RTL-SDR: not fast enough without FPU?

[Sergeanter]

I am trying to use Pogo V4 with RTL2832U USB receiver to run dump1090 (Mode S decoder) and acarsdec or rtl_acars_ng.
Dump1090 works well with about 30% load. Both ACARS decoders choke after 4-5 messages since CPU load hits 99%.
I have recompiled both ACARS decoders with profiling option and they seem to be limited by floating point multiplication .
I wonder if anything can be done to optimize executables to make floating point emulation just a tad faster.
I ran compiler with -march=armv5te and -mtune=native. Is this all that can be done?

[moonman]

I think native defaults to armv4. Run this to find out
$this->bbcode_second_pass_code('', 'gcc -c -Q -march=native --help=target | grep march')

You could use O3 optimizations (or even Ofast if it dorsn't break anything) but i doubt any of this will help much

[Sergeanter]

The combo of -march=armv5te -mtune=native results in "armv5te" setting. I am not sure if -mtune does anything after -march is set. It was already compiled with -Ofast . I will try to overclock it today by pulling serial port up ( I do not need serial anymore since I used it only to reflash V4 to boot from SD card). I am assuming DSP code is optimized already and nothing can be done.
I had really big hopes for this to work.... Dump1090 is just awesome by the way. Not only it does DSP with just 30% load , but also it runs a simple web server to show planes on Google map in browser .

[Sergeanter]

acarsdec works when Pogo V4 is overclocked from 800MHz to 1000MHz. CPU load is 85% with one ACARS channel being decoded. I am looking into re-writing parts of DSP code using CORDIC instead of floating point trig.
I wonder if anybody else is playing with RTL SDR on arm5.