Re: [NEW PORT: audio/rubberband]

[Tobias Brodel]

ping

On 23 Apr. 2017 8:45 pm, "Tobias Brodel"  wrote:

> necroping,
>
> i've made a few attempts to get the inline assembly built on armv7 with
> clang,
> but it fails in the linking stage, unable to find the `main' function.
>
> so for now i'm just patching that part out, any comments?
>
> please find a tarball attached.
>
> t/
>
> On 01/26/17 18:59, Tobias Brodel wrote:
>
>> On 01/18/17 22:55, Tobias Brodel wrote:
>>
>>> hi stuart, thanks for your response.
>>>
>>> On 01/17/17 23:10, Stuart Henderson wrote:
>>>
 Hi, some quick feedback :

 Makefile:
 - "ONLY_FOR_ARCHS =  amd64 i386", why?

>>>
>>> this was due to a build i tried on armv7 which failed with:
>>>
>>> Error: selected processor does not support `fmrx r3,fpscr'
>>>
>>> i figured this was inline x86 assembly but your question promted
>>> metolook further. turns out its an issue with floating point
>>> instructionson arm.
>>>
>>> i tried copying other ports with `--target=generic' and
>>> `--arch=generic' in CONFIGURE_ARGS with no success. then i tried
>>> CFLAGS+='-mfloat-abi=hard' which the base gcc must not support?
>>> pulling in ports' gcc got me a bit further:
>>>
>>> {standard input}: Assembler messages:
>>> {standard input}:18: Error: selected processor does not support
>>> `fstmfdd sp!,{d8,d9}'
>>> {standard input}:19: Error: unknown pseudo-op: `.vsave'
>>> {standard input}:24: Error: selected processor does not support
>>> `fcpyd d9,d0'
>>> {standard input}:25: Error: selected processor does not support
>>> `fcpyd d8,d1'
>>> {standard input}:37: Error: selected processor does not support
>>> `fcpyd d0,d9'
>>> {standard input}:38: Error: selected processor does not support
>>> `fcpyd d1,d8'
>>> {standard input}:46: Error: selected processor does not support
>>> `fldmfdd ip!,{d8-d9}'
>>>
>>> unsure how to proceed, uncertain what OpenBSD support
>>> for hardware floating point is like on armv7.
>>>
>>>
>> updated tarball attached, simply removed the offending
>> lines of assembly, runs _far_ slower on armv7 but
>> produces expected results. some research suggests that
>> our older binutils in base could be the culprit.
>>
>> perhaps when clang/llvm get enabled on armv7 we can
>> lose this patch.
>>
>> tested on armv7, macppc and amd64
>>
>> - "#GPLv2 only" please add a space, "# GPLv2 only"
 - "COMMENT = Library for ..." lower-case first letter -> "library for"

 pkg/DESCR:
 - don't list WWW, it comes automatically from HOMEPAGE in Makefile

 pkg/PLIST, Makefile:
 - the linux-style shared library handling needs modifying.

>>>
>>> the attached tarball should have fixed these issues.
>>>
>>>
>> are things shaping up?
>>
>> cheers,
>> toby/
>>
>
>

Re: [NEW PORT: audio/rubberband]

[Tobias Brodel]

necroping,

i've made a few attempts to get the inline assembly built on armv7 with 
clang,

but it fails in the linking stage, unable to find the `main' function.

so for now i'm just patching that part out, any comments?

please find a tarball attached.

t/

On 01/26/17 18:59, Tobias Brodel wrote:

On 01/18/17 22:55, Tobias Brodel wrote:

hi stuart, thanks for your response.

On 01/17/17 23:10, Stuart Henderson wrote:

Hi, some quick feedback :

Makefile:
- "ONLY_FOR_ARCHS =  amd64 i386", why?


this was due to a build i tried on armv7 which failed with:

Error: selected processor does not support `fmrx r3,fpscr'

i figured this was inline x86 assembly but your question promted
metolook further. turns out its an issue with floating point
instructionson arm.

i tried copying other ports with `--target=generic' and
`--arch=generic' in CONFIGURE_ARGS with no success. then i tried
CFLAGS+='-mfloat-abi=hard' which the base gcc must not support?
pulling in ports' gcc got me a bit further:

{standard input}: Assembler messages:
{standard input}:18: Error: selected processor does not support 
`fstmfdd sp!,{d8,d9}'

{standard input}:19: Error: unknown pseudo-op: `.vsave'
{standard input}:24: Error: selected processor does not support 
`fcpyd d9,d0'
{standard input}:25: Error: selected processor does not support 
`fcpyd d8,d1'
{standard input}:37: Error: selected processor does not support 
`fcpyd d0,d9'
{standard input}:38: Error: selected processor does not support 
`fcpyd d1,d8'
{standard input}:46: Error: selected processor does not support 
`fldmfdd ip!,{d8-d9}'


unsure how to proceed, uncertain what OpenBSD support
for hardware floating point is like on armv7.



updated tarball attached, simply removed the offending
lines of assembly, runs _far_ slower on armv7 but
produces expected results. some research suggests that
our older binutils in base could be the culprit.

perhaps when clang/llvm get enabled on armv7 we can
lose this patch.

tested on armv7, macppc and amd64


- "#GPLv2 only" please add a space, "# GPLv2 only"
- "COMMENT = Library for ..." lower-case first letter -> "library for"

pkg/DESCR:
- don't list WWW, it comes automatically from HOMEPAGE in Makefile

pkg/PLIST, Makefile:
- the linux-style shared library handling needs modifying.


the attached tarball should have fixed these issues.



are things shaping up?

cheers,
toby/




rubberband.tar.gz
Description: application/gzip

Re: [NEW PORT: audio/rubberband]

[Tobias Brodel]

ping

On 2 Feb. 2017 8:46 am, "Tobias Brodel"  wrote:

> ping
>
> On 26 Jan. 2017 7:02 pm, "Tobias Brodel"  wrote:
>
>> whoops, wrong tarball. sorry for the noise!
>>
>> On 01/26/17 18:59, Tobias Brodel wrote:
>>
>>> On 01/18/17 22:55, Tobias Brodel wrote:
>>>
 hi stuart, thanks for your response.

 On 01/17/17 23:10, Stuart Henderson wrote:

> Hi, some quick feedback :
>
> Makefile:
> - "ONLY_FOR_ARCHS =  amd64 i386", why?
>

 this was due to a build i tried on armv7 which failed with:

 Error: selected processor does not support `fmrx r3,fpscr'

 i figured this was inline x86 assembly but your question promted
 metolook further. turns out its an issue with floating point
 instructionson arm.

 i tried copying other ports with `--target=generic' and
 `--arch=generic' in CONFIGURE_ARGS with no success. then i tried
 CFLAGS+='-mfloat-abi=hard' which the base gcc must not support?
 pulling in ports' gcc got me a bit further:

 {standard input}: Assembler messages:
 {standard input}:18: Error: selected processor does not support
 `fstmfdd sp!,{d8,d9}'
 {standard input}:19: Error: unknown pseudo-op: `.vsave'
 {standard input}:24: Error: selected processor does not support
 `fcpyd d9,d0'
 {standard input}:25: Error: selected processor does not support
 `fcpyd d8,d1'
 {standard input}:37: Error: selected processor does not support
 `fcpyd d0,d9'
 {standard input}:38: Error: selected processor does not support
 `fcpyd d1,d8'
 {standard input}:46: Error: selected processor does not support
 `fldmfdd ip!,{d8-d9}'

 unsure how to proceed, uncertain what OpenBSD support
 for hardware floating point is like on armv7.


>>> updated tarball attached, simply removed the offending
>>> lines of assembly, runs _far_ slower on armv7 but
>>> produces expected results. some research suggests that
>>> our older binutils in base could be the culprit.
>>>
>>> perhaps when clang/llvm get enabled on armv7 we can
>>> lose this patch.
>>>
>>> tested on armv7, macppc and amd64
>>>
>>> - "#GPLv2 only" please add a space, "# GPLv2 only"
> - "COMMENT = Library for ..." lower-case first letter -> "library for"
>
> pkg/DESCR:
> - don't list WWW, it comes automatically from HOMEPAGE in Makefile
>
> pkg/PLIST, Makefile:
> - the linux-style shared library handling needs modifying.
>

 the attached tarball should have fixed these issues.


>>> are things shaping up?
>>>
>>> cheers,
>>> toby/
>>>
>>
>>

Re: [NEW PORT: audio/rubberband]

[Tobias Brodel]

ping

On 26 Jan. 2017 7:02 pm, "Tobias Brodel"  wrote:

> whoops, wrong tarball. sorry for the noise!
>
> On 01/26/17 18:59, Tobias Brodel wrote:
>
>> On 01/18/17 22:55, Tobias Brodel wrote:
>>
>>> hi stuart, thanks for your response.
>>>
>>> On 01/17/17 23:10, Stuart Henderson wrote:
>>>
 Hi, some quick feedback :

 Makefile:
 - "ONLY_FOR_ARCHS =  amd64 i386", why?

>>>
>>> this was due to a build i tried on armv7 which failed with:
>>>
>>> Error: selected processor does not support `fmrx r3,fpscr'
>>>
>>> i figured this was inline x86 assembly but your question promted
>>> metolook further. turns out its an issue with floating point
>>> instructionson arm.
>>>
>>> i tried copying other ports with `--target=generic' and
>>> `--arch=generic' in CONFIGURE_ARGS with no success. then i tried
>>> CFLAGS+='-mfloat-abi=hard' which the base gcc must not support?
>>> pulling in ports' gcc got me a bit further:
>>>
>>> {standard input}: Assembler messages:
>>> {standard input}:18: Error: selected processor does not support
>>> `fstmfdd sp!,{d8,d9}'
>>> {standard input}:19: Error: unknown pseudo-op: `.vsave'
>>> {standard input}:24: Error: selected processor does not support
>>> `fcpyd d9,d0'
>>> {standard input}:25: Error: selected processor does not support
>>> `fcpyd d8,d1'
>>> {standard input}:37: Error: selected processor does not support
>>> `fcpyd d0,d9'
>>> {standard input}:38: Error: selected processor does not support
>>> `fcpyd d1,d8'
>>> {standard input}:46: Error: selected processor does not support
>>> `fldmfdd ip!,{d8-d9}'
>>>
>>> unsure how to proceed, uncertain what OpenBSD support
>>> for hardware floating point is like on armv7.
>>>
>>>
>> updated tarball attached, simply removed the offending
>> lines of assembly, runs _far_ slower on armv7 but
>> produces expected results. some research suggests that
>> our older binutils in base could be the culprit.
>>
>> perhaps when clang/llvm get enabled on armv7 we can
>> lose this patch.
>>
>> tested on armv7, macppc and amd64
>>
>> - "#GPLv2 only" please add a space, "# GPLv2 only"
 - "COMMENT = Library for ..." lower-case first letter -> "library for"

 pkg/DESCR:
 - don't list WWW, it comes automatically from HOMEPAGE in Makefile

 pkg/PLIST, Makefile:
 - the linux-style shared library handling needs modifying.

>>>
>>> the attached tarball should have fixed these issues.
>>>
>>>
>> are things shaping up?
>>
>> cheers,
>> toby/
>>
>
>

Re: [NEW PORT: audio/rubberband]

[Tobias Brodel]

whoops, wrong tarball. sorry for the noise!

On 01/26/17 18:59, Tobias Brodel wrote:

On 01/18/17 22:55, Tobias Brodel wrote:

hi stuart, thanks for your response.

On 01/17/17 23:10, Stuart Henderson wrote:

Hi, some quick feedback :

Makefile:
- "ONLY_FOR_ARCHS =  amd64 i386", why?


this was due to a build i tried on armv7 which failed with:

Error: selected processor does not support `fmrx r3,fpscr'

i figured this was inline x86 assembly but your question promted
metolook further. turns out its an issue with floating point
instructionson arm.

i tried copying other ports with `--target=generic' and
`--arch=generic' in CONFIGURE_ARGS with no success. then i tried
CFLAGS+='-mfloat-abi=hard' which the base gcc must not support?
pulling in ports' gcc got me a bit further:

{standard input}: Assembler messages:
{standard input}:18: Error: selected processor does not support 
`fstmfdd sp!,{d8,d9}'

{standard input}:19: Error: unknown pseudo-op: `.vsave'
{standard input}:24: Error: selected processor does not support 
`fcpyd d9,d0'
{standard input}:25: Error: selected processor does not support 
`fcpyd d8,d1'
{standard input}:37: Error: selected processor does not support 
`fcpyd d0,d9'
{standard input}:38: Error: selected processor does not support 
`fcpyd d1,d8'
{standard input}:46: Error: selected processor does not support 
`fldmfdd ip!,{d8-d9}'


unsure how to proceed, uncertain what OpenBSD support
for hardware floating point is like on armv7.



updated tarball attached, simply removed the offending
lines of assembly, runs _far_ slower on armv7 but
produces expected results. some research suggests that
our older binutils in base could be the culprit.

perhaps when clang/llvm get enabled on armv7 we can
lose this patch.

tested on armv7, macppc and amd64


- "#GPLv2 only" please add a space, "# GPLv2 only"
- "COMMENT = Library for ..." lower-case first letter -> "library for"

pkg/DESCR:
- don't list WWW, it comes automatically from HOMEPAGE in Makefile

pkg/PLIST, Makefile:
- the linux-style shared library handling needs modifying.


the attached tarball should have fixed these issues.



are things shaping up?

cheers,
toby/




rubberband.tar.gz
Description: application/gzip

Re: [NEW PORT: audio/rubberband]

[Tobias Brodel]

On 01/18/17 22:55, Tobias Brodel wrote:

hi stuart, thanks for your response.

On 01/17/17 23:10, Stuart Henderson wrote:

Hi, some quick feedback :

Makefile:
- "ONLY_FOR_ARCHS =  amd64 i386", why?


this was due to a build i tried on armv7 which failed with:

Error: selected processor does not support `fmrx r3,fpscr'

i figured this was inline x86 assembly but your question promted
metolook further. turns out its an issue with floating point
instructionson arm.

i tried copying other ports with `--target=generic' and
`--arch=generic' in CONFIGURE_ARGS with no success. then i tried
CFLAGS+='-mfloat-abi=hard' which the base gcc must not support?
pulling in ports' gcc got me a bit further:

{standard input}: Assembler messages:
{standard input}:18: Error: selected processor does not support 
`fstmfdd sp!,{d8,d9}'

{standard input}:19: Error: unknown pseudo-op: `.vsave'
{standard input}:24: Error: selected processor does not support 
`fcpyd d9,d0'
{standard input}:25: Error: selected processor does not support 
`fcpyd d8,d1'
{standard input}:37: Error: selected processor does not support 
`fcpyd d0,d9'
{standard input}:38: Error: selected processor does not support 
`fcpyd d1,d8'
{standard input}:46: Error: selected processor does not support 
`fldmfdd ip!,{d8-d9}'


unsure how to proceed, uncertain what OpenBSD support
for hardware floating point is like on armv7.



updated tarball attached, simply removed the offending
lines of assembly, runs _far_ slower on armv7 but
produces expected results. some research suggests that
our older binutils in base could be the culprit.

perhaps when clang/llvm get enabled on armv7 we can
lose this patch.

tested on armv7, macppc and amd64


- "#GPLv2 only" please add a space, "# GPLv2 only"
- "COMMENT = Library for ..." lower-case first letter -> "library for"

pkg/DESCR:
- don't list WWW, it comes automatically from HOMEPAGE in Makefile

pkg/PLIST, Makefile:
- the linux-style shared library handling needs modifying.


the attached tarball should have fixed these issues.



are things shaping up?

cheers,
toby/


rubberband.tar.gz
Description: application/gzip

Re: [NEW PORT: audio/rubberband]

[Tobias Brodel]

hi stuart, thanks for your response.

On 01/17/17 23:10, Stuart Henderson wrote:

Hi, some quick feedback :

Makefile:
- "ONLY_FOR_ARCHS =  amd64 i386", why?


this was due to a build i tried on armv7 which failed with:

Error: selected processor does not support `fmrx r3,fpscr'

i figured this was inline x86 assembly but your question promted
metolook further. turns out its an issue with floating point
instructionson arm.

i tried copying other ports with `--target=generic' and
`--arch=generic' in CONFIGURE_ARGS with no success. then i tried
CFLAGS+='-mfloat-abi=hard' which the base gcc must not support?
pulling in ports' gcc got me a bit further:

{standard input}: Assembler messages:
{standard input}:18: Error: selected processor does not support 
`fstmfdd sp!,{d8,d9}'

{standard input}:19: Error: unknown pseudo-op: `.vsave'
{standard input}:24: Error: selected processor does not support 
`fcpyd d9,d0'
{standard input}:25: Error: selected processor does not support 
`fcpyd d8,d1'
{standard input}:37: Error: selected processor does not support 
`fcpyd d0,d9'
{standard input}:38: Error: selected processor does not support 
`fcpyd d1,d8'
{standard input}:46: Error: selected processor does not support 
`fldmfdd ip!,{d8-d9}'


unsure how to proceed, uncertain what OpenBSD support
for hardware floating point is like on armv7.


- "#GPLv2 only" please add a space, "# GPLv2 only"
- "COMMENT = Library for ..." lower-case first letter -> "library for"

pkg/DESCR:
- don't list WWW, it comes automatically from HOMEPAGE in Makefile

pkg/PLIST, Makefile:
- the linux-style shared library handling needs modifying.


the attached tarball should have fixed these issues.



Since you are pretty new to ports on OpenBSD, please get the first
one in good shape before sending others, as most of my comments apply
to the others too.



sure thing, sorry for the noise, i now realise it was
poor etiquette.

thanks for having a look at this,
toby/


rubberband.tar.gz
Description: application/gzip

Re: [NEW PORT: audio/rubberband]

[Stuart Henderson]

On 2017/01/17 08:15, Tobias Brodel wrote:
> Please find attached a port for rubberband, a library and utility for
> timestretching and repitching audio. It is one of a few new dependencies
> required for an update to audio/ardour. I'm pretty new to ports so any
> feedback would be much appreciated. Tested on amd64. Cheers, toby/

Hi, some quick feedback :

Makefile:
- "ONLY_FOR_ARCHS =  amd64 i386", why?
- "#GPLv2 only" please add a space, "# GPLv2 only"
- "COMMENT = Library for ..." lower-case first letter -> "library for"

pkg/DESCR:
- don't list WWW, it comes automatically from HOMEPAGE in Makefile

pkg/PLIST, Makefile:
- the linux-style shared library handling needs modifying. you have:

lib/librubberband.so
lib/librubberband.so.2
@bin lib/librubberband.so.2.1.0

there should be a single entry,

@lib lib/librubberband.so.0.0

and the ".0.0" needs to be under control of SHARED_LIBS lines in
Makefile, so you'll need

SHARED_LIBS= rubberband 0.0

and arrange things so you can pass this in to the build via
LIBrubberband_VERSION (maybe via patching, maybe MAKE_FLAGS overrides
etc). If you grep other port Makefiles for "LIB[a-z]*_VERSION" you'll
find various ways that this is handled; the :R and :E constructs
you'll see in some of these are to split up the "major" and "minor"
versions for ports that need it.


On 2017/01/17 14:20, Tobias Brodel wrote:
> Lv2 is an audio plugin standard designed to replace LADSPA.
(etc)

Since you are pretty new to ports on OpenBSD, please get the first
one in good shape before sending others, as most of my comments apply
to the others too.