Support sRGB dsts in opt code

201295.jpg on HP z620 (300x280)

QCMS Xform                    0.418 ms
Skia NEW Xform                0.378 ms

Vs QCMS                       1.11x

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2078623002
CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot

Review-Url: https://codereview.chromium.org/2078623002

7 files changed, 101 insertions, 35 deletions

diff --git a/src/core/SkColorSpaceXform.cpp b/src/core/SkColorSpaceXform.cpp
index 4c67e8d3f0..f42811a549 100644
--- a/src/core/SkColorSpaceXform.cpp
+++ b/src/core/SkColorSpaceXform.cpp
@@ -37,15 +37,35 @@ std::unique_ptr<SkColorSpaceXform> SkColorSpaceXform::New(const sk_sp<SkColorSpa
         return nullptr;
     }
 
-    if (SkColorSpace::k2Dot2Curve_GammaNamed == dstSpace->gammaNamed() &&
-        0.0f == srcToDst.getFloat(3, 0) &&
+    if (0.0f == srcToDst.getFloat(3, 0) &&
         0.0f == srcToDst.getFloat(3, 1) &&
         0.0f == srcToDst.getFloat(3, 2))
     {
-        if (SkColorSpace::kSRGB_GammaNamed == srcSpace->gammaNamed()) {
-            return std::unique_ptr<SkColorSpaceXform>(new SkSRGBTo2Dot2Xform(srcToDst));
-        } else if (SkColorSpace::k2Dot2Curve_GammaNamed == srcSpace->gammaNamed()) {
-            return std::unique_ptr<SkColorSpaceXform>(new Sk2Dot2To2Dot2Xform(srcToDst));
+        switch (srcSpace->gammaNamed()) {
+            case SkColorSpace::kSRGB_GammaNamed:
+                if (SkColorSpace::kSRGB_GammaNamed == dstSpace->gammaNamed()) {
+                    return std::unique_ptr<SkColorSpaceXform>(
+                            new SkFastXform<SkColorSpace::kSRGB_GammaNamed,
+                                            SkColorSpace::kSRGB_GammaNamed>(srcToDst));
+                } else if (SkColorSpace::k2Dot2Curve_GammaNamed == dstSpace->gammaNamed()) {
+                    return std::unique_ptr<SkColorSpaceXform>(
+                            new SkFastXform<SkColorSpace::kSRGB_GammaNamed,
+                                            SkColorSpace::k2Dot2Curve_GammaNamed>(srcToDst));
+                }
+                break;
+            case SkColorSpace::k2Dot2Curve_GammaNamed:
+                if (SkColorSpace::kSRGB_GammaNamed == dstSpace->gammaNamed()) {
+                    return std::unique_ptr<SkColorSpaceXform>(
+                            new SkFastXform<SkColorSpace::k2Dot2Curve_GammaNamed,
+                                            SkColorSpace::kSRGB_GammaNamed>(srcToDst));
+                } else if (SkColorSpace::k2Dot2Curve_GammaNamed == dstSpace->gammaNamed()) {
+                    return std::unique_ptr<SkColorSpaceXform>(
+                            new SkFastXform<SkColorSpace::k2Dot2Curve_GammaNamed,
+                                            SkColorSpace::k2Dot2Curve_GammaNamed>(srcToDst));
+                }
+                break;
+            default:
+                break;
         }
     }
 
@@ -90,23 +110,37 @@ static void build_src_to_dst(float srcToDstArray[12], const SkMatrix44& srcToDst
 #endif
 }
 
-SkSRGBTo2Dot2Xform::SkSRGBTo2Dot2Xform(const SkMatrix44& srcToDst)
+template <SkColorSpace::GammaNamed Src, SkColorSpace::GammaNamed Dst>
+SkFastXform<Src, Dst>::SkFastXform(const SkMatrix44& srcToDst)
 {
     build_src_to_dst(fSrcToDst, srcToDst);
 }
 
-void SkSRGBTo2Dot2Xform::xform_RGB1_8888(uint32_t* dst, const uint32_t* src, uint32_t len) const {
-    SkOpts::color_xform_RGB1_srgb_to_2dot2(dst, src, len, fSrcToDst);
+template <>
+void SkFastXform<SkColorSpace::kSRGB_GammaNamed, SkColorSpace::kSRGB_GammaNamed>
+::xform_RGB1_8888(uint32_t* dst, const uint32_t* src, uint32_t len) const
+{
+    SkOpts::color_xform_RGB1_srgb_to_srgb(dst, src, len, fSrcToDst);
 }
 
-///////////////////////////////////////////////////////////////////////////////////////////////////
+template <>
+void SkFastXform<SkColorSpace::kSRGB_GammaNamed, SkColorSpace::k2Dot2Curve_GammaNamed>
+::xform_RGB1_8888(uint32_t* dst, const uint32_t* src, uint32_t len) const
+{
+    SkOpts::color_xform_RGB1_srgb_to_2dot2(dst, src, len, fSrcToDst);
+}
 
-Sk2Dot2To2Dot2Xform::Sk2Dot2To2Dot2Xform(const SkMatrix44& srcToDst)
+template <>
+void SkFastXform<SkColorSpace::k2Dot2Curve_GammaNamed, SkColorSpace::kSRGB_GammaNamed>
+::xform_RGB1_8888(uint32_t* dst, const uint32_t* src, uint32_t len) const
 {
-    build_src_to_dst(fSrcToDst, srcToDst);
+    SkOpts::color_xform_RGB1_2dot2_to_srgb(dst, src, len, fSrcToDst);
 }
 
-void Sk2Dot2To2Dot2Xform::xform_RGB1_8888(uint32_t* dst, const uint32_t* src, uint32_t len) const {
+template <>
+void SkFastXform<SkColorSpace::k2Dot2Curve_GammaNamed, SkColorSpace::k2Dot2Curve_GammaNamed>
+::xform_RGB1_8888(uint32_t* dst, const uint32_t* src, uint32_t len) const
+{
     SkOpts::color_xform_RGB1_2dot2_to_2dot2(dst, src, len, fSrcToDst);
 }
 
diff --git a/src/core/SkColorSpaceXform.h b/src/core/SkColorSpaceXform.h
index 1ea608094f..e6abdbb992 100644
--- a/src/core/SkColorSpaceXform.h
+++ b/src/core/SkColorSpaceXform.h
@@ -34,26 +34,14 @@ public:
     virtual ~SkColorSpaceXform() {}
 };
 
-class SkSRGBTo2Dot2Xform : public SkColorSpaceXform {
+template <SkColorSpace::GammaNamed Src, SkColorSpace::GammaNamed Dst>
+class SkFastXform : public SkColorSpaceXform {
 public:
 
     void xform_RGB1_8888(uint32_t* dst, const uint32_t* src, uint32_t len) const override;
 
 private:
-    SkSRGBTo2Dot2Xform(const SkMatrix44& srcToDst);
-
-    float fSrcToDst[12];
-
-    friend class SkColorSpaceXform;
-};
-
-class Sk2Dot2To2Dot2Xform : public SkColorSpaceXform {
-public:
-
-    void xform_RGB1_8888(uint32_t* dst, const uint32_t* src, uint32_t len) const override;
-
-private:
-    Sk2Dot2To2Dot2Xform(const SkMatrix44& srcToDst);
+    SkFastXform(const SkMatrix44& srcToDst);
 
     float fSrcToDst[12];
 
diff --git a/src/core/SkOpts.cpp b/src/core/SkOpts.cpp
index b4145acb7a..a2de1f59d2 100644
--- a/src/core/SkOpts.cpp
+++ b/src/core/SkOpts.cpp
@@ -81,6 +81,10 @@ namespace SkOpts {
             sk_default::color_xform_RGB1_srgb_to_2dot2;
     decltype(color_xform_RGB1_2dot2_to_2dot2) color_xform_RGB1_2dot2_to_2dot2 =
             sk_default::color_xform_RGB1_2dot2_to_2dot2;
+    decltype(color_xform_RGB1_srgb_to_srgb)   color_xform_RGB1_srgb_to_srgb   =
+            sk_default::color_xform_RGB1_srgb_to_srgb;
+    decltype(color_xform_RGB1_2dot2_to_srgb)  color_xform_RGB1_2dot2_to_srgb  =
+            sk_default::color_xform_RGB1_2dot2_to_srgb;
 
     // 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 1c33529d52..186fa73d4c 100644
--- a/src/core/SkOpts.h
+++ b/src/core/SkOpts.h
@@ -74,6 +74,10 @@ namespace SkOpts {
                                                    const float srcToDstMatrix[16]);
     extern void (*color_xform_RGB1_2dot2_to_2dot2)(uint32_t* dst, const uint32_t* src, int len,
                                                    const float srcToDstMatrix[16]);
+    extern void (*color_xform_RGB1_srgb_to_srgb) (uint32_t* dst, const uint32_t* src, int len,
+                                                  const float srcToDstMatrix[16]);
+    extern void (*color_xform_RGB1_2dot2_to_srgb)(uint32_t* dst, const uint32_t* src, int len,
+                                                  const float srcToDstMatrix[16]);
 }
 
 #endif//SkOpts_DEFINED
diff --git a/src/opts/SkColorXform_opts.h b/src/opts/SkColorXform_opts.h
index 5af5c2aed7..2c14c80283 100644
--- a/src/opts/SkColorXform_opts.h
+++ b/src/opts/SkColorXform_opts.h
@@ -157,13 +157,33 @@ static Sk4f linear_to_2dot2(const Sk4f& x) {
     return 255.0f * x2.invert() * x32 * x64.invert();
 }
 
+static Sk4f linear_to_srgb(const Sk4f& x) {
+    // Approximation of the sRGB gamma curve (within 1 when scaled to 8-bit pixels).
+    // For 0.00000f <= x <  0.00349f,    12.92 * x
+    // For 0.00349f <= x <= 1.00000f,    0.679*(x.^0.5) + 0.423*x.^(0.25) - 0.101
+    // Note that 0.00349 was selected because it is a point where both functions produce the
+    // same pixel value when rounded.
+    auto rsqrt = x.rsqrt(),
+         sqrt  = rsqrt.invert(),
+         ftrt  = rsqrt.rsqrt();
+
+    auto hi = (-0.101115084998961f * 255.0f) +
+              (+0.678513029959381f * 255.0f) * sqrt +
+              (+0.422602055039580f * 255.0f) * ftrt;
+
+    auto lo = (12.92f * 255.0f) * x;
+
+    auto mask = (x < 0.00349f);
+    return mask.thenElse(lo, hi);
+}
+
 static Sk4f clamp_0_to_255(const Sk4f& x) {
     // The order of the arguments is important here.  We want to make sure that NaN
     // clamps to zero.  Note that max(NaN, 0) = 0, while max(0, NaN) = NaN.
     return Sk4f::Min(Sk4f::Max(x, 0.0f), 255.0f);
 }
 
-template <const float (&linear_from_curve)[256]>
+template <const float (&linear_from_curve)[256], Sk4f (*linear_to_curve)(const Sk4f&)>
 static void color_xform_RGB1(uint32_t* dst, const uint32_t* src, int len,
                              const float matrix[16]) {
     // Load transformation matrix.
@@ -192,9 +212,9 @@ static void color_xform_RGB1(uint32_t* dst, const uint32_t* src, int len,
              dstBlues  = rXgXbX[2]*reds + rYgYbY[2]*greens + rZgZbZ[2]*blues;
 
         // Convert to dst gamma.
-        dstReds   = linear_to_2dot2(dstReds);
-        dstGreens = linear_to_2dot2(dstGreens);
-        dstBlues  = linear_to_2dot2(dstBlues);
+        dstReds   = linear_to_curve(dstReds);
+        dstGreens = linear_to_curve(dstGreens);
+        dstBlues  = linear_to_curve(dstBlues);
 
         // Clamp floats to byte range.
         dstReds   = clamp_0_to_255(dstReds);
@@ -223,7 +243,7 @@ static void color_xform_RGB1(uint32_t* dst, const uint32_t* src, int len,
         auto dstPixel = rXgXbX*r + rYgYbY*g + rZgZbZ*b;
 
         // Convert to dst gamma.
-        dstPixel = linear_to_2dot2(dstPixel);
+        dstPixel = linear_to_curve(dstPixel);
 
         // Clamp floats to byte range.
         dstPixel = clamp_0_to_255(dstPixel);
@@ -242,12 +262,22 @@ static void color_xform_RGB1(uint32_t* dst, const uint32_t* src, int len,
 
 static void color_xform_RGB1_srgb_to_2dot2(uint32_t* dst, const uint32_t* src, int len,
                                            const float matrix[16]) {
-    color_xform_RGB1<linear_from_srgb>(dst, src, len, matrix);
+    color_xform_RGB1<linear_from_srgb, linear_to_2dot2>(dst, src, len, matrix);
 }
 
 static void color_xform_RGB1_2dot2_to_2dot2(uint32_t* dst, const uint32_t* src, int len,
                                            const float matrix[16]) {
-    color_xform_RGB1<linear_from_2dot2>(dst, src, len, matrix);
+    color_xform_RGB1<linear_from_2dot2, linear_to_2dot2>(dst, src, len, matrix);
+}
+
+static void color_xform_RGB1_srgb_to_srgb(uint32_t* dst, const uint32_t* src, int len,
+                                           const float matrix[16]) {
+    color_xform_RGB1<linear_from_srgb, linear_to_srgb>(dst, src, len, matrix);
+}
+
+static void color_xform_RGB1_2dot2_to_srgb(uint32_t* dst, const uint32_t* src, int len,
+                                           const float matrix[16]) {
+    color_xform_RGB1<linear_from_2dot2, linear_to_srgb>(dst, src, len, matrix);
 }
 
 }  // namespace SK_OPTS_NS
diff --git a/src/opts/SkNx_sse.h b/src/opts/SkNx_sse.h
index bdc6d77ce7..9751c4db91 100644
--- a/src/opts/SkNx_sse.h
+++ b/src/opts/SkNx_sse.h
@@ -124,8 +124,12 @@ public:
     bool anyTrue() const { return 0x0000 != _mm_movemask_epi8(_mm_castps_si128(fVec)); }
 
     SkNx thenElse(const SkNx& t, const SkNx& e) const {
+#if SK_CPU_SSE_LEVEL >= SK_CPU_SSE_LEVEL_SSE41
+        return _mm_blendv_ps(e.fVec, t.fVec, fVec);
+#else
         return _mm_or_ps(_mm_and_ps   (fVec, t.fVec),
                          _mm_andnot_ps(fVec, e.fVec));
+#endif
     }
 
     __m128 fVec;
diff --git a/src/opts/SkOpts_sse41.cpp b/src/opts/SkOpts_sse41.cpp
index 416157125f..4489242de7 100644
--- a/src/opts/SkOpts_sse41.cpp
+++ b/src/opts/SkOpts_sse41.cpp
@@ -23,5 +23,7 @@ namespace SkOpts {
 
         color_xform_RGB1_srgb_to_2dot2  = sk_sse41::color_xform_RGB1_srgb_to_2dot2;
         color_xform_RGB1_2dot2_to_2dot2 = sk_sse41::color_xform_RGB1_2dot2_to_2dot2;
+        color_xform_RGB1_srgb_to_srgb   = sk_sse41::color_xform_RGB1_srgb_to_srgb;
+        color_xform_RGB1_2dot2_to_srgb  = sk_sse41::color_xform_RGB1_2dot2_to_srgb;
     }
 }