Open beardedN5rd opened 5 years ago
Hi Joseph, maybe it is an issue with NEON on ARMv7-A architecture, like stated in the gcc options
If the selected floating-point hardware includes the NEON extension (e.g. -mfpu=‘neon’), note that floating-point operations are not generated by GCC's auto-vectorization pass unless -funsafe-math-optimizations is also specified. This is because NEON hardware does not fully implement the IEEE 754 standard for floating-point arithmetic (in particular denormal values are treated as zero), so the use of NEON instructions may lead to a loss of precision.
In the meanwhile I tried to use softfp, but got errors as well. autotest.softfp.json.txt
I'd be interested in seeing why those particular (fm demod) tests fail and what the issues are there.
Can I provide any data to clarify?
I am not sure if running the check routine on an Raspberry 2 should result in similar issues, maybe this could be an approach to identify if it is problem that may occur on all ARMv7-A based processors or if it is only a problem on Cortex-A9/ Tegra T30
I was actually going to suggest that! Any samples you could send me would help immensely!
Hi Joseph, I am trying to do some rds decoding using https://github.com/windytan/redsea and only get a subset of the messages. Digging deeper I noticed that liquid-dsp have some issues in its checks. I added autotest.json below. autotest.json.txt
Initially with the auto generated build options (CONFIG_CFLAGS = -g -O2 -ffast-math -mcpu=cortex-a7 -mfloat-abi=hard -mfpu=neon-vfpv4) it failed to run due to illegal arguments. With corrected options (CONFIG_CFLAGS = -g -O2 -ffast-math -mcpu=cortex-a9 -mfloat-abi=hard -mfpu=neon) it works partly. Most tests run positively, some rds messages get decoded.
My environment is mostly fixed and can not be updated essentially:
Any idea how I can get it work? Do you need some further information?
Hagen