Hi python / CUDA folks,
I’m searching for a (local) maxima finder for 3D data that can make use of GPUs. Ideally it should be using dask-image, cupy or any other compatible sibling. I found cupy.argmax and cusignal.peak_finding.peak_finding. argrelmax but both have an axis parameter suggesting that these only work in one dimension…
And hint / link is appreciated! Is there by chance any spot-detection + region growing tutorial for CUDA based python libraries by chance?
Thanks!
Robert
Thanks @VolkerH , that was a great hint!
I just tried and receive an error. In short:
res = peak_local_max(blurred, min_distance=1)
results in
catastrophic error: cannot open source file "type_traits"
1 catastrophic error detected in the compilation of [...]_2.cubin.cu.
Compilation terminated.
I’ve had another issue of similar kind today …
At the moment I run on CUDA 10.2, cupy 0.8.5, Windows 10, NVidia 2080 Ti.
The issue is not of high priority. I was just exploring what’s possible with CUDA. I may continue that exploration later when things are more mature. But thanks for the hint anyway!
Cheers,
Robert
Full code:
import cupy
import numpy as np
from cupyx.scipy.ndimage import gaussian_filter as cupy_gaussian_filter
image = np.random.random((200,200,200))
from cupyimg.skimage.feature.peak import peak_local_max
cuda_image = cupy.asarray(image)
cuda_blurred = cupy_gaussian_filter(cuda_image, sigma=20)
res = peak_local_max(cuda_blurred, min_distance=1)
result1 = cupy.asnumpy(res)
Full error message:
---------------------------------------------------------------------------
NVRTCError Traceback (most recent call last)
C:\Programs\miniconda3\envs\beetlesafari\lib\site-packages\cupy\cuda\compiler.py in compile(self, options, log_stream)
548 nvrtc.addAddNameExpression(self.ptr, ker)
--> 549 nvrtc.compileProgram(self.ptr, options)
550 mapping = None
cupy_backends\cuda\libs\nvrtc.pyx in cupy_backends.cuda.libs.nvrtc.compileProgram()
cupy_backends\cuda\libs\nvrtc.pyx in cupy_backends.cuda.libs.nvrtc.compileProgram()
cupy_backends\cuda\libs\nvrtc.pyx in cupy_backends.cuda.libs.nvrtc.check_status()
NVRTCError: NVRTC_ERROR_COMPILATION (6)
During handling of the above exception, another exception occurred:
CompileException Traceback (most recent call last)
<ipython-input-1-859328117964> in <module>
9 cuda_blurred = cupy_gaussian_filter(cuda_image, sigma=20)
10
---> 11 res = peak_local_max(cuda_blurred, min_distance=1)
12
13 result1 = cupy.asnumpy(res)
C:\Programs\miniconda3\envs\beetlesafari\lib\site-packages\cupyimg\skimage\_shared\utils.py in fixed_func(*args, **kwargs)
118 # warn that arg_name is deprecated
119 warnings.warn(warning_msg, FutureWarning, stacklevel=2)
--> 120 return func(*args, **kwargs)
121
122 return fixed_func
C:\Programs\miniconda3\envs\beetlesafari\lib\site-packages\cupyimg\skimage\feature\peak.py in peak_local_max(image, min_distance, threshold_abs, threshold_rel, exclude_border, indices, num_peaks, footprint, labels, num_peaks_per_label, p_norm)
273 if labels is None:
274 # Non maximum filter
--> 275 mask = _get_peak_mask(image, footprint, threshold)
276
277 mask = _exclude_border(mask, border_width)
C:\Programs\miniconda3\envs\beetlesafari\lib\site-packages\cupyimg\skimage\feature\peak.py in _get_peak_mask(image, footprint, threshold, mask)
42 return image > threshold
43
---> 44 image_max = ndi.maximum_filter(image, footprint=footprint, mode="constant")
45
46 out = image == image_max
C:\Programs\miniconda3\envs\beetlesafari\lib\site-packages\cupyimg\scipy\ndimage\filters.py in maximum_filter(input, size, footprint, output, mode, cval, origin)
1366 .. seealso:: :func:`scipy.ndimage.maximum_filter`
1367 """
-> 1368 return _min_or_max_filter(
1369 input, size, footprint, None, output, mode, cval, origin, "max"
1370 )
C:\Programs\miniconda3\envs\beetlesafari\lib\site-packages\cupyimg\scipy\ndimage\filters.py in _min_or_max_filter(input, size, ftprnt, structure, output, mode, cval, origin, func)
1386 # Seperable filter, run as a series of 1D filters
1387 fltr = minimum_filter1d if func == "min" else maximum_filter1d
-> 1388 return _filters_core._run_1d_filters(
1389 [fltr if size > 1 else None for size in sizes],
1390 input,
C:\Programs\miniconda3\envs\beetlesafari\lib\site-packages\cupyimg\scipy\ndimage\_filters_core.py in _run_1d_filters(filters, input, args, output, mode, cval, origin, dtype_mode)
101 if fltr is None:
102 continue
--> 103 fltr(input, arg, axis, output, mode, cval, origin)
104 input, output = output, temp if first else input
105 first = False
C:\Programs\miniconda3\envs\beetlesafari\lib\site-packages\cupyimg\scipy\ndimage\filters.py in maximum_filter1d(input, size, axis, output, mode, cval, origin)
1473 .. seealso:: :func:`scipy.ndimage.maximum_filter1d`
1474 """
-> 1475 return _min_or_max_1d(input, size, axis, output, mode, cval, origin, "max")
1476
1477
C:\Programs\miniconda3\envs\beetlesafari\lib\site-packages\cupyimg\scipy\ndimage\filters.py in _min_or_max_1d(input, size, axis, output, mode, cval, origin, func)
1503 has_weights=False,
1504 )
-> 1505 return _filters_core._call_kernel(
1506 kernel, input, None, output, weights_dtype=bool
1507 )
C:\Programs\miniconda3\envs\beetlesafari\lib\site-packages\cupyimg\scipy\ndimage\_filters_core.py in _call_kernel(kernel, input, weights, output, structure, weights_dtype, structure_dtype)
150 output, temp = _util._get_output(output.dtype, input), output
151 args.append(output)
--> 152 kernel(*args)
153 if needs_temp:
154 temp[...] = output[...]
cupy\core\_kernel.pyx in cupy.core._kernel.ElementwiseKernel.__call__()
cupy\core\_kernel.pyx in cupy.core._kernel.ElementwiseKernel._get_elementwise_kernel()
cupy\_util.pyx in cupy._util.memoize.decorator.ret()
cupy\core\_kernel.pyx in cupy.core._kernel._get_elementwise_kernel()
cupy\core\_kernel.pyx in cupy.core._kernel._get_simple_elementwise_kernel()
cupy\core\_kernel.pyx in cupy.core._kernel._get_simple_elementwise_kernel()
cupy\core\core.pyx in cupy.core.core.compile_with_cache()
C:\Programs\miniconda3\envs\beetlesafari\lib\site-packages\cupy\cuda\compiler.py in compile_with_cache(source, options, arch, cache_dir, extra_source, backend, enable_cooperative_groups, name_expressions, log_stream)
368 name_expressions, log_stream, cache_in_memory)
369 else:
--> 370 return _compile_with_cache_cuda(
371 source, options, arch, cache_dir, extra_source, backend,
372 enable_cooperative_groups, name_expressions, log_stream,
C:\Programs\miniconda3\envs\beetlesafari\lib\site-packages\cupy\cuda\compiler.py in _compile_with_cache_cuda(source, options, arch, cache_dir, extra_source, backend, enable_cooperative_groups, name_expressions, log_stream, cache_in_memory)
433 if backend == 'nvrtc':
434 cu_name = '' if cache_in_memory else name + '.cu'
--> 435 ptx, mapping = compile_using_nvrtc(
436 source, options, arch, cu_name, name_expressions,
437 log_stream, cache_in_memory)
C:\Programs\miniconda3\envs\beetlesafari\lib\site-packages\cupy\cuda\compiler.py in compile_using_nvrtc(source, options, arch, filename, name_expressions, log_stream, cache_in_memory)
209 cu_file.write(source)
210
--> 211 return _compile(source, options, cu_path,
212 name_expressions, log_stream)
213 else:
C:\Programs\miniconda3\envs\beetlesafari\lib\site-packages\cupy\cuda\compiler.py in _compile(source, options, cu_path, name_expressions, log_stream)
193 name_expressions=name_expressions)
194 try:
--> 195 ptx, mapping = prog.compile(options, log_stream)
196 except CompileException as e:
197 dump = _get_bool_env_variable(
C:\Programs\miniconda3\envs\beetlesafari\lib\site-packages\cupy\cuda\compiler.py in compile(self, options, log_stream)
558 except nvrtc.NVRTCError:
559 log = nvrtc.getProgramLog(self.ptr)
--> 560 raise CompileException(log, self.src, self.name, options,
561 'nvrtc' if not runtime.is_hip else 'hiprtc')
562
CompileException: C:\Programs\miniconda3\envs\beetlesafari\lib\site-packages\cupy\core\include\cupy/complex/complex.h(94): warning: __host__ annotation is ignored on a function("complex") that is explicitly defaulted on its first declaration
C:\Programs\miniconda3\envs\beetlesafari\lib\site-packages\cupy\core\include\cupy/complex/complex.h(94): warning: __device__ annotation is ignored on a function("complex") that is explicitly defaulted on its first declaration
C:\Programs\miniconda3\envs\beetlesafari\lib\site-packages\cupy\core\include\cupy/complex/complex.h(101): warning: __host__ annotation is ignored on a function("complex") that is explicitly defaulted on its first declaration
C:\Programs\miniconda3\envs\beetlesafari\lib\site-packages\cupy\core\include\cupy/complex/complex.h(101): warning: __device__ annotation is ignored on a function("complex") that is explicitly defaulted on its first declaration
C:\Users\ROBERT~1\AppData\Local\Temp\tmpsxqh350g\7dc0f5bf0f642b710c0b8a179916eb98_2.cubin.cu(10): catastrophic error: cannot open source file "type_traits"
1 catastrophic error detected in the compilation of "C:\Users\ROBERT~1\AppData\Local\Temp\tmpsxqh350g\7dc0f5bf0f642b710c0b8a179916eb98_2.cubin.cu".
Compilation terminated.
Yes, getting a working version of cupyimg is still a bit tricky as it and all its dependencies are under constant developement.
I had very similar problems until a few days ago.
You may want to refer to this issue:
For me, some of the recent changes that Greg pushed together with cupy build 9.0.0b3 from conda-forge made it work (on Ubuntu/conda). He also announced that this will be released more publicly soon, probably also with easier installation.
that definitely will no longer work (see issue above)
Yeah, I will wait for that. The list of supported functions looks amazing and good things take a while. Thanks again for the support!
Yes, I am patiently waiting for the cupy/numpy compatible API to clEsperanto.
I was considering to work on API-compatibility with scikit-image as soon as its version 1.0 is out.