Locate `nvvm`, `libdevice` and `nvrtc` from `nvidia-cuda-nvcc-cu12` wheels #155

brandon-b-miller · 2025-03-11T19:58:46Z

Closes #66
Closes #65

WIP, current code finds nvvm/libdevice which is enough to launch kernels, nvrtc support is next. Logic vendored from nvmath-python

kkraus14 · 2025-03-11T21:01:10Z

numba_cuda/numba/cuda/cuda_paths.py

+        if sp is not None:
+            dso_dir = os.path.join(
+                sp,
+                "nvidia",
+                "cuda_nvcc",
+                "nvvm",
+                dso_dir
+            )
+            dso_path = os.path.join(dso_dir, dso_path)
+            if os.path.exists(dso_path):
+                return str(Path(dso_path).parent)


I commented on this in NVIDIA/cuda-python#441 (comment), but we may want to just consider trying to import the nvidia package and then subsequently trying to import the cuda_nvcc package from the nvidia package instead of manually traversing the paths? We can then use nvidia.cuda_nvcc.__path__ which always resolve to sp/nvidia/cuda_nvcc and will follow the general python rules for which package takes priority properly.

kkraus14 · 2025-03-11T21:06:17Z

numba_cuda/numba/cuda/cuda_paths.py

+def _get_nvvm_wheel():
+    site_paths = [
+        site.getusersitepackages()
+    ] + site.getsitepackages() + ["conda", None]


Why do we need to add a conda path here? If someone installed the wheel in a conda environment it would presumably be in the site.getsitepackages()?

kkraus14 · 2025-03-11T21:07:15Z

numba_cuda/numba/cuda/cuda_paths.py

+        # The SONAME is taken based on public CTK 12.x releases
+        if sys.platform.startswith("linux"):
+            dso_dir = "lib64"
+            # Hack: libnvvm from Linux wheel
+            # does not have any soname (CUDAINST-3183)
+            dso_path = "libnvvm.so"
+        elif sys.platform.startswith("win32"):
+            dso_dir = "bin"
+            dso_path = "nvvm64_40_0.dll"
+        else:
+            raise AssertionError()


Can pull this out of the site_paths loop I think?

It might also be good to raise the exception with some explanation of what is wrong?

gmarkall · 2025-03-13T17:26:36Z

numba_cuda/numba/cuda/cuda_paths.py

        ('Debian package', get_debian_pkg_libdevice()),
+        ('NVIDIA NVCC Wheel', get_libdevice_wheel()),


I think we want to be looking for this ahead of the Debian package, otherwise Debian-packaged versions will always get in front of the wheel.

gmarkall · 2025-03-13T17:27:47Z

numba_cuda/numba/cuda/cuda_paths.py

    ]
+    libdevice_ctk_dir = get_system_ctk('nvvm', 'libdevice')


Why did we move the system toolkit after the Debian-packaged version? I think we want to preserve the order if we can.

gmarkall · 2025-03-13T17:32:01Z

numba_cuda/numba/cuda/cuda_paths.py

-    # Keep only the max (most recent version) of the bitcode files.
-    out = max(candidates, default=None)
+    if by == "NVIDIA NVCC Wheel":
+        # The NVVM path is a directory, not a file


What's the relevance of the NVVM path here?

I just realised this is a copy/paste error from below.

gmarkall · 2025-03-13T17:35:09Z

numba_cuda/numba/cuda/cuda_paths.py

+        # The NVVM path is a directory, not a file
+        out = os.path.join(libdir, "libdevice.10.bc")
+    else:
+        # Search for pattern


I think it's called libdevice.10.bc in all supported toolkit versions, so this logic is probably no longer needed - will check and update here.

I just checked - even back in 11.2 it is called libdevice.10.bc - so we don't need to search and choose from a set of candidates anymore.

gmarkall · 2025-03-13T17:37:13Z

numba_cuda/numba/cuda/cuda_paths.py

-    candidates = find_lib('nvvm', path)
-    path = max(candidates) if candidates else None
+    if by == "NVIDIA NVCC Wheel":
+        # The NVVM path is a directory, not a file


I can't figure out what this comment means / adds - can you explain / reword / delete it?

gmarkall

A few questions on the diff - in addition, do we plan to add a CI config that installs these from wheels so that we know it will continue to work?

brandon-b-miller · 2025-03-18T14:21:54Z

A few questions on the diff - in addition, do we plan to add a CI config that installs these from wheels so that we know it will continue to work?

Yes, I'll see about adding a separate CI job for this

brandon-b-miller · 2025-03-18T16:07:37Z

ci/test_wheel_deps_wheels.sh

+
+# remove cuda-nvvm-12-5 leaving libnvvm.so from nvidia-cuda-nvcc-cu12 only 
+apt-get update
+apt remove --purge cuda-nvvm-12-5 -y


This combined with the addition of nvidia-cuda-nvcc-cu12 was the easiest way I could think of to get to the relevant test environment, but I'm by no means married to it, this would have to be dynamic wrt the minor version as well.

You can get the installed package name with something like

CUDA_NVVM_PACKAGE=`dpkg --get-selections | grep cuda-nvvm | awk '{print $1}'`

…a-cuda#155 AS-IS

…py from NVIDIA/numba-cuda#155

ZzEeKkAa · 2025-03-21T13:50:41Z

I've merged this branch with main (fbbc040) and tested on nvmath-python. I was able successfully get rid of this patch:

    # our device apis only support cuda 12+
    _utils.force_loading_nvrtc("12")
    nvrtc.NVRTC.__new__ = __nvrtc_new__

But can't get rid of

    # Patch Numba to support wheels
    _utils.patch_numba_nvvm(nvvm)

I'm getting the error:

> python ./examples/device/cublasdx_simple_gemm_fp32.py
/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/numba-cuda/numba_cuda/numba/cuda/dispatcher.py:663: NumbaPerformanceWarning: Grid size 1 will likely result in GPU under-utilization due to low occupancy.
  warn(NumbaPerformanceWarning(msg))
Traceback (most recent call last):
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/./examples/device/cublasdx_simple_gemm_fp32.py", line 78, in <module>
    main()
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/./examples/device/cublasdx_simple_gemm_fp32.py", line 68, in main
    f[1, block_dim](a_d, b_d, c_d, alpha, beta, o_d)
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/numba-cuda/numba_cuda/numba/cuda/dispatcher.py", line 666, in __call__
    return self.dispatcher.call(args, self.griddim, self.blockdim,
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/numba-cuda/numba_cuda/numba/cuda/dispatcher.py", line 808, in call
    kernel = _dispatcher.Dispatcher._cuda_call(self, *args)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/numba-cuda/numba_cuda/numba/cuda/dispatcher.py", line 816, in _compile_for_args
    return self.compile(tuple(argtypes))
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/numba-cuda/numba_cuda/numba/cuda/dispatcher.py", line 1065, in compile
    kernel = _Kernel(self.py_func, argtypes, **self.targetoptions)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/venv/lib/python3.12/site-packages/numba/core/compiler_lock.py", line 35, in _acquire_compile_lock
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/numba-cuda/numba_cuda/numba/cuda/dispatcher.py", line 156, in __init__
    cres = compile_cuda(self.py_func, types.void, self.argtypes,
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/venv/lib/python3.12/site-packages/numba/core/compiler_lock.py", line 35, in _acquire_compile_lock
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/numba-cuda/numba_cuda/numba/cuda/compiler.py", line 290, in compile_cuda
    cres = compiler.compile_extra(typingctx=typingctx,
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/venv/lib/python3.12/site-packages/numba/core/compiler.py", line 739, in compile_extra
    return pipeline.compile_extra(func)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/venv/lib/python3.12/site-packages/numba/core/compiler.py", line 439, in compile_extra
    return self._compile_bytecode()
           ^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/venv/lib/python3.12/site-packages/numba/core/compiler.py", line 505, in _compile_bytecode
    return self._compile_core()
           ^^^^^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/venv/lib/python3.12/site-packages/numba/core/compiler.py", line 481, in _compile_core
    raise e
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/venv/lib/python3.12/site-packages/numba/core/compiler.py", line 473, in _compile_core
    pm.run(self.state)
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/venv/lib/python3.12/site-packages/numba/core/compiler_machinery.py", line 363, in run
    raise e
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/venv/lib/python3.12/site-packages/numba/core/compiler_machinery.py", line 356, in run
    self._runPass(idx, pass_inst, state)
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/venv/lib/python3.12/site-packages/numba/core/compiler_lock.py", line 35, in _acquire_compile_lock
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/venv/lib/python3.12/site-packages/numba/core/compiler_machinery.py", line 311, in _runPass
    mutated |= check(pss.run_pass, internal_state)
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/venv/lib/python3.12/site-packages/numba/core/compiler_machinery.py", line 272, in check
    mangled = func(compiler_state)
              ^^^^^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/venv/lib/python3.12/site-packages/numba/core/typed_passes.py", line 112, in run_pass
    typemap, return_type, calltypes, errs = type_inference_stage(
                                            ^^^^^^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/venv/lib/python3.12/site-packages/numba/core/typed_passes.py", line 93, in type_inference_stage
    errs = infer.propagate(raise_errors=raise_errors)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/venv/lib/python3.12/site-packages/numba/core/typeinfer.py", line 1066, in propagate
    errors = self.constraints.propagate(self)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/venv/lib/python3.12/site-packages/numba/core/typeinfer.py", line 160, in propagate
    constraint(typeinfer)
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/venv/lib/python3.12/site-packages/numba/core/typeinfer.py", line 566, in __call__
    self.resolve(typeinfer, typevars, fnty)
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/venv/lib/python3.12/site-packages/numba/core/typeinfer.py", line 589, in resolve
    sig = typeinfer.resolve_call(fnty, pos_args, kw_args)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/venv/lib/python3.12/site-packages/numba/core/typeinfer.py", line 1560, in resolve_call
    return self.context.resolve_function_type(fnty, pos_args, kw_args)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/venv/lib/python3.12/site-packages/numba/core/typing/context.py", line 195, in resolve_function_type
    res = self._resolve_user_function_type(func, args, kws)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/venv/lib/python3.12/site-packages/numba/core/typing/context.py", line 247, in _resolve_user_function_type
    return func.get_call_type(self, args, kws)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/venv/lib/python3.12/site-packages/numba/core/types/functions.py", line 538, in get_call_type
    self.dispatcher.get_call_template(args, kws)
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/numba-cuda/numba_cuda/numba/cuda/dispatcher.py", line 979, in get_call_template
    self.compile_device(tuple(args))
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/numba-cuda/numba_cuda/numba/cuda/dispatcher.py", line 1016, in compile_device
    cres = compile_cuda(self.py_func, return_type, args,
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/venv/lib/python3.12/site-packages/numba/core/compiler_lock.py", line 35, in _acquire_compile_lock
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/numba-cuda/numba_cuda/numba/cuda/compiler.py", line 290, in compile_cuda
    cres = compiler.compile_extra(typingctx=typingctx,
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/venv/lib/python3.12/site-packages/numba/core/compiler.py", line 739, in compile_extra
    return pipeline.compile_extra(func)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/venv/lib/python3.12/site-packages/numba/core/compiler.py", line 439, in compile_extra
    return self._compile_bytecode()
           ^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/venv/lib/python3.12/site-packages/numba/core/compiler.py", line 505, in _compile_bytecode
    return self._compile_core()
           ^^^^^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/venv/lib/python3.12/site-packages/numba/core/compiler.py", line 481, in _compile_core
    raise e
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/venv/lib/python3.12/site-packages/numba/core/compiler.py", line 473, in _compile_core
    pm.run(self.state)
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/venv/lib/python3.12/site-packages/numba/core/compiler_machinery.py", line 363, in run
    raise e
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/venv/lib/python3.12/site-packages/numba/core/compiler_machinery.py", line 356, in run
    self._runPass(idx, pass_inst, state)
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/venv/lib/python3.12/site-packages/numba/core/compiler_lock.py", line 35, in _acquire_compile_lock
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/venv/lib/python3.12/site-packages/numba/core/compiler_machinery.py", line 311, in _runPass
    mutated |= check(pss.run_pass, internal_state)
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/venv/lib/python3.12/site-packages/numba/core/compiler_machinery.py", line 272, in check
    mangled = func(compiler_state)
              ^^^^^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/venv/lib/python3.12/site-packages/numba/core/typed_passes.py", line 466, in run_pass
    lower = self.lowering_class(targetctx, library, fndesc, interp,
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/venv/lib/python3.12/site-packages/numba/core/lowering.py", line 40, in __init__
    self.module = self.library.create_ir_module(self.fndesc.unique_name)
                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/nvmath-python/venv/lib/python3.12/site-packages/numba/core/codegen.py", line 574, in create_ir_module
    ir_module = self._codegen._create_empty_module(name)
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/numba-cuda/numba_cuda/numba/cuda/codegen.py", line 399, in _create_empty_module
    ir_module.data_layout = nvvm.NVVM().data_layout
                            ^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/numba-cuda/numba_cuda/numba/cuda/cudadrv/nvvm.py", line 139, in __new__
    inst.driver = open_cudalib('nvvm')
                  ^^^^^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/numba-cuda/numba_cuda/numba/cuda/cudadrv/libs.py", line 83, in open_cudalib
    path = get_cudalib(lib)
           ^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/numba-cuda/numba_cuda/numba/cuda/cudadrv/libs.py", line 54, in get_cudalib
    return get_cuda_paths()['nvvm'].info or _dllnamepattern % 'nvvm'
           ^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/numba-cuda/numba_cuda/numba/cuda/cuda_paths.py", line 290, in get_cuda_paths
    'nvvm': _get_nvvm_path(),
            ^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/numba-cuda/numba_cuda/numba/cuda/cuda_paths.py", line 263, in _get_nvvm_path
    by, path = _get_nvvm_path_decision()
               ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/scratch.yhavrylko_ent/Projects/nvidia/clean_test/numba-cuda/numba_cuda/numba/cuda/cuda_paths.py", line 60, in _get_nvvm_path_decision
    if os.path.exists(nvvm_ctk_dir):
       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "<frozen genericpath>", line 19, in exists
TypeError: stat: path should be string, bytes, os.PathLike or integer, not NoneType

Context:
pynvjitlink is on and lto set to True

brandon-b-miller · 2025-03-21T13:57:18Z

Hi @ZzEeKkAa , there's a couple pieces of this that are still WIP, I think you'll probably run into bugs right now. I'm working this PR over the next few days so hopefully some more updates soon.

brandon-b-miller added 8 commits March 10, 2025 12:47

initial

b8238f9

slightly refactor cuda_paths

9c56e55

refactor libdevice search mechanism

886b9b0

debug get_cuda_paths

7ffef77

can launch kernel

fcedb13

cleanup

151e565

style

b4ededf

reset files

4f2bc2b

kkraus14 reviewed Mar 11, 2025

View reviewed changes

gmarkall added the 3 - Ready for Review Ready for review by team label Mar 13, 2025

rwgk mentioned this pull request Mar 13, 2025

EPIC: Path finder for CUDA components NVIDIA/cuda-python#451

Open

gmarkall reviewed Mar 13, 2025

View reviewed changes

gmarkall requested changes Mar 13, 2025

View reviewed changes

gmarkall added 4 - Waiting on author Waiting for author to respond to review and removed 3 - Ready for Review Ready for review by team labels Mar 13, 2025

initial ci scripts

5f4ed8f

brandon-b-miller commented Mar 18, 2025

View reviewed changes

add pynvjitlink to tests and enable

d4bf113

gmarkall added 4 - Waiting on reviewer Waiting for reviewer to respond to author and removed 4 - Waiting on author Waiting for author to respond to review labels Mar 19, 2025

rwgk added a commit to rwgk/cuda-python that referenced this pull request Mar 19, 2025

Fetch numba-cuda/numba_cuda/numba/cuda/cuda_paths.py from NVIDIA/numb…

d31920c

…a-cuda#155 AS-IS

rwgk added a commit to rwgk/cuda-python that referenced this pull request Mar 19, 2025

Minimal changes to adapt numba-cuda/numba_cuda/numba/cuda/cuda_paths.…

0c5aca5

…py from NVIDIA/numba-cuda#155

rwgk mentioned this pull request Mar 19, 2025

[Experimental] Adopt numba/cuda/cuda_paths.py NVIDIA/cuda-python#447

Draft

brandon-b-miller added 5 commits March 22, 2025 05:35

locate nvrtc

443e998

working inside container

1b436c6

somewhat roundabout logic works for system/wheel

532d864

skip tests with no set bin dir

59bb493

ensure builtins on windows

f5dbee6

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Locate `nvvm`, `libdevice` and `nvrtc` from `nvidia-cuda-nvcc-cu12` wheels #155

Locate `nvvm`, `libdevice` and `nvrtc` from `nvidia-cuda-nvcc-cu12` wheels #155

brandon-b-miller commented Mar 11, 2025

kkraus14 Mar 11, 2025

kkraus14 Mar 11, 2025

kkraus14 Mar 11, 2025

gmarkall Mar 13, 2025

gmarkall Mar 13, 2025

gmarkall Mar 13, 2025

gmarkall Mar 13, 2025

gmarkall Mar 13, 2025

gmarkall Mar 13, 2025

gmarkall Mar 13, 2025

gmarkall Mar 13, 2025

gmarkall left a comment

brandon-b-miller commented Mar 18, 2025

brandon-b-miller Mar 18, 2025

gmarkall Mar 19, 2025

ZzEeKkAa commented Mar 21, 2025 •

edited

Loading

brandon-b-miller commented Mar 21, 2025

		('Debian package', get_debian_pkg_libdevice()),
		('NVIDIA NVCC Wheel', get_libdevice_wheel()),

Locate nvvm, libdevice and nvrtc from nvidia-cuda-nvcc-cu12 wheels #155

Are you sure you want to change the base?

Locate nvvm, libdevice and nvrtc from nvidia-cuda-nvcc-cu12 wheels #155

Conversation

brandon-b-miller commented Mar 11, 2025

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

gmarkall left a comment

Choose a reason for hiding this comment

brandon-b-miller commented Mar 18, 2025

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ZzEeKkAa commented Mar 21, 2025 • edited Loading

brandon-b-miller commented Mar 21, 2025

Locate `nvvm`, `libdevice` and `nvrtc` from `nvidia-cuda-nvcc-cu12` wheels #155

Locate `nvvm`, `libdevice` and `nvrtc` from `nvidia-cuda-nvcc-cu12` wheels #155

ZzEeKkAa commented Mar 21, 2025 •

edited

Loading