Ben Sander
@@ -0,0 +1,189 @@
|
||||
<!-- START doctoc generated TOC please keep comment here to allow auto update -->
|
||||
<!-- DON'T EDIT THIS SECTION, INSTEAD RE-RUN doctoc TO UPDATE -->
|
||||
**Installation**
|
||||
|
||||
- [Installing pre-built packages:](#installing-pre-built-packages)
|
||||
- [Prerequisites](#prerequisites)
|
||||
- [AMD (hcc)](#amd-hcc)
|
||||
- [NVIDIA (nvcc)](#nvidia-nvcc)
|
||||
- [Verify your installation](#verify-your-installation)
|
||||
- [Building HIP from source](#building-hip-from-source)
|
||||
- [HCC Options](#hcc-options)
|
||||
- [Using HIP with the AMD Native-GCN compiler.](#using-hip-with-the-amd-native-gcn-compiler)
|
||||
- [Compiling CodeXL markers for HIP Functions](#compiling-codexl-markers-for-hip-functions)
|
||||
- [Using clang-hipify](#using-clang-hipify)
|
||||
- [Building](#building)
|
||||
- [Running and using clang-hipify](#running-and-using-clang-hipify)
|
||||
|
||||
<!-- END doctoc generated TOC please keep comment here to allow auto update -->
|
||||
|
||||
# Installing pre-built packages
|
||||
|
||||
HIP can be easily installed using pre-built binary packages using the package manager for your platform.
|
||||
|
||||
## Prerequisites
|
||||
HIP code can be developed either on AMD ROCm platform using hcc compiler, or a CUDA platform with nvcc installed:
|
||||
|
||||
## AMD-hcc
|
||||
|
||||
* Install the [rocm](http://gpuopen.com/getting-started-with-boltzmann-components-platforms-installation/) packages. Rocm will install all of the necessary components, including the kernel driver, runtime software, HCC compiler, and HIP.
|
||||
|
||||
* Default paths and environment variables:
|
||||
|
||||
* By default HIP looks for hcc in /opt/rocm/hcc (can be overridden by setting HCC_HOME environment variable)
|
||||
* By default HIP looks for HSA in /opt/rocm/hsa (can be overridden by setting HSA_PATH environment variable)
|
||||
* By default HIP is installed into /opt/rocm/hip (can be overridden by setting HIP_PATH environment variable).
|
||||
* Optionally, consider adding /opt/rocm/bin to your path to make it easier to use the tools.
|
||||
|
||||
|
||||
## NVIDIA-nvcc
|
||||
* Configure the additional package server as described [here](http://gpuopen.com/getting-started-with-boltzmann-components-platforms-installation/).
|
||||
* Install the "hip_nvcc" package. This will install CUDA SDK and the HIP porting layer.
|
||||
```
|
||||
apt-get install hip_nvcc
|
||||
```
|
||||
|
||||
* Default paths and environment variables:
|
||||
* By default HIP looks for CUDA SDK in /usr/local/cuda (can be overriden by setting CUDA_PATH env variable)
|
||||
* By default HIP is installed into /opt/rocm/hip (can be overridden by setting HIP_PATH environment variable).
|
||||
* Optionally, consider adding /opt/rocm/bin to your path to make it easier to use the tools.
|
||||
|
||||
|
||||
## Verify your installation
|
||||
Run hipconfig (instructions below assume default installation path) :
|
||||
```shell
|
||||
/opt/rocm/bin/hipconfig --full
|
||||
```
|
||||
|
||||
Compile and run the [square sample](https://github.com/GPUOpen-ProfessionalCompute-Tools/HIP/tree/master/samples/0_Intro/square).
|
||||
|
||||
|
||||
# Building HIP from source
|
||||
HIP source code is available and the project can be built from source on the HCC platform.
|
||||
|
||||
1. Follow the above steps to install and validate the binary packages.
|
||||
2. Download HIP source code (from the [GitHub repot](https://github.com/GPUOpen-ProfessionalCompute-Tools/HIP).)
|
||||
3. Build and install HIP (This is the simple version assuming default paths ; see below for additional options.)
|
||||
```
|
||||
cd HIP-privatestaging
|
||||
mkdir build
|
||||
cd build
|
||||
cmake ..
|
||||
make
|
||||
make install
|
||||
```
|
||||
|
||||
* Default paths:
|
||||
* By default cmake looks for hcc in /opt/rocm/hcc (can be overridden by setting ```-DHCC_HOME=/path/to/hcc``` in the cmake step).*
|
||||
* By default cmake looks for HSA in /opt/rocm/hsa (can be overridden by setting ```-DHSA_PATH=/path/to/hsa``` in the cmake step).*
|
||||
* By default cmake installs HIP to /opt/rocm/hip (can be overridden by setting ```-DCMAKE_INSTALL_PREFIX=/where/to/install/hip``` in the cmake step).*
|
||||
|
||||
Here's a richer command-line that overrides the default paths:
|
||||
|
||||
```shell
|
||||
cd HIP-privatestaging
|
||||
mkdir build
|
||||
cd build
|
||||
cmake -DHSA_PATH=/path/to/hsa -DHCC_HOME=/path/to/hcc -DCMAKE_INSTALL_PREFIX=/where/to/install/hip -DCMAKE_BUILD_TYPE=Release ..
|
||||
make
|
||||
make install
|
||||
```
|
||||
|
||||
* After installation, make sure HIP_PATH is pointed to `/where/to/install/hip`.
|
||||
|
||||
## HCC Options
|
||||
|
||||
### Using HIP with the AMD Native-GCN compiler.
|
||||
AMD recently released a direct-to-GCN-ISA target. This compiler generates GCN ISA directly from LLVM, without going through an intermediate compiler
|
||||
IR such as HSAIL or PTX.
|
||||
The native GCN target is included with upstream LLVM, and has also been integrated with HCC compiler and can be used to compiler HIP programs for AMD.
|
||||
Binary packages for the direct-to-isa package are included with the [rocm](http://gpuopen.com/getting-started-with-boltzmann-components-platforms-installation/) package.
|
||||
Alternatively, this sections describes how to build it from source:
|
||||
|
||||
1. Install the rocm packages as described above.
|
||||
2. Follow the instructions [here](https://github.com/RadeonOpenCompute/HCC-Native-GCN-ISA/wiki)
|
||||
* In the make step for HCC, we recommend setting -DCMAKE_INSTALL_PREFIX.
|
||||
* Set HCC_HOME environment variable before compiling HIP program to point to the native compiler:
|
||||
```shell
|
||||
export HCC_HOME=/path/to/native/hcc
|
||||
```
|
||||
|
||||
|
||||
### Compiling CodeXL markers for HIP Functions
|
||||
HIP can generate markers at function begin/end which are displayed on the CodeXL timeline view. To do this, you need to install CodeXL, tell HIP
|
||||
where the CodeXL install directory lives, and enable HIP to generate the markers:
|
||||
|
||||
1. Install CodeXL
|
||||
See [CodeXL Download](http://developer.amd.com/tools-and-sdks/opencl-zone/codexl/?webSyncID=9d9c2cb9-3d73-5e65-268a-c7b06428e5e0&sessionGUID=29beacd0-d654-ddc6-a3e2-b9e6c0b0cc77) for the installation file.
|
||||
Also this [blog](http://gpuopen.com/getting-up-to-speed-with-the-codexl-gpu-profiler-and-radeon-open-compute/) provides more information and tips for using CodeXL. In addition to installing the CodeXL profiling
|
||||
and visualization tools, CodeXL also comes with an SDK that allow applications to add markers to the timeline viewer. We'll be linking HIP against this library.
|
||||
|
||||
2. Set CODEXL_PATH
|
||||
```shell
|
||||
# set to your code-xl installation location:
|
||||
export CODEXL_PATH=/opt/AMD/CodeXL
|
||||
```
|
||||
|
||||
3. Enable in source code.
|
||||
In src/hip_hcc.cpp, enable the define
|
||||
```c
|
||||
#define COMPILE_TRACE_MARKER 1
|
||||
```
|
||||
|
||||
|
||||
Then recompile the target application, run with profiler enabled to generate ATP file or trace log.
|
||||
```shell
|
||||
# Use profiler to generate timeline view:
|
||||
$CODEXL_PATH/CodeXLGpuProfiler -A -o ./myHipApp
|
||||
...
|
||||
Session output path: /home/me/HIP-privatestaging/tests/b1/mytrace.atp
|
||||
```
|
||||
|
||||
You can also print the HIP function strings to stderr using HIP_TRACE_API environment variable. This can be useful for tracing application flow. Also can be combined with the more detailed debug information provided
|
||||
by the HIP_DB switch. For example:
|
||||
```shell
|
||||
# Trace to stderr showing begin/end of each function (with arguments) + intermediate debug trace during the execution of each function.
|
||||
HIP_TRACE_API=1 HIP_DB=0x2 ./myHipApp
|
||||
```
|
||||
|
||||
Note this trace mode uses colors. "less -r" can handle raw control characters and will display the debug output in proper colors.
|
||||
|
||||
|
||||
### Using clang-hipify
|
||||
|
||||
Clang-hipify is a clang-based tool which can automate the translation of CUDA source code into portable HIP C++.
|
||||
The clang-hipify tool can automatically add extra HIP arguments (notably the "hipLaunchParm" required at the
|
||||
beginning of every HIP kernel call). Clang-hipify has some additional dependencies explained below and
|
||||
can be built as a separate make step.
|
||||
|
||||
|
||||
#### Building
|
||||
|
||||
1. Download and unpack clang+llvm 3.8 binary package preqrequisite:
|
||||
```
|
||||
wget http://llvm.org/releases/3.8.0/clang+llvm-3.8.0-x86_64-linux-gnu-ubuntu-14.04.tar.xz
|
||||
tar xvfJ clang+llvm-3.8.0-x86_64-linux-gnu-ubuntu-14.04.tar.xz
|
||||
```
|
||||
|
||||
2. Enable build of clang-hipify and specify path to LLVM:
|
||||
Note LLVM_DIR must be a full absolute path (not relative) to the location extracted above. Here's an example assuming we
|
||||
extract the clang 3.8 package into ~/HIP-privatestaging/clang+llvm-3.8.0-x86_64-linux-gnu-ubuntu-14.04/.
|
||||
```
|
||||
cd HIP-privatestaging
|
||||
mkdir build.clang-hipify
|
||||
cd build.clang-hipify
|
||||
cmake -DBUILD_CLANG_HIPIFY=1 -DLLVM_DIR=~/HIP-privatestaging/clang+llvm-3.8.0-x86_64-linux-gnu-ubuntu-14.04/ -DCMAKE_BUILD_TYPE=Release ..
|
||||
make
|
||||
make install
|
||||
```
|
||||
|
||||
#### Running and using clang-hipify
|
||||
clang-hipify performs an initial compile of the CUDA source code into a "symbol tree", and thus needs access to the appropriate header files:
|
||||
1. Download "deb(network)" variant of target installer from https://developer.nvidia.com/cuda-downloads. The commands below show how to download and install a recent version from the http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1404/x86_64/cuda-repo-ubuntu1404_7.5-18_amd64.deb.
|
||||
|
||||
```
|
||||
wget http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1404/x86_64/cuda-repo-ubuntu1404_7.5-18_amd64.deb
|
||||
sudo dpkg -i cuda-repo-ubuntu1404_7.5-18_amd64.deb
|
||||
sudo apt-get update && sudo apt-get install cuda-minimal-build-7-5 cuda-curand-dev-7-5
|
||||
```
|
||||
|
||||
@@ -10,32 +10,9 @@ Key features include:
|
||||
* Developers can specialize for the platform (CUDA or hcc) to tune for performance or handle tricky cases
|
||||
|
||||
New projects can be developed directly in the portable HIP C++ language and can run on either NVIDIA or AMD platforms. Additionally, HIP provides porting tools which make it easy to port existing CUDA codes to the HIP layer, with no loss of performance as compared to the original CUDA application. HIP is not intended to be a drop-in replacement for CUDA, and developers should expect to do some manual coding and performance tuning work to complete the port.
|
||||
## Installation
|
||||
```shell
|
||||
cd HIP-privatestaging
|
||||
mkdir build
|
||||
cd build
|
||||
<<<<<<< HEAD
|
||||
cmake ..
|
||||
make
|
||||
make install
|
||||
```
|
||||
*By default cmake looks for hcc in /opt/rocm/hcc (can be overridden by setting ```-DHCC_HOME=/path/to/hcc``` in the cmake step).*
|
||||
|
||||
*By default cmake looks for HSA in /opt/rocm/hsa (can be overridden by setting ```-DHSA_PATH=/path/to/hsa``` in the cmake step).*
|
||||
|
||||
*By default cmake installs HIP to /opt/rocm/hip (can be overridden by setting ```-DCMAKE_INSTALL_PREFIX=/where/to/install/hip``` in the cmake step).*
|
||||
|
||||
*Make sure HIP_PATH is pointed to `/where/to/install/hip` and PATH includes `$HIP_PATH/bin`. This requirement is optional, but required to run any HIP test infrastructure.*
|
||||
=======
|
||||
cmake -DHSA_PATH=/path/to/hsa -DHCC_HOME=/path/to/hcc -DCMAKE_INSTALL_PREFIX=/where/to/install/hip -DLLVM_DIR=/path/to/clang-llvm-3.8 -DCMAKE_BUILD_TYPE=Release ..
|
||||
make
|
||||
make install
|
||||
```
|
||||
Make sure HIP_PATH is pointed to `/where/to/install/hip` and PATH includes `$HIP_PATH/bin`. This requirement is optional, but required to run any HIP test infrastructure. The path `/path/to/clang-llvm-3.8` should be specified for [clang-hipify](README.md#clang-hipify) utility build.
|
||||
>>>>>>> clang-hipify
|
||||
|
||||
## More Info:
|
||||
- [Installation](INSTALL.md) and [clang-hipify](INSTALL.md#use-clang-hipify.md)
|
||||
- [HIP FAQ](docs/markdown/hip_faq.md)
|
||||
- [HIP Kernel Language](docs/markdown/hip_kernel_language.md)
|
||||
- [HIP Runtime API (Doxygen)](http://gpuopen-professionalcompute-tools.github.io/HIP)
|
||||
@@ -47,108 +24,8 @@ Make sure HIP_PATH is pointed to `/where/to/install/hip` and PATH includes `$HIP
|
||||
|
||||
## How do I get set up?
|
||||
|
||||
### Prerequisites - Choose Your Platform
|
||||
HIP code can be developed either on AMD ROCm platform using hcc compiler, or a CUDA platform with nvcc installed:
|
||||
See the [Installation](INSTALL.md) notes.
|
||||
|
||||
#### AMD (hcc):
|
||||
|
||||
* Install [hcc](https://bitbucket.org/multicoreware/hcc/wiki/Home) including supporting HSA kernel and runtime driver stack
|
||||
* By default HIP looks for hcc in /opt/rocm/hcc (can be overridden by setting HCC_HOME environment variable)
|
||||
* By default HIP looks for HSA in /opt/rocm/hsa (can be overridden by setting HSA_PATH environment variable)
|
||||
* Ensure that ROCR runtime is installed and added to LD_LIBRARY_PATH
|
||||
* Install HIP (from this GitHub repot). By default HIP is installed into /opt/rocm/hip (can be overridden by setting HIP_PATH environment variable).
|
||||
|
||||
* Optionally, consider adding /opt/rocm/bin to your path to make it easier to use the tools.
|
||||
|
||||
|
||||
#####clang-hipify
|
||||
To build and run clang based hipify utiliy a set of CUDA headers and clang+llvm 3.8 binary package are required:
|
||||
- download and install CUDA minimal prerequisites:
|
||||
1. Download "deb(network)" variant of target installer from https://developer.nvidia.com/cuda-downloads. E.g. at the moment the link is http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1404/x86_64/cuda-repo-ubuntu1404_7.5-18_amd64.deb
|
||||
2. install clang prerequisites with the following commands:
|
||||
```
|
||||
wget http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1404/x86_64/cuda-repo-ubuntu1404_7.5-18_amd64.deb
|
||||
sudo dpkg -i cuda-repo-ubuntu1404_7.5-18_amd64.deb
|
||||
sudo apt-get update && sudo apt-get install cuda-minimal-build-7-5 cuda-curand-dev-7-5
|
||||
```
|
||||
- download and unpack clang+llvm 3.8 binary package:
|
||||
```
|
||||
wget http://llvm.org/releases/3.8.0/clang+llvm-3.8.0-x86_64-linux-gnu-ubuntu-14.04.tar.xz
|
||||
tar xvfJ clang+llvm-3.8.0-x86_64-linux-gnu-ubuntu-14.04.tar.xz -C /path/to/clang-llvm-3.8
|
||||
```
|
||||
#### NVIDIA (nvcc)
|
||||
* Install CUDA SDK from manufacturer website
|
||||
* By default HIP looks for CUDA SDK in /usr/local/cuda (can be overriden by setting CUDA_PATH env variable)
|
||||
* Install HIP (from this GitHub repot). By default HIP is installed into /opt/rocm/hip (can be overridden by setting HIP_PATH environment variable).
|
||||
|
||||
* Optionally, consider adding /opt/rocm/bin to your path to make it easier to use the tools.
|
||||
|
||||
#### Verify your installation
|
||||
Run hipconfig (instructions below assume default installation path) :
|
||||
```shell
|
||||
/opt/rocm/bin/hipconfig --full
|
||||
```
|
||||
|
||||
Compile and run the [square sample](https://github.com/GPUOpen-ProfessionalCompute-Tools/HIP/tree/master/samples/0_Intro/square).
|
||||
|
||||
|
||||
|
||||
### HCC Options
|
||||
|
||||
#### Compiling CodeXL markers for HIP Functions
|
||||
HIP can generate markers at function begin/end which are displayed on the CodeXL timeline view. To do this, you need to install CodeXL, tell HIP
|
||||
where the CodeXL install directory lives, and enable HIP to generate the markers:
|
||||
|
||||
1. Install CodeXL
|
||||
See [CodeXL Download](http://developer.amd.com/tools-and-sdks/opencl-zone/codexl/?webSyncID=9d9c2cb9-3d73-5e65-268a-c7b06428e5e0&sessionGUID=29beacd0-d654-ddc6-a3e2-b9e6c0b0cc77) for the installation file.
|
||||
Also this [blog](http://gpuopen.com/getting-up-to-speed-with-the-codexl-gpu-profiler-and-radeon-open-compute/) provides more information and tips for using CodeXL. In addition to installing the CodeXL profiling
|
||||
and visualization tools, CodeXL also comes with an SDK that allow applications to add markers to the timeline viewer. We'll be linking HIP against this library.
|
||||
|
||||
2. Set CODEXL_PATH
|
||||
```shell
|
||||
# set to your code-xl installation location:
|
||||
export CODEXL_PATH=/opt/AMD/CodeXL
|
||||
```
|
||||
|
||||
3. Enable in source code.
|
||||
In src/hip_hcc.cpp, enable the define
|
||||
```c
|
||||
#define COMPILE_TRACE_MARKER 1
|
||||
```
|
||||
|
||||
|
||||
Then recompile the target application, run with profiler enabled to generate ATP file or trace log.
|
||||
```shell
|
||||
# Use profiler to generate timeline view:
|
||||
$CODEXL_PATH/CodeXLGpuProfiler -A -o ./myHipApp
|
||||
...
|
||||
Session output path: /home/me/HIP-privatestaging/tests/b1/mytrace.atp
|
||||
```
|
||||
|
||||
You can also print the HIP function strings to stderr using HIP_TRACE_API environment variable. This can be useful for tracing application flow. Also can be combined with the more detailed debug information provided
|
||||
by the HIP_DB switch. For example:
|
||||
```shell
|
||||
# Trace to stderr showing begin/end of each function (with arguments) + intermediate debug trace during the execution of each function.
|
||||
HIP_TRACE_API=1 HIP_DB=0x2 ./myHipApp
|
||||
```
|
||||
|
||||
Note this trace mode uses colors. "less -r" can handle raw control characters and will display the debug output in proper colors.
|
||||
|
||||
|
||||
#### Using HIP with the AMD Native-GCN compiler.
|
||||
AMD recently released a direct-to-GCN-ISA target. This compiler generates GCN ISA directly from LLVM, without going through an intermediate compiler
|
||||
IR such as HSAIL or PTX.
|
||||
The native GCN target is included with upstream LLVM, and has also been integrated with HCC compiler and can be used to compiler HIP programs for AMD.
|
||||
Here's how to use it with HIP:
|
||||
|
||||
- Follow the instructions here to compile the HCC and native LLVM compiler:
|
||||
> https://github.com/RadeonOpenCompute/HCC-Native-GCN-ISA/wiki
|
||||
> (In the make step for HCC, we recommend setting -DCMAKE_INSTALL_PREFIX=/opt/hcc-native)
|
||||
|
||||
Set HCC_HOME environment variable before compiling HIP program to point to the native compiler:
|
||||
```shell
|
||||
export HCC_HOME=/opt/hcc-native
|
||||
```
|
||||
|
||||
## Examples and Getting Started:
|
||||
|
||||
|
||||
Referencia en una nueva incidencia
Block a user