LZX: Allow max_block_size not a power of 2

[wimlib] / src / lzx-common.c

diff --git a/src/lzx-common.c b/src/lzx-common.c
index 24bf9974c4ca5783c68c7f0ca0e4479fd23755dc..566161c3585abf4130ed265a270aa7e09c8cba8d 100644 (file)
--- a/src/lzx-common.c
+++ b/src/lzx-common.c
@@ -33,14 +33,8 @@
 #  include <emmintrin.h>
 #endif
 
-/* LZX uses what it calls 'position slots' to represent match offsets.
- * What this means is that a small 'position slot' number and a small
- * offset from that slot are encoded instead of one large offset for
- * every match.
- * - lzx_position_base is an index to the position slot bases
- * - lzx_extra_bits states how many bits of offset-from-base data is needed.
+/* Mapping: position slot => first match offset that uses that position slot.
  */
-
 const u32 lzx_position_base[LZX_MAX_POSITION_SLOTS] = {
        0      , 1      , 2      , 3      , 4      ,    /* 0  --- 4  */
        6      , 8      , 12     , 16     , 24     ,    /* 5  --- 9  */
@@ -55,6 +49,8 @@ const u32 lzx_position_base[LZX_MAX_POSITION_SLOTS] = {
        2097152                                         /* 50        */
 };
 
+/* Mapping: position slot => how many extra bits must be read and added to the
+ * corresponding position base to decode the match offset.  */
 #ifdef USE_LZX_EXTRA_BITS_ARRAY
 const u8 lzx_extra_bits[LZX_MAX_POSITION_SLOTS] = {
        0 , 0 , 0 , 0 , 1 ,
@@ -71,23 +67,32 @@ const u8 lzx_extra_bits[LZX_MAX_POSITION_SLOTS] = {
 };
 #endif
 
-/* LZX window size must be a power of 2 between 2^15 and 2^21, inclusively.  */
-bool
-lzx_window_size_valid(size_t window_size)
+/* Round the specified compression block size (not LZX block size) up to the
+ * next valid LZX window size, and return its order (log2).  Or, if the block
+ * size is 0 or greater than the largest valid LZX window size, return 0.  */
+unsigned
+lzx_get_window_order(size_t max_block_size)
 {
-       if (window_size == 0 || (u32)window_size != window_size)
-               return false;
-       u32 order = bsr32(window_size);
-       if (window_size != 1U << order)
-               return false;
-       return (order >= LZX_MIN_WINDOW_ORDER && order <= LZX_MAX_WINDOW_ORDER);
+       unsigned order;
+
+       if (max_block_size == 0 || max_block_size > (1 << LZX_MAX_WINDOW_ORDER))
+               return 0;
+
+       order = bsr32(max_block_size);
+
+       if ((1 << order) != max_block_size)
+               order++;
+
+       return max(order, LZX_MIN_WINDOW_ORDER);
 }
 
-/* Given a valid LZX window size, return the number of symbols that will exist
+/* Given a valid LZX window order, return the number of symbols that will exist
  * in the main Huffman code.  */
 unsigned
-lzx_get_num_main_syms(u32 window_size)
+lzx_get_num_main_syms(unsigned window_order)
 {
+       u32 window_size = 1 << window_order;
+
        /* NOTE: the calculation *should* be as follows:
         *
         * u32 max_offset = window_size - LZX_MIN_MATCH_LEN;
@@ -119,7 +124,7 @@ lzx_get_num_main_syms(u32 window_size)
 }
 
 static void
-do_translate_target(s32 *target, s32 input_pos)
+do_translate_target(sle32 *target, s32 input_pos)
 {
        s32 abs_offset, rel_offset;
 
@@ -138,7 +143,7 @@ do_translate_target(s32 *target, s32 input_pos)
 }
 
 static void
-undo_translate_target(s32 *target, s32 input_pos)
+undo_translate_target(sle32 *target, s32 input_pos)
 {
        s32 abs_offset, rel_offset;
 
@@ -193,7 +198,7 @@ inline  /* Although inlining the 'process_target' function still speeds up the
           SSE2 case, it bloats the binary more.  */
 #endif
 void
-lzx_e8_filter(u8 *data, s32 size, void (*process_target)(s32 *, s32))
+lzx_e8_filter(u8 *data, u32 size, void (*process_target)(sle32 *, s32))
 {
 #ifdef __SSE2__
        /* SSE2 vectorized implementation for x86_64.  This speeds up LZX
@@ -248,7 +253,7 @@ lzx_e8_filter(u8 *data, s32 size, void (*process_target)(s32 *, s32))
 
                                        /* Do (or undo) the e8 translation.  */
                                        u8 *p8 = (u8 *)p128 + bit;
-                                       (*process_target)((s32 *)(p8 + 1),
+                                       (*process_target)((sle32 *)(p8 + 1),
                                                          p8 - data);
 
                                        /* Don't start an e8 translation in the
@@ -278,7 +283,7 @@ lzx_e8_filter(u8 *data, s32 size, void (*process_target)(s32 *, s32))
                u8 *p8_end = data + size - 10;
                do {
                        if (*p8 == 0xe8) {
-                               (*process_target)((s32 *)(p8 + 1), p8 - data);
+                               (*process_target)((sle32 *)(p8 + 1), p8 - data);
                                p8 += 5;
                        } else {
                                p8++;
@@ -288,13 +293,13 @@ lzx_e8_filter(u8 *data, s32 size, void (*process_target)(s32 *, s32))
 }
 
 void
-lzx_do_e8_preprocessing(u8 *data, s32 size)
+lzx_do_e8_preprocessing(u8 *data, u32 size)
 {
        lzx_e8_filter(data, size, do_translate_target);
 }
 
 void
-lzx_undo_e8_preprocessing(u8 *data, s32 size)
+lzx_undo_e8_preprocessing(u8 *data, u32 size)
 {
        lzx_e8_filter(data, size, undo_translate_target);
 }