Skip to content

[Relay] Change when int8 operations are converted to int16 on Arm #12671

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
Mousius merged 2 commits into from
Sep 8, 2022
Merged

[Relay] Change when int8 operations are converted to int16 on Arm #12671

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
d65b3f4
Allow int8 operations for Cortex-M cores
guberti Sep 1, 2022
7a4c52d
Fix conversion requirements
guberti Sep 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
22 changes: 14 additions & 8 deletions python/tvm/relay/qnn/op/legalizations.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -424,26 +424,32 @@ def is_aarch64_arm():

@qnn_conv2d_legalize.register("arm_cpu")
def _qnn_conv2d_legalize_arm_cpu(attrs, inputs, types):
# ARM prefers the dtypes to be same.
target = tvm.target.Target.current(allow_none=False)
has_asimd = is_aarch64_arm() or "+neon" in target.mattr
is_depthwise = relay.op.strategy.is_depthwise_conv2d(
types[0].shape,
attrs["data_layout"],
types[1].shape,
attrs["kernel_layout"],
attrs["groups"],
)
use_int8_on_arm = (not is_depthwise) and is_aarch64_arm() and attrs["data_layout"] == "NHWC"
if use_int8_on_arm or is_fast_int8_on_arm():
return helper_change_dtypes_to_be_same(attrs, inputs, types, relay.qnn.op.conv2d)
return helper_no_fast_int8_hw_legalization(attrs, inputs, types, relay.nn.conv2d)
use_int8_on_arm = (not is_depthwise) and attrs["data_layout"] == "NHWC"
has_dotprod = is_fast_int8_on_arm()
other_options = use_int8_on_arm or has_dotprod
if has_asimd and not other_options:
return helper_no_fast_int8_hw_legalization(attrs, inputs, types, relay.nn.conv2d)
# ARM prefers the dtypes to be same.
return helper_change_dtypes_to_be_same(attrs, inputs, types, relay.qnn.op.conv2d)


@qnn_dense_legalize.register("arm_cpu")
def _qnn_dense_legalize_arm_cpu(attrs, inputs, types):
target = tvm.target.Target.current(allow_none=False)
has_asimd = is_aarch64_arm() or "+neon" in target.mattr
if has_asimd and not is_fast_int8_on_arm():
return helper_no_fast_int8_hw_legalization(attrs, inputs, types, relay.nn.dense)
# ARM prefers the dtypes to be same.
if is_fast_int8_on_arm():
return helper_change_dtypes_to_be_same(attrs, inputs, types, relay.qnn.op.dense)
return helper_no_fast_int8_hw_legalization(attrs, inputs, types, relay.nn.dense)
return helper_change_dtypes_to_be_same(attrs, inputs, types, relay.qnn.op.dense)


##########################
Expand Down