1.9.1 Release

[SergioRAgostinho]

Task list

• Go through all PRs for the milestone
• Update Changelog
• Start a release branch
• (If needed) Revert whatever unintentional API/ABI breakage that might have occurred in the release branch
• Bump version to x.x.x in CMakeLists.txt
• Bump version off the release badge in README.md
• Find replace all occurrences of x.x.x.99 CMake requirements in the tutorials and examples
• Try to build and test on all relevant platforms
• Finalize changelog and fix the release date
• Tag
• Remove the release branch
• Add binaries to the GitHub release page.
• Add link doc link to tutorials page
• Bump version to x.x.x-dev in CMakeLists.txt
• Send announcements to the mailing lists regarding 1.9.0 and 1.9.1
• Throw two parties: 1.9.0 and 1.9.1

[taketwo]

I propose to include #2612 in this release.

[SergioRAgostinho]

I've created the milestone and added it.

[SergioRAgostinho]

@claudiofantacci you are able to build with GPU on mac right? You haven't seen these before?

[  1%] Building CXX object gpu/containers/CMakeFiles/pcl_gpu_containers.dir/src/device_memory.cpp.o
nvcc fatal   : Unsupported gpu architecture 'compute_72'
nvcc fatal   : Unsupported gpu architecture 'compute_72'
CMake Error at pcl_cuda_features_generated_normal_3d.cu.o.RelWithDebInfo.cmake:219 (message):
  Error generating
  /Users/neglective/Development/3rdparty/pcl/build/cuda/features/CMakeFiles/pcl_cuda_features.dir/src/./pcl_cuda_features_generated_normal_3d.cu.o


CMake Error at pcl_cuda_segmentation_generated_connected_components.cu.o.RelWithDebInfo.cmake:219 (message):
  Error generating
  /Users/neglective/Development/3rdparty/pcl/build/cuda/segmentation/CMakeFiles/pcl_cuda_segmentation.dir/src/./pcl_cuda_segmentation_generated_connected_components.cu.o


make[2]: *** [cuda/features/CMakeFiles/pcl_cuda_features.dir/src/pcl_cuda_features_generated_normal_3d.cu.o] Error 1
make[2]: *** [cuda/segmentation/CMakeFiles/pcl_cuda_segmentation.dir/src/pcl_cuda_segmentation_generated_connected_components.cu.o] Error 1
make[1]: *** [cuda/features/CMakeFiles/pcl_cuda_features.dir/all] Error 2
make[1]: *** Waiting for unfinished jobs....
make[1]: *** [cuda/segmentation/CMakeFiles/pcl_cuda_segmentation.dir/all] Error 2

[claudiofantacci]

@SergioRAgostinho yes, I have that too. I did not notice this before because I always manually set CUDA_ARCH_BIN to my local architecture that is 3.0. I don't understand why that error occurs, the most reasonable reason is that our version of nvcc simply does not support it. I'm still investigating this.

[claudiofantacci]

I updated my answer.

[SergioRAgostinho]

So basically this..

diff --git a/cmake/pcl_find_cuda.cmake b/cmake/pcl_find_cuda.cmake
index ee70fdb43..f6daee7d1 100644
--- a/cmake/pcl_find_cuda.cmake
+++ b/cmake/pcl_find_cuda.cmake
@@ -45,9 +45,17 @@ if(CUDA_FOUND)
   if(NOT ${CUDA_VERSION_STRING} VERSION_LESS "10.0")
     set(__cuda_arch_bin "3.0 3.5 5.0 5.2 5.3 6.0 6.1 7.0 7.2 7.5")
   elseif(NOT ${CUDA_VERSION_STRING} VERSION_LESS "9.1")
-    set(__cuda_arch_bin "3.0 3.5 5.0 5.2 5.3 6.0 6.1 7.0 7.2")
+    if(NOT APPLE)
+      set(__cuda_arch_bin "3.0 3.5 5.0 5.2 5.3 6.0 6.1 7.0 7.2")
+    else()
+      set(__cuda_arch_bin "3.0 3.5 5.0 5.2 5.3 6.0 6.1")
+    endif()
   elseif(NOT ${CUDA_VERSION_STRING} VERSION_LESS "9.0")
-    set(__cuda_arch_bin "3.0 3.5 5.0 5.2 5.3 6.0 6.1 7.0")  
+    if(NOT APPLE)
+      set(__cuda_arch_bin "3.0 3.5 5.0 5.2 5.3 6.0 6.1 7.0")
+    else()
+      set(__cuda_arch_bin "3.0 3.5 5.0 5.2 5.3 6.0 6.1")
+    endif()
   elseif(NOT ${CUDA_VERSION_STRING} VERSION_LESS "8.0")
     set(__cuda_arch_bin "2.0 2.1(2.0) 3.0 3.5 5.0 5.2 5.3 6.0 6.1")
   elseif(NOT ${CUDA_VERSION_STRING} VERSION_LESS "6.5")

[claudiofantacci]

Yes, but I’m not sure that we are clearly understand the issue. I fear there is something I don’t understand here. Also, I have a CMake procedure to automatically set the cuda architecture based on the host GPU. Do you think it might be interesting for PCL?

[taketwo]

Do you think it might be interesting for PCL?

I'd suggest we look at the advanced stuff that OpenCV has.

[SergioRAgostinho]

Thanks for the input, both. This is going to be more involved than I thought. Probably not a good idea to target it for 1.9.1.

[claudiofantacci]

Do you think it might be interesting for PCL?

I'd suggest we look at the advanced stuff that OpenCV has.

Yes, we can have a look into it. First thing to notice by just CTRL+F-ing is that architecture 7.2 is not present. That is very suspicious and it is worth investigating. I can have a look at what OpenCV does and let you know.

[SergioRAgostinho]

compat_reports.zip

The compat reports. Nothing changed as expected.

[UnaNancyOwen]

I was tested this current release candidate #2631. It is occurred an error in CMake.
Perhaps, I think the policy change is affecting this.

CMake Error at cmake/pcl_utils.cmake:63 (if):
  if given arguments:

    "EQUAL" "99"

  Unknown arguments specified

It seems to be solved by enclosing variable with double quotes.

- if(${PCL_VERSION_TWEAK} EQUAL 99)
+ if("${PCL_VERSION_TWEAK}" EQUAL 99)

Do you have any ideas?

[SergioRAgostinho]

Yep. I just pushed the fix to the release branch PR.

[claudiofantacci]

@SergioRAgostinho here is the Volta support guidelines. Basically there is no support to compute_72 at all. Instead of adding an exception for APPLE I would remove compute_72 altogether.

[SergioRAgostinho]

This is only for cuda 9.0. That particular branch doesn't have compute_72.

For 9.1, seems to not exist as well
https://docs.nvidia.com/cuda/archive/9.1/cuda-compiler-driver-nvcc/index.html#virtual-architecture-feature-list

For 9.2 the same
https://docs.nvidia.com/cuda/archive/9.2/cuda-compiler-driver-nvcc/index.html#virtual-architecture-feature-list

@PointCloudLibrary/maintainers we should probably remove them before the patch release and have people test the RC.

If believe I will also push separate PRs, to the fixes I included in #2631, so that they show up on the change log properly.