aboutsummaryrefslogtreecommitdiff
path: root/src/liblzma/lzma
diff options
context:
space:
mode:
authorJia Tan <jiat0218@gmail.com>2024-02-12 17:09:10 +0200
committerLasse Collin <lasse.collin@tukaani.org>2024-02-14 18:31:16 +0200
commite446ab7a18abfde18f8d1cf02a914df72b1370e3 (patch)
tree9fe4636a1fa93454b71143159429a3f22b59d5bc /src/liblzma/lzma
parentdoxygen/footer.html: Add missing closing tags and don't open a new tab. (diff)
downloadxz-e446ab7a18abfde18f8d1cf02a914df72b1370e3.tar.xz
liblzma: Creates separate "safe" range decoder mode.
The new "safe" range decoder mode is the same as old range decoder, but now the default behavior of the range decoder will not check if there is enough input or output to complete the operation. When the buffers are close to fully consumed, the "safe" operations must be used instead. This will improve speed because it will reduce the number of branches needed for most of the range decoder operations.
Diffstat (limited to 'src/liblzma/lzma')
-rw-r--r--src/liblzma/lzma/lzma_decoder.c108
1 files changed, 25 insertions, 83 deletions
diff --git a/src/liblzma/lzma/lzma_decoder.c b/src/liblzma/lzma/lzma_decoder.c
index cf437d88..2e8393d6 100644
--- a/src/liblzma/lzma/lzma_decoder.c
+++ b/src/liblzma/lzma/lzma_decoder.c
@@ -25,21 +25,13 @@
#ifdef HAVE_SMALL
// Macros for (somewhat) size-optimized code.
-#define seq_4(seq) seq
-
-#define seq_6(seq) seq
-
-#define seq_8(seq) seq
-
-#define seq_len(seq) \
- seq ## _CHOICE, \
- seq ## _CHOICE2, \
- seq ## _BITTREE
-
+// This is used to decode the match length (how many bytes must be repeated
+// from the dictionary). This version is used in the Resumable mode and
+// does not unroll any loops.
#define len_decode(target, ld, pos_state, seq) \
do { \
case seq ## _CHOICE: \
- rc_if_0(ld.choice, seq ## _CHOICE) { \
+ rc_if_0_safe(ld.choice, seq ## _CHOICE) { \
rc_update_0(ld.choice); \
probs = ld.low[pos_state];\
limit = LEN_LOW_SYMBOLS; \
@@ -47,7 +39,7 @@ case seq ## _CHOICE: \
} else { \
rc_update_1(ld.choice); \
case seq ## _CHOICE2: \
- rc_if_0(ld.choice2, seq ## _CHOICE2) { \
+ rc_if_0_safe(ld.choice2, seq ## _CHOICE2) { \
rc_update_0(ld.choice2); \
probs = ld.mid[pos_state]; \
limit = LEN_MID_SYMBOLS; \
@@ -63,89 +55,42 @@ case seq ## _CHOICE2: \
symbol = 1; \
case seq ## _BITTREE: \
do { \
- rc_bit(probs[symbol], , , seq ## _BITTREE); \
+ rc_bit_safe(probs[symbol], , , seq ## _BITTREE); \
} while (symbol < limit); \
target += symbol - limit; \
} while (0)
-#else // HAVE_SMALL
-
-// Unrolled versions
-#define seq_4(seq) \
- seq ## 0, \
- seq ## 1, \
- seq ## 2, \
- seq ## 3
-
-#define seq_6(seq) \
- seq ## 0, \
- seq ## 1, \
- seq ## 2, \
- seq ## 3, \
- seq ## 4, \
- seq ## 5
-
-#define seq_8(seq) \
- seq ## 0, \
- seq ## 1, \
- seq ## 2, \
- seq ## 3, \
- seq ## 4, \
- seq ## 5, \
- seq ## 6, \
- seq ## 7
-
-#define seq_len(seq) \
- seq ## _CHOICE, \
- seq ## _LOW0, \
- seq ## _LOW1, \
- seq ## _LOW2, \
- seq ## _CHOICE2, \
- seq ## _MID0, \
- seq ## _MID1, \
- seq ## _MID2, \
- seq ## _HIGH0, \
- seq ## _HIGH1, \
- seq ## _HIGH2, \
- seq ## _HIGH3, \
- seq ## _HIGH4, \
- seq ## _HIGH5, \
- seq ## _HIGH6, \
- seq ## _HIGH7
-#define len_decode(target, ld, pos_state, seq) \
+// This is the faster version of the match length decoder that does not
+// worry about being resumable. It unrolls the bittree decoding loop.
+#define len_decode_fast(target, ld, pos_state) \
do { \
symbol = 1; \
-case seq ## _CHOICE: \
- rc_if_0(ld.choice, seq ## _CHOICE) { \
+ rc_if_0(ld.choice) { \
rc_update_0(ld.choice); \
- rc_bit_case(ld.low[pos_state][symbol], , , seq ## _LOW0); \
- rc_bit_case(ld.low[pos_state][symbol], , , seq ## _LOW1); \
- rc_bit_case(ld.low[pos_state][symbol], , , seq ## _LOW2); \
+ rc_bit(ld.low[pos_state][symbol], , ); \
+ rc_bit(ld.low[pos_state][symbol], , ); \
+ rc_bit(ld.low[pos_state][symbol], , ); \
target = symbol - LEN_LOW_SYMBOLS + MATCH_LEN_MIN; \
} else { \
rc_update_1(ld.choice); \
-case seq ## _CHOICE2: \
- rc_if_0(ld.choice2, seq ## _CHOICE2) { \
+ rc_if_0(ld.choice2) { \
rc_update_0(ld.choice2); \
- rc_bit_case(ld.mid[pos_state][symbol], , , \
- seq ## _MID0); \
- rc_bit_case(ld.mid[pos_state][symbol], , , \
- seq ## _MID1); \
- rc_bit_case(ld.mid[pos_state][symbol], , , \
- seq ## _MID2); \
+ rc_bit(ld.mid[pos_state][symbol], , ); \
+ rc_bit(ld.mid[pos_state][symbol], , ); \
+ rc_bit(ld.mid[pos_state][symbol], , ); \
target = symbol - LEN_MID_SYMBOLS \
+ MATCH_LEN_MIN + LEN_LOW_SYMBOLS; \
} else { \
rc_update_1(ld.choice2); \
- rc_bit_case(ld.high[symbol], , , seq ## _HIGH0); \
- rc_bit_case(ld.high[symbol], , , seq ## _HIGH1); \
- rc_bit_case(ld.high[symbol], , , seq ## _HIGH2); \
- rc_bit_case(ld.high[symbol], , , seq ## _HIGH3); \
- rc_bit_case(ld.high[symbol], , , seq ## _HIGH4); \
- rc_bit_case(ld.high[symbol], , , seq ## _HIGH5); \
- rc_bit_case(ld.high[symbol], , , seq ## _HIGH6); \
- rc_bit_case(ld.high[symbol], , , seq ## _HIGH7); \
+ rc_bit(ld.high[symbol], , ); \
+ rc_bit(ld.high[symbol], , ); \
+ rc_bit(ld.high[symbol], , ); \
+ rc_bit(ld.high[symbol], , ); \
+ rc_bit(ld.high[symbol], , ); \
+ rc_bit(ld.high[symbol], , ); \
+ rc_bit(ld.high[symbol], , ); \
+ rc_bit(ld.high[symbol], , ); \
target = symbol - LEN_HIGH_SYMBOLS \
+ MATCH_LEN_MIN \
+ LEN_LOW_SYMBOLS + LEN_MID_SYMBOLS; \
@@ -153,8 +98,6 @@ case seq ## _CHOICE2: \
} \
} while (0)
-#endif // HAVE_SMALL
-
/// Length decoder probabilities; see comments in lzma_common.h.
typedef struct {
@@ -889,7 +832,6 @@ out:
}
-
static void
lzma_decoder_uncompressed(void *coder_ptr, lzma_vli uncompressed_size,
bool allow_eopm)