When is TIRToRuntime called for OpenCL?

Questions
#1

Hi all,

I want to know where this code is called. Specially when we use OpenCL. (The TIRToRuntime function) path/to/tvm/src/relay/backend/contrib/uma/tir_to_runtime.cc

I am trying to understand how we can generate OpenCL kernel code in TVM workflow and finally I got this part return what I want but I do not know when it is used.

Best, Hadi

#2

This code must not be called for OpenCL flow.

I am trying to understand how we can generate OpenCL kernel code in TVM workflow and finally I got this part return what I want but I do not know when it is used.

The generic approach to use target=“opencl” and target_host=“llvm …” where ... is something for your platform. For example for native it is enough to have target_host jsut “llvm” in case of cross compilation for android “llvm -mtriple=arm64-linux-android”

And regarding the target - there are some vendor dependent keys like

target = "opencl -device=adreno" for Qualcomm Adreno

target="opencl -device=mali" for ARM Mali graphics

target="opencl -device=intel" for Intel graphics

2 Likes