Updating README by introducing HaloBox into OS Environment setup

Author: ldokovic-personal

playbooks/supplemental/pytorch-kernels/README.md

@@ -120,13 +120,18 @@ PyTorch also exposes `torch.cuda._compile_kernel()`, a high-level shortcut to JI
 ---
 
 ## Setup
-<!-- @os:windows -->
-### Prerequisites - Windows
-- Install latest: [AMD Adrenalin Software](https://www.amd.com/en/products/software/adrenalin.html)
+
+### System Detection
+Before setting up, the script checks if the system is a HaloBox. If it is, some steps, such as creating a virtual environment, are skipped as HaloBox comes with pre-installed driver.
+
+<!-- @os:halobox -->
+### HaloBox: Skipping Virtual Environment Setup
+No need to set up a virtual environment as necessary driver and configurations are pre-installed.
 <!-- @os:end -->
 
 ### Create a Virtual Environment
 <!-- @os:linux -->
+#### Linux
 ```bash
 sudo apt install -y python3-venv
 python3 -m venv ~/rocm-env
@@ -136,6 +141,12 @@ source ~/rocm-env/bin/activate
 <!-- @os:end -->
 
 <!-- @os:windows -->
+### Prerequisites - Windows
+- Install latest: [AMD Adrenalin Software](https://www.amd.com/en/products/software/adrenalin.html)
+<!-- @os:end -->
+
+<!-- @os:windows -->
+#### Windows
 ```bash
 python -m venv rocm-env
 rocm-env\Scripts\activate
@@ -146,7 +157,15 @@ rocm-env\Scripts\activate
 ---
 
 ### Installing Dependencies
+<!-- @os:halobox -->
+#### HaloBox
+```bash
+pip install --upgrade pip setuptools wheel --break-system-packages
+```
+<!-- @os:end -->
+
 <!-- @os:linux -->
+#### Linux
 ```bash
 source ~/rocm-env/bin/activate
 
@@ -161,6 +180,7 @@ pip install --pre --index-url https://rocm.nightlies.amd.com/v2/gfx1151/ torch==
 <!-- @os:end -->
 
 <!-- @os:windows -->
+#### Windows
 ```bash
 rocm-env\Scripts\activate
 
@@ -179,6 +199,7 @@ pip install --pre --index-url https://rocm.nightlies.amd.com/v2/gfx1151/ torch==
 
 #### Set Environment Variables
 <!-- @os:linux -->
+#### Linux
 ```bash
 rocm-sdk init # Initialize the devel libraries
 
@@ -188,7 +209,18 @@ export PATH = "$ROCM_HOME/bin:$PATH"
 ```
 <!-- @os:end -->
 
+<!-- @os:linux&halobox -->
+#### Linux & HaloBox
+```bash
+# Set compiler and build settings
+export CC=clang
+export CXX=clang
+export DISTUTILS_USE_SDK=1
+```
+<!-- @os:end -->
+
 <!-- @os:windows -->
+#### Windows
 ```bash
 rocm-sdk init # Initialize the devel libraries
 
@@ -299,13 +331,14 @@ On Windows, `rocm-smi` is not supported. To track GPU utilization, you can use T
 The full manual path: write the kernel and Python binding in a single `.cu` file, compile it as a native extension using PyTorch's build system, then import and call it from Python.
 
 **Files:**
-
+#### Windows
 <!-- @os:windows -->
 | File | Role |
 |---|---|
 | [add_one_kernel.cu](assets/Vector_Addition/add_one_kernel.cu) | Kernel + launcher + pybind11 binding, everything in one file |
 | [setup.py](assets/Vector_Addition/setup.py) | Build script, uses `CUDAExtension` to compile the `.cu` into a `.pyd` |
 <!-- @os:end -->
+#### Linux
 <!-- @os:linux -->
 | File | Role |
 |---|---|
@@ -349,16 +382,26 @@ the CPU immediately continues executing the next instruction without waiting for
 
 **Step 2: Build**
 
+<!-- @os:halobox -->
+#### HaloBox
+```bash
+pip install --no-build-isolation -v . --break-system-packages
+```
+<!-- @os:end -->
+
+#### Linux & Windows
 ```bash
 pip install --no-build-isolation -v .
 ```
 
 `CUDAExtension` is a CUDA build helper from `torch.utils.cpp_extension`. On AMD with ROCm, PyTorch **remaps `CUDAExtension` to use `hipcc`** instead of `nvcc`, so the same `setup.py` that would build a CUDA extension on NVIDIA compiles to AMD GPU code without any changes. This is the key mechanism that makes CUDA extension code portable to AMD: PyTorch's ROCm build intercepts the build path and routes it through the HIP compiler. Produces these in the same directory:
 <!-- @os:windows -->
+#### Windows
 - `build/`:  directory with the `.pyd` files
 - `add_one_kernel.hip`:  the HIP source generated by hipifying the `.cu` file; this is what `hipcc` actually compiled
 <!-- @os:end -->
 <!-- @os:linux -->
+#### Linux
 - `build/`:  directory with the `.so` files
 - `add_one_kernel.hip`:  the HIP source generated by hipifying the `.cu` file; this is what `hipcc` actually compiled
 <!-- @os:end -->
@@ -530,12 +573,14 @@ Average GPU Utilization: 55.00%
 The full manual path: write the kernel and Python binding in a `.cu` file, compile it as a native extension, then import and call it from Python. Mirrors the structure of `add_one_kernel.cu` exactly, only the kernel signature and launcher logic differ.
 
 **Files:**
+#### Windows
 <!-- @os:windows -->
 | File | Role |
 |---|---|
 | [matmul_kernel.cu](assets/Matrix_Multiplication/matmul_kernel.cu) | Kernel + launcher + pybind11 binding |
 | [setup.py](assets/Matrix_Multiplication/setup.py) | Build script, uses `CUDAExtension` to compile the `.cu` into a `.pyd` |
 <!-- @os:end -->
+#### Linux
 <!-- @os:linux -->
 | File | Role |
 |---|---|
@@ -591,16 +636,26 @@ Compared to `add_one_launcher` in Walkthrough 1, the launcher here:
 
 **Step 2: Build**
 
+<!-- @os:halobox -->
+#### HaloBox
+```bash
+pip install --no-build-isolation -v . --break-system-packages
+```
+<!-- @os:end -->
+
+#### Linux & Windows
 ```bash
 pip install --no-build-isolation -v .
 ```
 
 Produces these in the same directory:
 <!-- @os:windows -->
+#### Windows
 - `build/`:  directory with the `.pyd` files
 - `matmul_kernel.hip`:  the HIP source generated by hipifying the `.cu` file; this is what `hipcc` actually compiled
 <!-- @os:end -->
 <!-- @os:linux -->
+#### Linux
 - `build/`:  directory with the `.so` files
 - `matmul_kernel.hip`:  the HIP source generated by hipifying the `.cu` file; this is what `hipcc` actually compiled
 <!-- @os:end -->