AI/rocm-systems
2
0
Fork 0
Code Issues Pull-Requests Actions Pakete Projekte Releases Wiki Aktivität

Merge branch 'rocm-rel-1.5'

Quellcode durchsuchen 
Change-Id: Ib2318f9c0d01a1bc8be2fcb172a3075e82851877


[ROCm/hip commit: dbe049937e]
Dieser Commit ist enthalten in:
Maneesh Gupta
2017-05-02 09:06:49 +05:30
Ursprung 73dad41dd7 b2f561e1ee
Commit 0ac1c42de3
280 geänderte Dateien mit 35694 neuen und 6493 gelöschten Zeilen
Patch-Datei herunterladen Vergleichs-Datei herunterladen Alle Dateien ausklappen Alle Dateien einklappen
projects/hip/.vimrc
-1
Datei anzeigen
@@ -1 +0,0 @@
:set makeprg=make\ -C\ build.hcc-LC.db
projects/hip/CMakeLists.txt
+85 -24
Datei anzeigen
@@ -33,7 +33,7 @@ execute_process(COMMAND git show -s --format=@%ct
OUTPUT_VARIABLE HIP_VERSION_PATCH
OUTPUT_STRIP_TRAILING_WHITESPACE)
set(HIP_VERSION $HIP_VERSION_MAJOR.$HIP_VERSION_MINOR.$HIP_VERSION_PATCH)
set(HIP_VERSION ${HIP_VERSION_MAJOR}.${HIP_VERSION_MINOR}.${HIP_VERSION_PATCH})
add_to_config(_versionInfo HIP_VERSION_MAJOR)
add_to_config(_versionInfo HIP_VERSION_MINOR)
add_to_config(_versionInfo HIP_VERSION_PATCH)
@@ -63,14 +63,16 @@ if(HIP_PLATFORM STREQUAL "hcc")
set(HCC_HOME $ENV{HCC_HOME} CACHE PATH "Path to which HCC has been installed")
endif()
endif()
if(DEFINED ENV{HIP_DEVELOPER})
add_to_config(_buildInfo HCC_HOME)
endif()
if(IS_ABSOLUTE ${HCC_HOME} AND EXISTS ${HCC_HOME} AND IS_DIRECTORY ${HCC_HOME})
execute_process(COMMAND ${HCC_HOME}/bin/hcc --version
COMMAND cut -d\ -f9
OUTPUT_VARIABLE HCC_VERSION
OUTPUT_STRIP_TRAILING_WHITESPACE)
string(REGEX REPLACE ".*based on HCC " "" HCC_VERSION ${HCC_VERSION})
string(REGEX REPLACE " .*" "" HCC_VERSION ${HCC_VERSION})
message(STATUS "Looking for HCC in: " ${HCC_HOME} ". Found version: " ${HCC_VERSION})
else()
message(FATAL_ERROR "Don't know where to find HCC. Please specify abolute path using -DHCC_HOME")
@@ -78,6 +80,9 @@ if(HIP_PLATFORM STREQUAL "hcc")
add_to_config(_buildInfo HCC_VERSION)
string(REPLACE "-" ";" HCC_VERSION_LIST ${HCC_VERSION})
list(GET HCC_VERSION_LIST 0 HCC_PACKAGE_VERSION)
string(REPLACE "." ";" HCC_VERSION_LIST ${HCC_PACKAGE_VERSION})
list(GET HCC_VERSION_LIST 0 HCC_VERSION_MAJOR)
list(GET HCC_VERSION_LIST 1 HCC_VERSION_MINOR)
# Determine HSA_PATH
if(NOT DEFINED HSA_PATH)
@@ -137,7 +142,7 @@ add_to_config(_buildInfo COMPILE_HIP_ATP_MARKER)
# Build steps
#############################
# Rebuild cmake cache updates .hipInfo and .hipVersion
add_custom_target(update_build_and_version_info ALL COMMAND make rebuild_cache)
add_custom_target(update_build_and_version_info COMMAND make rebuild_cache)
# Build clang hipify if enabled
add_subdirectory(hipify-clang)
@@ -155,7 +160,8 @@ if(HIP_PLATFORM STREQUAL "hcc")
set(HIP_HCC_BUILD_FLAGS "${HIP_HCC_BUILD_FLAGS} -DHIP_VERSION_MAJOR=${HIP_VERSION_MAJOR} -DHIP_VERSION_MINOR=${HIP_VERSION_MINOR} -DHIP_VERSION_PATCH=${HIP_VERSION_PATCH}")
# Add remaining flags
set(HIP_HCC_BUILD_FLAGS "${HIP_HCC_BUILD_FLAGS} -fPIC -hc -I${HCC_HOME}/include -I${HSA_PATH}/include -I/opt/rocm/libhsakmt/include -I/usr/local/include/c++/v1 -stdlib=libc++")
execute_process(COMMAND ${HCC_HOME}/bin/hcc-config --cxxflags OUTPUT_VARIABLE HCC_CXX_FLAGS)
set(HIP_HCC_BUILD_FLAGS "${HIP_HCC_BUILD_FLAGS} -fPIC ${HCC_CXX_FLAGS} -I${HSA_PATH}/include")
# Set compiler and compiler flags
set(CMAKE_CXX_COMPILER "${HCC_HOME}/bin/hcc")
@@ -172,20 +178,36 @@ if(HIP_PLATFORM STREQUAL "hcc")
src/hip_memory.cpp
src/hip_peer.cpp
src/hip_stream.cpp
src/hip_module.cpp)
src/hip_module.cpp
src/grid_launch.cpp
src/env.cpp)
set(SOURCE_FILES_DEVICE
src/device_util.cpp
src/hip_ldg.cpp
src/hip_fp16.cpp)
src/hip_fp16.cpp
src/device_functions.cpp
src/math_functions.cpp)
set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} -L${HCC_HOME}/lib -lmcwamp -Wl,-Bsymbolic")
execute_process(COMMAND ${HCC_HOME}/bin/hcc-config --ldflags OUTPUT_VARIABLE HCC_LD_FLAGS)
set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} ${HCC_LD_FLAGS} -Wl,-Bsymbolic")
#find_package(LLVM HINTS ${HCC_HOME}/compiler/lib/cmake)
set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} --amdgpu-target=gfx701 --amdgpu-target=gfx801 --amdgpu-target=gfx802 --amdgpu-target=gfx803 --amdgpu-target=gfx900")
add_library(hip_hcc SHARED ${SOURCE_FILES_RUNTIME})
target_link_libraries(hip_hcc PRIVATE hc_am)
#target_link_libraries(hip_hcc PUBLIC LLVMAMDGPUUtils)
add_library(hip_hcc_static STATIC ${SOURCE_FILES_RUNTIME})
target_link_libraries(hip_hcc_static PRIVATE hc_am)
add_dependencies(hip_hcc_static hip_hcc)
add_library(hip_device STATIC ${SOURCE_FILES_DEVICE})
add_dependencies(hip_device hip_hcc)
string(REPLACE " " ";" HCC_CXX_FLAGS_LIST ${HCC_CXX_FLAGS})
foreach(TARGET hip_hcc hip_hcc_static hip_device)
target_include_directories(${TARGET} SYSTEM INTERFACE $<INSTALL_INTERFACE:$<INSTALL_PREFIX>/include>;${HSA_PATH}/include)
endforeach()
target_link_libraries(hip_hcc INTERFACE hcc::hccrt;hcc::hc_am)
# Generate hcc_version.txt
add_custom_target(query_hcc_version COMMAND ${HCC_HOME}/bin/hcc --version > ${PROJECT_BINARY_DIR}/hcc_version.tmp)
add_custom_target(check_hcc_version COMMAND ${CMAKE_COMMAND} -E copy_if_different ${PROJECT_BINARY_DIR}/hcc_version.tmp ${PROJECT_BINARY_DIR}/hcc_version.txt DEPENDS query_hcc_version)
@@ -210,7 +232,7 @@ add_custom_target(doc COMMAND HIP_PATH=${CMAKE_CURRENT_SOURCE_DIR} doxygen ${CMA
# Install hip_hcc if platform is hcc
if(HIP_PLATFORM STREQUAL "hcc")
install(TARGETS hip_hcc_static hip_hcc hip_device DESTINATION lib)
install(FILES ${CMAKE_CURRENT_SOURCE_DIR}/src/hip_ir.ll DESTINATION lib)
install(FILES ${CMAKE_CURRENT_SOURCE_DIR}/src/hip_hc.ll ${CMAKE_CURRENT_SOURCE_DIR}/src/hip_hc_gfx803.ll DESTINATION lib)
# Install .hipInfo
install(FILES ${PROJECT_BINARY_DIR}/.hipInfo DESTINATION lib)
@@ -229,6 +251,39 @@ if(NOT ${INSTALL_SOURCE} EQUAL 0)
install(DIRECTORY cmake DESTINATION .)
endif()
#############################
# hip-config
#############################
set(LIB_INSTALL_DIR ${CMAKE_INSTALL_PREFIX}/lib)
set(INCLUDE_INSTALL_DIR ${CMAKE_INSTALL_PREFIX}/include)
set(BIN_INSTALL_DIR ${CMAKE_INSTALL_PREFIX}/bin)
set(CONFIG_PACKAGE_INSTALL_DIR ${LIB_INSTALL_DIR}/cmake/hip)
if(HIP_PLATFORM STREQUAL "hcc")
install(TARGETS hip_hcc_static hip_hcc hip_device EXPORT hip-targets DESTINATION ${LIB_INSTALL_DIR})
install(EXPORT hip-targets DESTINATION ${CONFIG_PACKAGE_INSTALL_DIR} NAMESPACE hip::)
include(CMakePackageConfigHelpers)
configure_package_config_file(
hip-config.cmake.in
${CMAKE_CURRENT_BINARY_DIR}/hip-config.cmake
INSTALL_DESTINATION ${CONFIG_PACKAGE_INSTALL_DIR}
PATH_VARS LIB_INSTALL_DIR INCLUDE_INSTALL_DIR BIN_INSTALL_DIR
)
write_basic_package_version_file(
${CMAKE_CURRENT_BINARY_DIR}/hip-config-version.cmake
VERSION "${HIP_VERSION}"
COMPATIBILITY SameMajorVersion
)
install(
FILES
${CMAKE_CURRENT_BINARY_DIR}/hip-config.cmake
${CMAKE_CURRENT_BINARY_DIR}/hip-config-version.cmake
DESTINATION
${CONFIG_PACKAGE_INSTALL_DIR}
)
endif()
#############################
# Packaging steps
#############################
@@ -268,7 +323,7 @@ add_custom_target(pkg_hip_hcc COMMAND ${CMAKE_COMMAND} .
COMMAND cp *.rpm ${PROJECT_BINARY_DIR}
COMMAND cp *.tar.gz ${PROJECT_BINARY_DIR}
WORKING_DIRECTORY ${BUILD_DIR}
DEPENDS hip_hcc)
DEPENDS hip_hcc hip_device hip_hcc_static)
# Package: hip_nvcc
set(BUILD_DIR ${CMAKE_CURRENT_BINARY_DIR}/packages/hip_nvcc)
@@ -317,22 +372,28 @@ endif()
# Testing steps
#############################
# Target: test
set(BUILD_DIR ${CMAKE_CURRENT_BINARY_DIR}/hip_tests)
configure_file(tests/hip_tests.txt ${BUILD_DIR}/CMakeLists.txt @ONLY)
if(POLICY CMP0037)
cmake_policy(PUSH)
cmake_policy(SET CMP0037 OLD)
set(HIP_ROOT_DIR ${CMAKE_INSTALL_PREFIX})
set(HIP_SRC_PATH ${CMAKE_CURRENT_SOURCE_DIR})
execute_process(COMMAND "${CMAKE_COMMAND}" -E copy_directory "${HIP_SRC_PATH}/cmake" "${HIP_ROOT_DIR}/cmake" RESULT_VARIABLE RUN_HIT ERROR_QUIET)
if(${RUN_HIT} EQUAL 0)
execute_process(COMMAND "${CMAKE_COMMAND}" -E copy_directory "${HIP_SRC_PATH}/bin" "${HIP_ROOT_DIR}/bin" RESULT_VARIABLE RUN_HIT ERROR_QUIET)
endif()
add_custom_target(install_for_test COMMAND "${CMAKE_COMMAND}" --build . --target install
WORKING_DIRECTORY ${CMAKE_BINARY_DIR})
execute_process(COMMAND getconf _NPROCESSORS_ONLN OUTPUT_VARIABLE DASH_JAY OUTPUT_STRIP_TRAILING_WHITESPACE)
add_custom_target(test COMMAND ${CMAKE_COMMAND} -DCMAKE_BUILD_TYPE=${CMAKE_BUILD_TYPE} .
COMMAND make -j ${DASH_JAY}
COMMAND make test
WORKING_DIRECTORY ${BUILD_DIR}
DEPENDS install_for_test)
if(POLICY CMP0037)
cmake_policy(POP)
if(${RUN_HIT} EQUAL 0)
set(CMAKE_MODULE_PATH "${HIP_ROOT_DIR}/cmake" ${CMAKE_MODULE_PATH})
include(${HIP_SRC_PATH}/tests/hit/HIT.cmake)
# Add tests
include_directories(${HIP_SRC_PATH}/tests/src)
hit_add_directory_recursive(${HIP_SRC_PATH}/tests/src "directed_tests")
# Add top-level tests to build_tests
add_custom_target(build_tests DEPENDS directed_tests)
# Add custom target: check
add_custom_target(check COMMAND "${CMAKE_COMMAND}" --build . --target test DEPENDS build_tests)
else()
message(STATUS "Testing targets will not be available. To enable them please ensure that the HIP installation directory is writeable. Use -DCMAKE_INSTALL_PREFIX to specify a suitable location")
endif()
# vim: ts=4:sw=4:expandtab:smartindent
projects/hip/CONTRIBUTING.md
+4 -4
Datei anzeigen
@@ -93,7 +93,7 @@ Differences or limitations of HIP APIs as compared to CUDA APIs should be clearl
## Coding Guidelines (in brief)
- Code Indentation:
- Tabs should be expanded to spaces.
- Use 4 spaces indendation.
- Use 4 spaces indentation.
- Capitalization and Naming
- Prefer camelCase for HIP interfaces and internal symbols. Note HCC uses _ for separator.
This guideline is not yet consistently followed in HIP code - eventual compliance is aspirational.
@@ -120,13 +120,13 @@ Differences or limitations of HIP APIs as compared to CUDA APIs should be clearl
- HIP_INIT_API() should be placed at the start of each top-level HIP API. This function will make sure the HIP runtime is initialized,
and also constructs an appropriate API string for tracing and CodeXL marker tracing. The arguments to HIP_INIT_API should match
those of the parent fucntion.
those of the parent function.
- ihipLogStatus should only be called from top-level HIP APIs,and should be called to log and return the error code. The error code
is used by the GetLastError and PeekLastError functions - if a HIP API simply returns, then the error will not be logged correctly.
- All HIP environment variables should begin with the keyword HIP_
Environment variables should be long enough to describe their purpose but short enough so they can be remembered - perhaps 10-20 characters, with 3-4 parts separated by underscores.
To see the list of current environment variables, along with their values, set HIP_PRINT_ENV and run any hip applications on ROCM platform .
To see the list of current environment variables, along with their values, set HIP_PRINT_ENV and run any hip applications on ROCm platform .
HIPCC or other tools may support additional environment variables which should follow the above convention.
@@ -161,4 +161,4 @@ doxygen bug list.
## Other Tips:
### Markdown Editing
Recommended to use an offline Markown viewer to review documentation, such as Markdown Preview Plus extension in Chrome browser, or Remarkable.
Recommended to use an offline Markdown viewer to review documentation, such as Markdown Preview Plus extension in Chrome browser, or Remarkable.
projects/hip/INSTALL.md
+3 -3
Datei anzeigen
@@ -22,14 +22,14 @@ HIP code can be developed either on AMD ROCm platform using hcc compiler, or a C
## 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.
* 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.
* Optionally, consider adding /opt/rocm/bin to your PATH to make it easier to use the tools.
## NVIDIA-nvcc
@@ -96,7 +96,7 @@ The native GCN target is included with upstream LLVM, and has also been integrat
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.
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:
projects/hip/README.md
+3 -3
Datei anzeigen
@@ -88,8 +88,8 @@ The HIP Runtime API code and compute kernel definition can exist in the same sou
## HIP Portability and Compiler Technology
HIP C++ code can be compiled with either :
- On the Nvidia CUDA platform, HIP provides header file which translate from the HIP runtime APIs to CUDA runtime APIs. The header file contains mostly inlined
functions and thus has very low overhead - developers coding in HIP should expect the same perforamnce as coding in native CUDA. The code is then
- On the NVIDIA CUDA platform, HIP provides header file which translate from the HIP runtime APIs to CUDA runtime APIs. The header file contains mostly inlined
functions and thus has very low overhead - developers coding in HIP should expect the same performance as coding in native CUDA. The code is then
compiled with nvcc, the standard C++ compiler provided with the CUDA SDK. Developers can use any tools supported by the CUDA SDK including the CUDA
profiler and debugger.
- On the AMD ROCm platform, HIP provides a header and runtime library built on top of hcc compiler. The HIP runtime implements HIP streams, events, and memory APIs,
@@ -120,7 +120,7 @@ make
## More Examples
The GitHub repot [HIP-Examples](https://github.com/GPUOpen-ProfessionalCompute-Tools/HIP-Examples.git) contains a hipified vesion of the popular Rodinia benchmark suite.
The GitHub repository [HIP-Examples](https://github.com/GPUOpen-ProfessionalCompute-Tools/HIP-Examples.git) contains a hipified version of the popular Rodinia benchmark suite.
The README with the procedures and tips the team used during this porting effort is here: [Rodinia Porting Guide](https://github.com/GPUOpen-ProfessionalCompute-Tools/HIP-Examples/blob/master/rodinia_3.0/hip/README.hip_porting)
## Tour of the HIP Directories
projects/hip/RELEASE.md
+48 -2
Datei anzeigen
@@ -3,7 +3,7 @@
We have attempted to document known bugs and limitations - in particular the [HIP Kernel Language](docs/markdown/hip_kernel_language.md) document uses the phrase "Under Development", and the [HIP Runtime API bug list](http://gpuopen-professionalcompute-tools.github.io/HIP/bug.html) lists known bugs.
Upcoming:
- Stability: Enforce perioidic host synchronization to reclaim resources if the application has launched a large
- Stability: Enforce periodic host synchronization to reclaim resources if the application has launched a large
number of commands (>1K) without synchronizing.
- Register keyword now silently ignored on HCC (previously would emit warning).
- Doc updates: Add some more frequently asked questions to FAQ, fix TOC in some files, review.
@@ -13,6 +13,52 @@ Upcoming:
## Revision History:
===================================================================================================
Release: 1.0.17102
Date: 2017.03.07
- Lots of improvements to hipify-clang.
- Added HIP package config for cmake.
- Several bug fixes and documentation updates.
===================================================================================================
Release: 1.0.17066
Date: 2017.02.11
- Improved support for math device functions.
- Added several half math device functions.
- Enabled support for CUDA 8.0 in hipify-clang.
- Lots of bug fixes and documentation updates.
===================================================================================================
Release: 1.0.17015
Date: 2017.01.06
- Several improvements to the hipify-clang infrastructure.
- Refactored module and function APIs.
- HIP now defaults to linking against the shared runtime library.
- Documentation updates.
===================================================================================================
Release: 1.0.16502
Date: 2016.12.13
- Added several fast math and packaged math instrincs
- Improved debug and profiler documentation
- Support for building and linking to HIP shared library
- Several improvements to hipify-clang
- Several bug fixes
===================================================================================================
Release: 1.0.16461
Date: 2016.11.14
- Significant changes to the HIP Profiling APIs. Refer to the documentation for details
- Improvements to P2P support
- New API: hipDeviceGetByPCIBusId
- Several bug fixes in NV path
- hipModuleLaunch now works for multi-dim kernels
===================================================================================================
Release:1.0
Date: 2016.11.8
@@ -73,7 +119,7 @@ Date: 2016.04.25
- Create static library and link.
- Set HIP_PATH to install.
- Make hipDevice and hipStream thread-safe.
- Prefered hipStream usage is still to create new streams for each new thread, but it works even if you don;t.
- Preferred hipStream usage is still to create new streams for each new thread, but it works even if you don;t.
- Improve automated platform detection: If AMD GPU is installed and detected by driver, default HIP_PLATFORM to hcc.
- HIP_TRACE_API now prints arguments to the HIP function (in addition to name of function).
- Deprecate hipDeviceGetProp (Replace with hipGetDeviceProp)
projects/hip/bin/hccgenco.sh
+7 -8
Datei anzeigen
@@ -10,7 +10,7 @@ if [ $# = 0 ]; then
fi
: ${ROCM_PATH:=/opt/rocm}
: ${ROCM_TARGET:=fiji}
: ${ROCM_TARGET:=gfx803}
INPUT_FILES=""
OUTPUT_FILE=""
@@ -45,18 +45,17 @@ for inputfile in $INPUT_FILES; do
done
printf "\nint main(){}\n" >> $hipgenisa_main
$HIP_PATH/bin/hipcc $hipgenisa_files -o $hipgenisa_dir/a.out
mv dump.* $hipgenisa_dir
$ROCM_PATH/hcc-lc/bin/llvm-mc -arch=amdgcn -mcpu=$ROCM_TARGET -filetype=obj $hipgenisa_dir/dump.isa -o $hipgenisa_dir/dump.o
$ROCM_PATH/llvm/bin/clang -target amdgcn--amdhsa $hipgenisa_dir/dump.o -o $hipgenisa_dir/dump.co
$HIP_PATH/bin/hipcc -DGENERIC_GRID_LAUNCH=0 $hipgenisa_files -o $hipgenisa_dir/a.out
mv dump* $hipgenisa_dir
hsaco_file="dump-$ROCM_TARGET.hsaco"
map_sym=""
kernels=$(objdump -t $hipgenisa_dir/dump.co | grep grid_launch_parm | sed 's/ \+/ /g; s/\t/ /g' | cut -d" " -f6)
kernels=$(objdump -t $hipgenisa_dir/$hsaco_file | grep grid_launch_parm | sed 's/ \+/ /g; s/\t/ /g' | cut -d" " -f6)
for mangled_sym in $kernels; do
real_sym=$(c++filt $(c++filt _$mangled_sym | cut -d: -f3 | sed 's/_functor//g') | cut -d\( -f1)
real_sym=$(c++filt $(c++filt $mangled_sym | cut -d: -f3 | sed 's/_functor//g') | cut -d\( -f1)
map_sym="--redefine-sym $mangled_sym=$real_sym $map_sym"
done
objcopy -F elf64-little $map_sym $hipgenisa_dir/dump.co $OUTPUT_FILE
objcopy -F elf64-little $map_sym $hipgenisa_dir/$hsaco_file $OUTPUT_FILE
rm $hipgenisa_files
rm -r $hipgenisa_dir
projects/hip/bin/hipcc
+159 -83
Datei anzeigen
@@ -23,8 +23,8 @@ use File::Basename;
# HSA_PATH : Path to HSA dir (default /opt/rocm/hsa). Used on AMD platforms only.
if(scalar @ARGV == 0){
print "No Arguments passed, exiting ...\n";
exit(-1);
print "No Arguments passed, exiting ...\n";
exit(-1);
}
#---
@@ -70,28 +70,40 @@ if ($verbose & 0x2) {
# set if user explicitly requests -stdlib=libc++. (else we default to libstdc++ for better interop with g++):
$setStdLib = 0; # TODO - set to 0
$target_gfx701 = 0;
$target_gfx801 = 0;
$target_gfx802 = 0;
$target_gfx803 = 0;
$target_gfx900 = 0;
if ($HIP_PLATFORM eq "hcc") {
$HSA_PATH=$ENV{'HSA_PATH'} // "/opt/rocm/hsa";
$HCC_HOME=$ENV{'HCC_HOME'} // $hipConfig{'HCC_HOME'} // "/opt/rocm/hcc";
$HCC_VERSION=`${HCC_HOME}/bin/hcc --version | cut -d" " -f9 | tr -d "\n"`;
$HCC_VERSION_MAJOR=`${HCC_HOME}/bin/hcc --version | cut -d" " -f9 | cut -d"." -f1 | tr -d "\n"`;
$HCC_VERSION=`${HCC_HOME}/bin/hcc --version`;
$HCC_VERSION=~/.*based on HCC ([^ ]+).*/;
$HCC_VERSION=$1;
$HCC_VERSION_MAJOR=$HCC_VERSION;
$HCC_VERSION_MAJOR=~s/\..*//;
$ROCM_PATH=$ENV{'ROCM_PATH'} // "/opt/rocm";
$HIP_ATP_MARKER=$ENV{'HIP_ATP_MARKER'} // 1;
$marker_path = "$ROCM_PATH/profiler/CXLActivityLogger";
$ROCM_TARGET=$ENV{'ROCM_TARGET'} // "fiji";
$ROCM_TARGET=$ENV{'ROCM_TARGET'} // "gfx803";
# HCC* may be used to compile src/hip_hcc.o (and also feed the HIPCXXFLAGS below)
$HCC = "$HCC_HOME/bin/hcc";
$HCCFLAGS = "-hc -I$HCC_HOME/include ";
$HCCFLAGS = "-hc -D__HIPCC__ -I$HCC_HOME/include ";
$HIPCC=$HCC;
$HIPCXXFLAGS = $HCCFLAGS;
$HIPLDFLAGS = `${HCC_HOME}/bin/hcc-config --ldflags`;
#### GCC system includes workaround ####
$HCC_WA_FLAGS = " ";
if ($HCC_VERSION_MAJOR eq 1) {
@@ -99,49 +111,33 @@ if ($HIP_PLATFORM eq "hcc") {
my $GPP_CUR_VER = `g++ -dumpversion`;
$GCC_CUR_VER =~ s/\R//g;
$GPP_CUR_VER =~ s/\R//g;
if (${GCC_CUR_VER} eq ${GPP_CUR_VER}) {
$HCC_WA_FLAGS .= " -I/usr/include/x86_64-linux-gnu -I/usr/include/x86_64-linux-gnu/c++/${GCC_CUR_VER} -I/usr/include/c++/${GCC_CUR_VER} ";
my @GPP_VER_FIELDS = split('\.', $GPP_CUR_VER);
# Only include the libstdc++ headers and libraries flags explicitly if the g++ is older than version 5.
# That's because HCC already uses libstdc++ by default if a newer g++/libstdc++ is available
if (${GCC_CUR_VER} eq ${GPP_CUR_VER} and $GPP_VER_FIELDS[0] < 5) {
$HCC_WA_FLAGS .= " -stdlib=libstdc++ -I/usr/include/x86_64-linux-gnu -I/usr/include/x86_64-linux-gnu/c++/${GCC_CUR_VER} -I/usr/include/c++/${GCC_CUR_VER} ";
# Add C++ libs for GCC.
$HIPLDFLAGS .= " -lstdc++";
}
}
# Force -stdlib=libc++ on UB14.04
$HOST_OSNAME= `cat /etc/os-release | grep "^ID\=" | cut -d= -f2 | tr -d '\n'`;
$HOST_OSVER= `cat /etc/os-release | grep "^VERSION_ID\=" | cut -d= -f2 | tr -d '\n'`;
if ($HOST_OSNAME eq "ubuntu" and $HOST_OSVER eq "\"14.04\"") {
$HIPCXXFLAGS .= " -stdlib=libc++";
$setStdLib = 1;
}
$HIPCXXFLAGS .= " -I$HIP_PATH/include/hip/hcc_detail/cuda";
$HIPCXXFLAGS .= " -I$HSA_PATH/include";
$HIPCXXFLAGS .= " -Wno-deprecated-register";
$HIPLDFLAGS = "-hc -L$HCC_HOME/lib -Wl,--rpath=$HCC_HOME/lib -lc++ -ldl -lpthread -Wl,--whole-archive -lmcwamp -Wl,--no-whole-archive";
# Suppress linker warnings in case HCC distribution contains OpenCL/SPIR symbols
$HOST_OSNAME= `cat /etc/os-release | grep "^ID\=" | cut -d= -f2 | tr -d '\n'`;
$HOST_OSVER= `cat /etc/os-release | grep "^VERSION_ID\=" | cut -d= -f2 | tr -d '\n'`;
if ($HOST_OSNAME eq "ubuntu" and $HOST_OSVER eq "\"16.04\"") {
# No additional flags required
} else {
$HIPLDFLAGS .= " -Wl,--defsym=_binary_kernel_spir_end=1 -Wl,--defsym=_binary_kernel_spir_start=1 -Wl,--defsym=_binary_kernel_cl_start=1 -Wl,--defsym=_binary_kernel_cl_end=1";
}
if ($HOST_OSNAME eq "fedora") {
$HIPCXXFLAGS .= " -I/usr/local/include/c++/v1";
}
# Satisfy HCC dependencies
if ($HOST_OSNAME eq "fedora") {
$HIPLDFLAGS .= " -lc++abi";
} else {
$HIPLDFLAGS .= " -lc++abi -lsupc++";
}
$HIPLDFLAGS .= " -L$HSA_PATH/lib -L$ROCM_PATH/lib -lhsa-runtime64 -lhc_am -lhsakmt";
# Handle ROCm target platform
if ($ROCM_TARGET eq "fiji") {
$HIPLDFLAGS .= " -amdgpu-target=AMD:AMDGPU:8:0:3";
}
if ($ROCM_TARGET eq "carrizo") {
$HIPLDFLAGS .= " -amdgpu-target=AMD:AMDGPU:8:0:1";
}
if ($ROCM_TARGET eq "hawaii") {
$HIPLDFLAGS .= " -amdgpu-target=AMD:AMDGPU:7:0:1";
}
if ($ROCM_TARGET eq "polaris") {
$HIPLDFLAGS .= " -amdgpu-target=AMD:AMDGPU:8:0:3";
}
$HIPLDFLAGS .= " -lsupc++";
$HIPLDFLAGS .= " -L$HSA_PATH/lib -L$ROCM_PATH/lib -lhsa-runtime64 -lhc_am -lhsakmt ";
# $HIPLDFLAGS .= " -L$HCC_HOME/compiler/lib -lLLVMAMDGPUDesc -lLLVMAMDGPUUtils -lLLVMMC -lLLVMCore -lLLVMSupport ";
# Add trace marker library:
# TODO - once we cleanly separate the HIP API headers from HIP library headers this logic should move to CMakebuild option - apps do not need to see the marker library.
@@ -157,8 +153,6 @@ if ($HIP_PLATFORM eq "hcc") {
$HIPLDFLAGS .= " -L$marker_lib_path -lCXLActivityLogger -Wl,--rpath=$marker_lib_path";
}
# Add C++ libs for GCC.
$HIPLDFLAGS .= " -lstdc++";
$HIPLDFLAGS .= " -lm";
if ($verbose & 0x2) {
@@ -193,7 +187,7 @@ my $needHipHcc = ($HIP_PLATFORM eq 'hcc'); # set if we need to link hip_hcc
my $printHipVersion = 0; # print HIP version
my $runCmd = 1;
my $buildDeps = 0;
my $linkType = 0;
my $linkType = 1;
my $setLinkType = 0;
my @options = ();
@@ -206,44 +200,47 @@ if ($verbose & 0x4) {
# Handle code object generation
my $ISACMD="";
if($HIP_PLATFORM eq "hcc"){
$ISACMD .= "set ROCM_PATH=$ROCM_PATH && set ROCM_TARGET=$ROCM_TARGET && $HIP_PATH/bin/hccgenco.sh ";
if($ARGV[0] eq "--genco"){
foreach $isaarg (@ARGV[1..$#ARGV]){
$ISACMD .= " ";
$ISACMD .= $isaarg;
$ISACMD .= "set ROCM_PATH=$ROCM_PATH && set ROCM_TARGET=$ROCM_TARGET && $HIP_PATH/bin/hccgenco.sh ";
if($ARGV[0] eq "--genco"){
foreach $isaarg (@ARGV[1..$#ARGV]){
$ISACMD .= " ";
$ISACMD .= $isaarg;
}
if ($verbose & 0x1) {
print "hipcc-cmd: ", $ISACMD, "\n";
}
system($ISACMD) and die();
exit(0);
}
if ($verbose & 0x1) {
print "hipcc-cmd: ", $ISACMD, "\n";
}
system($ISACMD) and die();
exit(0);
}
}
if(($HIP_PLATFORM eq "hcc") and defined $ENV{HIP_EXPERIMENTAL}){
if(($HIP_PLATFORM eq "hcc")){
$EXPORT_LL=" ";
$ENV{HCC_EXTRA_LIBRARIES}="$HIP_PATH/lib/hip_ir.ll\n";
$ENV{HCC_EXTRA_LIBRARIES}="$HIP_PATH/lib/hip_hc.ll\n";
$ENV{HIP_HC_IR_FILE}="";
}
if($HIP_PLATFORM eq "nvcc"){
$ISACMD .= "$HIP_PATH/bin/hipcc -ptx ";
if($ARGV[0] eq "--genco"){
foreach $isaarg (@ARGV[1..$#ARGV]){
$ISACMD .= " ";
$ISACMD .= $isaarg;
$ISACMD .= "$HIP_PATH/bin/hipcc -ptx ";
if($ARGV[0] eq "--genco"){
foreach $isaarg (@ARGV[1..$#ARGV]){
$ISACMD .= " ";
$ISACMD .= $isaarg;
}
if ($verbose & 0x1) {
print "hipcc-cmd: ", $ISACMD, "\n";
}
system($ISACMD) and die();
exit(0);
}
if ($verbose & 0x1) {
print "hipcc-cmd: ", $ISACMD, "\n";
}
system($ISACMD) and die();
exit(0);
}
}
my $toolArgs = ""; # arguments to pass to the hcc or nvcc tool
foreach $arg (@ARGV)
{
$trimarg = $arg;
$trimarg =~ s/^\s+|\s+$//g; # Remive whitespace
my $swallowArg = 0;
if ($arg eq '-c') {
$compileOnly = 1;
@@ -254,38 +251,57 @@ foreach $arg (@ARGV)
$needLDFLAGS = 1;
}
if(($arg eq '-stdlib=libc++') and ($setStdLib eq 0))
if(($trimarg eq '-stdlib=libc++') and ($setStdLib eq 0))
{
$HIPCXXFLAGS .= " -stdlib=libc++";
$setStdLib = 1;
}
if(($arg eq '-stdlib=libstdc++') and ($setStdLib eq 0))
if($arg eq '--amdgpu-target=gfx701')
{
$target_gfx701 = 1;
}
if($arg eq '--amdgpu-target=gfx801')
{
$target_gfx801 = 1;
}
if($arg eq '--amdgpu-target=gfx802')
{
$target_gfx802 = 1;
}
if($arg eq '--amdgpu-target=gfx803')
{
$target_gfx803 = 1;
}
if($arg eq '--amdgpu-target=gfx900')
{
$target_gfx900 = 1;
}
if(($trimarg eq '-stdlib=libstdc++') and ($setStdLib eq 0))
{
$HIPCXXFLAGS .= " -stdlib=libstdc++";
$HIPCXXFLAGS .= $HCC_WA_FLAGS;
$setStdLib = 1;
}
if($arg eq '--version') {
if($trimarg eq '--version') {
$printHipVersion = 1;
}
if($arg eq '--short-version') {
if($trimarg eq '--short-version') {
$printHipVersion = 1;
$runCmd = 0;
}
if($arg eq '-M') {
if($trimarg eq '-M') {
$compileOnly = 1;
$buildDeps = 1;
}
if($arg eq '-use_fast_math') {
print "In fast Math";
if($trimarg eq '-use_fast_math') {
$HIPCXXFLAGS .= " -DHIP_FAST_MATH ";
}
if(($arg eq '-use-staticlib') and ($setLinkType eq 0))
if(($trimarg eq '-use-staticlib') and ($setLinkType eq 0))
{
$linkType = 0;
$setLinkType = 1;
}
if(($arg eq '-use-sharedlib') and ($setLinkType eq 0))
if(($trimarg eq '-use-sharedlib') and ($setLinkType eq 0))
{
$linkType = 1;
$setLinkType = 1;
@@ -320,6 +336,64 @@ foreach $arg (@ARGV)
}
$toolArgs .= " $arg" unless $swallowArg;
}
if(defined $ENV{HCC_AMDGPU_TARGET})
{
foreach my $target (split(/,/, $ENV{HCC_AMDGPU_TARGET}))
{
if($target eq 'gfx701')
{
$target_gfx701 = 1;
}
if($target eq 'gfx801')
{
$target_gfx801 = 1;
}
if($target eq 'gfx802')
{
$target_gfx802 = 1;
}
if($target eq 'gfx803')
{
$target_gfx803 = 1;
}
if($target eq 'gfx900')
{
$target_gfx900 = 1;
}
}
}
if ($target_gfx701 eq 0 and $target_gfx801 eq 0 and $target_gfx802 eq 0 and $target_gfx803 eq 0 and $target_gfx900 eq 0)
{
$target_gfx803 = 1;
}
if($HIP_PLATFORM eq "hcc"){
$ENV{HCC_EXTRA_LIBRARIES}="$HIP_PATH/lib/hip_hc.ll\n";
# Handle ROCm target platform
if ($target_gfx701 eq 1) {
$HIPLDFLAGS .= " --amdgpu-target=gfx701";
$HIPCXXFLAGS .= " -D__HIP_ARCH_GFX701__=1 ";
}
if ($target_gfx801 eq 1) {
$HIPLDFLAGS .= " --amdgpu-target=gfx801";
$HIPCXXFLAGS .= " -D__HIP_ARCH_GFX801__=1 ";
}
if ($target_gfx802 eq 1) {
$HIPLDFLAGS .= " --amdgpu-target=gfx802";
$HIPCXXFLAGS .= " -D__HIP_ARCH_GFX802__=1 ";
}
if ($target_gfx803 eq 1) {
$HIPLDFLAGS .= " --amdgpu-target=gfx803";
$HIPCXXFLAGS .= " -D__HIP_ARCH_GFX803__=1 ";
$ENV{HCC_EXTRA_LIBRARIES_GFX803}="$HIP_PATH/lib/hip_hc_gfx803.ll\n";
}
if ($target_gfx900 eq 1) {
$HIPLDFLAGS .= " --amdgpu-target=gfx900";
$HIPCXXFLAGS .= " -D__HIP_ARCH_GFX900__=1 ";
}
}
if ($hasC and $HIP_PLATFORM eq 'nvcc') {
$HIPCXXFLAGS .= " -x cu";
@@ -333,15 +407,14 @@ if ($buildDeps and $HIP_PLATFORM eq 'nvcc') {
if ($setStdLib eq 0 and $HIP_PLATFORM eq 'hcc')
{
$HIPCXXFLAGS .= " -stdlib=libstdc++";
$HIPCXXFLAGS .= $HCC_WA_FLAGS;
}
if ($needHipHcc) {
if ($linkType eq 0) {
substr($HIPLDFLAGS,0,0) = " -L$HIP_PATH/lib -lhip_hcc_static -lhip_device " ;
substr($HIPLDFLAGS,0,0) = " $HIP_PATH/lib/libhip_hcc_static.a $HIP_PATH/lib/libhip_device.a " ;
} else {
substr($HIPLDFLAGS,0,0) = " -L$HIP_PATH/lib -Wl,--rpath=$HIP_PATH/lib -lhip_hcc -lhip_device ";
substr($HIPLDFLAGS,0,0) = " -Wl,--rpath=$HIP_PATH/lib $HIP_PATH/lib/libhip_hcc.so $HIP_PATH/lib/libhip_device.a ";
}
}
@@ -373,7 +446,10 @@ if ($printHipVersion) {
}
if ($runCmd) {
if ($HIP_PLATFORM eq "hcc" and exists($hipConfig{'HCC_VERSION'}) and $HCC_VERSION ne $hipConfig{'HCC_VERSION'}) {
print ("HIP ($HIP_PATH) was built using hcc $hipConfig{'HCC_VERSION'}, but you are using $HCC_HOME/hcc with version $HCC_VERSION from hipcc. Please rebuild HIP including cmake or update HCC_HOME variable.\n") && die ();
print ("HIP ($HIP_PATH) was built using hcc $hipConfig{'HCC_VERSION'}, but you are using $HCC_HOME/hcc with version $HCC_VERSION from hipcc. Please rebuild HIP including cmake or update HCC_HOME variable.\n") ;
die unless $ENV{'HIP_IGNORE_HCC_VERSION'};
}
system ("$CMD") and die ();
}
# vim: ts=4:sw=4:expandtab:smartindent
projects/hip/bin/hipcc_cmake_linker_helper
Ausführbare Datei
+5
Datei anzeigen
@@ -0,0 +1,5 @@
#!/bin/bash
SOURCE="${BASH_SOURCE[0]}"
HIP_PATH="$( command cd -P "$( dirname "$SOURCE" )/.." && pwd )"
HCC_HOME=$1 $HIP_PATH/bin/hipcc "${@:2}"
projects/hip/bin/hipconvertinplace-perl.sh
Ausführbare Datei
+18
Datei anzeigen
@@ -0,0 +1,18 @@
#!/bin/bash
#usage : hipconvertinplace.sh [DIRNAME] [HIPIFY_OPTIONS]
#hipify "inplace" all code files in specified directory.
# This can be quite handy when dealing with an existing CUDA code base since the script
# preserves the existing directory structure.
# For each code file, this script will:
# - If ".prehip file does not exist, copy the original code to a new file with extension ".prehip". Then Hipify the code file.
# - If ".prehip" file exists, this is used as input to hipify.
# (this is useful for testing improvements to the hipify toolset).
SCRIPT_DIR=`dirname $0`
SEARCH_DIR=$1
shift
$SCRIPT_DIR/hipify -inplace -print-stats "$@" `$SCRIPT_DIR/findcode.sh $SEARCH_DIR`
projects/hip/bin/hipconvertinplace.sh
+16 -10
Datei anzeigen
@@ -1,18 +1,24 @@
#!/bin/bash
#usage : hipconvertinplace.sh [DIRNAME] [HIPIFY_OPTIONS]
#usage : hipconvertinplace.sh DIRNAME [hipify options] [--] [clang options]
#hipify "inplace" all code files in specified directory.
#hipify "inplace" all code files in specified directory.
# This can be quite handy when dealing with an existing CUDA code base since the script
# preserves the existing directory structure.
# For each code file, this script will:
# - If ".prehip file does not exist, copy the original code to a new file with extension ".prehip". Then Hipify the code file.
# - If ".prehip" file exists, this is used as input to hipify.
# (this is useful for testing improvements to the hipify toolset).
SCRIPT_DIR=`dirname $0`
SEARCH_DIR=$1
shift
$SCRIPT_DIR/hipify -inplace -print-stats "$@" `$SCRIPT_DIR/findcode.sh $SEARCH_DIR`
hipify_args=''
while (( "$#" )); do
shift
if [ "$1" != "--" ]; then
hipify_args="$hipify_args $1"
else
shift
break
fi
done
clang_args="$@"
$SCRIPT_DIR/hipify-clang -inplace -print-stats $hipify_args `$SCRIPT_DIR/findcode.sh $SEARCH_DIR` -- -x cuda $clang_args
projects/hip/bin/hipexamine-perl.sh
Ausführbare Datei
+12
Datei anzeigen
@@ -0,0 +1,12 @@
#!/bin/bash
#usage : hipexamine.sh DIRNAME [hipify.pl options]
# Generate HIP stats (LOC, CUDA->API conversions, missing functionality) for all the code files
# in the specified directory.
SCRIPT_DIR=`dirname $0`
SEARCH_DIR=$1
shift
$SCRIPT_DIR/hipify -no-output -print-stats "$@" `$SCRIPT_DIR/findcode.sh $SEARCH_DIR`
projects/hip/bin/hipexamine.sh
+16 -6
Datei anzeigen
@@ -1,12 +1,22 @@
#!/bin/bash
#usage : hipexamine.sh DIRNAME [hipify.pl options]
# Generate HIP stats (LOC, CUDA->API conversions, missing functionality) for all the code files
# in the specified directory.
#usage : hipexamine2.sh DIRNAME [hipify options] [--] [clang options]
# Generate CUDA->HIP conversion statistics for all the code files in the specified directory.
SCRIPT_DIR=`dirname $0`
SEARCH_DIR=$1
shift
$SCRIPT_DIR/hipify -no-output -print-stats "$@" `$SCRIPT_DIR/findcode.sh $SEARCH_DIR`
hipify_args=''
while (( "$#" )); do
shift
if [ "$1" != "--" ]; then
hipify_args="$hipify_args $1"
else
shift
break
fi
done
clang_args="$@"
$SCRIPT_DIR/hipify-clang -examine $hipify_args `$SCRIPT_DIR/findcode.sh $SEARCH_DIR` -- -x cuda $clang_args
projects/hip/cmake/FindHIP.cmake
+44 -3
Datei anzeigen
@@ -105,6 +105,25 @@ if(UNIX AND NOT APPLE AND NOT CYGWIN)
endif()
mark_as_advanced(HIP_HIPCONFIG_EXECUTABLE)
# Find HIPCC_CMAKE_LINKER_HELPER executable
find_program(
HIP_HIPCC_CMAKE_LINKER_HELPER
NAMES hipcc_cmake_linker_helper
PATHS
"${HIP_ROOT_DIR}"
ENV ROCM_PATH
ENV HIP_PATH
/opt/rocm
/opt/rocm/hip
PATH_SUFFIXES bin
NO_DEFAULT_PATH
)
if(NOT HIP_HIPCC_CMAKE_LINKER_HELPER)
# Now search in default paths
find_program(HIP_HIPCC_CMAKE_LINKER_HELPER hipcc_cmake_linker_helper)
endif()
mark_as_advanced(HIP_HIPCC_CMAKE_LINKER_HELPER)
if(HIP_HIPCONFIG_EXECUTABLE AND NOT HIP_VERSION)
# Compute the version
execute_process(
@@ -178,6 +197,21 @@ hip_find_helper_file(run_make2cmake cmake)
hip_find_helper_file(run_hipcc cmake)
###############################################################################
###############################################################################
# MACRO: Reset compiler flags
###############################################################################
macro(HIP_RESET_FLAGS)
unset(HIP_HIPCC_FLAGS)
unset(HIP_HCC_FLAGS)
unset(HIP_NVCC_FLAGS)
foreach(config ${_hip_configuration_types})
string(TOUPPER ${config} config_upper)
unset(HIP_HIPCC_FLAGS_${config_upper})
unset(HIP_HCC_FLAGS_${config_upper})
unset(HIP_NVCC_FLAGS_${config_upper})
endforeach()
endmacro()
###############################################################################
# MACRO: Separate the options from the sources
###############################################################################
@@ -480,8 +514,13 @@ macro(HIP_ADD_EXECUTABLE hip_target)
# Separate the sources from the options
HIP_GET_SOURCES_AND_OPTIONS(_sources _cmake_options _hipcc_options _hcc_options _nvcc_options ${ARGN})
HIP_PREPARE_TARGET_COMMANDS(${hip_target} OBJ _generated_files _source_files ${_sources} HIPCC_OPTIONS ${_hipcc_options} HCC_OPTIONS ${_hcc_options} NVCC_OPTIONS ${_nvcc_options})
list(REMOVE_ITEM _sources ${_source_files})
set(CMAKE_HIP_LINK_EXECUTABLE "${HIP_HIPCC_EXECUTABLE} <FLAGS> <CMAKE_CXX_LINK_FLAGS> <LINK_FLAGS> <OBJECTS> -o <TARGET>")
if(_source_files)
list(REMOVE_ITEM _sources ${_source_files})
endif()
if("x${HCC_HOME}" STREQUAL "x")
set(HCC_HOME "/opt/rocm/hcc")
endif()
set(CMAKE_HIP_LINK_EXECUTABLE "${HIP_HIPCC_CMAKE_LINKER_HELPER} ${HCC_HOME} <FLAGS> <CMAKE_CXX_LINK_FLAGS> <LINK_FLAGS> <OBJECTS> -o <TARGET>")
add_executable(${hip_target} ${_cmake_options} ${_generated_files} ${_sources})
set_target_properties(${hip_target} PROPERTIES LINKER_LANGUAGE HIP)
endmacro()
@@ -493,7 +532,9 @@ macro(HIP_ADD_LIBRARY hip_target)
# Separate the sources from the options
HIP_GET_SOURCES_AND_OPTIONS(_sources _cmake_options _hipcc_options _hcc_options _nvcc_options ${ARGN})
HIP_PREPARE_TARGET_COMMANDS(${hip_target} OBJ _generated_files _source_files ${_sources} ${_cmake_options} HIPCC_OPTIONS ${_hipcc_options} HCC_OPTIONS ${_hcc_options} NVCC_OPTIONS ${_nvcc_options})
list(REMOVE_ITEM _sources ${_source_files})
if(_source_files)
list(REMOVE_ITEM _sources ${_source_files})
endif()
add_library(${hip_target} ${_cmake_options} ${_generated_files} ${_sources})
set_target_properties(${hip_target} PROPERTIES LINKER_LANGUAGE ${HIP_C_OR_CXX})
endmacro()
projects/hip/cmake/FindHIP/run_hipcc.cmake
+4
Datei anzeigen
@@ -26,6 +26,7 @@ set(HIP_HIPCONFIG_EXECUTABLE "@HIP_HIPCONFIG_EXECUTABLE@") #path
set(HIP_HOST_COMPILER "@HIP_HOST_COMPILER@") # path
set(CMAKE_COMMAND "@CMAKE_COMMAND@") # path
set(HIP_run_make2cmake "@HIP_run_make2cmake@") # path
set(HCC_HOME "@HCC_HOME@") #path
@HIP_HOST_FLAGS@
@_HIP_HIPCC_FLAGS@
@@ -42,6 +43,9 @@ execute_process(COMMAND ${HIP_HIPCONFIG_EXECUTABLE} --platform OUTPUT_VARIABLE H
if(NOT host_flag)
set(__CC ${HIP_HIPCC_EXECUTABLE})
if(HIP_PLATFORM STREQUAL "hcc")
if(NOT "x${HCC_HOME}" STREQUAL "x")
set(ENV{HCC_HOME} ${HCC_HOME})
endif()
set(__CC_FLAGS ${HIP_HIPCC_FLAGS} ${HIP_HCC_FLAGS} ${HIP_HIPCC_FLAGS_${build_configuration}} ${HIP_HCC_FLAGS_${build_configuration}})
else()
set(__CC_FLAGS ${HIP_HIPCC_FLAGS} ${HIP_NVCC_FLAGS} ${HIP_HIPCC_FLAGS_${build_configuration}} ${HIP_NVCC_FLAGS_${build_configuration}})
projects/hip/docs/markdown/CUDA_Driver_API_functions_supported_by_HIP.md
+499
Datei anzeigen
@@ -0,0 +1,499 @@
# CUDA Driver API functions supported by HIP
## **1. Data types used by CUDA driver**
| **type** | **CUDA** | **HIP** | **CUDA description** |
|-------------:|---------------------------------------------------------------|------------------------------------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
| struct | `CUDA_ARRAY3D_DESCRIPTOR` | | |
| struct | `CUDA_ARRAY_DESCRIPTOR` | | |
| struct | `CUDA_MEMCPY2D` | | |
| struct | `CUDA_MEMCPY3D` | | |
| struct | `CUDA_MEMCPY3D_PEER` | | |
| struct | `CUDA_POINTER_ATTRIBUTE_P2P_TOKENS` | | |
| struct | `CUDA_RESOURCE_DESC` | | |
| struct | `CUDA_RESOURCE_VIEW_DESC` | | |
| struct | `CUdevprop` | `hipDeviceProp_t` | |
| struct | `CUipcEventHandle` | | |
| struct | `CUipcMemHandle` | | |
| enum |***`CUaddress_mode`*** | | Texture reference addressing modes |
| 0 |*`CU_TR_ADDRESS_MODE_WRAP`* | | Wrapping address mode |
| 1 |*`CU_TR_ADDRESS_MODE_CLAMP`* | | Clamp to edge address mode |
| 2 |*`CU_TR_ADDRESS_MODE_MIRROR`* | | Mirror address mode |
| 3 |*`CU_TR_ADDRESS_MODE_BORDER`* | | Border address mode |
| enum |***`CUarray_cubemap_face`*** | | Array indices for cube faces |
| 0x00 |*`CU_CUBEMAP_FACE_POSITIVE_X`* | | Positive X face of cubemap |
| 0x01 |*`CU_CUBEMAP_FACE_NEGATIVE_X`* | | Negative X face of cubemap |
| 0x02 |*`CU_CUBEMAP_FACE_POSITIVE_Y`* | | Positive Y face of cubemap |
| 0x03 |*`CU_CUBEMAP_FACE_NEGATIVE_Y`* | | Negative Y face of cubemap |
| 0x04 |*`CU_CUBEMAP_FACE_POSITIVE_Z`* | | Positive Z face of cubemap |
| 0x05 |*`CU_CUBEMAP_FACE_NEGATIVE_Z`* | | Negative Z face of cubemap |
| enum |***`CUarray_format`*** | | Array formats |
| 0x01 |*`CU_AD_FORMAT_UNSIGNED_INT8`* | | Unsigned 8-bit integers |
| 0x02 |*`CU_AD_FORMAT_UNSIGNED_INT16`* | | Unsigned 16-bit integers |
| 0x03 |*`CU_AD_FORMAT_UNSIGNED_INT32`* | | Unsigned 32-bit integers |
| 0x08 |*`CU_AD_FORMAT_SIGNED_INT8`* | | Signed 8-bit integers |
| 0x09 |*`CU_AD_FORMAT_SIGNED_INT16`* | | Signed 16-bit integers |
| 0x0a |*`CU_AD_FORMAT_SIGNED_INT32`* | | Signed 32-bit integers |
| 0x10 |*`CU_AD_FORMAT_HALF`* | | 16-bit floating point |
| 0x20 |*`CU_AD_FORMAT_FLOAT`* | | 32-bit floating point |
| enum |***`CUctx_flags`*** | | Context creation flags |
| 0x00 |*`CU_CTX_SCHED_AUTO`* | | Automatic scheduling |
| 0x01 |*`CU_CTX_SCHED_SPIN`* | | Set spin as default scheduling |
| 0x02 |*`CU_CTX_SCHED_YIELD`* | | Set yield as default scheduling |
| 0x04 |*`CU_CTX_SCHED_BLOCKING_SYNC`* | | Set blocking synchronization as default scheduling |
| 0x04 |*`CU_CTX_BLOCKING_SYNC`* | | Set blocking synchronization as default scheduling Deprecated. This flag was deprecated as of CUDA 4.0 and was replaced with CU_CTX_SCHED_BLOCKING_SYNC.|
| 0x07 |*`CU_CTX_SCHED_MASK`* | | |
| 0x08 |*`CU_CTX_MAP_HOST`* | | Support mapped pinned allocations |
| 0x10 |*`CU_CTX_LMEM_RESIZE_TO_MAX`* | | Keep local memory allocation after launch |
| 0x1f |*`CU_CTX_FLAGS_MASK`* | | |
| enum |***`CUdevice_attribute`*** | | Device properties |
| 1 |*`CU_DEVICE_ATTRIBUTE_MAX_THREADS_PER_BLOCK`* |*`hipDeviceAttributeMaxThreadsPerBlock`* | Maximum number of threads per block |
| 2 |*`CU_DEVICE_ATTRIBUTE_MAX_BLOCK_DIM_X`* |*`hipDeviceAttributeMaxBlockDimX`* | Maximum block dimension X |
| 3 |*`CU_DEVICE_ATTRIBUTE_MAX_BLOCK_DIM_Y`* |*`hipDeviceAttributeMaxBlockDimY`* | Maximum block dimension Y |
| 4 |*`CU_DEVICE_ATTRIBUTE_MAX_BLOCK_DIM_Z`* |*`hipDeviceAttributeMaxBlockDimZ`* | Maximum block dimension Z |
| 5 |*`CU_DEVICE_ATTRIBUTE_MAX_GRID_DIM_X`* |*`hipDeviceAttributeMaxGridDimX`* | Maximum grid dimension X |
| 6 |*`CU_DEVICE_ATTRIBUTE_MAX_GRID_DIM_Y`* |*`hipDeviceAttributeMaxGridDimY`* | Maximum grid dimension Y |
| 7 |*`CU_DEVICE_ATTRIBUTE_MAX_GRID_DIM_Z`* |*`hipDeviceAttributeMaxGridDimZ`* | Maximum grid dimension Y |
| 8 |*`CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_BLOCK`* |*`hipDeviceAttributeMaxSharedMemoryPerBlock`* | Maximum shared memory available per block in bytes |
| 8 |*`CU_DEVICE_ATTRIBUTE_SHARED_MEMORY_PER_BLOCK`* |*`hipDeviceAttributeMaxSharedMemoryPerBlock`* | Deprecated, use CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_BLOCK |
| 9 |*`CU_DEVICE_ATTRIBUTE_TOTAL_CONSTANT_MEMORY`* |*`hipDeviceAttributeTotalConstantMemory`* | Memory available on device for __constant__ variables in a CUDA C kernel in bytes |
| 10 |*`CU_DEVICE_ATTRIBUTE_WARP_SIZE`* |*`hipDeviceAttributeWarpSize`* | Warp size in threads |
| 11 |*`CU_DEVICE_ATTRIBUTE_MAX_PITCH`* | | Maximum pitch in bytes allowed by memory copies |
| 12 |*`CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_BLOCK`* |*`hipDeviceAttributeMaxRegistersPerBlock`* | Maximum number of 32-bit registers available per block |
| 12 |*`CU_DEVICE_ATTRIBUTE_REGISTERS_PER_BLOCK`* |*`hipDeviceAttributeMaxRegistersPerBlock`* | Deprecated, use CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_BLOCK |
| 13 |*`CU_DEVICE_ATTRIBUTE_CLOCK_RATE`* |*`hipDeviceAttributeClockRate`* | Typical clock frequency in kilohertz |
| 14 |*`CU_DEVICE_ATTRIBUTE_TEXTURE_ALIGNMENT`* | | Alignment requirement for textures |
| 15 |*`CU_DEVICE_ATTRIBUTE_GPU_OVERLAP`* | | Device can possibly copy memory and execute a kernel concurrently. Deprecated. Use instead CU_DEVICE_ATTRIBUTE_ASYNC_ENGINE_COUNT|
| 16 |*`CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT`* |*`hipDeviceAttributeMultiprocessorCount`* | Number of multiprocessors on device |
| 17 |*`CU_DEVICE_ATTRIBUTE_KERNEL_EXEC_TIMEOUT`* | | Specifies whether there is a run time limit on kernels |
| 18 |*`CU_DEVICE_ATTRIBUTE_INTEGRATED`* | | Device is integrated with host memory |
| 19 |*`CU_DEVICE_ATTRIBUTE_CAN_MAP_HOST_MEMORY`* | | Device can map host memory into CUDA address space |
| 20 |*`CU_DEVICE_ATTRIBUTE_COMPUTE_MODE`* |*`hipDeviceAttributeComputeMode`* | Compute mode (See CUcomputemode for details) |
| 21 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_WIDTH`* | | Maximum 1D texture width |
| 22 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_WIDTH`* | | Maximum 2D texture width |
| 23 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_HEIGHT`* | | Maximum 2D texture height |
| 24 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_WIDTH`* | | Maximum 3D texture width |
| 25 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_HEIGHT`* | | Maximum 3D texture height |
| 26 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_DEPTH`* | | Maximum 3D texture depth |
| 27 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LAYERED_WIDTH`* | | Maximum 2D layered texture width |
| 28 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LAYERED_HEIGHT`* | | Maximum 2D layered texture height |
| 29 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LAYERED_LAYERS`* | | Maximum layers in a 2D layered texture |
| 27 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_ARRAY_WIDTH`* | | Deprecated, use CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LAYERED_WIDTH |
| 28 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_ARRAY_HEIGHT`* | | Deprecated, use CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LAYERED_HEIGHT |
| 29 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_ARRAY_NUMSLICES`* | | Deprecated, use CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LAYERED_LAYERS |
| 30 |*`CU_DEVICE_ATTRIBUTE_SURFACE_ALIGNMENT`* | | Alignment requirement for surfaces |
| 31 |*`CU_DEVICE_ATTRIBUTE_CONCURRENT_KERNELS`* |*`hipDeviceAttributeConcurrentKernels`* | Device can possibly execute multiple kernels concurrently |
| 32 |*`CU_DEVICE_ATTRIBUTE_ECC_ENABLED`* | | Device has ECC support enabled |
| 33 |*`CU_DEVICE_ATTRIBUTE_PCI_BUS_ID`* |*`hipDeviceAttributePciBusId`* | PCI bus ID of the device |
| 34 |*`CU_DEVICE_ATTRIBUTE_PCI_DEVICE_ID`* |*`hipDeviceAttributePciDeviceId`* | PCI device ID of the device |
| 35 |*`CU_DEVICE_ATTRIBUTE_TCC_DRIVER`* | | Device is using TCC driver model |
| 36 |*`CU_DEVICE_ATTRIBUTE_MEMORY_CLOCK_RATE`* |*`hipDeviceAttributeMemoryClockRate`* | Peak memory clock frequency in kilohertz |
| 37 |*`CU_DEVICE_ATTRIBUTE_GLOBAL_MEMORY_BUS_WIDTH`* |*`hipDeviceAttributeMemoryBusWidth`* | Global memory bus width in bits |
| 38 |*`CU_DEVICE_ATTRIBUTE_L2_CACHE_SIZE`* |*`hipDeviceAttributeL2CacheSize`* | Size of L2 cache in bytes |
| 39 |*`CU_DEVICE_ATTRIBUTE_MAX_THREADS_PER_MULTIPROCESSOR`* |*`hipDeviceAttributeMaxThreadsPerMultiProcessor`* | Maximum resident threads per multiprocessor |
| 40 |*`CU_DEVICE_ATTRIBUTE_ASYNC_ENGINE_COUNT`* | | Number of asynchronous engines |
| 41 |*`CU_DEVICE_ATTRIBUTE_UNIFIED_ADDRESSING`* | | Device shares a unified address space with the host |
| 42 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_LAYERED_WIDTH`* | | Maximum 1D layered texture width |
| 43 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_LAYERED_LAYERS`* | | Maximum layers in a 1D layered texture |
| 44 |*`CU_DEVICE_ATTRIBUTE_CAN_TEX2D_GATHER`* | | Deprecated, do not use |
| 45 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_GATHER_WIDTH`* | | Maximum 2D texture width if CUDA_ARRAY3D_TEXTURE_GATHER is set |
| 46 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_GATHER_HEIGHT`* | | Maximum 2D texture height if CUDA_ARRAY3D_TEXTURE_GATHER is set |
| 47 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_WIDTH_ALTERNATE`* | | Alternate maximum 3D texture width |
| 48 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_HEIGHT_ALTERNATE`* | | Alternate maximum 3D texture height |
| 49 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_DEPTH_ALTERNATE`* | | Alternate maximum 3D texture depth |
| 50 |*`CU_DEVICE_ATTRIBUTE_PCI_DOMAIN_ID`* | | PCI domain ID of the device |
| 51 |*`CU_DEVICE_ATTRIBUTE_TEXTURE_PITCH_ALIGNMENT`* | | Pitch alignment requirement for textures |
| 52 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURECUBEMAP_WIDTH`* | | Maximum cubemap texture width/height |
| 53 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURECUBEMAP_LAYERED_WIDTH`* | | Maximum cubemap layered texture width/height |
| 54 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURECUBEMAP_LAYERED_LAYERS`* | | Maximum layers in a cubemap layered texture |
| 55 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE1D_WIDTH`* | | Maximum 1D surface width |
| 56 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE2D_WIDTH`* | | Maximum 2D surface width |
| 57 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE2D_HEIGHT`* | | Maximum 2D surface height |
| 58 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE3D_WIDTH`* | | Maximum 3D surface width |
| 59 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE3D_HEIGHT`* | | Maximum 3D surface height |
| 60 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE3D_DEPTH`* | | Maximum 3D surface depth |
| 61 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE1D_LAYERED_WIDTH`* | | Maximum 1D layered surface width |
| 62 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE1D_LAYERED_LAYERS`* | | Maximum layers in a 1D layered surface |
| 63 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE2D_LAYERED_WIDTH`* | | Maximum 2D layered surface width |
| 64 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE2D_LAYERED_HEIGHT`* | | Maximum 2D layered surface height |
| 65 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE2D_LAYERED_LAYERS`* | | Maximum layers in a 2D layered surface |
| 66 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACECUBEMAP_WIDTH`* | | Maximum cubemap surface width |
| 67 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACECUBEMAP_LAYERED_WIDTH`* | | Maximum cubemap layered surface width |
| 68 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACECUBEMAP_LAYERED_LAYERS`* | | Maximum layers in a cubemap layered surface |
| 69 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_LINEAR_WIDTH`* | | Maximum 1D linear texture width |
| 70 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LINEAR_WIDTH`* | | Maximum 2D linear texture width |
| 71 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LINEAR_HEIGHT`* | | Maximum 2D linear texture height |
| 72 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LINEAR_PITCH`* | | Maximum 2D linear texture pitch in bytes |
| 73 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_MIPMAPPED_WIDTH`* | | Maximum mipmapped 2D texture width |
| 74 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_MIPMAPPED_HEIGHT`* | | Maximum mipmapped 2D texture height |
| 75 |*`CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR`* |*`hipDeviceAttributeComputeCapabilityMajor`* | Major compute capability version number |
| 76 |*`CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR`* |*`hipDeviceAttributeComputeCapabilityMinor`* | Minor compute capability version number |
| 77 |*`CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_MIPMAPPED_WIDTH`* | | Maximum mipmapped 1D texture width |
| 78 |*`CU_DEVICE_ATTRIBUTE_STREAM_PRIORITIES_SUPPORTED`* | | Device supports stream priorities |
| 79 |*`CU_DEVICE_ATTRIBUTE_GLOBAL_L1_CACHE_SUPPORTED`* | | Device supports caching globals in L1 |
| 80 |*`CU_DEVICE_ATTRIBUTE_LOCAL_L1_CACHE_SUPPORTED`* | | Device supports caching locals in L1 |
| 81 |*`CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_MULTIPROCESSOR`* |*`hipDeviceAttributeMaxSharedMemoryPerMultiprocessor`* | Maximum shared memory available per multiprocessor in bytes |
| 82 |*`CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_MULTIPROCESSOR`* | | Maximum number of 32-bit registers available per multiprocessor |
| 83 |*`CU_DEVICE_ATTRIBUTE_MANAGED_MEMORY`* |*`hipDeviceAttributeManagedMemory`* | Device can allocate managed memory on this system |
| 84 |*`CU_DEVICE_ATTRIBUTE_MULTI_GPU_BOARD`* | | Device is on a multi-GPU board |
| 85 |*`CU_DEVICE_ATTRIBUTE_MULTI_GPU_BOARD_GROUP_ID`* | | Unique id for a group of devices on the same multi-GPU board |
| 86 |*`CU_DEVICE_ATTRIBUTE_MAX`* | | |
| enum |***`CUevent_flags`*** | | Event creation flags |
| 0x00 |*`CU_EVENT_DEFAULT`* |*`hipEventDefault`* | Default event flag |
| 0x01 |*`CU_EVENT_BLOCKING_SYNC`* |*`hipEventBlockingSync`* | Event uses blocking synchronization |
| 0x02 |*`CU_EVENT_DISABLE_TIMING`* |*`hipEventDisableTiming`* | Event will not record timing data |
| 0x04 |*`CU_EVENT_INTERPROCESS`* |*`hipEventInterprocess`* | Event is suitable for interprocess use. CU_EVENT_DISABLE_TIMING must be set |
| enum |***`CUfilter_mode`*** |***`hipTextureFilterMode`*** | Texture reference filtering modes |
| 0 |*`CU_TR_FILTER_MODE_POINT`* |*`hipFilterModePoint`* | Point filter mode |
| 1 |*`CU_TR_FILTER_MODE_LINEAR`* |*`hipFilterModeLinear`* | Linear filter mode |
| enum |***`CUfunc_cache`*** |***`hipFuncCache`*** | Function cache configurations |
| 0x00 |*`CU_FUNC_CACHE_PREFER_NONE`* |*`hipFuncCachePreferNone`* | no preference for shared memory or L1 (default) |
| 0x01 |*`CU_FUNC_CACHE_PREFER_SHARED`* |*`hipFuncCachePreferShared`* | prefer larger shared memory and smaller L1 cache |
| 0x02 |*`CU_FUNC_CACHE_PREFER_L1`* |*`hipFuncCachePreferL1`* | prefer larger L1 cache and smaller shared memory |
| 0x03 |*`CU_FUNC_CACHE_PREFER_EQUAL`* |*`hipFuncCachePreferEqual`* | prefer equal sized L1 cache and shared memory |
| enum |***`CUfunction_attribute`*** | | Function properties |
| 0 |*`CU_FUNC_ATTRIBUTE_MAX_THREADS_PER_BLOCK`* | | The maximum number of threads per block, beyond which a launch of the function would fail. This number depends on both the function and the device on which the function is currently loaded. |
| 1 |*`CU_FUNC_ATTRIBUTE_SHARED_SIZE_BYTES`* | | The size in bytes of statically-allocated shared memory required by this function. This does not include dynamically-allocated shared memory requested by the user at runtime. |
| 2 |*`CU_FUNC_ATTRIBUTE_CONST_SIZE_BYTES`* | | The size in bytes of user-allocated constant memory required by this function. |
| 3 |*`CU_FUNC_ATTRIBUTE_LOCAL_SIZE_BYTES`* | | The size in bytes of local memory used by each thread of this function. |
| 4 |*`CU_FUNC_ATTRIBUTE_NUM_REGS`* | | The number of registers used by each thread of this function. |
| 5 |*`CU_FUNC_ATTRIBUTE_PTX_VERSION`* | | The PTX virtual architecture version for which the function was compiled. This value is the major PTX version * 10 + the minor PTX version, so a PTX version 1.3 function would return the value 13. Note that this may return the undefined value of 0 for cubins compiled prior to CUDA 3.0. |
| 6 |*`CU_FUNC_ATTRIBUTE_BINARY_VERSION`* | | The binary architecture version for which the function was compiled. This value is the major binary version * 10 + the minor binary version, so a binary version 1.3 function would return the value 13. Note that this will return a value of 10 for legacy cubins that do not have a properly-encoded binary architecture version. |
| 7 |*`CU_FUNC_ATTRIBUTE_CACHE_MODE_CA`* | | The attribute to indicate whether the function has been compiled with user specified option "-Xptxas --dlcm=ca" set. |
| 8 |*`CU_FUNC_ATTRIBUTE_MAX`* | | |
| enum |***`CUgraphicsMapResourceFlags`*** | | Flags for mapping and unmapping interop resources |
| 0x00 |*`CU_GRAPHICS_MAP_RESOURCE_FLAGS_NONE`* | | |
| 0x01 |*`CU_GRAPHICS_MAP_RESOURCE_FLAGS_READ_ONLY`* | | |
| 0x02 |*`CU_GRAPHICS_MAP_RESOURCE_FLAGS_WRITE_DISCARD`* | | |
| enum |***`CUgraphicsRegisterFlags`*** | | Flags to register a graphics resource |
| 0x00 |*`CU_GRAPHICS_REGISTER_FLAGS_NONE`* | | |
| 0x01 |*`CU_GRAPHICS_REGISTER_FLAGS_READ_ONLY`* | | |
| 0x02 |*`CU_GRAPHICS_REGISTER_FLAGS_WRITE_DISCARD`* | | |
| 0x04 |*`CU_GRAPHICS_REGISTER_FLAGS_SURFACE_LDST`* | | |
| 0x08 |*`CU_GRAPHICS_REGISTER_FLAGS_TEXTURE_GATHER`* | | |
| enum |***`CUipcMem_flags`*** | | CUDA Ipc Mem Flags |
| 0x1 |*`CU_IPC_MEM_LAZY_ENABLE_PEER_ACCESS`* |*`hipIpcMemLazyEnablePeerAccess`* | Automatically enable peer access between remote devices as needed |
| enum |***`CUjit_cacheMode`*** | | Caching modes for dlcm |
| 0 |*`CU_JIT_CACHE_OPTION_NONE`* | | Compile with no -dlcm flag specified |
| |*`CU_JIT_CACHE_OPTION_CG`* | | Compile with L1 cache disabled |
| |*`CU_JIT_CACHE_OPTION_CA`* | | Compile with L1 cache enabled |
| enum |***`CUjit_fallback`*** | | Cubin matching fallback strategies |
| 0 |*`CU_PREFER_PTX`* | | Prefer to compile ptx if exact binary match not found |
| |*`CU_PREFER_BINARY`* | | Prefer to fall back to compatible binary code if exact match not found |
| enum |***`CUjit_option`*** | | Online compiler and linker options |
| 0 |*`CU_JIT_MAX_REGISTERS`* | | Max number of registers that a thread may use. Option type: unsigned int Applies to: compiler only. |
| |*`CU_JIT_THREADS_PER_BLOCK`* | | IN: Specifies minimum number of threads per block to target compilation for OUT: Returns the number of threads the compiler actually targeted. This restricts the resource utilization fo the compiler (e.g. max registers) such that a block with the given number of threads should be able to launch based on register limitations. Note, this option does not currently take into account any other resource limitations, such as shared memory utilization. Cannot be combined with CU_JIT_TARGET. Option type: unsigned int Applies to: compiler only. |
| |*`CU_JIT_WALL_TIME`* | | Overwrites the option value with the total wall clock time, in milliseconds, spent in the compiler and linker Option type: float Applies to: compiler and linker. |
| |*`CU_JIT_INFO_LOG_BUFFER`* | | Pointer to a buffer in which to print any log messages that are informational in nature (the buffer size is specified via option CU_JIT_INFO_LOG_BUFFER_SIZE_BYTES) Option type: char * Applies to: compiler and linker. |
| |*`CU_JIT_INFO_LOG_BUFFER_SIZE_BYTES`* | | IN: Log buffer size in bytes. Log messages will be capped at this size (including null terminator) OUT: Amount of log buffer filled with messages Option type: unsigned int Applies to: compiler and linker. |
| |*`CU_JIT_OPTIMIZATION_LEVEL`* | | Level of optimizations to apply to generated code (0 - 4), with 4 being the default and highest level of optimizations. Option type: unsigned int Applies to: compiler only. |
| |*`CU_JIT_TARGET_FROM_CUCONTEXT`* | | No option value required. Determines the target based on the current attached context (default) Option type: No option value needed Applies to: compiler and linker. |
| |*`CU_JIT_TARGET`* | | Target is chosen based on supplied CUjit_target. Cannot be combined with CU_JIT_THREADS_PER_BLOCK. Option type: unsigned int for enumerated type CUjit_target Applies to: compiler and linker. |
| |*`CU_JIT_FALLBACK_STRATEGY`* | | Specifies choice of fallback strategy if matching cubin is not found. Choice is based on supplied CUjit_fallback. This option cannot be used with cuLink* APIs as the linker requires exact matches. Option type: unsigned int for enumerated type CUjit_fallback Applies to: compiler only. |
| |*`CU_JIT_GENERATE_DEBUG_INFO`* | | Specifies whether to create debug information in output (-g) (0: false, default) Option type: int Applies to: compiler and linker. |
| |*`CU_JIT_LOG_VERBOSE`* | | Generate verbose log messages (0: false, default) Option type: int Applies to: compiler and linker. |
| |*`CU_JIT_GENERATE_LINE_INFO`* | | Generate line number information (-lineinfo) (0: false, default) Option type: int Applies to: compiler only. |
| |*`CU_JIT_CACHE_MODE`* | | Specifies whether to enable caching explicitly (-dlcm) Choice is based on supplied CUjit_cacheMode_enum. Option type: unsigned int for enumerated type CUjit_cacheMode_enum Applies to: compiler only. |
| |*`CU_JIT_NUM_OPTIONS`* | | |
| enum |***`CUjit_target`*** | | Online compilation targets |
| 10 |*`CU_TARGET_COMPUTE_10`* | | Compute device class 1.0. |
| 11 |*`CU_TARGET_COMPUTE_11`* | | Compute device class 1.1. |
| 12 |*`CU_TARGET_COMPUTE_12`* | | Compute device class 1.2. |
| 13 |*`CU_TARGET_COMPUTE_13`* | | Compute device class 1.3. |
| 20 |*`CU_TARGET_COMPUTE_20`* | | Compute device class 2.0. |
| 21 |*`CU_TARGET_COMPUTE_21`* | | Compute device class 2.1. |
| 30 |*`CU_TARGET_COMPUTE_30`* | | Compute device class 3.0. |
| 32 |*`CU_TARGET_COMPUTE_32`* | | Compute device class 3.2. |
| 35 |*`CU_TARGET_COMPUTE_35`* | | Compute device class 3.5. |
| 37 |*`CU_TARGET_COMPUTE_37`* | | Compute device class 3.7. |
| 50 |*`CU_TARGET_COMPUTE_50`* | | Compute device class 5.0. |
| 52 |*`CU_TARGET_COMPUTE_52`* | | Compute device class 5.2. |
| enum |***`CUjitInputType`*** | | Device code formats |
| 0 |*`CU_JIT_INPUT_CUBIN`* | | Compiled device-class-specific device code Applicable options: none. |
| |*`CU_JIT_INPUT_PTX`* | | PTX source code Applicable options: PTX compiler options. |
| |*`CU_JIT_INPUT_FATBINARY`* | | Bundle of multiple cubins and/or PTX of some device code Applicable options: PTX compiler options, CU_JIT_FALLBACK_STRATEGY. |
| |*`CU_JIT_INPUT_OBJECT`* | | Host object with embedded device code Applicable options: PTX compiler options, CU_JIT_FALLBACK_STRATEGY. |
| |*`CU_JIT_INPUT_LIBRARY`* | | Archive of host objects with embedded device code Applicable options: PTX compiler options, CU_JIT_FALLBACK_STRATEGY. |
| |*`CU_JIT_NUM_INPUT_TYPES`* | | |
| enum |***`CUlimit`*** |***`hipLimit_t`*** | Limits |
| 0x00 |*`CU_LIMIT_STACK_SIZE`* | | GPU thread stack size. |
| 0x01 |*`CU_LIMIT_PRINTF_FIFO_SIZE`* | | GPU printf FIFO size. |
| 0x02 |*`CU_LIMIT_MALLOC_HEAP_SIZE`* |*`hipLimitMallocHeapSize`* | GPU malloc heap size. |
| 0x03 |*`CU_LIMIT_DEV_RUNTIME_SYNC_DEPTH`* | | GPU device runtime launch synchronize depth. |
| 0x04 |*`CU_LIMIT_DEV_RUNTIME_PENDING_LAUNCH_COUNT`* | | GPU device runtime pending launch count. |
| |*`CU_LIMIT_MAX`* | | |
| enum |***`CUmemAttach_flags`*** | | CUDA Mem Attach Flags |
| 0x1 |*`CU_MEM_ATTACH_GLOBAL`* | | Memory can be accessed by any stream on any device. |
| 0x2 |*`CU_MEM_ATTACH_HOST`* | | Memory cannot be accessed by any stream on any device. |
| 0x4 |*`CU_MEM_ATTACH_SINGLE`* | | Memory can only be accessed by a single stream on the associated device. |
| enum |***`CUmemorytype`*** | | Memory types |
| 0x01 |*`CU_MEMORYTYPE_HOST`* | | Host memory |
| 0x02 |*`CU_MEMORYTYPE_DEVICE`* | | Device memory |
| 0x03 |*`CU_MEMORYTYPE_ARRAY`* | | Array memory |
| 0x04 |*`CU_MEMORYTYPE_UNIFIED`* | | Unified device or host memory |
| enum |***`CUoccupancy_flags`*** | | Occupancy calculator flag |
| 0x00 |*`CU_OCCUPANCY_DEFAULT`* | | Default behavior |
| 0x01 |*`CU_OCCUPANCY_DISABLE_CACHING_OVERRIDE`* | | Assume global caching is enabled and cannot be automatically turned off |
| enum |***`CUpointer_attribute`*** | | Pointer information |
| 1 |*`CU_POINTER_ATTRIBUTE_CONTEXT`* | | The CUcontext on which a pointer was allocated or registered |
| 2 |*`CU_POINTER_ATTRIBUTE_MEMORY_TYPE`* | | The CUmemorytype describing the physical location of a pointer |
| 3 |*`CU_POINTER_ATTRIBUTE_DEVICE_POINTER`* | | The address at which a pointer's memory may be accessed on the device |
| 4 |*`CU_POINTER_ATTRIBUTE_HOST_POINTER`* | | The address at which a pointer's memory may be accessed on the host |
| 5 |*`CU_POINTER_ATTRIBUTE_P2P_TOKENS`* | | A pair of tokens for use with the nv-p2p.h Linux kernel interface |
| 6 |*`CU_POINTER_ATTRIBUTE_SYNC_MEMOPS`* | | Synchronize every synchronous memory operation initiated on this region |
| 7 |*`CU_POINTER_ATTRIBUTE_BUFFER_ID`* | | A process-wide unique ID for an allocated memory region |
| 8 |*`CU_POINTER_ATTRIBUTE_IS_MANAGED`* | | Indicates if the pointer points to managed memory |
| enum |***`CUmemorytype`*** | | Resource types |
| 0x00 |*`CU_RESOURCE_TYPE_ARRAY`* | | Array resoure |
| 0x01 |*`CU_RESOURCE_TYPE_MIPMAPPED_ARRAY`* | | Mipmapped array resource |
| 0x02 |*`CU_RESOURCE_TYPE_LINEAR`* | | Linear resource |
| 0x03 |*`CU_RESOURCE_TYPE_PITCH2D`* | | Pitch 2D resource |
| enum |***`CUresourceViewFormat`*** | | Resource view format |
| 0x00 |*`CU_RES_VIEW_FORMAT_NONE`* | | No resource view format (use underlying resource format) |
| 0x01 |*`CU_RES_VIEW_FORMAT_UINT_1X8`* | | 1 channel unsigned 8-bit integers |
| 0x02 |*`CU_RES_VIEW_FORMAT_UINT_2X8`* | | 2 channel unsigned 8-bit integers |
| 0x03 |*`CU_RES_VIEW_FORMAT_UINT_4X8`* | | 4 channel unsigned 8-bit integers |
| 0x04 |*`CU_RES_VIEW_FORMAT_SINT_1X8`* | | 1 channel signed 8-bit integers |
| 0x05 |*`CU_RES_VIEW_FORMAT_SINT_2X8`* | | 2 channel signed 8-bit integers |
| 0x06 |*`CU_RES_VIEW_FORMAT_SINT_4X8`* | | 4 channel signed 8-bit integers |
| 0x07 |*`CU_RES_VIEW_FORMAT_UINT_1X16`* | | 1 channel unsigned 16-bit integers |
| 0x08 |*`CU_RES_VIEW_FORMAT_UINT_2X16`* | | 2 channel unsigned 16-bit integers |
| 0x09 |*`CU_RES_VIEW_FORMAT_UINT_4X16`* | | 4 channel unsigned 16-bit integers |
| 0x0a |*`CU_RES_VIEW_FORMAT_SINT_1X16`* | | 1 channel signed 16-bit integers |
| 0x0b |*`CU_RES_VIEW_FORMAT_SINT_2X16`* | | 2 channel signed 16-bit integers |
| 0x0c |*`CU_RES_VIEW_FORMAT_SINT_4X16`* | | 4 channel signed 16-bit integers |
| 0x0d |*`CU_RES_VIEW_FORMAT_UINT_1X32`* | | 1 channel unsigned 32-bit integers |
| 0x0e |*`CU_RES_VIEW_FORMAT_UINT_2X32`* | | 2 channel unsigned 32-bit integers |
| 0x0f |*`CU_RES_VIEW_FORMAT_UINT_4X32`* | | 4 channel unsigned 32-bit integers |
| 0x10 |*`CU_RES_VIEW_FORMAT_SINT_1X32`* | | 1 channel signed 32-bit integers |
| 0x11 |*`CU_RES_VIEW_FORMAT_SINT_2X32`* | | 2 channel signed 32-bit integers |
| 0x12 |*`CU_RES_VIEW_FORMAT_SINT_4X32`* | | 4 channel signed 32-bit integers |
| 0x13 |*`CU_RES_VIEW_FORMAT_FLOAT_1X16`* | | 1 channel 16-bit floating point |
| 0x14 |*`CU_RES_VIEW_FORMAT_FLOAT_2X16`* | | 2 channel 16-bit floating point |
| 0x15 |*`CU_RES_VIEW_FORMAT_FLOAT_4X16`* | | 4 channel 16-bit floating point |
| 0x16 |*`CU_RES_VIEW_FORMAT_FLOAT_1X32`* | | 1 channel 32-bit floating point |
| 0x17 |*`CU_RES_VIEW_FORMAT_FLOAT_2X32`* | | 2 channel 32-bit floating point |
| 0x18 |*`CU_RES_VIEW_FORMAT_FLOAT_4X32`* | | 4 channel 32-bit floating point |
| 0x19 |*`CU_RES_VIEW_FORMAT_UNSIGNED_BC1`* | | Block compressed 1 |
| 0x1a |*`CU_RES_VIEW_FORMAT_UNSIGNED_BC3`* | | Block compressed 2 |
| 0x1b |*`CU_RES_VIEW_FORMAT_UNSIGNED_BC3`* | | Block compressed 3 |
| 0x1c |*`CU_RES_VIEW_FORMAT_UNSIGNED_BC4`* | | Block compressed 4 unsigned |
| 0x1d |*`CU_RES_VIEW_FORMAT_SIGNED_BC4`* | | Block compressed 4 signed |
| 0x1e |*`CU_RES_VIEW_FORMAT_UNSIGNED_BC5`* | | Block compressed 5 unsigned |
| 0x1f |*`CU_RES_VIEW_FORMAT_SIGNED_BC5`* | | Block compressed 5 signed |
| 0x20 |*`CU_RES_VIEW_FORMAT_UNSIGNED_BC6H`* | | Block compressed 6 unsigned half-float |
| 0x21 |*`CU_RES_VIEW_FORMAT_SIGNED_BC6H`* | | Block compressed 6 signed half-float |
| 0x22 |*`CU_RES_VIEW_FORMAT_UNSIGNED_BC7`* | | Block compressed 7 |
| enum |***`CUresult`*** |***`hipError_t`*** | Error codes |
| 0 |*`CUDA_SUCCESS`* |*`hipSuccess`* | The API call returned with no errors. In the case of query calls, this can also mean that the operation being queried is complete (see cuEventQuery() and cuStreamQuery()). |
| 1 |*`CUDA_ERROR_INVALID_VALUE`* |*`hipErrorInvalidValue`* | This indicates that one or more of the parameters passed to the API call is not within an acceptable range of values. |
| 2 |*`CUDA_ERROR_OUT_OF_MEMORY`* |*`hipErrorMemoryAllocation`* | The API call failed because it was unable to allocate enough memory to perform the requested operation. |
| 3 |*`CUDA_ERROR_NOT_INITIALIZED`* |*`hipErrorNotInitialized`* | This indicates that the CUDA driver has not been initialized with cuInit() or that initialization has failed. |
| 4 |*`CUDA_ERROR_DEINITIALIZED`* |*`hipErrorDeinitialized`* | This indicates that the CUDA driver is in the process of shutting down. |
| 5 |*`CUDA_ERROR_PROFILER_DISABLED`* |*`hipErrorProfilerDisabled`* | This indicates profiler is not initialized for this run. This can happen when the application is running with external profiling tools like visual profiler. |
| 6 |*`CUDA_ERROR_PROFILER_NOT_INITIALIZED`* |*`hipErrorProfilerNotInitialized`* | Deprecated This error return is deprecated as of CUDA 5.0. It is no longer an error to attempt to enable/disable the profiling via cuProfilerStart or cuProfilerStop without initialization. |
| 7 |*`CUDA_ERROR_PROFILER_ALREADY_STARTED`* |*`hipErrorProfilerAlreadyStarted`* | Deprecated This error return is deprecated as of CUDA 5.0. It is no longer an error to call cuProfilerStart() when profiling is already enabled. |
| 8 |*`CUDA_ERROR_PROFILER_ALREADY_STOPPED`* |*`hipErrorProfilerAlreadyStopped`* | Deprecated This error return is deprecated as of CUDA 5.0. It is no longer an error to call cuProfilerStop() when profiling is already disabled. |
| 100 |*`CUDA_ERROR_NO_DEVICE`* |*`hipErrorNoDevice`* | This indicates that no CUDA-capable devices were detected by the installed CUDA driver. |
| 101 |*`CUDA_ERROR_INVALID_DEVICE`* |*`hipErrorInvalidDevice`* | This indicates that the device ordinal supplied by the user does not correspond to a valid CUDA device. |
| 200 |*`CUDA_ERROR_INVALID_IMAGE`* |*`hipErrorInvalidImage`* | This indicates that the device kernel image is invalid. This can also indicate an invalid CUDA module. |
| 201 |*`CUDA_ERROR_INVALID_CONTEXT`* |*`hipErrorInvalidContext`* | This most frequently indicates that there is no context bound to the current thread. This can also be returned if the context passed to an API call is not a valid handle (such as a context that has had cuCtxDestroy() invoked on it). This can also be returned if a user mixes different API versions (i.e. 3010 context with 3020 API calls). See cuCtxGetApiVersion() for more details. |
| 202 |*`CUDA_ERROR_CONTEXT_ALREADY_CURRENT`* |*`hipErrorContextAlreadyCurrent`* | This indicated that the context being supplied as a parameter to the API call was already the active context. Deprecated This error return is deprecated as of CUDA 3.2. It is no longer an error to attempt to push the active context via cuCtxPushCurrent(). |
| 205 |*`CUDA_ERROR_MAP_FAILED`* |*`hipErrorMapFailed`* | This indicates that a map or register operation has failed. |
| 206 |*`CUDA_ERROR_UNMAP_FAILED`* |*`hipErrorUnmapFailed`* | This indicates that an unmap or unregister operation has failed. |
| 207 |*`CUDA_ERROR_ARRAY_IS_MAPPED`* |*`hipErrorArrayIsMapped`* | This indicates that the specified array is currently mapped and thus cannot be destroyed. |
| 208 |*`CUDA_ERROR_ALREADY_MAPPED`* |*`hipErrorAlreadyMapped`* | This indicates that the resource is already mapped. |
| 209 |*`CUDA_ERROR_NO_BINARY_FOR_GPU`* |*`hipErrorNoBinaryForGpu* | This indicates that there is no kernel image available that is suitable for the device. This can occur when a user specifies code generation options for a particular CUDA source file that do not include the corresponding device configuration. |
| 210 |*`CUDA_ERROR_ALREADY_ACQUIRED`* |*`hipErrorAlreadyAcquired* | This indicates that a resource has already been acquired. |
| 211 |*`CUDA_ERROR_NOT_MAPPED`* |*`hipErrorNotMapped`* | This indicates that a resource is not mapped. |
| 212 |*`CUDA_ERROR_NOT_MAPPED_AS_ARRAY`* |*`hipErrorNotMappedAsArray`* | This indicates that a mapped resource is not available for access as an array. |
| 213 |*`CUDA_ERROR_NOT_MAPPED_AS_POINTER`* |*`hipErrorNotMappedAsPointer`* | This indicates that a mapped resource is not available for access as a pointer. |
| 214 |*`CUDA_ERROR_ECC_UNCORRECTABLE`* |*`hipErrorECCNotCorrectable`* | This indicates that an uncorrectable ECC error was detected during execution. |
| 215 |*`CUDA_ERROR_UNSUPPORTED_LIMIT`* |*`hipErrorUnsupportedLimit`* | This indicates that the CUlimit passed to the API call is not supported by the active device. |
| 216 |*`CUDA_ERROR_CONTEXT_ALREADY_IN_USE`* |*`hipErrorContextAlreadyInUse`* | This indicates that the CUcontext passed to the API call can only be bound to a single CPU thread at a time but is already bound to a CPU thread. |
| 217 |*`CUDA_ERROR_PEER_ACCESS_UNSUPPORTED`* |*`hipErrorPeerAccessUnsupported`* | This indicates that peer access is not supported across the given devices. |
| 218 |*`CUDA_ERROR_INVALID_PTX`* |*`hipErrorInvalidKernelFile`* | This indicates that a PTX JIT compilation failed. |
| 219 |*`CUDA_ERROR_INVALID_GRAPHICS_CONTEXT`* |*`hipErrorInvalidGraphicsContext`* | This indicates an error with OpenGL or DirectX context. |
| 300 |*`CUDA_ERROR_INVALID_SOURCE`* |*`hipErrorInvalidSource`* | This indicates that the device kernel source is invalid. |
| 301 |*`CUDA_ERROR_FILE_NOT_FOUND`* |*`hipErrorFileNotFound`* | This indicates that the file specified was not found. |
| 302 |*`CUDA_ERROR_SHARED_OBJECT_SYMBOL_NOT_FOUND`* |*`hipErrorSharedObjectSymbolNotFound`* | This indicates that a link to a shared object failed to resolve. |
| 303 |*`CUDA_ERROR_SHARED_OBJECT_INIT_FAILED`* |*`hipErrorSharedObjectInitFailed`* | This indicates that initialization of a shared object failed. |
| 304 |*`CUDA_ERROR_OPERATING_SYSTEM`* |*`hipErrorOperatingSystem`* | This indicates that an OS call failed. |
| 400 |*`CUDA_ERROR_INVALID_HANDLE`* |*`hipErrorInvalidResourceHandle`* | This indicates that a resource handle passed to the API call was not valid. Resource handles are opaque types like CUstream and CUevent. |
| 500 |*`CUDA_ERROR_NOT_FOUND`* |*`hipErrorNotFound`* | This indicates that a named symbol was not found. Examples of symbols are global/constant variable names, texture names, and surface names. |
| 600 |*`CUDA_ERROR_NOT_READY`* |*`hipErrorNotReady`* | This indicates that asynchronous operations issued previously have not completed yet. This result is not actually an error, but must be indicated differently than CUDA_SUCCESS (which indicates completion). Calls that may return this value include cuEventQuery() and cuStreamQuery(). |
| 700 |*`CUDA_ERROR_ILLEGAL_ADDRESS`* |*`hipErrorIllegalAddress`* | While executing a kernel, the device encountered a load or store instruction on an invalid memory address. The context cannot be used, so it must be destroyed (and a new one should be created). All existing device memory allocations from this context are invalid and must be reconstructed if the program is to continue using CUDA. |
## **2. Error Handling**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
## **3. Initialization**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
## **4. Version Management**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
## **5. Device Management**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
## **6. Device Management [DEPRECATED]**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
## **7. Primary Context Management**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
## **8. Context Management**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
## **9. Context Management [DEPRECATED]**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
## **10. Module Management**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
## **11. Memory Management**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
## **12. Unified Addressing**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
## **13. Stream Management**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
## **14. Event Management**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
## **15. Execution Control**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
## **16. Execution Control [DEPRECATED]**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
## **17. Occupancy**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
## **18. Texture Reference Management**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
## **19. Texture Reference Management [DEPRECATED]**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
## **20. Surface Reference Management**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
## **21. Texture Object Management**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
## **22. Surface Object Management**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
## **23. Peer Context Memory Access**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
## **24. Graphics Interoperability**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
## **25. Profiler Control**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
## **26. OpenGL Interoperability**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
## **27. Direct3D 9 Interoperability**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
## **28. Direct3D 10 Interoperability**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
## **29. Direct3D 11 Interoperability**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
## **30. VDPAU Interoperability**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
projects/hip/docs/markdown/CUDA_Runtime_API_functions_supported_by_HIP.md
+421 -54
Datei anzeigen
@@ -1,34 +1,36 @@
**1. Device Management**
# CUDA Runtime API functions supported by HIP
## **1. Device Management**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
| `cudaChooseDevice` | `hipChooseDevice` | Select compute-device which best matches criteria. |
| `cudaChooseDevice` | `hipChooseDevice` | Select compute-device which best matches criteria. |
| `cudaDeviceGetAttribute` | `hipDeviceGetAttribute` | Returns information about the device. |
| `cudaDeviceGetByPCIBusId` | | Returns a handle to a compute device. |
| `cudaDeviceGetByPCIBusId` | `hipDeviceGetByPCIBusId` | Returns a handle to a compute device. |
| `cudaDeviceGetCacheConfig` | `hipDeviceGetCacheConfig` | Returns the preferred cache configuration for the current device. |
| `cudaDeviceGetLimit` | `hipDeviceGetLimit` | Returns resource limits. |
| `cudaDeviceGetPCIBusId` | | Returns a PCI Bus Id string for the device. |
| `cudaDeviceGetLimit` | `hipDeviceGetLimit` | Returns resource limits. |
| `cudaDeviceGetPCIBusId` | `hipDeviceGetPCIBusId` | Returns a PCI Bus Id string for the device. |
| `cudaDeviceGetSharedMemConfig` | `hipDeviceGetSharedMemConfig` | Returns the shared memory configuration for the current device. |
| `cudaDeviceGetStreamPriorityRange` | | Returns numerical values that correspond to the least and greatest stream priorities. |
| `cudaDeviceReset` | `hipDeviceReset` | Destroy all allocations and reset all state on the current device in the current process. |
| `cudaDeviceSetCacheConfig` | `hipDeviceSetCacheConfig` | Sets the preferred cache configuration for the current device. |
| `cudaDeviceSetLimit` | `hipDeviceSetLimit` | Set resource limits. |
| `cudaDeviceSetLimit` | `hipDeviceSetLimit` | Set resource limits. |
| `cudaDeviceSetSharedMemConfig` | `hipDeviceSetSharedMemConfig` | Sets the shared memory configuration for the current device. |
| `cudaDeviceSynchronize` | `hipDeviceSynchronize` | Wait for compute device to finish. |
| `cudaGetDevice` | `hipGetDevice` | Returns which device is currently being used. |
| `cudaGetDeviceCount` | `hipGetDeviceCount` | Returns the number of compute-capable devices. |
| `cudaGetDeviceFlags` | | Gets the flags for the current device. |
| `cudaGetDeviceProperties` | `hipGetDeviceProperties` | Returns information about the compute-device. |
| `cudaIpcCloseMemHandle` | | Close memory mapped with cudaIpcOpenMemHandle. |
| `cudaIpcGetEventHandle` | | Gets an interprocess handle for a previously allocated event. |
| `cudaIpcGetMemHandle` | | Gets an interprocess memory handle for an existing device memory allocation. |
| `cudaIpcOpenEventHandle` | | Opens an interprocess event handle for use in the current process. |
| `cudaIpcOpenMemHandle` | | Opens an interprocess memory handle exported from another process and returns a device pointer usable in the local process. |
| `cudaIpcCloseMemHandle` | `hipIpcCloseMemHandle` | Close memory mapped with cudaIpcOpenMemHandle. |
| `cudaIpcGetEventHandle` | `hipIpcGetEventHandle` | Gets an interprocess handle for a previously allocated event. |
| `cudaIpcGetMemHandle` | `hipIpcGetMemHandle` | Gets an interprocess memory handle for an existing device memory allocation. |
| `cudaIpcOpenEventHandle` | `hipIpcOpenEventHandle` | Opens an interprocess event handle for use in the current process. |
| `cudaIpcOpenMemHandle` | `hipIpcOpenMemHandle` | Opens an interprocess memory handle exported from another process and returns a device pointer usable in the local process. |
| `cudaSetDevice` | `hipSetDevice` | Set device to be used for GPU executions. |
| `cudaSetDeviceFlags` | `hipSetDeviceFlags` | Sets flags to be used for device executions. |
| `cudaSetValidDevices` | | Set a list of devices that can be used for CUDA. |`
| `cudaSetValidDevices` | | Set a list of devices that can be used for CUDA. |
**2. Error Handling**
## **2. Error Handling**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
@@ -37,12 +39,12 @@
| `cudaGetLastError` | `hipGetLastError` | Returns the last error from a runtime call. |
| `cudaPeekAtLastError` | `hipPeekAtLastError` | Returns the last error from a runtime call. |
**3. Stream Management**
## **3. Stream Management**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
| `cudaStreamAddCallback` | | Add a callback to a compute stream. |
| `cudaStreamAttachMemAsync` | | Attach managed memory to a stream asynchronously. |
| `cudaStreamAddCallback` | `hipStreamAddCallback` | Add a callback to a compute stream. |
| `cudaStreamAttachMemAsync` | | Attach managed memory to a stream asynchronously. |
| `cudaStreamCreate` | `hipStreamCreate` | Create an asynchronous stream. |
| `cudaStreamCreateWithFlags` | `hipStreamCreateWithFlags` | Create an asynchronous stream. |
| `cudaStreamCreateWithPriority` | | Create an asynchronous stream with the specified priority. |
@@ -53,7 +55,7 @@
| `cudaStreamSynchronize` | `hipStreamSynchronize` | Waits for stream tasks to complete. |
| `cudaStreamWaitEvent` | `hipStreamWaitEvent` | Make a compute stream wait on an event. |
**4. Event Management**
## **4. Event Management**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
@@ -65,7 +67,7 @@
| `cudaEventRecord` | `hipEventRecord` | Records an event. |
| `cudaEventSynchronize` | `hipEventSynchronize` | Waits for an event to complete. |
**5. Execution Control**
## **5. Execution Control**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
@@ -78,46 +80,54 @@
| `cudaSetDoubleForDevice` | | Converts a double argument to be executed on a device. |
| `cudaSetDoubleForHost` | | Converts a double argument after execution on a device. |
**6. Occupancy**
## **6. Occupancy**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
| `cudaOccupancyMaxActiveBlocksPerMultiprocessor` | | Returns occupancy for a device function. |
| `cudaOccupancyMaxActiveBlocksPerMultiprocessor` | `hipOccupancyMaxActiveBlocksPerMultiprocessor`| Returns occupancy for a device function. |
| `cudaOccupancyMaxActiveBlocksPerMultiprocessorWithFlags` | | Returns occupancy for a device function with the specified flags. |
**7. Memory Management**
## **7. Execution Control [deprecated since 7.0]**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
| `cudaConfigureCall` | | Configure a device-launch. |
| `cudaLaunch` | | Launches a device function. |
| `cudaSetupArgument` | | Configure a device launch. |
## **8. Memory Management**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
| `cudaArrayGetInfo` | | Gets info about the specified cudaArray. |
| `cudaFree` | `hipFree` | Frees memory on the device. |
| `cudaFreeArray` | | Frees an array on the device. |
| `cudaFreeArray` | `hipFreeArray` | Frees an array on the device. |
| `cudaFreeHost` | `hipHostFree` | Frees page-locked memory. |
| `cudaFreeMipmappedArray` | | Frees a mipmapped array on the device. |
| `cudaGetMipmappedArrayLevel` | | Gets a mipmap level of a CUDA mipmapped array. |
| `cudaGetSymbolAddress` | | Finds the address associated with a CUDA symbol. |
| `cudaGetSymbolSize` | | Finds the size of the object associated with a CUDA symbol. |
| `cudaHostAlloc` | `hipHostMalloc` | Allocates page-locked memory on the host. |
| `cudaHostGetDevicePointer` | `hipHostGetDevicePointer` | Passes back device pointer of mapped host memory allocated by cudaHostAlloc or registered by cudaHostRegister. |
| `cudaHostGetFlags` | `hipHostGetFlags` | Passes back flags used to allocate pinned host memory allocated by cudaHostAlloc. |
| `cudaHostAlloc` | `hipHostMalloc` | Allocates page-locked memory on the host. |
| `cudaHostGetDevicePointer` | `hipHostGetDevicePointer` | Passes back device pointer of mapped host memory allocated by cudaHostAlloc or registered by cudaHostRegister. |
| `cudaHostGetFlags` | `hipHostGetFlags` | Passes back flags used to allocate pinned host memory allocated by cudaHostAlloc. |
| `cudaHostRegister` | `hipHostRegister` | Registers an existing host memory range for use by CUDA. |
| `cudaHostUnregister` | `hipHostUnregister` | Unregisters a memory range that was registered with cudaHostRegister. |
| `cudaMalloc` | `hipMalloc` | Allocate memory on the device. |
| `cudaMalloc3D` | | Allocates logical 1D, 2D, or 3D memory objects on the device. |
| `cudaMalloc3DArray` | | Allocate an array on the device. |
| `cudaMallocArray` | | Allocate an array on the device. |
| `cudaMallocHost` | `hipHostMalloc` | Allocates page-locked memory on the host. |
| `cudaMallocArray` | `hipMallocArray` | Allocate an array on the device. |
| `cudaMallocHost` | `hipHostMalloc` | Allocates page-locked memory on the host. |
| `cudaMallocManaged` | | Allocates memory that will be automatically managed by the Unified Memory system. |
| `cudaMallocMipmappedArray` | | Allocate a mipmapped array on the device. |
| `cudaMallocPitch` | | Allocates pitched memory on the device. |
| `cudaMemGetInfo` | | Gets free and total device memory. |
| `cudaMemGetInfo` | `hipMemGetInfo` | Gets free and total device memory. |
| `cudaMemcpy` | `hipMemcpy` | Copies data between host and device. |
| `cudaMemcpy2D` | | Copies data between host and device. |
| `cudaMemcpy2D` | `hipMemcpy2D` | Copies data between host and device. |
| `cudaMemcpy2DArrayToArray` | | Copies data between host and device. |
| `cudaMemcpy2DAsync` | | Copies data between host and device. |
| `cudaMemcpy2DFromArray` | | Copies data between host and device. |
| `cudaMemcpy2DFromArrayAsync` | | Copies data between host and device. |
| `cudaMemcpy2DToArray` | | Copies data between host and device. |
| `cudaMemcpy2DToArray` | `hipMemcpy2DToArray` | Copies data between host and device. |
| `cudaMemcpy2DToArrayAsync` | | Copies data between host and device. |
| `cudaMemcpy3D` | | Copies data between 3D objects. |
| `cudaMemcpy3DAsync` | | Copies data between 3D objects. |
@@ -131,27 +141,27 @@
| `cudaMemcpyFromSymbolAsync` | | Copies data from the given symbol on the device. |
| `cudaMemcpyPeer` | `hipMemcpyPeer` | Copies memory between two devices. |
| `cudaMemcpyPeerAsync` | `hipMemcpyPeerAsync` | Copies memory between two devices asynchronously. |
| `cudaMemcpyToArray` | | Copies data between host and device. |
| `cudaMemcpyToArray` | `hipMemcpyToArray` | Copies data between host and device. |
| `cudaMemcpyToArrayAsync` | | Copies data between host and device. |
| `cudaMemcpyToSymbol` | `hipMemcpyToSymbol` | Copies data to the given symbol on the device. |
| `cudaMemcpyToSymbolAsync` | | Copies data to the given symbol on the device. |
| `cudaMemcpyToSymbolAsync` | `hipMemcpyToSymbolAsync` | Copies data to the given symbol on the device. |
| `cudaMemset` | `hipMemset` | Initializes or sets device memory to a value. |
| `cudaMemset2D` | | Initializes or sets device memory to a value. |
| `cudaMemset2DAsync` | | Initializes or sets device memory to a value. |
| `cudaMemset3D` | | Initializes or sets device memory to a value. |
| `cudaMemset3DAsync` | | Initializes or sets device memory to a value. |
| `cudaMemsetAsync` | `hipMemsetAsync` | Initializes or sets device memory to a value. |
| `make\_cudaExtent` | | Returns a cudaExtent based on input parameters. |
| `make\_cudaPitchedPtr` | | Returns a cudaPitchedPtr based on input parameters. |
| `make\_cudaPos` | | Returns a cudaPos based on input parameters. |
| `make_cudaExtent` | | Returns a cudaExtent based on input parameters. |
| `make_cudaPitchedPtr` | | Returns a cudaPitchedPtr based on input parameters. |
| `make_cudaPos` | | Returns a cudaPos based on input parameters. |
**8. Unified Addressing**
## **9. Unified Addressing**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
| `cudaPointerGetAttributes` | `hipPointerGetAttributes` | Returns attributes about a specified pointer. |
**9. Peer Device Memory Access**
## **10. Peer Device Memory Access**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
@@ -159,7 +169,7 @@
| `cudaDeviceDisablePeerAccess` | `hipDeviceDisablePeerAccess` | Disables direct access to memory allocations on a peer device. |
| `cudaDeviceEnablePeerAccess` | `hipDeviceEnablePeerAccess` | Enables direct access to memory allocations on a peer device. |
**10. OpenGL Interoperability**
## **11. OpenGL Interoperability**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
@@ -168,7 +178,7 @@
| `cudaGraphicsGLRegisterImage` | | Register an OpenGL texture or renderbuffer object. |
| `cudaWGLGetDevice` | | Gets the CUDA device associated with hGpu. |
**11. Graphics Interoperability**
## **12. Graphics Interoperability**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
@@ -180,7 +190,7 @@
| `cudaGraphicsUnmapResources` | | Unmap graphics resources. |
| `cudaGraphicsUnregisterResource` | | Unregisters a graphics resource for access by CUDA. |
**12. Texture Reference Management**
## **13. Texture Reference Management**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
@@ -194,14 +204,14 @@
| `cudaGetTextureReference` | | Get the texture reference associated with a symbol. |
| `cudaUnbindTexture` | | Unbinds a texture. |
**13. Surface Reference Management**
## **14. Surface Reference Management**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
| `cudaBindSurfaceToArray` | | Binds an array to a surface. |
| `cudaGetSurfaceReference` | | Get the surface reference associated with a symbol. |
**14. Texture Object Management**
## **15. Texture Object Management**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
@@ -211,7 +221,7 @@
| `cudaGetTextureObjectResourceViewDesc` | | Returns a texture object's resource view descriptor. |
| `cudaGetTextureObjectTextureDesc` | | Returns a texture object's texture descriptor. |
**15. Surface Object Management**
## **16. Surface Object Management**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
@@ -219,25 +229,24 @@
| `cudaDestroySurfaceObject` | | Destroys a surface object. |
| `cudaGetSurfaceObjectResourceDesc` | | Returns a surface object's resource descriptor Returns the resource descriptor for the surface object specified by surfObject. |
**16. Version Management**
## **17. Version Management**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
| `cudaDriverGetVersion` | `hipDriverGetVersion` | Returns the CUDA driver version. |
| `cudaRuntimeGetVersion` | | Returns the CUDA Runtime version. |
| `cudaRuntimeGetVersion` | `hipRuntimeGetVersion` | Returns the CUDA Runtime version. |
**17. C++ API Routines (7.0 contains, 7.5 doesn’t)**
> Will not support for HIP (probably)
## **18. C++ API Routines**
*(7.0 contains, 7.5 doesn’t)*
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
| `cudaBindSurfaceToArray` | | Binds an array to a surface. |
| `cudaBindTexture` | | Binds a memory area to a texture. |
| `cudaBindTexture` | `hipBindTexture` | Binds a memory area to a texture. |
| `cudaBindTexture2D` | | Binds a 2D memory area to a texture. |
| `cudaBindTextureToArray` | | Binds an array to a texture. |
| `cudaBindTextureToMipmappedArray` | | Binds a mipmapped array to a texture. |
| `cudaCreateChannelDesc` | | Returns a channel descriptor using the specified format. |
| `cudaEventCreate` | | Creates an event object with the specified flags. |
| `cudaCreateChannelDesc` | `hipCreateChannelDesc` | Returns a channel descriptor using the specified format. |
| `cudaFuncGetAttributes` | | Find out attributes for a given function. |
| `cudaFuncSetCacheConfig` | | Sets the preferred cache configuration for a device function. |
| `cudaGetSymbolAddress` | | Finds the address associated with a CUDA symbol |
@@ -251,20 +260,378 @@
| `cudaMemcpyFromSymbolAsync` | | Copies data from the given symbol on the device. |
| `cudaMemcpyToSymbol` | | Copies data to the given symbol on the device. |
| `cudaMemcpyToSymbolAsync` | | Async copies data to the given symbol on the device. |
| `cudaOccupancyMaxActiveBlocksPerMultiprocessor` | | Returns occupancy for a device function. |
| `cudaOccupancyMaxActiveBlocksPerMultiprocessor` | `hipOccupancyMaxActiveBlocksPerMultiprocessor` | Returns occupancy for a device function. |
| `cudaOccupancyMaxActiveBlocksPerMultiprocessorWithFlags` | | Returns occupancy for a device function with the specified flags. |
| `cudaOccupancyMaxPotentialBlockSize` | | Returns grid and block size that achieves maximum potential occupancy for a device function. |
| `cudaOccupancyMaxPotentialBlockSize` | `hipOccupancyMaxPotentialBlockSize` | Returns grid and block size that achieves maximum potential occupancy for a device function. |
| `cudaOccupancyMaxPotentialBlockSizeVariableSMem` | | Returns grid and block size that achieves maximum potential occupancy for a device function. |
| `cudaOccupancyMaxPotentialBlockSizeVariableSMemWithFlags` | | Returns grid and block size that achieves maximum potential occupancy for a device function. |
| `cudaOccupancyMaxPotentialBlockSizeWithFlags` | | Returns grid and block size that achived maximum potential occupancy for a device function with the specified flags. |
| `cudaSetupArgument` | | Configure a device launch. |
| `cudaStreamAttachMemAsync` | | Attach memory to a stream asynchronously. |
| `cudaUnbindTexture` | | Unbinds a texture. |
| `cudaUnbindTexture` | `hipUnbindTexture` | Unbinds a texture. |
**18. Profiler Control**
## **19. Profiler Control**
| **CUDA** | **HIP** | **CUDA description** |
|-----------------------------------------------------------|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
| `cudaProfilerInitialize` | | Initialize the CUDA profiler. |
| `cudaProfilerStart` | `hipProfilerStart` | Enable profiling. |
| `cudaProfilerStop` | `hipProfilerStop` | Disable profiling. |
# Data types used by CUDA Runtime API and supported by HIP
## **20. Data types**
| **type** | **CUDA** | **HIP** | **CUDA description** |
|-------------:|-----------------------------------------------|------------------------------------------------------|--------------------------------------------------------------------------------------------------------------------------------|
| struct | `cudaChannelFormatDesc` | `hipChannelFormatDesc` | CUDA Channel format descriptor. |
| struct | `cudaDeviceProp` | `hipDeviceProp_t` | CUDA device properties. |
| struct | `cudaExtent` | | CUDA extent (width, height, depth). |
| struct | `cudaFuncAttributes` | | CUDA function attributes. |
| struct | `cudaIpcEventHandle_t` | `hipIpcEventHandle_t` | CUDA IPC event handle. |
| struct | `cudaIpcMemHandle_t` | `hipIpcMemHandle_t` | CUDA IPC memory handle. |
| struct | `cudaMemcpy3DParms` | | CUDA 3D memory copying parameters. |
| struct | `cudaMemcpy3DPeerParms` | | CUDA 3D cross-device memory copying parameters. |
| struct | `cudaPitchedPtr` | | CUDA Pitched memory pointer. |
| struct | `cudaPointerAttributes` | `hipPointerAttribute_t` | CUDA pointer attributes. |
| struct | `cudaPos` | | CUDA 3D position. |
| struct | `cudaResourceDesc` | | CUDA resource descriptor. |
| struct | `cudaResourceViewDesc` | | CUDA resource view descriptor. |
| struct | `cudaTextureDesc` | | CUDA texture descriptor. |
| struct | `surfaceReference` | | CUDA Surface reference. |
| struct | `textureReference` | `textureReference` | CUDA texture reference. |
| enum |***`cudaChannelFormatKind`*** |***`hipChannelFormatKind`*** | Channel format kind. |
| 0 |*`cudaChannelFormatKindSigned`* |*`hipChannelFormatKindSigned`* | Signed channel format. |
| 1 |*`cudaChannelFormatKindUnsigned`* |*`hipChannelFormatKindUnsigned`* | Unsigned channel format. |
| 2 |*`cudaChannelFormatKindFloat`* |*`hipChannelFormatKindFloat`* | Float channel format. |
| 3 |*`cudaChannelFormatKindNone`* |*`hipChannelFormatKindNone`* | No channel format. |
| enum |***`cudaComputeMode`*** | | CUDA device compute modes. |
| 0 |*`cudaComputeModeDefault`* | | Default compute mode (Multiple threads can use ::cudaSetDevice() with this device). |
| 1 |*`cudaComputeModeExclusive`* | | Compute-exclusive-thread mode (Only one thread in one process will be able to use ::cudaSetDevice() with this device). |
| 2 |*`cudaComputeModeProhibited`* | | Compute-prohibited mode (No threads can use ::cudaSetDevice() with this device). |
| 3 |*`cudaComputeModeExclusiveProcess`* | | Compute-exclusive-process mode (Many threads in one process will be able to use ::cudaSetDevice() with this device). |
| enum |***`cudaDeviceAttr`*** |***`hipDeviceAttribute_t`*** | CUDA device attributes. |
| 1 |*`cudaDevAttrMaxThreadsPerBlock`* |*`hipDeviceAttributeMaxThreadsPerBlock`* | Maximum number of threads per block. |
| 2 |*`cudaDevAttrMaxBlockDimX`* |*`hipDeviceAttributeMaxBlockDimX`* | Maximum block dimension X. |
| 3 |*`cudaDevAttrMaxBlockDimY`* |*`hipDeviceAttributeMaxBlockDimY`* | Maximum block dimension Y. |
| 4 |*`cudaDevAttrMaxBlockDimZ`* |*`hipDeviceAttributeMaxBlockDimZ`* | Maximum block dimension Y. |
| 5 |*`cudaDevAttrMaxGridDimX`* |*`hipDeviceAttributeMaxGridDimX`* | Maximum grid dimension X. |
| 6 |*`cudaDevAttrMaxGridDimY`* |*`hipDeviceAttributeMaxGridDimY`* | Maximum grid dimension Y. |
| 7 |*`cudaDevAttrMaxGridDimZ`* |*`hipDeviceAttributeMaxGridDimZ`* | Maximum grid dimension Y. |
| 8 |*`cudaDevAttrMaxSharedMemoryPerBlock`* |*`hipDeviceAttributeMaxSharedMemoryPerBlock`* | Maximum shared memory available per block in bytes. |
| 9 |*`cudaDevAttrTotalConstantMemory`* |*`hipDeviceAttributeTotalConstantMemory`* | Memory available on device for \__constant__ variables in a CUDA C kernel in bytes. |
| 10 |*`cudaDevAttrWarpSize`* |*`hipDeviceAttributeWarpSize`* | Warp size in threads. |
| 11 |*`cudaDevAttrMaxPitch`* | | Maximum pitch in bytes allowed by memory copies. |
| 12 |*`cudaDevAttrMaxRegistersPerBlock`* |*`hipDeviceAttributeMaxRegistersPerBlock`* | Maximum number of 32-bit registers available per block. |
| 13 |*`cudaDevAttrClockRate`* |*`hipDeviceAttributeClockRate`* | Peak clock frequency in kilohertz. |
| 14 |*`cudaDevAttrTextureAlignment`* | | Alignment requirement for textures. |
| 15 |*`cudaDevAttrGpuOverlap`* | | Device can possibly copy memory and execute a kernel concurrently. |
| 16 |*`cudaDevAttrMultiProcessorCount`* |*`hipDeviceAttributeMultiprocessorCount`* | Number of multiprocessors on device. |
| 17 |*`cudaDevAttrKernelExecTimeout`* | | Specifies whether there is a run time limit on kernels. |
| 18 |*`cudaDevAttrIntegrated`* | | Device is integrated with host memory. |
| 19 |*`cudaDevAttrCanMapHostMemory`* | | Device can map host memory into CUDA address space. |
| 20 |*`cudaDevAttrComputeMode`* |*`hipDeviceAttributeComputeMode`* | Compute mode (See cudaComputeMode for details). |
| 21 |*`cudaDevAttrMaxTexture1DWidth`* | | Maximum 1D texture width. |
| 22 |*`cudaDevAttrMaxTexture2DWidth`* | | Maximum 2D texture width. |
| 23 |*`cudaDevAttrMaxTexture2DHeight`* | | Maximum 2D texture height. |
| 24 |*`cudaDevAttrMaxTexture3DWidth`* | | Maximum 3D texture width. |
| 25 |*`cudaDevAttrMaxTexture3DHeight`* | | Maximum 3D texture height. |
| 26 |*`cudaDevAttrMaxTexture3DDepth`* | | Maximum 3D texture depth. |
| 27 |*`cudaDevAttrMaxTexture2DLayeredWidth`* | | Maximum 2D layered texture width. |
| 28 |*`cudaDevAttrMaxTexture2DLayeredHeight`* | | Maximum 2D layered texture height. |
| 29 |*`cudaDevAttrMaxTexture2DLayeredLayers`* | | Maximum layers in a 2D layered texture. |
| 30 |*`cudaDevAttrSurfaceAlignment`* | | Alignment requirement for surfaces. |
| 31 |*`cudaDevAttrConcurrentKernels`* |*`hipDeviceAttributeConcurrentKernels`* | Device can possibly execute multiple kernels concurrently. |
| 32 |*`cudaDevAttrEccEnabled`* | | Device has ECC support enabled. |
| 33 |*`cudaDevAttrPciBusId`* |*`hipDeviceAttributePciBusId`* | PCI bus ID of the device. |
| 34 |*`cudaDevAttrPciDeviceId`* |*`hipDeviceAttributePciDeviceId`* | PCI device ID of the device. |
| 35 |*`cudaDevAttrTccDriver`* | | Device is using TCC driver model. |
| 36 |*`cudaDevAttrMemoryClockRate`* |*`hipDeviceAttributeMemoryClockRate`* | Peak memory clock frequency in kilohertz. |
| 37 |*`cudaDevAttrGlobalMemoryBusWidth`* |*`hipDeviceAttributeMemoryBusWidth`* | Global memory bus width in bits. |
| 38 |*`cudaDevAttrL2CacheSize`* |*`hipDeviceAttributeL2CacheSize`* | Size of L2 cache in bytes. |
| 39 |*`cudaDevAttrMaxThreadsPerMultiProcessor`* |*`hipDeviceAttributeMaxThreadsPerMultiProcessor`* | Maximum resident threads per multiprocessor. |
| 40 |*`cudaDevAttrAsyncEngineCount`* | | Number of asynchronous engines. |
| 41 |*`cudaDevAttrUnifiedAddressing`* | | Device shares a unified address space with the host. |
| 42 |*`cudaDevAttrMaxTexture1DLayeredWidth`* | | Maximum 1D layered texture width. |
| 43 |*`cudaDevAttrMaxTexture1DLayeredLayers`* | | Maximum layers in a 1D layered texture. |
| 44 | | | *reserved* |
| 45 |*`cudaDevAttrMaxTexture2DGatherWidth`* | | Maximum 2D texture width if cudaArrayTextureGather is set. |
| 46 |*`cudaDevAttrMaxTexture2DGatherHeight`* | | Maximum 2D texture height if cudaArrayTextureGather is set. |
| 47 |*`cudaDevAttrMaxTexture3DWidthAlt`* | | Alternate maximum 3D texture width. |
| 48 |*`cudaDevAttrMaxTexture3DHeightAlt`* | | Alternate maximum 3D texture height. |
| 49 |*`cudaDevAttrMaxTexture3DDepthAlt`* | | Alternate maximum 3D texture depth. |
| 50 |*`cudaDevAttrPciDomainId`* | | PCI domain ID of the device. |
| 51 |*`cudaDevAttrTexturePitchAlignment`* | | Pitch alignment requirement for textures. |
| 52 |*`cudaDevAttrMaxTextureCubemapWidth`* | | Maximum cubemap texture width/height. |
| 53 |*`cudaDevAttrMaxTextureCubemapLayeredWidth`* | | Maximum cubemap layered texture width/height. |
| 54 |*`cudaDevAttrMaxTextureCubemapLayeredLayers`* | | Maximum layers in a cubemap layered texture. |
| 55 |*`cudaDevAttrMaxSurface1DWidth`* | | Maximum 1D surface width. |
| 56 |*`cudaDevAttrMaxSurface2DWidth`* | | Maximum 2D surface width. |
| 57 |*`cudaDevAttrMaxSurface2DHeight`* | | Maximum 2D surface height. |
| 58 |*`cudaDevAttrMaxSurface3DWidth`* | | Maximum 3D surface width. |
| 59 |*`cudaDevAttrMaxSurface3DHeight`* | | Maximum 3D surface height. |
| 60 |*`cudaDevAttrMaxSurface3DDepth`* | | Maximum 3D surface depth. |
| 61 |*`cudaDevAttrMaxSurface1DLayeredWidth`* | | Maximum 1D layered surface width. |
| 62 |*`cudaDevAttrMaxSurface1DLayeredLayers`* | | Maximum layers in a 1D layered surface. |
| 63 |*`cudaDevAttrMaxSurface2DLayeredWidth`* | | Maximum 2D layered surface width. |
| 64 |*`cudaDevAttrMaxSurface2DLayeredHeight`* | | Maximum 2D layered surface height. |
| 65 |*`cudaDevAttrMaxSurface2DLayeredLayers`* | | Maximum layers in a 2D layered surface. |
| 66 |*`cudaDevAttrMaxSurfaceCubemapWidth`* | | Maximum cubemap surface width. |
| 67 |*`cudaDevAttrMaxSurfaceCubemapLayeredWidth`* | | Maximum cubemap layered surface width. |
| 68 |*`cudaDevAttrMaxSurfaceCubemapLayeredLayers`* | | Maximum layers in a cubemap layered surface. |
| 69 |*`cudaDevAttrMaxTexture1DLinearWidth`* | | Maximum 1D linear texture width. |
| 70 |*`cudaDevAttrMaxTexture2DLinearWidth`* | | Maximum 2D linear texture width. |
| 71 |*`cudaDevAttrMaxTexture2DLinearHeight`* | | Maximum 2D linear texture height. |
| 72 |*`cudaDevAttrMaxTexture2DLinearPitch`* | | Maximum 2D linear texture pitch in bytes. |
| 73 |*`cudaDevAttrMaxTexture2DMipmappedWidth`* | | Maximum mipmapped 2D texture width. |
| 74 |*`cudaDevAttrMaxTexture2DMipmappedHeight`* | | Maximum mipmapped 2D texture height. |
| 75 |*`cudaDevAttrComputeCapabilityMajor`* |*`hipDeviceAttributeComputeCapabilityMajor`* | Major compute capability version number. |
| 76 |*`cudaDevAttrComputeCapabilityMinor`* |*`hipDeviceAttributeComputeCapabilityMinor`* | Minor compute capability version number. |
| 77 |*`cudaDevAttrMaxTexture1DMipmappedWidth`* | | Maximum mipmapped 1D texture width. |
| 78 |*`cudaDevAttrStreamPrioritiesSupported`* | | Device supports stream priorities. |
| 79 |*`cudaDevAttrGlobalL1CacheSupported`* | | Device supports caching globals in L1. |
| 80 |*`cudaDevAttrLocalL1CacheSupported`* | | Device supports caching locals in L1. |
| 81 |*`cudaDevAttrMaxSharedMemoryPerMultiprocessor`*|*`hipDeviceAttributeMaxSharedMemoryPerMultiprocessor`*| Maximum shared memory available per multiprocessor in bytes. |
| 82 |*`cudaDevAttrMaxRegistersPerMultiprocessor`* | | Maximum number of 32-bit registers available per multiprocessor. |
| 83 |*`cudaDevAttrManagedMemory`* | | Device can allocate managed memory on this system. |
| 84 |*`cudaDevAttrIsMultiGpuBoard`* |*`hipDeviceAttributeIsMultiGpuBoard`* | Device is on a multi-GPU board. |
| 85 |*`cudaDevAttrMultiGpuBoardGroupID`* | | Unique identifier for a group of devices on the same multi-GPU board. |
| enum |***`cudaError`*** |***`hipError_t`*** | CUDA Error types. |
| enum |***`cudaError_t`*** |***`hipError_t`*** | CUDA Error types. |
| 0 |*`cudaSuccess`* |*`hipSuccess`* | The API call returned with no errors. In the case of query calls, this can also mean that the operation being queried is complete.|
| 1 |*`cudaErrorMissingConfiguration`* | | The device function being invoked (usually via cudaLaunchKernel()) was not previously configured via the cudaConfigureCall() function.|
| 2 |*`cudaErrorMemoryAllocation`* |*`hipErrorMemoryAllocation`* | The API call failed because it was unable to allocate enough memory to perform the requested operation. |
| 3 |*`cudaErrorInitializationError`* |*`hipErrorInitializationError`* | The API call failed because the CUDA driver and runtime could not be initialized. |
| 4 |*`cudaErrorLaunchFailure`* | | An exception occurred on the device while executing a kernel. Common causes include dereferencing an invalid device pointer and accessing out of bounds shared memory. The device cannot be used until cudaThreadExit() is called. All existing device memory allocations are invalid and must be reconstructed if the program is to continue using CUDA. |
| 5 |*`cudaErrorPriorLaunchFailure`* | | This indicated that a previous kernel launch failed. This was previously used for device emulation of kernel launches. Deprecated This error return is deprecated as of CUDA 3.1. Device emulation mode was removed with the CUDA 3.1 release.|
| 6 |*`cudaErrorLaunchTimeout`* | | This indicates that the device kernel took too long to execute. This can only occur if timeouts are enabled - see the device property kernelExecTimeoutEnabled for more information. The device cannot be used until cudaThreadExit() is called. All existing device memory allocations are invalid and must be reconstructed if the program is to continue using CUDA.|
| 7 |*`cudaErrorLaunchOutOfResources`* |*`hipErrorLaunchOutOfResources`* | This indicates that a launch did not occur because it did not have appropriate resources. Although this error is similar to cudaErrorInvalidConfiguration, this error usually indicates that the user has attempted to pass too many arguments to the device kernel, or the kernel launch specifies too many threads for the kernel's register count.|
| 8 |*`cudaErrorInvalidDeviceFunction`* | | The requested device function does not exist or is not compiled for the proper device architecture. |
| 9 |*`cudaErrorInvalidConfiguration`* | | This indicates that a kernel launch is requesting resources that can never be satisfied by the current device. Requesting more shared memory per block than the device supports will trigger this error, as will requesting too many threads or blocks. See cudaDeviceProp for more device limitations.|
| 10 |*`cudaErrorInvalidDevice`* |*`hipErrorInvalidDevice`* | This indicates that the device ordinal supplied by the user does not correspond to a valid CUDA device. |
| 11 |*`cudaErrorInvalidValue`* |*`hipErrorInvalidValue`* | This indicates that one or more of the parameters passed to the API call is not within an acceptable range of values. |
| 12 |*`cudaErrorInvalidPitchValue`* | | This indicates that one or more of the pitch-related parameters passed to the API call is not within the acceptable range for pitch.|
| 13 |*`cudaErrorInvalidSymbol`* | | This indicates that the symbol name/identifier passed to the API call is not a valid name or identifier. |
| 14 |*`cudaErrorMapBufferObjectFailed`* | | This indicates that the buffer object could not be mapped. |
| 15 |*`cudaErrorUnmapBufferObjectFailed`* | | This indicates that the buffer object could not be unmapped. |
| 16 |*`cudaErrorInvalidHostPointer`* | | This indicates that at least one host pointer passed to the API call is not a valid host pointer. |
| 17 |*`cudaErrorInvalidDevicePointer`* |*`hipErrorInvalidDevicePointer`* | This indicates that at least one host pointer passed to the API call is not a valid host pointer. |
| 18 |*`cudaErrorInvalidTexture`* | | This indicates that the texture passed to the API call is not a valid texture. |
| 19 |*`cudaErrorInvalidTextureBinding`* | | This indicates that the texture binding is not valid. This occurs if you call cudaGetTextureAlignmentOffset() with an unbound texture.|
| 20 |*`cudaErrorInvalidChannelDescriptor`* | | This indicates that the channel descriptor passed to the API call is not valid. This occurs if the format is not one of the formats specified by cudaChannelFormatKind, or if one of the dimensions is invalid.|
| 21 |*`cudaErrorInvalidMemcpyDirection`* | | This indicates that the direction of the memcpy passed to the API call is not one of the types specified by cudaMemcpyKind. |
| 22 |*`cudaErrorAddressOfConstant`* | | This indicated that the user has taken the address of a constant variable, which was forbidden up until the CUDA 3.1 release. Deprecated This error return is deprecated as of CUDA 3.1. Variables in constant memory may now have their address taken by the runtime via cudaGetSymbolAddress().|
| 23 |*`cudaErrorTextureFetchFailed`* | | This indicated that a texture fetch was not able to be performed. This was previously used for device emulation of texture operations. Deprecated This error return is deprecated as of CUDA 3.1. Device emulation mode was removed with the CUDA 3.1 release.|
| 24 |*`cudaErrorTextureNotBound`* | | This indicated that a texture was not bound for access. This was previously used for device emulation of texture operations. Deprecated This error return is deprecated as of CUDA 3.1. Device emulation mode was removed with the CUDA 3.1 release.|
| 25 |*`cudaErrorSynchronizationError`* | | This indicated that a synchronization operation had failed. This was previously used for some device emulation functions. Deprecated This error return is deprecated as of CUDA 3.1. Device emulation mode was removed with the CUDA 3.1 release.|
| 26 |*`cudaErrorInvalidFilterSetting`* | | This indicates that a non-float texture was being accessed with linear filtering. This is not supported by CUDA. |
| 27 |*`cudaErrorInvalidNormSetting`* | | This indicates that an attempt was made to read a non-float texture as a normalized float. This is not supported by CUDA. |
| 28 |*`cudaErrorMixedDeviceExecution`* | | Mixing of device and device emulation code was not allowed. Deprecated This error return is deprecated as of CUDA 3.1. Device emulation mode was removed with the CUDA 3.1 release.|
| 29 |*`cudaErrorCudartUnloading`* | | This indicates that a CUDA Runtime API call cannot be executed because it is being called during process shut down, at a point in time after CUDA driver has been unloaded.|
| 30 |*`cudaErrorUnknown`* |*`hipErrorUnknown`* | This indicates that an unknown internal error has occurred. |
| 31 |*`cudaErrorNotYetImplemented`* | | This indicates that the API call is not yet implemented. Production releases of CUDA will never return this error. Deprecated This error return is deprecated as of CUDA 4.1.|
| 32 |*`cudaErrorMemoryValueTooLarge`* | | This indicated that an emulated device pointer exceeded the 32-bit address range. Deprecated This error return is deprecated as of CUDA 3.1. Device emulation mode was removed with the CUDA 3.1 release.|
| 33 |*`cudaErrorInvalidResourceHandle`* |*`hipErrorInvalidResourceHandle`* | This indicates that a resource handle passed to the API call was not valid. Resource handles are opaque types like cudaStream_t and cudaEvent_t.|
| 34 |*`cudaErrorNotReady`* |*`hipErrorNotReady`* | This indicates that asynchronous operations issued previously have not completed yet. This result is not actually an error, but must be indicated differently than cudaSuccess (which indicates completion). Calls that may return this value include cudaEventQuery() and cudaStreamQuery().|
| 35 |*`cudaErrorInsufficientDriver`* | | This indicates that the installed NVIDIA CUDA driver is older than the CUDA runtime library. This is not a supported configuration. Users should install an updated NVIDIA display driver to allow the application to run.|
| 36 |*`cudaErrorSetOnActiveProcess`* | | This indicates that the user has called cudaSetValidDevices(), cudaSetDeviceFlags(), cudaD3D9SetDirect3DDevice(), cudaD3D10SetDirect3DDevice, cudaD3D11SetDirect3DDevice(), or cudaVDPAUSetVDPAUDevice() after initializing the CUDA runtime by calling non-device management operations (allocating memory and launching kernels are examples of non-device management operations). This error can also be returned if using runtime/driver interoperability and there is an existing CUcontext active on the host thread.|
| 37 |*`cudaErrorInvalidSurface`* | | This indicates that the surface passed to the API call is not a valid surface. |
| 38 |*`cudaErrorNoDevice`* |*`hipErrorNoDevice`* | This indicates that no CUDA-capable devices were detected by the installed CUDA driver. |
| 39 |*`cudaErrorECCUncorrectable`* | | This indicates that an uncorrectable ECC error was detected during execution. |
| 40 |*`cudaErrorSharedObjectSymbolNotFound`* | | This indicates that a link to a shared object failed to resolve. |
| 41 |*`cudaErrorSharedObjectInitFailed`* | | This indicates that initialization of a shared object failed. |
| 42 |*`cudaErrorUnsupportedLimit`* |*`hipErrorUnsupportedLimit`* | This indicates that the cudaLimit passed to the API call is not supported by the active device. |
| 43 |*`cudaErrorDuplicateVariableName`* | | This indicates that multiple global or constant variables (across separate CUDA source files in the application) share the same string name.|
| 44 |*`cudaErrorDuplicateTextureName`* | | This indicates that multiple textures (across separate CUDA source files in the application) share the same string name. |
| 45 |*`cudaErrorDuplicateSurfaceName`* | | This indicates that multiple surfaces (across separate CUDA source files in the application) share the same string name. |
| 46 |*`cudaErrorDevicesUnavailable`* | | This indicates that all CUDA devices are busy or unavailable at the current time. Devices are often busy/unavailable due to use of cudaComputeModeExclusive, cudaComputeModeProhibited or when long running CUDA kernels have filled up the GPU and are blocking new work from starting. They can also be unavailable due to memory constraints on a device that already has active CUDA work being performed.|
| 47 |*`cudaErrorInvalidKernelImage`* | | This indicates that the device kernel image is invalid. |
| 48 |*`cudaErrorNoKernelImageForDevice`* | | This indicates that there is no kernel image available that is suitable for the device. This can occur when a user specifies code generation options for a particular CUDA source file that do not include the corresponding device configuration.|
| 49 |*`cudaErrorIncompatibleDriverContext`* | | This indicates that the current context is not compatible with this the CUDA Runtime. This can only occur if you are using CUDA Runtime/Driver interoperability and have created an existing Driver context using the driver API. The Driver context may be incompatible either because the Driver context was created using an older version of the API, because the Runtime API call expects a primary driver context and the Driver context is not primary, or because the Driver context has been destroyed. Please see Interactions with the CUDA Driver API" for more information.|
| 50 |*`cudaErrorPeerAccessAlreadyEnabled`* |*`hipErrorPeerAccessAlreadyEnabled`* | This error indicates that a call to cudaDeviceEnablePeerAccess() is trying to re-enable peer addressing on from a context which has already had peer addressing enabled.|
| 51 |*`cudaErrorPeerAccessNotEnabled`* |*`hipErrorPeerAccessNotEnabled`* | This error indicates that a call to cudaDeviceEnablePeerAccess() is trying to re-enable peer addressing on from a context which has already had peer addressing enabled.|
| 52 | | | *reserved* |
| 53 | | | *reserved* |
| 54 |*`cudaErrorDeviceAlreadyInUse`* | | This indicates that a call tried to access an exclusive-thread device that is already in use by a different thread. |
| 55 |*`cudaErrorProfilerDisabled`* | | This indicates profiler is not initialized for this run. This can happen when the application is running with external profiling tools like visual profiler.|
| 56 |*`cudaErrorProfilerNotInitialized`* | | Deprecated This error return is deprecated as of CUDA 5.0. It is no longer an error to attempt to enable/disable the profiling via cudaProfilerStart or cudaProfilerStop without initialization.|
| 57 |*`cudaErrorProfilerAlreadyStarted`* | | Deprecated This error return is deprecated as of CUDA 5.0. It is no longer an error to call cudaProfilerStart() when profiling is already enabled.|
| 58 |*`cudaErrorProfilerAlreadyStopped`* | | Deprecated This error return is deprecated as of CUDA 5.0. It is no longer an error to call cudaProfilerStop() when profiling is already disabled.|
| 59 |*`cudaErrorAssert`* | | An assert triggered in device code during kernel execution. The device cannot be used again until cudaThreadExit() is called. All existing allocations are invalid and must be reconstructed if the program is to continue using CUDA.|
| 60 |*`cudaErrorTooManyPeers`* | | This error indicates that the hardware resources required to enable peer access have been exhausted for one or more of the devices passed to cudaEnablePeerAccess().|
| 61 |*`cudaErrorHostMemoryAlreadyRegistered`* | *`hipErrorHostMemoryAlreadyRegistered`* | This error indicates that the memory range passed to cudaHostRegister() has already been registered. |
| 62 |*`cudaErrorHostMemoryNotRegistered`* | *`hipErrorHostMemoryNotRegistered`* | This error indicates that the pointer passed to cudaHostUnregister() does not correspond to any currently registered memory region.|
| 63 |*`cudaErrorOperatingSystem`* | | This error indicates that an OS call failed. |
| 64 |*`cudaErrorPeerAccessUnsupported`* | | This error indicates that P2P access is not supported across the given devices. |
| 65 |*`cudaErrorLaunchMaxDepthExceeded`* | | This error indicates that a device runtime grid launch did not occur because the depth of the child grid would exceed the maximum supported number of nested grid launches.|
| 66 |*`cudaErrorLaunchFileScopedTex`* | | This error indicates that a grid launch did not occur because the kernel uses filescoped textures which are unsupported by the device runtime. Kernels launched via the device runtime only support textures created with the Texture Object API's.|
| 67 |*`cudaErrorLaunchFileScopedSurf`* | | This error indicates that a grid launch did not occur because the kernel uses filescoped surfaces which are unsupported by the device runtime. Kernels launched via the device runtime only support surfaces created with the Surface Object API's.|
| 68 |*`cudaErrorSyncDepthExceeded`* | | This error indicates that a call to cudaDeviceSynchronize made from the device runtime failed because the call was made at grid depth greater than than either the default (2 levels of grids) or user specified device limit cudaLimitDevRuntimeSyncDepth. To be able to synchronize on launched grids at a greater depth successfully, the maximum nested depth at which cudaDeviceSynchronize will be called must be specified with the cudaLimitDevRuntimeSyncDepth limit to the cudaDeviceSetLimit api before the host-side launch of a kernel using the device runtime. Keep in mind that additional levels of sync depth require the runtime to reserve large amounts of device memory that cannot be used for user allocations.|
| 69 |*`cudaErrorLaunchPendingCountExceeded`* | | This error indicates that a device runtime grid launch failed because the launch would exceed the limit cudaLimitDevRuntimePendingLaunchCount. For this launch to proceed successfully, cudaDeviceSetLimit must be called to set the cudaLimitDevRuntimePendingLaunchCount to be higher than the upper bound of outstanding launches that can be issued to the device runtime. Keep in mind that raising the limit of pending device runtime launches will require the runtime to reserve device memory that cannot be used for user allocations.|
| 70 |*`cudaErrorNotPermitted`* | | This error indicates the attempted operation is not permitted. |
| 71 |*`cudaErrorNotSupported`* | | This error indicates the attempted operation is not supported on the current system or device. |
| 72 |*`cudaErrorHardwareStackError`* | | Device encountered an error in the call stack during kernel execution, possibly due to stack corruption or exceeding the stack size limit. The context cannot be used, so it must be destroyed (and a new one should be created). All existing device memory allocations from this context are invalid and must be reconstructed if the program is to continue using CUDA.|
| 73 |*`cudaErrorIllegalInstruction`* | | The device encountered an illegal instruction during kernel execution The context cannot be used, so it must be destroyed (and a new one should be created). All existing device memory allocations from this context are invalid and must be reconstructed if the program is to continue using CUDA.|
| 74 |*`cudaErrorMisalignedAddress`* | | The device encountered a load or store instruction on a memory address which is not aligned. The context cannot be used, so it must be destroyed (and a new one should be created). All existing device memory allocations from this context are invalid and must be reconstructed if the program is to continue using CUDA.|
| 75 |*`cudaErrorInvalidAddressSpace`* | | While executing a kernel, the device encountered an instruction which can only operate on memory locations in certain address spaces (global, shared, or local), but was supplied a memory address not belonging to an allowed address space. The context cannot be used, so it must be destroyed (and a new one should be created). All existing device memory allocations from this context are invalid and must be reconstructed if the program is to continue using CUDA.|
| 76 |*`cudaErrorInvalidPc`* | | The device encountered an invalid program counter. The context cannot be used, so it must be destroyed (and a new one should be created). All existing device memory allocations from this context are invalid and must be reconstructed if the program is to continue using CUDA.|
| 77 |*`cudaErrorIllegalAddress`* | | The device encountered a load or store instruction on an invalid memory address. The context cannot be used, so it must be destroyed (and a new one should be created). All existing device memory allocations from this context are invalid and must be reconstructed if the program is to continue using CUDA.|
| 78 |*`cudaErrorInvalidPtx`* | | A PTX compilation failed. The runtime may fall back to compiling PTX if an application does not contain a suitable binary for the current device.|
| 79 |*`cudaErrorInvalidGraphicsContext`* | | This indicates an error with the OpenGL or DirectX context. |
| 0x7f |*`cudaErrorStartupFailure`* | | This indicates an internal startup failure in the CUDA runtime. |
| 1000 |*`cudaErrorApiFailureBase`* | | Any unhandled CUDA driver error is added to this value and returned via the runtime. Production releases of CUDA should not return such errors. Deprecated This error return is deprecated as of CUDA 4.1.|
| enum |***`cudaFuncCache`*** |***`hipFuncCache_t`*** | CUDA function cache configurations. |
| 0 |*`cudaFuncCachePreferNone`* |*`hipFuncCachePreferNone`* | Default function cache configuration, no preference. |
| 1 |*`cudaFuncCachePreferShared`* |*`hipFuncCachePreferShared`* | Prefer larger shared memory and smaller L1 cache. |
| 2 |*`cudaFuncCachePreferL1`* |*`hipFuncCachePreferL1`* | Prefer larger L1 cache and smaller shared memory. |
| 3 |*`cudaFuncCachePreferEqual`* |*`hipFuncCachePreferEqual`* | Prefer equal size L1 cache and shared memory. |
| enum |***`cudaGraphicsCubeFace`*** | | CUDA graphics interop array indices for cube maps. |
| 0x00 |*`cudaGraphicsCubeFacePositiveX`* | | Positive X face of cubemap. |
| 0x01 |*`cudaGraphicsCubeFaceNegativeX`* | | Negative X face of cubemap. |
| 0x02 |*`cudaGraphicsCubeFacePositiveY`* | | Positive Y face of cubemap. |
| 0x03 |*`cudaGraphicsCubeFaceNegativeY`* | | Negative Y face of cubemap. |
| 0x04 |*`cudaGraphicsCubeFacePositiveZ`* | | Positive Z face of cubemap. |
| 0x05 |*`cudaGraphicsCubeFaceNegativeZ`* | | Negative Z face of cubemap. |
| enum |***`cudaGraphicsMapFlags`*** | | CUDA graphics interop map flags. |
| 0 |*`cudaGraphicsMapFlagsNone`* | | Default; Assume resource can be read/written. |
| 1 |*`cudaGraphicsMapFlagsReadOnly`* | | CUDA will not write to this resource. |
| 2 |*`cudaGraphicsMapFlagsWriteDiscard`* | | CUDA will only write to and will not read from this resource. |
| enum |***`cudaGraphicsRegisterFlags`*** | | CUDA graphics interop register flags. |
| 0 |*`cudaGraphicsRegisterFlagsNone`* | | Default. |
| 1 |*`cudaGraphicsRegisterFlagsReadOnly`* | | CUDA will not write to this resource. |
| 2 |*`cudaGraphicsRegisterFlagsWriteDiscard`* | | CUDA will only write to and will not read from this resource. |
| 4 |*`cudaGraphicsRegisterFlagsSurfaceLoadStore`* | | CUDA will bind this resource to a surface reference. |
| 8 |*`cudaGraphicsRegisterFlagsTextureGather`* | | CUDA will perform texture gather operations on this resource. |
| enum |***`cudaLimit`*** |***`hipLimit_t`*** | CUDA Limits. |
| 0x00 |*`cudaLimitStackSize`* | | GPU thread stack size. |
| 0x01 |*`cudaLimitPrintfFifoSize`* | | GPU printf/fprintf FIFO size. |
| 0x02 |*`cudaLimitMallocHeapSize`* |*`hipLimitMallocHeapSize`* | GPU malloc heap size. |
| 0x03 |*`cudaLimitDevRuntimeSyncDepth`* | | GPU device runtime synchronize depth. |
| 0x04 |*`cudaLimitDevRuntimePendingLaunchCount`* | | GPU device runtime pending launch count. |
| enum |***`cudaMemcpyKind`*** |***`hipMemcpyKind`*** | CUDA memory copy types. |
| 0 |*`cudaMemcpyHostToHost`* |*`hipMemcpyHostToHost`* | Host -> Host. |
| 1 |*`cudaMemcpyHostToDevice`* |*`hipMemcpyHostToDevice`* | Host -> Device. |
| 2 |*`cudaMemcpyDeviceToHost`* |*`hipMemcpyDeviceToHost`* | Device -> Host. |
| 3 |*`cudaMemcpyDeviceToDevice`* |*`hipMemcpyDeviceToDevice`* | Device -> Device. |
| 4 |*`cudaMemcpyDefault`* |*`hipMemcpyDefault`* | Default based unified virtual address space. |
| enum |***`cudaMemoryType`*** |***`hipMemoryType`*** | CUDA memory types. |
| 1 |*`cudaMemoryTypeHost`* |*`hipMemoryTypeHost`* | Host memory. |
| 2 |*`cudaMemoryTypeDevice`* |*`hipMemoryTypeDevice`* | Device memory. |
| enum |***`cudaResourceType`*** | | CUDA resource types. |
| 0 |*`cudaResourceTypeArray`* | | Array resource. |
| 1 |*`cudaResourceTypeMipmappedArray`* | | Mipmapped array resource. |
| 2 |*`cudaResourceTypeLinear`* | | Linear resource. |
| 3 |*`cudaResourceTypePitch2D`* | | Pitch 2D resource. |
| enum |***`cudaResourceViewFormat`*** | | CUDA texture resource view formats. |
| 0x00 |*`cudaResViewFormatNone`* | | No resource view format (use underlying resource format). |
| 0x01 |*`cudaResViewFormatUnsignedChar1`* | | 1 channel unsigned 8-bit integers. |
| 0x02 |*`cudaResViewFormatUnsignedChar2`* | | 2 channel unsigned 8-bit integers. |
| 0x03 |*`cudaResViewFormatUnsignedChar4`* | | 4 channel unsigned 8-bit integers. |
| 0x04 |*`cudaResViewFormatSignedChar1`* | | 1 channel signed 8-bit integers. |
| 0x05 |*`cudaResViewFormatSignedChar2`* | | 2 channel signed 8-bit integers. |
| 0x06 |*`cudaResViewFormatSignedChar4`* | | 4 channel signed 8-bit integers. |
| 0x07 |*`cudaResViewFormatUnsignedShort1`* | | 1 channel unsigned 16-bit integers. |
| 0x08 |*`cudaResViewFormatUnsignedShort2`* | | 2 channel unsigned 16-bit integers. |
| 0x09 |*`cudaResViewFormatUnsignedShort4`* | | 4 channel unsigned 16-bit integers. |
| 0x0a |*`cudaResViewFormatSignedShort1`* | | 1 channel signed 16-bit integers. |
| 0x0b |*`cudaResViewFormatSignedShort2`* | | 2 channel signed 16-bit integers. |
| 0x0c |*`cudaResViewFormatSignedShort4`* | | 4 channel signed 16-bit integers. |
| 0x0d |*`cudaResViewFormatUnsignedInt1`* | | 1 channel unsigned 32-bit integers. |
| 0x0e |*`cudaResViewFormatUnsignedInt2`* | | 2 channel unsigned 32-bit integers. |
| 0x0f |*`cudaResViewFormatUnsignedInt4`* | | 4 channel unsigned 32-bit integers. |
| 0x10 |*`cudaResViewFormatSignedInt1`* | | 1 channel signed 32-bit integers. |
| 0x11 |*`cudaResViewFormatSignedInt2`* | | 2 channel signed 32-bit integers. |
| 0x12 |*`cudaResViewFormatSignedInt4`* | | 4 channel signed 32-bit integers. |
| 0x13 |*`cudaResViewFormatHalf1`* | | 1 channel 16-bit floating point. |
| 0x14 |*`cudaResViewFormatHalf2`* | | 2 channel 16-bit floating point. |
| 0x15 |*`cudaResViewFormatHalf4`* | | 4 channel 16-bit floating point. |
| 0x16 |*`cudaResViewFormatFloat1`* | | 1 channel 32-bit floating point. |
| 0x17 |*`cudaResViewFormatFloat2`* | | 2 channel 32-bit floating point. |
| 0x18 |*`cudaResViewFormatFloat4`* | | 4 channel 32-bit floating point. |
| 0x19 |*`cudaResViewFormatUnsignedBlockCompressed1`* | | Block compressed 1. |
| 0x1a |*`cudaResViewFormatUnsignedBlockCompressed2`* | | Block compressed 2. |
| 0x1b |*`cudaResViewFormatUnsignedBlockCompressed3`* | | Block compressed 3. |
| 0x1c |*`cudaResViewFormatUnsignedBlockCompressed4`* | | Block compressed 4 unsigned. |
| 0x1d |*`cudaResViewFormatSignedBlockCompressed4`* | | Block compressed 4 signed. |
| 0x1e |*`cudaResViewFormatUnsignedBlockCompressed5`* | | Block compressed 5 unsigned. |
| 0x1f |*`cudaResViewFormatSignedBlockCompressed5`* | | Block compressed 5 signed. |
| 0x20 |*`cudaResViewFormatUnsignedBlockCompressed6H`* | | Block compressed 6 unsigned half-float. |
| 0x21 |*`cudaResViewFormatSignedBlockCompressed6H`* | | Block compressed 6 signed half-float. |
| 0x22 |*`cudaResViewFormatUnsignedBlockCompressed7`* | | Block compressed 7. |
| enum |***`cudaSharedMemConfig`*** |***`hipSharedMemConfig`*** | CUDA shared memory configuration. |
| 0 |*`cudaSharedMemBankSizeDefault`* |*`hipSharedMemBankSizeDefault`* | |
| 1 |*`cudaSharedMemBankSizeFourByte`* |*`hipSharedMemBankSizeFourByte`* | |
| 2 |*`cudaSharedMemBankSizeEightByte`* |*`hipSharedMemBankSizeEightByte`* | |
| enum |***`cudaSurfaceBoundaryMode`*** | | CUDA Surface boundary modes. |
| 0 |*`cudaBoundaryModeZero`* | | Zero boundary mode. |
| 1 |*`cudaBoundaryModeClamp`* | | Clamp boundary mode. |
| 2 |*`cudaBoundaryModeTrap`* | | Trap boundary mode. |
| enum |***`cudaSurfaceFormatMode`*** | | CUDA Surface format modes. |
| 0 |*`cudaFormatModeForced`* | | Forced format mode. |
| 1 |*`cudaFormatModeAuto`* | | Auto format mode. |
| enum |***`cudaTextureAddressMode`*** | | CUDA texture address modes. |
| 0 |*`cudaAddressModeWrap`* | | Wrapping address mode. |
| 1 |*`cudaAddressModeClamp`* | | Clamp to edge address mode. |
| 2 |*`cudaAddressModeMirror`* | | Mirror address mode. |
| 3 |*`cudaAddressModeBorder`* | | Border address mode. |
| enum |***`cudaTextureFilterMode`*** |***`hipTextureFilterMode`*** | Point filter mode. |
| 0 |*`cudaFilterModePoint`* |*`hipFilterModePoint`* | Linear filter mode. |
| 1 |*`cudaFilterModeLinear`* | | Clamp to edge address mode. |
| enum |***`cudaTextureReadMode`*** |***`hipTextureReadMode`*** | CUDA texture read modes. |
| 0 |*`cudaReadModeElementType`* |*`hipReadModeElementType`* | Read texture as specified element type. |
| 1 |*`cudaReadModeNormalizedFloat`* | | Read texture as normalized float. |
| struct | `cudaArray` | `hipArray` | CUDA array [opaque]. |
| typedef | `cudaArray_t` | `hipArray *` | CUDA array pointer. |
| typedef | `cudaArray_const_t` | `const hipArray *` | CUDA array (as source copy argument). |
| enum | `cudaError` | `hipError_t` | CUDA Error types. |
| typedef | `cudaError_t` | `hipError_t` | CUDA Error types. |
| typedef | `cudaEvent_t` | `hipEvent_t` | CUDA event types. |
| typedef | `cudaGraphicsResource_t` | | CUDA graphics resource types. |
| typedef | `cudaMipmappedArray_t` | | CUDA mipmapped array. |
| typedef | `cudaMipmappedArray_const_t` | | CUDA mipmapped array (as source argument). |
| enum |***`cudaOutputMode`*** | | CUDA Profiler Output modes. |
| 0x00 |*`cudaKeyValuePair`* | | Output mode Key-Value pair format. |
| 0x01 |*`cudaCSV`* | | Output mode Comma separated values format. |
| typedef | `cudaOutputMode_t` | | CUDA output file modes. |
| typedef | `cudaStream_t` | `hipStream_t` | CUDA stream. |
| typedef | `cudaSurfaceObject_t` | | An opaque value that represents a CUDA Surface object. |
| typedef | `cudaTextureObject_t` | | An opaque value that represents a CUDA texture object. |
| typedef | `CUuuid_stcudaUUID_t` | | CUDA UUID types. |
| define | `CUDA_IPC_HANDLE_SIZE` | | CUDA IPC Handle Size. |
| define | `cudaArrayCubemap` | | Must be set in cudaMalloc3DArray to create a cubemap CUDA array. |
| define | `cudaArrayDefault` | | Default CUDA array allocation flag. |
| define | `cudaArrayLayered` | | Must be set in cudaMalloc3DArray to create a layered CUDA array. |
| define | `cudaArraySurfaceLoadStore` | | Must be set in cudaMallocArray or cudaMalloc3DArray in order to bind surfaces to the CUDA array. |
| define | `cudaArrayTextureGather` | | Must be set in cudaMallocArray or cudaMalloc3DArray in order to perform texture gather operations on the CUDA array. |
| define | `cudaDeviceBlockingSync` | `hipDeviceScheduleBlockingSync` | Device flag - Use blocking synchronization. Deprecated as of CUDA 4.0 and replaced with cudaDeviceScheduleBlockingSync. |
| define | `cudaDeviceLmemResizeToMax` | | Device flag - Keep local memory allocation after launch. |
| define | `cudaDeviceMapHost` | | Device flag - Support mapped pinned allocations. |
| define | `cudaDeviceMask` | | Device flags mask. |
| define | `cudaDevicePropDontCare` | | Empty device properties. |
| define | `cudaDeviceScheduleAuto` | `hipDeviceScheduleAuto` | Device flag - Automatic scheduling. |
| define | `cudaDeviceScheduleBlockingSync` | `hipDeviceScheduleBlockingSync` | Device flag - Use blocking synchronization. |
| define | `cudaDeviceScheduleMask` | `hipDeviceScheduleMask` | Device schedule flags mask. |
| define | `cudaDeviceScheduleSpin` | `hipDeviceScheduleSpin` | Device flag - Spin default scheduling. |
| define | `cudaDeviceScheduleYield` | `hipDeviceScheduleYield` | Device flag - Yield default scheduling. |
| define | `cudaEventDefault` | `hipEventDefault` | Default event flag. |
| define | `cudaEventDisableTiming` | `hipEventDisableTiming` | Event will not record timing data. |
| define | `cudaEventInterprocess` | `hipEventInterprocess` | Event is suitable for interprocess use. cudaEventDisableTiming must be set. |
| define | `cudaHostAllocDefault` | `hipHostMallocDefault` | Default page-locked allocation flag. |
| define | `cudaHostAllocMapped` | `hipHostMallocMapped` | Map allocation into device space. |
| define | `cudaHostAllocPortable` | `hipHostMallocPortable` | Pinned memory accessible by all CUDA contexts. |
| define | `cudaHostAllocWriteCombined` | `hipHostMallocWriteCombined` | Write-combined memory. |
| define | `cudaHostRegisterDefault` | `hipHostRegisterDefault` | Default host memory registration flag. |
| define | `cudaHostRegisterIoMemory` | `hipHostRegisterIoMemory` | Memory-mapped I/O space. |
| define | `cudaHostRegisterMapped` | `hipHostRegisterMapped` | Map registered memory into device space. |
| define | `cudaHostRegisterPortable` | `hipHostRegisterPortable` | Pinned memory accessible by all CUDA contexts. |
| define | `cudaIpcMemLazyEnablePeerAccess` | `hipIpcMemLazyEnablePeerAccess` | Automatically enable peer access between remote devices as needed. |
| define | `cudaMemAttachGlobal` | | Memory can be accessed by any stream on any device. |
| define | `cudaMemAttachHost` | | Memory cannot be accessed by any stream on any device. |
| define | `cudaMemAttachSingle` | | Memory can only be accessed by a single stream on the associated device. |
| define | `cudaOccupancyDefault` | | Default behavior. |
| define | `cudaOccupancyDisableCachingOverride` | | Assume global caching is enabled and cannot be automatically turned off. |
| define | `cudaPeerAccessDefault` | | Default peer addressing enable flag. |
| define | `cudaStreamDefault` | `hipStreamDefault` | Default stream flag. |
| define | `cudaStreamLegacy` | | Default stream flag. |
| define | `cudaStreamNonBlocking` | `hipStreamNonBlocking` | Stream does not synchronize with stream 0 (the NULL stream). |
| define | `cudaStreamPerThread` | | Per-thread stream handle. |
projects/hip/docs/markdown/device_md_gen.py
+508
Datei anzeigen
@@ -0,0 +1,508 @@
"""
Copyright (c) 2015-2017 Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
"""
"""
1. This files uses Python3 to run
List of device functions:
acosf
acoshf
asinf
asinhf
atan2f
atanf
atanhf
cbrtf
ceilf
copysignf
cosf
coshf
cospif
cyl_bessel_i0f
cyl_bessel_i1f
erfcf
erfcinvf
erfcxf
erff
erfinvf
exp10f
exp2f
expf
expm1f
fabsf
fdimf
fdividef
floorf
fmaf
fmaxf
fminf
fmodf
frexpf
hypotf
ilogbf
isfinite
isinf
isnan
j0f
j1f
jnf
ldexpf
lgammaf
llrintf
llroundf
log10f
log1pf
logbf
lrintf
lroundf
modff
nanf
nearbyintf
nextafterf
norm3df
norm4df
normcdff
normcdfinvf
normf
powf
rcbrtf
remainderf
remquof
rhypotf
rintf
rnorm3df
rnorm4df
rnormf
roundf
rsqrtf
scalblnf
scalbnf
signbit
sincosf
sincospif
sinf
sinhf
sinpif
sqrtf
tanf
tanhf
tgammaf
truncf
y0f
y1f
ynf
acos
acosh
asin
asinh
atan
atan2
atanh
cbrt
ceil
copysign
cos
cosh
cospi
cyl_bessel_i0
cyl_bessel_i1
erf
erfc
erfcinv
erfcx
erfinv
exp
exp10
exp2
expm1
fabs
fdim
floor
fma
fmax
fmin
fmod
frexp
hypot
ilogb
isfinite
isinf
isnan
j0
j1
jn
ldexp
lgamma
llrint
llround
log
log10
log1p
log2
logb
lrint
lround
modf
nan
nearbyint
nextafter
norm
norm3d
norm4d
normcdf
normcdfinv
pow
rcbrt
remainder
remquo
rhypot
rint
rnorm
rnorm3d
rnorm4d
round
rsqrt
scalbln
scalbn
signbit
sin
sincos
sincospi
sinh
sinpi
sqrt
tan
tanh
tgamma
trunc
y0
y1
yn
__cosf
__exp10f
__expf
__fadd_rd
__fadd_rn
__fadd_ru
__fadd_rz
__fdiv_rd
__fdiv_rn
__fdiv_ru
__fdiv_rz
__fdividef
__fmaf_rd
__fmaf_rn
__fmaf_ru
__fmaf_rz
__fmul_rd
__fmul_rn
__fmul_ru
__fmul_rz
__frcp_rd
__frcp_rn
__frcp_ru
__frcp_rz
__frsqrt_rn
__fsqrt_rd
__fsqrt_rn
__fsqrt_ru
__fsqrt_rz
__fsub_rd
__fsub_rn
__fsub_ru
__log10f
__log2f
__logf
__powf
__saturatef
__sincosf
__sinf
__tanf
__dadd_rd
__dadd_rn
__dadd_ru
__dadd_rz
__ddiv_rd
__ddiv_rn
__ddiv_ru
__ddiv_rz
__dmul_rd
__dmul_rn
__dmul_ru
__dmul_rz
__drcp_rd
__drcp_rn
__drcp_ru
__drcp_rz
__dsqrt_rd
__dsqrt_rn
__dsqrt_ru
__dsqrt_rz
__dsub_rd
__dsub_rn
__dsub_ru
__dsub_rz
__fma_rd
__fma_rn
__fma_ru
__fma_rz
__brev
__brevll
__byte_perm
__clz
__clzll
__ffs
__ffsll
__hadd
__mul24
__mul64hi
__mulhi
__popc
__popcll
__rhadd
__sad
__uhadd
__umul24
__umul64hi
__umulhi
__urhadd
__usad
__double2float_rd
__double2float_rn
__double2float_ru
__double2float_rz
__double2hiint
__double2int_rd
__double2int_rn
__double2int_ru
__double2int_rz
__double2ll_rd
__double2ll_rn
__double2ll_ru
__double2ll_rz
__double2loint
__double2uint_rd
__double2uint_rn
__double2uint_ru
__double2uint_rz
__double2ull_rd
__double2ull_rn
__double2ull_ru
__double2ull_rz
__double_as_longlong
__float2half_rn
__half2float
__float2half_rn
__half2float
__float2int_rd
__float2int_rn
__float2int_ru
__float2int_rz
__float2ll_rd
__float2ll_rn
__float2ll_ru
__float2ll_rz
__float2uint_rd
__float2uint_rn
__float2uint_ru
__float2uint_rz
__float2ull_rd
__float2ull_rn
__float2ull_ru
__float2ull_rz
__float_as_int
__float_as_uint
__hiloint2double
__int2double_rn
__int2float_rd
__int2float_rn
__int2float_ru
__int2float_rz
__int_as_float
__ll2double_rd
__ll2double_rn
__ll2double_ru
__ll2double_rz
__ll2float_rd
__ll2float_rn
__ll2float_ru
__ll2float_rz
__longlong_as_double
__uint2double_rn
__uint2float_rd
__uint2float_rn
__uint2float_ru
__uint2float_rz
__uint_as_float
__ull2double_rd
__ull2double_rn
__ull2double_ru
__ull2double_rz
__ull2float_rd
__ull2float_rn
__ull2float_ru
__ull2float_rz
__heq
__hge
__hgt
__hisinf
__hisnan
__hle
__hlt
__hne
__hbeq2
__hbge2
__hbgt2
__hble2
__hblt2
__hbne2
__heq2
__hge2
__hgt2
__hisnan2
__hle2
__hlt2
__hne2
__float22half2_rn
__float2half
__float2half2_rn
__float2half_rd
__float2half_rn
__float2half_ru
__float2half_rz
__floats2half2_rn
__half22float2
__half2float
half2half2
__half2int_rd
__half2int_rn
__half2int_ru
__half2int_rz
__half2ll_rd
__half2ll_rn
__half2ll_ru
__half2ll_rz
__half2short_rd
__half2short_rn
__half2short_ru
__half2short_rz
__half2uint_rd
__half2uint_rn
__half2uint_ru
__half2uint_rz
__half2ull_rd
__half2ull_rn
__half2ull_ru
__half2ull_rz
__half2ushort_rd
__half2ushort_rn
__half2ushort_ru
__half2ushort_rz
__half_as_short
__half_as_ushort
__halves2half2
__high2float
__high2half
__high2half2
__highs2half2
__int2half_rd
__int2half_rn
__int2half_ru
__int2half_rz
__ll2half_rd
__ll2half_rn
__ll2half_ru
__ll2half_rz
__low2float
__low2half
__low2half2
__low2half2
__lowhigh2highlow
__lows2half2
__short2half_rd
__short2half_rn
__short2half_ru
__short2half_rz
__uint2half_rd
__uint2half_rn
__uint2half_ru
__uint2half_rz
__ull2half_rd
__ull2half_rn
__ull2half_ru
__ull2half_rz
__ushort2half_rd
__ushort2half_rn
__ushort2half_ru
__ushort2half_rz
__ushort_as_half
"""
# The dictionary is to place description of each device function. Expand it to all the device functions
deviceFuncDesc = {'acosf': "This function returns floating point of arc cosine from a floating point input"}
fnames = ["../../include/hip/hcc_detail/math_functions.h","../../include/hip/hcc_detail/device_functions.h","../../include/hip/hcc_detail/hip_fp16.h"]
markdownFileName = "./hip-math-api.md"
preamble = "# HIP MATH APIs Documentation \n"+\
"HIP supports most of the device functions supported by CUDA. Way to find the unsupported one is to search for the function and check its description\n" + \
"Note: This document is not human generated. Any changes to this file will be discarded. Please make changes to Python3 script docs/markdown/device_md_gen.py\n\n" + \
"## For Developers \n" + \
"If you add or fixed a device function, make sure to add a signature of the function and definition later.\n" + \
"For example, if you want to add `__device__ float __dotf(float4, float4)`, which does a dot product on 4 float vector components \n" + \
"The way to add to the header is, \n" + \
"```cpp \n" + \
"__device__ static float __dotf(float4, float4); \n" + \
"/*Way down in the file....*/\n" + \
"__device__ static inline float __dotf(float4 x, float4 y) { \n" + \
" /*implementation*/\n}\n" + \
"```\n\n" + \
"This helps python script to add the device function newly declared into markdown documentation (as it looks at functions with `;` at the end and `__device__` at the beginning)\n\n" + \
"The next step would be to add Description to `deviceFuncDesc` dictionary in python script.\n" + \
"From the above example, it can be writtern as,\n`deviceFuncDesc['__dotf'] = 'This functions takes 2 4 component float vector and outputs dot product across them'`\n\n"
def generateSnippet(name, description, signature):
return "### " + name + "\n" + \
"```cpp \n" + signature + "\n```\n" + \
"**Description:** " + description + "\n\n\n"
def getName(line):
l1 = line.split('(')
l2 = l1[0].split(' ')
return l2[-1]
with open(markdownFileName, 'w') as mdfd:
mdfd.truncate()
mdfd.write(preamble)
for fname in fnames:
with open(fname) as fd:
lines = fd.readlines()
for line in lines:
if line.find('HIP_FAST_MATH') != -1:
break;
if line.find('__device__') != -1 and line.find(';') != -1 and line.find('hip') == -1:
name = getName(line)
if line.find('//') == -1:
if name in deviceFuncDesc:
mdfd.write(generateSnippet(name, deviceFuncDesc[name], line))
else:
mdfd.write(generateSnippet(name, "Supported", line))
else:
mdfd.write(generateSnippet(name, "**NOT Supported**", line))
fd.close()
mdfd.close()
projects/hip/docs/markdown/hip-math-api.md
+3838
Datei anzeigen
Datei-Diff unterdrückt, da er zu groß ist Diff laden
projects/hip/docs/markdown/hip_bugs.md
+221
Datei anzeigen
@@ -0,0 +1,221 @@
# HIP Bugs
<!-- toc -->
- [Errors related to undefined reference to `__hcLaunchKernel__***__grid_launch_parm**`](#errors-related-to-undefined-reference-to-__hclaunchkernel____grid_launch_parm)
- [What is the current limitation of HIP Generic Grid Launch method?](#what-is-the-current-limitation-of-hip-generic-grid-launch-method)
- [Errors related to `no matching constructor`](#errors-related-to-no-matching-constructor)
- [HIP is more restrictive in enforcing restrictions](#hip-is-more-restrictive-in-enforcing-restrictions)
<!-- tocstop -->
### Errors related to undefined reference to `__hcLaunchKernel__***__grid_launch_parm**`
Some common code practices may lead to hipcc generating a error with the form :
```
undefined reference to `__hcLaunchKernel__ZN15vecAddNamespace6vecAddIidEEv16grid_launch_parmPT0_S3_S3_T_
```
Or:
```
error: weak declaration cannot have internal linkage
```
Suggested workarounds:
- Avoid use of static with kernel definition:
```c++
static __global__ MyKernel
```
- Avoid defining kernels in anonymous namespace :
```c++
namespace {
__global__ MyKernel
}
```
### Can't find kernels inside dynamic linked library
HCC requires use of the "-Bdynamic" flag when creating a dynamic library which contains kernels. The dynamic flag causes the symbols to be created with a signature which allows HCC to discover and load the kernels in the dynamic library. This flag is often not set by default and must be added to the link step of the library. If not done, HCC will be unable to find the kernels defined in the library, and will emit a message such as:
```
HSADevice::CreateKernel(): Unable to create kernel"
```
To correct, add the following flag to hcc or hipcc:
```
$ hipcc -Wl,-Bsymbolic ...
```
### What is the current limitation of HIP Generic Grid Launch method?
1. __global__ functions cannot be marked as static or put in an unnamed namespace i.e. they cannot be given internal linkage (this would clash with __attribute__((weak)));
2. using the macro based dispatch mechanism i.e. hipLaunchKernel* only works for functions that take no more than 20 arguments (this limit can be increased up to 126, and is temporary until we can enable C++14 mode and use variadic generic lambdas); no such limitation applies do dispatching directly through grid_launch.
### Errors related to `no matching constructor`
The symptom is the compiler would complain about errors like `no matching constructor` for classes/structs passed as arguments into a GPU kernel. Often, this is caused by a design limitation in HCC where array-typed member variables inside a class/struct can’t be correctly passed into GPU kernels. To mitigate this issue, a custom serializer/deserializer pair is provided.
For example, `Foo` in the code snippets below contains an array-typed member variable `table`, which would fail the compiler if used as a kernel argument.
```
struct Foo {
float _data;
// table is an array, which makes foo
int table[3];
};
```
A workaround is to provide a custom serializer on host side which appends the contents of the array as kernel arguments, and a custome deserializaer on the device path to reconstruct the array inside the GPU kernels.
The deserializer can not be a function template, and should have scalar-typed parameters of the number equals to the length of the array-typed member variable. For example:
```
struct Foo {
float _data;
int _table[3];
#ifdef __HCC__
// user-provided CPU serializer
// Append the contents of the array member as kernel arguments
__attribute__((annotate(“serialize”)))
void __cxxamp_serialize(Kalmar::Serialize &s) const {
s.Append(sizeof(float), &_data);
for (int i = 0; i < 3; ++i)
s.Append(sizeof(int), &_table[i]);
}
// user-provided GPU deserializer
// table has 3 int elements, so deserializer must have 3 int parameters.
__attribute__((annotate(“user_deserialize”)))
Foo(float d, int x0, int x1, int x2) [[cpu]][[hc]] {
_data = d;
_table[0] = x0;
_table[1] = x1;
_table[2] = x2;
}
#endif
};
```
Rather than create serializer functions, another workaround is to pass the member fields from the structure as simple data types.
Note a class or struct can contain only one "user_deserialize" constructor.
For types which contain arrays which are based on template parameter, you can use partial template instantiation to implement one constructor per specialization.
However, an easier approach may be to create one user_deserializer which processes the maximum supported dimension.
This will take more memory in the structure and also require additional kernel arguments, but this may have little performance impact and the conversion is easier than partial template specialization. An example:
```
#define MAX_Dim 4
template<typename T, int Dim> struct MyArray {
T* dataPtr_;
//int size_[Dim]; // Original code with template-sized Dims
int size_[MAX_dim]; // Workaround code - allocate an array big enough for all dims so one serializer works.
...
#ifdef __HCC__
__attribute__((annotate("serialize")))
void __cxxamp_serialize(Kalmar::Serialize &s) const {
s.Append(sizeof(float), &_dataPtr);
for (int i=0; i<MAX_Dim; i++) {
s.Append(sizeof(size_[0]), &size_[i]);
}
}
__attribute__((annotate("user_deserialize")))
MyArray(T* data, int size0, int size1, int size2, int size3) [[cpu]][[hc]] {
data_ = data;
size_[0] = size0;
size_[1] = size1;
size_[2] = size2;
size_[3] = size3;
}
#endif
```
### HIP is more restrictive in enforcing restrictions
The language specification for HIP and CUDA forbid calling a
`__device__` function in a `__host__` context. In practice, you may observe
differences in the strictness of this restriction, with HIP exhibiting a tighter
adherence to the specification and thus less tolerant of infringing code. The
solution is to ensure that all functions which are called in a
`__device__` context are correctly annotated to reflect it. An interesting case
where these differences emerge is shown below. This relies on a the common
[C++ Member Detector idiom][1], as it would be implemented pre C++11):
```c++
#include <cassert>
#include <type_traits>
struct aye { bool a[1]; };
struct nay { bool a[2]; };
// Dual restriction is necessary in HIP if the detector is to work for
// __device__ contexts as well as __host__ ones. NVCC is less strict.
template<typename T>
__host__ __device__
const T& cref_t();
template<typename T>
struct Has_call_operator {
// Dual restriction is necessary in HIP if the detector is to work for
// __device__ contexts as well as __host__ ones. NVCC is less strict.
template<typename C>
__host__ __device__
static
aye test(
C const *,
typename std::enable_if<
(sizeof(cref_t<C>().operator()()) > 0)>::type* = nullptr);
static
nay test(...);
enum { value = sizeof(test(static_cast<T*>(0))) == sizeof(aye) };
};
template<typename T, typename U, bool callable = has_call_operator<U>::value>
struct Wrapper {
template<typename V>
V f() const { return T{1}; }
};
template<typename T, typename U>
struct Wrapper<T, U, true> {
template<typename V>
V f() const { return T{10}; }
};
// This specialisation will yield a compile-time error, if selected.
template<typename T, typename U>
struct Wrapper<T, U, false> {};
template<typename T>
struct Functor;
template<> struct Functor<float> {
__device__
float operator()() const { return 42.0f; }
};
__device__
void this_will_not_compile_if_detector_is_not_marked_device()
{
float f = Wrapper<float, Functor<float>>().f<float>();
}
__host__
void this_will_not_compile_if_detector_is_marked_device_only()
{
float f = Wrapper<float, Functor<float>>().f<float>();
}
```
[1]: https://en.wikibooks.org/wiki/More_C%2B%2B_Idioms/Member_Detector
projects/hip/docs/markdown/hip_faq.md
+39 -32
Datei anzeigen
@@ -4,13 +4,13 @@
- [What APIs and features does HIP support?](#what-apis-and-features-does-hip-support)
- [What is not supported?](#what-is-not-supported)
* [Run-time features](#run-time-features)
* [Runtime/Driver API features](#runtimedriver-api-features)
* [Kernel language features](#kernel-language-features)
- [Is HIP a drop-in replacement for CUDA?](#is-hip-a-drop-in-replacement-for-cuda)
- [What specific version of CUDA does HIP support?](#what-specific-version-of-cuda-does-hip-support)
- [What libraries does HIP support?](#what-libraries-does-hip-support)
- [How does HIP compare with OpenCL?](#how-does-hip-compare-with-opencl)
- [How does porting CUDA to HIP compare to porting CUDA to OpenCL?]
- [How does porting CUDA to HIP compare to porting CUDA to OpenCL?](#how-does-porting-cuda-to-hip-compare-to-porting-cuda-to-opencl)
- [What hardware does HIP support?](#what-hardware-does-hip-support)
- [Does Hipify automatically convert all source code?](#does-hipify-automatically-convert-all-source-code)
- [What is NVCC?](#what-is-nvcc)
@@ -23,9 +23,11 @@
- [On HCC, can I link HIP code with host code compiled with another compiler such as gcc, icc, or clang ?](#on-hcc-can-i-link-hip-code-with-host-code-compiled-with-another-compiler-such-as-gcc-icc-or-clang-)
- [HIP detected my platform (hcc vs nvcc) incorrectly - what should I do?](#hip-detected-my-platform-hcc-vs-nvcc-incorrectly---what-should-i-do)
- [Can I install both CUDA SDK and HCC on same machine?](#can-i-install-both-cuda-sdk-and-hcc-on-same-machine)
- [On CUDA, can I mix CUDA code with HIP code?](#on-cuda-can-i-mix-cuda-code-with-hip-code)
- [On HCC, can I use HC functionality with HIP?](#on-hcc-can-i-use-hc-functionality-with-hip)
- [How do I trace HIP application flow?](#how-do-i-trace-hip-application-flow)
* [Using CodeXL markers for HIP Functions](#using-codexl-markers-for-hip-functions)
* [Using HIP_TRACE_API](#using-hip_trace_api)
- [What if HIP generates error of "symbol multiply defined!" only on AMD machine?](#what-if-hip-generates-error-of-symbol-multiply-defined-only-on-amd-machine)
- [How do I disable HIP Generic Grid Launch option?](#how-do-i-disable-hip-generic-grid-launch-option)
<!-- tocstop -->
@@ -50,10 +52,9 @@ At a high-level, the following features are not supported:
- Textures
- Dynamic parallelism (CUDA 5.0)
- Managed memory (CUDA 6.5)
- Graphics interoperation with OpenGL or Direct3D
- Graphics interoperability with OpenGL or Direct3D
- CUDA Driver API (Under Development)
- CUDA IPC Functions (Under Development)
- CUDA array, mipmappedArray and pitched memory
- MemcpyToSymbol functions
- Queue priority controls
@@ -76,13 +77,13 @@ See the [API Support Table](CUDA_Runtime_API_functions_supported_by_HIP.md) for
### Is HIP a drop-in replacement for CUDA?
No. HIP provides porting tools which do most of the work do convert CUDA code into portable C++ code that uses the HIP APIs.
No. HIP provides porting tools which do most of the work to convert CUDA code into portable C++ code that uses the HIP APIs.
Most developers will port their code from CUDA to HIP and then maintain the HIP version.
HIP code provides the same performance as native CUDA code, plus the benefits of running on AMD platforms.
### What specific version of CUDA does HIP support?
HIP APIs and features do not map to a specific CUDA version. HIP provides a strong subset of functionality provided in CUDA, and the hipify tools can
scan code to identify any unsupported CUDA functions - this is very useful for identifying the specific features required by a given application.
HIP APIs and features do not map to a specific CUDA version. HIP provides a strong subset of functionality provided in CUDA, and the hipify tools can
scan code to identify any unsupported CUDA functions - this is useful for identifying the specific features required by a given application.
However, we can provide a rough summary of the features included in each CUDA SDK and the support level in HIP:
@@ -106,8 +107,8 @@ However, we can provide a rough summary of the features included in each CUDA SD
- TBD.
### What libraries does HIP support?
HIP includes growing support for the 4 key math libraries using hcBlas, hcFft, hcrng, and hcsparse).
These offer pointer-based memory interfaces (as opposed to opaque buffers) and can be easily interfaces with other HCC code. Developers should use conditional compliation if portability to nvcc systems is desired - using calls to cu* routines on one path and hc* routines on the other.
HIP includes growing support for the 4 key math libraries using hcBlas, hcFft, hcrng and hcsparse.
These offer pointer-based memory interfaces (as opposed to opaque buffers) and can be easily interfaced with other HCC applications. Developers should use conditional compilation if portability to nvcc systems is desired - using calls to cu* routines on one path and hc* routines on the other.
- [hcblas](https://bitbucket.org/multicoreware/hcblas)
- [hcfft](https://bitbucket.org/multicoreware/hcfft)
@@ -131,47 +132,47 @@ HIP offers several benefits over OpenCL:
### How does porting CUDA to HIP compare to porting CUDA to OpenCL?
Both HIP and CUDA are dialects of C++, and thus porting between them is relatively straightforward.
Both dialects support templates, classes, lambdas, and other C++ constructs.
As one example, the hipify tool was originally a perl script that used simple text conversions from CUDA to HIP.
HIP and CUDA provide similar math library calls as well. In summary, the HIP philospohy was to make the HIP language close enough to CUDA that the porting effort is relatively simple.
As one example, the hipify tool was originally a Perl script that used simple text conversions from CUDA to HIP.
HIP and CUDA provide similar math library calls as well. In summary, the HIP philosophy was to make the HIP language close enough to CUDA that the porting effort is relatively simple.
This reduces the potential for error, and also makes it easy to automate the translation. HIP's goal is to quickly get the ported program running on both platforms with little manual intervention,
so that the programmer can focus on performance optimizations.
There have been several tools that have attempted to convert CUDA into OpenCL, such as CU2CL. OpenCL is a C99-based kernel language (rather than C++) and also does not support single-source compilation.
As a result, the OpenCL syntax is quite different than CUDA, and the porting tools have to perform some heroic transformations to bridge this gap.
As a result, the OpenCL syntax is different from CUDA, and the porting tools have to perform some heroic transformations to bridge this gap.
The tools also struggle with more complex CUDA applications, in particular those that use templates, classes, or other C++ features inside the kernel.
### What hardware does HIP support?
- For AMD platforms, HIP runs on the same hardware that the HCC "hc" mode supports. See the ROCM documentation for the list of supported platforms.
- For Nvidia platforms, HIP requires Unified Memory and should run on a device which runs the CUDA SDK 6.0 or newer. We have tested the Nvidia Titan and K40.
- For AMD platforms, HIP runs on the same hardware that the HCC "hc" mode supports. See the ROCm documentation for the list of supported platforms.
- For Nvidia platforms, HIP requires Unified Memory and should run on any device supporting CUDA SDK 6.0 or newer. We have tested the Nvidia Titan and Tesla K40.
### Does Hipify automatically convert all source code?
Typically, Hipify can automatically convert almost all run-time code, and the coordinate indexing device code (i.e. threadIdx.x -> hipThreadIdx_x).
Typically, hipify can automatically convert almost all run-time code, and the coordinate indexing device code ( threadIdx.x -> hipThreadIdx_x ).
Most device code needs no additional conversion, since HIP and CUDA have similar names for math and built-in functions.
The hipify-clang tool will automatically modify the kernel signature as needed (automating a step that used to be done manually)
Additional porting may be required to deal with architecture feature queries or with CUDA capabilities that HIP doesn't support.
In general, developers should always expect to perform some platform-specific tuning and optimization.
### What is NVCC?
NVCC is Nvidia's compiler driver for compiling "CUDA C++" code into PTX or device code for Nvidia GPUs. It's a closed-source binary product that comes with CUDA SDKs.
NVCC is Nvidia's compiler driver for compiling "CUDA C++" code into PTX or device code for Nvidia GPUs. It's a closed-source binary compiler that is provided by the CUDA SDK.
### What is HCC?
HCC is AMD's compiler driver which compiles "heterogenous C++" code into HSAIL or GCN device code for AMD GPUs. It's an open-source compiler based on recent versions of CLANG/LLVM.
HCC is AMD's compiler driver which compiles "heterogeneous C++" code into HSAIL or GCN device code for AMD GPUs. It's an open-source compiler based on recent versions of CLANG/LLVM.
### Why use HIP rather than supporting CUDA directly?
While HIP is a strong subset of the CUDA, it is a subset. The HIP layer allows that subset to be clearly defined and documented.
Developers who code to the HIP API can be assured there code will remain portable across Nvidia and AMD platforms.
Developers who code to the HIP API can be assured their code will remain portable across Nvidia and AMD platforms.
In addition, HIP defines portable mechanisms to query architectural features, and supports a larger 64-bit wavesize which expands the return type for cross-lane functions like ballot and shuffle from 32-bit ints to 64-bit ints.
### Can I develop HIP code on an Nvidia CUDA platform?
Yes! HIP's CUDA path only exposes the APIs and functionality that work on both NVCC and HCC back-ends.
"Extra" APIs, parameters, and features which exist in CUDA but not in HCC will typically result in compile-time or run-time errors.
Developers need to use the HIP API for most accelerator code, and bracket any CUDA-specific code with appropriate ifdefs.
Yes. HIP's CUDA path only exposes the APIs and functionality that work on both NVCC and HCC back-ends.
"Extra" APIs, parameters, and features which exist in CUDA but not in HCC will typically result in compile- or run-time errors.
Developers need to use the HIP API for most accelerator code, and bracket any CUDA-specific code with preprocessor conditionals.
Developers concerned about portability should of course run on both platforms, and should expect to tune for performance.
In some cases CUDA has a richer set of modes for some APIs, and some C++ capabilities such as virtual functions - see the HIP @API documentation for more details.
### Can I develop HIP code on an AMD HCC platform?
Yes! HIP's HCC path only exposes the APIs and functions that work on both NVCC and HCC back ends. "Extra" APIs, parameters and features that appear in HCC but not CUDA will typically cause compile- or run-time errors. Developers must use the HIP API for most accelerator code and bracket any HCC-specific code with appropriate ifdefs. Those concerned about portability should, of course, test their code on both platforms and should tune it for performance. Typically, HCC supports a more modern set of C++11/C++14/C++17 features, so HIP developers who want portability should be careful when using advanced C++ features on the hc path.
Yes. HIP's HCC path only exposes the APIs and functions that work on both NVCC and HCC back ends. "Extra" APIs, parameters and features that appear in HCC but not CUDA will typically cause compile- or run-time errors. Developers must use the HIP API for most accelerator code and bracket any HCC-specific code with preprocessor conditionals. Those concerned about portability should, of course, test their code on both platforms and should tune it for performance. Typically, HCC supports a more modern set of C++11/C++14/C++17 features, so HIP developers who want portability should be careful when using advanced C++ features on the hc path.
### Can a HIP binary run on both AMD and Nvidia platforms?
HIP is a source-portable language that can be compiled to run on either the HCC or NVCC platform. HIP tools don't create a "fat binary" that can run on either platform, however.
@@ -184,9 +185,9 @@ A C++ dialect, hc is supported by the AMD HCC compiler. It provides C++ run time
### On HCC, can I link HIP code with host code compiled with another compiler such as gcc, icc, or clang ?
Yes! HIP/HCC generates the object code which conforms to the GCC ABI, and also links with libstdc++. This means you can compile host code with the compiler of your choice and link this
with GPU code compiler with HIP. Larger projects often contain a mixture of accelerator code (initially written in CUDA with nvcc) plus host code (compiled with gcc, icc, or clang). These projects
can convert the accelerator code to HIP, compile that code with hipcc, and link with object code from the preferred compiler.S
Yes. HIP/HCC generates the object code which conforms to the GCC ABI, and also links with libstdc++. This means you can compile host code with the compiler of your choice and link the generated object code
with GPU code compiled with HIP. Larger projects often contain a mixture of accelerator code (initially written in CUDA with nvcc) and host code (compiled with gcc, icc, or clang). These projects
can convert the accelerator code to HIP, compile that code with hipcc, and link with object code from their preferred compiler.
@@ -196,9 +197,9 @@ HIP will set the platform to HCC if it sees that the AMD graphics driver is inst
Sometimes this isn't what you want - you can force HIP to recognize the platform by setting HIP_PLATFORM to hcc (or nvcc)
```
export HIP_PLATFORM=hcc
```
One symptom of this problem is the message "error: 'unknown error'(11) at square.hipref.cpp:56". This can occur if you have a CUDA installation on an AMD platform, and HIP incorrectly detects the platform as nvcc. HIP may be able to compile the application using the nvcc tool-chain, but will generate this error at runtime since the platform does not have a CUDA device. The fix is to set HIP_PLATFORM=hcc and rebuild the issue.
One symptom of this problem is the message "error: 'unknown error'(11) at square.hipref.cpp:56". This can occur if you have a CUDA installation on an AMD platform, and HIP incorrectly detects the platform as nvcc. HIP may be able to compile the application using the nvcc tool-chain, but will generate this error at runtime since the platform does not have a CUDA device. The fix is to set HIP_PLATFORM=hcc and rebuild.
If you see issues related to incorrect platform detection, please file an issue with the GitHub issue tracker so we can improve HIP's platform detection logic.
@@ -207,7 +208,7 @@ Yes. You can use HIP_PLATFORM to choose which path hipcc targets. This configur
### On CUDA, can I mix CUDA code with HIP code?
Yes. Most HIP data structures (hipStream_t, hipEvent_t) are typedefs to CUDA equivalents and can be intermixed. Both CUDA and HIP use integer device ids .
Yes. Most HIP data structures (hipStream_t, hipEvent_t) are typedefs to CUDA equivalents and can be intermixed. Both CUDA and HIP use integer device ids.
One notable exception is that hipError_t is a new type, and cannot be used where a cudaError_t is expected. In these cases, refactor the code to remove the expectation. Alternatively, hip_runtime_api.h defines functions which convert between the error code spaces:
hipErrorToCudaError
@@ -218,10 +219,10 @@ If platform portability is important, use #ifdef __HIP_PLATFORM_NVCC__ to guard
### On HCC, can I use HC functionality with HIP?
Yes.
The code can include hc.hpp and use HC functions inside the kernel. A typical use case is to use AMD-specific hardware features such as the permute, swizzle, or DPP operations.
The code can include hc.hpp and use HC functions inside the kernel. A typical use-case is to use AMD-specific hardware features such as the permute, swizzle, or DPP operations.
The "-stdlib=libc++" must be passed to hipcc in order to compile hc.hpp. See the 'bit_extract' sample for an example.
Also these functions can be used to extract HCC acclerator and accelerator_view structures from the HIP deviceId and hipStream_t:
Also these functions can be used to extract HCC accelerator and accelerator_view structures from the HIP deviceId and hipStream_t:
hipHccGetAccelerator(int deviceId, hc::accelerator *acc);
hipError_t hipHccGetAcceleratorView(hipStream_t stream, hc::accelerator_view **av);
@@ -236,3 +237,9 @@ Unlike CUDA, in HCC, for functions defined in the header files, the keyword of "
Thus, if failed to define "static" keyword, you might see a lot of "symbol multiply defined!" errors at compilation.
The workaround is to explicitly add the keyword of "static" before any functions that were defined as "__forceinline__".
### How do I disable HIP Generic Grid Launch option?
Generic Grid Launch(GGL) is currently the default method for hip kernel launch.
To disable it and use the legancy grid launch method, please either change the default value of GENERIC_GRID_LAUNCH to 0 in the following to header files and rebuild HIP:
$HIP/include/hip/hcc_detail/hip_runtime_api.h
$HIP/include/hip/hcc_detail/host_defines.h
Or pass "-DGENERIC_GRID_LAUNCH=0" to hipcc at application compilation time.
projects/hip/docs/markdown/hip_kernel_language.md
+3 -2
Datei anzeigen
@@ -44,6 +44,7 @@
- [Pragma Unroll](#pragma-unroll)
- [In-Line Assembly](#in-line-assembly)
- [C++ Support](#c-support)
- [Kernel Compilation](#kernel-compilation)
<!-- tocstop -->
@@ -105,7 +106,7 @@ HIP parses the `__noinline__` and `__forceinline__` keywords and converts them t
```
// Example psuedocode introducing hipLaunchKernel:
// Example pseudo code introducing hipLaunchKernel:
__global__ MyKernel(hipLaunchParm lp, float *A, float *B, float *C, size_t N)
{
...
@@ -680,7 +681,7 @@ The user can specify the target for which the binary can be generated. HIP/HCC d
The file format for binary is `.co` which means Code Object. The following command builds the code object using `hipcc`.
`hipcc --genco --target-isa=[TARGET GPU] [INPUT FILE] -o [OUTPUT FILE]`
```[TARGET GPU] = fiji/hawaii
```[TARGET GPU] = gfx803/gfx701
[INPUT FILE] = Name of the file containing kernels
[OUTPUT FILE] = Name of the generated code object file```
projects/hip/docs/markdown/hip_performance.md
+3 -3
Datei anzeigen
@@ -6,18 +6,18 @@ Please note that this document lists possible ways for experimenting with HIP st
#### On Small BAR Setup
There are two possible ways to transfer data from Host to Device (H2D) and Device to Host(D2H)
There are two possible ways to transfer data from host-to-device (H2D) and device-to-host(D2H)
* Using Staging Buffers
* Using PinInPlace
#### On Large BAR Setup
There are three possible ways to transfer data from Host to Device (H2D)
There are three possible ways to transfer data from host-to-device (H2D)
* Using Staging Buffers
* Using PinInPlace
* Direct Memcpy
And there are two possible ways to transfer data from Device to Host (D2H)
And there are two possible ways to transfer data from device-to-host (D2H)
* Using Staging Buffers
* Using PinInPlace
projects/hip/docs/markdown/hip_porting_driver_api.md
+58 -91
Datei anzeigen
@@ -1,151 +1,119 @@
# Porting CUDA Driver API
## Introduction to the CUDA Driver and Runtime APIs
CUDA provides a separate CUDA Driver and Runtime APIs. The two APis have significant overlap in functionality:
CUDA provides a separate CUDA Driver and Runtime APIs. The two APIs have significant overlap in functionality:
- Both APIs support events, streams, memory management, memory copy, and error handling.
- Both APIs deliver similar performance.
- Driver APIs calls begin with the prefix `cu` while Runtime APIs begin with the prefix `cuda`. For example, the Driver API API contains `cuEventCreate` while the Runtime API contains `cudaEventCreate`, with similar functionality.
- The Driver API defines a different but largely overlapping error code code space than the Runtime API, and uses a different coding convention. For example, Driver API defines `CUDA_ERROR_INVALID_VALUE` while the Runtime API defines `cudaErrorInvalidValue`
- Both APIs deliver similar performance.
- Driver APIs calls begin with the prefix `cu` while Runtime APIs begin with the prefix `cuda`. For example, the Driver API API contains `cuEventCreate` while the Runtime API contains `cudaEventCreate`, with similar functionality.
- The Driver API defines a different but largely overlapping error code space than the Runtime API, and uses a different coding convention. For example, Driver API defines `CUDA_ERROR_INVALID_VALUE` while the Runtime API defines `cudaErrorInvalidValue`
The Driver API offers two additional pieces of functionality not provided by the Runtime API: cuModule and cuCtx APIs.
### cuModule API
The Module section of the Driver API provides additional control over how and when accelerator
code objects are loaded.
The Module section of the Driver API provides additional control over how and when accelerator code objects are loaded.
For example, the driver API allows code objects to be loaded from files or memory pointers.
Symbols for kernels or global data can be extracted from the loaded code objects.
In contrast, the Runtime API automatically loads and (if necessary) compiles all of the
kernels from a executable binary when run.
In contrast, the Runtime API automatically loads and (if necessary) compiles all of the kernels from an executable binary when run.
In this mode, NVCC must be used to compile kernel code so the automatic loading can function correctly.
Both Driver and Runtime APIs define a function for launching kernels (called `cuLaunchKernel` or `cudaLaunchKernel`.
The kernel arguments and the execution configuration (grid dimensions, group dimensions, dynamic shared memory, and stream) are passed as arguments to the launch function.
The Runtime additionally provides the `<<< >>>` syntax for launching kernels, which resembles
a special function call and is easier to use than explicit launch API (in particular with respect to
handling of kernel arguments).
However, this syntax is not standard C++ and is available only when NVCC is used to compile the host code.
The Runtime additionally provides the `<<< >>>` syntax for launching kernels, which resembles a special function call and is easier to use than explicit launch API (in particular with respect to handling of kernel arguments).
However, this syntax is not standard C++ and is available only when NVCC is used to compile the host code.
The Module features are useful in an environment which generate the code objects directly, such as a new
accelerator language front-end. Here, NVCC is not used. Instead, the environment may have a
different kernel language or different compilation flow.
The Module features are useful in an environment which generates the code objects directly, such as a new accelerator language front-end.
Here, NVCC is not used. Instead, the environment may have a different kernel language or different compilation flow.
Other environments have many kernels and do not want them to be all loaded automatically.
The Module functions can be used to load the generated code objects and launch kernels.
As we will see below, HIP defines a Module API which provides similar explict control over code
object managemenet.
As we will see below, HIP defines a Module API which provides similar explicit control over code object management.
### cuCtx API
The Driver API defines "Context" and "Devices" as separate entities.
Contexts contain a single device, and a device can theoretically have multiple contexts.
Each context contains a set of streams and events specific to the context.
Historically contexts also defined a unique address space for the GPU, though this may not longer be the case in Unified Memory platforms (since the CPU and all the devices in the same process share a single unified address space).
The Context APIs also provide a mechanism to switch between devices, which allowed
a single CPU thread to send commands to different GPUs. HIP as well as a recent versions
of CUDA Runtime provide other mechanisms to accomplish this feat - for example using streams or
`cudaSetDevice`.
The Driver API defines "Context" and "Devices" as separate entities.
Contexts contain a single device, and a device can theoretically have multiple contexts.
Each context contains a set of streams and events specific to the context.
Historically contexts also defined a unique address space for the GPU, though this may no longer be the case in Unified Memory platforms (since the CPU and all the devices in the same process share a single unified address space).
The Context APIs also provide a mechanism to switch between devices, which allowed a single CPU thread to send commands to different GPUs.
HIP as well as a recent versions of CUDA Runtime provide other mechanisms to accomplish this feat - for example using streams or `cudaSetDevice`.
The CUDA Runtime API unifies the Context API with the Device API. This simplifies the APIs
and has little loss of functionality since each Context can contain a single device,
and the benefits of multiple contexts has been replaced with other interfaces. HIP provides
a context API to facilitate easy porting from existing Driver codes.
In HIP, the Ctx functions largely provide an alternate syntax for changing the active device.
The CUDA Runtime API unifies the Context API with the Device API. This simplifies the APIs and has little loss of functionality since each Context can contain a single device, and the benefits of multiple contexts has been replaced with other interfaces.
HIP provides a context API to facilitate easy porting from existing Driver codes.
In HIP, the Ctx functions largely provide an alternate syntax for changing the active device.
Most new applications will prefer to use `hipSetDevice` or the stream APIs.
## HIP Module and Ctx APIs
Rather than present two separate APIs, HIP extends the HIP API with new APIs for Modules
and Ctx control.
Rather than present two separate APIs, HIP extends the HIP API with new APIs for Modules and Ctx control.
### hipModule API
Like the CUDA Driver API, the Module API provides additional control
over how code is loaded, including options to load code from files or from in-memory pointers.
NVCC and HCC target different architectures and use different code object formats : NVCC
is `cubin` or `ptx` files, while the HCC path is the `hsaco` format. The external
compilers which generate these code objects are responsible for generating and loading
the correct code object for each platform. Notably, there is not a fat binary format that
can contain code for both NVCC and HCC platforms. The following table summarizes the
formats used on each platform:
Like the CUDA Driver API, the Module API provides additional control over how code is loaded, including options to load code from files or from in-memory pointers.
NVCC and HCC target different architectures and use different code object formats: NVCC is `cubin` or `ptx` files, while the HCC path is the `hsaco` format.
The external compilers which generate these code objects are responsible for generating and loading the correct code object for each platform.
Notably, there is not a fat binary format that can contain code for both NVCC and HCC platforms. The following table summarizes the formats used on each platform:
| Format | APIs | NVCC | HCC |
| --- | --- | --- | --- |
| Format | APIs | NVCC | HCC |
| --- | --- | --- | --- |
| Code Object | hipModuleLoad, hipModuleLoadData | .cubin or PTX text | .hsaco |
| Fat Binary | hipModuleLoadFatBin | .fatbin | Under Development |
| Fat Binary | hipModuleLoadFatBin | .fatbin | Under Development |
hipcc uses NVCC and HCC to compile host codes. Both of these may embed code objects
into the final executable, and these code objects will be automatically loaded when
the application starts.
The hipModule API can be used to load additional code objects, and in this way
provides an extended capability to the automatically loaded code objects. HCC allows
both of these capabilities to be used together, if desired. Of course it is possible
to create a program with no kernels and thus no automatic loading.
`hipcc` uses NVCC and HCC to compile host codes. Both of these may embed code objects into the final executable, and these code objects will be automatically loaded when the application starts.
The hipModule API can be used to load additional code objects, and in this way provides an extended capability to the automatically loaded code objects.
HCC allows both of these capabilities to be used together, if desired. Of course it is possible to create a program with no kernels and thus no automatic loading.
### hipCtx API
HIP provides a `Ctx` API as a thin layer over the existing Device functions. This Ctx API
can be used to set the current context, or to query properties of the device associated with
the context. The current context is implicitly used by other APIs such as `hipStreamCreate`.
HIP provides a `Ctx` API as a thin layer over the existing Device functions. This Ctx API can be used to set the current context, or to query properties of the device associated with the context.
The current context is implicitly used by other APIs such as `hipStreamCreate`.
### hipify translation of CUDA Driver API
The hipify tool will convert CUDA Driver APIs for streams, events, memory management to
the equivalent HIP driver calls. For example, `cuEventCreate` will be translated to
`hipEventCreate`. Hipify also converts error code from the Driver namespace and coding
convention to the equivalent HIP error code. Thus, HIP unifies the APis for these common functions.
[hipify support for translating driver API is Under Development]
The hipify tool converts CUDA Driver APIs for streams, events, modules, devices, memory management, context, profiler to the equivalent HIP driver calls. For example, `cuEventCreate` will be translated to `hipEventCreate`.
Hipify also converts error code from the Driver namespace and coding convention to the equivalent HIP error code. Thus, HIP unifies the APIs for these common functions.
The memory copy APIs require additional explanation. The CUDA driver includes the memory
direction in the name of the API (ie `cuMemcpyH2D`) while the CUDA driver API provides
a single memory copy API with a parameter that specifies the direction and additionally
supports a "default" direction where the runtime determines the direction automatically.
HIP provides APis with both styles: for example, `hipMemcpyH2D` as well as `hipMemcpy`.
The first flavor may be faster in some cases since they avoid host overhead to detect the
different memory directions.
The memory copy API requires additional explanation. The CUDA driver includes the memory direction in the name of the API (ie `cuMemcpyH2D`) while the CUDA driver API provides a single memory copy API with a parameter that specifies the direction and additionally supports a "default" direction where the runtime determines the direction automatically.
HIP provides APIs with both styles: for example, `hipMemcpyH2D` as well as `hipMemcpy`.
The first flavor may be faster in some cases since they avoid host overhead to detect the different memory directions.
HIP defines a single error space, and uses camel-case for all errors (i.e. `hipErrorInvalidValue`).
HIP defines a single error space, and uses camel-case for all errors (i.e. `hipErrorInvalidValue`).
### HCC Implementation Notes
#### .hsaco
The .hsaco format used by HCC is described in more detail [here](https://github.com/RadeonOpenCompute/ROCm-ComputeABI-Doc).
An example and blog that show how to use the format is [here](http://gpuopen.com/rocm-with-harmony-combining-opencl-hcc-hsa-in-a-single-program). hsaco can be generated by hcc + extractkernel tool,
cloc, the GCN assembler, or other tools.
An example and blog that show how to use the format is [here](http://gpuopen.com/rocm-with-harmony-combining-opencl-hcc-hsa-in-a-single-program). hsaco can be generated by hcc + extractkernel tool, cloc, the GCN assembler, or other tools.
#### Address Spaces
HCC defines a process-wide address space where the CPU and all devices allocate
addresses from a single unified pool. Thus addresses may be shared between contexts, and
unlike the original CUDA definition a new context does not create a new address space for
the device.
HCC defines a process-wide address space where the CPU and all devices allocate addresses from a single unified pool.
Thus addresses may be shared between contexts, and unlike the original CUDA definition a new context does not create a new address space for the device.
#### Using hipModuleLaunchKernel
`hipModuleLaunchKernel` is `cuLaunchKernel` in HIP world. It takes the same arguments as `cuLaunchKernel`. The argument `kernelParams` is not fully implemented for HCC. The workaround for it is, to use platform specific macros for each target. Or, `extra` argument can be used which works on both the platforms.
#### Additional Information
- HCC allocates staging buffers (used for unpinned copies) on a per-device basis.
- HCC creates a primary context when the HIP API is called. So in a pure driver API code, HIP/HCC will create a primary context while HIP/NVCC will have empty context stack. HIP/HCC will push primary context to context stack when it is empty. This can have subtle differences on applications which mix the runtime and driver APIs.
- HCC creates a primary context when the HIP API is called. So in a pure driver API code, HIP/HCC will create a primary context while HIP/NVCC will have empty context stack.
HIP/HCC will push primary context to context stack when it is empty. This can have subtle differences on applications which mix the runtime and driver APIs.
### NVCC Implementation Notes
#### Interoperation between HIP and CUDA Driver
CUDA applications may want to mix CUDA driver code with HIP code (see example below). This
table shows the type equivalence to enable this interaction.
|**HIP Type** |**CU Driver Type**| **CUDA Runtime Type** |
| ---- | ---- | ---- |
| hipModule | CUmodule | |
| hipFunction | CUfunction | |
| hipCtx_t | CUcontext | |
| hipDevice_t | CUdevice | |
| hipStream_t | CUstream | cudaStream_t |
| hipEvent_t | CUevent | cudaEvent_t |
| hipArray | CUarray | cudaArray |
CUDA applications may want to mix CUDA driver code with HIP code (see example below). This table shows the type equivalence to enable this interaction.
|**HIP Type** |**CU Driver Type**|**CUDA Runtime Type**|
| ---- | ---- | ---- |
| hipModule | CUmodule | |
| hipFunction | CUfunction | |
| hipCtx_t | CUcontext | |
| hipDevice_t | CUdevice | |
| hipStream_t | CUstream | cudaStream_t |
| hipEvent_t | CUevent | cudaEvent_t |
| hipArray | CUarray | cudaArray |
#### Compilation Flags
The hipModule interface does not support the hipModuleLoadEx function, which is used to control PTX compilaton options.
The hipModule interface does not support the `cuModuleLoadDataEx` function, which is used to control PTX compilation options.
HCC does not use PTX and does not support the same compilation options.
In fact, HCC code objects always contain fully compiled ISA and do not require additional compilation as part of the load step.
Code which requires this functionaly should use platform-specific coding, calling `cuModuleLoadEx`
on the NVCC path and hipModuleLoad on the hcc path. For example:
Code which requires this functionally should use platform-specific coding, calling `cuModuleLoadDataEx` on the NVCC path and `hipModuleLoadData` on the hcc path.
For example:
```
hipModule module;
@@ -240,7 +208,7 @@ int main(){
HIP_LAUNCH_PARAM_END
};
hipModuleLaunchKernel(Function, 1, 1, 1, LEN, 1, 1, 0, 0, NULL, (void**)&con fig);
hipModuleLaunchKernel(Function, 1, 1, 1, LEN, 1, 1, 0, 0, NULL, (void**)&config);
hipMemcpyDtoH(B, Bd, SIZE);
for(uint32_t i=0;i<LEN;i++){
@@ -253,5 +221,4 @@ int main(){
return 0;
}
```
projects/hip/docs/markdown/hip_porting_guide.md
+12 -9
Datei anzeigen
@@ -1,5 +1,5 @@
# HIP Porting Guide
In addition to providing a portable C++ programmming environement for GPUs, HIP is designed to ease
In addition to providing a portable C++ programming environment for GPUs, HIP is designed to ease
the porting of existing CUDA code into the HIP environment. This section describes the available tools
and provides practical suggestions on how to port CUDA code and work through common issues.
@@ -21,6 +21,7 @@ and provides practical suggestions on how to port CUDA code and work through com
* [Device-Architecture Properties](#device-architecture-properties)
* [Table of Architecture Properties](#table-of-architecture-properties)
- [Finding HIP](#finding-hip)
- [hipLaunchKernel](#hiplaunchkernel)
- [Compiler Options](#compiler-options)
- [Linking Issues](#linking-issues)
* [Linking With hipcc](#linking-with-hipcc)
@@ -31,9 +32,11 @@ and provides practical suggestions on how to port CUDA code and work through com
* [Using a Standard C++ Compiler](#using-a-standard-c-compiler)
+ [cuda.h](#cudah)
* [Choosing HIP File Extensions](#choosing-hip-file-extensions)
* [Workarounds](#workarounds)
+ [warpSize](#warpsize)
+ [Textures and Cache Control](#textures-and-cache-control)
- [Workarounds](#workarounds)
* [warpSize](#warpsize)
- [memcpyToSymbol](#memcpytosymbol)
- [threadfence_system](#threadfence_system)
* [Textures and Cache Control](#textures-and-cache-control)
- [More Tips](#more-tips)
* [HIPTRACE Mode](#hiptrace-mode)
* [Environment Variables](#environment-variables)
@@ -166,10 +169,10 @@ Both nvcc and hcc make two passes over the code: one for host code and one for d
```
// #ifdef __CUDA_ARCH__
#ifdef __HIP_DEVICE_COMPILE__
#if __HIP_DEVICE_COMPILE__
```
Unlike `__CUDA_ARCH__`, the `__HIP_DEVICE_COMPILE__` value is 0 or 1, and it doesn’t represent the feature capability of the target device.
Unlike `__CUDA_ARCH__`, the `__HIP_DEVICE_COMPILE__` value is 1 or undefined, and it doesn’t represent the feature capability of the target device.
### Compiler Defines: Summary
@@ -178,11 +181,11 @@ Unlike `__CUDA_ARCH__`, the `__HIP_DEVICE_COMPILE__` value is 0 or 1, and it doe
|HIP-related defines:|
|`__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
|`__HIP_DEVICE_COMPILE__` | 1 if compiling for device; undefined if compiling for host |1 if compiling for device; undefined 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
|`__CUDACC__` | Undefined | Defined if source code is compiled by nvcc; 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:|
@@ -553,7 +556,7 @@ If you pass a ".cu" file, hcc will attempt to compile it as a Cuda language file
### HIP Environment Variables
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),
On the HCC path, HIP provides a number of environment variables that control the behavior of HIP. Some of these are useful for application 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:
projects/hip/docs/markdown/hip_profiling.md
+120 -27
Datei anzeigen
@@ -4,26 +4,32 @@ This section describes the profiling and debugging capabilities that HIP provide
Profiling information can viewed in the CodeXL visualization tool or printed directly to stderr as the application runs.
This document starts with some of the general capabilities of CodeXL and then describes some of the additional HIP marker and debug features.
* [CodeXL Profiling](#codexl-profiling)
* [Collecting and Viewing Traces](#collecting-and-viewing-traces)
* [Using rocm-profiler timestamp profiling](#using-rocm-profiler-timestamp-profiling)
* [Using rocm-profiler performance counter collection:](#using-rocm-profiler-performance-counter-collection)
* [Using CodeXL to view profiling results:](#using-codexl-to-view-profiling-results)
* [More information on CodeXL](#more-information-on-codexl)
* [HIP Markers](#hip-markers)
* [Profiling HIP APIs](#profiling-hip-apis)
* [Adding markers to applications](#adding-markers-to-applications)
* [Additional HIP Profiling Features](#additional-hip-profiling-features)
* [Demangling C Kernel Names](#demangling-c-kernel-names)
* [Controlling when profiling starts and ends](#controlling-when-profiling-starts-and-ends)
* [Reducing timeline trace output file size](#reducing-timeline-trace-output-file-size)
* [How to enable profiling at HIP build time](#how-to-enable-profiling-at-hip-build-time)
* [Tracing and Debug](#tracing-and-debug)
* [Tracing HIP APIs](#tracing-hip-apis)
* [Color](#color)
* [Using HIP_DB](#using-hip_db)
* [Using ltrace](#using-ltrace)
* [Chicken bits](#chicken-bits)
<!-- toc -->
- [CodeXL Profiling](#codexl-profiling)
* [Collecting and Viewing Traces](#collecting-and-viewing-traces)
+ [Using rocm-profiler timestamp profiling](#using-rocm-profiler-timestamp-profiling)
+ [Using rocm-profiler performance counter collection:](#using-rocm-profiler-performance-counter-collection)
+ [Using CodeXL to view profiling results:](#using-codexl-to-view-profiling-results)
+ [More information on CodeXL](#more-information-on-codexl)
* [HIP Markers](#hip-markers)
+ [Profiling HIP APIs](#profiling-hip-apis)
+ [Adding markers to applications](#adding-markers-to-applications)
* [Additional HIP Profiling Features](#additional-hip-profiling-features)
+ [Demangling C++ Kernel Names](#demangling-c-kernel-names)
+ [Controlling when profiling starts and ends](#controlling-when-profiling-starts-and-ends)
+ [Reducing timeline trace output file size](#reducing-timeline-trace-output-file-size)
+ [How to enable profiling at HIP build time](#how-to-enable-profiling-at-hip-build-time)
- [Tracing and Debug](#tracing-and-debug)
* [Tracing HIP APIs](#tracing-hip-apis)
+ [Color](#color)
* [Using HIP_DB](#using-hip_db)
* [Using ltrace](#using-ltrace)
* [Chicken bits](#chicken-bits)
* [Debugging HIP Applications](#debugging-hip-applications)
* [General Debugging Tips](#general-debugging-tips)
<!-- tocstop -->
## CodeXL Profiling
@@ -92,11 +98,11 @@ HIP_PROFILE_API supports two levels of information.
#### Adding markers to applications
Markers can be used to define application-specific events that will be recorded in the ATP file and displayed in the CodeXL gui.
Markers can be used to define application-specific events that will be recorded in the ATP file and displayed in the CodeXL GUI.
This can be particularly useful for visualizing how the higher-level phases of application behavior relate to the lower level HIP APIs, kernel launches, and data transfers.
For example, an instrumented machine learning framework could show the beginning and ending of each layer in the network.
Markers have a specific begin and end time, and can be nested. Nested calls are displayed hierarchically in the CodeXL gui, with each level of the hierarchy occupying a different row.
Markers have a specific begin and end time, and can be nested. Nested calls are displayed hierarchically in the CodeXL GUI, with each level of the hierarchy occupying a different row.
The HIP APis are defined in "hip_profile.h":
```
@@ -131,7 +137,7 @@ The HIP marker API is only supported on ROCm platform. The marker macros are de
This [HIP sample](samples/2_Cookbook/2_Profiler/) shows the profiler marker API used in a small application.
More information on the marker API can be found in the profiler header file and PDF in a ROCM installation:
More information on the marker API can be found in the profiler header file and PDF in a ROCm installation:
- /opt/rocm/profiler/CXLActivityLogger/include/CXLActivityLogger.h
- /opt/rocm/profiler/CXLActivityLogger/doc/CXLActivityLogger.pdf
@@ -185,7 +191,7 @@ $ nvprof --profile-from-start-off ...
This feature is under development.
#### Reducing timeline trace output file size
If the application is already recording the HIP APIs, the HSA APIs are somewhat redundant and the ATP file size can be substantially reduced by not recording these APIs. HIP includes a text file that lists all of the HSA APis and can assist in this filtering:
If the application is already recording the HIP APIs, the HSA APIs are somewhat redundant and the ATP file size can be substantially reduced by not recording these APIs. HIP includes a text file that lists all of the HSA APIs and can assist in this filtering:
```
$ rocm-profiler -F hip/bin/hsa-api-filter-cxl.txt
@@ -273,7 +279,7 @@ None will disable use of color control codes for both the opening and closing an
This flag is primarily targeted to assist HIP development team in the development of the HIP runtime, but in some situations may be useful to HIP application developers as well.
The HIP debug information is designed to print important information during the execution of a HIP API. HIP provides
different color-coded levels of debug informaton:
different color-coded levels of debug information:
- api : Print the beginning and end of each HIP API, including the arguments and return codes. This is equivalent to setting HIP_TRACE_API=1.
- sync : Print multi-thread and other synchronization debug information.
- copy : Print which engine is doing the copy, which copy flavor is selected, information on source and destination memory.
@@ -315,7 +321,7 @@ libhsa-runtime64.so.1->hsaKmtDeregisterMemory(0x7f7776d3e010, 0x7f7776d3e010, 0x
```
Some key information from the trace above.
- The trace snippet shows the execution of a hipMemcpy API, bracketed by the first and last message in the trace output. The messages show the thread id and API sequence number (`1.17`). ltrace output intermixes messages from all threads, so the HIP debug information can be useful to determine which threads are executing.
- Thy trace snippet shows the execution of a hipMemcpy API, bracketed by the first and last message in the trace output. The messages show the thread id and API sequence number (`1.17`). ltrace output intermixes messages from all threads, so the HIP debug information can be useful to determine which threads are executing.
- The code flows through HIP APIs into ROCr (HSA) APIs (hsa*) and into the thunk (hsaKmt*) calls.
- The HCC runtime is "libmcwamp_hsa.so" and the HSA/ROCr runtime is "libhsa-runtime64.so".
- In this particular case, the memory copy is for unpinned memory, and the selected copy algorithm is to pin the host memory "in-place" before performing the copy. The signaling APIs and calls to pin ("lock", "register") the memory are readily apparent in the trace output.
@@ -335,8 +341,95 @@ HIP provides 3 environment variables in the HIP_*_BLOCKING family. These introd
These options cause HCC to serialize. Useful if you have libraries or code which is calling HCC kernels directly rather than using HIP.
- HCC_SERIALZIE_KERNELS : 0x1=pre-serialize before each kernel launch, 0x2=post-serialize after each kernel launch., 0x3= pre- and post- serialize.
- HCC_SERIALIZE_COPY : 0x1=pre-serialize before each async copy, 0x2=post-serialize after each async copy., 0x3= pre- and post- serialize.0
- HCC_SERIALIZE_COPY : 0x1=pre-serialize before each async copy, 0x2=post-serialize after each async copy., 0x3= pre- and post- serialize.
- HSA_ENABLE_SDMA=0 : Causes host-to-device and device-to-host copies to use compute shader blit kernels rather than the dedicated DMA copy engines. Compute shader copies have low latency (typically < 5us) and can achieve approximately 80% of the bandwidth of the DMA copy engine. This flag is useful to isolate issues with the hardware copy engines.
- HSA_ENABLE_INTERRUPT=0 : Causes completion signals to be detected with memory-based polling rather than interrupts. Can be useful to diagnose interrupt storm issues in the driver.
- HSA_DISABLE_CACHE=1 : Disables the GPU L2 data cache.
### Debugging HIP Applications
- The variable "tls_tidInfo" contains the API sequence number (_apiSeqNum)- a monotonically increasing count of the HIP APIs called from this thread. This can be useful for setting conditional breakpoints. Also, each new HIP thread is mapped to monotically increasing shortTid ID. Both of these fields are displayed in the HIP debug info.
```
(gdb) p tls_tidInfo
$32 = {_shortTid = 1, _apiSeqNum = 803}
```
- HCC tracks all of the application memory allocations, including those from HIP and HC's "am_alloc".
If the HCC runtime is built with debug information (HCC_RUNTIME_DEBUG=ON when building HCC), then calling the function 'hc::am_memtracker_print()' will show all memory allocations.
An optional argument specifies a void * targetPointer - the print routine will mark the allocation which contains the specified pointer with "-->" in the printed output.
This example shows a sample GDB session where we print the memory allocated by this process and mark a specified address by using the gdb "call" function..
The gdb syntax also supports using the variable name (in this case 'dst'):
```
(gdb) p dst
$33 = (void *) 0x5ec7e9000
(gdb) call hc::am_memtracker_print(dst)
TargetAddress:0x5ec7e9000
0x504cfc000-0x504cfc00f:: allocSeqNum:1 hostPointer:0x504cfc000 devicePointer:0x504cfc000 sizeBytes:16 isInDeviceMem:0 isAmManaged:1 appId:0 appAllocFlags:0 appPtr:(nil)
...
-->0x5ec7e9000-0x5f7e28fff:: allocSeqNum:488 hostPointer:(nil) devicePointer:0x5ec7e9000 sizeBytes:191102976 isInDeviceMem:1 isAmManaged:1 appId:0 appAllocFlags:0 appPtr:(nil)
```
To debug an explicit address, cast the address to (void*) :
```
(gdb) call hc::am_memtracker_print((void*)0x508c7f000)
```
- Debugging GPUVM fault.
For example:
```
Memory access fault by GPU node-1 on address 0x5924000. Reason: Page not present or supervisor privilege.
Program received signal SIGABRT, Aborted.
[Switching to Thread 0x7fffdffb5700 (LWP 14893)]
0x00007ffff2057c37 in __GI_raise (sig=sig@entry=6) at ../nptl/sysdeps/unix/sysv/linux/raise.c:56
56 ../nptl/sysdeps/unix/sysv/linux/raise.c: No such file or directory.
(gdb) bt
#0 0x00007ffff2057c37 in __GI_raise (sig=sig@entry=6) at ../nptl/sysdeps/unix/sysv/linux/raise.c:56
#1 0x00007ffff205b028 in __GI_abort () at abort.c:89
#2 0x00007ffff6f960eb in ?? () from /opt/rocm/hsa/lib/libhsa-runtime64.so.1
#3 0x00007ffff6f99ea5 in ?? () from /opt/rocm/hsa/lib/libhsa-runtime64.so.1
#4 0x00007ffff6f78107 in ?? () from /opt/rocm/hsa/lib/libhsa-runtime64.so.1
#5 0x00007ffff744f184 in start_thread (arg=0x7fffdffb5700) at pthread_create.c:312
#6 0x00007ffff211b37d in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:111
(gdb) info threads
Id Target Id Frame
4 Thread 0x7fffdd521700 (LWP 14895) "caffe" pthread_cond_wait@@GLIBC_2.3.2 () at ../nptl/sysdeps/unix/sysv/linux/x86_64/pthread_cond_wait.S:185
3 Thread 0x7fffddd22700 (LWP 14894) "caffe" pthread_cond_wait@@GLIBC_2.3.2 () at ../nptl/sysdeps/unix/sysv/linux/x86_64/pthread_cond_wait.S:185
* 2 Thread 0x7fffdffb5700 (LWP 14893) "caffe" 0x00007ffff2057c37 in __GI_raise (sig=sig@entry=6) at ../nptl/sysdeps/unix/sysv/linux/raise.c:56
1 Thread 0x7ffff7fa6ac0 (LWP 14892) "caffe" 0x00007ffff6f934d5 in ?? () from /opt/rocm/hsa/lib/libhsa-runtime64.so.1
(gdb) thread 1
[Switching to thread 1 (Thread 0x7ffff7fa6ac0 (LWP 14892))]
#0 0x00007ffff6f934d5 in ?? () from /opt/rocm/hsa/lib/libhsa-runtime64.so.1
(gdb) bt
#0 0x00007ffff6f934d5 in ?? () from /opt/rocm/hsa/lib/libhsa-runtime64.so.1
#1 0x00007ffff6f929ba in ?? () from /opt/rocm/hsa/lib/libhsa-runtime64.so.1
#2 0x00007fffe080beca in HSADispatch::waitComplete() () from /opt/rocm/hcc/lib/libmcwamp_hsa.so
#3 0x00007fffe080415f in HSADispatch::dispatchKernelAsync(Kalmar::HSAQueue*, void const*, int, bool) () from /opt/rocm/hcc/lib/libmcwamp_hsa.so
#4 0x00007fffe080238e in Kalmar::HSAQueue::dispatch_hsa_kernel(hsa_kernel_dispatch_packet_s const*, void const*, unsigned long, hc::completion_future*) () from /opt/rocm/hcc/lib/libmcwamp_hsa.so
#5 0x00007ffff7bb7559 in hipModuleLaunchKernel () from /opt/rocm/hip/lib/libhip_hcc.so
#6 0x00007ffff2e6cd2c in mlopen::HIPOCKernel::run (this=0x7fffffffb5a8, args=0x7fffffffb2a8, size=80) at /root/MIOpen/src/hipoc/hipoc_kernel.cpp:15
...
```
### General Debugging Tips
- The fault will be caught by the runtime but was actually generated by an asynchronous command running on the GPU. So, the GDB backtrace will show a path in the runtime, ie inside "GI_Raise" as shown in the example above.
- To determine the true location of the fault, force the kernels to execute synchronously by seeing the environment variables HCC_SERIALIZE_KERNEL=3 HCC_SERIALIZE_COPY=3. This will force HCC to wait for the kernel to finish executing before retuning. If the fault occurs during the execution of a kernel, you can see the code which launched the kernel inside the backtrace. A bit of guesswork is required to determine which thread is actually causing the issue - typically it will the thread which is waiting inside the libhsa-runtime64.so.
- VM faults inside kernels can be caused byi:
- incorrect code (ie a for loop which extends past array boundaries), i
- memory issues - kernel arguments which are invalid (null pointers, unregistered host pointers, bad pointers).
- synchronization issues
- compiler issues (incorrect code generation from the compiler)
- runtime issues
-- General debug tips:
- 'gdb --args' can be used to conviently pass the executable and arguments to gdb.
- From inside GDB, you can set environment variables "set env". Note the command does not use an '=' sign:
```
(gdb) set env HIP_DB 1
```
Setting HIP_PRINT_ENV=1 and then running a HIP application will print the HIP environment variables, their current values, and usage info.
Setting HCC_PRINT_ENV=1 and then running a HCC application will print the HCC environment variables, their current values, and usage info.
projects/hip/docs/markdown/hip_terms2.md
+1 -1
Datei anzeigen
@@ -13,7 +13,7 @@ The default device can be set with hipSetDevice.
- hcc = Heterogeneous Compute Compiler (https://bitbucket.org/multicoreware/hcc/wiki/Home).
- hipify - tool to convert CUDA(R) code to portable C++ code.
- hipconfig - tool to report various confoguration properties of the target platform.
- hipconfig - tool to report various configuration properties of the target platform.
- nvcc = nvcc compiler, do not capitalize.
- hcc = heterogeneous compute compiler, do not capitalize.
projects/hip/hip-config.cmake.in
+65
Datei anzeigen
@@ -0,0 +1,65 @@
@PACKAGE_INIT@
include(CMakeFindDependencyMacro OPTIONAL RESULT_VARIABLE _CMakeFindDependencyMacro_FOUND)
if (NOT _CMakeFindDependencyMacro_FOUND)
macro(find_dependency dep)
if (NOT ${dep}_FOUND)
set(cmake_fd_version)
if (${ARGC} GREATER 1)
set(cmake_fd_version ${ARGV1})
endif()
set(cmake_fd_exact_arg)
if(${CMAKE_FIND_PACKAGE_NAME}_FIND_VERSION_EXACT)
set(cmake_fd_exact_arg EXACT)
endif()
set(cmake_fd_quiet_arg)
if(${CMAKE_FIND_PACKAGE_NAME}_FIND_QUIETLY)
set(cmake_fd_quiet_arg QUIET)
endif()
set(cmake_fd_required_arg)
if(${CMAKE_FIND_PACKAGE_NAME}_FIND_REQUIRED)
set(cmake_fd_required_arg REQUIRED)
endif()
find_package(${dep} ${cmake_fd_version}
${cmake_fd_exact_arg}
${cmake_fd_quiet_arg}
${cmake_fd_required_arg}
)
string(TOUPPER ${dep} cmake_dep_upper)
if (NOT ${dep}_FOUND AND NOT ${cmake_dep_upper}_FOUND)
set(${CMAKE_FIND_PACKAGE_NAME}_NOT_FOUND_MESSAGE "${CMAKE_FIND_PACKAGE_NAME} could not be found because dependency ${dep} could not be found.")
set(${CMAKE_FIND_PACKAGE_NAME}_FOUND False)
return()
endif()
set(cmake_fd_version)
set(cmake_fd_required_arg)
set(cmake_fd_quiet_arg)
set(cmake_fd_exact_arg)
endif()
endmacro()
endif()
set_and_check( hip_INCLUDE_DIR "@PACKAGE_INCLUDE_INSTALL_DIR@" )
set_and_check( hip_INCLUDE_DIRS "${hip_INCLUDE_DIR}" )
set_and_check( hip_LIB_INSTALL_DIR "@PACKAGE_LIB_INSTALL_DIR@" )
set_and_check( hip_BIN_INSTALL_DIR "@PACKAGE_BIN_INSTALL_DIR@" )
set_and_check(hip_HIPCC_EXECUTABLE "${hip_BIN_INSTALL_DIR}/hipcc")
set_and_check(hip_HIPCONFIG_EXECUTABLE "${hip_BIN_INSTALL_DIR}/hipconfig")
find_dependency(hcc)
include( "${CMAKE_CURRENT_LIST_DIR}/hip-targets.cmake" )
set( hip_LIBRARIES hip::hip_hcc)
set( hip_LIBRARY ${hip_LIBRARIES})
set(HIP_INCLUDE_DIR ${hip_INCLUDE_DIR})
set(HIP_INCLUDE_DIRS ${hip_INCLUDE_DIRS})
set(HIP_LIB_INSTALL_DIR ${hip_LIB_INSTALL_DIR})
set(HIP_BIN_INSTALL_DIR ${hip_BIN_INSTALL_DIR})
set(HIP_LIBRARIES ${hip_LIBRARIES})
set(HIP_LIBRARY ${hip_LIBRARY})
set(HIP_HIPCC_EXECUTABLE ${hip_HIPCC_EXECUTABLE})
set(HIP_HIPCONFIG_EXECUTABLE ${hip_HIPCONFIG_EXECUTABLE})
projects/hip/hipify-clang/CMakeLists.txt
+3
Datei anzeigen
@@ -19,6 +19,9 @@ else()
add_llvm_executable(hipify-clang src/Cuda2Hip.cpp)
find_program(LIT_COMMAND lit)
set(CMAKE_CXX_COMPILER ${LLVM_TOOLS_BINARY_DIR}/clang++)
set(CMAKE_C_COMPILER ${LLVM_TOOLS_BINARY_DIR}/clang)
# Link against LLVM and CLANG tools libraries
target_link_libraries(hipify-clang
clangASTMatchers
projects/hip/hipify-clang/README.md
+31 -6
Datei anzeigen
@@ -13,24 +13,31 @@
`hipify-clang` is a clang-based tool which can automate the translation of CUDA source code into portable HIP C++.
The tool can automatically add extra HIP arguments (notably the "hipLaunchParm" required at the beginning of every HIP kernel call).
`hipify-clang` has some additional dependencies explained below and can be built as a separate make step. The instructions below are specifically for **Ubuntu 14.04**
`hipify-clang` has some additional dependencies explained below and can be built as a separate make step. The instructions below are specifically for **Ubuntu 14.04** and **Ubuntu 16.04**.
### Build and install
- Download and unpack clang+llvm 3.8 binary package preqrequisite.
**Ubuntu 14.04**:
```shell
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
```
**Ubuntu 16.04**:
```shell
wget http://llvm.org/releases/3.8.0/clang+llvm-3.8.0-x86_64-linux-gnu-ubuntu-16.04.tar.xz
tar xvfJ clang+llvm-3.8.0-x86_64-linux-gnu-ubuntu-16.04.tar.xz
```
- Enable build of hipify-clang and specify path to LLVM.
Note HIPIFY_CLANG_LLVM_DIR must be a full absolute path to the location extracted above. Here's an example assuming we extract the clang 3.8 package into ~/HIP/clang+llvm-3.8.0-x86_64-linux-gnu-ubuntu-14.04/
Note HIPIFY_CLANG_LLVM_DIR must be a full absolute path to the location extracted above. Here's an example assuming we extract the clang 3.8 package into ~/HIP/clang+llvm-3.8.0/
```shell
cd HIP
mkdir build
cd build
cmake -DHIPIFY_CLANG_LLVM_DIR=~/HIP/clang+llvm-3.8.0-x86_64-linux-gnu-ubuntu-14.04/ -DCMAKE_BUILD_TYPE=Release ..
cmake -DHIPIFY_CLANG_LLVM_DIR=~/HIP/clang+llvm-3.8.0/ -DCMAKE_BUILD_TYPE=Release ..
make
make install
```
@@ -41,17 +48,35 @@ make install
In the case when `hipify-clang` doesn't find cuda headers, it reports various errors about unknown keywords (e.g. '\__global\__'), API function names (e.g. 'cudaMalloc'), syntax (e.g. 'foo<<<1,n>>>(...)'), etc.
To install CUDA headers, download the "deb(network)" variant of the target installer from https://developer.nvidia.com/cuda-downloads. The commands below show how to download and install a recent version from http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1404/x86_64/cuda-repo-ubuntu1404_7.5-18_amd64.deb.
To install CUDA headers, download the "deb(network)" variant of the target installer.
**Ubuntu 14.04**:
```shell
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
```
**Ubuntu 16.04**:
```shell
wget http://archive.ubuntu.com/ubuntu/pool/multiverse/n/nvidia-cuda-toolkit/nvidia-cuda-toolkit_7.5.18-0ubuntu1_amd64.deb
sudo dpkg -i nvidia-cuda-toolkit_7.5.18-0ubuntu1_amd64.deb
sudo apt-get update && sudo apt-get install cuda-minimal-build-7-5 cuda-curand-dev-7-5
```
To set additional options like Language Selection (only "-x cuda" is supported), Preprocessor Definition (-D), Include Path (-I), etc., options delimiter "--" should be used before them, for instance:
#### Disclaimer
```shell
./hipify-clang -print-stats sort_kernel.cu -- -x cuda -I/srv/git/HIP/include -I/usr/local/cuda-7.5/include -DX=1
```
Delimiter "--" is used to separate hipify-clang options (before the delimiter) from clang options (after the delimiter). It is strongly recommended to always specify the delimiter, even if there are no clang specific options at all, in order to avoid possible errors regarding compilation database; in such case delimeter should be the last option in hipify-clang's command line.
Option "-x clang" is also worth specifying in order to convert source CUDA files with extensions other than standard extensions (*.cu, *.cuh).
## Disclaimer
The information contained herein is for informational purposes only, and is subject to change without notice. While every precaution has been taken in the preparation of this document, it may contain technical inaccuracies, omissions and typographical errors, and AMD is under no obligation to update or otherwise correct this information. Advanced Micro Devices, Inc. makes no representations or warranties with respect to the accuracy or completeness of the contents of this document, and assumes no liability of any kind, including the implied warranties of noninfringement, merchantability or fitness for particular purposes, with respect to the operation or use of AMD hardware, software or other products described herein. No license, including implied or arising by estoppel, to any intellectual property rights is granted by this document. Terms and limitations applicable to the purchase or use of AMD's products are as set forth in a signed agreement between the parties or in AMD's Standard Terms and Conditions of Sale.
AMD, the AMD Arrow logo, and combinations thereof are trademarks of Advanced Micro Devices, Inc. Other product names used in this publication are for identification purposes only and may be trademarks of their respective companies.
Copyright (c) 2014-2016 Advanced Micro Devices, Inc. All rights reserved.
Copyright (c) 2014-2017 Advanced Micro Devices, Inc. All rights reserved.
projects/hip/hipify-clang/src/Cuda2Hip.cpp
+2534 -1127
Datei anzeigen
Datei-Diff unterdrückt, da er zu groß ist Diff laden
projects/hip/include/hip/channel_descriptor.h
+39
Datei anzeigen
@@ -0,0 +1,39 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#ifndef HIP_INCLUDE_HIP_CHANNEL_DESCRIPTOR_H
#define HIP_INCLUDE_HIP_CHANNEL_DESCRIPTOR_H
// Some standard header files, these are included by hc.hpp and so want to make them avail on both
// paths to provide a consistent include env and avoid "missing symbol" errors that only appears
// on NVCC path:
#if defined(__HIP_PLATFORM_HCC__) && !defined (__HIP_PLATFORM_NVCC__)
#include <hip/hcc_detail/channel_descriptor.h>
#elif defined(__HIP_PLATFORM_NVCC__) && !defined (__HIP_PLATFORM_HCC__)
#include <hip/nvcc_detail/channel_descriptor.h>
#else
#error("Must define exactly one of __HIP_PLATFORM_HCC__ or __HIP_PLATFORM_NVCC__");
#endif
#endif
projects/hip/include/hip/device_functions.h
+36
Datei anzeigen
@@ -0,0 +1,36 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#ifndef HIP_INCLUDE_HIP_DEVICE_FUNCTIONS_H
#define HIP_INCLUDE_HIP_DEVICE_FUNCTIONS_H
#include<hip/hip_common.h>
#if defined(__HIP_PLATFORM_HCC__) && !defined (__HIP_PLATFORM_NVCC__)
#include <hip/hcc_detail/device_functions.h>
#elif defined(__HIP_PLATFORM_NVCC__) && !defined (__HIP_PLATFORM_HCC__)
#include <device_functions.h>
#else
#error("Must define exactly one of __HIP_PLATFORM_HCC__ or __HIP_PLATFORM_NVCC__");
#endif
#endif
projects/hip/include/hip/driver_types.h
+36
Datei anzeigen
@@ -0,0 +1,36 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#ifndef HIP_INCLUDE_HIP_DRIVER_TYPES_H
#define HIP_INCLUDE_HIP_DRIVER_TYPES_H
#include <hip/hip_common.h>
#if defined(__HIP_PLATFORM_HCC__) && !defined (__HIP_PLATFORM_NVCC__)
#include <hip/hcc_detail/driver_types.h>
#elif defined(__HIP_PLATFORM_NVCC__) && !defined (__HIP_PLATFORM_HCC__)
#include "driver_types.h"
#else
#error("Must define exactly one of __HIP_PLATFORM_HCC__ or __HIP_PLATFORM_NVCC__");
#endif
#endif
projects/hip/include/hip/hcc_detail/channel_descriptor.h
+382
Datei anzeigen
@@ -0,0 +1,382 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#ifndef HIP_INCLUDE_HIP_HCC_DETAIL_CHANNEL_DESCRIPTOR_H
#define HIP_INCLUDE_HIP_HCC_DETAIL_CHANNEL_DESCRIPTOR_H
#include<hip/hcc_detail/driver_types.h>
#include<hip/hcc_detail/hip_vector_types.h>
#ifdef __cplusplus
hipChannelFormatDesc hipCreateChannelDesc(int x, int y, int z, int w, hipChannelFormatKind f);
static inline hipChannelFormatDesc hipCreateChannelDescHalf() {
int e = (int)sizeof(unsigned short) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindFloat);
}
static inline hipChannelFormatDesc hipCreateChannelDescHalf1() {
int e = (int)sizeof(unsigned short) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindFloat);
}
static inline hipChannelFormatDesc hipCreateChannelDescHalf2()
{
int e = (int)sizeof(unsigned short) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindFloat);
}
template<typename T>
static inline hipChannelFormatDesc hipCreateChannelDesc() {
return hipCreateChannelDesc(0, 0, 0, 0, hipChannelFormatKindNone);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<char>()
{
int e = (int)sizeof(char) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindSigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<signed char>()
{
int e = (int)sizeof(signed char) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindUnsigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<unsigned char>()
{
int e = (int)sizeof(unsigned char) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindUnsigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<uchar1>()
{
int e = (int)sizeof(unsigned char) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindSigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<char1>()
{
int e = (int)sizeof(signed char) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindSigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<uchar2>()
{
int e = (int)sizeof(unsigned char) * 8;
return hipCreateChannelDesc(e, e, 0, 0, hipChannelFormatKindSigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<char2>()
{
int e = (int)sizeof(signed char) * 8;
return hipCreateChannelDesc(e, e, 0, 0, hipChannelFormatKindSigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<uchar3>()
{
int e = (int)sizeof(unsigned char) * 8;
return hipCreateChannelDesc(e, e, e, 0, hipChannelFormatKindSigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<char3>()
{
int e = (int)sizeof(signed char) * 8;
return hipCreateChannelDesc(e, e, e, 0, hipChannelFormatKindSigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<uchar4>()
{
int e = (int)sizeof(unsigned char) * 8;
return hipCreateChannelDesc(e, e, e, e, hipChannelFormatKindSigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<char4>()
{
int e = (int)sizeof(signed char) * 8;
return hipCreateChannelDesc(e, e, e, e, hipChannelFormatKindSigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<unsigned short>()
{
int e = (int)sizeof(unsigned short) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindUnsigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<signed short>()
{
int e = (int)sizeof(signed short) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindSigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<ushort1>()
{
int e = (int)sizeof(unsigned short) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindUnsigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<short1>()
{
int e = (int)sizeof(signed short) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindSigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<ushort2>()
{
int e = (int)sizeof(unsigned short) * 8;
return hipCreateChannelDesc(e, e, 0, 0, hipChannelFormatKindUnsigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<short2>()
{
int e = (int)sizeof(signed short) * 8;
return hipCreateChannelDesc(e, e, 0, 0, hipChannelFormatKindSigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<ushort3>()
{
int e = (int)sizeof(unsigned short) * 8;
return hipCreateChannelDesc(e, e, e, 0, hipChannelFormatKindUnsigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<short3>()
{
int e = (int)sizeof(signed short) * 8;
return hipCreateChannelDesc(e, e, e, 0, hipChannelFormatKindSigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<ushort4>()
{
int e = (int)sizeof(unsigned short) * 8;
return hipCreateChannelDesc(e, e, e, e, hipChannelFormatKindUnsigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<short4>()
{
int e = (int)sizeof(signed short) * 8;
return hipCreateChannelDesc(e, e, e, e, hipChannelFormatKindSigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<unsigned int>()
{
int e = (int)sizeof(unsigned int) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindUnsigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<signed int>()
{
int e = (int)sizeof(signed int) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindSigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<uint1>()
{
int e = (int)sizeof(unsigned int) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindUnsigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<int1>()
{
int e = (int)sizeof(signed int) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindSigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<uint2>()
{
int e = (int)sizeof(unsigned int) * 8;
return hipCreateChannelDesc(e, e, 0, 0, hipChannelFormatKindUnsigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<int2>()
{
int e = (int)sizeof(signed int) * 8;
return hipCreateChannelDesc(e, e, 0, 0, hipChannelFormatKindSigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<uint3>()
{
int e = (int)sizeof(unsigned int) * 8;
return hipCreateChannelDesc(e, e, e, 0, hipChannelFormatKindUnsigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<int3>()
{
int e = (int)sizeof(signed int) * 8;
return hipCreateChannelDesc(e, e, e, 0, hipChannelFormatKindSigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<uint4>()
{
int e = (int)sizeof(unsigned int) * 8;
return hipCreateChannelDesc(e, e, e, e, hipChannelFormatKindUnsigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<int4>()
{
int e = (int)sizeof(signed int) * 8;
return hipCreateChannelDesc(e, e, e, e, hipChannelFormatKindSigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<float>()
{
int e = (int)sizeof(float) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindFloat);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<float1>()
{
int e = (int)sizeof(float) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindFloat);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<float2>()
{
int e = (int)sizeof(float) * 8;
return hipCreateChannelDesc(e, e, 0, 0, hipChannelFormatKindFloat);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<float3>()
{
int e = (int)sizeof(float) * 8;
return hipCreateChannelDesc(e, e, e, 0, hipChannelFormatKindFloat);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<float4>()
{
int e = (int)sizeof(float) * 8;
return hipCreateChannelDesc(e, e, e, e, hipChannelFormatKindFloat);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<unsigned long>()
{
int e = (int)sizeof(unsigned long) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindUnsigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<signed long>()
{
int e = (int)sizeof(signed long) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindSigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<ulong1>()
{
int e = (int)sizeof(unsigned long) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindUnsigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<long1>()
{
int e = (int)sizeof(signed long) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindSigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<ulong2>()
{
int e = (int)sizeof(unsigned long) * 8;
return hipCreateChannelDesc(e, e, 0, 0, hipChannelFormatKindUnsigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<long2>()
{
int e = (int)sizeof(signed long) * 8;
return hipCreateChannelDesc(e, e, 0, 0, hipChannelFormatKindSigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<ulong3>()
{
int e = (int)sizeof(unsigned long) * 8;
return hipCreateChannelDesc(e, e, e, 0, hipChannelFormatKindUnsigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<long3>()
{
int e = (int)sizeof(signed long) * 8;
return hipCreateChannelDesc(e, e, e, 0, hipChannelFormatKindSigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<ulong4>()
{
int e = (int)sizeof(unsigned long) * 8;
return hipCreateChannelDesc(e, e, e, e, hipChannelFormatKindUnsigned);
}
template<>
inline hipChannelFormatDesc hipCreateChannelDesc<long4>()
{
int e = (int)sizeof(signed long) * 8;
return hipCreateChannelDesc(e, e, e, e, hipChannelFormatKindSigned);
}
#else
struct hipChannelFormatDesc hipCreateChannelDesc(int x, int y, int z, int w, enum hipChannelFormatKind f);
#endif
#endif
projects/hip/include/hip/hcc.h → projects/hip/include/hip/hcc_detail/concepts.hpp
+6 -6
Datei anzeigen
@@ -20,11 +20,11 @@ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#ifndef HIP_HCC_H
#define HIP_HCC_H
#pragma once
#if defined(__HIP_PLATFORM_HCC__) && !defined (__HIP_PLATFORM_NVCC__)
#include "hip/hcc_detail/hcc_acc.h"
#endif
namespace hip_impl // Documentation only.
{
#define requires(...)
#endif
#define FunctionalProcedure typename
}
projects/hip/include/hip/hcc_detail/device_functions.h
+604
Datei anzeigen
@@ -0,0 +1,604 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#ifndef HIP_INCLUDE_HIP_HCC_DETAIL_DEVICE_FUNCTIONS_H
#define HIP_INCLUDE_HIP_HCC_DETAIL_DEVICE_FUNCTIONS_H
#include <hip/hip_runtime.h>
#include <hip/hip_vector_types.h>
// Single Precision Fast Math
__device__ float __cosf(float x);
__device__ float __exp10f(float x);
__device__ float __expf(float x);
__device__ static float __fadd_rd(float x, float y);
__device__ static float __fadd_rn(float x, float y);
__device__ static float __fadd_ru(float x, float y);
__device__ static float __fadd_rz(float x, float y);
__device__ static float __fdiv_rd(float x, float y);
__device__ static float __fdiv_rn(float x, float y);
__device__ static float __fdiv_ru(float x, float y);
__device__ static float __fdiv_rz(float x, float y);
__device__ static float __fdividef(float x, float y);
__device__ float __fmaf_rd(float x, float y, float z);
__device__ float __fmaf_rn(float x, float y, float z);
__device__ float __fmaf_ru(float x, float y, float z);
__device__ float __fmaf_rz(float x, float y, float z);
__device__ static float __fmul_rd(float x, float y);
__device__ static float __fmul_rn(float x, float y);
__device__ static float __fmul_ru(float x, float y);
__device__ static float __fmul_rz(float x, float y);
__device__ float __frcp_rd(float x);
__device__ float __frcp_rn(float x);
__device__ float __frcp_ru(float x);
__device__ float __frcp_rz(float x);
__device__ float __frsqrt_rn(float x);
__device__ float __fsqrt_rd(float x);
__device__ float __fsqrt_rn(float x);
__device__ float __fsqrt_ru(float x);
__device__ float __fsqrt_rz(float x);
__device__ static float __fsub_rd(float x, float y);
__device__ static float __fsub_rn(float x, float y);
__device__ static float __fsub_ru(float x, float y);
__device__ float __log10f(float x);
__device__ float __log2f(float x);
__device__ float __logf(float x);
__device__ float __powf(float base, float exponent);
__device__ static float __saturatef(float x);
__device__ void __sincosf(float x, float *s, float *c);
__device__ float __sinf(float x);
__device__ float __tanf(float x);
/*
Double Precision Intrinsics
*/
__device__ static double __dadd_rd(double x, double y);
__device__ static double __dadd_rn(double x, double y);
__device__ static double __dadd_ru(double x, double y);
__device__ static double __dadd_rz(double x, double y);
__device__ static double __ddiv_rd(double x, double y);
__device__ static double __ddiv_rn(double x, double y);
__device__ static double __ddiv_ru(double x, double y);
__device__ static double __ddiv_rz(double x, double y);
__device__ static double __dmul_rd(double x, double y);
__device__ static double __dmul_rn(double x, double y);
__device__ static double __dmul_ru(double x, double y);
__device__ static double __dmul_rz(double x, double y);
__device__ double __drcp_rd(double x);
__device__ double __drcp_rn(double x);
__device__ double __drcp_ru(double x);
__device__ double __drcp_rz(double x);
__device__ double __dsqrt_rd(double x);
__device__ double __dsqrt_rn(double x);
__device__ double __dsqrt_ru(double x);
__device__ double __dsqrt_rz(double x);
__device__ static double __dsub_rd(double x, double y);
__device__ static double __dsub_rn(double x, double y);
__device__ static double __dsub_ru(double x, double y);
__device__ static double __dsub_rz(double x, double y);
__device__ double __fma_rd(double x, double y, double z);
__device__ double __fma_rn(double x, double y, double z);
__device__ double __fma_ru(double x, double y, double z);
__device__ double __fma_rz(double x, double y, double z);
// Single Precision Fast Math
extern __attribute__((const)) float __hip_fast_cosf(float) __asm("llvm.cos.f32");
extern __attribute__((const)) float __hip_fast_exp2f(float) __asm("llvm.exp2.f32");
__device__ float __hip_fast_exp10f(float);
__device__ float __hip_fast_expf(float);
__device__ float __hip_fast_frsqrt_rn(float);
extern __attribute__((const)) float __hip_fast_fsqrt_rd(float) __asm("llvm.sqrt.f32");
__device__ float __hip_fast_fsqrt_rn(float);
__device__ float __hip_fast_fsqrt_ru(float);
__device__ float __hip_fast_fsqrt_rz(float);
__device__ float __hip_fast_log10f(float);
extern __attribute__((const)) float __hip_fast_log2f(float) __asm("llvm.log2.f32");
__device__ float __hip_fast_logf(float);
__device__ float __hip_fast_powf(float, float);
__device__ void __hip_fast_sincosf(float,float*,float*);
extern __attribute__((const)) float __hip_fast_sinf(float) __asm("llvm.sin.f32");
__device__ float __hip_fast_tanf(float);
extern __attribute__((const)) float __hip_fast_fmaf(float,float,float) __asm("llvm.fma.f32");
extern __attribute__((const)) float __hip_fast_frcp(float) __asm("llvm.amdgcn.rcp.f32");
extern __attribute__((const)) double __hip_fast_dsqrt(double) __asm("llvm.sqrt.f64");
extern __attribute__((const)) double __hip_fast_fma(double,double,double) __asm("llvm.fma.f64");
extern __attribute__((const)) double __hip_fast_drcp(double) __asm("llvm.amdgcn.rcp.f64");
// Single Precision Fast Math
__device__ inline float __cosf(float x) {
return __hip_fast_cosf(x);
}
__device__ inline float __exp10f(float x) {
return __hip_fast_exp10f(x);
}
__device__ inline float __expf(float x) {
return __hip_fast_expf(x);
}
__device__ static inline float __fadd_rd(float x, float y) {
return x + y;
}
__device__ static inline float __fadd_rn(float x, float y) {
return x + y;
}
__device__ static inline float __fadd_ru(float x, float y) {
return x + y;
}
__device__ static inline float __fadd_rz(float x, float y) {
return x + y;
}
__device__ static inline float __fdiv_rd(float x, float y) {
return x / y;
}
__device__ static inline float __fdiv_rn(float x, float y) {
return x / y;
}
__device__ static inline float __fdiv_ru(float x, float y) {
return x / y;
}
__device__ static inline float __fdiv_rz(float x, float y) {
return x / y;
}
__device__ static inline float __fdividef(float x, float y) {
return x / y;
}
__device__ inline float __fmaf_rd(float x, float y, float z) {
return __hip_fast_fmaf(x, y, z);
}
__device__ inline float __fmaf_rn(float x, float y, float z) {
return __hip_fast_fmaf(x, y, z);
}
__device__ inline float __fmaf_ru(float x, float y, float z) {
return __hip_fast_fmaf(x, y, z);
}
__device__ inline float __fmaf_rz(float x, float y, float z) {
return __hip_fast_fmaf(x, y, z);
}
__device__ static inline float __fmul_rd(float x, float y) {
return x * y;
}
__device__ static inline float __fmul_rn(float x, float y) {
return x * y;
}
__device__ static inline float __fmul_ru(float x, float y) {
return x * y;
}
__device__ static inline float __fmul_rz(float x, float y) {
return x * y;
}
__device__ inline float __frcp_rd(float x) {
return __hip_fast_frcp(x);
}
__device__ inline float __frcp_rn(float x) {
return __hip_fast_frcp(x);
}
__device__ inline float __frcp_ru(float x) {
return __hip_fast_frcp(x);
}
__device__ inline float __frcp_rz(float x) {
return __hip_fast_frcp(x);
}
__device__ inline float __frsqrt_rn(float x) {
return __hip_fast_frsqrt_rn(x);
}
__device__ inline float __fsqrt_rd(float x) {
return __hip_fast_fsqrt_rd(x);
}
__device__ inline float __fsqrt_rn(float x) {
return __hip_fast_fsqrt_rn(x);
}
__device__ inline float __fsqrt_ru(float x) {
return __hip_fast_fsqrt_ru(x);
}
__device__ inline float __fsqrt_rz(float x) {
return __hip_fast_fsqrt_rz(x);
}
__device__ static inline float __fsub_rd(float x, float y) {
return x - y;
}
__device__ static inline float __fsub_rn(float x, float y) {
return x - y;
}
__device__ static inline float __fsub_ru(float x, float y) {
return x - y;
}
__device__ static inline float __fsub_rz(float x, float y) {
return x - y;
}
__device__ inline float __log10f(float x) {
return __hip_fast_log10f(x);
}
__device__ inline float __log2f(float x) {
return __hip_fast_log2f(x);
}
__device__ inline float __logf(float x) {
return __hip_fast_logf(x);
}
__device__ inline float __powf(float base, float exponent) {
return __hip_fast_powf(base, exponent);
}
__device__ static inline float __saturatef(float x) {
x = x > 1.0f ? 1.0f : x;
x = x < 0.0f ? 0.0f : x;
return x;
}
__device__ inline void __sincosf(float x, float *s, float *c) {
return __hip_fast_sincosf(x, s, c);
}
__device__ inline float __sinf(float x) {
return __hip_fast_sinf(x);
}
__device__ inline float __tanf(float x) {
return __hip_fast_tanf(x);
}
/*
Double Precision Intrinsics
*/
__device__ static inline double __dadd_rd(double x, double y) {
return x + y;
}
__device__ static inline double __dadd_rn(double x, double y) {
return x + y;
}
__device__ static inline double __dadd_ru(double x, double y) {
return x + y;
}
__device__ static inline double __dadd_rz(double x, double y) {
return x + y;
}
__device__ static inline double __ddiv_rd(double x, double y) {
return x / y;
}
__device__ static inline double __ddiv_rn(double x, double y) {
return x / y;
}
__device__ static inline double __ddiv_ru(double x, double y) {
return x / y;
}
__device__ static inline double __ddiv_rz(double x, double y) {
return x / y;
}
__device__ static inline double __dmul_rd(double x, double y) {
return x * y;
}
__device__ static inline double __dmul_rn(double x, double y) {
return x * y;
}
__device__ static inline double __dmul_ru(double x, double y) {
return x * y;
}
__device__ static inline double __dmul_rz(double x, double y) {
return x * y;
}
__device__ inline double __drcp_rd(double x) {
return __hip_fast_drcp(x);
}
__device__ inline double __drcp_rn(double x) {
return __hip_fast_drcp(x);
}
__device__ inline double __drcp_ru(double x) {
return __hip_fast_drcp(x);
}
__device__ inline double __drcp_rz(double x) {
return __hip_fast_drcp(x);
}
__device__ inline double __dsqrt_rd(double x) {
return __hip_fast_dsqrt(x);
}
__device__ inline double __dsqrt_rn(double x) {
return __hip_fast_dsqrt(x);
}
__device__ inline double __dsqrt_ru(double x) {
return __hip_fast_dsqrt(x);
}
__device__ inline double __dsqrt_rz(double x) {
return __hip_fast_dsqrt(x);
}
__device__ static inline double __dsub_rd(double x, double y) {
return x - y;
}
__device__ static inline double __dsub_rn(double x, double y) {
return x - y;
}
__device__ static inline double __dsub_ru(double x, double y) {
return x - y;
}
__device__ static inline double __dsub_rz(double x, double y) {
return x - y;
}
__device__ inline double __fma_rd(double x, double y, double z) {
return __hip_fast_fma(x, y, z);
}
__device__ inline double __fma_rn(double x, double y, double z) {
return __hip_fast_fma(x, y, z);
}
__device__ inline double __fma_ru(double x, double y, double z) {
return __hip_fast_fma(x, y, z);
}
__device__ inline double __fma_rz(double x, double y, double z) {
return __hip_fast_fma(x, y, z);
}
extern "C" unsigned int __hip_hc_ir_umul24_int(unsigned int, unsigned int);
extern "C" signed int __hip_hc_ir_mul24_int(signed int, signed int);
extern "C" signed int __hip_hc_ir_mulhi_int(signed int, signed int);
extern "C" unsigned int __hip_hc_ir_umulhi_int(unsigned int, unsigned int);
extern "C" unsigned int __hip_hc_ir_usad_int(unsigned int, unsigned int, unsigned int);
// integer intrinsic function __poc __clz __ffs __brev
__device__ unsigned int __brev( unsigned int x);
__device__ unsigned long long int __brevll( unsigned long long int x);
__device__ unsigned int __byte_perm(unsigned int x, unsigned int y, unsigned int s);
__device__ unsigned int __clz(int x);
__device__ unsigned int __clzll(long long int x);
__device__ unsigned int __ffs(int x);
__device__ unsigned int __ffsll(long long int x);
__device__ static unsigned int __hadd(int x, int y);
__device__ static int __mul24(int x, int y);
__device__ long long int __mul64hi(long long int x, long long int y);
__device__ static int __mulhi(int x, int y);
__device__ unsigned int __popc(unsigned int x);
__device__ unsigned int __popcll(unsigned long long int x);
__device__ static int __rhadd(int x, int y);
__device__ static unsigned int __sad(int x, int y, int z);
__device__ static unsigned int __uhadd(unsigned int x, unsigned int y);
__device__ static int __umul24(unsigned int x, unsigned int y);
__device__ unsigned long long int __umul64hi(unsigned long long int x, unsigned long long int y);
__device__ static unsigned int __umulhi(unsigned int x, unsigned int y);
__device__ static unsigned int __urhadd(unsigned int x, unsigned int y);
__device__ static unsigned int __usad(unsigned int x, unsigned int y, unsigned int z);
__device__ static inline unsigned int __hadd(int x, int y) {
int z = x + y;
int sign = z & 0x8000000;
int value = z & 0x7FFFFFFF;
return ((value) >> 1 || sign);
}
__device__ static inline int __mul24(int x, int y) {
return __hip_hc_ir_mul24_int(x, y);
}
__device__ static inline int __mulhi(int x, int y) {
return __hip_hc_ir_mulhi_int(x, y);
}
__device__ static inline int __rhadd(int x, int y) {
int z = x + y + 1;
int sign = z & 0x8000000;
int value = z & 0x7FFFFFFF;
return ((value) >> 1 || sign);
}
__device__ static inline unsigned int __sad(int x, int y, int z) {
return x > y ? x - y + z : y - x + z;
}
__device__ static inline unsigned int __uhadd(unsigned int x, unsigned int y) {
return (x + y) >> 1;
}
__device__ static inline int __umul24(unsigned int x, unsigned int y) {
return __hip_hc_ir_umul24_int(x, y);
}
__device__ static inline unsigned int __umulhi(unsigned int x, unsigned int y) {
return __hip_hc_ir_umulhi_int(x, y);
}
__device__ static inline unsigned int __urhadd(unsigned int x, unsigned int y) {
return (x + y + 1) >> 1;
}
__device__ static inline unsigned int __usad(unsigned int x, unsigned int y, unsigned int z)
{
return __hip_hc_ir_usad_int(x, y, z);
}
/*
Rounding modes are not yet supported in HIP
*/
__device__ float __double2float_rd(double x);
__device__ float __double2float_rn(double x);
__device__ float __double2float_ru(double x);
__device__ float __double2float_rz(double x);
__device__ int __double2hiint(double x);
__device__ int __double2int_rd(double x);
__device__ int __double2int_rn(double x);
__device__ int __double2int_ru(double x);
__device__ int __double2int_rz(double x);
__device__ long long int __double2ll_rd(double x);
__device__ long long int __double2ll_rn(double x);
__device__ long long int __double2ll_ru(double x);
__device__ long long int __double2ll_rz(double x);
__device__ int __double2loint(double x);
__device__ unsigned int __double2uint_rd(double x);
__device__ unsigned int __double2uint_rn(double x);
__device__ unsigned int __double2uint_ru(double x);
__device__ unsigned int __double2uint_rz(double x);
__device__ unsigned long long int __double2ull_rd(double x);
__device__ unsigned long long int __double2ull_rn(double x);
__device__ unsigned long long int __double2ull_ru(double x);
__device__ unsigned long long int __double2ull_rz(double x);
__device__ long long int __double_as_longlong(double x);
/*
__device__ unsigned short __float2half_rn(float x);
__device__ float __half2float(unsigned short);
The above device function are not a valid .
Use
__device__ __half __float2half_rn(float x);
__device__ float __half2float(__half);
from hip_fp16.h
CUDA implements half as unsigned short whereas, HIP doesn't.
*/
__device__ int __float2int_rd(float x);
__device__ int __float2int_rn(float x);
__device__ int __float2int_ru(float x);
__device__ int __float2int_rz(float x);
__device__ long long int __float2ll_rd(float x);
__device__ long long int __float2ll_rn(float x);
__device__ long long int __float2ll_ru(float x);
__device__ long long int __float2ll_rz(float x);
__device__ unsigned int __float2uint_rd(float x);
__device__ unsigned int __float2uint_rn(float x);
__device__ unsigned int __float2uint_ru(float x);
__device__ unsigned int __float2uint_rz(float x);
__device__ unsigned long long int __float2ull_rd(float x);
__device__ unsigned long long int __float2ull_rn(float x);
__device__ unsigned long long int __float2ull_ru(float x);
__device__ unsigned long long int __float2ull_rz(float x);
__device__ int __float_as_int(float x);
__device__ unsigned int __float_as_uint(float x);
__device__ double __hiloint2double(int hi, int lo);
__device__ double __int2double_rn(int x);
__device__ float __int2float_rd(int x);
__device__ float __int2float_rn(int x);
__device__ float __int2float_ru(int x);
__device__ float __int2float_rz(int x);
__device__ float __int_as_float(int x);
__device__ double __ll2double_rd(long long int x);
__device__ double __ll2double_rn(long long int x);
__device__ double __ll2double_ru(long long int x);
__device__ double __ll2double_rz(long long int x);
__device__ float __ll2float_rd(long long int x);
__device__ float __ll2float_rn(long long int x);
__device__ float __ll2float_ru(long long int x);
__device__ float __ll2float_rz(long long int x);
__device__ double __longlong_as_double(long long int x);
__device__ double __uint2double_rn(int x);
__device__ float __uint2float_rd(unsigned int x);
__device__ float __uint2float_rn(unsigned int x);
__device__ float __uint2float_ru(unsigned int x);
__device__ float __uint2float_rz(unsigned int x);
__device__ float __uint_as_float(unsigned int x);
__device__ double __ull2double_rd(unsigned long long int x);
__device__ double __ull2double_rn(unsigned long long int x);
__device__ double __ull2double_ru(unsigned long long int x);
__device__ double __ull2double_rz(unsigned long long int x);
__device__ float __ull2float_rd(unsigned long long int x);
__device__ float __ull2float_rn(unsigned long long int x);
__device__ float __ull2float_ru(unsigned long long int x);
__device__ float __ull2float_rz(unsigned long long int x);
__device__ char4 __hip_hc_add8pk(char4, char4);
__device__ char4 __hip_hc_sub8pk(char4, char4);
__device__ char4 __hip_hc_mul8pk(char4, char4);
#endif
projects/hip/include/hip/hcc_detail/driver_types.h
+44
Datei anzeigen
@@ -0,0 +1,44 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#ifndef HIP_INCLUDE_HIP_HCC_DETAIL_DRIVER_TYPES_H
#define HIP_INCLUDE_HIP_HCC_DETAIL_DRIVER_TYPES_H
enum hipChannelFormatKind
{
hipChannelFormatKindSigned = 0,
hipChannelFormatKindUnsigned = 1,
hipChannelFormatKindFloat = 2,
hipChannelFormatKindNone = 3
};
struct hipChannelFormatDesc
{
int x;
int y;
int z;
int w;
enum hipChannelFormatKind f;
};
#endif
projects/hip/include/hip/hcc_detail/grid_launch_GGL.hpp
+863
Datei anzeigen
@@ -0,0 +1,863 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#pragma once
#if GENERIC_GRID_LAUNCH == 1
#include "concepts.hpp"
#include "helpers.hpp"
#include "hc.hpp"
#include "hip_hcc.h"
#include "hip_runtime.h"
#include <functional>
#include <iostream>
#include <stdexcept>
#include <type_traits>
#include <utility>
namespace hip_impl
{
namespace
{
struct New_grid_launch_tag {};
struct Old_grid_launch_tag {};
template<typename C, typename D>
class RAII_guard {
D dtor_;
public:
RAII_guard() = default;
RAII_guard(const C& ctor, D dtor) : dtor_{std::move(dtor)}
{
ctor();
}
RAII_guard(const RAII_guard&) = default;
RAII_guard(RAII_guard&&) = default;
RAII_guard& operator=(const RAII_guard&) = default;
RAII_guard& operator=(RAII_guard&&) = default;
~RAII_guard() { dtor_(); }
};
template<typename C, typename D>
RAII_guard<C, D> make_RAII_guard(const C& ctor, D dtor)
{
return RAII_guard<C, D>{ctor, std::move(dtor)};
}
template<FunctionalProcedure F, typename... Ts>
using is_new_grid_launch_t = typename std::conditional<
is_callable<F(Ts...)>{},
New_grid_launch_tag,
Old_grid_launch_tag>::type;
}
// TODO: - dispatch rank should be derived from the domain dimensions passed
// in, and not always assumed to be 3;
template<FunctionalProcedure K, typename... Ts>
requires(Domain<K> == {Ts...})
inline
void grid_launch_hip_impl_(
New_grid_launch_tag,
dim3 num_blocks,
dim3 dim_blocks,
int group_mem_bytes,
const hc::accelerator_view& acc_v,
K k,
Ts&&... args)
{
const auto d = hc::extent<3>{
num_blocks.z * dim_blocks.z,
num_blocks.y * dim_blocks.y,
num_blocks.x * dim_blocks.x}.tile_with_dynamic(
dim_blocks.z,
dim_blocks.y,
dim_blocks.x,
group_mem_bytes);
try {
hc::parallel_for_each(
acc_v,
d,
[=](const hc::tiled_index<3>& idx) [[hc]] {
k(args...);
});
}
catch (std::exception& ex) {
std::cerr << "Failed in " << __FUNCTION__ << ", with exception: "
<< ex.what() << std::endl;
throw;
}
}
// TODO: these are workarounds, they should be removed.
hc::accelerator_view lock_stream_hip_(hipStream_t&, void*&);
void print_prelaunch_trace_(const char*, dim3, dim3, int, hipStream_t);
void unlock_stream_hip_(
hipStream_t, void*, const char*, hc::accelerator_view*);
template<FunctionalProcedure K, typename... Ts>
requires(Domain<K> == {Ts...})
inline
void grid_launch_hip_impl_(
New_grid_launch_tag,
dim3 num_blocks,
dim3 dim_blocks,
int group_mem_bytes,
hipStream_t stream,
const char* kernel_name,
K k,
Ts&&... args)
{
void* lck_stream = nullptr;
auto acc_v = lock_stream_hip_(stream, lck_stream);
auto stream_guard = make_RAII_guard(
std::bind(
print_prelaunch_trace_,
kernel_name,
num_blocks,
dim_blocks,
group_mem_bytes,
stream),
std::bind(
unlock_stream_hip_, stream, lck_stream, kernel_name, &acc_v));
try {
grid_launch_hip_impl_(
New_grid_launch_tag{},
std::move(num_blocks),
std::move(dim_blocks),
group_mem_bytes,
acc_v,
std::move(k),
std::forward<Ts>(args)...);
}
catch (std::exception& ex) {
std::cerr << "Failed in " << __FUNCTION__ << ", with exception: "
<< ex.what() << std::endl;
throw;
}
}
template<FunctionalProcedure K, typename... Ts>
requires(Domain<K> == {hipLaunchParm, Ts...})
inline
void grid_launch_hip_impl_(
Old_grid_launch_tag,
dim3 num_blocks,
dim3 dim_blocks,
int group_mem_bytes,
hipStream_t stream,
K k,
Ts&&... args)
{
grid_launch_hip_impl_(
New_grid_launch_tag{},
std::move(num_blocks),
std::move(dim_blocks),
group_mem_bytes,
std::move(stream),
std::move(k),
hipLaunchParm{},
std::forward<Ts>(args)...);
}
template<FunctionalProcedure K, typename... Ts>
requires(Domain<K> == {hipLaunchParm, Ts...})
inline
void grid_launch_hip_impl_(
Old_grid_launch_tag,
dim3 num_blocks,
dim3 dim_blocks,
int group_mem_bytes,
hipStream_t stream,
const char* kernel_name,
K k,
Ts&&... args)
{
grid_launch_hip_impl_(
New_grid_launch_tag{},
std::move(num_blocks),
std::move(dim_blocks),
group_mem_bytes,
std::move(stream),
kernel_name,
std::move(k),
hipLaunchParm{},
std::forward<Ts>(args)...);
}
template<FunctionalProcedure K, typename... Ts>
requires(Domain<K> == {Ts...})
inline
std::enable_if_t<!std::is_function<K>::value> grid_launch_hip_(
dim3 num_blocks,
dim3 dim_blocks,
int group_mem_bytes,
hipStream_t stream,
const char* kernel_name,
K k,
Ts&& ... args)
{
grid_launch_hip_impl_(
is_new_grid_launch_t<K, Ts...>{},
std::move(num_blocks),
std::move(dim_blocks),
group_mem_bytes,
std::move(stream),
kernel_name,
std::move(k),
std::forward<Ts>(args)...);
}
template<FunctionalProcedure K, typename... Ts>
requires(Domain<K> == {Ts...})
inline
std::enable_if_t<!std::is_function<K>::value> grid_launch_hip_(
dim3 num_blocks,
dim3 dim_blocks,
int group_mem_bytes,
hipStream_t stream,
K k,
Ts&& ... args)
{
grid_launch_hip_impl_(
is_new_grid_launch_t<K, Ts...>{},
std::move(num_blocks),
std::move(dim_blocks),
group_mem_bytes,
std::move(stream),
std::move(k),
std::forward<Ts>(args)...);
}
namespace
{
template<typename T>
constexpr
inline
T&& forward_(std::remove_reference_t<T>& x) [[hc]]
{
return static_cast<T&&>(x);
}
template<FunctionalProcedure K, K* k>
struct Forwarder {
template<typename... Ts>
void operator()(Ts&&...args) const [[hc]]
{
k(forward_<Ts>(args)...);
}
};
}
template<FunctionalProcedure K, K* k, typename... Ts>
requires(Domain<K> == {Ts...})
inline
void grid_launch(
New_grid_launch_tag,
dim3 num_blocks,
dim3 dim_blocks,
int group_mem_bytes,
hipStream_t stream,
Ts&&... args)
{
grid_launch_hip_impl_(
New_grid_launch_tag{},
std::move(num_blocks),
std::move(dim_blocks),
group_mem_bytes,
std::move(stream),
Forwarder<K, k>{},
std::forward<Ts>(args)...);
}
template<FunctionalProcedure K, K* k, typename... Ts>
requires(Domain<K> == {Ts...})
inline
void grid_launch(
Old_grid_launch_tag,
dim3 num_blocks,
dim3 dim_blocks,
int group_mem_bytes,
hipStream_t stream,
Ts&&... args)
{
grid_launch_hip_<K, k>(
New_grid_launch_tag{},
std::move(num_blocks),
std::move(dim_blocks),
group_mem_bytes,
std::move(stream),
hipLaunchParm{},
std::forward<Ts>(args)...);
}
template<FunctionalProcedure K, K* k, typename... Ts>
requires(Domain<K> == {Ts...})
inline
std::enable_if_t<std::is_function<K>::value> grid_launch_hip_(
dim3 num_blocks,
dim3 dim_blocks,
int group_mem_bytes,
hipStream_t stream,
Ts&&... args)
{
grid_launch_hip_<K, k>(
is_new_grid_launch_t<K*, Ts...>{},
std::move(num_blocks),
std::move(dim_blocks),
group_mem_bytes,
std::move(stream),
std::forward<Ts>(args)...);
}
// TODO: these are temporary, they need to be completely removed once we
// enable C++14 support and can have proper generic, variadic lambdas.
#define make_kernel_lambda_hip_26(\
kernel_name,\
p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, p10, p11, p12, p13, p14, p15,\
p16, p17, p18, p19, p20, p21, p22, p23, p24)\
[](const std::decay_t<decltype(p0)>& _p0_,\
const std::decay_t<decltype(p1)>& _p1_,\
const std::decay_t<decltype(p2)>& _p2_,\
const std::decay_t<decltype(p3)>& _p3_,\
const std::decay_t<decltype(p4)>& _p4_,\
const std::decay_t<decltype(p5)>& _p5_,\
const std::decay_t<decltype(p6)>& _p6_,\
const std::decay_t<decltype(p7)>& _p7_,\
const std::decay_t<decltype(p8)>& _p8_,\
const std::decay_t<decltype(p9)>& _p9_,\
const std::decay_t<decltype(p10)>& _p10_,\
const std::decay_t<decltype(p11)>& _p11_,\
const std::decay_t<decltype(p12)>& _p12_,\
const std::decay_t<decltype(p13)>& _p13_,\
const std::decay_t<decltype(p14)>& _p14_,\
const std::decay_t<decltype(p15)>& _p15_,\
const std::decay_t<decltype(p16)>& _p16_,\
const std::decay_t<decltype(p17)>& _p17_,\
const std::decay_t<decltype(p18)>& _p18_,\
const std::decay_t<decltype(p19)>& _p19_,\
const std::decay_t<decltype(p20)>& _p20_,\
const std::decay_t<decltype(p21)>& _p21_,\
const std::decay_t<decltype(p22)>& _p22_,\
const std::decay_t<decltype(p23)>& _p23_,\
const std::decay_t<decltype(p24)>& _p24_) [[hc]] {\
kernel_name(\
_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_,\
_p9_, _p10_, _p11_, _p12_, _p13_, _p14_, _p15_, _p16_, _p17_,\
_p18_, _p19_, _p20_, _p21_, _p22_, _p23_, _p24_);\
}
#define make_kernel_lambda_hip_25(\
kernel_name,\
p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, p10, p11, p12, p13, p14, p15,\
p16, p17, p18, p19, p20, p21, p22, p23)\
[](const std::decay_t<decltype(p0)>& _p0_,\
const std::decay_t<decltype(p1)>& _p1_,\
const std::decay_t<decltype(p2)>& _p2_,\
const std::decay_t<decltype(p3)>& _p3_,\
const std::decay_t<decltype(p4)>& _p4_,\
const std::decay_t<decltype(p5)>& _p5_,\
const std::decay_t<decltype(p6)>& _p6_,\
const std::decay_t<decltype(p7)>& _p7_,\
const std::decay_t<decltype(p8)>& _p8_,\
const std::decay_t<decltype(p9)>& _p9_,\
const std::decay_t<decltype(p10)>& _p10_,\
const std::decay_t<decltype(p11)>& _p11_,\
const std::decay_t<decltype(p12)>& _p12_,\
const std::decay_t<decltype(p13)>& _p13_,\
const std::decay_t<decltype(p14)>& _p14_,\
const std::decay_t<decltype(p15)>& _p15_,\
const std::decay_t<decltype(p16)>& _p16_,\
const std::decay_t<decltype(p17)>& _p17_,\
const std::decay_t<decltype(p18)>& _p18_,\
const std::decay_t<decltype(p19)>& _p19_,\
const std::decay_t<decltype(p20)>& _p20_,\
const std::decay_t<decltype(p21)>& _p21_,\
const std::decay_t<decltype(p22)>& _p22_,\
const std::decay_t<decltype(p23)>& _p23_) [[hc]] {\
kernel_name(\
_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_,\
_p9_, _p10_, _p11_, _p12_, _p13_, _p14_, _p15_, _p16_, _p17_,\
_p18_, _p19_, _p20_, _p21_, _p22_, _p23_);\
}
#define make_kernel_lambda_hip_24(\
kernel_name,\
p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, p10, p11, p12, p13, p14, p15,\
p16, p17, p18, p19, p20, p21, p22)\
[](const std::decay_t<decltype(p0)>& _p0_,\
const std::decay_t<decltype(p1)>& _p1_,\
const std::decay_t<decltype(p2)>& _p2_,\
const std::decay_t<decltype(p3)>& _p3_,\
const std::decay_t<decltype(p4)>& _p4_,\
const std::decay_t<decltype(p5)>& _p5_,\
const std::decay_t<decltype(p6)>& _p6_,\
const std::decay_t<decltype(p7)>& _p7_,\
const std::decay_t<decltype(p8)>& _p8_,\
const std::decay_t<decltype(p9)>& _p9_,\
const std::decay_t<decltype(p10)>& _p10_,\
const std::decay_t<decltype(p11)>& _p11_,\
const std::decay_t<decltype(p12)>& _p12_,\
const std::decay_t<decltype(p13)>& _p13_,\
const std::decay_t<decltype(p14)>& _p14_,\
const std::decay_t<decltype(p15)>& _p15_,\
const std::decay_t<decltype(p16)>& _p16_,\
const std::decay_t<decltype(p17)>& _p17_,\
const std::decay_t<decltype(p18)>& _p18_,\
const std::decay_t<decltype(p19)>& _p19_,\
const std::decay_t<decltype(p20)>& _p20_,\
const std::decay_t<decltype(p21)>& _p21_,\
const std::decay_t<decltype(p22)>& _p22_) [[hc]] {\
kernel_name(\
_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_,\
_p9_, _p10_, _p11_, _p12_, _p13_, _p14_, _p15_, _p16_, _p17_,\
_p18_, _p19_, _p20_, _p21_, _p22_);\
}
#define make_kernel_lambda_hip_23(\
kernel_name,\
p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, p10, p11, p12, p13, p14, p15,\
p16, p17, p18, p19, p20, p21)\
[](const std::decay_t<decltype(p0)>& _p0_,\
const std::decay_t<decltype(p1)>& _p1_,\
const std::decay_t<decltype(p2)>& _p2_,\
const std::decay_t<decltype(p3)>& _p3_,\
const std::decay_t<decltype(p4)>& _p4_,\
const std::decay_t<decltype(p5)>& _p5_,\
const std::decay_t<decltype(p6)>& _p6_,\
const std::decay_t<decltype(p7)>& _p7_,\
const std::decay_t<decltype(p8)>& _p8_,\
const std::decay_t<decltype(p9)>& _p9_,\
const std::decay_t<decltype(p10)>& _p10_,\
const std::decay_t<decltype(p11)>& _p11_,\
const std::decay_t<decltype(p12)>& _p12_,\
const std::decay_t<decltype(p13)>& _p13_,\
const std::decay_t<decltype(p14)>& _p14_,\
const std::decay_t<decltype(p15)>& _p15_,\
const std::decay_t<decltype(p16)>& _p16_,\
const std::decay_t<decltype(p17)>& _p17_,\
const std::decay_t<decltype(p18)>& _p18_,\
const std::decay_t<decltype(p19)>& _p19_,\
const std::decay_t<decltype(p20)>& _p20_,\
const std::decay_t<decltype(p21)>& _p21_) [[hc]] {\
kernel_name(\
_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_,\
_p9_, _p10_, _p11_, _p12_, _p13_, _p14_, _p15_, _p16_, _p17_,\
_p18_, _p19_, _p20_, _p21_);\
}
#define make_kernel_lambda_hip_22(\
kernel_name,\
p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, p10, p11, p12, p13, p14, p15,\
p16, p17, p18, p19, p20)\
[](const std::decay_t<decltype(p0)>& _p0_,\
const std::decay_t<decltype(p1)>& _p1_,\
const std::decay_t<decltype(p2)>& _p2_,\
const std::decay_t<decltype(p3)>& _p3_,\
const std::decay_t<decltype(p4)>& _p4_,\
const std::decay_t<decltype(p5)>& _p5_,\
const std::decay_t<decltype(p6)>& _p6_,\
const std::decay_t<decltype(p7)>& _p7_,\
const std::decay_t<decltype(p8)>& _p8_,\
const std::decay_t<decltype(p9)>& _p9_,\
const std::decay_t<decltype(p10)>& _p10_,\
const std::decay_t<decltype(p11)>& _p11_,\
const std::decay_t<decltype(p12)>& _p12_,\
const std::decay_t<decltype(p13)>& _p13_,\
const std::decay_t<decltype(p14)>& _p14_,\
const std::decay_t<decltype(p15)>& _p15_,\
const std::decay_t<decltype(p16)>& _p16_,\
const std::decay_t<decltype(p17)>& _p17_,\
const std::decay_t<decltype(p18)>& _p18_,\
const std::decay_t<decltype(p19)>& _p19_,\
const std::decay_t<decltype(p20)>& _p20_) [[hc]] {\
kernel_name(\
_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_,\
_p9_, _p10_, _p11_, _p12_, _p13_, _p14_, _p15_, _p16_, _p17_,\
_p18_, _p19_, _p20_);\
}
#define make_kernel_lambda_hip_21(\
kernel_name,\
p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, p10, p11, p12, p13, p14, p15,\
p16, p17, p18, p19)\
[](const std::decay_t<decltype(p0)>& _p0_,\
const std::decay_t<decltype(p1)>& _p1_,\
const std::decay_t<decltype(p2)>& _p2_,\
const std::decay_t<decltype(p3)>& _p3_,\
const std::decay_t<decltype(p4)>& _p4_,\
const std::decay_t<decltype(p5)>& _p5_,\
const std::decay_t<decltype(p6)>& _p6_,\
const std::decay_t<decltype(p7)>& _p7_,\
const std::decay_t<decltype(p8)>& _p8_,\
const std::decay_t<decltype(p9)>& _p9_,\
const std::decay_t<decltype(p10)>& _p10_,\
const std::decay_t<decltype(p11)>& _p11_,\
const std::decay_t<decltype(p12)>& _p12_,\
const std::decay_t<decltype(p13)>& _p13_,\
const std::decay_t<decltype(p14)>& _p14_,\
const std::decay_t<decltype(p15)>& _p15_,\
const std::decay_t<decltype(p16)>& _p16_,\
const std::decay_t<decltype(p17)>& _p17_,\
const std::decay_t<decltype(p18)>& _p18_,\
const std::decay_t<decltype(p19)>& _p19_) [[hc]] {\
kernel_name(\
_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_,\
_p9_, _p10_, _p11_, _p12_, _p13_, _p14_, _p15_, _p16_, _p17_,\
_p18_, _p19_);\
}
#define make_kernel_lambda_hip_20(\
kernel_name,\
p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, p10, p11, p12, p13, p14, p15,\
p16, p17, p18)\
[](const std::decay_t<decltype(p0)>& _p0_,\
const std::decay_t<decltype(p1)>& _p1_,\
const std::decay_t<decltype(p2)>& _p2_,\
const std::decay_t<decltype(p3)>& _p3_,\
const std::decay_t<decltype(p4)>& _p4_,\
const std::decay_t<decltype(p5)>& _p5_,\
const std::decay_t<decltype(p6)>& _p6_,\
const std::decay_t<decltype(p7)>& _p7_,\
const std::decay_t<decltype(p8)>& _p8_,\
const std::decay_t<decltype(p9)>& _p9_,\
const std::decay_t<decltype(p10)>& _p10_,\
const std::decay_t<decltype(p11)>& _p11_,\
const std::decay_t<decltype(p12)>& _p12_,\
const std::decay_t<decltype(p13)>& _p13_,\
const std::decay_t<decltype(p14)>& _p14_,\
const std::decay_t<decltype(p15)>& _p15_,\
const std::decay_t<decltype(p16)>& _p16_,\
const std::decay_t<decltype(p17)>& _p17_,\
const std::decay_t<decltype(p18)>& _p18_) [[hc]] {\
kernel_name(\
_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_,\
_p9_, _p10_, _p11_, _p12_, _p13_, _p14_, _p15_, _p16_, _p17_,\
_p18_);\
}
#define make_kernel_lambda_hip_19(\
kernel_name,\
p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, p10, p11, p12, p13, p14, p15,\
p16, p17)\
[](const std::decay_t<decltype(p0)>& _p0_,\
const std::decay_t<decltype(p1)>& _p1_,\
const std::decay_t<decltype(p2)>& _p2_,\
const std::decay_t<decltype(p3)>& _p3_,\
const std::decay_t<decltype(p4)>& _p4_,\
const std::decay_t<decltype(p5)>& _p5_,\
const std::decay_t<decltype(p6)>& _p6_,\
const std::decay_t<decltype(p7)>& _p7_,\
const std::decay_t<decltype(p8)>& _p8_,\
const std::decay_t<decltype(p9)>& _p9_,\
const std::decay_t<decltype(p10)>& _p10_,\
const std::decay_t<decltype(p11)>& _p11_,\
const std::decay_t<decltype(p12)>& _p12_,\
const std::decay_t<decltype(p13)>& _p13_,\
const std::decay_t<decltype(p14)>& _p14_,\
const std::decay_t<decltype(p15)>& _p15_,\
const std::decay_t<decltype(p16)>& _p16_,\
const std::decay_t<decltype(p17)>& _p17_) [[hc]] {\
kernel_name(\
_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_,\
_p9_, _p10_, _p11_, _p12_, _p13_, _p14_, _p15_, _p16_, _p17_);\
}
#define make_kernel_lambda_hip_18(\
kernel_name,\
p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, p10, p11, p12, p13, p14, p15,\
p16)\
[](const std::decay_t<decltype(p0)>& _p0_,\
const std::decay_t<decltype(p1)>& _p1_,\
const std::decay_t<decltype(p2)>& _p2_,\
const std::decay_t<decltype(p3)>& _p3_,\
const std::decay_t<decltype(p4)>& _p4_,\
const std::decay_t<decltype(p5)>& _p5_,\
const std::decay_t<decltype(p6)>& _p6_,\
const std::decay_t<decltype(p7)>& _p7_,\
const std::decay_t<decltype(p8)>& _p8_,\
const std::decay_t<decltype(p9)>& _p9_,\
const std::decay_t<decltype(p10)>& _p10_,\
const std::decay_t<decltype(p11)>& _p11_,\
const std::decay_t<decltype(p12)>& _p12_,\
const std::decay_t<decltype(p13)>& _p13_,\
const std::decay_t<decltype(p14)>& _p14_,\
const std::decay_t<decltype(p15)>& _p15_,\
const std::decay_t<decltype(p16)>& _p16_) [[hc]] {\
kernel_name(\
_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_,\
_p9_, _p10_, _p11_, _p12_, _p13_, _p14_, _p15_, _p16_);\
}
#define make_kernel_lambda_hip_17(\
kernel_name,\
p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, p10, p11, p12, p13, p14, p15)\
[](const std::decay_t<decltype(p0)>& _p0_,\
const std::decay_t<decltype(p1)>& _p1_,\
const std::decay_t<decltype(p2)>& _p2_,\
const std::decay_t<decltype(p3)>& _p3_,\
const std::decay_t<decltype(p4)>& _p4_,\
const std::decay_t<decltype(p5)>& _p5_,\
const std::decay_t<decltype(p6)>& _p6_,\
const std::decay_t<decltype(p7)>& _p7_,\
const std::decay_t<decltype(p8)>& _p8_,\
const std::decay_t<decltype(p9)>& _p9_,\
const std::decay_t<decltype(p10)>& _p10_,\
const std::decay_t<decltype(p11)>& _p11_,\
const std::decay_t<decltype(p12)>& _p12_,\
const std::decay_t<decltype(p13)>& _p13_,\
const std::decay_t<decltype(p14)>& _p14_,\
const std::decay_t<decltype(p15)>& _p15_) [[hc]] {\
kernel_name(\
_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_,\
_p9_, _p10_, _p11_, _p12_, _p13_, _p14_, _p15_);\
}
#define make_kernel_lambda_hip_16(\
kernel_name,\
p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, p10, p11, p12, p13, p14)\
[](const std::decay_t<decltype(p0)>& _p0_,\
const std::decay_t<decltype(p1)>& _p1_,\
const std::decay_t<decltype(p2)>& _p2_,\
const std::decay_t<decltype(p3)>& _p3_,\
const std::decay_t<decltype(p4)>& _p4_,\
const std::decay_t<decltype(p5)>& _p5_,\
const std::decay_t<decltype(p6)>& _p6_,\
const std::decay_t<decltype(p7)>& _p7_,\
const std::decay_t<decltype(p8)>& _p8_,\
const std::decay_t<decltype(p9)>& _p9_,\
const std::decay_t<decltype(p10)>& _p10_,\
const std::decay_t<decltype(p11)>& _p11_,\
const std::decay_t<decltype(p12)>& _p12_,\
const std::decay_t<decltype(p13)>& _p13_,\
const std::decay_t<decltype(p14)>& _p14_) [[hc]] {\
kernel_name(\
_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_,\
_p9_, _p10_, _p11_, _p12_, _p13_, _p14_);\
}
#define make_kernel_lambda_hip_15(\
kernel_name,\
p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, p10, p11, p12, p13)\
[](const std::decay_t<decltype(p0)>& _p0_,\
const std::decay_t<decltype(p1)>& _p1_,\
const std::decay_t<decltype(p2)>& _p2_,\
const std::decay_t<decltype(p3)>& _p3_,\
const std::decay_t<decltype(p4)>& _p4_,\
const std::decay_t<decltype(p5)>& _p5_,\
const std::decay_t<decltype(p6)>& _p6_,\
const std::decay_t<decltype(p7)>& _p7_,\
const std::decay_t<decltype(p8)>& _p8_,\
const std::decay_t<decltype(p9)>& _p9_,\
const std::decay_t<decltype(p10)>& _p10_,\
const std::decay_t<decltype(p11)>& _p11_,\
const std::decay_t<decltype(p12)>& _p12_,\
const std::decay_t<decltype(p13)>& _p13_) [[hc]] {\
kernel_name(\
_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_,\
_p9_, _p10_, _p11_, _p12_, _p13_);\
}
#define make_kernel_lambda_hip_14(\
kernel_name, p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, p10, p11, p12)\
[](const std::decay_t<decltype(p0)>& _p0_,\
const std::decay_t<decltype(p1)>& _p1_,\
const std::decay_t<decltype(p2)>& _p2_,\
const std::decay_t<decltype(p3)>& _p3_,\
const std::decay_t<decltype(p4)>& _p4_,\
const std::decay_t<decltype(p5)>& _p5_,\
const std::decay_t<decltype(p6)>& _p6_,\
const std::decay_t<decltype(p7)>& _p7_,\
const std::decay_t<decltype(p8)>& _p8_,\
const std::decay_t<decltype(p9)>& _p9_,\
const std::decay_t<decltype(p10)>& _p10_,\
const std::decay_t<decltype(p11)>& _p11_,\
const std::decay_t<decltype(p12)>& _p12_) [[hc]] {\
kernel_name(\
_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_,\
_p9_, _p10_, _p11_, _p12_);\
}
#define make_kernel_lambda_hip_13(\
kernel_name, p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, p10, p11)\
[](const std::decay_t<decltype(p0)>& _p0_,\
const std::decay_t<decltype(p1)>& _p1_,\
const std::decay_t<decltype(p2)>& _p2_,\
const std::decay_t<decltype(p3)>& _p3_,\
const std::decay_t<decltype(p4)>& _p4_,\
const std::decay_t<decltype(p5)>& _p5_,\
const std::decay_t<decltype(p6)>& _p6_,\
const std::decay_t<decltype(p7)>& _p7_,\
const std::decay_t<decltype(p8)>& _p8_,\
const std::decay_t<decltype(p9)>& _p9_,\
const std::decay_t<decltype(p10)>& _p10_,\
const std::decay_t<decltype(p11)>& _p11_) [[hc]] {\
kernel_name(\
_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_,\
_p9_, _p10_, _p11_);\
}
#define make_kernel_lambda_hip_12(\
kernel_name, p0, p1, p2, p3, p4, p5, p6, p7, p8, p9, p10)\
[](const std::decay_t<decltype(p0)>& _p0_,\
const std::decay_t<decltype(p1)>& _p1_,\
const std::decay_t<decltype(p2)>& _p2_,\
const std::decay_t<decltype(p3)>& _p3_,\
const std::decay_t<decltype(p4)>& _p4_,\
const std::decay_t<decltype(p5)>& _p5_,\
const std::decay_t<decltype(p6)>& _p6_,\
const std::decay_t<decltype(p7)>& _p7_,\
const std::decay_t<decltype(p8)>& _p8_,\
const std::decay_t<decltype(p9)>& _p9_,\
const std::decay_t<decltype(p10)>& _p10_) [[hc]] {\
kernel_name(\
_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_, _p9_,\
_p10_);\
}
#define make_kernel_lambda_hip_11(\
kernel_name, p0, p1, p2, p3, p4, p5, p6, p7, p8, p9)\
[](const std::decay_t<decltype(p0)>& _p0_,\
const std::decay_t<decltype(p1)>& _p1_,\
const std::decay_t<decltype(p2)>& _p2_,\
const std::decay_t<decltype(p3)>& _p3_,\
const std::decay_t<decltype(p4)>& _p4_,\
const std::decay_t<decltype(p5)>& _p5_,\
const std::decay_t<decltype(p6)>& _p6_,\
const std::decay_t<decltype(p7)>& _p7_,\
const std::decay_t<decltype(p8)>& _p8_,\
const std::decay_t<decltype(p9)>& _p9_) [[hc]] {\
kernel_name(\
_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_, _p9_);\
}
#define make_kernel_lambda_hip_10(\
kernel_name, p0, p1, p2, p3, p4, p5, p6, p7, p8)\
[](const std::decay_t<decltype(p0)>& _p0_,\
const std::decay_t<decltype(p1)>& _p1_,\
const std::decay_t<decltype(p2)>& _p2_,\
const std::decay_t<decltype(p3)>& _p3_,\
const std::decay_t<decltype(p4)>& _p4_,\
const std::decay_t<decltype(p5)>& _p5_,\
const std::decay_t<decltype(p6)>& _p6_,\
const std::decay_t<decltype(p7)>& _p7_,\
const std::decay_t<decltype(p8)>& _p8_) [[hc]] {\
kernel_name(_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_, _p8_);\
}
#define make_kernel_lambda_hip_9(\
kernel_name, p0, p1, p2, p3, p4, p5, p6, p7)\
[](const std::decay_t<decltype(p0)>& _p0_,\
const std::decay_t<decltype(p1)>& _p1_,\
const std::decay_t<decltype(p2)>& _p2_,\
const std::decay_t<decltype(p3)>& _p3_,\
const std::decay_t<decltype(p4)>& _p4_,\
const std::decay_t<decltype(p5)>& _p5_,\
const std::decay_t<decltype(p6)>& _p6_,\
const std::decay_t<decltype(p7)>& _p7_) [[hc]] {\
kernel_name(_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_, _p7_);\
}
#define make_kernel_lambda_hip_8(kernel_name, p0, p1, p2, p3, p4, p5, p6)\
[](const std::decay_t<decltype(p0)>& _p0_,\
const std::decay_t<decltype(p1)>& _p1_,\
const std::decay_t<decltype(p2)>& _p2_,\
const std::decay_t<decltype(p3)>& _p3_,\
const std::decay_t<decltype(p4)>& _p4_,\
const std::decay_t<decltype(p5)>& _p5_,\
const std::decay_t<decltype(p6)>& _p6_) [[hc]] {\
kernel_name(_p0_, _p1_, _p2_, _p3_, _p4_, _p5_, _p6_);\
}
#define make_kernel_lambda_hip_7(kernel_name, p0, p1, p2, p3, p4, p5)\
[](const std::decay_t<decltype(p0)>& _p0_,\
const std::decay_t<decltype(p1)>& _p1_,\
const std::decay_t<decltype(p2)>& _p2_,\
const std::decay_t<decltype(p3)>& _p3_,\
const std::decay_t<decltype(p4)>& _p4_,\
const std::decay_t<decltype(p5)>& _p5_) [[hc]] {\
kernel_name(_p0_, _p1_, _p2_, _p3_, _p4_, _p5_);\
}
#define make_kernel_lambda_hip_6(kernel_name, p0, p1, p2, p3, p4)\
[](const std::decay_t<decltype(p0)>& _p0_,\
const std::decay_t<decltype(p1)>& _p1_,\
const std::decay_t<decltype(p2)>& _p2_,\
const std::decay_t<decltype(p3)>& _p3_,\
const std::decay_t<decltype(p4)>& _p4_) [[hc]] {\
kernel_name(_p0_, _p1_, _p2_, _p3_, _p4_);\
}
#define make_kernel_lambda_hip_5(kernel_name, p0, p1, p2, p3)\
[](const std::decay_t<decltype(p0)>& _p0_,\
const std::decay_t<decltype(p1)>& _p1_,\
const std::decay_t<decltype(p2)>& _p2_,\
const std::decay_t<decltype(p3)>& _p3_) [[hc]] {\
kernel_name(_p0_, _p1_, _p2_, _p3_);\
}
#define make_kernel_lambda_hip_4(kernel_name, p0, p1, p2)\
[](const std::decay_t<decltype(p0)>& _p0_,\
const std::decay_t<decltype(p1)>& _p1_,\
const std::decay_t<decltype(p2)>& _p2_) [[hc]] {\
kernel_name(_p0_, _p1_, _p2_);\
}
#define make_kernel_lambda_hip_3(kernel_name, p0, p1)\
[](const std::decay_t<decltype(p0)>& _p0_,\
const std::decay_t<decltype(p1)>& _p1_) [[hc]] {\
kernel_name(_p0_, _p1_);\
}
#define make_kernel_lambda_hip_2(kernel_name, p0)\
[](const std::decay_t<decltype(p0)>& _p0_) [[hc]] {\
kernel_name(_p0_);\
}
#define make_kernel_lambda_hip_1(kernel_name)\
[]() [[hc]] { return kernel_name(hipLaunchParm{}); }
#define make_kernel_lambda_hip_(...)\
overload_macro_hip_(make_kernel_lambda_hip_, __VA_ARGS__)
#define hipLaunchKernelGGL(\
kernel_name,\
num_blocks,\
dim_blocks,\
group_mem_bytes,\
stream,\
...)\
do {\
hip_impl::grid_launch_hip_(\
num_blocks,\
dim_blocks,\
group_mem_bytes,\
stream,\
#kernel_name,\
make_kernel_lambda_hip_(kernel_name, __VA_ARGS__),\
##__VA_ARGS__);\
} while(0)
#define hipLaunchKernel(\
kernel_name,\
num_blocks,\
dim_blocks,\
group_mem_bytes,\
stream,\
...)\
do {\
hipLaunchKernelGGL(\
kernel_name,\
num_blocks,\
dim_blocks,\
group_mem_bytes,\
stream,\
hipLaunchParm{},\
##__VA_ARGS__);\
} while(0)
}
#endif //GENERIC_GRID_LAUNCH
projects/hip/include/hip/hcc_detail/helpers.hpp
+134
Datei anzeigen
@@ -0,0 +1,134 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#pragma once
#include "concepts.hpp"
#include <type_traits> // For std::conditional, std::decay, std::enable_if,
// std::false_type, std result_of and std::true_type.
#include <utility> // For std::declval.
namespace std
{ // TODO: these should be removed as soon as possible.
#if (__cplusplus < 201406L)
#if (__cplusplus < 201402L)
template<bool cond, typename T = void>
using enable_if_t = typename enable_if<cond, T>::type;
template<bool cond, typename T, typename U>
using conditional_t = typename conditional<cond, T, U>::type;
template<typename T>
using decay_t = typename decay<T>::type;
template<FunctionalProcedure F, typename... Ts>
using result_of_t = typename result_of<F(Ts...)>::type;
template<typename T>
using remove_reference_t = typename remove_reference<T>::type;
#endif
#endif
}
namespace hip_impl
{
template<typename...>
using void_t_ = void;
#if (__cplusplus < 201402L)
template<
FunctionalProcedure F,
unsigned int n = 0u,
typename = void>
struct is_callable_impl : is_callable_impl<F, n + 1u> {};
// Pointer to member function, call through non-pointer.
template<FunctionalProcedure F, typename C, typename... Ts>
struct is_callable_impl<
F(C, Ts...),
0u,
void_t_<decltype((std::declval<C>().*std::declval<F>())(
std::declval<Ts>()...))>
> : std::true_type {};
// Pointer to member function, call through pointer.
template<FunctionalProcedure F, typename C, typename... Ts>
struct is_callable_impl<
F(C, Ts...),
1u,
void_t_<decltype(((*std::declval<C>()).*std::declval<F>())(
std::declval<Ts>()...))>
> : std::true_type {};
// Pointer to member data, call through non-pointer, no args.
template<FunctionalProcedure F, typename C>
struct is_callable_impl<
F(C),
2u,
void_t_<decltype(std::declval<C>().*std::declval<F>())>
> : std::true_type {};
// Pointer to member data, call through pointer, no args.
template<FunctionalProcedure F, typename C>
struct is_callable_impl<
F(C),
3u,
void_t_<decltype(*std::declval<C>().*std::declval<F>())>
> : std::true_type {};
// General call, n args.
template<FunctionalProcedure F, typename... Ts>
struct is_callable_impl<
F(Ts...),
4u,
void_t_<decltype(std::declval<F>()(std::declval<Ts>()...))>
> : std::true_type {};
// Not callable.
template<FunctionalProcedure F>
struct is_callable_impl<F, 5u> : std::false_type {};
template<typename Call>
struct is_callable : is_callable_impl<Call> {};
#else
template<typename, typename = void>
struct is_callable_impl : std::false_type {};
template<FunctionalProcedure F, typename... Ts>
struct is_callable_impl<
F(Ts...),
void_t_<std::result_of_t<F(Ts...)>>> : std::true_type {};
#endif
#define count_macro_args_impl_hip_(\
_0, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, _15,\
_16, _17, _18, _19, _20, _21, _22, _23, _24, _25, _26, _27, _28, _29,\
_30, _31, _n, ...)\
_n
#define count_macro_args_hip_(...)\
count_macro_args_impl_hip_(\
, ##__VA_ARGS__, 31, 30, 29, 28, 27, 26, 25, 24, 23, 22, 21, 20, 19,\
18, 17, 16, 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0)
#define overloaded_macro_expand_hip_(macro, arg_cnt) macro##arg_cnt
#define overload_macro_impl_hip_(macro, arg_cnt)\
overloaded_macro_expand_hip_(macro, arg_cnt)
#define overload_macro_hip_(macro, ...)\
overload_macro_impl_hip_(macro, count_macro_args_hip_(__VA_ARGS__))\
(__VA_ARGS__)
}
projects/hip/include/hip/hcc_detail/hip_complex.h
+181 -36
Datei anzeigen
@@ -1,5 +1,5 @@
/*
Copyright (c) 2015-2016 Advanced Micro Devices, Inc. All rights reserved.
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
@@ -20,53 +20,202 @@ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#ifndef HIPCOMPLEX_H
#define HIPCOMPLEX_H
#ifndef HIP_INCLUDE_HIP_HCC_DETAIL_HIP_COMPLEX_H
#define HIP_INCLUDE_HIP_HCC_DETAIL_HIP_COMPLEX_H
typedef struct{
float x;
float y;
}hipFloatComplex;
#include "hip/hcc_detail/hip_vector_types.h"
#include <math.h>
__device__ static inline float hipCrealf(hipFloatComplex z){
#if __cplusplus
#define COMPLEX_ADD_OP_OVERLOAD(type) \
__device__ __host__ static inline type operator + (const type& lhs, const type& rhs) { \
type ret; \
ret.x = lhs.x + rhs.x ; \
ret.y = lhs.y + rhs.y ; \
return ret; \
}
#define COMPLEX_SUB_OP_OVERLOAD(type) \
__device__ __host__ static inline type operator - (const type& lhs, const type& rhs) { \
type ret; \
ret.x = lhs.x - rhs.x; \
ret.y = lhs.y - rhs.y; \
return ret; \
}
#define COMPLEX_MUL_OP_OVERLOAD(type) \
__device__ __host__ static inline type operator * (const type& lhs, const type& rhs) { \
type ret; \
ret.x = lhs.x * rhs.x - lhs.y * rhs.y; \
ret.y = lhs.x * rhs.y + lhs.y * rhs.x; \
return ret; \
}
#define COMPLEX_DIV_OP_OVERLOAD(type) \
__device__ __host__ static inline type operator / (const type& lhs, const type& rhs) { \
type ret; \
ret.x = (lhs.x * rhs.x + lhs.y * rhs.y); \
ret.y = (rhs.x * lhs.y - lhs.x * rhs.y); \
ret.x = ret.x / (rhs.x * rhs.x + rhs.y * rhs.y); \
ret.y = ret.y / (rhs.x * rhs.x + rhs.y * rhs.y); \
return ret; \
}
#define COMPLEX_ADD_PREOP_OVERLOAD(type) \
__device__ __host__ static inline type& operator += (type& lhs, const type& rhs) { \
lhs.x += rhs.x; \
lhs.y += rhs.y; \
return lhs; \
}
#define COMPLEX_SUB_PREOP_OVERLOAD(type) \
__device__ __host__ static inline type& operator -= (type& lhs, const type& rhs) { \
lhs.x -= rhs.x; \
lhs.y -= rhs.y; \
return lhs; \
}
#define COMPLEX_MUL_PREOP_OVERLOAD(type) \
__device__ __host__ static inline type& operator *= (type& lhs, const type& rhs) { \
lhs = lhs * rhs; \
return lhs; \
}
#define COMPLEX_DIV_PREOP_OVERLOAD(type) \
__device__ __host__ static inline type& operator /= (type& lhs, const type& rhs) { \
lhs = lhs / rhs; \
return lhs; \
}
#define COMPLEX_SCALAR_PRODUCT(type, type1) \
__device__ __host__ static inline type operator * (const type& lhs, type1 rhs) { \
type ret; \
ret.x = lhs.x * rhs; \
ret.y = lhs.y * rhs; \
return ret; \
}
#endif
struct hipFloatComplex {
#ifdef __cplusplus
public:
__device__ __host__ hipFloatComplex() : x(0.0f), y(0.0f) {}
__device__ __host__ hipFloatComplex(float x) : x(x), y(0.0f) {}
__device__ __host__ hipFloatComplex(float x, float y) : x(x), y(y) {}
MAKE_COMPONENT_CONSTRUCTOR_TWO_COMPONENT(hipFloatComplex, unsigned short)
MAKE_COMPONENT_CONSTRUCTOR_TWO_COMPONENT(hipFloatComplex, signed short)
MAKE_COMPONENT_CONSTRUCTOR_TWO_COMPONENT(hipFloatComplex, unsigned int)
MAKE_COMPONENT_CONSTRUCTOR_TWO_COMPONENT(hipFloatComplex, signed int)
MAKE_COMPONENT_CONSTRUCTOR_TWO_COMPONENT(hipFloatComplex, double)
MAKE_COMPONENT_CONSTRUCTOR_TWO_COMPONENT(hipFloatComplex, unsigned long)
MAKE_COMPONENT_CONSTRUCTOR_TWO_COMPONENT(hipFloatComplex, signed long)
MAKE_COMPONENT_CONSTRUCTOR_TWO_COMPONENT(hipFloatComplex, unsigned long long)
MAKE_COMPONENT_CONSTRUCTOR_TWO_COMPONENT(hipFloatComplex, signed long long)
#endif
float x, y;
} __attribute__((aligned(8)));
struct hipDoubleComplex {
#ifdef __cplusplus
public:
__device__ __host__ hipDoubleComplex() : x(0.0f), y(0.0f) {}
__device__ __host__ hipDoubleComplex(double x) : x(x), y(0.0f) {}
__device__ __host__ hipDoubleComplex(double x, double y) : x(x), y(y) {}
MAKE_COMPONENT_CONSTRUCTOR_TWO_COMPONENT(hipDoubleComplex, unsigned short)
MAKE_COMPONENT_CONSTRUCTOR_TWO_COMPONENT(hipDoubleComplex, signed short)
MAKE_COMPONENT_CONSTRUCTOR_TWO_COMPONENT(hipDoubleComplex, unsigned int)
MAKE_COMPONENT_CONSTRUCTOR_TWO_COMPONENT(hipDoubleComplex, signed int)
MAKE_COMPONENT_CONSTRUCTOR_TWO_COMPONENT(hipDoubleComplex, float)
MAKE_COMPONENT_CONSTRUCTOR_TWO_COMPONENT(hipDoubleComplex, unsigned long)
MAKE_COMPONENT_CONSTRUCTOR_TWO_COMPONENT(hipDoubleComplex, signed long)
MAKE_COMPONENT_CONSTRUCTOR_TWO_COMPONENT(hipDoubleComplex, unsigned long long)
MAKE_COMPONENT_CONSTRUCTOR_TWO_COMPONENT(hipDoubleComplex, signed long long)
#endif
double x, y;
} __attribute__((aligned(16)));
#if __cplusplus
COMPLEX_ADD_OP_OVERLOAD(hipFloatComplex)
COMPLEX_SUB_OP_OVERLOAD(hipFloatComplex)
COMPLEX_MUL_OP_OVERLOAD(hipFloatComplex)
COMPLEX_DIV_OP_OVERLOAD(hipFloatComplex)
COMPLEX_ADD_PREOP_OVERLOAD(hipFloatComplex)
COMPLEX_SUB_PREOP_OVERLOAD(hipFloatComplex)
COMPLEX_MUL_PREOP_OVERLOAD(hipFloatComplex)
COMPLEX_DIV_PREOP_OVERLOAD(hipFloatComplex)
COMPLEX_SCALAR_PRODUCT(hipFloatComplex, unsigned short)
COMPLEX_SCALAR_PRODUCT(hipFloatComplex, signed short)
COMPLEX_SCALAR_PRODUCT(hipFloatComplex, unsigned int)
COMPLEX_SCALAR_PRODUCT(hipFloatComplex, signed int)
COMPLEX_SCALAR_PRODUCT(hipFloatComplex, float)
COMPLEX_SCALAR_PRODUCT(hipFloatComplex, unsigned long)
COMPLEX_SCALAR_PRODUCT(hipFloatComplex, signed long)
COMPLEX_SCALAR_PRODUCT(hipFloatComplex, double)
COMPLEX_SCALAR_PRODUCT(hipFloatComplex, signed long long)
COMPLEX_SCALAR_PRODUCT(hipFloatComplex, unsigned long long)
COMPLEX_ADD_OP_OVERLOAD(hipDoubleComplex)
COMPLEX_SUB_OP_OVERLOAD(hipDoubleComplex)
COMPLEX_MUL_OP_OVERLOAD(hipDoubleComplex)
COMPLEX_DIV_OP_OVERLOAD(hipDoubleComplex)
COMPLEX_ADD_PREOP_OVERLOAD(hipDoubleComplex)
COMPLEX_SUB_PREOP_OVERLOAD(hipDoubleComplex)
COMPLEX_MUL_PREOP_OVERLOAD(hipDoubleComplex)
COMPLEX_DIV_PREOP_OVERLOAD(hipDoubleComplex)
COMPLEX_SCALAR_PRODUCT(hipDoubleComplex, unsigned short)
COMPLEX_SCALAR_PRODUCT(hipDoubleComplex, signed short)
COMPLEX_SCALAR_PRODUCT(hipDoubleComplex, unsigned int)
COMPLEX_SCALAR_PRODUCT(hipDoubleComplex, signed int)
COMPLEX_SCALAR_PRODUCT(hipDoubleComplex, float)
COMPLEX_SCALAR_PRODUCT(hipDoubleComplex, unsigned long)
COMPLEX_SCALAR_PRODUCT(hipDoubleComplex, signed long)
COMPLEX_SCALAR_PRODUCT(hipDoubleComplex, double)
COMPLEX_SCALAR_PRODUCT(hipDoubleComplex, signed long long)
COMPLEX_SCALAR_PRODUCT(hipDoubleComplex, unsigned long long)
#endif
__device__ __host__ static inline float hipCrealf(hipFloatComplex z){
return z.x;
}
__device__ static inline float hipCimagf(hipFloatComplex z){
__device__ __host__ static inline float hipCimagf(hipFloatComplex z){
return z.y;
}
__device__ static inline hipFloatComplex make_hipFloatComplex(float a, float b){
__device__ __host__ static inline hipFloatComplex make_hipFloatComplex(float a, float b){
hipFloatComplex z;
z.x = a;
z.y = b;
return z;
}
__device__ static inline hipFloatComplex hipConjf(hipFloatComplex z){
__device__ __host__ static inline hipFloatComplex hipConjf(hipFloatComplex z){
hipFloatComplex ret;
ret.x = z.x;
ret.y = -z.y;
return ret;
}
__device__ static inline float hipCsqabsf(hipFloatComplex z){
__device__ __host__ static inline float hipCsqabsf(hipFloatComplex z){
return z.x * z.x + z.y * z.y;
}
__device__ static inline hipFloatComplex hipCaddf(hipFloatComplex p, hipFloatComplex q){
__device__ __host__ static inline hipFloatComplex hipCaddf(hipFloatComplex p, hipFloatComplex q){
return make_hipFloatComplex(p.x + q.x, p.y + q.y);
}
__device__ static inline hipFloatComplex hipCsubf(hipFloatComplex p, hipFloatComplex q){
__device__ __host__ static inline hipFloatComplex hipCsubf(hipFloatComplex p, hipFloatComplex q){
return make_hipFloatComplex(p.x - q.x, p.y - q.y);
}
__device__ static inline hipFloatComplex hipCmulf(hipFloatComplex p, hipFloatComplex q){
__device__ __host__ static inline hipFloatComplex hipCmulf(hipFloatComplex p, hipFloatComplex q){
return make_hipFloatComplex(p.x * q.x - p.y * q.y, p.y * q.x + p.x * q.y);
}
__device__ static inline hipFloatComplex hipCdivf(hipFloatComplex p, hipFloatComplex q){
__device__ __host__ static inline hipFloatComplex hipCdivf(hipFloatComplex p, hipFloatComplex q){
float sqabs = hipCsqabsf(q);
hipFloatComplex ret;
ret.x = (p.x * q.x + p.y * q.y)/sqabs;
@@ -74,55 +223,51 @@ __device__ static inline hipFloatComplex hipCdivf(hipFloatComplex p, hipFloatCom
return ret;
}
__device__ static inline float hipCabsf(hipFloatComplex z){
__device__ __host__ static inline float hipCabsf(hipFloatComplex z){
return sqrtf(hipCsqabsf(z));
}
typedef struct{
double x;
double y;
}hipDoubleComplex;
__device__ static inline double hipCreal(hipDoubleComplex z){
__device__ __host__ static inline double hipCreal(hipDoubleComplex z){
return z.x;
}
__device__ static inline double hipCimag(hipDoubleComplex z){
__device__ __host__ static inline double hipCimag(hipDoubleComplex z){
return z.y;
}
__device__ static inline hipDoubleComplex make_hipDoubleComplex(double a, double b){
__device__ __host__ static inline hipDoubleComplex make_hipDoubleComplex(double a, double b){
hipDoubleComplex z;
z.x = a;
z.y = b;
return z;
}
__device__ static inline hipDoubleComplex hipConj(hipDoubleComplex z){
__device__ __host__ static inline hipDoubleComplex hipConj(hipDoubleComplex z){
hipDoubleComplex ret;
ret.x = z.x;
ret.y = z.y;
return ret;
}
__device__ static inline double hipCsqabs(hipDoubleComplex z){
__device__ __host__ static inline double hipCsqabs(hipDoubleComplex z){
return z.x * z.x + z.y * z.y;
}
__device__ static inline hipDoubleComplex hipCadd(hipDoubleComplex p, hipDoubleComplex q){
__device__ __host__ static inline hipDoubleComplex hipCadd(hipDoubleComplex p, hipDoubleComplex q){
return make_hipDoubleComplex(p.x + q.x, p.y + q.y);
}
__device__ static inline hipDoubleComplex hipCsub(hipDoubleComplex p, hipDoubleComplex q){
__device__ __host__ static inline hipDoubleComplex hipCsub(hipDoubleComplex p, hipDoubleComplex q){
return make_hipDoubleComplex(p.x - q.x, p.y - q.y);
}
__device__ static inline hipDoubleComplex hipCmul(hipDoubleComplex p, hipDoubleComplex q){
__device__ __host__ static inline hipDoubleComplex hipCmul(hipDoubleComplex p, hipDoubleComplex q){
return make_hipDoubleComplex(p.x * q.x - p.y * q.y, p.y * q.x + p.x * q.y);
}
__device__ static inline hipDoubleComplex hipCdiv(hipDoubleComplex p, hipDoubleComplex q){
__device__ __host__ static inline hipDoubleComplex hipCdiv(hipDoubleComplex p, hipDoubleComplex q){
double sqabs = hipCsqabs(q);
hipDoubleComplex ret;
ret.x = (p.x * q.x + p.y * q.y)/sqabs;
@@ -130,28 +275,28 @@ __device__ static inline hipDoubleComplex hipCdiv(hipDoubleComplex p, hipDoubleC
return ret;
}
__device__ static inline double hipCabs(hipDoubleComplex z){
__device__ __host__ static inline double hipCabs(hipDoubleComplex z){
return sqrtf(hipCsqabs(z));
}
typedef hipFloatComplex hipComplex;
__device__ static inline hipComplex make_hipComplex(float x,
__device__ __host__ static inline hipComplex make_hipComplex(float x,
float y){
return make_hipFloatComplex(x, y);
}
__device__ static inline hipFloatComplex hipComplexDoubleToFloat
__device__ __host__ static inline hipFloatComplex hipComplexDoubleToFloat
(hipDoubleComplex z){
return make_hipFloatComplex((float)z.x, (float)z.y);
}
__device__ static inline hipDoubleComplex hipComplexFloatToDouble
__device__ __host__ static inline hipDoubleComplex hipComplexFloatToDouble
(hipFloatComplex z){
return make_hipDoubleComplex((double)z.x, (double)z.y);
}
__device__ static inline hipComplex hipCfmaf(hipComplex p, hipComplex q, hipComplex r){
__device__ __host__ static inline hipComplex hipCfmaf(hipComplex p, hipComplex q, hipComplex r){
float real = (p.x * q.x) + r.x;
float imag = (q.x * p.y) + r.y;
@@ -161,7 +306,7 @@ __device__ static inline hipComplex hipCfmaf(hipComplex p, hipComplex q, hipComp
return make_hipComplex(real, imag);
}
__device__ static inline hipDoubleComplex hipCfma(hipDoubleComplex p, hipDoubleComplex q, hipDoubleComplex r){
__device__ __host__ static inline hipDoubleComplex hipCfma(hipDoubleComplex p, hipDoubleComplex q, hipDoubleComplex r){
float real = (p.x * q.x) + r.x;
float imag = (q.x * p.y) + r.y;
projects/hip/include/hip/hcc_detail/hip_fp16.h
+427 -118
Datei anzeigen
@@ -1,5 +1,5 @@
/*
Copyright (c) 2015-2016 Advanced Micro Devices, Inc. All rights reserved.
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
@@ -20,158 +20,467 @@ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#ifndef HIP_FP16_H
#define HIP_FP16_H
#ifndef HIP_INCLUDE_HIP_HCC_DETAIL_HIP_FP16_H
#define HIP_INCLUDE_HIP_HCC_DETAIL_HIP_FP16_H
#include "hip/hip_runtime.h"
#include "hip/hcc_detail/hip_vector_types.h"
typedef struct{
unsigned x: 16;
} __half;
#if __clang_major__ > 3
typedef __fp16 __half;
typedef struct __attribute__((aligned(4))){
__half p,q;
union {
__half p[2];
unsigned int q;
};
} __half2;
typedef __half half;
typedef __half2 half2;
/*
Arithmetic functions
Half Arithmetic Functions
*/
__device__ __half __hadd(const __half a, const __half b);
__device__ __half __hadd_sat(const __half a, const __half b);
__device__ __half __hfma(const __half a, const __half b, const __half c);
__device__ __half __hfma_sat(const __half a, const __half b, const __half c);
__device__ __half __hmul(const __half a, const __half b);
__device__ __half __hmul_sat(const __half a, const __half b);
__device__ __half __hneq(const __half a);
__device__ __half __hsub(const __half a, const __half b);
__device__ __half __hsub_sat(const __half a, const __half b);
__device__ __half __hadd_sat(__half a, __half b);
__device__ __half __hfma(__half a, __half b, __half c);
__device__ __half __hfma_sat(__half a, __half b, __half c);
__device__ __half __hmul(__half a, __half b);
__device__ __half __hmul_sat(__half a, __half b);
__device__ __half __hneg(__half a);
__device__ __half __hsub(__half a, __half b);
__device__ __half __hsub_sat(__half a, __half b);
__device__ __half hdiv(__half a, __half b);
/*
Half2 Arithmetic Instructions
Half2 Arithmetic Functions
*/
__device__ __half2 __hadd2(const __half2 a, const __half2 b);
__device__ __half2 __hadd2_sat(const __half2 a, const __half2 b);
__device__ __half2 __hfma2(const __half2 a, const __half2 b, const __half2 c);
__device__ __half2 __hfma2_sat(const __half2 a, const __half2 b, const __half2 c);
__device__ __half2 __hmul2(const __half2 a, const __half2 b);
__device__ __half2 __hmul2_sat(const __half2 a, const __half2 b);
__device__ __half2 __hneq2(const __half2 a);
__device__ __half2 __hsub2(const __half2 a, const __half2 b);
__device__ __half2 __hsub2_sat(const __half2 a, const __half2 b);
__device__ static __half2 __hadd2(__half2 a, __half2 b);
__device__ static __half2 __hadd2_sat(__half2 a, __half2 b);
__device__ static __half2 __hfma2(__half2 a, __half2 b, __half2 c);
__device__ static __half2 __hfma2_sat(__half2 a, __half2 b, __half2 c);
__device__ static __half2 __hmul2(__half2 a, __half2 b);
__device__ static __half2 __hmul2_sat(__half2 a, __half2 b);
__device__ static __half2 __hsub2(__half2 a, __half2 b);
__device__ static __half2 __hneg2(__half2 a);
__device__ static __half2 __hsub2_sat(__half2 a, __half2 b);
__device__ static __half2 h2div(__half2 a, __half2 b);
/*
Half Cmps
Half Comparision Functions
*/
__device__ bool __heq(const __half a, const __half b);
__device__ bool __hge(const __half a, const __half b);
__device__ bool __hgt(const __half a, const __half b);
__device__ bool __hisinf(const __half a);
__device__ bool __hisnan(const __half a);
__device__ bool __hle(const __half a, const __half b);
__device__ bool __hlt(const __half a, const __half b);
__device__ bool __hne(const __half a, const __half b);
__device__ bool __heq(__half a, __half b);
__device__ bool __hge(__half a, __half b);
__device__ bool __hgt(__half a, __half b);
__device__ bool __hisinf(__half a);
__device__ bool __hisnan(__half a);
__device__ bool __hle(__half a, __half b);
__device__ bool __hlt(__half a, __half b);
__device__ bool __hne(__half a, __half b);
/*
Half2 Cmps
Half2 Comparision Functions
*/
__device__ bool __hbeq2(const __half2 a, const __half2 b);
__device__ bool __hbge2(const __half2 a, const __half2 b);
__device__ bool __hbgt2(const __half2 a, const __half2 b);
__device__ bool __hble2(const __half2 a, const __half2 b);
__device__ bool __hblt2(const __half2 a, const __half2 b);
__device__ bool __hbne2(const __half2 a, const __half2 b);
__device__ __half2 __heq2(const __half2 a, const __half2 b);
__device__ __half2 __hge2(const __half2 a, const __half2 b);
__device__ __half2 __hgt2(const __half2 a, const __half2 b);
__device__ __half2 __hisnan2(const __half2 a);
__device__ __half2 __hle2(const __half2 a, const __half2 b);
__device__ __half2 __hlt2(const __half2 a, const __half2 b);
__device__ __half2 __hne2(const __half2 a, const __half2 b);
__device__ bool __hbeq2(__half2 a, __half2 b);
__device__ bool __hbge2(__half2 a, __half2 b);
__device__ bool __hbgt2(__half2 a, __half2 b);
__device__ bool __hble2(__half2 a, __half2 b);
__device__ bool __hblt2(__half2 a, __half2 b);
__device__ bool __hbne2(__half2 a, __half2 b);
__device__ __half2 __heq2(__half2 a, __half2 b);
__device__ __half2 __hge2(__half2 a, __half2 b);
__device__ __half2 __hgt2(__half2 a, __half2 b);
__device__ __half2 __hisnan2(__half2 a);
__device__ __half2 __hle2(__half2 a, __half2 b);
__device__ __half2 __hlt2(__half2 a, __half2 b);
__device__ __half2 __hne2(__half2 a, __half2 b);
/*
Half Cnvs and Data Mvmnt
Half Math Functions
*/
__device__ __half2 __float22half2_rn(const float2 a);
__device__ static __half hceil(const __half h);
__device__ static __half hcos(const __half h);
__device__ static __half hexp(const __half h);
__device__ static __half hexp10(const __half h);
__device__ static __half hexp2(const __half h);
__device__ static __half hfloor(const __half h);
__device__ static __half hlog(const __half h);
__device__ static __half hlog10(const __half h);
__device__ static __half hlog2(const __half h);
//__device__ static __half hrcp(const __half h);
__device__ static __half hrint(const __half h);
__device__ static __half hsin(const __half h);
__device__ static __half hsqrt(const __half a);
__device__ static __half htrunc(const __half a);
__device__ __half __float2half(const float a);
/*
Half2 Math Functions
*/
__device__ __half2 __float2half2_rn(const float a);
__device__ static __half2 h2ceil(const __half2 h);
__device__ static __half2 h2exp(const __half2 h);
__device__ static __half2 h2exp10(const __half2 h);
__device__ static __half2 h2exp2(const __half2 h);
__device__ static __half2 h2floor(const __half2 h);
__device__ static __half2 h2log(const __half2 h);
__device__ static __half2 h2log10(const __half2 h);
__device__ static __half2 h2log2(const __half2 h);
__device__ static __half2 h2rcp(const __half2 h);
__device__ static __half2 h2rsqrt(const __half2 h);
__device__ static __half2 h2sin(const __half2 h);
__device__ static __half2 h2sqrt(const __half2 h);
__device__ __half2 __floats2half2_rn(const float a, const float b);
/*
Half Conversion And Data Movement
*/
__device__ float2 __half22float2(const __half2 a);
__device__ float __half2float(const __half a);
__device__ __half2 __half2half2(const __half a);
__device__ __half2 __halves2half2(const __half a, const __half b);
__device__ float __high2float(const __half2 a);
__device__ __half __high2half(const __half2 a);
__device__ __half2 __high2half2(const __half2 a);
__device__ __half2 __highs2half2(const __half2 a, const __half2 b);
__device__ float __low2float(const __half2 a);
__device__ __half2 __float22half2_rn(const float2 a);
__device__ __half __float2half(const float a);
__device__ __half2 __float2half2_rn(const float a);
__device__ __half __float2half_rd(const float a);
__device__ __half __float2half_rn(const float a);
__device__ __half __float2half_ru(const float a);
__device__ __half __float2half_rz(const float a);
__device__ __half2 __floats2half2_rn(const float a, const float b);
__device__ float2 __half22float2(const __half2 a);
__device__ float __half2float(const __half a);
__device__ __half2 half2half2(const __half a);
__device__ int __half2int_rd(__half h);
__device__ int __half2int_rn(__half h);
__device__ int __half2int_ru(__half h);
__device__ int __half2int_rz(__half h);
__device__ long long int __half2ll_rd(__half h);
__device__ long long int __half2ll_rn(__half h);
__device__ long long int __half2ll_ru(__half h);
__device__ long long int __half2ll_rz(__half h);
__device__ short __half2short_rd(__half h);
__device__ short __half2short_rn(__half h);
__device__ short __half2short_ru(__half h);
__device__ short __half2short_rz(__half h);
__device__ unsigned int __half2uint_rd(__half h);
__device__ unsigned int __half2uint_rn(__half h);
__device__ unsigned int __half2uint_ru(__half h);
__device__ unsigned int __half2uint_rz(__half h);
__device__ unsigned long long int __half2ull_rd(__half h);
__device__ unsigned long long int __half2ull_rn(__half h);
__device__ unsigned long long int __half2ull_ru(__half h);
__device__ unsigned long long int __half2ull_rz(__half h);
__device__ unsigned short int __half2ushort_rd(__half h);
__device__ unsigned short int __half2ushort_rn(__half h);
__device__ unsigned short int __half2ushort_ru(__half h);
__device__ unsigned short int __half2ushort_rz(__half h);
__device__ short int __half_as_short(const __half h);
__device__ unsigned short int __half_as_ushort(const __half h);
__device__ __half2 __halves2half2(const __half a, const __half b);
__device__ float __high2float(const __half2 a);
__device__ __half __high2half(const __half2 a);
__device__ __half2 __high2half2(const __half2 a);
__device__ __half2 __highs2half2(const __half2 a, const __half2 b);
__device__ __half __int2half_rd(int i);
__device__ __half __int2half_rn(int i);
__device__ __half __int2half_ru(int i);
__device__ __half __int2half_rz(int i);
__device__ __half __ll2half_rd(long long int i);
__device__ __half __ll2half_rn(long long int i);
__device__ __half __ll2half_ru(long long int i);
__device__ __half __ll2half_rz(long long int i);
__device__ float __low2float(const __half2 a);
__device__ __half __low2half(const __half2 a);
__device__ __half2 __low2half2(const __half2 a);
__device__ __half2 __lows2half2(const __half2 a, const __half2 b);
__device__ __half2 __lowhigh2highlow(const __half2 a);
__device__ __half2 __low2half2(const __half2 a, const __half2 b);
__device__ __half2 __low2half2(const __half2 a);
__device__ __half2 __lowhigh2highlow(const __half2 a);
__device__ __half2 __lows2half2(const __half2 a, const __half2 b);
__device__ __half __short2half_rd(short int i);
__device__ __half __short2half_rn(short int i);
__device__ __half __short2half_ru(short int i);
__device__ __half __short2half_rz(short int i);
__device__ __half __uint2half_rd(unsigned int i);
__device__ __half __uint2half_rn(unsigned int i);
__device__ __half __uint2half_ru(unsigned int i);
__device__ __half __uint2half_rz(unsigned int i);
__device__ __half __ull2half_rd(unsigned long long int i);
__device__ __half __ull2half_rn(unsigned long long int i);
__device__ __half __ull2half_ru(unsigned long long int i);
__device__ __half __ull2half_rz(unsigned long long int i);
__device__ __half __ushort2half_rd(unsigned short int i);
__device__ __half __ushort2half_rn(unsigned short int i);
__device__ __half __ushort2half_ru(unsigned short int i);
__device__ __half __ushort2half_rz(unsigned short int i);
__device__ __half __ushort_as_half(const unsigned short int i);
extern "C" int __hip_hc_ir_hadd2_int(int, int);
extern "C" int __hip_hc_ir_hfma2_int(int, int, int);
extern "C" int __hip_hc_ir_hmul2_int(int, int);
extern "C" int __hip_hc_ir_hsub2_int(int, int);
extern "C" __half __hip_hc_ir_hceil_half(__half) __asm("llvm.ceil.f16");
extern "C" __half __hip_hc_ir_hcos_half(__half) __asm("llvm.cos.f16");
extern "C" __half __hip_hc_ir_hexp2_half(__half) __asm("llvm.exp2.f16");
extern "C" __half __hip_hc_ir_hfloor_half(__half) __asm("llvm.floor.f16");
extern "C" __half __hip_hc_ir_hlog2_half(__half) __asm("llvm.log2.f16");
extern "C" __half __hip_hc_ir_hrcp_half(__half) __asm("llvm.amdgcn.rcp.f16");
extern "C" __half __hip_hc_ir_hrint_half(__half) __asm("llvm.rint.f16");
extern "C" __half __hip_hc_ir_hrsqrt_half(__half) __asm("llvm.sqrt.f16");
extern "C" __half __hip_hc_ir_hsin_half(__half) __asm("llvm.sin.f16");
extern "C" __half __hip_hc_ir_hsqrt_half(__half) __asm("llvm.sqrt.f16");
extern "C" __half __hip_hc_ir_htrunc_half(__half) __asm("llvm.trunc.f16");
extern "C" int __hip_hc_ir_h2ceil_int(int);
extern "C" int __hip_hc_ir_h2cos_int(int);
extern "C" int __hip_hc_ir_h2exp2_int(int);
extern "C" int __hip_hc_ir_h2floor_int(int);
extern "C" int __hip_hc_ir_h2log2_int(int);
extern "C" int __hip_hc_ir_h2rcp_int(int);
extern "C" int __hip_hc_ir_h2rsqrt_int(int);
extern "C" int __hip_hc_ir_h2sin_int(int);
extern "C" int __hip_hc_ir_h2sqrt_int(int);
extern "C" int __hip_hc_ir_h2trunc_int(int);
/*
Half2 Arithmetic Functions
*/
__device__ static inline __half2 __hadd2(__half2 a, __half2 b) {
__half2 c;
c.q = __hip_hc_ir_hadd2_int(a.q, b.q);
return c;
}
__device__ static inline __half2 __hadd2_sat(__half2 a, __half2 b) {
__half2 c;
c.q = __hip_hc_ir_hadd2_int(a.q, b.q);
return c;
}
__device__ static inline __half2 __hfma2(__half2 a, __half2 b, __half2 c) {
__half2 d;
d.q = __hip_hc_ir_hfma2_int(a.q, b.q, c.q);
return d;
}
__device__ static inline __half2 __hfma2_sat(__half2 a, __half2 b, __half2 c) {
__half2 d;
d.q = __hip_hc_ir_hfma2_int(a.q, b.q, c.q);
return d;
}
__device__ static inline __half2 __hmul2(__half2 a, __half2 b) {
__half2 c;
c.q = __hip_hc_ir_hmul2_int(a.q, b.q);
return c;
}
__device__ static inline __half2 __hmul2_sat(__half2 a, __half2 b) {
__half2 c;
c.q = __hip_hc_ir_hmul2_int(a.q, b.q);
return c;
}
__device__ static inline __half2 __hsub2(__half2 a, __half2 b) {
__half2 c;
c.q = __hip_hc_ir_hsub2_int(a.q, b.q);
return c;
}
__device__ static inline __half2 __hneg2(__half2 a) {
__half2 c;
c.p[0] = - a.p[0];
c.p[1] = - a.p[1];
return c;
}
__device__ static inline __half2 __hsub2_sat(__half2 a, __half2 b) {
__half2 c;
c.q = __hip_hc_ir_hsub2_int(a.q, b.q);
return c;
}
__device__ static inline __half2 h2div(__half2 a, __half2 b) {
__half2 c;
c.p[0] = a.p[0] / b.p[0];
c.p[1] = a.p[1] / b.p[1];
return c;
}
__device__ static inline __half hceil(const __half h) {
return __hip_hc_ir_hceil_half(h);
}
__device__ static inline __half hcos(const __half h) {
return __hip_hc_ir_hcos_half(h);
}
__device__ static inline __half hexp(const __half h) {
return __hip_hc_ir_hexp2_half(__hmul(h, 1.442694));
}
__device__ static inline __half hexp10(const __half h) {
return __hip_hc_ir_hexp2_half(__hmul(h, 3.3219281));
}
__device__ static inline __half hexp2(const __half h) {
return __hip_hc_ir_hexp2_half(h);
}
__device__ static inline __half hfloor(const __half h) {
return __hip_hc_ir_hfloor_half(h);
}
__device__ static inline __half hlog(const __half h) {
return __hmul(__hip_hc_ir_hlog2_half(h), 0.693147);
}
__device__ static inline __half hlog10(const __half h) {
return __hmul(__hip_hc_ir_hlog2_half(h), 0.301029);
}
__device__ static inline __half hlog2(const __half h) {
return __hip_hc_ir_hlog2_half(h);
}
/*
__device__ static inline __half hrcp(const __half h) {
return __hip_hc_ir_hrcp_half(h);
}
*/
__device__ static inline __half hrint(const __half h) {
return __hip_hc_ir_hrint_half(h);
}
__device__ static inline __half hrsqrt(const __half h) {
return __hip_hc_ir_hrsqrt_half(h);
}
__device__ static inline __half hsin(const __half h) {
return __hip_hc_ir_hsin_half(h);
}
__device__ static inline __half hsqrt(const __half a) {
return __hip_hc_ir_hsqrt_half(a);
}
__device__ static inline __half htrunc(const __half a) {
return __hip_hc_ir_htrunc_half(a);
}
/*
Half2 Math Operations
*/
__device__ static inline __half2 h2ceil(const __half2 h) {
__half2 a;
a.q = __hip_hc_ir_h2ceil_int(h.q);
return a;
}
__device__ static inline __half2 h2cos(const __half2 h) {
__half2 a;
a.q = __hip_hc_ir_h2cos_int(h.q);
return a;
}
__device__ static inline __half2 h2exp(const __half2 h) {
__half2 factor;
factor.p[0] = 1.442694;
factor.p[1] = 1.442694;
factor.q = __hip_hc_ir_h2exp2_int(__hip_hc_ir_hmul2_int(h.q, factor.q));
return factor;
}
__device__ static inline __half2 h2exp10(const __half2 h) {
__half2 factor;
factor.p[0] = 3.3219281;
factor.p[1] = 3.3219281;
factor.q = __hip_hc_ir_h2exp2_int(__hip_hc_ir_hmul2_int(h.q, factor.q));
return factor;
}
__device__ static inline __half2 h2exp2(const __half2 h) {
__half2 a;
a.q = __hip_hc_ir_h2exp2_int(h.q);
return a;
}
__device__ static inline __half2 h2floor(const __half2 h) {
__half2 a;
a.q = __hip_hc_ir_h2floor_int(h.q);
return a;
}
__device__ static inline __half2 h2log(const __half2 h) {
__half2 factor;
factor.p[0] = 0.693147;
factor.p[1] = 0.693147;
factor. q = __hip_hc_ir_hmul2_int(__hip_hc_ir_h2log2_int(h.q), factor.q);
return factor;
}
__device__ static inline __half2 h2log10(const __half2 h) {
__half2 factor;
factor.p[0] = 0.301029;
factor.p[1] = 0.301029;
factor.q = __hip_hc_ir_hmul2_int(__hip_hc_ir_h2log2_int(h.q), factor.q);
return factor;
}
__device__ static inline __half2 h2log2(const __half2 h) {
__half2 a;
a.q = __hip_hc_ir_h2log2_int(h.q);
return a;
}
__device__ static inline __half2 h2rcp(const __half2 h) {
__half2 a;
a.q = __hip_hc_ir_h2rcp_int(h.q);
return a;
}
__device__ static inline __half2 h2rsqrt(const __half2 h) {
__half2 a;
a.q = __hip_hc_ir_h2rsqrt_int(h.q);
return a;
}
__device__ static inline __half2 h2sin(const __half2 h) {
__half2 a;
a.q = __hip_hc_ir_h2sin_int(h.q);
return a;
}
__device__ static inline __half2 h2sqrt(const __half2 h) {
__half2 a;
a.q = __hip_hc_ir_h2sqrt_int(h.q);
return a;
}
__device__ static inline __half2 h2trunc(const __half2 h) {
__half2 a;
a.q = __hip_hc_ir_h2trunc_int(h.q);
return a;
}
#endif
#if __clang_major__ == 3
typedef struct {
unsigned x: 16;
} __half;
typedef struct __attribute__((aligned(4))){
union {
__half p[2];
unsigned int q;
};
} __half2;
#endif
#endif
projects/hip/include/hip/hcc_detail/hip_hcc.h
+103
Datei anzeigen
@@ -0,0 +1,103 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#ifndef HIP_INCLUDE_HIP_HCC_DETAIL_HIP_HCC_H
#define HIP_INCLUDE_HIP_HCC_DETAIL_HIP_HCC_H
#include "hip/hip_runtime_api.h"
#if __cplusplus
#ifdef __HCC__
#include <hc.hpp>
/**
*-------------------------------------------------------------------------------------------------
*-------------------------------------------------------------------------------------------------
* @defgroup HCC-specific features
* @warning These APIs provide access to special features of HCC compiler and are not available through the CUDA path.
* @{
*/
/**
* @brief Return hc::accelerator associated with the specified deviceId
* @return #hipSuccess, #hipErrorInvalidDevice
*/
hipError_t hipHccGetAccelerator(int deviceId, hc::accelerator *acc);
/**
* @brief Return hc::accelerator_view associated with the specified stream
*
* If stream is 0, the accelerator_view for the default stream is returned.
* @return #hipSuccess
*/
hipError_t hipHccGetAcceleratorView(hipStream_t stream, hc::accelerator_view **av);
#endif // #ifdef __HCC__
/**
* @brief launches kernel f with launch parameters and shared memory on stream with arguments passed to kernelparams or extra
*
* @param [in[ f Kernel to launch.
* @param [in] gridDimX X grid dimension specified in work-items
* @param [in] gridDimY Y grid dimension specified in work-items
* @param [in] gridDimZ Z grid dimension specified in work-items
* @param [in] blockDimX X block dimensions specified in work-items
* @param [in] blockDimY Y grid dimension specified in work-items
* @param [in] blockDimZ Z grid dimension specified in work-items
* @param [in] sharedMemBytes Amount of dynamic shared memory to allocate for this kernel. The kernel can access this with HIP_DYNAMIC_SHARED.
* @param [in] stream Stream where the kernel should be dispatched. May be 0, in which case th default stream is used with associated synchronization rules.
* @param [in] kernelParams
* @param [in] extra Pointer to kernel arguments. These are passed directly to the kernel and must be in the memory layout and alignment expected by the kernel.
* @param [in] startEvent If non-null, specified event will be updated to track the start time of the kernel launch. The event must be created before calling this API.
* @param [in] stopEvent If non-null, specified event will be updated to track the stop time of the kernel launch. The event must be created before calling this API.
*
* @returns hipSuccess, hipInvalidDevice, hipErrorNotInitialized, hipErrorInvalidValue
*
* @warning kernellParams argument is not yet implemented in HIP. Please use extra instead. Please refer to hip_porting_driver_api.md for sample usage.
* HIP/ROCm actually updates the start event when the associated kernel completes.
*/
hipError_t hipHccModuleLaunchKernel(hipFunction_t f,
uint32_t globalWorkSizeX,
uint32_t globalWorkSizeY,
uint32_t globalWorkSizeZ,
uint32_t localWorkSizeX,
uint32_t localWorkSizeY,
uint32_t localWorkSizeZ,
size_t sharedMemBytes,
hipStream_t hStream,
void **kernelParams,
void **extra,
hipEvent_t startEvent=nullptr,
hipEvent_t stopEvent=nullptr
);
// doxygen end HCC-specific features
/**
* @}
*/
#endif // #if __cplusplus
#endif //
projects/hip/include/hip/hcc_detail/hip_ldg.h
+6 -8
Datei anzeigen
@@ -1,5 +1,5 @@
/*
Copyright (c) 2015-2016 Advanced Micro Devices, Inc. All rights reserved.
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
@@ -20,14 +20,13 @@ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#ifndef HIP_LDG_H
#define HIP_LDG_H
#ifndef HIP_INCLUDE_HIP_HCC_DETAIL_HIP_LDG_H
#define HIP_INCLUDE_HIP_HCC_DETAIL_HIP_LDG_H
#if __HCC__
#if defined __HCC__
#if __hcc_workweek__ >= 16164
#include "hip/hip_vector_types.h"
#include "hip/hcc_detail/host_defines.h"
#include "hip_vector_types.h"
#include "host_defines.h"
__device__ char __ldg(const char* );
__device__ char2 __ldg(const char2* );
@@ -75,4 +74,3 @@ __device__ double2 __ldg(const double2* );
#endif // __HCC__
#endif // HIP_LDG_H
projects/hip/include/hip/hcc_detail/hip_runtime.h
+111 -513
Datei anzeigen
@@ -1,5 +1,5 @@
/*
Copyright (c) 2015-2016 Advanced Micro Devices, Inc. All rights reserved.
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
@@ -26,13 +26,12 @@ THE SOFTWARE.
*/
//#pragma once
#ifndef HIP_RUNTIME_H
#define HIP_RUNTIME_H
#ifndef HIP_INCLUDE_HIP_HCC_DETAIL_HIP_RUNTIME_H
#define HIP_INCLUDE_HIP_HCC_DETAIL_HIP_RUNTIME_H
//---
// Top part of file can be compiled with any compiler
//#include <cstring>
#if __cplusplus
#include <cmath>
@@ -40,39 +39,77 @@ THE SOFTWARE.
#include <math.h>
#include <string.h>
#include <stddef.h>
#endif
#endif//__cplusplus
// Define NVCC_COMPAT for CUDA compatibility
#define NVCC_COMPAT
#define CUDA_SUCCESS hipSuccess
#include <hip/hip_runtime_api.h>
//#include "hip/hcc_detail/hip_hcc.h"
#if USE_PROMOTE_FREE_HCC == 1
#define ADDRESS_SPACE_1
#define ADDRESS_SPACE_3
#else
#define ADDRESS_SPACE_1 __attribute__((address_space(1)))
#define ADDRESS_SPACE_3 __attribute__((address_space(3)))
#endif
//---
// Remainder of this file only compiles with HCC
#ifdef __HCC__
#if defined __HCC__
#include <grid_launch.h>
//TODO-HCC-GL - change this to typedef.
//typedef grid_launch_parm hipLaunchParm ;
#if defined (GRID_LAUNCH_VERSION) and (GRID_LAUNCH_VERSION >= 20)
// Use field names for grid_launch 2.0 structure, if HCC supports GL 2.0.
#if GENERIC_GRID_LAUNCH == 0
#define hipLaunchParm grid_launch_parm
#else
namespace hip_impl
{
struct Empty_launch_parm {};
}
#define hipLaunchParm hip_impl::Empty_launch_parm
#endif //GENERIC_GRID_LAUNCH
#if defined (GRID_LAUNCH_VERSION) and (GRID_LAUNCH_VERSION >= 20) || GENERIC_GRID_LAUNCH == 1
#else // Use field names for grid_launch 2.0 structure, if HCC supports GL 2.0.
#error (HCC must support GRID_LAUNCH_20)
#endif
#endif //GRID_LAUNCH_VERSION
#endif //HCC
#if GENERIC_GRID_LAUNCH==1 && defined __HCC__
#include "grid_launch_GGL.hpp"
#endif//GENERIC_GRID_LAUNCH
extern int HIP_TRACE_API;
//TODO-HCC-GL - change this to typedef.
//typedef grid_launch_parm hipLaunchParm ;
#define hipLaunchParm grid_launch_parm
#ifdef __cplusplus
//#include <hip/hcc_detail/hip_texture.h>
#include <hip/hcc_detail/hip_ldg.h>
#endif
#include <hip/hcc_detail/host_defines.h>
#include <hip/hcc_detail/math_functions.h>
#include <hip/hcc_detail/device_functions.h>
// TODO-HCC remove old definitions ; ~1602 hcc supports __HCC_ACCELERATOR__ define.
#if defined (__KALMAR_ACCELERATOR__) && !defined (__HCC_ACCELERATOR__)
#define __HCC_ACCELERATOR__ __KALMAR_ACCELERATOR__
#endif
// TODO-HCC add a dummy implementation of assert, need to replace with a proper kernel exit call.
#if __HIP_DEVICE_COMPILE__ == 1
#undef assert
#define assert(COND) { if (COND) {} }
#endif
// Feature tests:
#if defined(__HCC_ACCELERATOR__) && (__HCC_ACCELERATOR__ != 0)
// Device compile and not host compile:
@@ -120,210 +157,9 @@ extern int HIP_TRACE_API;
#define __HCC_C__
#endif
__device__ float acosf(float x);
__device__ float acoshf(float x);
__device__ float asinf(float x);
__device__ float asinhf(float x);
__device__ float atan2f(float y, float x);
__device__ float atanf(float x);
__device__ float atanhf(float x);
__device__ float cbrtf(float x);
__device__ float ceilf(float x);
__device__ float copysignf(float x, float y);
__device__ float coshf(float x);
__device__ float cyl_bessel_i0f(float x);
__device__ float cyl_bessel_i1f(float x);
__device__ float erfcf(float x);
__device__ float erfcinvf(float y);
__host__ float erfcinvf(float y);
__device__ float erfcxf(float x);
__host__ float erfcxf(float x);
__device__ float erff(float x);
__device__ float erfinvf(float y);
__host__ float erfinvf(float y);
__device__ float exp2f(float x);
__device__ float expm1f(float x);
__device__ float fabsf(float x);
__device__ float fdimf(float x, float y);
__device__ __host__ float fdividef(float x, float y);
__device__ float floorf(float x);
__device__ float fmaf(float x, float y, float z);
__device__ float fmaxf(float x, float y);
__device__ float fminf(float x, float y);
__device__ float fmodf(float x, float y);
__device__ float frexpf(float x, float y);
__device__ float hypotf(float x, float y);
__device__ float ilogbf(float x);
__host__ __device__ unsigned isfinite(float a);
__device__ unsigned isinf(float a);
__device__ unsigned isnan(float a);
__device__ float j0f(float x);
__device__ float j1f(float x);
__device__ float jnf(int n, float x);
__device__ float ldexpf(float x, int exp);
__device__ float lgammaf(float x);
__device__ long long int llrintf(float x);
__device__ long long int llroundf(float x);
__device__ float log1pf(float x);
__device__ float logbf(float x);
__device__ long int lrintf(float x);
__device__ long int lroundf(float x);
__device__ float modff(float x, float *iptr);
__device__ float nanf(const char* tagp);
__device__ float nearbyintf(float x);
__device__ float nextafterf(float x, float y);
__device__ float norm3df(float a, float b, float c);
__host__ float norm3df(float a, float b, float c);
__device__ float norm4df(float a, float b, float c, float d);
__host__ float norm4df(float a, float b, float c, float d);
__device__ float normcdff(float y);
__host__ float normcdff(float y);
__device__ float normcdfinvf(float y);
__host__ float normcdfinvf(float y);
__device__ float normf(int dim, const float *a);
__device__ float rcbrtf(float x);
__host__ float rcbrtf(float x);
__device__ float remainderf(float x, float y);
__device__ float remquof(float x, float y, int *quo);
__device__ float rhypotf(float x, float y);
__host__ float rhypotf(float x, float y);
__device__ float rintf(float x);
__device__ float rnorm3df(float a, float b, float c);
__host__ float rnorm3df(float a, float b, float c);
__device__ float rnorm4df(float a, float b, float c, float d);
__host__ float rnorm4df(float a, float b, float c, float d);
__device__ float rnormf(int dim, const float* a);
__host__ float rnormf(int dim, const float* a);
__device__ float roundf(float x);
__device__ float rsqrtf(float x);
__device__ float scalblnf(float x, long int n);
__device__ float scalbnf(float x, int n);
__host__ __device__ unsigned signbit(float a);
__device__ void sincospif(float x, float *sptr, float *cptr);
__host__ void sincospif(float x, float *sptr, float *cptr);
__device__ float sinhf(float x);
__device__ float sinpif(float x);
__device__ float sqrtf(float x);
__device__ float tanhf(float x);
__device__ float tgammaf(float x);
__device__ float truncf(float x);
__device__ float y0f(float x);
__device__ float y1f(float x);
__device__ float ynf(int n, float x);
__host__ __device__ float cospif(float x);
__host__ __device__ float sinpif(float x);
__device__ float sqrtf(float x);
__host__ __device__ float rsqrtf(float x);
__device__ double acos(double x);
__device__ double acosh(double x);
__device__ double asin(double x);
__device__ double asinh(double x);
__device__ double atan(double x);
__device__ double atan2(double y, double x);
__device__ double atanh(double x);
__device__ double cbrt(double x);
__device__ double ceil(double x);
__device__ double copysign(double x, double y);
__device__ double cos(double x);
__device__ double cosh(double x);
__host__ __device__ double cospi(double x);
__device__ double cyl_bessel_i0(double x);
__device__ double cyl_bessel_i1(double x);
__device__ double erf(double x);
__device__ double erfc(double x);
__device__ double erfcinv(double y);
__device__ double erfcx(double x);
__device__ double erfinv(double x);
__device__ double exp(double x);
__device__ double exp10(double x);
__device__ double exp2(double x);
__device__ double expm1(double x);
__device__ double fabs(double x);
__device__ double fdim(double x, double y);
__device__ double fdivide(double x, double y);
__device__ double floor(double x);
__device__ double fma(double x, double y, double z);
__device__ double fmax(double x, double y);
__device__ double fmin(double x, double y);
__device__ double fmod(double x, double y);
__device__ double frexp(double x, int *nptr);
__device__ double hypot(double x, double y);
__device__ double ilogb(double x);
__host__ __device__ unsigned isfinite(double x);
__device__ unsigned isinf(double x);
__device__ unsigned isnan(double x);
__device__ double j0(double x);
__device__ double j1(double x);
__device__ double jn(int n, double x);
__device__ double ldexp(double x, int exp);
__device__ double lgamma(double x);
__device__ long long llrint(double x);
__device__ long long llround(double x);
__device__ double log(double x);
__device__ double log10(double x);
__device__ double log1p(double x);
__device__ double log2(double x);
__device__ double logb(double x);
__device__ long int lrint(double x);
__device__ long int lround(double x);
__device__ double modf(double x, double *iptr);
__device__ double nan(const char* tagp);
__device__ double nearbyint(double x);
__device__ double nextafter(double x, double y);
__device__ double norm(int dim, const double* t);
__device__ double norm3d(double a, double b, double c);
__host__ double norm3d(double a, double b, double c);
__device__ double norm4d(double a, double b, double c, double d);
__host__ double norm4d(double a, double b, double c, double d);
__device__ double normcdf(double y);
__host__ double normcdf(double y);
__device__ double normcdfinv(double y);
__host__ double normcdfinv(double y);
__device__ double pow(double x, double y);
__device__ double rcbrt(double x);
__host__ double rcbrt(double x);
__device__ double remainder(double x, double y);
__device__ double remquo(double x, double y, int *quo);
__device__ double rhypot(double x, double y);
__host__ double rhypot(double x, double y);
__device__ double rint(double x);
__device__ double rnorm(int dim, const double* t);
__host__ double rnorm(int dim, const double* t);
__device__ double rnorm3d(double a, double b, double c);
__host__ double rnorm3d(double a, double b, double c);
__device__ double rnorm4d(double a, double b, double c, double d);
__host__ double rnorm4d(double a, double b, double c, double d);
__device__ double round(double x);
__host__ __device__ double rsqrt(double x);
__device__ double scalbln(double x, long int n);
__device__ double scalbn(double x, int n);
__host__ __device__ unsigned signbit(double a);
__device__ double sin(double a);
__device__ void sincos(double x, double *sptr, double *cptr);
__device__ void sincospi(double x, double *sptr, double *cptr);
__host__ void sincospi(double x, double *sptr, double *cptr);
__device__ double sinh(double x);
__host__ __device__ double sinpi(double x);
__device__ double sqrt(double x);
__device__ double tan(double x);
__device__ double tanh(double x);
__device__ double tgamma(double x);
__device__ double trunc(double x);
__device__ double y0(double x);
__device__ double y1(double y);
__device__ double yn(int n, double x);
__host__ double erfcinv(double y);
__host__ double erfcx(double x);
__host__ double erfinv(double y);
__host__ double fdivide(double x, double y);
// TODO - hipify-clang - change to use the function call.
//#define warpSize hc::__wavesize()
extern const int warpSize;
static constexpr int warpSize = 64;
#define clock_t long long int
__device__ long long int clock64();
@@ -417,30 +253,13 @@ __device__ unsigned int atomicInc(unsigned int* address,
__device__ unsigned int atomicDec(unsigned int* address,
unsigned int val);
//__mul24 __umul24
__device__ int __mul24(int arg1, int arg2);
__device__ unsigned int __umul24(unsigned int arg1, unsigned int arg2);
// integer intrinsic function __poc __clz __ffs __brev
__device__ unsigned int __popc( unsigned int input);
__device__ unsigned int __popcll( unsigned long long int input);
__device__ unsigned int __clz(unsigned int input);
__device__ unsigned int __clzll(unsigned long long int input);
__device__ unsigned int __clz(int input);
__device__ unsigned int __clzll(long long int input);
__device__ unsigned int __ffs(unsigned int input);
__device__ unsigned int __ffsll(unsigned long long int input);
__device__ unsigned int __ffs(int input);
__device__ unsigned int __ffsll(long long int input);
__device__ unsigned int __brev( unsigned int input);
__device__ unsigned long long int __brevll( unsigned long long int input);
// warp vote function __all __any __ballot
// warp vote function __all __any __ballot
__device__ int __all( int input);
__device__ int __any( int input);
__device__ unsigned long long int __ballot( int input);
#if __HIP_ARCH_GFX701__ == 0
// warp shuffle functions
#ifdef __cplusplus
__device__ int __shfl(int input, int lane, int width=warpSize);
@@ -460,259 +279,25 @@ __device__ float __shfl(float input, int lane, int width);
__device__ float __shfl_up(float input, unsigned int lane_delta, int width);
__device__ float __shfl_down(float input, unsigned int lane_delta, int width);
__device__ float __shfl_xor(float input, int lane_mask, int width);
#endif
#endif //__cplusplus
__device__ unsigned __hip_ds_bpermute(int index, unsigned src);
__device__ float __hip_ds_bpermutef(int index, float src);
__device__ unsigned __hip_ds_permute(int index, unsigned src);
__device__ float __hip_ds_permutef(int index, float src);
__device__ unsigned __hip_ds_swizzle(unsigned int src, int pattern);
__device__ float __hip_ds_swizzlef(float src, int pattern);
__device__ int __hip_move_dpp(int src, int dpp_ctrl, int row_mask, int bank_mask, bool bound_ctrl);
#endif //__HIP_ARCH_GFX803__ == 1
__host__ __device__ int min(int arg1, int arg2);
__host__ __device__ int max(int arg1, int arg2);
__device__ __attribute__((address_space(3))) void* __get_dynamicgroupbaseptr();
__device__ ADDRESS_SPACE_3 void* __get_dynamicgroupbaseptr();
//TODO - add a couple fast math operations here, the set here will grow :
// Single Precision Precise Math
__device__ float __hip_precise_cosf(float);
__device__ float __hip_precise_exp10f(float);
__device__ float __hip_precise_expf(float);
__device__ float __hip_precise_frsqrt_rn(float);
__device__ float __hip_precise_fsqrt_rd(float);
__device__ float __hip_precise_fsqrt_rn(float);
__device__ float __hip_precise_fsqrt_ru(float);
__device__ float __hip_precise_fsqrt_rz(float);
__device__ float __hip_precise_log10f(float);
__device__ float __hip_precise_log2f(float);
__device__ float __hip_precise_logf(float);
__device__ float __hip_precise_powf(float, float);
__device__ void __hip_precise_sincosf(float,float*,float*);
__device__ float __hip_precise_sinf(float);
__device__ float __hip_precise_tanf(float);
// Double Precision Precise Math
__device__ double __hip_precise_dsqrt_rd(double);
__device__ double __hip_precise_dsqrt_rn(double);
__device__ double __hip_precise_dsqrt_ru(double);
__device__ double __hip_precise_dsqrt_rz(double);
// Single Precision Fast Math
extern __attribute__((const)) float __hip_fast_cosf(float) __asm("llvm.cos.f32");
extern __attribute__((const)) float __hip_fast_exp2f(float) __asm("llvm.exp2.f32");
__device__ float __hip_fast_exp10f(float);
__device__ float __hip_fast_expf(float);
__device__ float __hip_fast_frsqrt_rn(float);
extern __attribute__((const)) float __hip_fast_fsqrt_rd(float) __asm("llvm.sqrt.f32");
__device__ float __hip_fast_fsqrt_rn(float);
__device__ float __hip_fast_fsqrt_ru(float);
__device__ float __hip_fast_fsqrt_rz(float);
__device__ float __hip_fast_log10f(float);
extern __attribute__((const)) float __hip_fast_log2f(float) __asm("llvm.log2.f32");
__device__ float __hip_fast_logf(float);
__device__ float __hip_fast_powf(float, float);
__device__ void __hip_fast_sincosf(float,float*,float*);
extern __attribute__((const)) float __hip_fast_sinf(float) __asm("llvm.sin.f32");
__device__ float __hip_fast_tanf(float);
extern __attribute__((const)) float __hip_fast_fmaf(float,float,float) __asm("llvm.fma.f32");
extern __attribute__((const)) float __hip_fast_frcp(float) __asm("llvm.amdgcn.rcp.f32");
extern __attribute__((const)) double __hip_fast_dsqrt(double) __asm("llvm.sqrt.f64");
extern __attribute__((const)) double __hip_fast_fma(double,double,double) __asm("llvm.fma.f64");
extern __attribute__((const)) double __hip_fast_drcp(double) __asm("llvm.amdgcn.rcp.f64");
#ifdef HIP_FAST_MATH
// Single Precision Precise Math when enabled
__device__ inline float cosf(float x) {
return __hip_fast_cosf(x);
}
__device__ inline float exp10f(float x) {
return __hip_fast_exp10f(x);
}
__device__ inline float expf(float x) {
return __hip_fast_expf(x);
}
__device__ inline float log10f(float x) {
return __hip_fast_log10f(x);
}
__device__ inline float log2f(float x) {
return __hip_fast_log2f(x);
}
__device__ inline float logf(float x) {
return __hip_fast_logf(x);
}
__device__ inline float powf(float base, float exponent) {
return __hip_fast_powf(base, exponent);
}
__device__ inline void sincosf(float x, float *s, float *c) {
return __hip_fast_sincosf(x, s, c);
}
__device__ inline float sinf(float x) {
return __hip_fast_sinf(x);
}
__device__ inline float tanf(float x) {
return __hip_fast_tanf(x);
}
#else
__device__ float sinf(float);
__device__ float cosf(float);
__device__ float tanf(float);
__device__ void sincosf(float, float*, float*);
__device__ float logf(float);
__device__ float log2f(float);
__device__ float log10f(float);
__device__ float expf(float);
__device__ float exp10f(float);
__device__ float powf(float, float);
#endif
// Single Precision Fast Math
__device__ inline float __cosf(float x) {
return __hip_fast_cosf(x);
}
__device__ inline float __exp10f(float x) {
return __hip_fast_exp10f(x);
}
__device__ inline float __expf(float x) {
return __hip_fast_expf(x);
}
__device__ inline float __frsqrt_rn(float x) {
return __hip_fast_frsqrt_rn(x);
}
__device__ inline float __fsqrt_rd(float x) {
return __hip_fast_fsqrt_rd(x);
}
__device__ inline float __fsqrt_rn(float x) {
return __hip_fast_fsqrt_rn(x);
}
__device__ inline float __fsqrt_ru(float x) {
return __hip_fast_fsqrt_ru(x);
}
__device__ inline float __fsqrt_rz(float x) {
return __hip_fast_fsqrt_rz(x);
}
__device__ inline float __log10f(float x) {
return __hip_fast_log10f(x);
}
__device__ inline float __log2f(float x) {
return __hip_fast_log2f(x);
}
__device__ inline float __logf(float x) {
return __hip_fast_logf(x);
}
__device__ inline float __powf(float base, float exponent) {
return __hip_fast_powf(base, exponent);
}
__device__ inline void __sincosf(float x, float *s, float *c) {
return __hip_fast_sincosf(x, s, c);
}
__device__ inline float __sinf(float x) {
return __hip_fast_sinf(x);
}
__device__ inline float __tanf(float x) {
return __hip_fast_tanf(x);
}
__device__ inline float __fmaf_rd(float x, float y, float z) {
return __hip_fast_fmaf(x, y, z);
}
__device__ inline float __fmaf_rn(float x, float y, float z) {
return __hip_fast_fmaf(x, y, z);
}
__device__ inline float __fmaf_ru(float x, float y, float z) {
return __hip_fast_fmaf(x, y, z);
}
__device__ inline float __fmaf_rz(float x, float y, float z) {
return __hip_fast_fmaf(x, y, z);
}
__device__ inline float __frcp_rd(float x) {
return __hip_fast_frcp(x);
}
__device__ inline float __frcp_rn(float x) {
return __hip_fast_frcp(x);
}
__device__ inline float __frcp_ru(float x) {
return __hip_fast_frcp(x);
}
__device__ inline float __frcp_rz(float x) {
return __hip_fast_frcp(x);
}
__device__ inline double __dsqrt_rd(double x) {
return __hip_fast_dsqrt(x);
}
__device__ inline double __dsqrt_rn(double x) {
return __hip_fast_dsqrt(x);
}
__device__ inline double __dsqrt_ru(double x) {
return __hip_fast_dsqrt(x);
}
__device__ inline double __dsqrt_rz(double x) {
return __hip_fast_dsqrt(x);
}
__device__ inline double __fma_rd(double x, double y, double z) {
return __hip_fast_fma(x, y, z);
}
__device__ inline double __fma_rn(double x, double y, double z) {
return __hip_fast_fma(x, y, z);
}
__device__ inline double __fma_ru(double x, double y, double z) {
return __hip_fast_fma(x, y, z);
}
__device__ inline double __fma_rz(double x, double y, double z) {
return __hip_fast_fma(x, y, z);
}
__device__ inline double __drcp_rd(double x) {
return __hip_fast_drcp(x);
}
__device__ inline double __drcp_rn(double x) {
return __hip_fast_drcp(x);
}
__device__ inline double __drcp_ru(double x) {
return __hip_fast_drcp(x);
}
__device__ inline double __drcp_rz(double x) {
return __hip_fast_drcp(x);
}
/**
* CUDA 8 device function features
@@ -738,6 +323,10 @@ __device__ inline double __drcp_rz(double x) {
* On AMD platforms, the threadfence* routines are currently empty stubs.
*/
extern __attribute__((const)) __device__ void __hip_hc_threadfence() __asm("__llvm_fence_sc_dev");
extern __attribute__((const)) __device__ void __hip_hc_threadfence_block() __asm("__llvm_fence_sc_wg");
/**
* @brief threadfence_block makes writes visible to threads running in same block.
*
@@ -748,7 +337,9 @@ __device__ inline double __drcp_rz(double x) {
* @warning __threadfence_block is a stub and map to no-op.
*/
// __device__ void __threadfence_block(void);
extern "C" __device__ void __threadfence_block(void);
__device__ static inline void __threadfence_block(void) {
return __hip_hc_threadfence_block();
}
/**
* @brief threadfence makes wirtes visible to other threads running on same GPU.
@@ -760,7 +351,9 @@ extern "C" __device__ void __threadfence_block(void);
* @warning __threadfence is a stub and map to no-op, application should set "export HSA_DISABLE_CACHE=1" to disable both L1 and L2 caches.
*/
// __device__ void __threadfence(void) __attribute__((deprecated("Provided for compile-time compatibility, not yet functional")));
extern "C" __device__ void __threadfence(void);
__device__ static inline void __threadfence(void) {
return __hip_hc_threadfence();
}
/**
* @brief threadfence_system makes writes to pinned system memory visible on host CPU.
@@ -774,16 +367,6 @@ extern "C" __device__ void __threadfence(void);
//__device__ void __threadfence_system(void) __attribute__((deprecated("Provided with workaround configuration, see hip_kernel_language.md for details")));
__device__ void __threadfence_system(void) ;
__device__ unsigned __hip_ds_bpermute(int index, unsigned src);
__device__ float __hip_ds_bpermutef(int index, float src);
__device__ unsigned __hip_ds_permute(int index, unsigned src);
__device__ float __hip_ds_permutef(int index, float src);
__device__ unsigned __hip_ds_swizzle(unsigned int src, int pattern);
__device__ float __hip_ds_swizzlef(float src, int pattern);
__device__ int __hip_move_dpp(int src, int dpp_ctrl, int row_mask, int bank_mask, bool bound_ctrl);
// doxygen end Fence Fence
/**
* @}
@@ -806,32 +389,48 @@ __device__ int __hip_move_dpp(int src, int dpp_ctrl, int row_mask, int bank_mask
#define hipGridDim_y (hc_get_num_groups(1))
#define hipGridDim_z (hc_get_num_groups(2))
//extern "C" __device__ void* memcpy(void* dst, void* src, size_t size);
//extern "C" __device__ void* memset(void* ptr, uint8_t val, size_t size);
extern "C" __device__ void* __hip_hc_memcpy(void* dst, const void* src, size_t size);
extern "C" __device__ void* __hip_hc_memset(void* ptr, uint8_t val, size_t size);
extern "C" __device__ void* __hip_hc_malloc(size_t);
extern "C" __device__ void* __hip_hc_free(void *ptr);
//extern "C" __device__ void* malloc(size_t size);
//extern "C" __device__ void* free(void *ptr);
static inline __device__ void* malloc(size_t size)
{
return __hip_hc_malloc(size);
}
static inline __device__ void* free(void *ptr)
{
return __hip_hc_free(ptr);
}
static inline __device__ void* memcpy(void* dst, const void* src, size_t size)
{
return __hip_hc_memcpy(dst, src, size);
}
static inline __device__ void* memset(void* ptr, int val, size_t size)
{
uint8_t val8 = static_cast <uint8_t> (val);
return __hip_hc_memset(ptr, val8, size);
}
extern "C" __device__ char4 __hip_hc_add8pk(char4, char4);
extern "C" __device__ char4 __hip_hc_sub8pk(char4, char4);
extern "C" __device__ char4 __hip_hc_mul8pk(char4, char4);
#define __syncthreads() hc_barrier(CLK_LOCAL_MEM_FENCE)
#define HIP_KERNEL_NAME(...) __VA_ARGS__
#define HIP_KERNEL_NAME(...) (__VA_ARGS__)
#define HIP_SYMBOL(X) #X
#ifdef __HCC_CPP__
#if defined __HCC_CPP__
extern hipStream_t ihipPreLaunchKernel(hipStream_t stream, dim3 grid, dim3 block, grid_launch_parm *lp, const char *kernelNameStr);
extern hipStream_t ihipPreLaunchKernel(hipStream_t stream, dim3 grid, size_t block, grid_launch_parm *lp, const char *kernelNameStr);
extern hipStream_t ihipPreLaunchKernel(hipStream_t stream, size_t grid, dim3 block, grid_launch_parm *lp, const char *kernelNameStr);
extern hipStream_t ihipPreLaunchKernel(hipStream_t stream, size_t grid, size_t block, grid_launch_parm *lp, const char *kernelNameStr);
extern void ihipPostLaunchKernel(const char *kernelName, hipStream_t stream, grid_launch_parm &lp);
#if GENERIC_GRID_LAUNCH == 0
//#warning "Original hipLaunchKernel defined"
// Due to multiple overloaded versions of ihipPreLaunchKernel, the numBlocks3D and blockDim3D can be either size_t or dim3 types
#define hipLaunchKernel(_kernelName, _numBlocks3D, _blockDim3D, _groupMemBytes, _stream, ...) \
do {\
@@ -841,13 +440,13 @@ do {\
_kernelName (lp, ##__VA_ARGS__);\
ihipPostLaunchKernel(#_kernelName, trueStream, lp);\
} while(0)
#endif //GENERIC_GRID_LAUNCH
#elif defined (__HCC_C__)
//TODO - develop C interface.
#endif
#endif //__HCC_CPP__
/**
* extern __shared__
@@ -856,12 +455,11 @@ do {\
// Macro to replace extern __shared__ declarations
// to local variable definitions
#define HIP_DYNAMIC_SHARED(type, var) \
__attribute__((address_space(3))) type* var = \
(__attribute__((address_space(3))) type*)__get_dynamicgroupbaseptr(); \
ADDRESS_SPACE_3 type* var = \
(ADDRESS_SPACE_3 type*)__get_dynamicgroupbaseptr(); \
#define HIP_DYNAMIC_SHARED_ATTRIBUTE __attribute__((address_space(3)))
#define HIP_DYNAMIC_SHARED_ATTRIBUTE ADDRESS_SPACE_3
#endif // __HCC__
/**
@@ -885,4 +483,4 @@ do {\
#endif
#endif//HIP_HCC_DETAIL_RUNTIME_H
projects/hip/include/hip/hcc_detail/hip_runtime_api.h
+286 -42
Datei anzeigen
@@ -1,5 +1,5 @@
/*
Copyright (c) 2015-2016 Advanced Micro Devices, Inc. All rights reserved.
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
@@ -21,19 +21,23 @@ THE SOFTWARE.
*/
//#pragma once
#ifndef HIP_RUNTIME_API_H
#define HIP_RUNTIME_API_H
#ifndef HIP_INCLUDE_HIP_HCC_DETAIL_HIP_RUNTIME_API_H
#define HIP_INCLUDE_HIP_HCC_DETAIL_HIP_RUNTIME_API_H
/**
* @file hcc_detail/hip_runtime_api.h
* @brief Contains C function APIs for HIP runtime. This file does not use any HCC builtin or special language extensions (-hc mode) ; those functions in hip_runtime.h.
*/
#include <stdint.h>
#include <stddef.h>
#include <iostream>
#ifndef GENERIC_GRID_LAUNCH
#define GENERIC_GRID_LAUNCH 1
#endif
#include <hip/hcc_detail/host_defines.h>
#include <hip/hip_runtime_api.h>
//#include "hip/hip_hcc.h"
#include <hip/hip_texture.h>
#if defined (__HCC__) && (__hcc_workweek__ < 16155)
#error("This version of HIP requires a newer version of HCC.");
@@ -53,7 +57,7 @@ extern "C" {
typedef struct ihipCtx_t *hipCtx_t;
// Note many APIs also use integer deviceIds as an alternative to the device pointer:
typedef struct ihipDevice_t *hipDevice_t;
typedef int hipDevice_t;
typedef struct ihipStream_t *hipStream_t;
@@ -61,7 +65,12 @@ typedef struct ihipStream_t *hipStream_t;
#define hipIpcMemLazyEnablePeerAccess 0
typedef struct ihipIpcMemHandle_t *hipIpcMemHandle_t;
#define HIP_IPC_HANDLE_SIZE 64
typedef struct hipIpcMemHandle_st
{
char reserved[HIP_IPC_HANDLE_SIZE];
}hipIpcMemHandle_t;
//TODO: IPC event handle currently unsupported
struct ihipIpcEventHandle_t;
@@ -72,7 +81,7 @@ typedef struct ihipIpcEventHandle_t *hipIpcEventHandle_t;
typedef struct ihipModule_t *hipModule_t;
typedef struct ihipFunction_t *hipFunction_t;
typedef struct ihipModuleSymbol_t *hipFunction_t;
typedef void* hipDeviceptr_t;
@@ -170,6 +179,12 @@ typedef enum hipMemcpyKind {
,hipMemcpyDefault = 4, ///< Runtime will automatically determine copy-kind based on virtual addresses.
} hipMemcpyKind;
typedef struct {
unsigned int width;
unsigned int height;
enum hipChannelFormatKind f;
void* data; //FIXME: generalize this
} hipArray;
@@ -358,7 +373,7 @@ hipError_t hipDeviceGetCacheConfig ( hipFuncCache_t *cacheConfig );
* Note: Currently, only hipLimitMallocHeapSize is available
*
*/
hipError_t hipDeviceGetLimit(size_t *pValue, hipLimit_t limit);
hipError_t hipDeviceGetLimit(size_t *pValue, enum hipLimit_t limit);
/**
@@ -504,12 +519,10 @@ const char *hipGetErrorString(hipError_t hipError);
* @{
*
* The following Stream APIs are not (yet) supported in HIP:
* - cudaStreamAddCallback
* - cudaStreamAttachMemAsync
* - cudaStreamCreateWithPriority
* - cudaStreamGetPriority
* - cudaStreamWaitEvent
*/
*/
/**
@@ -840,7 +853,7 @@ hipError_t hipEventQuery(hipEvent_t event) ;
*
* @see hipGetDeviceCount, hipGetDevice, hipSetDevice, hipChooseDevice
*/
hipError_t hipPointerGetAttributes(hipPointerAttribute_t *attributes, void* ptr);
hipError_t hipPointerGetAttributes(hipPointerAttribute_t *attributes, const void* ptr);
/**
* @brief Allocate memory on the default accelerator
@@ -848,7 +861,9 @@ hipError_t hipPointerGetAttributes(hipPointerAttribute_t *attributes, void* ptr)
* @param[out] ptr Pointer to the allocated memory
* @param[in] size Requested memory size
*
* @return #hipSuccess
* If size is 0, no memory is allocated, *ptr returns nullptr, and hipSuccess is returned.
*
* @return #hipSuccess, #hipErrorMemoryAllocation, #hipErrorInvalidValue (bad context, null *ptr)
*
* @see hipMallocPitch, hipFree, hipMallocArray, hipFreeArray, hipMalloc3D, hipMalloc3DArray, hipHostFree, hipHostMalloc
*/
@@ -860,6 +875,8 @@ hipError_t hipMalloc(void** ptr, size_t size) ;
* @param[out] ptr Pointer to the allocated host pinned memory
* @param[in] size Requested memory size
*
* If size is 0, no memory is allocated, *ptr returns nullptr, and hipSuccess is returned.
*
* @return #hipSuccess, #hipErrorMemoryAllocation
*
* @deprecated use hipHostMalloc() instead
@@ -873,6 +890,8 @@ hipError_t hipMallocHost(void** ptr, size_t size) __attribute__((deprecated("use
* @param[in] size Requested memory size
* @param[in] flags Type of host memory allocation
*
* If size is 0, no memory is allocated, *ptr returns nullptr, and hipSuccess is returned.
*
* @return #hipSuccess, #hipErrorMemoryAllocation
*
* @see hipSetDeviceFlags, hipHostFree
@@ -886,6 +905,8 @@ hipError_t hipHostMalloc(void** ptr, size_t size, unsigned int flags) ;
* @param[in] size Requested memory size
* @param[in] flags Type of host memory allocation
*
* If size is 0, no memory is allocated, *ptr returns nullptr, and hipSuccess is returned.
*
* @return #hipSuccess, #hipErrorMemoryAllocation
*
* @deprecated use hipHostMalloc() instead
@@ -970,6 +991,9 @@ hipError_t hipHostUnregister(void* hostPtr) ;
* @param[out] pitch Pitch for allocation (in bytes)
* @param[in] width Requested pitched allocation width (in bytes)
* @param[in] height Requested pitched allocation height
*
* If size is 0, no memory is allocated, *ptr returns nullptr, and hipSuccess is returned.
*
* @return Error code
*
* @see hipMalloc, hipFree, hipMallocArray, hipFreeArray, hipHostFree, hipMalloc3D, hipMalloc3DArray, hipHostMalloc
@@ -1132,7 +1156,7 @@ hipError_t hipMemcpyDtoDAsync(hipDeviceptr_t dst, hipDeviceptr_t src, size_t siz
*
* @see hipMemcpy, hipMemcpy2D, hipMemcpyToArray, hipMemcpy2DToArray, hipMemcpyFromArray, hipMemcpy2DFromArray, hipMemcpyArrayToArray, hipMemcpy2DArrayToArray, hipMemcpyFromSymbol, hipMemcpyAsync, hipMemcpy2DAsync, hipMemcpyToArrayAsync, hipMemcpy2DToArrayAsync, hipMemcpyFromArrayAsync, hipMemcpy2DFromArrayAsync, hipMemcpyToSymbolAsync, hipMemcpyFromSymbolAsync
*/
hipError_t hipMemcpyToSymbol(const char* symbolName, const void *src, size_t sizeBytes, size_t offset, hipMemcpyKind kind);
hipError_t hipMemcpyToSymbol(const void* symbolName, const void *src, size_t sizeBytes, size_t offset, hipMemcpyKind kind);
/**
@@ -1152,9 +1176,11 @@ hipError_t hipMemcpyToSymbol(const char* symbolName, const void *src, size_t siz
*
* @see hipMemcpy, hipMemcpy2D, hipMemcpyToArray, hipMemcpy2DToArray, hipMemcpyFromArray, hipMemcpy2DFromArray, hipMemcpyArrayToArray, hipMemcpy2DArrayToArray, hipMemcpyFromSymbol, hipMemcpyAsync, hipMemcpy2DAsync, hipMemcpyToArrayAsync, hipMemcpy2DToArrayAsync, hipMemcpyFromArrayAsync, hipMemcpy2DFromArrayAsync, hipMemcpyToSymbolAsync, hipMemcpyFromSymbolAsync
*/
hipError_t hipMemcpyToSymbolAsync(const char* symbolName, const void *src, size_t sizeBytes, size_t offset, hipMemcpyKind kind, hipStream_t stream);
hipError_t hipMemcpyToSymbolAsync(const void* symbolName, const void *src, size_t sizeBytes, size_t offset, hipMemcpyKind kind, hipStream_t stream);
hipError_t hipMemcpyFromSymbol(void *dst, const void* symbolName, size_t sizeBytes, size_t offset, hipMemcpyKind kind);
hipError_t hipMemcpyFromSymbolAsync(void *dst, const void* symbolName, size_t sizeBytes, size_t offset, hipMemcpyKind kind, hipStream_t stream);
/**
* @brief Copy data from src to dst asynchronously.
@@ -1185,19 +1211,24 @@ hipError_t hipMemcpyAsync(void* dst, const void* src, size_t sizeBytes, hipMemcp
#endif
/**
* @brief Copy data from src to dst asynchronously.
* @brief Fills the first sizeBytes bytes of the memory area pointed to by dest with the constant byte value value.
*
* It supports memory from host to device,
* device to host, device to device and host to host.
*
* @param[out] dst Data being copy to
* @param[in] src Data being copy from
* @param[out] dst Data being filled
* @param[in] constant value to be set
* @param[in] sizeBytes Data size in bytes
* @param[in] accelerator_view Accelerator view which the copy is being enqueued
* @return #hipSuccess, #hipErrorInvalidValue, #hipErrorMemoryFree
* @return #hipSuccess, #hipErrorInvalidValue, #hipErrorNotInitialized
*/
hipError_t hipMemset(void* dst, int value, size_t sizeBytes );
/**
* @brief Fills the first sizeBytes bytes of the memory area pointed to by dest with the constant byte value value.
*
* @param[out] dst Data ptr to be filled
* @param[in] constant value to be set
* @param[in] sizeBytes Data size in bytes
* @return #hipSuccess, #hipErrorInvalidValue, #hipErrorNotInitialized
*/
hipError_t hipMemsetD8(hipDeviceptr_t dest, unsigned char value, size_t sizeBytes );
/**
* @brief Fills the first sizeBytes bytes of the memory area pointed to by dev with the constant byte value value.
@@ -1228,6 +1259,91 @@ hipError_t hipMemsetAsync(void* dst, int value, size_t sizeBytes, hipStream_t st
**/
hipError_t hipMemGetInfo (size_t * free, size_t * total) ;
hipError_t hipMemPtrGetInfo(void *ptr, size_t *size);
/**
* @brief Allocate an array on the device.
*
* @param[out] array Pointer to allocated array in device memory
* @param[in] desc Requested channel format
* @param[in] width Requested array allocation width
* @param[in] height Requested array allocation height
* @param[in] flags Requested properties of allocated array
* @return #hipSuccess, #hipErrorMemoryAllocation
*
* @see hipMalloc, hipMallocPitch, hipFree, hipFreeArray, hipHostMalloc, hipHostFree
*/
#if __cplusplus
hipError_t hipMallocArray(hipArray** array, const hipChannelFormatDesc* desc,
size_t width, size_t height = 0, unsigned int flags = 0);
#else
hipError_t hipMallocArray(hipArray** array, const struct hipChannelFormatDesc* desc,
size_t width, size_t height, unsigned int flags);
#endif
/**
* @brief Frees an array on the device.
*
* @param[in] array Pointer to array to free
* @return #hipSuccess, #hipErrorInvalidValue, #hipErrorInitializationError
*
* @see hipMalloc, hipMallocPitch, hipFree, hipMallocArray, hipHostMalloc, hipHostFree
*/
hipError_t hipFreeArray(hipArray* array);
/**
* @brief Copies data between host and device.
*
* @param[in] dst Destination memory address
* @param[in] dpitch Pitch of destination memory
* @param[in] src Source memory address
* @param[in] spitch Pitch of source memory
* @param[in] width Width of matrix transfer (columns in bytes)
* @param[in] height Height of matrix transfer (rows)
* @param[in] kind Type of transfer
* @return #hipSuccess, #hipErrorInvalidValue, #hipErrorInvalidPitchValue, #hipErrorInvalidDevicePointer, #hipErrorInvalidMemcpyDirection
*
* @see hipMemcpy, hipMemcpyToArray, hipMemcpy2DToArray, hipMemcpyFromArray, hipMemcpyToSymbol, hipMemcpyAsync
*/
hipError_t hipMemcpy2D(void* dst, size_t dpitch, const void* src, size_t spitch, size_t width, size_t height, hipMemcpyKind kind);
/**
* @brief Copies data between host and device.
*
* @param[in] dst Destination memory address
* @param[in] dpitch Pitch of destination memory
* @param[in] src Source memory address
* @param[in] spitch Pitch of source memory
* @param[in] width Width of matrix transfer (columns in bytes)
* @param[in] height Height of matrix transfer (rows)
* @param[in] kind Type of transfer
* @return #hipSuccess, #hipErrorInvalidValue, #hipErrorInvalidPitchValue, #hipErrorInvalidDevicePointer, #hipErrorInvalidMemcpyDirection
*
* @see hipMemcpy, hipMemcpyToArray, hipMemcpy2D, hipMemcpyFromArray, hipMemcpyToSymbol, hipMemcpyAsync
*/
hipError_t hipMemcpy2DToArray(hipArray* dst, size_t wOffset, size_t hOffset, const void* src,
size_t spitch, size_t width, size_t height, hipMemcpyKind kind);
/**
* @brief Copies data between host and device.
*
* @param[in] dst Destination memory address
* @param[in] dpitch Pitch of destination memory
* @param[in] src Source memory address
* @param[in] spitch Pitch of source memory
* @param[in] width Width of matrix transfer (columns in bytes)
* @param[in] height Height of matrix transfer (rows)
* @param[in] kind Type of transfer
* @return #hipSuccess, #hipErrorInvalidValue, #hipErrorInvalidPitchValue, #hipErrorInvalidDevicePointer, #hipErrorInvalidMemcpyDirection
*
* @see hipMemcpy, hipMemcpy2DToArray, hipMemcpy2D, hipMemcpyFromArray, hipMemcpyToSymbol, hipMemcpyAsync
*/
hipError_t hipMemcpyToArray(hipArray* dst, size_t wOffset, size_t hOffset,
const void* src, size_t count, hipMemcpyKind kind);
// doxygen end Memory
/**
* @}
@@ -1580,6 +1696,66 @@ hipError_t hipCtxEnablePeerAccess (hipCtx_t peerCtx, unsigned int flags);
*/
hipError_t hipCtxDisablePeerAccess (hipCtx_t peerCtx);
/**
* @brief Get the state of the primary context.
*
* @param [in] Device to get primary context flags for
* @param [out] Pointer to store flags
* @param [out] Pointer to store context state; 0 = inactive, 1 = active
*
* @returns #hipSuccess
*
* @see hipCtxCreate, hipCtxDestroy, hipCtxGetFlags, hipCtxPopCurrent, hipCtxGetCurrent, hipCtxSetCurrent, hipCtxPushCurrent, hipCtxSetCacheConfig, hipCtxSynchronize, hipCtxGetDevice
*/
hipError_t hipDevicePrimaryCtxGetState ( hipDevice_t dev, unsigned int* flags, int* active );
/**
* @brief Release the primary context on the GPU.
*
* @param [in] Device which primary context is released
*
* @returns #hipSuccess
*
* @see hipCtxCreate, hipCtxDestroy, hipCtxGetFlags, hipCtxPopCurrent, hipCtxGetCurrent, hipCtxSetCurrent, hipCtxPushCurrent, hipCtxSetCacheConfig, hipCtxSynchronize, hipCtxGetDevice
* @warning This function return #hipSuccess though doesn't release the primaryCtx by design on HIP/HCC path.
*/
hipError_t hipDevicePrimaryCtxRelease ( hipDevice_t dev);
/**
* @brief Retain the primary context on the GPU.
*
* @param [out] Returned context handle of the new context
* @param [in] Device which primary context is released
*
* @returns #hipSuccess
*
* @see hipCtxCreate, hipCtxDestroy, hipCtxGetFlags, hipCtxPopCurrent, hipCtxGetCurrent, hipCtxSetCurrent, hipCtxPushCurrent, hipCtxSetCacheConfig, hipCtxSynchronize, hipCtxGetDevice
*/
hipError_t hipDevicePrimaryCtxRetain ( hipCtx_t* pctx, hipDevice_t dev );
/**
* @brief Resets the primary context on the GPU.
*
* @param [in] Device which primary context is reset
*
* @returns #hipSuccess
*
* @see hipCtxCreate, hipCtxDestroy, hipCtxGetFlags, hipCtxPopCurrent, hipCtxGetCurrent, hipCtxSetCurrent, hipCtxPushCurrent, hipCtxSetCacheConfig, hipCtxSynchronize, hipCtxGetDevice
*/
hipError_t hipDevicePrimaryCtxReset ( hipDevice_t dev );
/**
* @brief Set flags for the primary context.
*
* @param [in] Device for which the primary context flags are set
* @param [in] New flags for the device
*
* @returns #hipSuccess, #hipErrorContextAlreadyInUse
*
* @see hipCtxCreate, hipCtxDestroy, hipCtxGetFlags, hipCtxPopCurrent, hipCtxGetCurrent, hipCtxSetCurrent, hipCtxPushCurrent, hipCtxSetCacheConfig, hipCtxSynchronize, hipCtxGetDevice
*/
hipError_t hipDevicePrimaryCtxSetFlags ( hipDevice_t dev, unsigned int flags );
// doxygen end Context Management
/**
* @}
@@ -1737,19 +1913,18 @@ hipError_t hipModuleLoadData(hipModule_t *module, const void *image);
/**
* @brief launches kernel f with launch parameters and shared memory on stream with arguments passed to kernelparams or extra
*
* @param [in[ f
* @param [in] gridDimX
* @param [in] gridDimY
* @param [in] gridDimZ
* @param [in] blockDimX
* @param [in] blockDimY
* @param [in] blockDimZ
* @param [in] sharedMemBytes
* @param [in] stream
* @param [in[ f Kernel to launch.
* @param [in] gridDimX X grid dimension specified as multiple of blockDimX.
* @param [in] gridDimY Y grid dimension specified as multiple of blockDimY.
* @param [in] gridDimZ Z grid dimension specified as multiple of blockDimZ.
* @param [in] blockDimX X block dimensions specified in work-items
* @param [in] blockDimY Y grid dimension specified in work-items
* @param [in] blockDimZ Z grid dimension specified in work-items
* @param [in] sharedMemBytes Amount of dynamic shared memory to allocate for this kernel. The kernel can access this with HIP_DYNAMIC_SHARED.
* @param [in] stream Stream where the kernel should be dispatched. May be 0, in which case th default stream is used with associated synchronization rules.
* @param [in] kernelParams
* @param [in] extraa
* @param [in] extra Pointer to kernel arguments. These are passed directly to the kernel and must be in the memory layout and alignment expected by the kernel.
*
* The function takes the above arguments and run the kernel in hipFunction_t f. with launch parameters specified in gridDimX, gridDimY, gridDimZ, blockDimX, blockDimY and blockDimmZ. The amount of shared memory is specificed and can be used with HIP_DYNAMIC_SHARED. The arguemt extra is used to pass in the arguments for the kernel.
* @returns hipSuccess, hipInvalidDevice, hipErrorNotInitialized, hipErrorInvalidValue
*
* @warning kernellParams argument is not yet implemented in HIP. Please use extra instead. Please refer to hip_porting_driver_api.md for sample usage.
@@ -1905,17 +2080,86 @@ hipError_t hipIpcCloseMemHandle(void *devPtr);
#endif
#ifdef __cplusplus
/**
* @brief Returns a PCI Bus Id string for the device.
* @param [out] pciBusId
* @param [in] len
* @param [hipDevice_t] device
/*
* @brief hipBindTexture Binds size bytes of the memory area pointed to by @p devPtr to the texture reference tex.
*
* @returns #hipSuccess, #hipErrorInavlidDevice
* @p desc describes how the memory is interpreted when fetching values from the texture. The @p offset parameter is an optional byte offset as with the low-level
* hipBindTexture() function. Any memory previously bound to tex is unbound.
*
* @param[in] offset - Offset in bytes
* @param[out] tex - texture to bind
* @param[in] devPtr - Memory area on device
* @param[in] desc - Channel format
* @param[in] size - Size of the memory area pointed to by devPtr
* @return #hipSuccess, #hipErrorInvalidValue, #hipErrorMemoryFree, #hipErrorUnknown
**/
template <class T, int dim, enum hipTextureReadMode readMode>
hipError_t hipBindTexture(size_t *offset,
struct texture<T, dim, readMode> &tex,
const void *devPtr,
const struct hipChannelFormatDesc *desc,
size_t size=UINT_MAX)
{
tex._dataPtr = static_cast<const T*>(devPtr);
return hipSuccess;
}
/*
* @brief hipBindTexture Binds size bytes of the memory area pointed to by @p devPtr to the texture reference tex.
*
* @p desc describes how the memory is interpreted when fetching values from the texture. The @p offset parameter is an optional byte offset as with the low-level
* hipBindTexture() function. Any memory previously bound to tex is unbound.
*
* @param[in] offset - Offset in bytes
* @param[in] tex - texture to bind
* @param[in] devPtr - Memory area on device
* @param[in] size - Size of the memory area pointed to by devPtr
* @return #hipSuccess, #hipErrorInvalidValue, #hipErrorMemoryFree, #hipErrorUnknown
**/
template <class T, int dim, enum hipTextureReadMode readMode>
hipError_t hipBindTexture(size_t *offset,
struct texture<T, dim, readMode> &tex,
const void *devPtr,
size_t size=UINT_MAX)
{
return hipBindTexture(offset, tex, devPtr, &tex.channelDesc, size);
}
template <class T, int dim, enum hipTextureReadMode readMode>
hipError_t hipBindTextureToArray(struct texture<T, dim, readMode> &tex, hipArray* array) {
tex.width = array->width;
tex.height = array->height;
tex._dataPtr = static_cast<const T*>(array->data);
return hipSuccess;
}
/*
* @brief Unbinds the textuer bound to @p tex
*
* @param[in] tex - texture to unbind
*
* @return #hipSuccess
**/
template <class T, int dim, enum hipTextureReadMode readMode>
hipError_t hipUnbindTexture(struct texture<T, dim, readMode> &tex)
{
tex._dataPtr = NULL;
return hipSuccess;
}
// doxygen end Texture
/**
* @}
*/
hipError_t hipDeviceGetPCIBusId (char *pciBusId,int len,hipDevice_t device);
#endif
/**
*-------------------------------------------------------------------------------------------------
*-------------------------------------------------------------------------------------------------
projects/hip/include/hip/hcc_detail/hip_texture.h
+7 -236
Datei anzeigen
@@ -1,5 +1,5 @@
/*
Copyright (c) 2015-2016 Advanced Micro Devices, Inc. All rights reserved.
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
@@ -22,8 +22,8 @@ THE SOFTWARE.
//#pragma once
#ifndef HIP_HCC_DETAIL_TEXTURE_H
#define HIP_HCC_DETAIL_TEXTURE_H
#ifndef HIP_INCLUDE_HIP_HCC_DETAIL_HIP_TEXTURE_H
#define HIP_INCLUDE_HIP_HCC_DETAIL_HIP_TEXTURE_H
/**
* @file hcc_detail/hip_texture.h
@@ -31,48 +31,17 @@ THE SOFTWARE.
*/
#include <limits.h>
#include <hip/hcc_detail/driver_types.h>
#include <hip/hcc_detail/channel_descriptor.h>
#include <hip/hcc_detail/texture_types.h>
//#include <hip/hcc_detail/hip_runtime.h>
//----
//Texture - TODO - likely need to move this to a separate file only included with kernel compilation.
#define hipTextureType1D 1
typedef enum {
hipChannelFormatKindSigned = 0,
hipChannelFormatKindUnsigned,
hipChannelFormatKindFloat,
hipChannelFormatKindNone
} hipChannelFormatKind;
typedef struct hipChannelFormatDesc {
int x;
int y;
int z;
int w;
hipChannelFormatKind f;
} hipChannelFormatDesc;
typedef enum hipTextureReadMode
{
hipReadModeElementType, ///< Read texture as specified element type
//! @warning cudaReadModeNormalizedFloat is not supported.
} hipTextureReadMode;
typedef enum hipTextureFilterMode
{
hipFilterModePoint, ///< Point filter mode.
//! @warning cudaFilterModeLinear is not supported.
} hipTextureFilterMode;
struct textureReference {
hipTextureFilterMode filterMode;
bool normalized;
hipChannelFormatDesc channelDesc;
};
#if __cplusplus
template <class T, int texType=hipTextureType1D, enum hipTextureReadMode=hipReadModeElementType>
template <class T, int texType=hipTextureType1D, hipTextureReadMode readMode=hipReadModeElementType>
struct texture : public textureReference {
const T * _dataPtr; // pointer to underlying data.
@@ -84,95 +53,12 @@ struct texture : public textureReference {
};
#endif
typedef struct {
unsigned int width;
unsigned int height;
hipChannelFormatKind f;
void* data; //FIXME: generalize this
} hipArray;
#define tex1Dfetch(_tex, _addr) (_tex._dataPtr[_addr])
#define tex2D(_tex, _dx, _dy) \
_tex._dataPtr[(unsigned int)_dx + (unsigned int)_dy*(_tex.width)]
/**
* @brief Allocate an array on the device.
*
* @param[out] array Pointer to allocated array in device memory
* @param[in] desc Requested channel format
* @param[in] width Requested array allocation width
* @param[in] height Requested array allocation height
* @param[in] flags Requested properties of allocated array
* @return #hipSuccess, #hipErrorMemoryAllocation
*
* @see hipMalloc, hipMallocPitch, hipFree, hipFreeArray, hipHostMalloc, hipHostFree
*/
hipError_t hipMallocArray(hipArray** array, const hipChannelFormatDesc* desc,
size_t width, size_t height = 0, unsigned int flags = 0);
/**
* @brief Frees an array on the device.
*
* @param[in] array Pointer to array to free
* @return #hipSuccess, #hipErrorInvalidValue, #hipErrorInitializationError
*
* @see hipMalloc, hipMallocPitch, hipFree, hipMallocArray, hipHostMalloc, hipHostFree
*/
hipError_t hipFreeArray(hipArray* array);
/**
* @brief Copies data between host and device.
*
* @param[in] dst Destination memory address
* @param[in] dpitch Pitch of destination memory
* @param[in] src Source memory address
* @param[in] spitch Pitch of source memory
* @param[in] width Width of matrix transfer (columns in bytes)
* @param[in] height Height of matrix transfer (rows)
* @param[in] kind Type of transfer
* @return #hipSuccess, #hipErrorInvalidValue, #hipErrorInvalidPitchValue, #hipErrorInvalidDevicePointer, #hipErrorInvalidMemcpyDirection
*
* @see hipMemcpy, hipMemcpyToArray, hipMemcpy2DToArray, hipMemcpyFromArray, hipMemcpyToSymbol, hipMemcpyAsync
*/
hipError_t hipMemcpy2D(void* dst, size_t dpitch, const void* src, size_t spitch, size_t width, size_t height, hipMemcpyKind kind);
/**
* @brief Copies data between host and device.
*
* @param[in] dst Destination memory address
* @param[in] dpitch Pitch of destination memory
* @param[in] src Source memory address
* @param[in] spitch Pitch of source memory
* @param[in] width Width of matrix transfer (columns in bytes)
* @param[in] height Height of matrix transfer (rows)
* @param[in] kind Type of transfer
* @return #hipSuccess, #hipErrorInvalidValue, #hipErrorInvalidPitchValue, #hipErrorInvalidDevicePointer, #hipErrorInvalidMemcpyDirection
*
* @see hipMemcpy, hipMemcpyToArray, hipMemcpy2D, hipMemcpyFromArray, hipMemcpyToSymbol, hipMemcpyAsync
*/
hipError_t hipMemcpy2DToArray(hipArray* dst, size_t wOffset, size_t hOffset, const void* src,
size_t spitch, size_t width, size_t height, hipMemcpyKind kind);
/**
* @brief Copies data between host and device.
*
* @param[in] dst Destination memory address
* @param[in] dpitch Pitch of destination memory
* @param[in] src Source memory address
* @param[in] spitch Pitch of source memory
* @param[in] width Width of matrix transfer (columns in bytes)
* @param[in] height Height of matrix transfer (rows)
* @param[in] kind Type of transfer
* @return #hipSuccess, #hipErrorInvalidValue, #hipErrorInvalidPitchValue, #hipErrorInvalidDevicePointer, #hipErrorInvalidMemcpyDirection
*
* @see hipMemcpy, hipMemcpy2DToArray, hipMemcpy2D, hipMemcpyFromArray, hipMemcpyToSymbol, hipMemcpyAsync
*/
hipError_t hipMemcpyToArray(hipArray* dst, size_t wOffset, size_t hOffset,
const void* src, size_t count, hipMemcpyKind kind);
/**
* @addtogroup API HIP API
* @{
@@ -212,120 +98,6 @@ hipChannelFormatDesc hipBindTexture(size_t *offset, struct textureReference *te
}
#endif
/**
* @brief Returns a channel descriptor using the specified format.
*
* @param[in] x X component
* @param[in] y Y component
* @param[in] z Z component
* @param[in] w W component
* @param[in] f Channel format
* @return Channel descriptor with format f
*
*/
hipChannelFormatDesc hipCreateChannelDesc(int x, int y, int z, int w, hipChannelFormatKind f);
// descriptors
template <typename T> inline hipChannelFormatDesc hipCreateChannelDesc() {
return hipCreateChannelDesc(0, 0, 0, 0, hipChannelFormatKindNone);
}
template <> inline hipChannelFormatDesc hipCreateChannelDesc<int>() {
int e = (int)sizeof(int) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindSigned);
}
template <> inline hipChannelFormatDesc hipCreateChannelDesc<unsigned int>() {
int e = (int)sizeof(unsigned int) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindUnsigned);
}
template <> inline hipChannelFormatDesc hipCreateChannelDesc<long>() {
int e = (int)sizeof(long) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindSigned);
}
template <> inline hipChannelFormatDesc hipCreateChannelDesc<unsigned long>() {
int e = (int)sizeof(unsigned long) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindUnsigned);
}
template <> inline hipChannelFormatDesc hipCreateChannelDesc<float>() {
int e = (int)sizeof(float) * 8;
return hipCreateChannelDesc(e, 0, 0, 0, hipChannelFormatKindFloat);
}
/*
* @brief hipBindTexture Binds size bytes of the memory area pointed to by @p devPtr to the texture reference tex.
*
* @p desc describes how the memory is interpreted when fetching values from the texture. The @p offset parameter is an optional byte offset as with the low-level
* hipBindTexture() function. Any memory previously bound to tex is unbound.
*
* @param[in] offset - Offset in bytes
* @param[out] tex - texture to bind
* @param[in] devPtr - Memory area on device
* @param[in] desc - Channel format
* @param[in] size - Size of the memory area pointed to by devPtr
* @return #hipSuccess, #hipErrorInvalidValue, #hipErrorMemoryFree, #hipErrorUnknown
**/
template <class T, int dim, enum hipTextureReadMode readMode>
hipError_t hipBindTexture(size_t *offset,
struct texture<T, dim, readMode> &tex,
const void *devPtr,
const struct hipChannelFormatDesc *desc,
size_t size=UINT_MAX)
{
tex._dataPtr = static_cast<const T*>(devPtr);
return hipSuccess;
}
/*
* @brief hipBindTexture Binds size bytes of the memory area pointed to by @p devPtr to the texture reference tex.
*
* @p desc describes how the memory is interpreted when fetching values from the texture. The @p offset parameter is an optional byte offset as with the low-level
* hipBindTexture() function. Any memory previously bound to tex is unbound.
*
* @param[in] offset - Offset in bytes
* @param[in] tex - texture to bind
* @param[in] devPtr - Memory area on device
* @param[in] size - Size of the memory area pointed to by devPtr
* @return #hipSuccess, #hipErrorInvalidValue, #hipErrorMemoryFree, #hipErrorUnknown
**/
template <class T, int dim, enum hipTextureReadMode readMode>
hipError_t hipBindTexture(size_t *offset,
struct texture<T, dim, readMode> &tex,
const void *devPtr,
size_t size=UINT_MAX)
{
return hipBindTexture(offset, tex, devPtr, &tex.channelDesc, size);
}
template <class T, int dim, enum hipTextureReadMode readMode>
hipError_t hipBindTextureToArray(struct texture<T, dim, readMode> &tex, hipArray* array) {
tex.width = array->width;
tex.height = array->height;
tex._dataPtr = static_cast<const T*>(array->data);
return hipSuccess;
}
/*
* @brief Unbinds the textuer bound to @p tex
*
* @param[in] tex - texture to unbind
*
* @return #hipSuccess
**/
template <class T, int dim, enum hipTextureReadMode readMode>
hipError_t hipUnbindTexture(struct texture<T, dim, readMode> &tex)
{
tex._dataPtr = NULL;
return hipSuccess;
}
// doxygen end Texture
/**
* @}
*/
// End doxygen API:
/**
@@ -333,4 +105,3 @@ hipError_t hipUnbindTexture(struct texture<T, dim, readMode> &tex)
*/
#endif
projects/hip/include/hip/hcc_detail/hip_vector_types.h
+4010 -328
Datei anzeigen
Datei-Diff unterdrückt, da er zu groß ist Diff laden
projects/hip/include/hip/hcc_detail/host_defines.h
+23 -5
Datei anzeigen
@@ -1,5 +1,5 @@
/*
Copyright (c) 2015-2016 Advanced Micro Devices, Inc. All rights reserved.
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
@@ -25,8 +25,15 @@ THE SOFTWARE.
* @brief TODO-doc
*/
#ifndef HOST_DEFINES_H
#define HOST_DEFINES_H
#ifndef HIP_INCLUDE_HIP_HCC_DETAIL_HOST_DEFINES_H
#define HIP_INCLUDE_HIP_HCC_DETAIL_HOST_DEFINES_H
#define USE_PROMOTE_FREE_HCC 1
// Add guard to Generic Grid Launch method
#ifndef GENERIC_GRID_LAUNCH
#define GENERIC_GRID_LAUNCH 1
#endif
#ifdef __HCC__
/**
@@ -35,7 +42,14 @@ THE SOFTWARE.
#define __host__ __attribute__((cpu))
#define __device__ __attribute__((hc))
#define __global__ __attribute__((hc_grid_launch))
//#warning "HOST DEFINE header included"
#if GENERIC_GRID_LAUNCH == 0
//#warning "original global define reached"
#define __global__ __attribute__((hc_grid_launch)) __attribute__((used))
#else
//#warning "GGL global define reached"
#define __global__ __attribute__((hc, weak))
#endif //GENERIC_GRID_LAUNCH
#define __noinline__ __attribute__((noinline))
#define __forceinline__ __attribute__((always_inline))
@@ -47,7 +61,11 @@ THE SOFTWARE.
*/
// _restrict is supported by the compiler
#define __shared__ tile_static
#define __constant__ __attribute__((address_space(1)))
#if USE_PROMOTE_FREE_HCC==1
#define __constant__ __attribute__((hc))
#else
#define __constant__ ADDRESS_SPACE_1
#endif
#else
// Non-HCC compiler
projects/hip/include/hip/hcc_detail/math_functions.h
+267
Datei anzeigen
@@ -0,0 +1,267 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#ifndef HIP_INCLUDE_HIP_HCC_DETAIL_MATH_FUNCTIONS_H
#define HIP_INCLUDE_HIP_HCC_DETAIL_MATH_FUNCTIONS_H
#include <hip/hip_runtime.h>
#include <hip/hip_vector_types.h>
#include <hip/hcc_detail/device_functions.h>
__device__ float acosf(float x);
__device__ float acoshf(float x);
__device__ float asinf(float x);
__device__ float asinhf(float x);
__device__ float atan2f(float y, float x);
__device__ float atanf(float x);
__device__ float atanhf(float x);
__device__ float cbrtf(float x);
__device__ float ceilf(float x);
__device__ float copysignf(float x, float y);
__device__ float cosf(float x);
__device__ float coshf(float x);
__device__ float cospif(float x);
//__device__ float cyl_bessel_i0f(float x);
//__device__ float cyl_bessel_i1f(float x);
__device__ float erfcf(float x);
__device__ float erfcinvf(float y);
__device__ float erfcxf(float x);
__device__ float erff(float x);
__device__ float erfinvf(float y);
__device__ float exp10f(float x);
__device__ float exp2f(float x);
__device__ float expf(float x);
__device__ float expm1f(float x);
__device__ int abs(int x);
__device__ float fabsf(float x);
__device__ float fdimf(float x, float y);
__device__ float fdividef(float x, float y);
__device__ float floorf(float x);
__device__ float fmaf(float x, float y, float z);
__device__ float fmaxf(float x, float y);
__device__ float fminf(float x, float y);
__device__ float fmodf(float x, float y);
//__device__ float frexpf(float x, int* nptr);
__device__ float hypotf(float x, float y);
__device__ float ilogbf(float x);
__device__ int isfinite(float a);
__device__ unsigned isinf(float a);
__device__ unsigned isnan(float a);
__device__ float j0f(float x);
__device__ float j1f(float x);
__device__ float jnf(int n, float x);
__device__ float ldexpf(float x, int exp);
__device__ float lgammaf(float x);
__device__ long long int llrintf(float x);
__device__ long long int llroundf(float x);
__device__ float log10f(float x);
__device__ float log1pf(float x);
__device__ float logbf(float x);
__device__ long int lrintf(float x);
__device__ long int lroundf(float x);
//__device__ float modff(float x, float *iptr);
__device__ float nanf(const char* tagp);
__device__ float nearbyintf(float x);
//__device__ float nextafterf(float x, float y);
__device__ float norm3df(float a, float b, float c);
__device__ float norm4df(float a, float b, float c, float d);
__device__ float normcdff(float y);
__device__ float normcdfinvf(float y);
__device__ float normf(int dim, const float *a);
__device__ float powf(float x, float y);
__device__ float rcbrtf(float x);
__device__ float remainderf(float x, float y);
__device__ float remquof(float x, float y, int *quo);
__device__ float rhypotf(float x, float y);
__device__ float rintf(float x);
__device__ float rnorm3df(float a, float b, float c);
__device__ float rnorm4df(float a, float b, float c, float d);
__device__ float rnormf(int dim, const float* a);
__device__ float roundf(float x);
__device__ float rsqrtf(float x);
__device__ float scalblnf(float x, long int n);
__device__ float scalbnf(float x, int n);
__device__ int signbit(float a);
__device__ void sincosf(float x, float *sptr, float *cptr);
__device__ void sincospif(float x, float *sptr, float *cptr);
__device__ float sinf(float x);
__device__ float sinhf(float x);
__device__ float sinpif(float x);
__device__ float sqrtf(float x);
__device__ float tanf(float x);
__device__ float tanhf(float x);
__device__ float tgammaf(float x);
__device__ float truncf(float x);
__device__ float y0f(float x);
__device__ float y1f(float x);
__device__ float ynf(int n, float x);
__device__ double acos(double x);
__device__ double acosh(double x);
__device__ double asin(double x);
__device__ double asinh(double x);
__device__ double atan(double x);
__device__ double atan2(double y, double x);
__device__ double atanh(double x);
__device__ double cbrt(double x);
__device__ double ceil(double x);
__device__ double copysign(double x, double y);
__device__ double cos(double x);
__device__ double cosh(double x);
__device__ double cospi(double x);
//__device__ double cyl_bessel_i0(double x);
//__device__ double cyl_bessel_i1(double x);
__device__ double erf(double x);
__device__ double erfc(double x);
__device__ double erfcinv(double y);
__device__ double erfcx(double x);
__device__ double erfinv(double x);
__device__ double exp(double x);
__device__ double exp10(double x);
__device__ double exp2(double x);
__device__ double expm1(double x);
__device__ double fabs(double x);
__device__ double fdim(double x, double y);
__device__ double floor(double x);
__device__ double fma(double x, double y, double z);
__device__ double fmax(double x, double y);
__device__ double fmin(double x, double y);
__device__ double fmod(double x, double y);
//__device__ double frexp(double x, int *nptr);
__device__ double hypot(double x, double y);
__device__ double ilogb(double x);
__device__ int isfinite(double x);
__device__ unsigned isinf(double x);
__device__ unsigned isnan(double x);
__device__ double j0(double x);
__device__ double j1(double x);
__device__ double jn(int n, double x);
__device__ double ldexp(double x, int exp);
__device__ double lgamma(double x);
__device__ long long llrint(double x);
__device__ long long llround(double x);
__device__ double log(double x);
__device__ double log10(double x);
__device__ double log1p(double x);
__device__ double log2(double x);
__device__ double logb(double x);
__device__ long int lrint(double x);
__device__ long int lround(double x);
//__device__ double modf(double x, double *iptr);
__device__ double nan(const char* tagp);
__device__ double nearbyint(double x);
__device__ double nextafter(double x, double y);
__device__ double norm(int dim, const double* t);
__device__ double norm3d(double a, double b, double c);
__device__ double norm4d(double a, double b, double c, double d);
__device__ double normcdf(double y);
__device__ double normcdfinv(double y);
__device__ double pow(double x, double y);
__device__ double rcbrt(double x);
__device__ double remainder(double x, double y);
//__device__ double remquo(double x, double y, int *quo);
__device__ double rhypot(double x, double y);
__device__ double rint(double x);
__device__ double rnorm(int dim, const double* t);
__device__ double rnorm3d(double a, double b, double c);
__device__ double rnorm4d(double a, double b, double c, double d);
__device__ double round(double x);
__device__ double rsqrt(double x);
__device__ double scalbln(double x, long int n);
__device__ double scalbn(double x, int n);
__device__ int signbit(double a);
__device__ double sin(double a);
__device__ void sincos(double x, double *sptr, double *cptr);
__device__ void sincospi(double x, double *sptr, double *cptr);
__device__ double sinh(double x);
__device__ double sinpi(double x);
__device__ double sqrt(double x);
__device__ double tan(double x);
__device__ double tanh(double x);
__device__ double tgamma(double x);
__device__ double trunc(double x);
__device__ double y0(double x);
__device__ double y1(double y);
__device__ double yn(int n, double x);
// ENDPARSER
#ifdef HIP_FAST_MATH
// Single Precision Precise Math when enabled
__device__ inline float cosf(float x) {
return __hip_fast_cosf(x);
}
__device__ inline float exp10f(float x) {
return __hip_fast_exp10f(x);
}
__device__ inline float expf(float x) {
return __hip_fast_expf(x);
}
__device__ inline float log10f(float x) {
return __hip_fast_log10f(x);
}
__device__ inline float log2f(float x) {
return __hip_fast_log2f(x);
}
__device__ inline float logf(float x) {
return __hip_fast_logf(x);
}
__device__ inline float powf(float base, float exponent) {
return __hip_fast_powf(base, exponent);
}
__device__ inline void sincosf(float x, float *s, float *c) {
return __hip_fast_sincosf(x, s, c);
}
__device__ inline float sinf(float x) {
return __hip_fast_sinf(x);
}
__device__ inline float tanf(float x) {
return __hip_fast_tanf(x);
}
#else
__device__ float sinf(float);
__device__ float cosf(float);
__device__ float tanf(float);
__device__ void sincosf(float, float*, float*);
__device__ float logf(float);
__device__ float log2f(float);
__device__ float log10f(float);
__device__ float expf(float);
__device__ float exp10f(float);
__device__ float powf(float, float);
#endif
#endif
projects/hip/include/hip/hcc_detail/texture_types.h
+45
Datei anzeigen
@@ -0,0 +1,45 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#ifndef HIP_INCLUDE_HIP_HCC_DETAIL_TEXTURE_TYPES_H
#define HIP_INCLUDE_HIP_HCC_DETAIL_TEXTURE_TYPES_H
#include<hip/hcc_detail/driver_types.h>
enum hipTextureReadMode
{
hipReadModeElementType = 0
};
enum hipTextureFilterMode
{
hipFilterModePoint = 0
};
struct textureReference {
enum hipTextureFilterMode filterMode;
unsigned normalized;
struct hipChannelFormatDesc channelDesc;
};
#endif
projects/hip/include/hip/hip_common.h
+15 -23
Datei anzeigen
@@ -1,5 +1,5 @@
/*
Copyright (c) 2015-2016 Advanced Micro Devices, Inc. All rights reserved.
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
@@ -20,40 +20,30 @@ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#pragma once
#ifndef HIP_INCLUDE_HIP_HIP_COMMON_H
#define HIP_INCLUDE_HIP_HIP_COMMON_H
// Common code included at start of every hip file.
// Auto enable __HIP_PLATFORM_HCC__ if compiling with HCC
// Other compiler (GCC,ICC,etc) need to set one of these macros explicitly
#if defined(__HCC__)
#if defined(__HCC__)
#define __HIP_PLATFORM_HCC__
#define __HIPCC__
#if defined(__HCC_ACCELERATOR__) && (__HCC_ACCELERATOR__ != 0)
#define __HIP_DEVICE_COMPILE__ 1
#else
#define __HIP_DEVICE_COMPILE__ 0
#endif
#endif
#endif //__HCC__
// Auto enable __HIP_PLATFORM_NVCC__ if compiling with NVCC
#if defined(__NVCC__)
#if defined(__NVCC__)
#define __HIP_PLATFORM_NVCC__
# ifdef __CUDACC__
# define __HIPCC__
# endif
#if defined(__CUDA_ARCH__) && (__CUDA_ARCH__ != 0)
#define __HIP_DEVICE_COMPILE__ 1
#else
#define __HIP_DEVICE_COMPILE__ 0
#ifdef __CUDACC__
#define __HIPCC__
#endif
#endif //__NVCC__
// Auto enable __HIP_DEVICE_COMPILE__ if compiled in HCC or NVCC device path
#if (defined(__HCC_ACCELERATOR__) && __HCC_ACCELERATOR__ != 0) || (defined(__CUDA_ARCH__) && __CUDA_ARCH__ != 0)
#define __HIP_DEVICE_COMPILE__ 1
#endif
#if __HIP_DEVICE_COMPILE__ == 0
// 32-bit Atomics
#define __HIP_ARCH_HAS_GLOBAL_INT32_ATOMICS__ (0)
@@ -84,3 +74,5 @@ THE SOFTWARE.
#define __HIP_ARCH_HAS_3DGRID__ (0)
#define __HIP_ARCH_HAS_DYNAMIC_PARALLEL__ (0)
#endif
#endif
projects/hip/include/hip/hip_complex.h
+4 -2
Datei anzeigen
@@ -1,5 +1,5 @@
/*
Copyright (c) 2015-2016 Advanced Micro Devices, Inc. All rights reserved.
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
@@ -20,7 +20,8 @@ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#pragma once
#ifndef HIP_INCLUDE_HIP_HIP_COMPLEX_H
#define HIP_INCLUDE_HIP_HIP_COMPLEX_H
#include <hip/hip_common.h>
@@ -32,3 +33,4 @@ THE SOFTWARE.
#error("Must define exactly one of __HIP_PLATFORM_HCC__ or __HIP_PLATFORM_NVCC__");
#endif
#endif
projects/hip/include/hip/hip_fp16.h
+5 -2
Datei anzeigen
@@ -1,5 +1,5 @@
/*
Copyright (c) 2015-2016 Advanced Micro Devices, Inc. All rights reserved.
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
@@ -20,7 +20,8 @@ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#pragma once
#ifndef HIP_INCLUDE_HIP_HIP_FP16_H
#define HIP_INCLUDE_HIP_HIP_FP16_H
#include <hip/hip_common.h>
@@ -31,3 +32,5 @@ THE SOFTWARE.
#else
#error("Must define exactly one of __HIP_PLATFORM_HCC__ or __HIP_PLATFORM_NVCC__");
#endif
#endif
projects/hip/include/hip/hip_hcc.h
+30
Datei anzeigen
@@ -0,0 +1,30 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#ifndef HIP_INCLUDE_HIP_HIP_HCC_H
#define HIP_INCLUDE_HIP_HIP_HCC_H
#if defined(__HIP_PLATFORM_HCC__) && !defined (__HIP_PLATFORM_NVCC__)
#include "hip/hcc_detail/hip_hcc.h"
#endif
#endif
projects/hip/include/hip/hip_profile.h
+7 -4
Datei anzeigen
@@ -1,5 +1,5 @@
/*
Copyright (c) 2015-2016 Advanced Micro Devices, Inc. All rights reserved.
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
@@ -20,7 +20,8 @@ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#pragma once
#ifndef HIP_INCLUDE_HIP_HIP_PROFILE_H
#define HIP_INCLUDE_HIP_HIP_PROFILE_H
#if not defined (ENABLE_HIP_PROFILE)
#define ENABLE_HIP_PROFILE 1
@@ -32,7 +33,9 @@ THE SOFTWARE.
#define HIP_BEGIN_MARKER(markerName, group) amdtBeginMarker(markerName, group, nullptr);
#define HIP_END_MARKER() amdtEndMarker();
#else
#define HIP_SCOPED_MARKER(markerName, group)
#define HIP_SCOPED_MARKER(markerName, group)
#define HIP_BEGIN_MARKER(markerName, group)
#define HIP_END_MARKER()
#define HIP_END_MARKER()
#endif
#endif
projects/hip/include/hip/hip_runtime.h
+7 -5
Datei anzeigen
@@ -1,5 +1,5 @@
/*
Copyright (c) 2015-2016 Advanced Micro Devices, Inc. All rights reserved.
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
@@ -22,7 +22,7 @@ THE SOFTWARE.
//! HIP = Heterogeneous-compute Interface for Portability
//!
//! Define a extremely thin runtime layer that allows source code to be compiled unmodified
//! Define a extremely thin runtime layer that allows source code to be compiled unmodified
//! through either AMD HCC or NVCC. Key features tend to be in the spirit
//! and terminology of CUDA, but with a portable path to other accelerators as well:
//
@@ -33,7 +33,8 @@ THE SOFTWARE.
//! hip_runtime.h : includes everything in hip_api.h, plus math builtins and kernel launch macros.
//! hip_runtime_api.h : Defines HIP API. This is a C header file and does not use any C++ features.
#pragma once
#ifndef HIP_INCLUDE_HIP_HIP_RUNTIME_H
#define HIP_INCLUDE_HIP_HIP_RUNTIME_H
// Some standard header files, these are included by hc.hpp and so want to make them avail on both
// paths to provide a consistent include env and avoid "missing symbol" errors that only appears
@@ -54,11 +55,12 @@ THE SOFTWARE.
#include <hip/hcc_detail/hip_runtime.h>
#elif defined(__HIP_PLATFORM_NVCC__) && !defined (__HIP_PLATFORM_HCC__)
#include <hip/nvcc_detail/hip_runtime.h>
#else
#else
#error("Must define exactly one of __HIP_PLATFORM_HCC__ or __HIP_PLATFORM_NVCC__");
#endif
#endif
#include <hip/hip_runtime_api.h>
#include <hip/hip_vector_types.h>
#endif
projects/hip/include/hip/hip_runtime_api.h
+7 -2
Datei anzeigen
@@ -1,5 +1,5 @@
/*
Copyright (c) 2015-2016 Advanced Micro Devices, Inc. All rights reserved.
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
@@ -27,7 +27,8 @@ THE SOFTWARE.
* This file can be compiled with a standard compiler.
*/
#pragma once
#ifndef HIP_INCLUDE_HIP_HIP_RUNTIME_API_H
#define HIP_INCLUDE_HIP_HIP_RUNTIME_API_H
#include <string.h> // for getDeviceProp
@@ -106,6 +107,7 @@ typedef struct hipDeviceProp_t {
size_t maxSharedMemoryPerMultiProcessor; ///< Maximum Shared Memory Per Multiprocessor.
int isMultiGpuBoard; ///< 1 if device is on a multi-GPU board, 0 if not.
int canMapHostMemory; ///< Check whether HIP can map host memory
int gcnArch; ///< AMD GCN Arch Value. Eg: 803, 701
} hipDeviceProp_t;
@@ -181,6 +183,7 @@ typedef enum hipError_t {
hipErrorSharedObjectSymbolNotFound = 302,
hipErrorSharedObjectInitFailed = 303,
hipErrorOperatingSystem = 304,
hipErrorSetOnActiveProcess = 305,
hipErrorInvalidHandle = 400,
hipErrorNotFound = 500,
hipErrorIllegalAddress = 700,
@@ -281,3 +284,5 @@ static inline hipError_t hipHostMalloc( T** ptr, size_t size, unsigned int flags
return hipHostMalloc((void**)ptr, size, flags);
}
#endif
#endif
projects/hip/include/hip/hip_texture.h
+5 -7
Datei anzeigen
@@ -1,5 +1,5 @@
/*
Copyright (c) 2015-2016 Advanced Micro Devices, Inc. All rights reserved.
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
@@ -20,18 +20,16 @@ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#ifndef HIP_HIP_TEXTURE_H
#define HIP_HIP_TEXTURE_H
#ifndef HIP_INCLUDE_HIP_HIP_TEXTURE_H
#define HIP_INCLUDE_HIP_HIP_TEXTURE_H
#if defined(__HIP_PLATFORM_HCC__) && !defined (__HIP_PLATFORM_NVCC__)
#include <hip/hcc_detail/hip_texture.h>
#elif defined(__HIP_PLATFORM_NVCC__) && !defined (__HIP_PLATFORM_HCC__)
#include <hip/nvcc_detail/hip_texture.h>
#else
#else
#error("Must define exactly one of __HIP_PLATFORM_HCC__ or __HIP_PLATFORM_NVCC__");
#endif
#endif
#endif
projects/hip/include/hip/hip_vector_types.h
+7 -4
Datei anzeigen
@@ -1,5 +1,5 @@
/*
Copyright (c) 2015-2016 Advanced Micro Devices, Inc. All rights reserved.
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
@@ -22,7 +22,8 @@ THE SOFTWARE.
//! hip_vector_types.h : Defines the HIP vector types.
#pragma once
#ifndef HIP_INCLUDE_HIP_HIP_VECTOR_TYPES_H
#define HIP_INCLUDE_HIP_HIP_VECTOR_TYPES_H
#include <hip/hip_common.h>
@@ -33,6 +34,8 @@ THE SOFTWARE.
#endif
#elif defined(__HIP_PLATFORM_NVCC__) && !defined (__HIP_PLATFORM_HCC__)
#include <vector_types.h>
#else
#else
#error("Must define exactly one of __HIP_PLATFORM_HCC__ or __HIP_PLATFORM_NVCC__");
#endif
#endif
#endif
projects/hip/include/hip/math_functions.h
+39
Datei anzeigen
@@ -0,0 +1,39 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#ifndef HIP_INCLUDE_HIP_MATH_FUNCTIONS_H
#define HIP_INCLUDE_HIP_MATH_FUNCTIONS_H
// Some standard header files, these are included by hc.hpp and so want to make them avail on both
// paths to provide a consistent include env and avoid "missing symbol" errors that only appears
// on NVCC path:
#if defined(__HIP_PLATFORM_HCC__) && !defined (__HIP_PLATFORM_NVCC__)
#include <hip/hcc_detail/math_functions.h>
#elif defined(__HIP_PLATFORM_NVCC__) && !defined (__HIP_PLATFORM_HCC__)
//#include <hip/nvcc_detail/math_functions.h>
#else
#error("Must define exactly one of __HIP_PLATFORM_HCC__ or __HIP_PLATFORM_NVCC__");
#endif
#endif
projects/hip/include/hip/nvcc_detail/channel_descriptor.h
+28
Datei anzeigen
@@ -0,0 +1,28 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#ifndef HIP_INCLUDE_HIP_NVCC_DETAIL_CHANNEL_DESCRIPTOR_H
#define HIP_INCLUDE_HIP_NVCC_DETAIL_CHANNEL_DESCRIPTOR_H
#include"channel_descriptor.h"
#endif
projects/hip/include/hip/nvcc_detail/hip_complex.h
+6 -6
Datei anzeigen
@@ -1,5 +1,5 @@
/*
Copyright (c) 2015-2016 Advanced Micro Devices, Inc. All rights reserved.
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
@@ -20,8 +20,8 @@ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#ifndef HIPCOMPLEX_H
#define HIPCOMPLEX_H
#ifndef HIP_INCLUDE_HIP_NVCC_DETAIL_HIP_COMPLEX_H
#define HIP_INCLUDE_HIP_NVCC_DETAIL_HIP_COMPLEX_H
#include"cuComplex.h"
@@ -64,7 +64,7 @@ __device__ __host__ static inline hipFloatComplex hipCdivf(hipFloatComplex p, hi
}
__device__ __host__ static inline float hipCabsf(hipFloatComplex z){
return cuCabsf(p, q);
return cuCabsf(z);
}
typedef cuDoubleComplex hipDoubleComplex;
@@ -85,7 +85,7 @@ __device__ __host__ static inline hipDoubleComplex hipConj(hipDoubleComplex z){
return cuConj(z);
}
__device__ __host__ static inline hipDoubleComplex hipCsqabs(hipDoubleComplex z){
__device__ __host__ static inline double hipCsqabs(hipDoubleComplex z){
return cuCabs(z) * cuCabs(z);
}
@@ -123,7 +123,7 @@ __device__ __host__ static inline hipComplex hipCfmaf(hipComplex p, hipComplex q
return cuCfmaf(p, q, r);
}
__device__ __host__ static inline hipDoubleComplex hipCfma(hipComplex p, hipComplex q, hipComplex r){
__device__ __host__ static inline hipDoubleComplex hipCfma(hipDoubleComplex p, hipDoubleComplex q, hipDoubleComplex r){
return cuCfma(p, q, r);
}
projects/hip/include/hip/nvcc_detail/hip_runtime.h
+9 -2
Datei anzeigen
@@ -1,5 +1,5 @@
/*
Copyright (c) 2015-2016 Advanced Micro Devices, Inc. All rights reserved.
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
@@ -20,7 +20,8 @@ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#pragma once
#ifndef HIP_INCLUDE_HIP_NVCC_DETAIL_HIP_RUNTIME_H
#define HIP_INCLUDE_HIP_NVCC_DETAIL_HIP_RUNTIME_H
#include <cuda_runtime.h>
@@ -35,6 +36,10 @@ do {\
kernelName<<<numblocks,numthreads,memperblock,streamId>>>(0, ##__VA_ARGS__);\
} while(0)
#define hipLaunchKernelGGL(kernelName, numblocks, numthreads, memperblock, streamId, ...) \
do {\
kernelName<<<numblocks,numthreads,memperblock,streamId>>>(__VA_ARGS__);\
} while(0)
#define hipReadModeElementType cudaReadModeElementType
@@ -105,3 +110,5 @@ kernelName<<<numblocks,numthreads,memperblock,streamId>>>(0, ##__VA_ARGS__);\
#define HIP_DYNAMIC_SHARED_ATTRIBUTE
#endif
#endif
projects/hip/include/hip/nvcc_detail/hip_runtime_api.h
+74 -10
Datei anzeigen
@@ -1,5 +1,5 @@
/*
Copyright (c) 2015-2016 Advanced Micro Devices, Inc. All rights reserved.
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
@@ -20,7 +20,8 @@ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#pragma once
#ifndef HIP_INCLUDE_HIP_NVCC_DETAIL_HIP_RUNTIME_API_H
#define HIP_INCLUDE_HIP_NVCC_DETAIL_HIP_RUNTIME_API_H
#include <cuda_runtime_api.h>
#include <cuda.h>
@@ -85,6 +86,10 @@ typedef CUdevice hipDevice_t;
typedef CUmodule hipModule_t;
typedef CUfunction hipFunction_t;
typedef CUdeviceptr hipDeviceptr_t;
typedef cudaChannelFormatKind hipChannelFormatKind;
typedef cudaChannelFormatDesc hipChannelFormatDesc;
typedef cudaTextureReadMode hipTextureReadMode;
typedef cudaArray hipArray;
// Flags that can be used with hipStreamCreateWithFlags
#define hipStreamDefault cudaStreamDefault
@@ -215,6 +220,14 @@ inline static hipError_t hipHostMalloc(void** ptr, size_t size, unsigned int fla
return hipCUDAErrorTohipError(cudaHostAlloc(ptr, size, flags));
}
inline static hipError_t hipMallocArray(hipArray** array, const hipChannelFormatDesc* desc, size_t width, size_t height, unsigned int flags) {
return hipCUDAErrorTohipError(cudaMallocArray(array, desc, width, height, flags));
}
inline static hipError_t hipFreeArray(hipArray* array) {
return hipCUDAErrorTohipError(cudaFreeArray(array));
}
inline static hipError_t hipHostGetDevicePointer(void** devPtr, void* hostPtr, unsigned int flags){
return hipCUDAErrorTohipError(cudaHostGetDevicePointer(devPtr, hostPtr, flags));
}
@@ -294,13 +307,13 @@ inline static hipError_t hipMemcpyHtoDAsync(hipDeviceptr_t dst,
inline static hipError_t hipMemcpyDtoHAsync(void* dst,
hipDeviceptr_t src, size_t size, hipStream_t stream)
{
return hipCUResultTohipError(cuMemcpyDtoH(dst, src, size));
return hipCUResultTohipError(cuMemcpyDtoHAsync(dst, src, size, stream));
}
inline static hipError_t hipMemcpyDtoDAsync(hipDeviceptr_t dst,
hipDeviceptr_t src, size_t size, hipStream_t stream)
{
return hipCUResultTohipError(cuMemcpyDtoD(dst, src, size));
return hipCUResultTohipError(cuMemcpyDtoDAsync(dst, src, size, stream));
}
inline static hipError_t hipMemcpy(void* dst, const void* src, size_t sizeBytes, hipMemcpyKind copyKind) {
@@ -321,6 +334,28 @@ inline static hipError_t hipMemcpyToSymbolAsync(const void* symbol, const void*
return hipCUDAErrorTohipError(cudaMemcpyToSymbolAsync(symbol, src, sizeBytes, offset, hipMemcpyKindToCudaMemcpyKind(copyType)));
}
inline static hipError_t hipMemcpyFromSymbol(void *dst, const void* symbolName, size_t sizeBytes, size_t offset, hipMemcpyKind kind)
{
return hipCUDAErrorTohipError(cudaMemcpyFromSymbol(dst, symbolName, sizeBytes, offset, hipMemcpyKindToCudaMemcpyKind(kind)));
}
inline static hipError_t hipMemcpyFromSymbolAsync(void *dst, const void* symbolName, size_t sizeBytes, size_t offset, hipMemcpyKind kind, hipStream_t stream)
{
return hipCUDAErrorTohipError(cudaMemcpyFromSymbolAsync(dst, symbolName, sizeBytes, offset, hipMemcpyKindToCudaMemcpyKind(kind), stream));
}
inline static hipError_t hipMemcpy2D(void* dst, size_t dpitch, const void* src, size_t spitch, size_t width, size_t height, hipMemcpyKind kind){
return hipCUDAErrorTohipError(cudaMemcpy2D(dst, dpitch, src, spitch, width, height, hipMemcpyKindToCudaMemcpyKind(kind)));
}
inline static hipError_t hipMemcpy2DToArray(hipArray *dst, size_t wOffset, size_t hOffset, const void* src, size_t spitch, size_t width, size_t height, hipMemcpyKind kind){
return hipCUDAErrorTohipError(cudaMemcpy2DToArray(dst, wOffset, hOffset, src, spitch, width, height, hipMemcpyKindToCudaMemcpyKind(kind)));
}
inline static hipError_t hipMemcpyToArray(hipArray* dst, size_t wOffset, size_t hOffset, const void* src, size_t count, hipMemcpyKind kind) {
return hipCUDAErrorTohipError(cudaMemcpyToArray(dst, wOffset, hOffset, src, count, hipMemcpyKindToCudaMemcpyKind(kind)));
}
inline static hipError_t hipDeviceSynchronize() {
return hipCUDAErrorTohipError(cudaDeviceSynchronize());
}
@@ -373,6 +408,11 @@ inline static hipError_t hipMemsetAsync(void* devPtr,int value, size_t count, hi
return hipCUDAErrorTohipError(cudaMemsetAsync(devPtr, value, count, stream));
}
inline static hipError_t hipMemsetD8(hipDeviceptr_t dest, unsigned char value, size_t sizeBytes )
{
return hipCUResultTohipError(cuMemsetD8(dest, value, sizeBytes));
}
inline static hipError_t hipGetDeviceProperties(hipDeviceProp_t *p_prop, int device)
{
cudaDeviceProp cdprop;
@@ -478,6 +518,8 @@ inline static hipError_t hipDeviceGetAttribute(int* pi, hipDeviceAttribute_t att
cdattr = cudaDevAttrMaxThreadsPerMultiProcessor; break;
case hipDeviceAttributeComputeCapabilityMajor:
cdattr = cudaDevAttrComputeCapabilityMajor; break;
case hipDeviceAttributeComputeCapabilityMinor:
cdattr = cudaDevAttrComputeCapabilityMinor; break;
case hipDeviceAttributeConcurrentKernels:
cdattr = cudaDevAttrConcurrentKernels; break;
case hipDeviceAttributePciBusId:
@@ -639,6 +681,31 @@ inline static hipError_t hipCtxEnablePeerAccess ( hipCtx_t peerCtx, unsigned in
return hipCUResultTohipError(cuCtxEnablePeerAccess(peerCtx, flags));
}
inline static hipError_t hipDevicePrimaryCtxGetState ( hipDevice_t dev, unsigned int* flags, int* active )
{
return hipCUResultTohipError(cuDevicePrimaryCtxGetState(dev, flags, active));
}
inline static hipError_t hipDevicePrimaryCtxRelease ( hipDevice_t dev)
{
return hipCUResultTohipError(cuDevicePrimaryCtxRelease(dev));
}
inline static hipError_t hipDevicePrimaryCtxRetain ( hipCtx_t* pctx, hipDevice_t dev )
{
return hipCUResultTohipError(cuDevicePrimaryCtxRetain(pctx, dev));
}
inline static hipError_t hipDevicePrimaryCtxReset ( hipDevice_t dev )
{
return hipCUResultTohipError(cuDevicePrimaryCtxReset(dev));
}
inline static hipError_t hipDevicePrimaryCtxSetFlags ( hipDevice_t dev, unsigned int flags )
{
return hipCUResultTohipError(cuDevicePrimaryCtxSetFlags(dev, flags));
}
inline static hipError_t hipMemGetAddressRange ( hipDeviceptr_t* pbase, size_t* psize, hipDeviceptr_t dptr )
{
return hipCUResultTohipError(cuMemGetAddressRange( pbase , psize , dptr));
@@ -770,11 +837,6 @@ inline static hipError_t hipDeviceGetName(char *name,int len,hipDevice_t device)
return hipCUResultTohipError(cuDeviceGetName(name,len,device));
}
inline static hipError_t hipDeviceGetPCIBusId(char* pciBusId,int len,int device)
{
return hipCUDAErrorTohipError(cudaDeviceGetPCIBusId(pciBusId,len,device));
}
inline static hipError_t hipDeviceGetPCIBusId(char* pciBusId,int len,hipDevice_t device)
{
return hipCUResultTohipError(cuDeviceGetPCIBusId(pciBusId,len,device));
@@ -886,4 +948,6 @@ inline static hipChannelFormatDesc hipCreateChannelDesc()
{
return cudaCreateChannelDesc<T>();
}
#endif
#endif //__CUDACC__
#endif //HIP_INCLUDE_HIP_NVCC_DETAIL_HIP_RUNTIME_API_H
projects/hip/include/hip/nvcc_detail/hip_texture.h
+24 -2
Datei anzeigen
@@ -1,5 +1,27 @@
#ifndef HIP_TEXTURE_H
#define HIP_TEXTURE_H
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#ifndef HIP_INCLUDE_HIP_NVCC_DETAIL_HIP_TEXTURE_H
#define HIP_INCLUDE_HIP_NVCC_DETAIL_HIP_TEXTURE_H
#include <texture_types.h>
projects/hip/include/hip/texture_types.h
+36
Datei anzeigen
@@ -0,0 +1,36 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#ifndef HIP_INCLUDE_HIP_TEXTURE_TYPES_H
#define HIP_INCLUDE_HIP_TEXTURE_TYPES_H
#include <hip/hip_common.h>
#if defined(__HIP_PLATFORM_HCC__) && !defined (__HIP_PLATFORM_NVCC__)
#include <hip/hcc_detail/texture_types.h>
#elif defined(__HIP_PLATFORM_NVCC__) && !defined (__HIP_PLATFORM_HCC__)
#include "texture_types.h"
#else
#error("Must define exactly one of __HIP_PLATFORM_HCC__ or __HIP_PLATFORM_NVCC__");
#endif
#endif
projects/hip/packaging/create_hip_samples_installer.sh
-23
Datei anzeigen
@@ -1,23 +0,0 @@
#!/bin/bash
function die {
echo "${1-Died}." >&2
exit 1
}
payload=$1
script=$2
[ "$payload" != "" ] || [ "$script" != "" ] || die "Invalid arguments!"
tmp=__extract__$RANDOM
printf "#!/bin/bash
samples_dir=\$1
[ \"\$samples_dir\" != \"\" ] || read -e -p \"Enter the path to extract the HIP samples: \" samples_dir
mkdir -p \$samples_dir
PAYLOAD=\`awk '/^__PAYLOAD_BELOW__/ {print NR + 1; exit 0; }' \$0\`
tail -n+\$PAYLOAD \$0 | tar -xz -C \$samples_dir
echo \"HIP samples installed in \$samples_dir\"
exit 0
__PAYLOAD_BELOW__\n" > "$tmp"
cat "$tmp" "$payload" > "$script" && rm "$tmp"
chmod +x "$script"
projects/hip/packaging/hip-targets-release.cmake
+41
Datei anzeigen
@@ -0,0 +1,41 @@
#----------------------------------------------------------------
# Generated CMake target import file for configuration "Release".
#----------------------------------------------------------------
# Commands may need to know the format version.
set(CMAKE_IMPORT_FILE_VERSION 1)
# Import target "hip::hip_hcc_static" for configuration "Release"
set_property(TARGET hip::hip_hcc_static APPEND PROPERTY IMPORTED_CONFIGURATIONS RELEASE)
set_target_properties(hip::hip_hcc_static PROPERTIES
IMPORTED_LINK_INTERFACE_LANGUAGES_RELEASE "CXX"
IMPORTED_LINK_INTERFACE_LIBRARIES_RELEASE "hc_am"
IMPORTED_LOCATION_RELEASE "/opt/rocm/hip/lib/libhip_hcc_static.a"
)
list(APPEND _IMPORT_CHECK_TARGETS hip::hip_hcc_static )
list(APPEND _IMPORT_CHECK_FILES_FOR_hip::hip_hcc_static "/opt/rocm/hip/lib/libhip_hcc_static.a" )
# Import target "hip::hip_hcc" for configuration "Release"
set_property(TARGET hip::hip_hcc APPEND PROPERTY IMPORTED_CONFIGURATIONS RELEASE)
set_target_properties(hip::hip_hcc PROPERTIES
IMPORTED_LINK_INTERFACE_LIBRARIES_RELEASE "hcc::hccrt;hcc::hc_am"
IMPORTED_LOCATION_RELEASE "/opt/rocm/hip/lib/libhip_hcc.so"
IMPORTED_SONAME_RELEASE "libhip_hcc.so"
)
list(APPEND _IMPORT_CHECK_TARGETS hip::hip_hcc )
list(APPEND _IMPORT_CHECK_FILES_FOR_hip::hip_hcc "/opt/rocm/hip/lib/libhip_hcc.so" )
# Import target "hip::hip_device" for configuration "Release"
set_property(TARGET hip::hip_device APPEND PROPERTY IMPORTED_CONFIGURATIONS RELEASE)
set_target_properties(hip::hip_device PROPERTIES
IMPORTED_LINK_INTERFACE_LANGUAGES_RELEASE "CXX"
IMPORTED_LOCATION_RELEASE "/opt/rocm/hip/lib/libhip_device.a"
)
list(APPEND _IMPORT_CHECK_TARGETS hip::hip_device )
list(APPEND _IMPORT_CHECK_FILES_FOR_hip::hip_device "/opt/rocm/hip/lib/libhip_device.a" )
# Commands beyond this point should not need to know the version.
set(CMAKE_IMPORT_FILE_VERSION)
projects/hip/packaging/hip-targets.cmake
+102
Datei anzeigen
@@ -0,0 +1,102 @@
# Generated by CMake 3.5.1
if("${CMAKE_MAJOR_VERSION}.${CMAKE_MINOR_VERSION}" LESS 2.5)
message(FATAL_ERROR "CMake >= 2.6.0 required")
endif()
cmake_policy(PUSH)
cmake_policy(VERSION 2.6)
#----------------------------------------------------------------
# Generated CMake target import file.
#----------------------------------------------------------------
# Commands may need to know the format version.
set(CMAKE_IMPORT_FILE_VERSION 1)
# Protect against multiple inclusion, which would fail when already imported targets are added once more.
set(_targetsDefined)
set(_targetsNotDefined)
set(_expectedTargets)
foreach(_expectedTarget hip::hip_hcc_static hip::hip_hcc hip::hip_device)
list(APPEND _expectedTargets ${_expectedTarget})
if(NOT TARGET ${_expectedTarget})
list(APPEND _targetsNotDefined ${_expectedTarget})
endif()
if(TARGET ${_expectedTarget})
list(APPEND _targetsDefined ${_expectedTarget})
endif()
endforeach()
if("${_targetsDefined}" STREQUAL "${_expectedTargets}")
set(CMAKE_IMPORT_FILE_VERSION)
cmake_policy(POP)
return()
endif()
if(NOT "${_targetsDefined}" STREQUAL "")
message(FATAL_ERROR "Some (but not all) targets in this export set were already defined.\nTargets Defined: ${_targetsDefined}\nTargets not yet defined: ${_targetsNotDefined}\n")
endif()
unset(_targetsDefined)
unset(_targetsNotDefined)
unset(_expectedTargets)
# The installation prefix configured by this project.
set(_IMPORT_PREFIX "/opt/rocm/hip")
# Create imported target hip::hip_hcc_static
add_library(hip::hip_hcc_static STATIC IMPORTED)
set_target_properties(hip::hip_hcc_static PROPERTIES
INTERFACE_INCLUDE_DIRECTORIES "${_IMPORT_PREFIX}/include;/opt/rocm/hsa/include"
INTERFACE_SYSTEM_INCLUDE_DIRECTORIES "${_IMPORT_PREFIX}/include;/opt/rocm/hsa/include"
)
# Create imported target hip::hip_hcc
add_library(hip::hip_hcc SHARED IMPORTED)
set_target_properties(hip::hip_hcc PROPERTIES
INTERFACE_INCLUDE_DIRECTORIES "${_IMPORT_PREFIX}/include;/opt/rocm/hsa/include"
INTERFACE_SYSTEM_INCLUDE_DIRECTORIES "${_IMPORT_PREFIX}/include;/opt/rocm/hsa/include"
)
# Create imported target hip::hip_device
add_library(hip::hip_device STATIC IMPORTED)
set_target_properties(hip::hip_device PROPERTIES
INTERFACE_INCLUDE_DIRECTORIES "${_IMPORT_PREFIX}/include;/opt/rocm/hsa/include"
INTERFACE_SYSTEM_INCLUDE_DIRECTORIES "${_IMPORT_PREFIX}/include;/opt/rocm/hsa/include"
)
# Load information for each installed configuration.
get_filename_component(_DIR "${CMAKE_CURRENT_LIST_FILE}" PATH)
file(GLOB CONFIG_FILES "${_DIR}/hip-targets-*.cmake")
foreach(f ${CONFIG_FILES})
include(${f})
endforeach()
# Cleanup temporary variables.
set(_IMPORT_PREFIX)
# Loop over all imported files and verify that they actually exist
foreach(target ${_IMPORT_CHECK_TARGETS} )
foreach(file ${_IMPORT_CHECK_FILES_FOR_${target}} )
if(NOT EXISTS "${file}" )
message(FATAL_ERROR "The imported target \"${target}\" references the file
\"${file}\"
but this file does not exist. Possible reasons include:
* The file was deleted, renamed, or moved to another location.
* An install or uninstall procedure did not complete successfully.
* The installation package was faulty and contained
\"${CMAKE_CURRENT_LIST_FILE}\"
but not all the files it references.
")
endif()
endforeach()
unset(_IMPORT_CHECK_FILES_FOR_${target})
endforeach()
unset(_IMPORT_CHECK_TARGETS)
# This file does not depend on other imported targets which have
# been exported from the same project but in a separate export set.
# Commands beyond this point should not need to know the version.
set(CMAKE_IMPORT_FILE_VERSION)
cmake_policy(POP)
projects/hip/packaging/hip_base.txt
+1
Datei anzeigen
@@ -33,5 +33,6 @@ set(CPACK_RPM_POST_INSTALL_SCRIPT_FILE "${PROJECT_BINARY_DIR}/postinst")
set(CPACK_RPM_PRE_UNINSTALL_SCRIPT_FILE "${PROJECT_BINARY_DIR}/prerm")
set(CPACK_RPM_PACKAGE_AUTOREQPROV " no")
set(CPACK_RPM_PACKAGE_REQUIRES "perl >= 5.0")
set(CPACK_RPM_EXCLUDE_FROM_AUTO_FILELIST_ADDITION "/opt")
set(CPACK_SOURCE_GENERATOR "TGZ")
include(CPack)
projects/hip/packaging/hip_doc.txt
+12 -4
Datei anzeigen
@@ -1,12 +1,19 @@
cmake_minimum_required(VERSION 2.8.3)
project(hip_doc)
add_custom_target(build_doxygen ALL
find_program(DOXYGEN_EXE doxygen)
if(DOXYGEN_EXE)
add_custom_target(build_doxygen ALL
COMMAND HIP_PATH=@hip_SOURCE_DIR@ doxygen @hip_SOURCE_DIR@/docs/doxygen-input/doxy.cfg)
add_custom_target(convert_md_to_html ALL
install(DIRECTORY RuntimeAPI/html DESTINATION docs/docs/RuntimeAPI)
endif()
find_program(GRIP_EXE grip)
if(GRIP_EXE)
add_custom_target(convert_md_to_html ALL
COMMAND @hip_SOURCE_DIR@/packaging/convert_md_to_html.sh @hip_SOURCE_DIR@ ${PROJECT_BINARY_DIR}/md2html)
install(DIRECTORY RuntimeAPI/html DESTINATION docs/docs/RuntimeAPI)
install(DIRECTORY md2html/ DESTINATION docs)
install(DIRECTORY md2html/ DESTINATION docs)
endif()
#############################
# Packaging steps
@@ -29,5 +36,6 @@ set(CPACK_BINARY_RPM "ON")
set(CPACK_RPM_PACKAGE_ARCHITECTURE "x86_64")
set(CPACK_RPM_PACKAGE_AUTOREQPROV " no")
set(CPACK_RPM_PACKAGE_REQUIRES "hip_base = ${CPACK_PACKAGE_VERSION}")
set(CPACK_RPM_EXCLUDE_FROM_AUTO_FILELIST_ADDITION "/opt")
set(CPACK_SOURCE_GENERATOR "TGZ")
include(CPack)
projects/hip/packaging/hip_hcc.postinst
+11 -6
Datei anzeigen
@@ -8,17 +8,22 @@ popd () {
}
ROCMDIR=/opt/rocm
HIPDIR=$ROCMDIR/hip
# Soft-link to libraries
HIPLIBFILES=$HIPDIR/lib/*
ROCMLIBDIR=$ROCMDIR/lib
HIPDIR=$ROCMDIR/hip
HIPLIBDIR=$ROCMDIR/hip/lib
# Soft-link to library files
HIPLIBFILES=$(ls -aF $HIPLIBDIR | grep -v [-/$])
mkdir -p $ROCMLIBDIR
mkdir -p $ROCMLIBDIR/cmake
pushd $ROCMLIBDIR
for f in $HIPLIBFILES
do
ln -s $f $(basename $f)
ln -s $HIPLIBDIR/$f $(basename $f)
done
ln -s $HIPDIR/lib/.hipInfo .hipInfo
# Make the hip cmake directory link.
pushd cmake
ln -s $HIPLIBDIR/cmake/hip hip
popd
popd
projects/hip/packaging/hip_hcc.prerm
+7 -3
Datei anzeigen
@@ -9,17 +9,21 @@ popd () {
}
ROCMDIR=/opt/rocm
ROCMLIBDIR=$ROCMDIR/lib
HIPDIR=$ROCMDIR/hip
HIPLIBDIR=$ROCMDIR/hip/lib
# Remove soft-links to libraries
HIPLIBFILES=$HIPDIR/lib/*
ROCMLIBDIR=$ROCMDIR/lib
HIPLIBFILES=$(ls -aF $HIPLIBDIR | grep -v [-/$])
pushd $ROCMLIBDIR
for f in $HIPLIBFILES
do
rm $(basename $f)
done
rm .hipInfo
pushd cmake
unlink hip
popd
rmdir --ignore-fail-on-non-empty cmake
popd
rmdir --ignore-fail-on-non-empty $ROCMLIBDIR
projects/hip/packaging/hip_hcc.txt
+13 -5
Datei anzeigen
@@ -5,7 +5,9 @@ install(FILES @PROJECT_BINARY_DIR@/libhip_hcc.so DESTINATION lib)
install(FILES @PROJECT_BINARY_DIR@/libhip_hcc_static.a DESTINATION lib)
install(FILES @PROJECT_BINARY_DIR@/libhip_device.a DESTINATION lib)
install(FILES @PROJECT_BINARY_DIR@/.hipInfo DESTINATION lib)
install(FILES @hip_SOURCE_DIR@/src/hip_ir.ll DESTINATION lib)
install(FILES @hip_SOURCE_DIR@/src/hip_hc.ll @hip_SOURCE_DIR@/src/hip_hc_gfx803.ll DESTINATION lib)
install(FILES @PROJECT_BINARY_DIR@/hip-config.cmake @PROJECT_BINARY_DIR@/hip-config-version.cmake DESTINATION lib/cmake/hip)
install(FILES @hip_SOURCE_DIR@/packaging/hip-targets.cmake @hip_SOURCE_DIR@/packaging/hip-targets-release.cmake DESTINATION lib/cmake/hip)
#############################
# Packaging steps
@@ -13,6 +15,11 @@ install(FILES @hip_SOURCE_DIR@/src/hip_ir.ll DESTINATION lib)
set(CPACK_SET_DESTDIR TRUE)
set(CPACK_INSTALL_PREFIX "/opt/rocm/hip")
set(CPACK_PACKAGE_NAME "hip_hcc")
if(@HCC_VERSION_MAJOR@ EQUAL 0)
set(HCC_PACKAGE_NAME "hcc_lc")
else()
set(HCC_PACKAGE_NAME "hcc")
endif()
set(CPACK_PACKAGE_DESCRIPTION_SUMMARY "HIP: Heterogenous-computing Interface for Portability [HCC]")
set(CPACK_PACKAGE_VENDOR "Advanced Micro Devices, Inc.")
set(CPACK_PACKAGE_CONTACT "Maneesh Gupta <maneesh.gupta@amd.com>")
@@ -25,9 +32,9 @@ set(CPACK_GENERATOR "TGZ;DEB;RPM")
set(CPACK_BINARY_DEB "ON")
set(CPACK_DEBIAN_PACKAGE_CONTROL_EXTRA "${PROJECT_BINARY_DIR}/postinst;${PROJECT_BINARY_DIR}/prerm")
if(@COMPILE_HIP_ATP_MARKER@)
set(CPACK_DEBIAN_PACKAGE_DEPENDS "hip_base (= ${CPACK_PACKAGE_VERSION}), hcc_lc (= @HCC_PACKAGE_VERSION@), rocm-profiler")
set(CPACK_DEBIAN_PACKAGE_DEPENDS "hip_base (= ${CPACK_PACKAGE_VERSION}), ${HCC_PACKAGE_NAME} (= @HCC_PACKAGE_VERSION@), rocm-profiler")
else()
set(CPACK_DEBIAN_PACKAGE_DEPENDS "hip_base (= ${CPACK_PACKAGE_VERSION}), hcc_lc (= @HCC_PACKAGE_VERSION@)")
set(CPACK_DEBIAN_PACKAGE_DEPENDS "hip_base (= ${CPACK_PACKAGE_VERSION}), ${HCC_PACKAGE_NAME} (= @HCC_PACKAGE_VERSION@)")
endif()
set(CPACK_BINARY_RPM "ON")
set(CPACK_RPM_PACKAGE_ARCHITECTURE "x86_64")
@@ -35,9 +42,10 @@ set(CPACK_RPM_POST_INSTALL_SCRIPT_FILE "${PROJECT_BINARY_DIR}/postinst")
set(CPACK_RPM_PRE_UNINSTALL_SCRIPT_FILE "${PROJECT_BINARY_DIR}/prerm")
set(CPACK_RPM_PACKAGE_AUTOREQPROV " no")
if(@COMPILE_HIP_ATP_MARKER@)
set(CPACK_RPM_PACKAGE_REQUIRES "hip_base = ${CPACK_PACKAGE_VERSION}, hcc_lc = @HCC_PACKAGE_VERSION@, rocm-profiler")
set(CPACK_RPM_PACKAGE_REQUIRES "hip_base = ${CPACK_PACKAGE_VERSION}, ${HCC_PACKAGE_NAME} = @HCC_PACKAGE_VERSION@, rocm-profiler")
else()
set(CPACK_RPM_PACKAGE_REQUIRES "hip_base = ${CPACK_PACKAGE_VERSION}, hcc_lc = @HCC_PACKAGE_VERSION@")
set(CPACK_RPM_PACKAGE_REQUIRES "hip_base = ${CPACK_PACKAGE_VERSION}, ${HCC_PACKAGE_NAME} = @HCC_PACKAGE_VERSION@")
endif()
set(CPACK_RPM_EXCLUDE_FROM_AUTO_FILELIST_ADDITION "/opt")
set(CPACK_SOURCE_GENERATOR "TGZ")
include(CPack)
projects/hip/packaging/hip_nvcc.txt
+1
Datei anzeigen
@@ -25,5 +25,6 @@ set(CPACK_RPM_PACKAGE_ARCHITECTURE "x86_64")
#set(CPACK_RPM_PRE_UNINSTALL_SCRIPT_FILE "${PROJECT_BINARY_DIR}/prerm")
set(CPACK_RPM_PACKAGE_AUTOREQPROV " no")
set(CPACK_RPM_PACKAGE_REQUIRES "hip_base = ${CPACK_PACKAGE_VERSION}, cuda >= 7.5")
set(CPACK_RPM_EXCLUDE_FROM_AUTO_FILELIST_ADDITION "/opt")
set(CPACK_SOURCE_GENERATOR "TGZ")
include(CPack)
projects/hip/packaging/hip_samples.txt
+1
Datei anzeigen
@@ -24,5 +24,6 @@ set(CPACK_BINARY_RPM "ON")
set(CPACK_RPM_PACKAGE_ARCHITECTURE "x86_64")
set(CPACK_RPM_PACKAGE_AUTOREQPROV " no")
set(CPACK_RPM_PACKAGE_REQUIRES "hip_base = ${CPACK_PACKAGE_VERSION}")
set(CPACK_RPM_EXCLUDE_FROM_AUTO_FILELIST_ADDITION "/opt")
set(CPACK_SOURCE_GENERATOR "TGZ")
include(CPack)
projects/hip/samples/0_Intro/bit_extract/Makefile
-5
Datei anzeigen
@@ -11,10 +11,6 @@ HIPCC=$(HIP_PATH)/bin/hipcc
ifeq (${HIP_PLATFORM}, nvcc)
HIPCC_FLAGS = -gencode=arch=compute_20,code=sm_20
endif
ifeq (${HIP_PLATFORM}, hcc)
HIPCC_FLAGS = -stdlib=libc++
endif
EXE=bit_extract
@@ -24,4 +20,3 @@ $(EXE): bit_extract.cpp
clean:
rm -f *.o $(EXE)
projects/hip/samples/0_Intro/bit_extract/bit_extract.cpp
+3 -3
Datei anzeigen
@@ -37,7 +37,7 @@ THE SOFTWARE.
}\
}
void __global__
__global__ void
bit_extract_kernel(hipLaunchParm lp, uint32_t *C_d, const uint32_t *A_d, size_t N)
{
size_t offset = (hipBlockIdx_x * hipBlockDim_x + hipThreadIdx_x);
@@ -45,7 +45,7 @@ bit_extract_kernel(hipLaunchParm lp, uint32_t *C_d, const uint32_t *A_d, size_t
for (size_t i=offset; i<N; i+=stride) {
#ifdef __HIP_PLATFORM_HCC__
C_d[i] = hc::__bitextract_u32(A_d[i], 8, 4);
C_d[i] = hc::__bitextract_u32(A_d[i], 8, 4);
#else /* defined __HIP_PLATFORM_NVCC__ or other path */
C_d[i] = ((A_d[i] & 0xf00) >> 8);
#endif
@@ -73,7 +73,7 @@ int main(int argc, char *argv[])
C_h = (uint32_t*)malloc(Nbytes);
CHECK(C_h == 0 ? hipErrorMemoryAllocation : hipSuccess );
for (size_t i=0; i<N; i++)
for (size_t i=0; i<N; i++)
{
A_h[i] = i;
}
projects/hip/samples/0_Intro/module_api/Makefile
+8 -3
Datei anzeigen
@@ -5,14 +5,19 @@ endif
HIPCC=$(HIP_PATH)/bin/hipcc
HIP_PLATFORM=$(shell $(HIP_PATH)/bin/hipconfig --compiler)
all: vcpy_kernel.code runKernel.hip.out
all: vcpy_kernel.code runKernel.hip.out defaultDriver.hip.out
runKernel.hip.out: runKernel.cpp
$(HIPCC) $(HIPCC_FLAGS) $< -o $@
launchKernelHcc.hip.out: launchKernelHcc.cpp
$(HIPCC) $(HIPCC_FLAGS) $< -o $@
defaultDriver.hip.out: defaultDriver.cpp
$(HIPCC) $(HIPCC_FLAGS) $< -o $@
vcpy_kernel.code: vcpy_kernel.cpp
$(HIPCC) --genco $(GENCO_FLAGS) $^ -o $@
$(HIPCC) --genco $(GENCO_FLAGS) $^ -o $@
clean:
rm -f *.code *.out
projects/hip/samples/0_Intro/module_api/defaultDriver.cpp
+89
Datei anzeigen
@@ -0,0 +1,89 @@
/*
Copyright (c) 2015-2017 Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#include "hip/hip_runtime.h"
#include "hip/hip_runtime_api.h"
#include<iostream>
#include<fstream>
#include<vector>
#define LEN 64
#define SIZE LEN<<2
#define fileName "test.co"
#define kernel_name "vadd"
int main(){
float *A, *B, *C;
hipDeviceptr_t Ad, Bd, Cd;
A = new float[LEN];
B = new float[LEN];
C = new float[LEN];
for(uint32_t i=0;i<LEN;i++){
A[i] = i*1.0f;
B[i] = 1.0f;
C[i] = 0.0f;
}
hipInit(0);
hipDevice_t device;
hipCtx_t context;
hipDeviceGet(&device, 0);
hipCtxCreate(&context, 0, device);
hipMalloc((void**)&Ad, SIZE);
hipMalloc((void**)&Bd, SIZE);
hipMalloc((void**)&Cd, SIZE);
hipMemcpyHtoD(Ad, A, SIZE);
hipMemcpyHtoD(Bd, B, SIZE);
hipMemcpyHtoD(Cd, C, SIZE);
hipModule_t Module;
hipFunction_t Function;
hipModuleLoad(&Module, fileName);
hipModuleGetFunction(&Function, Module, kernel_name);
int n = LEN;
void * args[4] = {&Ad, &Bd, &Cd, &n};
hipModuleLaunchKernel(Function, 1, 1, 1, LEN, 1, 1, 0, 0, args, nullptr);
hipMemcpyDtoH(C, Cd, SIZE);
int mismatchCount = 0;
for(uint32_t i=0;i<LEN;i++){
if (A[i] + B[i] != C[i]) {
mismatchCount++;
std::cout<<"error: mismatch " << A[i]<<" + "<<B[i]<<" != "<<C[i]<<std::endl;
}
}
if (mismatchCount == 0) {
std::cout << "PASSED!\n";
} else {
std::cout << "FAILED!\n";
};
hipCtxDestroy(context);
return 0;
}
projects/hip/samples/0_Intro/module_api/launchKernelHcc.cpp
+112
Datei anzeigen
@@ -0,0 +1,112 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#include "hip/hip_runtime.h"
#include "hip/hip_runtime_api.h"
#include <iostream>
#include <fstream>
#include <vector>
#ifdef __HIP_PLATFORM_HCC__
#include <hip/hip_hcc.h>
#endif
#define LEN 64
#define SIZE LEN<<2
#define fileName "vcpy_kernel.code"
#define kernel_name "hello_world"
#define HIP_CHECK(status) \
if(status != hipSuccess) {std::cout<<"Got Status: "<<status<<" at Line: "<<__LINE__<<std::endl;exit(0);}
int main(){
float *A, *B;
hipDeviceptr_t Ad, Bd;
A = new float[LEN];
B = new float[LEN];
for(uint32_t i=0;i<LEN;i++){
A[i] = i*1.0f;
B[i] = 0.0f;
}
hipInit(0);
hipDevice_t device;
hipCtx_t context;
hipDeviceGet(&device, 0);
hipCtxCreate(&context, 0, device);
hipMalloc((void**)&Ad, SIZE);
hipMalloc((void**)&Bd, SIZE);
hipMemcpyHtoD(Ad, A, SIZE);
hipMemcpyHtoD(Bd, B, SIZE);
hipModule_t Module;
hipFunction_t Function;
HIP_CHECK(hipModuleLoad(&Module, fileName));
HIP_CHECK(hipModuleGetFunction(&Function, Module, kernel_name));
uint32_t len = LEN;
uint32_t one = 1;
struct {
void * _Ad;
void * _Bd;
} args;
args._Ad = Ad;
args._Bd = Bd;
size_t size = sizeof(args);
void *config[] = {
HIP_LAUNCH_PARAM_BUFFER_POINTER, &args,
HIP_LAUNCH_PARAM_BUFFER_SIZE, &size,
HIP_LAUNCH_PARAM_END
};
HIP_CHECK(hipHccModuleLaunchKernel(Function, LEN, 1, 1, LEN, 1, 1, 0, 0, NULL, (void**)&config));
hipMemcpyDtoH(B, Bd, SIZE);
int mismatchCount = 0;
for(uint32_t i=0;i<LEN;i++){
if (A[i] != B[i]) {
mismatchCount++;
std::cout<<"error: mismatch " << A[i]<<" != "<<B[i]<<std::endl;
}
}
if (mismatchCount == 0) {
std::cout << "PASSED!\n";
} else {
std::cout << "FAILED!\n";
};
hipCtxDestroy(context);
return 0;
}
projects/hip/samples/0_Intro/module_api/runKernel.cpp
+16 -15
Datei anzeigen
@@ -1,5 +1,5 @@
/*
Copyright (c) 2015-2016 Advanced Micro Devices, Inc. All rights reserved.
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
@@ -22,9 +22,10 @@ THE SOFTWARE.
#include "hip/hip_runtime.h"
#include "hip/hip_runtime_api.h"
#include<iostream>
#include<fstream>
#include<vector>
#include <iostream>
#include <fstream>
#include <vector>
#include <hip/hip_hcc.h>
#define LEN 64
#define SIZE LEN<<2
@@ -32,6 +33,9 @@ THE SOFTWARE.
#define fileName "vcpy_kernel.code"
#define kernel_name "hello_world"
#define HIP_CHECK(status) \
if(status != hipSuccess) {std::cout<<"Got Status: "<<status<<" at Line: "<<__LINE__<<std::endl;exit(0);}
int main(){
float *A, *B;
hipDeviceptr_t Ad, Bd;
@@ -43,10 +47,10 @@ int main(){
B[i] = 0.0f;
}
hipInit(0);
hipDevice_t device;
hipCtx_t context;
hipDeviceGet(&device, 0);
hipInit(0);
hipDevice_t device;
hipCtx_t context;
hipDeviceGet(&device, 0);
hipCtxCreate(&context, 0, device);
hipMalloc((void**)&Ad, SIZE);
@@ -56,22 +60,18 @@ int main(){
hipMemcpyHtoD(Bd, B, SIZE);
hipModule_t Module;
hipFunction_t Function;
hipModuleLoad(&Module, fileName);
hipModuleGetFunction(&Function, Module, kernel_name);
HIP_CHECK(hipModuleLoad(&Module, fileName));
HIP_CHECK(hipModuleGetFunction(&Function, Module, kernel_name));
#ifdef __HIP_PLATFORM_HCC__
uint32_t len = LEN;
uint32_t one = 1;
struct {
uint32_t _hidden[6];
void * _Ad;
void * _Bd;
} args;
for (int i=0; i<6; i++) {
args._hidden[i] = 0;
}
args._Ad = Ad;
args._Bd = Bd;
@@ -98,9 +98,10 @@ int main(){
HIP_LAUNCH_PARAM_END
};
hipModuleLaunchKernel(Function, 1, 1, 1, LEN, 1, 1, 0, 0, NULL, (void**)&config);
HIP_CHECK(hipModuleLaunchKernel(Function, 1, 1, 1, LEN, 1, 1, 0, 0, NULL, (void**)&config));
hipMemcpyDtoH(B, Bd, SIZE);
int mismatchCount = 0;
for(uint32_t i=0;i<LEN;i++){
if (A[i] != B[i]) {
projects/hip/samples/0_Intro/module_api/test.cl
+12
Datei anzeigen
@@ -0,0 +1,12 @@
__kernel void memset(char in, __global int* out) {
int tx = get_global_id(0);
out[tx] = in;
}
__kernel void vadd(__global float *Ad, __global float *Bd, __global float *Cd, int N){
int tx = get_global_id(0);
if(tx < N){
Cd[tx] = Ad[tx] + Bd[tx];
}
}
projects/hip/samples/0_Intro/module_api/test.co
Ausführbare Datei
BIN
Datei anzeigen
Binäre Datei nicht angezeigt.
projects/hip/samples/0_Intro/module_api/vcpy_kernel.cpp
+1 -2
Datei anzeigen
@@ -1,5 +1,5 @@
/*
Copyright (c) 2015-2016 Advanced Micro Devices, Inc. All rights reserved.
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
@@ -27,4 +27,3 @@ extern "C" __global__ void hello_world(hipLaunchParm lp, float *a, float *b)
int tx = hipThreadIdx_x;
b[tx] = a[tx];
}
projects/hip/samples/0_Intro/square/Makefile
+1
Datei anzeigen
@@ -15,5 +15,6 @@ square.hip.out: square.hipref.cpp
clean:
rm -f *.o *.out
projects/hip/samples/0_Intro/square/square.hipref.cpp
+7 -5
Datei anzeigen
@@ -32,7 +32,7 @@ THE SOFTWARE.
}\
}
/*
/*
* Square each element in the array A and write to array C.
*/
template <typename T>
@@ -58,16 +58,18 @@ int main(int argc, char *argv[])
hipDeviceProp_t props;
CHECK(hipGetDeviceProperties(&props, 0/*deviceID*/));
printf ("info: running on device %s\n", props.name);
#ifdef __HIP_PLATFORM_HCC__
printf ("info: architecture on AMD GPU device is: %d\n",props.gcnArch);
#endif
printf ("info: allocate host mem (%6.2f MB)\n", 2*Nbytes/1024.0/1024.0);
A_h = (float*)malloc(Nbytes);
CHECK(A_h == 0 ? hipErrorMemoryAllocation : hipSuccess );
C_h = (float*)malloc(Nbytes);
CHECK(C_h == 0 ? hipErrorMemoryAllocation : hipSuccess );
// Fill with Phi + i
for (size_t i=0; i<N; i++)
for (size_t i=0; i<N; i++)
{
A_h[i] = 1.618f + i;
A_h[i] = 1.618f + i;
}
printf ("info: allocate device mem (%6.2f MB)\n", 2*Nbytes/1024.0/1024.0);
@@ -81,7 +83,7 @@ int main(int argc, char *argv[])
const unsigned threadsPerBlock = 256;
printf ("info: launch 'vector_square' kernel\n");
hipLaunchKernel(HIP_KERNEL_NAME(vector_square), dim3(blocks), dim3(threadsPerBlock), 0, 0, C_d, A_d, N);
hipLaunchKernel(vector_square, dim3(blocks), dim3(threadsPerBlock), 0, 0, C_d, A_d, N);
printf ("info: copy Device2Host\n");
CHECK ( hipMemcpy(C_h, C_d, Nbytes, hipMemcpyDeviceToHost));
projects/hip/samples/1_Utils/hipBusBandwidth/hipBusBandwidth.cpp
+154 -59
Datei anzeigen
@@ -6,9 +6,12 @@
#include "ResultDatabase.h"
enum MallocMode {MallocPinned, MallocUnpinned, MallocRegistered};
// Cmdline parms:
bool p_verbose = false;
bool p_pinned = true;
MallocMode p_malloc_mode = MallocPinned;
int p_numa_ctl = -1;
int p_iterations = 10;
int p_beatsperiteration=1;
int p_device = 0;
@@ -21,7 +24,7 @@ bool p_h2d = true;
bool p_d2h = true;
bool p_bidir = true;
//#define NO_CHECK
#define CHECK_HIP_ERROR() \
@@ -36,6 +39,14 @@ bool p_bidir = true;
}
std::string mallocModeString(int mallocMode) {
switch (mallocMode) {
case MallocPinned : return "pinned";
case MallocUnpinned: return "unpinned";
case MallocRegistered: return "registered";
default: return "mallocmode-UNKNOWN";
};
};
// ****************************************************************************
int sizeToBytes(int size) {
@@ -106,7 +117,7 @@ void RunBenchmark_H2D(ResultDatabase &resultDB)
// Create some host memory pattern
float *hostMem = NULL;
if (p_pinned)
if (p_malloc_mode == MallocPinned)
{
hipHostMalloc((void**)&hostMem, sizeof(float) * numMaxFloats);
while (hipGetLastError() != hipSuccess)
@@ -116,20 +127,33 @@ void RunBenchmark_H2D(ResultDatabase &resultDB)
--nSizes;
if (nSizes < 1)
{
std::cerr << "Error: Couldn't allocated any pinned buffer\n";
std::cerr << "Error: Couldn't allocate any pinned buffer\n";
return;
}
numMaxFloats = 1024 * (sizes[nSizes-1]) / 4;
hipHostMalloc((void**)&hostMem, sizeof(float) * numMaxFloats);
}
}
else
else if (p_malloc_mode == MallocUnpinned)
{
if (p_alignedhost) {
hostMem = (float*)aligned_alloc(p_alignedhost, numMaxFloats*sizeof(float));
} else {
hostMem = new float[numMaxFloats];
}
}
else if (p_malloc_mode == MallocRegistered)
{
if (p_numa_ctl == -1) {
hostMem = (float*)malloc(numMaxFloats*sizeof(float));
}
hipHostRegister(hostMem, numMaxFloats * sizeof(float), 0);
CHECK_HIP_ERROR();
}
else
{
assert(0);
}
for (int i = 0; i < numMaxFloats; i++)
@@ -146,7 +170,7 @@ void RunBenchmark_H2D(ResultDatabase &resultDB)
--nSizes;
if (nSizes < 1)
{
std::cerr << "Error: Couldn't allocated any device buffer\n";
std::cerr << "Error: Couldn't allocate any device buffer\n";
return;
}
numMaxFloats = 1024 * (sizes[nSizes-1]) / 4;
@@ -199,8 +223,8 @@ void RunBenchmark_H2D(ResultDatabase &resultDB)
} else {
sprintf(sizeStr, "%9s", sizeToString(thisSize).c_str());
}
resultDB.AddResult(std::string("H2D_Bandwidth") + (p_pinned ? "_Pinned" : "_Unpinned"), sizeStr, "GB/sec", speed);
resultDB.AddResult(std::string("H2D_Time") + (p_pinned ? "_Pinned" : "_Unpinned"), sizeStr, "ms", t);
resultDB.AddResult(std::string("H2D_Bandwidth") + "_" + mallocModeString(p_malloc_mode), sizeStr, "GB/sec", speed);
resultDB.AddResult(std::string("H2D_Time") + mallocModeString(p_malloc_mode), sizeStr, "ms", t);
if (p_onesize) {
break;
@@ -212,6 +236,8 @@ void RunBenchmark_H2D(ResultDatabase &resultDB)
numMaxFloats = sizeToBytes(p_onesize) / sizeof(float);
}
#ifndef NO_CHECK
// Check. First reset the host memory, then copy-back result. Then compare against original ref value.
for (int i = 0; i < numMaxFloats; i++)
{
@@ -225,24 +251,36 @@ void RunBenchmark_H2D(ResultDatabase &resultDB)
printf ("error: H2D. i=%d reference:%6.f != copyback:%6.2f\n", i, ref, hostMem[i]);
}
}
#endif
// Cleanup
hipFree((void*)device);
CHECK_HIP_ERROR();
if (p_pinned)
{
switch (p_malloc_mode) {
case MallocPinned:
hipHostFree((void*)hostMem);
CHECK_HIP_ERROR();
}
else
{
break;
case MallocUnpinned:
if (p_alignedhost) {
delete[] hostMem;
} else {
free(hostMem);
}
break;
case MallocRegistered:
hipHostUnregister(hostMem);
CHECK_HIP_ERROR();
free(hostMem);
break;
default:
assert(0);
}
hipEventDestroy(start);
hipEventDestroy(stop);
}
@@ -257,38 +295,56 @@ void RunBenchmark_D2H(ResultDatabase &resultDB)
// Create some host memory pattern
float *hostMem1;
float *hostMem2;
if (p_pinned)
if (p_malloc_mode == MallocPinned)
{
hipHostMalloc((void**)&hostMem1, sizeof(float)*numMaxFloats);
hipError_t err1 = hipGetLastError();
hipHostMalloc((void**)&hostMem2, sizeof(float)*numMaxFloats);
hipError_t err2 = hipGetLastError();
while (err1 != hipSuccess || err2 != hipSuccess)
{
// free the first buffer if only the second failed
if (err1 == hipSuccess)
hipHostFree((void*)hostMem1);
while (err1 != hipSuccess || err2 != hipSuccess)
{
// free the first buffer if only the second failed
if (err1 == hipSuccess)
hipHostFree((void*)hostMem1);
// drop the size and try again
if (p_verbose) std::cout << " - dropping size allocating pinned mem\n";
--nSizes;
if (nSizes < 1)
{
std::cerr << "Error: Couldn't allocated any pinned buffer\n";
return;
}
numMaxFloats = 1024 * (sizes[nSizes-1]) / 4;
hipHostMalloc((void**)&hostMem1, sizeof(float)*numMaxFloats);
err1 = hipGetLastError();
hipHostMalloc((void**)&hostMem2, sizeof(float)*numMaxFloats);
err2 = hipGetLastError();
}
}
else
// drop the size and try again
if (p_verbose) std::cout << " - dropping size allocating pinned mem\n";
--nSizes;
if (nSizes < 1)
{
std::cerr << "Error: Couldn't allocate any pinned buffer\n";
return;
}
numMaxFloats = 1024 * (sizes[nSizes-1]) / 4;
hipHostMalloc((void**)&hostMem1, sizeof(float)*numMaxFloats);
err1 = hipGetLastError();
hipHostMalloc((void**)&hostMem2, sizeof(float)*numMaxFloats);
err2 = hipGetLastError();
}
}
else if (p_malloc_mode == MallocUnpinned)
{
hostMem1 = new float[numMaxFloats];
hostMem2 = new float[numMaxFloats];
}
else if (p_malloc_mode == MallocRegistered)
{
if (p_numa_ctl == -1) {
hostMem1 = (float*)malloc(numMaxFloats*sizeof(float));
hostMem2 = (float*)malloc(numMaxFloats*sizeof(float));
}
hipHostRegister(hostMem1, numMaxFloats * sizeof(float), 0);
CHECK_HIP_ERROR();
hipHostRegister(hostMem2, numMaxFloats * sizeof(float), 0);
CHECK_HIP_ERROR();
}
else
{
assert(0);
}
for (int i=0; i<numMaxFloats; i++)
hostMem1[i] = i % 77;
@@ -301,7 +357,7 @@ void RunBenchmark_D2H(ResultDatabase &resultDB)
--nSizes;
if (nSizes < 1)
{
std::cerr << "Error: Couldn't allocated any device buffer\n";
std::cerr << "Error: Couldn't allocate any device buffer\n";
return;
}
numMaxFloats = 1024 * (sizes[nSizes-1]) / 4;
@@ -358,8 +414,8 @@ void RunBenchmark_D2H(ResultDatabase &resultDB)
} else {
sprintf(sizeStr, "%9s", sizeToString(thisSize).c_str());
}
resultDB.AddResult(std::string("D2H_Bandwidth") + (p_pinned ? "_Pinned" : "_Unpinned"), sizeStr, "GB/sec", speed);
resultDB.AddResult(std::string("D2H_Time") + (p_pinned ? "_Pinned" : "_Unpinned"), sizeStr, "ms", t);
resultDB.AddResult(std::string("D2H_Bandwidth") +"_" + mallocModeString(p_malloc_mode) , sizeStr, "GB/sec", speed);
resultDB.AddResult(std::string("D2H_Time") +"_" + mallocModeString(p_malloc_mode) , sizeStr, "ms", t);
if (p_onesize) {
break;
}
@@ -381,20 +437,31 @@ void RunBenchmark_D2H(ResultDatabase &resultDB)
// Cleanup
hipFree((void*)device);
CHECK_HIP_ERROR();
if (p_pinned)
{
switch (p_malloc_mode) {
case MallocPinned:
hipHostFree((void*)hostMem1);
CHECK_HIP_ERROR();
hipHostFree((void*)hostMem2);
CHECK_HIP_ERROR();
}
else
{
break;
case MallocUnpinned:
delete[] hostMem1;
delete[] hostMem2;
hipEventDestroy(start);
hipEventDestroy(stop);
break;
case MallocRegistered:
hipHostUnregister(hostMem1);
CHECK_HIP_ERROR();
free(hostMem1);
hipHostUnregister(hostMem2);
free(hostMem2);
break;
default:
assert(0);
}
hipEventDestroy(start);
hipEventDestroy(stop);
}
@@ -409,7 +476,7 @@ void RunBenchmark_Bidir(ResultDatabase &resultDB)
// Create some host memory pattern
float *hostMem[2] = {NULL, NULL};
if (p_pinned)
if (p_malloc_mode == MallocPinned)
{
while (1)
{
@@ -424,18 +491,34 @@ void RunBenchmark_Bidir(ResultDatabase &resultDB)
--nSizes;
if (nSizes < 1)
{
std::cerr << "Error: Couldn't allocated any pinned buffer\n";
std::cerr << "Error: Couldn't allocate any pinned buffer\n";
return;
}
numMaxFloats = 1024 * (sizes[nSizes-1]) / 4;
}
}
}
else
else if (p_malloc_mode == MallocUnpinned)
{
hostMem[0] = new float[numMaxFloats];
hostMem[1] = new float[numMaxFloats];
}
else if (p_malloc_mode == MallocRegistered)
{
if (p_numa_ctl == -1) {
hostMem[0] = (float*)malloc(numMaxFloats*sizeof(float));
hostMem[1] = (float*)malloc(numMaxFloats*sizeof(float));
}
hipHostRegister(hostMem[0], numMaxFloats * sizeof(float), 0);
CHECK_HIP_ERROR();
hipHostRegister(hostMem[1], numMaxFloats * sizeof(float), 0);
CHECK_HIP_ERROR();
}
else
{
assert(0);
}
for (int i = 0; i < numMaxFloats; i++)
{
@@ -459,7 +542,7 @@ void RunBenchmark_Bidir(ResultDatabase &resultDB)
--nSizes;
if (nSizes < 1)
{
std::cerr << "Error: Couldn't allocated any device buffer\n";
std::cerr << "Error: Couldn't allocate any device buffer\n";
return;
}
numMaxFloats = 1024 * (sizes[nSizes-1]) / 4;
@@ -512,8 +595,8 @@ void RunBenchmark_Bidir(ResultDatabase &resultDB)
double speed = (double(sizeToBytes(thisSize)) / (1000*1000)) / t;
char sizeStr[256];
sprintf(sizeStr, "%9s", sizeToString(thisSize).c_str());
resultDB.AddResult(std::string("Bidir_Bandwidth") + (p_pinned ? "_Pinned" : "_Unpinned"), sizeStr, "GB/sec", speed);
resultDB.AddResult(std::string("Bidir_Time") + (p_pinned ? "_Pinned" : "_Unpinned"), sizeStr, "ms", t);
resultDB.AddResult(std::string("Bidir_Bandwidth") + "_" + mallocModeString(p_malloc_mode), sizeStr, "GB/sec", speed);
resultDB.AddResult(std::string("Bidir_Time") + "_" + mallocModeString(p_malloc_mode), sizeStr, "ms", t);
}
}
@@ -521,17 +604,27 @@ void RunBenchmark_Bidir(ResultDatabase &resultDB)
hipFree((void*)deviceMem[0]);
hipFree((void*)deviceMem[1]);
CHECK_HIP_ERROR();
if (p_pinned)
{
switch (p_malloc_mode) {
case MallocPinned:
hipHostFree((void*)hostMem[0]);
hipHostFree((void*)hostMem[1]);
CHECK_HIP_ERROR();
}
else
{
break;
case MallocUnpinned:
delete[] hostMem[0];
delete[] hostMem[1];
}
break;
case MallocRegistered:
for (int i=0; i<2; i++) {
hipHostUnregister(hostMem[i]);
CHECK_HIP_ERROR();
free(hostMem[i]);
}
break;
default:
assert(0);
};
hipEventDestroy(start);
hipEventDestroy(stop);
hipStreamDestroy(stream[0]);
@@ -557,7 +650,7 @@ void printConfig() {
hipDeviceProp_t props;
hipGetDeviceProperties(&props, p_device);
printf ("Device:%s Mem=%.1fGB #CUs=%d Freq=%.0fMhz Pinned=%s\n", props.name, props.totalGlobalMem/1024.0/1024.0/1024.0, props.multiProcessorCount, props.clockRate/1000.0, p_pinned ? "YES" : "NO");
printf ("Device:%s Mem=%.1fGB #CUs=%d Freq=%.0fMhz MallocMode=%s\n", props.name, props.totalGlobalMem/1024.0/1024.0/1024.0, props.multiProcessorCount, props.clockRate/1000.0, mallocModeString(p_malloc_mode).c_str());
}
void help() {
@@ -601,7 +694,9 @@ int parseStandardArguments(int argc, char *argv[])
failed("Bad onesize argument");
}
} else if (!strcmp(arg, "--unpinned")) {
p_pinned = 0;
p_malloc_mode = MallocUnpinned;
} else if (!strcmp(arg, "--registered")) {
p_malloc_mode = MallocRegistered;
} else if (!strcmp(arg, "--h2d")) {
p_h2d = true;
p_d2h = false;
projects/hip/samples/1_Utils/hipCommander/Makefile
-3
Datei anzeigen
@@ -10,9 +10,6 @@ OPT=-O3
CXXFLAGS = $(OPT) --std=c++11
HIP_PLATFORM=$(shell $(HIP_PATH)/bin/hipconfig --platform)
ifeq (${HIP_PLATFORM}, hcc)
CXXFLAGS += " -stdlib=libc++"
endif
CODE_OBJECTS=nullkernel.hsaco
projects/hip/samples/1_Utils/hipCommander/TODO
-50
Datei anzeigen
@@ -1,50 +0,0 @@
_ Add AQL kernel.
_ Fix &*kernel command so the kernel name/type is an argument not a new command.
_ Add command to parse only.
_ Add regression to parse all the hcm files.
_ Partition HCC, HIP, HSA, OpenCL commands into separate files.
_ Show time for back-to-back copies.
_ Add variables.
%loopcnt
./hipCommander %loopcnt=4
_ Add datasize command.
_ Add ( ) to parsing.
_ Add argument parsing and checking.
_ Add verbose option to print each step of setup.
- print deliniater between setup and run. Add run start message.
- print sizes of all buffers.
- print each command before running.
- show start/stop of timer routine.
_
_ Clear documentation on what each oepration does.
_ Add time instrumentation for each command.
_ Add pcie atomic.
_ Add tests for negative cases, ie endloop w/o opening loop.
README tips
---
- HIP_API_TRACE combined with -v is useful to track the exact commands generates by hipCommander.
Other ideas:
---
[ ] Perf guide : stream creation very slow on HCC and should be avoided.
Scratch:
projects/hip/samples/1_Utils/hipCommander/hipCommander.cpp
+1 -2
Datei anzeigen
@@ -11,7 +11,6 @@
#include <elf.h>
#include <hsa/hsa.h>
#include <hc.hpp>
#include <hip/hcc_detail/hcc_acc.h>
#endif
#include <sys/time.h>
@@ -23,7 +22,7 @@ bool g_printedTiming = false;
// Cmdline parms:
int p_device = 0;
const char* p_command = "H2D; NullKernel; D2H";
const char* p_command = "setstream(1); H2D; NullKernel; D2H;";
const char* p_file = nullptr;
unsigned p_verbose = 0x0;
unsigned p_db = 0x0;
projects/hip/samples/1_Utils/hipCommander/perf/scripts/latency_2_d2h_h2d.hcm
+9
Datei anzeigen
@@ -0,0 +1,9 @@
setstream(1);
loop(10); D2H; H2D; streamsync;D2H; H2D; streamsync; endloop(1);
loop(10); D2H; H2D; streamsync;D2H; H2D; streamsync; endloop(1);
loop(100); D2H; H2D; streamsync;D2H; H2D; streamsync; endloop(1);
loop(100); D2H; H2D; streamsync;D2H; H2D; streamsync; endloop(1);
loop(1000); D2H; H2D; streamsync;D2H; H2D; streamsync; endloop(1);
loop(1000); D2H; H2D; streamsync;D2H; H2D; streamsync; endloop(1);
loop(10000); D2H; H2D; streamsync;D2H; H2D; streamsync; endloop(1);
loop(10000); D2H; H2D; streamsync;D2H; H2D; streamsync; endloop(1);
projects/hip/samples/1_Utils/hipCommander/perf/scripts/latency_2_d2h_kernel.hcm
+9
Datei anzeigen
@@ -0,0 +1,9 @@
setstream(1);
loop(10); D2H; NullKernel; streamsync;D2H; NullKernel; streamsync; endloop(1);
loop(10); D2H; NullKernel; streamsync;D2H; NullKernel; streamsync; endloop(1);
loop(100); D2H; NullKernel; streamsync;D2H; NullKernel; streamsync; endloop(1);
loop(100); D2H; NullKernel; streamsync;D2H; NullKernel; streamsync; endloop(1);
loop(1000); D2H; NullKernel; streamsync;D2H; NullKernel; streamsync; endloop(1);
loop(1000); D2H; NullKernel; streamsync;D2H; NullKernel; streamsync; endloop(1);
loop(10000); D2H; NullKernel; streamsync;D2H; NullKernel; streamsync; endloop(1);
loop(10000); D2H; NullKernel; streamsync;D2H; NullKernel; streamsync; endloop(1);
projects/hip/samples/1_Utils/hipCommander/perf/scripts/latency_2_d2h_sync_h2d.hcm
+9
Datei anzeigen
@@ -0,0 +1,9 @@
setstream(1);
loop(10); D2H; streamsync; H2D; streamsync;D2H; streamsync; H2D; streamsync; endloop(1);
loop(10); D2H; streamsync; H2D; streamsync;D2H; streamsync; H2D; streamsync; endloop(1);
loop(100); D2H; streamsync; H2D; streamsync;D2H; streamsync; H2D; streamsync; endloop(1);
loop(100); D2H; streamsync; H2D; streamsync;D2H; streamsync; H2D; streamsync; endloop(1);
loop(1000); D2H; streamsync; H2D; streamsync;D2H; streamsync; H2D; streamsync; endloop(1);
loop(1000); D2H; streamsync; H2D; streamsync;D2H; streamsync; H2D; streamsync; endloop(1);
loop(10000); D2H; streamsync; H2D; streamsync;D2H; streamsync; H2D; streamsync; endloop(1);
loop(10000); D2H; streamsync; H2D; streamsync;D2H; streamsync; H2D; streamsync; endloop(1);
projects/hip/samples/1_Utils/hipCommander/perf/scripts/latency_2_d2h_sync_kernel.hcm
+9
Datei anzeigen
@@ -0,0 +1,9 @@
setstream(1);
loop(10); D2H; streamsync; NullKernel; streamsync;D2H; streamsync; NullKernel; streamsync; endloop(1);
loop(10); D2H; streamsync; NullKernel; streamsync;D2H; streamsync; NullKernel; streamsync; endloop(1);
loop(100); D2H; streamsync; NullKernel; streamsync;D2H; streamsync; NullKernel; streamsync; endloop(1);
loop(100); D2H; streamsync; NullKernel; streamsync;D2H; streamsync; NullKernel; streamsync; endloop(1);
loop(1000); D2H; streamsync; NullKernel; streamsync;D2H; streamsync; NullKernel; streamsync; endloop(1);
loop(1000); D2H; streamsync; NullKernel; streamsync;D2H; streamsync; NullKernel; streamsync; endloop(1);
loop(10000); D2H; streamsync; NullKernel; streamsync;D2H; streamsync; NullKernel; streamsync; endloop(1);
loop(10000); D2H; streamsync; NullKernel; streamsync;D2H; streamsync; NullKernel; streamsync; endloop(1);
projects/hip/samples/1_Utils/hipCommander/perf/scripts/latency_2_h2d_d2h.hcm
+9
Datei anzeigen
@@ -0,0 +1,9 @@
setstream(1);
loop(10); H2D; D2H; streamsync;H2D; D2H; streamsync; endloop(1);
loop(10); H2D; D2H; streamsync;H2D; D2H; streamsync; endloop(1);
loop(100); H2D; D2H; streamsync;H2D; D2H; streamsync; endloop(1);
loop(100); H2D; D2H; streamsync;H2D; D2H; streamsync; endloop(1);
loop(1000); H2D; D2H; streamsync;H2D; D2H; streamsync; endloop(1);
loop(1000); H2D; D2H; streamsync;H2D; D2H; streamsync; endloop(1);
loop(10000); H2D; D2H; streamsync;H2D; D2H; streamsync; endloop(1);
loop(10000); H2D; D2H; streamsync;H2D; D2H; streamsync; endloop(1);
projects/hip/samples/1_Utils/hipCommander/perf/scripts/latency_2_h2d_kernel.hcm
+9
Datei anzeigen
@@ -0,0 +1,9 @@
setstream(1);
loop(10); H2D; NullKernel; streamsync;H2D; NullKernel; streamsync; endloop(1);
loop(10); H2D; NullKernel; streamsync; H2D; NullKernel; streamsync;endloop(1);
loop(100); H2D; NullKernel; streamsync; H2D; NullKernel; streamsync;endloop(1);
loop(100); H2D; NullKernel; streamsync; H2D; NullKernel; streamsync; endloop(1);
loop(1000); H2D; NullKernel; streamsync; H2D; NullKernel; streamsync; endloop(1);
loop(1000); H2D; NullKernel; streamsync; H2D; NullKernel; streamsync; endloop(1);
loop(10000); H2D; NullKernel; streamsync;H2D; NullKernel; streamsync; endloop(1);
loop(10000); H2D; NullKernel; streamsync; H2D; NullKernel; streamsync; endloop(1);
projects/hip/samples/1_Utils/hipCommander/perf/scripts/latency_2_h2d_sync_d2h.hcm
+9
Datei anzeigen
@@ -0,0 +1,9 @@
setstream(1);
loop(10); H2D; streamsync; D2H; streamsync;H2D; streamsync; D2H; streamsync; endloop(1);
loop(10); H2D; streamsync; D2H; streamsync;H2D; streamsync; D2H; streamsync; endloop(1);
loop(100); H2D; streamsync; D2H; streamsync;H2D; streamsync; D2H; streamsync; endloop(1);
loop(100); H2D; streamsync; D2H; streamsync;H2D; streamsync; D2H; streamsync; endloop(1);
loop(1000); H2D; streamsync; D2H; streamsync;H2D; streamsync; D2H; streamsync; endloop(1);
loop(1000); H2D; streamsync; D2H; streamsync;H2D; streamsync; D2H; streamsync; endloop(1);
loop(10000); H2D; streamsync; D2H; streamsync;H2D; streamsync; D2H; streamsync; endloop(1);
loop(10000); H2D; streamsync; D2H; streamsync;H2D; streamsync; D2H; streamsync; endloop(1);
projects/hip/samples/1_Utils/hipCommander/perf/scripts/latency_2_h2d_sync_kernel.hcm
+9
Datei anzeigen
@@ -0,0 +1,9 @@
setstream(1);
loop(10); H2D; streamsync; NullKernel; streamsync;H2D; streamsync; NullKernel; streamsync; endloop(1);
loop(10); H2D; streamsync; NullKernel; streamsync;H2D; streamsync; NullKernel; streamsync; endloop(1);
loop(100); H2D; streamsync; NullKernel; streamsync;H2D; streamsync; NullKernel; streamsync; endloop(1);
loop(100); H2D; streamsync; NullKernel; streamsync;H2D; streamsync; NullKernel; streamsync; endloop(1);
loop(1000); H2D; streamsync; NullKernel; streamsync;H2D; streamsync; NullKernel; streamsync; endloop(1);
loop(1000); H2D; streamsync; NullKernel; streamsync;H2D; streamsync; NullKernel; streamsync; endloop(1);
loop(10000); H2D; streamsync; NullKernel; streamsync;H2D; streamsync; NullKernel; streamsync; endloop(1);
loop(10000); H2D; streamsync; NullKernel; streamsync;H2D; streamsync; NullKernel; streamsync; endloop(1);

Einige Dateien werden nicht angezeigt, da zu viele Dateien in diesem Diff geändert wurden Mehr anzeigen