OpenBLAS

OpenBLAS copied to clipboard

martin-frbg

dpotrf + dpotri: Windows vs Linux

6

I've come across what looks like an anomalous difference in performance inverting a positive definite matrix using dpotrf() and dpotri(), on Windows as compared with Linux. This is on a...

AllinCottrell

cmake: set `CMP0042` to `NEW`

3

It's better to enable https://cmake.org/cmake/help/latest/policy/CMP0042.html#policy:CMP0042 than to explicitly set CMAKE_MACOSX_RPATH, since users may want to set the latter from the command line.

haampie

Make ICOPY_OPERATION threaded?

1

In some multi-threaded versions of OpenBLAS, the ICOPY_OPERATION (packing for A) seems to take 6-13X more time than the actual KERNEL_OPERATION(s). Is it possible to make this operation multi-threaded? ```...

ChipKerchner

Previously baseline AArch64 was left without `SWITCH_RATIO` or `GEMM_PREFERED_SIZE` and older default values, but it can be seen across other cores that these values seem to work for many devices....

Mousius

Remove unconditional `NO_AVX512=1` for flang builds

12

I want to figure out what's happening with the errors with flang when using the default build options on a platform that supports AVX512. This problem was already observed in...

h-vetinari

Fix AIX build

1

AIX build with OpenXL broke after PR #4944 with the below errors. In file included from ../kernel/power/sbgemv_t_vsx.c:31: In file included from ../kernel/power/sbgemv_common.c:30: ../kernel/power/gemm_common.c:31:18: error: type specifier missing, defaults to 'int';...

ayappanec

Performance regression in version 0.3.28 on Graviton4

7

Recently I reported a performance regression at https://github.com/OpenMathLib/OpenBLAS/commit/a4e56e045266b6049e336ff27db184fc42ba7ff1 There is some glitch in GitHub which prevents me from posting a follow up so I am opening this ticket. In my...

dnoan

Reordering matrix rows leads to output differences

2

Hello, while using OpenBLAS as a backend for PyTorch code, we found minor output differences, when reordering our input samples in a batch. In our code, we used a linear...

j-moeller

Very poor `gemmt` performance compared to `gemm` and `syrk`

3

I'm running some timings on operation `t(X)*X` on row-major matrices having many more rows than columns. I'm finding that for these types of inputs, function `gemmt` is much slower than...

david-cortes