Bug in GEMV: result wrong dimension

[chrism0dwk]

I think I've discovered a critical bug in the GEMV implementation in gpuR version 2.0.0.

Example: multiply 3x2 matrix 'x' by a vector of length 2 'y':

> x = vclMatrix(matrix(1:6,ncol=2),type='float')
> y = vclVector(1:2, type='float')
> z = x%*%y
> z[]
[1]  9 12

The result 'z' is of length length(y), rather than nrow(x).

If I change the dimensions of the data structures such that 'x' is 3x2 and 'y' is length 3:

> x = vclMatrix(matrix(1:6,ncol=3),type='float')
> y = vclVector(1:3, type='float')
> z = x%*%y
> z[]
[1] 22 28 0

Again, the result is of length(y), not nrow(x).

Cheers,

Chris

System spec

R version 3.4.1
gpuR version 2.0.0 (CRAN)
CentOS 7.5
CUDA 9.0 with NVIDIA Quadro GP100 card.

[cdeterman]

Thank you for reporting. I completely forgot my unit tests for gemv and this is what I get. I will commit a bug fix shortly.

[chrism0dwk]

Many thanks for fixing this so quickly! A great help to my research :)

Chris

[cdeterman]

No problem, in this case it was an easy fix that you pointed out. Please let me know if you encounter any other issues. Glad the package is useful to you.