Skip to content

fix: torch.std and torch.var support multi-dimensional reductions #1395

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 5 commits into from
Oct 12, 2022
Merged

fix: torch.std and torch.var support multi-dimensional reductions #1395

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
39246e6
Augment current tests to reproduce error generating bug with standard…
gs-olive Oct 8, 2022
c65ef4e
fix: Ensure torch.std (and torch.var) support multiple dimensions
gs-olive Oct 8, 2022
6a54c2b
Merge branch 'pytorch:master' into multi_dim_var_bugfix
gs-olive Oct 10, 2022
5bebd9c
Improved comments on IR for UnpackVar function
gs-olive Oct 10, 2022
49ff751
Merge branch 'multi_dim_var_bugfix' of https://github.com/gs-olive/Te…
gs-olive Oct 10, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
17 changes: 12 additions & 5 deletions core/lowering/passes/unpack_var.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,11 +26,18 @@ void UnpackVar(std::shared_ptr<torch::jit::Graph>& graph) {
%var: Tensor = aten::sub(%sqrdmean, %meansqrd, %1)
%varout : Tensor = prim::If(%unbiased)
block0():
%shape: int[] = aten::size(%input)
%shapet: Tensor = aten::tensor(%shape, %f32_dtype, %none, %false)
%dim: int = prim::ListUnpack(%dims)
%reduceddims: Tensor = aten::select(%shapet, %0, %dim)
%numel: Tensor = aten::prod(%reduceddims, %dim, %keepdim, %none)
# Compute number of elements in original input tensor
%originalshape: int[] = aten::size(%input)
%originalshapet: Tensor = aten::tensor(%originalshape, %f32_dtype, %none, %false)
%originalnumel: Tensor = aten::prod(%originalshapet, %0, %false, %none)
# Compute number of elements in resulting output tensor
%resultingshape: int[] = aten::size(%var)
%resultingshapet: Tensor = aten::tensor(%resultingshape, %f32_dtype, %none, %false)
%resultingnumel: Tensor = aten::prod(%resultingshapet, %0, %false, %none)
# Quotient of original number of elements and resulting number of elements
# is equal to the number of elements used per variance calculation
%numel: Tensor = aten::div(%originalnumel, %resultingnumel)
# Perform Bessel's correction on computed variance
%mul: Tensor = aten::mul(%var, %numel)
%sub: Tensor = aten::sub(%numel, %1, %1)
%v: Tensor = aten::div(%mul, %sub)
Expand Down
3 changes: 2 additions & 1 deletion tests/core/conversion/converters/test_reduce.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -465,7 +465,8 @@ TEST(Converters, UnpackStdUnbiasedKeepDimsLowersCorrectly) {
%5 : bool = prim::Constant[value=0]() # test_zeros.py:10:65
%4 : bool = prim::Constant[value=1]() # test_zeros.py:10:50
%3 : int = prim::Constant[value=0]() # test_zeros.py:10:39
%6 : int[] = prim::ListConstruct(%3)
%one : int = prim::Constant[value=1]()
%6 : int[] = prim::ListConstruct(%3, %one)
%7 : Tensor = aten::std(%x.1, %6, %4, %5) # test_zeros.py:10:26
return (%7))IR";

Expand Down
6 changes: 4 additions & 2 deletions tests/core/lowering/test_unpack_reduce_ops.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -134,7 +134,8 @@ TEST(LoweringPasses, UnpackStdKeepDimsLowersCorrectly) {
%5 : bool = prim::Constant[value=0]() # test_zeros.py:10:65
%4 : bool = prim::Constant[value=1]() # test_zeros.py:10:50
%3 : int = prim::Constant[value=0]() # test_zeros.py:10:39
%6 : int[] = prim::ListConstruct(%3)
%one : int = prim::Constant[value=1]()
%6 : int[] = prim::ListConstruct(%3, %one)
%7 : Tensor = aten::std(%x.1, %6, %5, %5) # test_zeros.py:10:26
return (%7))IR";

Expand Down Expand Up @@ -184,7 +185,8 @@ TEST(LoweringPasses, UnpackStdUnbiasedKeepDimsLowersCorrectly) {
%5 : bool = prim::Constant[value=0]() # test_zeros.py:10:65
%4 : bool = prim::Constant[value=1]() # test_zeros.py:10:50
%3 : int = prim::Constant[value=0]() # test_zeros.py:10:39
%6 : int[] = prim::ListConstruct(%3)
%one : int = prim::Constant[value=1]()
%6 : int[] = prim::ListConstruct(%3, %one)
%7 : Tensor = aten::std(%x.1, %6, %4, %5) # test_zeros.py:10:26
return (%7))IR";

Expand Down