diff --git a/tests/pytorch/matmul/config.nix b/tests/pytorch/matmul/config.nix
new file mode 100644
index 0000000000..d00359f79b
--- /dev/null
+++ b/tests/pytorch/matmul/config.nix
@@ -0,0 +1,30 @@
+{
+  includes = [
+    ../memref.h
+  ];
+
+  buddyOptArgs = [
+    [
+      "--pass-pipeline"
+      "builtin.module(func.func(tosa-to-linalg-named, tosa-to-linalg, tosa-to-tensor, tosa-to-arith), empty-tensor-to-alloc-tensor, convert-elementwise-to-linalg, arith-bufferize, func.func(linalg-bufferize, tensor-bufferize), func-bufferize)"
+    ]
+    [
+      "--pass-pipeline"
+      "builtin.module(func.func(buffer-deallocation-simplification, convert-linalg-to-loops), eliminate-empty-tensors, func.func(llvm-request-c-wrappers))"
+    ]
+    [
+      "--lower-affine"
+      "--convert-math-to-llvm"
+      "--convert-math-to-libm"
+      "--convert-scf-to-cf"
+      "--convert-arith-to-llvm"
+      "--expand-strided-metadata"
+      "--finalize-memref-to-llvm"
+      "--lower-vector-exp"
+      "--lower-rvv=rv32"
+      "--convert-vector-to-llvm"
+      "--convert-func-to-llvm"
+      "--reconcile-unrealized-casts"
+    ]
+  ];
+}
diff --git a/tests/pytorch/matmul/matmul.c b/tests/pytorch/matmul/matmul.c
new file mode 100644
index 0000000000..d8aa32497c
--- /dev/null
+++ b/tests/pytorch/matmul/matmul.c
@@ -0,0 +1,39 @@
+#include "memref.h"
+
+NEW_MEMREF(float, 3);
+
+extern void _mlir_ciface_forward(struct MemRef_float_dim1 *output,
+                                 struct MemRef_float_dim1 *arg1,
+                                 struct MemRef_float_dim1 *arg2);
+
+__attribute((section(".vdata"))) float input_float_0[64][32][2];
+struct MemRef_float_dim3 input1 = {
+    .allocatedPtr = input_float_0,
+    .alignedPtr = input_float_0,
+    .offset = 0,
+    .sizes = {64, 32, 2},
+    .strides = {2 * 32, 2, 1},
+};
+
+__attribute((section(".vdata"))) float input_float_1[64][2][8];
+struct MemRef_float_dim3 input2 = {
+    .allocatedPtr = input_float_1,
+    .alignedPtr = input_float_1,
+    .offset = 0,
+    .sizes = {64, 2, 8},
+    .strides = {2 * 8, 8, 1},
+};
+
+__attribute((section(".vdata"))) float output_float_0[64][32][8];
+struct MemRef_float_dim3 output = {
+    .allocatedPtr = output_float_0,
+    .alignedPtr = output_float_0,
+    .offset = 0,
+    .sizes = {64, 32, 8},
+    .strides = {32 * 8, 8, 1},
+};
+
+int test() {
+  _mlir_ciface_forward(&output, &input1, &input2);
+  return 0;
+}
diff --git a/tests/pytorch/matmul/matmul.py b/tests/pytorch/matmul/matmul.py
new file mode 100644
index 0000000000..86157e75db
--- /dev/null
+++ b/tests/pytorch/matmul/matmul.py
@@ -0,0 +1,26 @@
+import torch
+import torch._dynamo as dynamo
+from torch._inductor.decomposition import decompositions as inductor_decomp
+
+from buddy.compiler.frontend import DynamoCompiler
+from buddy.compiler.ops import tosa
+
+# Define the input data.
+float32_in1 = torch.randn(64, 32, 2).to(torch.float32)
+float32_in2 = torch.randn(64, 2, 8).to(torch.float32)
+
+# Initialize the dynamo compiler.
+dynamo_compiler = DynamoCompiler(
+    primary_registry=tosa.ops_registry,
+    aot_autograd_decomposition=inductor_decomp,
+)
+
+# Pass the function and input data to the dynamo compiler's importer, the 
+# importer will first build a graph. Then, lower the graph to top-level IR. 
+# (tosa, linalg, etc.). Finally, accepts the generated module and weight parameters.
+graphs = dynamo_compiler.importer(torch.matmul, *(float32_in1, float32_in2))
+graph = graphs[0]
+graph.lower_to_top_level_ir()
+
+with open("forward.mlir", "w") as mlir_module:
+    print(graph._imported_module, file = mlir_module)