3 - Commented-out uses of fbCompositeCopyAreasse2()
5 - Use MM_HINT_NTA instead of MM_HINT_T0
7 - Use of fbCompositeOver_x888x8x8888sse2()
9 - Use pixmanFillsse2 and pixmanBltsse2
11 - Update the RLEASING file
13 - Things to keep in mind if breaking ABI:
15 - There should be a guard #ifndef I_AM_EITHER_CAIRO_OR_THE_X_SERVER
17 - X server will require 16.16 essentially forever. Can we get
18 the required precision by simply adding offset_x/y to the
19 relevant rendering API?
21 - pixman_image_set_indexed() should copy its argument, and X
22 should be ported over to use a pixman_image as the
23 representation of a Picture, rather than creating one on each
26 - We should get rid of pixman_set_static_pointers()
28 - We should get rid of the various trapezoid helper functions().
29 (They only exist because they are theoretically available to
32 - Regions should become 32 bit
34 - There should only be one trap rasterization API.
36 - The PIXMAN_g8/c8/etc formats should use the A channel
37 to indicate the actual depth. That way PIXMAN_x4c4 and PIXMAN_c8
40 - Make pixman_region_point_in() survive a NULL box, then fix up
43 - Possibly look into inlining the fetch functions
47 - Add a general way of dealing with architecture specific
48 fast-paths. The current idea is to have each operation that can
49 be optimized is called through a function pointer that is
50 initially set to an initialization function that is responsible for
51 setting the function pointer to the appropriate fast-path.
53 - Go through things marked FIXME
55 - Add calls to prepare and finish access where necessary. grep for
56 ACCESS_MEM, and make sure they are correctly wrapped in prepare
59 - restore READ/WRITE in the fbcompose combiners since they sometimes
60 store directly to destination drawables.
62 - It probably makes sense to move the more strange X region API
63 into pixman as well, but guarded with PIXMAN_XORG_COMPATIBILITY
65 - Reinstate the FbBits typedef? At the moment we don't
66 even have the FbBits type; we just use uint32_t everywhere.
68 Keith says in bug 2335:
70 The 64-bit code in fb (pixman) is probably broken; it hasn't been
71 used in quite some time as PCI (and AGP) is 32-bits wide, so
72 doing things 64-bits at a time is a net loss. To quickly fix
73 this, I suggest just using 32-bit datatypes by setting
74 IC_SHIFT to 5 for all machines.
76 - Consider whether calling regions region16 is really such a great
77 idea. Vlad wants 32 bit regions for Cairo. This will break X server
78 ABI, but should otherwise be mostly harmless, though a
79 pixman_region_get_boxes16() may be useful.
81 - Consider optimizing the 8/16 bit solid fills in pixman-util.c by
82 storing more than one value at a time.
84 - Add an image cache to prevent excessive malloc/free. Note that pixman
85 needs to be thread safe when used from cairo.
87 - Moving to 24.8 coordinates. This is tricky because X is still
88 defined as 16.16 and will be basically forever. It's possible we
89 could do this by adding extra offset_x/y parameters to the
90 trapezoid calls. The X server could then just call the API with
91 (0, 0). Cairo would have to make sure that the delta *within* a
92 batch of trapezoids does not exceed 16 bit.
94 - Review the pixman_format_code_t enum to make sure it will support
95 future formats. Some formats we will probably need:
97 ARGB/ABGR with 16/32/64 bit integer/floating channels
101 Also we may need the ability to distinguish between PICT_c8 and
102 PICT_x4c4. (This could be done by interpreting the A channel as
103 the depth for TYPE_COLOR and TYPE_GRAY formats).
105 A possibility may be to reserve the two top bits and make them
106 encode "number of places to shift the channel widths given" Since
107 these bits are 00 at the moment everything will continue to work,
108 but these additional widths will be allowed:
110 All even widths between 18-32
111 All multiples of four widths between 33 and 64
112 All multiples of eight between 64 and 128
114 This means things like r21g22b21 won't work - is that worth
115 worrying about? I don't think so. And of course the bpp field
116 can't handle a depth of over 256, so > 64 bit channels arent'
117 really all that useful.
119 We could reserve one extra bit to indicate floating point, but
122 PIXMAN_TYPE_ARGB_FLOAT
123 PIXMAN_TYPE_BGRA_FLOAT
126 image types. With five bits we can support up to 32 different
127 format types, which should be enough for everybody, even if we
128 decide to support all the various video formats here:
130 http://www.fourcc.org/yuv.php
132 It may make sense to have a PIXMAN_TYPE_YUV, and then use the
133 channel bits to specify the exact subtype.
135 Another possibility is to add
140 where the channel widths would get 16 and 32 added to them,
143 What about color spaces such a linear vs. srGB etc.?
148 - Crashes in Mozilla because of unaligned stack. Possible fixes
149 - Make use of gcc 4.2 feature to align the stack
150 - Write some sort of trampoline that aligns the stack
151 before calling SSE functions.
153 - Get rid of the switch-of-doom; replace it with a big table
154 describing the various fast paths.
156 - Make source clipping optional.
157 - done: source clipping happens through an indirection.
158 still needs to make the indirection settable. (And call it
161 - Run cairo test suite; fix bugs
162 - one bug in source-scale-clip
164 - Remove the warning suppression in the ACCESS_MEM macro and fix the
165 warnings that are real
168 - make the wrapper functions global instead of image specific
169 - this won't work since pixman is linked to both fb and wfb
171 - Add non-mmx solid fill
173 - Make sure the endian-ness macros are defined correctly.
175 - The rectangles in a region probably shouldn't be returned const as
176 the X server will be changing them.
178 - Right now we _always_ have a clip region, which is empty by default.
179 Why does this work at all? It probably doesn't. The server
180 distinguishes two cases, one where nothing is clipped (CT_NONE), and
181 one where there is a clip region (CT_REGION).
183 - Default clip region should be the full image
185 - Test if pseudo color still works. It does, but it also shows that
186 copying a pixman_indexed_t on every composite operation is not
187 going to fly. So, for now set_indexed() does not copy the
190 Also just the malloc() to allocate a pixman image shows up pretty
195 - Make all the setters not copy their arguments
197 - Possibly combined with going back to the stack allocated
198 approach that we already use for regions.
200 - Keep a cached pixman_image_t around for every picture. It would
201 have to be kept uptodate every time something changes about the
204 - Break the X server ABI and simply have the relevant parameter
205 stored in the pixman image. This would have the additional benefits
208 - We can get rid of the annoying repeat field which is duplicated
211 - We can use pixman_color_t and pixman_gradient_stop_t
212 etc. instead of the types that are defined in