nschaeff comments

Results 15 comments of


                                            nschaeff

{Sym/Gen}GEigs{Real/Complex}Shift

I'm particularly interested in the last one: GenGEigsComplexShift

feature request: full diff between two commits

Here is a patch that provides full diff between two commits for git. It is used like : diffuse -c commit1..commit2 # shows the full diff between git commit1 and...

feature request: full diff between two commits

Same thing for mercurial (a bit more tricky, hopefully it does not break anything!) ```diff --- a/src/diffuse/vcs/hg.py +++ b/src/diffuse/vcs/hg.py @@ -66,6 +66,12 @@ class Hg(VcsInterface): args.append(utils.safeRelativePath(self.root, name, prefs, 'hg_cygwin')) #...

feature request: full diff between two commits

There is already one: https://github.com/MightyCreak/diffuse/pull/212

Full diff

Any chance to have this merge in the near future? What is missing?

Allow complex transforms to use non-interleaved memory layouts (separate I and Q buffers)

Hello, I'm also interested in this feature.

low performance on MI250X in certain cases

Thanks a lot! Without changing the batch size, I can increase the stride just a little (+8 works well) and VkFFT delivers the transform in 1.1ms. Rule of thumb that...

use fma for a faster and more accurate sqrt

It would also fix issue #166 for sqrt for free.

use fma for a faster and more accurate sqrt

For cbrt, you can't just use an fma to compute the error with respect to cbrt(hi), so it is a bit more involved. But once we have the error, the...

use fma for a faster and more accurate sqrt

Breaking news: I just learned from one of the authors, that the algorithm I propose for sqrt has been published and analyzed by serious people. It is algorithm 8 from...