I need an example on how to paramtrize Vector of registers in terms of bit-width and initial values which are not '0' and are different for each register.
My use-case is a generic filter coefficients bank with some unique reset values to each, and off course an option to override values.
I thought of something like the below code (not really sure how to write the iteration, so this is kind of pseudo):
class Coeffbank(bitWidth : UInt ,ncoeff : UInt, rstVal : Vec(SInt)) extends Module {
// how do iterate through the reset vector ?? //
val coeffs = Vec.fill(ncoeff) {Reg(init = SInt(rstVal(i),width = bitwidth))
}
Also, when new'ing the above (instantiating this module how do I pass the list of reset value in the argument list?
Hoping to get some help on how to write it properly.
The explanation should probably be a bit more thorough, but basically you need to create a Reg of Vec. Something like should do it:
val coeffs = RegInit(rstVal)
In this case, since you already have the Vec of reset values, you can just pass it to the Reg constructor.
I'm assuming that the size of rstVal is equal to ncoeff, otherwise you'll need to reduce the size of rstVal with something like rstVal.take(ncoeff). Also note that I'm using RegInit which is the preferred way to create a register with a reset value.
Let's start with the easy case. This would be much easier if instead of a Vec of SInts your rstVal array was instead a scala collection (Seq, Array, ...) of regular SInt. When possible it is best to save generation of actual hardware until you directly need them. If rstVal contains Int's. Your code would become
val newRstVals = VecInit(Seq.tabulate(ncoeff) { index => rstVals(index).S(bitWidth.W) })
val reg = RegInit(newRstVals)
If you really need to pass in a Vec then the right approach is to create a separate type instance and use the two argument call to RegInit
val vecType = Vec(ncoeff, SInt(bitWidth.W))
val newRstVals1 = VecInit(Seq.tabulate(ncoeff) { index => newRstVals(index) })
val reg = RegInit(vecType, newRstVals1)
There might be problems if the bitWidth you pass in is not big enough to contain the constants you have passed in. You probably should have some checks for that.
Related
I have, tab=Array(1.U, 6.U, 5.U, 2.U, 4.U, 3.U) and Y=Seq(b,g,g,g,b,g), tab is an array of UInt.
I want to do a map on tab as follows:
tab.map(case idx=>Y(idx))
But I keep getting the error: found chisel3.core.UInt, required Int.
I tried using the function peek() to convert idx to an Int by doing
tab.map(case idx=>Y(peek(idx).toInt)
but I get peek not found. I also saw that I cannot convert a chisel UInt to an Int here but did not understand the use of peek well with the example given. So please, is there another approach to do the above?
Thanks!
The immediate problem is that you cannot access the elements of scala collections using a hardware construct like UInt or SInt. It should work if you wrap Y in a Vec. Depending on your overall module this would probably look like
val YVec = VecInit(Y)
val mappedY = tab.map { case idx => YVec(idx) }
Is there any way to initialize a Sequential value not in one fellow swoop?
Like, can I declare it, then use a for loop to populate it, step by step?
As this could all happen inside a class body, the true immutability of the Sequential value could then kick in once the class instance construction phase has been completed.
Example:
Sequential<String> strSeq;
for (i in span(0,10)) {
strSeq[i] = "hello";
}
This code doesn't work, as I get this error:
Error:(12, 9) ceylon: illegal receiving type for index expression:
'Sequential' is not a subtype of 'KeyedCorrespondenceMutator' or
'IndexedCorrespondenceMutator'
So what I can conclude is that sequences must be assigned in one statement, right?
Yes, several language guarantees hinge on the immutability of sequential objects, so that immutability must be guaranteed by the language – it can’t just trust you that you won’t mutate it after the initialization is done :)
Typically, what you do in this situation is construct some sort of collection (e. g. an ArrayList from ceylon.collection), mutate it however you want, and then take its .sequence() when you’re done.
Your specific case can also be written as a comprehension in a sequential literal:
String[] strSeq = [for (i in 0..10) "hello"];
The square brackets used to create a sequence literal accept not only a comma-separated list of values, but also a for-comprehension:
String[] strSeq = [for (i in 0..10) "hello"];
You can also do both at the same time, as long as the for-comprehension comes last:
String[] strSeq = ["hello", "hello", for (i in 0..8) "hello"];
In this specific case, you could also do this:
String[] strSeq = ["hello"].repeat(11);
You can also get a sequence of sequences via nesting:
String[][] strSeqSeq = [for (i in 0..2) [for (j in 0..2) "hello"]];
And you can do the cartesian product (notice that the nested for-comprehension here isn't in square brackets):
[Integer, Character][] pairs = [for (i in 0..2) for (j in "abc") [i, j]];
Foo[] is an abbreviation for Sequential<Foo>, and x..y translates to span(x, y).
If you know upfront the size of the sequence you want to create, then a very efficient way is to use an Array:
value array = Array.ofSize(11, "");
for (i in 0:11) {
array[i] = "hello";
}
String[] strSeq = array.sequence();
On the other hand, if you don't know the size upfront, then, as described by Lucas, you need to use either:
a comprehension, or
some sort of growable array, like ArrayList.
I have defined a register of vectors like this
val my_reg = Reg(Vec(n, Bits(32.W)))
and I access the elements of this register in a for loop using my_reg(i).
Now, I like to initialize this register to zero, so I change the variable definition to this
val my_reg = Reg(Vec(n, Bits(32.W)), init = UInt(0))
However, I get the following compilation error when I want to access the elements of this register
chisel3.core.Data does not take parameters
my_reg(i) := io.a(i)
How can I define a register of vectors and properly initialize them synchronously?
Use RegInit instead. I believe the following statement will do what you want
val my_reg = RegInit(Vec(Seq.fill(n)(0.U(32.W))))
The Vector is initialized by a Seq of UInt zeros that are 32 bits wide
Looks like in Chisel 3 this should be VecInit instead of Vec:
val initRegOfVec = RegInit(VecInit(Seq.fill(4)(0.U(32.W))))
Source: https://github.com/freechipsproject/chisel3/wiki/Cookbook#how-do-i-create-a-reg-of-type-vec
I'm learning Programming Paradigms in my University and reading this course material provided by the lecturer that defined a function this way:
val double = (x: Int) => 2 * x
double: Int => Int = <function1>
But from my own studies I found and got used to defining the same function like this:
def d (x: Int) = 2 * x
d: (x: Int)Int
I'm new to Scala. And both definitions give a result of:
res21: Int = 8
Upon passing 4 as the parameter.
Now my main question is why would the lecturer prefer to use val to define a function? I see it as longer and not really necessary unless using val gives some added advantages that I don't know of. Besides I understand using val makes some name a placeholder so later in the program, I could mistakenly write val double = 5 and the function would be gone!
At this stage I'm quite convinced I learned a better way of defining a function unless someone would tell me otherwise.
Strictly speaking def d (x: Int) = 2 * x is a method, not a Function, however scala can transparently convert (lift) methods into Functions for us. So that means you can use the d method anywhere that requires a Int => Int Function.
There is a small overhead of performing this conversion, as a new Function instance is created every time. We can see this happening here:
val double = (x: Int) => 2 * x
def d (x: Int) = 2 * x
def printFunc(f: Int => Int) = println(f.hashCode())
printFunc(double)
printFunc(double)
printFunc(d)
printFunc(d)
Which results in output like so:
1477986427
1477986427
574533740
1102091268
You can see when explicitly defining a Function using a val, our program only creates a single Function and reuses it when we pass as an argument to printFunc (we see the same hash code). When we use a def, the conversion to a Function happens every time we pass it to printFunc and we create several instances of the Function with different hash codes. Try it
That said, the performance overhead is small and often doesn't make any real difference to our program, so defs are often used to define Functions as many people find them more concise and easier to read.
In Scala, function values are monomorphic (i.e. they can not have type parameters, aka "generics"). If you want a polymorphic function, you have to work around this, for example by defining it using a method:
def headOption[A]: List[A] => Option[A] = {
case Nil => None
case x::xs => Some(x)
}
It would not be valid syntax to write val headOption[A]. Note that this didn't make a polymorphic function value, it is just a polymorphic method, returning a monomorphic function value of the appropriate type.
Because you might have something like the following:
abstract class BaseClass {
val intToIntFunc: Int => Int
}
class A extends BaseClass {
override val intToIntFunc = (i: Int) => i * 2
}
So its purpose might not be obvious with a very simple example. But that Function value could itself be passed to higher order functions: functions that take functions as parameters. If you look in the Scala collections documentation you will see numerous methods that take functions as parameters. Its a very powerful and versatile tool, but you need to get to a certain complexity and familiarity with algorithms before the cost /benefit becomes obvious.
I would also suggest not using "double" as an identifier name. Although legal Scala, it is easy to confuse it with the type Double.
I will run a set of experiments. The main method evaluated has the following signature:
[Model threshold] = detect(...
TrainNeg, TrainPos, nf, nT, factors, ...
removeEachStage, applyEstEachStage, removeFeatures);
where removeEachStage, applyEstEachStage, and removeFeatures are booleans. You can see that if I reverse the order of any of these boolean parameters I may get wrong results.
Is there a method in MATLAB that allows better organization in order to minimize this kind of error? Or is there any tool I can use to protect me against these errors?
Organization with a struct
You could input a struct that has these parameters as it's fields.
For example a structure with fields
setts.TrainNeg
.TrainPos
.nf
.nT
.factors
.removeEachStage
.applyEstEachStage
.removeFeatures
That way when you set the fields it is clear what the field is, unlike a function call where you have to remember the order of the parameters.
Then your function call becomes
[Model threshold] = detect(setts);
and your function definition would be something like
function [model, threshold] = detect(setts)
Then simply replace the occurrences of e.g. param with setts.param.
Mixed approach
You can also mix this approach with your current one if you prefer, e.g.
[Model threshold] = detect(in1, in2, setts);
if you wanted to still explicitly include in1 and in2, and bundle the rest into setts.
OOP approach
Another option is to turn detect into a class. The benefit to this is that a detect object would then have member variables with fixed names, as opposed to structs where if you make a typo when setting a field you just create a new field with the misspelled name.
For example
classdef detect()
properties
TrainNeg = [];
TrainPos = [];
nf = [];
nT = [];
factors = [];
removeEachStage = [];
applyEstEachStage = [];
removeFeatures =[];
end
methods
function run(self)
% Put the old detect code in here, use e.g. self.TrainNeg to access member variables (aka properties)
end
end