Re: Challenge: CPU-optimized byte-wise or-equals (for a meter of beer)

[bduggan]

I've always wanted an opportunity to use Duff's device :

void bduggan_duff(SV *sv1, SV *sv2)
{ /* slightly modified avar_c_inplace */
    char *sv1p, *sv2p;
    STRLEN sv1len, sv2len;
    SV *sv1_sv;
    STRLEN i;

    if (!SvROK(sv1) || !SvPOK((SV*)SvRV(sv1)) || !SvPOK(sv2)) {
        croak("Usage: avar_cee(\$s1, $s2)");
    }
    
    sv1_sv = (SV*)SvRV(sv1);
    sv1p = SvPV(sv1_sv, sv1len);
    sv2p = SvPV(sv2,    sv2len);
    
    if (sv1len != sv2len) {
       croak("The given strings must be of the same length");
    }
    
    i = 0; 
    switch (sv2len % 8) {
        case 0 : do { if (sv1p[i] == '\0') sv1p[i] = sv2p[i]; i++;
        case 1 :      if (sv1p[i] == '\0') sv1p[i] = sv2p[i]; i++;
        case 2 :      if (sv1p[i] == '\0') sv1p[i] = sv2p[i]; i++;
        case 3 :      if (sv1p[i] == '\0') sv1p[i] = sv2p[i]; i++;
        case 4 :      if (sv1p[i] == '\0') sv1p[i] = sv2p[i]; i++;
        case 5 :      if (sv1p[i] == '\0') sv1p[i] = sv2p[i]; i++;
        case 6 :      if (sv1p[i] == '\0') sv1p[i] = sv2p[i]; i++;
        case 7 :      if (sv1p[i] == '\0') sv1p[i] = sv2p[i]; i++;
              } while (i<sv2len);
    }
}
void bduggan_duff2(SV *sv1, SV *sv2)
{ /* slightly modified avar_tye_c_inplace */
    char *sv1p, *sv2p;                                                
+                                                                     
+            
    STRLEN sv1len, sv2len;
    SV *sv1_sv;
    STRLEN i;

    if (!SvROK(sv1) || !SvPOK((SV*)SvRV(sv1)) || !SvPOK(sv2)) {
        croak("Usage: avar_cee(\$s1, $s2)");
    }

    sv1_sv = (SV*)SvRV(sv1);
    sv1p = SvPV(sv1_sv, sv1len);
    sv2p = SvPV(sv2,    sv2len);

    if (sv1len != sv2len) {
       croak("The given strings must be of the same length");
    }

    char *pSrc = sv2p + sv2len;
    char *pDst = sv1p + sv2len;

    switch (sv2len % 8) {
        case 0 : do { if (!*pDst) { *pDst = *pSrc; } --pSrc; --pDst;
        case 1 :      if (!*pDst) { *pDst = *pSrc; } --pSrc; --pDst;
        case 2 :      if (!*pDst) { *pDst = *pSrc; } --pSrc; --pDst;
        case 3 :      if (!*pDst) { *pDst = *pSrc; } --pSrc; --pDst;
        case 4 :      if (!*pDst) { *pDst = *pSrc; } --pSrc; --pDst;
        case 5 :      if (!*pDst) { *pDst = *pSrc; } --pSrc; --pDst;
        case 6 :      if (!*pDst) { *pDst = *pSrc; } --pSrc; --pDst;
        case 7 :      if (!*pDst) { *pDst = *pSrc; } --pSrc; --pDst;
                    } while (sv2p <= pSrc);
    }

Sadly (or maybe not), they can't compete with memchr. With gcc 3.3.3 the first one seems to come in second place. With gcc 4.1.2, the second one does.
-Brian