Merge remote-tracking branch 'origin/master' into pipe-video

Conflicts:
	src/gallium/drivers/r600/r600_state.c

6 files changed, 585 insertions, 9 deletions

diff --git a/src/gallium/auxiliary/util/u_format.csv b/src/gallium/auxiliary/util/u_format.csv
index e4b3f5df6ee..347e2beb8dd 100644
--- a/src/gallium/auxiliary/util/u_format.csv
+++ b/src/gallium/auxiliary/util/u_format.csv
@@ -147,7 +147,7 @@ PIPE_FORMAT_G8R8_G8B8_UNORM      , subsampled, 2, 1, x32 ,     ,     ,     , xyz
 
 # some special formats not fitting anywhere else
 PIPE_FORMAT_R10G10B10A2_USCALED   , plain,      1,  1, u10 , u10 , u10 , u2  , xyzw, rgb
-PIPE_FORMAT_R11G11B10_FLOAT       , plain,      1,  1, f11 , f11 , f10 ,     , xyz1, rgb
+PIPE_FORMAT_R11G11B10_FLOAT       , other,      1,  1, x32 ,     ,     ,     , xyz1, rgb
 PIPE_FORMAT_R9G9B9E5_FLOAT        , other,      1,  1, x32 ,     ,     ,     , xyz1, rgb
 PIPE_FORMAT_R1_UNORM              , other,      8,  1, x8  ,     ,     ,     , x001, rgb
 # A.k.a. D3DFMT_CxV8U8
diff --git a/src/gallium/auxiliary/util/u_format_other.c b/src/gallium/auxiliary/util/u_format_other.c
index fa42ec37138..c23f4ee4ac5 100644
--- a/src/gallium/auxiliary/util/u_format_other.c
+++ b/src/gallium/auxiliary/util/u_format_other.c
@@ -28,6 +28,8 @@
 
 #include "u_math.h"
 #include "u_format_other.h"
+#include "u_format_rgb9e5.h"
+#include "u_format_r11g11b10f.h"
 
 
 void
@@ -35,7 +37,23 @@ util_format_r9g9b9e5_float_unpack_rgba_float(float *dst_row, unsigned dst_stride
                                         const uint8_t *src_row, unsigned src_stride,
                                         unsigned width, unsigned height)
 {
-
+   unsigned x, y;
+   for(y = 0; y < height; y += 1) {
+      float *dst = dst_row;
+      const uint8_t *src = src_row;
+      for(x = 0; x < width; x += 1) {
+         uint32_t value = *(const uint32_t *)src;
+#ifdef PIPE_ARCH_BIG_ENDIAN
+         value = util_bswap32(value);
+#endif
+         rgb9e5_to_float3(value, dst);
+         dst[3] = 1; /* a */
+         src += 4;
+         dst += 4;
+      }
+      src_row += src_stride;
+      dst_row += dst_stride/sizeof(*dst_row);
+   }
 }
 
 void
@@ -43,14 +61,34 @@ util_format_r9g9b9e5_float_pack_rgba_float(uint8_t *dst_row, unsigned dst_stride
                                       const float *src_row, unsigned src_stride,
                                       unsigned width, unsigned height)
 {
-
+   unsigned x, y;
+   for(y = 0; y < height; y += 1) {
+      const float *src = src_row;
+      uint8_t *dst = dst_row;
+      for(x = 0; x < width; x += 1) {
+         uint32_t value = float3_to_rgb9e5(src);
+#ifdef PIPE_ARCH_BIG_ENDIAN
+         value = util_bswap32(value);
+#endif
+         *(uint32_t *)dst = value;
+         src += 4;
+         dst += 4;
+      }
+      dst_row += dst_stride;
+      src_row += src_stride/sizeof(*src_row);
+   }
 }
 
 void
 util_format_r9g9b9e5_float_fetch_rgba_float(float *dst, const uint8_t *src,
                                        unsigned i, unsigned j)
 {
-
+   uint32_t value = *(const uint32_t *)src;
+#ifdef PIPE_ARCH_BIG_ENDIAN
+   value = util_bswap32(value);
+#endif
+   rgb9e5_to_float3(value, dst);
+   dst[3] = 1; /* a */
 }
 
 
@@ -59,7 +97,27 @@ util_format_r9g9b9e5_float_unpack_rgba_8unorm(uint8_t *dst_row, unsigned dst_str
                                          const uint8_t *src_row, unsigned src_stride,
                                          unsigned width, unsigned height)
 {
-
+   unsigned x, y;
+   float p[3];
+   for(y = 0; y < height; y += 1) {
+      uint8_t *dst = dst_row;
+      const uint8_t *src = src_row;
+      for(x = 0; x < width; x += 1) {
+         uint32_t value = *(const uint32_t *)src;
+#ifdef PIPE_ARCH_BIG_ENDIAN
+         value = util_bswap32(value);
+#endif
+         rgb9e5_to_float3(value, p);
+         dst[0] = float_to_ubyte(p[0]); /* r */
+         dst[1] = float_to_ubyte(p[1]); /* g */
+         dst[2] = float_to_ubyte(p[2]); /* b */
+         dst[3] = 255; /* a */
+         src += 4;
+         dst += 4;
+      }
+      src_row += src_stride;
+      dst_row += dst_stride/sizeof(*dst_row);
+   }
 }
 
 
@@ -68,7 +126,145 @@ util_format_r9g9b9e5_float_pack_rgba_8unorm(uint8_t *dst_row, unsigned dst_strid
                                        const uint8_t *src_row, unsigned src_stride,
                                        unsigned width, unsigned height)
 {
+   unsigned x, y;
+   float p[3];
+   for(y = 0; y < height; y += 1) {
+      const uint8_t *src = src_row;
+      uint8_t *dst = dst_row;
+      for(x = 0; x < width; x += 1) {
+         uint32_t value;
+         p[0] = ubyte_to_float(src[0]);
+         p[1] = ubyte_to_float(src[1]);
+         p[2] = ubyte_to_float(src[2]);
+         value = float3_to_rgb9e5(p);
+#ifdef PIPE_ARCH_BIG_ENDIAN
+         value = util_bswap32(value);
+#endif
+         *(uint32_t *)dst = value;
+         src += 4;
+         dst += 4;
+      }
+      dst_row += dst_stride;
+      src_row += src_stride/sizeof(*src_row);
+   }
+}
+
 
+void
+util_format_r11g11b10_float_unpack_rgba_float(float *dst_row, unsigned dst_stride,
+                                        const uint8_t *src_row, unsigned src_stride,
+                                        unsigned width, unsigned height)
+{
+   unsigned x, y;
+   for(y = 0; y < height; y += 1) {
+      float *dst = dst_row;
+      const uint8_t *src = src_row;
+      for(x = 0; x < width; x += 1) {
+         uint32_t value = *(const uint32_t *)src;
+#ifdef PIPE_ARCH_BIG_ENDIAN
+         value = util_bswap32(value);
+#endif
+         r11g11b10f_to_float3(value, dst);
+         dst[3] = 1; /* a */
+         src += 4;
+         dst += 4;
+      }
+      src_row += src_stride;
+      dst_row += dst_stride/sizeof(*dst_row);
+   }
+}
+
+void
+util_format_r11g11b10_float_pack_rgba_float(uint8_t *dst_row, unsigned dst_stride,
+                                      const float *src_row, unsigned src_stride,
+                                      unsigned width, unsigned height)
+{
+   unsigned x, y;
+   for(y = 0; y < height; y += 1) {
+      const float *src = src_row;
+      uint8_t *dst = dst_row;
+      for(x = 0; x < width; x += 1) {
+         uint32_t value = float3_to_r11g11b10f(src);
+#ifdef PIPE_ARCH_BIG_ENDIAN
+         value = util_bswap32(value);
+#endif
+         *(uint32_t *)dst = value;
+         src += 4;
+         dst += 4;
+      }
+      dst_row += dst_stride;
+      src_row += src_stride/sizeof(*src_row);
+   }
+}
+
+void
+util_format_r11g11b10_float_fetch_rgba_float(float *dst, const uint8_t *src,
+                                       unsigned i, unsigned j)
+{
+   uint32_t value = *(const uint32_t *)src;
+#ifdef PIPE_ARCH_BIG_ENDIAN
+   value = util_bswap32(value);
+#endif
+   r11g11b10f_to_float3(value, dst);
+   dst[3] = 1; /* a */
+}
+
+
+void
+util_format_r11g11b10_float_unpack_rgba_8unorm(uint8_t *dst_row, unsigned dst_stride,
+                                         const uint8_t *src_row, unsigned src_stride,
+                                         unsigned width, unsigned height)
+{
+   unsigned x, y;
+   float p[3];
+   for(y = 0; y < height; y += 1) {
+      uint8_t *dst = dst_row;
+      const uint8_t *src = src_row;
+      for(x = 0; x < width; x += 1) {
+         uint32_t value = *(const uint32_t *)src;
+#ifdef PIPE_ARCH_BIG_ENDIAN
+         value = util_bswap32(value);
+#endif
+         r11g11b10f_to_float3(value, p);
+         dst[0] = float_to_ubyte(p[0]); /* r */
+         dst[1] = float_to_ubyte(p[1]); /* g */
+         dst[2] = float_to_ubyte(p[2]); /* b */
+         dst[3] = 255; /* a */
+         src += 4;
+         dst += 4;
+      }
+      src_row += src_stride;
+      dst_row += dst_stride/sizeof(*dst_row);
+   }
+}
+
+
+void
+util_format_r11g11b10_float_pack_rgba_8unorm(uint8_t *dst_row, unsigned dst_stride,
+                                       const uint8_t *src_row, unsigned src_stride,
+                                       unsigned width, unsigned height)
+{
+   unsigned x, y;
+   float p[3];
+   for(y = 0; y < height; y += 1) {
+      const uint8_t *src = src_row;
+      uint8_t *dst = dst_row;
+      for(x = 0; x < width; x += 1) {
+         uint32_t value;
+         p[0] = ubyte_to_float(src[0]);
+         p[1] = ubyte_to_float(src[1]);
+         p[2] = ubyte_to_float(src[2]);
+         value = float3_to_r11g11b10f(p);
+#ifdef PIPE_ARCH_BIG_ENDIAN
+         value = util_bswap32(value);
+#endif
+         *(uint32_t *)dst = value;
+         src += 4;
+         dst += 4;
+      }
+      dst_row += dst_stride;
+      src_row += src_stride/sizeof(*src_row);
+   }
 }
 
 
diff --git a/src/gallium/auxiliary/util/u_format_other.h b/src/gallium/auxiliary/util/u_format_other.h
index 98c97be33fa..2f6a908bbe8 100644
--- a/src/gallium/auxiliary/util/u_format_other.h
+++ b/src/gallium/auxiliary/util/u_format_other.h
@@ -57,6 +57,32 @@ util_format_r9g9b9e5_float_pack_rgba_8unorm(uint8_t *dst_row, unsigned dst_strid
                                        const uint8_t *src_row, unsigned src_stride,
                                        unsigned width, unsigned height);
 
+
+void
+util_format_r11g11b10_float_unpack_rgba_float(float *dst_row, unsigned dst_stride,
+                                        const uint8_t *src_row, unsigned src_stride,
+                                        unsigned width, unsigned height);
+
+void
+util_format_r11g11b10_float_pack_rgba_float(uint8_t *dst_row, unsigned dst_stride,
+                                      const float *src_row, unsigned src_stride,
+                                      unsigned width, unsigned height);
+
+void
+util_format_r11g11b10_float_fetch_rgba_float(float *dst, const uint8_t *src,
+                                       unsigned i, unsigned j);
+
+void
+util_format_r11g11b10_float_unpack_rgba_8unorm(uint8_t *dst_row, unsigned dst_stride,
+                                         const uint8_t *src_row, unsigned src_stride,
+                                         unsigned width, unsigned height);
+
+void
+util_format_r11g11b10_float_pack_rgba_8unorm(uint8_t *dst_row, unsigned dst_stride,
+                                       const uint8_t *src_row, unsigned src_stride,
+                                       unsigned width, unsigned height);
+
+
 void
 util_format_r1_unorm_unpack_rgba_float(float *dst_row, unsigned dst_stride,
                                   const uint8_t *src_row, unsigned src_stride,
diff --git a/src/gallium/auxiliary/util/u_format_r11g11b10f.h b/src/gallium/auxiliary/util/u_format_r11g11b10f.h
new file mode 100644
index 00000000000..c4181d0e34e
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_format_r11g11b10f.h
@@ -0,0 +1,190 @@
+/*
+ * Copyright (C) 2011 Marek Olšák <[email protected]>
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ */
+
+/* Based on code from The OpenGL Programming Guide / 7th Edition, Appendix J.
+ * Available here: http://www.opengl-redbook.com/appendices/
+ * The algorithm in the book contains a bug though, which is fixed in the code
+ * below.
+ */
+
+#define UF11_EXPONENT_BIAS   15
+#define UF11_EXPONENT_BITS   0x1F
+#define UF11_EXPONENT_SHIFT  6
+#define UF11_MANTISSA_BITS   0x3F
+#define UF11_MANTISSA_SHIFT  (23 - UF11_EXPONENT_SHIFT)
+#define UF11_MAX_EXPONENT    (UF11_EXPONENT_BITS << UF11_EXPONENT_SHIFT)
+
+#define UF10_EXPONENT_BIAS   15
+#define UF10_EXPONENT_BITS   0x1F
+#define UF10_EXPONENT_SHIFT  5
+#define UF10_MANTISSA_BITS   0x3F
+#define UF10_MANTISSA_SHIFT  (23 - UF10_EXPONENT_SHIFT)
+#define UF10_MAX_EXPONENT    (UF10_EXPONENT_BITS << UF10_EXPONENT_SHIFT)
+
+#define F32_INFINITY         0x7f800000
+
+static INLINE unsigned f32_to_uf11(float val)
+{
+   uint32_t f32 = (*(uint32_t *) &val);
+   uint16_t uf11 = 0;
+
+   /* Decode little-endian 32-bit floating-point value */
+   int sign = (f32 >> 16) & 0x8000;
+   /* Map exponent to the range [-127,128] */
+   int exponent = ((f32 >> 23) & 0xff) - 127;
+   int mantissa = f32 & 0x007fffff;
+
+   if (sign) return 0;
+
+   if (exponent == 128) { /* Infinity or NaN */
+      uf11 = UF11_MAX_EXPONENT;
+      if (mantissa) uf11 |= (mantissa & UF11_MANTISSA_BITS);
+   }
+   else if (exponent > 15) { /* Overflow - flush to Infinity */
+      uf11 = UF11_MAX_EXPONENT;
+   }
+   else if (exponent > -15) { /* Representable value */
+      exponent += UF11_EXPONENT_BIAS;
+      mantissa >>= UF11_MANTISSA_SHIFT;
+      uf11 = exponent << UF11_EXPONENT_SHIFT | mantissa;
+   }
+
+   return uf11;
+}
+
+static INLINE float uf11_to_f32(uint16_t val)
+{
+   union {
+      float f;
+      uint32_t ui;
+   } f32;
+
+   int exponent = (val & 0x07c0) >> UF11_EXPONENT_SHIFT;
+   int mantissa = (val & 0x003f);
+
+   f32.f = 0.0;
+
+   if (exponent == 0) {
+      if (mantissa != 0) {
+         const float scale = 1.0 / (1 << 20);
+         f32.f = scale * mantissa;
+      }
+   }
+   else if (exponent == 31) {
+      f32.ui = F32_INFINITY | mantissa;
+   }
+   else {
+      float scale, decimal;
+      exponent -= 15;
+      if (exponent < 0) {
+         scale = 1.0 / (1 << -exponent);
+      }
+      else {
+         scale = 1 << exponent;
+      }
+      decimal = 1.0 + (float) mantissa / 64;
+      f32.f = scale * decimal;
+   }
+
+   return f32.f;
+}
+
+static INLINE unsigned f32_to_uf10(float val)
+{
+   uint32_t f32 = (*(uint32_t *) &val);
+   uint16_t uf10 = 0;
+
+   /* Decode little-endian 32-bit floating-point value */
+   int sign = (f32 >> 16) & 0x8000;
+   /* Map exponent to the range [-127,128] */
+   int exponent = ((f32 >> 23) & 0xff) - 127;
+   int mantissa = f32 & 0x007fffff;
+
+   if (sign) return 0;
+
+   if (exponent == 128) { /* Infinity or NaN */
+      uf10 = UF10_MAX_EXPONENT;
+      if (mantissa) uf10 |= (mantissa & UF10_MANTISSA_BITS);
+   }
+   else if (exponent > 15) { /* Overflow - flush to Infinity */
+      uf10 = UF10_MAX_EXPONENT;
+   }
+   else if (exponent > -15) { /* Representable value */
+      exponent += UF10_EXPONENT_BIAS;
+      mantissa >>= UF10_MANTISSA_SHIFT;
+      uf10 = exponent << UF10_EXPONENT_SHIFT | mantissa;
+   }
+
+   return uf10;
+}
+
+static INLINE float uf10_to_f32(uint16_t val)
+{
+   union {
+      float f;
+      uint32_t ui;
+   } f32;
+
+   int exponent = (val & 0x07c0) >> UF10_EXPONENT_SHIFT;
+   int mantissa = (val & 0x003f);
+
+   f32.f = 0.0;
+
+   if (exponent == 0) {
+      if (mantissa != 0) {
+         const float scale = 1.0 / (1 << 20);
+         f32.f = scale * mantissa;
+      }
+   }
+   else if (exponent == 31) {
+      f32.ui = F32_INFINITY | mantissa;
+   }
+   else {
+      float scale, decimal;
+      exponent -= 15;
+      if (exponent < 0) {
+         scale = 1.0 / (1 << -exponent);
+      }
+      else {
+         scale = 1 << exponent;
+      }
+      decimal = 1.0 + (float) mantissa / 32;
+      f32.f = scale * decimal;
+   }
+
+   return f32.f;
+}
+
+static INLINE unsigned float3_to_r11g11b10f(const float rgb[3])
+{
+   return ( f32_to_uf11(rgb[0]) & 0x7ff) |
+          ((f32_to_uf11(rgb[1]) & 0x7ff) << 11) |
+          ((f32_to_uf10(rgb[2]) & 0x3ff) << 22);
+}
+
+static INLINE void r11g11b10f_to_float3(unsigned rgb, float retval[3])
+{
+   retval[0] = uf11_to_f32( rgb        & 0x7ff);
+   retval[1] = uf11_to_f32((rgb >> 11) & 0x7ff);
+   retval[2] = uf10_to_f32((rgb >> 22) & 0x3ff);
+}
diff --git a/src/gallium/auxiliary/util/u_format_rgb9e5.h b/src/gallium/auxiliary/util/u_format_rgb9e5.h
new file mode 100644
index 00000000000..c2a3f6f3e9d
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_format_rgb9e5.h
@@ -0,0 +1,164 @@
+/*
+ * Copyright (C) 2011 Marek Olšák <[email protected]>
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ */
+
+/* Copied from EXT_texture_shared_exponent and edited. */
+
+#ifndef RGB9E5_H
+#define RGB9E5_H
+
+#include <math.h>
+#include <assert.h>
+
+#define RGB9E5_EXPONENT_BITS          5
+#define RGB9E5_MANTISSA_BITS          9
+#define RGB9E5_EXP_BIAS               15
+#define RGB9E5_MAX_VALID_BIASED_EXP   31
+
+#define MAX_RGB9E5_EXP               (RGB9E5_MAX_VALID_BIASED_EXP - RGB9E5_EXP_BIAS)
+#define RGB9E5_MANTISSA_VALUES       (1<<RGB9E5_MANTISSA_BITS)
+#define MAX_RGB9E5_MANTISSA          (RGB9E5_MANTISSA_VALUES-1)
+#define MAX_RGB9E5                   (((float)MAX_RGB9E5_MANTISSA)/RGB9E5_MANTISSA_VALUES * (1<<MAX_RGB9E5_EXP))
+#define EPSILON_RGB9E5               ((1.0/RGB9E5_MANTISSA_VALUES) / (1<<RGB9E5_EXP_BIAS))
+
+typedef union {
+   unsigned int raw;
+   float value;
+   struct {
+#if defined(MESA_BIG_ENDIAN) || defined(PIPE_ARCH_BIG_ENDIAN)
+      unsigned int negative:1;
+      unsigned int biasedexponent:8;
+      unsigned int mantissa:23;
+#else
+      unsigned int mantissa:23;
+      unsigned int biasedexponent:8;
+      unsigned int negative:1;
+#endif
+   } field;
+} float754;
+
+typedef union {
+   unsigned int raw;
+   struct {
+#if defined(MESA_BIG_ENDIAN) || defined(PIPE_ARCH_BIG_ENDIAN)
+      unsigned int biasedexponent:RGB9E5_EXPONENT_BITS;
+      unsigned int b:RGB9E5_MANTISSA_BITS;
+      unsigned int g:RGB9E5_MANTISSA_BITS;
+      unsigned int r:RGB9E5_MANTISSA_BITS;
+#else
+      unsigned int r:RGB9E5_MANTISSA_BITS;
+      unsigned int g:RGB9E5_MANTISSA_BITS;
+      unsigned int b:RGB9E5_MANTISSA_BITS;
+      unsigned int biasedexponent:RGB9E5_EXPONENT_BITS;
+#endif
+   } field;
+} rgb9e5;
+
+static INLINE float rgb9e5_ClampRange(float x)
+{
+   if (x > 0.0) {
+      if (x >= MAX_RGB9E5) {
+         return MAX_RGB9E5;
+      } else {
+         return x;
+      }
+   } else {
+      /* NaN gets here too since comparisons with NaN always fail! */
+      return 0.0;
+   }
+}
+
+/* Ok, FloorLog2 is not correct for the denorm and zero values, but we
+   are going to do a max of this value with the minimum rgb9e5 exponent
+   that will hide these problem cases. */
+static INLINE int rgb9e5_FloorLog2(float x)
+{
+   float754 f;
+
+   f.value = x;
+   return (f.field.biasedexponent - 127);
+}
+
+static INLINE unsigned float3_to_rgb9e5(const float rgb[3])
+{
+   rgb9e5 retval;
+   float maxrgb;
+   int rm, gm, bm;
+   float rc, gc, bc;
+   int exp_shared, maxm;
+   double denom;
+
+   rc = rgb9e5_ClampRange(rgb[0]);
+   gc = rgb9e5_ClampRange(rgb[1]);
+   bc = rgb9e5_ClampRange(rgb[2]);
+
+   maxrgb = MAX3(rc, gc, bc);
+   exp_shared = MAX2(-RGB9E5_EXP_BIAS-1, rgb9e5_FloorLog2(maxrgb)) + 1 + RGB9E5_EXP_BIAS;
+   assert(exp_shared <= RGB9E5_MAX_VALID_BIASED_EXP);
+   assert(exp_shared >= 0);
+   /* This pow function could be replaced by a table. */
+   denom = pow(2, exp_shared - RGB9E5_EXP_BIAS - RGB9E5_MANTISSA_BITS);
+
+   maxm = (int) floor(maxrgb / denom + 0.5);
+   if (maxm == MAX_RGB9E5_MANTISSA+1) {
+      denom *= 2;
+      exp_shared += 1;
+      assert(exp_shared <= RGB9E5_MAX_VALID_BIASED_EXP);
+   } else {
+      assert(maxm <= MAX_RGB9E5_MANTISSA);
+   }
+
+   rm = (int) floor(rc / denom + 0.5);
+   gm = (int) floor(gc / denom + 0.5);
+   bm = (int) floor(bc / denom + 0.5);
+
+   assert(rm <= MAX_RGB9E5_MANTISSA);
+   assert(gm <= MAX_RGB9E5_MANTISSA);
+   assert(bm <= MAX_RGB9E5_MANTISSA);
+   assert(rm >= 0);
+   assert(gm >= 0);
+   assert(bm >= 0);
+
+   retval.field.r = rm;
+   retval.field.g = gm;
+   retval.field.b = bm;
+   retval.field.biasedexponent = exp_shared;
+
+   return retval.raw;
+}
+
+static INLINE void rgb9e5_to_float3(unsigned rgb, float retval[3])
+{
+   rgb9e5 v;
+   int exponent;
+   float scale;
+
+   v.raw = rgb;
+   exponent = v.field.biasedexponent - RGB9E5_EXP_BIAS - RGB9E5_MANTISSA_BITS;
+   scale = (float) pow(2, exponent);
+
+   retval[0] = v.field.r * scale;
+   retval[1] = v.field.g * scale;
+   retval[2] = v.field.b * scale;
+}
+
+#endif
diff --git a/src/gallium/auxiliary/util/u_math.h b/src/gallium/auxiliary/util/u_math.h
index e3d4c06b6f9..dad6a101fd7 100644
--- a/src/gallium/auxiliary/util/u_math.h
+++ b/src/gallium/auxiliary/util/u_math.h
@@ -581,11 +581,11 @@ util_bswap16(uint16_t n)
 #define MIN2( A, B )   ( (A)<(B) ? (A) : (B) )
 #define MAX2( A, B )   ( (A)>(B) ? (A) : (B) )
 
-#define MIN3( A, B, C ) MIN2( MIN2( A, B ), C )
-#define MAX3( A, B, C ) MAX2( MAX2( A, B ), C )
+#define MIN3( A, B, C ) ((A) < (B) ? MIN2(A, C) : MIN2(B, C))
+#define MAX3( A, B, C ) ((A) > (B) ? MAX2(A, C) : MAX2(B, C))
 
-#define MIN4( A, B, C, D ) MIN2( MIN2( A, B ), MIN2(C, D) )
-#define MAX4( A, B, C, D ) MAX2( MAX2( A, B ), MAX2(C, D) )
+#define MIN4( A, B, C, D ) ((A) < (B) ? MIN3(A, C, D) : MIN3(B, C, D))
+#define MAX4( A, B, C, D ) ((A) > (B) ? MAX3(A, C, D) : MAX3(B, C, D))
 
 
 /**