diff --git a/mlir/test/Integration/Dialect/SparseTensor/GPU/CUDA/sm80-lt/sparse-matmul-2-4-lib.mlir b/mlir/test/Integration/Dialect/SparseTensor/GPU/CUDA/sm80-lt/sparse-matmul-2-4-lib.mlir
--- a/mlir/test/Integration/Dialect/SparseTensor/GPU/CUDA/sm80-lt/sparse-matmul-2-4-lib.mlir
+++ b/mlir/test/Integration/Dialect/SparseTensor/GPU/CUDA/sm80-lt/sparse-matmul-2-4-lib.mlir
@@ -1,5 +1,5 @@
 //
-// NOTE: this test requires gpu-sm80
+// NOTE: this test requires gpu-sm80 and cusparselt
 //
 // RUN: mlir-opt --convert-scf-to-cf -convert-cf-to-llvm --convert-vector-to-llvm \
 // RUN:          --convert-arith-to-llvm --gpu-to-llvm --reconcile-unrealized-casts \
@@ -69,9 +69,9 @@
     %c64 = arith.constant 64  : index
 
     // Matrices A, B, C (16x32, 32x16, 16x16).
-    %a = memref.alloc() : memref<16x32xf16>  // 16x32 but 2:4, row-major
-    %b = memref.alloc() : memref<32x16xf16>   // regular dense  column-major
-    %c = memref.alloc() : memref<16x16xf16>   // accumulator    row-major
+    %a = memref.alloc() : memref<16x32xf16> // 16x32 with 2:4, row-major
+    %b = memref.alloc() : memref<32x16xf16> // regular dense   column-major
+    %c = memref.alloc() : memref<16x16xf16> // accumulator     row-major
 
     //
     // Setup matrix A.
@@ -181,27 +181,8 @@
       vector.print %pb0 : vector<16xf16>
     }
 
-    // Maps the provided host buffers into the device address space.
-    // Writes from the host are guaranteed to be visible to device
-    // kernels that are launched afterwards. Writes from the device
-    // are guaranteed to be visible on the host after synchronizing
-    // with the device kernel completion.
-    %cast_a = memref.cast %a : memref<16x32xf16> to memref<*xf16>
-    gpu.host_register %cast_a : memref<*xf16>
-    %cast_b = memref.cast %b : memref<32x16xf16> to memref<*xf16>
-    gpu.host_register %cast_b : memref<*xf16>
-    %cast_c = memref.cast %c : memref<16x16xf16> to memref<*xf16>
-    gpu.host_register %cast_c : memref<*xf16>
-
     // Call the kernel.
-    %t1  = arith.constant 1  : index
-    %t32 = arith.constant 32 : index
     call @sampled_matmul (%a, %b, %c): (memref<16x32xf16>, memref<32x16xf16>, memref<16x16xf16>) -> ()
-    
-    // Unmaps the host buffers.
-    gpu.host_unregister %cast_a : memref<*xf16>
-    gpu.host_unregister %cast_b : memref<*xf16>
-    gpu.host_unregister %cast_c : memref<*xf16>
 
     //
     // Verify computed matrix C.
@@ -227,7 +208,7 @@
       %pc0 = vector.transfer_read %c[%pci, %c0], %f0 : memref<16x16xf16>, vector<16xf16>
       vector.print %pc0 : vector<16xf16>
     }
-    
+
     llvm.call @mgpuDestroySparseLtEnv() : () -> ()
     return
   }