Hello all, I am facing problems in constraining Negative Slacks (for a 50Mhz clock and 150MhZ clock both generated clocks). I had a negative slack of -59 which was because of using a 44_24 LPM divider which I solved using 2 stages of pipeline. There were few other slack issues which I constrained using set_clock_groups and a set_false_path. Now, I have setup negative slack of -14 and -6 (two different generated clocks) within a module where the input to the module comes from a written register(in SW), using which a division is performed and the value is stored in a register. The same happens with one more module. I am not sure why lot of timing issues come up with Division (Be it using lpm or a simple arithemtic division). I am not very familiar with timing analysis, and I am not sure if I should alter the code which can solve the timing issues or use constraints set_false_path multicycle paths setting max delays to ignore the constraints. I dont want to alter the code much which has been already tested. I am attaching the Setup Timing Closure Report, the Source code of the module and the SDC File.(see attachments) Looking forward to your suggestions! Thank you.
Glen wrote: > I am not very familiar with timing analysis In such a simple design you should not need this tool at all. Lets lokk at this here:
1 | s_PWM_int <= std_logic_vector(unsigned(v_PWMvalue) * to_unsigned(i_PWM_Freq_Devider / 100, 9)); -- Timing Fails |
You like to cast and convert you to death? Why not doing the whole internal calculation in integer? Beside that moaning about formalisms your problem is this:
1 | i_PWM_Freq_Devider / 100 |
Here you invoke a combinatorial divider. Its eating up lots of ressources and addtitionally such a thing is really, really, really slow because it has lots of logic stages behind each other. Its strongly not recommended to use a divider on hardware. The only "simple" dividers are those like 2, 4, 8, 16, 32, ...32768, 65536, ... because these power of two vlaues are just a little rewiring on the FPGA... So lets pack it togehter: a number like 10 or 100 or 1000 is only good for man, not for computers. If you ABSOLUTELY want to kepp on with your 100% scale i would try that:
1 | s_PWM_int <= std_logic_vector(unsigned(v_PWMvalue) * to_unsigned((i_PWM_Freq_Devider*4100/100)/4096, 9)); -- a little bit "uprounding" here... |
In that way I dealt the division by a "none power of two value" against a (fast and uncritical) multiplication and a division by a power of two. Try it... And one thing more a beginner MUST use only 1 clock in his designs. A pro will TRY to use only 1 clock in his design.
:
Edited by Moderator
Thanks for the response. This module is just one sub module.I had constraints with other modules which have been resolved. The FPGA runs on only one clock. But few other clocks have been generated to support other modules. I was not aware about the division strategy in FPGAs. It makes sense. I tried incorporating it. Unfortunately, it did not help. The negative slack has just increased more. I am not sure if it requires more latency or I should pipeline further to reduce the slack.
It worked when I removed 100 from the divisor. Thanks a lot. But the module with 150MHz clock still gives negative slack. Though, the same calculation is used there (which I changed to the you suggested), it doesnt help. The latch points where there is negative slack are s_PWMCounter and v_updatePWM value.
Watch this topic
Disable multi-page viewPlease log in before posting. Registration is free and takes only a minute.
Existing account
Do you have a Google/GoogleMail account? No registration required!
Log in with Google account
Log in with Google account
No account? Register here.