对于早于4.0的Bash版本，该版本应该是最快的（因为它不会fork/exec任何命令）：

function string.monolithic.tolower
{
   local __word=$1
   local __len=${#__word}
   local __char
   local __octal
   local __decimal
   local __result

   for (( i=0; i<__len; i++ ))
   do
      __char=${__word:$i:1}
      case "$__char" in
         [A-Z] )
            printf -v __decimal '%d' "'$__char"
            printf -v __octal '%03o' $(( $__decimal ^ 0x20 ))
            printf -v __char \\$__octal
            ;;
      esac
      __result+="$__char"
   done
   REPLY="$__result"
}

技术龙的回答也很有潜力，尽管它确实适合mee。

tr:

a="$(tr [A-Z] [a-z] <<< "$a")"

AWK:

{ print tolower($0) }

sed:

y/ABCDEFGHIJKLMNOPQRSTUVWXYZ/abcdefghijklmnopqrstuvwxyz/

预冲击4.0

Bash降低字符串的大小写并赋值给变量

VARIABLE=$(echo "$VARIABLE" | tr '[:upper:]' '[:lower:]') 

echo "$VARIABLE"

从bash手册页：

$｛参数^模式｝$｛参数^^模式｝$｛参数，模式｝$｛参数，，模式｝案例修改。此扩展修改参数中字母字符的大小写。该模式被扩展以生成模式，就像路径名扩展一样。展开的每个字符根据模式测试参数的值，如果匹配模式，其大小写被转换。该模式不应尝试匹配多个字符。^运算符转换为小写字母匹配模式到大写；运算符转换将大写字母匹配为小写字母。^^和，，展开式转换展开值中的每个匹配字符；这个^并且，展开式仅匹配并转换展开值中的第一个字符。如果省略了模式，则将其视为?, 它匹配每个字符。如果参数为@或*，则大小写修改操作将依次应用于每个位置参数，展开即为结果列表。如果参数是下标为@或*的数组变量，大小写修改操作应用于旋转，展开即为结果列表。

尽管这个问题有多古老，而且与技术龙的答案相似。我很难找到一个能够跨大多数平台（我使用的平台）以及旧版本bash移植的解决方案。我还对数组、函数以及使用prints、echo和临时文件来检索琐碎变量感到失望。到目前为止，这对我来说非常有效，我想我会分享。我的主要测试环境是：

GNUbash，版本4.1.2（1）-发布（x86_64-redhat-linux-GNU）GNUbash，版本3.2.57（1）-发布（sparc-sun-solaris2.10）

lcs="abcdefghijklmnopqrstuvwxyz"
ucs="ABCDEFGHIJKLMNOPQRSTUVWXYZ"
input="Change Me To All Capitals"
for (( i=0; i<"${#input}"; i++ )) ; do :
    for (( j=0; j<"${#lcs}"; j++ )) ; do :
        if [[ "${input:$i:1}" == "${lcs:$j:1}" ]] ; then
            input="${input/${input:$i:1}/${ucs:$j:1}}" 
        fi
    done
done

循环的简单C样式，用于遍历字符串。如果您以前没有看到过类似的内容，请选择下面的行这是我学到的。在这种情况下，该行检查输入中是否存在字符${input:$i:1}（小写），如果存在，则用给定的字符${ucs:$j:1}（大写）替换它，并将其存储回输入中。

input="${input/${input:$i:1}/${ucs:$j:1}}"

因此，我尝试对每个实用程序使用共识方法执行一些更新的基准测试，但我没有多次重复一个小集合，而是。。。

以UTF-8编码的多字节Unicode字符填充到边缘的1.85 GB.txt文件中，为了均衡I/O方面，同时还强制所有人使用LC_ALL=C，以确保公平竞争

————————————————————————————————————————

准确地说，bsd-sed和gnu-sed都相当平庸。我甚至不知道bsd sed在做什么，因为他们的xxhash不匹配python3是否试图使用Unicode字母大小写？（即使我已经强制设置了区域设置LC_ALL=C）tr是最极端的到目前为止，gnutr是最快的bsd tr非常残暴perl5比我拥有的任何awk变体都快，除非你可以使用mawk2一次加载整个文件，以便稍微超过perl5：2.935秒mawk2对每15秒3.081秒在awk中，gnu-gawk的速度最慢，中间是mawk 1.3.4，最快是mawk1.9.9.6：比gawk节省50%以上的时间.（我没有把时间浪费在无用的macosx nawk上）

.

     out9: 1.85GiB 0:00:03 [ 568MiB/s] [ 568MiB/s] [ <=> ]
      in0: 1.85GiB 0:00:03 [ 568MiB/s] [ 568MiB/s] [============>] 100%            
( pvE 0.1 in0 < "${m3t}" | LC_ALL=C mawk2 '{ print tolower($_) }' FS='^$'; )  

mawk 1.9.9.6 (mawk2-beta)

3.07s user 0.66s system 111% cpu 3.348 total
85759a34df874966d096c6529dbfb9d5  stdin


     out9: 1.85GiB 0:00:06 [ 297MiB/s] [ 297MiB/s] [ <=> ]
      in0: 1.85GiB 0:00:06 [ 297MiB/s] [ 297MiB/s] [============>] 100%            
( pvE 0.1 in0 < "${m3t}" | LC_ALL=C mawk '{ print tolower($_) }' FS='^$'; )  

 mawk 1.3.4

6.01s user 0.83s system 107% cpu 6.368 total
85759a34df874966d096c6529dbfb9d5  stdin

     out9: 23.8MiB 0:00:00 [ 238MiB/s] [ 238MiB/s] [ <=> ]
      in0: 1.85GiB 0:00:07 [ 244MiB/s] [ 244MiB/s] [============>] 100%            
     out9: 1.85GiB 0:00:07 [ 244MiB/s] [ 244MiB/s] [ <=>                             ]
( pvE 0.1 in0 < "${m3t}" | LC_ALL=C gawk -be '{ print tolower($_) }' FS='^$';  

GNU Awk 5.1.1, API: 3.1 (GNU MPFR 4.1.0, GNU MP 6.2.1) 

7.49s user 0.78s system 106% cpu 7.763 total
85759a34df874966d096c6529dbfb9d5  stdin


     out9: 1.85GiB 0:00:03 [ 616MiB/s] [ 616MiB/s] [ <=> ]
      in0: 1.85GiB 0:00:03 [ 617MiB/s] [ 617MiB/s] [============>] 100%            
( pvE 0.1 in0 < "${m3t}" | LC_ALL=C perl -ne 'print lc'; )  

perl5 (revision 5 version 34 subversion 0)

2.70s user 0.85s system 115% cpu 3.081 total
85759a34df874966d096c6529dbfb9d5  stdin


     out9: 1.85GiB 0:00:32 [57.4MiB/s] [57.4MiB/s] [ <=> ]
      in0: 1.85GiB 0:00:32 [57.4MiB/s] [57.4MiB/s] [============>] 100%            
( pvE 0.1 in0 < "${m3t}" | LC_ALL=C gsed 's/.*/\L&/'; )  # GNU-sed


gsed (GNU sed) 4.8

32.57s user 0.97s system 101% cpu 32.982 total
85759a34df874966d096c6529dbfb9d5  stdin


     out9: 1.86GiB 0:00:38 [49.7MiB/s] [49.7MiB/s] [ <=> ]
      in0: 1.85GiB 0:00:38 [49.4MiB/s] [49.4MiB/s] [============>] 100%            
( pvE 0.1 in0 < "${m3t}" | LC_ALL=C sed 's/.*/\L&/'; )   # BSD-sed



37.94s user 0.86s system 101% cpu 38.318 total
d5e2d8487df1136db7c2334a238755c0  stdin



      in0:  313MiB 0:00:00 [3.06GiB/s] [3.06GiB/s] [=====>] 16% ETA 0:00:00
     out9: 1.85GiB 0:00:11 [ 166MiB/s] [ 166MiB/s] [ <=>]
      in0: 1.85GiB 0:00:00 [3.31GiB/s] [3.31GiB/s] [============>] 100%            
( pvE 0.1 in0 < "${m3t}" | LC_ALL=C python3 -c "print(open(0).read().lower())) 

Python 3.9.12 

9.04s user 2.18s system 98% cpu 11.403 total
7ddc0b5cbcfbbfac3c2b6da6731bd262  stdin

     out9: 2.51MiB 0:00:00 [25.1MiB/s] [25.1MiB/s] [ <=> ]
      in0: 1.85GiB 0:00:11 [ 171MiB/s] [ 171MiB/s] [============>] 100%            
     out9: 1.85GiB 0:00:11 [ 171MiB/s] [ 171MiB/s] [ <=> ]
( pvE 0.1 in0 < "${m3t}" | LC_ALL=C ruby -pe '$_.downcase!'; )


ruby 2.6.8p205 (2021-07-07 revision 67951) [universal.arm64e-darwin21]

10.46s user 1.23s system 105% cpu 11.073 total
85759a34df874966d096c6529dbfb9d5  stdin


      in0: 1.85GiB 0:00:01 [1.01GiB/s] [1.01GiB/s] [============>] 100%            
     out9: 1.85GiB 0:00:01 [1.01GiB/s] [1.01GiB/s] [ <=> ]
( pvE 0.1 in0 < "${m3t}" | LC_ALL=C gtr '[A-Z]' '[a-z]'; )  # GNU-tr


gtr (GNU coreutils) 9.1

1.11s user 1.21s system 124% cpu 1.855 total
85759a34df874966d096c6529dbfb9d5  stdin


     out9: 1.85GiB 0:01:19 [23.7MiB/s] [23.7MiB/s] [ <=> ]
      in0: 1.85GiB 0:01:19 [23.7MiB/s] [23.7MiB/s] [============>] 100%            
( pvE 0.1 in0 < "${m3t}" | LC_ALL=C tr '[A-Z]' '[a-z]'; ) # BSD-tr

78.94s user 1.50s system 100% cpu 1:19.67 total
85759a34df874966d096c6529dbfb9d5  stdin


( time ( pvE0 < "${m3t}" | LC_ALL=C   gdd  conv=lcase ) | pvE9 )  | xxh128sum | lgp3; sleep 3; 
     out9: 0.00 B 0:00:01 [0.00 B/s] [0.00 B/s] [<=> ]
      in0: 1.85GiB 0:00:06 [ 295MiB/s] [ 295MiB/s] [============>] 100%            
     out9: 1.81GiB 0:00:06 [ 392MiB/s] [ 294MiB/s] [ <=>   ]
3874110+1 records in
3874110+1 records out
     out9: 1.85GiB 0:00:06 [ 295MiB/s] [ 295MiB/s] [ <=>  ]
( pvE 0.1 in0 < "${m3t}" | LC_ALL=C gdd conv=lcase; )  # GNU-dd


gdd (coreutils) 9.1

1.93s user 4.35s system 97% cpu 6.413 total
85759a34df874966d096c6529dbfb9d5  stdin



%  ( time ( pvE0 < "${m3t}" | LC_ALL=C   dd  conv=lcase ) | pvE9 )  | xxh128sum | lgp3; sleep 3; 
     out9: 36.9MiB 0:00:00 [ 368MiB/s] [ 368MiB/s] [ <=> ]
      in0: 1.85GiB 0:00:04 [ 393MiB/s] [ 393MiB/s] [============>] 100%            
     out9: 1.85GiB 0:00:04 [ 393MiB/s] [ 393MiB/s] [ <=>   ]
3874110+1 records in
3874110+1 records out
     out9: 1.85GiB 0:00:04 [ 393MiB/s] [ 393MiB/s] [ <=>  ]
( pvE 0.1 in0 < "${m3t}" | LC_ALL=C dd conv=lcase; )  # BSD-dd


1.92s user 4.24s system 127% cpu 4.817 total
85759a34df874966d096c6529dbfb9d5  stdin

————————————————————————————————————————

通过一次加载所有文件，并在单个函数调用中对所有1.85 GB执行tolower（），可以人为地使mawk2比perl5更快：：

( time ( pvE0 < "${m3t}" | 

  LC_ALL=C mawk2 '
           BEGIN {            FS = RS = "^$"  } 
             END { print tolower($(ORS = "")) }' 

 ) | pvE9 ) | xxh128sum| lgp3 

      in0: 1.85GiB 0:00:00 [3.35GiB/s] [3.35GiB/s] [============>] 100%            
     out9: 1.85GiB 0:00:02 [ 647MiB/s] [ 647MiB/s] [ <=> ]
( pvE 0.1 in0 < "${m3t}" | LC_ALL=C mawk2 ; )


1.39s user 1.31s system 91% cpu 2.935 total
85759a34df874966d096c6529dbfb9d5  stdin