Build fails on big-endian PowerPC due to missing AltiVec

[glaubitz]

On Debian's big-endian PowerPC ports (powerpc and ppc64), openal-soft fails due to AltiVec not being available:

[  7%] Building CXX object CMakeFiles/alsoft.common.dir/common/ringbuffer.cpp.o
[  8%] Building CXX object CMakeFiles/alsoft.common.dir/common/strutils.cpp.o
/usr/bin/c++  -I/<<PKGBUILDDIR>>/include -I/<<PKGBUILDDIR>>/build-tree -I/<<PKGBUILDDIR>>/common -g -O2 -ffile-prefix-map=/<<PKGBUILDDIR>>=. -fstack-protector-strong -Wformat -Werror=format-security -Wdate-time -D_FORTIFY_SOURCE=2 -Wdate-time -D_FORTIFY_SOURCE=2 -std=gnu++17 -fPIC -fvisibility=hidden -Winline -Wunused -Wall -Wextra -Wshadow -Wconversion -Wcast-align -Wpedantic -Wold-style-cast -Wnon-virtual-dtor -Woverloaded-virtual -Wno-c++20-attribute-extensions -Wno-interference-size -Werror=undef -fno-math-errno -pthread -MD -MT CMakeFiles/alsoft.common.dir/common/ringbuffer.cpp.o -MF CMakeFiles/alsoft.common.dir/common/ringbuffer.cpp.o.d -o CMakeFiles/alsoft.common.dir/common/ringbuffer.cpp.o -c /<<PKGBUILDDIR>>/common/ringbuffer.cpp
/usr/bin/c++  -I/<<PKGBUILDDIR>>/include -I/<<PKGBUILDDIR>>/build-tree -I/<<PKGBUILDDIR>>/common -g -O2 -ffile-prefix-map=/<<PKGBUILDDIR>>=. -fstack-protector-strong -Wformat -Werror=format-security -Wdate-time -D_FORTIFY_SOURCE=2 -Wdate-time -D_FORTIFY_SOURCE=2 -std=gnu++17 -fPIC -fvisibility=hidden -Winline -Wunused -Wall -Wextra -Wshadow -Wconversion -Wcast-align -Wpedantic -Wold-style-cast -Wnon-virtual-dtor -Woverloaded-virtual -Wno-c++20-attribute-extensions -Wno-interference-size -Werror=undef -fno-math-errno -pthread -MD -MT CMakeFiles/alsoft.common.dir/common/strutils.cpp.o -MF CMakeFiles/alsoft.common.dir/common/strutils.cpp.o.d -o CMakeFiles/alsoft.common.dir/common/strutils.cpp.o -c /<<PKGBUILDDIR>>/common/strutils.cpp
In file included from /<<PKGBUILDDIR>>/common/pffft.cpp:102:
/usr/lib/gcc/powerpc64-linux-gnu/14/include/altivec.h:34:2: error: #error Use the "-maltivec" flag to enable PowerPC AltiVec support
   34 | #error Use the "-maltivec" flag to enable PowerPC AltiVec support
      |  ^~~~~
/<<PKGBUILDDIR>>/common/pffft.cpp:103:14: error: ‘__vector’ does not name a type; did you mean ‘vector’?
  103 | using v4sf = vector float;
      |              ^~~~~~

AltiVec should be available on these targets in general, although I'm not sure what we're currently using as the baseline on Debian for these two PowerPC targets.

Full log at: https://buildd.debian.org/status/fetch.php?pkg=openal-soft&arch=ppc64&ver=1%3A1.24.0-2&stamp=1732122545&raw=0

[glaubitz]

Ah, the change d6520c4 enables AltiVec unconditionally on PowerPC which it shouldn't do as not all PowerPC targets actually support it or the baseline may have disabled it.

I need to figure out what the currently baselines for these Debian targets are at the moment.

[kcat]

Looks like the __VEC__ and/or __ALTIVEC__ macros should be defined when AltiVec is enabled. Should be easy enough to add a check for them to avoid it when not available (done in commit 8474293). It would still be good to check up on what baseline you can use, enabling AltiVec/SIMD operations would help performance.

[glaubitz]

Thanks! The big-endian baseline was changed to a more generic one in 2016 and it's probably a good idea to raise it again to include AltiVec for both 32-bit and 64-bit PowerPC which we will discuss in Debian as I fully agree with your stance.

In the meantime, it would be great if this could be fixed so that the package will build again for the time being.

[glaubitz]

It should just be enough to check for __ALTIVEC__:

glaubitz@perotto:~$ echo | gcc -maltivec -E -dM - |grep -i altivec
#define __ALTIVEC__ 1
#define __pixel __attribute__((altivec(pixel__))) unsigned short
#define __vector __attribute__((altivec(vector__)))
#define __bool __attribute__((altivec(bool__))) unsigned
#define __APPLE_ALTIVEC__ 1
glaubitz@perotto:~$ echo | gcc -mno-altivec -E -dM - |grep -i altivec
#define __pixel __attribute__((altivec(pixel__))) unsigned short
#define __vector __attribute__((altivec(vector__)))
#define __bool __attribute__((altivec(bool__))) unsigned
glaubitz@perotto:~$

[smcv]

Was this fixed by 8474293?

[glaubitz]

Yes, this should fix it.