search

SuperScientificSoftwareLaboratory / TileSpGEMM

Source code of the PPoPP '22 paper: "TileSpGEMM: A Tiled Algorithm for Parallel Sparse General Matrix-Matrix Multiplication on GPUs" by Yuyao Niu, Zhengyang Lu, Haonan Ji, Shuhui Song, Zhou Jin, and Weifeng Liu.
38 stars 7 forks source link

run tilespgemm in cuda 12.2 #5

Open baharmahab opened 10 months ago

baharmahab commented 10 months ago

hi. I am a student who wants to use your "TileSPGEMM"code.

But unfortunately I am facing this error.

nvcc -O3 -w -arch=compute_61 -code=sm_86 -gencode=arch=compute_61,code=sm_86 -Xcompiler -fopenmp -Xcompiler -mfma main.cu -o test -I/usr/local/cuda-12.2/include -L/usr/local/cuda-12.2/lib64 -lcudart -lcusparse -D VALUE_TYPE=double /usr/local/cuda-12.2/include/cuda_bf16.hpp(575): error: no instance of overloaded function "half::half" matches the specified type attribute((host)) attribute((device)) inline attribute((always_inline)) half::half(const __nv_bfloat16 f) ^

1 error detected in the compilation of "main.cu". make: *** [Makefile:29: make] Error 2

Can you please guide me? my linux is ubuntu 22.04 and my gpu is GTX 1660 TI. and i have cuda toolkit 12.2

thanks.

yh-tian commented 2 months ago

i try to delete the file "cuda_fp16.h" and "cuda_fp16.hpp", and then it can run