aboutsummaryrefslogtreecommitdiffhomepage
diff options
context:
space:
mode:
Diffstat
-rw-r--r--bench/SwizzleBench.cpp2
-rw-r--r--src/codec/SkSwizzler.cpp17
-rw-r--r--src/core/SkOpts.cpp18
-rw-r--r--src/core/SkOpts.h18
-rw-r--r--src/opts/SkOpts_neon.cpp18
-rw-r--r--src/opts/SkOpts_ssse3.cpp18
-rw-r--r--src/opts/SkSwizzler_opts.h168
7 files changed, 227 insertions, 32 deletions
diff --git a/bench/SwizzleBench.cpp b/bench/SwizzleBench.cpp
index cf7a407f0e..5490a6c44b 100644
--- a/bench/SwizzleBench.cpp
+++ b/bench/SwizzleBench.cpp
@@ -35,3 +35,5 @@ DEF_BENCH(return new SwizzleBench("SkOpts::RGB_to_BGR1", SkOpts::RGB_to_BGR1));
DEF_BENCH(return new SwizzleBench("SkOpts::gray_to_RGB1", SkOpts::gray_to_RGB1));
DEF_BENCH(return new SwizzleBench("SkOpts::grayA_to_RGBA", SkOpts::grayA_to_RGBA));
DEF_BENCH(return new SwizzleBench("SkOpts::grayA_to_rgbA", SkOpts::grayA_to_rgbA));
+DEF_BENCH(return new SwizzleBench("SkOpts::inverted_CMYK_to_RGB1", SkOpts::inverted_CMYK_to_RGB1));
+DEF_BENCH(return new SwizzleBench("SkOpts::inverted_CMYK_to_BGR1", SkOpts::inverted_CMYK_to_BGR1));
diff --git a/src/codec/SkSwizzler.cpp b/src/codec/SkSwizzler.cpp
index debcd45c20..f9f212072e 100644
--- a/src/codec/SkSwizzler.cpp
+++ b/src/codec/SkSwizzler.cpp
@@ -592,6 +592,21 @@ static void swizzle_cmyk_to_n32(
}
}
+static void fast_swizzle_cmyk_to_n32(
+ void* dst, const uint8_t* src, int width, int bpp, int deltaSrc, int offset,
+ const SkPMColor ctable[]) {
+
+ // This function must not be called if we are sampling. If we are not
+ // sampling, deltaSrc should equal bpp.
+ SkASSERT(deltaSrc == bpp);
+
+#ifdef SK_PMCOLOR_IS_RGBA
+ SkOpts::inverted_CMYK_to_RGB1((uint32_t*) dst, src + offset, width);
+#else
+ SkOpts::inverted_CMYK_to_BGR1((uint32_t*) dst, src + offset, width);
+#endif
+}
+
static void swizzle_cmyk_to_565(
void* SK_RESTRICT dstRow, const uint8_t* SK_RESTRICT src, int dstWidth,
int bpp, int deltaSrc, int offset, const SkPMColor ctable[]) {
@@ -811,6 +826,7 @@ SkSwizzler* SkSwizzler::CreateSwizzler(SkSwizzler::SrcConfig sc,
break;
case kRGB_565_SkColorType:
proc = &swizzle_rgb_to_565;
+ break;
default:
break;
}
@@ -844,6 +860,7 @@ SkSwizzler* SkSwizzler::CreateSwizzler(SkSwizzler::SrcConfig sc,
switch (dstInfo.colorType()) {
case kN32_SkColorType:
proc = &swizzle_cmyk_to_n32;
+ fastProc = &fast_swizzle_cmyk_to_n32;
break;
case kRGB_565_SkColorType:
proc = &swizzle_cmyk_to_565;
diff --git a/src/core/SkOpts.cpp b/src/core/SkOpts.cpp
index 8065b77d85..56604171bb 100644
--- a/src/core/SkOpts.cpp
+++ b/src/core/SkOpts.cpp
@@ -79,14 +79,16 @@ namespace SkOpts {
decltype(matrix_scale_translate) matrix_scale_translate = sk_default::matrix_scale_translate;
decltype(matrix_affine) matrix_affine = sk_default::matrix_affine;
- decltype(RGBA_to_BGRA) RGBA_to_BGRA = sk_default::RGBA_to_BGRA;
- decltype(RGBA_to_rgbA) RGBA_to_rgbA = sk_default::RGBA_to_rgbA;
- decltype(RGBA_to_bgrA) RGBA_to_bgrA = sk_default::RGBA_to_bgrA;
- decltype(RGB_to_RGB1) RGB_to_RGB1 = sk_default::RGB_to_RGB1;
- decltype(RGB_to_BGR1) RGB_to_BGR1 = sk_default::RGB_to_BGR1;
- decltype(gray_to_RGB1) gray_to_RGB1 = sk_default::gray_to_RGB1;
- decltype(grayA_to_RGBA) grayA_to_RGBA = sk_default::grayA_to_RGBA;
- decltype(grayA_to_rgbA) grayA_to_rgbA = sk_default::grayA_to_rgbA;
+ decltype(RGBA_to_BGRA) RGBA_to_BGRA = sk_default::RGBA_to_BGRA;
+ decltype(RGBA_to_rgbA) RGBA_to_rgbA = sk_default::RGBA_to_rgbA;
+ decltype(RGBA_to_bgrA) RGBA_to_bgrA = sk_default::RGBA_to_bgrA;
+ decltype(RGB_to_RGB1) RGB_to_RGB1 = sk_default::RGB_to_RGB1;
+ decltype(RGB_to_BGR1) RGB_to_BGR1 = sk_default::RGB_to_BGR1;
+ decltype(gray_to_RGB1) gray_to_RGB1 = sk_default::gray_to_RGB1;
+ decltype(grayA_to_RGBA) grayA_to_RGBA = sk_default::grayA_to_RGBA;
+ decltype(grayA_to_rgbA) grayA_to_rgbA = sk_default::grayA_to_rgbA;
+ decltype(inverted_CMYK_to_RGB1) inverted_CMYK_to_RGB1 = sk_default::inverted_CMYK_to_RGB1;
+ decltype(inverted_CMYK_to_BGR1) inverted_CMYK_to_BGR1 = sk_default::inverted_CMYK_to_BGR1;
// Each Init_foo() is defined in src/opts/SkOpts_foo.cpp.
void Init_ssse3();
diff --git a/src/core/SkOpts.h b/src/core/SkOpts.h
index 69da36254c..2e8778e1a1 100644
--- a/src/core/SkOpts.h
+++ b/src/core/SkOpts.h
@@ -57,14 +57,16 @@ namespace SkOpts {
// Swizzle input into some sort of 8888 pixel, {premul,unpremul} x {rgba,bgra}.
typedef void (*Swizzle_8888)(uint32_t*, const void*, int);
- extern Swizzle_8888 RGBA_to_BGRA, // i.e. just swap RB
- RGBA_to_rgbA, // i.e. just premultiply
- RGBA_to_bgrA, // i.e. swap RB and premultiply
- RGB_to_RGB1, // i.e. insert an opaque alpha
- RGB_to_BGR1, // i.e. swap RB and insert an opaque alpha
- gray_to_RGB1, // i.e. expand to color channels + an opaque alpha
- grayA_to_RGBA, // i.e. expand to color channels
- grayA_to_rgbA; // i.e. expand to color channels and premultiply
+ extern Swizzle_8888 RGBA_to_BGRA, // i.e. just swap RB
+ RGBA_to_rgbA, // i.e. just premultiply
+ RGBA_to_bgrA, // i.e. swap RB and premultiply
+ RGB_to_RGB1, // i.e. insert an opaque alpha
+ RGB_to_BGR1, // i.e. swap RB and insert an opaque alpha
+ gray_to_RGB1, // i.e. expand to color channels + an opaque alpha
+ grayA_to_RGBA, // i.e. expand to color channels
+ grayA_to_rgbA, // i.e. expand to color channels and premultiply
+ inverted_CMYK_to_RGB1, // i.e. convert color space
+ inverted_CMYK_to_BGR1; // i.e. convert color space
}
#endif//SkOpts_DEFINED
diff --git a/src/opts/SkOpts_neon.cpp b/src/opts/SkOpts_neon.cpp
index 01f43ef3ed..80fb4e93d0 100644
--- a/src/opts/SkOpts_neon.cpp
+++ b/src/opts/SkOpts_neon.cpp
@@ -47,13 +47,15 @@ namespace SkOpts {
matrix_scale_translate = sk_neon::matrix_scale_translate;
matrix_affine = sk_neon::matrix_affine;
- RGBA_to_BGRA = sk_neon::RGBA_to_BGRA;
- RGBA_to_rgbA = sk_neon::RGBA_to_rgbA;
- RGBA_to_bgrA = sk_neon::RGBA_to_bgrA;
- RGB_to_RGB1 = sk_neon::RGB_to_RGB1;
- RGB_to_BGR1 = sk_neon::RGB_to_BGR1;
- gray_to_RGB1 = sk_neon::gray_to_RGB1;
- grayA_to_RGBA = sk_neon::grayA_to_RGBA;
- grayA_to_rgbA = sk_neon::grayA_to_rgbA;
+ RGBA_to_BGRA = sk_neon::RGBA_to_BGRA;
+ RGBA_to_rgbA = sk_neon::RGBA_to_rgbA;
+ RGBA_to_bgrA = sk_neon::RGBA_to_bgrA;
+ RGB_to_RGB1 = sk_neon::RGB_to_RGB1;
+ RGB_to_BGR1 = sk_neon::RGB_to_BGR1;
+ gray_to_RGB1 = sk_neon::gray_to_RGB1;
+ grayA_to_RGBA = sk_neon::grayA_to_RGBA;
+ grayA_to_rgbA = sk_neon::grayA_to_rgbA;
+ inverted_CMYK_to_RGB1 = sk_neon::inverted_CMYK_to_RGB1;
+ inverted_CMYK_to_BGR1 = sk_neon::inverted_CMYK_to_BGR1;
}
}
diff --git a/src/opts/SkOpts_ssse3.cpp b/src/opts/SkOpts_ssse3.cpp
index 13c8fbf8e2..94a26a1ca2 100644
--- a/src/opts/SkOpts_ssse3.cpp
+++ b/src/opts/SkOpts_ssse3.cpp
@@ -18,13 +18,15 @@ namespace SkOpts {
blit_mask_d32_a8 = sk_ssse3::blit_mask_d32_a8;
color_cube_filter_span = sk_ssse3::color_cube_filter_span;
- RGBA_to_BGRA = sk_ssse3::RGBA_to_BGRA;
- RGBA_to_rgbA = sk_ssse3::RGBA_to_rgbA;
- RGBA_to_bgrA = sk_ssse3::RGBA_to_bgrA;
- RGB_to_RGB1 = sk_ssse3::RGB_to_RGB1;
- RGB_to_BGR1 = sk_ssse3::RGB_to_BGR1;
- gray_to_RGB1 = sk_ssse3::gray_to_RGB1;
- grayA_to_RGBA = sk_ssse3::grayA_to_RGBA;
- grayA_to_rgbA = sk_ssse3::grayA_to_rgbA;
+ RGBA_to_BGRA = sk_ssse3::RGBA_to_BGRA;
+ RGBA_to_rgbA = sk_ssse3::RGBA_to_rgbA;
+ RGBA_to_bgrA = sk_ssse3::RGBA_to_bgrA;
+ RGB_to_RGB1 = sk_ssse3::RGB_to_RGB1;
+ RGB_to_BGR1 = sk_ssse3::RGB_to_BGR1;
+ gray_to_RGB1 = sk_ssse3::gray_to_RGB1;
+ grayA_to_RGBA = sk_ssse3::grayA_to_RGBA;
+ grayA_to_rgbA = sk_ssse3::grayA_to_rgbA;
+ inverted_CMYK_to_RGB1 = sk_ssse3::inverted_CMYK_to_RGB1;
+ inverted_CMYK_to_BGR1 = sk_ssse3::inverted_CMYK_to_BGR1;
}
}
diff --git a/src/opts/SkSwizzler_opts.h b/src/opts/SkSwizzler_opts.h
index 1d3cc51c37..15eec3a355 100644
--- a/src/opts/SkSwizzler_opts.h
+++ b/src/opts/SkSwizzler_opts.h
@@ -125,6 +125,41 @@ static void grayA_to_rgbA_portable(uint32_t dst[], const void* vsrc, int count)
}
}
+static void inverted_CMYK_to_RGB1_portable(uint32_t* dst, const void* vsrc, int count) {
+ const uint32_t* src = (const uint32_t*)vsrc;
+ for (int i = 0; i < count; i++) {
+ uint8_t k = src[i] >> 24,
+ y = src[i] >> 16,
+ m = src[i] >> 8,
+ c = src[i] >> 0;
+ // See comments in SkSwizzler.cpp for details on the conversion formula.
+ uint8_t b = (y*k+127)/255,
+ g = (m*k+127)/255,
+ r = (c*k+127)/255;
+ dst[i] = (uint32_t)0xFF << 24
+ | (uint32_t) b << 16
+ | (uint32_t) g << 8
+ | (uint32_t) r << 0;
+ }
+}
+
+static void inverted_CMYK_to_BGR1_portable(uint32_t* dst, const void* vsrc, int count) {
+ const uint32_t* src = (const uint32_t*)vsrc;
+ for (int i = 0; i < count; i++) {
+ uint8_t k = src[i] >> 24,
+ y = src[i] >> 16,
+ m = src[i] >> 8,
+ c = src[i] >> 0;
+ uint8_t b = (y*k+127)/255,
+ g = (m*k+127)/255,
+ r = (c*k+127)/255;
+ dst[i] = (uint32_t)0xFF << 24
+ | (uint32_t) r << 16
+ | (uint32_t) g << 8
+ | (uint32_t) b << 0;
+ }
+}
+
#if defined(SK_ARM_HAS_NEON)
// Rounded divide by 255, (x + 127) / 255
@@ -401,6 +436,54 @@ static void grayA_to_rgbA(uint32_t dst[], const void* src, int count) {
expand_grayA<true>(dst, src, count);
}
+enum Format { kRGB1, kBGR1 };
+template <Format format>
+static void inverted_cmyk_to(uint32_t* dst, const void* vsrc, int count) {
+ auto src = (const uint32_t*)vsrc;
+ while (count >= 8) {
+ // Load 8 cmyk pixels.
+ uint8x8x4_t pixels = vld4_u8((const uint8_t*) src);
+
+ uint8x8_t k = pixels.val[3],
+ y = pixels.val[2],
+ m = pixels.val[1],
+ c = pixels.val[0];
+
+ // Scale to r, g, b.
+ uint8x8_t b = scale(y, k);
+ uint8x8_t g = scale(m, k);
+ uint8x8_t r = scale(c, k);
+
+ // Store 8 rgba pixels.
+ if (kBGR1 == format) {
+ pixels.val[3] = vdup_n_u8(0xFF);
+ pixels.val[2] = r;
+ pixels.val[1] = g;
+ pixels.val[0] = b;
+ } else {
+ pixels.val[3] = vdup_n_u8(0xFF);
+ pixels.val[2] = b;
+ pixels.val[1] = g;
+ pixels.val[0] = r;
+ }
+ vst4_u8((uint8_t*) dst, pixels);
+ src += 8;
+ dst += 8;
+ count -= 8;
+ }
+
+ auto proc = (kBGR1 == format) ? inverted_CMYK_to_BGR1_portable : inverted_CMYK_to_RGB1_portable;
+ proc(dst, src, count);
+}
+
+static void inverted_CMYK_to_RGB1(uint32_t dst[], const void* src, int count) {
+ inverted_cmyk_to<kRGB1>(dst, src, count);
+}
+
+static void inverted_CMYK_to_BGR1(uint32_t dst[], const void* src, int count) {
+ inverted_cmyk_to<kBGR1>(dst, src, count);
+}
+
#elif SK_CPU_SSE_LEVEL >= SK_CPU_SSE_LEVEL_SSSE3
// Scale a byte by another.
@@ -631,6 +714,83 @@ static void grayA_to_rgbA(uint32_t dst[], const void* vsrc, int count) {
grayA_to_rgbA_portable(dst, src, count);
}
+enum Format { kRGB1, kBGR1 };
+template <Format format>
+static void inverted_cmyk_to(uint32_t* dst, const void* vsrc, int count) {
+ auto src = (const uint32_t*)vsrc;
+
+ auto convert8 = [](__m128i* lo, __m128i* hi) {
+ const __m128i zeros = _mm_setzero_si128();
+ __m128i planar;
+ if (kBGR1 == format) {
+ planar = _mm_setr_epi8(2,6,10,14, 1,5,9,13, 0,4,8,12, 3,7,11,15);
+ } else {
+ planar = _mm_setr_epi8(0,4,8,12, 1,5,9,13, 2,6,10,14, 3,7,11,15);
+ }
+
+ // Swizzle the pixels to 8-bit planar.
+ *lo = _mm_shuffle_epi8(*lo, planar); // ccccmmmm yyyykkkk
+ *hi = _mm_shuffle_epi8(*hi, planar); // CCCCMMMM YYYYKKKK
+ __m128i cm = _mm_unpacklo_epi32(*lo, *hi), // ccccCCCC mmmmMMMM
+ yk = _mm_unpackhi_epi32(*lo, *hi); // yyyyYYYY kkkkKKKK
+
+ // Unpack to 16-bit planar.
+ __m128i c = _mm_unpacklo_epi8(cm, zeros), // c_c_c_c_ C_C_C_C_
+ m = _mm_unpackhi_epi8(cm, zeros), // m_m_m_m_ M_M_M_M_
+ y = _mm_unpacklo_epi8(yk, zeros), // y_y_y_y_ Y_Y_Y_Y_
+ k = _mm_unpackhi_epi8(yk, zeros); // k_k_k_k_ K_K_K_K_
+
+ // Scale to r, g, b.
+ __m128i r = scale(c, k),
+ g = scale(m, k),
+ b = scale(y, k);
+
+ // Repack into interlaced pixels.
+ __m128i rg = _mm_or_si128(r, _mm_slli_epi16(g, 8)), // rgrgrgrg RGRGRGRG
+ ba = _mm_or_si128(b, _mm_set1_epi16((uint16_t) 0xFF00)); // b1b1b1b1 B1B1B1B1
+ *lo = _mm_unpacklo_epi16(rg, ba); // rgbargba rgbargba
+ *hi = _mm_unpackhi_epi16(rg, ba); // RGB1RGB1 RGB1RGB1
+ };
+
+ while (count >= 8) {
+ __m128i lo = _mm_loadu_si128((const __m128i*) (src + 0)),
+ hi = _mm_loadu_si128((const __m128i*) (src + 4));
+
+ convert8(&lo, &hi);
+
+ _mm_storeu_si128((__m128i*) (dst + 0), lo);
+ _mm_storeu_si128((__m128i*) (dst + 4), hi);
+
+ src += 8;
+ dst += 8;
+ count -= 8;
+ }
+
+ if (count >= 4) {
+ __m128i lo = _mm_loadu_si128((const __m128i*) src),
+ hi = _mm_setzero_si128();
+
+ convert8(&lo, &hi);
+
+ _mm_storeu_si128((__m128i*) dst, lo);
+
+ src += 4;
+ dst += 4;
+ count -= 4;
+ }
+
+ auto proc = (kBGR1 == format) ? inverted_CMYK_to_BGR1_portable : inverted_CMYK_to_RGB1_portable;
+ proc(dst, src, count);
+}
+
+static void inverted_CMYK_to_RGB1(uint32_t dst[], const void* src, int count) {
+ inverted_cmyk_to<kRGB1>(dst, src, count);
+}
+
+static void inverted_CMYK_to_BGR1(uint32_t dst[], const void* src, int count) {
+ inverted_cmyk_to<kBGR1>(dst, src, count);
+}
+
#else
static void RGBA_to_rgbA(uint32_t* dst, const void* src, int count) {
@@ -665,6 +825,14 @@ static void grayA_to_rgbA(uint32_t dst[], const void* src, int count) {
grayA_to_rgbA_portable(dst, src, count);
}
+static void inverted_CMYK_to_RGB1(uint32_t dst[], const void* src, int count) {
+ inverted_CMYK_to_RGB1_portable(dst, src, count);
+}
+
+static void inverted_CMYK_to_BGR1(uint32_t dst[], const void* src, int count) {
+ inverted_CMYK_to_BGR1_portable(dst, src, count);
+}
+
#endif
}
generated by cgit v1.2.3 (git 2.39.1) at 2024-05-30 23:48:41 +0000