A concat typically has multiple inputs. Whenever one of the input values
change the output value of the concat is updated and propagated to its
downstream consumers.
When multiple inputs change within the same cycle each input will cause a
update propagation. Depending of the overall structure of the design this
can cause a significant performance penalty.
E.g. the following synthetic structure has a exponential runtime increase
based on the value of N.
```
reg [N-1:0] x;
generate for (genvar i = 0; i < N - 1; i++)
assign x[i+1] = ^{x[i],x[i]};
endgenerate
```
To improve this defer the value propagation of the concat to the end of the
current cycle, this allows multiple input updates to be included in a
single output update.
For the example in report #1052 this reduced the runtime from 2 minutes to
essentially 0.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
The concat functors use individual bit access to initialize and copy
values.
For initialization pass the initial bit value to the constructor and for
coping use set_vec() instead. Both can be a fair bit faster since data is
copied word by word rather than bit by bit.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
The implementation for partial receive for concat only differs from the
regular receive in that it takes an additional offset.
The regular receive can easily be implemented by calling the partial
receive with an offset of 0. This allows to remove some duplicated code.
The overhead of this is negligible, but to help the compiler to optimize this
a bit better mark the `recv_vec()` and `recv_vec_pv()` functions as final.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
Before the start of simulation, functor update events resulting from
initial value propagation are added to the initialisation event queue
(schedule_init_list). Once simulation has started, they are added to
the main event queue (sched_list). The cbStartOfSimulation callbacks
are executed after the initialisation event queue has been emptied.
Currently, if these callbacks generate further functor update events,
those events are added to the initialisation event queue, but that
queue is not looked at again. Instead, make sure any new events are
added to the main event queue.
This issue and proposed fix was reported by gatk555 in PR #1065.
The part functor has no real typed state and the bitsr field of the state
struct is unused. Remove it.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
This causes tgt-vvp to use a lower case 'c' instead of an upper case
'C' as the prefix for constant values used to initialise undriven nets.
For use by the following commit.
vvp_net_ptr_t uses vvp_sub_pointer_t to implement a tagged pointer with the
tag containing the port number.
The size of the tagged pointer is that of a normal pointer and could easily
be passed in a register when passing it as an argument to a function.
But since the vvp_sub_pointer_t type has a non-standard destructor it is
instead passed on the stack with the register containing a pointer to the
stack location where the value is stored.
This creates extra boiler plate code when passing a vvp_net_ptr_t to a
function writing and reading the value to and from the stack.
Use the default destructor for vvp_sub_pointer_t to avoid this and have the
value passed in a register.
There isn't much of a performance gain but the change is simple enough to
do anyway.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
The repeat functor can receive a partial vector. Make sure this is handled.
Since the expectation is that will only happen if the input wire is driven
by a single partial selection the default recv_vec4_pv_() can be used which
replaces the missing bits by `z`.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
An initial value propagation should only be scheduled for variable array
ports, but not for net array port since those do not contain any values.
This got accidentally broken when fixing support for 2-state variable
array ports.
Add a check that only does the initial value propagation if the port is for
a variable array.
Fixes: 0eb01fff1e ("vvp: Fix initial value propagation for 2-state non-automatic arrays")
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
find_scope() requires the hierarchical path to use escaped identifiers;
without the leading '\' and trailing ' ', if there are any '.' characters,
the path will be broken up at the wrong point(s).
vvp is missing a parser rule for unsigned 2-state net arrays. E.g.
```
bit a[0:1];
assign a[0] = 1'b0;
```
will synthesize fine, but when running it with vvp cause a syntax error.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
For 2-state non-automatic arrays currently real 0.0 value is propagated as
the initial value. This will cause an assert at the downstream receivers
which expect a vector.
Make sure that the vector value 0 is propagated instead for 2-state
vectors.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
When vvp parses a command which it thinks is a system call it tries to
match the symbols to values in the scope. The typo uses the wrong index
variable to access the vpi table. This results in a failed dynamic cast
which goes unchecked until the value is dereferenced, resulting in a
segfault.
Automatic 2-state vectors currently get initialized to 'hx, while their
default value should be 0.
Make sure the vector is initialized to 0 at the beginning of the automatic
context.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
Partial out-of-bounds write to a function's return value will trigger an
assert, even though the operation is valid.
The assert checks that the truncated value has the expected width, but
instead it should check that the non-truncated value has the expected with.
Move the assert before the truncation to fix this.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
Directly casting a negative double to a unsigned integer type is undefined
behavior and has different results on x86 and ARM based platforms.
On x86 the behavior is similar to casting a signed int to an unsigned
integer, i.e. the sign bit will end up in the MSB. But on ARM the result
will be 0.
To get consistent behavior, first cast to signed integer and then cast the
signed integer value to an unsigned integer value.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
In the current implementation a `%fork` instruction in a final block will
get scheduled, but never executed.
And while SystemVerilog requires a `final` procedure to execute in 0 time
and so no SystemVerilog `fork` is allowed inside of it, there are some
other scenarios where iverilog generates `%fork` statements.
For example when declaring variables in a sub-block a sub-scope with its
own thread is is used to allocate the storage for those variables and
`%fork` is used to execute the child thread.
E.g. the following, while being valid SystemVerilog, will never execute the
loop because the generated code will implement the loop as a child thread
being executed by a `%fork` statement.
```
final for (int i = 0; i < 10; i++) $display(i);
```
To mitigate this treat final statements the same as functions and rather
than scheduling a child thread, execute it immediately when using the
`%fork` statement.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
Currently when reading a number literal through the VPI API as a
vpiStringVal all null-bytes in the literal get ignored. This behavior is
different from when reading a signal through the VPI API as a vpiStringVal.
The latter will only ignore leading null-bytes and replace other null-bytes
with a space. E.g. the following two will print different values.
```
$display("%s", "a\000b"); // -> " ab"
reg [23:0] x = "a\000b";
$display("%s", x); // -> "a b"
```
For consistency modify the number literal formatting code so that it has
the same behavior as the signal value formatting code and only replaces
leading null-bytes.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
The VPI API for string literals does not correctly handle the case where a
null-byte ('\0') appears in the string literal. It uses strlen() to
calculate the length of the literal, which will give the wrong result if
there is a null-byte in the string literal. Instead of using strlen() use
the stored length to fix this.
In addition when formatting a string literal as a string ignore any
null-bytes. The LRM is not entirely clear what should happen to null-bytes
when formatting a value as a string. But the behavior of ignoring the
null-bytes is consistent with the rules of SystemVerilog for converting a
string literal to a SV string.
This problem can occur when a string literal gets null-byte left-padded due
to width of its context of its expression, but then optimization removes
part of the expression and only leaves the padded string literal.
E.g.
```
$display(0 ? "Yes" : "No");
```
will be transformed into
```
$display("\000No");
```
There is also one subtle change in behavior associated with this. The empty
string ("") is supposed to be equivalent to 8'h00. So e.g.
`$display(":%s:", "")` should print ": :" since the width of the empty
string is 1 byte and the %s modifier prints a string with the width of the
value, left-padding with spaces if necessary. The current implementation
will print "::" though. This change requires to update the marco_with_args
gold file.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
Cary R
Lars-Peter Clausen
Stephen Williams
gatk555
Martin Whitaker
Marlon James
DeflateAwning
mole99
Mason Loyet
Ethan Sifferman