Vivado build : tcl shell mode vs GUI mode

[randikae]

Hi All,

Recently I came up with this problem that when the same project was build using GUI it has a better timing performance (High positive WNS value) against when I build the same project using TCL shell mode. Both builds are using the same configuration for implementation. Following is the Build configuration I use.

GUI Mode:

Build Configuration

Timing results on GUI

TCL shell mode:

 synth_design -top itch_top -part xcku095-ffva1156-2-e
 opt_design -directive Explore
 place_design -directive Explore
 phys_opt_design -directive Explore
 route_design -directive Explore -tns_cleanup
 phys_opt_design -directive AggressiveExplore
 write_checkpoint -force post_route.dcp

Timing results on tcl shell mode

FPGA: Xilinx kintex ultrascale

Is this usual ? As I use jenkins to automate the build, tcl shell mode is critically important to me. Anyone got some insight on this ?

Best Regards,
randika.

[rshekhaw]

Hi @randikae ,

This seems to tool repeatability issue.

Vivado should generate identical results between runs involving identical:

• Design sources
• Constraints
• Tcl scripts and command sequences
• Tool and command options
• Vivado software version
• Multi-threading settings
• Operating Systems
• use of the route_design -ultrathreads argument

Please check this AR#:

https://www.xilinx.com/support/answers/61599.html

Also which vivado version you are working with?

Thanks,

Raj

[randikae]

Hi @rshekhaw 

Thank you for the reply.

I am using Vivado 2017.2

Yes the build results are identical between runs but only when using the same method. The difference between these two methods are consistant.

[rshekhaw]

Hi @randikae ,

Will it be possible at your end to try it in Vivado 2019.2?

Or else will it be possible for you to share the post opt dcp?

Thanks,

Raj

Spoiler

[rshekhaw]

Hi @randikae ,

Is the issue resolved at your end?

Thanks,

Raj

[randikae]

Hi @rshekhaw

For now it is difficult for me to get through the permission to post the dcp you asked.

I Noticed that my GUI build had run with general.maxThreads = 2 and tcl shell mode run with general.maxThreads = 8

(Please note that the better performance was in GUI mode).

My CPU is intel Xeon E3-1220 (4 Core) and the RAM is 64GB.

Can lower general.maxThreads parameter resulted with higher performance, If so how to choose the best threadcount ?

[hemangd]

@randikae 

The lower the thread count the higher repeatability of results you get in each runs. Hence, try to set 1 max thread for both tcl mode and GUI mode and observe that you see different values or not.  - I believe it will help in this issue. But, it may affect your performance. (So if you are looking for repeatability only then you can use this suggest and also refer that AR which @rshekhaw  had mentioned above.)

Furthermore, 

To do optimum usage of your machine/cores, you can check how many cores are present in your server machine by doing nproc command. And then you can set set_param general.maxThreads <number of cores>. 

Also, now it depends that whether your design requires a high number of cores to run a specific phase of design or not. - This point will get decided by the tool itself.