Merge changes from The Witness.

src/nvmath/PackedFloat.cpp

@@ -0,0 +1,61 @@
+// This code is in the public domain -- Ignacio Casta<74>o <castano@gmail.com>
+
+#include "PackedFloat.h"
+#include "Vector.inl"
+#include "ftoi.h"
+
+using namespace nv;
+
+Vector3 nv::rgb9e5_to_vector3(FloatRGB9E5 v) {
+}
+
+FloatRGB9E5 nv::vector3_to_rgb9e5(const Vector3 & v) {
+}
+
+
+float nv::float11_to_float32(uint v) {
+}
+
+float nv::float10_to_float32(uint v) {
+}
+
+Vector3 nv::r11g11b10_to_vector3(FloatR11G11B10 v) {
+}
+
+FloatR11G11B10 nv::vector3_to_r11g11b10(const Vector3 & v) {
+}
+
+// These are based on: 
+// http://www.graphics.cornell.edu/~bjw/rgbe/rgbe.c
+// While this may not be the best way to encode/decode RGBE8, I'm not making any changes to maintain compatibility.
+FloatRGBE8 nv::vector3_to_rgbe8(const Vector3 & v) {
+
+    float m = max3(v.x, v.y, v.z);
+
+    FloatRGBE8 rgbe;
+
+    if (m < 1e-32) {
+        rgbe.v = 0;
+    }
+    else {
+        int e;
+        float scale = frexpf(m, &e) * 256.0f / m;
+        rgbe.r = U8(ftoi_round(v.x * scale));
+        rgbe.g = U8(ftoi_round(v.y * scale));
+        rgbe.b = U8(ftoi_round(v.z * scale));
+        rgbe.e = U8(e + 128);
+    }
+
+    return rgbe;
+}
+
+
+Vector3 nv::rgbe8_to_vector3(FloatRGBE8 v) {
+    if (v.e != 0) {
+        float scale = ldexpf(1.0f, v.e-(int)(128+8));             // +8 to divide by 256. @@ Shouldn't we divide by 255 instead?
+        return scale * Vector3(float(v.r), float(v.g), float(v.b));
+    }
+    
+    return Vector3(0);
+}
+

src/nvmath/PackedFloat.h

@@ -0,0 +1,79 @@
+// This code is in the public domain -- castano@gmail.com
+
+#pragma once
+#ifndef NV_MATH_PACKEDFLOAT_H
+#define NV_MATH_PACKEDFLOAT_H
+
+#include "nvmath.h"
+#include "Vector.h"
+
+namespace nv
+{
+
+    union FloatRGB9E5 {
+        uint32 v;
+        struct {
+        #if NV_BIG_ENDIAN
+            uint32 e : 5;
+            uint32 zm : 9;
+            uint32 ym : 9;
+            uint32 xm : 9;
+        #else
+            uint32 xm : 9;
+            uint32 ym : 9;
+            uint32 zm : 9;
+            uint32 e : 5;
+        #endif
+        };
+    };
+
+    union FloatR11G11B10 {
+        uint32 v;
+        struct {
+        #if NV_BIG_ENDIAN
+            uint32 ze : 5;
+            uint32 zm : 5;
+            uint32 ye : 5;
+            uint32 ym : 6;
+            uint32 xe : 5;
+            uint32 xm : 6;
+        #else
+            uint32 xm : 6;
+            uint32 xe : 5;
+            uint32 ym : 6;
+            uint32 ye : 5;
+            uint32 zm : 5;
+            uint32 ze : 5;
+        #endif
+        };
+    };
+
+    union FloatRGBE8 {
+        uint32 v;
+        struct {
+        #if NV_LITTLE_ENDIAN
+            uint8 r, g, b, e;
+        #else
+            uint8 e: 8;
+            uint8 b: 8;
+            uint8 g: 8;
+            uint8 r: 8;
+        #endif
+        };
+    };
+
+    NVMATH_API Vector3 rgb9e5_to_vector3(FloatRGB9E5 v);
+    NVMATH_API FloatRGB9E5 vector3_to_rgb9e5(const Vector3 & v);
+
+    NVMATH_API float float11_to_float32(uint v);
+    NVMATH_API float float10_to_float32(uint v);
+
+    NVMATH_API Vector3 r11g11b10_to_vector3(FloatR11G11B10 v);
+    NVMATH_API FloatR11G11B10 vector3_to_r11g11b10(const Vector3 & v);
+
+    NVMATH_API Vector3 rgbe8_to_vector3(FloatRGBE8 v);
+    NVMATH_API FloatRGBE8 vector3_to_rgbe8(const Vector3 & v);
+
+} // nv
+
+#endif // NV_MATH_PACKEDFLOAT_H

src/nvmath/SimdVector.h

@@ -2,29 +2,6 @@
 
 #include "Vector.h" // Vector3, Vector4
 
-// Set some reasonable defaults.
-#ifndef NV_USE_ALTIVEC
-#   define NV_USE_ALTIVEC NV_CPU_PPC
-//#   define NV_USE_ALTIVEC defined(__VEC__)
-#endif
-
-#ifndef NV_USE_SSE
-#   if NV_CPU_X86 || NV_CPU_X86_64
-#       define NV_USE_SSE 2
-#   endif
-#   if defined(__SSE2__)
-#       define NV_USE_SSE 2
-#   elif defined(__SSE__)
-#       define NV_USE_SSE 1
-#   else
-#       define NV_USE_SSE 0
-#   endif
-#endif
-
-// Internally set NV_USE_SIMD when either altivec or sse is available.
-#if NV_USE_ALTIVEC && NV_USE_SSE
-#	error "Cannot enable both altivec and sse!"
-#endif
 
 #if NV_USE_ALTIVEC
 #   include "SimdVector_VE.h"

src/nvmath/nvmath.h

@@ -283,6 +283,49 @@ namespace nv
         f.value = x;
         return (f.field.biasedexponent - 127);
     }
+
+
+    // FloatRGB9E5
+    union Float3SE {
+        uint32 v;
+        struct {
+        #if NV_BIG_ENDIAN
+            uint32 e : 5;
+            uint32 zm : 9;
+            uint32 ym : 9;
+            uint32 xm : 9;
+        #else
+            uint32 xm : 9;
+            uint32 ym : 9;
+            uint32 zm : 9;
+            uint32 e : 5;
+        #endif
+        };
+    };
+
+    // FloatR11G11B10
+    union Float3PK {
+        uint32 v;
+        struct {
+        #if NV_BIG_ENDIAN
+            uint32 ze : 5;
+            uint32 zm : 5;
+            uint32 ye : 5;
+            uint32 ym : 6;
+            uint32 xe : 5;
+            uint32 xm : 6;
+        #else
+            uint32 xm : 6;
+            uint32 xe : 5;
+            uint32 ym : 6;
+            uint32 ye : 5;
+            uint32 zm : 5;
+            uint32 ze : 5;
+        #endif
+        };
+    };
+
+
 } // nv
 
 #endif // NV_MATH_H