changed cast to round to zero if casting from float to integral

mindspore/ccsrc/backend/kernel_compiler/gpu/cuda_impl/cast_impl.cu

@@ -28,35 +28,35 @@ __device__ __forceinline__ void CastBase(const S *input_addr, T *output_addr) {
 
 // half --> integer
 __device__ __forceinline__ void CastBase(const half *input_addr, uint64_t *output_addr) {
-  *output_addr = __half2ull_rd((*input_addr));
+  *output_addr = __half2ull_rz((*input_addr));
 }
 
 __device__ __forceinline__ void CastBase(const half *input_addr, int64_t *output_addr) {
-  *output_addr = __half2ll_rd((*input_addr));
+  *output_addr = __half2ll_rz((*input_addr));
 }
 
 __device__ __forceinline__ void CastBase(const half *input_addr, uint32_t *output_addr) {
-  *output_addr = __half2uint_rd((*input_addr));
+  *output_addr = __half2uint_rz((*input_addr));
 }
 
 __device__ __forceinline__ void CastBase(const half *input_addr, int32_t *output_addr) {
-  *output_addr = __half2int_rd((*input_addr));
+  *output_addr = __half2int_rz((*input_addr));
 }
 
 __device__ __forceinline__ void CastBase(const half *input_addr, uint16_t *output_addr) {
-  *output_addr = __half2ushort_rd((*input_addr));
+  *output_addr = __half2ushort_rz((*input_addr));
 }
 
 __device__ __forceinline__ void CastBase(const half *input_addr, int16_t *output_addr) {
-  *output_addr = __half2short_rd((*input_addr));
+  *output_addr = __half2short_rz((*input_addr));
 }
 
 __device__ __forceinline__ void CastBase(const half *input_addr, uint8_t *output_addr) {
-  *output_addr = static_cast<uint8_t>(__half2ushort_rd((*input_addr)));
+  *output_addr = static_cast<uint8_t>(__half2ushort_rz((*input_addr)));
 }
 
 __device__ __forceinline__ void CastBase(const half *input_addr, int8_t *output_addr) {
-  *output_addr = static_cast<int8_t>(__half2short_rd((*input_addr)));
+  *output_addr = static_cast<int8_t>(__half2short_rz((*input_addr)));
 }
 
 // integer --> half

tests/st/ops/gpu/test_cast_op.py

@@ -604,7 +604,7 @@ def test_cast31():
 @pytest.mark.env_onecard
 def test_cast32():
     np.random.seed(10)
-    x = np.random.rand(*(3, 2)).astype(np.float16)
+    x = np.random.uniform(-5, 5, (3, 2)).astype(np.float16)
     x0 = Tensor(x)
     t0 = mstype.int32
     x1 = Tensor(x)