62fee66ff2
The existing workgroup calculation logic for GWS initialization is
incorrect. It tries to add together workgroups across dimensions,
leading to major under-count in 2D and 3D kernels. An (x,y,z) kernel
uses x * y * z blocks, not x + y + z.
In addition, the previous logic was incorrect for the case of launching
a single-threaded kernel. It calculated 0 workgroups, leading to
initializing GWS to -1.
Change-Id: I1bb20a0d5b6e0cc10ac55901c28d8f93aac61c09
[ROCm/clr commit: 54d1d69c0a]
ROCclr - Radeon Open Compute Common Language Runtime
ROCclr is a virtual device interface that compute runtimes interact with to different backends such as ROCr or PAL This abstraction allows runtimes to work on Windows as well as on Linux without much effort.
Repository branches
The repository maintains several branches. The branches that are of importance are:
- master: This is the default branch.
Building
Prerequisites
- Install mesa-common-dev
- Either build or install COMGR, CLANG and Device Library
Getting the source code
git clone https://github.com/ROCm-Developer-Tools/ROCclr.git
git clone -b master-next https://github.com/RadeonOpenCompute/ROCm-OpenCL-Runtime.git
Set the environment variables
export ROCclr_DIR="$(readlink -f ROCclr)"
export OPENCL_DIR="$(readlink -f ROCm-OpenCL-Runtime)"
Build ROCclr
Here is command to build ROCclr:
cd "$ROCclr_DIR"
mkdir -p build; cd build
cmake -DOPENCL_DIR="$OPENCL_DIR" -DCMAKE_INSTALL_PREFIX=/opt/rocm/rocclr ..
make -j$(nproc)
sudo make install
Optional steps to build HIP runtime
Enter the directory where git cloned the ROCClr and OpenCL. Run the following commands:
git clone -b master-next https://github.com/ROCm-Developer-Tools/HIP.git
export HIP_DIR="$(readlink -f HIP)"
cd "$HIP_DIR"
mkdir -p build; cd build
cmake -DHIP_COMPILER=clang -DHIP_PLATFORM=rocclr -DCMAKE_PREFIX_PATH="$ROCclr_DIR/build" ..
make -j$(nproc)
Release build
For release build, add "-DCMAKE_BUILD_TYPE=Release" to the cmake command line.