Bug 538710

Summary:	media-libs/spandsp-0.0.6_pre21: fails hard when a single 'lower' instruction set is missing
Product:	Gentoo Linux	Reporter:	Michał Górny <mgorny>
Component:	Current packages	Assignee:	No maintainer - Look at https://wiki.gentoo.org/wiki/Project:Proxy_Maintainers if you want to take care of it <maintainer-needed>
Status:	CONFIRMED ---
Severity:	normal	CC:	sam
Priority:	Normal
Version:	unspecified
Hardware:	All
OS:	Linux
Whiteboard:
Package list:		Runtime testing required:	---

Description Michał Górny archtester

2015-02-03 22:01:53 UTC

Original report: https://bugs.funtoo.org/browse/FL-2069

So upstream is insane here and pretty much:

1. has --enable for every instruction set out there, even if they don't use it. Even if it doesn't really exist, like sse5.

2. Every 'higher' instruction set enables lower one. To the point that sse4a implies sse4_2, and sse3 implies ssse3.

3. Every enabled impl puts -mXXX, unused automake conditional and semi-unused C conditional.

4. There's one src/mmx_sse_decs.h which includes all gcc intrinsic headers for enabled instruction sets.

In particular, even if no code makes use of SSSE3, enabling SSE3 automatically implies including the header for it. With user CFLAGS disabling SSSE3, this quickly results in:

In file included from mmx_sse_decs.h:42:0,
                 from complex_vector_float.c:45:
/usr/lib/gcc/x86_64-pc-linux-gnu/4.8.4/include/tmmintrin.h:31:3: error: #error "SSSE3 instruction set not enabled"

Comment 1 Michał Górny archtester

2015-02-03 22:14:56 UTC

(In reply to Michał Górny from comment #0)
> Original report: https://bugs.funtoo.org/browse/FL-2069
> 
> So upstream is insane here and pretty much:
> 
> 1. has --enable for every instruction set out there, even if they don't use
> it. Even if it doesn't really exist, like sse5.
> 
> 2. Every 'higher' instruction set enables lower one. To the point that sse4a
> implies sse4_2, and sse3 implies ssse3.

Was a bit unclear on this one. It only enables the define for the lower one, not the enable bit itself -- so no matching automake conditional and no -mXXX.

Comment 2 Michał Górny archtester

2015-02-03 23:14:42 UTC

I've restored the flags to the previous state to fix the intermediate issues. However, this is something worth fixing upstream. Not every SSE4* CPU will have SSSE3.

+  03 Feb 2015; Michał Górny <mgorny@gentoo.org> spandsp-0.0.6.ebuild:
+  Remove unused CPU flags that only implicitly enable other flags and cause the
+  build to fail. Add proper REQUIRED_USE for implicitly enabled instruction
+  sets. Bug #538710, fixes https://bugs.funtoo.org/browse/FL-2069.

Comment 3 Michał Górny archtester

2015-02-04 21:57:22 UTC

I've submitted this upstream.

[1]:http://lists.soft-switch.org/pipermail/spandsp/2015-February/000200.html