56fd68e7dd
CCCL (CUDA C++ Core Libraries) provides: - CUB: device/block/warp-level GPU primitives (reduce, scan, sort, topk) - Thrust: high-level parallel algorithms (transform_reduce, sort, scan) - libcudacxx: CUDA C++ standard library (atomics, barriers, memory) - cudax: experimental features (memory resources, allocators) - Tuning policies: per-SM hardware-specific algorithm parameters Competition optimization vectors mapped to CCCL: - Output TPS (83% weight): warp_reduce, block_reduce, device_topk - Input TPS (14% weight): device_scan, block_load, prefetch - Cache TPS (3% weight): prefix caching strategy patterns - Memory (0.9 util): pooled/cached/buddy allocators Source: https://github.com/NVIDIA/cccl (shallow clone, HEAD only) License: Apache-2.0
21 lines
872 B
Plaintext
21 lines
872 B
Plaintext
This example shows how to link a Thrust program contained in
|
|
a .cu file with a C++ program contained in a .cpp file. Note
|
|
that device_vector only appears in the .cu file while host_vector
|
|
appears in both. This relects the fact that algorithms on device
|
|
vectors are only available when the contents of the program are
|
|
located in a .cu file and compiled with the nvcc compiler.
|
|
|
|
On a Linux system where Thrust is installed in the default location
|
|
we can use the following procedure to compile the two parts of the
|
|
program and link them together.
|
|
|
|
$ nvcc -O2 -c device.cu
|
|
$ g++ -O2 -c host.cpp -I/usr/local/cuda/include/
|
|
$ nvcc -o tester device.o host.o
|
|
|
|
Alternatively, we can use g++ to perform final linking step.
|
|
|
|
$ nvcc -O2 -c device.cu
|
|
$ g++ -O2 -c host.cpp -I/usr/local/cuda/include/
|
|
$ g++ -o tester device.o host.o -L/usr/local/cuda/lib64 -lcudart
|