add new prebuilts for 3.13 and cu126

spconv/core_cc/csrc/sparse/inference.pyi

 from typing import overload, Any, Callable, Dict, List, Optional, Set, Tuple, Type, Union
-from pccm.stubs import EnumValue, EnumClassValue
+from pccm.stubs import EnumValue, EnumClassValue, enum
 from cumm.tensorview import Tensor
 from cumm.tensorview.gemm import Activation
 class InferenceOps:

spconv/core_cc/csrc/utils/boxops.pyi

 from typing import overload, Any, Callable, Dict, List, Optional, Set, Tuple, Type, Union
-from pccm.stubs import EnumValue, EnumClassValue
+from pccm.stubs import EnumValue, EnumClassValue, enum
 from cumm.tensorview import Tensor
 class BoxOps:
     @staticmethod

spconv/core_cc/csrc/utils/pcc.pyi

 from typing import overload, Any, Callable, Dict, List, Optional, Set, Tuple, Type, Union
-from pccm.stubs import EnumValue, EnumClassValue
+from pccm.stubs import EnumValue, EnumClassValue, enum
 from cumm.tensorview import Tensor
 class PointCloudCompress:
     @staticmethod
@@ -43,8 +43,8 @@ class PointCloudCompress:
             data: 
         """
         ...
-    class EncodeType:
-        XYZ_8 = EnumClassValue(0) # type: EnumClassValue
-        XYZI_8 = EnumClassValue(1) # type: EnumClassValue
+    class EncodeType(enum.Enum):
+        XYZ_8 = 0
+        XYZI_8 = 1
         @staticmethod
         def __members__() -> Dict[str, EnumClassValue]: ...

spconv/core_cc/cumm/common.pyi

 from typing import overload, Any, Callable, Dict, List, Optional, Set, Tuple, Type, Union
-from pccm.stubs import EnumValue, EnumClassValue
+from pccm.stubs import EnumValue, EnumClassValue, enum
 class CompileInfo:
     @staticmethod
     def get_compiled_cuda_version() -> Tuple[int, int]: ...

spconv/core_cc/cumm/conv/main.pyi

 from typing import overload, Any, Callable, Dict, List, Optional, Set, Tuple, Type, Union
-from pccm.stubs import EnumValue, EnumClassValue
+from pccm.stubs import EnumValue, EnumClassValue, enum
 from cumm.tensorview.gemm import ConvParams
 class ConvMainUnitTest:
     @staticmethod

spconv/core_cc/cumm/gemm/main.pyi

 from typing import overload, Any, Callable, Dict, List, Optional, Set, Tuple, Type, Union
-from pccm.stubs import EnumValue, EnumClassValue
+from pccm.stubs import EnumValue, EnumClassValue, enum
 from cumm.tensorview.gemm import GemmAlgoDesp
 from cumm.tensorview.gemm import GemmParams
 class GemmMainUnitTest:

spconv/csrc/utils/pcc.py

@@ -95,7 +95,7 @@ class PointCloudCompress(pccm.Class):
                             enc[3] = uint8_t(inten);
                         }}
                         auto pos_uint = pos_int + hash_t::direct_hash_offset();
-                        uint64_t scalar = hash_t::encode(pos_int[0], pos_int[1], pos_int[2]);
+                        uint64_t scalar = hash_t::encode(pos_uint[0], pos_uint[1], pos_uint[2]);
                         auto iter = hash.find(scalar);
                         if (iter == hash.end()){{
                             auto pos_offset = pos_int.cast<float>() * vsize;
@@ -116,6 +116,7 @@ class PointCloudCompress(pccm.Class):
                             }}
                         }}
                         points_data += point_stride;
+                        intensity_data += inten_stride;
 
                     }}
                     res = tv::empty({{final_size}}, tv::uint8, -1);

spconv/pytorch/functional.py

@@ -41,8 +41,12 @@ _T = TypeVar("_T")
 def identity_decorator(func: _T) -> _T:
     return func
 
+if PYTORCH_VERSION >= [2, 5, 0]:
+    import torch.amp as amp
+    _TORCH_CUSTOM_FWD = amp.custom_fwd(cast_inputs=torch.float16, device_type="cuda")
+    _TORCH_CUSTOM_BWD = amp.custom_bwd(device_type="cuda")
 
-if PYTORCH_VERSION >= [1, 6, 0]:
+elif PYTORCH_VERSION >= [1, 6, 0]:
     import torch.cuda.amp as amp
     _TORCH_CUSTOM_FWD = amp.custom_fwd(cast_inputs=torch.float16)
     _TORCH_CUSTOM_BWD = amp.custom_bwd

tools/build-wheels-dev.sh

@@ -27,24 +27,16 @@ function repair_wheel {
 gcc -v
 export SPCONV_DISABLE_JIT="1"
 export CUMM_CUDA_ARCH_LIST="8.6"
-# export SPCONV_PYTHON_LIST="3.7;3.8;3.9;3.10"
+export SPCONV_PYTHON_LIST="3.13"
 # Compile wheels, we only support 3.6-3.10.
 # "/opt/python/cp36-cp36m/bin/pip" wheel /io/ --no-deps -w /io/wheelhouse_tmp
 
 for PYVER in ${SPCONV_PYTHON_LIST//;/ }
 do
     PYVER2=`echo "$PYVER" | sed 's/\.//'`
-    PYVER_CP="cp$PYVER2-cp$PYVER2"
-    if [ "$PYVER2" = "36" ]; then
-        PYVER_CP="cp$PYVER2-cp${PYVER2}m"
-    fi
-    if [ "$PYVER2" = "37" ]; then
-        PYVER_CP="cp$PYVER2-cp${PYVER2}m"
-    fi
-    if [[ $PYVER2 == *"311"* ]]; then
-        PYVER_CP="cp311-cp311"
-    fi
+    PYVER_T=`echo "$PYVER2" | sed 's/t//'`
 
+    PYVER_CP="cp$PYVER_T-cp$PYVER2"
     "/opt/python/$PYVER_CP/bin/pip" wheel /io/  -v --no-deps -w /io/wheelhouse_tmp
 done
 

tools/build-wheels.sh

@@ -34,17 +34,9 @@ export CUMM_CUDA_ARCH_LIST="all"
 for PYVER in ${SPCONV_PYTHON_LIST//;/ }
 do
     PYVER2=`echo "$PYVER" | sed 's/\.//'`
-    PYVER_CP="cp$PYVER2-cp$PYVER2"
-    if [ "$PYVER2" = "36" ]; then
-        PYVER_CP="cp$PYVER2-cp${PYVER2}m"
-    fi
-    if [ "$PYVER2" = "37" ]; then
-        PYVER_CP="cp$PYVER2-cp${PYVER2}m"
-    fi
-    if [[ $PYVER2 == *"311"* ]]; then
-        PYVER_CP="cp311-cp311"
-    fi
+    PYVER_T=`echo "$PYVER2" | sed 's/\t//'`
 
+    PYVER_CP="cp$PYVER_T-cp$PYVER2"
     "/opt/python/$PYVER_CP/bin/pip" wheel /io/  -v --no-deps -w /io/wheelhouse_tmp
 done
 

tools/install_windows_cuda.ps1

@@ -13,7 +13,13 @@ $CUDA_KNOWN_URLS = @{
     "11.6" = "https://developer.download.nvidia.com/compute/cuda/11.6.2/network_installers/cuda_11.6.2_windows_network.exe";
     "11.7" = "https://developer.download.nvidia.com/compute/cuda/11.7.1/network_installers/cuda_11.7.1_windows_network.exe";
     "11.8" = "https://developer.download.nvidia.com/compute/cuda/11.8.0/network_installers/cuda_11.8.0_windows_network.exe";
-    "12.0" = "https://developer.download.nvidia.com/compute/cuda/12.0.0/network_installers/cuda_12.0.0_windows_network.exe";
+    "12.0" = "https://developer.download.nvidia.com/compute/cuda/12.0.1/network_installers/cuda_12.0.1_windows_network.exe";
+    "12.1" = "https://developer.download.nvidia.com/compute/cuda/12.1.1/network_installers/cuda_12.1.1_windows_network.exe";
+    "12.2" = "https://developer.download.nvidia.com/compute/cuda/12.2.2/network_installers/cuda_12.2.2_windows_network.exe";
+    "12.3" = "https://developer.download.nvidia.com/compute/cuda/12.3.2/network_installers/cuda_12.3.2_windows_network.exe";
+    "12.4" = "https://developer.download.nvidia.com/compute/cuda/12.4.1/network_installers/cuda_12.4.1_windows_network.exe";
+    "12.5" = "https://developer.download.nvidia.com/compute/cuda/12.5.1/network_installers/cuda_12.5.1_windows_network.exe";
+    "12.6" = "https://developer.download.nvidia.com/compute/cuda/12.6.1/network_installers/cuda_12.6.1_windows_network.exe";
 }
 
 # cuda_runtime.h is in nvcc <= 10.2, but cudart >= 11.0
@@ -34,46 +40,28 @@ $CUDA_KNOWN_URLS = @{
 $CUDA_VERSION_FULL = $env:cuda
 # Make sure CUDA_VERSION_FULL is set and valid, otherwise error.
 
-
-if (($CUDA_VERSION_FULL -eq "10.2") -or ($CUDA_VERSION_FULL -eq "11.0") -or ($CUDA_VERSION_FULL -eq "11.1") -or ($CUDA_VERSION_FULL -eq "11.2")){
+if (($CUDA_VERSION_FULL -eq "10.2") -or ($CUDA_VERSION_FULL -eq "11.0") -or ($CUDA_VERSION_FULL -eq "11.1") -or ($CUDA_VERSION_FULL -eq "11.2") -or ($CUDA_VERSION_FULL -eq "11.3")){
     $CUDA_PACKAGES_IN = @(
         "nvcc";
         "visual_studio_integration";
-        "nvrtc_dev";
-        "cudart";
         "curand_dev";
-        # before 11.3, thrust are included by default and no explicit package exists
-    )
-} elseif ($CUDA_VERSION_FULL -eq "11.3"){
-    $CUDA_PACKAGES_IN = @(
-        "nvcc";
-        "visual_studio_integration";
         "nvrtc_dev";
         "cudart";
-        "thrust";
-        "curand_dev";
     )
 } else {
     # after cuda 11.4
     $CUDA_PACKAGES_IN = @(
         "nvcc";
         "visual_studio_integration";
+        "curand_dev";
         "nvrtc_dev";
         "cudart";
-        "thrust"; 
-        "curand_dev";
         "cuxxfilt";
+        # we need this to get libcu++ headers in cuda 12.0.
+        "thrust";
     )
 }
 
-
-## -------------------
-## Select CUDA version
-## -------------------
-
-# Get the cuda version from the environment as env:cuda.
-# Make sure CUDA_VERSION_FULL is set and valid, otherwise error.
-
 # Validate CUDA version, extracting components via regex
 $cuda_ver_matched = $CUDA_VERSION_FULL -match "^(?<major>[1-9][0-9]*)\.(?<minor>[0-9]+)$"
 if(-not $cuda_ver_matched){
@@ -83,7 +71,6 @@ if(-not $cuda_ver_matched){
 $CUDA_MAJOR=$Matches.major
 $CUDA_MINOR=$Matches.minor
 
-
 ## ------------------------------------------------
 ## Select CUDA packages to install from environment
 ## ------------------------------------------------

tools/linux_test_build.sh

+sudo docker run --rm -it -e CUMM_CUDA_VERSION=126 -e PLAT=manylinux_2_28_x86_64 -v `pwd`:/io -v $HOME:/myhome manylinux2014-cuda:cu126-devel bash -c "source /etc/bashrc && /io/tools/build-wheels-dev.sh"
\ No newline at end of file

version.txt

-2.3.6
+2.3.7