Skip to content

[TOPI,CUDA] Don't enable cudnn conv2d kernel if is not supported #10021

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
masahi merged 2 commits into from
Jan 22, 2022
Merged

[TOPI,CUDA] Don't enable cudnn conv2d kernel if is not supported #10021

Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
db9815d
[TOPI,CUDA] Don't enable cudnn conv2d kernel if is not supported
Jan 21, 2022
1afa1ae
remove all conv2d_cudnn int8 support
Jan 21, 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: 7 additions & 1 deletion python/tvm/relay/op/strategy/cuda.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -313,6 +313,7 @@ def conv2d_strategy_cuda(attrs, inputs, out_type, target):
and layout in ["NCHW", "NHWC"]
and padding[0] == padding[2]
and padding[1] == padding[3]
and not ((data.dtype == "int8" or kernel.dtype == "int8") and layout == "NCHW")
):
# add cudnn implementation
if layout == "NHWC":
Expand Down Expand Up @@ -347,7 +348,12 @@ def conv2d_strategy_cuda(attrs, inputs, out_type, target):
# add cudnn implementation, if any
cudnn_impl = False
if target.kind.name == "cuda" and "cudnn" in target.libs:
if layout in ["NCHW", "NHWC"] and padding[0] == padding[2] and padding[1] == padding[3]:
if (
layout in ["NCHW", "NHWC"]
and padding[0] == padding[2]
and padding[1] == padding[3]
and not ((data.dtype == "int8" or kernel.dtype == "int8") and layout == "NCHW")
):
strategy.add_implementation(
wrap_compute_conv2d(
topi.cuda.conv2d_cudnn, need_data_layout=True, has_groups=True
Expand Down