[GitHub] [incubator-tvm] giuseros edited a comment on issue #6857: Not able to compile TVM with Arm Compute library

[GitBox <gi...@apache.org>]

giuseros edited a comment on issue #6857:
URL: https://github.com/apache/incubator-tvm/issues/6857#issuecomment-723035016


   Hi @joyalbin ,
   So Arm Compute Library is a runtime library, and it is meant to run on Arm machines. The usual setup is the following:
   - A: is the machine where you are actually compiling your network. Which is usually a x86 machine. 
   - B: is the machine where you run the network (e.g., through RPC). This should be an Arm machine. 
   
   About the flags:
   - On machine A, you need build TVM with USE_ARM_COMPUTE_LIB  to ON. In that way you tell the (Relay) compiler  to annotate the Arm Compute LIbrary supported operators. On machine A USE_ARM_COMPUTE_LIB_GRAPH_RUNTIME has to be set to OFF. This is because that machine won't be able to build Arm Compute Library, being an x86 machine. 
   
   - On machine B , you need to only build the runtime with  USE_ARM_COMPUTE_LIB_GRAPH_RUNTIME to ON. This is telling the runtime that whenever it sees an annotated operator, it should run it through the Arm Compute Library. Just to be clear, on machine B, you usually build only the runtime part of TVM (i.e., `make -j4 runtime`)
   
   Some more info about all this: 
   - https://tvm.apache.org/2020/07/15/how-to-bring-your-own-codegen-to-tvm
   - https://tvm.apache.org/docs/deploy/arm_compute_lib.html?highlight=acl
   - https://discuss.tvm.apache.org/t/rfc-byoc-arm-compute-library-integration/7082
   
   Hope this helps


----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

For queries about this service, please contact Infrastructure at:
users@infra.apache.org