🐛 "No module matching user provided call signature" on RevLib File Format doc example

[Geremia]

Environment information

This example from "RevLib – File Format Documentation" p. 7:

module counter(in reset(1), in enable(1), inout jump, inout countval)
  wire zero
  if ( reset = 1) then
    countval <=> zero
  else
    if ( enable = 1 ) then
      countval <=> jump
    else
      ++=countval
    fi ( enable = 1 )
  fi ( reset = 1 )

module main( in reset(1), in enable(1), in jump)
  state countval
  call counter(reset, enable, jump, countval)

gives this error:

-- line 15 col 7: No module matching user provided call signature

on the call counter(reset, enable, jump, countval) line.

Description

OS: Linux
C++ compiler: GCC 15.1.0
mqt.core version: 3.0.2

How to Reproduce

Run error = prog.read(syrec_code_file) and examine the output error.

[burgholzer]

Could you try with #220 and see if the issue still persists?

[TooMuchDakka]

@Geremia From your provided error message i am assuming that your are using the parser of #220 and will base my answer on that implementation in which we assume that module parameters of type in are read-only (a property that is not defined in the specification of SyReC in RevLib).

Thus, the signature of the module counter defined as a C++ function would be equal to void counter(const bool reset, const bool enable, bool& jump, bool& countval) that is then called in the main module where the read-only variable jump of type in is passed to the modifiable parameter jump of the module counter. Due to our intial assumption about variables of type in such a call is not allowed but whether this assumption is actually correct still needs to be decided since it did not exist in the previous parser implementation.

[burgholzer]

@Geremia From your provided error message i am assuming that your are using the parser of #220 and will base my answer on that implementation in which we assume that module parameters of type in are read-only (a property that is not defined in the specification of SyReC in RevLib).

Thus, the signature of the module counter defined as a C++ function would be equal to void counter(const bool reset, const bool enable, bool& jump, bool& countval) that is then called in the main module where the read-only variable jump of type in is passed to the modifiable parameter jump of the module counter. Due to our intial assumption about variables of type in such a call is not allowed but whether this assumption is actually correct still needs to be decided since it did not exist in the previous parser implementation.

Just a quick side comment here:
I think that assumption (in parameters being read-only) is reasonable.
The parameter would need to be defined inout instead.

[Geremia]

@burgholzer @TooMuchDakka Yes, I'm using #220 .

[Geremia]

@burgholzer Changing everything to inout still results in the same error.

[Geremia]

@burgholzer

The parameter would need to be defined inout instead.

Removing state, which appears to be const as well, fixes it:

module main( in reset(1), in enable(1), inout jump, inout countval)
  call counter(reset, enable, jump, countval)

(Changing state countval to wire countval is another way of correcting the error.)

So the question is: Why can't I pass a state variable to an inout argument of a function?

[TooMuchDakka]

@burgholzer

The parameter would need to be defined inout instead.

Removing state, which appears to be const as well, fixes it:

module main( in reset(1), in enable(1), inout jump, inout countval)
  call counter(reset, enable, jump, countval)

(Changing state countval to wire countval is another way of correcting the error.)

So the question is: Why can't I pass a state variable to an inout argument of a function?

In the PDF-version of the paper "SyReC: a hardware description language for the specification and synthesis of reversible circuits", listed in the publications section of the mqt-syrec documentation, a small footnote on page 42 regarding variables of type state is important for this issue:

"Note that depending on the application feedback and, hence, state signals might not be allowed in reversible circuits.Nevertheless, SyReC supports this concept in principle..."

Additionally, the RevLib specification states in section 4.4 "According to the basic definitions, feedback is not directly allowed in reversible circuits. Nevertheless, first steps towards the realization of sequential reversible circuits have been performed"

1. I am not sure whether the current synthesis algorithms handle variables of type state correctly.
2. Preliminary support for variables of type state was added to the parser in ♻️ Replacement of boost::fusion and boost::spirit with ANTLR4 components for SyReC grammar and parser #220 with the restriction that variables of this type are read-only due to neither the initially mentioned "SyReC" paper nor RevLib specifying whether assignments to variables of this type should be possible. Combined with 1. I think that this is a reasonable assumption for now. Thus variables of type state can only be passed to module parameters of type in which in turn explains why no matching module for your provided arguments in the call statement was found.

[Geremia]

@TooMuchDakka
Using

mqt-core                       3.0.3.dev94+ga06d0fdf
mqt.syrec                      1.2.2.dev26+gd192282

to process the *.src file above gives this error now:

Error reading syrec code file
In line 0: Unknown variable %szero

It trips up on the wire.

[TooMuchDakka]

@TooMuchDakka Using

mqt-core                       3.0.3.dev94+ga06d0fdf
mqt.syrec                      1.2.2.dev26+gd192282

to process the *.src file above gives this error now:

Error reading syrec code file
In line 0: Unknown variable %szero

It trips up on the wire.

Based on the format of the error message your build of mqt.syrec still uses the "old" boost parser. I had to perform a reinstall of the built from source package in my local virtual environment so that the "new" parser is used (with which your provided SyReC program could be built assuming that all semantic errors were fixed) but that could be due to my installation from source using pip not clearing the previous installation correctly.

[Geremia]

@TooMuchDakka The #270 merge fixed this problem.

[Geremia]

@TooMuchDakka I'm getting a regression on this issue using

mqt-core            3.2.2.dev110+g41b1d34d3
mqt.syrec           1.2.2.dev149+g89940586e

The sample SyReC code in my OP above gives this again:

Error reading syrec code file
-- line 15 col 7: No module matching user provided call signature

[TooMuchDakka]

I am assuming that you are using the version of your SyReC code in which the issue with the state variable as well as the assignment of the readonly parameter jump of the main module to the modifiable jump parameter of the counter module was resolved, i.e. the SyReC program that should be synthesized is equal to:

module counter(in reset(1), in enable(1), inout jump, inout countval)
  wire zero
  if ( reset = 1) then
    countval <=> zero
  else
    if ( enable = 1 ) then
      countval <=> jump
    else
      ++=countval
    fi ( enable = 1 )
  fi ( reset = 1 )

module main( in reset(1), in enable(1), inout jump, inout countval)
  call counter(reset, enable, jump, countval)

I have perform a rebuild from source for the most recent commit on the main branch and could synthesize the program without errors. Perhaps your mqt-syrec was built using an 'older' version of the library?