try Herb's new to_srgb

This was 6-8% faster than the previous code on my Trashcan.

Change-Id: I70081009e233c83226d6d302f871fb7e86cdc438
Reviewed-on: https://skia-review.googlesource.com/16986
Reviewed-by: Matt Sarett <msarett@google.com>
Reviewed-by: Herb Derby <herb@google.com>
Commit-Queue: Mike Klein <mtklein@chromium.org>

bench/SkRasterPipelineBench.cpp

@@ -125,3 +125,21 @@ public:
     }
 };
 DEF_BENCH( return (new SkRasterPipeline_2dot2); )
+
+class SkRasterPipelineToSRGB : public Benchmark {
+public:
+    bool isSuitableFor(Backend backend) override { return backend == kNonRendering_Backend; }
+    const char* onGetName() override {
+        return "SkRasterPipeline_to_srgb";
+    }
+
+    void onDraw(int loops, SkCanvas*) override {
+        SkRasterPipeline p;
+        p.append(SkRasterPipeline::to_srgb);
+
+        while (loops --> 0) {
+            p.run(0,N);
+        }
+    }
+};
+DEF_BENCH( return (new SkRasterPipelineToSRGB); )

src/jumper/SkJumper_stages.cpp

@@ -621,12 +621,11 @@ STAGE(from_srgb) {
 }
 STAGE(to_srgb) {
     auto fn = [&](F l) {
-        F sqrt = rcp  (rsqrt(l)),
-          ftrt = rsqrt(rsqrt(l));
-        auto lo = l * 12.46f;
-        auto hi = min(1.0f, mad(0.411192f, ftrt,
-                            mad(0.689206f, sqrt, -0.0988f)));
-        return if_then_else(l < 0.0043f, lo, hi);
+        F t = rsqrt(l);
+        auto lo = l * 12.92f;
+        auto hi = mad(t, mad(t, -0.0024542345f, 0.013832027f), 1.1334244f)
+                * rcp(0.14513608f + t);
+        return if_then_else(l < 0.00465985f, lo, hi);
     };
     r = fn(r);
     g = fn(g);