Skip to content

[jit] fix pow bug on overloads and clean up #20824

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

Closed
wanchaol wants to merge 2 commits into from
Closed

[jit] fix pow bug on overloads and clean up #20824

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
5a02ca5
[jit] fix pow bug on overloads and clean up
wanchaol May 22, 2019
3b106a4
Update on "[jit] fix pow bug on overloads and clean up"
wanchaol May 22, 2019
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
5 changes: 5 additions & 0 deletions test/test_jit.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3863,12 +3863,17 @@ def func(a, b):
def func2(a, b, c, d):
return c + a ** b ** d

def func3(a, b):
# type: (int, float) -> float
return a ** b

a = torch.rand(1, requires_grad=True)
b = torch.rand(1, requires_grad=True)
c = torch.rand(1, requires_grad=True)
d = torch.rand(1, requires_grad=True)
self.checkScript(func, (a, b), optimize=True)
self.checkScript(func2, (a, b, c, d), optimize=True)
self.checkScript(func3, (4, -0.5), optimize=True)

@unittest.skipIf(not RUN_CUDA, "device tests require CUDA")
def test_pow_scalar_backward_cuda(self):
Expand Down
20 changes: 1 addition & 19 deletions torch/csrc/jit/register_prim_ops.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1962,7 +1962,7 @@ RegisterOperators reg2({
DEFINE_BINARY_OP(aten::add, a + b),
DEFINE_BINARY_OP(aten::sub, a - b),
DEFINE_BINARY_OP(aten::mul, a* b),
DEFINE_BINARY_OP(aten::pow, static_cast<decltype(a)>(pow(a, b))),
DEFINE_BINARY_OP(aten::pow, pow(a, b)),
// min and max are in prim:: because there is a difference between
// the python builtin 'min' and 'torch.min'
DEFINE_BINARY_OP(prim::min, a < b ? a : b),
Expand Down Expand Up @@ -2037,24 +2037,6 @@ RegisterOperators reg2({
return 0;
}),

Operator(
Copy link
Contributor

@cpuhrsch cpuhrsch May 22, 2019

Choose a reason for hiding this comment

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

Why do you get to remove this?

Copy link
Collaborator Author

@wanchaol wanchaol May 22, 2019

Choose a reason for hiding this comment

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

DEFINE_BINARY_OPS already emit the int/float mixed ops (called DEFINE_INT_FLOAT_OP here), this is not necessary anymore. This was due to some merge conflicts that we did not aware of before..

"aten::pow(float a, float b) -> float",
[](Stack& stack) {
double a, b;
pop(stack, a, b);
push(stack, std::pow(a, b));
return 0;
}),
Operator(
"aten::pow(float a, int b) -> float",
[](Stack& stack) {
double a;
int b;
pop(stack, a, b);
push(stack, std::pow(a, b));
return 0;
}),

Operator(
"aten::floor(float a) -> float",
[](Stack& stack) {
Expand Down