MNN not working with MPS partitioned Nvidia GPU

[amarflybot]

Background

The Multi-Process Service (MPS) is an alternative, binary-compatible implementation of the CUDA Application Programming Interface (API). The MPS runtime architecture is designed to transparently enable co-operative multi-process CUDA applications, typically MPI jobs, to utilize Hyper-Q capabilities on A10 GPUs.
MPS Link

平台(如果交叉编译请再附上交叉编译目标平台):

Platform(Include target platform as well if cross-compiling):

CUDA 12.4
Platform: amd (x86-64)
GPU: Nvidia

Github版本:

Github Version:

version_mnn= 2.9.0

编译方式:

Compiling Method

cmake -DCMAKE_BUILD_TYPE=Release -DMNN_USE_THREAD_POOL=OFF -DMNN_CUDA=ON -DMNN_OPENMP=ON ..

编译日志:

Build Log:

Build is fine, and I can integrate the built MNN binary with a http

Issue:

MNN is not able to get CUDA memory when MPS was enabled.

Execution Log:

The device support i8sdot:0, support fp16:0, support i8mm: 0
Error for /build/MNN/source/backend/cuda/core/runtime/CUDARuntime.cpp, 47
face_detection: /build/MNN/source/backend/cuda/core/runtime/CUDARuntime.cpp:47: MNN::CUDARuntime::CUDARuntime(int): Assertion `res' failed.
Aborted (core dumped)