FWIW, the Nim version is (no deps besides stdlib): import os, strutils, tables, heapqueue iterator topByVal*[K, V](c: Table[K, V], n = 10, min = V.low): (

FWIW, the Nim version is (no deps besides stdlib): import os, strutils, tables, heapqueue iterator topByVal*[K, V](c: Table[K, V], n = 10, min = V.low): (K, V) = var q = initHeapQueue[(V, K)]() for key, val in c: if val >= min: let elem = (val, key) if q.len < n: q.push(elem) elif elem > q[0]: discard q.replace(elem) var y: (K, V) # yielded tuple while q.len > 0: # q now has top n entries let r = q.pop y[0] = r[1] y[1] = r[0] yield y # yield in ASCENDING order template maybeCount(count, word: untyped) = if word.len > 1: count.mgetOrPut(word, 0).inc proc main() = var count = initTable[string, uint]() for path in ".".walkDirRec: if not path.endsWith(".txt"): continue for line in path.lines: var word = "" for c in line.toLower: if c in {'a'..'z'}: word.add c else: count.maybeCount word if word.len > 0: word.setLen 0 count.maybeCount word # line ends in a word for word, cnt in count.topByVal: echo cnt, " ", word main() Doing a gcc-10 profile-guided-optimization build on this, I get the min time of 5 trials running on Charles Dickens' Tale Of Two Cities { opening paragraph is very optimization claims relevant! :-) } available here: http://www.textfiles.com/etext/AUTHORS/DICKENS/dickens-tale-126.txt http://www.textfiles.com/etext/AUTHORS/DICKENS/dickens-tale-... 7.3 ms EDIT- tptacek's C version (w/bug-fix & new tokenization) 15.8 ms Above Nim 1.4 PGO --gc:arc 24.8 ms Above Nim 1.4 --gc:arc 27.3 ms Original C++ PGO 30.2 ms Original C++ -O3 only 42.6 ms dga's Rust with rustc-1.47.0 build --release This is on a i7-6700k. So, Skylake core and probably more importantly 8 MiB L3. So, also FWIW, I could not reproduce the Rust/C++ timing ratios with the above mentioned input file. I would also point out that not only is globwalk unnecessary as @burntsushi points out, neither are regexes nor even sorting. A heap is a better way to do "top N", but you may need to reverse the answer if you care. I mean, maybe the point of the original C++ is to also bench/exhibit use of regex engines and part of the Rust point is utf8 or some such, but to me that seems more like a distraction. Anyway, maybe my rust build environment is screwy. Hence, my inclusion of an exact input .txt file for others to try out if they care. I had to add some extern crates and [dependencies] of anyhow, lazysort, regex, globwalk. I've never done a PGO build with rust, either. So, that may help.