diff --git a/internal/cgen/base/image-impl.c b/internal/cgen/base/image-impl.c index 00ef3633..9fdf7ca8 100644 --- a/internal/cgen/base/image-impl.c +++ b/internal/cgen/base/image-impl.c @@ -175,6 +175,18 @@ wuffs_base__pixel_buffer__set_color_u32_at( // Common formats above. Rarer formats below. + case WUFFS_BASE__PIXEL_FORMAT__Y: + wuffs_base__store_u8__no_bounds_check( + row + ((size_t)x), wuffs_base__color_u32_argb_premul__as_gray(color)); + break; + + case WUFFS_BASE__PIXEL_FORMAT__INDEXED__BGRA_BINARY: + wuffs_base__store_u8__no_bounds_check( + row + ((size_t)x), wuffs_base__pixel_palette__closest_element( + wuffs_base__pixel_buffer__palette(pb), + pb->pixcfg.private_impl.pixfmt, color)); + break; + case WUFFS_BASE__PIXEL_FORMAT__BGR_565: { uint32_t b5 = 0x1F & (color >> (8 - 5)); uint32_t g6 = 0x3F & (color >> (16 - 6)); @@ -218,6 +230,68 @@ wuffs_base__pixel_buffer__set_color_u32_at( // -------- +uint8_t // +wuffs_base__pixel_palette__closest_element( + wuffs_base__slice_u8 palette_slice, + wuffs_base__pixel_format palette_format, + wuffs_base__color_u32_argb_premul c) { + size_t n = palette_slice.len / 4; + if (n > 256) { + n = 256; + } + size_t best_index = 0; + uint64_t best_score = 0xFFFFFFFFFFFFFFFF; + + // Work in 16-bit color. + uint32_t ca = 0x101 * (0xFF & (c >> 24)); + uint32_t cr = 0x101 * (0xFF & (c >> 16)); + uint32_t cg = 0x101 * (0xFF & (c >> 8)); + uint32_t cb = 0x101 * (0xFF & (c >> 0)); + + switch (palette_format.repr) { + case WUFFS_BASE__PIXEL_FORMAT__INDEXED__BGRA_NONPREMUL: + case WUFFS_BASE__PIXEL_FORMAT__INDEXED__BGRA_PREMUL: + case WUFFS_BASE__PIXEL_FORMAT__INDEXED__BGRA_BINARY: { + bool nonpremul = palette_format.repr == + WUFFS_BASE__PIXEL_FORMAT__INDEXED__BGRA_NONPREMUL; + + size_t i; + for (i = 0; i < n; i++) { + // Work in 16-bit color. + uint32_t pb = 0x101 * ((uint32_t)(palette_slice.ptr[(4 * i) + 0])); + uint32_t pg = 0x101 * ((uint32_t)(palette_slice.ptr[(4 * i) + 1])); + uint32_t pr = 0x101 * ((uint32_t)(palette_slice.ptr[(4 * i) + 2])); + uint32_t pa = 0x101 * ((uint32_t)(palette_slice.ptr[(4 * i) + 3])); + + // Convert to premultiplied alpha. + if (nonpremul && (pa != 0xFFFF)) { + pb = (pb * pa) / 0xFFFF; + pg = (pg * pa) / 0xFFFF; + pr = (pr * pa) / 0xFFFF; + } + + // These deltas are conceptually int32_t (signed) but after squaring, + // it's equivalent to work in uint32_t (unsigned). + pb -= cb; + pg -= cg; + pr -= cr; + pa -= ca; + uint64_t score = ((uint64_t)(pb * pb)) + ((uint64_t)(pg * pg)) + + ((uint64_t)(pr * pr)) + ((uint64_t)(pa * pa)); + if (best_score > score) { + best_score = score; + best_index = i; + } + } + break; + } + } + + return best_index; +} + +// -------- + static uint64_t // wuffs_base__pixel_swizzler__bgra_premul__bgra_nonpremul__src( wuffs_base__slice_u8 dst, diff --git a/internal/cgen/base/image-public.h b/internal/cgen/base/image-public.h index 1fd3f283..b235f2a1 100644 --- a/internal/cgen/base/image-public.h +++ b/internal/cgen/base/image-public.h @@ -21,6 +21,23 @@ // 0xAARRGGBB (Alpha most significant, Blue least), regardless of endianness. typedef uint32_t wuffs_base__color_u32_argb_premul; +static inline uint8_t // +wuffs_base__color_u32_argb_premul__as_gray( + wuffs_base__color_u32_argb_premul c) { + // Work in 16-bit color. + uint32_t cr = 0x101 * (0xFF & (c >> 16)); + uint32_t cg = 0x101 * (0xFF & (c >> 8)); + uint32_t cb = 0x101 * (0xFF & (c >> 0)); + + // These coefficients (the fractions 0.299, 0.587 and 0.114) are the same + // as those given by the JFIF specification. + // + // Note that 19595 + 38470 + 7471 equals 65536, also known as (1 << 16). We + // shift by 24, not just by 16, because the return value is 8-bit color, not + // 16-bit color. + return ((19595 * cr) + (38470 * cg) + (7471 * cb) + 32768) >> 24; +} + // wuffs_base__premul_u32_axxx converts from non-premultiplied alpha to // premultiplied alpha. The "axxx" means either "argb" or "abgr". static inline uint32_t // @@ -1229,6 +1246,23 @@ typedef struct { // -------- +// wuffs_base__pixel_palette__closest_element returns the index of the palette +// element that minimizes the sum of squared differences of the four ARGB +// channels, working in premultiplied alpha. Ties favor the smaller index. +// +// The palette_slice.len may equal (N*4), for N less than 256, which means that +// only the first N palette elements are considered. It returns 0 when N is 0. +// +// Applying this function on a per-pixel basis will not produce whole-of-image +// dithering. +uint8_t // +wuffs_base__pixel_palette__closest_element( + wuffs_base__slice_u8 palette_slice, + wuffs_base__pixel_format palette_format, + wuffs_base__color_u32_argb_premul c); + +// -------- + // TODO: should the func type take restrict pointers? typedef uint64_t (*wuffs_base__pixel_swizzler__func)( wuffs_base__slice_u8 dst, diff --git a/internal/cgen/data.go b/internal/cgen/data.go index 2285cdf2..de410ef6 100644 --- a/internal/cgen/data.go +++ b/internal/cgen/data.go @@ -32,9 +32,12 @@ const baseImageImplC = "" + "te_impl.planes[0].stride;\n uint8_t* row = pb->private_impl.planes[0].ptr + (stride * ((size_t)y));\n\n switch (pb->pixcfg.private_impl.pixfmt.repr) {\n case WUFFS_BASE__PIXEL_FORMAT__BGRA_PREMUL:\n case WUFFS_BASE__PIXEL_FORMAT__BGRA_BINARY:\n return wuffs_base__load_u32le__no_bounds_check(row + (4 * ((size_t)x)));\n\n case WUFFS_BASE__PIXEL_FORMAT__INDEXED__BGRA_PREMUL:\n case WUFFS_BASE__PIXEL_FORMAT__INDEXED__BGRA_BINARY: {\n uint8_t* palette = pb->private_impl.planes[3].ptr;\n return wuffs_base__load_u32le__no_bounds_check(palette +\n (4 * ((size_t)row[x])));\n }\n\n // Common formats above. Rarer formats below.\n\n case WUFFS_BASE__PIXEL_FORMAT__Y:\n return 0xFF000000 | (0x00010101 * ((uint32_t)(row[x])));\n\n case WUFFS_BASE__PIXEL_FORMAT__INDEXED__BGRA_NONPREMUL: {\n uint8_t* palette = pb->private_impl.planes[3].ptr;\n return wuffs_base__premul_u32_axxx(\n wuffs_base__load_u32le__no_bounds_check(palette +\n " + " (4 * ((size_t)row[x]))));\n }\n\n case WUFFS_BASE__PIXEL_FORMAT__BGR_565: {\n uint16_t bgr =\n wuffs_base__load_u16le__no_bounds_check(row + (2 * ((size_t)x)));\n uint32_t b5 = 0x1F & (bgr >> 0);\n uint32_t b = (b5 << 3) | (b5 >> 2);\n uint32_t g6 = 0x3F & (bgr >> 5);\n uint32_t g = (g6 << 2) | (g6 >> 4);\n uint32_t r5 = 0x1F & (bgr >> 11);\n uint32_t r = (r5 << 3) | (r5 >> 2);\n return 0xFF000000 | (r << 16) | (g << 8) | (b << 0);\n }\n case WUFFS_BASE__PIXEL_FORMAT__BGR:\n return 0xFF000000 |\n wuffs_base__load_u24le__no_bounds_check(row + (3 * ((size_t)x)));\n case WUFFS_BASE__PIXEL_FORMAT__BGRA_NONPREMUL:\n return wuffs_base__premul_u32_axxx(\n wuffs_base__load_u32le__no_bounds_check(row + (4 * ((size_t)x))));\n case WUFFS_BASE__PIXEL_FORMAT__BGRX:\n return 0xFF000000 |\n wuffs_base__load_u32le__no_bounds_check(row + (4 * ((size_t)x)));\n\n case WUFFS_BASE__PIXEL_F" + "ORMAT__RGB:\n return wuffs_base__swap_u32_argb_abgr(\n 0xFF000000 |\n wuffs_base__load_u24le__no_bounds_check(row + (3 * ((size_t)x))));\n case WUFFS_BASE__PIXEL_FORMAT__RGBA_NONPREMUL:\n return wuffs_base__swap_u32_argb_abgr(wuffs_base__premul_u32_axxx(\n wuffs_base__load_u32le__no_bounds_check(row + (4 * ((size_t)x)))));\n case WUFFS_BASE__PIXEL_FORMAT__RGBA_PREMUL:\n case WUFFS_BASE__PIXEL_FORMAT__RGBA_BINARY:\n return wuffs_base__swap_u32_argb_abgr(\n wuffs_base__load_u32le__no_bounds_check(row + (4 * ((size_t)x))));\n case WUFFS_BASE__PIXEL_FORMAT__RGBX:\n return wuffs_base__swap_u32_argb_abgr(\n 0xFF000000 |\n wuffs_base__load_u32le__no_bounds_check(row + (4 * ((size_t)x))));\n\n default:\n // TODO: support more formats.\n break;\n }\n\n return 0;\n}\n\nwuffs_base__status //\nwuffs_base__pixel_buffer__set_color_u32_at(\n wuffs_base__pixel_buffer* pb,\n uint32_t x,\n uint32_t y,\n wuffs_base__color_u32_argb_premul color)" + - " {\n if (!pb) {\n return wuffs_base__make_status(wuffs_base__error__bad_receiver);\n }\n if ((x >= pb->pixcfg.private_impl.width) ||\n (y >= pb->pixcfg.private_impl.height)) {\n return wuffs_base__make_status(wuffs_base__error__bad_argument);\n }\n\n if (wuffs_base__pixel_format__is_planar(&pb->pixcfg.private_impl.pixfmt)) {\n // TODO: support planar formats.\n return wuffs_base__make_status(wuffs_base__error__unsupported_option);\n }\n\n size_t stride = pb->private_impl.planes[0].stride;\n uint8_t* row = pb->private_impl.planes[0].ptr + (stride * ((size_t)y));\n\n switch (pb->pixcfg.private_impl.pixfmt.repr) {\n case WUFFS_BASE__PIXEL_FORMAT__BGRA_PREMUL:\n case WUFFS_BASE__PIXEL_FORMAT__BGRX:\n wuffs_base__store_u32le__no_bounds_check(row + (4 * ((size_t)x)), color);\n break;\n\n // Common formats above. Rarer formats below.\n\n case WUFFS_BASE__PIXEL_FORMAT__BGR_565: {\n uint32_t b5 = 0x1F & (color >> (8 - 5));\n uint32_t g6 = 0x3F & (color >> (16 - 6));\n uint32_t r5 =" + - " 0x1F & (color >> (24 - 5));\n uint32_t bgr565 = (b5 << 0) | (g6 << 5) | (r5 << 11);\n wuffs_base__store_u16le__no_bounds_check(row + (2 * ((size_t)x)),\n (uint16_t)bgr565);\n break;\n }\n case WUFFS_BASE__PIXEL_FORMAT__BGR:\n wuffs_base__store_u24le__no_bounds_check(row + (3 * ((size_t)x)), color);\n break;\n case WUFFS_BASE__PIXEL_FORMAT__BGRA_NONPREMUL:\n wuffs_base__store_u32le__no_bounds_check(\n row + (4 * ((size_t)x)), wuffs_base__nonpremul_u32_axxx(color));\n break;\n\n case WUFFS_BASE__PIXEL_FORMAT__RGB:\n wuffs_base__store_u24le__no_bounds_check(\n row + (3 * ((size_t)x)), wuffs_base__swap_u32_argb_abgr(color));\n break;\n case WUFFS_BASE__PIXEL_FORMAT__RGBA_NONPREMUL:\n wuffs_base__store_u32le__no_bounds_check(\n row + (4 * ((size_t)x)), wuffs_base__nonpremul_u32_axxx(\n wuffs_base__swap_u32_argb_abgr(color)));\n break;\n case WUFFS_BASE__P" + - "IXEL_FORMAT__RGBA_PREMUL:\n WUFFS_BASE__FALLTHROUGH;\n case WUFFS_BASE__PIXEL_FORMAT__RGBX:\n wuffs_base__store_u32le__no_bounds_check(\n row + (4 * ((size_t)x)), wuffs_base__swap_u32_argb_abgr(color));\n break;\n\n default:\n // TODO: support more formats.\n return wuffs_base__make_status(wuffs_base__error__unsupported_option);\n }\n\n return wuffs_base__make_status(NULL);\n}\n\n" + + " {\n if (!pb) {\n return wuffs_base__make_status(wuffs_base__error__bad_receiver);\n }\n if ((x >= pb->pixcfg.private_impl.width) ||\n (y >= pb->pixcfg.private_impl.height)) {\n return wuffs_base__make_status(wuffs_base__error__bad_argument);\n }\n\n if (wuffs_base__pixel_format__is_planar(&pb->pixcfg.private_impl.pixfmt)) {\n // TODO: support planar formats.\n return wuffs_base__make_status(wuffs_base__error__unsupported_option);\n }\n\n size_t stride = pb->private_impl.planes[0].stride;\n uint8_t* row = pb->private_impl.planes[0].ptr + (stride * ((size_t)y));\n\n switch (pb->pixcfg.private_impl.pixfmt.repr) {\n case WUFFS_BASE__PIXEL_FORMAT__BGRA_PREMUL:\n case WUFFS_BASE__PIXEL_FORMAT__BGRX:\n wuffs_base__store_u32le__no_bounds_check(row + (4 * ((size_t)x)), color);\n break;\n\n // Common formats above. Rarer formats below.\n\n case WUFFS_BASE__PIXEL_FORMAT__Y:\n wuffs_base__store_u8__no_bounds_check(\n row + ((size_t)x), wuffs_base__color_u32_argb_premul__as_gray(color" + + "));\n break;\n\n case WUFFS_BASE__PIXEL_FORMAT__INDEXED__BGRA_BINARY:\n wuffs_base__store_u8__no_bounds_check(\n row + ((size_t)x), wuffs_base__pixel_palette__closest_element(\n wuffs_base__pixel_buffer__palette(pb),\n pb->pixcfg.private_impl.pixfmt, color));\n break;\n\n case WUFFS_BASE__PIXEL_FORMAT__BGR_565: {\n uint32_t b5 = 0x1F & (color >> (8 - 5));\n uint32_t g6 = 0x3F & (color >> (16 - 6));\n uint32_t r5 = 0x1F & (color >> (24 - 5));\n uint32_t bgr565 = (b5 << 0) | (g6 << 5) | (r5 << 11);\n wuffs_base__store_u16le__no_bounds_check(row + (2 * ((size_t)x)),\n (uint16_t)bgr565);\n break;\n }\n case WUFFS_BASE__PIXEL_FORMAT__BGR:\n wuffs_base__store_u24le__no_bounds_check(row + (3 * ((size_t)x)), color);\n break;\n case WUFFS_BASE__PIXEL_FORMAT__BGRA_NONPREMUL:\n wuffs_base__store_u32le__no_bounds_check(\n row + (4 * ((size_t)x))" + + ", wuffs_base__nonpremul_u32_axxx(color));\n break;\n\n case WUFFS_BASE__PIXEL_FORMAT__RGB:\n wuffs_base__store_u24le__no_bounds_check(\n row + (3 * ((size_t)x)), wuffs_base__swap_u32_argb_abgr(color));\n break;\n case WUFFS_BASE__PIXEL_FORMAT__RGBA_NONPREMUL:\n wuffs_base__store_u32le__no_bounds_check(\n row + (4 * ((size_t)x)), wuffs_base__nonpremul_u32_axxx(\n wuffs_base__swap_u32_argb_abgr(color)));\n break;\n case WUFFS_BASE__PIXEL_FORMAT__RGBA_PREMUL:\n WUFFS_BASE__FALLTHROUGH;\n case WUFFS_BASE__PIXEL_FORMAT__RGBX:\n wuffs_base__store_u32le__no_bounds_check(\n row + (4 * ((size_t)x)), wuffs_base__swap_u32_argb_abgr(color));\n break;\n\n default:\n // TODO: support more formats.\n return wuffs_base__make_status(wuffs_base__error__unsupported_option);\n }\n\n return wuffs_base__make_status(NULL);\n}\n\n" + + "" + + "// --------\n\nuint8_t //\nwuffs_base__pixel_palette__closest_element(\n wuffs_base__slice_u8 palette_slice,\n wuffs_base__pixel_format palette_format,\n wuffs_base__color_u32_argb_premul c) {\n size_t n = palette_slice.len / 4;\n if (n > 256) {\n n = 256;\n }\n size_t best_index = 0;\n uint64_t best_score = 0xFFFFFFFFFFFFFFFF;\n\n // Work in 16-bit color.\n uint32_t ca = 0x101 * (0xFF & (c >> 24));\n uint32_t cr = 0x101 * (0xFF & (c >> 16));\n uint32_t cg = 0x101 * (0xFF & (c >> 8));\n uint32_t cb = 0x101 * (0xFF & (c >> 0));\n\n switch (palette_format.repr) {\n case WUFFS_BASE__PIXEL_FORMAT__INDEXED__BGRA_NONPREMUL:\n case WUFFS_BASE__PIXEL_FORMAT__INDEXED__BGRA_PREMUL:\n case WUFFS_BASE__PIXEL_FORMAT__INDEXED__BGRA_BINARY: {\n bool nonpremul = palette_format.repr ==\n WUFFS_BASE__PIXEL_FORMAT__INDEXED__BGRA_NONPREMUL;\n\n size_t i;\n for (i = 0; i < n; i++) {\n // Work in 16-bit color.\n uint32_t pb = 0x101 * ((uint32_t)(palette_slice.ptr[(4 * i) + 0]))" + + ";\n uint32_t pg = 0x101 * ((uint32_t)(palette_slice.ptr[(4 * i) + 1]));\n uint32_t pr = 0x101 * ((uint32_t)(palette_slice.ptr[(4 * i) + 2]));\n uint32_t pa = 0x101 * ((uint32_t)(palette_slice.ptr[(4 * i) + 3]));\n\n // Convert to premultiplied alpha.\n if (nonpremul && (pa != 0xFFFF)) {\n pb = (pb * pa) / 0xFFFF;\n pg = (pg * pa) / 0xFFFF;\n pr = (pr * pa) / 0xFFFF;\n }\n\n // These deltas are conceptually int32_t (signed) but after squaring,\n // it's equivalent to work in uint32_t (unsigned).\n pb -= cb;\n pg -= cg;\n pr -= cr;\n pa -= ca;\n uint64_t score = ((uint64_t)(pb * pb)) + ((uint64_t)(pg * pg)) +\n ((uint64_t)(pr * pr)) + ((uint64_t)(pa * pa));\n if (best_score > score) {\n best_score = score;\n best_index = i;\n }\n }\n break;\n }\n }\n\n return best_index;\n}\n\n" + "" + "// --------\n\nstatic uint64_t //\nwuffs_base__pixel_swizzler__bgra_premul__bgra_nonpremul__src(\n wuffs_base__slice_u8 dst,\n wuffs_base__slice_u8 dst_palette,\n wuffs_base__slice_u8 src) {\n size_t dst_len4 = dst.len / 4;\n size_t src_len4 = src.len / 4;\n size_t len = dst_len4 < src_len4 ? dst_len4 : src_len4;\n uint8_t* d = dst.ptr;\n uint8_t* s = src.ptr;\n size_t n = len;\n\n // TODO: unroll.\n\n while (n >= 1) {\n uint32_t s0 = wuffs_base__premul_u32_axxx(\n wuffs_base__load_u32le__no_bounds_check(s + (0 * 4)));\n wuffs_base__store_u32le__no_bounds_check(d + (0 * 4), s0);\n\n s += 1 * 4;\n d += 1 * 4;\n n -= 1;\n }\n\n return len;\n}\n\nstatic uint64_t //\nwuffs_base__pixel_swizzler__bgra_premul__bgra_nonpremul__src_over(\n wuffs_base__slice_u8 dst,\n wuffs_base__slice_u8 dst_palette,\n wuffs_base__slice_u8 src) {\n size_t dst_len4 = dst.len / 4;\n size_t src_len4 = src.len / 4;\n size_t len = dst_len4 < src_len4 ? dst_len4 : src_len4;\n uint8_t* d = dst.ptr;\n uint8_t* s = src.ptr;\n" + " size_t n = len;\n\n // TODO: unroll.\n\n while (n >= 1) {\n uint32_t d0 = wuffs_base__load_u32le__no_bounds_check(d + (0 * 4));\n uint32_t s0 = wuffs_base__load_u32le__no_bounds_check(s + (0 * 4));\n wuffs_base__store_u32le__no_bounds_check(\n d + (0 * 4), wuffs_base__composite_premul_nonpremul_u32_axxx(d0, s0));\n\n s += 1 * 4;\n d += 1 * 4;\n n -= 1;\n }\n\n return len;\n}\n\nstatic uint64_t //\nwuffs_base__pixel_swizzler__copy_1_1(wuffs_base__slice_u8 dst,\n wuffs_base__slice_u8 dst_palette,\n wuffs_base__slice_u8 src) {\n return wuffs_base__slice_u8__copy_from_slice(dst, src);\n}\n\nstatic uint64_t //\nwuffs_base__pixel_swizzler__copy_4_4(wuffs_base__slice_u8 dst,\n wuffs_base__slice_u8 dst_palette,\n wuffs_base__slice_u8 src) {\n size_t dst_len4 = dst.len / 4;\n size_t src_len4 = src.len / 4;\n size_t len = dst_len4 < src_len4 ? dst_len4 : src_len4;\n if (l" + @@ -227,8 +230,9 @@ const baseImagePrivateH = "" + "" const baseImagePublicH = "" + - "// ---------------- Images\n\n// wuffs_base__color_u32_argb_premul is an 8 bit per channel premultiplied\n// Alpha, Red, Green, Blue color, as a uint32_t value. Its value is always\n// 0xAARRGGBB (Alpha most significant, Blue least), regardless of endianness.\ntypedef uint32_t wuffs_base__color_u32_argb_premul;\n\n// wuffs_base__premul_u32_axxx converts from non-premultiplied alpha to\n// premultiplied alpha. The \"axxx\" means either \"argb\" or \"abgr\".\nstatic inline uint32_t //\nwuffs_base__premul_u32_axxx(uint32_t nonpremul) {\n // Multiplying by 0x101 (twice, once for alpha and once for color) converts\n // from 8-bit to 16-bit color. Shifting right by 8 undoes that.\n //\n // Working in the higher bit depth can produce slightly different (and\n // arguably slightly more accurate) results. For example, given 8-bit blue\n // and alpha of 0x80 and 0x81:\n //\n // - ((0x80 * 0x81 ) / 0xFF ) = 0x40 = 0x40\n // - ((0x8080 * 0x8181) / 0xFFFF) >> 8 = 0x4101 >> 8 = 0x41\n uint32_t a = 0xFF & (nonpremul >> 2" + - "4);\n uint32_t a16 = a * (0x101 * 0x101);\n\n uint32_t r = 0xFF & (nonpremul >> 16);\n r = ((r * a16) / 0xFFFF) >> 8;\n uint32_t g = 0xFF & (nonpremul >> 8);\n g = ((g * a16) / 0xFFFF) >> 8;\n uint32_t b = 0xFF & (nonpremul >> 0);\n b = ((b * a16) / 0xFFFF) >> 8;\n\n return (a << 24) | (r << 16) | (g << 8) | (b << 0);\n}\n\n// wuffs_base__nonpremul_u32_axxx converts from premultiplied alpha to\n// non-premultiplied alpha. The \"axxx\" means either \"argb\" or \"abgr\".\nstatic inline uint32_t //\nwuffs_base__nonpremul_u32_axxx(uint32_t premul) {\n uint32_t a = 0xFF & (premul >> 24);\n if (a == 0xFF) {\n return premul;\n } else if (a == 0) {\n return 0;\n }\n uint32_t a16 = a * 0x101;\n\n uint32_t r = 0xFF & (premul >> 16);\n r = ((r * (0x101 * 0xFFFF)) / a16) >> 8;\n uint32_t g = 0xFF & (premul >> 8);\n g = ((g * (0x101 * 0xFFFF)) / a16) >> 8;\n uint32_t b = 0xFF & (premul >> 0);\n b = ((b * (0x101 * 0xFFFF)) / a16) >> 8;\n\n return (a << 24) | (r << 16) | (g << 8) | (b << 0);\n}\n\n" + + "// ---------------- Images\n\n// wuffs_base__color_u32_argb_premul is an 8 bit per channel premultiplied\n// Alpha, Red, Green, Blue color, as a uint32_t value. Its value is always\n// 0xAARRGGBB (Alpha most significant, Blue least), regardless of endianness.\ntypedef uint32_t wuffs_base__color_u32_argb_premul;\n\nstatic inline uint8_t //\nwuffs_base__color_u32_argb_premul__as_gray(\n wuffs_base__color_u32_argb_premul c) {\n // Work in 16-bit color.\n uint32_t cr = 0x101 * (0xFF & (c >> 16));\n uint32_t cg = 0x101 * (0xFF & (c >> 8));\n uint32_t cb = 0x101 * (0xFF & (c >> 0));\n\n // These coefficients (the fractions 0.299, 0.587 and 0.114) are the same\n // as those given by the JFIF specification.\n //\n // Note that 19595 + 38470 + 7471 equals 65536, also known as (1 << 16). We\n // shift by 24, not just by 16, because the return value is 8-bit color, not\n // 16-bit color.\n return ((19595 * cr) + (38470 * cg) + (7471 * cb) + 32768) >> 24;\n}\n\n// wuffs_base__premul_u32_axxx converts from non-premultiplied alpha " + + "to\n// premultiplied alpha. The \"axxx\" means either \"argb\" or \"abgr\".\nstatic inline uint32_t //\nwuffs_base__premul_u32_axxx(uint32_t nonpremul) {\n // Multiplying by 0x101 (twice, once for alpha and once for color) converts\n // from 8-bit to 16-bit color. Shifting right by 8 undoes that.\n //\n // Working in the higher bit depth can produce slightly different (and\n // arguably slightly more accurate) results. For example, given 8-bit blue\n // and alpha of 0x80 and 0x81:\n //\n // - ((0x80 * 0x81 ) / 0xFF ) = 0x40 = 0x40\n // - ((0x8080 * 0x8181) / 0xFFFF) >> 8 = 0x4101 >> 8 = 0x41\n uint32_t a = 0xFF & (nonpremul >> 24);\n uint32_t a16 = a * (0x101 * 0x101);\n\n uint32_t r = 0xFF & (nonpremul >> 16);\n r = ((r * a16) / 0xFFFF) >> 8;\n uint32_t g = 0xFF & (nonpremul >> 8);\n g = ((g * a16) / 0xFFFF) >> 8;\n uint32_t b = 0xFF & (nonpremul >> 0);\n b = ((b * a16) / 0xFFFF) >> 8;\n\n return (a << 24) | (r << 16) | (g << 8) | (b << 0);\n}\n\n// wuffs_base__nonpremul_u32_axxx converts from premultip" + + "lied alpha to\n// non-premultiplied alpha. The \"axxx\" means either \"argb\" or \"abgr\".\nstatic inline uint32_t //\nwuffs_base__nonpremul_u32_axxx(uint32_t premul) {\n uint32_t a = 0xFF & (premul >> 24);\n if (a == 0xFF) {\n return premul;\n } else if (a == 0) {\n return 0;\n }\n uint32_t a16 = a * 0x101;\n\n uint32_t r = 0xFF & (premul >> 16);\n r = ((r * (0x101 * 0xFFFF)) / a16) >> 8;\n uint32_t g = 0xFF & (premul >> 8);\n g = ((g * (0x101 * 0xFFFF)) / a16) >> 8;\n uint32_t b = 0xFF & (premul >> 0);\n b = ((b * (0x101 * 0xFFFF)) / a16) >> 8;\n\n return (a << 24) | (r << 16) | (g << 8) | (b << 0);\n}\n\n" + "" + "// --------\n\ntypedef uint8_t wuffs_base__pixel_blend;\n\n// wuffs_base__pixel_blend encodes how to blend source and destination pixels,\n// accounting for transparency. It encompasses the Porter-Duff compositing\n// operators as well as the other blending modes defined by PDF.\n//\n// TODO: implement the other modes.\n#define WUFFS_BASE__PIXEL_BLEND__SRC ((wuffs_base__pixel_blend)0)\n#define WUFFS_BASE__PIXEL_BLEND__SRC_OVER ((wuffs_base__pixel_blend)1)\n\n" + "" + @@ -282,6 +286,8 @@ const baseImagePublicH = "" + "" + "// --------\n\ntypedef struct {\n // Do not access the private_impl's fields directly. There is no API/ABI\n // compatibility or safety guarantee if you do so.\n struct {\n uint8_t TODO;\n } private_impl;\n\n#ifdef __cplusplus\n#endif // __cplusplus\n\n} wuffs_base__decode_frame_options;\n\n#ifdef __cplusplus\n\n#endif // __cplusplus\n\n" + "" + + "// --------\n\n// wuffs_base__pixel_palette__closest_element returns the index of the palette\n// element that minimizes the sum of squared differences of the four ARGB\n// channels, working in premultiplied alpha. Ties favor the smaller index.\n//\n// The palette_slice.len may equal (N*4), for N less than 256, which means that\n// only the first N palette elements are considered. It returns 0 when N is 0.\n//\n// Applying this function on a per-pixel basis will not produce whole-of-image\n// dithering.\nuint8_t //\nwuffs_base__pixel_palette__closest_element(\n wuffs_base__slice_u8 palette_slice,\n wuffs_base__pixel_format palette_format,\n wuffs_base__color_u32_argb_premul c);\n\n" + + "" + "// --------\n\n// TODO: should the func type take restrict pointers?\ntypedef uint64_t (*wuffs_base__pixel_swizzler__func)(\n wuffs_base__slice_u8 dst,\n wuffs_base__slice_u8 dst_palette,\n wuffs_base__slice_u8 src);\n\ntypedef struct {\n // Do not access the private_impl's fields directly. There is no API/ABI\n // compatibility or safety guarantee if you do so.\n struct {\n wuffs_base__pixel_swizzler__func func;\n } private_impl;\n\n#ifdef __cplusplus\n inline wuffs_base__status prepare(wuffs_base__pixel_format dst_format,\n wuffs_base__slice_u8 dst_palette,\n wuffs_base__pixel_format src_format,\n wuffs_base__slice_u8 src_palette,\n wuffs_base__pixel_blend blend);\n inline uint64_t swizzle_interleaved(wuffs_base__slice_u8 dst,\n wuffs_base__slice_u8 dst_palette,\n wuffs_base__slice_u8 src) const;\n#endi" + "f // __cplusplus\n\n} wuffs_base__pixel_swizzler;\n\nwuffs_base__status //\nwuffs_base__pixel_swizzler__prepare(wuffs_base__pixel_swizzler* p,\n wuffs_base__pixel_format dst_format,\n wuffs_base__slice_u8 dst_palette,\n wuffs_base__pixel_format src_format,\n wuffs_base__slice_u8 src_palette,\n wuffs_base__pixel_blend blend);\n\nuint64_t //\nwuffs_base__pixel_swizzler__swizzle_interleaved(\n const wuffs_base__pixel_swizzler* p,\n wuffs_base__slice_u8 dst,\n wuffs_base__slice_u8 dst_palette,\n wuffs_base__slice_u8 src);\n\n#ifdef __cplusplus\n\ninline wuffs_base__status //\nwuffs_base__pixel_swizzler::prepare(wuffs_base__pixel_format dst_format,\n wuffs_base__slice_u8 dst_palette,\n wuffs_base__pixel_format src_format,\n wuffs_base__slice_u8 src" + "_palette,\n wuffs_base__pixel_blend blend) {\n return wuffs_base__pixel_swizzler__prepare(this, dst_format, dst_palette,\n src_format, src_palette, blend);\n}\n\nuint64_t //\nwuffs_base__pixel_swizzler::swizzle_interleaved(\n wuffs_base__slice_u8 dst,\n wuffs_base__slice_u8 dst_palette,\n wuffs_base__slice_u8 src) const {\n return wuffs_base__pixel_swizzler__swizzle_interleaved(this, dst, dst_palette,\n src);\n}\n\n#endif // __cplusplus\n" + diff --git a/release/c/wuffs-unsupported-snapshot.c b/release/c/wuffs-unsupported-snapshot.c index d86140ef..ff2e2feb 100644 --- a/release/c/wuffs-unsupported-snapshot.c +++ b/release/c/wuffs-unsupported-snapshot.c @@ -2507,6 +2507,23 @@ wuffs_base__malloc_slice_u64(void* (*malloc_func)(size_t), uint64_t num_u64) { // 0xAARRGGBB (Alpha most significant, Blue least), regardless of endianness. typedef uint32_t wuffs_base__color_u32_argb_premul; +static inline uint8_t // +wuffs_base__color_u32_argb_premul__as_gray( + wuffs_base__color_u32_argb_premul c) { + // Work in 16-bit color. + uint32_t cr = 0x101 * (0xFF & (c >> 16)); + uint32_t cg = 0x101 * (0xFF & (c >> 8)); + uint32_t cb = 0x101 * (0xFF & (c >> 0)); + + // These coefficients (the fractions 0.299, 0.587 and 0.114) are the same + // as those given by the JFIF specification. + // + // Note that 19595 + 38470 + 7471 equals 65536, also known as (1 << 16). We + // shift by 24, not just by 16, because the return value is 8-bit color, not + // 16-bit color. + return ((19595 * cr) + (38470 * cg) + (7471 * cb) + 32768) >> 24; +} + // wuffs_base__premul_u32_axxx converts from non-premultiplied alpha to // premultiplied alpha. The "axxx" means either "argb" or "abgr". static inline uint32_t // @@ -3715,6 +3732,23 @@ typedef struct { // -------- +// wuffs_base__pixel_palette__closest_element returns the index of the palette +// element that minimizes the sum of squared differences of the four ARGB +// channels, working in premultiplied alpha. Ties favor the smaller index. +// +// The palette_slice.len may equal (N*4), for N less than 256, which means that +// only the first N palette elements are considered. It returns 0 when N is 0. +// +// Applying this function on a per-pixel basis will not produce whole-of-image +// dithering. +uint8_t // +wuffs_base__pixel_palette__closest_element( + wuffs_base__slice_u8 palette_slice, + wuffs_base__pixel_format palette_format, + wuffs_base__color_u32_argb_premul c); + +// -------- + // TODO: should the func type take restrict pointers? typedef uint64_t (*wuffs_base__pixel_swizzler__func)( wuffs_base__slice_u8 dst, @@ -8626,6 +8660,18 @@ wuffs_base__pixel_buffer__set_color_u32_at( // Common formats above. Rarer formats below. + case WUFFS_BASE__PIXEL_FORMAT__Y: + wuffs_base__store_u8__no_bounds_check( + row + ((size_t)x), wuffs_base__color_u32_argb_premul__as_gray(color)); + break; + + case WUFFS_BASE__PIXEL_FORMAT__INDEXED__BGRA_BINARY: + wuffs_base__store_u8__no_bounds_check( + row + ((size_t)x), wuffs_base__pixel_palette__closest_element( + wuffs_base__pixel_buffer__palette(pb), + pb->pixcfg.private_impl.pixfmt, color)); + break; + case WUFFS_BASE__PIXEL_FORMAT__BGR_565: { uint32_t b5 = 0x1F & (color >> (8 - 5)); uint32_t g6 = 0x3F & (color >> (16 - 6)); @@ -8669,6 +8715,68 @@ wuffs_base__pixel_buffer__set_color_u32_at( // -------- +uint8_t // +wuffs_base__pixel_palette__closest_element( + wuffs_base__slice_u8 palette_slice, + wuffs_base__pixel_format palette_format, + wuffs_base__color_u32_argb_premul c) { + size_t n = palette_slice.len / 4; + if (n > 256) { + n = 256; + } + size_t best_index = 0; + uint64_t best_score = 0xFFFFFFFFFFFFFFFF; + + // Work in 16-bit color. + uint32_t ca = 0x101 * (0xFF & (c >> 24)); + uint32_t cr = 0x101 * (0xFF & (c >> 16)); + uint32_t cg = 0x101 * (0xFF & (c >> 8)); + uint32_t cb = 0x101 * (0xFF & (c >> 0)); + + switch (palette_format.repr) { + case WUFFS_BASE__PIXEL_FORMAT__INDEXED__BGRA_NONPREMUL: + case WUFFS_BASE__PIXEL_FORMAT__INDEXED__BGRA_PREMUL: + case WUFFS_BASE__PIXEL_FORMAT__INDEXED__BGRA_BINARY: { + bool nonpremul = palette_format.repr == + WUFFS_BASE__PIXEL_FORMAT__INDEXED__BGRA_NONPREMUL; + + size_t i; + for (i = 0; i < n; i++) { + // Work in 16-bit color. + uint32_t pb = 0x101 * ((uint32_t)(palette_slice.ptr[(4 * i) + 0])); + uint32_t pg = 0x101 * ((uint32_t)(palette_slice.ptr[(4 * i) + 1])); + uint32_t pr = 0x101 * ((uint32_t)(palette_slice.ptr[(4 * i) + 2])); + uint32_t pa = 0x101 * ((uint32_t)(palette_slice.ptr[(4 * i) + 3])); + + // Convert to premultiplied alpha. + if (nonpremul && (pa != 0xFFFF)) { + pb = (pb * pa) / 0xFFFF; + pg = (pg * pa) / 0xFFFF; + pr = (pr * pa) / 0xFFFF; + } + + // These deltas are conceptually int32_t (signed) but after squaring, + // it's equivalent to work in uint32_t (unsigned). + pb -= cb; + pg -= cg; + pr -= cr; + pa -= ca; + uint64_t score = ((uint64_t)(pb * pb)) + ((uint64_t)(pg * pg)) + + ((uint64_t)(pr * pr)) + ((uint64_t)(pa * pa)); + if (best_score > score) { + best_score = score; + best_index = i; + } + } + break; + } + } + + return best_index; +} + +// -------- + static uint64_t // wuffs_base__pixel_swizzler__bgra_premul__bgra_nonpremul__src( wuffs_base__slice_u8 dst, diff --git a/test/c/std/wbmp.c b/test/c/std/wbmp.c index f5d9f7c9..11c51629 100644 --- a/test/c/std/wbmp.c +++ b/test/c/std/wbmp.c @@ -84,14 +84,6 @@ fill_palette_with_grays(wuffs_base__pixel_buffer* pb) { } } -void // -fill_table(wuffs_base__table_u8 t, uint8_t c) { - size_t y; - for (y = 0; y < t.height; y++) { - memset(t.ptr + (y * t.stride), c, t.width); - } -} - const char* // test_wuffs_pixel_swizzler_swizzle() { CHECK_FOCUS(__func__); @@ -150,19 +142,10 @@ test_wuffs_pixel_swizzler_swizzle() { &src_pixbuf, &src_pixcfg, g_src_slice_u8)); fill_palette_with_grays(&src_pixbuf); - // Set the middle src pixel. - if (srcs[s].pixfmt_repr == WUFFS_BASE__PIXEL_FORMAT__Y) { - fill_table(wuffs_base__pixel_buffer__plane(&src_pixbuf, 0), 0x44); - } else if (srcs[s].pixfmt_repr == - WUFFS_BASE__PIXEL_FORMAT__INDEXED__BGRA_BINARY) { - fill_table(wuffs_base__pixel_buffer__plane(&src_pixbuf, 0), 0x44); - } else { - CHECK_STATUS("set_color_u32_at", - wuffs_base__pixel_buffer__set_color_u32_at( - &src_pixbuf, width / 2, height / 2, srcs[s].pixel)); - } - - // Check the middle src pixel. + // Set and check the middle src pixel. + CHECK_STATUS("set_color_u32_at", + wuffs_base__pixel_buffer__set_color_u32_at( + &src_pixbuf, width / 2, height / 2, srcs[s].pixel)); wuffs_base__color_u32_argb_premul have_src_pixel = wuffs_base__pixel_buffer__color_u32_at(&src_pixbuf, width / 2, height / 2);