cmake vs. scons: slow analysis

[Daniel Schürmann]

While testing https://bugs.launchpad.net/mixxx/+bug/1879588
i have found a regression when switching to cmake.
The beat analysis performs significant slower with cmake
Scons 4 s
CMake 6 s

44.2 kHz and 48 kHz Tracks are effected in the same way.

[Daniel Schürmann (daschuer)]

Tested was the commit: b7bf705804d19375dc764468e1997fe25e647555
which is still suffering the regression fixed in
https://github.com/mixxxdj/mixxx/pull/2683

[Daniel Schürmann (daschuer)]

With current master 2.4 alpha 436f30d82f616d8d04912d1bb827532c777d3ec7, the thing is the same for
44.2k, but worse for 48k

44.2k Track
Scons 4 s
CMake 6 s

48k Track
Scons 25 s
CMake 133 s

The difference between 48k and 44.2 k is the topic of
https://bugs.launchpad.net/mixxx/+bug/1879588

[Daniel Schürmann (daschuer)]

The issue is that CMake explicit defines
-msse -mfpmath=sse
This is a de-optimizing for 64 bit builds where SSE2 is always available.

We build by default with OPTIMIZE "portable" which requires at least SSE on a 32 bit CPU.

But even my 2008 Intel Atom N270 features 32 bit only but has SSE2 registers.
I looks like one of the last 32 bit Processors release without was the Celeron (P6), produce until 2002

So I think we can safely advance and require SSE2 for "portable" 32 bit builds and let the others use legacy or native if there are any.

[Daniel Schürmann (daschuer)]

OK, Scons already requires SSE2 for portable builds. It is just a typo in CMakeList.txt

[Daniel Schürmann (daschuer)]

The typo is fixed here: https://github.com/mixxxdj/mixxx/pull/2808
But the performance difference not ... Ideas?

[Daniel Schürmann (daschuer)]

Here is the final fix:
https://github.com/mixxxdj/mixxx/pull/2812

[Swiftb0y (swiftb0y)]

Mixxx now uses GitHub for bug tracking. This bug has been migrated to:
https://github.com/mixxxdj/mixxx/issues/9986