OpenBLAS

OpenBLAS copied to clipboard

DYNAMIC_ARCH support on Android

12

I've been experimenting with using DYNAMIC_ARCH support on Android (aarch64) and had some questions about this. I am using 0.3.10 release (commit 63b03efc2af332c88b86d4fd8079d00f4b439adf). On couple of devices I've been using...

ognjentodic

[Documentation] Help understanding the build system - adding to wiki

9

I'm trying to get a clear idea of how the build system works and thankfully due to using Makefiles, it is a lot clearer than autotools and other cancerous build...

epsilon-0

Precision issues on POWER related to SWITCH_RATIO

16

I'm debugging a precision issue in PyTorch which I traced back to parallelization done in OpenBLAS. I.e. forcing OpenBLAS to use the serial algorithm (either by hacking the source or...

Flamefire

"BLAS: Bad memory unallocation!" errors on Power

6

We recently encountered some failures in some of our tests on Power, which a colleague of mine traced to an apparent issue in OpenBLAS. For reference, consider the attached test...

cparrott73

cblas_sgemv switches algorithm way too early

10

While benchmarking cblas_sgemv against other implementations, I found it apparently switches between algorithms at around a size of 96. That results in a massive drop in performance from being almost...

EricGrange

Performance issue with many cores

41

Running the following code on a machine with many cores will give worse performances than limiting the number of threads. ```python import numpy as np X = np.random.random_sample((2048, 2048)) %time...

jeremiedbb

DTBSM, triangular band matrix solve with multiple right-hand sides

2

Hi OpenBLAS community, I received an email from a BLAS/LAPACK user and they were wondering why there exists Level 2 DTBSV, and why there does not exist a Level 3...

langou

Hi All, I would like to request to know if there are plans for implementing binary gemm. Binary gemm is used in machine learning applications. I bring this to your...

rayjs

dsysv slower than dgesv?

2

I was trying to benchmark netlib Lapack and OpenBlas for `dgesv` and `dsysv` routines. I found that dsysv in OpenBlas was slower than dgesv although common sense tells me the...

theinlinaung2010

LU and eigen routines slower than MKL

9

Hi, I'm running some performance comparisons between OpenBLAS and MKL for LU and eigen routines. I see that OpenBLAS tests with, for example, `dgetrf` and `dsyevd`, are about 3 times...

youyu3