create a separate libwebpdsp under src/dsp

Gathers all DSP-related function (and SSE2 implementations).
Clean-up some unwanted symbolic dependencies so that webp_encode,
webp_decode and webp_dsp are truly independent libraries.

+ opportunistic clean-up:
  * remove unneeded VP8DspInitTables(), now integrated in VP8DspInit()
  * make consistent use of VP8GetCPUInfo() in the various DspInit() funcs
  * change OUT macro to DST
This commit is contained in:
Pascal Massimino
2011-09-02 21:30:08 +00:00
committed by James Zern
parent ebeb412aa5
commit e06ac0887f
28 changed files with 2217 additions and 500 deletions

View File

@ -11,224 +11,15 @@
#include <assert.h>
#include <stdlib.h>
#include "vp8i.h"
#include "webpi.h"
#include "yuv.h"
#include "../dec/vp8i.h"
#include "./webpi.h"
#include "../dsp/dsp.h"
#include "../dsp/yuv.h"
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
#endif
#define FANCY_UPSAMPLING // undefined to remove fancy upsampling support
//------------------------------------------------------------------------------
// Fancy upsampler
#ifdef FANCY_UPSAMPLING
// Given samples laid out in a square as:
// [a b]
// [c d]
// we interpolate u/v as:
// ([9*a + 3*b + 3*c + d 3*a + 9*b + 3*c + d] + [8 8]) / 16
// ([3*a + b + 9*c + 3*d a + 3*b + 3*c + 9*d] [8 8]) / 16
// We process u and v together stashed into 32bit (16bit each).
#define LOAD_UV(u,v) ((u) | ((v) << 16))
#define UPSAMPLE_FUNC(FUNC_NAME, FUNC, XSTEP) \
static void FUNC_NAME(const uint8_t* top_y, const uint8_t* bottom_y, \
const uint8_t* top_u, const uint8_t* top_v, \
const uint8_t* cur_u, const uint8_t* cur_v, \
uint8_t* top_dst, uint8_t* bottom_dst, int len) { \
int x; \
const int last_pixel_pair = (len - 1) >> 1; \
uint32_t tl_uv = LOAD_UV(top_u[0], top_v[0]); /* top-left sample */ \
uint32_t l_uv = LOAD_UV(cur_u[0], cur_v[0]); /* left-sample */ \
if (top_y) { \
const uint32_t uv0 = (3 * tl_uv + l_uv + 0x00020002u) >> 2; \
FUNC(top_y[0], uv0 & 0xff, (uv0 >> 16), top_dst); \
} \
if (bottom_y) { \
const uint32_t uv0 = (3 * l_uv + tl_uv + 0x00020002u) >> 2; \
FUNC(bottom_y[0], uv0 & 0xff, (uv0 >> 16), bottom_dst); \
} \
for (x = 1; x <= last_pixel_pair; ++x) { \
const uint32_t t_uv = LOAD_UV(top_u[x], top_v[x]); /* top sample */ \
const uint32_t uv = LOAD_UV(cur_u[x], cur_v[x]); /* sample */ \
/* precompute invariant values associated with first and second diagonals*/\
const uint32_t avg = tl_uv + t_uv + l_uv + uv + 0x00080008u; \
const uint32_t diag_12 = (avg + 2 * (t_uv + l_uv)) >> 3; \
const uint32_t diag_03 = (avg + 2 * (tl_uv + uv)) >> 3; \
if (top_y) { \
const uint32_t uv0 = (diag_12 + tl_uv) >> 1; \
const uint32_t uv1 = (diag_03 + t_uv) >> 1; \
FUNC(top_y[2 * x - 1], uv0 & 0xff, (uv0 >> 16), \
top_dst + (2 * x - 1) * XSTEP); \
FUNC(top_y[2 * x - 0], uv1 & 0xff, (uv1 >> 16), \
top_dst + (2 * x - 0) * XSTEP); \
} \
if (bottom_y) { \
const uint32_t uv0 = (diag_03 + l_uv) >> 1; \
const uint32_t uv1 = (diag_12 + uv) >> 1; \
FUNC(bottom_y[2 * x - 1], uv0 & 0xff, (uv0 >> 16), \
bottom_dst + (2 * x - 1) * XSTEP); \
FUNC(bottom_y[2 * x + 0], uv1 & 0xff, (uv1 >> 16), \
bottom_dst + (2 * x + 0) * XSTEP); \
} \
tl_uv = t_uv; \
l_uv = uv; \
} \
if (!(len & 1)) { \
if (top_y) { \
const uint32_t uv0 = (3 * tl_uv + l_uv + 0x00020002u) >> 2; \
FUNC(top_y[len - 1], uv0 & 0xff, (uv0 >> 16), \
top_dst + (len - 1) * XSTEP); \
} \
if (bottom_y) { \
const uint32_t uv0 = (3 * l_uv + tl_uv + 0x00020002u) >> 2; \
FUNC(bottom_y[len - 1], uv0 & 0xff, (uv0 >> 16), \
bottom_dst + (len - 1) * XSTEP); \
} \
} \
}
// All variants implemented.
UPSAMPLE_FUNC(UpsampleRgbLinePair, VP8YuvToRgb, 3)
UPSAMPLE_FUNC(UpsampleBgrLinePair, VP8YuvToBgr, 3)
UPSAMPLE_FUNC(UpsampleRgbaLinePair, VP8YuvToRgba, 4)
UPSAMPLE_FUNC(UpsampleBgraLinePair, VP8YuvToBgra, 4)
UPSAMPLE_FUNC(UpsampleArgbLinePair, VP8YuvToArgb, 4)
UPSAMPLE_FUNC(UpsampleRgba4444LinePair, VP8YuvToRgba4444, 2)
UPSAMPLE_FUNC(UpsampleRgb565LinePair, VP8YuvToRgb565, 2)
// These two don't erase the alpha value
UPSAMPLE_FUNC(UpsampleRgbKeepAlphaLinePair, VP8YuvToRgb, 4)
UPSAMPLE_FUNC(UpsampleBgrKeepAlphaLinePair, VP8YuvToBgr, 4)
UPSAMPLE_FUNC(UpsampleArgbKeepAlphaLinePair, VP8YuvToArgbKeepA, 4)
UPSAMPLE_FUNC(UpsampleRgba4444KeepAlphaLinePair, VP8YuvToRgba4444KeepA, 2)
#undef LOAD_UV
#undef UPSAMPLE_FUNC
// Fancy upsampling functions to convert YUV to RGB
WebPUpsampleLinePairFunc WebPUpsamplers[MODE_LAST];
WebPUpsampleLinePairFunc WebPUpsamplersKeepAlpha[MODE_LAST];
static void InitUpsamplers(void) {
WebPUpsamplers[MODE_RGB] = UpsampleRgbLinePair;
WebPUpsamplers[MODE_RGBA] = UpsampleRgbaLinePair;
WebPUpsamplers[MODE_BGR] = UpsampleBgrLinePair;
WebPUpsamplers[MODE_BGRA] = UpsampleBgraLinePair;
WebPUpsamplers[MODE_ARGB] = UpsampleArgbLinePair;
WebPUpsamplers[MODE_RGBA_4444] = UpsampleRgba4444LinePair;
WebPUpsamplers[MODE_RGB_565] = UpsampleRgb565LinePair;
WebPUpsamplersKeepAlpha[MODE_RGB] = UpsampleRgbLinePair;
WebPUpsamplersKeepAlpha[MODE_RGBA] = UpsampleRgbKeepAlphaLinePair;
WebPUpsamplersKeepAlpha[MODE_BGR] = UpsampleBgrLinePair;
WebPUpsamplersKeepAlpha[MODE_BGRA] = UpsampleBgrKeepAlphaLinePair;
WebPUpsamplersKeepAlpha[MODE_ARGB] = UpsampleArgbKeepAlphaLinePair;
WebPUpsamplersKeepAlpha[MODE_RGBA_4444] = UpsampleRgba4444KeepAlphaLinePair;
WebPUpsamplersKeepAlpha[MODE_RGB_565] = UpsampleRgb565LinePair;
// If defined, use CPUInfo() to overwrite some pointers with faster versions.
if (VP8DecGetCPUInfo) {
if (VP8DecGetCPUInfo(kSSE2)) {
#if defined(__SSE2__) || defined(_MSC_VER)
WebPInitUpsamplersSSE2();
#endif
}
}
}
#endif // FANCY_UPSAMPLING
//------------------------------------------------------------------------------
// simple point-sampling
#define SAMPLE_FUNC(FUNC_NAME, FUNC, XSTEP) \
static void FUNC_NAME(const uint8_t* top_y, const uint8_t* bottom_y, \
const uint8_t* u, const uint8_t* v, \
uint8_t* top_dst, uint8_t* bottom_dst, int len) { \
int i; \
for (i = 0; i < len - 1; i += 2) { \
FUNC(top_y[0], u[0], v[0], top_dst); \
FUNC(top_y[1], u[0], v[0], top_dst + XSTEP); \
FUNC(bottom_y[0], u[0], v[0], bottom_dst); \
FUNC(bottom_y[1], u[0], v[0], bottom_dst + XSTEP); \
top_y += 2; \
bottom_y += 2; \
u++; \
v++; \
top_dst += 2 * XSTEP; \
bottom_dst += 2 * XSTEP; \
} \
if (i == len - 1) { /* last one */ \
FUNC(top_y[0], u[0], v[0], top_dst); \
FUNC(bottom_y[0], u[0], v[0], bottom_dst); \
} \
}
// All variants implemented.
SAMPLE_FUNC(SampleRgbLinePair, VP8YuvToRgb, 3)
SAMPLE_FUNC(SampleBgrLinePair, VP8YuvToBgr, 3)
SAMPLE_FUNC(SampleRgbaLinePair, VP8YuvToRgba, 4)
SAMPLE_FUNC(SampleBgraLinePair, VP8YuvToBgra, 4)
SAMPLE_FUNC(SampleArgbLinePair, VP8YuvToArgb, 4)
SAMPLE_FUNC(SampleRgba4444LinePair, VP8YuvToRgba4444, 2)
SAMPLE_FUNC(SampleRgb565LinePair, VP8YuvToRgb565, 2)
#undef SAMPLE_FUNC
// Main methods.
typedef void (*SampleLinePairFunc)(
const uint8_t* top_y, const uint8_t* bottom_y,
const uint8_t* u, const uint8_t* v,
uint8_t* top_dst, uint8_t* bottom_dst, int len);
static const SampleLinePairFunc kSamplers[MODE_LAST] = {
SampleRgbLinePair, // MODE_RGB
SampleRgbaLinePair, // MODE_RGBA
SampleBgrLinePair, // MODE_BGR
SampleBgraLinePair, // MODE_BGRA
SampleArgbLinePair, // MODE_ARGB
SampleRgba4444LinePair, // MODE_RGBA_4444
SampleRgb565LinePair // MODE_RGB_565
};
//------------------------------------------------------------------------------
// YUV444 converter
#define YUV444_FUNC(FUNC_NAME, FUNC, XSTEP) \
static void FUNC_NAME(const uint8_t* y, const uint8_t* u, const uint8_t* v, \
uint8_t* dst, int len) { \
int i; \
for (i = 0; i < len; ++i) FUNC(y[i], u[i], v[i], &dst[i * XSTEP]); \
}
YUV444_FUNC(Yuv444ToRgb, VP8YuvToRgb, 3)
YUV444_FUNC(Yuv444ToBgr, VP8YuvToBgr, 3)
YUV444_FUNC(Yuv444ToRgba, VP8YuvToRgba, 4)
YUV444_FUNC(Yuv444ToBgra, VP8YuvToBgra, 4)
YUV444_FUNC(Yuv444ToArgb, VP8YuvToArgb, 4)
YUV444_FUNC(Yuv444ToRgba4444, VP8YuvToRgba4444, 2)
YUV444_FUNC(Yuv444ToRgb565, VP8YuvToRgb565, 2)
#undef YUV444_FUNC
typedef void (*YUV444Func)(const uint8_t* y, const uint8_t* u, const uint8_t* v,
uint8_t* dst, int len);
static const YUV444Func kYUV444Converters[MODE_LAST] = {
Yuv444ToRgb, // MODE_RGB
Yuv444ToRgba, // MODE_RGBA
Yuv444ToBgr, // MODE_BGR
Yuv444ToBgra, // MODE_BGRA
Yuv444ToArgb, // MODE_ARGB
Yuv444ToRgba4444, // MODE_RGBA_4444
Yuv444ToRgb565 // MODE_RGB_565
};
//------------------------------------------------------------------------------
// Main YUV<->RGB conversion functions
@ -260,7 +51,7 @@ static int EmitSampledRGB(const VP8Io* const io, WebPDecParams* const p) {
const uint8_t* y_src = io->y;
const uint8_t* u_src = io->u;
const uint8_t* v_src = io->v;
const SampleLinePairFunc sample = kSamplers[output->colorspace];
const WebPSampleLinePairFunc sample = WebPSamplers[output->colorspace];
const int mb_w = io->mb_w;
const int last = io->mb_h - 1;
int j;
@ -289,7 +80,7 @@ static int EmitRGB(const VP8Io* const io, WebPDecParams* const p) {
const uint8_t* y_src = io->y;
const uint8_t* u_src = io->u;
const uint8_t* v_src = io->v;
const YUV444Func convert = kYUV444Converters[output->colorspace];
const WebPYUV444Converter convert = WebPYUV444Converters[output->colorspace];
const int mb_w = io->mb_w;
const int last = io->mb_h;
int j;
@ -608,7 +399,8 @@ static int Import(const uint8_t* src, int src_stride,
}
static int ExportRGB(WebPDecParams* const p, int y_pos) {
const YUV444Func convert = kYUV444Converters[p->output->colorspace];
const WebPYUV444Converter convert =
WebPYUV444Converters[p->output->colorspace];
const WebPRGBABuffer* const buf = &p->output->u.RGBA;
uint8_t* dst = buf->rgba + (p->last_y + y_pos) * buf->stride;
int num_lines_out = 0;
@ -811,7 +603,7 @@ static int CustomSetup(VP8Io* io) {
p->tmp_u = p->tmp_y + io->mb_w;
p->tmp_v = p->tmp_u + uv_width;
p->emit = EmitFancyRGB;
InitUpsamplers();
WebPInitUpsamplers();
}
#endif
} else {