[Feat request] Allow multiple built-in function usage in .update_all

🚀 Feature

support multiple built-in function at once: g.update_all( [m_func1, m_func2] , [r_func1, r_func2] )

Motivation

Main motivation would be better performance, since I imagine it would eliminate considerable overhead. If there is no overhead, then the improvement would be only having a cleaner code.

Pitch

Desired solution

import dgl
import dgl.function as fn
import torch

g = dgl.heterograph({('user', 'follows', 'user'): ([0, 1, 2], [1, 2, 2])})
g.nodes['user'].data['h'] = torch.tensor([[0.], [1.], [2.]])
g.nodes['user'].data['x'] = torch.tensor([[0.], [1.], [2.]])

#CURRENT APPROACH
g['follows'].update_all(fn.copy_u('h', 'm'), fn.sum('m', 'h_new'), etype='follows')
g['follows'].update_all(fn.copy_u('x', 'm2'), fn.sum('m2', 'h_new'), etype='follows')

#DESIRED APPROACH
g['follows'].update_all([fn.copy_src('h', 'm'), fn.copy_src('x', 'm2')], [fn.sum('m', 'h_new'), fn.sum('m2', 'h_new2')], etype='follows')

#>>Not supported. Throws error: `TypeError: 'list' object is not callable`

[VoVAllen]

Hi, currently we won’t support this fusion since we don’t have much to improve on the performance.

[github-actions[bot]]

This issue is closed due to lack of activity. Feel free to reopen it if you still have questions.