summaryrefslogtreecommitdiff
path: root/media/libjpeg/jchuff.c
diff options
context:
space:
mode:
Diffstat (limited to 'media/libjpeg/jchuff.c')
-rw-r--r--media/libjpeg/jchuff.c620
1 files changed, 333 insertions, 287 deletions
diff --git a/media/libjpeg/jchuff.c b/media/libjpeg/jchuff.c
index fffaacebce..f4dfa1cb54 100644
--- a/media/libjpeg/jchuff.c
+++ b/media/libjpeg/jchuff.c
@@ -4,8 +4,10 @@
* This file was part of the Independent JPEG Group's software:
* Copyright (C) 1991-1997, Thomas G. Lane.
* libjpeg-turbo Modifications:
- * Copyright (C) 2009-2011, 2014-2016, D. R. Commander.
+ * Copyright (C) 2009-2011, 2014-2016, 2018-2022, D. R. Commander.
* Copyright (C) 2015, Matthieu Darbois.
+ * Copyright (C) 2018, Matthias Räncker.
+ * Copyright (C) 2020, Arm Limited.
* For conditions of distribution and use, see the accompanying README.ijg
* file.
*
@@ -16,6 +18,9 @@
* back up to the start of the current MCU. To do this, we copy state
* variables into local working storage, and update them back to the
* permanent JPEG objects only upon successful completion of an MCU.
+ *
+ * NOTE: All referenced figures are from
+ * Recommendation ITU-T T.81 (1992) | ISO/IEC 10918-1:1994.
*/
#define JPEG_INTERNALS
@@ -31,32 +36,33 @@
* memory footprint by 64k, which is important for some mobile applications
* that create many isolated instances of libjpeg-turbo (web browsers, for
* instance.) This may improve performance on some mobile platforms as well.
- * This feature is enabled by default only on ARM processors, because some x86
+ * This feature is enabled by default only on Arm processors, because some x86
* chips have a slow implementation of bsr, and the use of clz/bsr cannot be
* shown to have a significant performance impact even on the x86 chips that
- * have a fast implementation of it. When building for ARMv6, you can
+ * have a fast implementation of it. When building for Armv6, you can
* explicitly disable the use of clz/bsr by adding -mthumb to the compiler
* flags (this defines __thumb__).
*/
/* NOTE: Both GCC and Clang define __GNUC__ */
-#if defined __GNUC__ && (defined __arm__ || defined __aarch64__)
-#if !defined __thumb__ || defined __thumb2__
+#if (defined(__GNUC__) && (defined(__arm__) || defined(__aarch64__))) || \
+ defined(_M_ARM) || defined(_M_ARM64)
+#if !defined(__thumb__) || defined(__thumb2__)
#define USE_CLZ_INTRINSIC
#endif
#endif
#ifdef USE_CLZ_INTRINSIC
-#define JPEG_NBITS_NONZERO(x) (32 - __builtin_clz(x))
-#define JPEG_NBITS(x) (x ? JPEG_NBITS_NONZERO(x) : 0)
+#if defined(_MSC_VER) && !defined(__clang__)
+#define JPEG_NBITS_NONZERO(x) (32 - _CountLeadingZeros(x))
#else
-#include "jpeg_nbits_table.h"
-#define JPEG_NBITS(x) (jpeg_nbits_table[x])
-#define JPEG_NBITS_NONZERO(x) JPEG_NBITS(x)
+#define JPEG_NBITS_NONZERO(x) (32 - __builtin_clz(x))
#endif
-
-#ifndef min
- #define min(a,b) ((a)<(b)?(a):(b))
+#define JPEG_NBITS(x) (x ? JPEG_NBITS_NONZERO(x) : 0)
+#else
+#include "jpeg_nbits_table.h"
+#define JPEG_NBITS(x) (jpeg_nbits_table[x])
+#define JPEG_NBITS_NONZERO(x) JPEG_NBITS(x)
#endif
@@ -66,31 +72,42 @@
* but must not be updated permanently until we complete the MCU.
*/
-typedef struct {
- size_t put_buffer; /* current bit-accumulation buffer */
- int put_bits; /* # of bits now in it */
- int last_dc_val[MAX_COMPS_IN_SCAN]; /* last DC coef for each component */
-} savable_state;
+#if defined(__x86_64__) && defined(__ILP32__)
+typedef unsigned long long bit_buf_type;
+#else
+typedef size_t bit_buf_type;
+#endif
-/* This macro is to work around compilers with missing or broken
- * structure assignment. You'll need to fix this code if you have
- * such a compiler and you change MAX_COMPS_IN_SCAN.
+/* NOTE: The more optimal Huffman encoding algorithm is only used by the
+ * intrinsics implementation of the Arm Neon SIMD extensions, which is why we
+ * retain the old Huffman encoder behavior when using the GAS implementation.
*/
-
-#ifndef NO_STRUCT_ASSIGN
-#define ASSIGN_STATE(dest,src) ((dest) = (src))
+#if defined(WITH_SIMD) && !(defined(__arm__) || defined(__aarch64__) || \
+ defined(_M_ARM) || defined(_M_ARM64))
+typedef unsigned long long simd_bit_buf_type;
#else
-#if MAX_COMPS_IN_SCAN == 4
-#define ASSIGN_STATE(dest,src) \
- ((dest).put_buffer = (src).put_buffer, \
- (dest).put_bits = (src).put_bits, \
- (dest).last_dc_val[0] = (src).last_dc_val[0], \
- (dest).last_dc_val[1] = (src).last_dc_val[1], \
- (dest).last_dc_val[2] = (src).last_dc_val[2], \
- (dest).last_dc_val[3] = (src).last_dc_val[3])
+typedef bit_buf_type simd_bit_buf_type;
#endif
+
+#if (defined(SIZEOF_SIZE_T) && SIZEOF_SIZE_T == 8) || defined(_WIN64) || \
+ (defined(__x86_64__) && defined(__ILP32__))
+#define BIT_BUF_SIZE 64
+#elif (defined(SIZEOF_SIZE_T) && SIZEOF_SIZE_T == 4) || defined(_WIN32)
+#define BIT_BUF_SIZE 32
+#else
+#error Cannot determine word size
#endif
+#define SIMD_BIT_BUF_SIZE (sizeof(simd_bit_buf_type) * 8)
+typedef struct {
+ union {
+ bit_buf_type c;
+ simd_bit_buf_type simd;
+ } put_buffer; /* current bit accumulation buffer */
+ int free_bits; /* # of bits available in it */
+ /* (Neon GAS: # of bits now in it) */
+ int last_dc_val[MAX_COMPS_IN_SCAN]; /* last DC coef for each component */
+} savable_state;
typedef struct {
struct jpeg_entropy_encoder pub; /* public fields */
@@ -124,16 +141,17 @@ typedef struct {
size_t free_in_buffer; /* # of byte spaces remaining in buffer */
savable_state cur; /* Current bit buffer & DC state */
j_compress_ptr cinfo; /* dump_buffer needs access to this */
+ int simd;
} working_state;
/* Forward declarations */
-METHODDEF(boolean) encode_mcu_huff (j_compress_ptr cinfo, JBLOCKROW *MCU_data);
-METHODDEF(void) finish_pass_huff (j_compress_ptr cinfo);
+METHODDEF(boolean) encode_mcu_huff(j_compress_ptr cinfo, JBLOCKROW *MCU_data);
+METHODDEF(void) finish_pass_huff(j_compress_ptr cinfo);
#ifdef ENTROPY_OPT_SUPPORTED
-METHODDEF(boolean) encode_mcu_gather (j_compress_ptr cinfo,
- JBLOCKROW *MCU_data);
-METHODDEF(void) finish_pass_gather (j_compress_ptr cinfo);
+METHODDEF(boolean) encode_mcu_gather(j_compress_ptr cinfo,
+ JBLOCKROW *MCU_data);
+METHODDEF(void) finish_pass_gather(j_compress_ptr cinfo);
#endif
@@ -144,9 +162,9 @@ METHODDEF(void) finish_pass_gather (j_compress_ptr cinfo);
*/
METHODDEF(void)
-start_pass_huff (j_compress_ptr cinfo, boolean gather_statistics)
+start_pass_huff(j_compress_ptr cinfo, boolean gather_statistics)
{
- huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
+ huff_entropy_ptr entropy = (huff_entropy_ptr)cinfo->entropy;
int ci, dctbl, actbl;
jpeg_component_info *compptr;
@@ -180,30 +198,39 @@ start_pass_huff (j_compress_ptr cinfo, boolean gather_statistics)
/* Note that jpeg_gen_optimal_table expects 257 entries in each table! */
if (entropy->dc_count_ptrs[dctbl] == NULL)
entropy->dc_count_ptrs[dctbl] = (long *)
- (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+ (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE,
257 * sizeof(long));
- MEMZERO(entropy->dc_count_ptrs[dctbl], 257 * sizeof(long));
+ memset(entropy->dc_count_ptrs[dctbl], 0, 257 * sizeof(long));
if (entropy->ac_count_ptrs[actbl] == NULL)
entropy->ac_count_ptrs[actbl] = (long *)
- (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+ (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE,
257 * sizeof(long));
- MEMZERO(entropy->ac_count_ptrs[actbl], 257 * sizeof(long));
+ memset(entropy->ac_count_ptrs[actbl], 0, 257 * sizeof(long));
#endif
} else {
/* Compute derived values for Huffman tables */
/* We may do this more than once for a table, but it's not expensive */
jpeg_make_c_derived_tbl(cinfo, TRUE, dctbl,
- & entropy->dc_derived_tbls[dctbl]);
+ &entropy->dc_derived_tbls[dctbl]);
jpeg_make_c_derived_tbl(cinfo, FALSE, actbl,
- & entropy->ac_derived_tbls[actbl]);
+ &entropy->ac_derived_tbls[actbl]);
}
/* Initialize DC predictions to 0 */
entropy->saved.last_dc_val[ci] = 0;
}
/* Initialize bit buffer to empty */
- entropy->saved.put_buffer = 0;
- entropy->saved.put_bits = 0;
+ if (entropy->simd) {
+ entropy->saved.put_buffer.simd = 0;
+#if defined(__aarch64__) && !defined(NEON_INTRINSICS)
+ entropy->saved.free_bits = 0;
+#else
+ entropy->saved.free_bits = SIMD_BIT_BUF_SIZE;
+#endif
+ } else {
+ entropy->saved.put_buffer.c = 0;
+ entropy->saved.free_bits = BIT_BUF_SIZE;
+ }
/* Initialize restart stuff */
entropy->restarts_to_go = cinfo->restart_interval;
@@ -219,8 +246,8 @@ start_pass_huff (j_compress_ptr cinfo, boolean gather_statistics)
*/
GLOBAL(void)
-jpeg_make_c_derived_tbl (j_compress_ptr cinfo, boolean isDC, int tblno,
- c_derived_tbl **pdtbl)
+jpeg_make_c_derived_tbl(j_compress_ptr cinfo, boolean isDC, int tblno,
+ c_derived_tbl **pdtbl)
{
JHUFF_TBL *htbl;
c_derived_tbl *dtbl;
@@ -244,7 +271,7 @@ jpeg_make_c_derived_tbl (j_compress_ptr cinfo, boolean isDC, int tblno,
/* Allocate a workspace if we haven't already done so. */
if (*pdtbl == NULL)
*pdtbl = (c_derived_tbl *)
- (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+ (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE,
sizeof(c_derived_tbl));
dtbl = *pdtbl;
@@ -252,11 +279,11 @@ jpeg_make_c_derived_tbl (j_compress_ptr cinfo, boolean isDC, int tblno,
p = 0;
for (l = 1; l <= 16; l++) {
- i = (int) htbl->bits[l];
+ i = (int)htbl->bits[l];
if (i < 0 || p + i > 256) /* protect against table overrun */
ERREXIT(cinfo, JERR_BAD_HUFF_TABLE);
while (i--)
- huffsize[p++] = (char) l;
+ huffsize[p++] = (char)l;
}
huffsize[p] = 0;
lastp = p;
@@ -268,14 +295,14 @@ jpeg_make_c_derived_tbl (j_compress_ptr cinfo, boolean isDC, int tblno,
si = huffsize[0];
p = 0;
while (huffsize[p]) {
- while (((int) huffsize[p]) == si) {
+ while (((int)huffsize[p]) == si) {
huffcode[p++] = code;
code++;
}
/* code is now 1 more than the last code used for codelength si; but
* it must still fit in si bits, since no code is allowed to be all ones.
*/
- if (((JLONG) code) >= (((JLONG) 1) << si))
+ if (((JLONG)code) >= (((JLONG)1) << si))
ERREXIT(cinfo, JERR_BAD_HUFF_TABLE);
code <<= 1;
si++;
@@ -288,7 +315,8 @@ jpeg_make_c_derived_tbl (j_compress_ptr cinfo, boolean isDC, int tblno,
* this lets us detect duplicate VAL entries here, and later
* allows emit_bits to detect any attempt to emit such symbols.
*/
- MEMZERO(dtbl->ehufsi, sizeof(dtbl->ehufsi));
+ memset(dtbl->ehufco, 0, sizeof(dtbl->ehufco));
+ memset(dtbl->ehufsi, 0, sizeof(dtbl->ehufsi));
/* This is also a convenient place to check for out-of-range
* and duplicated VAL entries. We allow 0..255 for AC symbols
@@ -310,20 +338,21 @@ jpeg_make_c_derived_tbl (j_compress_ptr cinfo, boolean isDC, int tblno,
/* Outputting bytes to the file */
/* Emit a byte, taking 'action' if must suspend. */
-#define emit_byte(state,val,action) \
- { *(state)->next_output_byte++ = (JOCTET) (val); \
- if (--(state)->free_in_buffer == 0) \
- if (! dump_buffer(state)) \
- { action; } }
+#define emit_byte(state, val, action) { \
+ *(state)->next_output_byte++ = (JOCTET)(val); \
+ if (--(state)->free_in_buffer == 0) \
+ if (!dump_buffer(state)) \
+ { action; } \
+}
LOCAL(boolean)
-dump_buffer (working_state *state)
+dump_buffer(working_state *state)
/* Empty the output buffer; return TRUE if successful, FALSE if must suspend */
{
struct jpeg_destination_mgr *dest = state->cinfo->dest;
- if (! (*dest->empty_output_buffer) (state->cinfo))
+ if (!(*dest->empty_output_buffer) (state->cinfo))
return FALSE;
/* After a successful buffer dump, must reset buffer pointers */
state->next_output_byte = dest->next_output_byte;
@@ -334,89 +363,93 @@ dump_buffer (working_state *state)
/* Outputting bits to the file */
-/* These macros perform the same task as the emit_bits() function in the
- * original libjpeg code. In addition to reducing overhead by explicitly
- * inlining the code, additional performance is achieved by taking into
- * account the size of the bit buffer and waiting until it is almost full
- * before emptying it. This mostly benefits 64-bit platforms, since 6
- * bytes can be stored in a 64-bit bit buffer before it has to be emptied.
+/* Output byte b and, speculatively, an additional 0 byte. 0xFF must be
+ * encoded as 0xFF 0x00, so the output buffer pointer is advanced by 2 if the
+ * byte is 0xFF. Otherwise, the output buffer pointer is advanced by 1, and
+ * the speculative 0 byte will be overwritten by the next byte.
*/
-
-#define EMIT_BYTE() { \
- JOCTET c; \
- put_bits -= 8; \
- c = (JOCTET)GETJOCTET(put_buffer >> put_bits); \
- *buffer++ = c; \
- if (c == 0xFF) /* need to stuff a zero byte? */ \
- *buffer++ = 0; \
- }
-
-#define PUT_BITS(code, size) { \
- put_bits += size; \
- put_buffer = (put_buffer << size) | code; \
+#define EMIT_BYTE(b) { \
+ buffer[0] = (JOCTET)(b); \
+ buffer[1] = 0; \
+ buffer -= -2 + ((JOCTET)(b) < 0xFF); \
}
-#define CHECKBUF15() { \
- if (put_bits > 15) { \
- EMIT_BYTE() \
- EMIT_BYTE() \
+/* Output the entire bit buffer. If there are no 0xFF bytes in it, then write
+ * directly to the output buffer. Otherwise, use the EMIT_BYTE() macro to
+ * encode 0xFF as 0xFF 0x00.
+ */
+#if BIT_BUF_SIZE == 64
+
+#define FLUSH() { \
+ if (put_buffer & 0x8080808080808080 & ~(put_buffer + 0x0101010101010101)) { \
+ EMIT_BYTE(put_buffer >> 56) \
+ EMIT_BYTE(put_buffer >> 48) \
+ EMIT_BYTE(put_buffer >> 40) \
+ EMIT_BYTE(put_buffer >> 32) \
+ EMIT_BYTE(put_buffer >> 24) \
+ EMIT_BYTE(put_buffer >> 16) \
+ EMIT_BYTE(put_buffer >> 8) \
+ EMIT_BYTE(put_buffer ) \
+ } else { \
+ buffer[0] = (JOCTET)(put_buffer >> 56); \
+ buffer[1] = (JOCTET)(put_buffer >> 48); \
+ buffer[2] = (JOCTET)(put_buffer >> 40); \
+ buffer[3] = (JOCTET)(put_buffer >> 32); \
+ buffer[4] = (JOCTET)(put_buffer >> 24); \
+ buffer[5] = (JOCTET)(put_buffer >> 16); \
+ buffer[6] = (JOCTET)(put_buffer >> 8); \
+ buffer[7] = (JOCTET)(put_buffer); \
+ buffer += 8; \
} \
}
-#define CHECKBUF31() { \
- if (put_bits > 31) { \
- EMIT_BYTE() \
- EMIT_BYTE() \
- EMIT_BYTE() \
- EMIT_BYTE() \
- } \
-}
+#else
-#define CHECKBUF47() { \
- if (put_bits > 47) { \
- EMIT_BYTE() \
- EMIT_BYTE() \
- EMIT_BYTE() \
- EMIT_BYTE() \
- EMIT_BYTE() \
- EMIT_BYTE() \
+#define FLUSH() { \
+ if (put_buffer & 0x80808080 & ~(put_buffer + 0x01010101)) { \
+ EMIT_BYTE(put_buffer >> 24) \
+ EMIT_BYTE(put_buffer >> 16) \
+ EMIT_BYTE(put_buffer >> 8) \
+ EMIT_BYTE(put_buffer ) \
+ } else { \
+ buffer[0] = (JOCTET)(put_buffer >> 24); \
+ buffer[1] = (JOCTET)(put_buffer >> 16); \
+ buffer[2] = (JOCTET)(put_buffer >> 8); \
+ buffer[3] = (JOCTET)(put_buffer); \
+ buffer += 4; \
} \
}
-#if !defined(_WIN32) && !defined(SIZEOF_SIZE_T)
-#error Cannot determine word size
#endif
-#if SIZEOF_SIZE_T==8 || defined(_WIN64)
-
-#define EMIT_BITS(code, size) { \
- CHECKBUF47() \
- PUT_BITS(code, size) \
+/* Fill the bit buffer to capacity with the leading bits from code, then output
+ * the bit buffer and put the remaining bits from code into the bit buffer.
+ */
+#define PUT_AND_FLUSH(code, size) { \
+ put_buffer = (put_buffer << (size + free_bits)) | (code >> -free_bits); \
+ FLUSH() \
+ free_bits += BIT_BUF_SIZE; \
+ put_buffer = code; \
}
-#define EMIT_CODE(code, size) { \
- temp2 &= (((JLONG) 1)<<nbits) - 1; \
- CHECKBUF31() \
- PUT_BITS(code, size) \
- PUT_BITS(temp2, nbits) \
- }
-
-#else
-
-#define EMIT_BITS(code, size) { \
- PUT_BITS(code, size) \
- CHECKBUF15() \
+/* Insert code into the bit buffer and output the bit buffer if needed.
+ * NOTE: We can't flush with free_bits == 0, since the left shift in
+ * PUT_AND_FLUSH() would have undefined behavior.
+ */
+#define PUT_BITS(code, size) { \
+ free_bits -= size; \
+ if (free_bits < 0) \
+ PUT_AND_FLUSH(code, size) \
+ else \
+ put_buffer = (put_buffer << size) | code; \
}
-#define EMIT_CODE(code, size) { \
- temp2 &= (((JLONG) 1)<<nbits) - 1; \
- PUT_BITS(code, size) \
- CHECKBUF15() \
- PUT_BITS(temp2, nbits) \
- CHECKBUF15() \
- }
-
-#endif
+#define PUT_CODE(code, size) { \
+ temp &= (((JLONG)1) << nbits) - 1; \
+ temp |= code << nbits; \
+ nbits += size; \
+ PUT_BITS(temp, nbits) \
+}
/* Although it is exceedingly rare, it is possible for a Huffman-encoded
@@ -428,55 +461,81 @@ dump_buffer (working_state *state)
* scanning order-- 1, 8, 16, etc.), then this will produce an encoded block
* larger than 200 bytes.
*/
-#define BUFSIZE (DCTSIZE2 * 4)
+#define BUFSIZE (DCTSIZE2 * 8)
#define LOAD_BUFFER() { \
if (state->free_in_buffer < BUFSIZE) { \
localbuf = 1; \
buffer = _buffer; \
- } \
- else buffer = state->next_output_byte; \
- }
+ } else \
+ buffer = state->next_output_byte; \
+}
#define STORE_BUFFER() { \
if (localbuf) { \
+ size_t bytes, bytestocopy; \
bytes = buffer - _buffer; \
buffer = _buffer; \
while (bytes > 0) { \
- bytestocopy = min(bytes, state->free_in_buffer); \
- MEMCOPY(state->next_output_byte, buffer, bytestocopy); \
+ bytestocopy = MIN(bytes, state->free_in_buffer); \
+ memcpy(state->next_output_byte, buffer, bytestocopy); \
state->next_output_byte += bytestocopy; \
buffer += bytestocopy; \
state->free_in_buffer -= bytestocopy; \
if (state->free_in_buffer == 0) \
- if (! dump_buffer(state)) return FALSE; \
+ if (!dump_buffer(state)) return FALSE; \
bytes -= bytestocopy; \
} \
- } \
- else { \
+ } else { \
state->free_in_buffer -= (buffer - state->next_output_byte); \
state->next_output_byte = buffer; \
} \
- }
+}
LOCAL(boolean)
-flush_bits (working_state *state)
+flush_bits(working_state *state)
{
- JOCTET _buffer[BUFSIZE], *buffer;
- size_t put_buffer; int put_bits;
- size_t bytes, bytestocopy; int localbuf = 0;
+ JOCTET _buffer[BUFSIZE], *buffer, temp;
+ simd_bit_buf_type put_buffer; int put_bits;
+ int localbuf = 0;
+
+ if (state->simd) {
+#if defined(__aarch64__) && !defined(NEON_INTRINSICS)
+ put_bits = state->cur.free_bits;
+#else
+ put_bits = SIMD_BIT_BUF_SIZE - state->cur.free_bits;
+#endif
+ put_buffer = state->cur.put_buffer.simd;
+ } else {
+ put_bits = BIT_BUF_SIZE - state->cur.free_bits;
+ put_buffer = state->cur.put_buffer.c;
+ }
- put_buffer = state->cur.put_buffer;
- put_bits = state->cur.put_bits;
LOAD_BUFFER()
- /* fill any partial byte with ones */
- PUT_BITS(0x7F, 7)
- while (put_bits >= 8) EMIT_BYTE()
+ while (put_bits >= 8) {
+ put_bits -= 8;
+ temp = (JOCTET)(put_buffer >> put_bits);
+ EMIT_BYTE(temp)
+ }
+ if (put_bits) {
+ /* fill partial byte with ones */
+ temp = (JOCTET)((put_buffer << (8 - put_bits)) | (0xFF >> put_bits));
+ EMIT_BYTE(temp)
+ }
- state->cur.put_buffer = 0; /* and reset bit-buffer to empty */
- state->cur.put_bits = 0;
+ if (state->simd) { /* and reset bit buffer to empty */
+ state->cur.put_buffer.simd = 0;
+#if defined(__aarch64__) && !defined(NEON_INTRINSICS)
+ state->cur.free_bits = 0;
+#else
+ state->cur.free_bits = SIMD_BIT_BUF_SIZE;
+#endif
+ } else {
+ state->cur.put_buffer.c = 0;
+ state->cur.free_bits = BIT_BUF_SIZE;
+ }
STORE_BUFFER()
return TRUE;
@@ -486,11 +545,11 @@ flush_bits (working_state *state)
/* Encode a single block's worth of coefficients */
LOCAL(boolean)
-encode_one_block_simd (working_state *state, JCOEFPTR block, int last_dc_val,
- c_derived_tbl *dctbl, c_derived_tbl *actbl)
+encode_one_block_simd(working_state *state, JCOEFPTR block, int last_dc_val,
+ c_derived_tbl *dctbl, c_derived_tbl *actbl)
{
JOCTET _buffer[BUFSIZE], *buffer;
- size_t bytes, bytestocopy; int localbuf = 0;
+ int localbuf = 0;
LOAD_BUFFER()
@@ -503,108 +562,91 @@ encode_one_block_simd (working_state *state, JCOEFPTR block, int last_dc_val,
}
LOCAL(boolean)
-encode_one_block (working_state *state, JCOEFPTR block, int last_dc_val,
- c_derived_tbl *dctbl, c_derived_tbl *actbl)
+encode_one_block(working_state *state, JCOEFPTR block, int last_dc_val,
+ c_derived_tbl *dctbl, c_derived_tbl *actbl)
{
- int temp, temp2, temp3;
- int nbits;
- int r, code, size;
+ int temp, nbits, free_bits;
+ bit_buf_type put_buffer;
JOCTET _buffer[BUFSIZE], *buffer;
- size_t put_buffer; int put_bits;
- int code_0xf0 = actbl->ehufco[0xf0], size_0xf0 = actbl->ehufsi[0xf0];
- size_t bytes, bytestocopy; int localbuf = 0;
+ int localbuf = 0;
- put_buffer = state->cur.put_buffer;
- put_bits = state->cur.put_bits;
+ free_bits = state->cur.free_bits;
+ put_buffer = state->cur.put_buffer.c;
LOAD_BUFFER()
/* Encode the DC coefficient difference per section F.1.2.1 */
- temp = temp2 = block[0] - last_dc_val;
-
- /* This is a well-known technique for obtaining the absolute value without a
- * branch. It is derived from an assembly language technique presented in
- * "How to Optimize for the Pentium Processors", Copyright (c) 1996, 1997 by
- * Agner Fog.
- */
- temp3 = temp >> (CHAR_BIT * sizeof(int) - 1);
- temp ^= temp3;
- temp -= temp3;
+ temp = block[0] - last_dc_val;
- /* For a negative input, want temp2 = bitwise complement of abs(input) */
- /* This code assumes we are on a two's complement machine */
- temp2 += temp3;
+ /* This is a well-known technique for obtaining the absolute value without a
+ * branch. It is derived from an assembly language technique presented in
+ * "How to Optimize for the Pentium Processors", Copyright (c) 1996, 1997 by
+ * Agner Fog. This code assumes we are on a two's complement machine.
+ */
+ nbits = temp >> (CHAR_BIT * sizeof(int) - 1);
+ temp += nbits;
+ nbits ^= temp;
/* Find the number of bits needed for the magnitude of the coefficient */
- nbits = JPEG_NBITS(temp);
-
- /* Emit the Huffman-coded symbol for the number of bits */
- code = dctbl->ehufco[nbits];
- size = dctbl->ehufsi[nbits];
- EMIT_BITS(code, size)
+ nbits = JPEG_NBITS(nbits);
- /* Mask off any extra bits in code */
- temp2 &= (((JLONG) 1)<<nbits) - 1;
-
- /* Emit that number of bits of the value, if positive, */
- /* or the complement of its magnitude, if negative. */
- EMIT_BITS(temp2, nbits)
+ /* Emit the Huffman-coded symbol for the number of bits.
+ * Emit that number of bits of the value, if positive,
+ * or the complement of its magnitude, if negative.
+ */
+ PUT_CODE(dctbl->ehufco[nbits], dctbl->ehufsi[nbits])
/* Encode the AC coefficients per section F.1.2.2 */
- r = 0; /* r = run length of zeros */
+ {
+ int r = 0; /* r = run length of zeros */
/* Manually unroll the k loop to eliminate the counter variable. This
* improves performance greatly on systems with a limited number of
* registers (such as x86.)
*/
-#define kloop(jpeg_natural_order_of_k) { \
+#define kloop(jpeg_natural_order_of_k) { \
if ((temp = block[jpeg_natural_order_of_k]) == 0) { \
- r++; \
+ r += 16; \
} else { \
- temp2 = temp; \
/* Branch-less absolute value, bitwise complement, etc., same as above */ \
- temp3 = temp >> (CHAR_BIT * sizeof(int) - 1); \
- temp ^= temp3; \
- temp -= temp3; \
- temp2 += temp3; \
- nbits = JPEG_NBITS_NONZERO(temp); \
+ nbits = temp >> (CHAR_BIT * sizeof(int) - 1); \
+ temp += nbits; \
+ nbits ^= temp; \
+ nbits = JPEG_NBITS_NONZERO(nbits); \
/* if run length > 15, must emit special run-length-16 codes (0xF0) */ \
- while (r > 15) { \
- EMIT_BITS(code_0xf0, size_0xf0) \
- r -= 16; \
+ while (r >= 16 * 16) { \
+ r -= 16 * 16; \
+ PUT_BITS(actbl->ehufco[0xf0], actbl->ehufsi[0xf0]) \
} \
/* Emit Huffman symbol for run length / number of bits */ \
- temp3 = (r << 4) + nbits; \
- code = actbl->ehufco[temp3]; \
- size = actbl->ehufsi[temp3]; \
- EMIT_CODE(code, size) \
- r = 0; \
+ r += nbits; \
+ PUT_CODE(actbl->ehufco[r], actbl->ehufsi[r]) \
+ r = 0; \
} \
}
- /* One iteration for each value in jpeg_natural_order[] */
- kloop(1); kloop(8); kloop(16); kloop(9); kloop(2); kloop(3);
- kloop(10); kloop(17); kloop(24); kloop(32); kloop(25); kloop(18);
- kloop(11); kloop(4); kloop(5); kloop(12); kloop(19); kloop(26);
- kloop(33); kloop(40); kloop(48); kloop(41); kloop(34); kloop(27);
- kloop(20); kloop(13); kloop(6); kloop(7); kloop(14); kloop(21);
- kloop(28); kloop(35); kloop(42); kloop(49); kloop(56); kloop(57);
- kloop(50); kloop(43); kloop(36); kloop(29); kloop(22); kloop(15);
- kloop(23); kloop(30); kloop(37); kloop(44); kloop(51); kloop(58);
- kloop(59); kloop(52); kloop(45); kloop(38); kloop(31); kloop(39);
- kloop(46); kloop(53); kloop(60); kloop(61); kloop(54); kloop(47);
- kloop(55); kloop(62); kloop(63);
-
- /* If the last coef(s) were zero, emit an end-of-block code */
- if (r > 0) {
- code = actbl->ehufco[0];
- size = actbl->ehufsi[0];
- EMIT_BITS(code, size)
+ /* One iteration for each value in jpeg_natural_order[] */
+ kloop(1); kloop(8); kloop(16); kloop(9); kloop(2); kloop(3);
+ kloop(10); kloop(17); kloop(24); kloop(32); kloop(25); kloop(18);
+ kloop(11); kloop(4); kloop(5); kloop(12); kloop(19); kloop(26);
+ kloop(33); kloop(40); kloop(48); kloop(41); kloop(34); kloop(27);
+ kloop(20); kloop(13); kloop(6); kloop(7); kloop(14); kloop(21);
+ kloop(28); kloop(35); kloop(42); kloop(49); kloop(56); kloop(57);
+ kloop(50); kloop(43); kloop(36); kloop(29); kloop(22); kloop(15);
+ kloop(23); kloop(30); kloop(37); kloop(44); kloop(51); kloop(58);
+ kloop(59); kloop(52); kloop(45); kloop(38); kloop(31); kloop(39);
+ kloop(46); kloop(53); kloop(60); kloop(61); kloop(54); kloop(47);
+ kloop(55); kloop(62); kloop(63);
+
+ /* If the last coef(s) were zero, emit an end-of-block code */
+ if (r > 0) {
+ PUT_BITS(actbl->ehufco[0], actbl->ehufsi[0])
+ }
}
- state->cur.put_buffer = put_buffer;
- state->cur.put_bits = put_bits;
+ state->cur.put_buffer.c = put_buffer;
+ state->cur.free_bits = free_bits;
STORE_BUFFER()
return TRUE;
@@ -616,11 +658,11 @@ encode_one_block (working_state *state, JCOEFPTR block, int last_dc_val,
*/
LOCAL(boolean)
-emit_restart (working_state *state, int restart_num)
+emit_restart(working_state *state, int restart_num)
{
int ci;
- if (! flush_bits(state))
+ if (!flush_bits(state))
return FALSE;
emit_byte(state, 0xFF, return FALSE);
@@ -641,9 +683,9 @@ emit_restart (working_state *state, int restart_num)
*/
METHODDEF(boolean)
-encode_mcu_huff (j_compress_ptr cinfo, JBLOCKROW *MCU_data)
+encode_mcu_huff(j_compress_ptr cinfo, JBLOCKROW *MCU_data)
{
- huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
+ huff_entropy_ptr entropy = (huff_entropy_ptr)cinfo->entropy;
working_state state;
int blkn, ci;
jpeg_component_info *compptr;
@@ -651,13 +693,14 @@ encode_mcu_huff (j_compress_ptr cinfo, JBLOCKROW *MCU_data)
/* Load up working state */
state.next_output_byte = cinfo->dest->next_output_byte;
state.free_in_buffer = cinfo->dest->free_in_buffer;
- ASSIGN_STATE(state.cur, entropy->saved);
+ state.cur = entropy->saved;
state.cinfo = cinfo;
+ state.simd = entropy->simd;
/* Emit restart marker if needed */
if (cinfo->restart_interval) {
if (entropy->restarts_to_go == 0)
- if (! emit_restart(&state, entropy->next_restart_num))
+ if (!emit_restart(&state, entropy->next_restart_num))
return FALSE;
}
@@ -666,10 +709,10 @@ encode_mcu_huff (j_compress_ptr cinfo, JBLOCKROW *MCU_data)
for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
ci = cinfo->MCU_membership[blkn];
compptr = cinfo->cur_comp_info[ci];
- if (! encode_one_block_simd(&state,
- MCU_data[blkn][0], state.cur.last_dc_val[ci],
- entropy->dc_derived_tbls[compptr->dc_tbl_no],
- entropy->ac_derived_tbls[compptr->ac_tbl_no]))
+ if (!encode_one_block_simd(&state,
+ MCU_data[blkn][0], state.cur.last_dc_val[ci],
+ entropy->dc_derived_tbls[compptr->dc_tbl_no],
+ entropy->ac_derived_tbls[compptr->ac_tbl_no]))
return FALSE;
/* Update last_dc_val */
state.cur.last_dc_val[ci] = MCU_data[blkn][0][0];
@@ -678,10 +721,10 @@ encode_mcu_huff (j_compress_ptr cinfo, JBLOCKROW *MCU_data)
for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
ci = cinfo->MCU_membership[blkn];
compptr = cinfo->cur_comp_info[ci];
- if (! encode_one_block(&state,
- MCU_data[blkn][0], state.cur.last_dc_val[ci],
- entropy->dc_derived_tbls[compptr->dc_tbl_no],
- entropy->ac_derived_tbls[compptr->ac_tbl_no]))
+ if (!encode_one_block(&state,
+ MCU_data[blkn][0], state.cur.last_dc_val[ci],
+ entropy->dc_derived_tbls[compptr->dc_tbl_no],
+ entropy->ac_derived_tbls[compptr->ac_tbl_no]))
return FALSE;
/* Update last_dc_val */
state.cur.last_dc_val[ci] = MCU_data[blkn][0][0];
@@ -691,7 +734,7 @@ encode_mcu_huff (j_compress_ptr cinfo, JBLOCKROW *MCU_data)
/* Completed MCU, so update state */
cinfo->dest->next_output_byte = state.next_output_byte;
cinfo->dest->free_in_buffer = state.free_in_buffer;
- ASSIGN_STATE(entropy->saved, state.cur);
+ entropy->saved = state.cur;
/* Update restart-interval state too */
if (cinfo->restart_interval) {
@@ -712,25 +755,26 @@ encode_mcu_huff (j_compress_ptr cinfo, JBLOCKROW *MCU_data)
*/
METHODDEF(void)
-finish_pass_huff (j_compress_ptr cinfo)
+finish_pass_huff(j_compress_ptr cinfo)
{
- huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
+ huff_entropy_ptr entropy = (huff_entropy_ptr)cinfo->entropy;
working_state state;
/* Load up working state ... flush_bits needs it */
state.next_output_byte = cinfo->dest->next_output_byte;
state.free_in_buffer = cinfo->dest->free_in_buffer;
- ASSIGN_STATE(state.cur, entropy->saved);
+ state.cur = entropy->saved;
state.cinfo = cinfo;
+ state.simd = entropy->simd;
/* Flush out the last data */
- if (! flush_bits(&state))
+ if (!flush_bits(&state))
ERREXIT(cinfo, JERR_CANT_SUSPEND);
/* Update state */
cinfo->dest->next_output_byte = state.next_output_byte;
cinfo->dest->free_in_buffer = state.free_in_buffer;
- ASSIGN_STATE(entropy->saved, state.cur);
+ entropy->saved = state.cur;
}
@@ -751,8 +795,8 @@ finish_pass_huff (j_compress_ptr cinfo)
/* Process a single block's worth of coefficients */
LOCAL(void)
-htest_one_block (j_compress_ptr cinfo, JCOEFPTR block, int last_dc_val,
- long dc_counts[], long ac_counts[])
+htest_one_block(j_compress_ptr cinfo, JCOEFPTR block, int last_dc_val,
+ long dc_counts[], long ac_counts[])
{
register int temp;
register int nbits;
@@ -773,7 +817,7 @@ htest_one_block (j_compress_ptr cinfo, JCOEFPTR block, int last_dc_val,
/* Check for out-of-range coefficient values.
* Since we're encoding a difference, the range limit is twice as much.
*/
- if (nbits > MAX_COEF_BITS+1)
+ if (nbits > MAX_COEF_BITS + 1)
ERREXIT(cinfo, JERR_BAD_DCT_COEF);
/* Count the Huffman symbol for the number of bits */
@@ -824,9 +868,9 @@ htest_one_block (j_compress_ptr cinfo, JCOEFPTR block, int last_dc_val,
*/
METHODDEF(boolean)
-encode_mcu_gather (j_compress_ptr cinfo, JBLOCKROW *MCU_data)
+encode_mcu_gather(j_compress_ptr cinfo, JBLOCKROW *MCU_data)
{
- huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
+ huff_entropy_ptr entropy = (huff_entropy_ptr)cinfo->entropy;
int blkn, ci;
jpeg_component_info *compptr;
@@ -863,13 +907,14 @@ encode_mcu_gather (j_compress_ptr cinfo, JBLOCKROW *MCU_data)
* one bits (so that padding bits added at the end of a compressed segment
* can't look like a valid code). Because of the canonical ordering of
* codewords, this just means that there must be an unused slot in the
- * longest codeword length category. Section K.2 of the JPEG spec suggests
- * reserving such a slot by pretending that symbol 256 is a valid symbol
- * with count 1. In theory that's not optimal; giving it count zero but
- * including it in the symbol set anyway should give a better Huffman code.
- * But the theoretically better code actually seems to come out worse in
- * practice, because it produces more all-ones bytes (which incur stuffed
- * zero bytes in the final file). In any case the difference is tiny.
+ * longest codeword length category. Annex K (Clause K.2) of
+ * Rec. ITU-T T.81 (1992) | ISO/IEC 10918-1:1994 suggests reserving such a slot
+ * by pretending that symbol 256 is a valid symbol with count 1. In theory
+ * that's not optimal; giving it count zero but including it in the symbol set
+ * anyway should give a better Huffman code. But the theoretically better code
+ * actually seems to come out worse in practice, because it produces more
+ * all-ones bytes (which incur stuffed zero bytes in the final file). In any
+ * case the difference is tiny.
*
* The JPEG standard requires Huffman codes to be no more than 16 bits long.
* If some symbols have a very small but nonzero probability, the Huffman tree
@@ -884,10 +929,10 @@ encode_mcu_gather (j_compress_ptr cinfo, JBLOCKROW *MCU_data)
*/
GLOBAL(void)
-jpeg_gen_optimal_table (j_compress_ptr cinfo, JHUFF_TBL *htbl, long freq[])
+jpeg_gen_optimal_table(j_compress_ptr cinfo, JHUFF_TBL *htbl, long freq[])
{
-#define MAX_CLEN 32 /* assumed maximum initial code length */
- UINT8 bits[MAX_CLEN+1]; /* bits[k] = # of symbols with code length k */
+#define MAX_CLEN 32 /* assumed maximum initial code length */
+ UINT8 bits[MAX_CLEN + 1]; /* bits[k] = # of symbols with code length k */
int codesize[257]; /* codesize[k] = code length of symbol k */
int others[257]; /* next symbol in current branch of tree */
int c1, c2;
@@ -896,8 +941,8 @@ jpeg_gen_optimal_table (j_compress_ptr cinfo, JHUFF_TBL *htbl, long freq[])
/* This algorithm is explained in section K.2 of the JPEG standard */
- MEMZERO(bits, sizeof(bits));
- MEMZERO(codesize, sizeof(codesize));
+ memset(bits, 0, sizeof(bits));
+ memset(codesize, 0, sizeof(codesize));
for (i = 0; i < 257; i++)
others[i] = -1; /* init links to empty */
@@ -971,13 +1016,13 @@ jpeg_gen_optimal_table (j_compress_ptr cinfo, JHUFF_TBL *htbl, long freq[])
/* JPEG doesn't allow symbols with code lengths over 16 bits, so if the pure
* Huffman procedure assigned any such lengths, we must adjust the coding.
- * Here is what the JPEG spec says about how this next bit works:
- * Since symbols are paired for the longest Huffman code, the symbols are
- * removed from this length category two at a time. The prefix for the pair
- * (which is one bit shorter) is allocated to one of the pair; then,
- * skipping the BITS entry for that prefix length, a code word from the next
- * shortest nonzero BITS entry is converted into a prefix for two code words
- * one bit longer.
+ * Here is what Rec. ITU-T T.81 | ISO/IEC 10918-1 says about how this next
+ * bit works: Since symbols are paired for the longest Huffman code, the
+ * symbols are removed from this length category two at a time. The prefix
+ * for the pair (which is one bit shorter) is allocated to one of the pair;
+ * then, skipping the BITS entry for that prefix length, a code word from the
+ * next shortest nonzero BITS entry is converted into a prefix for two code
+ * words one bit longer.
*/
for (i = MAX_CLEN; i > 16; i--) {
@@ -987,8 +1032,8 @@ jpeg_gen_optimal_table (j_compress_ptr cinfo, JHUFF_TBL *htbl, long freq[])
j--;
bits[i] -= 2; /* remove two symbols */
- bits[i-1]++; /* one goes in this length */
- bits[j+1] += 2; /* two new symbols in this length */
+ bits[i - 1]++; /* one goes in this length */
+ bits[j + 1] += 2; /* two new symbols in this length */
bits[j]--; /* symbol of this length is now a prefix */
}
}
@@ -999,17 +1044,18 @@ jpeg_gen_optimal_table (j_compress_ptr cinfo, JHUFF_TBL *htbl, long freq[])
bits[i]--;
/* Return final symbol counts (only for lengths 0..16) */
- MEMCOPY(htbl->bits, bits, sizeof(htbl->bits));
+ memcpy(htbl->bits, bits, sizeof(htbl->bits));
/* Return a list of the symbols sorted by code length */
/* It's not real clear to me why we don't need to consider the codelength
- * changes made above, but the JPEG spec seems to think this works.
+ * changes made above, but Rec. ITU-T T.81 | ISO/IEC 10918-1 seems to think
+ * this works.
*/
p = 0;
for (i = 1; i <= MAX_CLEN; i++) {
for (j = 0; j <= 255; j++) {
if (codesize[j] == i) {
- htbl->huffval[p] = (UINT8) j;
+ htbl->huffval[p] = (UINT8)j;
p++;
}
}
@@ -1025,9 +1071,9 @@ jpeg_gen_optimal_table (j_compress_ptr cinfo, JHUFF_TBL *htbl, long freq[])
*/
METHODDEF(void)
-finish_pass_gather (j_compress_ptr cinfo)
+finish_pass_gather(j_compress_ptr cinfo)
{
- huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
+ huff_entropy_ptr entropy = (huff_entropy_ptr)cinfo->entropy;
int ci, dctbl, actbl;
jpeg_component_info *compptr;
JHUFF_TBL **htblptr;
@@ -1037,24 +1083,24 @@ finish_pass_gather (j_compress_ptr cinfo)
/* It's important not to apply jpeg_gen_optimal_table more than once
* per table, because it clobbers the input frequency counts!
*/
- MEMZERO(did_dc, sizeof(did_dc));
- MEMZERO(did_ac, sizeof(did_ac));
+ memset(did_dc, 0, sizeof(did_dc));
+ memset(did_ac, 0, sizeof(did_ac));
for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
compptr = cinfo->cur_comp_info[ci];
dctbl = compptr->dc_tbl_no;
actbl = compptr->ac_tbl_no;
- if (! did_dc[dctbl]) {
- htblptr = & cinfo->dc_huff_tbl_ptrs[dctbl];
+ if (!did_dc[dctbl]) {
+ htblptr = &cinfo->dc_huff_tbl_ptrs[dctbl];
if (*htblptr == NULL)
- *htblptr = jpeg_alloc_huff_table((j_common_ptr) cinfo);
+ *htblptr = jpeg_alloc_huff_table((j_common_ptr)cinfo);
jpeg_gen_optimal_table(cinfo, *htblptr, entropy->dc_count_ptrs[dctbl]);
did_dc[dctbl] = TRUE;
}
- if (! did_ac[actbl]) {
- htblptr = & cinfo->ac_huff_tbl_ptrs[actbl];
+ if (!did_ac[actbl]) {
+ htblptr = &cinfo->ac_huff_tbl_ptrs[actbl];
if (*htblptr == NULL)
- *htblptr = jpeg_alloc_huff_table((j_common_ptr) cinfo);
+ *htblptr = jpeg_alloc_huff_table((j_common_ptr)cinfo);
jpeg_gen_optimal_table(cinfo, *htblptr, entropy->ac_count_ptrs[actbl]);
did_ac[actbl] = TRUE;
}
@@ -1070,15 +1116,15 @@ finish_pass_gather (j_compress_ptr cinfo)
*/
GLOBAL(void)
-jinit_huff_encoder (j_compress_ptr cinfo)
+jinit_huff_encoder(j_compress_ptr cinfo)
{
huff_entropy_ptr entropy;
int i;
entropy = (huff_entropy_ptr)
- (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+ (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE,
sizeof(huff_entropy_encoder));
- cinfo->entropy = (struct jpeg_entropy_encoder *) entropy;
+ cinfo->entropy = (struct jpeg_entropy_encoder *)entropy;
entropy->pub.start_pass = start_pass_huff;
/* Mark tables unallocated */
generated by cgit v1.2.3 (git 2.26.2) at 2024-11-09 16:54:32 +0000