Skip to content

Albert/fix conv2d #289

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
real-zhangzhe merged 2 commits into from
Jun 23, 2026
Merged

Albert/fix conv2d #289

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
a02c830
Fix(Conv2d)
Jun 23, 2026
0a007af
Fix(Conv2d)
Jun 23, 2026
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
14 changes: 2 additions & 12 deletions musa_ext/kernels/math/musa_conv2d_op.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -160,16 +160,7 @@ Status RunMusaConv2D(OpKernelContext* ctx, const Tensor& input,
static_cast<int>(status));
}

mConvolution::Algorithm algo;
status = conv.GetRecommendForwardAlgorithm(handle, algo, y, x, w);
if (status != mStatus::SUCCESS) {
return errors::Internal(
"muDNN Convolution::GetRecommendForwardAlgorithm failed. status=",
static_cast<int>(status), ", data_format=NHWC",
", input_shape=", input.shape().DebugString(),
", filter_shape=", filter.shape().DebugString(),
", output_shape=", output->shape().DebugString());
}
const mConvolution::Algorithm algo = mConvolution::Algorithm::IMPLICIT_GEMM;

size_t workspace_size = 0;
status = conv.GetForwardWorkspaceSize(handle, workspace_size, y, x, w, algo);
Expand Down Expand Up @@ -325,8 +316,7 @@ class MusaConv2DOp : public MusaOpKernel {
return;
}

const bool use_tf32 =
tf32_enabled_ && dilation_h_ == 1 && dilation_w_ == 1;
const bool use_tf32 = tf32_enabled_ && dilation_h_ == 1 && dilation_w_ == 1;

if (data_format_ == FORMAT_NHWC) {
OP_REQUIRES_OK(
Expand Down
11 changes: 11 additions & 0 deletions test/ops/conv2d_op_test.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -155,6 +155,17 @@ def testConv2DPointwise1x1NHWC(self):
padding="SAME",
data_format="NHWC")

def testConv2DResNet50Conv4ShapeNHWC(self):
"""ResNet50 conv4_block1_2 shape should not use an unstable algorithm."""
self._test_conv2d(
input_shape=[1, 14, 14, 256],
filter_shape=[3, 3, 256, 256],
dtype=tf.float32,
strides=[1, 1, 1, 1],
padding="SAME",
data_format="NHWC",
seed=2027)

def testConv2DValidNCHW(self):
"""NCHW + VALID + stride=1."""
for dtype in [tf.float32, tf.float16]:
Expand Down
Loading