langou

For a software like OpenBLAS, during the execution of a program that calls twice GEMM (say) with twice the exact same parameters (but at different times in the execution) then...

> Repeat calls to GEMM with the exact same parameters should produce the exact same computation and result, This answers my question. > "similar" parameters might result in a different...

> While it is not directly related to this issue, I would like to highlight that deviations between job configurations are already accepted for some routines in LAPACK. GESVD, for...

@thijssteel wrote "A possible solution would be to split the gemm call into two pieces, one of which only gets called if wantt is true. Then the gemm call that...

To summarize where we are at: @thijssteel established that, even with a BLAS that returns bit-to-bit-similar answers for similar calls, GEEV from LAPACK might not return bit-to-bit-similar eigenvalues when called...

It's great. Thanks. I see that there is still one issue somewhere. But we had the same issue this Summer and we were using the CI to tell us whether...

OK. Cool. I will try this trick on my laptop too. Thanks. That was a pain. I had not thought about changing the clangformat version on my laptop. Good idea....

Hi @TiborGY, (*) You are correct about the behavior of GEMM in the presence of alpha = 0 or beta = 0 and in the presence of NaNs/Infs in the...

Argh. Thanks for the report. Shoo. That's going to be a tough one to debug. We'll have a look. @jprhyne

> On a related note, are there guidelines on how to direct the make file to use the macOS Accelerate for BLAS instead of the inbuilt librefblas? In make.inc, https://github.com/Reference-LAPACK/lapack/blob/9128e2051a922c3889e3c1fccaf4ce0d230b2785/make.inc.example#L77...