CUDA latest 11.8 to support PyTorch 2.0

Fix CUDAARCHS in Windows CI, since no GPU present.
ARM Linux: Fix for "cannot allocate memory in TLS blck"

.github/workflows/windows.yml

@@ -50,6 +50,7 @@ jobs:
             CUDA_VERSION: 11.0.3
     env:
       BUILD_WEBRTC: ${{ ( matrix.BUILD_SHARED_LIBS == 'OFF' && matrix.STATIC_RUNTIME == 'ON' ) && 'ON' || 'OFF' }}
+      CUDAARCHS: 75     # Turing, oldest supported CUDA arch for CUDA v12. Required if no CUDA GPU on build machine.
 
     steps:
       - name: Disk space used

CMakeLists.txt

@@ -510,6 +510,12 @@ macro(add_source_group module_name)
     source_group("Source Files\\Material" FILES ${MODULE_MATERIAL_FILES})
 endmacro()
 
+if (LINUX_AARCH64)
+# Fix for ImportError: ... /pybind.cpython-310-aarch64-linux-gnu.so: cannot allocate memory in static TLS block
+# https://bugs.launchpad.net/ubuntu/+source/mysql-8.0/+bug/1889851
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -ftls-model=global-dynamic")
+endif()
+
 # Include convenience functions
 include(Open3DLink3rdpartyLibraries)
 include(Open3DSetGlobalProperties)

docker/docker_build.sh

@@ -79,7 +79,7 @@ CCACHE_VERSION=4.3
 CMAKE_VERSION=cmake-3.24.4-linux-x86_64
 CMAKE_VERSION_AARCH64=cmake-3.24.4-linux-aarch64
 CUDA_VERSION=11.7.1-cudnn8
-CUDA_VERSION_LATEST=12.1.1-cudnn8
+CUDA_VERSION_LATEST=11.8.0-cudnn8
 
 print_usage_and_exit_docker_build() {
     echo "$__usage_docker_build"