[ARM64]Fix the bug right shift uint64_t by 64 generates incorrect result.

llvm-svn: 208761

2 files changed, 40 insertions, 15 deletions

diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 824de8ccbb9..f77aeecd292 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -5580,38 +5580,41 @@ Value *CodeGenFunction::EmitARM64BuiltinExpr(unsigned BuiltinID,
   case NEON::BI__builtin_neon_vshld_n_u64: {
     llvm::ConstantInt *Amt = cast<ConstantInt>(EmitScalarExpr(E->getArg(1)));
     return Builder.CreateShl(
-        Ops[0], ConstantInt::get(Int64Ty, std::min(static_cast<uint64_t>(63),
-                                                   Amt->getZExtValue())),
-        "vshr_n");
+        Ops[0], ConstantInt::get(Int64Ty, Amt->getZExtValue()), "shld_n");
   }
   case NEON::BI__builtin_neon_vshrd_n_s64: {
     llvm::ConstantInt *Amt = cast<ConstantInt>(EmitScalarExpr(E->getArg(1)));
     return Builder.CreateAShr(
         Ops[0], ConstantInt::get(Int64Ty, std::min(static_cast<uint64_t>(63),
                                                    Amt->getZExtValue())),
-        "vshr_n");
+        "shrd_n");
   }
   case NEON::BI__builtin_neon_vshrd_n_u64: {
     llvm::ConstantInt *Amt = cast<ConstantInt>(EmitScalarExpr(E->getArg(1)));
-    return Builder.CreateLShr(
-        Ops[0], ConstantInt::get(Int64Ty, std::min(static_cast<uint64_t>(63),
-                                                   Amt->getZExtValue())),
-        "vshr_n");
+    uint64_t ShiftAmt = Amt->getZExtValue();
+    // Right-shifting an unsigned value by its size yields 0.
+    if (ShiftAmt == 64)
+      return ConstantInt::get(Int64Ty, 0);
+    return Builder.CreateLShr(Ops[0], ConstantInt::get(Int64Ty, ShiftAmt),
+                              "shrd_n");
   }
   case NEON::BI__builtin_neon_vsrad_n_s64: {
     llvm::ConstantInt *Amt = cast<ConstantInt>(EmitScalarExpr(E->getArg(2)));
     Ops[1] = Builder.CreateAShr(
         Ops[1], ConstantInt::get(Int64Ty, std::min(static_cast<uint64_t>(63),
                                                    Amt->getZExtValue())),
-        "vshr_n");
+        "shrd_n");
     return Builder.CreateAdd(Ops[0], Ops[1]);
   }
   case NEON::BI__builtin_neon_vsrad_n_u64: {
     llvm::ConstantInt *Amt = cast<ConstantInt>(EmitScalarExpr(E->getArg(2)));
-    Ops[1] = Builder.CreateLShr(
-        Ops[1], ConstantInt::get(Int64Ty, std::min(static_cast<uint64_t>(63),
-                                                   Amt->getZExtValue())),
-        "vshr_n");
+    uint64_t ShiftAmt = Amt->getZExtValue();
+    // Right-shifting an unsigned value by its size yields 0.
+    // As Op + 0 = Op, return Ops[0] directly.
+    if (ShiftAmt == 64)
+      return Ops[0];
+    Ops[1] = Builder.CreateLShr(Ops[1], ConstantInt::get(Int64Ty, ShiftAmt),
+                                "shrd_n");
     return Builder.CreateAdd(Ops[0], Ops[1]);
   }
   case NEON::BI__builtin_neon_vqdmlalh_lane_s16:
diff --git a/clang/test/CodeGen/aarch64-neon-intrinsics.c b/clang/test/CodeGen/aarch64-neon-intrinsics.c
index e7fa048881f..83214f1f3ed 100644
--- a/clang/test/CodeGen/aarch64-neon-intrinsics.c
+++ b/clang/test/CodeGen/aarch64-neon-intrinsics.c
@@ -8572,11 +8572,24 @@ int64x1_t test_vshr_n_s64(int64x1_t a) {
 }
 
 uint64_t test_vshrd_n_u64(uint64_t a) {
-// CHECK-LABEL: test_vshrd_n_u64
-// CHECK: {{ushr d[0-9]+, d[0-9]+, #64|lsr x0, x0, #63}}
+// CHECK-AARCH64-LABEL: test_vshrd_n_u64
+// CHECK-AARCH64: {{ushr d[0-9]+, d[0-9]+, #64}}
+
+// CHECK-ARM64-LABEL: test_vshrd_n_u64
+// CHECK-ARM64: mov x0, xzr
   return (uint64_t)vshrd_n_u64(a, 64);
 }
 
+uint64_t test_vshrd_n_u64_2() {
+// CHECK-AARCH64-LABEL: test_vshrd_n_u64_2
+// CHECK-AARCH64: {{ushr d[0-9]+, d[0-9]+, #64}}
+
+// CHECK-ARM64-LABEL: test_vshrd_n_u64_2
+// CHECK-ARM64: mov x0, xzr
+  uint64_t a = UINT64_C(0xf000000000000000);
+  return vshrd_n_u64(a, 64);
+}
+
 uint64x1_t test_vshr_n_u64(uint64x1_t a) {
 // CHECK-LABEL: test_vshr_n_u64
 // CHECK: ushr {{d[0-9]+}}, {{d[0-9]+}}, #1
@@ -8625,6 +8638,15 @@ uint64_t test_vsrad_n_u64(uint64_t a, uint64_t b) {
   return (uint64_t)vsrad_n_u64(a, b, 63);
 }
 
+uint64_t test_vsrad_n_u64_2(uint64_t a, uint64_t b) {
+// CHECK-AARCH64-LABEL: test_vsrad_n_u64_2
+// CHECK-AARCH64: {{usra d[0-9]+, d[0-9]+, #64}}
+
+// CHECK-ARM64-LABEL: test_vsrad_n_u64_2
+// CHECK-ARM64-NOT: add
+  return (uint64_t)vsrad_n_u64(a, b, 64);
+}
+
 uint64x1_t test_vsra_n_u64(uint64x1_t a, uint64x1_t b) {
 // CHECK-LABEL: test_vsra_n_u64
 // CHECK: usra d{{[0-9]+}}, d{{[0-9]+}}, #1