std/jpeg: decode CMYK and YCbCrK JPEGs

Once again, the benchmarks don't exercise the CMYK / YCbCrK code paths
per se, but here are the numbers:

name                                             old speed      new speed      delta

wuffs_jpeg_decode_19k_8bpp/clang11                376MB/s ± 0%   376MB/s ± 0%    ~     (p=0.310 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/clang11   341MB/s ± 0%   340MB/s ± 0%  -0.42%  (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/clang11   1.01GB/s ± 0%  1.01GB/s ± 0%  +0.27%  (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/clang11               836MB/s ± 1%   839MB/s ± 0%  +0.37%  (p=0.016 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/clang11          912MB/s ± 0%   914MB/s ± 0%  +0.28%  (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/clang11         1.10GB/s ± 0%  1.11GB/s ± 0%  +0.82%  (p=0.008 n=5+5)
wuffs_jpeg_decode_4002k_24bpp/clang11             916MB/s ± 0%   916MB/s ± 0%    ~     (p=0.841 n=5+5)

wuffs_jpeg_decode_19k_8bpp/gcc10                  368MB/s ± 0%   361MB/s ± 0%  -1.87%  (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_progressive/gcc10     279MB/s ± 0%   292MB/s ± 0%  +4.68%  (p=0.008 n=5+5)
wuffs_jpeg_decode_30k_24bpp_sequential/gcc10      947MB/s ± 1%   930MB/s ± 0%  -1.74%  (p=0.008 n=5+5)
wuffs_jpeg_decode_77k_24bpp/gcc10                 770MB/s ± 0%   760MB/s ± 0%  -1.32%  (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_420/gcc10            838MB/s ± 0%   826MB/s ± 0%  -1.36%  (p=0.008 n=5+5)
wuffs_jpeg_decode_552k_24bpp_444/gcc10           1.00GB/s ± 0%  1.01GB/s ± 0%  +1.04%  (p=0.008 n=5+5)
wuffs_jpeg_decode_4002k_24bpp/gcc10               847MB/s ± 0%   833MB/s ± 0%  -1.67%  (p=0.008 n=5+5)
This commit is contained in:
Nigel Tao committed 2023-08-21 22:13:39 +10:00
1 parent f569699221
commit 6489915ed3
4 files changed
+741 -45

No files matched your search

+367 -19
View File
@@ -77,6 +77,71 @@ wuffs_base__u32__min_of_5(uint32_t a,
// --------
typedef void (*wuffs_base__pixel_swizzler__swizzle_ycc__convert_4_func)(
wuffs_base__pixel_buffer* dst,
uint32_t x,
uint32_t x_end,
uint32_t y,
const uint8_t* up0,
const uint8_t* up1,
const uint8_t* up2,
const uint8_t* up3);
static void //
wuffs_base__pixel_swizzler__swizzle_cmyk__convert_4_general(
wuffs_base__pixel_buffer* dst,
uint32_t x,
uint32_t x_end,
uint32_t y,
const uint8_t* up0,
const uint8_t* up1,
const uint8_t* up2,
const uint8_t* up3) {
for (; x < x_end; x++) {
// It's called CMYK but, but for Adobe CMYK JPEG images in practice, it's
// RGBW: 0xFFu means no ink instead of full ink. Note that a double
// inversion is a no-op, so inversions might be implicit in the code below.
uint32_t r = ((uint32_t)(*up0++));
uint32_t g = ((uint32_t)(*up1++));
uint32_t b = ((uint32_t)(*up2++));
uint32_t w = ((uint32_t)(*up3++));
r = ((r * w) + 0x7Fu) / 0xFFu;
g = ((g * w) + 0x7Fu) / 0xFFu;
b = ((b * w) + 0x7Fu) / 0xFFu;
wuffs_base__pixel_buffer__set_color_u32_at(
dst, x, y, 0xFF000000u | (r << 16u) | (g << 8u) | (b << 0u));
}
}
static void //
wuffs_base__pixel_swizzler__swizzle_ycck__convert_4_general(
wuffs_base__pixel_buffer* dst,
uint32_t x,
uint32_t x_end,
uint32_t y,
const uint8_t* up0,
const uint8_t* up1,
const uint8_t* up2,
const uint8_t* up3) {
for (; x < x_end; x++) {
// We invert once again: 0xFFu means no ink instead of full ink.
uint32_t color = //
wuffs_base__color_ycc__as__color_u32( //
*up0++, *up1++, *up2++);
uint32_t r = 0xFFu - (0xFFu & (color >> 16u));
uint32_t g = 0xFFu - (0xFFu & (color >> 8u));
uint32_t b = 0xFFu - (0xFFu & (color >> 0u));
uint32_t w = ((uint32_t)(*up3++));
r = ((r * w) + 0x7Fu) / 0xFFu;
g = ((g * w) + 0x7Fu) / 0xFFu;
b = ((b * w) + 0x7Fu) / 0xFFu;
wuffs_base__pixel_buffer__set_color_u32_at(
dst, x, y, 0xFF000000u | (r << 16u) | (g << 8u) | (b << 0u));
}
}
// --------
typedef void (*wuffs_base__pixel_swizzler__swizzle_ycc__convert_3_func)(
wuffs_base__pixel_buffer* dst,
uint32_t x,
@@ -96,10 +161,10 @@ wuffs_base__pixel_swizzler__swizzle_rgb__convert_3_general(
const uint8_t* up1,
const uint8_t* up2) {
for (; x < x_end; x++) {
uint32_t color = 0xFF000000 | //
(((uint32_t)(*up0++)) << 16) | //
(((uint32_t)(*up1++)) << 8) | //
(((uint32_t)(*up2++)) << 0);
uint32_t color = 0xFF000000u | //
(((uint32_t)(*up0++)) << 16u) | //
(((uint32_t)(*up1++)) << 8u) | //
(((uint32_t)(*up2++)) << 0u);
wuffs_base__pixel_buffer__set_color_u32_at(dst, x, y, color);
}
}
@@ -132,14 +197,14 @@ wuffs_base__pixel_swizzler__swizzle_ycc__convert_3_bgrx(
const uint8_t* up2) {
size_t dst_stride = dst->private_impl.planes[0].stride;
uint8_t* dst_iter = dst->private_impl.planes[0].ptr +
(dst_stride * ((size_t)y)) + (4 * ((size_t)x));
(dst_stride * ((size_t)y)) + (4u * ((size_t)x));
for (; x < x_end; x++) {
uint32_t color = //
wuffs_base__color_ycc__as__color_u32( //
*up0++, *up1++, *up2++);
wuffs_base__poke_u32le__no_bounds_check(dst_iter, color);
dst_iter += 4;
dst_iter += 4u;
}
}
@@ -154,14 +219,14 @@ wuffs_base__pixel_swizzler__swizzle_ycc__convert_3_rgbx(
const uint8_t* up2) {
size_t dst_stride = dst->private_impl.planes[0].stride;
uint8_t* dst_iter = dst->private_impl.planes[0].ptr +
(dst_stride * ((size_t)y)) + (4 * ((size_t)x));
(dst_stride * ((size_t)y)) + (4u * ((size_t)x));
for (; x < x_end; x++) {
uint32_t color = //
wuffs_base__color_ycc__as__color_u32_abgr( //
*up0++, *up1++, *up2++);
wuffs_base__poke_u32le__no_bounds_check(dst_iter, color);
dst_iter += 4;
dst_iter += 4u;
}
}
@@ -461,6 +526,260 @@ wuffs_base__pixel_swizzler__has_triangle_upsampler(uint32_t inv_h,
// wuffs_base__pixel_swizzler__swizzle_ycck before calling these functions. For
// example, (width > 0) is a precondition, but there are many more.
static void //
wuffs_base__pixel_swizzler__swizzle_ycck__general__triangle_filter_edge_row(
wuffs_base__pixel_buffer* dst,
uint32_t width,
uint32_t y,
const uint8_t* src_ptr0,
const uint8_t* src_ptr1,
const uint8_t* src_ptr2,
const uint8_t* src_ptr3,
uint32_t stride0,
uint32_t stride1,
uint32_t stride2,
uint32_t stride3,
uint32_t inv_h0,
uint32_t inv_h1,
uint32_t inv_h2,
uint32_t inv_h3,
uint32_t inv_v0,
uint32_t inv_v1,
uint32_t inv_v2,
uint32_t inv_v3,
uint32_t half_width_for_2to1,
uint32_t h1v2_bias,
uint8_t* scratch_buffer_2k_ptr,
wuffs_base__pixel_swizzler__swizzle_ycc__upsample_func upfunc0,
wuffs_base__pixel_swizzler__swizzle_ycc__upsample_func upfunc1,
wuffs_base__pixel_swizzler__swizzle_ycc__upsample_func upfunc2,
wuffs_base__pixel_swizzler__swizzle_ycc__upsample_func upfunc3,
wuffs_base__pixel_swizzler__swizzle_ycc__convert_4_func conv4func) {
const uint8_t* src0 = src_ptr0 + ((y / inv_v0) * (size_t)stride0);
const uint8_t* src1 = src_ptr1 + ((y / inv_v1) * (size_t)stride1);
const uint8_t* src2 = src_ptr2 + ((y / inv_v2) * (size_t)stride2);
const uint8_t* src3 = src_ptr3 + ((y / inv_v3) * (size_t)stride3);
uint32_t total_src_len0 = 0u;
uint32_t total_src_len1 = 0u;
uint32_t total_src_len2 = 0u;
uint32_t total_src_len3 = 0u;
uint32_t x = 0u;
while (x < width) {
bool first_column = x == 0u;
uint32_t end = x + 480u;
if (end > width) {
end = width;
}
uint32_t src_len0 = ((end - x) + inv_h0 - 1u) / inv_h0;
uint32_t src_len1 = ((end - x) + inv_h1 - 1u) / inv_h1;
uint32_t src_len2 = ((end - x) + inv_h2 - 1u) / inv_h2;
uint32_t src_len3 = ((end - x) + inv_h3 - 1u) / inv_h3;
total_src_len0 += src_len0;
total_src_len1 += src_len1;
total_src_len2 += src_len2;
total_src_len3 += src_len3;
const uint8_t* src_ptr_x0 = src0 + (x / inv_h0);
const uint8_t* up0 = (*upfunc0)( //
scratch_buffer_2k_ptr + (0u * 480u), //
src_ptr_x0, //
src_ptr_x0, //
src_len0, //
h1v2_bias, //
first_column, //
(total_src_len0 >= half_width_for_2to1));
const uint8_t* src_ptr_x1 = src1 + (x / inv_h1);
const uint8_t* up1 = (*upfunc1)( //
scratch_buffer_2k_ptr + (1u * 480u), //
src_ptr_x1, //
src_ptr_x1, //
src_len1, //
h1v2_bias, //
first_column, //
(total_src_len1 >= half_width_for_2to1));
const uint8_t* src_ptr_x2 = src2 + (x / inv_h2);
const uint8_t* up2 = (*upfunc2)( //
scratch_buffer_2k_ptr + (2u * 480u), //
src_ptr_x2, //
src_ptr_x2, //
src_len2, //
h1v2_bias, //
first_column, //
(total_src_len2 >= half_width_for_2to1));
const uint8_t* src_ptr_x3 = src3 + (x / inv_h3);
const uint8_t* up3 = (*upfunc3)( //
scratch_buffer_2k_ptr + (3u * 480u), //
src_ptr_x3, //
src_ptr_x3, //
src_len3, //
h1v2_bias, //
first_column, //
(total_src_len3 >= half_width_for_2to1));
(*conv4func)(dst, x, end, y, up0, up1, up2, up3);
x = end;
}
}
static void //
wuffs_base__pixel_swizzler__swizzle_ycck__general__triangle_filter(
wuffs_base__pixel_buffer* dst,
uint32_t width,
uint32_t height,
const uint8_t* src_ptr0,
const uint8_t* src_ptr1,
const uint8_t* src_ptr2,
const uint8_t* src_ptr3,
uint32_t stride0,
uint32_t stride1,
uint32_t stride2,
uint32_t stride3,
uint32_t inv_h0,
uint32_t inv_h1,
uint32_t inv_h2,
uint32_t inv_h3,
uint32_t inv_v0,
uint32_t inv_v1,
uint32_t inv_v2,
uint32_t inv_v3,
uint32_t half_width_for_2to1,
uint32_t half_height_for_2to1,
uint8_t* scratch_buffer_2k_ptr,
wuffs_base__pixel_swizzler__swizzle_ycc__upsample_func (*upfuncs)[4][4],
wuffs_base__pixel_swizzler__swizzle_ycc__convert_4_func conv4func) {
wuffs_base__pixel_swizzler__swizzle_ycc__upsample_func upfunc0 =
(*upfuncs)[(inv_h0 - 1u) & 3u][(inv_v0 - 1u) & 3u];
wuffs_base__pixel_swizzler__swizzle_ycc__upsample_func upfunc1 =
(*upfuncs)[(inv_h1 - 1u) & 3u][(inv_v1 - 1u) & 3u];
wuffs_base__pixel_swizzler__swizzle_ycc__upsample_func upfunc2 =
(*upfuncs)[(inv_h2 - 1u) & 3u][(inv_v2 - 1u) & 3u];
wuffs_base__pixel_swizzler__swizzle_ycc__upsample_func upfunc3 =
(*upfuncs)[(inv_h3 - 1u) & 3u][(inv_v3 - 1u) & 3u];
// First row.
uint32_t h1v2_bias = 1u;
wuffs_base__pixel_swizzler__swizzle_ycck__general__triangle_filter_edge_row(
dst, width, 0u, //
src_ptr0, src_ptr1, src_ptr2, src_ptr3, //
stride0, stride1, stride2, stride3, //
inv_h0, inv_h1, inv_h2, inv_h3, //
inv_v0, inv_v1, inv_v2, inv_v3, //
half_width_for_2to1, //
h1v2_bias, //
scratch_buffer_2k_ptr, //
upfunc0, upfunc1, upfunc2, upfunc3, conv4func);
h1v2_bias = 2u;
// Middle rows.
bool last_row = height == 2u * half_height_for_2to1;
uint32_t y_max_excl = last_row ? (height - 1u) : height;
uint32_t y;
for (y = 1u; y < y_max_excl; y++) {
const uint8_t* src0_major = src_ptr0 + ((y / inv_v0) * (size_t)stride0);
const uint8_t* src0_minor =
(inv_v0 != 2u)
? src0_major
: ((y & 1u) ? (src0_major + stride0) : (src0_major - stride0));
const uint8_t* src1_major = src_ptr1 + ((y / inv_v1) * (size_t)stride1);
const uint8_t* src1_minor =
(inv_v1 != 2u)
? src1_major
: ((y & 1u) ? (src1_major + stride1) : (src1_major - stride1));
const uint8_t* src2_major = src_ptr2 + ((y / inv_v2) * (size_t)stride2);
const uint8_t* src2_minor =
(inv_v2 != 2u)
? src2_major
: ((y & 1u) ? (src2_major + stride2) : (src2_major - stride2));
const uint8_t* src3_major = src_ptr3 + ((y / inv_v3) * (size_t)stride3);
const uint8_t* src3_minor =
(inv_v3 != 2u)
? src3_major
: ((y & 1u) ? (src3_major + stride3) : (src3_major - stride3));
uint32_t total_src_len0 = 0u;
uint32_t total_src_len1 = 0u;
uint32_t total_src_len2 = 0u;
uint32_t total_src_len3 = 0u;
uint32_t x = 0u;
while (x < width) {
bool first_column = x == 0u;
uint32_t end = x + 480u;
if (end > width) {
end = width;
}
uint32_t src_len0 = ((end - x) + inv_h0 - 1u) / inv_h0;
uint32_t src_len1 = ((end - x) + inv_h1 - 1u) / inv_h1;
uint32_t src_len2 = ((end - x) + inv_h2 - 1u) / inv_h2;
uint32_t src_len3 = ((end - x) + inv_h3 - 1u) / inv_h3;
total_src_len0 += src_len0;
total_src_len1 += src_len1;
total_src_len2 += src_len2;
total_src_len3 += src_len3;
const uint8_t* up0 = (*upfunc0)( //
scratch_buffer_2k_ptr + (0u * 480u), //
src0_major + (x / inv_h0), //
src0_minor + (x / inv_h0), //
src_len0, //
h1v2_bias, //
first_column, //
(total_src_len0 >= half_width_for_2to1));
const uint8_t* up1 = (*upfunc1)( //
scratch_buffer_2k_ptr + (1u * 480u), //
src1_major + (x / inv_h1), //
src1_minor + (x / inv_h1), //
src_len1, //
h1v2_bias, //
first_column, //
(total_src_len1 >= half_width_for_2to1));
const uint8_t* up2 = (*upfunc2)( //
scratch_buffer_2k_ptr + (2u * 480u), //
src2_major + (x / inv_h2), //
src2_minor + (x / inv_h2), //
src_len2, //
h1v2_bias, //
first_column, //
(total_src_len2 >= half_width_for_2to1));
const uint8_t* up3 = (*upfunc3)( //
scratch_buffer_2k_ptr + (3u * 480u), //
src3_major + (x / inv_h3), //
src3_minor + (x / inv_h3), //
src_len3, //
h1v2_bias, //
first_column, //
(total_src_len3 >= half_width_for_2to1));
(*conv4func)(dst, x, end, y, up0, up1, up2, up3);
x = end;
}
h1v2_bias ^= 3u;
}
// Last row.
if (y_max_excl != height) {
wuffs_base__pixel_swizzler__swizzle_ycck__general__triangle_filter_edge_row(
dst, width, height - 1u, //
src_ptr0, src_ptr1, src_ptr2, src_ptr3, //
stride0, stride1, stride2, stride3, //
inv_h0, inv_h1, inv_h2, inv_h3, //
inv_v0, inv_v1, inv_v2, inv_v3, //
half_width_for_2to1, //
h1v2_bias, //
scratch_buffer_2k_ptr, //
upfunc0, upfunc1, upfunc2, upfunc3, conv4func);
}
}
static void //
wuffs_base__pixel_swizzler__swizzle_ycc__general__triangle_filter_edge_row(
wuffs_base__pixel_buffer* dst,
@@ -801,10 +1120,6 @@ wuffs_base__pixel_swizzler__swizzle_ycck(
wuffs_base__slice_u8 scratch_buffer_2k) {
if (!p) {
return wuffs_base__make_status(wuffs_base__error__bad_receiver);
} else if ((h3 != 0u) || (v3 != 0u)) {
// TODO: support the K in YCCK.
return wuffs_base__make_status(
wuffs_base__error__unsupported_pixel_swizzler_option);
} else if (!dst || (width > 0xFFFFu) || (height > 0xFFFFu) || //
(4u <= ((unsigned int)h0 - 1u)) || //
(4u <= ((unsigned int)h1 - 1u)) || //
@@ -815,6 +1130,12 @@ wuffs_base__pixel_swizzler__swizzle_ycck(
(scratch_buffer_2k.len < 2048u)) {
return wuffs_base__make_status(wuffs_base__error__bad_argument);
}
if ((h3 != 0u) || (v3 != 0u)) {
if ((4u <= ((unsigned int)h3 - 1u)) || //
(4u <= ((unsigned int)v3 - 1u))) {
return wuffs_base__make_status(wuffs_base__error__bad_argument);
}
}
uint32_t max_incl_h = wuffs_base__u32__max_of_4(h0, h1, h2, h3);
uint32_t max_incl_v = wuffs_base__u32__max_of_4(v0, v1, v2, v3);
@@ -826,9 +1147,11 @@ wuffs_base__pixel_swizzler__swizzle_ycck(
uint32_t inv_h0 = max_incl_h / h0;
uint32_t inv_h1 = max_incl_h / h1;
uint32_t inv_h2 = max_incl_h / h2;
uint32_t inv_h3 = h3 ? (max_incl_h / h3) : 0u;
uint32_t inv_v0 = max_incl_v / v0;
uint32_t inv_v1 = max_incl_v / v1;
uint32_t inv_v2 = max_incl_v / v2;
uint32_t inv_v3 = v3 ? (max_incl_v / v3) : 0u;
uint32_t half_width_for_2to1 = (width + 1u) / 2u;
uint32_t half_height_for_2to1 = (height + 1u) / 2u;
@@ -860,6 +1183,14 @@ wuffs_base__pixel_swizzler__swizzle_ycck(
width, height, stride2, inv_h2, inv_v2))) {
return wuffs_base__make_status(wuffs_base__error__bad_argument);
}
if ((h3 != 0u) || (v3 != 0u)) {
if (((h3 * inv_h3) != max_incl_h) || //
((v3 * inv_v3) != max_incl_v) || //
(src3.len < wuffs_base__pixel_swizzler__flattened_length(
width, height, stride3, inv_h3, inv_v3))) {
return wuffs_base__make_status(wuffs_base__error__bad_argument);
}
}
if (wuffs_base__pixel_format__is_planar(&dst->pixcfg.private_impl.pixfmt)) {
// TODO: see wuffs_base__pixel_buffer__set_color_u32_at's TODO.
@@ -994,13 +1325,30 @@ wuffs_base__pixel_swizzler__swizzle_ycck(
#endif
}
(*func)(dst, width, height, //
src0.ptr, src1.ptr, src2.ptr, //
stride0, stride1, stride2, //
inv_h0, inv_h1, inv_h2, //
inv_v0, inv_v1, inv_v2, //
half_width_for_2to1, half_height_for_2to1, //
scratch_buffer_2k.ptr, &upfuncs, conv3func);
if ((h3 != 0u) || (v3 != 0u)) {
wuffs_base__pixel_swizzler__swizzle_ycc__convert_4_func conv4func =
is_rgb_or_cmyk
? &wuffs_base__pixel_swizzler__swizzle_cmyk__convert_4_general
: &wuffs_base__pixel_swizzler__swizzle_ycck__convert_4_general;
wuffs_base__pixel_swizzler__swizzle_ycck__general__triangle_filter( //
dst, width, height, //
src0.ptr, src1.ptr, src2.ptr, src3.ptr, //
stride0, stride1, stride2, stride3, //
inv_h0, inv_h1, inv_h2, inv_h3, //
inv_v0, inv_v1, inv_v2, inv_v3, //
half_width_for_2to1, half_height_for_2to1, //
scratch_buffer_2k.ptr, &upfuncs, conv4func);
} else {
(*func)( //
dst, width, height, //
src0.ptr, src1.ptr, src2.ptr, //
stride0, stride1, stride2, //
inv_h0, inv_h1, inv_h2, //
inv_v0, inv_v1, inv_v2, //
half_width_for_2to1, half_height_for_2to1, //
scratch_buffer_2k.ptr, &upfuncs, conv3func);
}
return wuffs_base__make_status(NULL);
}
+367 -19
View File
@@ -23641,6 +23641,71 @@ wuffs_base__u32__min_of_5(uint32_t a,
// --------
typedef void (*wuffs_base__pixel_swizzler__swizzle_ycc__convert_4_func)(
wuffs_base__pixel_buffer* dst,
uint32_t x,
uint32_t x_end,
uint32_t y,
const uint8_t* up0,
const uint8_t* up1,
const uint8_t* up2,
const uint8_t* up3);
static void //
wuffs_base__pixel_swizzler__swizzle_cmyk__convert_4_general(
wuffs_base__pixel_buffer* dst,
uint32_t x,
uint32_t x_end,
uint32_t y,
const uint8_t* up0,
const uint8_t* up1,
const uint8_t* up2,
const uint8_t* up3) {
for (; x < x_end; x++) {
// It's called CMYK but, but for Adobe CMYK JPEG images in practice, it's
// RGBW: 0xFFu means no ink instead of full ink. Note that a double
// inversion is a no-op, so inversions might be implicit in the code below.
uint32_t r = ((uint32_t)(*up0++));
uint32_t g = ((uint32_t)(*up1++));
uint32_t b = ((uint32_t)(*up2++));
uint32_t w = ((uint32_t)(*up3++));
r = ((r * w) + 0x7Fu) / 0xFFu;
g = ((g * w) + 0x7Fu) / 0xFFu;
b = ((b * w) + 0x7Fu) / 0xFFu;
wuffs_base__pixel_buffer__set_color_u32_at(
dst, x, y, 0xFF000000u | (r << 16u) | (g << 8u) | (b << 0u));
}
}
static void //
wuffs_base__pixel_swizzler__swizzle_ycck__convert_4_general(
wuffs_base__pixel_buffer* dst,
uint32_t x,
uint32_t x_end,
uint32_t y,
const uint8_t* up0,
const uint8_t* up1,
const uint8_t* up2,
const uint8_t* up3) {
for (; x < x_end; x++) {
// We invert once again: 0xFFu means no ink instead of full ink.
uint32_t color = //
wuffs_base__color_ycc__as__color_u32( //
*up0++, *up1++, *up2++);
uint32_t r = 0xFFu - (0xFFu & (color >> 16u));
uint32_t g = 0xFFu - (0xFFu & (color >> 8u));
uint32_t b = 0xFFu - (0xFFu & (color >> 0u));
uint32_t w = ((uint32_t)(*up3++));
r = ((r * w) + 0x7Fu) / 0xFFu;
g = ((g * w) + 0x7Fu) / 0xFFu;
b = ((b * w) + 0x7Fu) / 0xFFu;
wuffs_base__pixel_buffer__set_color_u32_at(
dst, x, y, 0xFF000000u | (r << 16u) | (g << 8u) | (b << 0u));
}
}
// --------
typedef void (*wuffs_base__pixel_swizzler__swizzle_ycc__convert_3_func)(
wuffs_base__pixel_buffer* dst,
uint32_t x,
@@ -23660,10 +23725,10 @@ wuffs_base__pixel_swizzler__swizzle_rgb__convert_3_general(
const uint8_t* up1,
const uint8_t* up2) {
for (; x < x_end; x++) {
uint32_t color = 0xFF000000 | //
(((uint32_t)(*up0++)) << 16) | //
(((uint32_t)(*up1++)) << 8) | //
(((uint32_t)(*up2++)) << 0);
uint32_t color = 0xFF000000u | //
(((uint32_t)(*up0++)) << 16u) | //
(((uint32_t)(*up1++)) << 8u) | //
(((uint32_t)(*up2++)) << 0u);
wuffs_base__pixel_buffer__set_color_u32_at(dst, x, y, color);
}
}
@@ -23696,14 +23761,14 @@ wuffs_base__pixel_swizzler__swizzle_ycc__convert_3_bgrx(
const uint8_t* up2) {
size_t dst_stride = dst->private_impl.planes[0].stride;
uint8_t* dst_iter = dst->private_impl.planes[0].ptr +
(dst_stride * ((size_t)y)) + (4 * ((size_t)x));
(dst_stride * ((size_t)y)) + (4u * ((size_t)x));
for (; x < x_end; x++) {
uint32_t color = //
wuffs_base__color_ycc__as__color_u32( //
*up0++, *up1++, *up2++);
wuffs_base__poke_u32le__no_bounds_check(dst_iter, color);
dst_iter += 4;
dst_iter += 4u;
}
}
@@ -23718,14 +23783,14 @@ wuffs_base__pixel_swizzler__swizzle_ycc__convert_3_rgbx(
const uint8_t* up2) {
size_t dst_stride = dst->private_impl.planes[0].stride;
uint8_t* dst_iter = dst->private_impl.planes[0].ptr +
(dst_stride * ((size_t)y)) + (4 * ((size_t)x));
(dst_stride * ((size_t)y)) + (4u * ((size_t)x));
for (; x < x_end; x++) {
uint32_t color = //
wuffs_base__color_ycc__as__color_u32_abgr( //
*up0++, *up1++, *up2++);
wuffs_base__poke_u32le__no_bounds_check(dst_iter, color);
dst_iter += 4;
dst_iter += 4u;
}
}
@@ -24025,6 +24090,260 @@ wuffs_base__pixel_swizzler__has_triangle_upsampler(uint32_t inv_h,
// wuffs_base__pixel_swizzler__swizzle_ycck before calling these functions. For
// example, (width > 0) is a precondition, but there are many more.
static void //
wuffs_base__pixel_swizzler__swizzle_ycck__general__triangle_filter_edge_row(
wuffs_base__pixel_buffer* dst,
uint32_t width,
uint32_t y,
const uint8_t* src_ptr0,
const uint8_t* src_ptr1,
const uint8_t* src_ptr2,
const uint8_t* src_ptr3,
uint32_t stride0,
uint32_t stride1,
uint32_t stride2,
uint32_t stride3,
uint32_t inv_h0,
uint32_t inv_h1,
uint32_t inv_h2,
uint32_t inv_h3,
uint32_t inv_v0,
uint32_t inv_v1,
uint32_t inv_v2,
uint32_t inv_v3,
uint32_t half_width_for_2to1,
uint32_t h1v2_bias,
uint8_t* scratch_buffer_2k_ptr,
wuffs_base__pixel_swizzler__swizzle_ycc__upsample_func upfunc0,
wuffs_base__pixel_swizzler__swizzle_ycc__upsample_func upfunc1,
wuffs_base__pixel_swizzler__swizzle_ycc__upsample_func upfunc2,
wuffs_base__pixel_swizzler__swizzle_ycc__upsample_func upfunc3,
wuffs_base__pixel_swizzler__swizzle_ycc__convert_4_func conv4func) {
const uint8_t* src0 = src_ptr0 + ((y / inv_v0) * (size_t)stride0);
const uint8_t* src1 = src_ptr1 + ((y / inv_v1) * (size_t)stride1);
const uint8_t* src2 = src_ptr2 + ((y / inv_v2) * (size_t)stride2);
const uint8_t* src3 = src_ptr3 + ((y / inv_v3) * (size_t)stride3);
uint32_t total_src_len0 = 0u;
uint32_t total_src_len1 = 0u;
uint32_t total_src_len2 = 0u;
uint32_t total_src_len3 = 0u;
uint32_t x = 0u;
while (x < width) {
bool first_column = x == 0u;
uint32_t end = x + 480u;
if (end > width) {
end = width;
}
uint32_t src_len0 = ((end - x) + inv_h0 - 1u) / inv_h0;
uint32_t src_len1 = ((end - x) + inv_h1 - 1u) / inv_h1;
uint32_t src_len2 = ((end - x) + inv_h2 - 1u) / inv_h2;
uint32_t src_len3 = ((end - x) + inv_h3 - 1u) / inv_h3;
total_src_len0 += src_len0;
total_src_len1 += src_len1;
total_src_len2 += src_len2;
total_src_len3 += src_len3;
const uint8_t* src_ptr_x0 = src0 + (x / inv_h0);
const uint8_t* up0 = (*upfunc0)( //
scratch_buffer_2k_ptr + (0u * 480u), //
src_ptr_x0, //
src_ptr_x0, //
src_len0, //
h1v2_bias, //
first_column, //
(total_src_len0 >= half_width_for_2to1));
const uint8_t* src_ptr_x1 = src1 + (x / inv_h1);
const uint8_t* up1 = (*upfunc1)( //
scratch_buffer_2k_ptr + (1u * 480u), //
src_ptr_x1, //
src_ptr_x1, //
src_len1, //
h1v2_bias, //
first_column, //
(total_src_len1 >= half_width_for_2to1));
const uint8_t* src_ptr_x2 = src2 + (x / inv_h2);
const uint8_t* up2 = (*upfunc2)( //
scratch_buffer_2k_ptr + (2u * 480u), //
src_ptr_x2, //
src_ptr_x2, //
src_len2, //
h1v2_bias, //
first_column, //
(total_src_len2 >= half_width_for_2to1));
const uint8_t* src_ptr_x3 = src3 + (x / inv_h3);
const uint8_t* up3 = (*upfunc3)( //
scratch_buffer_2k_ptr + (3u * 480u), //
src_ptr_x3, //
src_ptr_x3, //
src_len3, //
h1v2_bias, //
first_column, //
(total_src_len3 >= half_width_for_2to1));
(*conv4func)(dst, x, end, y, up0, up1, up2, up3);
x = end;
}
}
static void //
wuffs_base__pixel_swizzler__swizzle_ycck__general__triangle_filter(
wuffs_base__pixel_buffer* dst,
uint32_t width,
uint32_t height,
const uint8_t* src_ptr0,
const uint8_t* src_ptr1,
const uint8_t* src_ptr2,
const uint8_t* src_ptr3,
uint32_t stride0,
uint32_t stride1,
uint32_t stride2,
uint32_t stride3,
uint32_t inv_h0,
uint32_t inv_h1,
uint32_t inv_h2,
uint32_t inv_h3,
uint32_t inv_v0,
uint32_t inv_v1,
uint32_t inv_v2,
uint32_t inv_v3,
uint32_t half_width_for_2to1,
uint32_t half_height_for_2to1,
uint8_t* scratch_buffer_2k_ptr,
wuffs_base__pixel_swizzler__swizzle_ycc__upsample_func (*upfuncs)[4][4],
wuffs_base__pixel_swizzler__swizzle_ycc__convert_4_func conv4func) {
wuffs_base__pixel_swizzler__swizzle_ycc__upsample_func upfunc0 =
(*upfuncs)[(inv_h0 - 1u) & 3u][(inv_v0 - 1u) & 3u];
wuffs_base__pixel_swizzler__swizzle_ycc__upsample_func upfunc1 =
(*upfuncs)[(inv_h1 - 1u) & 3u][(inv_v1 - 1u) & 3u];
wuffs_base__pixel_swizzler__swizzle_ycc__upsample_func upfunc2 =
(*upfuncs)[(inv_h2 - 1u) & 3u][(inv_v2 - 1u) & 3u];
wuffs_base__pixel_swizzler__swizzle_ycc__upsample_func upfunc3 =
(*upfuncs)[(inv_h3 - 1u) & 3u][(inv_v3 - 1u) & 3u];
// First row.
uint32_t h1v2_bias = 1u;
wuffs_base__pixel_swizzler__swizzle_ycck__general__triangle_filter_edge_row(
dst, width, 0u, //
src_ptr0, src_ptr1, src_ptr2, src_ptr3, //
stride0, stride1, stride2, stride3, //
inv_h0, inv_h1, inv_h2, inv_h3, //
inv_v0, inv_v1, inv_v2, inv_v3, //
half_width_for_2to1, //
h1v2_bias, //
scratch_buffer_2k_ptr, //
upfunc0, upfunc1, upfunc2, upfunc3, conv4func);
h1v2_bias = 2u;
// Middle rows.
bool last_row = height == 2u * half_height_for_2to1;
uint32_t y_max_excl = last_row ? (height - 1u) : height;
uint32_t y;
for (y = 1u; y < y_max_excl; y++) {
const uint8_t* src0_major = src_ptr0 + ((y / inv_v0) * (size_t)stride0);
const uint8_t* src0_minor =
(inv_v0 != 2u)
? src0_major
: ((y & 1u) ? (src0_major + stride0) : (src0_major - stride0));
const uint8_t* src1_major = src_ptr1 + ((y / inv_v1) * (size_t)stride1);
const uint8_t* src1_minor =
(inv_v1 != 2u)
? src1_major
: ((y & 1u) ? (src1_major + stride1) : (src1_major - stride1));
const uint8_t* src2_major = src_ptr2 + ((y / inv_v2) * (size_t)stride2);
const uint8_t* src2_minor =
(inv_v2 != 2u)
? src2_major
: ((y & 1u) ? (src2_major + stride2) : (src2_major - stride2));
const uint8_t* src3_major = src_ptr3 + ((y / inv_v3) * (size_t)stride3);
const uint8_t* src3_minor =
(inv_v3 != 2u)
? src3_major
: ((y & 1u) ? (src3_major + stride3) : (src3_major - stride3));
uint32_t total_src_len0 = 0u;
uint32_t total_src_len1 = 0u;
uint32_t total_src_len2 = 0u;
uint32_t total_src_len3 = 0u;
uint32_t x = 0u;
while (x < width) {
bool first_column = x == 0u;
uint32_t end = x + 480u;
if (end > width) {
end = width;
}
uint32_t src_len0 = ((end - x) + inv_h0 - 1u) / inv_h0;
uint32_t src_len1 = ((end - x) + inv_h1 - 1u) / inv_h1;
uint32_t src_len2 = ((end - x) + inv_h2 - 1u) / inv_h2;
uint32_t src_len3 = ((end - x) + inv_h3 - 1u) / inv_h3;
total_src_len0 += src_len0;
total_src_len1 += src_len1;
total_src_len2 += src_len2;
total_src_len3 += src_len3;
const uint8_t* up0 = (*upfunc0)( //
scratch_buffer_2k_ptr + (0u * 480u), //
src0_major + (x / inv_h0), //
src0_minor + (x / inv_h0), //
src_len0, //
h1v2_bias, //
first_column, //
(total_src_len0 >= half_width_for_2to1));
const uint8_t* up1 = (*upfunc1)( //
scratch_buffer_2k_ptr + (1u * 480u), //
src1_major + (x / inv_h1), //
src1_minor + (x / inv_h1), //
src_len1, //
h1v2_bias, //
first_column, //
(total_src_len1 >= half_width_for_2to1));
const uint8_t* up2 = (*upfunc2)( //
scratch_buffer_2k_ptr + (2u * 480u), //
src2_major + (x / inv_h2), //
src2_minor + (x / inv_h2), //
src_len2, //
h1v2_bias, //
first_column, //
(total_src_len2 >= half_width_for_2to1));
const uint8_t* up3 = (*upfunc3)( //
scratch_buffer_2k_ptr + (3u * 480u), //
src3_major + (x / inv_h3), //
src3_minor + (x / inv_h3), //
src_len3, //
h1v2_bias, //
first_column, //
(total_src_len3 >= half_width_for_2to1));
(*conv4func)(dst, x, end, y, up0, up1, up2, up3);
x = end;
}
h1v2_bias ^= 3u;
}
// Last row.
if (y_max_excl != height) {
wuffs_base__pixel_swizzler__swizzle_ycck__general__triangle_filter_edge_row(
dst, width, height - 1u, //
src_ptr0, src_ptr1, src_ptr2, src_ptr3, //
stride0, stride1, stride2, stride3, //
inv_h0, inv_h1, inv_h2, inv_h3, //
inv_v0, inv_v1, inv_v2, inv_v3, //
half_width_for_2to1, //
h1v2_bias, //
scratch_buffer_2k_ptr, //
upfunc0, upfunc1, upfunc2, upfunc3, conv4func);
}
}
static void //
wuffs_base__pixel_swizzler__swizzle_ycc__general__triangle_filter_edge_row(
wuffs_base__pixel_buffer* dst,
@@ -24365,10 +24684,6 @@ wuffs_base__pixel_swizzler__swizzle_ycck(
wuffs_base__slice_u8 scratch_buffer_2k) {
if (!p) {
return wuffs_base__make_status(wuffs_base__error__bad_receiver);
} else if ((h3 != 0u) || (v3 != 0u)) {
// TODO: support the K in YCCK.
return wuffs_base__make_status(
wuffs_base__error__unsupported_pixel_swizzler_option);
} else if (!dst || (width > 0xFFFFu) || (height > 0xFFFFu) || //
(4u <= ((unsigned int)h0 - 1u)) || //
(4u <= ((unsigned int)h1 - 1u)) || //
@@ -24379,6 +24694,12 @@ wuffs_base__pixel_swizzler__swizzle_ycck(
(scratch_buffer_2k.len < 2048u)) {
return wuffs_base__make_status(wuffs_base__error__bad_argument);
}
if ((h3 != 0u) || (v3 != 0u)) {
if ((4u <= ((unsigned int)h3 - 1u)) || //
(4u <= ((unsigned int)v3 - 1u))) {
return wuffs_base__make_status(wuffs_base__error__bad_argument);
}
}
uint32_t max_incl_h = wuffs_base__u32__max_of_4(h0, h1, h2, h3);
uint32_t max_incl_v = wuffs_base__u32__max_of_4(v0, v1, v2, v3);
@@ -24390,9 +24711,11 @@ wuffs_base__pixel_swizzler__swizzle_ycck(
uint32_t inv_h0 = max_incl_h / h0;
uint32_t inv_h1 = max_incl_h / h1;
uint32_t inv_h2 = max_incl_h / h2;
uint32_t inv_h3 = h3 ? (max_incl_h / h3) : 0u;
uint32_t inv_v0 = max_incl_v / v0;
uint32_t inv_v1 = max_incl_v / v1;
uint32_t inv_v2 = max_incl_v / v2;
uint32_t inv_v3 = v3 ? (max_incl_v / v3) : 0u;
uint32_t half_width_for_2to1 = (width + 1u) / 2u;
uint32_t half_height_for_2to1 = (height + 1u) / 2u;
@@ -24424,6 +24747,14 @@ wuffs_base__pixel_swizzler__swizzle_ycck(
width, height, stride2, inv_h2, inv_v2))) {
return wuffs_base__make_status(wuffs_base__error__bad_argument);
}
if ((h3 != 0u) || (v3 != 0u)) {
if (((h3 * inv_h3) != max_incl_h) || //
((v3 * inv_v3) != max_incl_v) || //
(src3.len < wuffs_base__pixel_swizzler__flattened_length(
width, height, stride3, inv_h3, inv_v3))) {
return wuffs_base__make_status(wuffs_base__error__bad_argument);
}
}
if (wuffs_base__pixel_format__is_planar(&dst->pixcfg.private_impl.pixfmt)) {
// TODO: see wuffs_base__pixel_buffer__set_color_u32_at's TODO.
@@ -24558,13 +24889,30 @@ wuffs_base__pixel_swizzler__swizzle_ycck(
#endif
}
(*func)(dst, width, height, //
src0.ptr, src1.ptr, src2.ptr, //
stride0, stride1, stride2, //
inv_h0, inv_h1, inv_h2, //
inv_v0, inv_v1, inv_v2, //
half_width_for_2to1, half_height_for_2to1, //
scratch_buffer_2k.ptr, &upfuncs, conv3func);
if ((h3 != 0u) || (v3 != 0u)) {
wuffs_base__pixel_swizzler__swizzle_ycc__convert_4_func conv4func =
is_rgb_or_cmyk
? &wuffs_base__pixel_swizzler__swizzle_cmyk__convert_4_general
: &wuffs_base__pixel_swizzler__swizzle_ycck__convert_4_general;
wuffs_base__pixel_swizzler__swizzle_ycck__general__triangle_filter( //
dst, width, height, //
src0.ptr, src1.ptr, src2.ptr, src3.ptr, //
stride0, stride1, stride2, stride3, //
inv_h0, inv_h1, inv_h2, inv_h3, //
inv_v0, inv_v1, inv_v2, inv_v3, //
half_width_for_2to1, half_height_for_2to1, //
scratch_buffer_2k.ptr, &upfuncs, conv4func);
} else {
(*func)( //
dst, width, height, //
src0.ptr, src1.ptr, src2.ptr, //
stride0, stride1, stride2, //
inv_h0, inv_h1, inv_h2, //
inv_v0, inv_v1, inv_v2, //
half_width_for_2to1, half_height_for_2to1, //
scratch_buffer_2k.ptr, &upfuncs, conv3func);
}
return wuffs_base__make_status(NULL);
}
+6 -6
View File
@@ -37,7 +37,7 @@ ffc8a627 test/3pdata/blinksuite/bug106024.jpg
18cee321 test/3pdata/blinksuite/cHRM_color_spin.png
3e460fe4 test/3pdata/blinksuite/cat.jpg
e9151f0e test/3pdata/blinksuite/cicp_pq.png
e00ca328 test/3pdata/blinksuite/cmyk-jpeg.jpg
ab6d7d2e test/3pdata/blinksuite/cmyk-jpeg.jpg
5ec78d7c test/3pdata/blinksuite/color-checker-adobe-color-profile.png
b3389858 test/3pdata/blinksuite/color-checker-srgb-color-profile.png
10721531 test/3pdata/blinksuite/count-down-color-test.gif
@@ -46,10 +46,10 @@ b3389858 test/3pdata/blinksuite/color-checker-srgb-color-profile.png
ae034abe test/3pdata/blinksuite/crbug807324.png
4f50fdb0 test/3pdata/blinksuite/crbug827754.png
42e5598b test/3pdata/blinksuite/cropped_mandrill.jpg
e00ca328 test/3pdata/blinksuite/cs-uma-cmyk-jfif-marker.jpg
e00ca328 test/3pdata/blinksuite/cs-uma-cmyk-no-jfif-or-adobe-markers.jpg
e00ca328 test/3pdata/blinksuite/cs-uma-cmyk-unknown-transform.jpg
e00ca328 test/3pdata/blinksuite/cs-uma-cmyk.jpg
fe2877fe test/3pdata/blinksuite/cs-uma-cmyk-jfif-marker.jpg
fe2877fe test/3pdata/blinksuite/cs-uma-cmyk-no-jfif-or-adobe-markers.jpg
f8341b72 test/3pdata/blinksuite/cs-uma-cmyk-unknown-transform.jpg
fe2877fe test/3pdata/blinksuite/cs-uma-cmyk.jpg
1ec303fd test/3pdata/blinksuite/cs-uma-grayscale.jpg
c9eafb54 test/3pdata/blinksuite/cs-uma-rgb-non-interleaved.jpg
528a67f1 test/3pdata/blinksuite/cs-uma-rgb-unknown-transform.jpg
@@ -63,7 +63,7 @@ a027c965 test/3pdata/blinksuite/cs-uma-ycbcr-420-non-interleaved.jpg
8bd535aa test/3pdata/blinksuite/cs-uma-ycbcr-440.jpg
1bc16d1a test/3pdata/blinksuite/cs-uma-ycbcr-444.jpg
d97ce4e8 test/3pdata/blinksuite/cs-uma-ycbcr-other.jpg
e00ca328 test/3pdata/blinksuite/cs-uma-ycck.jpg
a2a940bc test/3pdata/blinksuite/cs-uma-ycck.jpg
b6000be6 test/3pdata/blinksuite/dice.png
1023c20a test/3pdata/blinksuite/empty-frame.png
32a03152 test/3pdata/blinksuite/flowchart.jpg
+1 -1
View File
@@ -3,7 +3,7 @@
f47bba01 test/3pdata/golangsuite/video-001.221212.jpeg
f887663d test/3pdata/golangsuite/video-001.221212.png
9db259ab test/3pdata/golangsuite/video-001.5bpp.gif
1263dc39 test/3pdata/golangsuite/video-001.cmyk.jpeg
7b281f70 test/3pdata/golangsuite/video-001.cmyk.jpeg
f543e98a test/3pdata/golangsuite/video-001.cmyk.png
0548cd56 test/3pdata/golangsuite/video-001.gif
0548cd56 test/3pdata/golangsuite/video-001.interlaced.gif