experimental support for parsing from a float16 mesh (or a float32-quantized-as-ushort mesh)

GL/draw.c

@@ -8,6 +8,31 @@
 #include "private.h"
 #include "platform.h"
 
+GLushort _quantize( GLfloat v ) {
+    union { GLfloat f; GLuint ui; } u = {v};
+    GLuint ui = u.ui;
+
+    int s = (ui >> 16) & 0x8000;
+    int em = ui & 0x7fffffff;
+
+    int h = (em - (112 << 23) + (1 << 12)) >> 13;
+    h = (em < (113 << 23)) ? 0 : h;
+    h = (em >= (143 << 23)) ? 0x7c00 : h;
+    h = (em > (255 << 23)) ? 0x7e00 : h;
+
+    return (GLushort)(s | h);
+}
+GLfloat _dequantize( GLushort h ) {
+    GLuint s = (GLuint) (h & 0x8000) << 16;
+    int em = h & 0x7fff;
+    int r = (em + (112 << 10)) << 13;
+    r = (em < (1 << 10)) ? 0 : r;
+    r += (em >= (31 << 10)) ? (112 << 23) : 0;
+
+    union { GLfloat f; GLuint ui; } u;
+    u.ui = s | r;
+    return u.f;
+}
 
 AttribPointerList ATTRIB_POINTERS;
 GLuint ENABLED_VERTEX_ATTRIBUTES = 0;
@@ -63,6 +88,7 @@ GL_FORCE_INLINE GLsizei byte_size(GLenum type) {
     case GL_INT: return sizeof(GLint);
     case GL_UNSIGNED_INT: return sizeof(GLuint);
     case GL_DOUBLE: return sizeof(GLdouble);
+    case GL_HALF_FLOAT: return sizeof(GLhalf);
     case GL_UNSIGNED_INT_2_10_10_10_REV: return sizeof(GLuint);
     case GL_FLOAT:
     default: return sizeof(GLfloat);
@@ -109,6 +135,15 @@ static void _readVertexData3us3f(const GLubyte* in, GLubyte* out) {
     output[2] = input[2];
 }
 
+static void _readVertexData3usq3f(const GLubyte* in, GLubyte* out) {
+    const GLushort* input = (const GLushort*) in;
+    float* output = (float*) out;
+
+    output[0] = _dequantize(input[0]);
+    output[1] = _dequantize(input[1]);
+    output[2] = _dequantize(input[2]);
+}
+
 static void _readVertexData3ui3f(const GLubyte* in, GLubyte* out) {
     const GLuint* input = (const GLuint*) in;
     float* output = (float*) out;
@@ -127,6 +162,15 @@ static void _readVertexData3ub3f(const GLubyte* input, GLubyte* out) {
     output[2] = input[2] * ONE_OVER_TWO_FIVE_FIVE;
 }
 
+static void _readVertexData3f16_3f(const GLubyte* in, GLubyte* out) {
+    const GLhalf* input = (const GLhalf*) in;
+    float* output = (float*) out;
+
+    output[0] = input[0];
+    output[1] = input[1];
+    output[2] = input[2];
+}
+
 static void _readVertexData2f2f(const GLubyte* in, GLubyte* out) {
     vec2cpy(out, in);
 }
@@ -164,6 +208,23 @@ static void _readVertexData2us2f(const GLubyte* in, GLubyte* out) {
     output[1] = (float)input[1] / SHRT_MAX;
 }
 
+static void _readVertexData2usq3f(const GLubyte* in, GLubyte* out) {
+    const GLushort* input = (const GLushort*) in;
+    float* output = (float*) out;
+
+    output[0] = _dequantize(input[0]);
+    output[1] = _dequantize(input[1]);
+    output[2] = 0.0f;
+}
+
+static void _readVertexData2usq2f(const GLubyte* in, GLubyte* out) {
+    const GLushort* input = (const GLushort*) in;
+    float* output = (float*) out;
+
+    output[0] = _dequantize(input[0]);
+    output[1] = _dequantize(input[1]);
+}
+
 static void _readVertexData2ui2f(const GLubyte* in, GLubyte* out) {
     const GLuint* input = (const GLuint*) in;
     float* output = (float*) out;
@@ -179,6 +240,14 @@ static void _readVertexData2ub2f(const GLubyte* input, GLubyte* out) {
     output[1] = input[1] * ONE_OVER_TWO_FIVE_FIVE;
 }
 
+static void _readVertexData2f16_2f(const GLubyte* in, GLubyte* out) {
+    const GLhalf* input = (const GLhalf*) in;
+    float* output = (float*) out;
+
+    output[0] = input[0];
+    output[1] = input[1];
+}
+
 static void _readVertexData2ui3f(const GLubyte* in, GLubyte* out) {
     const GLuint* input = (const GLuint*) in;
     float* output = (float*) out;
@@ -188,6 +257,15 @@ static void _readVertexData2ui3f(const GLubyte* in, GLubyte* out) {
     output[2] = 0.0f;
 }
 
+static void _readVertexData2f16_3f(const GLubyte* in, GLubyte* out) {
+    const GLhalf* input = (const GLhalf*) in;
+    float* output = (float*) out;
+
+    output[0] = input[0];
+    output[1] = input[1];
+    output[2] = 0.0f;
+}
+
 static void _readVertexData4ubARGB(const GLubyte* input, GLubyte* output) {
     output[R8IDX] = input[0];
     output[G8IDX] = input[1];
@@ -493,14 +571,17 @@ ReadPositionFunc calcReadPositionFunc() {
         case GL_FLOAT:
             return (ATTRIB_POINTERS.vertex.size == 3) ? _readVertexData3f3f:
                     _readVertexData2f3f;
+        case GL_HALF_FLOAT:
+            return (ATTRIB_POINTERS.vertex.size == 3) ? _readVertexData3f16_3f:
+                    _readVertexData2f16_3f;
         case GL_BYTE:
         case GL_UNSIGNED_BYTE:
             return (ATTRIB_POINTERS.vertex.size == 3) ? _readVertexData3ub3f:
                     _readVertexData2ub3f;
         case GL_SHORT:
         case GL_UNSIGNED_SHORT:
-            return (ATTRIB_POINTERS.vertex.size == 3) ? _readVertexData3us3f:
+            return (ATTRIB_POINTERS.vertex.size == 3) ? _readVertexData3usq3f:
-                    _readVertexData2us3f;
+                    _readVertexData2usq3f;
         case GL_INT:
         case GL_UNSIGNED_INT:
             return (ATTRIB_POINTERS.vertex.size == 3) ? _readVertexData3ui3f:
@@ -518,12 +599,14 @@ ReadUVFunc calcReadUVFunc() {
         case GL_DOUBLE:
         case GL_FLOAT:
             return _readVertexData2f2f;
+        case GL_HALF_FLOAT:
+            return _readVertexData2f16_2f;
         case GL_BYTE:
         case GL_UNSIGNED_BYTE:
             return _readVertexData2ub2f;
         case GL_SHORT:
         case GL_UNSIGNED_SHORT:
-            return _readVertexData2us2f;
+            return _readVertexData2usq2f;
         case GL_INT:
         case GL_UNSIGNED_INT:
             return _readVertexData2ui2f;
@@ -540,12 +623,14 @@ ReadUVFunc calcReadSTFunc() {
         case GL_DOUBLE:
         case GL_FLOAT:
             return _readVertexData2f2f;
+        case GL_HALF_FLOAT:
+            return _readVertexData2f16_2f;
         case GL_BYTE:
         case GL_UNSIGNED_BYTE:
             return _readVertexData2ub2f;
         case GL_SHORT:
         case GL_UNSIGNED_SHORT:
-            return _readVertexData2us2f;
+            return _readVertexData2usq2f;
         case GL_INT:
         case GL_UNSIGNED_INT:
             return _readVertexData2ui2f;
@@ -562,6 +647,8 @@ ReadNormalFunc calcReadNormalFunc() {
         case GL_DOUBLE:
         case GL_FLOAT:
             return _readVertexData3f3f;
+        case GL_HALF_FLOAT:
+            return _readVertexData3f16_3f;
         break;
         case GL_BYTE:
         case GL_UNSIGNED_BYTE:
@@ -569,7 +656,7 @@ ReadNormalFunc calcReadNormalFunc() {
         break;
         case GL_SHORT:
         case GL_UNSIGNED_SHORT:
-            return _readVertexData3us3f;
+            return _readVertexData3usq3f;
         break;
         case GL_INT:
         case GL_UNSIGNED_INT:

include/GL/gl.h

@@ -19,6 +19,10 @@ __BEGIN_DECLS
 
 #include <math.h>
 
+#if __STDCPP_FLOAT16_T__
+   #include <stdfloat>
+#endif
+
 /* Primitive Types taken from GL for compatability */
 /* Not all types are implemented in Open GL DC V.1.0 */
 #define GL_POINTS                               0x0000
@@ -305,6 +309,7 @@ __BEGIN_DECLS
 #define GL_UNSIGNED_INT                         0x1405
 #define GL_FLOAT                                0x1406
 #define GL_DOUBLE                               0x140A
+#define GL_HALF_FLOAT                           0x140B
 #define GL_2_BYTES                              0x1407
 #define GL_3_BYTES                              0x1408
 #define GL_4_BYTES                              0x1409
@@ -461,6 +466,12 @@ __BEGIN_DECLS
 #define GL_FALSE   0
 #define GL_TRUE    1
 
+#if __STDCPP_FLOAT16_T__
+#define GLhalf std::float16_t
+#else
+#define GLhalf unsigned short
+#endif
+
 /* Stubs for portability */
 #define GL_LINE_SMOOTH                    0x0B20
 #define GL_ALPHA_TEST                     0x0BC0