random postings from Xilinx forums

This finds all the cells of a certain type but it doesn’t check if the CE is connected to VCC

show_objects -name find_1 [get_cells -hierarchical -filter { PRIMITIVE_TYPE == REGISTER.SDR.FDCE } ]

This however is more through: finds all flops with an enable pin and then verifies that it’s not connected to power (vcc)
set ffs_with_ce {}

foreach ff [get_cells -hier -filter {PRIMITIVE_SUBGROUP == “flop”}] {
if {[get_property TYPE [get_nets -of_objects [get_pins -of_objects $ff -filter IS_ENABLE]]] != “POWER”} {lappend ffs_with_ce $ff}

HARD_SYNC Dedicated Synchronizers

In recent generations of Xilinx FPGAs a new feature has been introduced: HARD_SYNC dedicated synchronization registers.
These are hardwired/dedicated blocks with 3 registers specifically designed with synchronization purposes.
It’s possible to configure a hard_sync synchronizer for 2 or 3 stages of shift registers. These should give the best defence against meta-stability while sampling asynchronous signals.

Here is how they are used in Verilog:

.INIT(1’b0), // Initial value
.IS_CLK_INVERTED(1’b0), // configurable inverted clock
.LATENCY(2) // 2-3 stage shift register
uhardsync (
.DIN(DIN), // input Data
.DOUT(DOUT), // output Data
.CLK(CLK) // Clock

how to find all the cells in a timing path in Xilinx Vivado

In Vivado, one way to cleanup timing after final place & route is to unplace the cells in the worst negative slack path and just place & route that portion. I have had great success with this in the GUI flow. It’s quite easy there. You just do a timing report (report_timing_summary), right-click on the timing path and say unplace cells. Then you run the cleanup commands.

Of course it’s tedious to do this manually but one has to find all the cells involved in a timing path. Here is what I came up with:

set string [report_timing -no_header -path_type full -return_string]
set items [split string]
foreach item $items {if [regexp {TOP\/.*$} item] {lappend cells [get_cells -filter {IS_PRIMITIVE==1} -of_objects [get_pins $item]]))

So here we get a timing report, find all fields in it then if the item starts with our TOP module name, we assume it’s a pin name and get the cell of that pin. This finds some of the same cells twice but unplacing an already unplaced cell is fine so we don’t have to worry about duplicates.

Finally one can do

unplace_cell $cells
place_design -directive Explore
route_design -directive Explore
route_design -tns_cleanup
phys_opt_design -directive Explore

for final rip & re-route stage. I usually get very good timing results from this.

Vivado Synthesis: RLOC and BEL attributes not supported when accessing array of strings

AR #65413 is a really annoying answer record from Xilinx. They are basically saying that a perfectly legitimate piece of code won’t work in Vivado synthesis and they have no intention of fixing it. Alas the following code will not work:

  for (g=0; g < D; g++) begin: regs
    //string rlocval;
    //localparam string rlocpre = "X0Y";
    //localparam string rloc = {rlocpre, rlocval.itoa(g/8)};
    localparam rlocpre = "X0Y";
    localparam byte rlocbyte = g/8 + 48;
    localparam rloc = {rlocpre, rlocbyte};
    localparam bel[8] = {"AFF", "A5FF", "BFF", "B5FF", "CFF", "C5FF", "DFF", "D5FF"};
    (* RLOC = rloc, BEL = bel[g%8] *) FDRE u(.C(clk), .D(d[g]), .Q(q[g]), .CE(1'b1), .R(1'b0));

how to use Xilinx AXI IIC (I2C) controller with SCCB devices and no pull-up on SCL

Omnivision Image sensors support a serial control interface called SCCB Omnivision SCCB spec

The main difference between SCCB & I2C is that it’s specified as SCLK being output only and active driver instead of open-drain (or tri-state or open-collector). If one actually wants to remove the pull-up on the SCLK and use this feature, it causes a minor issue with some I2C IP blocks in FPGA as they assume open-drain SCLK. Specifically in Xilinx AXI IIC controller, the IP generates 3 signals (scl_i, scl_o, scl_t) which are normally used to control a set of IBUF & OBUFT pads to drive scl_io PAD. The way to use this IP in an SCCB setting is to use the scl_t signal to drive the scl_io. Xilinx OBUFT has an active high tri-state so scl_t is high when scl_io needs to be pulled-up and it’s low with scl_o being also low. So scl_t carries the intent of scl_io signal when active driven. Also scl_i should be connected to scl_t for the IP to sense the driven signal.

One final complication is that if one makes the IIC interface external and uses some of the signals in the Vivado block diagram, Vivado makes all remaining signals external too which creates non-existing pins so all scl_x signals (other than scl_t) should be consumed inside the block diagram. scl_i being connected to scl_t takes care of that signal but scl_o should also be consumed. One can use it as an active low source or “or” it to another signal (it’s always zero so it can’t hurt) for it to be not exported out of the top level wrapper.