FreeBSD support #819
FreeBSD support #819
Conversation
|
@iamleot, @iarebatman, can you test on NetBSD and OpenBSD? intel-vaapi-driver is deprecated in favor of media-driver. This PR fixes both DragonFly and FreeBSD but if it also fixes NetBSD and OpenBSD it'd increase upstream buy in. ;) |
|
Hello Jan, Both intel/gmmlib#68 and #819 NetBSD parts looks good to me after quick skimming them. |
media_driver/media_driver_next/linux/common/os/mos_utilities_specific_next.cpp
Outdated
Show resolved
Hide resolved
|
Hello Jan,
Jan Beich writes:
@iamleot, did you forget to test media-driver after gmmlib built fine? gmmlib is ~15x smaller than media-driver, so I expect more fun here.
(Just a reminder. I'll remove this comment shortly after.)
No, I haven't and now that gmmlib seems ready I can finally test it.
I hope to test and share news this evening!
Thanks for the remainder and for working on that!
|
|
@iamleot wrote:
Do you still plan to? See also intel/libva#363 |
|
Hello Jan,
Jan Beich writes:
@iamleot wrote:
> I hope to test and share news this evening!
Do you still plan to? See also intel/libva#363
Sorry for the delay and (upstream) please consider my test not blocking
(I was pretty busy in the last weeks).
I'll hopefully have more time this weekend but I couldn't promise
anything.
|
jbeich
force-pushed
the
freebsd
branch
5 times, most recently
from
9488259
to
1792489
Compare
|
Porting 6cafa0b is harder than expected due to bundled libdrm. Intel didn't bother documenting exact revision the code was taken from and BSDs didn't bother upstreaming all of their patches. I've dropped my incomplete attempt to add missing pieces, leaving cmrtlib partially broken on BSDs. The regressing commit can be reverted downstream until a proper fix is found. Note, cmrtlib can be disabled via |
|
|
jbeich
force-pushed
the
freebsd
branch
3 times, most recently
from
5117b26
to
017c1db
Compare
|
casher was fixed, so
|
|
As much as I wish to take this PR in, there is an issue that we have unpatched dependency (gmmlib). This means that the best we can do (w/o having permanently patched gmmlib) would be to setup ci based on PR against gmmlib - that's what is actually done in this PR. This has a problem however that we risk to have ci failures on gmmlib updates. Someone will need to fix PR against gmmlib. And ci fill be down during the time gmmlib PR is getting fixed. To be honest, this is not something I wish to setup and track. I would suggest however to start picking up patches in this PR one by one reducing number of patches we have here. Some of them look to be ready to go, some might need further discussion. |
In file included from media_driver/agnostic/common/cm/cm_hal.cpp:26:
In file included from media_driver/agnostic/common/os/mos_os.h:30:
In file included from media_driver/agnostic/common/os/mos_defs.h:38:
In file included from /usr/include/c++/v1/stdio.h:101:
/usr/include/c++/v1/__config:219:12: fatal error: 'features.h' file not found
# include <features.h>
^~~~~~~~~~~~
In file included from media_driver/agnostic/common/cm/cm_kernel_rt.cpp:35:
In file included from media_driver/agnostic/common/cm/cm_mov_inst.h:31:
In file included from media_driver/agnostic/common/cm/cm_mem.h:128:
media_driver/linux/common/cm/hal/cm_mem_os.h:33:9: error: '__fastcall' macro redefined [-Werror,-Wmacro-redefined]
#define __fastcall
^
/usr/include/sys/cdefs.h:372:9: note: previous definition is here
#define __fastcall __attribute__((__fastcall__))
^
In file included from media_driver/linux/gen12/ddi/media_libva_caps_g12.cpp:38:
media_driver/linux/common/os/i915/include/mos_bufmgr_priv.h:351:9: error: 'ALIGN' macro redefined [-Werror,-Wmacro-redefined]
#define ALIGN(value, alignment) ((value + alignment - 1) & ~(alignment - 1))
^
/usr/include/machine/param.h:79:9: note: previous definition is here
#define ALIGN(p) _ALIGN(p)
^
In file included from media_driver/linux/gen12/ddi/media_libva_caps_g12.cpp:30:
In file included from media_driver/linux/common/ddi/media_libva.h:52:
In file included from media_driver/agnostic/common/codec/hal/codechal_decoder.h:40:
In file included from media_driver/agnostic/common/cm/cm_wrapper.h:31:
In file included from media_driver/linux/common/cm/hal/cm_wrapper_os.h:31:
In file included from media_driver/agnostic/common/cm/cm_def.h:31:
In file included from media_driver/agnostic/common/cm/cm_common.h:31:
In file included from media_driver/linux/common/cm/hal/cm_innerdef_os.h:42:
In file included from /usr/include/pthread_np.h:36:
In file included from /usr/include/sys/param.h:162:
/usr/include/machine/param.h:79:2: error: bootlegging test
ld: error: unable to find library -ldl
media_driver/media_driver_next/linux/common/os/mos_utilities_specific_next.cpp:1693:12: error: cast from pointer to smaller type 'uint32_t' (aka 'unsigned int') loses information
return (uint32_t)pthread_self();
^~~~~~~~~~~~~~~~~~~~~~~~
…laris
In file included from cmrtlib/linux/hardware/cm_device_os.cpp:24:
cmrtlib/linux/../linux/hardware/drm_device.h:42:10: fatal error: 'sys/sysmacros.h' file not found
#include <sys/sysmacros.h> //<sys/types.h>
^~~~~~~~~~~~~~~~~
media_driver/linux/common/os/i915/xf86drm.c:50:10: fatal error: 'sys/sysmacros.h' file not found
#include <sys/sysmacros.h>
^~~~~~~~~~~~~~~~~
media_driver/linux/ult/libdrm_mock/xf86drm_mock.c:50:10: fatal error: 'sys/sysmacros.h' file not found
#include <sys/sysmacros.h>
^~~~~~~~~~~~~~~~~
In file included from cmrtlib/agnostic/hardware/cm_queue.cpp:22:
In file included from cmrtlib/linux/../agnostic/hardware/cm_queue.h:26:
cmrtlib/linux/../linux/share/cm_def_os.h:37:10: fatal error: 'malloc.h' file not found
#include <malloc.h>
^~~~~~~~~~
In file included from cmrtlib/agnostic/share/cm_avs_state_msg_ex.cpp:22:
In file included from cmrtlib/linux/../agnostic/share/cm_def.h:26:
cmrtlib/linux/../linux/share/cm_def_os.h:104:10: error: use of undeclared identifier 'memalign'
return memalign(alignment, size);
^
media_driver/agnostic/common/os/mos_utilities.c:36:10: fatal error: 'malloc.h' file not found
#include <malloc.h> // For memalign
^~~~~~~~~~
media_driver/agnostic/common/os/mos_utilities.c:4047:11: error: use of undeclared identifier 'memalign'
ptr = _aligned_malloc(size, alignment);
^
media_driver/linux/common/os/mos_utilities_specific.h:91:43: note: expanded from macro '_aligned_malloc'
#define _aligned_malloc(size, alignment) memalign(alignment, size)
^
media_driver/agnostic/common/cm/cm_task_internal.cpp:896:39: error: use of undeclared identifier 'SYS_gettid'
m_taskProfilingInfo.threadID = CmGetCurThreadId(); // Get Thread ID
^
media_driver/linux/common/cm/hal/cm_innerdef_os.h:98:37: note: expanded from macro 'CmGetCurThreadId'
#define CmGetCurThreadId() syscall(SYS_gettid)
^
media_driver/agnostic/common/cm/cm_task_internal.cpp:997:39: error: use of undeclared identifier 'SYS_gettid'
m_taskProfilingInfo.threadID = CmGetCurThreadId(); // Get Thread ID
^
media_driver/linux/common/cm/hal/cm_innerdef_os.h:98:37: note: expanded from macro 'CmGetCurThreadId'
#define CmGetCurThreadId() syscall(SYS_gettid)
^
media_driver/linux/common/codec/ddi/media_libva_decoder.cpp:48:10: fatal error: 'linux/fb.h' file not found
#include <linux/fb.h>
^~~~~~~~~~~~
media_driver/linux/common/codec/ddi/media_libva_decoder.cpp:55:43: error: variable has incomplete type 'struct fb_var_screeninfo'
struct fb_var_screeninfo vsinfo;
^
media_driver/linux/common/codec/ddi/media_libva_decoder.cpp:62:22: error: use of undeclared identifier 'FBIOGET_VSCREENINFO'
if(ioctl(fd, FBIOGET_VSCREENINFO, &vsinfo) < 0)
^
media_driver/linux/common/ddi/media_libva.cpp:39:10: fatal error: 'linux/fb.h' file not found
#include <linux/fb.h>
^~~~~~~~~~~~
This reverts commit 53405da.
ld: error: undefined symbol: backtrace >>> referenced by mos_utilities_specific.cpp:2504 (media_softlet/linux/common/os/osservice/mos_utilities_specific.cpp:2504) >>> media_driver/CMakeFiles/iHD_drv_video_COMMON.dir/__/media_softlet/linux/common/os/osservice/mos_utilities_specific.cpp.o:(MosUtilities::MosTraceEvent(unsigned short, unsigned char, void const*, unsigned int, void const*, unsigned int)) c++: error: linker command failed with exit code 1 (use -v to see invocation)
|
Obsoleted by #1785. Already used downstream. Personally, I gave up and use Vulkan Video (via anv from Mesa) or legacy intel-vaapi-driver (for hybrid VP9 decode) on my aging Skylake iGPU. |
Depends on intel/libva#363 and intel/gmmlib#68
Build logs
http://beefy9.nyi.freebsd.org/data/latest-per-pkg/libva-intel-media-driver/21.1.1/114amd64-default.log
http://beefy10.nyi.freebsd.org/data/latest-per-pkg/libva-intel-media-driver/21.1.1/114i386-default.log
http://beefy6.nyi.freebsd.org/data/latest-per-pkg/libva-intel-media-driver/21.1.1/122amd64-default.log
http://beefy5.nyi.freebsd.org/data/latest-per-pkg/libva-intel-media-driver/21.1.1/122i386-default.log
http://beefy18.nyi.freebsd.org/data/latest-per-pkg/libva-intel-media-driver/21.1.1/main-amd64-default.log
http://beefy17.nyi.freebsd.org/data/latest-per-pkg/libva-intel-media-driver/21.1.1/main-i386-default.log
https://sting.dragonflybsd.org/dports/logs/multimedia___libva-intel-media-driver.log
Tests output
Note, to avoid regressing build in future consider FreeBSD-friendly CI.