mirror of
https://github.com/webmproject/libwebp.git
synced 2024-12-24 12:48:23 +01:00
Make libwebp depend on libsharpyuv.
Change-Id: I6d8ebfe1f855024fc0694b1aa584f71fa27b83ae
This commit is contained in:
Maryla
parent
e4cbcdd2b5
commit
d55d447c9a
@ -33,6 +33,12 @@ else
|
||||
NEON := c
|
||||
endif
|
||||
|
||||
sharpyuv_srcs := \
|
||||
sharpyuv/sharpyuv.c \
|
||||
sharpyuv/sharpyuv_dsp.c \
|
||||
sharpyuv/sharpyuv_neon.$(NEON) \
|
||||
sharpyuv/sharpyuv_sse2.c \
|
||||
|
||||
dec_srcs := \
|
||||
src/dec/alpha_dec.c \
|
||||
src/dec/buffer_dec.c \
|
||||
@ -204,6 +210,7 @@ endif # ENABLE_SHARED=1
|
||||
include $(CLEAR_VARS)
|
||||
|
||||
LOCAL_SRC_FILES := \
|
||||
$(sharpyuv_srcs) \
|
||||
$(dsp_enc_srcs) \
|
||||
$(enc_srcs) \
|
||||
$(utils_enc_srcs) \
|
||||
|
||||
@ -31,7 +31,6 @@ option(WEBP_BUILD_VWEBP "Build the vwebp viewer tool." ON)
|
||||
option(WEBP_BUILD_WEBPINFO "Build the webpinfo command line tool." ON)
|
||||
option(WEBP_BUILD_LIBWEBPMUX "Build the libwebpmux library." ON)
|
||||
option(WEBP_BUILD_WEBPMUX "Build the webpmux command line tool." ON)
|
||||
option(WEBP_BUILD_SHARPYUV "Build the sharpyuv library." OFF)
|
||||
option(WEBP_BUILD_EXTRAS "Build extras." ON)
|
||||
option(WEBP_BUILD_WEBP_JS "Emscripten build of webp.js." OFF)
|
||||
option(WEBP_USE_THREAD "Enable threading support" ON)
|
||||
@ -212,18 +211,16 @@ function(libwebp_add_stub_file TARGET)
|
||||
target_sources(${TARGET} PRIVATE ${stub_source_file})
|
||||
endfunction()
|
||||
|
||||
if(WEBP_BUILD_SHARPYUV)
|
||||
parse_makefile_am(${CMAKE_CURRENT_SOURCE_DIR}/sharpyuv "WEBP_SHARPYUV_SRCS"
|
||||
"")
|
||||
add_library(sharpyuv OBJECT ${WEBP_SHARPYUV_SRCS})
|
||||
target_include_directories(sharpyuv
|
||||
PRIVATE ${CMAKE_CURRENT_BINARY_DIR}
|
||||
${CMAKE_CURRENT_SOURCE_DIR})
|
||||
set_target_properties(
|
||||
sharpyuv
|
||||
PROPERTIES PUBLIC_HEADER "${CMAKE_CURRENT_SOURCE_DIR}/sharpyuv/sharpyuv.h;\
|
||||
parse_makefile_am(${CMAKE_CURRENT_SOURCE_DIR}/sharpyuv "WEBP_SHARPYUV_SRCS"
|
||||
"")
|
||||
add_library(sharpyuv OBJECT ${WEBP_SHARPYUV_SRCS})
|
||||
target_include_directories(sharpyuv
|
||||
PRIVATE ${CMAKE_CURRENT_BINARY_DIR}
|
||||
${CMAKE_CURRENT_SOURCE_DIR})
|
||||
set_target_properties(
|
||||
sharpyuv
|
||||
PROPERTIES PUBLIC_HEADER "${CMAKE_CURRENT_SOURCE_DIR}/sharpyuv/sharpyuv.h;\
|
||||
${CMAKE_CURRENT_SOURCE_DIR}/src/webp/types.h")
|
||||
endif()
|
||||
|
||||
if(MSVC)
|
||||
# avoid security warnings for e.g., fopen() used in the examples.
|
||||
@ -289,6 +286,7 @@ target_include_directories(webputils
|
||||
PRIVATE ${CMAKE_CURRENT_BINARY_DIR}
|
||||
${CMAKE_CURRENT_SOURCE_DIR})
|
||||
add_library(webp
|
||||
$<TARGET_OBJECTS:sharpyuv>
|
||||
$<TARGET_OBJECTS:webpdecode>
|
||||
$<TARGET_OBJECTS:webpdsp>
|
||||
$<TARGET_OBJECTS:webpencode>
|
||||
@ -311,7 +309,8 @@ ${CMAKE_CURRENT_SOURCE_DIR}/src/webp/types.h")
|
||||
|
||||
# Make sure the OBJECT libraries are built with position independent code (it is
|
||||
# not ON by default).
|
||||
set_target_properties(webpdecode
|
||||
set_target_properties(sharpyuv
|
||||
webpdecode
|
||||
webpdspdecode
|
||||
webputilsdecode
|
||||
webpencode
|
||||
|
||||
@ -1,11 +1,7 @@
|
||||
ACLOCAL_AMFLAGS = -I m4
|
||||
SUBDIRS = src imageio man
|
||||
SUBDIRS = sharpyuv src imageio man
|
||||
EXTRA_DIST = COPYING autogen.sh
|
||||
|
||||
if BUILD_SHARPYUV
|
||||
SUBDIRS += sharpyuv
|
||||
endif
|
||||
|
||||
if BUILD_EXTRAS
|
||||
SUBDIRS += extras
|
||||
endif
|
||||
|
||||
13
Makefile.vc
13
Makefile.vc
@ -81,6 +81,7 @@ OUTPUT_DIRS = $(DIRBIN) $(DIRINC) $(DIRLIB) \
|
||||
$(DIROBJ)\extras \
|
||||
$(DIROBJ)\imageio \
|
||||
$(DIROBJ)\mux \
|
||||
$(DIROBJ)\sharpyuv \
|
||||
$(DIROBJ)\utils \
|
||||
|
||||
# Target configuration
|
||||
@ -173,6 +174,12 @@ CFLAGS = $(CFLAGS) /D_UNICODE /DUNICODE
|
||||
# A config was provided, so the library can be built.
|
||||
#
|
||||
|
||||
SHARPYUV_OBJS = \
|
||||
$(DIROBJ)\sharpyuv\sharpyuv.obj \
|
||||
$(DIROBJ)\sharpyuv\sharpyuv_dsp.obj \
|
||||
$(DIROBJ)\sharpyuv\sharpyuv_neon.obj \
|
||||
$(DIROBJ)\sharpyuv\sharpyuv_sse2.obj \
|
||||
|
||||
DEC_OBJS = \
|
||||
$(DIROBJ)\dec\alpha_dec.obj \
|
||||
$(DIROBJ)\dec\buffer_dec.obj \
|
||||
@ -334,8 +341,8 @@ UTILS_ENC_OBJS = \
|
||||
$(DIROBJ)\utils\quant_levels_utils.obj \
|
||||
|
||||
LIBWEBPDECODER_OBJS = $(DEC_OBJS) $(DSP_DEC_OBJS) $(UTILS_DEC_OBJS)
|
||||
LIBWEBP_OBJS = $(LIBWEBPDECODER_OBJS) $(ENC_OBJS) $(DSP_ENC_OBJS) \
|
||||
$(UTILS_ENC_OBJS) $(DLL_OBJS)
|
||||
LIBWEBP_OBJS = $(LIBWEBPDECODER_OBJS) $(SHARPYUV_OBJS) $(ENC_OBJS) \
|
||||
$(DSP_ENC_OBJS) $(UTILS_ENC_OBJS) $(DLL_OBJS)
|
||||
LIBWEBPMUX_OBJS = $(MUX_OBJS) $(LIBWEBPMUX_OBJS)
|
||||
LIBWEBPDEMUX_OBJS = $(DEMUX_OBJS) $(LIBWEBPDEMUX_OBJS)
|
||||
|
||||
@ -481,6 +488,8 @@ $(DIROBJ)\examples\gifdec.obj: examples\gifdec.c
|
||||
$(CC) $(CFLAGS) /Fd$(DIROBJ)\extras\ /Fo$(DIROBJ)\extras\ $<
|
||||
{imageio}.c{$(DIROBJ)\imageio}.obj::
|
||||
$(CC) $(CFLAGS) /Fd$(DIROBJ)\imageio\ /Fo$(DIROBJ)\imageio\ $<
|
||||
{sharpyuv}.c{$(DIROBJ)\sharpyuv}.obj::
|
||||
$(CC) $(CFLAGS) /Fd$(DIROBJ)\sharpyuv\ /Fo$(DIROBJ)\sharpyuv\ $<
|
||||
{src\dec}.c{$(DIROBJ)\dec}.obj::
|
||||
$(CC) $(CFLAGS) /Fd$(LIBWEBP_PDBNAME) /Fo$(DIROBJ)\dec\ $<
|
||||
{src\demux}.c{$(DIROBJ)\demux}.obj::
|
||||
|
||||
@ -105,6 +105,11 @@ model {
|
||||
sources {
|
||||
c {
|
||||
source {
|
||||
srcDir "sharpyuv"
|
||||
include "sharpyuv.c"
|
||||
include "sharpyuv_dsp.c"
|
||||
include "sharpyuv_neon.c"
|
||||
include "sharpyuv_sse2.c"
|
||||
srcDir "src/dec"
|
||||
include "alpha_dec.c"
|
||||
include "buffer_dec.c"
|
||||
|
||||
@ -67,14 +67,6 @@ AC_ARG_ENABLE([libwebpextras],
|
||||
AC_MSG_RESULT(${enable_libwebpextras-no})
|
||||
AM_CONDITIONAL([BUILD_EXTRAS], [test "$enable_libwebpextras" = "yes"])
|
||||
|
||||
dnl === Check whether libsharpyuv should be built
|
||||
AC_MSG_CHECKING(whether libsharpyuv is to be built)
|
||||
AC_ARG_ENABLE([libsharpyuv],
|
||||
AS_HELP_STRING([--enable-libsharpyuv],
|
||||
[Build libsharpyuv @<:@default=no@:>@]))
|
||||
AC_MSG_RESULT(${enable_libsharpyuv-no})
|
||||
AM_CONDITIONAL([BUILD_SHARPYUV], [test "$enable_libsharpyuv" = "yes"])
|
||||
|
||||
dnl === If --enable-asserts is not defined, define NDEBUG
|
||||
|
||||
AC_MSG_CHECKING(whether asserts are enabled)
|
||||
@ -782,7 +774,6 @@ libwebpdecoder: ${enable_libwebpdecoder-no}
|
||||
libwebpdemux: ${enable_libwebpdemux-no}
|
||||
libwebpmux: ${enable_libwebpmux-no}
|
||||
libwebpextras: ${enable_libwebpextras-no}
|
||||
libsharpyuv: ${enable_libsharpyuv-no}
|
||||
|
||||
Tools:
|
||||
cwebp : ${enable_libwebpdemux-no}
|
||||
|
||||
@ -125,6 +125,12 @@ endif
|
||||
ANIM_UTIL_OBJS = \
|
||||
examples/anim_util.o \
|
||||
|
||||
SHARPYUV_OBJS = \
|
||||
sharpyuv/sharpyuv.o \
|
||||
sharpyuv/sharpyuv_dsp.o \
|
||||
sharpyuv/sharpyuv_neon.o \
|
||||
sharpyuv/sharpyuv_sse2.o \
|
||||
|
||||
DEC_OBJS = \
|
||||
src/dec/alpha_dec.o \
|
||||
src/dec/buffer_dec.o \
|
||||
@ -282,8 +288,8 @@ EXTRA_OBJS = \
|
||||
extras/quality_estimate.o \
|
||||
|
||||
LIBWEBPDECODER_OBJS = $(DEC_OBJS) $(DSP_DEC_OBJS) $(UTILS_DEC_OBJS)
|
||||
LIBWEBP_OBJS = $(LIBWEBPDECODER_OBJS) $(ENC_OBJS) $(DSP_ENC_OBJS) \
|
||||
$(UTILS_ENC_OBJS)
|
||||
LIBWEBP_OBJS = $(SHARPYUV_OBJS) $(LIBWEBPDECODER_OBJS) $(ENC_OBJS) \
|
||||
$(DSP_ENC_OBJS) $(UTILS_ENC_OBJS)
|
||||
LIBWEBPMUX_OBJS = $(MUX_OBJS)
|
||||
LIBWEBPDEMUX_OBJS = $(DEMUX_OBJS)
|
||||
LIBWEBPEXTRA_OBJS = $(EXTRA_OBJS)
|
||||
|
||||
@ -332,15 +332,6 @@ extern void WebPConvertARGBToUV_C(const uint32_t* argb, uint8_t* u, uint8_t* v,
|
||||
extern void WebPConvertRGBA32ToUV_C(const uint16_t* rgb,
|
||||
uint8_t* u, uint8_t* v, int width);
|
||||
|
||||
// utilities for accurate RGB->YUV conversion
|
||||
extern uint64_t (*WebPSharpYUVUpdateY)(const uint16_t* src, const uint16_t* ref,
|
||||
uint16_t* dst, int len);
|
||||
extern void (*WebPSharpYUVUpdateRGB)(const int16_t* src, const int16_t* ref,
|
||||
int16_t* dst, int len);
|
||||
extern void (*WebPSharpYUVFilterRow)(const int16_t* A, const int16_t* B,
|
||||
int len,
|
||||
const uint16_t* best_y, uint16_t* out);
|
||||
|
||||
// Must be called before using the above.
|
||||
void WebPInitConvertARGBToYUV(void);
|
||||
|
||||
|
||||
@ -194,50 +194,6 @@ void WebPConvertRGBA32ToUV_C(const uint16_t* rgb,
|
||||
|
||||
//-----------------------------------------------------------------------------
|
||||
|
||||
#if !WEBP_NEON_OMIT_C_CODE
|
||||
#define MAX_Y ((1 << 10) - 1) // 10b precision over 16b-arithmetic
|
||||
static uint16_t clip_y(int v) {
|
||||
return (v < 0) ? 0 : (v > MAX_Y) ? MAX_Y : (uint16_t)v;
|
||||
}
|
||||
|
||||
static uint64_t SharpYUVUpdateY_C(const uint16_t* ref, const uint16_t* src,
|
||||
uint16_t* dst, int len) {
|
||||
uint64_t diff = 0;
|
||||
int i;
|
||||
for (i = 0; i < len; ++i) {
|
||||
const int diff_y = ref[i] - src[i];
|
||||
const int new_y = (int)dst[i] + diff_y;
|
||||
dst[i] = clip_y(new_y);
|
||||
diff += (uint64_t)abs(diff_y);
|
||||
}
|
||||
return diff;
|
||||
}
|
||||
|
||||
static void SharpYUVUpdateRGB_C(const int16_t* ref, const int16_t* src,
|
||||
int16_t* dst, int len) {
|
||||
int i;
|
||||
for (i = 0; i < len; ++i) {
|
||||
const int diff_uv = ref[i] - src[i];
|
||||
dst[i] += diff_uv;
|
||||
}
|
||||
}
|
||||
|
||||
static void SharpYUVFilterRow_C(const int16_t* A, const int16_t* B, int len,
|
||||
const uint16_t* best_y, uint16_t* out) {
|
||||
int i;
|
||||
for (i = 0; i < len; ++i, ++A, ++B) {
|
||||
const int v0 = (A[0] * 9 + A[1] * 3 + B[0] * 3 + B[1] + 8) >> 4;
|
||||
const int v1 = (A[1] * 9 + A[0] * 3 + B[1] * 3 + B[0] + 8) >> 4;
|
||||
out[2 * i + 0] = clip_y(best_y[2 * i + 0] + v0);
|
||||
out[2 * i + 1] = clip_y(best_y[2 * i + 1] + v1);
|
||||
}
|
||||
}
|
||||
#endif // !WEBP_NEON_OMIT_C_CODE
|
||||
|
||||
#undef MAX_Y
|
||||
|
||||
//-----------------------------------------------------------------------------
|
||||
|
||||
void (*WebPConvertRGB24ToY)(const uint8_t* rgb, uint8_t* y, int width);
|
||||
void (*WebPConvertBGR24ToY)(const uint8_t* bgr, uint8_t* y, int width);
|
||||
void (*WebPConvertRGBA32ToUV)(const uint16_t* rgb,
|
||||
@ -247,18 +203,9 @@ void (*WebPConvertARGBToY)(const uint32_t* argb, uint8_t* y, int width);
|
||||
void (*WebPConvertARGBToUV)(const uint32_t* argb, uint8_t* u, uint8_t* v,
|
||||
int src_width, int do_store);
|
||||
|
||||
uint64_t (*WebPSharpYUVUpdateY)(const uint16_t* ref, const uint16_t* src,
|
||||
uint16_t* dst, int len);
|
||||
void (*WebPSharpYUVUpdateRGB)(const int16_t* ref, const int16_t* src,
|
||||
int16_t* dst, int len);
|
||||
void (*WebPSharpYUVFilterRow)(const int16_t* A, const int16_t* B, int len,
|
||||
const uint16_t* best_y, uint16_t* out);
|
||||
|
||||
extern void WebPInitConvertARGBToYUVSSE2(void);
|
||||
extern void WebPInitConvertARGBToYUVSSE41(void);
|
||||
extern void WebPInitConvertARGBToYUVNEON(void);
|
||||
extern void WebPInitSharpYUVSSE2(void);
|
||||
extern void WebPInitSharpYUVNEON(void);
|
||||
|
||||
WEBP_DSP_INIT_FUNC(WebPInitConvertARGBToYUV) {
|
||||
WebPConvertARGBToY = ConvertARGBToY_C;
|
||||
@ -269,17 +216,10 @@ WEBP_DSP_INIT_FUNC(WebPInitConvertARGBToYUV) {
|
||||
|
||||
WebPConvertRGBA32ToUV = WebPConvertRGBA32ToUV_C;
|
||||
|
||||
#if !WEBP_NEON_OMIT_C_CODE
|
||||
WebPSharpYUVUpdateY = SharpYUVUpdateY_C;
|
||||
WebPSharpYUVUpdateRGB = SharpYUVUpdateRGB_C;
|
||||
WebPSharpYUVFilterRow = SharpYUVFilterRow_C;
|
||||
#endif
|
||||
|
||||
if (VP8GetCPUInfo != NULL) {
|
||||
#if defined(WEBP_HAVE_SSE2)
|
||||
if (VP8GetCPUInfo(kSSE2)) {
|
||||
WebPInitConvertARGBToYUVSSE2();
|
||||
WebPInitSharpYUVSSE2();
|
||||
}
|
||||
#endif // WEBP_HAVE_SSE2
|
||||
#if defined(WEBP_HAVE_SSE41)
|
||||
@ -293,7 +233,6 @@ WEBP_DSP_INIT_FUNC(WebPInitConvertARGBToYUV) {
|
||||
if (WEBP_NEON_OMIT_C_CODE ||
|
||||
(VP8GetCPUInfo != NULL && VP8GetCPUInfo(kNEON))) {
|
||||
WebPInitConvertARGBToYUVNEON();
|
||||
WebPInitSharpYUVNEON();
|
||||
}
|
||||
#endif // WEBP_HAVE_NEON
|
||||
|
||||
@ -302,7 +241,4 @@ WEBP_DSP_INIT_FUNC(WebPInitConvertARGBToYUV) {
|
||||
assert(WebPConvertRGB24ToY != NULL);
|
||||
assert(WebPConvertBGR24ToY != NULL);
|
||||
assert(WebPConvertRGBA32ToUV != NULL);
|
||||
assert(WebPSharpYUVUpdateY != NULL);
|
||||
assert(WebPSharpYUVUpdateRGB != NULL);
|
||||
assert(WebPSharpYUVFilterRow != NULL);
|
||||
}
|
||||
|
||||
@ -173,116 +173,8 @@ WEBP_TSAN_IGNORE_FUNCTION void WebPInitConvertARGBToYUVNEON(void) {
|
||||
WebPConvertRGBA32ToUV = ConvertRGBA32ToUV_NEON;
|
||||
}
|
||||
|
||||
//------------------------------------------------------------------------------
|
||||
|
||||
#define MAX_Y ((1 << 10) - 1) // 10b precision over 16b-arithmetic
|
||||
static uint16_t clip_y_NEON(int v) {
|
||||
return (v < 0) ? 0 : (v > MAX_Y) ? MAX_Y : (uint16_t)v;
|
||||
}
|
||||
|
||||
static uint64_t SharpYUVUpdateY_NEON(const uint16_t* ref, const uint16_t* src,
|
||||
uint16_t* dst, int len) {
|
||||
int i;
|
||||
const int16x8_t zero = vdupq_n_s16(0);
|
||||
const int16x8_t max = vdupq_n_s16(MAX_Y);
|
||||
uint64x2_t sum = vdupq_n_u64(0);
|
||||
uint64_t diff;
|
||||
|
||||
for (i = 0; i + 8 <= len; i += 8) {
|
||||
const int16x8_t A = vreinterpretq_s16_u16(vld1q_u16(ref + i));
|
||||
const int16x8_t B = vreinterpretq_s16_u16(vld1q_u16(src + i));
|
||||
const int16x8_t C = vreinterpretq_s16_u16(vld1q_u16(dst + i));
|
||||
const int16x8_t D = vsubq_s16(A, B); // diff_y
|
||||
const int16x8_t F = vaddq_s16(C, D); // new_y
|
||||
const uint16x8_t H =
|
||||
vreinterpretq_u16_s16(vmaxq_s16(vminq_s16(F, max), zero));
|
||||
const int16x8_t I = vabsq_s16(D); // abs(diff_y)
|
||||
vst1q_u16(dst + i, H);
|
||||
sum = vpadalq_u32(sum, vpaddlq_u16(vreinterpretq_u16_s16(I)));
|
||||
}
|
||||
diff = vgetq_lane_u64(sum, 0) + vgetq_lane_u64(sum, 1);
|
||||
for (; i < len; ++i) {
|
||||
const int diff_y = ref[i] - src[i];
|
||||
const int new_y = (int)(dst[i]) + diff_y;
|
||||
dst[i] = clip_y_NEON(new_y);
|
||||
diff += (uint64_t)(abs(diff_y));
|
||||
}
|
||||
return diff;
|
||||
}
|
||||
|
||||
static void SharpYUVUpdateRGB_NEON(const int16_t* ref, const int16_t* src,
|
||||
int16_t* dst, int len) {
|
||||
int i;
|
||||
for (i = 0; i + 8 <= len; i += 8) {
|
||||
const int16x8_t A = vld1q_s16(ref + i);
|
||||
const int16x8_t B = vld1q_s16(src + i);
|
||||
const int16x8_t C = vld1q_s16(dst + i);
|
||||
const int16x8_t D = vsubq_s16(A, B); // diff_uv
|
||||
const int16x8_t E = vaddq_s16(C, D); // new_uv
|
||||
vst1q_s16(dst + i, E);
|
||||
}
|
||||
for (; i < len; ++i) {
|
||||
const int diff_uv = ref[i] - src[i];
|
||||
dst[i] += diff_uv;
|
||||
}
|
||||
}
|
||||
|
||||
static void SharpYUVFilterRow_NEON(const int16_t* A, const int16_t* B, int len,
|
||||
const uint16_t* best_y, uint16_t* out) {
|
||||
int i;
|
||||
const int16x8_t max = vdupq_n_s16(MAX_Y);
|
||||
const int16x8_t zero = vdupq_n_s16(0);
|
||||
for (i = 0; i + 8 <= len; i += 8) {
|
||||
const int16x8_t a0 = vld1q_s16(A + i + 0);
|
||||
const int16x8_t a1 = vld1q_s16(A + i + 1);
|
||||
const int16x8_t b0 = vld1q_s16(B + i + 0);
|
||||
const int16x8_t b1 = vld1q_s16(B + i + 1);
|
||||
const int16x8_t a0b1 = vaddq_s16(a0, b1);
|
||||
const int16x8_t a1b0 = vaddq_s16(a1, b0);
|
||||
const int16x8_t a0a1b0b1 = vaddq_s16(a0b1, a1b0); // A0+A1+B0+B1
|
||||
const int16x8_t a0b1_2 = vaddq_s16(a0b1, a0b1); // 2*(A0+B1)
|
||||
const int16x8_t a1b0_2 = vaddq_s16(a1b0, a1b0); // 2*(A1+B0)
|
||||
const int16x8_t c0 = vshrq_n_s16(vaddq_s16(a0b1_2, a0a1b0b1), 3);
|
||||
const int16x8_t c1 = vshrq_n_s16(vaddq_s16(a1b0_2, a0a1b0b1), 3);
|
||||
const int16x8_t d0 = vaddq_s16(c1, a0);
|
||||
const int16x8_t d1 = vaddq_s16(c0, a1);
|
||||
const int16x8_t e0 = vrshrq_n_s16(d0, 1);
|
||||
const int16x8_t e1 = vrshrq_n_s16(d1, 1);
|
||||
const int16x8x2_t f = vzipq_s16(e0, e1);
|
||||
const int16x8_t g0 = vreinterpretq_s16_u16(vld1q_u16(best_y + 2 * i + 0));
|
||||
const int16x8_t g1 = vreinterpretq_s16_u16(vld1q_u16(best_y + 2 * i + 8));
|
||||
const int16x8_t h0 = vaddq_s16(g0, f.val[0]);
|
||||
const int16x8_t h1 = vaddq_s16(g1, f.val[1]);
|
||||
const int16x8_t i0 = vmaxq_s16(vminq_s16(h0, max), zero);
|
||||
const int16x8_t i1 = vmaxq_s16(vminq_s16(h1, max), zero);
|
||||
vst1q_u16(out + 2 * i + 0, vreinterpretq_u16_s16(i0));
|
||||
vst1q_u16(out + 2 * i + 8, vreinterpretq_u16_s16(i1));
|
||||
}
|
||||
for (; i < len; ++i) {
|
||||
const int a0b1 = A[i + 0] + B[i + 1];
|
||||
const int a1b0 = A[i + 1] + B[i + 0];
|
||||
const int a0a1b0b1 = a0b1 + a1b0 + 8;
|
||||
const int v0 = (8 * A[i + 0] + 2 * a1b0 + a0a1b0b1) >> 4;
|
||||
const int v1 = (8 * A[i + 1] + 2 * a0b1 + a0a1b0b1) >> 4;
|
||||
out[2 * i + 0] = clip_y_NEON(best_y[2 * i + 0] + v0);
|
||||
out[2 * i + 1] = clip_y_NEON(best_y[2 * i + 1] + v1);
|
||||
}
|
||||
}
|
||||
#undef MAX_Y
|
||||
|
||||
//------------------------------------------------------------------------------
|
||||
|
||||
extern void WebPInitSharpYUVNEON(void);
|
||||
|
||||
WEBP_TSAN_IGNORE_FUNCTION void WebPInitSharpYUVNEON(void) {
|
||||
WebPSharpYUVUpdateY = SharpYUVUpdateY_NEON;
|
||||
WebPSharpYUVUpdateRGB = SharpYUVUpdateRGB_NEON;
|
||||
WebPSharpYUVFilterRow = SharpYUVFilterRow_NEON;
|
||||
}
|
||||
|
||||
#else // !WEBP_USE_NEON
|
||||
|
||||
WEBP_DSP_INIT_STUB(WebPInitConvertARGBToYUVNEON)
|
||||
WEBP_DSP_INIT_STUB(WebPInitSharpYUVNEON)
|
||||
|
||||
#endif // WEBP_USE_NEON
|
||||
|
||||
@ -747,128 +747,9 @@ WEBP_TSAN_IGNORE_FUNCTION void WebPInitConvertARGBToYUVSSE2(void) {
|
||||
WebPConvertRGBA32ToUV = ConvertRGBA32ToUV_SSE2;
|
||||
}
|
||||
|
||||
//------------------------------------------------------------------------------
|
||||
|
||||
#define MAX_Y ((1 << 10) - 1) // 10b precision over 16b-arithmetic
|
||||
static uint16_t clip_y(int v) {
|
||||
return (v < 0) ? 0 : (v > MAX_Y) ? MAX_Y : (uint16_t)v;
|
||||
}
|
||||
|
||||
static uint64_t SharpYUVUpdateY_SSE2(const uint16_t* ref, const uint16_t* src,
|
||||
uint16_t* dst, int len) {
|
||||
uint64_t diff = 0;
|
||||
uint32_t tmp[4];
|
||||
int i;
|
||||
const __m128i zero = _mm_setzero_si128();
|
||||
const __m128i max = _mm_set1_epi16(MAX_Y);
|
||||
const __m128i one = _mm_set1_epi16(1);
|
||||
__m128i sum = zero;
|
||||
|
||||
for (i = 0; i + 8 <= len; i += 8) {
|
||||
const __m128i A = _mm_loadu_si128((const __m128i*)(ref + i));
|
||||
const __m128i B = _mm_loadu_si128((const __m128i*)(src + i));
|
||||
const __m128i C = _mm_loadu_si128((const __m128i*)(dst + i));
|
||||
const __m128i D = _mm_sub_epi16(A, B); // diff_y
|
||||
const __m128i E = _mm_cmpgt_epi16(zero, D); // sign (-1 or 0)
|
||||
const __m128i F = _mm_add_epi16(C, D); // new_y
|
||||
const __m128i G = _mm_or_si128(E, one); // -1 or 1
|
||||
const __m128i H = _mm_max_epi16(_mm_min_epi16(F, max), zero);
|
||||
const __m128i I = _mm_madd_epi16(D, G); // sum(abs(...))
|
||||
_mm_storeu_si128((__m128i*)(dst + i), H);
|
||||
sum = _mm_add_epi32(sum, I);
|
||||
}
|
||||
_mm_storeu_si128((__m128i*)tmp, sum);
|
||||
diff = tmp[3] + tmp[2] + tmp[1] + tmp[0];
|
||||
for (; i < len; ++i) {
|
||||
const int diff_y = ref[i] - src[i];
|
||||
const int new_y = (int)dst[i] + diff_y;
|
||||
dst[i] = clip_y(new_y);
|
||||
diff += (uint64_t)abs(diff_y);
|
||||
}
|
||||
return diff;
|
||||
}
|
||||
|
||||
static void SharpYUVUpdateRGB_SSE2(const int16_t* ref, const int16_t* src,
|
||||
int16_t* dst, int len) {
|
||||
int i = 0;
|
||||
for (i = 0; i + 8 <= len; i += 8) {
|
||||
const __m128i A = _mm_loadu_si128((const __m128i*)(ref + i));
|
||||
const __m128i B = _mm_loadu_si128((const __m128i*)(src + i));
|
||||
const __m128i C = _mm_loadu_si128((const __m128i*)(dst + i));
|
||||
const __m128i D = _mm_sub_epi16(A, B); // diff_uv
|
||||
const __m128i E = _mm_add_epi16(C, D); // new_uv
|
||||
_mm_storeu_si128((__m128i*)(dst + i), E);
|
||||
}
|
||||
for (; i < len; ++i) {
|
||||
const int diff_uv = ref[i] - src[i];
|
||||
dst[i] += diff_uv;
|
||||
}
|
||||
}
|
||||
|
||||
static void SharpYUVFilterRow_SSE2(const int16_t* A, const int16_t* B, int len,
|
||||
const uint16_t* best_y, uint16_t* out) {
|
||||
int i;
|
||||
const __m128i kCst8 = _mm_set1_epi16(8);
|
||||
const __m128i max = _mm_set1_epi16(MAX_Y);
|
||||
const __m128i zero = _mm_setzero_si128();
|
||||
for (i = 0; i + 8 <= len; i += 8) {
|
||||
const __m128i a0 = _mm_loadu_si128((const __m128i*)(A + i + 0));
|
||||
const __m128i a1 = _mm_loadu_si128((const __m128i*)(A + i + 1));
|
||||
const __m128i b0 = _mm_loadu_si128((const __m128i*)(B + i + 0));
|
||||
const __m128i b1 = _mm_loadu_si128((const __m128i*)(B + i + 1));
|
||||
const __m128i a0b1 = _mm_add_epi16(a0, b1);
|
||||
const __m128i a1b0 = _mm_add_epi16(a1, b0);
|
||||
const __m128i a0a1b0b1 = _mm_add_epi16(a0b1, a1b0); // A0+A1+B0+B1
|
||||
const __m128i a0a1b0b1_8 = _mm_add_epi16(a0a1b0b1, kCst8);
|
||||
const __m128i a0b1_2 = _mm_add_epi16(a0b1, a0b1); // 2*(A0+B1)
|
||||
const __m128i a1b0_2 = _mm_add_epi16(a1b0, a1b0); // 2*(A1+B0)
|
||||
const __m128i c0 = _mm_srai_epi16(_mm_add_epi16(a0b1_2, a0a1b0b1_8), 3);
|
||||
const __m128i c1 = _mm_srai_epi16(_mm_add_epi16(a1b0_2, a0a1b0b1_8), 3);
|
||||
const __m128i d0 = _mm_add_epi16(c1, a0);
|
||||
const __m128i d1 = _mm_add_epi16(c0, a1);
|
||||
const __m128i e0 = _mm_srai_epi16(d0, 1);
|
||||
const __m128i e1 = _mm_srai_epi16(d1, 1);
|
||||
const __m128i f0 = _mm_unpacklo_epi16(e0, e1);
|
||||
const __m128i f1 = _mm_unpackhi_epi16(e0, e1);
|
||||
const __m128i g0 = _mm_loadu_si128((const __m128i*)(best_y + 2 * i + 0));
|
||||
const __m128i g1 = _mm_loadu_si128((const __m128i*)(best_y + 2 * i + 8));
|
||||
const __m128i h0 = _mm_add_epi16(g0, f0);
|
||||
const __m128i h1 = _mm_add_epi16(g1, f1);
|
||||
const __m128i i0 = _mm_max_epi16(_mm_min_epi16(h0, max), zero);
|
||||
const __m128i i1 = _mm_max_epi16(_mm_min_epi16(h1, max), zero);
|
||||
_mm_storeu_si128((__m128i*)(out + 2 * i + 0), i0);
|
||||
_mm_storeu_si128((__m128i*)(out + 2 * i + 8), i1);
|
||||
}
|
||||
for (; i < len; ++i) {
|
||||
// (9 * A0 + 3 * A1 + 3 * B0 + B1 + 8) >> 4 =
|
||||
// = (8 * A0 + 2 * (A1 + B0) + (A0 + A1 + B0 + B1 + 8)) >> 4
|
||||
// We reuse the common sub-expressions.
|
||||
const int a0b1 = A[i + 0] + B[i + 1];
|
||||
const int a1b0 = A[i + 1] + B[i + 0];
|
||||
const int a0a1b0b1 = a0b1 + a1b0 + 8;
|
||||
const int v0 = (8 * A[i + 0] + 2 * a1b0 + a0a1b0b1) >> 4;
|
||||
const int v1 = (8 * A[i + 1] + 2 * a0b1 + a0a1b0b1) >> 4;
|
||||
out[2 * i + 0] = clip_y(best_y[2 * i + 0] + v0);
|
||||
out[2 * i + 1] = clip_y(best_y[2 * i + 1] + v1);
|
||||
}
|
||||
}
|
||||
|
||||
#undef MAX_Y
|
||||
|
||||
//------------------------------------------------------------------------------
|
||||
|
||||
extern void WebPInitSharpYUVSSE2(void);
|
||||
|
||||
WEBP_TSAN_IGNORE_FUNCTION void WebPInitSharpYUVSSE2(void) {
|
||||
WebPSharpYUVUpdateY = SharpYUVUpdateY_SSE2;
|
||||
WebPSharpYUVUpdateRGB = SharpYUVUpdateRGB_SSE2;
|
||||
WebPSharpYUVFilterRow = SharpYUVFilterRow_SSE2;
|
||||
}
|
||||
|
||||
#else // !WEBP_USE_SSE2
|
||||
|
||||
WEBP_DSP_INIT_STUB(WebPInitSamplersSSE2)
|
||||
WEBP_DSP_INIT_STUB(WebPInitConvertARGBToYUVSSE2)
|
||||
WEBP_DSP_INIT_STUB(WebPInitSharpYUVSSE2)
|
||||
|
||||
#endif // WEBP_USE_SSE2
|
||||
|
||||
@ -37,6 +37,7 @@ libwebpencodeinclude_HEADERS += ../webp/types.h
|
||||
noinst_HEADERS =
|
||||
noinst_HEADERS += ../webp/format_constants.h
|
||||
|
||||
libwebpencode_la_LIBADD = ../../sharpyuv/libsharpyuv.la
|
||||
libwebpencode_la_LDFLAGS = -lm
|
||||
libwebpencode_la_CPPFLAGS = $(AM_CPPFLAGS)
|
||||
libwebpencodeincludedir = $(includedir)/webp
|
||||
|
||||
@ -15,6 +15,7 @@
|
||||
#include <stdlib.h>
|
||||
#include <math.h>
|
||||
|
||||
#include "sharpyuv/sharpyuv.h"
|
||||
#include "src/enc/vp8i_enc.h"
|
||||
#include "src/utils/random_utils.h"
|
||||
#include "src/utils/utils.h"
|
||||
@ -158,7 +159,6 @@ static int RGBToV(int r, int g, int b, VP8Random* const rg) {
|
||||
//------------------------------------------------------------------------------
|
||||
// Sharp RGB->YUV conversion
|
||||
|
||||
static const int kNumIterations = 4;
|
||||
static const int kMinDimensionIterativeConversion = 4;
|
||||
|
||||
// We could use SFIX=0 and only uint8_t for fixed_y_t, but it produces some
|
||||
@ -171,402 +171,23 @@ typedef uint16_t fixed_y_t; // unsigned type with extra SFIX precision for W
|
||||
#define MAX_Y_T ((256 << SFIX) - 1)
|
||||
#define SROUNDER (1 << (YUV_FIX + SFIX - 1))
|
||||
|
||||
#if defined(USE_GAMMA_COMPRESSION)
|
||||
|
||||
// We use tables of different size and precision for the Rec709 / BT2020
|
||||
// transfer function.
|
||||
#define kGammaF (1./0.45)
|
||||
static uint32_t kLinearToGammaTabS[kGammaTabSize + 2];
|
||||
#define GAMMA_TO_LINEAR_BITS 14
|
||||
static uint32_t kGammaToLinearTabS[MAX_Y_T + 1]; // size scales with Y_FIX
|
||||
static volatile int kGammaTablesSOk = 0;
|
||||
static void InitGammaTablesS(void);
|
||||
|
||||
WEBP_DSP_INIT_FUNC(InitGammaTablesS) {
|
||||
assert(2 * GAMMA_TO_LINEAR_BITS < 32); // we use uint32_t intermediate values
|
||||
if (!kGammaTablesSOk) {
|
||||
int v;
|
||||
const double norm = 1. / MAX_Y_T;
|
||||
const double scale = 1. / kGammaTabSize;
|
||||
const double a = 0.09929682680944;
|
||||
const double thresh = 0.018053968510807;
|
||||
const double final_scale = 1 << GAMMA_TO_LINEAR_BITS;
|
||||
for (v = 0; v <= MAX_Y_T; ++v) {
|
||||
const double g = norm * v;
|
||||
double value;
|
||||
if (g <= thresh * 4.5) {
|
||||
value = g / 4.5;
|
||||
} else {
|
||||
const double a_rec = 1. / (1. + a);
|
||||
value = pow(a_rec * (g + a), kGammaF);
|
||||
}
|
||||
kGammaToLinearTabS[v] = (uint32_t)(value * final_scale + .5);
|
||||
}
|
||||
for (v = 0; v <= kGammaTabSize; ++v) {
|
||||
const double g = scale * v;
|
||||
double value;
|
||||
if (g <= thresh) {
|
||||
value = 4.5 * g;
|
||||
} else {
|
||||
value = (1. + a) * pow(g, 1. / kGammaF) - a;
|
||||
}
|
||||
// we already incorporate the 1/2 rounding constant here
|
||||
kLinearToGammaTabS[v] =
|
||||
(uint32_t)(MAX_Y_T * value) + (1 << GAMMA_TO_LINEAR_BITS >> 1);
|
||||
}
|
||||
// to prevent small rounding errors to cause read-overflow:
|
||||
kLinearToGammaTabS[kGammaTabSize + 1] = kLinearToGammaTabS[kGammaTabSize];
|
||||
kGammaTablesSOk = 1;
|
||||
}
|
||||
}
|
||||
|
||||
// return value has a fixed-point precision of GAMMA_TO_LINEAR_BITS
|
||||
static WEBP_INLINE uint32_t GammaToLinearS(int v) {
|
||||
return kGammaToLinearTabS[v];
|
||||
}
|
||||
|
||||
static WEBP_INLINE uint32_t LinearToGammaS(uint32_t value) {
|
||||
// 'value' is in GAMMA_TO_LINEAR_BITS fractional precision
|
||||
const uint32_t v = value * kGammaTabSize;
|
||||
const uint32_t tab_pos = v >> GAMMA_TO_LINEAR_BITS;
|
||||
// fractional part, in GAMMA_TO_LINEAR_BITS fixed-point precision
|
||||
const uint32_t x = v - (tab_pos << GAMMA_TO_LINEAR_BITS); // fractional part
|
||||
// v0 / v1 are in GAMMA_TO_LINEAR_BITS fixed-point precision (range [0..1])
|
||||
const uint32_t v0 = kLinearToGammaTabS[tab_pos + 0];
|
||||
const uint32_t v1 = kLinearToGammaTabS[tab_pos + 1];
|
||||
// Final interpolation. Note that rounding is already included.
|
||||
const uint32_t v2 = (v1 - v0) * x; // note: v1 >= v0.
|
||||
const uint32_t result = v0 + (v2 >> GAMMA_TO_LINEAR_BITS);
|
||||
return result;
|
||||
}
|
||||
|
||||
#else
|
||||
|
||||
static void InitGammaTablesS(void) {}
|
||||
static WEBP_INLINE uint32_t GammaToLinearS(int v) {
|
||||
return (v << GAMMA_TO_LINEAR_BITS) / MAX_Y_T;
|
||||
}
|
||||
static WEBP_INLINE uint32_t LinearToGammaS(uint32_t value) {
|
||||
return (MAX_Y_T * value) >> GAMMA_TO_LINEAR_BITS;
|
||||
}
|
||||
|
||||
#endif // USE_GAMMA_COMPRESSION
|
||||
|
||||
//------------------------------------------------------------------------------
|
||||
|
||||
static uint8_t clip_8b(fixed_t v) {
|
||||
return (!(v & ~0xff)) ? (uint8_t)v : (v < 0) ? 0u : 255u;
|
||||
}
|
||||
|
||||
static fixed_y_t clip_y(int y) {
|
||||
return (!(y & ~MAX_Y_T)) ? (fixed_y_t)y : (y < 0) ? 0 : MAX_Y_T;
|
||||
}
|
||||
|
||||
//------------------------------------------------------------------------------
|
||||
|
||||
static int RGBToGray(int r, int g, int b) {
|
||||
const int luma = 13933 * r + 46871 * g + 4732 * b + YUV_HALF;
|
||||
return (luma >> YUV_FIX);
|
||||
}
|
||||
|
||||
static uint32_t ScaleDown(int a, int b, int c, int d) {
|
||||
const uint32_t A = GammaToLinearS(a);
|
||||
const uint32_t B = GammaToLinearS(b);
|
||||
const uint32_t C = GammaToLinearS(c);
|
||||
const uint32_t D = GammaToLinearS(d);
|
||||
return LinearToGammaS((A + B + C + D + 2) >> 2);
|
||||
}
|
||||
|
||||
static WEBP_INLINE void UpdateW(const fixed_y_t* src, fixed_y_t* dst, int w) {
|
||||
int i;
|
||||
for (i = 0; i < w; ++i) {
|
||||
const uint32_t R = GammaToLinearS(src[0 * w + i]);
|
||||
const uint32_t G = GammaToLinearS(src[1 * w + i]);
|
||||
const uint32_t B = GammaToLinearS(src[2 * w + i]);
|
||||
const uint32_t Y = RGBToGray(R, G, B);
|
||||
dst[i] = (fixed_y_t)LinearToGammaS(Y);
|
||||
}
|
||||
}
|
||||
|
||||
static void UpdateChroma(const fixed_y_t* src1, const fixed_y_t* src2,
|
||||
fixed_t* dst, int uv_w) {
|
||||
int i;
|
||||
for (i = 0; i < uv_w; ++i) {
|
||||
const int r = ScaleDown(src1[0 * uv_w + 0], src1[0 * uv_w + 1],
|
||||
src2[0 * uv_w + 0], src2[0 * uv_w + 1]);
|
||||
const int g = ScaleDown(src1[2 * uv_w + 0], src1[2 * uv_w + 1],
|
||||
src2[2 * uv_w + 0], src2[2 * uv_w + 1]);
|
||||
const int b = ScaleDown(src1[4 * uv_w + 0], src1[4 * uv_w + 1],
|
||||
src2[4 * uv_w + 0], src2[4 * uv_w + 1]);
|
||||
const int W = RGBToGray(r, g, b);
|
||||
dst[0 * uv_w] = (fixed_t)(r - W);
|
||||
dst[1 * uv_w] = (fixed_t)(g - W);
|
||||
dst[2 * uv_w] = (fixed_t)(b - W);
|
||||
dst += 1;
|
||||
src1 += 2;
|
||||
src2 += 2;
|
||||
}
|
||||
}
|
||||
|
||||
static void StoreGray(const fixed_y_t* rgb, fixed_y_t* y, int w) {
|
||||
int i;
|
||||
for (i = 0; i < w; ++i) {
|
||||
y[i] = RGBToGray(rgb[0 * w + i], rgb[1 * w + i], rgb[2 * w + i]);
|
||||
}
|
||||
}
|
||||
|
||||
//------------------------------------------------------------------------------
|
||||
|
||||
static WEBP_INLINE fixed_y_t Filter2(int A, int B, int W0) {
|
||||
const int v0 = (A * 3 + B + 2) >> 2;
|
||||
return clip_y(v0 + W0);
|
||||
}
|
||||
|
||||
//------------------------------------------------------------------------------
|
||||
|
||||
static WEBP_INLINE fixed_y_t UpLift(uint8_t a) { // 8bit -> SFIX
|
||||
return ((fixed_y_t)a << SFIX) | SHALF;
|
||||
}
|
||||
|
||||
static void ImportOneRow(const uint8_t* const r_ptr,
|
||||
const uint8_t* const g_ptr,
|
||||
const uint8_t* const b_ptr,
|
||||
int step,
|
||||
int pic_width,
|
||||
fixed_y_t* const dst) {
|
||||
int i;
|
||||
const int w = (pic_width + 1) & ~1;
|
||||
for (i = 0; i < pic_width; ++i) {
|
||||
const int off = i * step;
|
||||
dst[i + 0 * w] = UpLift(r_ptr[off]);
|
||||
dst[i + 1 * w] = UpLift(g_ptr[off]);
|
||||
dst[i + 2 * w] = UpLift(b_ptr[off]);
|
||||
}
|
||||
if (pic_width & 1) { // replicate rightmost pixel
|
||||
dst[pic_width + 0 * w] = dst[pic_width + 0 * w - 1];
|
||||
dst[pic_width + 1 * w] = dst[pic_width + 1 * w - 1];
|
||||
dst[pic_width + 2 * w] = dst[pic_width + 2 * w - 1];
|
||||
}
|
||||
}
|
||||
|
||||
static void InterpolateTwoRows(const fixed_y_t* const best_y,
|
||||
const fixed_t* prev_uv,
|
||||
const fixed_t* cur_uv,
|
||||
const fixed_t* next_uv,
|
||||
int w,
|
||||
fixed_y_t* out1,
|
||||
fixed_y_t* out2) {
|
||||
const int uv_w = w >> 1;
|
||||
const int len = (w - 1) >> 1; // length to filter
|
||||
int k = 3;
|
||||
while (k-- > 0) { // process each R/G/B segments in turn
|
||||
// special boundary case for i==0
|
||||
out1[0] = Filter2(cur_uv[0], prev_uv[0], best_y[0]);
|
||||
out2[0] = Filter2(cur_uv[0], next_uv[0], best_y[w]);
|
||||
|
||||
WebPSharpYUVFilterRow(cur_uv, prev_uv, len, best_y + 0 + 1, out1 + 1);
|
||||
WebPSharpYUVFilterRow(cur_uv, next_uv, len, best_y + w + 1, out2 + 1);
|
||||
|
||||
// special boundary case for i == w - 1 when w is even
|
||||
if (!(w & 1)) {
|
||||
out1[w - 1] = Filter2(cur_uv[uv_w - 1], prev_uv[uv_w - 1],
|
||||
best_y[w - 1 + 0]);
|
||||
out2[w - 1] = Filter2(cur_uv[uv_w - 1], next_uv[uv_w - 1],
|
||||
best_y[w - 1 + w]);
|
||||
}
|
||||
out1 += w;
|
||||
out2 += w;
|
||||
prev_uv += uv_w;
|
||||
cur_uv += uv_w;
|
||||
next_uv += uv_w;
|
||||
}
|
||||
}
|
||||
|
||||
static WEBP_INLINE uint8_t ConvertRGBToY(int r, int g, int b) {
|
||||
const int luma = 16839 * r + 33059 * g + 6420 * b + SROUNDER;
|
||||
return clip_8b(16 + (luma >> (YUV_FIX + SFIX)));
|
||||
}
|
||||
|
||||
static WEBP_INLINE uint8_t ConvertRGBToU(int r, int g, int b) {
|
||||
const int u = -9719 * r - 19081 * g + 28800 * b + SROUNDER;
|
||||
return clip_8b(128 + (u >> (YUV_FIX + SFIX)));
|
||||
}
|
||||
|
||||
static WEBP_INLINE uint8_t ConvertRGBToV(int r, int g, int b) {
|
||||
const int v = +28800 * r - 24116 * g - 4684 * b + SROUNDER;
|
||||
return clip_8b(128 + (v >> (YUV_FIX + SFIX)));
|
||||
}
|
||||
|
||||
static int ConvertWRGBToYUV(const fixed_y_t* best_y, const fixed_t* best_uv,
|
||||
WebPPicture* const picture) {
|
||||
int i, j;
|
||||
uint8_t* dst_y = picture->y;
|
||||
uint8_t* dst_u = picture->u;
|
||||
uint8_t* dst_v = picture->v;
|
||||
const fixed_t* const best_uv_base = best_uv;
|
||||
const int w = (picture->width + 1) & ~1;
|
||||
const int h = (picture->height + 1) & ~1;
|
||||
const int uv_w = w >> 1;
|
||||
const int uv_h = h >> 1;
|
||||
for (best_uv = best_uv_base, j = 0; j < picture->height; ++j) {
|
||||
for (i = 0; i < picture->width; ++i) {
|
||||
const int off = (i >> 1);
|
||||
const int W = best_y[i];
|
||||
const int r = best_uv[off + 0 * uv_w] + W;
|
||||
const int g = best_uv[off + 1 * uv_w] + W;
|
||||
const int b = best_uv[off + 2 * uv_w] + W;
|
||||
dst_y[i] = ConvertRGBToY(r, g, b);
|
||||
}
|
||||
best_y += w;
|
||||
best_uv += (j & 1) * 3 * uv_w;
|
||||
dst_y += picture->y_stride;
|
||||
}
|
||||
for (best_uv = best_uv_base, j = 0; j < uv_h; ++j) {
|
||||
for (i = 0; i < uv_w; ++i) {
|
||||
const int off = i;
|
||||
const int r = best_uv[off + 0 * uv_w];
|
||||
const int g = best_uv[off + 1 * uv_w];
|
||||
const int b = best_uv[off + 2 * uv_w];
|
||||
dst_u[i] = ConvertRGBToU(r, g, b);
|
||||
dst_v[i] = ConvertRGBToV(r, g, b);
|
||||
}
|
||||
best_uv += 3 * uv_w;
|
||||
dst_u += picture->uv_stride;
|
||||
dst_v += picture->uv_stride;
|
||||
}
|
||||
return 1;
|
||||
}
|
||||
|
||||
//------------------------------------------------------------------------------
|
||||
// Main function
|
||||
|
||||
#define SAFE_ALLOC(W, H, T) ((T*)WebPSafeMalloc((W) * (H), sizeof(T)))
|
||||
|
||||
static int PreprocessARGB(const uint8_t* r_ptr,
|
||||
const uint8_t* g_ptr,
|
||||
const uint8_t* b_ptr,
|
||||
int step, int rgb_stride,
|
||||
WebPPicture* const picture) {
|
||||
// we expand the right/bottom border if needed
|
||||
const int w = (picture->width + 1) & ~1;
|
||||
const int h = (picture->height + 1) & ~1;
|
||||
const int uv_w = w >> 1;
|
||||
const int uv_h = h >> 1;
|
||||
uint64_t prev_diff_y_sum = ~0;
|
||||
int j, iter;
|
||||
|
||||
// TODO(skal): allocate one big memory chunk. But for now, it's easier
|
||||
// for valgrind debugging to have several chunks.
|
||||
fixed_y_t* const tmp_buffer = SAFE_ALLOC(w * 3, 2, fixed_y_t); // scratch
|
||||
fixed_y_t* const best_y_base = SAFE_ALLOC(w, h, fixed_y_t);
|
||||
fixed_y_t* const target_y_base = SAFE_ALLOC(w, h, fixed_y_t);
|
||||
fixed_y_t* const best_rgb_y = SAFE_ALLOC(w, 2, fixed_y_t);
|
||||
fixed_t* const best_uv_base = SAFE_ALLOC(uv_w * 3, uv_h, fixed_t);
|
||||
fixed_t* const target_uv_base = SAFE_ALLOC(uv_w * 3, uv_h, fixed_t);
|
||||
fixed_t* const best_rgb_uv = SAFE_ALLOC(uv_w * 3, 1, fixed_t);
|
||||
fixed_y_t* best_y = best_y_base;
|
||||
fixed_y_t* target_y = target_y_base;
|
||||
fixed_t* best_uv = best_uv_base;
|
||||
fixed_t* target_uv = target_uv_base;
|
||||
const uint64_t diff_y_threshold = (uint64_t)(3.0 * w * h);
|
||||
int ok;
|
||||
|
||||
if (best_y_base == NULL || best_uv_base == NULL ||
|
||||
target_y_base == NULL || target_uv_base == NULL ||
|
||||
best_rgb_y == NULL || best_rgb_uv == NULL ||
|
||||
tmp_buffer == NULL) {
|
||||
int ok = SharpArgbToYuv(r_ptr, g_ptr, b_ptr, step, rgb_stride, picture->y,
|
||||
picture->y_stride, picture->u, picture->uv_stride,
|
||||
picture->v, picture->uv_stride, picture->width,
|
||||
picture->height);
|
||||
if (!ok) {
|
||||
ok = WebPEncodingSetError(picture, VP8_ENC_ERROR_OUT_OF_MEMORY);
|
||||
goto End;
|
||||
}
|
||||
assert(picture->width >= kMinDimensionIterativeConversion);
|
||||
assert(picture->height >= kMinDimensionIterativeConversion);
|
||||
|
||||
WebPInitConvertARGBToYUV();
|
||||
|
||||
// Import RGB samples to W/RGB representation.
|
||||
for (j = 0; j < picture->height; j += 2) {
|
||||
const int is_last_row = (j == picture->height - 1);
|
||||
fixed_y_t* const src1 = tmp_buffer + 0 * w;
|
||||
fixed_y_t* const src2 = tmp_buffer + 3 * w;
|
||||
|
||||
// prepare two rows of input
|
||||
ImportOneRow(r_ptr, g_ptr, b_ptr, step, picture->width, src1);
|
||||
if (!is_last_row) {
|
||||
ImportOneRow(r_ptr + rgb_stride, g_ptr + rgb_stride, b_ptr + rgb_stride,
|
||||
step, picture->width, src2);
|
||||
} else {
|
||||
memcpy(src2, src1, 3 * w * sizeof(*src2));
|
||||
}
|
||||
StoreGray(src1, best_y + 0, w);
|
||||
StoreGray(src2, best_y + w, w);
|
||||
|
||||
UpdateW(src1, target_y, w);
|
||||
UpdateW(src2, target_y + w, w);
|
||||
UpdateChroma(src1, src2, target_uv, uv_w);
|
||||
memcpy(best_uv, target_uv, 3 * uv_w * sizeof(*best_uv));
|
||||
best_y += 2 * w;
|
||||
best_uv += 3 * uv_w;
|
||||
target_y += 2 * w;
|
||||
target_uv += 3 * uv_w;
|
||||
r_ptr += 2 * rgb_stride;
|
||||
g_ptr += 2 * rgb_stride;
|
||||
b_ptr += 2 * rgb_stride;
|
||||
}
|
||||
|
||||
// Iterate and resolve clipping conflicts.
|
||||
for (iter = 0; iter < kNumIterations; ++iter) {
|
||||
const fixed_t* cur_uv = best_uv_base;
|
||||
const fixed_t* prev_uv = best_uv_base;
|
||||
uint64_t diff_y_sum = 0;
|
||||
|
||||
best_y = best_y_base;
|
||||
best_uv = best_uv_base;
|
||||
target_y = target_y_base;
|
||||
target_uv = target_uv_base;
|
||||
for (j = 0; j < h; j += 2) {
|
||||
fixed_y_t* const src1 = tmp_buffer + 0 * w;
|
||||
fixed_y_t* const src2 = tmp_buffer + 3 * w;
|
||||
{
|
||||
const fixed_t* const next_uv = cur_uv + ((j < h - 2) ? 3 * uv_w : 0);
|
||||
InterpolateTwoRows(best_y, prev_uv, cur_uv, next_uv, w, src1, src2);
|
||||
prev_uv = cur_uv;
|
||||
cur_uv = next_uv;
|
||||
}
|
||||
|
||||
UpdateW(src1, best_rgb_y + 0 * w, w);
|
||||
UpdateW(src2, best_rgb_y + 1 * w, w);
|
||||
UpdateChroma(src1, src2, best_rgb_uv, uv_w);
|
||||
|
||||
// update two rows of Y and one row of RGB
|
||||
diff_y_sum += WebPSharpYUVUpdateY(target_y, best_rgb_y, best_y, 2 * w);
|
||||
WebPSharpYUVUpdateRGB(target_uv, best_rgb_uv, best_uv, 3 * uv_w);
|
||||
|
||||
best_y += 2 * w;
|
||||
best_uv += 3 * uv_w;
|
||||
target_y += 2 * w;
|
||||
target_uv += 3 * uv_w;
|
||||
}
|
||||
// test exit condition
|
||||
if (iter > 0) {
|
||||
if (diff_y_sum < diff_y_threshold) break;
|
||||
if (diff_y_sum > prev_diff_y_sum) break;
|
||||
}
|
||||
prev_diff_y_sum = diff_y_sum;
|
||||
}
|
||||
// final reconstruction
|
||||
ok = ConvertWRGBToYUV(best_y_base, best_uv_base, picture);
|
||||
|
||||
End:
|
||||
WebPSafeFree(best_y_base);
|
||||
WebPSafeFree(best_uv_base);
|
||||
WebPSafeFree(target_y_base);
|
||||
WebPSafeFree(target_uv_base);
|
||||
WebPSafeFree(best_rgb_y);
|
||||
WebPSafeFree(best_rgb_uv);
|
||||
WebPSafeFree(tmp_buffer);
|
||||
return ok;
|
||||
}
|
||||
#undef SAFE_ALLOC
|
||||
|
||||
//------------------------------------------------------------------------------
|
||||
// "Fast" regular RGB->YUV
|
||||
@ -874,7 +495,6 @@ static int ImportYUVAFromRGBA(const uint8_t* r_ptr,
|
||||
}
|
||||
|
||||
if (use_iterative_conversion) {
|
||||
InitGammaTablesS();
|
||||
if (!PreprocessARGB(r_ptr, g_ptr, b_ptr, step, rgb_stride, picture)) {
|
||||
return 0;
|
||||
}
|
||||
|
||||
Loading…
Reference in New Issue
Block a user