This patch reworks the latch synthesis that Alan submitted to be
more in line with the rest of the code and simplified it in a
few places. It also adds a warning that we do not currently
support synthesizing asynchronous set/clear signals for a latch.
In synthesis we transform registers into wires, but if there is bad
Verilog code where the register is written from non-synthesized code
then we can't do this! This patch adds a check in the code generator
for this and prints a error message instead of generating a .resolv
to combine the two, which is certainly not correct.
This patch reworks the NetLatch class to better match the NetFF class
(e.g. it adds a place for the async controls and value, adds a dump
routine, etc.) and moves it into the normal netlist.{h,cc} files. It
removes the unused Aload and Sload routines, etc. from the NetFF class.
It switches the latch to use the existing ff object to pass information
to the target and uses the existing FF routines as applicable.
ivl_lpm_async_clr()
ivl_lpm_async_set()
ivl_lpm_aset_value()
ivl_lpm_clk() for the gate
ivl_lpm_data()
ivl_lpm_q()
It adds support for generating a latch UDP just like is done for the
D flip-flop in the tgt-vvp code generator and fixes a bug in the
generation of the clear signal when both the Aclr and Aset signals
clear the register.
This patch fixes the bugs reported in pr1757438. These are fpga/edif
code generation bugs. For a DFF we need to add an inverter for a
negedge device and if both the Aclr and Aset signals clear the FF
we need to create the OR of these signals to drive the Aclr pin.
If both the async. clr and set pins clear a flip-flop then we need to
create an OR gate to combine the two signals to connect to the single
DFF clr pin. This patch fixes the tgt-vvp code generator to implement
this functionality.
This patch removes #ident since it is deprecated in recent gcc
versions. It also removes all the CVS comments and fixes a few
other compilation warnings. There are still a few compilation
warnings, but this takes care of most of them. Fixing the rest
will require a bit more thought.
System tasks and functions can be defined to have NULL arguments.
This patch makes the @* sensitivity list calculation skip these
NULL (constant) arguments.
This check could be useful, but it was making make check fail
since it was before base was redefined by -B. This check is
not in V0.9, so it was easier to just remove it from V0.8. This
would only happen if <path>/lib/ivl-0.8/ivl did not already
exist (on a pristine system).
Using auto_ptr in elab_net.cc (added in commit 15702f1, "Fix bad
memory reference." on 2008-11-28) requires #include <memory> when
compiling with gcc-4.3.
This is a back port from development of the code to put and check
that the vvp input file is the right version. It also adds the -V
flag to vvp. It does not add the VERSION_TAG information so will
only produce the single string ivl_version tag. The runtime will
read both forms since if it is feed a V0.9 file we want it to
complain nicely (no syntax error). To change the displayed version
for the various programs and checks change version.h.
This is a back port from development of the code to resize an
unsized constant value to 32 bits. Development used integer_width,
but that is not supported in V0.8. Development also supports using
the expression width which V0.8 does not.
This patch mirrors what was done in development to fail for zero
width repeats. Actually development supports them in some contexts,
this is allowed in 1364-2005, but V0.8 doesn't support them at all.
Cary R
Larry Doolittle
Alan M. Feldstein
Stephen Williams
Werner Hoch
Jared Casper