xserver xkeyboar-config mesa git update 16 May 2011

2 files changed, 1289 insertions, 1273 deletions

diff --git a/mesalib/src/gallium/auxiliary/util/u_math.h b/mesalib/src/gallium/auxiliary/util/u_math.h
index 16a9eab6d..2ecade5f7 100644
--- a/mesalib/src/gallium/auxiliary/util/u_math.h
+++ b/mesalib/src/gallium/auxiliary/util/u_math.h
@@ -1,660 +1,644 @@
-/**************************************************************************

- * 

- * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.

- * All Rights Reserved.

- * 

- * 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, sub license, 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 NON-INFRINGEMENT.

- * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS 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.

- * 

- **************************************************************************/

-

-

-/**

- * Math utilities and approximations for common math functions.

- * Reduced precision is usually acceptable in shaders...

- *

- * "fast" is used in the names of functions which are low-precision,

- * or at least lower-precision than the normal C lib functions.

- */

-

-

-#ifndef U_MATH_H

-#define U_MATH_H

-

-

-#include "pipe/p_compiler.h"

-#include "util/u_debug.h"

-

-

-#ifdef __cplusplus

-extern "C" {

-#endif

-

-

-#if defined(PIPE_SUBSYSTEM_WINDOWS_MINIPORT)

-__inline double ceil(double val)

-{

-   double ceil_val;

-

-   if ((val - (long) val) == 0) {

-      ceil_val = val;

-   }

-   else {

-      if (val > 0) {

-         ceil_val = (long) val + 1;

-      }

-      else {

-         ceil_val = (long) val;

-      }

-   }

-

-   return ceil_val;

-}

-

-#ifndef PIPE_SUBSYSTEM_WINDOWS_CE_OGL

-__inline double floor(double val)

-{

-   double floor_val;

-

-   if ((val - (long) val) == 0) {

-      floor_val = val;

-   }

-   else {

-      if (val > 0) {

-         floor_val = (long) val;

-      }

-      else {

-         floor_val = (long) val - 1;

-      }

-   }

-

-   return floor_val;

-}

-#endif

-

-#pragma function(pow)

-__inline double __cdecl pow(double val, double exponent)

-{

-   /* XXX */

-   assert(0);

-   return 0;

-}

-

-#pragma function(log)

-__inline double __cdecl log(double val)

-{

-   /* XXX */

-   assert(0);

-   return 0;

-}

-

-#pragma function(atan2)

-__inline double __cdecl atan2(double val)

-{

-   /* XXX */

-   assert(0);

-   return 0;

-}

-#else

-#include <math.h>

-#include <stdarg.h>

-#endif

-

-

-#ifndef M_SQRT2

-#define M_SQRT2 1.41421356237309504880

-#endif

-

-

-#if defined(_MSC_VER) 

-

-#if _MSC_VER < 1400 && !defined(__cplusplus) || defined(PIPE_SUBSYSTEM_WINDOWS_CE)

- 

-static INLINE float cosf( float f ) 

-{

-   return (float) cos( (double) f );

-}

-

-static INLINE float sinf( float f ) 

-{

-   return (float) sin( (double) f );

-}

-

-static INLINE float ceilf( float f ) 

-{

-   return (float) ceil( (double) f );

-}

-

-static INLINE float floorf( float f ) 

-{

-   return (float) floor( (double) f );

-}

-

-static INLINE float powf( float f, float g ) 

-{

-   return (float) pow( (double) f, (double) g );

-}

-

-static INLINE float sqrtf( float f ) 

-{

-   return (float) sqrt( (double) f );

-}

-

-static INLINE float fabsf( float f ) 

-{

-   return (float) fabs( (double) f );

-}

-

-static INLINE float logf( float f ) 

-{

-   return (float) log( (double) f );

-}

-

-#else

-/* Work-around an extra semi-colon in VS 2005 logf definition */

-#ifdef logf

-#undef logf

-#define logf(x) ((float)log((double)(x)))

-#endif /* logf */

-

-#define isfinite(x) _finite((double)(x))

-#define isnan(x) _isnan((double)(x))

-#endif /* _MSC_VER < 1400 && !defined(__cplusplus) */

-

-static INLINE double log2( double x )

-{

-   const double invln2 = 1.442695041;

-   return log( x ) * invln2;

-}

-

-static INLINE double

-round(double x)

-{

-   return x >= 0.0 ? floor(x + 0.5) : ceil(x - 0.5);

-}

-

-static INLINE float

-roundf(float x)

-{

-   return x >= 0.0f ? floorf(x + 0.5f) : ceilf(x - 0.5f);

-}

-

-#endif /* _MSC_VER */

-

-

-

-

-

-#define POW2_TABLE_SIZE_LOG2 9

-#define POW2_TABLE_SIZE (1 << POW2_TABLE_SIZE_LOG2)

-#define POW2_TABLE_OFFSET (POW2_TABLE_SIZE/2)

-#define POW2_TABLE_SCALE ((float)(POW2_TABLE_SIZE/2))

-extern float pow2_table[POW2_TABLE_SIZE];

-

-

-/**

- * Initialize math module.  This should be called before using any

- * other functions in this module.

- */

-extern void

-util_init_math(void);

-

-

-union fi {

-   float f;

-   int32_t i;

-   uint32_t ui;

-};

-

-

-/**

- * Fast version of 2^x

- * Identity: exp2(a + b) = exp2(a) * exp2(b)

- * Let ipart = int(x)

- * Let fpart = x - ipart;

- * So, exp2(x) = exp2(ipart) * exp2(fpart)

- * Compute exp2(ipart) with i << ipart

- * Compute exp2(fpart) with lookup table.

- */

-static INLINE float

-util_fast_exp2(float x)

-{

-   int32_t ipart;

-   float fpart, mpart;

-   union fi epart;

-

-   if(x > 129.00000f)

-      return 3.402823466e+38f;

-

-   if (x < -126.99999f)

-      return 0.0f;

-

-   ipart = (int32_t) x;

-   fpart = x - (float) ipart;

-

-   /* same as

-    *   epart.f = (float) (1 << ipart)

-    * but faster and without integer overflow for ipart > 31

-    */

-   epart.i = (ipart + 127 ) << 23;

-

-   mpart = pow2_table[POW2_TABLE_OFFSET + (int)(fpart * POW2_TABLE_SCALE)];

-

-   return epart.f * mpart;

-}

-

-

-/**

- * Fast approximation to exp(x).

- */

-static INLINE float

-util_fast_exp(float x)

-{

-   const float k = 1.44269f; /* = log2(e) */

-   return util_fast_exp2(k * x);

-}

-

-

-#define LOG2_TABLE_SIZE_LOG2 16

-#define LOG2_TABLE_SCALE (1 << LOG2_TABLE_SIZE_LOG2)

-#define LOG2_TABLE_SIZE (LOG2_TABLE_SCALE + 1)

-extern float log2_table[LOG2_TABLE_SIZE];

-

-

-/**

- * Fast approximation to log2(x).

- */

-static INLINE float

-util_fast_log2(float x)

-{

-   union fi num;

-   float epart, mpart;

-   num.f = x;

-   epart = (float)(((num.i & 0x7f800000) >> 23) - 127);

-   /* mpart = log2_table[mantissa*LOG2_TABLE_SCALE + 0.5] */

-   mpart = log2_table[((num.i & 0x007fffff) + (1 << (22 - LOG2_TABLE_SIZE_LOG2))) >> (23 - LOG2_TABLE_SIZE_LOG2)];

-   return epart + mpart;

-}

-

-

-/**

- * Fast approximation to x^y.

- */

-static INLINE float

-util_fast_pow(float x, float y)

-{

-   return util_fast_exp2(util_fast_log2(x) * y);

-}

-

-/* Note that this counts zero as a power of two.

- */

-static INLINE boolean

-util_is_power_of_two( unsigned v )

-{

-   return (v & (v-1)) == 0;

-}

-

-

-/**

- * Floor(x), returned as int.

- */

-static INLINE int

-util_ifloor(float f)

-{

-   int ai, bi;

-   double af, bf;

-   union fi u;

-   af = (3 << 22) + 0.5 + (double) f;

-   bf = (3 << 22) + 0.5 - (double) f;

-   u.f = (float) af;  ai = u.i;

-   u.f = (float) bf;  bi = u.i;

-   return (ai - bi) >> 1;

-}

-

-

-/**

- * Round float to nearest int.

- */

-static INLINE int

-util_iround(float f)

-{

-#if defined(PIPE_CC_GCC) && defined(PIPE_ARCH_X86) 

-   int r;

-   __asm__ ("fistpl %0" : "=m" (r) : "t" (f) : "st");

-   return r;

-#elif defined(PIPE_CC_MSVC) && defined(PIPE_ARCH_X86)

-   int r;

-   _asm {

-      fld f

-      fistp r

-   }

-   return r;

-#else

-   if (f >= 0.0f)

-      return (int) (f + 0.5f);

-   else

-      return (int) (f - 0.5f);

-#endif

-}

-

-

-/**

- * Approximate floating point comparison

- */

-static INLINE boolean

-util_is_approx(float a, float b, float tol)

-{

-   return fabs(b - a) <= tol;

-}

-

-

-/**

- * Test if x is NaN or +/- infinity.

- */

-static INLINE boolean

-util_is_inf_or_nan(float x)

-{

-   union fi tmp;

-   tmp.f = x;

-   return !(int)((unsigned int)((tmp.i & 0x7fffffff)-0x7f800000) >> 31);

-}

-

-

-/**

- * Find first bit set in word.  Least significant bit is 1.

- * Return 0 if no bits set.

- */

-#if defined(_MSC_VER) && _MSC_VER >= 1300 && (_M_IX86 || _M_AMD64 || _M_IA64)

-unsigned char _BitScanForward(unsigned long* Index, unsigned long Mask);

-#pragma intrinsic(_BitScanForward)

-static INLINE

-unsigned long ffs( unsigned long u )

-{

-   unsigned long i;

-   if (_BitScanForward(&i, u))

-      return i + 1;

-   else

-      return 0;

-}

-#elif defined(PIPE_CC_MSVC) && defined(PIPE_ARCH_X86)

-static INLINE

-unsigned ffs( unsigned u )

-{

-   unsigned i;

-

-   if (u == 0) {

-      return 0;

-   }

-

-   __asm bsf eax, [u]

-   __asm inc eax

-   __asm mov [i], eax

-

-   return i;

-}

-#elif defined(__MINGW32__)

-#define ffs __builtin_ffs

-#endif

-

-#ifdef __MINGW32__

-#define ffs __builtin_ffs

-#endif

-

-

-/* Could also binary search for the highest bit.

- */

-static INLINE unsigned

-util_unsigned_logbase2(unsigned n)

-{

-   unsigned log2 = 0;

-   while (n >>= 1)

-      ++log2;

-   return log2;

-}

-

-

-/**

- * Return float bits.

- */

-static INLINE unsigned

-fui( float f )

-{

-   union fi fi;

-   fi.f = f;

-   return fi.ui;

-}

-

-

-/**

- * Convert ubyte to float in [0, 1].

- * XXX a 256-entry lookup table would be slightly faster.

- */

-static INLINE float

-ubyte_to_float(ubyte ub)

-{

-   return (float) ub * (1.0f / 255.0f);

-}

-

-

-/**

- * Convert float in [0,1] to ubyte in [0,255] with clamping.

- */

-static INLINE ubyte

-float_to_ubyte(float f)

-{

-   const int ieee_0996 = 0x3f7f0000;   /* 0.996 or so */

-   union fi tmp;

-

-   tmp.f = f;

-   if (tmp.i < 0) {

-      return (ubyte) 0;

-   }

-   else if (tmp.i >= ieee_0996) {

-      return (ubyte) 255;

-   }

-   else {

-      tmp.f = tmp.f * (255.0f/256.0f) + 32768.0f;

-      return (ubyte) tmp.i;

-   }

-}

-

-static INLINE float

-byte_to_float_tex(int8_t b)

-{

-   return (b == -128) ? -1.0F : b * 1.0F / 127.0F;

-}

-

-static INLINE int8_t

-float_to_byte_tex(float f)

-{

-   return (int8_t) (127.0F * f);

-}

-

-/**

- * Calc log base 2

- */

-static INLINE unsigned

-util_logbase2(unsigned n)

-{

-   unsigned log2 = 0;

-   while (n >>= 1)

-      ++log2;

-   return log2;

-}

-

-

-/**

- * Returns the smallest power of two >= x

- */

-static INLINE unsigned

-util_next_power_of_two(unsigned x)

-{

-   unsigned i;

-

-   if (x == 0)

-      return 1;

-

-   --x;

-

-   for (i = 1; i < sizeof(unsigned) * 8; i <<= 1)

-      x |= x >> i;

-

-   return x + 1;

-}

-

-

-/**

- * Return number of bits set in n.

- */

-static INLINE unsigned

-util_bitcount(unsigned n)

-{

-#if defined(PIPE_CC_GCC)

-   return __builtin_popcount(n);

-#else

-   /* K&R classic bitcount.

-    *

-    * For each iteration, clear the LSB from the bitfield.

-    * Requires only one iteration per set bit, instead of

-    * one iteration per bit less than highest set bit.

-    */

-   unsigned bits = 0;

-   for (bits; n; bits++) {

-      n &= n - 1;

-   }

-   return bits;

-#endif

-}

-

-

-/**

- * Reverse byte order of a 32 bit word.

- */

-static INLINE uint32_t

-util_bswap32(uint32_t n)

-{

-#if defined(PIPE_CC_GCC) && (PIPE_CC_GCC_VERSION >= 403)

-   return __builtin_bswap32(n);

-#else

-   return (n >> 24) |

-          ((n >> 8) & 0x0000ff00) |

-          ((n << 8) & 0x00ff0000) |

-          (n << 24);

-#endif

-}

-

-

-/**

- * Reverse byte order of a 16 bit word.

- */

-static INLINE uint16_t

-util_bswap16(uint16_t n)

-{

-   return (n >> 8) |

-          (n << 8);

-}

-

-

-/**

- * Clamp X to [MIN, MAX].

- * This is a macro to allow float, int, uint, etc. types.

- */

-#define CLAMP( X, MIN, MAX )  ( (X)<(MIN) ? (MIN) : ((X)>(MAX) ? (MAX) : (X)) )

-

-#define MIN2( A, B )   ( (A)<(B) ? (A) : (B) )

-#define MAX2( A, B )   ( (A)>(B) ? (A) : (B) )

-

-#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 ) ((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))

-

-

-/**

- * Align a value, only works pot alignemnts.

- */

-static INLINE int

-align(int value, int alignment)

-{

-   return (value + alignment - 1) & ~(alignment - 1);

-}

-

-/**

- * Works like align but on npot alignments.

- */

-static INLINE size_t

-util_align_npot(size_t value, size_t alignment)

-{

-   if (value % alignment)

-      return value + (alignment - (value % alignment));

-   return value;

-}

-

-static INLINE unsigned

-u_minify(unsigned value, unsigned levels)

-{

-    return MAX2(1, value >> levels);

-}

-

-#ifndef COPY_4V

-#define COPY_4V( DST, SRC )         \

-do {                                \

-   (DST)[0] = (SRC)[0];             \

-   (DST)[1] = (SRC)[1];             \

-   (DST)[2] = (SRC)[2];             \

-   (DST)[3] = (SRC)[3];             \

-} while (0)

-#endif

-

-

-#ifndef COPY_4FV

-#define COPY_4FV( DST, SRC )  COPY_4V(DST, SRC)

-#endif

-

-

-#ifndef ASSIGN_4V

-#define ASSIGN_4V( DST, V0, V1, V2, V3 ) \

-do {                                     \

-   (DST)[0] = (V0);                      \

-   (DST)[1] = (V1);                      \

-   (DST)[2] = (V2);                      \

-   (DST)[3] = (V3);                      \

-} while (0)

-#endif

-

-

-static INLINE uint32_t util_unsigned_fixed(float value, unsigned frac_bits)

-{

-   return value < 0 ? 0 : (uint32_t)(value * (1<<frac_bits));

-}

-

-static INLINE int32_t util_signed_fixed(float value, unsigned frac_bits)

-{

-   return (int32_t)(value * (1<<frac_bits));

-}

-

-

-

-#ifdef __cplusplus

-}

-#endif

-

-#endif /* U_MATH_H */

+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * 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, sub license, 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 NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS 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.
+ * 
+ **************************************************************************/
+
+
+/**
+ * Math utilities and approximations for common math functions.
+ * Reduced precision is usually acceptable in shaders...
+ *
+ * "fast" is used in the names of functions which are low-precision,
+ * or at least lower-precision than the normal C lib functions.
+ */
+
+
+#ifndef U_MATH_H
+#define U_MATH_H
+
+
+#include "pipe/p_compiler.h"
+#include "util/u_debug.h"
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+
+#if defined(PIPE_SUBSYSTEM_WINDOWS_MINIPORT)
+__inline double ceil(double val)
+{
+   double ceil_val;
+
+   if ((val - (long) val) == 0) {
+      ceil_val = val;
+   }
+   else {
+      if (val > 0) {
+         ceil_val = (long) val + 1;
+      }
+      else {
+         ceil_val = (long) val;
+      }
+   }
+
+   return ceil_val;
+}
+
+#ifndef PIPE_SUBSYSTEM_WINDOWS_CE_OGL
+__inline double floor(double val)
+{
+   double floor_val;
+
+   if ((val - (long) val) == 0) {
+      floor_val = val;
+   }
+   else {
+      if (val > 0) {
+         floor_val = (long) val;
+      }
+      else {
+         floor_val = (long) val - 1;
+      }
+   }
+
+   return floor_val;
+}
+#endif
+
+#pragma function(pow)
+__inline double __cdecl pow(double val, double exponent)
+{
+   /* XXX */
+   assert(0);
+   return 0;
+}
+
+#pragma function(log)
+__inline double __cdecl log(double val)
+{
+   /* XXX */
+   assert(0);
+   return 0;
+}
+
+#pragma function(atan2)
+__inline double __cdecl atan2(double val)
+{
+   /* XXX */
+   assert(0);
+   return 0;
+}
+#else
+#include <math.h>
+#include <stdarg.h>
+#endif
+
+
+#ifndef M_SQRT2
+#define M_SQRT2 1.41421356237309504880
+#endif
+
+
+#if defined(_MSC_VER) 
+
+#if _MSC_VER < 1400 && !defined(__cplusplus) || defined(PIPE_SUBSYSTEM_WINDOWS_CE)
+ 
+static INLINE float cosf( float f ) 
+{
+   return (float) cos( (double) f );
+}
+
+static INLINE float sinf( float f ) 
+{
+   return (float) sin( (double) f );
+}
+
+static INLINE float ceilf( float f ) 
+{
+   return (float) ceil( (double) f );
+}
+
+static INLINE float floorf( float f ) 
+{
+   return (float) floor( (double) f );
+}
+
+static INLINE float powf( float f, float g ) 
+{
+   return (float) pow( (double) f, (double) g );
+}
+
+static INLINE float sqrtf( float f ) 
+{
+   return (float) sqrt( (double) f );
+}
+
+static INLINE float fabsf( float f ) 
+{
+   return (float) fabs( (double) f );
+}
+
+static INLINE float logf( float f ) 
+{
+   return (float) log( (double) f );
+}
+
+#else
+/* Work-around an extra semi-colon in VS 2005 logf definition */
+#ifdef logf
+#undef logf
+#define logf(x) ((float)log((double)(x)))
+#endif /* logf */
+
+#define isfinite(x) _finite((double)(x))
+#define isnan(x) _isnan((double)(x))
+#endif /* _MSC_VER < 1400 && !defined(__cplusplus) */
+
+static INLINE double log2( double x )
+{
+   const double invln2 = 1.442695041;
+   return log( x ) * invln2;
+}
+
+static INLINE double
+round(double x)
+{
+   return x >= 0.0 ? floor(x + 0.5) : ceil(x - 0.5);
+}
+
+static INLINE float
+roundf(float x)
+{
+   return x >= 0.0f ? floorf(x + 0.5f) : ceilf(x - 0.5f);
+}
+
+#endif /* _MSC_VER */
+
+
+
+
+
+#define POW2_TABLE_SIZE_LOG2 9
+#define POW2_TABLE_SIZE (1 << POW2_TABLE_SIZE_LOG2)
+#define POW2_TABLE_OFFSET (POW2_TABLE_SIZE/2)
+#define POW2_TABLE_SCALE ((float)(POW2_TABLE_SIZE/2))
+extern float pow2_table[POW2_TABLE_SIZE];
+
+
+/**
+ * Initialize math module.  This should be called before using any
+ * other functions in this module.
+ */
+extern void
+util_init_math(void);
+
+
+union fi {
+   float f;
+   int32_t i;
+   uint32_t ui;
+};
+
+
+/**
+ * Fast version of 2^x
+ * Identity: exp2(a + b) = exp2(a) * exp2(b)
+ * Let ipart = int(x)
+ * Let fpart = x - ipart;
+ * So, exp2(x) = exp2(ipart) * exp2(fpart)
+ * Compute exp2(ipart) with i << ipart
+ * Compute exp2(fpart) with lookup table.
+ */
+static INLINE float
+util_fast_exp2(float x)
+{
+   int32_t ipart;
+   float fpart, mpart;
+   union fi epart;
+
+   if(x > 129.00000f)
+      return 3.402823466e+38f;
+
+   if (x < -126.99999f)
+      return 0.0f;
+
+   ipart = (int32_t) x;
+   fpart = x - (float) ipart;
+
+   /* same as
+    *   epart.f = (float) (1 << ipart)
+    * but faster and without integer overflow for ipart > 31
+    */
+   epart.i = (ipart + 127 ) << 23;
+
+   mpart = pow2_table[POW2_TABLE_OFFSET + (int)(fpart * POW2_TABLE_SCALE)];
+
+   return epart.f * mpart;
+}
+
+
+/**
+ * Fast approximation to exp(x).
+ */
+static INLINE float
+util_fast_exp(float x)
+{
+   const float k = 1.44269f; /* = log2(e) */
+   return util_fast_exp2(k * x);
+}
+
+
+#define LOG2_TABLE_SIZE_LOG2 16
+#define LOG2_TABLE_SCALE (1 << LOG2_TABLE_SIZE_LOG2)
+#define LOG2_TABLE_SIZE (LOG2_TABLE_SCALE + 1)
+extern float log2_table[LOG2_TABLE_SIZE];
+
+
+/**
+ * Fast approximation to log2(x).
+ */
+static INLINE float
+util_fast_log2(float x)
+{
+   union fi num;
+   float epart, mpart;
+   num.f = x;
+   epart = (float)(((num.i & 0x7f800000) >> 23) - 127);
+   /* mpart = log2_table[mantissa*LOG2_TABLE_SCALE + 0.5] */
+   mpart = log2_table[((num.i & 0x007fffff) + (1 << (22 - LOG2_TABLE_SIZE_LOG2))) >> (23 - LOG2_TABLE_SIZE_LOG2)];
+   return epart + mpart;
+}
+
+
+/**
+ * Fast approximation to x^y.
+ */
+static INLINE float
+util_fast_pow(float x, float y)
+{
+   return util_fast_exp2(util_fast_log2(x) * y);
+}
+
+/* Note that this counts zero as a power of two.
+ */
+static INLINE boolean
+util_is_power_of_two( unsigned v )
+{
+   return (v & (v-1)) == 0;
+}
+
+
+/**
+ * Floor(x), returned as int.
+ */
+static INLINE int
+util_ifloor(float f)
+{
+   int ai, bi;
+   double af, bf;
+   union fi u;
+   af = (3 << 22) + 0.5 + (double) f;
+   bf = (3 << 22) + 0.5 - (double) f;
+   u.f = (float) af;  ai = u.i;
+   u.f = (float) bf;  bi = u.i;
+   return (ai - bi) >> 1;
+}
+
+
+/**
+ * Round float to nearest int.
+ */
+static INLINE int
+util_iround(float f)
+{
+#if defined(PIPE_CC_GCC) && defined(PIPE_ARCH_X86) 
+   int r;
+   __asm__ ("fistpl %0" : "=m" (r) : "t" (f) : "st");
+   return r;
+#elif defined(PIPE_CC_MSVC) && defined(PIPE_ARCH_X86)
+   int r;
+   _asm {
+      fld f
+      fistp r
+   }
+   return r;
+#else
+   if (f >= 0.0f)
+      return (int) (f + 0.5f);
+   else
+      return (int) (f - 0.5f);
+#endif
+}
+
+
+/**
+ * Approximate floating point comparison
+ */
+static INLINE boolean
+util_is_approx(float a, float b, float tol)
+{
+   return fabs(b - a) <= tol;
+}
+
+
+/**
+ * Test if x is NaN or +/- infinity.
+ */
+static INLINE boolean
+util_is_inf_or_nan(float x)
+{
+   union fi tmp;
+   tmp.f = x;
+   return !(int)((unsigned int)((tmp.i & 0x7fffffff)-0x7f800000) >> 31);
+}
+
+
+/**
+ * Find first bit set in word.  Least significant bit is 1.
+ * Return 0 if no bits set.
+ */
+#if defined(_MSC_VER) && _MSC_VER >= 1300 && (_M_IX86 || _M_AMD64 || _M_IA64)
+unsigned char _BitScanForward(unsigned long* Index, unsigned long Mask);
+#pragma intrinsic(_BitScanForward)
+static INLINE
+unsigned long ffs( unsigned long u )
+{
+   unsigned long i;
+   if (_BitScanForward(&i, u))
+      return i + 1;
+   else
+      return 0;
+}
+#elif defined(PIPE_CC_MSVC) && defined(PIPE_ARCH_X86)
+static INLINE
+unsigned ffs( unsigned u )
+{
+   unsigned i;
+
+   if (u == 0) {
+      return 0;
+   }
+
+   __asm bsf eax, [u]
+   __asm inc eax
+   __asm mov [i], eax
+
+   return i;
+}
+#elif defined(__MINGW32__)
+#define ffs __builtin_ffs
+#endif
+
+
+/**
+ * Return float bits.
+ */
+static INLINE unsigned
+fui( float f )
+{
+   union fi fi;
+   fi.f = f;
+   return fi.ui;
+}
+
+
+/**
+ * Convert ubyte to float in [0, 1].
+ * XXX a 256-entry lookup table would be slightly faster.
+ */
+static INLINE float
+ubyte_to_float(ubyte ub)
+{
+   return (float) ub * (1.0f / 255.0f);
+}
+
+
+/**
+ * Convert float in [0,1] to ubyte in [0,255] with clamping.
+ */
+static INLINE ubyte
+float_to_ubyte(float f)
+{
+   const int ieee_0996 = 0x3f7f0000;   /* 0.996 or so */
+   union fi tmp;
+
+   tmp.f = f;
+   if (tmp.i < 0) {
+      return (ubyte) 0;
+   }
+   else if (tmp.i >= ieee_0996) {
+      return (ubyte) 255;
+   }
+   else {
+      tmp.f = tmp.f * (255.0f/256.0f) + 32768.0f;
+      return (ubyte) tmp.i;
+   }
+}
+
+static INLINE float
+byte_to_float_tex(int8_t b)
+{
+   return (b == -128) ? -1.0F : b * 1.0F / 127.0F;
+}
+
+static INLINE int8_t
+float_to_byte_tex(float f)
+{
+   return (int8_t) (127.0F * f);
+}
+
+/**
+ * Calc log base 2
+ */
+static INLINE unsigned
+util_logbase2(unsigned n)
+{
+   unsigned log2 = 0;
+   while (n >>= 1)
+      ++log2;
+   return log2;
+}
+
+
+/**
+ * Returns the smallest power of two >= x
+ */
+static INLINE unsigned
+util_next_power_of_two(unsigned x)
+{
+   unsigned i;
+
+   if (x == 0)
+      return 1;
+
+   --x;
+
+   for (i = 1; i < sizeof(unsigned) * 8; i <<= 1)
+      x |= x >> i;
+
+   return x + 1;
+}
+
+
+/**
+ * Return number of bits set in n.
+ */
+static INLINE unsigned
+util_bitcount(unsigned n)
+{
+#if defined(PIPE_CC_GCC)
+   return __builtin_popcount(n);
+#else
+   /* K&R classic bitcount.
+    *
+    * For each iteration, clear the LSB from the bitfield.
+    * Requires only one iteration per set bit, instead of
+    * one iteration per bit less than highest set bit.
+    */
+   unsigned bits = 0;
+   for (bits; n; bits++) {
+      n &= n - 1;
+   }
+   return bits;
+#endif
+}
+
+
+/**
+ * Reverse byte order of a 32 bit word.
+ */
+static INLINE uint32_t
+util_bswap32(uint32_t n)
+{
+#if defined(PIPE_CC_GCC) && (PIPE_CC_GCC_VERSION >= 403)
+   return __builtin_bswap32(n);
+#else
+   return (n >> 24) |
+          ((n >> 8) & 0x0000ff00) |
+          ((n << 8) & 0x00ff0000) |
+          (n << 24);
+#endif
+}
+
+
+/**
+ * Reverse byte order of a 16 bit word.
+ */
+static INLINE uint16_t
+util_bswap16(uint16_t n)
+{
+   return (n >> 8) |
+          (n << 8);
+}
+
+
+/**
+ * Clamp X to [MIN, MAX].
+ * This is a macro to allow float, int, uint, etc. types.
+ */
+#define CLAMP( X, MIN, MAX )  ( (X)<(MIN) ? (MIN) : ((X)>(MAX) ? (MAX) : (X)) )
+
+#define MIN2( A, B )   ( (A)<(B) ? (A) : (B) )
+#define MAX2( A, B )   ( (A)>(B) ? (A) : (B) )
+
+#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 ) ((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))
+
+
+/**
+ * Align a value, only works pot alignemnts.
+ */
+static INLINE int
+align(int value, int alignment)
+{
+   return (value + alignment - 1) & ~(alignment - 1);
+}
+
+/**
+ * Works like align but on npot alignments.
+ */
+static INLINE size_t
+util_align_npot(size_t value, size_t alignment)
+{
+   if (value % alignment)
+      return value + (alignment - (value % alignment));
+   return value;
+}
+
+static INLINE unsigned
+u_minify(unsigned value, unsigned levels)
+{
+    return MAX2(1, value >> levels);
+}
+
+#ifndef COPY_4V
+#define COPY_4V( DST, SRC )         \
+do {                                \
+   (DST)[0] = (SRC)[0];             \
+   (DST)[1] = (SRC)[1];             \
+   (DST)[2] = (SRC)[2];             \
+   (DST)[3] = (SRC)[3];             \
+} while (0)
+#endif
+
+
+#ifndef COPY_4FV
+#define COPY_4FV( DST, SRC )  COPY_4V(DST, SRC)
+#endif
+
+
+#ifndef ASSIGN_4V
+#define ASSIGN_4V( DST, V0, V1, V2, V3 ) \
+do {                                     \
+   (DST)[0] = (V0);                      \
+   (DST)[1] = (V1);                      \
+   (DST)[2] = (V2);                      \
+   (DST)[3] = (V3);                      \
+} while (0)
+#endif
+
+
+static INLINE uint32_t util_unsigned_fixed(float value, unsigned frac_bits)
+{
+   return value < 0 ? 0 : (uint32_t)(value * (1<<frac_bits));
+}
+
+static INLINE int32_t util_signed_fixed(float value, unsigned frac_bits)
+{
+   return (int32_t)(value * (1<<frac_bits));
+}
+
+
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* U_MATH_H */
diff --git a/mesalib/src/gallium/auxiliary/util/u_vbuf_mgr.c b/mesalib/src/gallium/auxiliary/util/u_vbuf_mgr.c
index 9306fcd5c..a034483ee 100644
--- a/mesalib/src/gallium/auxiliary/util/u_vbuf_mgr.c
+++ b/mesalib/src/gallium/auxiliary/util/u_vbuf_mgr.c
@@ -1,613 +1,645 @@
-/**************************************************************************

- *

- * Copyright 2011 Marek Olšák <maraeo@gmail.com>

- * All Rights Reserved.

- *

- * 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, sub license, 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 NON-INFRINGEMENT.

- * IN NO EVENT SHALL AUTHORS AND/OR ITS SUPPLIERS 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.

- *

- **************************************************************************/

-

-#include "util/u_vbuf_mgr.h"

-

-#include "util/u_format.h"

-#include "util/u_inlines.h"

-#include "util/u_memory.h"

-#include "util/u_upload_mgr.h"

-#include "translate/translate.h"

-#include "translate/translate_cache.h"

-

-/* Hardware vertex fetcher limitations can be described by this structure. */

-struct u_vbuf_caps {

-   /* Vertex format CAPs. */

-   /* TRUE if hardware supports it. */

-   unsigned format_fixed32:1;    /* PIPE_FORMAT_*32*_FIXED */

-   unsigned format_float16:1;    /* PIPE_FORMAT_*16*_FLOAT */

-   unsigned format_float64:1;    /* PIPE_FORMAT_*64*_FLOAT */

-   unsigned format_norm32:1;     /* PIPE_FORMAT_*32*NORM */

-   unsigned format_scaled32:1;   /* PIPE_FORMAT_*32*SCALED */

-

-   /* Whether vertex fetches don't have to be dword-aligned. */

-   /* TRUE if hardware supports it. */

-   unsigned fetch_dword_unaligned:1;

-};

-

-struct u_vbuf_mgr_elements {

-   unsigned count;

-   struct pipe_vertex_element ve[PIPE_MAX_ATTRIBS];

-

-   /* If (velem[i].src_format != real_format[i]), the vertex buffer

-    * referenced by the vertex element cannot be used for rendering and

-    * its vertex data must be translated to real_format[i]. */

-   enum pipe_format native_format[PIPE_MAX_ATTRIBS];

-   unsigned native_format_size[PIPE_MAX_ATTRIBS];

-

-   /* This might mean two things:

-    * - src_format != native_format, as discussed above.

-    * - src_offset % 4 != 0 (if the caps don't allow such an offset). */

-   boolean incompatible_layout;

-};

-

-struct u_vbuf_mgr_priv {

-   struct u_vbuf_mgr b;

-   struct u_vbuf_caps caps;

-   struct pipe_context *pipe;

-

-   struct translate_cache *translate_cache;

-   unsigned translate_vb_slot;

-

-   struct u_vbuf_mgr_elements *ve;

-   void *saved_ve, *fallback_ve;

-   boolean ve_binding_lock;

-

-   boolean any_user_vbs;

-   boolean incompatible_vb_layout;

-};

-

-static void u_vbuf_mgr_init_format_caps(struct u_vbuf_mgr_priv *mgr)

-{

-   struct pipe_screen *screen = mgr->pipe->screen;

-

-   mgr->caps.format_fixed32 =

-      screen->is_format_supported(screen, PIPE_FORMAT_R32_FIXED, PIPE_BUFFER,

-                                  0, PIPE_BIND_VERTEX_BUFFER);

-

-   mgr->caps.format_float16 =

-      screen->is_format_supported(screen, PIPE_FORMAT_R16_FLOAT, PIPE_BUFFER,

-                                  0, PIPE_BIND_VERTEX_BUFFER);

-

-   mgr->caps.format_float64 =

-      screen->is_format_supported(screen, PIPE_FORMAT_R64_FLOAT, PIPE_BUFFER,

-                                  0, PIPE_BIND_VERTEX_BUFFER);

-

-   mgr->caps.format_norm32 =

-      screen->is_format_supported(screen, PIPE_FORMAT_R32_UNORM, PIPE_BUFFER,

-                                  0, PIPE_BIND_VERTEX_BUFFER) &&

-      screen->is_format_supported(screen, PIPE_FORMAT_R32_SNORM, PIPE_BUFFER,

-                                  0, PIPE_BIND_VERTEX_BUFFER);

-

-   mgr->caps.format_scaled32 =

-      screen->is_format_supported(screen, PIPE_FORMAT_R32_USCALED, PIPE_BUFFER,

-                                  0, PIPE_BIND_VERTEX_BUFFER) &&

-      screen->is_format_supported(screen, PIPE_FORMAT_R32_SSCALED, PIPE_BUFFER,

-                                  0, PIPE_BIND_VERTEX_BUFFER);

-}

-

-struct u_vbuf_mgr *

-u_vbuf_mgr_create(struct pipe_context *pipe,

-                  unsigned upload_buffer_size,

-                  unsigned upload_buffer_alignment,

-                  unsigned upload_buffer_bind,

-                  enum u_fetch_alignment fetch_alignment)

-{

-   struct u_vbuf_mgr_priv *mgr = CALLOC_STRUCT(u_vbuf_mgr_priv);

-

-   mgr->pipe = pipe;

-   mgr->translate_cache = translate_cache_create();

-

-   mgr->b.uploader = u_upload_create(pipe, upload_buffer_size,

-                                     upload_buffer_alignment,

-                                     upload_buffer_bind);

-

-   mgr->caps.fetch_dword_unaligned =

-         fetch_alignment == U_VERTEX_FETCH_BYTE_ALIGNED;

-

-   u_vbuf_mgr_init_format_caps(mgr);

-

-   return &mgr->b;

-}

-

-void u_vbuf_mgr_destroy(struct u_vbuf_mgr *mgrb)

-{

-   struct u_vbuf_mgr_priv *mgr = (struct u_vbuf_mgr_priv*)mgrb;

-   unsigned i;

-

-   for (i = 0; i < mgr->b.nr_real_vertex_buffers; i++) {

-      pipe_resource_reference(&mgr->b.vertex_buffer[i].buffer, NULL);

-      pipe_resource_reference(&mgr->b.real_vertex_buffer[i], NULL);

-   }

-

-   translate_cache_destroy(mgr->translate_cache);

-   u_upload_destroy(mgr->b.uploader);

-   FREE(mgr);

-}

-

-

-static void u_vbuf_translate_begin(struct u_vbuf_mgr_priv *mgr,

-                                   int min_index, int max_index,

-                                   boolean *upload_flushed)

-{

-   struct translate_key key;

-   struct translate_element *te;

-   unsigned tr_elem_index[PIPE_MAX_ATTRIBS];

-   struct translate *tr;

-   boolean vb_translated[PIPE_MAX_ATTRIBS] = {0};

-   uint8_t *vb_map[PIPE_MAX_ATTRIBS] = {0}, *out_map;

-   struct pipe_transfer *vb_transfer[PIPE_MAX_ATTRIBS] = {0};

-   struct pipe_resource *out_buffer = NULL;

-   unsigned i, num_verts, out_offset;

-   struct pipe_vertex_element new_velems[PIPE_MAX_ATTRIBS];

-

-   memset(&key, 0, sizeof(key));

-   memset(tr_elem_index, 0xff, sizeof(tr_elem_index));

-

-   /* Initialize the translate key, i.e. the recipe how vertices should be

-     * translated. */

-   memset(&key, 0, sizeof key);

-   for (i = 0; i < mgr->ve->count; i++) {

-      struct pipe_vertex_buffer *vb =

-            &mgr->b.vertex_buffer[mgr->ve->ve[i].vertex_buffer_index];

-      enum pipe_format output_format = mgr->ve->native_format[i];

-      unsigned output_format_size = mgr->ve->native_format_size[i];

-

-      /* Check for support. */

-      if (mgr->ve->ve[i].src_format == mgr->ve->native_format[i] &&

-          (mgr->caps.fetch_dword_unaligned ||

-           (vb->buffer_offset % 4 == 0 &&

-            vb->stride % 4 == 0 &&

-            mgr->ve->ve[i].src_offset % 4 == 0))) {

-         continue;

-      }

-

-      /* Workaround for translate: output floats instead of halfs. */

-      switch (output_format) {

-      case PIPE_FORMAT_R16_FLOAT:

-         output_format = PIPE_FORMAT_R32_FLOAT;

-         output_format_size = 4;

-         break;

-      case PIPE_FORMAT_R16G16_FLOAT:

-         output_format = PIPE_FORMAT_R32G32_FLOAT;

-         output_format_size = 8;

-         break;

-      case PIPE_FORMAT_R16G16B16_FLOAT:

-         output_format = PIPE_FORMAT_R32G32B32_FLOAT;

-         output_format_size = 12;

-         break;

-      case PIPE_FORMAT_R16G16B16A16_FLOAT:

-         output_format = PIPE_FORMAT_R32G32B32A32_FLOAT;

-         output_format_size = 16;

-         break;

-      default:;

-      }

-

-      /* Add this vertex element. */

-      te = &key.element[key.nr_elements];

-      /*te->type;

-        te->instance_divisor;*/

-      te->input_buffer = mgr->ve->ve[i].vertex_buffer_index;

-      te->input_format = mgr->ve->ve[i].src_format;

-      te->input_offset = mgr->ve->ve[i].src_offset;

-      te->output_format = output_format;

-      te->output_offset = key.output_stride;

-

-      key.output_stride += output_format_size;

-      vb_translated[mgr->ve->ve[i].vertex_buffer_index] = TRUE;

-      tr_elem_index[i] = key.nr_elements;

-      key.nr_elements++;

-   }

-

-   /* Get a translate object. */

-   tr = translate_cache_find(mgr->translate_cache, &key);

-

-   /* Map buffers we want to translate. */

-   for (i = 0; i < mgr->b.nr_vertex_buffers; i++) {

-      if (vb_translated[i]) {

-         struct pipe_vertex_buffer *vb = &mgr->b.vertex_buffer[i];

-

-         vb_map[i] = pipe_buffer_map(mgr->pipe, vb->buffer,

-                                     PIPE_TRANSFER_READ, &vb_transfer[i]);

-

-         tr->set_buffer(tr, i,

-                        vb_map[i] + vb->buffer_offset + vb->stride * min_index,

-                        vb->stride, ~0);

-      }

-   }

-

-   /* Create and map the output buffer. */

-   num_verts = max_index + 1 - min_index;

-

-   u_upload_alloc(mgr->b.uploader,

-                  key.output_stride * min_index,

-                  key.output_stride * num_verts,

-                  &out_offset, &out_buffer, upload_flushed,

-                  (void**)&out_map);

-

-   out_offset -= key.output_stride * min_index;

-

-   /* Translate. */

-   tr->run(tr, 0, num_verts, 0, out_map);

-

-   /* Unmap all buffers. */

-   for (i = 0; i < mgr->b.nr_vertex_buffers; i++) {

-      if (vb_translated[i]) {

-         pipe_buffer_unmap(mgr->pipe, vb_transfer[i]);

-      }

-   }

-

-   /* Setup the new vertex buffer in the first free slot. */

-   mgr->translate_vb_slot = ~0;

-   for (i = 0; i < PIPE_MAX_ATTRIBS; i++) {

-      if (!mgr->b.vertex_buffer[i].buffer) {

-         mgr->translate_vb_slot = i;

-

-         if (i >= mgr->b.nr_vertex_buffers) {

-            mgr->b.nr_real_vertex_buffers = i+1;

-         }

-         break;

-      }

-   }

-

-   if (mgr->translate_vb_slot != ~0) {

-      /* Setup the new vertex buffer. */

-      pipe_resource_reference(

-            &mgr->b.real_vertex_buffer[mgr->translate_vb_slot], out_buffer);

-      mgr->b.vertex_buffer[mgr->translate_vb_slot].buffer_offset = out_offset;

-      mgr->b.vertex_buffer[mgr->translate_vb_slot].stride = key.output_stride;

-

-      /* Setup new vertex elements. */

-      for (i = 0; i < mgr->ve->count; i++) {

-         if (tr_elem_index[i] < key.nr_elements) {

-            te = &key.element[tr_elem_index[i]];

-            new_velems[i].instance_divisor = mgr->ve->ve[i].instance_divisor;

-            new_velems[i].src_format = te->output_format;

-            new_velems[i].src_offset = te->output_offset;

-            new_velems[i].vertex_buffer_index = mgr->translate_vb_slot;

-         } else {

-            memcpy(&new_velems[i], &mgr->ve->ve[i],

-                   sizeof(struct pipe_vertex_element));

-         }

-      }

-

-      mgr->fallback_ve =

-            mgr->pipe->create_vertex_elements_state(mgr->pipe, mgr->ve->count,

-                                                    new_velems);

-

-      /* Preserve saved_ve. */

-      mgr->ve_binding_lock = TRUE;

-      mgr->pipe->bind_vertex_elements_state(mgr->pipe, mgr->fallback_ve);

-      mgr->ve_binding_lock = FALSE;

-   }

-

-   pipe_resource_reference(&out_buffer, NULL);

-}

-

-static void u_vbuf_translate_end(struct u_vbuf_mgr_priv *mgr)

-{

-   if (mgr->fallback_ve == NULL) {

-      return;

-   }

-

-   /* Restore vertex elements. */

-   /* Note that saved_ve will be overwritten in bind_vertex_elements_state. */

-   mgr->pipe->bind_vertex_elements_state(mgr->pipe, mgr->saved_ve);

-   mgr->pipe->delete_vertex_elements_state(mgr->pipe, mgr->fallback_ve);

-   mgr->fallback_ve = NULL;

-

-   /* Delete the now-unused VBO. */

-   pipe_resource_reference(&mgr->b.real_vertex_buffer[mgr->translate_vb_slot],

-                           NULL);

-   mgr->b.nr_real_vertex_buffers = mgr->b.nr_vertex_buffers;

-}

-

-#define FORMAT_REPLACE(what, withwhat) \

-    case PIPE_FORMAT_##what: format = PIPE_FORMAT_##withwhat; break

-

-struct u_vbuf_mgr_elements *

-u_vbuf_mgr_create_vertex_elements(struct u_vbuf_mgr *mgrb,

-                                  unsigned count,

-                                  const struct pipe_vertex_element *attribs,

-                                  struct pipe_vertex_element *native_attribs)

-{

-   struct u_vbuf_mgr_priv *mgr = (struct u_vbuf_mgr_priv*)mgrb;

-   unsigned i;

-   struct u_vbuf_mgr_elements *ve = CALLOC_STRUCT(u_vbuf_mgr_elements);

-

-   ve->count = count;

-

-   if (!count) {

-      return ve;

-   }

-

-   memcpy(ve->ve, attribs, sizeof(struct pipe_vertex_element) * count);

-   memcpy(native_attribs, attribs, sizeof(struct pipe_vertex_element) * count);

-

-   /* Set the best native format in case the original format is not

-    * supported. */

-   for (i = 0; i < count; i++) {

-      enum pipe_format format = ve->ve[i].src_format;

-

-      /* Choose a native format.

-       * For now we don't care about the alignment, that's going to

-       * be sorted out later. */

-      if (!mgr->caps.format_fixed32) {

-         switch (format) {

-            FORMAT_REPLACE(R32_FIXED,           R32_FLOAT);

-            FORMAT_REPLACE(R32G32_FIXED,        R32G32_FLOAT);

-            FORMAT_REPLACE(R32G32B32_FIXED,     R32G32B32_FLOAT);

-            FORMAT_REPLACE(R32G32B32A32_FIXED,  R32G32B32A32_FLOAT);

-            default:;

-         }

-      }

-      if (!mgr->caps.format_float16) {

-         switch (format) {

-            FORMAT_REPLACE(R16_FLOAT,           R32_FLOAT);

-            FORMAT_REPLACE(R16G16_FLOAT,        R32G32_FLOAT);

-            FORMAT_REPLACE(R16G16B16_FLOAT,     R32G32B32_FLOAT);

-            FORMAT_REPLACE(R16G16B16A16_FLOAT,  R32G32B32A32_FLOAT);

-            default:;

-         }

-      }

-      if (!mgr->caps.format_float64) {

-         switch (format) {

-            FORMAT_REPLACE(R64_FLOAT,           R32_FLOAT);

-            FORMAT_REPLACE(R64G64_FLOAT,        R32G32_FLOAT);

-            FORMAT_REPLACE(R64G64B64_FLOAT,     R32G32B32_FLOAT);

-            FORMAT_REPLACE(R64G64B64A64_FLOAT,  R32G32B32A32_FLOAT);

-            default:;

-         }

-      }

-      if (!mgr->caps.format_norm32) {

-         switch (format) {

-            FORMAT_REPLACE(R32_UNORM,           R32_FLOAT);

-            FORMAT_REPLACE(R32G32_UNORM,        R32G32_FLOAT);

-            FORMAT_REPLACE(R32G32B32_UNORM,     R32G32B32_FLOAT);

-            FORMAT_REPLACE(R32G32B32A32_UNORM,  R32G32B32A32_FLOAT);

-            FORMAT_REPLACE(R32_SNORM,           R32_FLOAT);

-            FORMAT_REPLACE(R32G32_SNORM,        R32G32_FLOAT);

-            FORMAT_REPLACE(R32G32B32_SNORM,     R32G32B32_FLOAT);

-            FORMAT_REPLACE(R32G32B32A32_SNORM,  R32G32B32A32_FLOAT);

-            default:;

-         }

-      }

-      if (!mgr->caps.format_scaled32) {

-         switch (format) {

-            FORMAT_REPLACE(R32_USCALED,         R32_FLOAT);

-            FORMAT_REPLACE(R32G32_USCALED,      R32G32_FLOAT);

-            FORMAT_REPLACE(R32G32B32_USCALED,   R32G32B32_FLOAT);

-            FORMAT_REPLACE(R32G32B32A32_USCALED,R32G32B32A32_FLOAT);

-            FORMAT_REPLACE(R32_SSCALED,         R32_FLOAT);

-            FORMAT_REPLACE(R32G32_SSCALED,      R32G32_FLOAT);

-            FORMAT_REPLACE(R32G32B32_SSCALED,   R32G32B32_FLOAT);

-            FORMAT_REPLACE(R32G32B32A32_SSCALED,R32G32B32A32_FLOAT);

-            default:;

-         }

-      }

-

-      native_attribs[i].src_format = format;

-      ve->native_format[i] = format;

-      ve->native_format_size[i] =

-            util_format_get_blocksize(ve->native_format[i]);

-

-      ve->incompatible_layout =

-            ve->incompatible_layout ||

-            ve->ve[i].src_format != ve->native_format[i] ||

-            (!mgr->caps.fetch_dword_unaligned && ve->ve[i].src_offset % 4 != 0);

-   }

-

-   /* Align the formats to the size of DWORD if needed. */

-   if (!mgr->caps.fetch_dword_unaligned) {

-      for (i = 0; i < count; i++) {

-         ve->native_format_size[i] = align(ve->native_format_size[i], 4);

-      }

-   }

-

-   return ve;

-}

-

-void u_vbuf_mgr_bind_vertex_elements(struct u_vbuf_mgr *mgrb,

-                                     void *cso,

-                                     struct u_vbuf_mgr_elements *ve)

-{

-   struct u_vbuf_mgr_priv *mgr = (struct u_vbuf_mgr_priv*)mgrb;

-

-   if (!cso) {

-      return;

-   }

-

-   if (!mgr->ve_binding_lock) {

-      mgr->saved_ve = cso;

-      mgr->ve = ve;

-   }

-}

-

-void u_vbuf_mgr_destroy_vertex_elements(struct u_vbuf_mgr *mgr,

-                                        struct u_vbuf_mgr_elements *ve)

-{

-   FREE(ve);

-}

-

-void u_vbuf_mgr_set_vertex_buffers(struct u_vbuf_mgr *mgrb,

-                                   unsigned count,

-                                   const struct pipe_vertex_buffer *bufs)

-{

-   struct u_vbuf_mgr_priv *mgr = (struct u_vbuf_mgr_priv*)mgrb;

-   unsigned i;

-

-   mgr->b.max_index = ~0;

-   mgr->any_user_vbs = FALSE;

-   mgr->incompatible_vb_layout = FALSE;

-

-   if (!mgr->caps.fetch_dword_unaligned) {

-      /* Check if the strides and offsets are aligned to the size of DWORD. */

-      for (i = 0; i < count; i++) {

-         if (bufs[i].buffer) {

-            if (bufs[i].stride % 4 != 0 ||

-                bufs[i].buffer_offset % 4 != 0) {

-               mgr->incompatible_vb_layout = TRUE;

-               break;

-            }

-         }

-      }

-   }

-

-   for (i = 0; i < count; i++) {

-      const struct pipe_vertex_buffer *vb = &bufs[i];

-

-      pipe_resource_reference(&mgr->b.vertex_buffer[i].buffer, vb->buffer);

-      pipe_resource_reference(&mgr->b.real_vertex_buffer[i], NULL);

-

-      if (u_vbuf_resource(vb->buffer)->user_ptr) {

-         mgr->any_user_vbs = TRUE;

-         continue;

-      }

-

-      pipe_resource_reference(&mgr->b.real_vertex_buffer[i], vb->buffer);

-

-      /* The stride of zero means we will be fetching only the first

-       * vertex, so don't care about max_index. */

-      if (!vb->stride) {

-         continue;

-      }

-

-      /* Update the maximum index. */

-      mgr->b.max_index =

-            MIN2(mgr->b.max_index,

-                 (vb->buffer->width0 - vb->buffer_offset) / vb->stride - 1);

-   }

-

-   for (; i < mgr->b.nr_real_vertex_buffers; i++) {

-      pipe_resource_reference(&mgr->b.vertex_buffer[i].buffer, NULL);

-      pipe_resource_reference(&mgr->b.real_vertex_buffer[i], NULL);

-   }

-

-   memcpy(mgr->b.vertex_buffer, bufs,

-          sizeof(struct pipe_vertex_buffer) * count);

-

-   mgr->b.nr_vertex_buffers = count;

-   mgr->b.nr_real_vertex_buffers = count;

-}

-

-static void u_vbuf_upload_buffers(struct u_vbuf_mgr_priv *mgr,

-                                  int min_index, int max_index,

-                                  unsigned instance_count,

-                                  boolean *upload_flushed)

-{

-   int i, nr = mgr->ve->count;

-   unsigned count = max_index + 1 - min_index;

-   boolean uploaded[PIPE_MAX_ATTRIBS] = {0};

-

-   for (i = 0; i < nr; i++) {

-      unsigned index = mgr->ve->ve[i].vertex_buffer_index;

-      struct pipe_vertex_buffer *vb = &mgr->b.vertex_buffer[index];

-

-      if (vb->buffer &&

-          u_vbuf_resource(vb->buffer)->user_ptr &&

-          !uploaded[index]) {

-         unsigned first, size;

-         boolean flushed;

-         unsigned instance_div = mgr->ve->ve[i].instance_divisor;

-

-         if (instance_div) {

-            first = 0;

-            size = vb->stride *

-                   ((instance_count + instance_div - 1) / instance_div);

-         } else if (vb->stride) {

-            first = vb->stride * min_index;

-            size = vb->stride * count;

-         } else {

-            first = 0;

-            size = mgr->ve->native_format_size[i];

-         }

-

-         u_upload_data(mgr->b.uploader, first, size,

-                       u_vbuf_resource(vb->buffer)->user_ptr + first,

-                       &vb->buffer_offset,

-                       &mgr->b.real_vertex_buffer[index],

-                       &flushed);

-

-         vb->buffer_offset -= first;

-

-         uploaded[index] = TRUE;

-         *upload_flushed = *upload_flushed || flushed;

-      } else {

-         assert(mgr->b.real_vertex_buffer[index]);

-      }

-   }

-}

-

-void u_vbuf_mgr_draw_begin(struct u_vbuf_mgr *mgrb,

-                           const struct pipe_draw_info *info,

-                           boolean *buffers_updated,

-                           boolean *uploader_flushed)

-{

-   struct u_vbuf_mgr_priv *mgr = (struct u_vbuf_mgr_priv*)mgrb;

-   boolean bufs_updated = FALSE, upload_flushed = FALSE;

-   int min_index, max_index;

-

-   min_index = info->min_index - info->index_bias;

-   if (info->max_index == ~0) {

-      max_index = mgr->b.max_index;

-   } else {

-      max_index = MIN2(info->max_index - info->index_bias, mgr->b.max_index);

-   }

-

-   /* Translate vertices with non-native layouts or formats. */

-   if (mgr->incompatible_vb_layout || mgr->ve->incompatible_layout) {

-      u_vbuf_translate_begin(mgr, min_index, max_index, &upload_flushed);

-

-      if (mgr->fallback_ve) {

-         bufs_updated = TRUE;

-      }

-   }

-

-   /* Upload user buffers. */

-   if (mgr->any_user_vbs) {

-      u_vbuf_upload_buffers(mgr, min_index, max_index, info->instance_count,

-                            &upload_flushed);

-      bufs_updated = TRUE;

-   }

-

-   /* Set the return values. */

-   if (buffers_updated) {

-      *buffers_updated = bufs_updated;

-   }

-   if (uploader_flushed) {

-      *uploader_flushed = upload_flushed;

-   }

-}

-

-void u_vbuf_mgr_draw_end(struct u_vbuf_mgr *mgrb)

-{

-   struct u_vbuf_mgr_priv *mgr = (struct u_vbuf_mgr_priv*)mgrb;

-

-   if (mgr->fallback_ve) {

-      u_vbuf_translate_end(mgr);

-   }

-}

+/**************************************************************************
+ *
+ * Copyright 2011 Marek Olšák <maraeo@gmail.com>
+ * All Rights Reserved.
+ *
+ * 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, sub license, 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 NON-INFRINGEMENT.
+ * IN NO EVENT SHALL AUTHORS AND/OR ITS SUPPLIERS 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.
+ *
+ **************************************************************************/
+
+#include "util/u_vbuf_mgr.h"
+
+#include "util/u_format.h"
+#include "util/u_inlines.h"
+#include "util/u_memory.h"
+#include "util/u_upload_mgr.h"
+#include "translate/translate.h"
+#include "translate/translate_cache.h"
+
+/* Hardware vertex fetcher limitations can be described by this structure. */
+struct u_vbuf_caps {
+   /* Vertex format CAPs. */
+   /* TRUE if hardware supports it. */
+   unsigned format_fixed32:1;    /* PIPE_FORMAT_*32*_FIXED */
+   unsigned format_float16:1;    /* PIPE_FORMAT_*16*_FLOAT */
+   unsigned format_float64:1;    /* PIPE_FORMAT_*64*_FLOAT */
+   unsigned format_norm32:1;     /* PIPE_FORMAT_*32*NORM */
+   unsigned format_scaled32:1;   /* PIPE_FORMAT_*32*SCALED */
+
+   /* Whether vertex fetches don't have to be dword-aligned. */
+   /* TRUE if hardware supports it. */
+   unsigned fetch_dword_unaligned:1;
+};
+
+struct u_vbuf_mgr_elements {
+   unsigned count;
+   struct pipe_vertex_element ve[PIPE_MAX_ATTRIBS];
+
+   unsigned src_format_size[PIPE_MAX_ATTRIBS];
+
+   /* If (velem[i].src_format != native_format[i]), the vertex buffer
+    * referenced by the vertex element cannot be used for rendering and
+    * its vertex data must be translated to native_format[i]. */
+   enum pipe_format native_format[PIPE_MAX_ATTRIBS];
+   unsigned native_format_size[PIPE_MAX_ATTRIBS];
+
+   /* This might mean two things:
+    * - src_format != native_format, as discussed above.
+    * - src_offset % 4 != 0 (if the caps don't allow such an offset). */
+   boolean incompatible_layout;
+};
+
+struct u_vbuf_mgr_priv {
+   struct u_vbuf_mgr b;
+   struct u_vbuf_caps caps;
+   struct pipe_context *pipe;
+
+   struct translate_cache *translate_cache;
+   unsigned translate_vb_slot;
+
+   struct u_vbuf_mgr_elements *ve;
+   void *saved_ve, *fallback_ve;
+   boolean ve_binding_lock;
+
+   boolean any_user_vbs;
+   boolean incompatible_vb_layout;
+};
+
+static void u_vbuf_mgr_init_format_caps(struct u_vbuf_mgr_priv *mgr)
+{
+   struct pipe_screen *screen = mgr->pipe->screen;
+
+   mgr->caps.format_fixed32 =
+      screen->is_format_supported(screen, PIPE_FORMAT_R32_FIXED, PIPE_BUFFER,
+                                  0, PIPE_BIND_VERTEX_BUFFER);
+
+   mgr->caps.format_float16 =
+      screen->is_format_supported(screen, PIPE_FORMAT_R16_FLOAT, PIPE_BUFFER,
+                                  0, PIPE_BIND_VERTEX_BUFFER);
+
+   mgr->caps.format_float64 =
+      screen->is_format_supported(screen, PIPE_FORMAT_R64_FLOAT, PIPE_BUFFER,
+                                  0, PIPE_BIND_VERTEX_BUFFER);
+
+   mgr->caps.format_norm32 =
+      screen->is_format_supported(screen, PIPE_FORMAT_R32_UNORM, PIPE_BUFFER,
+                                  0, PIPE_BIND_VERTEX_BUFFER) &&
+      screen->is_format_supported(screen, PIPE_FORMAT_R32_SNORM, PIPE_BUFFER,
+                                  0, PIPE_BIND_VERTEX_BUFFER);
+
+   mgr->caps.format_scaled32 =
+      screen->is_format_supported(screen, PIPE_FORMAT_R32_USCALED, PIPE_BUFFER,
+                                  0, PIPE_BIND_VERTEX_BUFFER) &&
+      screen->is_format_supported(screen, PIPE_FORMAT_R32_SSCALED, PIPE_BUFFER,
+                                  0, PIPE_BIND_VERTEX_BUFFER);
+}
+
+struct u_vbuf_mgr *
+u_vbuf_mgr_create(struct pipe_context *pipe,
+                  unsigned upload_buffer_size,
+                  unsigned upload_buffer_alignment,
+                  unsigned upload_buffer_bind,
+                  enum u_fetch_alignment fetch_alignment)
+{
+   struct u_vbuf_mgr_priv *mgr = CALLOC_STRUCT(u_vbuf_mgr_priv);
+
+   mgr->pipe = pipe;
+   mgr->translate_cache = translate_cache_create();
+
+   mgr->b.uploader = u_upload_create(pipe, upload_buffer_size,
+                                     upload_buffer_alignment,
+                                     upload_buffer_bind);
+
+   mgr->caps.fetch_dword_unaligned =
+         fetch_alignment == U_VERTEX_FETCH_BYTE_ALIGNED;
+
+   u_vbuf_mgr_init_format_caps(mgr);
+
+   return &mgr->b;
+}
+
+void u_vbuf_mgr_destroy(struct u_vbuf_mgr *mgrb)
+{
+   struct u_vbuf_mgr_priv *mgr = (struct u_vbuf_mgr_priv*)mgrb;
+   unsigned i;
+
+   for (i = 0; i < mgr->b.nr_real_vertex_buffers; i++) {
+      pipe_resource_reference(&mgr->b.vertex_buffer[i].buffer, NULL);
+      pipe_resource_reference(&mgr->b.real_vertex_buffer[i], NULL);
+   }
+
+   translate_cache_destroy(mgr->translate_cache);
+   u_upload_destroy(mgr->b.uploader);
+   FREE(mgr);
+}
+
+
+static void u_vbuf_translate_begin(struct u_vbuf_mgr_priv *mgr,
+                                   int min_index, int max_index,
+                                   boolean *upload_flushed)
+{
+   struct translate_key key;
+   struct translate_element *te;
+   unsigned tr_elem_index[PIPE_MAX_ATTRIBS];
+   struct translate *tr;
+   boolean vb_translated[PIPE_MAX_ATTRIBS] = {0};
+   uint8_t *vb_map[PIPE_MAX_ATTRIBS] = {0}, *out_map;
+   struct pipe_transfer *vb_transfer[PIPE_MAX_ATTRIBS] = {0};
+   struct pipe_resource *out_buffer = NULL;
+   unsigned i, num_verts, out_offset;
+   struct pipe_vertex_element new_velems[PIPE_MAX_ATTRIBS];
+
+   memset(&key, 0, sizeof(key));
+   memset(tr_elem_index, 0xff, sizeof(tr_elem_index));
+
+   /* Initialize the translate key, i.e. the recipe how vertices should be
+     * translated. */
+   memset(&key, 0, sizeof key);
+   for (i = 0; i < mgr->ve->count; i++) {
+      struct pipe_vertex_buffer *vb =
+            &mgr->b.vertex_buffer[mgr->ve->ve[i].vertex_buffer_index];
+      enum pipe_format output_format = mgr->ve->native_format[i];
+      unsigned output_format_size = mgr->ve->native_format_size[i];
+
+      /* Check for support. */
+      if (mgr->ve->ve[i].src_format == mgr->ve->native_format[i] &&
+          (mgr->caps.fetch_dword_unaligned ||
+           (vb->buffer_offset % 4 == 0 &&
+            vb->stride % 4 == 0 &&
+            mgr->ve->ve[i].src_offset % 4 == 0))) {
+         continue;
+      }
+
+      /* Workaround for translate: output floats instead of halfs. */
+      switch (output_format) {
+      case PIPE_FORMAT_R16_FLOAT:
+         output_format = PIPE_FORMAT_R32_FLOAT;
+         output_format_size = 4;
+         break;
+      case PIPE_FORMAT_R16G16_FLOAT:
+         output_format = PIPE_FORMAT_R32G32_FLOAT;
+         output_format_size = 8;
+         break;
+      case PIPE_FORMAT_R16G16B16_FLOAT:
+         output_format = PIPE_FORMAT_R32G32B32_FLOAT;
+         output_format_size = 12;
+         break;
+      case PIPE_FORMAT_R16G16B16A16_FLOAT:
+         output_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
+         output_format_size = 16;
+         break;
+      default:;
+      }
+
+      /* Add this vertex element. */
+      te = &key.element[key.nr_elements];
+      /*te->type;
+        te->instance_divisor;*/
+      te->input_buffer = mgr->ve->ve[i].vertex_buffer_index;
+      te->input_format = mgr->ve->ve[i].src_format;
+      te->input_offset = mgr->ve->ve[i].src_offset;
+      te->output_format = output_format;
+      te->output_offset = key.output_stride;
+
+      key.output_stride += output_format_size;
+      vb_translated[mgr->ve->ve[i].vertex_buffer_index] = TRUE;
+      tr_elem_index[i] = key.nr_elements;
+      key.nr_elements++;
+   }
+
+   /* Get a translate object. */
+   tr = translate_cache_find(mgr->translate_cache, &key);
+
+   /* Map buffers we want to translate. */
+   for (i = 0; i < mgr->b.nr_vertex_buffers; i++) {
+      if (vb_translated[i]) {
+         struct pipe_vertex_buffer *vb = &mgr->b.vertex_buffer[i];
+
+         vb_map[i] = pipe_buffer_map(mgr->pipe, vb->buffer,
+                                     PIPE_TRANSFER_READ, &vb_transfer[i]);
+
+         tr->set_buffer(tr, i,
+                        vb_map[i] + vb->buffer_offset + vb->stride * min_index,
+                        vb->stride, ~0);
+      }
+   }
+
+   /* Create and map the output buffer. */
+   num_verts = max_index + 1 - min_index;
+
+   u_upload_alloc(mgr->b.uploader,
+                  key.output_stride * min_index,
+                  key.output_stride * num_verts,
+                  &out_offset, &out_buffer, upload_flushed,
+                  (void**)&out_map);
+
+   out_offset -= key.output_stride * min_index;
+
+   /* Translate. */
+   tr->run(tr, 0, num_verts, 0, out_map);
+
+   /* Unmap all buffers. */
+   for (i = 0; i < mgr->b.nr_vertex_buffers; i++) {
+      if (vb_translated[i]) {
+         pipe_buffer_unmap(mgr->pipe, vb_transfer[i]);
+      }
+   }
+
+   /* Setup the new vertex buffer in the first free slot. */
+   mgr->translate_vb_slot = ~0;
+   for (i = 0; i < PIPE_MAX_ATTRIBS; i++) {
+      if (!mgr->b.vertex_buffer[i].buffer) {
+         mgr->translate_vb_slot = i;
+
+         if (i >= mgr->b.nr_vertex_buffers) {
+            mgr->b.nr_real_vertex_buffers = i+1;
+         }
+         break;
+      }
+   }
+
+   if (mgr->translate_vb_slot != ~0) {
+      /* Setup the new vertex buffer. */
+      pipe_resource_reference(
+            &mgr->b.real_vertex_buffer[mgr->translate_vb_slot], out_buffer);
+      mgr->b.vertex_buffer[mgr->translate_vb_slot].buffer_offset = out_offset;
+      mgr->b.vertex_buffer[mgr->translate_vb_slot].stride = key.output_stride;
+
+      /* Setup new vertex elements. */
+      for (i = 0; i < mgr->ve->count; i++) {
+         if (tr_elem_index[i] < key.nr_elements) {
+            te = &key.element[tr_elem_index[i]];
+            new_velems[i].instance_divisor = mgr->ve->ve[i].instance_divisor;
+            new_velems[i].src_format = te->output_format;
+            new_velems[i].src_offset = te->output_offset;
+            new_velems[i].vertex_buffer_index = mgr->translate_vb_slot;
+         } else {
+            memcpy(&new_velems[i], &mgr->ve->ve[i],
+                   sizeof(struct pipe_vertex_element));
+         }
+      }
+
+      mgr->fallback_ve =
+            mgr->pipe->create_vertex_elements_state(mgr->pipe, mgr->ve->count,
+                                                    new_velems);
+
+      /* Preserve saved_ve. */
+      mgr->ve_binding_lock = TRUE;
+      mgr->pipe->bind_vertex_elements_state(mgr->pipe, mgr->fallback_ve);
+      mgr->ve_binding_lock = FALSE;
+   }
+
+   pipe_resource_reference(&out_buffer, NULL);
+}
+
+static void u_vbuf_translate_end(struct u_vbuf_mgr_priv *mgr)
+{
+   if (mgr->fallback_ve == NULL) {
+      return;
+   }
+
+   /* Restore vertex elements. */
+   /* Note that saved_ve will be overwritten in bind_vertex_elements_state. */
+   mgr->pipe->bind_vertex_elements_state(mgr->pipe, mgr->saved_ve);
+   mgr->pipe->delete_vertex_elements_state(mgr->pipe, mgr->fallback_ve);
+   mgr->fallback_ve = NULL;
+
+   /* Delete the now-unused VBO. */
+   pipe_resource_reference(&mgr->b.real_vertex_buffer[mgr->translate_vb_slot],
+                           NULL);
+   mgr->b.nr_real_vertex_buffers = mgr->b.nr_vertex_buffers;
+}
+
+#define FORMAT_REPLACE(what, withwhat) \
+    case PIPE_FORMAT_##what: format = PIPE_FORMAT_##withwhat; break
+
+struct u_vbuf_mgr_elements *
+u_vbuf_mgr_create_vertex_elements(struct u_vbuf_mgr *mgrb,
+                                  unsigned count,
+                                  const struct pipe_vertex_element *attribs,
+                                  struct pipe_vertex_element *native_attribs)
+{
+   struct u_vbuf_mgr_priv *mgr = (struct u_vbuf_mgr_priv*)mgrb;
+   unsigned i;
+   struct u_vbuf_mgr_elements *ve = CALLOC_STRUCT(u_vbuf_mgr_elements);
+
+   ve->count = count;
+
+   if (!count) {
+      return ve;
+   }
+
+   memcpy(ve->ve, attribs, sizeof(struct pipe_vertex_element) * count);
+   memcpy(native_attribs, attribs, sizeof(struct pipe_vertex_element) * count);
+
+   /* Set the best native format in case the original format is not
+    * supported. */
+   for (i = 0; i < count; i++) {
+      enum pipe_format format = ve->ve[i].src_format;
+
+      ve->src_format_size[i] = util_format_get_blocksize(format);
+
+      /* Choose a native format.
+       * For now we don't care about the alignment, that's going to
+       * be sorted out later. */
+      if (!mgr->caps.format_fixed32) {
+         switch (format) {
+            FORMAT_REPLACE(R32_FIXED,           R32_FLOAT);
+            FORMAT_REPLACE(R32G32_FIXED,        R32G32_FLOAT);
+            FORMAT_REPLACE(R32G32B32_FIXED,     R32G32B32_FLOAT);
+            FORMAT_REPLACE(R32G32B32A32_FIXED,  R32G32B32A32_FLOAT);
+            default:;
+         }
+      }
+      if (!mgr->caps.format_float16) {
+         switch (format) {
+            FORMAT_REPLACE(R16_FLOAT,           R32_FLOAT);
+            FORMAT_REPLACE(R16G16_FLOAT,        R32G32_FLOAT);
+            FORMAT_REPLACE(R16G16B16_FLOAT,     R32G32B32_FLOAT);
+            FORMAT_REPLACE(R16G16B16A16_FLOAT,  R32G32B32A32_FLOAT);
+            default:;
+         }
+      }
+      if (!mgr->caps.format_float64) {
+         switch (format) {
+            FORMAT_REPLACE(R64_FLOAT,           R32_FLOAT);
+            FORMAT_REPLACE(R64G64_FLOAT,        R32G32_FLOAT);
+            FORMAT_REPLACE(R64G64B64_FLOAT,     R32G32B32_FLOAT);
+            FORMAT_REPLACE(R64G64B64A64_FLOAT,  R32G32B32A32_FLOAT);
+            default:;
+         }
+      }
+      if (!mgr->caps.format_norm32) {
+         switch (format) {
+            FORMAT_REPLACE(R32_UNORM,           R32_FLOAT);
+            FORMAT_REPLACE(R32G32_UNORM,        R32G32_FLOAT);
+            FORMAT_REPLACE(R32G32B32_UNORM,     R32G32B32_FLOAT);
+            FORMAT_REPLACE(R32G32B32A32_UNORM,  R32G32B32A32_FLOAT);
+            FORMAT_REPLACE(R32_SNORM,           R32_FLOAT);
+            FORMAT_REPLACE(R32G32_SNORM,        R32G32_FLOAT);
+            FORMAT_REPLACE(R32G32B32_SNORM,     R32G32B32_FLOAT);
+            FORMAT_REPLACE(R32G32B32A32_SNORM,  R32G32B32A32_FLOAT);
+            default:;
+         }
+      }
+      if (!mgr->caps.format_scaled32) {
+         switch (format) {
+            FORMAT_REPLACE(R32_USCALED,         R32_FLOAT);
+            FORMAT_REPLACE(R32G32_USCALED,      R32G32_FLOAT);
+            FORMAT_REPLACE(R32G32B32_USCALED,   R32G32B32_FLOAT);
+            FORMAT_REPLACE(R32G32B32A32_USCALED,R32G32B32A32_FLOAT);
+            FORMAT_REPLACE(R32_SSCALED,         R32_FLOAT);
+            FORMAT_REPLACE(R32G32_SSCALED,      R32G32_FLOAT);
+            FORMAT_REPLACE(R32G32B32_SSCALED,   R32G32B32_FLOAT);
+            FORMAT_REPLACE(R32G32B32A32_SSCALED,R32G32B32A32_FLOAT);
+            default:;
+         }
+      }
+
+      native_attribs[i].src_format = format;
+      ve->native_format[i] = format;
+      ve->native_format_size[i] =
+            util_format_get_blocksize(ve->native_format[i]);
+
+      ve->incompatible_layout =
+            ve->incompatible_layout ||
+            ve->ve[i].src_format != ve->native_format[i] ||
+            (!mgr->caps.fetch_dword_unaligned && ve->ve[i].src_offset % 4 != 0);
+   }
+
+   /* Align the formats to the size of DWORD if needed. */
+   if (!mgr->caps.fetch_dword_unaligned) {
+      for (i = 0; i < count; i++) {
+         ve->native_format_size[i] = align(ve->native_format_size[i], 4);
+      }
+   }
+
+   return ve;
+}
+
+void u_vbuf_mgr_bind_vertex_elements(struct u_vbuf_mgr *mgrb,
+                                     void *cso,
+                                     struct u_vbuf_mgr_elements *ve)
+{
+   struct u_vbuf_mgr_priv *mgr = (struct u_vbuf_mgr_priv*)mgrb;
+
+   if (!cso) {
+      return;
+   }
+
+   if (!mgr->ve_binding_lock) {
+      mgr->saved_ve = cso;
+      mgr->ve = ve;
+   }
+}
+
+void u_vbuf_mgr_destroy_vertex_elements(struct u_vbuf_mgr *mgr,
+                                        struct u_vbuf_mgr_elements *ve)
+{
+   FREE(ve);
+}
+
+void u_vbuf_mgr_set_vertex_buffers(struct u_vbuf_mgr *mgrb,
+                                   unsigned count,
+                                   const struct pipe_vertex_buffer *bufs)
+{
+   struct u_vbuf_mgr_priv *mgr = (struct u_vbuf_mgr_priv*)mgrb;
+   unsigned i;
+
+   mgr->any_user_vbs = FALSE;
+   mgr->incompatible_vb_layout = FALSE;
+
+   if (!mgr->caps.fetch_dword_unaligned) {
+      /* Check if the strides and offsets are aligned to the size of DWORD. */
+      for (i = 0; i < count; i++) {
+         if (bufs[i].buffer) {
+            if (bufs[i].stride % 4 != 0 ||
+                bufs[i].buffer_offset % 4 != 0) {
+               mgr->incompatible_vb_layout = TRUE;
+               break;
+            }
+         }
+      }
+   }
+
+   for (i = 0; i < count; i++) {
+      const struct pipe_vertex_buffer *vb = &bufs[i];
+
+      pipe_resource_reference(&mgr->b.vertex_buffer[i].buffer, vb->buffer);
+      pipe_resource_reference(&mgr->b.real_vertex_buffer[i], NULL);
+
+      if (!vb->buffer) {
+         continue;
+      }
+
+      if (u_vbuf_resource(vb->buffer)->user_ptr) {
+         mgr->any_user_vbs = TRUE;
+         continue;
+      }
+
+      pipe_resource_reference(&mgr->b.real_vertex_buffer[i], vb->buffer);
+   }
+
+   for (; i < mgr->b.nr_real_vertex_buffers; i++) {
+      pipe_resource_reference(&mgr->b.vertex_buffer[i].buffer, NULL);
+      pipe_resource_reference(&mgr->b.real_vertex_buffer[i], NULL);
+   }
+
+   memcpy(mgr->b.vertex_buffer, bufs,
+          sizeof(struct pipe_vertex_buffer) * count);
+
+   mgr->b.nr_vertex_buffers = count;
+   mgr->b.nr_real_vertex_buffers = count;
+}
+
+static void u_vbuf_upload_buffers(struct u_vbuf_mgr_priv *mgr,
+                                  int min_index, int max_index,
+                                  unsigned instance_count,
+                                  boolean *upload_flushed)
+{
+   unsigned i, nr = mgr->ve->count;
+   unsigned count = max_index + 1 - min_index;
+   boolean uploaded[PIPE_MAX_ATTRIBS] = {0};
+
+   for (i = 0; i < nr; i++) {
+      unsigned index = mgr->ve->ve[i].vertex_buffer_index;
+      struct pipe_vertex_buffer *vb = &mgr->b.vertex_buffer[index];
+
+      if (vb->buffer &&
+          u_vbuf_resource(vb->buffer)->user_ptr &&
+          !uploaded[index]) {
+         unsigned first, size;
+         boolean flushed;
+         unsigned instance_div = mgr->ve->ve[i].instance_divisor;
+
+         if (instance_div) {
+            first = 0;
+            size = vb->stride *
+                   ((instance_count + instance_div - 1) / instance_div);
+         } else if (vb->stride) {
+            first = vb->stride * min_index;
+            size = vb->stride * count;
+         } else {
+            first = 0;
+            size = mgr->ve->native_format_size[i];
+         }
+
+         u_upload_data(mgr->b.uploader, first, size,
+                       u_vbuf_resource(vb->buffer)->user_ptr + first,
+                       &vb->buffer_offset,
+                       &mgr->b.real_vertex_buffer[index],
+                       &flushed);
+
+         vb->buffer_offset -= first;
+
+         uploaded[index] = TRUE;
+         *upload_flushed = *upload_flushed || flushed;
+      } else {
+         assert(mgr->b.real_vertex_buffer[index]);
+      }
+   }
+}
+
+static void u_vbuf_mgr_compute_max_index(struct u_vbuf_mgr_priv *mgr)
+{
+   unsigned i, nr = mgr->ve->count;
+
+   mgr->b.max_index = ~0;
+
+   for (i = 0; i < nr; i++) {
+      struct pipe_vertex_buffer *vb =
+            &mgr->b.vertex_buffer[mgr->ve->ve[i].vertex_buffer_index];
+      int unused;
+      unsigned max_index;
+
+      if (!vb->buffer ||
+          !vb->stride ||
+          u_vbuf_resource(vb->buffer)->user_ptr) {
+         continue;
+      }
+
+      /* How many bytes is unused after the last vertex.
+       * width0 may be "count*stride - unused" and we have to compensate
+       * for that when dividing by stride. */
+      unused = vb->stride -
+               (mgr->ve->ve[i].src_offset + mgr->ve->src_format_size[i]);
+      assert(unused >= 0);
+
+      /* Compute the maximum index for this vertex element. */
+      max_index =
+         (vb->buffer->width0 - vb->buffer_offset + (unsigned)unused) /
+         vb->stride - 1;
+
+      mgr->b.max_index = MIN2(mgr->b.max_index, max_index);
+   }
+}
+
+void u_vbuf_mgr_draw_begin(struct u_vbuf_mgr *mgrb,
+                           const struct pipe_draw_info *info,
+                           boolean *buffers_updated,
+                           boolean *uploader_flushed)
+{
+   struct u_vbuf_mgr_priv *mgr = (struct u_vbuf_mgr_priv*)mgrb;
+   boolean bufs_updated = FALSE, upload_flushed = FALSE;
+   int min_index, max_index;
+
+   u_vbuf_mgr_compute_max_index(mgr);
+
+   min_index = info->min_index - info->index_bias;
+   if (info->max_index == ~0) {
+      max_index = mgr->b.max_index;
+   } else {
+      max_index = MIN2(info->max_index - info->index_bias, mgr->b.max_index);
+   }
+
+   /* Translate vertices with non-native layouts or formats. */
+   if (mgr->incompatible_vb_layout || mgr->ve->incompatible_layout) {
+      u_vbuf_translate_begin(mgr, min_index, max_index, &upload_flushed);
+
+      if (mgr->fallback_ve) {
+         bufs_updated = TRUE;
+      }
+   }
+
+   /* Upload user buffers. */
+   if (mgr->any_user_vbs) {
+      u_vbuf_upload_buffers(mgr, min_index, max_index, info->instance_count,
+                            &upload_flushed);
+      bufs_updated = TRUE;
+   }
+
+   /* Set the return values. */
+   if (buffers_updated) {
+      *buffers_updated = bufs_updated;
+   }
+   if (uploader_flushed) {
+      *uploader_flushed = upload_flushed;
+   }
+}
+
+void u_vbuf_mgr_draw_end(struct u_vbuf_mgr *mgrb)
+{
+   struct u_vbuf_mgr_priv *mgr = (struct u_vbuf_mgr_priv*)mgrb;
+
+   if (mgr->fallback_ve) {
+      u_vbuf_translate_end(mgr);
+   }
+}