Skip to content
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

Add support of int16 for gather op. #40052

Merged
merged 8 commits into from
Mar 3, 2022
Merged

Add support of int16 for gather op. #40052

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
1620e68
add support of int16 for gather op.
limin2021 Mar 2, 2022
c7ac4f8
Merge remote-tracking branch 'upstream/develop' into add_gather_int16
limin2021 Mar 2, 2022
8ce6e16
Recover formats.
limin2021 Mar 2, 2022
f3b20ac
Recover formats.
limin2021 Mar 2, 2022
df2f454
fix.
limin2021 Mar 2, 2022
8854e55
Fix format.
limin2021 Mar 2, 2022
99e9777
Fix format.
limin2021 Mar 2, 2022
f98cd51
Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into…
limin2021 Mar 2, 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
8 changes: 8 additions & 0 deletions paddle/fluid/operators/gather_op.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -45,6 +45,8 @@ class GatherOpCUDAKernel : public framework::OpKernel<T> {
axis = static_cast<int>(cpu_axis.data<int32_t>()[0]);
} else if (axis_type == framework::proto::VarType::INT64) {
axis = static_cast<int>(cpu_axis.data<int64_t>()[0]);
} else if (axis_type == framework::proto::VarType::INT16) {
axis = static_cast<int>(cpu_axis.data<int16_t>()[0]);
}
}
const auto &place = ctx.GetPlace();
Expand All @@ -57,6 +59,9 @@ class GatherOpCUDAKernel : public framework::OpKernel<T> {
} else if (index_type == framework::proto::VarType::INT64) {
phi::funcs::GatherV2CUDAFunction<T, int64_t>(x, index, axis, output,
dev_ctx);
} else if (index_type == framework::proto::VarType::INT16) {
phi::funcs::GatherV2CUDAFunction<T, int16_t>(x, index, axis, output,
dev_ctx);
}
return;
}
Expand All @@ -67,6 +72,8 @@ class GatherOpCUDAKernel : public framework::OpKernel<T> {
phi::funcs::GPUGather<T, int>(dev_ctx, *x, *index, output);
} else if (index_type == framework::proto::VarType::INT64) {
phi::funcs::GPUGather<T, int64_t>(dev_ctx, *x, *index, output);
} else if (index_type == framework::proto::VarType::INT16) {
phi::funcs::GPUGather<T, int16_t>(dev_ctx, *x, *index, output);
}
}
};
Expand Down Expand Up @@ -134,6 +141,7 @@ REGISTER_OP_CUDA_KERNEL(gather, ops::GatherOpCUDAKernel<float>,
ops::GatherOpCUDAKernel<double>,
ops::GatherOpCUDAKernel<int64_t>,
ops::GatherOpCUDAKernel<int>,
ops::GatherOpCUDAKernel<int16_t>,
ops::GatherOpCUDAKernel<plat::float16>,
ops::GatherOpCUDAKernel<plat::bfloat16>);
REGISTER_OP_CUDA_KERNEL(gather_grad, ops::GatherGradOpCUDAKernel<float>,
Expand Down
3 changes: 2 additions & 1 deletion python/paddle/tensor/manipulation.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1402,7 +1402,8 @@ def gather(x, index, axis=None, name=None):
return _C_ops.gather(x, index, None, "axis", axis, "overwrite", False)

check_variable_and_dtype(
x, 'x', ['float16', 'float32', 'float64', 'int32', 'int64', 'uint8'],
x, 'x',
['float16', 'float32', 'float64', 'int16', 'int32', 'int64', 'uint8'],
'gather')
check_variable_and_dtype(index, 'index', ['int32', 'int64'], 'gather')

Expand Down