From a1c54bc7f7b2c882139e21dc6bd1c65ba0d66225 Mon Sep 17 00:00:00 2001
From: Ignacio <castano@gmail.com>
Date: Wed, 10 Oct 2018 14:04:13 -0700
Subject: [PATCH] Add fast sRGB conversion.

---
 src/nvtt/Surface.cpp | 83 ++++++++++++++++++++++++++++++++++----------
 src/nvtt/nvtt.h      |  2 ++
 2 files changed, 67 insertions(+), 18 deletions(-)

diff --git a/src/nvtt/Surface.cpp b/src/nvtt/Surface.cpp
index eac61ba..2d64dea 100644
--- a/src/nvtt/Surface.cpp
+++ b/src/nvtt/Surface.cpp
@@ -1499,24 +1499,47 @@ static float toSrgb(float f) {
     return f;
 }
 
-void Surface::toSrgb()
-{
+// sRGB approximation from: http://chilliant.blogspot.com/2012/08/srgb-approximations-for-hlsl.html
+static float toSrgbFast(float f) {
+    f = saturate(f);
+    float s1 = sqrtf(f);
+    float s2 = sqrtf(s1);
+    float s3 = sqrtf(s2);
+    return 0.662002687f * s1 + 0.684122060f * s2 - 0.323583601f * s3 - 0.0225411470f * f;
+}
+
+void Surface::toSrgb() {
     if (isNull()) return;
 
     detach();
 
     FloatImage * img = m->image;
 
-    const uint count = img->pixelCount();
-    for (uint c = 0; c < 3; c++) {
-        float * channel = img->channel(c);
-        for (uint i = 0; i < count; i++) {
-        //parallel_for(count, 128, [=](int i) {
-            channel[i] = ::toSrgb(channel[i]);
-        }//);
+    const uint count = 3 * img->pixelCount();
+    float * channel = img->channel(0);
+    for (uint i = 0; i < count; i++) {
+        channel[i] = ::toSrgb(channel[i]);
+    }
+}
+
+void Surface::toSrgbFast() {
+    if (isNull()) return;
+
+    detach();
+
+    FloatImage * img = m->image;
+
+    const uint count = 3 * img->pixelCount();
+    float * channel = img->channel(0);
+    for (uint i = 0; i < count; i++) {
+        channel[i] = ::toSrgbFast(channel[i]);
     }
+    //parallel_for(count, 128, [=](int i) {
+    //    channel[i] = toSrgbFast(channel[i]);
+    //});
 }
 
+
 static float fromSrgb(float f) {
     if (f < 0.0f)           f = 0.0f;
     else if (f < 0.04045f)  f = f / 12.92f;
@@ -1525,24 +1548,48 @@ static float fromSrgb(float f) {
     return f;
 }
 
-void Surface::toLinearFromSrgb()
-{
+// sRGB approximation from: http://chilliant.blogspot.com/2012/08/srgb-approximations-for-hlsl.html
+static float fromSrgbFast(float f) {
+    f = saturate(f);
+    return f * (f * (f * 0.305306011f + 0.682171111f) + 0.012522878f);
+}
+
+
+void Surface::toLinearFromSrgb() {
     if (isNull()) return;
 
     detach();
 
     FloatImage * img = m->image;
 
-    const uint count = img->pixelCount();
-    for (uint c = 0; c < 3; c++) {
-        float * channel = img->channel(c);
-        for (uint i = 0; i < count; i++) {
-        //parallel_for(count, 128, [=](int i) {
-            channel[i] = ::fromSrgb(channel[i]);
-        }//);
+    const uint count = 3  *img->pixelCount();
+    float * channel = img->channel(0);
+    for (uint i = 0; i < count; i++) {
+        channel[i] = ::fromSrgb(channel[i]);
     }
+    //parallel_for(count, 128, [=](int i) {
+    //    channel[i] = ::fromSrgb(channel[i]);
+    //});
 }
 
+void Surface::toLinearFromSrgbFast() {
+    if (isNull()) return;
+
+    detach();
+
+    FloatImage * img = m->image;
+
+    const uint count = 3 * img->pixelCount();
+    float * channel = img->channel(0);
+    for (uint i = 0; i < count; i++) {
+        channel[i] = ::fromSrgbFast(channel[i]);
+    }
+    //parallel_for(count, 128, [=](int i) {
+    //    channel[i] = ::fromSrgbFast(channel[i]);
+    //});
+}
+
+
 static float toXenonSrgb(float f) {
     if (f < 0)                  f = 0;
     else if (f < (1.0f/16.0f))  f = 4.0f * f;
diff --git a/src/nvtt/nvtt.h b/src/nvtt/nvtt.h
index d86a503..303a1b4 100644
--- a/src/nvtt/nvtt.h
+++ b/src/nvtt/nvtt.h
@@ -529,7 +529,9 @@ namespace nvtt
         NVTT_API void toLinear(int channel, float gamma);
         NVTT_API void toGamma(int channel, float gamma);
         NVTT_API void toSrgb();
+        NVTT_API void toSrgbFast();
         NVTT_API void toLinearFromSrgb();
+        NVTT_API void toLinearFromSrgbFast();
         NVTT_API void toXenonSrgb();
         NVTT_API void transform(const float w0[4], const float w1[4], const float w2[4], const float w3[4], const float offset[4]);
         NVTT_API void swizzle(int r, int g, int b, int a);