Skip to content

Sparse Library#333

Merged
apaszke merged 7 commits intopytorch:masterfrom
ebetica:new-sparse
Jan 4, 2017
Merged

Sparse Library#333
apaszke merged 7 commits intopytorch:masterfrom
ebetica:new-sparse

Conversation

@ebetica
Copy link
Contributor

@ebetica ebetica commented Dec 20, 2016

I redid this PR, since the last one was old and I did not want to rebase everything just to change up the module format. For reference: #116

Having sparse modules in csrc/Modules.cpp allows us unify calling functions that require sparse and non-sparse arguments. In the previous PR, sparse.addmm is now just torch.addmm and torch will call the correct functions.

This is added as a "Sparse=True" flag to the Embedding Module. The modifications for the training loop is given in ebetica/examples@3c41e57

The list of implemented functions are in torch/csrc/generic/methods/SparseTensor.cwrap

@ebetica ebetica mentioned this pull request Dec 20, 2016
Closed
6 tasks
apaszke
apaszke suggested changes Dec 20, 2016
View reviewed changes
Copy link
Contributor

@apaszke apaszke left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think it would be better to move most of the C bindings to torch/csrc/sparse

torch/csrc/Module.cpp Outdated

This comment was marked as off-topic.

Sign in to view

torch/csrc/autograd/variable.h Outdated

This comment was marked as off-topic.

Sign in to view

torch/csrc/cuda/Module.cpp Outdated

This comment was marked as off-topic.

Sign in to view

This comment was marked as off-topic.

Sign in to view

torch/csrc/cuda/Module.cpp Outdated

This comment was marked as off-topic.

Sign in to view

This comment was marked as off-topic.

Sign in to view

torch/csrc/generic/SparseTensor.cpp Outdated

This comment was marked as off-topic.

Sign in to view

This comment was marked as off-topic.

Sign in to view

torch/lib/THS/generic/THSTensor.c Outdated

This comment was marked as off-topic.

Sign in to view

This comment was marked as off-topic.

Sign in to view

torch/lib/THS/generic/THSTensor.c Outdated

This comment was marked as off-topic.

Sign in to view

This comment was marked as off-topic.

Sign in to view

torch/csrc/generic/SparseTensor.cpp Outdated

This comment was marked as off-topic.

Sign in to view

test/test_sparse.py Outdated

This comment was marked as off-topic.

Sign in to view

torch/csrc/generic/SparseTensor.cpp Outdated

This comment was marked as off-topic.

Sign in to view

apaszke
apaszke suggested changes Dec 21, 2016
View reviewed changes
Copy link
Contributor

@apaszke apaszke left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Not sure how hard would that be, but I think we shouldn't make Tensor.cpp any bigger. It already takes ages to compile. Apart from this and minor comments from my review, looks good to me now.

torch/lib/THCS/generic/THCSTensorMath.cu Outdated

This comment was marked as off-topic.

Sign in to view

torch/lib/THS/generic/THSTensor.c Outdated

This comment was marked as off-topic.

Sign in to view

torch/lib/THS/generic/THSTensor.h Outdated

This comment was marked as off-topic.

Sign in to view

torch/nn/functions/thnn/sparse.py Outdated

This comment was marked as off-topic.

Sign in to view

@soumith
Copy link
Collaborator

soumith commented Dec 21, 2016

@apaszke zeming is on vacation till end of the year on the beaches of costa rica. if the last changes are minor, can you quickly locally do them and push it in?

@adamlerer
Copy link
Contributor

adamlerer commented Dec 29, 2016

cc @alexpeys

@ebetica ebetica mentioned this pull request Jan 4, 2017
Merged
apaszke
apaszke reviewed Jan 4, 2017
View reviewed changes
tools/cwrap/cwrap.py Outdated
def set_declaration_defaults(self, declaration):
declaration.setdefault('arguments', [])
declaration.setdefault('return', 'void')
if 'sparse' not in declaration:

This comment was marked as off-topic.

Sign in to view

tools/cwrap/plugins/THPPlugin.py Outdated
$init

$options
PyObject * $name(PyObject *self, PyObject *args, PyObject *kwargs)

This comment was marked as off-topic.

Sign in to view

torch/csrc/Module.cpp Outdated
}

/***
* SPARSE STATELESS FUNCTIONS

This comment was marked as off-topic.

Sign in to view

torch/csrc/generic/methods/SparseTensor.cwrap

[[
name: size
defined_if: "!IS_CUDA"

This comment was marked as off-topic.

Sign in to view

This comment was marked as off-topic.

Sign in to view

torch/csrc/generic/methods/TensorMath.cwrap
- arg: real value
default: AS_REAL(1)
- THTensor* other
- sparse: True

This comment was marked as off-topic.

Sign in to view

This comment was marked as off-topic.

Sign in to view

torch/csrc/sparse.cpp
PyObject* sparse_tensor_classes;

////////////////////////////////////////////////////////////////////////////////
// SPARSE MODULE INITIALIZATION

This comment was marked as off-topic.

Sign in to view

torch/lib/THS/generic/THSTensorMath.c
THLongTensor_zero(csr);

// Convert the sparse matrix to CSR format
#pragma omp parallel for private(i, h, hp0, hp1) schedule(static) if (nnz > 10000)

This comment was marked as off-topic.

Sign in to view

This comment was marked as off-topic.

Sign in to view

test/test_sparse.py
test_shape(5, 6)
test_shape(10, 10, 10)
test_shape(50, 30, 20)
test_shape(5, 5, 5, 5, 5, 5)

This comment was marked as off-topic.

Sign in to view

This comment was marked as off-topic.

Sign in to view

@apaszke apaszke merged commit 59d66e6 into pytorch:master Jan 4, 2017
mrshenli pushed a commit to mrshenli/pytorch that referenced this pull request Apr 11, 2020
@ericnakagawa
Merge pull request pytorch#333 from ericnakagawa/master
8bcd0ac 
Merging RC1 into master
jjsjann123 pushed a commit to jjsjann123/pytorch that referenced this pull request Sep 23, 2020
@kevinstephano
Fix Inner Dimension Reductions for FP16 to perform just as well as TI. (
0fbfa90 
pytorch#333)

Add Executor method to compile from a string for debug usage.  Fix Reduction Scheduler to have TI level perf for FP16 inner dimension reductions. Fix tests to use randn() so large reductions aren't matching on inf.
jjsjann123 pushed a commit to jjsjann123/pytorch that referenced this pull request Sep 24, 2020
@kevinstephano @jjsjann123
Fix Inner Dimension Reductions for FP16 to perform just as well as TI. (
46c0f22 
pytorch#333)

Add Executor method to compile from a string for debug usage.  Fix Reduction Scheduler to have TI level perf for FP16 inner dimension reductions. Fix tests to use randn() so large reductions aren't matching on inf.
KsenijaS pushed a commit to KsenijaS/pytorch that referenced this pull request Dec 14, 2020
@M-Tonin
GoogLenet Pre and post process (pytorch#333)
16ab8f8 
* Update README.md
KyleCZH pushed a commit to KyleCZH/pytorch that referenced this pull request Sep 20, 2021
@peterjc123 @ezyang
Move nightlies to the new channel for Windows (pytorch#333)
0032cad
eellison pushed a commit to eellison/pytorch that referenced this pull request Jun 29, 2022
@jansel
Fix test skip when sympy is missing (pytorch#333)
df47817
hubertlu-tw pushed a commit to hubertlu-tw/pytorch that referenced this pull request Nov 1, 2022
Give multi-tensor L2 norm the ability to compute norms per-tensor as …
93338e6 
…well as globally (pytorch#333)

* Existing tests passing, still need to add per-tensor tests

* Test is passing, still need to measure performance

* ILP for l2norm functor
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

1 more reviewer

@apaszke apaszke apaszke approved these changes

Reviewers whose approvals may not affect merge requirements

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

4 participants

@ebetica @soumith @adamlerer @apaszke