diff --git a/test/lowering/gemm.c b/test/lowering/gemm.c
new file mode 100644
index 0000000..eabf328
--- /dev/null
+++ b/test/lowering/gemm.c
@@ -0,0 +1,113 @@
+#include <math.h>
+#include <stdio.h>
+#include <stdlib.h>
+
+#define TOLERANCE 1e-5
+
+// Define the memref struct
+typedef struct {
+  float *allocated;  // Pointer to allocated memory
+  float *aligned;    // Pointer to aligned data (if applicable)
+  size_t offset;     // Offset from aligned pointer to actual data
+  size_t sizes[2];   // Array to hold sizes for the 2D dimensions
+  size_t strides[2]; // Array to hold strides for the 2D dimensions
+} memref_t;
+
+// External declaration of the gemm function
+extern void gemm(memref_t A, memref_t B, memref_t C, float alpha, float beta);
+
+void initialize_matrix(memref_t *matrix, size_t rows, size_t cols) {
+  matrix->allocated = (float *)malloc(rows * cols * sizeof(float));
+  matrix->aligned = matrix->allocated; // Assume no special alignment needed
+  matrix->offset = 0;
+  matrix->sizes[0] = rows;
+  matrix->sizes[1] = cols;
+  matrix->strides[0] = cols; // Row-major layout
+  matrix->strides[1] = 1;
+}
+
+void free_matrix(memref_t *matrix) { free(matrix->allocated); }
+
+// Simple C implementation of gemm for verification
+void gemm_verify(memref_t *A, memref_t *B, memref_t *C, float alpha,
+                 float beta) {
+  size_t rows = C->sizes[0];
+  size_t cols = C->sizes[1];
+  size_t K = A->sizes[1];
+
+  for (size_t i = 0; i < rows; i++) {
+    for (size_t j = 0; j < cols; j++) {
+      float sum = 0.0f;
+      for (size_t k = 0; k < K; k++) {
+        float a_val = A->allocated[i * A->strides[0] + k * A->strides[1]];
+        float b_val = B->allocated[k * B->strides[0] + j * B->strides[1]];
+        sum += a_val * b_val;
+      }
+      float c_val = C->allocated[i * C->strides[0] + j * C->strides[1]];
+      C->allocated[i * C->strides[0] + j * C->strides[1]] =
+          alpha * sum + beta * c_val;
+    }
+  }
+}
+
+// Function to check if two matrices are approximately equal
+int verify_result(memref_t *C, memref_t *C_ref) {
+  size_t rows = C->sizes[0];
+  size_t cols = C->sizes[1];
+  for (size_t i = 0; i < rows; i++) {
+    for (size_t j = 0; j < cols; j++) {
+      float val = C->allocated[i * C->strides[0] + j * C->strides[1]];
+      float ref_val =
+          C_ref->allocated[i * C_ref->strides[0] + j * C_ref->strides[1]];
+      if (fabs(val - ref_val) > TOLERANCE) {
+        printf("Mismatch at C[%zu][%zu]: %f (expected %f)\n", i, j, val,
+               ref_val);
+        return 0;
+      }
+    }
+  }
+  return 1;
+}
+
+int main() {
+  memref_t A, B, C, C_ref;
+  float alpha = 1.0f, beta = 1.0f;
+  size_t rows = 100, cols = 100;
+
+  // Initialize matrices A, B, C, and C_ref with 100x100 dimensions
+  initialize_matrix(&A, rows, cols);
+  initialize_matrix(&B, rows, cols);
+  initialize_matrix(&C, rows, cols);
+  initialize_matrix(&C_ref, rows, cols);
+
+  // Fill matrices A and B with some values and initialize C and C_ref
+  for (size_t i = 0; i < rows; i++) {
+    for (size_t j = 0; j < cols; j++) {
+      A.allocated[i * A.strides[0] + j * A.strides[1]] = (float)(i + j);
+      B.allocated[i * B.strides[0] + j * B.strides[1]] = (float)(i - j);
+      C.allocated[i * C.strides[0] + j * C.strides[1]] = 0.0f;
+      C_ref.allocated[i * C_ref.strides[0] + j * C_ref.strides[1]] = 0.0f;
+    }
+  }
+
+  // Call the external gemm function
+  gemm(A, B, C, alpha, beta);
+
+  // Call the verification gemm function
+  gemm_verify(&A, &B, &C_ref, alpha, beta);
+
+  // Verify the results
+  if (verify_result(&C, &C_ref)) {
+    printf("Verification passed! The matrices match.\n");
+  } else {
+    printf("Verification failed! The matrices do not match.\n");
+  }
+
+  // Free allocated memory for matrices
+  free_matrix(&A);
+  free_matrix(&B);
+  free_matrix(&C);
+  free_matrix(&C_ref);
+
+  return 0;
+}
diff --git a/test/lowering/gemm.mlir b/test/lowering/gemm.mlir
index 099fdf4..7ff0822 100644
--- a/test/lowering/gemm.mlir
+++ b/test/lowering/gemm.mlir
@@ -7,8 +7,14 @@
 // RUN:     -reconcile-unrealized-casts -mlir-print-debuginfo -o %t.mlir 
 // RUN: %filecheck --input-file=%t.mlir %s
 // RUN: %gccjit-translate %t.mlir -mlir-to-gccjit-gimple | %filecheck %s --check-prefix=CHECK-GIMPLE
+// RUN: %gccjit-translate %t.mlir -mlir-to-gccjit-dylib -o %t.so
+// RUN: cc -O3 %p/gemm.c %t.so -Wl,-rpath,%T -o %t.exe
+// RUN: %t.exe | %filecheck %s --check-prefix=CHECK-OUTPUT
+
+// CHECK-OUTPUT: Verification passed! The matrices match.
 module @test attributes {
-      gccjit.opt_level = #gccjit.opt_level<O3>
+      gccjit.opt_level = #gccjit.opt_level<O3>,
+      gccjit.debug_info = true
 }
 {
   // CHECK-NOT: func.func