Performance improvements for linear encoding and raster output

- use fixed-length string tables (mostly) instead of (char *) pointer ones
   (saves ~40K)
 - re-use C128Table for CODABLOCKF and CODE16K
   (required removal of Stop character and extra CODE16K-only entry)
 - use pointer to destination and copy (memcpy/strcpy(), bin_append_posn())
   instead of concatenating (strcat()) (mostly)
 - replace last remaining bin_append()s with bin_append_posn();
   bin_append() removed
 - add length arg to toupper() and expand() (avoids strlen())
 - change is_sane() to use table-based flags (avoids an iteration)
 - rename lookup() to is_sane_lookup() and change to check and return posns
   and use in pointer to destination loops (avoids strcat()s)
 - remove special case PHARMA in expand() (dealt with in pharma())
 - make #define SILVER/CALCIUM/TECHNETIUM/KRSET etc static strings
 - replace strchr() -> posn()
 - CODE128: populate destination once in checksum loop; re-use and export
   some more routines (c128_set_a/b/c(), c128_put_in_set()) for sharing;
   prefix defines (SHIFTA -> C128_SHIFTA etc) and existing exported routines
 - use factor XOR toggle trick in checksum calcs (avoids branch)
 - raster.c: fill out single 1-pixel row and copy using new draw_bar_line(),
   copy_bar_line() routines; similarly in buffer_plot compare previous line &
   copy if same (same technique as used to improve non-half-integer scaling,
   significant performance increase, (c) codemonkey82);
   also done for PNG (BMP/GIF/PCX/TIFF not done)
 - raster/vector/output.c: shorten "output_" prefix -> "out_";
   sync vector to other raster changes to try to keep source files similar
 - 2of5.c: prefix "c25_"
JAPANPOST: return error if input data truncated (backward incompatible)
DAFT: max chars 50 -> 100
common.c: istwodigit() -> is_twodigit()
common.c/emf.c/output.c: use some further stripf()s (MSVC6 float variations)
library.c: new check_output_args() helper
zint.h: add BARCODE_LAST marker and use in library.c
QRCODE: remove a NOLINT (requires clang-tidy-13), one remaining
CMake: separate no-optimize from ZINT_DEBUG into new ZINT_NOOPT option
This commit is contained in:
gitlost 2021-10-20 23:05:30 +01:00
parent e8b59aa696
commit fab7435fac
72 changed files with 3501 additions and 2380 deletions

View file

@ -1447,7 +1447,7 @@ INTERNAL int aztec(struct zint_symbol *symbol, unsigned char source[], int lengt
/* Encodes Aztec runes as specified in ISO/IEC 24778:2008 Annex A */
INTERNAL int azrune(struct zint_symbol *symbol, unsigned char source[], int length) {
unsigned int input_value;
int error_number, i, y, x, r;
int i, y, x, r;
char binary_string[28];
unsigned char data_codewords[3], ecc_codewords[6];
int bp = 0;
@ -1459,8 +1459,7 @@ INTERNAL int azrune(struct zint_symbol *symbol, unsigned char source[], int leng
strcpy(symbol->errtxt, "507: Input too large (3 character maximum)");
return ZINT_ERROR_TOO_LONG;
}
error_number = is_sane(NEON, source, length);
if (error_number != 0) {
if (!is_sane(NEON_F, source, length)) {
strcpy(symbol->errtxt, "508: Invalid character in data (digits only)");
return ZINT_ERROR_INVALID_DATA;
}