views:

185

answers:

4

I have translated the following code using ternary. However, I knew there was something wrong with it. Can someone please point me into the right direction?

ForwardA = 0;
ForwardB = 0;

//EX Hazard
if (EXMEMRegWrite == 1) begin
 if (EXMEMrd != 0)
    if (EXMEMrd == IDEXrs)
        ForwardA = 2'b10;
   if (EXMEMrd == IDEXrt && IDEXTest == 0)
        ForwardB = 2'b10;
end


//MEM Hazard

if (MEMWBRegWrite == 1) begin
 if (MEMWBrd != 0) begin
    if (!(EXMEMRegWrite == 1 && EXMEMrd != 0 && (EXMEMrd == IDEXrs)))
            if (MEMWBrd == IDEXrs)
                ForwardA = 2'b01;
    if (IDEXTest == 0) begin
        if (!(EXMEMRegWrite == 1 && EXMEMrd != 0 && (EXMEMrd == IDEXrt)))
            if (MEMWBrd == IDEXrt)
                ForwardB = 2'b01;
    end
 end
end



ForwardA = (MEMWBRegWrite && MEMWBrd != 0 && (!(EXMEMRegWrite == 1 && EXMEMrd != 0 && (EXMEMrd == IDEXrs))) && (MEMWBrd == IDEXrs)) ?  
                                    2'b01 : ((EXMEMRegWrite && EXMEMrd != 0 && EXMEMrd == IDEXrs) ? 2'b10 : 0);

ForwardB = (IDEXTest == 0 && MEMWBRegWrite && MEMWBrd != 0 && (!(EXMEMRegWrite == 1 && EXMEMrd != 0 && (EXMEMrd == IDEXrt))) && (MEMWBrd == IDEXrs)) ?  
                                    2'b01 : ((EXMEMRegWrite && EXMEMrd != 0 && EXMEMrd == IDEXrt && IDEXTest == 0) ? 2'b10 : 0);
+10  A: 

Surprisingly enough, I'm going to risk downvotes and tell you that the right direction is to leave your code in its relatively readable state.

I suspect the only thing you could do that would be worse would be to do it as a regular expression or convert it to inline assembly :-)

The fact that it's not converting easily should tell you something about the wisdom in what you're attempting.


Based on your comment elsewhere:

This is verilog and therefore I need to do it in ternary and can't have an if else, otherwise I would need an always block before and I don't want that... I want the remaining to be 0 if none of the conditions in the if else above is satisfied

Well, if you must do it, against my advice (and I'm not alone here in offering this advice), here's the method you should use (I have no idea what an "always block" even is so I'm not qualified to argue the point with you).

Since your current code is setting ForwardA and ForwardB to values then only changing them under certain conditions, you can transform that into a ternary by reversing the order. That's because, in your if version, later code takes precedence but earlier code takes precedence in the ternary.

Find out under what circumstances ForwardA and ForwardB are set in reverse order and reconstruct those conditions.

Here's your original code, compressed a bit. I've also changed your 2'b10 things into 2'b10' so we still get nice formatting in the SO rendering engine - don't forget to change them back.

ForwardA = 0;
ForwardB = 0;
if (EXMEMRegWrite == 1) begin
 if (EXMEMrd != 0)
    if (EXMEMrd == IDEXrs)
        ForwardA = 2'b10';
   if (EXMEMrd == IDEXrt && IDEXTest == 0)
        ForwardB = 2'b10';
end
if (MEMWBRegWrite == 1) begin
 if (MEMWBrd != 0) begin
    if (!(EXMEMRegWrite == 1 && EXMEMrd != 0 && (EXMEMrd == IDEXrs)))
            if (MEMWBrd == IDEXrs)
                ForwardA = 2'b01';
    if (IDEXTest == 0) begin
        if (!(EXMEMRegWrite == 1 && EXMEMrd != 0 && (EXMEMrd == IDEXrt)))
            if (MEMWBrd == IDEXrt)
                ForwardB = 2'b01';
    end
 end
end

You can see B is set in three places. It's set to 2'b01 in the bottom if, 2'b10 in the top one and 0 at the start. Converting the conditions:

ForwardB = ((MEMWBRegWrite == 1) &&
            (MEMWBrd != 0) &&
            (IDEXTest == 0) &&
            (!(EXMEMRegWrite == 1 && EXMEMrd != 0 && (EXMEMrd == IDEXrt))) &&
            (MEMWBrd == IDEXrt))
            ? 2'b01'
            : ((EXMEMRegWrite == 1) &&
               (EXMEMrd != 0) &&
               (EXMEMrd == IDEXrt && IDEXTest == 0))
               ? 2'b10'
               : 0;

Similarly for A:

ForwardA = ((MEMWBRegWrite == 1) &&
            (MEMWBrd != 0) &&
            (!(EXMEMRegWrite == 1 && EXMEMrd != 0 && (EXMEMrd == IDEXrs))) &&
            (MEMWBrd == IDEXrs))
            ? 2'b01'
            : ((EXMEMRegWrite == 1) &&
               (EXMEMrd != 0) &&
               (EXMEMrd == IDEXrs))
               ? 2'b10'
               : 0;

Now the theory behind that is good but I wouldn't be the least bit surprised if I'd made an error in the transcription, or if Verilog just threw its hands up in disgust, picked up its ball, and trotted off home :-)

Can I at least suggest, if you must follow this path, you both:

  • try to leave the ternary expressions at least a little readable, with all that nice white space and multiple lines; and
  • keep the original code in a comment so at least you can go back to it if you have problems or want to change the logic?

Seriously, you'll thank me in six months time when you're looking over this again, trying to figure out what on Earth you were thinking :-)

paxdiablo
Why would someone downvote you for saying what SHOULD be said?
Paul
There's always a risk - I'm not providing the answer requested after all. However, since I've been in the industry since the early 80s, I've become quite adept at telling people when they're wrong, though rather more tactfully now than I used to :-)
paxdiablo
You're absolutely right. There's no point in converting that with ternaries. It doesn't give better code and it's less readable. The fact that the OP made a error in transcripting it and not being able to correct it, is a good indicator he shouldn't do it.
tristopia
+1  A: 

Well, assuming that you insist on keeping it in ternary form for whatever reason, your readability would go up considerably if you'd just format it correctly.

const bool cond1 = MEMWBRegWrite && MEMWBrd != 0 &&
                   !(EXMEMRegWrite == 1 && EXMEMrd != 0 && EXMEMrd == IDEXrs) &&
                   MEMWBrd == IDEXrs;
ForwardA = cond1
           ? 2'b01
           : ((EXMEMRegWrite && EXMEMrd != 0 && EXMEMrd == IDEXrs) ? 2'b10 : 0);

const bool cond2 = IDEXTest == 0 &&
                   MEMWBRegWrite && MEMWBrd != 0 &&
                   !(EXMEMRegWrite == 1 && EXMEMrd != 0 && EXMEMrd == IDEXrt) &&
                   MEMWBrd == IDEXrs;
ForwardB =  cond2
            ? 2'b01
            : ((EXMEMRegWrite && EXMEMrd != 0 && EXMEMrd == IDEXrt && IDEXTest == 0) ? 2'b10 : 0);

Now, that code is formatted as if it were C++ rather than whatever you're actually using, but it becomes much easier to figure out what's going on.

However, I would point out that your if-statements can't possibly match your ternary expressions. Your if statements have no else clause, and ternary expressions always have else clauses. However, since your question doesn't even make it entirely clear whether you're trying to convert the if-statements into ternary expressions or the ternary expressions into if-statements, it's a bit hard to give you exactly what you want.

EDIT: Ternary expressions always have both an if and an else clause. You cannot directly turn an if statement without an else clause into a ternary because you wouldn't have the else portion of the ternary. Now, you can pull some tricks in some cases if you need to, like setting a variable to itself. For instance,

ForwardA = cond1 ? newValue : FordwardA;

You're basically saying not to change the value in the else clause - but that's assuming that you're assigning the result to a variable. The more complicated the expression, the harder it is to pull that sort of trick, and the more convoluted the code becomes when you do. Not to mention, depending on what optimizations that the compiler does or doesn't do, it could be assigning the variable to itself, which isn't terribly efficient.

Generally-speaking, translating if-statements with no else clauses into ternary expressions is a bad idea. It can only be done by pulling tricks rather than directly saying what you mean, and it just complicates things. And this code is complicated enough as it is.

I'd advise not using a ternary here unless you really need it. And if you do, at least break down the expression. Even if your ternary expression were correct, it's much harder to read than the if-statements.

EDIT 2: If you really do need this to be a ternary expression, then I'd advise that you sit down and figure out the exact conditions under which ForwardA should be what set of values and create a ternary expression based on that rather than trying to directly convert the if-statements that you have (and the same for ForwardB). Your if-statments are not only deciding what value to assign to each variable, but which variable to assign that value to, and that complicates things considerably.

In other languages (I don't know about verilog), you can use a ternary expression for choosing which variable to assign the value to in addition to whatever you're doing on the right side of the expression, but that's getting really complicated. It might be best to create a temporary which holds the value which is to be assigned and a separate ternary to determine which variable to assign it to.

Not knowing verilog, I really don't know what you can and can't do with if-statements and ternary expression, but I would think that there's got to be a better way to handle this than using a ternary. Maybe not, but what you're trying to do is very difficult and error-prone.

Jonathan M Davis
I am actually converting from the if else to ternary... so are you saying it's impossible to have the if else break down into ternary in this case?
EquinoX
I managed to figure out the problem where MEMWBrd == IDEXrs; should be MEMWBrd == IDEXrt; for forward B
EquinoX
+1  A: 

First don't do it! there's no point, in doing so. It doesn't compile to better code and is less readable, as you noticed in your tries to correct it. If you need it as an expression it would be better to code it as an inline function.

tristopia
+1  A: 

You don't need to do this. Stick the code in an 'always @*' block, and declare anything you're assigning to as 'reg'.

reg [1:0] ForwardA;
reg [1:0] ForwardB;

always @(*) begin
   // Your combo logic here..
end
Marty