diff --git a/mlir/include/mlir/Dialect/LLVMIR/NVVMOps.td b/mlir/include/mlir/Dialect/LLVMIR/NVVMOps.td
index 14880a1a66ba5..6f8971ca3c00c 100644
--- a/mlir/include/mlir/Dialect/LLVMIR/NVVMOps.td
+++ b/mlir/include/mlir/Dialect/LLVMIR/NVVMOps.td
@@ -2017,10 +2017,10 @@ def NVVM_CpAsyncBulkTensorPrefetchOp :
       translatedOperands.push_back(moduleTranslation.lookupValue(v));
 
     llvm::LLVMContext &ctx = moduleTranslation.getLLVMContext();
-    auto *i64Undef = llvm::UndefValue::get(llvm::IntegerType::get(ctx, 64));
+    auto *i64Unused = llvm::ConstantInt::get(llvm::Type::getInt64Ty(ctx), 0);
 
     bool isCacheHint = op.getL2CacheHint() ? true : false;
-    translatedOperands.push_back(isCacheHint ? $l2CacheHint : i64Undef);
+    translatedOperands.push_back(isCacheHint ? $l2CacheHint : i64Unused);
     translatedOperands.push_back(builder.getInt1(isCacheHint));
 
     auto intId = NVVM::CpAsyncBulkTensorPrefetchOp::getIntrinsicID(
diff --git a/mlir/test/Target/LLVMIR/nvvm/tma_prefetch.mlir b/mlir/test/Target/LLVMIR/nvvm/tma_prefetch.mlir
new file mode 100644
index 0000000000000..7be29fd616a6f
--- /dev/null
+++ b/mlir/test/Target/LLVMIR/nvvm/tma_prefetch.mlir
@@ -0,0 +1,61 @@
+// RUN: mlir-translate -mlir-to-llvmir %s  -split-input-file --verify-diagnostics | FileCheck %s
+
+// CHECK-LABEL: @tma_prefetch_1d
+llvm.func @tma_prefetch_1d(%tma_desc : !llvm.ptr, %d0 : i32, %ch : i64) {
+  // CHECK: call void @llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.1d(ptr %0, i32 %{{.*}}, i64 0, i1 false)
+  // CHECK: call void @llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.1d(ptr %0, i32 %{{.*}}, i64 %{{.*}}, i1 true)
+  nvvm.cp.async.bulk.tensor.prefetch %tma_desc, box[%d0] : !llvm.ptr
+  nvvm.cp.async.bulk.tensor.prefetch %tma_desc, box[%d0] l2_cache_hint = %ch : !llvm.ptr
+  llvm.return
+}
+
+// CHECK-LABEL: @tma_prefetch_2d
+llvm.func @tma_prefetch_2d(%tma_desc : !llvm.ptr, %d0 : i32, %d1 : i32, %ch : i64) {
+  // CHECK: call void @llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.2d(ptr %0, i32 %{{.*}}, i32 %{{.*}}, i64 0, i1 false)
+  // CHECK: call void @llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.2d(ptr %0, i32 %{{.*}}, i32 %{{.*}}, i64 %{{.*}}, i1 true)
+  nvvm.cp.async.bulk.tensor.prefetch %tma_desc, box[%d0, %d1] : !llvm.ptr
+  nvvm.cp.async.bulk.tensor.prefetch %tma_desc, box[%d0, %d1] l2_cache_hint = %ch : !llvm.ptr
+  llvm.return
+}
+
+// CHECK-LABEL: @tma_prefetch_3d
+llvm.func @tma_prefetch_3d(%tma_desc : !llvm.ptr, %d0 : i32, %d1 : i32, %d2 : i32, %off0 : i16, %ch : i64) {
+  // CHECK: call void @llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.3d(ptr %0, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i64 0, i1 false)
+  // CHECK: call void @llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.3d(ptr %0, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i64 %{{.*}}, i1 true)
+  nvvm.cp.async.bulk.tensor.prefetch %tma_desc, box[%d0, %d1, %d2] : !llvm.ptr
+  nvvm.cp.async.bulk.tensor.prefetch %tma_desc, box[%d0, %d1, %d2] l2_cache_hint = %ch : !llvm.ptr
+
+  // CHECK: call void @llvm.nvvm.cp.async.bulk.tensor.prefetch.im2col.3d(ptr %0, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i16 %{{.*}}, i64 0, i1 false)
+  // CHECK: call void @llvm.nvvm.cp.async.bulk.tensor.prefetch.im2col.3d(ptr %0, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i16 %{{.*}}, i64 %{{.*}}, i1 true)
+  nvvm.cp.async.bulk.tensor.prefetch %tma_desc, box[%d0, %d1, %d2] im2col[%off0] : !llvm.ptr
+  nvvm.cp.async.bulk.tensor.prefetch %tma_desc, box[%d0, %d1, %d2] im2col[%off0] l2_cache_hint = %ch : !llvm.ptr
+  llvm.return
+}
+
+// CHECK-LABEL: @tma_prefetch_4d
+llvm.func @tma_prefetch_4d(%tma_desc : !llvm.ptr, %d0 : i32, %d1 : i32, %d2 : i32, %d3 : i32, %off0 : i16, %off1 : i16, %ch : i64) {
+  // CHECK: call void @llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.4d(ptr %0, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i64 0, i1 false)
+  // CHECK: call void @llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.4d(ptr %0, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i64 %{{.*}}, i1 true)
+  nvvm.cp.async.bulk.tensor.prefetch %tma_desc, box[%d0, %d1, %d2, %d3] : !llvm.ptr
+  nvvm.cp.async.bulk.tensor.prefetch %tma_desc, box[%d0, %d1, %d2, %d3] l2_cache_hint = %ch : !llvm.ptr
+
+  // CHECK: call void @llvm.nvvm.cp.async.bulk.tensor.prefetch.im2col.4d(ptr %0, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i16 %{{.*}}, i16 %{{.*}}, i64 0, i1 false)
+  // CHECK: call void @llvm.nvvm.cp.async.bulk.tensor.prefetch.im2col.4d(ptr %0, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i16 %{{.*}}, i16 %{{.*}}, i64 %{{.*}}, i1 true)
+  nvvm.cp.async.bulk.tensor.prefetch %tma_desc, box[%d0, %d1, %d2, %d3] im2col[%off0, %off1] : !llvm.ptr
+  nvvm.cp.async.bulk.tensor.prefetch %tma_desc, box[%d0, %d1, %d2, %d3] im2col[%off0, %off1] l2_cache_hint = %ch : !llvm.ptr
+  llvm.return
+}
+
+// CHECK-LABEL: @tma_prefetch_5d
+llvm.func @tma_prefetch_5d(%tma_desc : !llvm.ptr, %d0 : i32, %d1 : i32, %d2 : i32, %d3 : i32, %d4 : i32, %off0 : i16, %off1 : i16, %off2 : i16, %ch : i64) {
+  // CHECK: call void @llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.5d(ptr %0, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i64 0, i1 false)
+  // CHECK: call void @llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.5d(ptr %0, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i64 %{{.*}}, i1 true)
+  nvvm.cp.async.bulk.tensor.prefetch %tma_desc, box[%d0, %d1, %d2, %d3, %d4] : !llvm.ptr
+  nvvm.cp.async.bulk.tensor.prefetch %tma_desc, box[%d0, %d1, %d2, %d3, %d4] l2_cache_hint = %ch : !llvm.ptr
+
+  // CHECK: call void @llvm.nvvm.cp.async.bulk.tensor.prefetch.im2col.5d(ptr %0, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i16 %{{.*}}, i16 %{{.*}}, i16 %{{.*}}, i64 0, i1 false)
+  // CHECK: call void @llvm.nvvm.cp.async.bulk.tensor.prefetch.im2col.5d(ptr %0, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i16 %{{.*}}, i16 %{{.*}}, i16 %{{.*}}, i64 %{{.*}}, i1 true)
+  nvvm.cp.async.bulk.tensor.prefetch %tma_desc, box[%d0, %d1, %d2, %d3, %d4] im2col[%off0, %off1, %off2] : !llvm.ptr
+  nvvm.cp.async.bulk.tensor.prefetch %tma_desc, box[%d0, %d1, %d2, %d3, %d4] im2col[%off0, %off1, %off2] l2_cache_hint = %ch : !llvm.ptr
+  llvm.return
+}
diff --git a/mlir/test/Target/LLVMIR/nvvmir-invalid.mlir b/mlir/test/Target/LLVMIR/nvvmir-invalid.mlir
index 194011d1621a1..44c7126255dc4 100644
--- a/mlir/test/Target/LLVMIR/nvvmir-invalid.mlir
+++ b/mlir/test/Target/LLVMIR/nvvmir-invalid.mlir
@@ -2,6 +2,41 @@
 
 // -----
 
+llvm.func @kernel_func(%numberOfThreads : i32) {
+  // expected-error @below {{'nvvm.barrier' op barrier id is missing, it should be set between 0 to 15}}
+  nvvm.barrier number_of_threads = %numberOfThreads
+}
+
+// -----
+
+// expected-error @below {{'"nvvm.minctasm"' attribute must be integer constant}}
+llvm.func @kernel_func() attributes {nvvm.kernel, nvvm.minctasm = "foo"} {
+  llvm.return
+}
+
+// -----
+
+// expected-error @below {{'"nvvm.maxnreg"' attribute must be integer constant}}
+llvm.func @kernel_func() attributes {nvvm.kernel, nvvm.maxnreg = "boo"} {
+  llvm.return
+}
+
+// -----
+
+// expected-error @below {{'"nvvm.reqntid"' attribute must be integer array with maximum 3 index}}
+llvm.func @kernel_func() attributes {nvvm.kernel, nvvm.reqntid = array<i32: 3, 4, 5, 6>} {
+  llvm.return
+}
+
+// -----
+
+// expected-error @below {{'"nvvm.maxntid"' attribute must be integer array with maximum 3 index}}
+llvm.func @kernel_func() attributes {nvvm.kernel, nvvm.maxntid = array<i32: 3, 4, 5, 6>} {
+  llvm.return
+}
+
+// -----
+
 llvm.func @nvvm_fence_proxy_acquire(%addr : !llvm.ptr, %size : i32) {
   // expected-error @below {{'nvvm.fence.proxy.acquire' op uni-directional proxies only support generic for from_proxy attribute}}
   nvvm.fence.proxy.acquire #nvvm.mem_scope<cta> %addr, %size from_proxy=#nvvm.proxy_kind<tensormap> to_proxy=#nvvm.proxy_kind<generic>
diff --git a/mlir/test/Target/LLVMIR/nvvmir.mlir b/mlir/test/Target/LLVMIR/nvvmir.mlir
index a4a3581d6b759..2749e42c40bc3 100644
--- a/mlir/test/Target/LLVMIR/nvvmir.mlir
+++ b/mlir/test/Target/LLVMIR/nvvmir.mlir
@@ -642,39 +642,6 @@ llvm.func @kernel_func() attributes {nvvm.kernel, nvvm.maxntid = array<i32: 1, 2
 // CHECK:     {ptr @kernel_func, !"maxntidz", i32 32}
 // CHECK:     {ptr @kernel_func, !"minctasm", i32 16}
 
-// -----
-
-llvm.func @kernel_func(%numberOfThreads : i32) {
-  // expected-error @below {{'nvvm.barrier' op barrier id is missing, it should be set between 0 to 15}}
-  nvvm.barrier number_of_threads = %numberOfThreads
-}
-
-// -----
-// expected-error @below {{'"nvvm.minctasm"' attribute must be integer constant}}
-llvm.func @kernel_func() attributes {nvvm.kernel,
-nvvm.minctasm = "foo"} {
-  llvm.return
-}
-
-
-// -----
-// expected-error @below {{'"nvvm.maxnreg"' attribute must be integer constant}}
-llvm.func @kernel_func() attributes {nvvm.kernel,
-nvvm.maxnreg = "boo"} {
-  llvm.return
-}
-// -----
-// expected-error @below {{'"nvvm.reqntid"' attribute must be integer array with maximum 3 index}}
-llvm.func @kernel_func() attributes {nvvm.kernel, nvvm.reqntid = array<i32: 3, 4, 5, 6>} {
-  llvm.return
-}
-
-// -----
-// expected-error @below {{'"nvvm.maxntid"' attribute must be integer array with maximum 3 index}}
-llvm.func @kernel_func() attributes {nvvm.kernel, nvvm.maxntid = array<i32: 3, 4, 5, 6>} {
-  llvm.return
-}
-
 // -----
 // CHECK: !nvvm.annotations =
 // CHECK: !1 = !{ptr @kernel_func, !"grid_constant", !2}
@@ -737,65 +704,3 @@ llvm.func @nvvm_breakpoint() {
   nvvm.breakpoint
   llvm.return
 }
-
-// -----
-
-// CHECK-LABEL: @tma_prefetch_1d
-llvm.func @tma_prefetch_1d(%tma_desc : !llvm.ptr, %d0 : i32, %ch : i64) {
-  // CHECK-LLVM: call void @llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.1d(ptr %0, i32 %{{.*}}, i64 undef, i1 false)
-  // CHECK-LLVM: call void @llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.1d(ptr %0, i32 %{{.*}}, i64 %{{.*}}, i1 true)
-  nvvm.cp.async.bulk.tensor.prefetch %tma_desc, box[%d0] : !llvm.ptr
-  nvvm.cp.async.bulk.tensor.prefetch %tma_desc, box[%d0] l2_cache_hint = %ch : !llvm.ptr
-  llvm.return
-}
-
-// CHECK-LABEL: @tma_prefetch_2d
-llvm.func @tma_prefetch_2d(%tma_desc : !llvm.ptr, %d0 : i32, %d1 : i32, %ch : i64) {
-  // CHECK-LLVM: call void @llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.2d(ptr %0, i32 %{{.*}}, i32 %{{.*}}, i64 undef, i1 false)
-  // CHECK-LLVM: call void @llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.2d(ptr %0, i32 %{{.*}}, i32 %{{.*}}, i64 %{{.*}}, i1 true)
-  nvvm.cp.async.bulk.tensor.prefetch %tma_desc, box[%d0, %d1] : !llvm.ptr
-  nvvm.cp.async.bulk.tensor.prefetch %tma_desc, box[%d0, %d1] l2_cache_hint = %ch : !llvm.ptr
-  llvm.return
-}
-
-// CHECK-LABEL: @tma_prefetch_3d
-llvm.func @tma_prefetch_3d(%tma_desc : !llvm.ptr, %d0 : i32, %d1 : i32, %d2 : i32, %off0 : i16, %ch : i64) {
-  // CHECK-LLVM: call void @llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.3d(ptr %0, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i64 undef, i1 false)
-  // CHECK-LLVM: call void @llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.3d(ptr %0, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i64 %{{.*}}, i1 true)
-  nvvm.cp.async.bulk.tensor.prefetch %tma_desc, box[%d0, %d1, %d2] : !llvm.ptr
-  nvvm.cp.async.bulk.tensor.prefetch %tma_desc, box[%d0, %d1, %d2] l2_cache_hint = %ch : !llvm.ptr
-
-  // CHECK-LLVM: call void @llvm.nvvm.cp.async.bulk.tensor.prefetch.im2col.3d(ptr %0, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i16 %{{.*}}, i64 undef, i1 false)
-  // CHECK-LLVM: call void @llvm.nvvm.cp.async.bulk.tensor.prefetch.im2col.3d(ptr %0, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i16 %{{.*}}, i64 %{{.*}}, i1 true)
-  nvvm.cp.async.bulk.tensor.prefetch %tma_desc, box[%d0, %d1, %d2] im2col[%off0] : !llvm.ptr
-  nvvm.cp.async.bulk.tensor.prefetch %tma_desc, box[%d0, %d1, %d2] im2col[%off0] l2_cache_hint = %ch : !llvm.ptr
-  llvm.return
-}
-
-// CHECK-LABEL: @tma_prefetch_4d
-llvm.func @tma_prefetch_4d(%tma_desc : !llvm.ptr, %d0 : i32, %d1 : i32, %d2 : i32, %d3 : i32, %off0 : i16, %off1 : i16, %ch : i64) {
-  // CHECK-LLVM: call void @llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.4d(ptr %0, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i64 undef, i1 false)
-  // CHECK-LLVM: call void @llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.4d(ptr %0, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i64 %{{.*}}, i1 true)
-  nvvm.cp.async.bulk.tensor.prefetch %tma_desc, box[%d0, %d1, %d2, %d3] : !llvm.ptr
-  nvvm.cp.async.bulk.tensor.prefetch %tma_desc, box[%d0, %d1, %d2, %d3] l2_cache_hint = %ch : !llvm.ptr
-
-  // CHECK-LLVM: call void @llvm.nvvm.cp.async.bulk.tensor.prefetch.im2col.4d(ptr %0, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i16 %{{.*}}, i16 %{{.*}}, i64 undef, i1 false)
-  // CHECK-LLVM: call void @llvm.nvvm.cp.async.bulk.tensor.prefetch.im2col.4d(ptr %0, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i16 %{{.*}}, i16 %{{.*}}, i64 %{{.*}}, i1 true)
-  nvvm.cp.async.bulk.tensor.prefetch %tma_desc, box[%d0, %d1, %d2, %d3] im2col[%off0, %off1] : !llvm.ptr
-  nvvm.cp.async.bulk.tensor.prefetch %tma_desc, box[%d0, %d1, %d2, %d3] im2col[%off0, %off1] l2_cache_hint = %ch : !llvm.ptr
-  llvm.return
-}
-
-// CHECK-LABEL: @tma_prefetch_5d
-llvm.func @tma_prefetch_5d(%tma_desc : !llvm.ptr, %d0 : i32, %d1 : i32, %d2 : i32, %d3 : i32, %d4 : i32, %off0 : i16, %off1 : i16, %off2 : i16, %ch : i64) {
-  // CHECK-LLVM: call void @llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.5d(ptr %0, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i64 undef, i1 false)
-  // CHECK-LLVM: call void @llvm.nvvm.cp.async.bulk.tensor.prefetch.tile.5d(ptr %0, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i64 %{{.*}}, i1 true)
-  nvvm.cp.async.bulk.tensor.prefetch %tma_desc, box[%d0, %d1, %d2, %d3, %d4] : !llvm.ptr
-  nvvm.cp.async.bulk.tensor.prefetch %tma_desc, box[%d0, %d1, %d2, %d3, %d4] l2_cache_hint = %ch : !llvm.ptr
-
-  // CHECK-LLVM: call void @llvm.nvvm.cp.async.bulk.tensor.prefetch.im2col.5d(ptr %0, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i16 %{{.*}}, i16 %{{.*}}, i16 %{{.*}}, i64 undef, i1 false)
-  // CHECK-LLVM: call void @llvm.nvvm.cp.async.bulk.tensor.prefetch.im2col.5d(ptr %0, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i32 %{{.*}}, i16 %{{.*}}, i16 %{{.*}}, i16 %{{.*}}, i64 %{{.*}}, i1 true)
-  nvvm.cp.async.bulk.tensor.prefetch %tma_desc, box[%d0, %d1, %d2, %d3, %d4] im2col[%off0, %off1, %off2] : !llvm.ptr
-  nvvm.cp.async.bulk.tensor.prefetch %tma_desc, box[%d0, %d1, %d2, %d3, %d4] im2col[%off0, %off1, %off2] l2_cache_hint = %ch : !llvm.ptr
-  llvm.return
-}