Remove CUDA 11 references #905

jakirkham · 2025-07-16T01:39:57Z

Fixes #904

Removes remaining CUDA 11 references.

copy-pr-bot · 2025-07-16T01:40:01Z

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

jakirkham · 2025-07-16T01:40:59Z

/ok to test

jakirkham · 2025-07-16T01:41:39Z

Also seeing some references in the notebooks

cucim/notebooks/Accessing_File_with_GDS.ipynb

Lines 30 to 38 in 44c5592

    
           "!conda install -c pytorch -c conda-forge pytorch==1.7.1 torchvision==0.8.2 torchaudio==0.7.2 cudatoolkit=11.0\n", 
        
           "(If executing `import torch; torch.cuda.is_available()` doesn't show 'True', use `pip` PyTorch installation method.)\n", 
        
           "```\n", 
        
           "\n", 
        
           "\n", 
        
           "or\n", 
        
           "```\n", 
        
           "!pip install cupy-cuda110\n", 
        
           "!pip install torch==1.7.1+cu110 torchvision==0.8.2+cu110 torchaudio==0.7.2 -f https://download.pytorch.org/whl/torch_stable.html\n",

https://github.com/rapidsai/cucim/blob/44c559284714c79c8a5af2a90212106f6b2a89f6/notebooks/Basic_Usage.ipynb#L25

Though have not addressed those

jakirkham · 2025-07-16T01:45:16Z

/ok to test

jakirkham · 2025-07-16T01:53:40Z

/ok to test

jakirkham · 2025-07-16T01:57:57Z

/ok to test

jakirkham · 2025-07-16T01:59:18Z

/ok to test

jakirkham · 2025-07-16T02:12:17Z

/ok to test

jakirkham · 2025-07-16T02:21:17Z

/ok to test

jakirkham · 2025-07-16T02:24:14Z

/ok to test

jakirkham · 2025-07-16T04:02:22Z

/ok to test

jakirkham · 2025-07-16T06:08:55Z

/ok to test

gds/CMakeLists.txt

grlee77 · 2025-08-20T16:21:42Z

/ok to test

bdice · 2025-08-20T16:33:06Z

/ok to test

grlee77 · 2025-08-20T17:13:04Z

Some notebooks also have outdated pip install commands using cuda 11 versions of CuPy. It looks like we should replace cupy-cuda110 with cupy-cuda12x in the following places:

notebooks/Basic_Usage.ipynb:25:    "!pip install numpy scipy scikit-image cupy-cuda110  # for cucim dependency (assuming that CUDA 11.0 is used for CuPy)\n",
notebooks/Basic_Usage.ipynb:40:    "# !pip install numpy scipy scikit-image cupy-cuda110  # for cucim dependency (assuming that CUDA 11.0 is used for CuPy)"
notebooks/Accessing_File_with_GDS.ipynb:37:    "!pip install cupy-cuda110\n",
notebooks/Accessing_File_with_GDS.ipynb:52:    "#!pip install cupy-cuda110\n",
notebooks/Working_with_DALI.ipynb:23:    "!pip install --extra-index-url https://developer.download.nvidia.com/compute/redist --upgrade nvidia-dali-cuda110\n",
notebooks/Working_with_DALI.ipynb:27:    "!pip install cucim scipy scikit-image cupy-cuda110 matplotlib\n",
notebooks/Working_with_DALI.ipynb:28:    "!pip install --extra-index-url https://developer.download.nvidia.com/compute/redist --upgrade nvidia-dali-cuda110\n",
notebooks/Working_with_DALI.ipynb:39:    "#!pip install --extra-index-url https://developer.download.nvidia.com/compute/redist --upgrade nvidia-dali-cuda110\n",
notebooks/Working_with_DALI.ipynb:43:    "#!pip install cucim scipy scikit-image cupy-cuda110 matplotlib\n",
notebooks/Working_with_DALI.ipynb:45:    "#!pip install --extra-index-url https://developer.download.nvidia.com/compute/redist --upgrade nvidia-dali-cuda110"

grlee77

Thanks @jakirkham, it looks good to me.

I pushed a 1-line commit that seems to have resolved finding the CUDA include path for GDS. Other changes look good to me, but @gigony, please confirm if you have a chance as you are the original author of most of the modified files.

I left a couple of comments for additional minor residual cleanups that could be made.

gds/CMakeLists.txt

jameslamb

Looks good to me, this simplification will help with CUDA 13 support (#926).

jameslamb · 2025-08-20T23:10:02Z

After handling culibos being removed from cuda-cudart, I didn't face any other issues adding CUDA 13 support: #926

If we can merge this soon, I'll update the CUDA 13 PR and ensure everything is still working together. If more work is needed here, we should merge the CUDA 13 PR first.

Co-authored-by: Bradley Dice <bdice@bradleydice.com>

grlee77 · 2025-08-21T00:19:06Z

I was able to push to John's branch, so just pushed the notebook commit here. Should be able to just merge once CI passes

bdice · 2025-08-21T00:50:10Z

Great. I'll do that now, to help along CUDA 13 work.

bdice · 2025-08-21T00:50:13Z

/merge

jakirkham

Just wanted to say thank you all for pushing this across the line ❤️

Had the sense this got 80-90% of the way during 25.08. However wasn't able to finish in that cycle so just split out the doc fixes and moved this to 25.10

There were some remaining things that I hoped to do here. Mainly fixing CI, cleaning up more CUDA 11 references, and some build fixes for issues I caught along the way

Think the changes that all of you made are good enough for this to be in. Will leave some comments here to discuss the remaining work that can be split out. Also to leave some context (especially for Gigon when he looks at this)

jakirkham · 2025-08-21T01:37:08Z

cpp/plugins/cucim.kit.cuslide/CMakeLists.txt

+if (TARGET CUDA::nvjpeg_static)
+        target_link_libraries(${CUCIM_PLUGIN_NAME}
+            PRIVATE
+		# Add nvjpeg before cudart so that nvjpeg.h in static library takes precedence.
+                CUDA::nvjpeg_static
+                # Add CUDA::culibos to link necessary methods for 'deps::nvjpeg_static'
+                CUDA::culibos
+                CUDA::cudart
+        )
+else()
+        target_link_libraries(${CUCIM_PLUGIN_NAME}
+            PRIVATE
+                CUDA::nvjpeg
+                CUDA::cudart
+        )
+endif()


Right now this checks for nvJPEG static and uses it if it finds it. Otherwise dynamically linking to nvJPEG

This works for the needs of cuCIM. Namely wheels continue to statically link to nvJPEG. Meanwhile Conda packages can dynamically link to nvJPEG

However what would be better is to add an option to build nvJPEG either dynamically or statically (like with cuFile's option). The code here could similarly be updated to check that option (just like cuFile's linking logic). Then wheel and Conda builds could pass that flag based on their needs

Raised this in a new issue: #932

jakirkham · 2025-08-22T02:39:31Z

gds/CMakeLists.txt

+if (CUCIM_STATIC_GDS)
+    # Enabling CUCIM_STATIC_GDS statically links cuFile
+    target_link_libraries(cufile_stub
        PUBLIC
            ${CMAKE_DL_LIBS}
        PRIVATE


This diff (and the diff behind the fold) looks more exciting than it is

All this does is reproduce the previous logic

cucim/gds/CMakeLists.txt

Lines 96 to 101 in 54cc342

target_link_libraries(cufile_stub

PUBLIC

${CMAKE_DL_LIBS}

PRIVATE

deps::gds_static

)

It just replaces deps::gds_static with CUDA::cuFile_static. The former was a stand-in for what CMake now provides. So we just switch to the latter

jakirkham · 2025-08-22T02:40:54Z

gds/CMakeLists.txt

+            CUDA::cuFile_static
    )


Here is where deps::gds_static was replaced with CUDA::cuFile_static . For some reason GitHub put a large number of deleted lines between the two. Unclear why it did that. In any event, this is connected with the above content

jakirkham · 2025-08-22T02:46:19Z

gds/CMakeLists.txt

+if (CUCIM_STATIC_GDS)
+    # Enabling CUCIM_STATIC_GDS statically links cuFile
+    target_link_libraries(cufile_stub
        PUBLIC
            ${CMAKE_DL_LIBS}


As another note, it appears we define a bunch of stuff unconditionally that needs dlopen, etc. and thus need to link to ${CMAKE_DL_LIBS}. However we don't actually need to use it in the static case. So it seems this linkage could be dropped (after adding some of the C++ code in guards)

Raised in issue: #933

jakirkham · 2025-08-22T02:49:58Z

gds/CMakeLists.txt

-# Set GDS include path (cufile.h)
-if (DEFINED ENV{CONDA_BUILD} AND EXISTS $ENV{PREFIX}/include/cufile.h)
-    set(GDS_INCLUDE_PATH "$ENV{PREFIX}/include")
-elseif (DEFINED ENV{CONDA_PREFIX} AND EXISTS $ENV{CONDA_PREFIX}/include/cufile.h)
-    set(GDS_INCLUDE_PATH "$ENV{CONDA_PREFIX}/include")
-elseif (EXISTS ${CMAKE_CURRENT_SOURCE_DIR}/../temp/cuda/include/cufile.h)
-    set(GDS_INCLUDE_PATH ${CMAKE_CURRENT_SOURCE_DIR}/../temp/cuda/include)
-else ()
-    set(GDS_INCLUDE_PATH /usr/local/cuda/include)
-endif ()
-
-message("Set GDS_INCLUDE_PATH to '${GDS_INCLUDE_PATH}'.")


This was just to find the path to the cuFile download we did on CI

Using CMake's CUDA::cuFile* targets to pick up the cuFile in the CUDA Toolkit obviates this need

jakirkham · 2025-08-22T03:23:08Z

conda/recipes/libcucim/conda_build_config.yaml

-# The CTK libraries below are missing from the conda-forge::cudatoolkit package
-# for CUDA 11. The "*_host_*" version specifiers correspond to `11.8` packages
-# and the "*_run_*" version specifiers correspond to `11.x` packages.
-
-cuda11_libcufile_host_version:
-  - "1.4.0.31"
-
-cuda11_libcufile_run_version:
-  - ">=1.0.0.82,<=1.4.0.31"
-
-cuda11_libnvjpeg_host_version:
-  - "11.6.0.55"


These are CUDA 11 pinnings for dependencies outside the Conda cudatoolkit package

None of that is relevant for CUDA 12+, which has the full CTK as packages with associated pinning

Also the usage of these values was dropped already with PR: #889

Hence these can just be dropped

jakirkham · 2025-08-22T03:27:17Z

conda/recipes/libcucim/meta.yaml

    - cuda-version ={{ cuda_version }}
    - cuda-cudart-dev
-    - libcufile-dev     # [linux64]
+    - libcufile-dev


cuFile is available on all platforms. Hence it is added as a dependency here

jakirkham · 2025-08-22T03:27:55Z

conda/recipes/libcucim/meta.yaml

    - libnvjpeg-dev
-    - libnvjpeg-static


nvJPEG is now dynamically linked to in the Conda case. Hence no need for the static library

jakirkham · 2025-08-22T03:31:52Z

conda/recipes/libcucim/meta.yaml

    - {{ pin_compatible('cuda-version', max_pin='x', min_pin='x') }}
    - cuda-cudart
-    - libcufile  # [linux64]
+    - libcufile


This adds the cuFile dependency on ARM

However the CTK only added cuFile for ARM in CUDA 12.2: conda-forge/libcufile-feedstock#9

So this won't work for ARM with CUDA 12.0 or 12.1. Meaning we need to use a similar workaround as KvikIO did: rapidsai/kvikio#754

Another option would be to just start depending on KvikIO

Addressing in PR: #930

jakirkham · 2025-08-22T03:33:52Z

cpp/plugins/cucim.kit.cuslide/CMakeLists.txt

-superbuild_depend(nvjpeg)
-superbuild_depend(libculibos)


These handled finding the nvJPEG and cuLIBOS dependencies via custom CMake logic

That is no longer needed now that they come from the CTK

Hence this and the associated files are dropped

Following up on this thread: #905 (comment) The indentation here was off. Think the editor may have used tabs. This fixes that spacing issue Authors: - https://github.com/jakirkham Approvers: - Kyle Edwards (https://github.com/KyleFromNVIDIA) URL: #931

Following up on this thread: #905 (comment) The CUDA Toolkit (CTK) added cuFile for ARM in CTK 12.2. As a result always requiring `libcufile` with `libcucim` can make it uninstallable on ARM with CTK 12.0 & 12.1. To fix this, build two variants of the `libcucim` package. One with `libcufile` and one without. Constrain `cuda-version` accordingly so that `libcufile` is installed whenever possible. Only when `libcufile` cannot be installed, fallback to the `libcucim` package where the `libcufile` dependency is left off. The `x86_64` CTK has had cuFile since the CTK 11.4. As RAPIDS now requires CUDA 12.0+, there is no issue always requiring `libcufile` on `x86_64`. This issue will not be present in CUDA 13+. So always require `libcufile` in that case. Authors: - https://github.com/jakirkham - James Lamb (https://github.com/jameslamb) - Gregory Lee (https://github.com/grlee77) Approvers: - James Lamb (https://github.com/jameslamb) - Bradley Dice (https://github.com/bdice) URL: #930

jakirkham added 5 commits July 15, 2025 18:36

Drop cuFile workaround for pre-CUDA 11.8

e456cd3

Use dynamic library for nvJPEG

5a6b778

Remove lingering CUDA 11 checks from Conda recipes

0ca8ef7

Drop CUDA 11 trove classifier

b94d4e2

Drop CUDA 11 from docs

9e46e6f

jakirkham added improvement Improves an existing functionality non-breaking Introduces a non-breaking change labels Jul 16, 2025

Drop unused variable CUDA_MAJOR_VERSION

19d6af2

Use CMake's CUDA::nvjpeg target directly

6cc92a6

Drop workarounds to retrieve cuFile and nvJPEG

e899a6b

Drop more libnvjpeg-static occurences

47602ba

Move cuFile include to stub source

0d94e84

Move cuFile include back to header

0b5fb05

jakirkham force-pushed the drop_cuda_11 branch from 0b5fb05 to 0d94e84 Compare July 16, 2025 02:20

Always include cuFile header

6abd4b5

Use CMake's CUDA::cuFile target

3a296bd

Find CMake's CUDAToolkit for CUDA dependencies

6ba441a

Use libcufile on ARM

64a69c7

grlee77 reviewed Aug 20, 2025

View reviewed changes

gds/CMakeLists.txt Outdated Show resolved Hide resolved

Update gds/CMakeLists.txt

ad02b70

Merge branch 'branch-25.10' into drop_cuda_11

f06bdc9

grlee77 approved these changes Aug 20, 2025

View reviewed changes

bdice marked this pull request as ready for review August 20, 2025 17:29

bdice requested review from a team as code owners August 20, 2025 17:29

bdice requested a review from jameslamb August 20, 2025 17:29

bdice approved these changes Aug 20, 2025

View reviewed changes

gds/CMakeLists.txt Outdated Show resolved Hide resolved

jameslamb approved these changes Aug 20, 2025

View reviewed changes

grlee77 and others added 2 commits August 20, 2025 17:14

Update gds/CMakeLists.txt

1b5a984

Co-authored-by: Bradley Dice <bdice@bradleydice.com>

update cupy install strings in notebooks

d42ec91

rapids-bot bot merged commit b738d2c into rapidsai:branch-25.10 Aug 21, 2025
37 checks passed

jakirkham mentioned this pull request Aug 21, 2025

Build and test with CUDA 13.0.0, use GCC 14 for conda builds #926

Merged

jakirkham commented Aug 22, 2025

View reviewed changes

This was referenced Aug 22, 2025

Relax libcufile dependency for old CTK 12 on ARM #930

Merged

Fix indentation #931

Merged

Add option to build with nvJPEG dynamically or statically #932

Open

Drop ${CMAKE_DL_LIBS} linkage from static cuFile (GDS) build #933

Open

jakirkham deleted the drop_cuda_11 branch August 22, 2025 05:09

jameslamb mentioned this pull request Sep 15, 2025

Add CUDA 13.0 Support across all of RAPIDS rapidsai/build-planning#208

Open

	target_link_libraries(cufile_stub
	PUBLIC
	${CMAKE_DL_LIBS}
	PRIVATE
	deps::gds_static
	)

		superbuild_depend(nvjpeg)
		superbuild_depend(libculibos)

Remove CUDA 11 references #905

Remove CUDA 11 references #905

Uh oh!

Conversation

jakirkham commented Jul 16, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

copy-pr-bot bot commented Jul 16, 2025

Uh oh!

jakirkham commented Jul 16, 2025

Uh oh!

jakirkham commented Jul 16, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jakirkham commented Jul 16, 2025

Uh oh!

jakirkham commented Jul 16, 2025

Uh oh!

jakirkham commented Jul 16, 2025

Uh oh!

jakirkham commented Jul 16, 2025

Uh oh!

jakirkham commented Jul 16, 2025

Uh oh!

jakirkham commented Jul 16, 2025

Uh oh!

jakirkham commented Jul 16, 2025

Uh oh!

jakirkham commented Jul 16, 2025

Uh oh!

jakirkham commented Jul 16, 2025

Uh oh!

Uh oh!

grlee77 commented Aug 20, 2025

Uh oh!

bdice commented Aug 20, 2025

Uh oh!

grlee77 commented Aug 20, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

grlee77 left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

jameslamb left a comment

Choose a reason for hiding this comment

Uh oh!

jameslamb commented Aug 20, 2025

Uh oh!

grlee77 commented Aug 21, 2025

Uh oh!

bdice commented Aug 21, 2025

Uh oh!

bdice commented Aug 21, 2025

Uh oh!

Uh oh!

jakirkham left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jakirkham Aug 22, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jakirkham commented Jul 16, 2025 •

edited

Loading

jakirkham commented Jul 16, 2025 •

edited

Loading

grlee77 commented Aug 20, 2025 •

edited

Loading

grlee77 left a comment •

edited

Loading

jakirkham Aug 22, 2025 •

edited

Loading