add new prebuilts for 3.13 and cu126

.github/workflows/build.yaml

@@ -15,8 +15,8 @@ jobs:
     runs-on: windows-2019
     strategy:
       matrix:
-        python-version: ['3.7', '3.8', '3.9', '3.10', '3.11'] 
-        cuda-version: ['10.2', '11.3', '11.4', '11.6', '11.7', '11.8', '12.0']
+        python-version: ['3.9', '3.10', '3.11', '3.12', '3.13'] 
+        cuda-version: ['11.4', '11.8', '12.1', '12.4', '12.6']
     steps:
       - uses: actions/checkout@master
       - uses: dorny/paths-filter@v2
@@ -41,7 +41,7 @@ jobs:
             (github.event_name == 'push' && (startsWith(github.ref, 'refs/tags')) ) || 
             (
               (steps.changes.outputs.needbuild == 'true') && 
-              (env.PYTHON_VERSION == '3.10')
+              (env.PYTHON_VERSION == '3.12')
             )
           )
         shell: powershell
@@ -60,7 +60,7 @@ jobs:
             (github.event_name == 'push' && (startsWith(github.ref, 'refs/tags')) ) || 
             (
               (steps.changes.outputs.needbuild == 'true') && 
-              (env.PYTHON_VERSION == '3.10')
+              (env.PYTHON_VERSION == '3.12')
             )
           )
         shell: powershell
@@ -89,7 +89,7 @@ jobs:
             (github.event_name == 'push' && (startsWith(github.ref, 'refs/tags')) ) || 
             (
               (steps.changes.outputs.needbuild == 'true') && 
-              (env.PYTHON_VERSION == '3.10')
+              (env.PYTHON_VERSION == '3.12')
             )
           )
         run: |
@@ -115,8 +115,8 @@ jobs:
     runs-on: ubuntu-20.04
     strategy:
       matrix:
-        python-version: ['3.7', '3.8', '3.9', '3.10', '3.11'] # this version is only used for upload.
-        cuda-version: ['102', '113', '114', '116', '117', '118', '120', '']
+        python-version: ['3.9', '3.10', '3.11', '3.12', '3.13'] # this version is only used for upload.
+        cuda-version: ['114', '118', '121', '124', '126', '']
 
     steps:
       - uses: actions/checkout@master
@@ -160,7 +160,7 @@ jobs:
             (github.event_name == 'push' && (startsWith(github.ref, 'refs/tags')) ) || 
             (
               (steps.changes.outputs.needbuild == 'true') && 
-              (env.PYTHON_VERSION == '3.10')
+              (env.PYTHON_VERSION == '3.12')
             )
           )
         run: |
@@ -174,15 +174,15 @@ jobs:
         env:
           CUDA_VERSION: ${{ matrix.cuda-version }}
           PYTHON_VERSION: ${{ matrix.python-version }}
-          DOCKER_IMAGE: scrin/manylinux2014-cuda:cu120-devel-1.0.0
+          DOCKER_IMAGE: scrin/manylinux2014-cuda:cu126-devel-1.0.0
           PLAT: manylinux2014_x86_64
           BOOST_VERSION: boost_1_77_0
         if: |
           (env.CUDA_VERSION == '') && (
             (github.event_name == 'push' && (startsWith(github.ref, 'refs/tags')) ) || 
             (
               (steps.changes.outputs.needbuild == 'true') && 
-              (env.PYTHON_VERSION == '3.10')
+              (env.PYTHON_VERSION == '3.12')
             )
           )
         run: |

CHANGELOG.md

@@ -1,4 +1,8 @@
 # Changelog
+## [2.3.7] - 2024-12-08
+### Changed
+- add python 3.13 and cuda 12.1/12.4/12.6 prebuilts, drop python 3.7/3.8 prebuilts, change manylinux glibc from 2_17 to 2_28 if cuda >= 12.4, which requires recent OS such as ubuntu 18.10+.
+
 ## [2.3.6] - 2023-04-19
 ### Fixed 
 - Fix a CI bug that cpu cumm and spconv use different gcc compiler, must be same.

pyproject.toml

@@ -1,5 +1,5 @@
 [build-system]
-requires = ["setuptools>=41.0", "wheel", "pccm>=0.4.0", "cumm>=0.4.8"]
+requires = ["setuptools>=41.0", "wheel", "pccm>=0.4.16", "cumm>=0.7.5"]
 # requires = ["setuptools>=41.0", "wheel", "pccm>=0.4.0", "cumm @ file:///io/dist/cumm_cu120-0.4.2-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl"]
-# requires = ["setuptools>=41.0", "wheel", "pccm>=0.4.0", "cumm @ file:///io/dist/cumm_cu117-0.4.6-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl"]
+# requires = ["setuptools>=41.0", "wheel", "pccm>=0.4.0", "cumm-cu126 @ file:///io/dist/cumm_cu126-0.7.3-cp313-cp313-manylinux_2_28_x86_64.whl"]
 build-backend = "setuptools.build_meta"

setup.py

@@ -39,21 +39,21 @@
     cuda_ver_str = cuda_ver.replace(".", "") # 10.2 to 102
 
     RELEASE_NAME += "-cu{}".format(cuda_ver_str)
-    deps = ["cumm-cu{}>=0.4.5, <0.5.0".format(cuda_ver_str)]
+    deps = ["cumm-cu{}>=0.7.5, <0.8.0".format(cuda_ver_str)]
 else:
-    deps = ["cumm>=0.4.5, <0.5.0"]
+    deps = ["cumm>=0.7.5, <0.8.0"]
 
 
 
 DESCRIPTION = 'spatial sparse convolution'
 URL = 'https://github.com/traveller59/spconv'
 EMAIL = 'yanyan.sub@outlook.com'
 AUTHOR = 'Yan Yan'
-REQUIRES_PYTHON = '>=3.7'
+REQUIRES_PYTHON = '>=3.9'
 VERSION = None
 
 # What packages are required for this module to be executed?
-REQUIRED = ["pccm>=0.4.0", "ccimport>=0.4.0", "pybind11>=2.6.0", "fire", "numpy", *deps]
+REQUIRED = ["pccm>=0.4.16", "ccimport>=0.4.4", "pybind11>=2.6.0", "fire", "numpy", *deps]
 
 # What packages are optional?
 EXTRAS = {
@@ -249,11 +249,11 @@ def run(self):
         'License :: OSI Approved :: Apache Software License',
         'Programming Language :: Python',
         'Programming Language :: Python :: 3',
-        'Programming Language :: Python :: 3.7',
-        'Programming Language :: Python :: 3.8',
         'Programming Language :: Python :: 3.9',
         'Programming Language :: Python :: 3.10',
         'Programming Language :: Python :: 3.11',
+        'Programming Language :: Python :: 3.12',
+        'Programming Language :: Python :: 3.13',
         'Programming Language :: Python :: Implementation :: CPython',
     ],
     # $ setup.py publish support.

spconv/core_cc/csrc/hash/core.pyi

@@ -1,5 +1,5 @@
 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 HashTable:
     key_itemsize: int

spconv/core_cc/csrc/sparse/all/__init__.pyi

@@ -1,5 +1,5 @@
 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 import CUDAKernelTimer
 class ThrustCustomAllocatorV2:

spconv/core_cc/csrc/sparse/all/ops1d.pyi

@@ -1,5 +1,5 @@
 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 Point2Voxel:
     hashdata: Tensor

spconv/core_cc/csrc/sparse/all/ops2d.pyi

@@ -1,5 +1,5 @@
 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 Point2Voxel:
     hashdata: Tensor

spconv/core_cc/csrc/sparse/all/ops3d.pyi

@@ -1,5 +1,5 @@
 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 Point2Voxel:
     hashdata: Tensor

spconv/core_cc/csrc/sparse/all/ops4d.pyi

@@ -1,5 +1,5 @@
 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 Point2Voxel:
     hashdata: Tensor

spconv/core_cc/csrc/sparse/all/ops_cpu1d.pyi

@@ -1,5 +1,5 @@
 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 Point2VoxelCPU:
     densehashdata: Tensor

spconv/core_cc/csrc/sparse/all/ops_cpu2d.pyi

@@ -1,5 +1,5 @@
 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 Point2VoxelCPU:
     densehashdata: Tensor

spconv/core_cc/csrc/sparse/all/ops_cpu3d.pyi

@@ -1,5 +1,5 @@
 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 Point2VoxelCPU:
     densehashdata: Tensor

spconv/core_cc/csrc/sparse/all/ops_cpu4d.pyi

@@ -1,5 +1,5 @@
 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 Point2VoxelCPU:
     densehashdata: Tensor

spconv/core_cc/csrc/sparse/alloc.pyi

@@ -1,5 +1,5 @@
 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 ExternalAllocator:
     def zeros(self, name: str, shape: List[int], dtype: int, device: int, stream: int = 0, is_temp_memory: bool = False, scale: float = 1.0) -> Tensor: 

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

@@ -1,5 +1,5 @@
 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 ConvAlgoDesp
 from cumm.tensorview import Tensor

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

@@ -1,5 +1,5 @@
 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 ConvAlgoDesp
 from cumm.tensorview import Tensor
 from cumm.tensorview.gemm import NVRTCParams

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

@@ -1,5 +1,5 @@
 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 import Tensor
 from cumm.tensorview.gemm import NVRTCParams

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

@@ -1,5 +1,5 @@
 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
 from cumm.tensorview import CUDAKernelTimer

spconv/core_cc/csrc/sparse/inference.pyi

@@ -1,5 +1,5 @@
 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

@@ -1,5 +1,5 @@
 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

@@ -1,5 +1,5 @@
 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

@@ -1,5 +1,5 @@
 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

@@ -1,5 +1,5 @@
 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

@@ -1,5 +1,5 @@
 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 @@ def encode_with_order(self):
                             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 @@ def encode_with_order(self):
                             }}
                         }}
                         points_data += point_stride;
+                        intensity_data += inten_stride;
 
                     }}
                     res = tv::empty({{final_size}}, tv::uint8, -1);

spconv/pytorch/functional.py

@@ -41,8 +41,12 @@
 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