Steal highly optimizable and smart algorithm to do LWS detection

invented by the clang project. It does LWS detection in just a
handful instructions and no branches, as compared to the dozen
instructions and 4 branches currently.

Some info on how it works here:

https://pdimov.github.io/blog/2020/07/19/llvm-and-memchr/

==== C code ===
int
is_ws_original(char cp)
{

    return ((cp == ' ') || (cp == '\r') || (cp == 'n') || (cp == '\t') || (cp == ','));
}

int
is_ws_new(unsigned char ch)
{
    const unsigned int mask = (1 << (' ' - 1)) | (1 << ('\r' - 1)) |
        (1 << ('\n' - 1)) | (1 << ('\t' - 1));
    ch--;
    return ch < ' ' && ((1 << ch) & mask);
}
==== Compiled, latest clang/trunk, -O2, x86_64 ====
is_ws_original:                         # @is_ws_original
        mov     eax, 1
        cmp     dil, 32
        ja      .LBB6_1
        movzx   ecx, dil
        movabs  rdx, 4294976000
        bt      rdx, rcx
        jb      .LBB6_4
.LBB6_1:
        cmp     dil, 110
        jne     .LBB6_2
.LBB6_4:
        ret
.LBB6_2:
        xor     eax, eax
        cmp     dil, 44
        sete    al
        ret

is_ws_new:                              # @is_ws_new
        add     dil, -1
        cmp     dil, 32
        setb    al
        movzx   ecx, dil
        mov     edx, -2147478784
        bt      edx, ecx
        setb    cl
        and     cl, al
        movzx   eax, cl
        ret
====

Author: sobomax

trim.h

@@ -25,7 +25,14 @@
 #include "str.h"
 
 /* whitespace */
-#define is_ws(c) ((c) == ' ' || (c) == '\r' || (c) == '\n' || (c) == '\t')
+static inline int
+is_ws(unsigned char ch)
+{
+    const unsigned int mask = (1 << (' ' - 1)) | (1 << ('\r' - 1)) |
+        (1 << ('\n' - 1)) | (1 << ('\t' - 1));
+    ch--;
+    return ch < ' ' && ((1 << ch) & mask);
+}
 
 /*
  * trim leading ws