How_to_optimize_in_GPU

How_to_optimize_in_GPU copied to clipboard

- 报错现象： `./How_to_optimize_in_GPU/reduce# make ` ` nvcc -o bin/reduce_v0 reduce_v0_baseline.cu ` ` /usr/bin/ld: cannot open output file bin/reduce_v0: No such file or directory ` ` collect2: error: ld returned 1...

shifang99

Sgemv_v0里 “我们将每个block设置为256个线程” 是不是应该是128个线程

xxz-wow

which modifications are necessary for arbitrary input size N are necessary for reduce_v7? I changed it to ```400*500``` and get wrong answer.

hygxy

question on reduce

3

hello, May I ask a question; why N must be 32X1024X1024? I reduced it to 1024, I hit a deadlock.

eric1hello

sgemm 与cublas求的结果不同

6

当K=64，128等的时候，结果不同。 例如： size_t M = 8000; size_t K = 64; size_t N = 8000; 

952480831

wplf

Performance doubles by only changing one line of code

1

Hi xiandong, Thanks for providing this amazing tutorial! Recently I am working on reduce0 and I found that I can double the performance of reduce_v0_baseline.cu kernel by simply changing a...

xiefan46