Sorry, hopefully this will be the last instalment on that same PWC 342-2 -- because mission (kind of) accomplished, with symbolic and "all-important" order-of-magnitude speed gain over "plain" C on a single core, through a little re-arrangement, partial loop unrolling (step in 96 bytes instead of 32: sums of "-1" and "1"s won't overflow and can be kept as bytes a little longer) and different choice for some instructions (TIMTOWTDI, there's real zoo of them). Apologies (can't edit parent), also, for calling AVX2 as "2017+", of course it's older; and using "len" in place of "aligned_len" above, once: it doesn't affect speed nor result, but is just unclean.

String length: 10,000
             Rate/s    %
 c            88080   100
 va_single   811975   922

String length: 100,000
             Rate/s    %
 c             8945   100
 va_single   108085  1208

String length: 1,000,000
             Rate/s    %
 c              894   100
 va_single    10802  1208

String length: 10,000,000
             Rate/s    %
 c             88.4   100
 va_single    913.5  1033

String length: 100,000,000
             Rate/s    %
 c              8.8   100
 va_single     83.4   946

[download]

However, no gain (compared to "unoptimised" version in parent node) with OMP (same CPU with 4 cores):

String length: 100,000,000
             Rate/s    %
 c              8.8   100
 va_omp       168.7  1910

[download]

That, and relative decrease in advantage for very long input (no additional memory allocation occurs, but only the same simple processing of string chunks, sequentially), I can only speculate is result of throttling of some kind.

And by the way, I also did try to place "mscore_c" C function in separate file, then compile it with "-O3 -march=native" (can't do it with Inline::C, can I?) which would optimise/vectorise to the best of compiler's "voodoo", as googling suggests; then link the library and call the wrapped function from Inline::C. Well, for uniform '"1"s only' string it gave ~25% gain, but for a "random" string it is 3 times slower (than "-O2" i.e. compiled directly from within Inline::C.) This is ridiculous, "voodoo", "A.I", "free" optimisations, and what not.

In reply to Re^4: Weird performance issue with Strawberries and Inline::C by Anonymous Monk
in thread Weird performance issue with Strawberries and Inline::C by Anonymous Monk

Title:
Use:  <p> text here (a paragraph) </p>
and:  <code> code here </code>
to format your post, it's "PerlMonks-approved HTML":


  • Posts are HTML formatted. Put <p> </p> tags around your paragraphs. Put <code> </code> tags around your code and data!
  • Titles consisting of a single word are discouraged, and in most cases are disallowed outright.
  • Read Where should I post X? if you're not absolutely sure you're posting in the right place.
  • Please read these before you post! —
  • Posts may use any of the Perl Monks Approved HTML tags:
    a, abbr, b, big, blockquote, br, caption, center, col, colgroup, dd, del, details, div, dl, dt, em, font, h1, h2, h3, h4, h5, h6, hr, i, ins, li, ol, p, pre, readmore, small, span, spoiler, strike, strong, sub, summary, sup, table, tbody, td, tfoot, th, thead, tr, tt, u, ul, wbr
  • You may need to use entities for some characters, as follows. (Exception: Within code tags, you can put the characters literally.)
            For:     Use:
    & &amp;
    < &lt;
    > &gt;
    [ &#91;
    ] &#93;
  • Link using PerlMonks shortcuts! What shortcuts can I use for linking?
  • See Writeup Formatting Tips and other pages linked from there for more info.