diff options
Diffstat (limited to 'src/pl/plperl')
| -rw-r--r-- | src/pl/plperl/ppport.h | 6 |
1 files changed, 3 insertions, 3 deletions
diff --git a/src/pl/plperl/ppport.h b/src/pl/plperl/ppport.h index 31d06cb3b0b..8c2365674f9 100644 --- a/src/pl/plperl/ppport.h +++ b/src/pl/plperl/ppport.h @@ -6205,10 +6205,10 @@ DPPP_(my_grok_number)(pTHX_ const char *pv, STRLEN len, UV *valuep) /* UVs are at least 32 bits, so the first 9 decimal digits cannot overflow. */ UV value = *s - '0'; - /* This construction seems to be more optimiser friendly. + /* This construction seems to be more optimizer friendly. (without it gcc does the isDIGIT test and the *s - '0' separately) With it gcc on arm is managing 6 instructions (6 cycles) per digit. - In theory the optimiser could deduce how far to unroll the loop + In theory the optimizer could deduce how far to unroll the loop before checking for overflow. */ if (++s < send) { int digit = *s - '0'; @@ -6606,7 +6606,7 @@ DPPP_(my_grok_oct)(pTHX_ const char *start, STRLEN *len_p, I32 *flags, NV *resul bool overflowed = FALSE; for (; len-- && *s; s++) { - /* gcc 2.95 optimiser not smart enough to figure that this subtraction + /* gcc 2.95 optimizer not smart enough to figure that this subtraction out front allows slicker code. */ int digit = *s - '0'; if (digit >= 0 && digit <= 7) { |