From 1cfb30faea60169c7b9cb3a08610247047d8a639 Mon Sep 17 00:00:00 2001
From: hyzboy <hyzboy@gmail.com>
Date: Thu, 2 Jan 2020 21:40:48 +0800
Subject: [PATCH] update Spheremap Transform algorithm

---
 SpheremapNormal.cpp | 30 +++++++++++++++++++-----------
 1 file changed, 19 insertions(+), 11 deletions(-)

diff --git a/SpheremapNormal.cpp b/SpheremapNormal.cpp
index c0d8914..ddcc795 100644
--- a/SpheremapNormal.cpp
+++ b/SpheremapNormal.cpp
@@ -6,7 +6,7 @@
 
     Normal to GBufffer:
 
-    G=normalize(N.xy)*sqrt(N.z*0.5+0.5)
+    G=normalize(N.xy)*sqrt(-N.z*0.5+0.5)
 
     GBuffer to Normal:
 
@@ -19,7 +19,7 @@ namespace hgl
     void normal_compress(uint8 *x,uint8 *y,const uint8 *rgb,const uint count)
     {
         Vector3f in;
-        Vector2f in_xy;
+        float f;
         Vector2f out;
 
         for(uint i=0;i<count;i++)
@@ -28,8 +28,11 @@ namespace hgl
             in.y=float(*rgb)/255.0f;++rgb;
             in.z=float(*rgb)/255.0f;++rgb;
 
-            in_xy=in.xy().Normalized();
-            out=in_xy*sqrt(in.z*0.5+0.5);
+//Spheremap Transform,PSNR 48.071 dB,15 GPU cycles
+            f=sqrt(8*in.z+8);
+            out=in.xy()/f+Vector2f(0.5);
+
+            out=out.Clamp01();
 
             *x=out.x*255;++x;
             *y=out.y*255;++y;
@@ -39,20 +42,25 @@ namespace hgl
     void normal_decompress(uint8 *rgb,const uint8 *x,const uint8 *y,const uint count)
     {
         Vector2f in;
-        Vector2f in_normal;
+        Vector2f fenc;
         Vector3f out;
-        float s;
+        float f,g;
 
         for(uint i=0;i<count;i++)
         {
             in.x=(*x)/255.0f;++x;
             in.y=(*y)/255.0f;++y;
+            
+//Spheremap Transform,PSNR 48.071 dB,15 GPU cycles
+            fenc=in*4.0f-Vector2f(2.0f);
+            f=dot(fenc,fenc);
+            g=sqrt(1.0-f/4.0f);
 
-            out.z=length(in)*2.0-1;
-            in_normal=in.Normalized();
-            s=sqrt(1-out.z*out.z);
-            out.x=in_normal.x*s;
-            out.y=in_normal.y*s;
+            out.x=fenc.x*g;
+            out.y=fenc.y*g;
+            out.z=1-f/2.0f;
+            
+            out=out.Clamp01();
 
             *rgb=out.x*255;++rgb;
             *rgb=out.y*255;++rgb;