I have a few questions about the core simple_spi_top.v


begin
bcnt treg sck_o
if (~wfempty) begin
wfre state if (cpha) sck_o end
end

This is inside an always.
Here sck_o is assigned twice with two concurrent statements.
Is this correct and unambiguous ?
In Verilog all statements in a begin/end block are executed in parallel.
So I guess that when the "if" conditions are true, it is undetermined
which statement will be implemented by the synthsizer.
There is a similar problem in the same module:
the signals wfre and rfwe are assigned twice with two concurrent
statements.

Comments ?

tullio.grassi at reicom.it wrote:

Short version:
It might be OK, depending on which tools you use, but I would certainly
not recommend it.

Longer version:
According to the 'original' Verilog LRM, the order of multiple
non-blocking assignments, a) with intra-assignment delay, and b) ending
up in the same time slot, cannot be guaranteed. Ie the outcome of the
code above would be 'random' (see page 8-8 of "OVI Verilog hardware LRM
1.0" (my file is called ovi.verilog.lrm.1.0.pdf)).

But, in the newer Verilog 2001 standard, it seems that this has been
resolved in cases where there are a clear order of execution (as in the
example). I.e. the code above will work as intended in a Verilog
2001-compatible simulator (see section 9.2.2, p. 123, of IEEE
P1364-2005/D3 (a draft revision of IEEE std 1364-2001) (my document is
called 1364-2005_D3.pdf)).

Note however that repeated nonblocking assignments to the same variable
_without_ intra-assignment delay are OK with all tools that I know of.



Since delay controls are discarded by normal synthesis tools, I would
guess that most synthesis tools would produce the expected results.

Best regards
Gunnar Dahlgren
ASIC designer, Axis Communications AB, Sweden

Short version:
It might be OK, depending on which tools you use, but I would certainly
not recommend it.
[rih] I tried this code on multiple simulators and synthesis tools; even
before Verilog2001 standard was official. They all yield the same result.
You're welcome to rewrite the code if you feel uncomfortable with it.
Always remember to run gate-level simulations!!
If gate-level sim passes, the core probably works ;)

Richard




Longer version:
According to the 'original' Verilog LRM, the order of multiple
non-blocking assignments, a) with intra-assignment delay, and b) ending
up in the same time slot, cannot be guaranteed. Ie the outcome of the
code above would be 'random' (see page 8-8 of "OVI Verilog hardware LRM
1.0" (my file is called ovi.verilog.lrm.1.0.pdf)).

But, in the newer Verilog 2001 standard, it seems that this has been
resolved in cases where there are a clear order of execution (as in the
example). I.e. the code above will work as intended in a Verilog
2001-compatible simulator (see section 9.2.2, p. 123, of IEEE
P1364-2005/D3 (a draft revision of IEEE std 1364-2001) (my document is
called 1364-2005_D3.pdf)).

Note however that repeated nonblocking assignments to the same variable
_without_ intra-assignment delay are OK with all tools that I know of.



Since delay controls are discarded by normal synthesis tools, I would
guess that most synthesis tools would produce the expected results.

Best regards
Gunnar Dahlgren
ASIC designer, Axis Communications AB, Sweden

_______________________________________________ http://www.opencores.org/mailman/listinfo/cores