Based on ensembl's hg38:

$ parallel -j8 "samtools faidx Homo_sapiens.GRCh38.dna.primary_assembly.fa {2} \
| seqkit seq -w0 \
| tail -n+2 \
| LC_ALL=C grep -io  {1} \
| wc -l \
| awk -v kmer={1} '{print kmer,\$0}'" ::: `echo {C,A,G,T}{C,A,G,T}{C,A,G,T}|tr " " "\n"` ::: {1..22} X Y \
| awk -v OFS="\t" 'BEGIN {print "kmer", "count"} {kmer[$1] += $2} END {for (k in kmer) {print k,kmer[k]}}' \
| sort -k1 > kmer_counts.tsv

Please notice that grep matches are not overlapping. This means in case of homopolymer stretches like TTTTTT, this will be count as 2 and not 4.