Some groundwork for supporting GLhalf datatype.

3 files changed, 311 insertions, 3 deletions

diff --git a/src/mesa/main/image.c b/src/mesa/main/image.c
index b9f026d7e6..955f14f6b1 100644
--- a/src/mesa/main/image.c
+++ b/src/mesa/main/image.c
@@ -161,6 +161,8 @@ GLint _mesa_sizeof_type( GLenum type )
 	 return sizeof(GLint);
       case GL_FLOAT:
 	 return sizeof(GLfloat);
+      case GL_HALF_FLOAT_NV:
+	 return sizeof(GLhalfNV);
       default:
          return -1;
    }
@@ -188,6 +190,8 @@ GLint _mesa_sizeof_packed_type( GLenum type )
 	 return sizeof(GLuint);
       case GL_INT:
 	 return sizeof(GLint);
+      case GL_HALF_FLOAT_NV:
+	 return sizeof(GLhalfNV);
       case GL_FLOAT:
 	 return sizeof(GLfloat);
       case GL_UNSIGNED_BYTE_3_3_2:
@@ -297,6 +301,8 @@ GLint _mesa_bytes_per_pixel( GLenum format, GLenum type )
          return comps * sizeof(GLint);
       case GL_FLOAT:
          return comps * sizeof(GLfloat);
+      case GL_HALF_FLOAT_NV:
+         return comps * sizeof(GLhalfNV);
       case GL_UNSIGNED_BYTE_3_3_2:
       case GL_UNSIGNED_BYTE_2_3_3_REV:
          if (format == GL_RGB || format == GL_BGR)
@@ -361,6 +367,7 @@ _mesa_is_legal_format_and_type( GLenum format, GLenum type )
             case GL_INT:
             case GL_UNSIGNED_INT:
             case GL_FLOAT:
+            case GL_HALF_FLOAT_NV:
                return GL_TRUE;
             default:
                return GL_FALSE;
@@ -381,6 +388,7 @@ _mesa_is_legal_format_and_type( GLenum format, GLenum type )
             case GL_INT:
             case GL_UNSIGNED_INT:
             case GL_FLOAT:
+            case GL_HALF_FLOAT_NV:
                return GL_TRUE;
             default:
                return GL_FALSE;
@@ -395,6 +403,7 @@ _mesa_is_legal_format_and_type( GLenum format, GLenum type )
             case GL_INT:
             case GL_UNSIGNED_INT:
             case GL_FLOAT:
+            case GL_HALF_FLOAT_NV:
             case GL_UNSIGNED_BYTE_3_3_2:
             case GL_UNSIGNED_BYTE_2_3_3_REV:
             case GL_UNSIGNED_SHORT_5_6_5:
@@ -414,6 +423,7 @@ _mesa_is_legal_format_and_type( GLenum format, GLenum type )
             case GL_INT:
             case GL_UNSIGNED_INT:
             case GL_FLOAT:
+            case GL_HALF_FLOAT_NV:
             case GL_UNSIGNED_SHORT_4_4_4_4:
             case GL_UNSIGNED_SHORT_4_4_4_4_REV:
             case GL_UNSIGNED_SHORT_5_5_5_1:
@@ -1538,6 +1548,82 @@ _mesa_pack_float_rgba_span( GLcontext *ctx,
             }
          }
          break;
+      case GL_HALF_FLOAT_NV:
+         {
+            GLhalfNV *dst = (GLhalfNV *) dstAddr;
+            switch (dstFormat) {
+               case GL_RED:
+                  for (i=0;i<n;i++)
+                     dst[i] = _mesa_float_to_half(rgba[i][RCOMP]);
+                  break;
+               case GL_GREEN:
+                  for (i=0;i<n;i++)
+                     dst[i] = _mesa_float_to_half(rgba[i][GCOMP]);
+                  break;
+               case GL_BLUE:
+                  for (i=0;i<n;i++)
+                     dst[i] = _mesa_float_to_half(rgba[i][BCOMP]);
+                  break;
+               case GL_ALPHA:
+                  for (i=0;i<n;i++)
+                     dst[i] = _mesa_float_to_half(rgba[i][ACOMP]);
+                  break;
+               case GL_LUMINANCE:
+                  for (i=0;i<n;i++)
+                     dst[i] = _mesa_float_to_half(luminance[i]);
+                  break;
+               case GL_LUMINANCE_ALPHA:
+                  for (i=0;i<n;i++) {
+                     dst[i*2+0] = _mesa_float_to_half(luminance[i]);
+                     dst[i*2+1] = _mesa_float_to_half(rgba[i][ACOMP]);
+                  }
+                  break;
+               case GL_RGB:
+                  for (i=0;i<n;i++) {
+                     dst[i*3+0] = _mesa_float_to_half(rgba[i][RCOMP]);
+                     dst[i*3+1] = _mesa_float_to_half(rgba[i][GCOMP]);
+                     dst[i*3+2] = _mesa_float_to_half(rgba[i][BCOMP]);
+                  }
+                  break;
+               case GL_RGBA:
+                  for (i=0;i<n;i++) {
+                     dst[i*4+0] = _mesa_float_to_half(rgba[i][RCOMP]);
+                     dst[i*4+1] = _mesa_float_to_half(rgba[i][GCOMP]);
+                     dst[i*4+2] = _mesa_float_to_half(rgba[i][BCOMP]);
+                     dst[i*4+3] = _mesa_float_to_half(rgba[i][ACOMP]);
+                  }
+                  break;
+               case GL_BGR:
+                  for (i=0;i<n;i++) {
+                     dst[i*3+0] = _mesa_float_to_half(rgba[i][BCOMP]);
+                     dst[i*3+1] = _mesa_float_to_half(rgba[i][GCOMP]);
+                     dst[i*3+2] = _mesa_float_to_half(rgba[i][RCOMP]);
+                  }
+                  break;
+               case GL_BGRA:
+                  for (i=0;i<n;i++) {
+                     dst[i*4+0] = _mesa_float_to_half(rgba[i][BCOMP]);
+                     dst[i*4+1] = _mesa_float_to_half(rgba[i][GCOMP]);
+                     dst[i*4+2] = _mesa_float_to_half(rgba[i][RCOMP]);
+                     dst[i*4+3] = _mesa_float_to_half(rgba[i][ACOMP]);
+                  }
+                  break;
+               case GL_ABGR_EXT:
+                  for (i=0;i<n;i++) {
+                     dst[i*4+0] = _mesa_float_to_half(rgba[i][ACOMP]);
+                     dst[i*4+1] = _mesa_float_to_half(rgba[i][BCOMP]);
+                     dst[i*4+2] = _mesa_float_to_half(rgba[i][GCOMP]);
+                     dst[i*4+3] = _mesa_float_to_half(rgba[i][RCOMP]);
+                  }
+                  break;
+               default:
+                  _mesa_problem(ctx, "bad format in _mesa_pack_rgba_span\n");
+            }
+            if (dstPacking->SwapBytes) {
+               _mesa_swap2( (GLushort *) dst, n * comps );
+            }
+         }
+         break;
       case GL_UNSIGNED_BYTE_3_3_2:
          if (dstFormat == GL_RGB) {
             GLubyte *dst = (GLubyte *) dstAddr;
@@ -1925,6 +2011,7 @@ extract_uint_indexes(GLuint n, GLuint indexes[],
           srcType == GL_SHORT ||
           srcType == GL_UNSIGNED_INT ||
           srcType == GL_INT ||
+          srcType == GL_HALF_FLOAT_NV ||
           srcType == GL_FLOAT);
 
    switch (srcType) {
@@ -2062,6 +2149,23 @@ extract_uint_indexes(GLuint n, GLuint indexes[],
             }
          }
          break;
+      case GL_HALF_FLOAT_NV:
+         {
+            GLuint i;
+            const GLhalfNV *s = (const GLhalfNV *) src;
+            if (unpack->SwapBytes) {
+               for (i = 0; i < n; i++) {
+                  GLhalfNV value = s[i];
+                  SWAP2BYTE(value);
+                  indexes[i] = (GLuint) _mesa_half_to_float(value);
+               }
+            }
+            else {
+               for (i = 0; i < n; i++)
+                  indexes[i] = (GLuint) _mesa_half_to_float(s[i]);
+            }
+         }
+         break;
       default:
          _mesa_problem(NULL, "bad srcType in extract_uint_indexes");
          return;
@@ -2113,6 +2217,7 @@ extract_float_rgba(GLuint n, GLfloat rgba[][4],
           srcType == GL_SHORT ||
           srcType == GL_UNSIGNED_INT ||
           srcType == GL_INT ||
+          srcType == GL_HALF_FLOAT_NV ||
           srcType == GL_FLOAT ||
           srcType == GL_UNSIGNED_BYTE_3_3_2 ||
           srcType == GL_UNSIGNED_BYTE_2_3_3_REV ||
@@ -2291,6 +2396,12 @@ extract_float_rgba(GLuint n, GLfloat rgba[][4],
          PROCESS(blueIndex,  BCOMP, 0.0F, GLfloat, (GLfloat));
          PROCESS(alphaIndex, ACOMP, 1.0F, GLfloat, (GLfloat));
          break;
+      case GL_HALF_FLOAT_NV:
+         PROCESS(redIndex,   RCOMP, 0.0F, GLhalfNV, _mesa_half_to_float);
+         PROCESS(greenIndex, GCOMP, 0.0F, GLhalfNV, _mesa_half_to_float);
+         PROCESS(blueIndex,  BCOMP, 0.0F, GLhalfNV, _mesa_half_to_float);
+         PROCESS(alphaIndex, ACOMP, 1.0F, GLhalfNV, _mesa_half_to_float);
+         break;
       case GL_UNSIGNED_BYTE_3_3_2:
          {
             const GLubyte *ubsrc = (const GLubyte *) src;
@@ -2626,6 +2737,7 @@ _mesa_unpack_chan_color_span( GLcontext *ctx,
           srcType == GL_SHORT ||
           srcType == GL_UNSIGNED_INT ||
           srcType == GL_INT ||
+          srcType == GL_HALF_FLOAT_NV ||
           srcType == GL_FLOAT ||
           srcType == GL_UNSIGNED_BYTE_3_3_2 ||
           srcType == GL_UNSIGNED_BYTE_2_3_3_REV ||
@@ -3019,6 +3131,7 @@ _mesa_unpack_float_color_span( GLcontext *ctx,
           srcType == GL_SHORT ||
           srcType == GL_UNSIGNED_INT ||
           srcType == GL_INT ||
+          srcType == GL_HALF_FLOAT_NV ||
           srcType == GL_FLOAT ||
           srcType == GL_UNSIGNED_BYTE_3_3_2 ||
           srcType == GL_UNSIGNED_BYTE_2_3_3_REV ||
@@ -3285,6 +3398,7 @@ _mesa_unpack_index_span( const GLcontext *ctx, GLuint n,
           srcType == GL_SHORT ||
           srcType == GL_UNSIGNED_INT ||
           srcType == GL_INT ||
+          srcType == GL_HALF_FLOAT_NV ||
           srcType == GL_FLOAT);
 
    ASSERT(dstType == GL_UNSIGNED_BYTE ||
@@ -3457,6 +3571,18 @@ _mesa_pack_index_span( const GLcontext *ctx, GLuint n,
          }
       }
       break;
+   case GL_HALF_FLOAT_NV:
+      {
+         GLhalfNV *dst = (GLhalfNV *) dest;
+         GLuint i;
+         for (i = 0; i < n; i++) {
+            dst[i] = _mesa_float_to_half((GLfloat) source[i]);
+         }
+         if (dstPacking->SwapBytes) {
+            _mesa_swap2( (GLushort *) dst, n );
+         }
+      }
+      break;
    default:
       _mesa_problem(ctx, "bad type in _mesa_pack_index_span");
    }
@@ -3491,6 +3617,7 @@ _mesa_unpack_stencil_span( const GLcontext *ctx, GLuint n,
           srcType == GL_SHORT ||
           srcType == GL_UNSIGNED_INT ||
           srcType == GL_INT ||
+          srcType == GL_HALF_FLOAT_NV ||
           srcType == GL_FLOAT);
 
    ASSERT(dstType == GL_UNSIGNED_BYTE ||
@@ -3677,6 +3804,18 @@ _mesa_pack_stencil_span( const GLcontext *ctx, GLuint n,
          }
       }
       break;
+   case GL_HALF_FLOAT_NV:
+      {
+         GLhalfNV *dst = (GLhalfNV *) dest;
+         GLuint i;
+         for (i=0;i<n;i++) {
+            dst[i] = _mesa_half_to_float(source[i]);
+         }
+         if (dstPacking->SwapBytes) {
+            _mesa_swap2( (GLushort *) dst, n );
+         }
+      }
+      break;
    case GL_BITMAP:
       if (dstPacking->LsbFirst) {
          GLubyte *dst = (GLubyte *) dest;
@@ -3778,6 +3917,15 @@ _mesa_unpack_depth_span( const GLcontext *ctx, GLuint n, GLfloat *dest,
       case GL_FLOAT:
          MEMCPY(dest, source, n * sizeof(GLfloat));
          break;
+      case GL_HALF_FLOAT_NV:
+         {
+            GLuint i;
+            const GLhalfNV *src = (const GLhalfNV *) source;
+            for (i = 0; i < n; i++) {
+               dest[i] = _mesa_half_to_float(src[i]);
+            }
+         }
+         break;
       default:
          _mesa_problem(NULL, "bad type in _mesa_unpack_depth_span()");
          return;
@@ -3898,6 +4046,18 @@ _mesa_pack_depth_span( const GLcontext *ctx, GLuint n, GLvoid *dest,
          }
       }
       break;
+   case GL_HALF_FLOAT_NV:
+      {
+         GLhalfNV *dst = (GLhalfNV *) dest;
+         GLuint i;
+         for (i = 0; i < n; i++) {
+            dst[i] = _mesa_float_to_half(depthSpan[i]);
+         }
+         if (dstPacking->SwapBytes) {
+            _mesa_swap2( (GLushort *) dst, n );
+         }
+      }
+      break;
    default:
       _mesa_problem(ctx, "bad type in _mesa_pack_depth_span");
    }
diff --git a/src/mesa/main/imports.c b/src/mesa/main/imports.c
index 6420303a4e..d8549f2b10 100644
--- a/src/mesa/main/imports.c
+++ b/src/mesa/main/imports.c
@@ -353,6 +353,9 @@ static void init_sqrt_table(void)
 }
 
 
+/**
+ * Single precision square root.
+ */
 float
 _mesa_sqrtf( float x )
 {
@@ -497,7 +500,9 @@ _mesa_inv_sqrtf(float n)
 }
 
 
-/** Wrapper around either pow() or xf86pow() */
+/**
+ * Wrapper around either pow() or xf86pow().
+ */
 double
 _mesa_pow(double x, double y)
 {
@@ -509,7 +514,7 @@ _mesa_pow(double x, double y)
 }
 
 
-/*
+/**
  * Return number of bits set in given GLuint.
  */
 unsigned int
@@ -522,6 +527,141 @@ _mesa_bitcount(unsigned int n)
    return bits;
 }
 
+
+/**
+ * Convert a 4-byte float to a 2-byte half float.
+ */
+GLhalfNV
+_mesa_float_to_half(float val)
+{
+   const int flt = *((int *) &val);
+   const int flt_m = flt & 0x7fffff;
+   const int flt_e = (flt >> 23) & 0xff;
+   const int flt_s = (flt >> 31) & 0x1;
+   int s, e, m;
+   GLhalfNV result;
+   
+   /* sign bit */
+   s = flt_s;
+
+   /* handle special cases */
+   if ((flt_e == 0) && (flt_m == 0)) {
+      /* zero */
+      m = 0;
+      e = 0;
+   }
+   else if ((flt_e == 0) && (flt_m != 0)) {
+      /* denorm -- denorm float maps to 0 half */
+      m = 0;
+      e = 0;
+   }
+   else if ((flt_e == 0xff) && (flt_m == 0)) {
+      /* infinity */
+      m = 0;
+      e = 31;
+   }
+   else if ((flt_e == 0xff) && (flt_m != 0)) {
+      /* NaN */
+      m = 1;
+      e = 31;
+   }
+   else {
+      /* regular number */
+      const int new_exp = flt_e - 127;
+      if (new_exp < -24) {
+         /* this maps to 0 */
+         m = 0;
+         e = 0;
+      }
+      else if (new_exp < -14) {
+         /* this maps to a denorm */
+         unsigned int exp_val = (unsigned int) (-14 - new_exp); /* 2^-exp_val*/
+         e = 0;
+         switch (exp_val) {
+            case 0:
+               _mesa_warning(NULL,
+                   "float_to_half: logical error in denorm creation!\n");
+               m = 0;
+               break;
+            case 1: m = 512 + (flt_m >> 14); break;
+            case 2: m = 256 + (flt_m >> 15); break;
+            case 3: m = 128 + (flt_m >> 16); break;
+            case 4: m = 64 + (flt_m >> 17); break;
+            case 5: m = 32 + (flt_m >> 18); break;
+            case 6: m = 16 + (flt_m >> 19); break;
+            case 7: m = 8 + (flt_m >> 20); break;
+            case 8: m = 4 + (flt_m >> 21); break;
+            case 9: m = 2 + (flt_m >> 22); break;
+            case 10: m = 1; break;
+         }
+      }
+      else if (new_exp > 15) {
+         /* map this value to infinity */
+         m = 0;
+         e = 31;
+      }
+      else {
+         /* regular */
+         e = new_exp + 15;
+         m = flt_m >> 13;
+      }
+   }
+
+   result = (s << 15) | (e << 10) | m;
+   return result;
+}
+
+
+/**
+ * Convert a 2-byte half float to a 4-byte float.
+ */
+float
+_mesa_half_to_float(GLhalfNV val)
+{
+   /* XXX could also use a 64K-entry lookup table */
+   const int m = val & 0x3ff;
+   const int e = (val >> 10) & 0x1f;
+   const int s = (val >> 15) & 0x1;
+   int flt_m, flt_e, flt_s, flt;
+   float result;
+
+   /* sign bit */
+   flt_s = s;
+
+   /* handle special cases */
+   if ((e == 0) && (m == 0)) {
+      /* zero */
+      flt_m = 0;
+      flt_e = 0;
+   }
+   else if ((e == 0) && (m != 0)) {
+      /* denorm -- denorm half will fit in non-denorm single */
+      const float half_denorm = 1.0f / 16384.0f; /* 2^-14 */
+      float mantissa = ((float) (m)) / 1024.0f;
+      float sign = s ? -1.0f : 1.0f;
+      return sign * mantissa * half_denorm;
+   }
+   else if ((e == 31) && (m == 0)) {
+      /* infinity */
+      flt_e = 0xff;
+      flt_m = 0;
+   }
+   else if ((e == 31) && (m != 0)) {
+      /* NaN */
+      flt_e = 0xff;
+      flt_m = 1;
+   }
+   else {
+      /* regular */
+      flt_e = e + 112;
+      flt_m = m << 13;
+   }
+
+   flt = (flt_s << 31) | (flt_e << 23) | flt_m;
+   result = *((float *) &flt);
+   return result;
+}
+
 /*@}*/
 
 
@@ -529,7 +669,9 @@ _mesa_bitcount(unsigned int n)
 /** \name Environment vars */
 /*@{*/
 
-/** Wrapper around either () or xf86() */
+/**
+ * Wrapper for getenv().
+ */
 char *
 _mesa_getenv( const char *var )
 {
diff --git a/src/mesa/main/imports.h b/src/mesa/main/imports.h
index 2c9c2fc342..0c0a291e16 100644
--- a/src/mesa/main/imports.h
+++ b/src/mesa/main/imports.h
@@ -696,6 +696,12 @@ _mesa_log2(float x);
 extern unsigned int
 _mesa_bitcount(unsigned int n);
 
+extern GLhalfNV
+_mesa_float_to_half(float f);
+
+extern float
+_mesa_half_to_float(GLhalfNV h);
+
 
 extern char *
 _mesa_getenv( const char *var );