Deion When using in-place operations for float arrays, the <

Thanks for reporting <a class="user-mention notranslate" data-hovercard-type="user" da

`ndimage.shift` results in incorrect values for inplace for float arrays about cupy HOT 2 OPEN

JoOkuma commented on June 18, 2024

`ndimage.shift` results in incorrect values for inplace for float arrays

from cupy.

Comments (2)

JoOkuma commented on June 18, 2024 1

Making a copy and updating the docs to say it's not in place would be excellent!
Thanks for the quick reply @kmaehashi

from cupy.

kmaehashi commented on June 18, 2024

Thanks for reporting @JoOkuma! The shift operation (and many ndimage features) rely on raw CUDA kernel generated on-the-fly. I haven't fully checked the code but it looks it does not support cases s.t. input and output memory overlap.

Supporting in-place will need almost full rewrite of the kernel so I'm thinking of adding a check and making a copy automatically if there is an overlap.

cupy/cupyx/scipy/ndimage/_interpolation.py

Lines 640 to 655 in 8fb6d04

 output = _util._get_output(output, input) 

 if input.dtype.kind in 'iu': 

 input = input.astype(cupy.float32) 

 filtered, nprepad = _filter_input(input, prefilter, mode, cval, order) 

 integer_output = output.dtype.kind in 'iu' 

 _util._check_cval(mode, cval, integer_output) 

 large_int = _prod(input.shape) > 1 << 31 

 kern = _interp_kernels._get_shift_kernel( 

 input.ndim, large_int, input.shape, mode, cval=cval, order=order, 

 integer_output=integer_output, nprepad=nprepad) 

 shift = cupy.asarray(shift, dtype=cupy.float64, order='C') 

 if shift.ndim != 1: 

 raise ValueError('shift must be 1d') 

 if shift.size != filtered.ndim: 

 raise ValueError('len(shift) must equal input.ndim') 

 kern(filtered, shift, output)

from cupy.

Recommend Projects

`ndimage.shift` results in incorrect values for inplace for float arrays about cupy HOT 2 OPEN

Comments (2)

Related Issues (20)

Recommend Projects

React

Vue.js

Typescript

TensorFlow

Django

Laravel

D3

Recommend Topics

javascript

web

server

Machine learning

Visualization

Game

Recommend Org

Facebook

Microsoft

Google

Alibaba

D3

Tencent

	output = _util._get_output(output, input)
	if input.dtype.kind in 'iu':
	input = input.astype(cupy.float32)
	filtered, nprepad = _filter_input(input, prefilter, mode, cval, order)
	integer_output = output.dtype.kind in 'iu'
	_util._check_cval(mode, cval, integer_output)
	large_int = _prod(input.shape) > 1 << 31
	kern = _interp_kernels._get_shift_kernel(
	input.ndim, large_int, input.shape, mode, cval=cval, order=order,
	integer_output=integer_output, nprepad=nprepad)
	shift = cupy.asarray(shift, dtype=cupy.float64, order='C')
	if shift.ndim != 1:
	raise ValueError('shift must be 1d')
	if shift.size != filtered.ndim:
	raise ValueError('len(shift) must equal input.ndim')
	kern(filtered, shift, output)