p4n3, float* %p
+  %p3n4p5 = call float @llvm.amdgcn.cubesc(float +3.0, float -4.0, float +5.0)
+  store volatile float %p3n4p5, float* %p
+  %p3n5p4 = call float @llvm.amdgcn.cubesc(float +3.0, float -5.0, float +4.0)
+  store volatile float %p3n5p4, float* %p
+  %p4n3p5 = call float @llvm.amdgcn.cubesc(float +4.0, float -3.0, float +5.0)
+  store volatile float %p4n3p5, float* %p
+  %p4n5p3 = call float @llvm.amdgcn.cubesc(float +4.0, float -5.0, float +3.0)
+  store volatile float %p4n5p3, float* %p
+  %p5n3p4 = call float @llvm.amdgcn.cubesc(float +5.0, float -3.0, float +4.0)
+  store volatile float %p5n3p4, float* %p
+  %p5n4p3 = call float @llvm.amdgcn.cubesc(float +5.0, float -4.0, float +3.0)
+  store volatile float %p5n4p3, float* %p
+  %p3n4n5 = call float @llvm.amdgcn.cubesc(float +3.0, float -4.0, float -5.0)
+  store volatile float %p3n4n5, float* %p
+  %p3n5n4 = call float @llvm.amdgcn.cubesc(float +3.0, float -5.0, float -4.0)
+  store volatile float %p3n5n4, float* %p
+  %p4n3n5 = call float @llvm.amdgcn.cubesc(float +4.0, float -3.0, float -5.0)
+  store volatile float %p4n3n5, float* %p
+  %p4n5n3 = call float @llvm.amdgcn.cubesc(float +4.0, float -5.0, float -3.0)
+  store volatile float %p4n5n3, float* %p
+  %p5n3n4 = call float @llvm.amdgcn.cubesc(float +5.0, float -3.0, float -4.0)
+  store volatile float %p5n3n4, float* %p
+  %p5n4n3 = call float @llvm.amdgcn.cubesc(float +5.0, float -4.0, float -3.0)
+  store volatile float %p5n4n3, float* %p
+  %n3p4p5 = call float @llvm.amdgcn.cubesc(float -3.0, float +4.0, float +5.0)
+  store volatile float %n3p4p5, float* %p
+  %n3p5p4 = call float @llvm.amdgcn.cubesc(float -3.0, float +5.0, float +4.0)
+  store volatile float %n3p5p4, float* %p
+  %n4p3p5 = call float @llvm.amdgcn.cubesc(float -4.0, float +3.0, float +5.0)
+  store volatile float %n4p3p5, float* %p
+  %n4p5p3 = call float @llvm.amdgcn.cubesc(float -4.0, float +5.0, float +3.0)
+  store volatile float %n4p5p3, float* %p
+  %n5p3p4 = call float @llvm.amdgcn.cubesc(float -5.0, float +3.0, float +4.0)
+  store volatile float %n5p3p4, float* %p
+  %n5p4p3 = call float @llvm.amdgcn.cubesc(float -5.0, float +4.0, float +3.0)
+  store volatile float %n5p4p3, float* %p
+  %n3p4n5 = call float @llvm.amdgcn.cubesc(float -3.0, float +4.0, float -5.0)
+  store volatile float %n3p4n5, float* %p
+  %n3p5n4 = call float @llvm.amdgcn.cubesc(float -3.0, float +5.0, float -4.0)
+  store volatile float %n3p5n4, float* %p
+  %n4p3n5 = call float @llvm.amdgcn.cubesc(float -4.0, float +3.0, float -5.0)
+  store volatile float %n4p3n5, float* %p
+  %n4p5n3 = call float @llvm.amdgcn.cubesc(float -4.0, float +5.0, float -3.0)
+  store volatile float %n4p5n3, float* %p
+  %n5p3n4 = call float @llvm.amdgcn.cubesc(float -5.0, float +3.0, float -4.0)
+  store volatile float %n5p3n4, float* %p
+  %n5p4n3 = call float @llvm.amdgcn.cubesc(float -5.0, float +4.0, float -3.0)
+  store volatile float %n5p4n3, float* %p
+  %n3n4p5 = call float @llvm.amdgcn.cubesc(float -3.0, float -4.0, float +5.0)
+  store volatile float %n3n4p5, float* %p
+  %n3n5p4 = call float @llvm.amdgcn.cubesc(float -3.0, float -5.0, float +4.0)
+  store volatile float %n3n5p4, float* %p
+  %n4n3p5 = call float @llvm.amdgcn.cubesc(float -4.0, float -3.0, float +5.0)
+  store volatile float %n4n3p5, float* %p
+  %n4n5p3 = call float @llvm.amdgcn.cubesc(float -4.0, float -5.0, float +3.0)
+  store volatile float %n4n5p3, float* %p
+  %n5n3p4 = call float @llvm.amdgcn.cubesc(float -5.0, float -3.0, float +4.0)
+  store volatile float %n5n3p4, float* %p
+  %n5n4p3 = call float @llvm.amdgcn.cubesc(float -5.0, float -4.0, float +3.0)
+  store volatile float %n5n4p3, float* %p
+  %n3n4n5 = call float @llvm.amdgcn.cubesc(float -3.0, float -4.0, float -5.0)
+  store volatile float %n3n4n5, float* %p
+  %n3n5n4 = call float @llvm.amdgcn.cubesc(float -3.0, float -5.0, float -4.0)
+  store volatile float %n3n5n4, float* %p
+  %n4n3n5 = call float @llvm.amdgcn.cubesc(float -4.0, float -3.0, float -5.0)
+  store volatile float %n4n3n5, float* %p
+  %n4n5n3 = call float @llvm.amdgcn.cubesc(float -4.0, float -5.0, float -3.0)
+  store volatile float %n4n5n3, float* %p
+  %n5n3n4 = call float @llvm.amdgcn.cubesc(float -5.0, float -3.0, float -4.0)
+  store volatile float %n5n3n4, float* %p
+  %n5n4n3 = call float @llvm.amdgcn.cubesc(float -5.0, float -4.0, float -3.0)
+  store volatile float %n5n4n3, float* %p
+  ret void
+}
diff --git a/llvm/test/Analysis/ConstantFolding/AMDGPU/cubetc.ll b/llvm/test/Analysis/ConstantFolding/AMDGPU/cubetc.ll
new file mode 100644
--- /dev/null
+++ b/llvm/test/Analysis/ConstantFolding/AMDGPU/cubetc.ll
@@ -0,0 +1,155 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -instsimplify -S | FileCheck %s
+
+declare float @llvm.amdgcn.cubetc(float, float, float)
+
+define void @test(float* %p) {
+; CHECK-LABEL: @test(
+; CHECK-NEXT:    store volatile float -4.000000e+00, float* [[P:%.*]]
+; CHECK-NEXT:    store volatile float 4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float -3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 4.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 3.000000e+00, float* [[P]]
+; CHECK-NEXT:    store volatile float 4.000000e+00, float* [[P]]
+; CHECK-NEXT:    ret void
+;
+  %p3p4p5 = call float @llvm.amdgcn.cubetc(float +3.0, float +4.0, float +5.0)
+  store volatile float %p3p4p5, float* %p
+  %p3p5p4 = call float @llvm.amdgcn.cubetc(float +3.0, float +5.0, float +4.0)
+  store volatile float %p3p5p4, float* %p
+  %p4p3p5 = call float @llvm.amdgcn.cubetc(float +4.0, float +3.0, float +5.0)
+  store volatile float %p4p3p5, float* %p
+  %p4p5p3 = call float @llvm.amdgcn.cubetc(float +4.0, float +5.0, float +3.0)
+  store volatile float %p4p5p3, float* %p
+  %p5p3p4 = call float @llvm.amdgcn.cubetc(float +5.0, float +3.0, float +4.0)
+  store volatile float %p5p3p4, float* %p
+  %p5p4p3 = call float @llvm.amdgcn.cubetc(float +5.0, float +4.0, float +3.0)
+  store volatile float %p5p4p3, float* %p
+  %p3p4n5 = call float @llvm.amdgcn.cubetc(float +3.0, float +4.0, float -5.0)
+  store volatile float %p3p4n5, float* %p
+  %p3p5n4 = call float @llvm.amdgcn.cubetc(float +3.0, float +5.0, float -4.0)
+  store volatile float %p3p5n4, float* %p
+  %p4p3n5 = call float @llvm.amdgcn.cubetc(float +4.0, float +3.0, float -5.0)
+  store volatile float %p4p3n5, float* %p
+  %p4p5n3 = call float @llvm.amdgcn.cubetc(float +4.0, float +5.0, float -3.0)
+  store volatile float %p4p5n3, float* %p
+  %p5p3n4 = call float @llvm.amdgcn.cubetc(float +5.0, float +3.0, float -4.0)
+  store volatile float %p5p3n4, float* %p
+  %p5p4n3 = call float @llvm.amdgcn.cubetc(float +5.0, float +4.0, float -3.0)
+  store volatile float %p5p4n3, float* %p
+  %p3n4p5 = call float @llvm.amdgcn.cubetc(float +3.0, float -4.0, float +5.0)
+  store volatile float %p3n4p5, float* %p
+  %p3n5p4 = call float @llvm.amdgcn.cubetc(float +3.0, float -5.0, float +4.0)
+  store volatile float %p3n5p4, float* %p
+  %p4n3p5 = call float @llvm.amdgcn.cubetc(float +4.0, float -3.0, float +5.0)
+  store volatile float %p4n3p5, float* %p
+  %p4n5p3 = call float @llvm.amdgcn.cubetc(float +4.0, float -5.0, float +3.0)
+  store volatile float %p4n5p3, float* %p
+  %p5n3p4 = call float @llvm.amdgcn.cubetc(float +5.0, float -3.0, float +4.0)
+  store volatile float %p5n3p4, float* %p
+  %p5n4p3 = call float @llvm.amdgcn.cubetc(float +5.0, float -4.0, float +3.0)
+  store volatile float %p5n4p3, float* %p
+  %p3n4n5 = call float @llvm.amdgcn.cubetc(float +3.0, float -4.0, float -5.0)
+  store volatile float %p3n4n5, float* %p
+  %p3n5n4 = call float @llvm.amdgcn.cubetc(float +3.0, float -5.0, float -4.0)
+  store volatile float %p3n5n4, float* %p
+  %p4n3n5 = call float @llvm.amdgcn.cubetc(float +4.0, float -3.0, float -5.0)
+  store volatile float %p4n3n5, float* %p
+  %p4n5n3 = call float @llvm.amdgcn.cubetc(float +4.0, float -5.0, float -3.0)
+  store volatile float %p4n5n3, float* %p
+  %p5n3n4 = call float @llvm.amdgcn.cubetc(float +5.0, float -3.0, float -4.0)
+  store volatile float %p5n3n4, float* %p
+  %p5n4n3 = call float @llvm.amdgcn.cubetc(float +5.0, float -4.0, float -3.0)
+  store volatile float %p5n4n3, float* %p
+  %n3p4p5 = call float @llvm.amdgcn.cubetc(float -3.0, float +4.0, float +5.0)
+  store volatile float %n3p4p5, float* %p
+  %n3p5p4 = call float @llvm.amdgcn.cubetc(float -3.0, float +5.0, float +4.0)
+  store volatile float %n3p5p4, float* %p
+  %n4p3p5 = call float @llvm.amdgcn.cubetc(float -4.0, float +3.0, float +5.0)
+  store volatile float %n4p3p5, float* %p
+  %n4p5p3 = call float @llvm.amdgcn.cubetc(float -4.0, float +5.0, float +3.0)
+  store volatile float %n4p5p3, float* %p
+  %n5p3p4 = call float @llvm.amdgcn.cubetc(float -5.0, float +3.0, float +4.0)
+  store volatile float %n5p3p4, float* %p
+  %n5p4p3 = call float @llvm.amdgcn.cubetc(float -5.0, float +4.0, float +3.0)
+  store volatile float %n5p4p3, float* %p
+  %n3p4n5 = call float @llvm.amdgcn.cubetc(float -3.0, float +4.0, float -5.0)
+  store volatile float %n3p4n5, float* %p
+  %n3p5n4 = call float @llvm.amdgcn.cubetc(float -3.0, float +5.0, float -4.0)
+  store volatile float %n3p5n4, float* %p
+  %n4p3n5 = call float @llvm.amdgcn.cubetc(float -4.0, float +3.0, float -5.0)
+  store volatile float %n4p3n5, float* %p
+  %n4p5n3 = call float @llvm.amdgcn.cubetc(float -4.0, float +5.0, float -3.0)
+  store volatile float %n4p5n3, float* %p
+  %n5p3n4 = call float @llvm.amdgcn.cubetc(float -5.0, float +3.0, float -4.0)
+  store volatile float %n5p3n4, float* %p
+  %n5p4n3 = call float @llvm.amdgcn.cubetc(float -5.0, float +4.0, float -3.0)
+  store volatile float %n5p4n3, float* %p
+  %n3n4p5 = call float @llvm.amdgcn.cubetc(float -3.0, float -4.0, float +5.0)
+  store volatile float %n3n4p5, float* %p
+  %n3n5p4 = call float @llvm.amdgcn.cubetc(float -3.0, float -5.0, float +4.0)
+  store volatile float %n3n5p4, float* %p
+  %n4n3p5 = call float @llvm.amdgcn.cubetc(float -4.0, float -3.0, float +5.0)
+  store volatile float %n4n3p5, float* %p
+  %n4n5p3 = call float @llvm.amdgcn.cubetc(float -4.0, float -5.0, float +3.0)
+  store volatile float %n4n5p3, float* %p
+  %n5n3p4 = call float @llvm.amdgcn.cubetc(float -5.0, float -3.0, float +4.0)
+  store volatile float %n5n3p4, float* %p
+  %n5n4p3 = call float @llvm.amdgcn.cubetc(float -5.0, float -4.0, float +3.0)
+  store volatile float %n5n4p3, float* %p
+  %n3n4n5 = call float @llvm.amdgcn.cubetc(float -3.0, float -4.0, float -5.0)
+  store volatile float %n3n4n5, float* %p
+  %n3n5n4 = call float @llvm.amdgcn.cubetc(float -3.0, float -5.0, float -4.0)
+  store volatile float %n3n5n4, float* %p
+  %n4n3n5 = call float @llvm.amdgcn.cubetc(float -4.0, float -3.0, float -5.0)
+  store volatile float %n4n3n5, float* %p
+  %n4n5n3 = call float @llvm.amdgcn.cubetc(float -4.0, float -5.0, float -3.0)
+  store volatile float %n4n5n3, float* %p
+  %n5n3n4 = call float @llvm.amdgcn.cubetc(float -5.0, float -3.0, float -4.0)
+  store volatile float %n5n3n4, float* %p
+  %n5n4n3 = call float @llvm.amdgcn.cubetc(float -5.0, float -4.0, float -3.0)
+  store volatile float %n5n4n3, float* %p
+  ret void
+}