Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

proc_dff: process sync rules in reverse input order #4568

Closed
wants to merge 2 commits into from

Conversation

widlarizer
Copy link
Collaborator

@widlarizer widlarizer commented Aug 27, 2024

What are the reasons/motivation for this change?

It's assumed that sync rules are in some kind of priority ordering. See here

Explain how this is achieved.

Since there is no "insertion ordered map" in STL, I provide a minimal one of my own

If applicable, please suggest to reviewers how they can test the change.

The tests now pass including dfflibmap.ys even when our hash function is modified as in #4559 and #4524. More investigation is needed about proc_dff and into this particular test and its correctness

@whitequark
Copy link
Member

I wonder if this should be clarified in the RTLIL guide in the manual.

@widlarizer
Copy link
Collaborator Author

I think so, I checked docs/source/yosys_internals/formats/rtlil_rep.rst and it's not described there. I'd describe it if I had sufficient faith in my mental model.

Also as @georgerennie pointed out the reverse order would be the correct one, the CI only confirms this

@widlarizer widlarizer marked this pull request as ready for review August 27, 2024 20:16
@widlarizer widlarizer changed the title proc_dff: process sync rules in input order proc_dff: process sync rules in reverse input order Aug 27, 2024
@georgerennie
Copy link
Contributor

georgerennie commented Aug 27, 2024

I discussed this with @widlarizer by dm, but I think this solution doesn't fully capture the priority for the sync rules. As the syncrules are still being aggregated by the assign expression for their condition, later syncrules automatically get the same priority as earlier syncrules that assign the same value, rather than their correct priority. I assume this is meant to be an optimization to merge together conditions leading to the same value being loaded, but some more care needs to be taken to preserve priorities.

Consider the following example

read_verilog <<EOT
module dffsr(input CLK, D, A, B, C, output reg Q);

always @(posedge CLK, posedge A, posedge B, posedge C)
	if (A)      Q <= '0;
	else if (B) Q <= '1;
	else if (C) Q <= '0;
	else        Q <= D;
endmodule
EOT

proc -noopt

opt -mux_bool
write_verilog -noattr

This produces

assign _0_ = | { C, A };
assign _1_ = _0_ ? 1'h0 : B;
always @(posedge CLK, posedge _1_, posedge _0_)
  if (_0_) Q <= 1'b0;
  else if (_1_) Q <= 1'b1;
  else Q <= D;

The 0 case has taken the trigger |{C, A}, whereas really it should be something like |{A, C && !B). This means that if the inputs A = 0, B = 1, C = 1 are applied, in the first circuit Q = 1 whereas in the processed version Q = 0

I think a possible solution is to process the sync rules one by one in the order they appear from proc_arst, guarding each condition with all the previous sync rule conditions, and taking the disjunction of conditions that lead to the same value.

With this, you would therefore get for each sync rule A -> Q <= '0, B && !A -> Q <= '1, C && !B && !A -> Q <= '0. You can then merge the conditions leading to zero, so you get |{A, C && !B && !A} -> Q <= '0 and B && !A -> Q <= '1.

@widlarizer
Copy link
Collaborator Author

Superseded by #4569

@widlarizer widlarizer closed this Aug 28, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants