Refactor of SkColorSpaceXformOpts

src/core/SkColorSpaceXformOpts.h

+/*
+ * Copyright 2016 Google Inc.
+ *
+ * Use of this source code is governed by a BSD-style license that can be
+ * found in the LICENSE file.
+ */
+
+#ifndef SkColorSpaceXformOpts_DEFINED
+#define SkColorSpaceXformOpts_DEFINED
+
+#include "SkNx.h"
+#include "SkColorPriv.h"
+#include "SkHalf.h"
+#include "SkSRGB.h"
+#include "SkTemplates.h"
+
+static inline void load_matrix(const float matrix[16],
+                               Sk4f& rXgXbX, Sk4f& rYgYbY, Sk4f& rZgZbZ, Sk4f& rTgTbT) {
+    rXgXbX = Sk4f::Load(matrix +  0);
+    rYgYbY = Sk4f::Load(matrix +  4);
+    rZgZbZ = Sk4f::Load(matrix +  8);
+    rTgTbT = Sk4f::Load(matrix + 12);
+}
+
+static inline void load_rgb_from_tables(const uint32_t* src,
+                                        Sk4f& r, Sk4f& g, Sk4f& b, Sk4f&,
+                                        const float* const srcTables[3]) {
+    r = { srcTables[0][(src[0] >>  0) & 0xFF],
+          srcTables[0][(src[1] >>  0) & 0xFF],
+          srcTables[0][(src[2] >>  0) & 0xFF],
+          srcTables[0][(src[3] >>  0) & 0xFF], };
+    g = { srcTables[1][(src[0] >>  8) & 0xFF],
+          srcTables[1][(src[1] >>  8) & 0xFF],
+          srcTables[1][(src[2] >>  8) & 0xFF],
+          srcTables[1][(src[3] >>  8) & 0xFF], };
+    b = { srcTables[2][(src[0] >> 16) & 0xFF],
+          srcTables[2][(src[1] >> 16) & 0xFF],
+          srcTables[2][(src[2] >> 16) & 0xFF],
+          srcTables[2][(src[3] >> 16) & 0xFF], };
+}
+
+static inline void load_rgba_from_tables(const uint32_t* src,
+                                         Sk4f& r, Sk4f& g, Sk4f& b, Sk4f& a,
+                                         const float* const srcTables[3]) {
+    r = { srcTables[0][(src[0] >>  0) & 0xFF],
+          srcTables[0][(src[1] >>  0) & 0xFF],
+          srcTables[0][(src[2] >>  0) & 0xFF],
+          srcTables[0][(src[3] >>  0) & 0xFF], };
+    g = { srcTables[1][(src[0] >>  8) & 0xFF],
+          srcTables[1][(src[1] >>  8) & 0xFF],
+          srcTables[1][(src[2] >>  8) & 0xFF],
+          srcTables[1][(src[3] >>  8) & 0xFF], };
+    b = { srcTables[2][(src[0] >> 16) & 0xFF],
+          srcTables[2][(src[1] >> 16) & 0xFF],
+          srcTables[2][(src[2] >> 16) & 0xFF],
+          srcTables[2][(src[3] >> 16) & 0xFF], };
+    a = (1.0f / 255.0f) * SkNx_cast<float>(Sk4u::Load(src) >> 24);
+}
+
+static inline void load_rgb_from_tables_1(const uint32_t* src,
+                                          Sk4f& r, Sk4f& g, Sk4f& b, Sk4f&,
+                                          const float* const srcTables[3]) {
+    // Splat r,g,b across a register each.
+    r = Sk4f(srcTables[0][(*src >>  0) & 0xFF]);
+    g = Sk4f(srcTables[1][(*src >>  8) & 0xFF]);
+    b = Sk4f(srcTables[2][(*src >> 16) & 0xFF]);
+}
+
+static inline void load_rgba_from_tables_1(const uint32_t* src,
+                                           Sk4f& r, Sk4f& g, Sk4f& b, Sk4f& a,
+                                           const float* const srcTables[3]) {
+    // Splat r,g,b across a register each.
+    r = Sk4f(srcTables[0][(*src >>  0) & 0xFF]);
+    g = Sk4f(srcTables[1][(*src >>  8) & 0xFF]);
+    b = Sk4f(srcTables[2][(*src >> 16) & 0xFF]);
+    a = (1.0f / 255.0f) * Sk4f(*src >> 24);
+}
+
+static inline void transform_gamut(const Sk4f& r, const Sk4f& g, const Sk4f& b, const Sk4f& a,
+                                   const Sk4f& rXgXbX, const Sk4f& rYgYbY, const Sk4f& rZgZbZ,
+                                   Sk4f& dr, Sk4f& dg, Sk4f& db, Sk4f& da) {
+    dr = rXgXbX[0]*r + rYgYbY[0]*g + rZgZbZ[0]*b;
+    dg = rXgXbX[1]*r + rYgYbY[1]*g + rZgZbZ[1]*b;
+    db = rXgXbX[2]*r + rYgYbY[2]*g + rZgZbZ[2]*b;
+    da = a;
+}
+
+static inline void transform_gamut_1(const Sk4f& r, const Sk4f& g, const Sk4f& b,
+                                     const Sk4f& rXgXbX, const Sk4f& rYgYbY, const Sk4f& rZgZbZ,
+                                     Sk4f& rgba) {
+    rgba = rXgXbX*r + rYgYbY*g + rZgZbZ*b;
+}
+
+static inline void translate_gamut(const Sk4f& rTgTbT, Sk4f& dr, Sk4f& dg, Sk4f& db) {
+    dr = dr + rTgTbT[0];
+    dg = dg + rTgTbT[1];
+    db = db + rTgTbT[2];
+}
+
+static inline void translate_gamut_1(const Sk4f& rTgTbT, Sk4f& rgba) {
+    rgba = rgba + rTgTbT;
+}
+
+static inline void premultiply(Sk4f& dr, Sk4f& dg, Sk4f& db, const Sk4f& da) {
+    dr = da * dr;
+    dg = da * dg;
+    db = da * db;
+}
+
+static inline void premultiply_1(const Sk4f& a, Sk4f& rgba) {
+    rgba = a * rgba;
+}
+
+static inline void store_srgb(void* dst, const uint32_t* src,
+                              Sk4f& dr, Sk4f& dg, Sk4f& db, Sk4f&,
+                              const uint8_t* const[3], bool kSwapRB) {
+    int kRShift = 0;
+    int kGShift = 8;
+    int kBShift = 16;
+    int kAShift = 24;
+    if (kSwapRB) {
+        kBShift = 0;
+        kRShift = 16;
+    }
+
+    dr = sk_linear_to_srgb_needs_trunc(dr);
+    dg = sk_linear_to_srgb_needs_trunc(dg);
+    db = sk_linear_to_srgb_needs_trunc(db);
+
+    dr = sk_clamp_0_255(dr);
+    dg = sk_clamp_0_255(dg);
+    db = sk_clamp_0_255(db);
+
+    Sk4i da = SkNx_cast<int32_t>(Sk4u::Load(src) >> 24);
+
+    Sk4i rgba = (SkNx_cast<int>(dr) << kRShift)
+              | (SkNx_cast<int>(dg) << kGShift)
+              | (SkNx_cast<int>(db) << kBShift)
+              | (da                 << kAShift);
+    rgba.store(dst);
+}
+
+static inline void store_srgb_1(void* dst, const uint32_t* src,
+                                Sk4f& rgba, const Sk4f&,
+                                const uint8_t* const[3], bool kSwapRB) {
+    rgba = sk_clamp_0_255(sk_linear_to_srgb_needs_trunc(rgba));
+
+    uint32_t tmp;
+    SkNx_cast<uint8_t>(SkNx_cast<int32_t>(rgba)).store(&tmp);
+    tmp = (*src & 0xFF000000) | (tmp & 0x00FFFFFF);
+    if (kSwapRB) {
+        tmp = SkSwizzle_RB(tmp);
+    }
+
+    *(uint32_t*)dst = tmp;
+}
+
+static inline Sk4f linear_to_2dot2(const Sk4f& x) {
+    // x^(29/64) is a very good approximation of the true value, x^(1/2.2).
+    auto x2  = x.rsqrt(),                            // x^(-1/2)
+         x32 = x2.rsqrt().rsqrt().rsqrt().rsqrt(),   // x^(-1/32)
+         x64 = x32.rsqrt();                          // x^(+1/64)
+
+    // 29 = 32 - 2 - 1
+    return 255.0f * x2.invert() * x32 * x64.invert();
+}
+
+static inline void store_2dot2(void* dst, const uint32_t* src,
+                               Sk4f& dr, Sk4f& dg, Sk4f& db, Sk4f&,
+                               const uint8_t* const[3], bool kSwapRB) {
+    int kRShift = 0;
+    int kGShift = 8;
+    int kBShift = 16;
+    int kAShift = 24;
+    if (kSwapRB) {
+        kBShift = 0;
+        kRShift = 16;
+    }
+
+    dr = linear_to_2dot2(dr);
+    dg = linear_to_2dot2(dg);
+    db = linear_to_2dot2(db);
+
+    dr = sk_clamp_0_255(dr);
+    dg = sk_clamp_0_255(dg);
+    db = sk_clamp_0_255(db);
+
+    Sk4i da = SkNx_cast<int32_t>(Sk4u::Load(src) >> 24);
+
+    Sk4i rgba = (Sk4f_round(dr) << kRShift)
+              | (Sk4f_round(dg) << kGShift)
+              | (Sk4f_round(db) << kBShift)
+              | (da             << kAShift);
+    rgba.store(dst);
+}
+
+static inline void store_2dot2_1(void* dst, const uint32_t* src,
+                                 Sk4f& rgba, const Sk4f&,
+                                 const uint8_t* const[3], bool kSwapRB) {
+    rgba = sk_clamp_0_255(linear_to_2dot2(rgba));
+
+    uint32_t tmp;
+    SkNx_cast<uint8_t>(Sk4f_round(rgba)).store(&tmp);
+    tmp = (*src & 0xFF000000) | (tmp & 0x00FFFFFF);
+    if (kSwapRB) {
+        tmp = SkSwizzle_RB(tmp);
+    }
+
+    *(uint32_t*)dst = tmp;
+}
+
+static inline void store_f16(void* dst, const uint32_t* src,
+                             Sk4f& dr, Sk4f& dg, Sk4f& db, Sk4f& da,
+                             const uint8_t* const[3], bool kSwapRB) {
+    Sk4h_store4(dst, SkFloatToHalf_finite(dr),
+                     SkFloatToHalf_finite(dg),
+                     SkFloatToHalf_finite(db),
+                     SkFloatToHalf_finite(da));
+    dst = SkTAddOffset<void>(dst, 4 * sizeof(uint64_t));
+}
+
+static inline void store_f16_1(void* dst, const uint32_t* src,
+                               Sk4f& rgba, const Sk4f& a,
+                               const uint8_t* const[3], bool kSwapRB) {
+    rgba = Sk4f(rgba[0], rgba[1], rgba[2], a[3]);
+    SkFloatToHalf_finite(rgba).store((uint64_t*) dst);
+}
+
+static inline void store_generic(void* dst, const uint32_t* src,
+                                 Sk4f& dr, Sk4f& dg, Sk4f& db, Sk4f&,
+                                 const uint8_t* const dstTables[3], bool kSwapRB) {
+    int kRShift = 0;
+    int kGShift = 8;
+    int kBShift = 16;
+    int kAShift = 24;
+    if (kSwapRB) {
+        kBShift = 0;
+        kRShift = 16;
+    }
+
+    dr = Sk4f::Min(Sk4f::Max(1023.0f * dr, 0.0f), 1023.0f);
+    dg = Sk4f::Min(Sk4f::Max(1023.0f * dg, 0.0f), 1023.0f);
+    db = Sk4f::Min(Sk4f::Max(1023.0f * db, 0.0f), 1023.0f);
+
+    Sk4i ir = Sk4f_round(dr);
+    Sk4i ig = Sk4f_round(dg);
+    Sk4i ib = Sk4f_round(db);
+
+    Sk4i da = SkNx_cast<int32_t>(Sk4u::Load(src) >> 24);
+
+    uint32_t* dst32 = (uint32_t*) dst;
+    dst32[0] = dstTables[0][ir[0]] << kRShift
+             | dstTables[1][ig[0]] << kGShift
+             | dstTables[2][ib[0]] << kBShift
+             | da[0]               << kAShift;
+    dst32[1] = dstTables[0][ir[1]] << kRShift
+             | dstTables[1][ig[1]] << kGShift
+             | dstTables[2][ib[1]] << kBShift
+             | da[1]               << kAShift;
+    dst32[2] = dstTables[0][ir[2]] << kRShift
+             | dstTables[1][ig[2]] << kGShift
+             | dstTables[2][ib[2]] << kBShift
+             | da[2]               << kAShift;
+    dst32[3] = dstTables[0][ir[3]] << kRShift
+             | dstTables[1][ig[3]] << kGShift
+             | dstTables[2][ib[3]] << kBShift
+             | da[3]               << kAShift;
+}
+
+static inline void store_generic_1(void* dst, const uint32_t* src,
+                                   Sk4f& rgba, const Sk4f&,
+                                   const uint8_t* const dstTables[3], bool kSwapRB) {
+    rgba = Sk4f::Min(Sk4f::Max(1023.0f * rgba, 0.0f), 1023.0f);
+
+    Sk4i indices = Sk4f_round(rgba);
+
+    *((uint32_t*) dst) = dstTables[0][indices[0]] <<  0
+                       | dstTables[1][indices[1]] <<  8
+                       | dstTables[2][indices[2]] << 16
+                       | (*src & 0xFF000000);
+}
+
+template <SkColorSpace::GammaNamed kDstGamma, bool kPremul, bool kSwapRB>
+static void color_xform_RGBA(void* dst, const uint32_t* src, int len,
+                             const float* const srcTables[3], const float matrix[16],
+                             const uint8_t* const dstTables[3]) {
+    decltype(store_srgb            )* store;
+    decltype(store_srgb_1          )* store_1;
+    decltype(load_rgb_from_tables  )* load;
+    decltype(load_rgb_from_tables_1)* load_1;
+    size_t sizeOfDstPixel;
+    switch (kDstGamma) {
+        case SkColorSpace::kSRGB_GammaNamed:
+            load    = kPremul ? load_rgba_from_tables   : load_rgb_from_tables;
+            load_1  = kPremul ? load_rgba_from_tables_1 : load_rgb_from_tables_1;
+            store   = store_srgb;
+            store_1 = store_srgb_1;
+            sizeOfDstPixel = 4;
+            break;
+        case SkColorSpace::k2Dot2Curve_GammaNamed:
+            load    = kPremul ? load_rgba_from_tables   : load_rgb_from_tables;
+            load_1  = kPremul ? load_rgba_from_tables_1 : load_rgb_from_tables_1;
+            store   = store_2dot2;
+            store_1 = store_2dot2_1;
+            sizeOfDstPixel = 4;
+            break;
+        case SkColorSpace::kLinear_GammaNamed:
+            load    = load_rgba_from_tables;
+            load_1  = load_rgba_from_tables_1;
+            store   = store_f16;
+            store_1 = store_f16_1;
+            sizeOfDstPixel = 8;
+            break;
+        case SkColorSpace::kNonStandard_GammaNamed:
+            load    = kPremul ? load_rgba_from_tables   : load_rgb_from_tables;
+            load_1  = kPremul ? load_rgba_from_tables_1 : load_rgb_from_tables_1;
+            store   = store_generic;
+            store_1 = store_generic_1;
+            sizeOfDstPixel = 4;
+            break;
+    }
+
+    Sk4f rXgXbX, rYgYbY, rZgZbZ, rTgTbT;
+    load_matrix(matrix, rXgXbX, rYgYbY, rZgZbZ, rTgTbT);
+
+    if (len >= 4) {
+        // Naively this would be a loop of load-transform-store, but we found it faster to
+        // move the N+1th load ahead of the Nth store.  We don't bother doing this for N<4.
+        Sk4f r, g, b, a;
+        load(src, r, g, b, a, srcTables);
+        src += 4;
+        len -= 4;
+
+        Sk4f dr, dg, db, da;
+        while (len >= 4) {
+            transform_gamut(r, g, b, a, rXgXbX, rYgYbY, rZgZbZ, dr, dg, db, da);
+            translate_gamut(rTgTbT, dr, dg, db);
+
+            if (kPremul) {
+                premultiply(dr, dg, db, da);
+            }
+
+            load(src, r, g, b, a, srcTables);
+            src += 4;
+            len -= 4;
+
+            store(dst, src - 4, dr, dg, db, da, dstTables, kSwapRB);
+            dst = SkTAddOffset<void>(dst, 4 * sizeOfDstPixel);
+        }
+
+        transform_gamut(r, g, b, a, rXgXbX, rYgYbY, rZgZbZ, dr, dg, db, da);
+        translate_gamut(rTgTbT, dr, dg, db);
+
+        if (kPremul) {
+            premultiply(dr, dg, db, da);
+        }
+
+        store(dst, src - 4, dr, dg, db, da, dstTables, kSwapRB);
+        dst = SkTAddOffset<void>(dst, 4 * sizeOfDstPixel);
+    }
+
+    while (len > 0) {
+        Sk4f r, g, b, a;
+        load_1(src, r, g, b, a, srcTables);
+
+        Sk4f rgba;
+        transform_gamut_1(r, g, b, rXgXbX, rYgYbY, rZgZbZ, rgba);
+
+        translate_gamut_1(rTgTbT, rgba);
+
+        store_1(dst, src, rgba, a, dstTables, kSwapRB);
+
+        src += 1;
+        len -= 1;
+        dst = SkTAddOffset<void>(dst, sizeOfDstPixel);
+    }
+}
+
+#endif // SkColorSpaceXformOpts_DEFINED