Fabian Tschopp

@bhack No the way the code generation works in libDNN is too different from the CLtune approach I think. Yes, when you use CLCudaAPI you can also make use of...

@doe300 Thanks for your answer :) - You are not using the right OpenCL Caffe branch (you're using the discontinued one from AMD, I think). The official ones and my...

No... actually it fails: `./build/VC4C --hex -o ../example/test.clptx ../example/hello_world.cl` ``` [E] Tue Feb 13 00:59:56 2018: (1) /home/pi/eth_bsc/VC4/VC4C/build/build/libVC4CC.so.1.2 : vc4c::CompilationError::CompilationError(vc4c::CompilationStep, std::__cxx11::basic_string const&)+0xac [0x76c918c4] [E] Tue Feb 13 00:59:56 2018: (2)...

This is the result on Caffe: ``` I0213 20:23:48.559901 3692 caffe.cpp:397] Use GPU with device ID 0 I0213 20:23:48.621196 3692 ocl_device.cpp:61] CL_DEVICE_HOST_UNIFIED_MEMORY: 1 E0213 20:23:48.701433 3692 ocl_device_program.cpp:113] Failed to compile...

This happens: ``` [D] Wed Feb 14 19:39:21 2018: Mapping LLVM instructions to immediates: [D] Wed Feb 14 19:39:21 2018: Generating label label %0 [D] Wed Feb 14 19:39:21 2018:...

OK I managed to compile the kernel standalone. Still no luck with Caffe though (device not available, as above). I also noticed that `clinfo` reports: ``` Number of platforms 1...

@doe300 Minimal buffer size: No, that will depend entirely on what kind of network is used in Caffe. I do not know enough about VC4 GPUs to understand if there...

@beniz Sure, give it a go and ask if you need to know something. First you'll have to set up a cross-compiling environment for ARMv7, maybe with docker, and set...

@sh1r0 It should be possible to get BVLC/caffe#2610 to work on android. It can probably be done by replacing the Caffe used in this project by the https://github.com/naibaf7/caffe branch and...

@sh1r0 Yes and I guess you would need to use 32 bit indexing (pointers) instead of 64 bit indexing for Android devices. What requirements would you have to be able...