Prefix instance and register names with the given Symbol

Suffix instance and register names with the given Symbol

Suffix instance and register names with the given Symbol, but add it in front of other suffixes.

When you write

suffixName @"A" (suffixName @"B" f))

you get register and instance names inside f with the suffix: "_B_A"

However, if you want them in the other order you can write:

suffixNameP @"A" (suffixName @"B" f))

so that names inside f will have the suffix "_A_B"

Suffix instance and register names with the given Nat

Suffix instance and register names with the given Nat, but add it in front of other suffixes.

When you write

suffixNameNat @1 (suffixName @"B" f))

you get register and instance names inside f with the suffix: "_B_1"

However, if you want them in the other order you can write:

suffixNameNatP @1 (suffixName @"B" f))

so that names inside f will have the suffix "_1_B"

Name the instance or register with the given Symbol, instead of using an auto-generated name. Pre- and suffixes annotated with prefixName and suffixName will be added to both instances and registers named with setName and instances and registers that are auto-named.

Name a given term, such as one of type Signal, using the given SSymbol. Results in a declaration with the name used as the identifier in the generated HDL code.

Example usage:

nameHint (SSymbol @"identifier") term

NB: The given name should be considered a hint as it may be expanded, e.g. if it collides with existing identifiers.

Force deduplication, i.e. share a function or operator between multiple branches.

By default Clash converts

case x of
  A -> 3 * y
  B -> x * x

to

let f_arg0 = case x of {A -> 3; _ -> x}
    f_arg1 = case x of {A -> y; _ -> x}
    f_out  = f_arg0 * f_arg1
in  case x of
      A -> f_out
      B -> f_out

However, it won't do this for:

case x of
  A -> 3 + y
  B -> x + x

Because according to the internal heuristics the multiplexer introduced for the deduplication are more expensive than the addition. This might not be the case for your particular platform.

In these cases you can force Clash to deduplicate by:

case x of
  A -> deDup (3 + y)
  B -> deDup (x + x)

Do not deduplicate, i.e. keep, an applied function inside a case-alternative; do not try to share the function between multiple branches.

By default Clash converts

case x of
  A -> f 3 y
  B -> f x x
  C -> h x

to

let f_arg0 = case x of {A -> 3; _ -> x}
    f_arg1 = case x of {A -> y; _ -> x}
    f_out  = f f_arg0 f_arg1
in  case x of
      A -> f_out
      B -> f_out
      C -> h x

i.e. it deduplicates functions (and operators such as multiplication) between case-alternatives to save on area. This comes at the cost of multiplexing the arguments for the deduplicated function.

There are two reasons you would want to stop Clash from doing this:

1. The deduplicated function is in the critical path, and the addition of the multiplexers further increased the propagation delay.
2. Clash's heuristics were off, and the addition of the multiplexers actually made the final circuit larger instead of smaller.

In these cases you want to tell Clash not to deduplicate:

case x of
  A -> noDeDup f 3 y
  B -> f x x
  C -> h x

Where the application of f in the A-alternative is now explicitly not deduplicated, and given that the f in the B-alternative is the only remaining application of f in the case-expression it is also not deduplicated.

Note that if the C-alternative also had an application of f, then the applications of f in the B- and C-alternatives would have been deduplicated; i.e. the final circuit would have had two application of f.