Re: [Discuss-gnuradio] CUDA-Enabled GNURadio gr_benchmark10 possible improvements
Upon a closer look into cuda_muiltiply_const_ff_kernel.cu, there exists 5
different kernel functions to do the multiplication, where the default one,
__global__ void
cuda_multiply_const_ff_kernel(const float* g_idata, float* g_odata,const
int noutput_items,const float konst)
is completely blank. But regardless, nobody calls these kernel functions.
Then, in the same file, which is called by cuda_multiply_const_ff.cc, in
this function
int
get_cuda_multiply_const_ff_kernel_params (
cuda_multiply_const_ff_kernel_params *params )
{
int result=0;
//const unsigned int max_num_threads_per_block = MAX_NUM_THREADS_ALL;
//can use the maximum number of threads if wanted
//unsigned int max_num_blocks = MAX_NUM_BLOCKS_ALL;
unsigned int num_blocks=4096 ;// =
gridDim.x; //NUM_CUDABLOCKS
unsigned int num_threads_per_block=512;// =
blockDim.x; //NUM_THREADS;
unsigned int num_outputs_per_block=num_threads_per_block;
const unsigned int num_outputs_per_grid=
num_outputs_per_block*num_blocks; //(blockDim.x)*gridDim.x
size_t dynamic_shared_mem_size =
0;//256*sizeof(float);//0;//num_threads_per_block*sizeof(gr_complex);
dim3 griddim( num_blocks, 1, 1);
dim3 threaddim( num_threads_per_block, 1, 1);
params-griddim=griddim;
params-threaddim=threaddim;
params-dynamic_shared_mem_size=dynamic_shared_mem_size;
params-num_outputs_padded=num_outputs_per_grid;
params-num_inputs_padded=num_outputs_per_grid;
params-num_inputs=0;//num_outputs_per_grid;//num_outputs;
params-num_outputs=0;//num_outputs_per_grid;//num_outputs;
//Now you can do the kernel invocation like this:
//cuda_multiply_const_ff_filter_kernel params-griddim,
params-threaddim, params-dynamic_shared_mem_size (g_idata, g_odata,
params-num_outputs_padded*X,konst);
return result;
}
The kernel invocation is completely commented out! The result is initialized
as 0 at the top and returns it. All the work in between to specify and
allocate thread, block sizes does not seem to matter. Not sure why this code
exists this way, did someone make an edit or did Martin specifically
commented out the kernel invocation? Is this suppose to be this way? I don't
see how this can be a proper benchmarking if it seems that we just test
about allocating threads and blocks on the device and memory access times,
but really don't do any computation.
I am probably way off here, doesnt make any sense..someone please
clarify!
___
Discuss-gnuradio mailing list
Discuss-gnuradio@gnu.org
http://lists.gnu.org/mailman/listinfo/discuss-gnuradio
[Discuss-gnuradio] Re: usrp for sale
new price $1000 +shipping need gone asap On Sat, Jun 27, 2009 at 2:47 AM, davekdavidki...@gmail.com wrote: im gonna sell my usrp for now , need some quick cash Its virtually brand new, sat on my desk since march only touched a few times to swap cards. usrp 1 motherboard, case, powersupply, fan, screws. dbsrx w/cable tvrx w/cable rfx900 will include all original packaging bought new for $1255.00 in march price is 1125$ in united states paypal or check email me for details, if i dont have any takers i will put on ebay dave ___ Discuss-gnuradio mailing list Discuss-gnuradio@gnu.org http://lists.gnu.org/mailman/listinfo/discuss-gnuradio
[Discuss-gnuradio] possible to sweep USRP oscillator?
Hey all, Is it possible to increase and decrease the USRP or daughter-boards oscillator frequency in real time in order to transmit a frequency modulated continuous wave from the frontend? Bruh -- Posted via http://www.ruby-forum.com/. ___ Discuss-gnuradio mailing list Discuss-gnuradio@gnu.org http://lists.gnu.org/mailman/listinfo/discuss-gnuradio
Re: [Discuss-gnuradio] possible to sweep USRP oscillator?
On Tue, Jun 30, 2009 at 11:28 AM, B. Godanali...@ruby-forum.com wrote: Hey all, Is it possible to increase and decrease the USRP or daughter-boards oscillator frequency in real time in order to transmit a frequency modulated continuous wave from the frontend? Is there a reason you wouldn't just want to do this at baseband and modulate it? Brian ___ Discuss-gnuradio mailing list Discuss-gnuradio@gnu.org http://lists.gnu.org/mailman/listinfo/discuss-gnuradio
Re: [Discuss-gnuradio] possible to sweep USRP oscillator?
On Tue, Jun 30, 2009 at 08:28, B. Godanali...@ruby-forum.com wrote: Is it possible to increase and decrease the USRP or daughter-boards oscillator frequency in real time in order to transmit a frequency modulated continuous wave from the frontend? Not really, no. But it is easy to send a baseband FM waveform to the USRP to be upconverted to your center frequency of interest. Johnathan ___ Discuss-gnuradio mailing list Discuss-gnuradio@gnu.org http://lists.gnu.org/mailman/listinfo/discuss-gnuradio
[Discuss-gnuradio] Output Stream Data Rates
Hello, Is possible to increase the output stream data rates? In gr_block.h, it mentions that all output streams must produce data at the same rate, but is the output stream data rate fixed? I'm working with a USRP2 and using gnuradio revision 10689 with the bbn 80211 mac layer code. Thanks, Miklos Christine ___ Discuss-gnuradio mailing list Discuss-gnuradio@gnu.org http://lists.gnu.org/mailman/listinfo/discuss-gnuradio
Re: [Discuss-gnuradio] Output Stream Data Rates
On Tue, Jun 30, 2009 at 09:50, Miklos Christinemchrist...@berkeley.edu wrote: Is possible to increase the output stream data rates? In gr_block.h, it mentions that all output streams must produce data at the same rate, but is the output stream data rate fixed? GNU Radio streams don't have a formal concept of data rates; it is simply up to the connecting blocks to agree on how to interpret the items arriving over stream. What is it you're really trying to accomplish? I'm working with a USRP2 and using gnuradio revision 10689 with the bbn 80211 mac layer code. The USRP2 code came out of 'beta' with the release of 3.2. You may wish to upgrade. Johnathan ___ Discuss-gnuradio mailing list Discuss-gnuradio@gnu.org http://lists.gnu.org/mailman/listinfo/discuss-gnuradio
[Discuss-gnuradio] Re: possible to sweep USRP oscillator?
Brian Padalino wrote: On Tue, Jun 30, 2009 at 11:28 AM, B. Godanali...@ruby-forum.com wrote: Hey all, Is it possible to increase and decrease the USRP or daughter-boards oscillator frequency in real time in order to transmit a frequency modulated continuous wave from the frontend? Is there a reason you wouldn't just want to do this at baseband and modulate it? Brian Yes, I want to do that to utilize more bandwidth of the daughter-boards rather than being limited by the sampling rate of the ADC/DAC. But, it looks it is impossible? -- Posted via http://www.ruby-forum.com/. ___ Discuss-gnuradio mailing list Discuss-gnuradio@gnu.org http://lists.gnu.org/mailman/listinfo/discuss-gnuradio
Re: [Discuss-gnuradio] Re: possible to sweep USRP oscillator?
On Tue, Jun 30, 2009 at 10:12, B. Godanali...@ruby-forum.com wrote: Yes, I want to do that to utilize more bandwidth of the daughter-boards rather than being limited by the sampling rate of the ADC/DAC. But, it looks it is impossible? If you reprogram the USRP(1) FPGA code, you can generate waveforms up to 32 Msps into the AD9862, which uses 1:4 hardware upsampling to get to the 128 Msps of the DAC. Theoretically this gives you 64 MHz of (complex baseband) bandwidth, though the passband of the AD9862 upsampling filter will limit this to about 50 MHz of usable bandwidth. On the receive side, you do get the full 64 MHz downconverted passband, but the anti-aliasing filters on most of the daughterboards will effectively limit this to 60 MHz. Johnathan ___ Discuss-gnuradio mailing list Discuss-gnuradio@gnu.org http://lists.gnu.org/mailman/listinfo/discuss-gnuradio
Re: [Discuss-gnuradio] Re: USRP - No quadrature samples
On Mon, Jun 29, 2009 at 04:21, Nick Withersn...@nickwithers.com wrote: I'm a bit worried... I no longer seem to be able to get anything but zeroes on the quadrature channel out of gr-radar-mono (and my derived wind profiler code). If someone with an up-to-date Ubuntu 9.04 machine with GNUradio 3.2 could try to reproduce this I'd owe them a beer! I just tested gr-radar-mono with 3.2 and Ubuntu 9.04, and am getting echo data on both I and Q channels. The -D debugging option returns an FPGA counter value correctly. You can test your BasicRX with the signal generator and usrp_fft.py. Be sure to connect it to the RX-A channel. Johnathan ___ Discuss-gnuradio mailing list Discuss-gnuradio@gnu.org http://lists.gnu.org/mailman/listinfo/discuss-gnuradio
[Discuss-gnuradio] gnuradio.org down?
Can't seem to hit the SVN repository. --n _ Lauren found her dream laptop. Find the PC that’s right for you. http://www.microsoft.com/windows/choosepc/?ocid=ftp_val_wl_290___ Discuss-gnuradio mailing list Discuss-gnuradio@gnu.org http://lists.gnu.org/mailman/listinfo/discuss-gnuradio
Re: [Discuss-gnuradio] gnuradio.org down?
On Tue, Jun 30, 2009 at 13:06, Nick Fosterbistro...@hotmail.com wrote: Can't seem to hit the SVN repository. Yes, it is down, working on it. Johnathan ___ Discuss-gnuradio mailing list Discuss-gnuradio@gnu.org http://lists.gnu.org/mailman/listinfo/discuss-gnuradio
Re: [Discuss-gnuradio] gnuradio.org down?
On Tue, Jun 30, 2009 at 13:09, Johnathan Corganjcor...@corganenterprises.com wrote: Can't seem to hit the SVN repository. Yes, it is down, working on it. The site is back up and running. Johnathan ___ Discuss-gnuradio mailing list Discuss-gnuradio@gnu.org http://lists.gnu.org/mailman/listinfo/discuss-gnuradio
Re: [Discuss-gnuradio] CUDA-Enabled GNURadio gr_benchmark10 possible improvements
On Tue, 2009-06-30 at 02:52 -0400, Yu-Hua Yang wrote:
Upon a closer look into cuda_muiltiply_const_ff_kernel.cu, there
exists 5 different kernel functions to do the multiplication, where
the default one,
__global__ void
cuda_multiply_const_ff_kernel(const float* g_idata, float*
g_odata,const int noutput_items,const float konst)
is completely blank. But regardless, nobody calls these kernel
functions.
cuda_multiply_const_ff was never finished or I accidently removed some
code here.
Either way, this is clearly a bug.
The empty kernel should call one of the implemented kernels.
Then, in the same file, which is called by cuda_multiply_const_ff.cc,
in this function
int
get_cuda_multiply_const_ff_kernel_params
( cuda_multiply_const_ff_kernel_params *params )
{
int result=0;
//const unsigned int max_num_threads_per_block =
MAX_NUM_THREADS_ALL; //can use the maximum number of threads
if wanted
//unsigned int max_num_blocks = MAX_NUM_BLOCKS_ALL;
unsigned int num_blocks=4096 ;// =
gridDim.x; //NUM_CUDABLOCKS
unsigned int num_threads_per_block=512;// =
blockDim.x; //NUM_THREADS;
unsigned int num_outputs_per_block=num_threads_per_block;
const unsigned int num_outputs_per_grid=
num_outputs_per_block*num_blocks; //(blockDim.x)*gridDim.x
size_t dynamic_shared_mem_size =
0;//256*sizeof(float);//0;//num_threads_per_block*sizeof(gr_complex);
dim3 griddim( num_blocks, 1, 1);
dim3 threaddim( num_threads_per_block, 1, 1);
params-griddim=griddim;
params-threaddim=threaddim;
params-dynamic_shared_mem_size=dynamic_shared_mem_size;
params-num_outputs_padded=num_outputs_per_grid;
params-num_inputs_padded=num_outputs_per_grid;
params-num_inputs=0;//num_outputs_per_grid;//num_outputs;
params-num_outputs=0;//num_outputs_per_grid;//num_outputs;
//Now you can do the kernel invocation like this:
//cuda_multiply_const_ff_filter_kernel params-griddim,
params-threaddim, params-dynamic_shared_mem_size
(g_idata, g_odata, params-num_outputs_padded*X,konst);
return result;
}
The kernel invocation is completely commented out! The result is
initialized as 0 at the top and returns it. All the work in between to
specify and allocate thread, block sizes does not seem to matter. Not
sure why this code exists this way, did someone make an edit or did
Martin specifically commented out the kernel invocation?
Yes the kernel invocation is specifically commented out.
The commented out kernel invocation is only here as documentation on
how to do the actual kernel invocation.
There is still a small typo.
It should refer to cuda_multiply_const_ff_kernel and not
cuda_multiply_const_ff_filter_kernel
This methods name is get_cuda_multiply_const_ff_kernel_params
It does just that, determine the needed params for the kernel
invocation.
The actual kernel invocation is done in:
cuda_multiply_const_ff_work_device
Both get_cuda_multiply_const_ff_kernel_params and
cuda_multiply_const_ff_work_device are called from normal C++ code in
cuda_multiply_const_ff_kernel.cu
Better examples to look at and benchmark would be
cuda_quadrature_demod_cuda_cf
cuda_fir_filter_fff
These are both used in
testbed/wfm/cuda_wfm_rcv.py
Which calls the complete cuda implementation of a WFM receiver in
cuda_wfm_rcv_cuda.py
You will notice a cuda_multiply_const_ff block is instantiated as
volume_control but not used in the final connect because it didn't work.
Now I know this is because of the bug you found.
Is this suppose to be this way? I don't see how this can be a proper
benchmarking if it seems that we just test about allocating threads
and blocks on the device and memory access times, but really don't do
any computation.
That is not a good benchmark indeed.
cuda_multiply_const_ff should be fixed first.
The fir-filter is however the thing you really want to benchmark.
(Which is also in there)
The more taps the fir-filter has, the more computations it will have to
make.
Note that the cuda implementation is at the moment limited in the number
of taps it can support at a certain decimation factor.
If I remember well decimationfactor*numberoftaps should stay below 2048
(or was it 512)
Ways of improving the resuls include having gnuradio do calculations in
bigger chunks.
This can be done in several ways.
One way is to have a block at the end of the processing chain which has
a large output_multiple requirement which needs 1 input for every
output.
In other words. Connect a block to the end of the processing chain which
has set_output_multiple(large_number) in
[Discuss-gnuradio] Is the GNU Radio Site Down?
Hi all, I have been trying to access the site for the past couple of days but in vain. I am new to GNU Radio, I just started using it. Is there a problem with the site? plus subversion is giving me host not found errors Best, Ahmed -- Ahmed Ali ElDin M. Hassan Research Assistant Information Security Group Nile University Smart Village - km28 Cairo-Alexandria Desert Road Giza, EGYPT EMail: ahmed.aliel...@nileu.edu.eg ___ Discuss-gnuradio mailing list Discuss-gnuradio@gnu.org http://lists.gnu.org/mailman/listinfo/discuss-gnuradio
Re: [Discuss-gnuradio] Re: USRP - No quadrature samples
On Tue, 30 Jun 2009 11:00:56 -0700, Johnathan Corgan jcor...@corganenterprises.com wrote: On Mon, Jun 29, 2009 at 04:21, Nick Withersn...@nickwithers.com wrote: I'm a bit worried... I no longer seem to be able to get anything but zeroes on the quadrature channel out of gr-radar-mono (and my derived wind profiler code). If someone with an up-to-date Ubuntu 9.04 machine with GNUradio 3.2 could try to reproduce this I'd owe them a beer! I just tested gr-radar-mono with 3.2 and Ubuntu 9.04, and am getting echo data on both I and Q channels. Champion, thanks for testing that. The -D debugging option returns an FPGA counter value correctly. Yeah, that works on my side too. You can test your BasicRX with the signal generator and usrp_fft.py. Be sure to connect it to the RX-A channel. Righto, will do. Thanks very much for your help! Johnathan -- Nick Withers email: n...@nickwithers.com Web: http://www.nickwithers.com Mobile: +61 414 397 446 ___ Discuss-gnuradio mailing list Discuss-gnuradio@gnu.org http://lists.gnu.org/mailman/listinfo/discuss-gnuradio
Re: [Discuss-gnuradio] Is the GNU Radio Site Down?
On Tue, Jun 30, 2009 at 17:19, Ahmed Aliahmed.aliel...@nileu.edu.eg wrote: I have been trying to access the site for the past couple of days but in vain. I am new to GNU Radio, I just started using it. Is there a problem with the site? plus subversion is giving me host not found errors The gnuradio.org server was down for about 3 hours today due to a memory issue, but it has been fine otherwise including the last couple of days. Johnathan ___ Discuss-gnuradio mailing list Discuss-gnuradio@gnu.org http://lists.gnu.org/mailman/listinfo/discuss-gnuradio
[Discuss-gnuradio] Earn money for Description of codes
Dear all GNU Users We need some detail description of some prgrams like benchmark_tx/rx in GNU radio etc for out tutorial sessions. If some body can provide complete discription and do some coding , we have resonable amount to spend. Students are encouraged but professional can also participate. Interested can mail us directly at masseyj...@yahoo.co.nz other details can disuss via mail. Thnaks Jhon ___ Discuss-gnuradio mailing list Discuss-gnuradio@gnu.org http://lists.gnu.org/mailman/listinfo/discuss-gnuradio
[Discuss-gnuradio] RFX-1800 Tranciever
Hello Is it possible to run benchmark_tx and benchmark_rx on same computer with one USRP with RFX-1800 Transciever daughter board. ___ Discuss-gnuradio mailing list Discuss-gnuradio@gnu.org http://lists.gnu.org/mailman/listinfo/discuss-gnuradio
