Re: [Discuss-gnuradio] [VOLK] volk_profile failures

[Tom Rondeau]

On Sat, Jun 6, 2015 at 4:51 PM, Nowlan, Sean <address@hidden> wrote:

That makes sense. I installed ORC 0.4.23 and tried to recompile GR, but the build is failing on Volk. GCC segfaults a couple times before failing with an error on volk_16ic_magnitude_16i_a_orc_impl. Full output here: 

https://gist.github.com/nowls/2f6ddf5734e152ca21a7

ORC 0.4.23 was built with:

CFLAGS="-mcpu=cortex-a15.cortex-a7 -mtune=cortex-a15.cortex-a7 -mfpu=neon-vfpv4 -mfloat-abi=hard -funsafe-math-optimizations"

CPPFLAGS="-mcpu=cortex-a15.cortex-a7 -mtune=cortex-a15.cortex-a7 -mfpu=neon-vfpv4 -mfloat-abi=hard -funsafe-math-optimizations"

UHD built fine w/ ORC 0.4.23.

Sean

Just disable ORC Altogether in VOLK. Most of the kernels have replacement NEON proto-kernels now, anyways. Pass -DENABLE_ORC=False to cmake.

Tom

 

---

From: West, Nathan <address@hidden>
Sent: Saturday, June 6, 2015 12:57 PM

To: Nowlan, Sean
Cc: address@hidden
Subject: Re: [Discuss-gnuradio] [VOLK] volk_profile failures

 

Most likely.

What we've seen before is that ORC didn't preserve registers according to ARM specifications. This basically results in unpredictable data corruption. The commit I pointed to is the fix (vpush certain registers if used, then vpop after the routine).

The visible corruption is seen by corrupting the test tolerance. If you look at ctest -V you should see that the "error"s are actually the same result, but since tolerance gets corrupted the test erroneously is false. The only QA failures should be on kernels that have an ORC protokernel, but this bug can affect non-ORC protokernels for that kernel.

Nathan

On Sat, Jun 6, 2015 at 12:47 PM, Nowlan, Sean <address@hidden> wrote:

​Ok, I will install ORC 0.4.23 from source. Is ORC involved with all the other architectures that had failures?

Sean

---

From: West, Nathan <address@hidden>
Sent: Friday, June 5, 2015 6:48 PM
To: Nowlan, Sean
Cc: address@hidden
Subject: Re: [Discuss-gnuradio] [VOLK] volk_profile failures

 

Hi Sean,

I didn't read all of the debug info, but your symptoms are a result of ORC doing bad things. 

You can either uninstall ORC or upgrade to the latest release (0.4.23). The commit you really need is this one: http://cgit.freedesktop.org/gstreamer/orc/commit/?id=acdaac31c648fd10f1bd0a49c4c33b483bb9c35c

If you choose to upgrade ORC let me know how it goes. 

Cheers,

Nathan

On Friday, June 5, 2015, Nowlan, Sean <address@hidden> wrote:

I compiled GNU Radio 3.7.7.1 natively on an ODROID XU3-Lite with Ubuntu 14.04.2 LTS. I ran volk_profile and got a lot of failures for various embedded architectures. Any ideas what could be happening? Could it be something with the compile flags I chose?

volk_16ic_s32f_deinterleave_32f_x2: fail on arch neon

volk_16ic_s32f_deinterleave_32f_x2: fail on arch neon

volk_16ic_s32f_deinterleave_32f_x2: fail on arch u_orc

volk_16ic_s32f_deinterleave_32f_x2: fail on arch u_orc

volk_32fc_s32f_magnitude_16i: fail on arch u_orc

volk_32fc_x2_multiply_32fc: fail on arch a_generic

volk_32fc_x2_multiply_32fc: fail on arch a_generic

volk_32fc_x2_multiply_32fc: fail on arch a_generic

volk_32fc_x2_multiply_32fc: fail on arch neon

volk_32fc_x2_multiply_32fc: fail on arch neon

volk_32fc_x2_multiply_32fc: fail on arch neon

volk_32fc_x2_multiply_32fc: fail on arch neon_opttests

volk_32fc_x2_multiply_32fc: fail on arch neon_opttests

volk_32fc_x2_multiply_32fc: fail on arch neon_opttests

volk_32fc_x2_multiply_32fc: fail on arch neonasm

volk_32fc_x2_multiply_32fc: fail on arch neonasm

volk_32fc_x2_multiply_32fc: fail on arch neonasm

volk_32fc_x2_multiply_32fc: fail on arch u_orc

volk_32fc_x2_multiply_32fc: fail on arch u_orc

volk_32fc_x2_multiply_32fc: fail on arch u_orc

volk_32f_sqrt_32f: fail on arch neon

volk_32f_x3_sum_of_poly_32f: fail on arch a_neon

volk_32f_x3_sum_of_poly_32f: fail on arch neonvert

volk_32fc_s32fc_multiply_32fc: fail on arch neon

The full output of volk_profile is here: https://gist.github.com/nowls/37c78d89342053bae0ea

Resulting volk_config: https://gist.github.com/nowls/5894fbce1b29325b8f58

Some relevant configuration info taken from CMakeCache.txt (gcc/g++ version and C/CXX/ASM flags):

CMAKE_CXX_COMPILER:FILEPATH=/usr/bin/g++-4.9

CMAKE_CXX_COMPILER_WITH_PATH:FILEPATH=/usr/bin/g++-4.9

CMAKE_CXX_FLAGS:STRING=-mcpu=cortex-a15.cortex-a7 -mtune=cortex-a15.cortex-a7 -mfpu=neon-vfpv4 -mfloat-abi=hard -funsafe-math-optimizations

CMAKE_CXX_FLAGS_DEBUG:STRING=-g

CMAKE_CXX_FLAGS_DEBUGPARANOID:STRING=-Wall -Wextra -g -O0

CMAKE_CXX_FLAGS_MINSIZEREL:STRING=-Os -DNDEBUG

CMAKE_CXX_FLAGS_NOOPTWITHASM:STRING=-Wall -save-temps -g -O0

CMAKE_CXX_FLAGS_O2WITHASM:STRING=-Wall -save-temps -g -O2

CMAKE_CXX_FLAGS_O3WITHASM:STRING=-Wall -save-temps -g -O3

CMAKE_CXX_FLAGS_RELEASE:STRING=-O3 -DNDEBUG

CMAKE_CXX_FLAGS_RELWITHDEBINFO:STRING=-O2 -g -DNDEBUG

CMAKE_C_COMPILER:FILEPATH=/usr/bin/gcc-4.9

CMAKE_C_COMPILER_WITH_PATH:FILEPATH=/usr/bin/gcc-4.9

CMAKE_C_FLAGS:STRING=-mcpu=cortex-a15.cortex-a7 -mtune=cortex-a15.cortex-a7 -mfpu=neon-vfpv4 -mfloat-abi=hard -funsafe-math-optimizations

CMAKE_C_FLAGS_DEBUG:STRING=-g

CMAKE_C_FLAGS_DEBUGPARANOID:STRING=-Wall -Wextra -g -O0

CMAKE_C_FLAGS_MINSIZEREL:STRING=-Os -DNDEBUG

CMAKE_C_FLAGS_NOOPTWITHASM:STRING=-Wall -save-temps -g -O0

CMAKE_C_FLAGS_O2WITHASM:STRING=-Wall -save-temps -g -O2

CMAKE_C_FLAGS_O3WITHASM:STRING=-Wall -save-temps -g -O3

CMAKE_C_FLAGS_RELEASE:STRING=-O3 -DNDEBUG

CMAKE_C_FLAGS_RELWITHDEBINFO:STRING=-O2 -g -DNDEBUG

CMAKE_ASM_FLAGS:STRING=-mcpu=cortex-a15.cortex-a7 -mtune=cortex-a15.cortex-a7 -mfpu=neon-vfpv4 -mfloat-abi=hard -funsafe-math-optimizations

CMAKE_ASM_FLAGS_DEBUG:STRING=-g

CMAKE_ASM_FLAGS_MINSIZEREL:STRING=-Os -DNDEBUG

CMAKE_ASM_FLAGS_RELEASE:STRING=-O3 -DNDEBUG

CMAKE_ASM_FLAGS_RELWITHDEBINFO:STRING=-O2 -g -DNDEBUG

Sean Nowlan

_______________________________________________
Discuss-gnuradio mailing list
address@hidden
https://lists.gnu.org/mailman/listinfo/discuss-gnuradio