AMD HIP Programming Guide: The Path to Cross-Platform GPU Computing

The Heterogeneous-compute Interface for Portability (HIP) provides a C++ runtime API that abstracts hardware layers to offer a unified codebase for both NVIDIA and AMD GPUs. This path eliminates vendor lock-in by mirroring the CUDA environment while targeting the ROCm backend.

1. Environment Configuration

Initialization begins with setting environment variables to point the hipcc compiler to the correct toolchain:

$ export HIP_PATH=[MYHIP]
$ export PATH=$PATH:[MYHIP]/bin

2. Standardizing the Toolchain

The hipcc compiler acts as a smart wrapper. For build automation, the HIP_PATH ?= $(shell hipconfig --path) logic ensures your Makefile dynamically finds the HIP installation regardless of the system configuration.

3. Versioning Logic

HIP uses a deterministic versioning formula to handle feature availability programmatically across releases:

$$\text{HIP\_VERSION} = \text{HIP\_VERSION\_MAJOR} \times 10^7 + \text{HIP\_VERSION\_MINOR} \times 10^5 + \text{HIP\_VERSION\_PATCH}$$

4. Core Runtime Equivalents

HIP provides hipMalloc and hipLaunchKernel as functional mirrors to CUDA's memory and execution calls, enabling a "single-source" development philosophy.

TERMINAL bash — 80x24

> Ready. Click "Run" to execute.

QUESTION 1

Which compiler driver is used to abstract both NVCC and Clang for HIP development?

rocm-cc

hipcc

nvcc-wrap

clang-hip

✅ Correct!

The hipcc tool detects the platform and invokes the appropriate backend compiler.

❌ Incorrect

HIP uses hipcc as the primary compiler driver for both NVIDIA and AMD platforms.

QUESTION 2

What header is mandatory for basic HIP runtime functionality?

rocm_runtime.h

hip/hip_runtime.h

cuda_to_hip.h

hip_portability.h

✅ Correct!

Including hip/hip_runtime.h provides the equivalent types and functions found in the CUDA runtime.

❌ Incorrect

The standard HIP runtime entry point is hip/hip_runtime.h.

QUESTION 3

In the HIP versioning formula, what is the multiplier for the MINOR version?

1,000,000

100,000

10,000

1,000

QUESTION 4

Which command is used within a Makefile to dynamically find the HIP installation path?

hipconfig --path

rocm-path

which hipcc

hip-env --get-path

✅ Correct!

hipconfig --path is the standard way to retrieve the installation directory for build scripts.

❌ Incorrect

The hipconfig utility provides various configuration details, including the installation path.

QUESTION 5

What is the purpose of hipGetDeviceProperties?

To allocate memory on the GPU.

To compile a kernel at runtime.

To inspect GPU capabilities and hardware attributes.

To synchronize the host and device.

Porting the 'Square' Sample

Practical migration scenario from CUDA to ROCm.

A developer has a legacy CUDA application located at `~/cuda/samples/square`. They want to port this to an AMD system using HIP. They need to set up the environment, build the project using standard samples, and verify the runtime version.

1. What environment variables must be exported before running the 'make' command?

Solution:
The developer must export HIP_PATH (pointing to the ROCm/HIP installation) and update the PATH variable to include the /bin directory of the HIP installation so hipcc can be executed.

2. Inside the sample directory, what is the typical sequence of commands to compile and run the ported square application?

Solution:
The sequence is: cd ~/hip/samples/0_Intro/square, followed by make (which uses hipcc), and finally executing the binary with ./square.out.

3. How does the application identify its runtime environment programmatically?

Solution:
By calling hipRuntimeGetVersion(&runtimeVersion), the application can determine the specific version of the HIP backend it is executing on, allowing for conditional logic if certain features are version-dependent.