Skip to content

Add C++ unit tests for cadence::fully_connected.out (#19165) #19165

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

Open
hsharma35 wants to merge 2 commits into
base: main
Choose a base branch
from
+13 −9
Open

Add C++ unit tests for cadence::fully_connected.out (#19165) #19165

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
d3387fb
Add C++ unit tests for cadence::quantized_conv1d_ncl.out and quantize…
hsharma35 Apr 28, 2026
3f3d684
Add C++ unit tests for cadence::fully_connected.out (#19165)
hsharma35 Apr 28, 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
5 changes: 3 additions & 2 deletions backends/cadence/generic/operators/op_fully_connected.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -27,7 +27,8 @@ void linear(
Tensor& output) {
const float* __restrict__ input_data = input.const_data_ptr<float>();
const float* __restrict__ weight_data = weight.const_data_ptr<float>();
const float* __restrict__ bias_data = bias.value().const_data_ptr<float>();
const float* __restrict__ bias_data =
bias.has_value() ? bias.value().const_data_ptr<float>() : nullptr;
float* __restrict__ output_data = output.mutable_data_ptr<float>();

// input comes in shape [batch_size, in_dim]
Expand All @@ -43,7 +44,7 @@ void linear(

for (int i = 0; i < leading_dims; ++i) {
for (int j = 0; j < M; ++j) {
float sum = bias_data[j];
float sum = bias_data != nullptr ? bias_data[j] : 0.0f;
for (int k = 0; k < N; ++k) {
sum += input_data[i * N + k] * weight_data[j * N + k];
}
Expand Down
11 changes: 7 additions & 4 deletions backends/cadence/hifi/operators/op_quantized_conv1d_ncl.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -240,7 +240,10 @@ void xa_opt_quantized_conv1d_ncl_asym8uxsym8u_asym8u(
WORD32 x_stride = stride[0];
WORD32 x_padding = padding[0];
WORD32 input_zero_bias = -in_zero_point;
WORD32 out_multiplier32 = bias_scale * (1. / output_scale) * 2147483648;
const float eff_scale = bias_scale * (1.0f / output_scale);
WORD32 out_multiplier32 = (eff_scale >= 1.0f)
? static_cast<WORD32>(2147483647)
: static_cast<WORD32>(eff_scale * 2147483648.0f);
WORD32 out_shift32 = 0;
WORD32 kernel_zero_bias = -weight_zero_point;

Expand Down Expand Up @@ -419,9 +422,9 @@ void quantized_conv1d_ncl_per_tensor_out(
out);
}
} else if (dtype == ScalarType::Byte) {
// HiFi nnlib conv1d_std kernel does not support depthwise (groups > 1).
// Fall back to generic implementation.
if (groups > 1) {
// HiFi nnlib conv1d_std kernel does not support depthwise (groups > 1)
// or stride > 1. Fall back to generic implementation.
if (groups > 1 || stride[0] > 1) {
impl::generic::native::quantized_conv1d_ncl_per_tensor_out(
ctx,
input,
Expand Down
6 changes: 3 additions & 3 deletions backends/cadence/hifi/operators/op_quantized_conv1d_nlc.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -298,9 +298,9 @@ void quantized_conv1d_nlc_per_tensor_out(
out);
}
} else if (dtype == ScalarType::Byte) {
// HiFi nnlib conv1d_std kernel does not support depthwise (groups > 1).
// Fall back to generic implementation.
if (groups > 1) {
// HiFi nnlib conv1d_std kernel does not support depthwise (groups > 1)
// or stride > 1. Fall back to generic implementation.
if (groups > 1 || stride[0] > 1) {
impl::generic::native::quantized_conv1d_nlc_per_tensor_out(
ctx,
input,
Expand Down
Loading