Merge branch 'master' of https://github.com/traveller59/spconv

spconv/core_cc/csrc/sparse/convops/spops.pyi

@@ -56,7 +56,7 @@ class ConvGemmOps:
         """
         ...
     @staticmethod
-    def implicit_gemm(allocator, conv_tuner, features: Tensor, filters: Tensor, pair_fwd: Tensor, pair_mask_fwd_splits: List[Tensor], mask_argsort_fwd_splits: List[Tensor], num_activate_out: int, masks: Tensor, arch: Tuple[int, int], is_train: bool = False, is_subm: bool = False, stream_int: int = 0, timer: CUDAKernelTimer =  CUDAKernelTimer(False), auto_fp32_accum: bool = True, fp32_accum: bool = False) -> int: 
+    def implicit_gemm(allocator, conv_tuner, features: Tensor, filters: Tensor, pair_fwd: Tensor, pair_mask_fwd_splits: List[Tensor], mask_argsort_fwd_splits: List[Tensor], num_activate_out: int, masks: Tensor, arch: Tuple[int, int], is_train: bool = False, is_subm: bool = False, stream_int: int = 0, timer: CUDAKernelTimer =  CUDAKernelTimer(False), auto_fp32_accum: bool = True, fp32_accum: bool = False) -> Tuple[int, Any]: 
         """
         Args:
             allocator: 

spconv/csrc/sparse/convops.py

@@ -2013,9 +2013,9 @@ class ConvGemmOps(pccm.ParameterizedClass):
         // tv::ssprint(tune_res.algo_desp.__repr__(), "WTF", exists, 
         //     features.shape(), filters.shape(), out_features.shape(), tv::CUDAEvent::sync_and_duration(start_ev, end_ev));
 
-        return mask_width;
+        return std::make_tuple(mask_width, tune_res);
         """)
-        return code.ret("int")
+        return code.ret("std::tuple<int, ConvTuneResult>")
 
     @pccm.pybind.mark
     @pccm.static_function

spconv/pytorch/conv.py

@@ -136,6 +136,12 @@ class SparseConvolution(SparseModule):
             self._register_load_state_dict_pre_hook(
                 self._load_weight_different_layout)
 
+    def get_max_num_voxels(self) -> Optional[torch.Tensor]:
+        if hasattr(self, _MAX_NUM_VOXELS_DURING_TRAINING):
+            return getattr(self, _MAX_NUM_VOXELS_DURING_TRAINING)
+        return None 
+
+
     def _load_weight_different_layout(self, state_dict, prefix, local_metadata,
                                       strict, missing_keys, unexpected_keys,
                                       error_msgs):

spconv/pytorch/ops.py

@@ -1413,7 +1413,7 @@ def implicit_gemm(features: torch.Tensor,
             fp32_accum = False
         arch = get_arch()
 
-        mask_width = ConvGemmOps.implicit_gemm(
+        mask_width, tune_res_cpp = ConvGemmOps.implicit_gemm(
             alloc, CONV_CPP, features_tv, filters_tv, pair_fwd_tv,
             pair_mask_fwd_splits_tv, mask_argsort_fwd_splits_tv,
             num_activate_out, mask_tv, arch, is_train, is_subm, stream,

spconv/pytorch/pool.py

@@ -90,6 +90,12 @@ class SparseMaxPool(SparseModule):
             s += f', algo={self.algo}'
         return s.format(**self.__dict__)
 
+    def get_max_num_voxels(self) -> Optional[torch.Tensor]:
+        if hasattr(self, _MAX_NUM_VOXELS_DURING_TRAINING):
+            return getattr(self, _MAX_NUM_VOXELS_DURING_TRAINING)
+        return None 
+
+
     def forward(self, input):
         assert isinstance(input, spconv.SparseConvTensor)
         features = input.features
@@ -283,6 +289,11 @@ class SparseAvgPool(SparseModule):
             s += f', algo={self.algo}'
         return s.format(**self.__dict__)
         
+    def get_max_num_voxels(self) -> Optional[torch.Tensor]:
+        if hasattr(self, _MAX_NUM_VOXELS_DURING_TRAINING):
+            return getattr(self, _MAX_NUM_VOXELS_DURING_TRAINING)
+        return None 
+
     def forward(self, input):
         assert isinstance(input, spconv.SparseConvTensor)
         features = input.features