|
|
|
@@ -105,8 +105,8 @@ int Convolution_x86::create_pipeline(const Option& opt) |
|
|
|
// conv3x3s1_winograd23_transform_kernel_int8_sse(weight_data, weight_3x3_winograd23_data, num_input, num_output); |
|
|
|
conv3x3s1_winograd43_transform_kernel_int8_sse(weight_data, weight_3x3_winograd23_data, num_input, num_output); |
|
|
|
else |
|
|
|
// conv3x3s1_winograd23_transform_kernel_sse(weight_data, weight_3x3_winograd23_data, num_input, num_output); |
|
|
|
conv3x3s1_winograd43_transform_kernel_sse(weight_data, weight_3x3_winograd43_data, num_input, num_output); |
|
|
|
conv3x3s1_winograd23_transform_kernel_sse(weight_data, weight_3x3_winograd23_data, num_input, num_output); |
|
|
|
// conv3x3s1_winograd43_transform_kernel_sse(weight_data, weight_3x3_winograd43_data, num_input, num_output); |
|
|
|
} |
|
|
|
|
|
|
|
if (use_int8_inference == false) |
|
|
|
@@ -616,8 +616,8 @@ int Convolution_x86::forward(const Mat& bottom_blob, Mat& top_blob, const Option |
|
|
|
|
|
|
|
if (use_winograd3x3 && outw >= 8 && outh >=8) |
|
|
|
{ |
|
|
|
// conv3x3s1_winograd23_sse(bottom_blob_bordered, top_blob, weight_3x3_winograd23_data, bias_data, opt); |
|
|
|
conv3x3s1_winograd43_sse(bottom_blob_bordered, top_blob, weight_3x3_winograd43_data, bias_data, opt); |
|
|
|
conv3x3s1_winograd23_sse(bottom_blob_bordered, top_blob, weight_3x3_winograd23_data, bias_data, opt); |
|
|
|
// conv3x3s1_winograd43_sse(bottom_blob_bordered, top_blob, weight_3x3_winograd43_data, bias_data, opt); |
|
|
|
} |
|
|
|
else |
|
|
|
//conv(bottom_blob_bordered, top_blob, weight_data, bias_data, opt); |
|
|
|
|