diff --git a/src/dsp/lossless.c b/src/dsp/lossless.c index aad5f43e..46b220e2 100644 --- a/src/dsp/lossless.c +++ b/src/dsp/lossless.c @@ -107,62 +107,62 @@ static WEBP_INLINE uint32_t Select(uint32_t a, uint32_t b, uint32_t c) { //------------------------------------------------------------------------------ // Predictors -static uint32_t Predictor0_C(uint32_t left, const uint32_t* const top) { +uint32_t VP8LPredictor0_C(uint32_t left, const uint32_t* const top) { (void)top; (void)left; return ARGB_BLACK; } -static uint32_t Predictor1_C(uint32_t left, const uint32_t* const top) { +uint32_t VP8LPredictor1_C(uint32_t left, const uint32_t* const top) { (void)top; return left; } -static uint32_t Predictor2_C(uint32_t left, const uint32_t* const top) { +uint32_t VP8LPredictor2_C(uint32_t left, const uint32_t* const top) { (void)left; return top[0]; } -static uint32_t Predictor3_C(uint32_t left, const uint32_t* const top) { +uint32_t VP8LPredictor3_C(uint32_t left, const uint32_t* const top) { (void)left; return top[1]; } -static uint32_t Predictor4_C(uint32_t left, const uint32_t* const top) { +uint32_t VP8LPredictor4_C(uint32_t left, const uint32_t* const top) { (void)left; return top[-1]; } -static uint32_t Predictor5_C(uint32_t left, const uint32_t* const top) { +uint32_t VP8LPredictor5_C(uint32_t left, const uint32_t* const top) { const uint32_t pred = Average3(left, top[0], top[1]); return pred; } -static uint32_t Predictor6_C(uint32_t left, const uint32_t* const top) { +uint32_t VP8LPredictor6_C(uint32_t left, const uint32_t* const top) { const uint32_t pred = Average2(left, top[-1]); return pred; } -static uint32_t Predictor7_C(uint32_t left, const uint32_t* const top) { +uint32_t VP8LPredictor7_C(uint32_t left, const uint32_t* const top) { const uint32_t pred = Average2(left, top[0]); return pred; } -static uint32_t Predictor8_C(uint32_t left, const uint32_t* const top) { +uint32_t VP8LPredictor8_C(uint32_t left, const uint32_t* const top) { const uint32_t pred = Average2(top[-1], top[0]); (void)left; return pred; } -static uint32_t Predictor9_C(uint32_t left, const uint32_t* const top) { +uint32_t VP8LPredictor9_C(uint32_t left, const uint32_t* const top) { const uint32_t pred = Average2(top[0], top[1]); (void)left; return pred; } -static uint32_t Predictor10_C(uint32_t left, const uint32_t* const top) { +uint32_t VP8LPredictor10_C(uint32_t left, const uint32_t* const top) { const uint32_t pred = Average4(left, top[-1], top[0], top[1]); return pred; } -static uint32_t Predictor11_C(uint32_t left, const uint32_t* const top) { +uint32_t VP8LPredictor11_C(uint32_t left, const uint32_t* const top) { const uint32_t pred = Select(top[0], left, top[-1]); return pred; } -static uint32_t Predictor12_C(uint32_t left, const uint32_t* const top) { +uint32_t VP8LPredictor12_C(uint32_t left, const uint32_t* const top) { const uint32_t pred = ClampedAddSubtractFull(left, top[0], top[-1]); return pred; } -static uint32_t Predictor13_C(uint32_t left, const uint32_t* const top) { +uint32_t VP8LPredictor13_C(uint32_t left, const uint32_t* const top) { const uint32_t pred = ClampedAddSubtractHalf(left, top[0], top[-1]); return pred; } @@ -182,18 +182,18 @@ static void PredictorAdd1_C(const uint32_t* in, const uint32_t* upper, out[i] = left = VP8LAddPixels(in[i], left); } } -GENERATE_PREDICTOR_ADD(Predictor2_C, PredictorAdd2_C) -GENERATE_PREDICTOR_ADD(Predictor3_C, PredictorAdd3_C) -GENERATE_PREDICTOR_ADD(Predictor4_C, PredictorAdd4_C) -GENERATE_PREDICTOR_ADD(Predictor5_C, PredictorAdd5_C) -GENERATE_PREDICTOR_ADD(Predictor6_C, PredictorAdd6_C) -GENERATE_PREDICTOR_ADD(Predictor7_C, PredictorAdd7_C) -GENERATE_PREDICTOR_ADD(Predictor8_C, PredictorAdd8_C) -GENERATE_PREDICTOR_ADD(Predictor9_C, PredictorAdd9_C) -GENERATE_PREDICTOR_ADD(Predictor10_C, PredictorAdd10_C) -GENERATE_PREDICTOR_ADD(Predictor11_C, PredictorAdd11_C) -GENERATE_PREDICTOR_ADD(Predictor12_C, PredictorAdd12_C) -GENERATE_PREDICTOR_ADD(Predictor13_C, PredictorAdd13_C) +GENERATE_PREDICTOR_ADD(VP8LPredictor2_C, PredictorAdd2_C) +GENERATE_PREDICTOR_ADD(VP8LPredictor3_C, PredictorAdd3_C) +GENERATE_PREDICTOR_ADD(VP8LPredictor4_C, PredictorAdd4_C) +GENERATE_PREDICTOR_ADD(VP8LPredictor5_C, PredictorAdd5_C) +GENERATE_PREDICTOR_ADD(VP8LPredictor6_C, PredictorAdd6_C) +GENERATE_PREDICTOR_ADD(VP8LPredictor7_C, PredictorAdd7_C) +GENERATE_PREDICTOR_ADD(VP8LPredictor8_C, PredictorAdd8_C) +GENERATE_PREDICTOR_ADD(VP8LPredictor9_C, PredictorAdd9_C) +GENERATE_PREDICTOR_ADD(VP8LPredictor10_C, PredictorAdd10_C) +GENERATE_PREDICTOR_ADD(VP8LPredictor11_C, PredictorAdd11_C) +GENERATE_PREDICTOR_ADD(VP8LPredictor12_C, PredictorAdd12_C) +GENERATE_PREDICTOR_ADD(VP8LPredictor13_C, PredictorAdd13_C) //------------------------------------------------------------------------------ @@ -562,7 +562,6 @@ VP8LPredictorFunc VP8LPredictors[16]; // exposed plain-C implementations VP8LPredictorAddSubFunc VP8LPredictorsAdd_C[16]; -VP8LPredictorFunc VP8LPredictors_C[16]; VP8LTransformColorInverseFunc VP8LTransformColorInverse; @@ -600,8 +599,7 @@ extern void VP8LDspInitMSA(void); } while (0); WEBP_DSP_INIT_FUNC(VP8LDspInit) { - COPY_PREDICTOR_ARRAY(Predictor, VP8LPredictors) - COPY_PREDICTOR_ARRAY(Predictor, VP8LPredictors_C) + COPY_PREDICTOR_ARRAY(VP8LPredictor, VP8LPredictors) COPY_PREDICTOR_ARRAY(PredictorAdd, VP8LPredictorsAdd) COPY_PREDICTOR_ARRAY(PredictorAdd, VP8LPredictorsAdd_C) diff --git a/src/dsp/lossless.h b/src/dsp/lossless.h index f709cc86..ebd316d1 100644 --- a/src/dsp/lossless.h +++ b/src/dsp/lossless.h @@ -30,7 +30,22 @@ extern "C" { typedef uint32_t (*VP8LPredictorFunc)(uint32_t left, const uint32_t* const top); extern VP8LPredictorFunc VP8LPredictors[16]; -extern VP8LPredictorFunc VP8LPredictors_C[16]; + +uint32_t VP8LPredictor0_C(uint32_t left, const uint32_t* const top); +uint32_t VP8LPredictor1_C(uint32_t left, const uint32_t* const top); +uint32_t VP8LPredictor2_C(uint32_t left, const uint32_t* const top); +uint32_t VP8LPredictor3_C(uint32_t left, const uint32_t* const top); +uint32_t VP8LPredictor4_C(uint32_t left, const uint32_t* const top); +uint32_t VP8LPredictor5_C(uint32_t left, const uint32_t* const top); +uint32_t VP8LPredictor6_C(uint32_t left, const uint32_t* const top); +uint32_t VP8LPredictor7_C(uint32_t left, const uint32_t* const top); +uint32_t VP8LPredictor8_C(uint32_t left, const uint32_t* const top); +uint32_t VP8LPredictor9_C(uint32_t left, const uint32_t* const top); +uint32_t VP8LPredictor10_C(uint32_t left, const uint32_t* const top); +uint32_t VP8LPredictor11_C(uint32_t left, const uint32_t* const top); +uint32_t VP8LPredictor12_C(uint32_t left, const uint32_t* const top); +uint32_t VP8LPredictor13_C(uint32_t left, const uint32_t* const top); + // These Add/Sub function expects upper[-1] and out[-1] to be readable. typedef void (*VP8LPredictorAddSubFunc)(const uint32_t* in, const uint32_t* upper, int num_pixels, diff --git a/src/dsp/lossless_common.h b/src/dsp/lossless_common.h index 9c2ebe68..96a106f9 100644 --- a/src/dsp/lossless_common.h +++ b/src/dsp/lossless_common.h @@ -184,19 +184,6 @@ static void PREDICTOR_ADD(const uint32_t* in, const uint32_t* upper, \ } \ } -// It subtracts the prediction from the input pixel and stores the residual -// in the output pixel. -#define GENERATE_PREDICTOR_SUB(PREDICTOR, PREDICTOR_SUB) \ -static void PREDICTOR_SUB(const uint32_t* in, const uint32_t* upper, \ - int num_pixels, uint32_t* out) { \ - int x; \ - assert(upper != NULL); \ - for (x = 0; x < num_pixels; ++x) { \ - const uint32_t pred = (PREDICTOR)(in[x - 1], upper + x); \ - out[x] = VP8LSubPixels(in[x], pred); \ - } \ -} - #ifdef __cplusplus } // extern "C" #endif diff --git a/src/dsp/lossless_enc.c b/src/dsp/lossless_enc.c index 9c36055a..a0c7ab91 100644 --- a/src/dsp/lossless_enc.c +++ b/src/dsp/lossless_enc.c @@ -702,140 +702,6 @@ void VP8LHistogramAdd(const VP8LHistogram* const a, //------------------------------------------------------------------------------ // Image transforms. -static WEBP_INLINE uint32_t Average2(uint32_t a0, uint32_t a1) { - return (((a0 ^ a1) & 0xfefefefeu) >> 1) + (a0 & a1); -} - -static WEBP_INLINE uint32_t Average3(uint32_t a0, uint32_t a1, uint32_t a2) { - return Average2(Average2(a0, a2), a1); -} - -static WEBP_INLINE uint32_t Average4(uint32_t a0, uint32_t a1, - uint32_t a2, uint32_t a3) { - return Average2(Average2(a0, a1), Average2(a2, a3)); -} - -static WEBP_INLINE uint32_t Clip255(uint32_t a) { - if (a < 256) { - return a; - } - // return 0, when a is a negative integer. - // return 255, when a is positive. - return ~a >> 24; -} - -static WEBP_INLINE int AddSubtractComponentFull(int a, int b, int c) { - return Clip255(a + b - c); -} - -static WEBP_INLINE uint32_t ClampedAddSubtractFull(uint32_t c0, uint32_t c1, - uint32_t c2) { - const int a = AddSubtractComponentFull(c0 >> 24, c1 >> 24, c2 >> 24); - const int r = AddSubtractComponentFull((c0 >> 16) & 0xff, - (c1 >> 16) & 0xff, - (c2 >> 16) & 0xff); - const int g = AddSubtractComponentFull((c0 >> 8) & 0xff, - (c1 >> 8) & 0xff, - (c2 >> 8) & 0xff); - const int b = AddSubtractComponentFull(c0 & 0xff, c1 & 0xff, c2 & 0xff); - return ((uint32_t)a << 24) | (r << 16) | (g << 8) | b; -} - -static WEBP_INLINE int AddSubtractComponentHalf(int a, int b) { - return Clip255(a + (a - b) / 2); -} - -static WEBP_INLINE uint32_t ClampedAddSubtractHalf(uint32_t c0, uint32_t c1, - uint32_t c2) { - const uint32_t ave = Average2(c0, c1); - const int a = AddSubtractComponentHalf(ave >> 24, c2 >> 24); - const int r = AddSubtractComponentHalf((ave >> 16) & 0xff, (c2 >> 16) & 0xff); - const int g = AddSubtractComponentHalf((ave >> 8) & 0xff, (c2 >> 8) & 0xff); - const int b = AddSubtractComponentHalf((ave >> 0) & 0xff, (c2 >> 0) & 0xff); - return ((uint32_t)a << 24) | (r << 16) | (g << 8) | b; -} - -// gcc-4.9 on ARM generates incorrect code in Select() when Sub3() is inlined. -#if defined(__arm__) && \ - (LOCAL_GCC_VERSION == 0x409 || LOCAL_GCC_VERSION == 0x408) -# define LOCAL_INLINE __attribute__ ((noinline)) -#else -# define LOCAL_INLINE WEBP_INLINE -#endif - -static LOCAL_INLINE int Sub3(int a, int b, int c) { - const int pb = b - c; - const int pa = a - c; - return abs(pb) - abs(pa); -} - -#undef LOCAL_INLINE - -static WEBP_INLINE uint32_t Select(uint32_t a, uint32_t b, uint32_t c) { - const int pa_minus_pb = - Sub3((a >> 24) , (b >> 24) , (c >> 24) ) + - Sub3((a >> 16) & 0xff, (b >> 16) & 0xff, (c >> 16) & 0xff) + - Sub3((a >> 8) & 0xff, (b >> 8) & 0xff, (c >> 8) & 0xff) + - Sub3((a ) & 0xff, (b ) & 0xff, (c ) & 0xff); - return (pa_minus_pb <= 0) ? a : b; -} - -//------------------------------------------------------------------------------ -// Predictors - -static uint32_t Predictor2(uint32_t left, const uint32_t* const top) { - (void)left; - return top[0]; -} -static uint32_t Predictor3(uint32_t left, const uint32_t* const top) { - (void)left; - return top[1]; -} -static uint32_t Predictor4(uint32_t left, const uint32_t* const top) { - (void)left; - return top[-1]; -} -static uint32_t Predictor5(uint32_t left, const uint32_t* const top) { - const uint32_t pred = Average3(left, top[0], top[1]); - return pred; -} -static uint32_t Predictor6(uint32_t left, const uint32_t* const top) { - const uint32_t pred = Average2(left, top[-1]); - return pred; -} -static uint32_t Predictor7(uint32_t left, const uint32_t* const top) { - const uint32_t pred = Average2(left, top[0]); - return pred; -} -static uint32_t Predictor8(uint32_t left, const uint32_t* const top) { - const uint32_t pred = Average2(top[-1], top[0]); - (void)left; - return pred; -} -static uint32_t Predictor9(uint32_t left, const uint32_t* const top) { - const uint32_t pred = Average2(top[0], top[1]); - (void)left; - return pred; -} -static uint32_t Predictor10(uint32_t left, const uint32_t* const top) { - const uint32_t pred = Average4(left, top[-1], top[0], top[1]); - return pred; -} -static uint32_t Predictor11(uint32_t left, const uint32_t* const top) { - const uint32_t pred = Select(top[0], left, top[-1]); - return pred; -} -static uint32_t Predictor12(uint32_t left, const uint32_t* const top) { - const uint32_t pred = ClampedAddSubtractFull(left, top[0], top[-1]); - return pred; -} -static uint32_t Predictor13(uint32_t left, const uint32_t* const top) { - const uint32_t pred = ClampedAddSubtractHalf(left, top[0], top[-1]); - return pred; -} - -//------------------------------------------------------------------------------ - static void PredictorSub0_C(const uint32_t* in, const uint32_t* upper, int num_pixels, uint32_t* out) { int i; @@ -850,18 +716,33 @@ static void PredictorSub1_C(const uint32_t* in, const uint32_t* upper, (void)upper; } -GENERATE_PREDICTOR_SUB(Predictor2, PredictorSub2_C) -GENERATE_PREDICTOR_SUB(Predictor3, PredictorSub3_C) -GENERATE_PREDICTOR_SUB(Predictor4, PredictorSub4_C) -GENERATE_PREDICTOR_SUB(Predictor5, PredictorSub5_C) -GENERATE_PREDICTOR_SUB(Predictor6, PredictorSub6_C) -GENERATE_PREDICTOR_SUB(Predictor7, PredictorSub7_C) -GENERATE_PREDICTOR_SUB(Predictor8, PredictorSub8_C) -GENERATE_PREDICTOR_SUB(Predictor9, PredictorSub9_C) -GENERATE_PREDICTOR_SUB(Predictor10, PredictorSub10_C) -GENERATE_PREDICTOR_SUB(Predictor11, PredictorSub11_C) -GENERATE_PREDICTOR_SUB(Predictor12, PredictorSub12_C) -GENERATE_PREDICTOR_SUB(Predictor13, PredictorSub13_C) +// It subtracts the prediction from the input pixel and stores the residual +// in the output pixel. +#define GENERATE_PREDICTOR_SUB(PREDICTOR_I) \ +static void PredictorSub##PREDICTOR_I##_C(const uint32_t* in, \ + const uint32_t* upper, \ + int num_pixels, uint32_t* out) { \ + int x; \ + assert(upper != NULL); \ + for (x = 0; x < num_pixels; ++x) { \ + const uint32_t pred = \ + VP8LPredictor##PREDICTOR_I##_C(in[x - 1], upper + x); \ + out[x] = VP8LSubPixels(in[x], pred); \ + } \ +} + +GENERATE_PREDICTOR_SUB(2) +GENERATE_PREDICTOR_SUB(3) +GENERATE_PREDICTOR_SUB(4) +GENERATE_PREDICTOR_SUB(5) +GENERATE_PREDICTOR_SUB(6) +GENERATE_PREDICTOR_SUB(7) +GENERATE_PREDICTOR_SUB(8) +GENERATE_PREDICTOR_SUB(9) +GENERATE_PREDICTOR_SUB(10) +GENERATE_PREDICTOR_SUB(11) +GENERATE_PREDICTOR_SUB(12) +GENERATE_PREDICTOR_SUB(13) //------------------------------------------------------------------------------