diff --git a/src/dec/vp8l.c b/src/dec/vp8l.c index 8b0fa70a..01c3274f 100644 --- a/src/dec/vp8l.c +++ b/src/dec/vp8l.c @@ -712,13 +712,15 @@ static void ApplyInverseTransforms(VP8LDecoder* const dec, int num_rows, uint32_t* const rows_out = dec->argb_cache_; // Inverse transforms. - // TODO: most transforms only need to operate on the cropped region only. - memcpy(rows_out, rows_in, cache_pixs * sizeof(*rows_out)); while (n-- > 0) { VP8LTransform* const transform = &dec->transforms_[n]; VP8LInverseTransform(transform, start_row, end_row, rows_in, rows_out); rows_in = rows_out; } + if (rows_in != rows_out) { + // No transform called, hence just copy. + memcpy(rows_out, rows_in, cache_pixs * sizeof(*rows_out)); + } } // Processes (transforms, scales & color-converts) the rows decoded after the diff --git a/src/dsp/lossless.c b/src/dsp/lossless.c index 1a4c3ac6..2f438731 100644 --- a/src/dsp/lossless.c +++ b/src/dsp/lossless.c @@ -234,15 +234,16 @@ static void PredictorInverseTransform(const VP8LTransform* const transform, // Add green to blue and red channels (i.e. perform the inverse transform of // 'subtract green'). -void VP8LAddGreenToBlueAndRed_C(uint32_t* data, int num_pixels) { +void VP8LAddGreenToBlueAndRed_C(const uint32_t* const src, int num_pixels, + uint32_t* dst) { int i; for (i = 0; i < num_pixels; ++i) { - const uint32_t argb = data[i]; + const uint32_t argb = src[i]; const uint32_t green = ((argb >> 8) & 0xff); uint32_t red_blue = (argb & 0x00ff00ffu); red_blue += (green << 16) | green; red_blue &= 0x00ff00ffu; - data[i] = (argb & 0xff00ff00u) | red_blue; + dst[i] = (argb & 0xff00ff00u) | red_blue; } } @@ -258,11 +259,12 @@ static WEBP_INLINE void ColorCodeToMultipliers(uint32_t color_code, m->red_to_blue_ = (color_code >> 16) & 0xff; } -void VP8LTransformColorInverse_C(const VP8LMultipliers* const m, uint32_t* data, - int num_pixels) { +void VP8LTransformColorInverse_C(const VP8LMultipliers* const m, + const uint32_t* const src, int num_pixels, + uint32_t* const dst) { int i; for (i = 0; i < num_pixels; ++i) { - const uint32_t argb = data[i]; + const uint32_t argb = src[i]; const uint32_t green = argb >> 8; const uint32_t red = argb >> 16; int new_red = red; @@ -272,13 +274,14 @@ void VP8LTransformColorInverse_C(const VP8LMultipliers* const m, uint32_t* data, new_blue += ColorTransformDelta(m->green_to_blue_, green); new_blue += ColorTransformDelta(m->red_to_blue_, new_red); new_blue &= 0xff; - data[i] = (argb & 0xff00ff00u) | (new_red << 16) | (new_blue); + dst[i] = (argb & 0xff00ff00u) | (new_red << 16) | (new_blue); } } // Color space inverse transform. static void ColorSpaceInverseTransform(const VP8LTransform* const transform, - int y_start, int y_end, uint32_t* data) { + int y_start, int y_end, + const uint32_t* src, uint32_t* dst) { const int width = transform->xsize_; const int tile_width = 1 << transform->bits_; const int mask = tile_width - 1; @@ -292,17 +295,19 @@ static void ColorSpaceInverseTransform(const VP8LTransform* const transform, while (y < y_end) { const uint32_t* pred = pred_row; VP8LMultipliers m = { 0, 0, 0 }; - const uint32_t* const data_safe_end = data + safe_width; - const uint32_t* const data_end = data + width; - while (data < data_safe_end) { + const uint32_t* const src_safe_end = src + safe_width; + const uint32_t* const src_end = src + width; + while (src < src_safe_end) { ColorCodeToMultipliers(*pred++, &m); - VP8LTransformColorInverse(&m, data, tile_width); - data += tile_width; + VP8LTransformColorInverse(&m, src, tile_width, dst); + src += tile_width; + dst += tile_width; } - if (data < data_end) { // Left-overs using C-version. + if (src < src_end) { // Left-overs using C-version. ColorCodeToMultipliers(*pred++, &m); - VP8LTransformColorInverse(&m, data, remaining_width); - data += remaining_width; + VP8LTransformColorInverse(&m, src, remaining_width, dst); + src += remaining_width; + dst += remaining_width; } ++y; if ((y & mask) == 0) pred_row += tiles_per_row; @@ -367,9 +372,13 @@ void VP8LInverseTransform(const VP8LTransform* const transform, assert(row_end <= transform->ysize_); switch (transform->type_) { case SUBTRACT_GREEN: - VP8LAddGreenToBlueAndRed(out, (row_end - row_start) * width); + VP8LAddGreenToBlueAndRed(in, (row_end - row_start) * width, out); break; case PREDICTOR_TRANSFORM: + // TODO(vrabaud): parallelize transform predictors. + if (in != out) { + memcpy(out, in, (row_end - row_start) * width * sizeof(*out)); + } PredictorInverseTransform(transform, row_start, row_end, out); if (row_end != transform->ysize_) { // The last predicted row in this iteration will be the top-pred row @@ -379,7 +388,7 @@ void VP8LInverseTransform(const VP8LTransform* const transform, } break; case CROSS_COLOR_TRANSFORM: - ColorSpaceInverseTransform(transform, row_start, row_end, out); + ColorSpaceInverseTransform(transform, row_start, row_end, in, out); break; case COLOR_INDEXING_TRANSFORM: if (in == out && transform->bits_ > 0) { @@ -556,10 +565,10 @@ void VP8LConvertFromBGRA(const uint32_t* const in_data, int num_pixels, //------------------------------------------------------------------------------ -VP8LProcessBlueAndRedFunc VP8LAddGreenToBlueAndRed; +VP8LProcessDecBlueAndRedFunc VP8LAddGreenToBlueAndRed; VP8LPredictorFunc VP8LPredictors[16]; -VP8LTransformColorFunc VP8LTransformColorInverse; +VP8LTransformColorInverseFunc VP8LTransformColorInverse; VP8LConvertFunc VP8LConvertBGRAToRGB; VP8LConvertFunc VP8LConvertBGRAToRGBA; diff --git a/src/dsp/lossless.h b/src/dsp/lossless.h index d9d821e3..7a95024a 100644 --- a/src/dsp/lossless.h +++ b/src/dsp/lossless.h @@ -35,8 +35,9 @@ extern "C" { typedef uint32_t (*VP8LPredictorFunc)(uint32_t left, const uint32_t* const top); extern VP8LPredictorFunc VP8LPredictors[16]; -typedef void (*VP8LProcessBlueAndRedFunc)(uint32_t* argb_data, int num_pixels); -extern VP8LProcessBlueAndRedFunc VP8LAddGreenToBlueAndRed; +typedef void (*VP8LProcessDecBlueAndRedFunc)(const uint32_t* src, + int num_pixels, uint32_t* dst); +extern VP8LProcessDecBlueAndRedFunc VP8LAddGreenToBlueAndRed; typedef struct { // Note: the members are uint8_t, so that any negative values are @@ -45,9 +46,10 @@ typedef struct { uint8_t green_to_blue_; uint8_t red_to_blue_; } VP8LMultipliers; -typedef void (*VP8LTransformColorFunc)(const VP8LMultipliers* const m, - uint32_t* argb_data, int num_pixels); -extern VP8LTransformColorFunc VP8LTransformColorInverse; +typedef void (*VP8LTransformColorInverseFunc)(const VP8LMultipliers* const m, + const uint32_t* src, + int num_pixels, uint32_t* dst); +extern VP8LTransformColorInverseFunc VP8LTransformColorInverse; struct VP8LTransform; // Defined in dec/vp8li.h. @@ -93,7 +95,8 @@ void VP8LColorIndexInverseTransformAlpha( // Expose some C-only fallback functions void VP8LTransformColorInverse_C(const VP8LMultipliers* const m, - uint32_t* data, int num_pixels); + const uint32_t* src, int num_pixels, + uint32_t* dst); void VP8LConvertBGRAToRGB_C(const uint32_t* src, int num_pixels, uint8_t* dst); void VP8LConvertBGRAToRGBA_C(const uint32_t* src, int num_pixels, uint8_t* dst); @@ -102,7 +105,8 @@ void VP8LConvertBGRAToRGBA4444_C(const uint32_t* src, void VP8LConvertBGRAToRGB565_C(const uint32_t* src, int num_pixels, uint8_t* dst); void VP8LConvertBGRAToBGR_C(const uint32_t* src, int num_pixels, uint8_t* dst); -void VP8LAddGreenToBlueAndRed_C(uint32_t* data, int num_pixels); +void VP8LAddGreenToBlueAndRed_C(const uint32_t* src, int num_pixels, + uint32_t* dst); // Must be called before calling any of the above methods. void VP8LDspInit(void); @@ -110,7 +114,10 @@ void VP8LDspInit(void); //------------------------------------------------------------------------------ // Encoding -extern VP8LProcessBlueAndRedFunc VP8LSubtractGreenFromBlueAndRed; +typedef void (*VP8LProcessEncBlueAndRedFunc)(uint32_t* dst, int num_pixels); +extern VP8LProcessEncBlueAndRedFunc VP8LSubtractGreenFromBlueAndRed; +typedef void (*VP8LTransformColorFunc)(const VP8LMultipliers* const m, + uint32_t* const dst, int num_pixels); extern VP8LTransformColorFunc VP8LTransformColor; typedef void (*VP8LCollectColorBlueTransformsFunc)( const uint32_t* argb, int stride, diff --git a/src/dsp/lossless_enc.c b/src/dsp/lossless_enc.c index 932f7ec6..58c6fc78 100644 --- a/src/dsp/lossless_enc.c +++ b/src/dsp/lossless_enc.c @@ -665,7 +665,7 @@ static void HistogramAdd(const VP8LHistogram* const a, //------------------------------------------------------------------------------ -VP8LProcessBlueAndRedFunc VP8LSubtractGreenFromBlueAndRed; +VP8LProcessEncBlueAndRedFunc VP8LSubtractGreenFromBlueAndRed; VP8LTransformColorFunc VP8LTransformColor; diff --git a/src/dsp/lossless_mips_dsp_r2.c b/src/dsp/lossless_mips_dsp_r2.c index 694c1324..2984ce8d 100644 --- a/src/dsp/lossless_mips_dsp_r2.c +++ b/src/dsp/lossless_mips_dsp_r2.c @@ -228,25 +228,27 @@ static uint32_t Predictor13(uint32_t left, const uint32_t* const top) { // Add green to blue and red channels (i.e. perform the inverse transform of // 'subtract green'). -static void AddGreenToBlueAndRed(uint32_t* data, int num_pixels) { +static void AddGreenToBlueAndRed(const uint32_t* src, int num_pixels, + uint32_t* dst) { uint32_t temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7; - uint32_t* const p_loop1_end = data + (num_pixels & ~3); - uint32_t* const p_loop2_end = data + num_pixels; + const uint32_t* const p_loop1_end = src + (num_pixels & ~3); + const uint32_t* const p_loop2_end = src + num_pixels; __asm__ volatile ( ".set push \n\t" ".set noreorder \n\t" - "beq %[data], %[p_loop1_end], 3f \n\t" + "beq %[src], %[p_loop1_end], 3f \n\t" " nop \n\t" "0: \n\t" - "lw %[temp0], 0(%[data]) \n\t" - "lw %[temp1], 4(%[data]) \n\t" - "lw %[temp2], 8(%[data]) \n\t" - "lw %[temp3], 12(%[data]) \n\t" + "lw %[temp0], 0(%[src]) \n\t" + "lw %[temp1], 4(%[src]) \n\t" + "lw %[temp2], 8(%[src]) \n\t" + "lw %[temp3], 12(%[src]) \n\t" "ext %[temp4], %[temp0], 8, 8 \n\t" "ext %[temp5], %[temp1], 8, 8 \n\t" "ext %[temp6], %[temp2], 8, 8 \n\t" "ext %[temp7], %[temp3], 8, 8 \n\t" - "addiu %[data], %[data], 16 \n\t" + "addiu %[src], %[src], 16 \n\t" + "addiu %[dst], %[dst], 16 \n\t" "replv.ph %[temp4], %[temp4] \n\t" "replv.ph %[temp5], %[temp5] \n\t" "replv.ph %[temp6], %[temp6] \n\t" @@ -255,44 +257,47 @@ static void AddGreenToBlueAndRed(uint32_t* data, int num_pixels) { "addu.qb %[temp1], %[temp1], %[temp5] \n\t" "addu.qb %[temp2], %[temp2], %[temp6] \n\t" "addu.qb %[temp3], %[temp3], %[temp7] \n\t" - "sw %[temp0], -16(%[data]) \n\t" - "sw %[temp1], -12(%[data]) \n\t" - "sw %[temp2], -8(%[data]) \n\t" - "bne %[data], %[p_loop1_end], 0b \n\t" - " sw %[temp3], -4(%[data]) \n\t" + "sw %[temp0], -16(%[dst]) \n\t" + "sw %[temp1], -12(%[dst]) \n\t" + "sw %[temp2], -8(%[dst]) \n\t" + "bne %[src], %[p_loop1_end], 0b \n\t" + " sw %[temp3], -4(%[dst]) \n\t" "3: \n\t" - "beq %[data], %[p_loop2_end], 2f \n\t" + "beq %[src], %[p_loop2_end], 2f \n\t" " nop \n\t" "1: \n\t" - "lw %[temp0], 0(%[data]) \n\t" - "addiu %[data], %[data], 4 \n\t" + "lw %[temp0], 0(%[src]) \n\t" + "addiu %[src], %[src], 4 \n\t" + "addiu %[dst], %[dst], 4 \n\t" "ext %[temp4], %[temp0], 8, 8 \n\t" "replv.ph %[temp4], %[temp4] \n\t" "addu.qb %[temp0], %[temp0], %[temp4] \n\t" - "bne %[data], %[p_loop2_end], 1b \n\t" - " sw %[temp0], -4(%[data]) \n\t" + "bne %[src], %[p_loop2_end], 1b \n\t" + " sw %[temp0], -4(%[dst]) \n\t" "2: \n\t" ".set pop \n\t" - : [data]"+&r"(data), [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), - [temp2]"=&r"(temp2), [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), - [temp5]"=&r"(temp5), [temp6]"=&r"(temp6), [temp7]"=&r"(temp7) + : [dst]"+&r"(dst), [src]"+&r"(src), [temp0]"=&r"(temp0), + [temp1]"=&r"(temp1), [temp2]"=&r"(temp2), [temp3]"=&r"(temp3), + [temp4]"=&r"(temp4), [temp5]"=&r"(temp5), [temp6]"=&r"(temp6), + [temp7]"=&r"(temp7) : [p_loop1_end]"r"(p_loop1_end), [p_loop2_end]"r"(p_loop2_end) : "memory" ); } static void TransformColorInverse(const VP8LMultipliers* const m, - uint32_t* data, int num_pixels) { + const uint32_t* src, int num_pixels, + uint32_t* dst) { int temp0, temp1, temp2, temp3, temp4, temp5; uint32_t argb, argb1, new_red; const uint32_t G_to_R = m->green_to_red_; const uint32_t G_to_B = m->green_to_blue_; const uint32_t R_to_B = m->red_to_blue_; - uint32_t* const p_loop_end = data + (num_pixels & ~1); + const uint32_t* const p_loop_end = src + (num_pixels & ~1); __asm__ volatile ( ".set push \n\t" ".set noreorder \n\t" - "beq %[data], %[p_loop_end], 1f \n\t" + "beq %[src], %[p_loop_end], 1f \n\t" " nop \n\t" "replv.ph %[temp0], %[G_to_R] \n\t" "replv.ph %[temp1], %[G_to_B] \n\t" @@ -304,9 +309,12 @@ static void TransformColorInverse(const VP8LMultipliers* const m, "shra.ph %[temp1], %[temp1], 8 \n\t" "shra.ph %[temp2], %[temp2], 8 \n\t" "0: \n\t" - "lw %[argb], 0(%[data]) \n\t" - "lw %[argb1], 4(%[data]) \n\t" - "addiu %[data], %[data], 8 \n\t" + "lw %[argb], 0(%[src]) \n\t" + "lw %[argb1], 4(%[src]) \n\t" + "sw %[argb], 0(%[dst]) \n\t" + "sw %[argb1], 4(%[dst]) \n\t" + "addiu %[src], %[src], 8 \n\t" + "addiu %[dst], %[dst], 8 \n\t" "precrq.qb.ph %[temp3], %[argb], %[argb1] \n\t" "preceu.ph.qbra %[temp3], %[temp3] \n\t" "shll.ph %[temp3], %[temp3], 8 \n\t" @@ -323,29 +331,29 @@ static void TransformColorInverse(const VP8LMultipliers* const m, "shll.ph %[temp4], %[temp5], 8 \n\t" "shra.ph %[temp4], %[temp4], 8 \n\t" "mul.ph %[temp4], %[temp4], %[temp2] \n\t" - "sb %[temp5], -2(%[data]) \n\t" + "sb %[temp5], -2(%[dst]) \n\t" "sra %[temp5], %[temp5], 16 \n\t" "shra.ph %[temp4], %[temp4], 5 \n\t" "addu.ph %[argb1], %[argb1], %[temp4] \n\t" "preceu.ph.qbra %[temp3], %[argb1] \n\t" - "sb %[temp5], -6(%[data]) \n\t" - "sb %[temp3], -4(%[data]) \n\t" + "sb %[temp5], -6(%[dst]) \n\t" + "sb %[temp3], -4(%[dst]) \n\t" "sra %[temp3], %[temp3], 16 \n\t" - "bne %[data], %[p_loop_end], 0b \n\t" - " sb %[temp3], -8(%[data]) \n\t" + "bne %[src], %[p_loop_end], 0b \n\t" + " sb %[temp3], -8(%[dst]) \n\t" "1: \n\t" ".set pop \n\t" : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2), [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5), [new_red]"=&r"(new_red), [argb]"=&r"(argb), - [argb1]"=&r"(argb1), [data]"+&r"(data) + [argb1]"=&r"(argb1), [dst]"+&r"(dst), [src]"+&r"(src) : [G_to_R]"r"(G_to_R), [R_to_B]"r"(R_to_B), [G_to_B]"r"(G_to_B), [p_loop_end]"r"(p_loop_end) : "memory", "hi", "lo" ); // Fall-back to C-version for left-overs. - if (num_pixels & 1) VP8LTransformColorInverse_C(m, data, 1); + if (num_pixels & 1) VP8LTransformColorInverse_C(m, src, 1, dst); } static void ConvertBGRAToRGB(const uint32_t* src, diff --git a/src/dsp/lossless_msa.c b/src/dsp/lossless_msa.c index 26c5ab62..f6dd5649 100644 --- a/src/dsp/lossless_msa.c +++ b/src/dsp/lossless_msa.c @@ -244,44 +244,51 @@ static void ConvertBGRAToRGB(const uint32_t* src, } } -static void AddGreenToBlueAndRed(uint32_t* data, int num_pixels) { +static void AddGreenToBlueAndRed(const uint32_t* const src, int num_pixels, + uint32_t* dst) { int i; - uint8_t* ptemp_data = (uint8_t*)data; + const uint8_t* in = (const uint8_t*)src; + uint8_t* out = (uint8_t*)dst; v16u8 src0, dst0, tmp0; const v16u8 mask = { 1, 255, 1, 255, 5, 255, 5, 255, 9, 255, 9, 255, 13, 255, 13, 255 }; while (num_pixels >= 8) { v16u8 src1, dst1, tmp1; - LD_UB2(ptemp_data, 16, src0, src1); + LD_UB2(in, 16, src0, src1); VSHF_B2_UB(src0, src1, src1, src0, mask, mask, tmp0, tmp1); ADD2(src0, tmp0, src1, tmp1, dst0, dst1); - ST_UB2(dst0, dst1, ptemp_data, 16); - ptemp_data += 32; + ST_UB2(dst0, dst1, out, 16); + in += 32; + out += 32; num_pixels -= 8; } if (num_pixels > 0) { if (num_pixels >= 4) { - src0 = LD_UB(ptemp_data); + src0 = LD_UB(in); tmp0 = VSHF_UB(src0, src0, mask); dst0 = src0 + tmp0; - ST_UB(dst0, ptemp_data); - ptemp_data += 16; + ST_UB(dst0, out); + in += 16; + out += 16; num_pixels -= 4; } for (i = 0; i < num_pixels; i++) { - const uint8_t b = ptemp_data[0]; - const uint8_t g = ptemp_data[1]; - const uint8_t r = ptemp_data[2]; - ptemp_data[0] = (b + g) & 0xff; - ptemp_data[2] = (r + g) & 0xff; - ptemp_data += 4; + const uint8_t b = in[0]; + const uint8_t g = in[1]; + const uint8_t r = in[2]; + out[0] = (b + g) & 0xff; + out[1] = g; + out[2] = (r + g) & 0xff; + out[4] = in[4]; + out += 4; } } } static void TransformColorInverse(const VP8LMultipliers* const m, - uint32_t* data, int num_pixels) { + const uint32_t* src, int num_pixels, + uint32_t* dst) { v16u8 src0, dst0; const v16i8 g2br = (v16i8)__msa_fill_w(m->green_to_blue_ | (m->green_to_red_ << 16)); @@ -293,34 +300,36 @@ static void TransformColorInverse(const VP8LMultipliers* const m, while (num_pixels >= 8) { v16u8 src1, dst1; - LD_UB2(data, 4, src0, src1); + LD_UB2(src, 4, src0, src1); TRANSFORM_COLOR_INVERSE_8(src0, src1, dst0, dst1, g2br, r2b, mask0, mask1); - ST_UB2(dst0, dst1, data, 4); - data += 8; + ST_UB2(dst0, dst1, dst, 4); + src += 8; + dst += 8; num_pixels -= 8; } if (num_pixels > 0) { if (num_pixels >= 4) { - src0 = LD_UB(data); + src0 = LD_UB(src); TRANSFORM_COLOR_INVERSE_4(src0, dst0, g2br, r2b, mask0, mask1); - ST_UB(dst0, data); - data += 4; + ST_UB(dst0, dst); + src += 4; + dst += 4; num_pixels -= 4; } if (num_pixels > 0) { - src0 = LD_UB(data); + src0 = LD_UB(src); TRANSFORM_COLOR_INVERSE_4(src0, dst0, g2br, r2b, mask0, mask1); if (num_pixels == 3) { const uint64_t pix_d = __msa_copy_s_d((v2i64)dst0, 0); const uint32_t pix_w = __msa_copy_s_w((v4i32)dst0, 2); - SD(pix_d, data + 0); - SW(pix_w, data + 2); + SD(pix_d, dst + 0); + SW(pix_w, dst + 2); } else if (num_pixels == 2) { const uint64_t pix_d = __msa_copy_s_d((v2i64)dst0, 0); - SD(pix_d, data); + SD(pix_d, dst); } else { const uint32_t pix_w = __msa_copy_s_w((v4i32)dst0, 0); - SW(pix_w, data); + SW(pix_w, dst); } } } diff --git a/src/dsp/lossless_neon.c b/src/dsp/lossless_neon.c index 6faccb8f..509f0c52 100644 --- a/src/dsp/lossless_neon.c +++ b/src/dsp/lossless_neon.c @@ -171,28 +171,30 @@ static WEBP_INLINE uint8x16_t DoGreenShuffle(const uint8x16_t argb, } #endif // USE_VTBLQ -static void AddGreenToBlueAndRed(uint32_t* argb_data, int num_pixels) { - const uint32_t* const end = argb_data + (num_pixels & ~3); +static void AddGreenToBlueAndRed(const uint32_t* src, int num_pixels, + uint32_t* dst) { + const uint32_t* const end = src + (num_pixels & ~3); #ifdef USE_VTBLQ const uint8x16_t shuffle = vld1q_u8(kGreenShuffle); #else const uint8x8_t shuffle = vld1_u8(kGreenShuffle); #endif - for (; argb_data < end; argb_data += 4) { - const uint8x16_t argb = vld1q_u8((uint8_t*)argb_data); + for (; src < end; src += 4, dst += 4) { + const uint8x16_t argb = vld1q_u8((const uint8_t*)src); const uint8x16_t greens = DoGreenShuffle(argb, shuffle); - vst1q_u8((uint8_t*)argb_data, vaddq_u8(argb, greens)); + vst1q_u8((uint8_t*)dst, vaddq_u8(argb, greens)); } // fallthrough and finish off with plain-C - VP8LAddGreenToBlueAndRed_C(argb_data, num_pixels & 3); + VP8LAddGreenToBlueAndRed_C(src, num_pixels & 3, dst); } //------------------------------------------------------------------------------ // Color Transform static void TransformColorInverse(const VP8LMultipliers* const m, - uint32_t* argb_data, int num_pixels) { - // sign-extended multiplying constants, pre-shifted by 6. + const uint32_t* const src, int num_pixels, + uint32_t* dst) { +// sign-extended multiplying constants, pre-shifted by 6. #define CST(X) (((int16_t)(m->X << 8)) >> 6) const int16_t rb[8] = { CST(green_to_blue_), CST(green_to_red_), @@ -219,7 +221,7 @@ static void TransformColorInverse(const VP8LMultipliers* const m, const uint32x4_t mask_ag = vdupq_n_u32(0xff00ff00u); int i; for (i = 0; i + 4 <= num_pixels; i += 4) { - const uint8x16_t in = vld1q_u8((uint8_t*)(argb_data + i)); + const uint8x16_t in = vld1q_u8((const uint8_t*)(src + i)); const uint32x4_t a0g0 = vandq_u32(vreinterpretq_u32_u8(in), mask_ag); // 0 g 0 g const uint8x16_t greens = DoGreenShuffle(in, shuffle); @@ -240,10 +242,10 @@ static void TransformColorInverse(const VP8LMultipliers* const m, // 0 r' 0 b'' const uint16x8_t G = vshrq_n_u16(vreinterpretq_u16_s8(F), 8); const uint32x4_t out = vorrq_u32(vreinterpretq_u32_u16(G), a0g0); - vst1q_u32(argb_data + i, out); + vst1q_u32(dst + i, out); } // Fall-back to C-version for left-overs. - VP8LTransformColorInverse_C(m, argb_data + i, num_pixels - i); + VP8LTransformColorInverse_C(m, src + i, num_pixels - i, dst + i); } #undef USE_VTBLQ diff --git a/src/dsp/lossless_sse2.c b/src/dsp/lossless_sse2.c index 92dbdea9..fb09fe0c 100644 --- a/src/dsp/lossless_sse2.c +++ b/src/dsp/lossless_sse2.c @@ -157,26 +157,28 @@ static uint32_t Predictor13(uint32_t left, const uint32_t* const top) { //------------------------------------------------------------------------------ // Subtract-Green Transform -static void AddGreenToBlueAndRed(uint32_t* argb_data, int num_pixels) { +static void AddGreenToBlueAndRed(const uint32_t* const src, int num_pixels, + uint32_t* dst) { int i; for (i = 0; i + 4 <= num_pixels; i += 4) { - const __m128i in = _mm_loadu_si128((__m128i*)&argb_data[i]); // argb + const __m128i in = _mm_loadu_si128((const __m128i*)&src[i]); // argb const __m128i A = _mm_srli_epi16(in, 8); // 0 a 0 g const __m128i B = _mm_shufflelo_epi16(A, _MM_SHUFFLE(2, 2, 0, 0)); const __m128i C = _mm_shufflehi_epi16(B, _MM_SHUFFLE(2, 2, 0, 0)); // 0g0g const __m128i out = _mm_add_epi8(in, C); - _mm_storeu_si128((__m128i*)&argb_data[i], out); + _mm_storeu_si128((__m128i*)&dst[i], out); } // fallthrough and finish off with plain-C - VP8LAddGreenToBlueAndRed_C(argb_data + i, num_pixels - i); + VP8LAddGreenToBlueAndRed_C(src + i, num_pixels - i, dst + i); } //------------------------------------------------------------------------------ // Color Transform static void TransformColorInverse(const VP8LMultipliers* const m, - uint32_t* argb_data, int num_pixels) { - // sign-extended multiplying constants, pre-shifted by 5. + const uint32_t* const src, int num_pixels, + uint32_t* dst) { +// sign-extended multiplying constants, pre-shifted by 5. #define CST(X) (((int16_t)(m->X << 8)) >> 5) // sign-extend const __m128i mults_rb = _mm_set_epi16( CST(green_to_red_), CST(green_to_blue_), @@ -190,7 +192,7 @@ static void TransformColorInverse(const VP8LMultipliers* const m, const __m128i mask_ag = _mm_set1_epi32(0xff00ff00); // alpha-green masks int i; for (i = 0; i + 4 <= num_pixels; i += 4) { - const __m128i in = _mm_loadu_si128((__m128i*)&argb_data[i]); // argb + const __m128i in = _mm_loadu_si128((const __m128i*)&src[i]); // argb const __m128i A = _mm_and_si128(in, mask_ag); // a 0 g 0 const __m128i B = _mm_shufflelo_epi16(A, _MM_SHUFFLE(2, 2, 0, 0)); const __m128i C = _mm_shufflehi_epi16(B, _MM_SHUFFLE(2, 2, 0, 0)); // g0g0 @@ -202,10 +204,10 @@ static void TransformColorInverse(const VP8LMultipliers* const m, const __m128i I = _mm_add_epi8(H, F); // r' x b'' 0 const __m128i J = _mm_srli_epi16(I, 8); // 0 r' 0 b'' const __m128i out = _mm_or_si128(J, A); - _mm_storeu_si128((__m128i*)&argb_data[i], out); + _mm_storeu_si128((__m128i*)&dst[i], out); } // Fall-back to C-version for left-overs. - VP8LTransformColorInverse_C(m, argb_data + i, num_pixels - i); + VP8LTransformColorInverse_C(m, src + i, num_pixels - i, dst + i); } //------------------------------------------------------------------------------