/*
* Copyright (C) 2012, 2013 Eric Biggers
*
- * This file is part of wimlib, a library for working with WIM files.
+ * This file is free software; you can redistribute it and/or modify it under
+ * the terms of the GNU Lesser General Public License as published by the Free
+ * Software Foundation; either version 3 of the License, or (at your option) any
+ * later version.
*
- * wimlib is free software; you can redistribute it and/or modify it under the
- * terms of the GNU General Public License as published by the Free
- * Software Foundation; either version 3 of the License, or (at your option)
- * any later version.
- *
- * wimlib is distributed in the hope that it will be useful, but WITHOUT ANY
- * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
- * A PARTICULAR PURPOSE. See the GNU General Public License for more
+ * This file is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+ * FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more
* details.
*
- * You should have received a copy of the GNU General Public License
- * along with wimlib; if not, see http://www.gnu.org/licenses/.
+ * You should have received a copy of the GNU Lesser General Public License
+ * along with this file; if not, see http://www.gnu.org/licenses/.
*/
#ifdef HAVE_CONFIG_H
# include "config.h"
#endif
+#include "wimlib/bitops.h"
#include "wimlib/endianness.h"
#include "wimlib/lzx.h"
+#include "wimlib/unaligned.h"
#include "wimlib/util.h"
#ifdef __SSE2__
{
unsigned order;
- if (max_block_size == 0 || max_block_size > (1 << LZX_MAX_WINDOW_ORDER))
+ if (max_block_size == 0 || max_block_size > LZX_MAX_WINDOW_SIZE)
return 0;
- order = bsr32(max_block_size);
+ order = fls32(max_block_size);
- if ((1 << order) != max_block_size)
+ if (((u32)1 << order) != max_block_size)
order++;
return max(order, LZX_MIN_WINDOW_ORDER);
unsigned
lzx_get_num_main_syms(unsigned window_order)
{
- u32 window_size = 1 << window_order;
+ u32 window_size = (u32)1 << window_order;
/* NOTE: the calculation *should* be as follows:
*
}
static void
-do_translate_target(sle32 *target, s32 input_pos)
+do_translate_target(void *target, s32 input_pos)
{
s32 abs_offset, rel_offset;
- /* XXX: This assumes unaligned memory accesses are okay. */
- rel_offset = le32_to_cpu(*target);
+ rel_offset = get_unaligned_u32_le(target);
if (rel_offset >= -input_pos && rel_offset < LZX_WIM_MAGIC_FILESIZE) {
if (rel_offset < LZX_WIM_MAGIC_FILESIZE - input_pos) {
/* "good translation" */
/* "compensating translation" */
abs_offset = rel_offset - LZX_WIM_MAGIC_FILESIZE;
}
- *target = cpu_to_le32(abs_offset);
+ put_unaligned_u32_le(abs_offset, target);
}
}
static void
-undo_translate_target(sle32 *target, s32 input_pos)
+undo_translate_target(void *target, s32 input_pos)
{
s32 abs_offset, rel_offset;
- /* XXX: This assumes unaligned memory accesses are okay. */
- abs_offset = le32_to_cpu(*target);
+ abs_offset = get_unaligned_u32_le(target);
if (abs_offset >= 0) {
if (abs_offset < LZX_WIM_MAGIC_FILESIZE) {
/* "good translation" */
rel_offset = abs_offset - input_pos;
-
- *target = cpu_to_le32(rel_offset);
+ put_unaligned_u32_le(rel_offset, target);
}
} else {
if (abs_offset >= -input_pos) {
/* "compensating translation" */
rel_offset = abs_offset + LZX_WIM_MAGIC_FILESIZE;
-
- *target = cpu_to_le32(rel_offset);
+ put_unaligned_u32_le(rel_offset, target);
}
}
}
SSE2 case, it bloats the binary more. */
#endif
void
-lzx_e8_filter(u8 *data, u32 size, void (*process_target)(sle32 *, s32))
+lzx_e8_filter(u8 *data, u32 size, void (*process_target)(void *, s32))
{
+ u8 *p = data;
#ifdef __SSE2__
/* SSE2 vectorized implementation for x86_64. This speeds up LZX
* decompression by about 5-8% overall. (Usually --- the performance
* consists entirely of 0xe8 bytes. Also, this optimization affects
* compression as well, but the percentage improvement is less because
* LZX compression is much slower than LZX decompression. ) */
- __m128i *p128 = (__m128i *)data;
- u32 valid_mask = 0xFFFFFFFF;
+ if (size >= 32 && (uintptr_t)p % 16 == 0) {
- if (size >= 32 && (uintptr_t)data % 16 == 0) {
- __m128i * const end128 = p128 + size / 16 - 1;
+ u32 valid_mask = 0xFFFFFFFF;
+
+ u8 * const vec_end = p + (size & ~15) - 16;
/* Create a vector of all 0xe8 bytes */
const __m128i e8_bytes = _mm_set1_epi8(0xe8);
/* Compare the current 16-byte vector with the vector of
* all 0xe8 bytes. This produces 0xff where the byte is
* 0xe8 and 0x00 where it is not. */
- __m128i cmpresult = _mm_cmpeq_epi8(*p128, e8_bytes);
+ __m128i cmpresult = _mm_cmpeq_epi8(*(const __m128i *)p,
+ e8_bytes);
/* Map the comparison results into a single 16-bit
* number. It will contain a 1 bit when the
* index of the byte, within the 16, at
* which the next e8 translation should
* be done. */
- u32 bit = __builtin_ctz(e8_mask);
+ int bit = ffs32(e8_mask);
/* Do (or undo) the e8 translation. */
- u8 *p8 = (u8 *)p128 + bit;
- (*process_target)((sle32 *)(p8 + 1),
- p8 - data);
+ (*process_target)(p + bit + 1,
+ p + bit - data);
/* Don't start an e8 translation in the
* next 4 bytes. */
valid_mask >>= 16;
valid_mask |= 0xFFFF0000;
}
- } while (++p128 < end128);
- }
+ } while ((p += 16) < vec_end);
- u8 *p8 = (u8 *)p128;
- while (!(valid_mask & 1)) {
- p8++;
- valid_mask >>= 1;
+ while (!(valid_mask & 1)) {
+ p++;
+ valid_mask >>= 1;
+ }
}
-#else /* __SSE2__ */
- u8 *p8 = data;
#endif /* !__SSE2__ */
if (size > 10) {
/* Finish any bytes that weren't processed by the vectorized
* implementation. */
- u8 *p8_end = data + size - 10;
+ u8 *end = data + size - 10;
do {
- if (*p8 == 0xe8) {
- (*process_target)((sle32 *)(p8 + 1), p8 - data);
- p8 += 5;
+ if (*p == 0xe8) {
+ (*process_target)(p + 1, p - data);
+ p += 5;
} else {
- p8++;
+ p++;
}
- } while (p8 < p8_end);
+ } while (p < end);
}
}
git://wimlib.net / wimlib / blobdiff