Fix warning of CUDA ops (#2324)

mmcv/ops/csrc/common/cuda/diff_iou_rotated_cuda_kernel.cuh

@@ -44,6 +44,7 @@ __device__ bool compare_vertices(float x1, float y1, float x2, float y2) {
     else
       return false;
   }
+  return false;
 }
 
 __global__ void diff_iou_rotated_sort_vertices_forward_cuda_kernel(

mmcv/ops/csrc/common/utils/spconv/spconv/geometry.h

@@ -36,7 +36,7 @@ TV_HOST_DEVICE Index getValidOutPos(const Index *input_pos,
   Index m, offset;
   bool valid = false;
 #pragma unroll
-  for (int i = 0; i < NDim; ++i) {
+  for (unsigned i = 0; i < NDim; ++i) {
     lowers[i] = (input_pos[i] - (kernelSize[i] - 1) * dilation[i] - 1 +
                  stride[i] + padding[i]) /
                 stride[i];
@@ -50,7 +50,7 @@ TV_HOST_DEVICE Index getValidOutPos(const Index *input_pos,
   }
 
 #pragma unroll
-  for (int i = 0; i < NDim; ++i) {
+  for (unsigned i = 0; i < NDim; ++i) {
     counter[i] = 0;
   }
   for (int i = 0; i < numPoints; ++i) {
@@ -98,7 +98,7 @@ TV_HOST_DEVICE Index getValidOutPosTranspose(
   Index m, offset;
   bool valid = false;
 #pragma unroll
-  for (int i = 0; i < NDim; ++i) {
+  for (unsigned i = 0; i < NDim; ++i) {
     lowers[i] = input_pos[i] * stride[i] - padding[i];
     uppers[i] = lowers[i] + (kernelSize[i] - 1) * dilation[i];
   }
@@ -108,7 +108,7 @@ TV_HOST_DEVICE Index getValidOutPosTranspose(
     numPoints *= counterSize[i];
   }
 #pragma unroll
-  for (int i = 0; i < NDim; ++i) {
+  for (unsigned i = 0; i < NDim; ++i) {
     counter[i] = 0;
   }
   for (int i = 0; i < numPoints; ++i) {
@@ -251,9 +251,7 @@ Index getIndicePairsSubM(tv::TensorView<const Index> indicesIn,
                          const Index *const stride, const Index *const padding,
                          const Index *dilation,
                          const Index *const outSpatialShape) {
-  Index numAct = 0;
   auto numActIn = indicesIn.dim(0);
-  Index batchIdx = 0;
   Index spatialVolume = 1;
 #pragma unroll
   for (int i = 0; i < NDim; ++i) {

mmcv/ops/csrc/pytorch/cuda/fused_spconv_ops_cuda.cu

@@ -40,9 +40,6 @@ torch::Tensor FusedIndiceConvBatchnormCUDAKernelLauncher(
                // add.
     torch::mm_out(output, features, filters[indicePairMaxOffset]);
   }
-  double totalGatherTime = 0;
-  double totalGEMMTime = 0;
-  double totalSAddTime = 0;
   for (int i = 0; i < kernelVolume; ++i) {
     auto nHot = indicePairNumCpu.data_ptr<int>()[i];
     if (nHot <= 0 || (subM && i == indicePairMaxOffset)) {

mmcv/ops/csrc/pytorch/cuda/sparse_pool_ops_cuda.cu

@@ -17,7 +17,6 @@ torch::Tensor IndiceMaxpoolForwardCUDAKernelLauncher(torch::Tensor features,
   auto options =
       torch::TensorOptions().dtype(features.dtype()).device(features.device());
   torch::Tensor output = torch::zeros({numAct, numInPlanes}, options);
-  double totalTime = 0;
   for (int i = 0; i < kernelVolume; ++i) {
     auto nHot = indicePairNumCpu.data_ptr<int>()[i];
     if (nHot <= 0) {