search

gpgpu-sim / gpgpu-sim_distribution

GPGPU-Sim provides a detailed simulation model of contemporary NVIDIA GPUs running CUDA and/or OpenCL workloads. It includes support for features such as TensorCores and CUDA Dynamic Parallelism as well as a performance visualization tool, AerialVisoin, and an integrated energy model, GPUWattch.
Other
1.1k stars 505 forks source link

Supporting CUDA 7.5 #16

Open andrewboktor opened 8 years ago

andrewboktor commented 8 years ago

We have a couple of problems running with cuda 7.5 according to an email discussion I had with a GPGPU-Sim user. He was able to get his benchmark(s) running by doing some modifications which are available on his/her fork https://github.com/andpic/gpgpu-sim_distribution

Reported by GPGPU-Sim user: OpenCL 1.1 C++ API compatibility. I added to gpgpu-sim's source code a couple of API functions that are necessary to be able to compile code with the C++ interface. In particular, the "clRetain" functions and "clCreateKernelsInProgram". PTX parsing. The problem was due to a constant argument passed to the kernel. In other words, gpgpu doesn't like when kernel arguments are not passed with buffers. I fixed the problem by just changing that .const to blank space during the parsing phase. ptxas output parsing. Ptxas outputs an extra line with the gmem keyword, which is not recognised by gpgpu-sim's parser. I added a bash trick to exclude that line when the file is read.

The simulator compiles and runs well with the latest CUDA 7.5.

andrewboktor commented 8 years ago

15 could potentially be related to this

HuangGroupStorage commented 6 years ago

How to solve this error? Error: No PTX sections found with sm capability that is lower than current forced maximum capability minimum ptx capability found = 0, maximum forced ptx capability = 20 User might want to change either the forced maximum capability from gpgpusim configuration or update the compilation to generate the required PTX version