Marcus Müller <[email protected]> writes:
> place. I'll make a PR to that end; if you could test that, Greg, it'd
> be much appreciated.
Sure, please feel free to @gdt (github and codeberg) or email me a link
and I'll try it.
Johannes: I realize that having bougght into github CI that it's not a
great portability situation as the set of available runners is very
limited. I am struggling with this myself on an unrelated project, but
so far more inclined to put energy into leaving github and some
beyond-github future CI more than improving CI on github, even if it's
not so much either way :-(
Someone else updated volk to 3.3.0 and here's the patch they arrived at.
I'm not a fan of doing it this way (upstream), but it does more or less
show the issue well.
--- include/volk/volk_common.h.orig 2026-02-09 12:27:43.544508041 +0000
+++ include/volk/volk_common.h
@@ -150,7 +150,12 @@ union bit256 {
////////////////////////////////////////////////////////////////////////
// log2f
////////////////////////////////////////////////////////////////////////
+#ifdef __cplusplus
+#include <cmath>
+using std::isinf, std::isnan;
+#else
#include <math.h>
+#endif
// +-Inf -> +-127.0f in order to match the behaviour of the SIMD kernels
// NaN -> NaN (preserved for consistency)
static inline float log2f_non_ieee(float f)
