From d2b2fb02f09b22d29a68f5725d2ce9936daa234c Mon Sep 17 00:00:00 2001 From: Hermet Park Date: Fri, 11 Jul 2025 02:16:16 +0900 Subject: [PATCH] sw_engine: rectifiy logic wrt overall blendings - minor optimization of the unpremultiply logic - exception handling for the unpremultipy logic for anti-aliasing quality - appropriate alpha pre/unpre multiplication is applied - clean code++ - updated doc blending equation issue: https://github.com/thorvg/thorvg/issues/1944 --- inc/thorvg.h | 8 +- src/renderer/sw_engine/tvgSwCommon.h | 182 ++++++++++++------------- src/renderer/sw_engine/tvgSwRaster.cpp | 28 ++-- 3 files changed, 105 insertions(+), 113 deletions(-) diff --git a/inc/thorvg.h b/inc/thorvg.h index f760c812..638481f0 100644 --- a/inc/thorvg.h +++ b/inc/thorvg.h @@ -200,15 +200,15 @@ enum class BlendMethod : uint8_t Normal = 0, ///< Perform the alpha blending(default). S if (Sa == 255), otherwise (Sa * S) + (255 - Sa) * D Multiply, ///< Takes the RGB channel values from 0 to 255 of each pixel in the top layer and multiples them with the values for the corresponding pixel from the bottom layer. (S * D) Screen, ///< The values of the pixels in the two layers are inverted, multiplied, and then inverted again. (S + D) - (S * D) - Overlay, ///< Combines Multiply and Screen blend modes. (2 * S * D) if (2 * D < Da), otherwise (Sa * Da) - 2 * (Da - S) * (Sa - D) + Overlay, ///< Combines Multiply and Screen blend modes. (2 * S * D) if (D < 128), otherwise 255 - 2 * (255 - S) * (255 - D) Darken, ///< Creates a pixel that retains the smallest components of the top and bottom layer pixels. min(S, D) Lighten, ///< Only has the opposite action of Darken Only. max(S, D) ColorDodge, ///< Divides the bottom layer by the inverted top layer. D / (255 - S) ColorBurn, ///< Divides the inverted bottom layer by the top layer, and then inverts the result. 255 - (255 - D) / S - HardLight, ///< The same as Overlay but with the color roles reversed. (2 * S * D) if (S < Sa), otherwise (Sa * Da) - 2 * (Da - S) * (Sa - D) - SoftLight, ///< The same as Overlay but with applying pure black or white does not result in pure black or white. (1 - 2 * S) * (D ^ 2) + (2 * S * D) + HardLight, ///< The same as Overlay but with the color roles reversed. (2 * S * D) if (S < 128), otherwise 255 - 2 * (255 - S) * (255 - D) + SoftLight, ///< The same as Overlay but with applying pure black or white does not result in pure black or white. (255 - 2 * S) * (D * D) + (2 * S * D) Difference, ///< Subtracts the bottom layer from the top layer or the other way around, to always get a non-negative value. (S - D) if (S > D), otherwise (D - S) - Exclusion, ///< The result is twice the product of the top and bottom layers, subtracted from their sum. s + d - (2 * s * d) + Exclusion, ///< The result is twice the product of the top and bottom layers, subtracted from their sum. S + D - (2 * S * D) Hue, ///< Reserved. Not supported. Saturation, ///< Reserved. Not supported. Color, ///< Reserved. Not supported. diff --git a/src/renderer/sw_engine/tvgSwCommon.h b/src/renderer/sw_engine/tvgSwCommon.h index 6ef53448..77380731 100644 --- a/src/renderer/sw_engine/tvgSwCommon.h +++ b/src/renderer/sw_engine/tvgSwCommon.h @@ -373,18 +373,34 @@ static inline uint8_t C3(uint32_t c) return (c); } -static inline bool UNPREMULTIPLY(uint32_t color, RenderColor& out) +static inline uint32_t PREMULTIPLY(uint32_t c, uint8_t a) { - out.a = A(color); - if (out.a == 0) return false; + return (c & 0xff000000) + ((((c >> 8) & 0xff) * a) & 0xff00) + ((((c & 0x00ff00ff) * a) >> 8) & 0x00ff00ff); +} - out.r = C1(color) * 255 / out.a; - out.g = C2(color) * 255 / out.a; - out.b = C3(color) * 255 / out.a; +static inline bool BLEND_UPRE(uint32_t c, RenderColor& o) +{ + o.a = A(c); + if (o.a == 0) return false; + + o.r = C1(c); + o.g = C2(c); + o.b = C3(c); + + if (o.a < 255) { + o.r = std::min(o.r * 255u / o.a, 255u); + o.g = std::min(o.g * 255u / o.a, 255u); + o.b = std::min(o.b * 255u / o.a, 255u); + } return true; } +static inline uint32_t BLEND_PRE(uint32_t c1, uint32_t c2, uint8_t a) +{ + return ALPHA_BLEND(c1, a) + ALPHA_BLEND(c2, 255 - a); +} + static inline uint32_t opBlendInterp(uint32_t s, uint32_t d, uint8_t a) { return INTERPOLATE(s, d, a); @@ -411,160 +427,140 @@ static inline uint32_t opBlendDifference(uint32_t s, uint32_t d, TVG_UNUSED uint { if (d == 0) return s; - //if (s > d) => s - d - //else => d - s - auto c1 = (C1(s) > C1(d)) ? (C1(s) - C1(d)) : (C1(d) - C1(s)); - auto c2 = (C2(s) > C2(d)) ? (C2(s) - C2(d)) : (C2(d) - C2(s)); - auto c3 = (C3(s) > C3(d)) ? (C3(s) - C3(d)) : (C3(d) - C3(s)); - return JOIN(255, c1, c2, c3); + auto f = [](uint8_t s, uint8_t d) { + return (s > d) ? (s - d) : (d - s); + }; + + return JOIN(255, f(C1(s), C1(d)), f(C2(s), C2(d)), f(C3(s), C3(d))); } static inline uint32_t opBlendExclusion(uint32_t s, uint32_t d, TVG_UNUSED uint8_t a) { if (d == 0) return s; - // (s + d) - (2 * s * d) - auto c1 = tvg::clamp(C1(s) + C1(d) - 2 * MULTIPLY(C1(s), C1(d)), 0, 255); - auto c2 = tvg::clamp(C2(s) + C2(d) - 2 * MULTIPLY(C2(s), C2(d)), 0, 255); - auto c3 = tvg::clamp(C3(s) + C3(d) - 2 * MULTIPLY(C3(s), C3(d)), 0, 255); - return JOIN(255, c1, c2, c3); + auto f = [](uint8_t s, uint8_t d) { + return tvg::clamp(s + d - 2 * MULTIPLY(s, d), 0, 255); + }; + + return JOIN(255, f(C1(s), C1(d)), f(C2(s), C2(d)), f(C3(s), C3(d))); } static inline uint32_t opBlendAdd(uint32_t s, uint32_t d, TVG_UNUSED uint8_t a) { if (d == 0) return s; - // s + d - auto c1 = std::min(C1(s) + C1(d), 255); - auto c2 = std::min(C2(s) + C2(d), 255); - auto c3 = std::min(C3(s) + C3(d), 255); - return JOIN(255, c1, c2, c3); + auto f = [](uint8_t s, uint8_t d) { + return std::min(s + d, 255); + }; + + return JOIN(255, f(C1(s), C1(d)), f(C2(s), C2(d)), f(C3(s), C3(d))); } static inline uint32_t opBlendScreen(uint32_t s, uint32_t d, TVG_UNUSED uint8_t a) { if (d == 0) return s; - // s + d - s * d - auto c1 = C1(s) + C1(d) - MULTIPLY(C1(s), C1(d)); - auto c2 = C2(s) + C2(d) - MULTIPLY(C2(s), C2(d)); - auto c3 = C3(s) + C3(d) - MULTIPLY(C3(s), C3(d)); - return JOIN(255, c1, c2, c3); + auto f = [](uint8_t s, uint8_t d) { + return s + d - MULTIPLY(s, d); + }; + + return JOIN(255, f(C1(s), C1(d)), f(C2(s), C2(d)), f(C3(s), C3(d))); } static inline uint32_t opBlendMultiply(uint32_t s, uint32_t d, TVG_UNUSED uint8_t a) { - if (d == 0) return s; - RenderColor o; - if (!UNPREMULTIPLY(d, o)) return 0; + if (!BLEND_UPRE(d, o)) return s; - // s * d - auto c1 = MULTIPLY(C1(s), o.r); - auto c2 = MULTIPLY(C2(s), o.g); - auto c3 = MULTIPLY(C3(s), o.b); - return JOIN(255, c1, c2, c3); + auto f = [](uint8_t s, uint8_t d) { + return MULTIPLY(s, d); + }; + + return BLEND_PRE(JOIN(255, f(C1(s), o.r), f(C2(s), o.g), f(C3(s), o.b)), s, o.a); } static inline uint32_t opBlendOverlay(uint32_t s, uint32_t d, TVG_UNUSED uint8_t a) { - if (d == 0) return s; - RenderColor o; - if (!UNPREMULTIPLY(d, o)) return 0; + if (!BLEND_UPRE(d, o)) return s; - // if (2 * d < da) => 2 * s * d, - // else => 1 - 2 * (1 - s) * (1 - d) - auto c1 = (o.r < 128) ? std::min(255, 2 * MULTIPLY(C1(s), o.r)) : (255 - std::min(255, 2 * MULTIPLY(255 - C1(s), 255 - o.r))); - auto c2 = (o.g < 128) ? std::min(255, 2 * MULTIPLY(C2(s), o.g)) : (255 - std::min(255, 2 * MULTIPLY(255 - C2(s), 255 - o.g))); - auto c3 = (o.b < 128) ? std::min(255, 2 * MULTIPLY(C3(s), o.b)) : (255 - std::min(255, 2 * MULTIPLY(255 - C3(s), 255 - o.b))); - return JOIN(255, c1, c2, c3); + auto f = [](uint8_t s, uint8_t d) { + return (d < 128) ? std::min(255, 2 * MULTIPLY(s, d)) : (255 - std::min(255, 2 * MULTIPLY(255 - s, 255 - d))); + }; + + return BLEND_PRE(JOIN(255, f(C1(s), o.r), f(C2(s), o.g), f(C3(s), o.b)), s, o.a); } static inline uint32_t opBlendDarken(uint32_t s, uint32_t d, TVG_UNUSED uint8_t a) { - if (d == 0) return s; - RenderColor o; - if (!UNPREMULTIPLY(d, o)) return 0; + if (!BLEND_UPRE(d, o)) return s; - // min(s, d) - auto c1 = std::min(C1(s), o.r); - auto c2 = std::min(C2(s), o.g); - auto c3 = std::min(C3(s), o.b); - return JOIN(255, c1, c2, c3); + auto f = [](uint8_t s, uint8_t d) { + return std::min(s, d); + }; + + return BLEND_PRE(JOIN(255, f(C1(s), o.r), f(C2(s), o.g), f(C3(s), o.b)), s, o.a); } static inline uint32_t opBlendLighten(uint32_t s, uint32_t d, TVG_UNUSED uint8_t a) { if (d == 0) return s; - // max(s, d) - auto c1 = std::max(C1(s), C1(d)); - auto c2 = std::max(C2(s), C2(d)); - auto c3 = std::max(C3(s), C3(d)); - return JOIN(255, c1, c2, c3); + auto f = [](uint8_t s, uint8_t d) { + return std::max(s, d); + }; + + return JOIN(255, f(C1(s), C1(d)), f(C2(s), C2(d)), f(C3(s), C3(d))); } static inline uint32_t opBlendColorDodge(uint32_t s, uint32_t d, TVG_UNUSED uint8_t a) { - if (d == 0) return s; + RenderColor o; + if (!BLEND_UPRE(d, o)) return s; - // d / (1 - s) - s = 0xffffffff - s; - auto c1 = C1(d) == 0 ? 0 : (C1(s) == 0 ? 255 : std::min(C1(d) * 255 / C1(s), 255)); - auto c2 = C2(d) == 0 ? 0 : (C2(s) == 0 ? 255 : std::min(C2(d) * 255 / C2(s), 255)); - auto c3 = C3(d) == 0 ? 0 : (C3(s) == 0 ? 255 : std::min(C3(d) * 255 / C3(s), 255)); - return JOIN(255, c1, c2, c3); + auto f = [](uint8_t s, uint8_t d) { + return d == 0 ? 0 : (s == 255 ? 255 : std::min(d * 255 / (255 - s), 255)); + }; + + return BLEND_PRE(JOIN(255, f(C1(s), o.r), f(C2(s), o.g), f(C3(s), o.b)), s, o.a); } static inline uint32_t opBlendColorBurn(uint32_t s, uint32_t d, TVG_UNUSED uint8_t a) { - if (d == 0) return s; - RenderColor o; - if (!UNPREMULTIPLY(d, o)) o.r = o.g = o.b = 0; + if (!BLEND_UPRE(d, o)) return s; - // 1 - (1 - d) / s - auto ir = 255 - o.r; - auto ig = 255 - o.g; - auto ib = 255 - o.b; + auto f = [](uint8_t s, uint8_t d) { + return d == 255 ? 255 : (s == 0 ? 0 : 255 - std::min((255 - d) * 255 / s, 255)); + }; - auto c1 = o.r == 255 ? 255 : (C1(s) == 0 ? 0 : 255 - std::min(ir * 255 / C1(s), 255)); - auto c2 = o.g == 255 ? 255 : (C2(s) == 0 ? 0 : 255 - std::min(ig * 255 / C2(s), 255)); - auto c3 = o.b == 255 ? 255 : (C3(s) == 0 ? 0 : 255 - std::min(ib * 255 / C3(s), 255)); - - return JOIN(255, c1, c2, c3); + return BLEND_PRE(JOIN(255, f(C1(s), o.r), f(C2(s), o.g), f(C3(s), o.b)), s, o.a); } static inline uint32_t opBlendHardLight(uint32_t s, uint32_t d, TVG_UNUSED uint8_t a) { - if (d == 0) return s; - RenderColor o; - if (!UNPREMULTIPLY(d, o)) o.r = o.g = o.b = 0; + if (!BLEND_UPRE(d, o)) return s; - // if (s < sa), (2 * s * d) - // else (sa * da) - 2 * (da - s) * (sa - d) - auto c1 = (C1(s) < 128) ? std::min(255, 2 * MULTIPLY(C1(s), o.r)) : (255 - std::min(255, 2 * MULTIPLY(255 - C1(s), 255 - o.r))); - auto c2 = (C2(s) < 128) ? std::min(255, 2 * MULTIPLY(C2(s), o.g)) : (255 - std::min(255, 2 * MULTIPLY(255 - C2(s), 255 - o.g))); - auto c3 = (C3(s) < 128) ? std::min(255, 2 * MULTIPLY(C3(s), o.b)) : (255 - std::min(255, 2 * MULTIPLY(255 - C3(s), 255 - o.b))); - return JOIN(255, c1, c2, c3); + auto f = [](uint8_t s, uint8_t d) { + return (s < 128) ? std::min(255, 2 * MULTIPLY(s, d)) : (255 - std::min(255, 2 * MULTIPLY(255 - s, 255 - d))); + }; + + return BLEND_PRE(JOIN(255, f(C1(s), o.r), f(C2(s), o.g), f(C3(s), o.b)), s, o.a); } static inline uint32_t opBlendSoftLight(uint32_t s, uint32_t d, TVG_UNUSED uint8_t a) { - if (d == 0) return s; - RenderColor o; - if (!UNPREMULTIPLY(d, o)) return 0; + if (!BLEND_UPRE(d, o)) return s; - //(255 - 2 * s) * (d * d) + (2 * s * d) - auto c1 = MULTIPLY(255 - std::min(255, 2 * C1(s)), MULTIPLY(o.r, o.r)) + std::min(255, 2 * MULTIPLY(C1(s), o.r)); - auto c2 = MULTIPLY(255 - std::min(255, 2 * C2(s)), MULTIPLY(o.g, o.g)) + std::min(255, 2 * MULTIPLY(C2(s), o.g)); - auto c3 = MULTIPLY(255 - std::min(255, 2 * C3(s)), MULTIPLY(o.b, o.b)) + std::min(255, 2 * MULTIPLY(C3(s), o.b)); - return JOIN(255, c1, c2, c3); + auto f = [](uint8_t s, uint8_t d) { + return MULTIPLY(255 - std::min(255, 2 * s), MULTIPLY(d, d)) + std::min(255, 2 * MULTIPLY(s, d)); + }; + + return BLEND_PRE(JOIN(255, f(C1(s), o.r), f(C2(s), o.g), f(C3(s), o.b)), s, o.a); } diff --git a/src/renderer/sw_engine/tvgSwRaster.cpp b/src/renderer/sw_engine/tvgSwRaster.cpp index 20767463..c09d0070 100644 --- a/src/renderer/sw_engine/tvgSwRaster.cpp +++ b/src/renderer/sw_engine/tvgSwRaster.cpp @@ -591,8 +591,7 @@ static bool _rasterBlendingRle(SwSurface* surface, const SwRle* rle, const Rende } } else { for (auto x = 0; x < len; ++x, ++dst) { - auto tmp = surface->blender(color, *dst, 255); - *dst = INTERPOLATE(tmp, *dst, span->coverage); + *dst = INTERPOLATE(surface->blender(color, *dst, 255), *dst, span->coverage); } } } @@ -733,15 +732,13 @@ static bool _rasterScaledBlendingRleImage(SwSurface* surface, const SwImage& ima for (uint32_t x = static_cast(span->x); x < static_cast(span->x) + span->len; ++x, ++dst) { SCALED_IMAGE_RANGE_X auto src = scaleMethod(image.buf32, image.stride, image.w, image.h, sx, sy, miny, maxy, sampleSize); - auto tmp = surface->blender(src, *dst, 255); - *dst = INTERPOLATE(tmp, *dst, A(src)); + *dst = INTERPOLATE(surface->blender(src, *dst, 255), *dst, A(src)); } } else { for (uint32_t x = static_cast(span->x); x < static_cast(span->x) + span->len; ++x, ++dst) { SCALED_IMAGE_RANGE_X auto src = scaleMethod(image.buf32, image.stride, image.w, image.h, sx, sy, miny, maxy, sampleSize); - auto tmp = surface->blender(src, *dst, 255); - *dst = INTERPOLATE(tmp, *dst, MULTIPLY(alpha, A(src))); + *dst = INTERPOLATE(surface->blender(src, *dst, 255), *dst, MULTIPLY(alpha, A(src))); } } } @@ -822,8 +819,7 @@ static bool _rasterDirectBlendingRleImage(SwSurface* surface, const SwImage& ima } } else { for (auto x = 0; x < len; ++x, ++dst, ++img) { - auto tmp = surface->blender(*img, *dst, 255); - *dst = INTERPOLATE(tmp, *dst, MULTIPLY(alpha, A(*img))); + *dst = INTERPOLATE(surface->blender(*img, *dst, 255), *dst, MULTIPLY(alpha, A(*img))); } } } @@ -918,8 +914,7 @@ static bool _rasterScaledBlendingImage(SwSurface* surface, const SwImage& image, for (auto x = bbox.min.x; x < bbox.max.x; ++x, ++dst) { SCALED_IMAGE_RANGE_X auto src = scaleMethod(image.buf32, image.stride, image.w, image.h, sx, sy, miny, maxy, sampleSize); - auto tmp = surface->blender(src, *dst, 255); - *dst = INTERPOLATE(tmp, *dst, MULTIPLY(opacity, A(src))); + *dst = INTERPOLATE(surface->blender(src, *dst, 255), *dst, MULTIPLY(opacity, A(src))); } } return true; @@ -1533,9 +1528,11 @@ uint32_t rasterUnpremultiply(uint32_t data) { auto a = A(data); if (a == 255 || a == 0) return data; - auto r = C1(data) * 255 / a; - auto g = C2(data) * 255 / a; - auto b = C3(data) * 255 / a; + + uint8_t r = std::min(C1(data) * 255u / a, 255u); + uint8_t g = std::min(C2(data) * 255u / a, 255u); + uint8_t b = std::min(C3(data) * 255u / a, 255u); + return JOIN(a, r, g, b); } @@ -1571,9 +1568,8 @@ void rasterPremultiply(RenderSurface* surface) auto dst = buffer; for (uint32_t x = 0; x < surface->w; ++x, ++dst) { auto c = *dst; - auto a = (c >> 24); - if (a == 255) continue; - *dst = (c & 0xff000000) + ((((c >> 8) & 0xff) * a) & 0xff00) + ((((c & 0x00ff00ff) * a) >> 8) & 0x00ff00ff); + if (A(c) == 255) continue; + *dst = PREMULTIPLY(c, A(c)); } } }