v2.2.4: add prebuilts for cuda 11.6 and 11.8

.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', '3.11.0-rc.2'] 
-        cuda-version: ['10.2', '11.3', '11.4', '11.7']
+        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']
     steps:
       - uses: actions/checkout@master
       - uses: dorny/paths-filter@v2
@@ -115,8 +115,8 @@ jobs:
     runs-on: ubuntu-20.04
     strategy:
       matrix:
-        python-version: ['3.7', '3.8', '3.9', '3.10', '3.11.0-rc.2'] # this version is only used for upload.
-        cuda-version: ['102', '113', '114', '117', '']
+        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']
 
     steps:
       - uses: actions/checkout@master

CHANGELOG.md

 # Changelog
+## [2.2.4] - 2022-10-13
+### Added 
+- Add prebuilt for CUDA 11.8 (RTX 4090 and H100) and CUDA 11.6.
+### Fixed 
+- Fix small bugs
+
 ## [2.2.3] - 2022-9-28
 ### Fixed 
 - Fix missing .contiguous for input feature

README.md

@@ -17,6 +17,8 @@
 [pypi-ver-114]: https://img.shields.io/pypi/v/spconv-cu114
 [pypi-ver-111]: https://img.shields.io/pypi/v/spconv-cu111
 [pypi-ver-117]: https://img.shields.io/pypi/v/spconv-cu117
+[pypi-ver-116]: https://img.shields.io/pypi/v/spconv-cu116
+[pypi-ver-118]: https://img.shields.io/pypi/v/spconv-cu118
 
 [pypi-ver-113]: https://img.shields.io/pypi/v/spconv-cu113
 [pypi-ver-120]: https://img.shields.io/pypi/v/spconv-cu120
@@ -36,9 +38,15 @@
 [pypi-download-120]: https://img.shields.io/pypi/dm/spconv-cu120
 [pypi-url-cpu]: https://pypi.org/project/spconv/
 [pypi-download-cpu]: https://img.shields.io/pypi/dm/spconv
+[pypi-url-118]: https://pypi.org/project/spconv-cu118/
+[pypi-download-118]: https://img.shields.io/pypi/dm/spconv-cu118
+
+[pypi-url-116]: https://pypi.org/project/spconv-cu118/
+[pypi-download-116]: https://img.shields.io/pypi/dm/spconv-cu118
 
 # SpConv: Spatially Sparse Convolution Library
 [![Build Status](https://github.com/traveller59/spconv/workflows/build/badge.svg)](https://github.com/traveller59/spconv/actions?query=workflow%3Abuild) 
+![pypi versions](https://img.shields.io/pypi/pyversions/spconv-cu117)
 
 
 |                | PyPI   | Install  |Downloads  |
@@ -47,7 +55,10 @@
 | CUDA 10.2 | [![PyPI Version][pypi-ver-102]][pypi-url-102] | ```pip install spconv-cu102```| [![pypi monthly download][pypi-download-102]][pypi-url-102]| 
 | CUDA 11.3 | [![PyPI Version][pypi-ver-113]][pypi-url-113] | ```pip install spconv-cu113```| [![pypi monthly download][pypi-download-113]][pypi-url-113]| 
 | CUDA 11.4 | [![PyPI Version][pypi-ver-114]][pypi-url-114] | ```pip install spconv-cu114```| [![pypi monthly download][pypi-download-114]][pypi-url-114]|
+| CUDA 11.6 | [![PyPI Version][pypi-ver-116]][pypi-url-116] | ```pip install spconv-cu116```| [![pypi monthly download][pypi-download-116]][pypi-url-116]|
 | CUDA 11.7 | [![PyPI Version][pypi-ver-117]][pypi-url-117] | ```pip install spconv-cu117```| [![pypi monthly download][pypi-download-117]][pypi-url-117]| 
+| CUDA 11.8 | [![PyPI Version][pypi-ver-118]][pypi-url-118] | ```pip install spconv-cu118```| [![pypi monthly download][pypi-download-118]][pypi-url-118]| 
+
 <!-- | CUDA 12.0 | [![PyPI Version][pypi-ver-120]][pypi-url-120] | ```pip install spconv-cu120```| [![pypi monthly download][pypi-download-120]][pypi-url-120]| -->
 
 ```spconv``` is a project that provide heavily-optimized sparse convolution implementation with tensor core support. check [benchmark](docs/BENCHMARK.md) to see how fast spconv 2.x runs.
@@ -131,7 +142,7 @@ For Linux users, you need to install pip >= 20.3 first to install prebuilt.
 
 **NOTE** It's safe to have different **minor** cuda version between system and conda (pytorch) in **CUDA >= 11.0** because of [CUDA Minor Version Compatibility](https://docs.nvidia.com/deploy/cuda-compatibility/#minor-version-compatibility). For example, you can use spconv-cu114 with anaconda version of pytorch cuda 11.1 in a OS with CUDA 11.2 installed.
 
-**NOTE** In Linux, you can install spconv-cuxxx without install CUDA to system! only suitable NVIDIA driver is required. for CUDA 11, we need driver >= 450.82.
+**NOTE** In Linux, you can install spconv-cuxxx without install CUDA to system! only suitable NVIDIA driver is required. for CUDA 11, we need driver >= 450.82. You may need newer driver if you use newer CUDA. for cuda 11.8, you need to have driver >= 520 installed.
 
 #### Prebuilt GPU Support Matrix
 
@@ -141,8 +152,8 @@ If you use a GPU architecture that isn't compiled in prebuilt, spconv will use N
 
 | CUDA version | GPU Arch List  |
 | -------------- |:---------------------:|
-| 11.x       | 52,60,61,70,75,80,86     | 
-| 12.x       | 70,75,80,86,90     | 
+| 11.1~11.7       | 52,60,61,70,75,80,86     | 
+| 11.8+       | 60,70,75,80,86,89,90     | 
 
 ### Build from source for development (JIT, recommend)
 
@@ -191,6 +202,18 @@ You need to rebuild ```cumm``` first if you are build along a CUDA version that
 5. run ```pip install pccm cumm wheel```
 6. run ```python setup.py bdist_wheel```+```pip install dists/xxx.whl```
 
+## Citation
+
+If you find this project useful in your research, please consider cite:
+
+```latex
+@misc{spconv2022,
+    title={Spconv: Spatially Sparse Convolution Library},
+    author={Spconv Contributors},
+    howpublished = {\url{https://github.com/traveller59/spconv}},
+    year={2022}
+}
+```
 ## Contributers
 
 * [EvernightAurora](https://github.com/EvernightAurora): add ampere feature.

pyproject.toml

 [build-system]
-requires = ["setuptools>=41.0", "wheel", "pccm>=0.4.0", "cumm>=0.3.4"]
+requires = ["setuptools>=41.0", "wheel", "pccm>=0.4.0", "cumm>=0.3.5"]
+# requires = ["setuptools>=41.0", "wheel", "pccm>=0.4.0", "cumm @ file:///io/dist/cumm_cu118-0.3.4-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl"]
 build-backend = "setuptools.build_meta"

setup.py

@@ -249,6 +249,11 @@ setup(
         '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 :: Implementation :: CPython',
     ],
     # $ setup.py publish support.

spconv/csrc/sparse/convops.py

@@ -573,6 +573,7 @@ class GemmTunerSimple(pccm.ParameterizedClass):
             auto ldb = b.stride(0);
             auto ldc = c.stride(0);
             if (desp.supported_ldx(lda, ldb, ldc)){{
+                auto desp2 = desp;
                 if (desp.is_nvrtc){{
                     if (!CompileInfo::algo_can_be_nvrtc_compiled(desp.min_arch)){{
                         continue;
@@ -581,14 +582,13 @@ class GemmTunerSimple(pccm.ParameterizedClass):
                 if (!CompileInfo::arch_is_compiled_gemm(arch)){{
                     if (!CompileInfo::gemm_algo_can_use_ptx(desp.min_arch, arch)){{
                         if (CompileInfo::algo_can_be_nvrtc_compiled(desp.min_arch)){{
-                            auto desp2 = desp;
                             desp2.is_nvrtc = true;
                         }}else{{
                             continue;
                         }}
                     }}
                 }}
-                finally_algos.push_back(desp);
+                finally_algos.push_back(desp2);
             }}
         }}
         std::sort(finally_algos.begin(), finally_algos.end(), [](auto a, auto b){{return a.min_arch > b.min_arch;}});
@@ -1078,6 +1078,7 @@ class ConvTunerSimple(pccm.ParameterizedClass):
                 mask_width_valid = mask_width % desp.tile_shape[2] == 0;
             }}
             if (desp.supported_ldx_conv(ldi, ldw, ldo) && mask_width_valid){{
+                auto desp2 = desp;
                 if (desp.is_nvrtc){{
                     if (!CompileInfo::algo_can_be_nvrtc_compiled(desp.min_arch)){{
                         continue;
@@ -1086,14 +1087,13 @@ class ConvTunerSimple(pccm.ParameterizedClass):
                 if (!CompileInfo::arch_is_compiled_gemm(arch)){{
                     if (!CompileInfo::gemm_algo_can_use_ptx(desp.min_arch, arch)){{
                         if (CompileInfo::algo_can_be_nvrtc_compiled(desp.min_arch)){{
-                            auto desp2 = desp;
                             desp2.is_nvrtc = true;
                         }}else{{
                             continue;
                         }}
                     }}
                 }}
-                finally_algos.push_back(desp);
+                finally_algos.push_back(desp2);
             }}
         }}
         std::sort(finally_algos.begin(), finally_algos.end(), [](auto a, auto b){{return a.min_arch > b.min_arch;}});

tools/install_windows_cuda.ps1

@@ -11,6 +11,7 @@ $CUDA_KNOWN_URLS = @{
     "11.3" = "https://developer.download.nvidia.com/compute/cuda/11.3.1/network_installers/cuda_11.3.1_win10_network.exe";
     "11.4" = "https://developer.download.nvidia.com/compute/cuda/11.4.2/network_installers/cuda_11.4.2_win10_network.exe";
     "11.5" = "https://developer.download.nvidia.com/compute/cuda/11.5.0/network_installers/cuda_11.5.0_win10_network.exe";
+    "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";
 }

version.txt

-2.2.3
+2.2.4