Merge pull request #89 from koide3/cuda13

koide3 · web-flow · commit c4af3d161061 · 2026-01-05T02:00:19.000+09:00
CUDA13
diff --git a/.github/workflows/build.yml b/.github/workflows/build.yml
@@ -17,13 +17,15 @@ jobs:
     strategy:
       matrix:
         DISTRO: [
-          { "ubuntu": "noble_gtsam4.3a0", "suffix": "gcc" },
-          { "ubuntu": "noble_gtsam4.3a0", "suffix": "llvm" },
-          { "ubuntu": "noble_cuda12.5_gtsam4.3a0", "suffix": "gcc.cuda" },
-          { "ubuntu": "noble_cuda12.5_gtsam4.3a0", "suffix": "llvm.cuda" },
-          { "ubuntu": "jammy_gtsam4.3a0", "suffix": "gcc" },
-          { "ubuntu": "jammy_cuda12.2_gtsam4.3a0", "suffix": "gcc.cuda" },
-          { "ubuntu": "jammy_cuda12.5_gtsam4.3a0", "suffix": "gcc.cuda" },
+          { "tag": "jammy_gtsam4.3a0", "base_image": "ubuntu:jammy", "suffix": "gcc" },
+          { "tag": "jammy_gtsam4.3a0", "base_image": "ubuntu:jammy", "suffix": "llvm" },
+          { "tag": "jammy_cuda12.2_gtsam4.3a0", "base_image": "nvidia/cuda:12.2.2-devel-ubuntu22.04", "suffix": "gcc.cuda" },
+          { "tag": "jammy_cuda12.5_gtsam4.3a0", "base_image": "nvidia/cuda:12.5.1-devel-ubuntu22.04", "suffix": "gcc.cuda" },
+          { "tag": "jammy_cuda13.1_gtsam4.3a0", "base_image": "nvidia/cuda:13.1.0-devel-ubuntu22.04", "suffix": "gcc.cuda" },
+          { "tag": "noble_gtsam4.3a0", "base_image": "ubuntu:noble", "suffix": "gcc" },
+          { "tag": "noble_gtsam4.3a0", "base_image": "ubuntu:noble", "suffix": "llvm" },
+          { "tag": "noble_cuda12.5_gtsam4.3a0", "base_image": "nvidia/cuda:12.5.1-devel-ubuntu24.04", "suffix": "gcc.cuda" },
+          { "tag": "noble_cuda13.1_gtsam4.3a0", "base_image": "nvidia/cuda:13.1.0-devel-ubuntu24.04", "suffix": "gcc.cuda" },
         ]
         BUILD_WITH_TBB: [ "ON", "OFF" ]
 
@@ -42,13 +44,13 @@ jobs:
       - name: Docker build
         uses: docker/build-push-action@v6
         with:
-          tags: koide3/gtsam_points:${{ matrix.DISTRO.ubuntu }}
+          tags: koide3/gtsam_points:${{ matrix.DISTRO.tag }}
           file: ${{github.workspace}}/docker/ubuntu/Dockerfile.${{ matrix.DISTRO.suffix }}
           build-args: |
-            BASE_IMAGE=koide3/gtsam_docker:${{ matrix.DISTRO.ubuntu }}
+            BASE_IMAGE=${{ matrix.DISTRO.base_image }}
             BUILD_WITH_TBB=${{ matrix.BUILD_WITH_TBB }}
           context: .
-          push: ${{ (github.event_name == 'push' || github.event_name == 'workflow_dispatch') && (matrix.BUILD_WITH_TBB == 'ON') && (contains(matrix.DISTRO.suffix, 'gcc') || contains(matrix.DISTRO.suffix, 'focal')) }}
+          push: ${{ (github.event_name == 'push' || github.event_name == 'workflow_dispatch') && (matrix.BUILD_WITH_TBB == 'OFF') && (contains(matrix.DISTRO.suffix, 'gcc') || contains(matrix.DISTRO.suffix, 'focal')) }}
 
   dispatch:
     needs: build
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -65,8 +65,8 @@ if(BUILD_WITH_CUDA)
       set(CMAKE_CUDA_ARCHITECTURES 75 80 87 89 90 CACHE STRING "CUDA architectures" FORCE)
     else()
     if(NOT DEFINED CMAKE_CUDA_ARCHITECTURES OR CMAKE_CUDA_ARCHITECTURES STREQUAL "52")
-        message(WARNING "CMAKE_CUDA_ARCHITECTURES is not defined. Set to 80 by default")
-        set(CMAKE_CUDA_ARCHITECTURES 53 62 80 87 89 90 CACHE STRING "CUDA architectures" FORCE)
+        message(WARNING "CMAKE_CUDA_ARCHITECTURES is not defined. Set to defaults")
+        set(CMAKE_CUDA_ARCHITECTURES 75 80 87 89 90 CACHE STRING "CUDA architectures" FORCE)
       endif()
     endif()
   endif()
diff --git a/docker/ubuntu/Dockerfile.gcc b/docker/ubuntu/Dockerfile.gcc
@@ -1,8 +1,22 @@
-ARG BASE_IMAGE=koide3/gtsam_docker:jammy
-
+ARG BASE_IMAGE=ubuntu:jammy
 FROM ${BASE_IMAGE}
 
-ARG BUILD_WITH_TBB=ON
+ENV DEBIAN_FRONTEND=noninteractive
+ENV LD_LIBRARY_PATH=/usr/local/lib:$LD_LIBRARY_PATH
+
+RUN apt-get update \
+    && apt-get install -y --no-install-recommends \
+       cmake build-essential gpg-agent software-properties-common \
+       libboost-system-dev libboost-timer-dev libboost-thread-dev libboost-filesystem-dev libboost-graph-dev libboost-program-options-dev \
+       libeigen3-dev libomp-dev \
+       clang lld ca-certificates \
+       libtbb-dev libgtest-dev libmetis-dev \
+       libglm-dev libglfw3-dev libpng-dev libjpeg-dev
+
+RUN add-apt-repository ppa:koide3/iridescence -y && \
+    add-apt-repository ppa:koide3/gtsam -y && \
+    apt-get update && \
+    apt-get install -y libiridescence-dev libgtsam-no-tbb-dev=4.3.0-1*
 
 COPY . /root/gtsam_points
 WORKDIR /root/gtsam_points/build
@@ -15,9 +29,8 @@ RUN cmake .. \
   -DBUILD_WITH_TBB=${BUILD_WITH_TBB} \
   -DBUILD_WITH_CUDA=OFF \
   -DCMAKE_BUILD_TYPE=Release && \
-  make -j$(nproc) && \
-  make test && \
-  make install && \
-  rm -rf /root/gtsam_points/build
+  make -j$(nproc)
+RUN make test
+RUN make install
 
 CMD ["bash"]
diff --git a/docker/ubuntu/Dockerfile.gcc.cuda b/docker/ubuntu/Dockerfile.gcc.cuda
@@ -1,8 +1,22 @@
-ARG BASE_IMAGE=koide3/gtsam_docker:jammy_cuda12.2
-
+ARG BASE_IMAGE=nvidia/cuda:12.2.2-devel-ubuntu22.04
 FROM ${BASE_IMAGE}
 
-ARG BUILD_WITH_TBB=ON
+ENV DEBIAN_FRONTEND=noninteractive
+ENV LD_LIBRARY_PATH=/usr/local/lib:$LD_LIBRARY_PATH
+
+RUN apt-get update \
+    && apt-get install -y --no-install-recommends \
+       cmake build-essential gpg-agent software-properties-common \
+       libboost-system-dev libboost-timer-dev libboost-thread-dev libboost-filesystem-dev libboost-graph-dev libboost-program-options-dev \
+       libeigen3-dev libomp-dev \
+       clang lld ca-certificates \
+       libtbb-dev libgtest-dev libmetis-dev \
+       libglm-dev libglfw3-dev libpng-dev libjpeg-dev
+
+RUN add-apt-repository ppa:koide3/iridescence -y && \
+    add-apt-repository ppa:koide3/gtsam -y && \
+    apt-get update && \
+    apt-get install -y libiridescence-dev libgtsam-no-tbb-dev=4.3.0-1*
 
 COPY . /root/gtsam_points
 WORKDIR /root/gtsam_points/build
@@ -14,10 +28,8 @@ RUN cmake .. \
   -DBUILD_TOOLS=ON \
   -DBUILD_WITH_TBB=${BUILD_WITH_TBB} \
   -DBUILD_WITH_CUDA=ON \
-  -DBUILD_WITH_CUDA_MULTIARCH=ON \
   -DCMAKE_BUILD_TYPE=Release && \
-  make -j$(nproc) && \
-  make install && \
-  rm -rf /root/gtsam_points/build
+  make -j$(nproc)
+RUN make install
 
 CMD ["bash"]
diff --git a/docker/ubuntu/Dockerfile.llvm b/docker/ubuntu/Dockerfile.llvm
@@ -1,10 +1,22 @@
-ARG BASE_IMAGE=koide3/gtsam_docker:jammy
-
+ARG BASE_IMAGE=ubuntu:jammy
 FROM ${BASE_IMAGE}
 
-ARG BUILD_WITH_TBB=ON
+ENV DEBIAN_FRONTEND=noninteractive
+ENV LD_LIBRARY_PATH=/usr/local/lib:$LD_LIBRARY_PATH
+
+RUN apt-get update \
+    && apt-get install -y --no-install-recommends \
+       cmake build-essential gpg-agent software-properties-common \
+       libboost-system-dev libboost-timer-dev libboost-thread-dev libboost-filesystem-dev libboost-graph-dev libboost-program-options-dev \
+       libeigen3-dev libomp-dev \
+       clang lld ca-certificates \
+       libtbb-dev libgtest-dev libmetis-dev \
+       libglm-dev libglfw3-dev libpng-dev libjpeg-dev
 
-RUN update-alternatives --install /usr/bin/ld ld /usr/bin/ld.lld 50
+RUN add-apt-repository ppa:koide3/iridescence -y && \
+    add-apt-repository ppa:koide3/gtsam -y && \
+    apt-get update && \
+    apt-get install -y libiridescence-dev libgtsam-no-tbb-dev=4.3.0-1*
 
 COPY . /root/gtsam_points
 WORKDIR /root/gtsam_points/build
@@ -18,9 +30,8 @@ RUN CC=clang CXX=clang++ \
   -DBUILD_WITH_TBB=${BUILD_WITH_TBB} \
   -DBUILD_WITH_CUDA=OFF \
   -DCMAKE_BUILD_TYPE=Release && \
-  make -j$(nproc) && \
-  make test && \
-  make install && \
-  rm -rf /root/gtsam_points/build
+  make -j$(nproc)
+RUN make test
+RUN make install
 
 CMD ["bash"]
diff --git a/docker/ubuntu/Dockerfile.llvm.cuda b/docker/ubuntu/Dockerfile.llvm.cuda
@@ -1,10 +1,22 @@
-ARG BASE_IMAGE=koide3/gtsam_docker:jammy_cuda12.2
-
+ARG BASE_IMAGE=nvidia/cuda:12.2.2-devel-ubuntu22.04
 FROM ${BASE_IMAGE}
 
-ARG BUILD_WITH_TBB=ON
+ENV DEBIAN_FRONTEND=noninteractive
+ENV LD_LIBRARY_PATH=/usr/local/lib:$LD_LIBRARY_PATH
+
+RUN apt-get update \
+    && apt-get install -y --no-install-recommends \
+       cmake build-essential gpg-agent software-properties-common \
+       libboost-system-dev libboost-timer-dev libboost-thread-dev libboost-filesystem-dev libboost-graph-dev libboost-program-options-dev \
+       libeigen3-dev libomp-dev \
+       clang lld ca-certificates \
+       libtbb-dev libgtest-dev libmetis-dev \
+       libglm-dev libglfw3-dev libpng-dev libjpeg-dev
 
-RUN update-alternatives --install /usr/bin/ld ld /usr/bin/ld.lld 50
+RUN add-apt-repository ppa:koide3/iridescence -y && \
+    add-apt-repository ppa:koide3/gtsam -y && \
+    apt-get update && \
+    apt-get install -y libiridescence-dev libgtsam-no-tbb-dev=4.3.0-1*
 
 COPY . /root/gtsam_points
 WORKDIR /root/gtsam_points/build
@@ -17,10 +29,8 @@ RUN CC=clang CXX=clang++ \
   -DBUILD_TOOLS=ON \
   -DBUILD_WITH_TBB=${BUILD_WITH_TBB} \
   -DBUILD_WITH_CUDA=ON \
-  -DBUILD_WITH_CUDA_MULTIARCH=ON \
   -DCMAKE_BUILD_TYPE=Release && \
-  make -j$(nproc) && \
-  make install && \
-  rm -rf /root/gtsam_points/build
+  make -j$(nproc)
+RUN make install
 
 CMD ["bash"]
diff --git a/src/gtsam_points/cuda/cuda_graph.cu b/src/gtsam_points/cuda/cuda_graph.cu
@@ -1,6 +1,7 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2021  Kenji Koide (k.koide@aist.go.jp)
 
+#include <cuda.h>
 #include <gtsam_points/cuda/cuda_graph.cuh>
 
 namespace gtsam_points {
@@ -14,7 +15,11 @@ CUDAGraph::~CUDAGraph() {
 }
 
 void CUDAGraph::add_dependency(CUgraphNode_st* from, CUgraphNode_st* to) {
+#if CUDA_VERSION >= 13000
+  check_error << cudaGraphAddDependencies(graph, &from, &to, nullptr, 1);
+#else
   check_error << cudaGraphAddDependencies(graph, &from, &to, 1);
+#endif
 }
 
 std::shared_ptr<CUDAGraphExec> CUDAGraph::instantiate() {
diff --git a/src/gtsam_points/factors/integrated_vgicp_derivatives_compute.cu b/src/gtsam_points/factors/integrated_vgicp_derivatives_compute.cu
@@ -8,7 +8,6 @@
 #include <thrust/iterator/transform_iterator.h>
 
 #include <cub/device/device_reduce.cuh>
-#include <cub/iterator/transform_input_iterator.cuh>
 
 #include <gtsam_points/cuda/kernels/pose.cuh>
 #include <gtsam_points/cuda/kernels/untie.cuh>
@@ -24,10 +23,10 @@ namespace gtsam_points {
 void IntegratedVGICPDerivatives::issue_compute_error(const Eigen::Isometry3f* d_xl, const Eigen::Isometry3f* d_xe, float* d_output) {
   //
   lookup_voxels_kernel corr_kernel(enable_surface_validation, *target, source->points_gpu, source->normals_gpu, d_xl);
-  cub::TransformInputIterator<thrust::pair<int, int>, lookup_voxels_kernel, int*> corr_first(source_inliers, corr_kernel);
+  auto corr_first = thrust::make_transform_iterator(source_inliers, corr_kernel);
 
   vgicp_error_kernel error_kernel(d_xl, d_xe, *target, source->points_gpu, source->covs_gpu);
-  cub::TransformInputIterator<float, vgicp_error_kernel, decltype(corr_first)> first(corr_first, error_kernel);
+  auto first = thrust::make_transform_iterator(corr_first, error_kernel);
 
   void* temp_storage = nullptr;
   size_t temp_storage_bytes = 0;
diff --git a/src/gtsam_points/factors/integrated_vgicp_derivatives_inliers.cu b/src/gtsam_points/factors/integrated_vgicp_derivatives_inliers.cu
@@ -11,7 +11,6 @@
 
 #include <cub/device/device_reduce.cuh>
 #include <cub/device/device_select.cuh>
-#include <cub/iterator/transform_input_iterator.cuh>
 
 #include <gtsam_points/cuda/check_error.cuh>
 #include <gtsam_points/cuda/kernels/pose.cuh>
diff --git a/src/gtsam_points/factors/integrated_vgicp_derivatives_linearize.cu b/src/gtsam_points/factors/integrated_vgicp_derivatives_linearize.cu
@@ -8,7 +8,6 @@
 #include <thrust/iterator/transform_iterator.h>
 
 #include <cub/device/device_reduce.cuh>
-#include <cub/iterator/transform_input_iterator.cuh>
 
 #include <gtsam_points/cuda/kernels/pose.cuh>
 #include <gtsam_points/cuda/kernels/untie.cuh>
@@ -24,10 +23,10 @@ namespace gtsam_points {
 void IntegratedVGICPDerivatives::issue_linearize(const Eigen::Isometry3f* d_x, LinearizedSystem6* d_output) {
   //
   lookup_voxels_kernel corr_kernel(enable_surface_validation, *target, source->points_gpu, source->normals_gpu, d_x);
-  cub::TransformInputIterator<thrust::pair<int, int>, lookup_voxels_kernel, int*> corr_first(source_inliers, corr_kernel);
+  auto corr_first = thrust::make_transform_iterator(source_inliers, corr_kernel);
 
   vgicp_derivatives_kernel deriv_kernel(d_x, *target, source->points_gpu, source->covs_gpu);
-  cub::TransformInputIterator<LinearizedSystem6, vgicp_derivatives_kernel, decltype(corr_first)> first(corr_first, deriv_kernel);
+  auto first = thrust::make_transform_iterator(corr_first, deriv_kernel);
 
   void* temp_storage = nullptr;
   size_t temp_storage_bytes = 0;
diff --git a/src/gtsam_points/types/gaussian_voxelmap_gpu_funcs.cu b/src/gtsam_points/types/gaussian_voxelmap_gpu_funcs.cu
@@ -6,7 +6,6 @@
 #include <thrust/device_vector.h>
 
 #include <cub/device/device_reduce.cuh>
-#include <cub/iterator/transform_input_iterator.cuh>
 
 #include <gtsam_points/types/point_cloud_gpu.hpp>
 #include <gtsam_points/types/gaussian_voxelmap_cpu.hpp>
