-
Hello. I have tried many times, but I can not install such lib.
ubuntu 1604
gcc/g++ 5.4.0
python 3.6
It says that:
--------------------------------------------------
$ pip install --upgrade…
-
### **** Motivation
With new sparse operators being added to TVM on a weekly basis, we would like to propose a change that would address the lack of expressiveness in sparse tensors of the current re…
yy665 updated
4 years ago
-
I have been doing some profiling comparing cupy's sparse matrix `dot()` with PyTorch's `mm()` and I'm getting some very surprising results. I noticed that while cupy is using cusparse, PyTorch appears…
-
ACL_GEMM (1024 x 1024) execution time : 10ms
CSR_FLEX_GEMM (1024 x 1024) execution time : 60ms (best performance)
The difference between these two performances seems to be the tiling optimization …
-
Hello,
I tried to install merge-spmm based on the instructions provided in README.
But, I failed to install modern GPU as follows.
I am not sure, but password seems to be required to install …
-
SpMV, SpMM kernel code is fixed with the best performance loop unrolling factor.
However, in the deep learning model, fixed values can not be used because the matrix size varies.
So i have to modif…
-
Matrix size : 1024 by 1024
There are no difference execution time with different sparsity( number of non-zeros).
The expected results should increase the kernel resolution time as the sparsity inc…
-
In the ACL, the input matrix appears in two forms: NCHW or NHWC.
The algorithm we are currently using transposes the input matrix to the column major and proceeds to SpMM.
If you try to use this alg…
-
Hello,
To reproduce this error I'm using the following Dockerfile:
```
FROM nvidia/cuda:9.0-cudnn7-devel-ubuntu16.04
RUN apt-get update && apt-get install -y \
build-essential \
cmak…
-
Hi, rusty1s!
Thx for your excellent works! I've noticed that there are some updates of this lib attempting to build more built-in interfaces like those provided in pytorch. May I have your future…