diff --git a/src/dec/frame.c b/src/dec/frame.c index 541347ce..22d291d2 100644 --- a/src/dec/frame.c +++ b/src/dec/frame.c @@ -316,6 +316,9 @@ static void PrecomputeFilterStrengths(VP8Decoder* const dec) { //------------------------------------------------------------------------------ // Dithering +// minimal amp that will provide a non-zero dithering effect +#define MIN_DITHER_AMP 4 + #define DITHER_AMP_TAB_SIZE 12 static const int kQuantToDitherAmp[DITHER_AMP_TAB_SIZE] = { // roughly, it's dqm->uv_mat_[1] @@ -356,27 +359,14 @@ void VP8InitDithering(const WebPDecoderOptions* const options, } } -// minimal amp that will provide a non-zero dithering effect -#define MIN_DITHER_AMP 4 -#define DITHER_DESCALE 4 -#define DITHER_DESCALE_ROUNDER (1 << (DITHER_DESCALE - 1)) -#define DITHER_AMP_BITS 8 -#define DITHER_AMP_CENTER (1 << DITHER_AMP_BITS) - +// Convert to range: [-2,2] for dither=50, [-4,4] for dither=100 static void Dither8x8(VP8Random* const rg, uint8_t* dst, int bps, int amp) { - int i, j; - for (j = 0; j < 8; ++j) { - for (i = 0; i < 8; ++i) { - // TODO: could be made faster with SSE2 - const int bits = - VP8RandomBits2(rg, DITHER_AMP_BITS + 1, amp) - DITHER_AMP_CENTER; - // Convert to range: [-2,2] for dither=50, [-4,4] for dither=100 - const int delta = (bits + DITHER_DESCALE_ROUNDER) >> DITHER_DESCALE; - const int v = (int)dst[i] + delta; - dst[i] = (v < 0) ? 0 : (v > 255) ? 255u : (uint8_t)v; - } - dst += bps; + uint8_t dither[64]; + int i; + for (i = 0; i < 8 * 8; ++i) { + dither[i] = VP8RandomBits2(rg, VP8_DITHER_AMP_BITS + 1, amp); } + VP8DitherCombine8x8(dither, dst, bps); } static void DitherRow(VP8Decoder* const dec) { diff --git a/src/dsp/dec.c b/src/dsp/dec.c index 1ef22d3f..17647f0a 100644 --- a/src/dsp/dec.c +++ b/src/dsp/dec.c @@ -655,6 +655,23 @@ static void HFilter8i(uint8_t* u, uint8_t* v, int stride, //------------------------------------------------------------------------------ +static void DitherCombine8x8(const uint8_t* dither, uint8_t* dst, + int dst_stride) { + int i, j; + for (j = 0; j < 8; ++j) { + for (i = 0; i < 8; ++i) { + const int delta0 = dither[i] - VP8_DITHER_AMP_CENTER; + const int delta1 = + (delta0 + VP8_DITHER_DESCALE_ROUNDER) >> VP8_DITHER_DESCALE; + dst[i] = clip_8b((int)dst[i] + delta1); + } + dst += dst_stride; + dither += 8; + } +} + +//------------------------------------------------------------------------------ + VP8DecIdct2 VP8Transform; VP8DecIdct VP8TransformAC3; VP8DecIdct VP8TransformUV; @@ -674,6 +691,9 @@ VP8SimpleFilterFunc VP8SimpleHFilter16; VP8SimpleFilterFunc VP8SimpleVFilter16i; VP8SimpleFilterFunc VP8SimpleHFilter16i; +void (*VP8DitherCombine8x8)(const uint8_t* dither, uint8_t* dst, + int dst_stride); + extern void VP8DspInitSSE2(void); extern void VP8DspInitSSE41(void); extern void VP8DspInitNEON(void); @@ -735,6 +755,8 @@ WEBP_TSAN_IGNORE_FUNCTION void VP8DspInit(void) { VP8PredChroma8[5] = DC8uvNoLeft; VP8PredChroma8[6] = DC8uvNoTopLeft; + VP8DitherCombine8x8 = DitherCombine8x8; + // If defined, use CPUInfo() to overwrite some pointers with faster versions. if (VP8GetCPUInfo != NULL) { #if defined(WEBP_USE_SSE2) diff --git a/src/dsp/dsp.h b/src/dsp/dsp.h index 9838a364..a0c45fcc 100644 --- a/src/dsp/dsp.h +++ b/src/dsp/dsp.h @@ -321,6 +321,15 @@ extern VP8LumaFilterFunc VP8HFilter16i; extern VP8ChromaFilterFunc VP8VFilter8i; // filtering u and v altogether extern VP8ChromaFilterFunc VP8HFilter8i; +// Dithering. Combines dithering values (centered around 128) with dst[], +// according to: dst[] = clip(dst[] + (((dither[]-128) + 8) >> 4) +#define VP8_DITHER_DESCALE 4 +#define VP8_DITHER_DESCALE_ROUNDER (1 << (VP8_DITHER_DESCALE - 1)) +#define VP8_DITHER_AMP_BITS 7 +#define VP8_DITHER_AMP_CENTER (1 << VP8_DITHER_AMP_BITS) +extern void (*VP8DitherCombine8x8)(const uint8_t* dither, uint8_t* dst, + int dst_stride); + // must be called before anything using the above void VP8DspInit(void);