Skip to content

[PIR]Open uts for scatter #60694

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
0x45f merged 6 commits into from
Jan 18, 2024
Merged

[PIR]Open uts for scatter #60694

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
b42cc79
[PIR]Open uts for scatter
Jan 10, 2024
07ace71
Fix OOM
Jan 15, 2024
5dad790
Fix Alloc
Jan 15, 2024
0c59760
Fix Alloc
Jan 16, 2024
93a9bca
Remove Alloc
Jan 17, 2024
c915503
Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into…
Jan 17, 2024
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
4 changes: 2 additions & 2 deletions paddle/phi/kernels/impl/data_impl.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -27,11 +27,11 @@ template <typename T, typename Context>
void ShadowFeedKernel(const Context& ctx,
const DenseTensor& x,
DenseTensor* out) {
ctx.template Alloc<T>(out);
if (!x.initialized()) {
ctx.template Alloc<T>(out);
return;
}
if (x.place() == out->place()) {
if (x.place() == ctx.GetPlace()) {
Copy link
Contributor

@zhangbo9674 zhangbo9674 Jan 18, 2024

Choose a reason for hiding this comment

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

为什么要改成 ctx.GetPlace？

Copy link
Contributor Author

@0x45f 0x45f Jan 18, 2024

Choose a reason for hiding this comment

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

out未初始化是不是拿不到place？

out->ShareDataWith(x);
out->set_lod(x.lod());
} else {
Expand Down
49 changes: 30 additions & 19 deletions test/legacy_test/test_scatter_op.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -711,25 +711,36 @@ def test_static_graph():
with paddle.static.program_guard(
paddle.static.Program(), paddle.static.Program()
):
x_t = paddle.static.data(name="x", dtype=x.dtype, shape=x.shape)
index_t = paddle.static.data(
name="index", dtype=index.dtype, shape=index.shape
)
updates_t = paddle.static.data(
name="updates", dtype=updates.dtype, shape=updates.shape
)
out_t = paddle.scatter(x_t, index_t, updates_t)
feed = {
x_t.name: x,
index_t.name: index,
updates_t.name: updates,
}
fetch = [out_t]
gpu_exe = paddle.static.Executor(paddle.CUDAPlace(0))
gpu_value = gpu_exe.run(feed=feed, fetch_list=fetch)[0]
return gpu_value

np.testing.assert_array_equal(test_dygraph(), test_static_graph())
scope = paddle.static.Scope()
with paddle.static.scope_guard(scope):
x_t = paddle.static.data(
name="x", dtype=x.dtype, shape=x.shape
)
index_t = paddle.static.data(
name="index", dtype=index.dtype, shape=index.shape
)
updates_t = paddle.static.data(
name="updates", dtype=updates.dtype, shape=updates.shape
)
out_t = paddle.scatter(x_t, index_t, updates_t)
feed = {
x_t.name: x,
index_t.name: index,
updates_t.name: updates,
}
fetch = [out_t]
gpu_exe = paddle.static.Executor(paddle.CUDAPlace(0))
gpu_value = gpu_exe.run(feed=feed, fetch_list=fetch)[0]
scope._remove_from_pool()
return gpu_value

def test_pir_static_graph():
with paddle.pir_utils.IrGuard():
return test_static_graph()

dy_out = test_dygraph()
np.testing.assert_array_equal(dy_out, test_static_graph())
np.testing.assert_array_equal(dy_out, test_pir_static_graph())


@unittest.skipIf(
Expand Down