- Starting the port on a Cuda machine is often the easiest approach, since you can incrementally port pieces of the code to HIP while leaving the rest in Cuda. (Recall that on Cuda machines HIP is just a thin layer over Cuda, so the two code types can interoperate on nvcc platforms.) Also, the HIP port can be compared with the original Cuda code for function and performance.
- Once the Cuda code is ported to HIP and is running on the Cuda machine, compile the HIP code using hcc on an AMD machine.
- HIP ports can replace Cuda versions---HIP can deliver the same performance as a native Cuda implementation, with the benefit of portability to both Nvidia and AMD architectures as well as a path to future C++ standard support. You can handle platform-specific features through conditional compilation or by adding them to the open-source HIP infrastructure.
- Use **bin/hipconvertinplace.sh** to hipify all code files in the Cuda source directory.
### Scanning existing CUDA code to scope the porting effort
The hipexamine.sh tool will scan a source directory to determine which files contain CUDA code and how much of that code can be automatically hipified,
* Some of the most interesting information in kmeans_cuda_kernel.cu :
* How many CUDA calls were converted to HIP (40)
* Breakdown of the different CUDA functionality used (dev:0 mem:0 etc). This file uses many CUDA builtins (37) and texture functions (3).
* Warning for code that looks like CUDA API but was not converted (0 in this file).
* Count Lines-of-Code (LOC) - 185 for this file.
* hipexamine also presents a summary at the end of the process for the statistics collected across all files. This has similar format to the per-file reporting, and also includes a list of all kernels which have been called. An example from above:
All HIP projects target either the hcc or nvcc platform. The platform affects which headers are included and which libraries are used for linking.
-`HIPCC_PLATFORM_HCC` is defined if the HIP platform targets hcc
-`HIPCC_PLATFORM_NVCC` is defined if the HIP platform targets nvcc
Many projects use a mixture of an accelerator compiler (hcc or nvcc) and a standard compiler (e.g., g++). These defines are set for both accelerator and standard compilers and thus are often the best option when writing code that uses conditional compilation.
### Identifying the Compiler: hcc or nvcc
Often, its useful to know whether the underlying compiler is hcc or nvcc. This knowledge can guard platform-specific code (features that only work on the nvcc or hcc path but not both) or aid in platform-specific performance tuning.
```
#ifdef __HCC__
// Compiled with hcc
```
```
#ifdef __NVCC__
// Compiled with nvcc
// Could be compiling with Cuda language extensions enabled (for example, a ".cu file)
// Could be in pass-through mode to an underlying host compile OR (for example, a .cpp file)
```
```
#ifdef __CUDACC__
// Compiled with nvcc (Cuda language extensions enabled)
```
hcc directly generates the host code (using the Clang x86 target) and passes the code to another host compiler. Thus, it lacks the equivalent of the \__CUDA_ACC define.
The macro `__HIPCC__` is set if either `__HCC__` or `__CUDACC__` is defined. This configuration is useful in determining when code is being compiled using an accelerator-enabled compiler (hcc or nvcc) as opposed to a standard host compiler (GCC, ICC, Clang, etc.).
### Identifying Current Compilation Pass: Host or Device
Both nvcc and hcc make two passes over the code: one for host code and one for device code. `__HIP_DEVICE_COMPILE__` is set to a nonzero value when the compiler (hcc or nvcc) is compiling code for a device inside a `__global__` kernel or for a device function. `__HIP_DEVICE_COMPILE__` can replace #ifdef checks on the `__CUDA_ARCH__` define.
```
// #ifdef __CUDA_ARCH__
#ifdef __HIP_DEVICE_COMPILE__
```
Unlike `__CUDA_ARCH__`, the `__HIP_DEVICE_COMPILE__` value is 0 or 1, and it doesnt represent the feature capability of the target device.
|`__HIP_PLATFORM_HCC___`| Defined | Undefined | Defined if targeting hcc platform; undefined otherwise |
|`__HIP_PLATFORM_NVCC___`| Undefined | Defined | Defined if targeting nvcc platform; undefined otherwise |
|`__HIP_DEVICE_COMPILE__` | 1 if compiling for device; 0 if compiling for host |1 if compiling for device; 0 if compiling for host | Undefined
|`__HIPCC__` | Defined | Defined | Undefined
|`__HIP_ARCH_*` | 0 or 1 depending on feature support (see below) | 0 or 1 depending on feature support (see below) | 0
|nvcc-related defines:|
|`__CUDACC__` | Undefined | Defined if compiling for Cuda device; undefined otherwise | Undefined
|`__NVCC__` | Undefined | Defined | Undefined
|`__CUDA_ARCH__` | Undefined | Unsigned representing compute capability (e.g., "130") if in device code; 0 if in host code | Undefined
|hcc-related defines:|
|`__HCC__` | Defined | Undefined | Undefined
|`__HCC_ACCELERATOR__` | Nonzero if in device code; otherwise undefined | Undefined | Undefined
|`__clang__` | Defined | Undefined | Defined if using Clang; otherwise undefined
## Identifying Architecture Features
### HIP_ARCH Defines
Some Cuda code tests `__CUDA_ARCH__` for a specific value to determine whether the machine supports a certain architectural feature. For instance,
```
#if (__CUDA_ARCH__ >= 130)
// doubles are supported
```
This type of code requires special attention, since hcc/AMD and nvcc/Cuda devices have different architectural capabilities. Moreover, you cant determine the presence of a feature using a simple comparison against an architectures version number. HIP provides a set of defines and device properties to query whether a specific architectural feature is supported.
The `__HIP_ARCH_*` defines can replace comparisons of `__CUDA_ARCH__` values:
```
//#if (__CUDA_ARCH__ >= 130) // non-portable
if __HIP_ARCH_HAS_DOUBLES__ { // portable HIP feature query
// doubles are supported
}
```
For host code, the `__HIP_ARCH__*` defines are set to 0. You should only use the __HIP_ARCH__ fields in device code.
Host code should query the architecture feature flags in the device properties that hipGetDeviceProperties returns, rather than testing the "major" and "minor" fields directly:
hipLaunchKernel is a variadic macro which accepts as parameters the launch configurations (grid dims, group dims, stream, dynamic shared size) followed by a variable number of kernel arguments.
This sequence is then expanded into the appropriate kernel launch syntax depending on the platform.
While this can be a convenient single-line kernel launch syntax, the macro implementation can cause issues when nested inside other macros. For example, consider the following:
```
// Will cause compile error:
#define MY_LAUNCH(command, doTrace) \
{\
if (doTrace) printf ("TRACE: %s\n", #command); \
(command); /* The nested ( ) will cause compile error */\
hipcc is a portable compiler driver that will call nvcc or hcc (depending on the target system) and attach all required include and library options. It passes options through to the target compiler. Tools that call hipcc must ensure the compiler options are appropriate for the target compiler. The `hipconfig` script may helpful in making
infrastructure that identifies the target platform and sets options appropriately. It returns either "nvcc" or "hcc." The following sample shows the script in a makefile:
```
HIP_PLATFORM=$(shell hipconfig --compiler)
ifeq (${HIP_PLATFORM}, nvcc)
HIPCC_FLAGS = -gencode=arch=compute_20,code=sm_20
endif
ifeq (${HIP_PLATFORM}, hcc)
HIPCC_FLAGS = -Wno-deprecated-register
endif
```
## Linking Issues
### Linking With hipcc
hipcc adds the necessary libraries for HIP as well as for the accelerator compiler (nvcc or hcc). We recommend linking with hipcc.
Cuda code often uses nvcc for accelerator code (defining and launching kernels, typically defined in .cu or .cuh files).
It also uses a standard compiler (g++) for the rest of the application. nvcc is a preprocessor that employs a standard host compiler (e.g., gcc) to generate the host code.
Code compiled using this tool can employ only the intersection of language features supported by both nvcc and the host compiler.
In some cases, you must take care to ensure the data types and alignment of the host compiler are identical to those of the device compiler. Only some host compilers are supported---for example, recent nvcc versions lack Clang host-compiler capability.
hcc generates both device and host code using the same Clang-based compiler. The code uses the same API as gcc, which allows code generated by different gcc-compatible compilers to be linked together. For example, code compiled using hcc can link with code compiled using "standard" compilers (such as gcc, ICC and Clang). You must take care to ensure all compilers use the same standard C++ header and library formats.
Version 0.86 of hipcc now uses libstdc++ by default for the HCC platform. This improves cross-linking support between G++ and hcc, in particular for interfaces that use
standard C++ libraries (ie std::vector, std::string).
If you pass "--stdlib=libc++" to hipcc, hipcc will use the libc++ library. Generally, libc++ provides a broader set of C++ features while libstdc++ is the standard
When cross-linking C++ code, any C++ functions that use types from the C++ standard library (including std::string, std::vector and other containers) must use the same standard-library implementation. They include the following:
- Functions or kernels defined in hcc that are called from a standard compiler
- Functions defined in a standard compiler that are called from hcc.
Applications which are compiled entirely with hipcc, and which benefit from advanced C++ features not supported in libstdc++, and which do not require portability to nvcc, may choose to use libc++.
- hip_runtime_api.h: defines all the HIP runtime APIs (e.g., hipMalloc) and the types required to call them. A source file that is only calling HIP APIs but neither defines nor launches any kernels can include hip_runtime_api.h. hip_runtime_api.h uses no custom hc language features and can be compiled using a standard C++ compiler.
- hip_runtime.h: included in hip_runtime_api.h. It additionally provides the types and defines required to create and launch kernels. hip_runtime.h does use custom hc language features, but they are guarded by ifdef checks. It can be compiled using a standard C++ compiler but will expose a subset of the available functions.
Cuda has slightly different contents for these two files. In some cases you may need to convert hipified code to include the richer hip_runtime.h instead of hip_runtime_api.h.
You can compile hip\_runtime\_api.h using a standard C or C++ compiler (e.g., gcc or ICC). The HIP include paths and defines (`__HIP_PLATFORM_HCC__` or `__HIP_PLATFORM_NVCC__`) must pass to the standard compiler; hipconfig then returns the necessary options:
nvcc includes some headers by default. However, HIP does not include default headers, and instead all required files must be explicitly included.
Specifically, files that call HIP run-time APIs or define HIP kernels must explicitly include the appropriate HIP headers.
If the compilation process reports that it cannot find necessary APIs (for example, "error: identifier hipSetDevice is undefined"),
ensure that the file includes hip_runtime.h (or hip_runtime_api.h, if appropriate).
The hipify script automatically converts "cuda_runtime.h" to "hip_runtime.h," and it converts "cuda_runtime_api.h" to "hip_runtime_api.h", but it may miss nested headers or macros.
The hcc path provides an empty cuda.h file. Some existing Cuda programs include this file but don't require any of the functions.
### Choosing HIP File Extensions
Many existing Cuda projects use the ".cu" and ".cuh" file extensions to indicate code that should be run through the nvcc compiler.
For quick HIP ports, leaving these file extensions unchanged is often easier, as it minimizes the work required to change file names in the directory and #include statements in the files.
For new projects or ports which can be re-factored, we recommend the use of the extension ".hip.cpp" for header files, and
".hip.h" or ".hip.hpp" for header files.
This indicates that the code is standard C++ code, but also provides a unique indication for make tools to
Code should not assume a warp size of 32 or 64. See [Warp Cross-Lane Functions](hip_kernel_language.md#warp-cross-lane-functions) for information on how to write portable wave-aware code.
Threadfence_system makes all device memory writes, all writes to mapped host memory, and all writes to peer memory visible to CPU and other GPU devices.
Some implementations can provide this behavior by flushing the GPU L2 cache.
HIP/HCC does not provide this functionality. As a workaround, users can set the environment variable `HSA_DISABLE_CACHE=1` to
disable the GPU L2 cache. This will affect all accesses and for all kernels and so may have
Compute programs sometimes use textures either to access dedicated texture caches or to use the texture-sampling hardware for interpolation and clamping. The former approach uses simple point samplers with linear interpolation, essentially only reading a single point. The latter approach uses the sampler hardware to interpolate and combine multiple
point samples. AMD hardware, as well as recent competing hardware,
has a unified texture/L1 cache, so it no longer has a dedicated texture cache. But the nvcc path often caches global loads in the L2 cache, and some programs may benefit from explicit control of the L1 cache contents. We recommend the __ldg instruction for this purpose.
HIP currently lacks texture support; a future revision will add this capability. Also, AMD compilers currently load all data into both the L1 and L2 caches, so __ldg is treated as a no-op.
We recommend the following for functional portability:
- For programs that use textures only to benefit from improved caching, use the __ldg instruction
- Alternatively, use conditional compilation (see [Identify HIP Target Platform](#identify-hip-target-platform))
- For the `__HIP_PLATFORM_NVCC__` path, use the full texture path
- For the `__HIP_PLATFORM_HCC__` path, pass an additional pointer to the kernel and reference it using regular device memory-load instructions rather than texture loads. Some applications may already take this step, since it allows experimentation with caching behavior.
Additionally, many of the Rodinia benchmarks demonstrate how to modify hipified programs so that textures are not required - search for USE_TEXTURES define in the rodinia source directory.
Cuda programs that employ sampler hardware must either wait for hcc texture support or use more-sophisticated workarounds.
## More Tips
### HIPTRACE Mode
On an hcc/AMD platform, set the HIP_TRACE_API environment variable to see a textural API trace. Use the following bit mask:
- 0x1 = trace APIs
- 0x2 = trace synchronization operations
- 0x4 = trace memory allocation / deallocation
### Environment Variables
On hcc/AMD platforms, set the HIP_PRINT_ENV environment variable to 1 and run an application that calls a HIP API to see all HIP-supported environment variables and their current values:
- HIP_PRINT_ENV = 1: print HIP environment variables
- HIP_LAUNCH_BLOCKING = 0: make HIP APIs host-synchronous so they are blocked until any kernel launches or data-copy commands are complete (an alias is CUDA_LAUNCH_BLOCKING)
To see the detailed commands that hipcc issues, set the environment variable HIPCC_VERBOSE to 1. Doing so will print to stderr the hcc (or nvcc) commands that hipcc generates.
#### /usr/include/c++/v1/memory:5172:15: error: call to implicitly deleted default constructor of 'std::__1::bad_weak_ptr' throw bad_weak_ptr();
If you pass a ".cu" file, hcc will attempt to compile it as a Cuda language file. You must tell hcc that its in fact a C++ file: use the "-x c++" option.
On the HCC path, HIP provides a number of environment variables that control the behavior of HIP. Some of these are useful for appliction development (for example HIP_VISIBLE_DEVICES, HIP_LAUNCH_BLOCKING),
some are useful for performance tuning or experimentation (for example HIP_STAGING*), and some are useful for debugging (HIP_DB). You can see the environment variables supported by HIP as well as
their current values and usage with the environment var "HIP_PRINT_ENV" - set this and then run any HIP application. For example:
```
$ HIP_PRINT_ENV=1 ./myhipapp
HIP_PRINT_ENV = 1 : Print HIP environment variables.
HIP_LAUNCH_BLOCKING = 0 : Make HIP APIs 'host-synchronous', so they block until any kernel launches or data copy commands complete. Alias: CUDA_LAUNCH_BLOCKING.
HIP_DB = 0 : Print various debug info. Bitmask, see hip_hcc.cpp for more information.
HIP_TRACE_API = 0 : Trace each HIP API call. Print function name and return code to stderr as program executes.
HIP_STAGING_SIZE = 64 : Size of each staging buffer (in KB)
HIP_STAGING_BUFFERS = 2 : Number of staging buffers to use in each direction. 0=use hsa_memory_copy.
HIP_PININPLACE = 0 : For unpinned transfers, pin the memory in-place in chunks before doing the copy. Under development.
HIP_STREAM_SIGNALS = 2 : Number of signals to allocate when new stream is created (signal pool will grow on demand)
HIP_VISIBLE_DEVICES = 0 : Only devices whose index is present in the secquence are visible to HIP applications and they are enumerated in the order of secquence
HIP_DISABLE_HW_KERNEL_DEP = 1 : Disable HW dependencies before kernel commands - instead wait for dependency on host. -1 means ignore these dependencies. (debug mode)
HIP_DISABLE_HW_COPY_DEP = 1 : Disable HW dependencies before copy commands - instead wait for dependency on host. -1 means ifnore these dependencies (debug mode)
