NASP/cuda-samples
84
0
Fork 0
mirror of https://github.com/NVIDIA/cuda-samples.git synced 2026-10-13 00:08:26 +08:00
Code Issues Packages Projects Releases Wiki Activity

CUDA 13.4 samples update - v13.4-public

Browse Source 
Release 13.4 of the CUDA Samples supported by CUDA Toolkit 13.4.
See Changelog for more information.
This commit is contained in:
Dheemanth
2026-09-09 15:07:08 -07:00
committed by GitHub
parent b7c5481c55
commit 5443602d89
871 changed files with 6843 additions and 321849 deletions
Download Patch File Download Diff File Expand all files Collapse all files

27
cpp/4_CUDA_Libraries/libcuxxMdspan/CMakeLists.txt
View File

@@ -1,9 +1,16 @@
cmake_minimum_required(VERSION 3.20)
list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/../../../cmake/Modules")
include("${CMAKE_CURRENT_SOURCE_DIR}/../../../cmake/DetectCudaArch.cmake")
if(SAMPLE_SKIP_BUILD)
return()
endif()
project(libcuxxMdspan LANGUAGES C CXX CUDA)
include("${CMAKE_CURRENT_SOURCE_DIR}/../../../cmake/CudaSampleCommon.cmake")
# Disable response file for libraries on QNX as qcc does not support lib paths with double quotes
if(CMAKE_SYSTEM_NAME STREQUAL "QNX")
set(CMAKE_CUDA_USE_RESPONSE_FILE_FOR_LIBRARIES OFF)
@@ -11,16 +18,6 @@ endif()
find_package(CUDAToolkit REQUIRED)
set(CMAKE_POSITION_INDEPENDENT_CODE ON)
set(CMAKE_CUDA_ARCHITECTURES 75 80 86 87 89 90 100 110 120)
set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Wno-deprecated-gpu-targets")
if(ENABLE_CUDA_DEBUG)
set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -G") # enable cuda-gdb (may significantly affect performance on some targets)
else()
set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -lineinfo") # add line information to all builds for debug tools (exclusive to -G option)
endif()
# Fetch CCCL (CUB + libcu++ + Thrust) via CPM. The toolkit that ships
# with CUDA 13.2 bundles CCCL 3.2, but this sample uses APIs added in
# CCCL 3.3. Pinning the tag here lets the sample build on any toolkit
@@ -68,25 +65,19 @@ if(NOT TARGET dlpack::dlpack)
endif()
endif()
# Include directories and libraries
include_directories(../../../Common)
# Source file
# Add target for libcuxxMdspan
add_executable(libcuxxMdspan libcuxxMdspan.cu)
target_compile_options(libcuxxMdspan PRIVATE $<$<COMPILE_LANGUAGE:CUDA>:--extended-lambda>)
target_compile_features(libcuxxMdspan PRIVATE cxx_std_17 cuda_std_17)
set_target_properties(libcuxxMdspan PROPERTIES CUDA_SEPARABLE_COMPILATION ON)
target_link_libraries(libcuxxMdspan PRIVATE
CUDA::cudart
CCCL::CCCL
dlpack::dlpack
)
# Include installation configuration
include(${CMAKE_CURRENT_SOURCE_DIR}/../../../cmake/InstallSamples.cmake)
include(InstallSamples)
setup_samples_install()

20
cpp/4_CUDA_Libraries/libcuxxMdspan/libcuxxMdspan.cu
View File

@@ -113,6 +113,26 @@ struct DLTensorStorage
::DLTensor tensor{};
cuda::std::array<cuda::std::int64_t, 2> shape{};
cuda::std::array<cuda::std::int64_t, 2> strides{};
DLTensorStorage() = default;
/* tensor.shape / tensor.strides point back into this object's own shape /
* strides members, so the object is self-referential. Any copy or move
* (including the by-value return from make_row_major_dltensor) must re-seat
* those pointers at the new location. Relying on NRVO to elide the copy is
* not portable -- NRVO is not guaranteed for named returns -- and a stale
* pointer into a moved-from temporary makes cuda::to_device_mdspan read
* garbage extents and abort. */
DLTensorStorage(const DLTensorStorage &other) { *this = other; }
DLTensorStorage &operator=(const DLTensorStorage &other)
{
tensor = other.tensor;
shape = other.shape;
strides = other.strides;
tensor.shape = shape.data();
tensor.strides = strides.data();
return *this;
}
};
static DLTensorStorage make_row_major_dltensor(float *device_ptr, int rows, int cols, int device_ordinal)