util: Code generate functions to pack and unpack a single pixel.

Should work correctly for all pixel formats except SRGB formats.

Generated code made much simpler by defining the pixel format as
a C structure. For example this is the generated structure for
PIPE_FORMAT_B6UG5SR5S_NORM:

union util_format_b6ug5sr5s_norm {
   uint16_t value;
   struct {
      int r:5;
      int g:5;
      unsigned b:6;
   } chan;
};

Not used everywhere yet because it seems compiled code is slower than
bitshift arithmetic by some misterious reason. So we should generate
bitshift arithmetic at least for the simple UNORM pixel formats.

7 files changed, 507 insertions, 164 deletions

diff --git a/src/gallium/auxiliary/Makefile b/src/gallium/auxiliary/Makefile
index 2d41fb3dcae..39e328f5144 100644
--- a/src/gallium/auxiliary/Makefile
+++ b/src/gallium/auxiliary/Makefile
@@ -184,5 +184,10 @@ indices/u_unfilled_gen.c: indices/u_unfilled_gen.py
 util/u_format_table.c: util/u_format_table.py util/u_format_parse.py util/u_format.csv
 	python util/u_format_table.py util/u_format.csv > $@
 
+util/u_format_pack.h: util/u_format_pack.py util/u_format_parse.py util/u_format.csv
+	python util/u_format_pack.py util/u_format.csv > $@
+
 util/u_format_access.c: util/u_format_access.py util/u_format_parse.py util/u_format.csv
 	python util/u_format_access.py util/u_format.csv > $@
+
+default: util/u_format_pack.h
diff --git a/src/gallium/auxiliary/SConscript b/src/gallium/auxiliary/SConscript
index 8d2859fa711..47de50bf3ec 100644
--- a/src/gallium/auxiliary/SConscript
+++ b/src/gallium/auxiliary/SConscript
@@ -29,6 +29,13 @@ env.CodeGenerate(
 )
 
 env.CodeGenerate(
+    target = File('util/u_format_pack.h').srcnode(),
+    script = 'util/u_format_pack.py',
+    source = ['util/u_format.csv'],
+    command = 'python $SCRIPT $SOURCE > $TARGET'
+)
+
+env.CodeGenerate(
     target = 'util/u_format_access.c',
     script = 'util/u_format_access.py',
     source = ['util/u_format.csv'],
diff --git a/src/gallium/auxiliary/util/.gitignore b/src/gallium/auxiliary/util/.gitignore
index 29c586c9b51..448d2f304fb 100644
--- a/src/gallium/auxiliary/util/.gitignore
+++ b/src/gallium/auxiliary/util/.gitignore
@@ -1,2 +1,3 @@
 u_format_access.c
 u_format_table.c
+u_format_pack.h
diff --git a/src/gallium/auxiliary/util/u_format_access.py b/src/gallium/auxiliary/util/u_format_access.py
index a7a91ea652e..1c9be1b538e 100644
--- a/src/gallium/auxiliary/util/u_format_access.py
+++ b/src/gallium/auxiliary/util/u_format_access.py
@@ -39,7 +39,7 @@
 
 import sys
 
-from u_format_parse import *
+from u_format_pack import *
 
 
 def is_format_supported(format):
@@ -94,160 +94,6 @@ def native_type(format):
         assert False
 
 
-def intermediate_native_type(bits, sign):
-    '''Find a native type adequate to hold intermediate results of the request bit size.'''
-
-    bytes = 4 # don't use anything smaller than 32bits
-    while bytes * 8 < bits:
-        bytes *= 2
-    bits = bytes*8
-
-    if sign:
-        return 'int%u_t' % bits
-    else:
-        return 'uint%u_t' % bits
-
-
-def get_one_shift(channel):
-    '''Get the number of the bit that matches unity for this channel.'''
-    if channel.type == 'FLOAT':
-        assert False
-    if not channel.norm:
-        return 0
-    if channel.type == UNSIGNED:
-        return channel.size
-    if channel.type == SIGNED:
-        return channel.size - 1
-    if channel.type == FIXED:
-        return channel.size / 2
-    assert False
-
-
-def get_one(channel):
-    '''Get the value of unity for this channel.'''
-    if channel.type == 'FLOAT' or not channel.norm:
-        return 1
-    else:
-        return (1 << get_one_shift(channel)) - 1
-
-
-def generate_clamp():
-    '''Code generate the clamping functions for each type.
-
-    We don't use a macro so that arguments with side effects, 
-    like *src_pixel++ are correctly handled.
-    '''
-
-    for suffix, native_type in [
-        ('', 'double'),
-        ('f', 'float'),
-        ('ui', 'unsigned int'),
-        ('si', 'int'),
-    ]:
-        print 'static INLINE %s' % native_type
-        print 'clamp%s(%s value, %s lbound, %s ubound)' % (suffix, native_type, native_type, native_type)
-        print '{'
-        print '   if(value < lbound)'
-        print '      return lbound;'
-        print '   if(value > ubound)'
-        print '      return ubound;'
-        print '   return value;'
-        print '}'
-        print
-
-
-def clamp_expr(src_channel, dst_channel, dst_native_type, value):
-    '''Generate the expression to clamp the value in the source type to the
-    destination type range.'''
-
-    if src_channel == dst_channel:
-        return value
-
-    # Pick the approriate clamp function
-    if src_channel.type == FLOAT:
-        if src_channel.size == 32:
-            func = 'clampf'
-        elif src_channel.size == 64:
-            func = 'clamp'
-        else:
-            assert False
-    elif src_channel.type == UNSIGNED:
-        func = 'clampui'
-    elif src_channel.type == SIGNED:
-        func = 'clampsi'
-    else:
-        assert False
-
-    # Clamp floats to [-1, 1] or [0, 1] range
-    if src_channel.type == FLOAT and dst_channel.norm:
-        max = 1
-        if src_channel.sign and dst_channel.sign:
-            min = -1
-        else:
-            min = 0
-        return '%s(%s, %s, %s)' % (func, value, min, max)
-                
-    # FIXME: Also clamp scaled values
-
-    return value
-
-
-def conversion_expr(src_channel, dst_channel, dst_native_type, value):
-    '''Generate the expression to convert a value between two types.'''
-
-    if src_channel == dst_channel:
-        return value
-
-    if src_channel.type == FLOAT and dst_channel.type == FLOAT:
-        return '(%s)%s' % (dst_native_type, value)
-    
-    if not src_channel.norm and not dst_channel.norm:
-        return '(%s)%s' % (dst_native_type, value)
-
-    value = clamp_expr(src_channel, dst_channel, dst_native_type, value)
-
-    if dst_channel.type == FLOAT:
-        if src_channel.norm:
-            one = get_one(src_channel)
-            if src_channel.size <= 23:
-                scale = '(1.0f/0x%x)' % one
-            else:
-                # bigger than single precision mantissa, use double
-                scale = '(1.0/0x%x)' % one
-            value = '(%s * %s)' % (value, scale)
-        return '(%s)%s' % (dst_native_type, value)
-
-    if src_channel.type == FLOAT:
-        if dst_channel.norm:
-            dst_one = get_one(dst_channel)
-            if dst_channel.size <= 23:
-                scale = '0x%x' % dst_one
-            else:
-                # bigger than single precision mantissa, use double
-                scale = '(double)0x%x' % dst_one
-            value = '(%s * %s)' % (value, scale)
-        return '(%s)%s' % (dst_native_type, value)
-
-    if src_channel.type == dst_channel.type:
-        src_one = get_one(src_channel)
-        dst_one = get_one(dst_channel)
-
-        if src_one > dst_one and src_channel.norm and dst_channel.norm:
-            # We can just bitshift
-            src_shift = get_one_shift(src_channel)
-            dst_shift = get_one_shift(dst_channel)
-            value = '(%s >> %s)' % (value, src_shift - dst_shift)
-        else:
-            # We need to rescale using an intermediate type big enough to hold the multiplication of both
-            tmp_native_type = intermediate_native_type(src_channel.size + dst_channel.size, src_channel.sign and dst_channel.sign)
-            value = '(%s)%s' % (tmp_native_type, value)
-            value = '%s * 0x%x / 0x%x' % (value, dst_one, src_one)
-        value = '(%s)%s' % (dst_native_type, value)
-        return value
-
-    assert False
-
-
 def generate_format_read(format, dst_channel, dst_native_type, dst_suffix):
     '''Generate the function to read pixels from a particular format'''
 
@@ -302,9 +148,10 @@ def generate_format_read(format, dst_channel, dst_native_type, dst_suffix):
             for i in range(4):
                 src_channel = format.channels[i]
                 if names[i]:
-                    value = '(*src_pixel++)'
+                    value = 'src_pixel[%u]' % i
                     value = conversion_expr(src_channel, dst_channel, dst_native_type, value)
                     print '         %s %s = %s;' % (dst_native_type, names[i], value)
+            print '         src_pixel += %u;' % (format.nr_channels())
     else:
         assert False
 
@@ -454,12 +301,10 @@ def main():
     print __doc__.strip()
     print
     print '#include "pipe/p_compiler.h"'
-    print '#include "u_format.h"'
     print '#include "u_math.h"'
+    print '#include "u_format_pack.h"'
     print
 
-    generate_clamp()
-
     type = Channel(FLOAT, False, 32)
     native_type = 'float'
     suffix = '4f'
diff --git a/src/gallium/auxiliary/util/u_format_pack.py b/src/gallium/auxiliary/util/u_format_pack.py
new file mode 100644
index 00000000000..3f33f7cc021
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_format_pack.py
@@ -0,0 +1,484 @@
+#!/usr/bin/env python
+
+'''
+/**************************************************************************
+ *
+ * Copyright 2009 VMware, Inc.
+ * 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 VMWARE 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.
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * Pixel format packing and unpacking functions.
+ *
+ * @author Jose Fonseca <[email protected]>
+ */
+'''
+
+
+import sys
+
+from u_format_parse import *
+
+
+def generate_format_type(format):
+    '''Generate a structure that describes the format.'''
+
+    print 'union util_format_%s {' % format.short_name()
+    if format.is_bitmask():
+        print '   uint%u_t value;' % (format.block_size(),)
+    print '   struct {'
+    for channel in format.channels:
+        if format.is_bitmask() and not format.is_array():
+            if channel.type == VOID:
+                if channel.size:
+                    print '      unsigned %s:%u;' % (channel.name, channel.size)
+            elif channel.type == UNSIGNED:
+                print '      unsigned %s:%u;' % (channel.name, channel.size)
+            elif channel.type == SIGNED:
+                print '      int %s:%u;' % (channel.name, channel.size)
+            else:
+                assert 0
+        else:
+            assert channel.size % 8 == 0 and is_pot(channel.size)
+            if channel.type == VOID:
+                if channel.size:
+                    print '      uint%u_t %s;' % (channel.size, channel.name)
+            elif channel.type == UNSIGNED:
+                print '      uint%u_t %s;' % (channel.size, channel.name)
+            elif channel.type in (SIGNED, FIXED):
+                print '      int%u_t %s;' % (channel.size, channel.name)
+            elif channel.type == FLOAT:
+                if channel.size == 64:
+                    print '      double %s;' % (channel.name)
+                elif channel.size == 32:
+                    print '      float %s;' % (channel.name)
+                elif channel.size == 16:
+                    print '      uint16_t %s;' % (channel.name)
+                else:
+                    assert 0
+            else:
+                assert 0
+    print '   } chan;'
+    print '};'
+    print
+
+
+def bswap_format(format):
+    '''Generate a structure that describes the format.'''
+
+    if format.is_bitmask() and not format.is_array():
+        print '#ifdef PIPE_ARCH_BIG_ENDIAN'
+        print '   pixel.value = util_bswap%u(pixel.value);' % format.block_size()
+        print '#endif'
+
+
+def is_format_supported(format):
+    '''Determines whether we actually have the plumbing necessary to generate the 
+    to read/write to/from this format.'''
+
+    # FIXME: Ideally we would support any format combination here.
+
+    if format.layout != PLAIN:
+        return False
+
+    for i in range(4):
+        channel = format.channels[i]
+        if channel.type not in (VOID, UNSIGNED, SIGNED, FLOAT):
+            return False
+
+    # We can only read a color from a depth/stencil format if the depth channel is present
+    if format.colorspace == 'zs' and format.swizzles[0] == SWIZZLE_NONE:
+        return False
+
+    return True
+
+
+def native_type(format):
+    '''Get the native appropriate for a format.'''
+
+    if format.layout == PLAIN:
+        if not format.is_array():
+            # For arithmetic pixel formats return the integer type that matches the whole pixel
+            return 'uint%u_t' % format.block_size()
+        else:
+            # For array pixel formats return the integer type that matches the color channel
+            type = format.channels[0]
+            if type.type == UNSIGNED:
+                return 'uint%u_t' % type.size
+            elif type.type == SIGNED:
+                return 'int%u_t' % type.size
+            elif type.type == FLOAT:
+                if type.size == 32:
+                    return 'float'
+                elif type.size == 64:
+                    return 'double'
+                else:
+                    assert False
+            else:
+                assert False
+    else:
+        assert False
+
+
+def intermediate_native_type(bits, sign):
+    '''Find a native type adequate to hold intermediate results of the request bit size.'''
+
+    bytes = 4 # don't use anything smaller than 32bits
+    while bytes * 8 < bits:
+        bytes *= 2
+    bits = bytes*8
+
+    if sign:
+        return 'int%u_t' % bits
+    else:
+        return 'uint%u_t' % bits
+
+
+def get_one_shift(type):
+    '''Get the number of the bit that matches unity for this type.'''
+    if type.type == 'FLOAT':
+        assert False
+    if not type.norm:
+        return 0
+    if type.type == UNSIGNED:
+        return type.size
+    if type.type == SIGNED:
+        return type.size - 1
+    if type.type == FIXED:
+        return type.size / 2
+    assert False
+
+
+def get_one(type):
+    '''Get the value of unity for this type.'''
+    if type.type == 'FLOAT' or not type.norm:
+        return 1
+    else:
+        return (1 << get_one_shift(type)) - 1
+
+
+def generate_clamp():
+    '''Code generate the clamping functions for each type.
+
+    We don't use a macro so that arguments with side effects, 
+    like *src_pixel++ are correctly handled.
+    '''
+
+    for suffix, native_type in [
+        ('', 'double'),
+        ('f', 'float'),
+        ('ui', 'unsigned int'),
+        ('si', 'int'),
+    ]:
+        print 'static INLINE %s' % native_type
+        print 'clamp%s(%s value, %s lbound, %s ubound)' % (suffix, native_type, native_type, native_type)
+        print '{'
+        print '   if(value < lbound)'
+        print '      return lbound;'
+        print '   if(value > ubound)'
+        print '      return ubound;'
+        print '   return value;'
+        print '}'
+        print
+
+
+def clamp_expr(src_channel, dst_channel, dst_native_type, value):
+    '''Generate the expression to clamp the value in the source type to the
+    destination type range.'''
+
+    if src_channel == dst_channel:
+        return value
+
+    # Pick the approriate clamp function
+    if src_channel.type == FLOAT:
+        if src_channel.size == 32:
+            func = 'clampf'
+        elif src_channel.size == 64:
+            func = 'clamp'
+        else:
+            assert False
+    elif src_channel.type == UNSIGNED:
+        func = 'clampui'
+    elif src_channel.type == SIGNED:
+        func = 'clampsi'
+    else:
+        assert False
+
+    src_min = src_channel.min()
+    src_max = src_channel.max()
+    dst_min = dst_channel.min()
+    dst_max = dst_channel.max()
+
+    if src_min < dst_min and src_max > dst_max:
+        return 'CLAMP(%s, %s, %s)' % (value, dst_min, dst_max)
+
+    if src_max > dst_max:
+        return 'MIN2(%s, %s)' % (value, dst_max)
+        
+    if src_min < dst_min:
+        return 'MAX2(%s, %s)' % (value, dst_min)
+
+    return value
+
+
+def conversion_expr(src_channel, dst_channel, dst_native_type, value, clamp=True):
+    '''Generate the expression to convert a value between two types.'''
+
+    if src_channel == dst_channel:
+        return value
+
+    if src_channel.type == FLOAT and dst_channel.type == FLOAT:
+        return '(%s)%s' % (dst_native_type, value)
+    
+    if not src_channel.norm and not dst_channel.norm:
+        return '(%s)%s' % (dst_native_type, value)
+
+    if clamp:
+        value = clamp_expr(src_channel, dst_channel, dst_native_type, value)
+
+    if dst_channel.type == FLOAT:
+        if src_channel.norm:
+            one = get_one(src_channel)
+            if src_channel.size <= 23:
+                scale = '(1.0f/0x%x)' % one
+            else:
+                # bigger than single precision mantissa, use double
+                scale = '(1.0/0x%x)' % one
+            value = '(%s * %s)' % (value, scale)
+        return '(%s)%s' % (dst_native_type, value)
+
+    if src_channel.type == FLOAT:
+        if dst_channel.norm:
+            dst_one = get_one(dst_channel)
+            if dst_channel.size <= 23:
+                scale = '0x%x' % dst_one
+            else:
+                # bigger than single precision mantissa, use double
+                scale = '(double)0x%x' % dst_one
+            value = '(%s * %s)' % (value, scale)
+        return '(%s)%s' % (dst_native_type, value)
+
+    if not src_channel.norm and not dst_channel.norm:
+        # neither is normalized -- just cast
+        return '(%s)%s' % (dst_native_type, value)
+
+    if src_channel.type in (SIGNED, UNSIGNED) and dst_channel.type in (SIGNED, UNSIGNED):
+        src_one = get_one(src_channel)
+        dst_one = get_one(dst_channel)
+
+        if src_one > dst_one and src_channel.norm:
+            # We can just bitshift
+            src_shift = get_one_shift(src_channel)
+            dst_shift = get_one_shift(dst_channel)
+            value = '(%s >> %s)' % (value, src_shift - dst_shift)
+        else:
+            # We need to rescale using an intermediate type big enough to hold the multiplication of both
+            tmp_native_type = intermediate_native_type(src_channel.size + dst_channel.size, src_channel.sign and dst_channel.sign)
+            value = '(%s)%s' % (tmp_native_type, value)
+            value = '(%s * 0x%x / 0x%x)' % (value, dst_one, src_one)
+        value = '(%s)%s' % (dst_native_type, value)
+        return value
+
+    assert False
+
+
+def generate_format_unpack(format, dst_channel, dst_native_type, dst_suffix):
+    '''Generate the function to unpack pixels from a particular format'''
+
+    name = format.short_name()
+
+    src_native_type = native_type(format)
+
+    print 'static INLINE void'
+    print 'util_format_%s_unpack_%s(%s *dst, const void *src)' % (name, dst_suffix, dst_native_type)
+    print '{'
+    print '   union util_format_%s pixel;' % format.short_name()
+    print '   memcpy(&pixel, src, sizeof pixel);'
+    bswap_format(format)
+
+    assert format.layout == PLAIN
+
+    for i in range(4):
+        swizzle = format.swizzles[i]
+        if swizzle < 4:
+            src_channel = format.channels[swizzle]
+            value = 'pixel.chan.%s' % src_channel.name 
+            value = conversion_expr(src_channel, dst_channel, dst_native_type, value)
+        elif swizzle == SWIZZLE_0:
+            value = '0'
+        elif swizzle == SWIZZLE_1:
+            value = get_one(dst_channel)
+        elif swizzle == SWIZZLE_NONE:
+            value = '0'
+        else:
+            assert False
+        if format.colorspace == ZS:
+            if i == 3:
+                value = get_one(dst_channel)
+            elif i >= 1:
+                value = 'dst[0]'
+        print '   dst[%u] = %s; /* %s */' % (i, value, 'rgba'[i])
+
+    print '}'
+    print
+    
+
+def generate_format_pack(format, src_channel, src_native_type, src_suffix):
+    '''Generate the function to pack pixels to a particular format'''
+
+    name = format.short_name()
+
+    dst_native_type = native_type(format)
+
+    print 'static INLINE void'
+    print 'util_format_%s_pack_%s(void *dst, %s r, %s g, %s b, %s a)' % (name, src_suffix, src_native_type, src_native_type, src_native_type, src_native_type)
+    print '{'
+    print '   union util_format_%s pixel;' % format.short_name()
+
+    assert format.layout == PLAIN
+
+    inv_swizzle = format.inv_swizzles()
+
+    for i in range(4):
+        dst_channel = format.channels[i]
+        width = dst_channel.size
+        if inv_swizzle[i] is None:
+            continue
+        value = 'rgba'[inv_swizzle[i]]
+        value = conversion_expr(src_channel, dst_channel, dst_native_type, value)
+        if format.colorspace == ZS:
+            if i == 3:
+                value = get_one(dst_channel)
+            elif i >= 1:
+                value = '0'
+        print '   pixel.chan.%s = %s;' % (dst_channel.name, value)
+
+    bswap_format(format)
+    print '   memcpy(dst, &pixel, sizeof pixel);'
+    print '}'
+    print
+    
+
+def generate_unpack(formats, dst_channel, dst_native_type, dst_suffix):
+    '''Generate the dispatch function to unpack pixels from any format'''
+
+    for format in formats:
+        if is_format_supported(format):
+            generate_format_unpack(format, dst_channel, dst_native_type, dst_suffix)
+
+    print 'static INLINE void'
+    print 'util_format_unpack_%s(enum pipe_format format, %s *dst, const void *src)' % (dst_suffix, dst_native_type)
+    print '{'
+    print '   void (*func)(%s *dst, const void *src);' % dst_native_type
+    print '   switch(format) {'
+    for format in formats:
+        if is_format_supported(format):
+            print '   case %s:' % format.name
+            print '      func = &util_format_%s_unpack_%s;' % (format.short_name(), dst_suffix)
+            print '      break;'
+    print '   default:'
+    print '      debug_printf("unsupported format\\n");'
+    print '      return;'
+    print '   }'
+    print '   func(dst, src);'
+    print '}'
+    print
+
+
+def generate_pack(formats, src_channel, src_native_type, src_suffix):
+    '''Generate the dispatch function to pack pixels to any format'''
+
+    for format in formats:
+        if is_format_supported(format):
+            generate_format_pack(format, src_channel, src_native_type, src_suffix)
+
+    print 'static INLINE void'
+    print 'util_format_pack_%s(enum pipe_format format, void *dst, %s r, %s g, %s b, %s a)' % (src_suffix, src_native_type, src_native_type, src_native_type, src_native_type)
+    print '{'
+    print '   void (*func)(void *dst, %s r, %s g, %s b, %s a);' % (src_native_type, src_native_type, src_native_type, src_native_type)
+    print '   switch(format) {'
+    for format in formats:
+        if is_format_supported(format):
+            print '   case %s:' % format.name
+            print '      func = &util_format_%s_pack_%s;' % (format.short_name(), src_suffix)
+            print '      break;'
+    print '   default:'
+    print '      debug_printf("%s: unsupported format\\n", __FUNCTION__);'
+    print '      return;'
+    print '   }'
+    print '   func(dst, r, g, b, a);'
+    print '}'
+    print
+
+
+def main():
+    formats = []
+    for arg in sys.argv[1:]:
+        formats.extend(parse(arg))
+
+    print '/* This file is autogenerated by u_format_pack.py from u_format.csv. Do not edit directly. */'
+    print
+    # This will print the copyright message on the top of this file
+    print __doc__.strip()
+
+    print
+    print '#ifndef U_FORMAT_PACK_H'
+    print '#define U_FORMAT_PACK_H'
+    print
+    print '#include "pipe/p_compiler.h"'
+    print '#include "u_math.h"'
+    print '#include "u_format.h"'
+    print
+
+    generate_clamp()
+
+    for format in formats:
+        if format.layout == PLAIN:
+            generate_format_type(format)
+
+    channel = Channel(FLOAT, False, 32)
+    native_type = 'float'
+    suffix = '4f'
+
+    generate_unpack(formats, channel, native_type, suffix)
+    generate_pack(formats, channel, native_type, suffix)
+
+    channel = Channel(UNSIGNED, True, 8)
+    native_type = 'uint8_t'
+    suffix = '4ub'
+
+    generate_unpack(formats, channel, native_type, suffix)
+    generate_pack(formats, channel, native_type, suffix)
+
+    print
+    print '#ifdef __cplusplus'
+    print '}'
+    print '#endif'
+    print
+    print '#endif /* ! U_FORMAT_PACK_H */'
+
+
+if __name__ == '__main__':
+    main()
diff --git a/src/gallium/drivers/llvmpipe/SConscript b/src/gallium/drivers/llvmpipe/SConscript
index 71f9337422f..13c1a13e87a 100644
--- a/src/gallium/drivers/llvmpipe/SConscript
+++ b/src/gallium/drivers/llvmpipe/SConscript
@@ -21,6 +21,7 @@ env.CodeGenerate(
 # XXX: Our dependency scanner only finds depended modules in relative dirs.
 env.Depends('lp_tile_soa.c', [
     '#src/gallium/auxiliary/util/u_format_parse.py', 
+    '#src/gallium/auxiliary/util/u_format_pack.py', 
     '#src/gallium/auxiliary/util/u_format_access.py',
 ])
 
diff --git a/src/gallium/drivers/llvmpipe/lp_tile_soa.py b/src/gallium/drivers/llvmpipe/lp_tile_soa.py
index dc44d67d634..00b8d4fc382 100644
--- a/src/gallium/drivers/llvmpipe/lp_tile_soa.py
+++ b/src/gallium/drivers/llvmpipe/lp_tile_soa.py
@@ -90,7 +90,7 @@ def generate_format_read(format, dst_channel, dst_native_type, dst_suffix):
                         value = '(%s >> %u)' % (value, shift)
                     if shift + width < format.block_size():
                         value = '(%s & 0x%x)' % (value, mask)
-                    value = conversion_expr(src_channel, dst_channel, dst_native_type, value)
+                    value = conversion_expr(src_channel, dst_channel, dst_native_type, value, clamp=False)
                     print '         %s %s = %s;' % (dst_native_type, names[i], value)
                 shift += width
         else:
@@ -98,7 +98,7 @@ def generate_format_read(format, dst_channel, dst_native_type, dst_suffix):
                 src_channel = format.channels[i]
                 if names[i]:
                     value = '(*src_pixel++)'
-                    value = conversion_expr(src_channel, dst_channel, dst_native_type, value)
+                    value = conversion_expr(src_channel, dst_channel, dst_native_type, value, clamp=False)
                     print '         %s %s = %s;' % (dst_native_type, names[i], value)
     else:
         assert False
@@ -154,7 +154,7 @@ def pack_rgba(format, src_channel, r, g, b, a):
         if value:
             dst_channel = format.channels[i]
             dst_native_type = native_type(format)
-            value = conversion_expr(src_channel, dst_channel, dst_native_type, value)
+            value = conversion_expr(src_channel, dst_channel, dst_native_type, value, clamp=False)
             term = "((%s) << %d)" % (value, shift)
             if expr:
                 expr = expr + " | " + term
@@ -217,7 +217,7 @@ def emit_tile_pixel_write_code(format, src_channel):
                 width = dst_channel.size
                 if inv_swizzle[i] is not None:
                     value = 'TILE_PIXEL(src, x, y, %u)' % inv_swizzle[i]
-                    value = conversion_expr(src_channel, dst_channel, dst_native_type, value)
+                    value = conversion_expr(src_channel, dst_channel, dst_native_type, value, clamp=False)
                     if shift:
                         value = '(%s << %u)' % (value, shift)
                     print '         pixel |= %s;' % value
@@ -228,7 +228,7 @@ def emit_tile_pixel_write_code(format, src_channel):
                 dst_channel = format.channels[i]
                 if inv_swizzle[i] is not None:
                     value = 'TILE_PIXEL(src, x, y, %u)' % inv_swizzle[i]
-                    value = conversion_expr(src_channel, dst_channel, dst_native_type, value)
+                    value = conversion_expr(src_channel, dst_channel, dst_native_type, value, clamp=False)
                     print '         *dst_pixel++ = %s;' % value
     else:
         assert False