https://github.com/libsdl-org/libtiff/commit/a2aa4074e90b63adc5d8d06679dc567de36f74cd
From a2aa4074e90b63adc5d8d06679dc567de36f74cd Mon Sep 17 00:00:00 2001
From: Even Rouault <[EMAIL REDACTED]>
Date: Thu, 10 Mar 2022 12:19:01 +0100
Subject: [PATCH] LZWDecode(): modest speed improvement: fetch input data by
chunks of the largest natural integer of the architecture
---
libtiff/tif_lzw.c | 141 +++++++++++++++++++++++++++++++++++-----------
1 file changed, 107 insertions(+), 34 deletions(-)
diff --git a/libtiff/tif_lzw.c b/libtiff/tif_lzw.c
index ec976e1f..0124df9f 100644
--- a/libtiff/tif_lzw.c
+++ b/libtiff/tif_lzw.c
@@ -38,6 +38,10 @@
#include <stdio.h>
+/* Select the plausible largest natural integer type for the architecture */
+#define SIZEOF_WORDTYPE SIZEOF_SIZE_T
+typedef size_t WordType;
+
/*
* NB: The 5.0 spec describes a different algorithm than Aldus
* implements. Specifically, Aldus does code length transitions
@@ -85,7 +89,7 @@ typedef struct {
unsigned short nbits; /* # of bits/code */
unsigned short maxcode; /* maximum code for lzw_nbits */
unsigned short free_ent; /* next free entry in hash table */
- unsigned long nextdata; /* next bits of i/o */
+ WordType nextdata; /* next bits of i/o */
long nextbits; /* # of valid bits in lzw_nextdata */
int rw_mode; /* preserve rw_mode from init */
@@ -158,22 +162,6 @@ static void cl_hash(LZWCodecState*);
* LZW Decoder.
*/
-/*
- * This check shouldn't be necessary because each
- * strip is suppose to be terminated with CODE_EOI.
- */
-#define NextCode(_tif, _sp, _bp, _code, _get, dec_bitsleft) { \
- if (dec_bitsleft < (uint64_t)nbits) { \
- TIFFWarningExt(_tif->tif_clientdata, module, \
- "LZWDecode: Strip %"PRIu32" not terminated with EOI code", \
- _tif->tif_curstrip); \
- _code = CODE_EOI; \
- } else { \
- _get(_sp,_bp,_code); \
- dec_bitsleft -= nbits; \
- } \
-}
-
static int
LZWFixupTags(TIFF* tif)
{
@@ -322,16 +310,77 @@ LZWPreDecode(TIFF* tif, uint16_t s)
/*
* Decode a "hunk of data".
*/
-#define GetNextCode(sp, bp, code) { \
- nextdata = (nextdata<<8) | *(bp)++; \
- nextbits += 8; \
- if (nextbits < nbits) { \
- nextdata = (nextdata<<8) | *(bp)++; \
- nextbits += 8; \
- } \
- code = (hcode_t)((nextdata >> (nextbits-nbits)) & nbitsmask); \
- nextbits -= nbits; \
-}
+
+/* Get the next 32 or 64-bit from the input data */
+#ifdef WORDS_BIGENDIAN
+# define GetNextData(nextdata, bp) memcpy(&nextdata, bp, sizeof(nextdata))
+#elif SIZEOF_WORDTYPE == 8
+# if defined(__GNUC__) && defined(__x86_64__)
+# define GetNextData(nextdata, bp) nextdata = __builtin_bswap64(*(uint64_t*)(bp))
+# elif defined(_M_X64)
+# define GetNextData(nextdata, bp) nextdata = _byteswap_uint64(*(uint64_t*)(bp))
+# elif defined(__GNUC__)
+# define GetNextData(nextdata, bp) memcpy(&nextdata, bp, sizeof(nextdata)); \
+ nextdata = __builtin_bswap64(nextdata)
+# else
+# define GetNextData(nextdata, bp) nextdata = (((uint64_t)bp[0]) << 56) | \
+ (((uint64_t)bp[1]) << 48) | \
+ (((uint64_t)bp[2]) << 40) | \
+ (((uint64_t)bp[3]) << 32) | \
+ (((uint64_t)bp[4]) << 24) | \
+ (((uint64_t)bp[5]) << 16) | \
+ (((uint64_t)bp[6]) << 8) | \
+ (((uint64_t)bp[7]))
+# endif
+#elif SIZEOF_WORDTYPE == 4
+# if defined(__GNUC__) && defined(__i386__)
+# define GetNextData(nextdata, bp) nextdata = __builtin_bswap32(*(uint32_t*)(bp))
+# elif defined(_M_X86)
+# define GetNextData(nextdata, bp) nextdata = _byteswap_ulong(*(unsigned long*)(bp))
+# elif defined(__GNUC__)
+# define GetNextData(nextdata, bp) memcpy(&nextdata, bp, sizeof(nextdata)); \
+ nextdata = __builtin_bswap32(nextdata)
+# else
+# define GetNextData(nextdata, bp) nextdata = (((uint32_t)bp[0]) << 24) | \
+ (((uint32_t)bp[1]) << 16) | \
+ (((uint32_t)bp[2]) << 8) | \
+ (((uint32_t)bp[3]))
+# endif
+#else
+# error "Unhandled SIZEOF_WORDTYPE"
+#endif
+
+#define GetNextCodeLZW() do { \
+ nextbits -= nbits; \
+ if (nextbits < 0) { \
+ if (dec_bitsleft >= 8 * SIZEOF_WORDTYPE) { \
+ unsigned codetmp = (unsigned)(nextdata << (-nextbits)); \
+ GetNextData(nextdata, bp); \
+ bp += SIZEOF_WORDTYPE; \
+ nextbits += 8 * SIZEOF_WORDTYPE; \
+ dec_bitsleft -= 8 * SIZEOF_WORDTYPE; \
+ code = (WordType)((codetmp | (nextdata >> nextbits)) & nbitsmask); \
+ break; \
+ } \
+ else {\
+ if( dec_bitsleft < 8) { \
+ goto no_eoi; \
+ }\
+ nextdata = (nextdata<<8) | *(bp)++; \
+ nextbits += 8; \
+ dec_bitsleft -= 8; \
+ if( nextbits < 0 ) { \
+ if( dec_bitsleft < 8) { \
+ goto no_eoi; \
+ }\
+ nextdata = (nextdata<<8) | *(bp)++; \
+ nextbits += 8; \
+ dec_bitsleft -= 8; \
+ } \
+ } \
+ } \
+ code = (WordType)((nextdata >> nextbits) & nbitsmask); \
+} while(0)
static void
codeLoop(TIFF* tif, const char* module)
@@ -350,10 +399,10 @@ LZWDecode(TIFF* tif, uint8_t* op0, tmsize_t occ0, uint16_t s)
tmsize_t occ = occ0;
uint8_t *tp;
uint8_t *bp;
- hcode_t code;
+ WordType code;
int len;
long nbits, nextbits, nbitsmask;
- unsigned long nextdata;
+ WordType nextdata;
code_t *codep, *free_entp, *maxcodep, *oldcodep;
(void) s;
@@ -413,7 +462,7 @@ LZWDecode(TIFF* tif, uint8_t* op0, tmsize_t occ0, uint16_t s)
maxcodep = sp->dec_maxcodep;
while (occ > 0) {
- NextCode(tif, sp, bp, code, GetNextCode, dec_bitsleft);
+ GetNextCodeLZW();
if (code == CODE_EOI)
break;
if (code == CODE_CLEAR) {
@@ -424,7 +473,7 @@ LZWDecode(TIFF* tif, uint8_t* op0, tmsize_t occ0, uint16_t s)
nbits = BITS_MIN;
nbitsmask = MAXCODE(BITS_MIN);
maxcodep = sp->dec_codetab + nbitsmask-1;
- NextCode(tif, sp, bp, code, GetNextCode, dec_bitsleft);
+ GetNextCodeLZW();
} while (code == CODE_CLEAR); /* consecutive CODE_CLEAR codes */
if (code == CODE_EOI)
break;
@@ -544,9 +593,32 @@ LZWDecode(TIFF* tif, uint8_t* op0, tmsize_t occ0, uint16_t s)
return (0);
}
return (1);
+
+no_eoi:
+ TIFFErrorExt(tif->tif_clientdata, module,
+ "LZWDecode: Strip %"PRIu32" not terminated with EOI code",
+ tif->tif_curstrip);
+ return 0;
}
#ifdef LZW_COMPAT
+
+/*
+ * This check shouldn't be necessary because each
+ * strip is suppose to be terminated with CODE_EOI.
+ */
+#define NextCode(_tif, _sp, _bp, _code, _get, dec_bitsleft) { \
+ if (dec_bitsleft < (uint64_t)nbits) { \
+ TIFFWarningExt(_tif->tif_clientdata, module, \
+ "LZWDecode: Strip %"PRIu32" not terminated with EOI code", \
+ _tif->tif_curstrip); \
+ _code = CODE_EOI; \
+ } else { \
+ _get(_sp,_bp,_code); \
+ dec_bitsleft -= nbits; \
+ } \
+}
+
/*
* Decode a "hunk of data" for old images.
*/
@@ -573,7 +645,8 @@ LZWDecodeCompat(TIFF* tif, uint8_t* op0, tmsize_t occ0, uint16_t s)
uint8_t *bp;
int code, nbits;
int len;
- long nextbits, nextdata, nbitsmask;
+ long nextbits, nbitsmask;
+ WordType nextdata;
code_t *codep, *free_entp, *maxcodep, *oldcodep;
(void) s;
@@ -858,7 +931,7 @@ LZWEncode(TIFF* tif, uint8_t* bp, tmsize_t cc, uint16_t s)
hcode_t ent;
long disp;
tmsize_t incount, outcount, checkpoint;
- unsigned long nextdata;
+ WordType nextdata;
long nextbits;
int free_ent, maxcode, nbits;
uint8_t* op;
@@ -1022,7 +1095,7 @@ LZWPostEncode(TIFF* tif)
register LZWCodecState *sp = EncoderState(tif);
uint8_t* op = tif->tif_rawcp;
long nextbits = sp->lzw_nextbits;
- unsigned long nextdata = sp->lzw_nextdata;
+ WordType nextdata = sp->lzw_nextdata;
tmsize_t outcount = sp->enc_outcount;
int nbits = sp->lzw_nbits;