experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Fix depthwise_conv test. Looks somehow they are not really tested. :(

Browse Source 
PiperOrigin-RevId: 261816763
This commit is contained in:
Renjie Liu 2019-08-05 18:41:43 -07:00 committed by TensorFlower Gardener
parent 4add221b50
commit 2d6cca0122
1 changed files with 6 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
tensorflow/lite/kernels/depthwise_conv_test.cc
View File

@ -616,6 +616,8 @@ class QuantizedDepthwiseConvolutionOpTest : public SingleOpTest {
}
};
// Only enable this test for neon.
#ifdef USE_NEON
TEST_F(QuantizedDepthwiseConvolutionOpTest, LargeOutputChannelTest) {
const TensorData input({TensorType_UINT8, {1, 4, 4, 2400}, -63.5, 64});
const TensorData filter({TensorType_UINT8, {1, 3, 3, 2400}, -63.5, 64});
@ -646,6 +648,7 @@ TEST_F(QuantizedDepthwiseConvolutionOpTest, LargeOutputChannelTest) {
reference_impl.SetInput(input_data);
reference_impl.SetFilter(filter_data);
reference_impl.SetBias(bias_data);
reference_impl.Invoke();
QuantizedDepthwiseConvolutionOpModel optimized_impl(
ops::builtin::Register_DEPTHWISE_CONVOLUTION_GENERIC_OPT(), input, filter,
@ -653,9 +656,11 @@ TEST_F(QuantizedDepthwiseConvolutionOpTest, LargeOutputChannelTest) {
optimized_impl.SetInput(input_data);
optimized_impl.SetFilter(filter_data);
optimized_impl.SetBias(bias_data);
optimized_impl.Invoke();
// EXPECT_THAT(reference_impl.GetOutput(), optimized_impl.GetOutput());
EXPECT_THAT(reference_impl.GetOutput(), optimized_impl.GetOutput());
}
#endif
// In this test we set the input and output scales so that the results match
// exactly the 'non-quantized' version.