Skip to content

[jit][fuser] avoid kernel launches for zero-sized tensor inputs #22790

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][fuser] avoid kernel launches for zero-sized tensor inputs #22790

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
46c41ae
[jit][fuser] avoid kernel launches for zero-sized tensor inputs
wanchaol Jul 12, 2019
398310d
Update on "[jit][fuser] avoid kernel launches for zero-sized tensor i…
wanchaol Jul 12, 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
12 changes: 12 additions & 0 deletions test/test_jit_fuser.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -46,6 +46,18 @@ def test_abs_cpu(self):
def test_abs_cuda(self):
self._test_fused_abs(device="cuda")

@unittest.skipIf(not RUN_CUDA, "requires CUDA")
@skipIfRocm
def test_zero_element_tensors(self):
def decode(sin_t, cos_t):
theta = torch.atan2(sin_t.float(), cos_t.float())
return theta

sin = torch.zeros(0, device="cuda")
cos = torch.zeros(0, device="cuda")
inputs = [sin, cos]
ge = self.checkScript(decode, inputs)

@unittest.skipIf(not RUN_CUDA, "fuser requires CUDA")
def test_arg_configurations_smoke_cuda(self):
# A smoke test to make sure we won't use the same kernel for contiguous
Expand Down
7 changes: 5 additions & 2 deletions torch/csrc/jit/fuser/executor.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -313,8 +313,11 @@ void launchFusion(
}
}
}

fusion.launch_raw(numel, arguments);
// Skip launching the kernel for zero-element tensor inputs
// launches are skipped, empty zero-sized output is returned
if (numel > 0) {
fusion.launch_raw(numel, arguments);
}
}

bool runFusion(const int64_t key, Stack& stack, std::string* code_out) {
Expand Down