HIP Support

HIP (Heterogeneous-Compute Interface for Portability) https://github.com/ROCm-Developer-Tools/HIP is a C++ Runtime API and Kernel Language. It enables developers to create portable applications for offloading computation to different hardware platforms from a single source code.

AMD GPU Support

Clang provides HIP support on AMD GPUs via the ROCm platform https://rocm.docs.amd.com/en/latest/#. The ROCm runtime forms the base for HIP host APIs, while HIP device APIs are realized through HIP header files and the ROCm device library. The Clang driver uses the HIPAMD toolchain to compile HIP device code to AMDGPU ISA via the AMDGPU backend. The compiled code is then bundled and embedded in the host executables.

Intel GPU Support

Clang provides partial HIP support on Intel GPUs using the CHIP-Star project https://github.com/CHIP-SPV/chipStar. CHIP-Star implements the HIP runtime over oneAPI Level Zero or OpenCL runtime. The Clang driver uses the HIPSPV toolchain to compile HIP device code into LLVM IR, which is subsequently translated to SPIR-V via the SPIR-V backend or the out-of-tree LLVM-SPIRV translator. The SPIR-V is then bundled and embedded into the host executables.

Note

While Clang does not directly provide HIP support for NVIDIA GPUs and CPUs, these platforms are supported via other means:

Example Usage

To compile a HIP program, use the following command:

clang++ -c --offload-arch=gfx906 -xhip sample.cpp -o sample.o

The -xhip option indicates that the source is a HIP program. If the file has a .hip extension, Clang will automatically recognize it as a HIP program:

clang++ -c --offload-arch=gfx906 sample.hip -o sample.o

To link a HIP program, use this command:

clang++ --hip-link --offload-arch=gfx906 sample.o -o sample

In the above command, the --hip-link flag instructs Clang to link the HIP runtime library. However, the use of this flag is unnecessary if a HIP input file is already present in your program.

For convenience, Clang also supports compiling and linking in a single step:

clang++ --offload-arch=gfx906 -xhip sample.cpp -o sample

In the above commands, gfx906 is the GPU architecture that the code is being compiled for. The supported GPU architectures can be found in the AMDGPU Processor Table. Alternatively, you can use the amdgpu-arch tool that comes with Clang to list the GPU architecture on your system:

amdgpu-arch

You can use --offload-arch=native to automatically detect the GPU architectures on your system:

clang++ --offload-arch=native -xhip sample.cpp -o sample

Path Setting for Dependencies

Compiling a HIP program depends on the HIP runtime and device library. The paths to the HIP runtime and device libraries can be specified either using compiler options or environment variables. The paths can also be set through the ROCm path if they follow the ROCm installation directory structure.

Order of Precedence for HIP Path

  1. --hip-path compiler option

  2. HIP_PATH environment variable (use with caution)

  3. --rocm-path compiler option

  4. ROCM_PATH environment variable (use with caution)

  5. Default automatic detection (relative to Clang or at the default ROCm installation location)

Order of Precedence for Device Library Path

  1. --hip-device-lib-path compiler option

  2. HIP_DEVICE_LIB_PATH environment variable (use with caution)

  3. --rocm-path compiler option

  4. ROCM_PATH environment variable (use with caution)

  5. Default automatic detection (relative to Clang or at the default ROCm installation location)

Compiler Option

Environment Variable

Description

Default Value

--rocm-path=<path>

ROCM_PATH

Specifies the ROCm installation path.

Automatic detection

--hip-path=<path>

HIP_PATH

Specifies the HIP runtime installation path.

Determined by ROCm directory structure

--hip-device-lib-path=<path>

HIP_DEVICE_LIB_PATH

Specifies the HIP device library installation path.

Determined by ROCm directory structure

Note

We recommend using the compiler options as the primary method for specifying these paths. While the environment variables ROCM_PATH, HIP_PATH, and HIP_DEVICE_LIB_PATH are supported, their use can lead to implicit dependencies that might cause issues in the long run. Use them with caution.

Predefined Macros

Macro

Description

__CLANG_RDC__

Defined when Clang is compiling code in Relocatable Device Code (RDC) mode. RDC, enabled with the -fgpu-rdc compiler option, is necessary for linking device codes across translation units.

__HIP__

Defined when compiling with HIP language support, indicating that the code targets the HIP environment.

__HIPCC__

Alias to __HIP__.

__HIP_DEVICE_COMPILE__

Defined during device code compilation in Clang’s separate compilation process for the host and each offloading GPU architecture.

__HIP_MEMORY_SCOPE_SINGLETHREAD

Represents single-thread memory scope in HIP (value is 1).

__HIP_MEMORY_SCOPE_WAVEFRONT

Represents wavefront memory scope in HIP (value is 2).

__HIP_MEMORY_SCOPE_WORKGROUP

Represents workgroup memory scope in HIP (value is 3).

__HIP_MEMORY_SCOPE_AGENT

Represents agent memory scope in HIP (value is 4).

__HIP_MEMORY_SCOPE_SYSTEM

Represents system-wide memory scope in HIP (value is 5).

__HIP_NO_IMAGE_SUPPORT__

Defined with a value of 1 when the target device lacks support for HIP image functions.

__HIP_NO_IMAGE_SUPPORT

Alias to __HIP_NO_IMAGE_SUPPORT__. Deprecated.

__HIP_API_PER_THREAD_DEFAULT_STREAM__

Defined when the GPU default stream is set to per-thread mode.

HIP_API_PER_THREAD_DEFAULT_STREAM

Alias to __HIP_API_PER_THREAD_DEFAULT_STREAM__. Deprecated.

Note that some architecture specific AMDGPU macros will have default values when used from the HIP host compilation. Other AMDGPU macros like __AMDGCN_WAVEFRONT_SIZE__ will default to 64 for example.

Compilation Modes

Each HIP source file contains intertwined device and host code. Depending on the chosen compilation mode by the compiler options -fno-gpu-rdc and -fgpu-rdc, these portions of code are compiled differently.

Device Code Compilation

``-fno-gpu-rdc`` Mode (default):

  • Compiles to a self-contained, fully linked offloading device binary for each offloading device architecture.

  • Device code within a Translation Unit (TU) cannot call functions located in another TU.

``-fgpu-rdc`` Mode:

  • Compiles to a bitcode for each GPU architecture.

  • For each offloading device architecture, the bitcode from different TUs are linked together to create a single offloading device binary.

  • Device code in one TU can call functions located in another TU.

Host Code Compilation

Both Modes:

  • Compiles to a relocatable object for each TU.

  • These relocatable objects are then linked together.

  • Host code within a TU can call host functions and launch kernels from another TU.

Function Pointers Support

Function pointers’ support varies with the usage mode in Clang with HIP. The following table provides an overview of the support status across different use-cases and modes.

Function Pointers Support Overview

Use Case

-fno-gpu-rdc Mode (default)

-fgpu-rdc Mode

Defined and used in the same TU

Supported

Supported

Defined in one TU and used in another TU

Not Supported

Supported

In the -fno-gpu-rdc mode, the compiler calculates the resource usage of kernels based only on functions present within the same TU. This mode does not support the use of function pointers defined in a different TU due to the possibility of incorrect resource usage calculations, leading to undefined behavior.

On the other hand, the -fgpu-rdc mode allows the definition and use of function pointers across different TUs, as resource usage calculations can accommodate functions from disparate TUs.

Virtual Function Support

In Clang with HIP, support for calling virtual functions of an object in device or host code is contingent on where the object is constructed.

  • Constructed in Device Code: Virtual functions of an object can be called in device code on a specific offloading device if the object is constructed in device code on an offloading device with the same architecture.

  • Constructed in Host Code: Virtual functions of an object can be called in host code if the object is constructed in host code.

In other scenarios, calling virtual functions is not allowed.

Explanation

An object constructed on the device side contains a pointer to the virtual function table on the device side, which is not accessible in host code, and vice versa. Thus, trying to invoke virtual functions from a context different from where the object was constructed will be disallowed because the appropriate virtual table cannot be accessed. The virtual function tables for offloading devices with different architecures are different, therefore trying to invoke virtual functions from an offloading device with a different architecture than where the object is constructed is also disallowed.

Example Usage

class Base {
public:
   __device__ virtual void virtualFunction() {
      // Base virtual function implementation
   }
};

class Derived : public Base {
public:
   __device__ void virtualFunction() override {
      // Derived virtual function implementation
   }
};

__global__ void kernel() {
   Derived obj;
   Base* basePtr = &obj;
   basePtr->virtualFunction(); // Allowed since obj is constructed in device code
}