This paper pays close attention to the cross-modality visible-infrared person re-identification (VI Re-ID) task, which aims to match human samples between visible and infrared modes. In order to reduce the discrepancy between features of different modalities, most existing works usually use constraints based on Euclidean metric. Since the Euclidean based distance metric cannot effectively measure the internal angles between the embedded vectors, the above methods cannot learn the angularly discriminative feature embedding. Because the most important factor affecting the classification task based on embedding vector is whether there is an angularly discriminativ feature space, in this paper, we propose a new loss function called Enumerate Angular Triplet (EAT) loss. Also, motivated by the knowledge distillation, to narrow down the features between different modalities before feature embedding, we further present a new Cross-Modality Knowledge Distillation (CMKD) loss. The experimental results on RegDB and SYSU-MM01 datasets have shown that the proposed method is superior to the other most advanced methods in terms of impressive performance.