diff --git a/tests/unit/test_fmha_6warp_multirow.cu b/tests/unit/test_fmha_6warp_multirow.cu
index 8c0ddcd0..834cf8db 100644
--- a/tests/unit/test_fmha_6warp_multirow.cu
+++ b/tests/unit/test_fmha_6warp_multirow.cu
@@ -142,12 +142,12 @@ static int test_single_T(int T, int n_h = 1, int batch = 1) {
         pass = 0; goto cleanup;
     }
 
-    cudaMemcpy(h_o, d_o, total_heads * T * HD * sizeof(bf16_t), cudaMemcpyDeviceToHost);
-    cudaMemcpy(h_lse, d_lse, total_heads * T * sizeof(float), cudaMemcpyDeviceToHost);
-
-    // Verify each head
+    // Verify each head (declared before goto targets)
     int checked = 0, failed = 0;
     float min_cos = 1.0f;
+
+    cudaMemcpy(h_o, d_o, total_heads * T * HD * sizeof(bf16_t), cudaMemcpyDeviceToHost);
+    cudaMemcpy(h_lse, d_lse, total_heads * T * sizeof(float), cudaMemcpyDeviceToHost);
     for (int b = 0; b < batch; b++) {
         for (int h = 0; h < n_h; h++) {
             int idx = b * n_h + h;