From 83aa9208dab96f1b6affa8167522e9a94afd70a5 Mon Sep 17 00:00:00 2001
From: fmalita <fmalita@chromium.org>
Date: Wed, 23 Mar 2016 12:28:14 -0700
Subject: [PATCH] Planar ramp() for S32 linear gradients

R=mtklein@google.com,reed@google.com
GOLD_TRYBOT_URL= https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&issue=1825333003

Review URL: https://codereview.chromium.org/1825333003
---
 src/effects/gradients/Sk4fGradientPriv.h     | 10 ++++----
 src/effects/gradients/Sk4fLinearGradient.cpp | 36 ++++++++++++++++++++++++++++
 2 files changed, 41 insertions(+), 5 deletions(-)

diff --git a/src/effects/gradients/Sk4fGradientPriv.h b/src/effects/gradients/Sk4fGradientPriv.h
index 4f71d3f..ae6fe7c 100644
--- a/src/effects/gradients/Sk4fGradientPriv.h
+++ b/src/effects/gradients/Sk4fGradientPriv.h
@@ -108,19 +108,19 @@ struct DstTraits<DstType::S32, premul> {
     using PM   = PremulTraits<premul>;
     using Type = SkPMColor;
 
-    // TODO: prescale by something like (255^2, 255^2, 255^2, 255) and use
-    //       linear_to_srgb to save a mult in store?
     static Sk4f load(const SkPM4f& c) {
-        return c.to4f_pmorder();
+        // Prescaling by (255^2, 255^2, 255^2, 255) on load, to avoid a 255 multiply on
+        // each store (S32 conversion yields a uniform 255 factor).
+        return c.to4f_pmorder() * Sk4f(255 * 255, 255 * 255, 255 * 255, 255);
     }
 
     static void store(const Sk4f& c, Type* dst) {
         // FIXME: this assumes opaque colors.  Handle unpremultiplication.
-        *dst = Sk4f_toS32(PM::apply(c));
+        *dst = to_4b(linear_to_srgb(PM::apply(c)));
     }
 
     static void store(const Sk4f& c, Type* dst, int n) {
-        sk_memset32(dst, Sk4f_toS32(PM::apply(c)), n);
+        sk_memset32(dst, to_4b(linear_to_srgb(PM::apply(c))), n);
     }
 
     static void store4x(const Sk4f& c0, const Sk4f& c1,
diff --git a/src/effects/gradients/Sk4fLinearGradient.cpp b/src/effects/gradients/Sk4fLinearGradient.cpp
index f4bb4a7..4ea025d 100644
--- a/src/effects/gradients/Sk4fLinearGradient.cpp
+++ b/src/effects/gradients/Sk4fLinearGradient.cpp
@@ -6,6 +6,7 @@
  */
 
 #include "Sk4fLinearGradient.h"
+#include "Sk4x4f.h"
 #include "SkXfermode.h"
 
 namespace {
@@ -42,6 +43,41 @@ void ramp(const Sk4f& c, const Sk4f& dc, typename DstTraits<dstType, premul>::Ty
     }
 }
 
+// Planar version of ramp (S32 no-premul only).
+template<>
+void ramp<DstType::S32, ApplyPremul::False>(const Sk4f& c, const Sk4f& dc, SkPMColor dst[], int n) {
+    SkASSERT(n > 0);
+
+    const Sk4f    dc4 = dc * 4;
+    const Sk4x4f dc4x = { Sk4f(dc4[0]), Sk4f(dc4[1]), Sk4f(dc4[2]), Sk4f(dc4[3]) };
+    Sk4x4f        c4x = Sk4x4f::Transpose(c, c + dc, c + dc * 2, c + dc * 3);
+
+    while (n >= 4) {
+        const Sk4x4f cx4s32 = { c4x.r.sqrt(), c4x.g.sqrt(), c4x.b.sqrt(), c4x.a };
+        cx4s32.transpose((uint8_t*)dst);
+
+        c4x.r += dc4x.r;
+        c4x.g += dc4x.g;
+        c4x.b += dc4x.b;
+        c4x.a += dc4x.a;
+
+        dst += 4;
+        n   -= 4;
+    }
+
+    if (n & 2) {
+        DstTraits<DstType::S32, ApplyPremul::False>
+            ::store(Sk4f(c4x.r[0], c4x.g[0], c4x.b[0], c4x.a[0]), dst++);
+        DstTraits<DstType::S32, ApplyPremul::False>
+            ::store(Sk4f(c4x.r[1], c4x.g[1], c4x.b[1], c4x.a[1]), dst++);
+    }
+
+    if (n & 1) {
+        DstTraits<DstType::S32, ApplyPremul::False>
+            ::store(Sk4f(c4x.r[n & 2], c4x.g[n & 2], c4x.b[n & 2], c4x.a[n & 2]), dst);
+    }
+}
+
 template<SkShader::TileMode>
 SkScalar pinFx(SkScalar);
 
-- 
2.7.4