search

taijusti / sleep_apnea

0 stars 0 forks source link

Implement local manager #3

Closed taijusti closed 9 years ago

taijusti commented 9 years ago

-Currently trying to get two Zedboards to communicate over Ethernet. -Should we have a full network stack? Searching online and asking around, I think the only way to do this is to put the barebones Linux distro that Xilinx includes as one of the BSPs. -will propose protocol

taijusti commented 9 years ago

Apparently we only have simulation (no bitstream generation) licenses for Ethernet. Someone else from our research group figured it out and is going to approach Paul about it today. This explains the error that I've been getting when trying to generate my design in Vivado. I guess I'll try to make a guess at what the local manager will look like in the meantime.

taijusti commented 9 years ago

I'm code complete on the local manager and communication (including changes to device_tb). Everything is still commented out right now however since we have two problems preventing me from turning it on (+ the fact that we're trying to get a single FPGA working with Patrick's work): 1) FIFO's can only be of one data type (i chose unsigned int), but our data is of both float/int. Solution is just to go to fixed point right now. 2) it doesn't like the dataflow pragma I've put in. I think the implementations need to use the hls::stream template class. So I'm gonna give that a shot right now

I'm working on these things right now. I need to update device_tb to also check delta_e. Hopefully will get this done by end of day.

taijusti commented 9 years ago

ive made an intermediate commit in the form of cpp files. they do not affect the current c files in any way. to enable to the cpp files, just set define INTEG_FULL in common.h. right now there are still compile time issues. after that i need to run device_tb to make sure it works. i had to replace the gaussian radial kernel with a linear kernel for now since it does not pass device_tb.

taijusti commented 9 years ago

passes c-sim. needs a little bit more work: 1) convert everything to fixed point (i.e. use ap_fixed). i'll leave this as part of the issue regarding changing everything to fixed point. 2) make test bench to test delta e pipeline as well. 3) investigate why expf() takes forever to synthesize, if at all? need to look into this as a separate issue

i'll write the test bench for 2 then close this issue.

ghost commented 9 years ago

concerning the third point, I have synthesized expf and it worked fine, the RTL co-simulation passed too

On Wed, Apr 29, 2015 at 10:51 PM, Justin notifications@github.com wrote:

passes c-sim. needs a little bit more work: 1) convert everything to fixed point (i.e. use ap_fixed). i'll leave this as part of the issue regarding changing everything to fixed point. 2) make test bench to test delta e pipeline as well. 3) investigate why expf() takes forever to synthesize, if at all? need to look into this as a separate issue

i'll write the test bench for 2 then close this issue.

— Reply to this email directly or view it on GitHub https://github.com/taijusti/sleep_apnea/issues/3#issuecomment-97642387.

taijusti commented 9 years ago

hmm, alright... ill look into it

taijusti commented 9 years ago

synthesizes. passes c-sim. synthesis takes for ever. still need to add some directives so it passes timing closure. closing