2 - Test implementations against each other
3 - Test both with and without the operator strength reduction.
4 They shold be identical.
8 - Use MM_HINT_NTA instead of MM_HINT_T0
10 - Use of fbCompositeOver_x888x8x8888sse2()
12 - Update the RLEASING file
14 - Things to keep in mind if breaking ABI:
16 - There should be a guard #ifndef I_AM_EITHER_CAIRO_OR_THE_X_SERVER
18 - X server will require 16.16 essentially forever. Can we get
19 the required precision by simply adding offset_x/y to the
20 relevant rendering API?
22 - Get rid of workaround for X server bug.
24 - pixman_image_set_indexed() should copy its argument, and X
25 should be ported over to use a pixman_image as the
26 representation of a Picture, rather than creating one on each
29 - We should get rid of pixman_set_static_pointers()
31 - We should get rid of the various trapezoid helper functions().
32 (They only exist because they are theoretically available to
35 - 16 bit regions should be deleted
37 - There should only be one trap rasterization API.
39 - The PIXMAN_g8/c8/etc formats should use the A channel
40 to indicate the actual depth. That way PIXMAN_x4c4 and PIXMAN_c8
43 - Maybe bite the bullet and make configure.ac generate a pixman-types.h
44 file that can be included from pixman.h to avoid the #ifdef magic
47 - Make pixman_region_point_in() survive a NULL box, then fix up
50 - Possibly look into inlining the fetch functions
52 - There is a bug with source clipping demonstrated by clip-test in the
53 test directory. If we interprete source clipping as given in
54 destination coordinates, which is probably the only sane choice,
55 then the result should have two red bars down the sides.
59 - Add a general way of dealing with architecture specific
60 fast-paths. The current idea is to have each operation that can
61 be optimized is called through a function pointer that is
62 initially set to an initialization function that is responsible for
63 setting the function pointer to the appropriate fast-path.
65 - Go through things marked FIXME
67 - Add calls to prepare and finish access where necessary. grep for
68 ACCESS_MEM, and make sure they are correctly wrapped in prepare
71 - restore READ/WRITE in the fbcompose combiners since they sometimes
72 store directly to destination drawables.
74 - It probably makes sense to move the more strange X region API
75 into pixman as well, but guarded with PIXMAN_XORG_COMPATIBILITY
77 - Reinstate the FbBits typedef? At the moment we don't
78 even have the FbBits type; we just use uint32_t everywhere.
80 Keith says in bug 2335:
82 The 64-bit code in fb (pixman) is probably broken; it hasn't been
83 used in quite some time as PCI (and AGP) is 32-bits wide, so
84 doing things 64-bits at a time is a net loss. To quickly fix
85 this, I suggest just using 32-bit datatypes by setting
86 IC_SHIFT to 5 for all machines.
88 - Consider optimizing the 8/16 bit solid fills in pixman-util.c by
89 storing more than one value at a time.
91 - Add an image cache to prevent excessive malloc/free. Note that pixman
92 needs to be thread safe when used from cairo.
94 - Moving to 24.8 coordinates. This is tricky because X is still
95 defined as 16.16 and will be basically forever. It's possible we
96 could do this by adding extra offset_x/y parameters to the
97 trapezoid calls. The X server could then just call the API with
98 (0, 0). Cairo would have to make sure that the delta *within* a
99 batch of trapezoids does not exceed 16 bit.
101 - Consider adding actual backends. Brain dump:
103 A backend is something that knows how to
106 - Composite three images
107 - Rasterize trapezoids
108 - Do solid fills and blits
110 These operations are provided by a vtable that the backend will
111 create when it is initialized. Initial backends:
118 When the SIMD backends are initialized, they will be passed a
119 pointer to the Plain Old C backend that they can use for fallback
122 Images would gain a vtable as well that would contain things like
127 (Or even read_patch/write_patch as suggested by Keith a while
130 This could simplify the compositing code considerably.
132 - Review the pixman_format_code_t enum to make sure it will support
133 future formats. Some formats we will probably need:
135 ARGB/ABGR with 16/32/64 bit integer/floating channels
139 Also we may need the ability to distinguish between PICT_c8 and
140 PICT_x4c4. (This could be done by interpreting the A channel as
141 the depth for TYPE_COLOR and TYPE_GRAY formats).
143 A possibility may be to reserve the two top bits and make them
144 encode "number of places to shift the channel widths given" Since
145 these bits are 00 at the moment everything will continue to work,
146 but these additional widths will be allowed:
148 All even widths between 18-32
149 All multiples of four widths between 33 and 64
150 All multiples of eight between 64 and 128
152 This means things like r21g22b21 won't work - is that worth
153 worrying about? I don't think so. And of course the bpp field
154 can't handle a depth of over 256, so > 64 bit channels arent'
155 really all that useful.
157 We could reserve one extra bit to indicate floating point, but
160 PIXMAN_TYPE_ARGB_FLOAT
161 PIXMAN_TYPE_BGRA_FLOAT
164 image types. With five bits we can support up to 32 different
165 format types, which should be enough for everybody, even if we
166 decide to support all the various video formats here:
168 http://www.fourcc.org/yuv.php
170 It may make sense to have a PIXMAN_TYPE_YUV, and then use the
171 channel bits to specify the exact subtype.
173 Another possibility is to add
178 where the channel widths would get 16 and 32 added to them,
181 What about color spaces such a linear vs. srGB etc.?
186 - Use pixmanFillsse2 and pixmanBltsse2
188 - Be consistent about calling sse2 sse2
190 - Rename "SSE" to "MMX_EXTENSIONS". (Deleted mmx extensions).
192 - Commented-out uses of fbCompositeCopyAreasse2()
194 - Consider whether calling regions region16 is really such a great
195 idea. Vlad wants 32 bit regions for Cairo. This will break X server
196 ABI, but should otherwise be mostly harmless, though a
197 pixman_region_get_boxes16() may be useful.
199 - Altivec signal issue (Company has fix, there is also a patch by
202 - Behdad's MMX issue - see list
205 - Crashes in Mozilla because of unaligned stack. Possible fixes
206 - Make use of gcc 4.2 feature to align the stack
207 - Write some sort of trampoline that aligns the stack
208 before calling SSE functions.
210 - Get rid of the switch-of-doom; replace it with a big table
211 describing the various fast paths.
213 - Make source clipping optional.
214 - done: source clipping happens through an indirection.
215 still needs to make the indirection settable. (And call it
218 - Run cairo test suite; fix bugs
219 - one bug in source-scale-clip
221 - Remove the warning suppression in the ACCESS_MEM macro and fix the
222 warnings that are real
225 - make the wrapper functions global instead of image specific
226 - this won't work since pixman is linked to both fb and wfb
228 - Add non-mmx solid fill
230 - Make sure the endian-ness macros are defined correctly.
232 - The rectangles in a region probably shouldn't be returned const as
233 the X server will be changing them.
235 - Right now we _always_ have a clip region, which is empty by default.
236 Why does this work at all? It probably doesn't. The server
237 distinguishes two cases, one where nothing is clipped (CT_NONE), and
238 one where there is a clip region (CT_REGION).
240 - Default clip region should be the full image
242 - Test if pseudo color still works. It does, but it also shows that
243 copying a pixman_indexed_t on every composite operation is not
244 going to fly. So, for now set_indexed() does not copy the
247 Also just the malloc() to allocate a pixman image shows up pretty
252 - Make all the setters not copy their arguments
254 - Possibly combined with going back to the stack allocated
255 approach that we already use for regions.
257 - Keep a cached pixman_image_t around for every picture. It would
258 have to be kept uptodate every time something changes about the
261 - Break the X server ABI and simply have the relevant parameter
262 stored in the pixman image. This would have the additional benefits
265 - We can get rid of the annoying repeat field which is duplicated
268 - We can use pixman_color_t and pixman_gradient_stop_t
269 etc. instead of the types that are defined in