Build failure (acl_lrn_layer.o)

[kmonachopoulos]

Hello,
i am trying to deploy caffe on Firefly RK3288 following the steps from "release_notes" that you provide. I managed to install ARM ComputeLibrary successfully, but when I am trying to build caffe I am getting the following error:

CXX src/caffe/layers/acl_lrn_layer.cpp
src/caffe/layers/acl_lrn_layer.cpp: In member function 'virtual void caffe::ACLLRNLayer<Dtype>::SetupACLLayer(const std::vector<caffe::Blob<Dtype>*>&, const std::vector<caffe::Blob<Dtype>*>&)':
src/caffe/layers/acl_lrn_layer.cpp:31:45: error: 'IN_MAP' is not a member of 'arm_compute::NormType'
        norm_info=new NormalizationLayerInfo(NormType::IN_MAP, this->size_, this->alpha_, this->beta_, this->k_);
                                             ^~~~~~~~
Makefile:632: recipe for target '.build_release/src/caffe/layers/acl_lrn_layer.o' failed
make: *** [.build_release/src/caffe/layers/acl_lrn_layer.o] Error 1

The example that you are providing concerns firefly 3399 board and not the firefly 3288 that i have, so is there any extra dependencies that I am missing?

P.S Inside the enumerator class NormType there is not a IN_MAP member.

enum class NormType
{
    IN_MAP_1D, /**< Normalization applied within the same map in 1D region */
    IN_MAP_2D, /**< Normalization applied within the same map in 2D region */
    CROSS_MAP  /**< Normalization applied cross maps */
};

Solved : Using Version ComputeLibrary : v17.05 solves the problem (I should read the manual carefully). I just saw that recently it has been a change in the recently ComputeLibrary newest version.

Thanks

[joey2014]

kmonachopoulos, ARM new ACL changed the name IN_MAP to IN_MAP_1D/2D which causes the the code can't pass compile . The issues can be fixed as below:
diff --git a/src/caffe/layers/acl_lrn_layer.cpp b/src/caffe/layers/acl_lrn_layer.cpp
index 9c03cad..6e5c7bc 100644
--- a/src/caffe/layers/acl_lrn_layer.cpp
+++ b/src/caffe/layers/acl_lrn_layer.cpp
@@ -5,6 +5,7 @@

namespace caffe {

+const NormType IN_MAP=(arm_compute::NormType)0;
template
void ACLLRNLayer::LayerSetUp(const vector<Blob>& bottom,
const vector<Blob>& top) {
@@ -28,7 +29,7 @@ void ACLLRNLayer::SetupACLLayer(const vector<Blob*>& bottom,
//this->force_bypass_acl_path_=false;
NormalizationLayerInfo *norm_info;
if(this->layer_param_.lrn_param().norm_region() == LRNParameter_NormRegion_WITHIN_CHANNEL)
- norm_info=new NormalizationLayerInfo(NormType::IN_MAP, this->size_, this->alpha_, this->beta_, this->k_);
+norm_info=new NormalizationLayerInfo(IN_MAP, this->size_, this->alpha_, this->beta_, this->k_);
else
norm_info=new NormalizationLayerInfo(NormType::CROSS_MAP, this->size_, this->alpha_, this->beta_, this->k_);

[openailab-sh]

Thanks, kmonachopoulos, the root cause is just like what joey2014 said. To fix the issue with the new ACL version 17.06, you can check out the new code from master

[honggui]

I closed the issue due to it's fixed