fix #20997 calling system.card[T](x: set[T]) with T of int8 or uint8 … (#21010)

* fix #20997 calling system.card[T](x: set[T]) with T of int8 or uint8 uses mismatched C array sizes
* fullfil set variant

compiler/ccgexprs.nim

@@ -2126,7 +2126,7 @@ proc genSetOp(p: BProc, e: PNode, d: var TLoc, op: TMagic) =
     of mCard:
       var a: TLoc
       initLocExpr(p, e[1], a)
-      putIntoDest(p, d, e, ropecg(p.module, "#cardSet($1, $2)", [rdCharLoc(a), size]))
+      putIntoDest(p, d, e, ropecg(p.module, "#cardSet($1, $2)", [addrLoc(p.config, a), size]))
     of mLtSet, mLeSet:
       getTemp(p, getSysType(p.module.g.graph, unknownLineInfo, tyInt), i) # our counter
       initLocExpr(p, e[1], a)

lib/system/sets.nim

@@ -10,10 +10,9 @@
 # set handling
 
 type
-  NimSet = array[0..4*2048-1, uint8]
+  NimSet = array[0..8192-1, uint8]
 
-
-proc cardSet(s: NimSet, len: int): int {.compilerproc, inline.} =
+proc cardSetImpl(s: openArray[uint8], len: int): int {.inline.} =
   var i = 0
   result = 0
   when defined(x86) or defined(amd64):
@@ -24,3 +23,6 @@ proc cardSet(s: NimSet, len: int): int {.compilerproc, inline.} =
   while i < len:
     inc(result, countBits32(uint32(s[i])))
     inc(i, 1)
+
+proc cardSet(s: NimSet, len: int): int {.compilerproc, inline.} =
+  result = cardSetImpl(s, len)

tests/sets/t20997.nim

@@ -0,0 +1,18 @@
+discard """
+  joinable: false
+"""
+
+{.passC: "-flto".}
+{.passL: "-flto".}
+
+template f(n: int) = discard card(default(set[range[0 .. (1 shl n) - 1]]))
+f( 7)
+f( 8)
+f( 9)
+f(10)
+f(11)
+f(12)
+f(13)
+f(14)
+f(15)
+f(16)