Retiming option of Synthesis causes Synthesis failure - 2017.3.1

[sachinb_apt1]

Hello,

 I am using Vivado 2017.3.1

As soon as enable Retiming in Synthesis, Synthesis fails with

TclStackFree: incorrect freePtr. Call out of sequence?

I cant upgrade to 2017.4 for some reason.I am going ahead without using the retiming option.

Thanks,

Sachin B

[prathikm]

Hi @sachinb_apt1,

Do you have a test-case that we can try?

Regards,
Prathik
-----------------------------------------------------------------------------------------------
Please mark the appropriate post as an answer "Accept as solution" in case it helps to resolve your query.
Give Kudos to a post which you think is helpful and reply oriented.
-----------------------------------------------------------------------------------------------

[sachinb_apt1]

I will try to give a test case. BTW i tried with 2017.4 as well. The same issue.
With 2017.4, an extra line printed in the runme.log file terminate called after throwing an instance of 'boost::exception_detail::clone_impl<boost::exception_detail::error_info_injector<boost::lock_error> >'
what(): boost: mutex lock failed in pthread_mutex_lock: Invalid argument

And there are 2 hs_err_pid*.log files generated. As far as i know, one one should be generated for a vivado crash. But here i got 2 of them.
Getting Retiming to work is extremely critical for us. Thanks

[pulim]

Hi @sachinb_apt1

Can you share vivado log and hs_err_pid* files to ahve a look?

It will be helpful if you can provide testcase to debug and provide you a workaround.

Thanks,

Manusha

[sachinb_apt1]

The attachment's hs_err_pid96778.log content type (text/x-log) does not match its file extension and has been removed.
The attachment's hs_err_pid96904.log content type (text/x-log) does not match its file extension and has been removed.
The attachment's runme.log content type (text/x-log) does not match its file extension and has been removed.

[sachinb_apt1]

@pulim PFA files attached

[sachinb_apt1]

Is there a set_param in vivado, with which we can enable either the forward or backward register balancing like we would do them in ISE?

Well what i  really trying to get register balancing done using some micro level switches instead of a macro level retiming switch.

Also let me know if there is  way to apply retiming switch to certain module or instance?

Thanks,

Sachin B

[sachinb_apt1]

Hi @pulim

  I suspect if a Combinatorial loop in GT logic is causing it.

The powergood signal generation in the GTY IP has a combinatorial loop. Which is mentioned in the runme log file i posted

See the snapshot attached.

Also let me know if there was a way to apply retiming to specific instance (and downstream) instead of entire design.

[pulim]

Hi @sachinb_apt1

I did a quick search with the crash log but didnt find any known issues with similar log.

Can you share the project so that I can debug and report this to development team?

You can apply re-timing to specific module instead of entire design uisng block level synthesis.

Please check below.

set_property BLOCK_SYNTH.RETIMING 1 [get_cells  <instname>]

For more details on block level synthesis refer to chapter 3 in UG901.

https://www.xilinx.com/support/documentation/sw_manuals/xilinx2017_4/ug901-vivado-synthesis.pdf

[sachinb_apt1]

Well with the block level

    Numbers of forward move = 0, and backward move = 0
    Numbers of forward move = 0, and backward move = 0
    Numbers of forward move = 0, and backward move = 0
    Numbers of forward move = 0, and backward move = 0
    Numbers of forward move = 0, and backward move = 0

So pretty much it passed the synthesis but did not do any register balancing.

And not sure what such messages in synth log

Total number of crtical paths = 0
    Total number of crtical paths = 0
    Total number of crtical paths = 0
    Total number of crtical paths = 0
    Total number of crtical paths = 0
    Total number of crtical paths = 0
    Total number of crtical paths = 0
    Total number of crtical paths = 0
    Total number of crtical paths = 0
    Total number of crtical paths = 0

During Synthesis the tool probably thinks there are no critical path and hence it did not do any register balancing.

However my post synthesis Negative Slack is -1.136 ns for the block i did the retiming with 16 levels of logic.

How did during synthesis/Retiming the tool saw not even a single path critical. Strange.

I tried creating a sharable testcase but it doesnt fail retiming in sub block level.

So cannot share the project.

Closing Comments for now--- Whether a tool does the retiming on some block or not, it should not crash the way it is.