Merge pull request #1130 from InfiniTensor/issue/1129

issue/1129 - fix dtype generalization and used variables

Author: wooway777

src/infiniop/ops/huber_loss/nvidia/huber_loss_nvidia.cu

@@ -105,7 +105,7 @@ infiniStatus_t Descriptor::calculate(
         launch_kernel<half>(output, input, target, workspace, _info, stream);
         break;
     case INFINI_DTYPE_BF16:
-        launch_kernel<nv_bfloat16>(output, input, target, workspace, _info, stream);
+        launch_kernel<cuda_bfloat16>(output, input, target, workspace, _info, stream);
         break;
     case INFINI_DTYPE_F32:
         launch_kernel<float>(output, input, target, workspace, _info, stream);

src/infiniop/ops/kthvalue/cuda/kernel.cuh

@@ -115,7 +115,7 @@ __global__ void kthvalue_kernel(
     for (unsigned int size = 2; size <= power_of_2_dim; size <<= 1) {
         // Bitonic Merge
         // dir: 升序或降序交替，构造双调序列
-        bool dir = (tid & (size / 2)) == 0;
+        // bool dir = (tid & (size / 2)) == 0;
 
         // 这里的逻辑稍微复杂，为了简单和稳定，我们使用全升序排序逻辑
         // 标准 Bitonic Sort 代码如下：