[special] Add cupyx.scipy.special.expn

[khushi-411]

TODO

• add benchmarks

Benchmarks:

size	dtype	SciPy	CuPy
1000	int8	0.119 ms	0.203 ms
1000	int32	0.111 ms	0.201 ms
1000	uint32	0.115 ms	0.201 ms
1000	float32	0.283 ms	0.207 ms
1000	float64	0.257 ms	0.207 ms
100000	int8	11.396 ms	8.503 ms
100000	int32	11.705 ms	7.147 ms
100000	uint32	11.495 ms	6.202 ms
100000	float32	17.490 ms	6.235 ms
100000	float64	17.482 ms	6.227 ms
1000000	int8	114.541 ms	60.157 ms
1000000	int32	110.311 ms	58.221 ms
1000000	uint32	109.507 ms	58.139 ms
1000000	float32	164.692 ms	59.586 ms
1000000	float64	165.868 ms	62.705 ms

Script: Click Please

import cupy
import cupyx

import scipy

from cupy import testing
from cupyx.profiler import benchmark


n_warmup = 1
n_repeat = 10
dtype_set = ('int8', 'int32', 'uint32', 'float32', 'float64')
n_set = (1000, 100000, 1000000)


def get_time(pref):
    cpu_time = pref.cpu_times.mean()
    gpu_time = pref.gpu_times.mean()
    return max(cpu_time, gpu_time) * 1000  # ms


def time_expn(a, b):
    cp_a = cupy.array(a)
    cp_b = cupy.array(b)
    ref = scipy.special.expn(a, b)
    ret = cupyx.scipy.special.expn(cp_a, cp_b)
    times = []
    perf = benchmark(scipy.special.expn, (a, b),
                     n_warmup=n_warmup, n_repeat=n_repeat)
    times.append(get_time(perf))
    perf = benchmark(cupyx.scipy.special.expn, (cp_a, cp_b),
                     n_warmup=n_warmup, n_repeat=n_repeat)
    times.append(get_time(perf))
    return times


print('size | dtype | SciPy | CuPy')
print('-----|-------|-------|-----')
for n in n_set:
    for dtype in dtype_set:
        a = scipy.random.randint(0, 2, size=n).astype(dtype)
        b = scipy.random.randint(0, 10, size=n).astype(dtype)
        times = time_expn(a, b)
        print('{} | {} | {:.3f} ms | {:.3f} ms'.format(n, dtype, times[0], times[1]))

[takagi]

/test mini

[khushi-411]

I think this failure is unrelatable to the PR:

Last 10 lines from the test output: |  
-- | --
00:49:43.379276 STDOUT 3240] | FAILED cupy_tests\core_tests\test_ndarray_adv_indexing.py::TestArrayAdvancedIndexingSetitemScalarValueDeprecated::test_adv_setitem[_param_8_{indexes=[[1, 0], 2], shape=(2, 3, 4), value=1}] |  
00:49:43.379775 STDOUT 3240] | FAILED cupy_tests\core_tests\test_ndarray_adv_indexing.py::TestArrayAdvancedIndexingSetitemScalarValueDeprecated::test_adv_setitem[_param_9_{indexes=[[1], slice(1, 2, None)], shape=(2, 3, 4), value=1}] |  
00:49:43.379775 STDOUT 3240] | FAILED cupy_tests\core_tests\test_ndarray_adv_indexing.py::TestArrayAdvancedIndexingSetitemScalarValueDeprecated::test_adv_setitem[_param_10_{indexes=[[[1]], slice(1, 2, None)], shape=(2, 3, 4), value=1}] |  
00:49:43.380175 STDOUT 3240] | FAILED cupy_tests\core_tests\test_ndarray_copy_and_view.py::TestArrayCopyAndView::test_view_f_contiguous |  
00:49:43.380175 STDOUT 3240] | FAILED cupy_tests\core_tests\test_ndarray_copy_and_view.py::TestArrayCopyAndView::test_view_flags_smaller[F-(3, 5)] |  
00:49:43.380175 STDOUT 3240] | FAILED cupy_tests\core_tests\test_ndarray_copy_and_view.py::TestArrayCopyAndView::test_view_flags_larger[F-(6, 5)] |  
00:49:43.380175 STDOUT 3240] | FAILED cupy_tests\core_tests\test_ndarray_copy_and_view.py::TestArrayCopyAndView::test_view_flags_larger[F-(2, 3)] |  
00:49:43.380175 STDOUT 3240] | FAILED cupy_tests\creation_tests\test_basic.py::TestBasic::test_empty_zero_sized_array_strides |  
00:49:43.380175 STDOUT 3240] | FAILED cupy_tests\functional_tests\test_vectorize.py::TestVectorizeOps::test_vectorize_div |  
00:49:43.380175 STDOUT 3240] | = 28 failed, 96133 passed, 5852 skipped, 689 deselected, 151 xfailed, 2569 warnings in 2247.59s (0:37:27) =

Last 10 lines from the test output:	
[00:49:43.379276 STDOUT 3240]](https://ci.preferred.jp/cupy.win.cuda116/104629/#L4877)	FAILED cupy_tests\core_tests\test_ndarray_adv_indexing.py::TestArrayAdvancedIndexingSetitemScalarValueDeprecated::test_adv_setitem[_param_8_{indexes=[[1, 0], 2], shape=(2, 3, 4), value=1}]	
[00:49:43.379775 STDOUT 3240]](https://ci.preferred.jp/cupy.win.cuda116/104629/#L4879)	FAILED cupy_tests\core_tests\test_ndarray_adv_indexing.py::TestArrayAdvancedIndexingSetitemScalarValueDeprecated::test_adv_setitem[_param_9_{indexes=[[1], slice(1, 2, None)], shape=(2, 3, 4), value=1}]	
[00:49:43.379775 STDOUT 3240]](https://ci.preferred.jp/cupy.win.cuda116/104629/#L4881)	FAILED cupy_tests\core_tests\test_ndarray_adv_indexing.py::TestArrayAdvancedIndexingSetitemScalarValueDeprecated::test_adv_setitem[_param_10_{indexes=[[[1]], slice(1, 2, None)], shape=(2, 3, 4), value=1}]	
[00:49:43.380175 STDOUT 3240]](https://ci.preferred.jp/cupy.win.cuda116/104629/#L4883)	FAILED cupy_tests\core_tests\test_ndarray_copy_and_view.py::TestArrayCopyAndView::test_view_f_contiguous	
[00:49:43.380175 STDOUT 3240]](https://ci.preferred.jp/cupy.win.cuda116/104629/#L4884)	FAILED cupy_tests\core_tests\test_ndarray_copy_and_view.py::TestArrayCopyAndView::test_view_flags_smaller[F-(3, 5)]	
[00:49:43.380175 STDOUT 3240]](https://ci.preferred.jp/cupy.win.cuda116/104629/#L4885)	FAILED cupy_tests\core_tests\test_ndarray_copy_and_view.py::TestArrayCopyAndView::test_view_flags_larger[F-(6, 5)]	
[00:49:43.380175 STDOUT 3240]](https://ci.preferred.jp/cupy.win.cuda116/104629/#L4886)	FAILED cupy_tests\core_tests\test_ndarray_copy_and_view.py::TestArrayCopyAndView::test_view_flags_larger[F-(2, 3)]	
[00:49:43.380175 STDOUT 3240]](https://ci.preferred.jp/cupy.win.cuda116/104629/#L4887)	FAILED cupy_tests\creation_tests\test_basic.py::TestBasic::test_empty_zero_sized_array_strides	
[00:49:43.380175 STDOUT 3240]](https://ci.preferred.jp/cupy.win.cuda116/104629/#L4888)	FAILED cupy_tests\functional_tests\test_vectorize.py::TestVectorizeOps::test_vectorize_div	
[00:49:43.380175 STDOUT 3240]](https://ci.preferred.jp/cupy.win.cuda116/104629/#L4889)	= 28 failed, 96133 passed, 5852 skipped, 689 deselected, 151 xfailed, 2569 warnings in 2247.59s (0:37:27) =

What do you think? Thanks!

[takagi]

Yes, its because of NumPy 1.23 that is recently released, and I'm working on fixing it.

[takagi]

LGTM!

[takagi]

LGTM!

[khushi-411]

Thanks a lot for your suggestions and guidance! :)

[takagi]

Would you merge this as the failure of cupy.win.cuda116 is not related? @kmaehashi

[takagi]

Thanks for your work, @khushi-411!