Re: [PATCH v9 0/2] MR821: Replace obsolete RtlUniform tests that no longer work (currently skipped) since Windows Vista; update RtlUniform impl.

[Huw Davies (＠huw)]

Huw Davies (@huw) commented about dlls/ntdll/rtl.c:
> - result = *seed * 0xffffffed + 0x7fffffc3;
> - if (result == 0xffffffff || result == 0x7ffffffe) {
> - result = (result + 2) & MAXLONG;
> - } else if (result == 0x7fffffff) {
> - result = 0;
> - } else if ((result & 0x80000000) == 0) {
> - result = result + (~result & 1);
> - } else {
> - result = (result + (result & 1)) & MAXLONG;
> - } /* if */
> - /* The algorithm below is equivalent to the following expression:
> -
> *
> -
> * result = ((ULONGLONG)*seed * 0x7fffffed + 0x7fffffc3) % 0x7fffffff;
> -
> * return (*seed = result);
> -
> */
> -

The optimization below doesn't seem to be worth it.  On x86_64 I couldn't measure a difference between the algorithm above and the optimization below.  On i386, while the optimization was about 10% faster, the algorithm above matched native's performance, so there seems little point in adding the complexity.

If you really need the optimization, you could potentially introduce it in a later MR.