| Commit message (Collapse) | Author | Age | Files | Lines |
... | |
|
|
|
| |
This is the original check (isnan) before the change in table and so the behavior for this input should be the same.
|
|
|
|
|
|
|
|
| |
zero out into its own check.
This change mostly affects ARM processors. They seem to not do well with nest branches whereas x86 do fine. Breaking out this nested if makes no computational difference on x86 but makes a ~5-6% improvement for the ARMv7 tests.
Also updates the docs to reflect the new mean error from the function.
|
|
|
|
|
|
| |
atan2 for some NaN/Inf combinations the performance is better now.
This change should not have any impact in SDR related things. It is only for checking NaN/Inf values.
|
| |
|
| |
|
|
|
|
|
|
|
|
| |
some errors in the code.
Error before correction: <0.111 degrees.
Error after correction: <8.20E-5 degrees
Some QA tests failed after correcting fast_atan2f function. The tests were also fixed.
|
|
|
|
|
| |
* Fixed usage of Windows thread-naming API, changed minimum Windows version
* Fixed MSVC usage of isnan, round
|
|
|
|
| |
We should be more careful about letting these into the code in the future. In emacs, we can use (add-hook 'before-save-hook 'delete-trailing-whitespace).
|
| |
|
|
|
|
| |
When calculating with nan/inf values for x and/or y, make sure they all return and don't segfault. The results of gr_fast_atan2f in these cases may be different than atan2. But since we are concerned about speed with this block over accuracy (especially in these instances), we just want to make sure it doesn't crash.
|
|
|
|
| |
and y=inf.
|
|
|
|
| |
runtime: adds math functions to swig to make available as gr.function.
|
| |
|
| |
|
|
|