Home My Page Projects Code Snippets Project Openings diderot
Summary Activity Tracker Tasks SCM

SCM Repository

[diderot] Annotation of /branches/cuda/TODO
ViewVC logotype

Annotation of /branches/cuda/TODO

Parent Directory Parent Directory | Revision Log Revision Log


Revision 1442 - (view) (download)
Original Path: trunk/TODO

1 : glk 1162 NOTE: GLK's approximate ranking of 8 most important tagged with
2 : jhr 1115 [GLK:1], [GLK:2], ...
3 :    
4 : glk 1156 ========================
5 :     SHORT TERM ============= (*needed* for streamlines & tractography)
6 :     ========================
7 : jhr 1115
8 : glk 1338 [GLK:2] Add sequence types (needed for evals & evecs)
9 : jhr 1115 syntax
10 :     types: ty '{' INT '}'
11 :     value construction: '{' e1 ',' … ',' en '}'
12 :     indexing: e '{' e '}'
13 :    
14 : glk 1338 [GLK:3] evals & evecs for symmetric tensor[2,2] and
15 : glk 1162 tensor[3,3] (requires sequences)
16 :    
17 : glk 1156 ability to emit/track/record variables into dynamically re-sized
18 : glk 1389 runtime output buffer
19 : jhr 1115
20 : glk 1442 [GLK:4] tensor fields from tensor images: Initially need at least
21 :     convolution on tensor[2,2] and tensor[3,3] (the same component-wise
22 :     convolution as for vectors).
23 : jhr 1115
24 : glk 1156 ========================
25 : glk 1162 SHORT-ISH TERM ========= (to make using Diderot less annoying to
26 :     ======================== program in, and slow to execute)
27 : jhr 1115
28 : glk 1167 Allow ".ddro" file extensions in addition to ".diderot"
29 :    
30 : glk 1204 Be able to output values of type tensor[2,2] and tensor[3,3];
31 :     (currently only scalars & vectors). Want to add some regression tests
32 :     based on this and currently can't
33 : glk 1167
34 : glk 1338 [GLK:1] Proper handling of stabilize method
35 : jhr 1115
36 : glk 1442 Convolution on general tensor images (order > 2)
37 :    
38 : glk 1162 allow "*" to represent "modulate": per-component multiplication of
39 :     vectors, and vectors only (not tensors of order 2 or higher). Once
40 :     sequences are implemented this should be removed: the operation is not
41 :     invariant WRT basis so it is not a legit vector computation.
42 :    
43 :     implicit type promotion of integers to reals where reals are
44 :     required (e.g. not exponentiation "^")
45 :    
46 : glk 1442 [Nick working on this] Save Diderot output to nrrd, instead of "mip.txt"
47 : jhr 1115 For grid of strands, save to similarly-shaped array
48 :     For list of strands, save to long 1-D (or 2-D for non-scalar output) list
49 :     For ragged things (like tractography output), will need to save both
50 :     complete list of values, as well as list of start indices and lengths
51 :     to index into complete list
52 :    
53 : glk 1442 [GLK:6] ability to declare a field so that probe positions are
54 : glk 1120 *always* "inside"; with various ways of mapping the known image values
55 :     to non-existant index locations. One possible syntax emphasizes that
56 :     there is a index mapping function that logically precedes convolution:
57 : glk 1162 F = bspln3 ⊛ (img ◦ clamp)
58 : glk 1120 F = bspln3 ⊛ (img ◦ repeat)
59 :     F = bspln3 ⊛ (img ◦ mirror)
60 :     where "◦" or "∘" is used to indicate function composition
61 : jhr 1115
62 : glk 1162 Level of differentiability in field type should be statement about how
63 :     much differentiation the program *needs*, rather than what the kernel
64 :     *provides*. The needed differentiability can be less than or equal to
65 :     the provided differentiability.
66 :    
67 : glk 1156 Use ∇⊗ etc. syntax
68 :     syntax [DONE]
69 :     typechecking
70 :     IL and codegen
71 : jhr 1115
72 : glk 1156 Add type aliases for color types
73 :     rgb = real{3}
74 :     rgba = real{4}
75 : jhr 1115
76 : glk 1389 Revisit how images are created within the language.
77 :     The "load" operator should probably go away, and its strangs
78 :     that strings are there only as a way to refer to nrrd filenames
79 :    
80 : jhr 1115 ==============================
81 : glk 1156 MEDIUM TERM ================== (*needed* for particles)
82 : jhr 1115 ==============================
83 :    
84 : glk 1442 [Lamont working on this] run-time birth of strands
85 : jhr 1115
86 :     "initially" supports lists
87 :    
88 : glk 1254 "initially" supports lists of positions output from different
89 :     initalization Diderot program (or output from the same program;
90 :     e.g. using output of iso2d.diderot for one isovalue to seed the input
91 :     to another invocation of the same program)
92 : jhr 1115
93 : glk 1442 [Lamont working on this] Communication between strands: they have to
94 :     be able to learn each other's state (at the previous iteration).
95 :     Early version of this can have the network of neighbors be completely
96 :     static (for running one strand/pixel image computations). Later
97 :     version with strands moving through the domain will require some
98 :     spatial data structure to optimize discovery of neighbors.
99 : jhr 1115
100 : glk 1156 ============================
101 :     MEDIUM-ISH TERM ============ (to make Diderot more useful/effective)
102 :     ============================
103 : jhr 1115
104 : glk 1442 [GLK:5] Want code-generation working for tensors of order three.
105 :     Order three matters for edge detection in scalar fields (to get
106 :     second derivatives of gradient magnitude), second derivatives
107 :     of vector fields (for some feature extraction), and first
108 :     derivatives of diffusion tensor fields.
109 :    
110 : glk 1156 Python/ctypes interface to run-time
111 : jhr 1115
112 : glk 1156 support for Python interop and GUI
113 : jhr 1115
114 : glk 1162 Allow integer exponentiation ("^2") to apply to square matrices,
115 :     to represent repeated matrix multiplication
116 :    
117 : glk 1156 Put small 1-D and 2-D fields, when reconstructed specifically by tent
118 :     and when differentiation is not needed, into faster texture buffers.
119 :     test/illust-vr.diderot is good example of program that uses multiple
120 :     such 1-D fields basically as lookup-table-based function evaluation
121 :    
122 :     extend norm (|exp|) to all tensor types [DONE for vectors and matrices]
123 :    
124 :     determinant ("det") for tensor[3,3]
125 :    
126 : jhr 1115 add ":" for tensor dot product (contracts out two indices
127 :     instead of one like •), valid for all pairs of tensors with
128 :     at least two indices
129 :    
130 : glk 1156 test/uninit.diderot:
131 :     documents need for better compiler error messages when output variables
132 :     are not initialized; the current messages are very cryptic
133 : jhr 1115
134 :     want: warnings when "D" (reserved for differentiation) is declared as
135 :     a variable name (get confusing error messages now)
136 :    
137 : glk 1156 ==============================
138 :     LONG TERM ==================== (make Diderot more interesting/attractive from
139 :     ============================== a research standpoint)
140 : jhr 1115
141 : glk 1156 IL support for higher-order tensor values (matrices, etc).
142 :     tensor construction [DONE]
143 :     tensor indexing [DONE]
144 :     tensor slicing
145 : jhr 1115
146 : glk 1156 Better handling of variables that determines the scope of a variable
147 :     based on its actual use, instead of where the user defined it. So,
148 :     for example, we should lift strand-invariant variables to global
149 :     scope. Also prune out useless variables, which should include field
150 :     variables after the translation to mid-il.
151 :    
152 :     test/vr-kcomp2.diderot: Add support for code like
153 :     (F1 if x else F2)@pos
154 :     This will require duplication of the continuation of the conditional
155 :     (but we should only duplicate over the live-range of the result of the
156 :     conditional.
157 :    
158 : glk 1338 [GLK:7] Want: non-trivial field expressions & functions.
159 : glk 1162 scalar fields from scalar fields F and G:
160 :     field#0(2)[] X = (sin(F) + 1.0)/2;
161 :     field#0(2)[] X = F*G;
162 :     scalar field of vector field magnitude:
163 : glk 1156 image(2)[2] Vimg = load(...);
164 :     field#0(2)[] Vlen = |Vimg ⊛ bspln3|;
165 : glk 1162 field of normalized vectors (for LIC and vector field feature extraction)
166 :     field#2(2)[2] F = ...
167 :     field#0(2)[2] V = normalize(F);
168 :     scalar field of gradient magnitude (for edge detection))
169 : glk 1156 field#2(2)[] F = Fimg ⊛ bspln3;
170 :     field#0(2)[] Gmag = |∇F|;
171 : glk 1162 scalar field of squared gradient magnitude (simpler to differentiate):
172 : glk 1156 field#2(2)[] F = Fimg ⊛ bspln3;
173 :     field#0(2)[] Gmsq = ∇F•∇F;
174 : glk 1162 There is value in having these, even if the differentiation of them is
175 :     not supported (hence the indication of "field#0" for these above)
176 : glk 1156
177 : jhr 1195 Introduce region types (syntax region(d), where d is the dimension of the
178 :     region. One useful operator would be
179 :     dom : field#k(d)[s] -> region(d)
180 :     Then the inside test could be written as
181 :     pos ∈ dom(F)
182 :     We could further extend this approach to allow geometric definitions of
183 :     regions. It might also be useful to do inside tests in world space,
184 :     instead of image space.
185 :    
186 : glk 1162 co- vs contra- index distinction
187 : glk 1156
188 : glk 1162 Permit field composition:
189 : glk 1156 field#2(3)[3] warp = bspln3 ⊛ warpData;
190 :     field#2(3)[] F = bspln3 ⊛ img;
191 :     field#2(3)[] Fwarp = F ◦ warp;
192 : glk 1162 So Fwarp(x) = F(warp(X)). Chain rule can be used for differentation.
193 :     This will be instrumental for expressing non-rigid registration
194 :     methods (but those will require co-vs-contra index distinction)
195 : glk 1156
196 : glk 1155 Allow the convolution to be specified either as a single 1D kernel
197 :     (as we have it now):
198 :     field#2(3)[] F = bspln3 ⊛ img;
199 :     or, as a tensor product of kernels, one for each axis, e.g.
200 :     field#0(3)[] F = (bspln3 ⊗ bspln3 ⊗ tent) ⊛ img;
201 : glk 1212 This is especially important for things like time-varying fields
202 :     and the use of scale-space in field visualization: one axis of the
203 :     must be convolved with a different kernel during probing.
204 :     What is very unclear is how, in such cases, we should notate the
205 :     gradient, when we only want to differentiate with respect to some
206 :     subset of the axes. One ambitious idea would be:
207 : glk 1162 field#0(3)[] Ft = (bspln3 ⊗ bspln3 ⊗ tent) ⊛ img; // 2D time-varying field
208 : glk 1204 field#0(2)[] F = lambda([x,y], Ft([x,y,42.0])) // restriction to time=42.0
209 :     vec2 grad = ∇F([x,y]); // 2D gradient
210 : glk 1155
211 : glk 1162 representation of tensor symmetry
212 : jhr 1115 (have to identify the group of index permutations that are symmetries)
213 :    
214 :     dot works on all tensors
215 :    
216 :     outer works on all tensors
217 :    
218 : glk 1204 Help for debugging Diderot programs: need to be able to uniquely
219 :     identify strands, and for particular strands that are known to behave
220 :     badly, do something like printf or other logging of their computations
221 :     and updates.
222 :    
223 :     Permit writing dimensionally general code: Have some statement of the
224 :     dimension of the world "W" (or have it be learned from one particular
225 :     field of interest), and then able to write "vec" instead of
226 :     "vec2/vec3", and perhaps "tensor[W,W]" instead of
227 :     "tensor[2,2]/tensor[3,3]"
228 :    
229 :     Traits: all things things that have boilerplate code (especially
230 :     volume rendering) should be expressed in terms of the unique
231 :     computational core. Different kinds of streamline/tractography
232 :     computation will be another example, as well as particle systems.
233 :    
234 : jhr 1115 Einstein summation notation
235 :    
236 :     "tensor comprehension" (like list comprehension)
237 :    
238 : glk 1204 Fields coming from different sources of data:
239 :     * triangular or tetrahedral meshes over 2D or 3D domains (of the
240 :     source produced by finite-element codes; these will come with their
241 :     own specialized kinds of reconstruction kernels, called "basis
242 :     functions" in this context)
243 :     * Large point clouds, with some radial basis function around each point,
244 :     which will be tuned by parameters of the point (at least one parameter
245 :     giving some notion of radius)
246 :    
247 : jhr 1115 ======================
248 :     BUGS =================
249 :     ======================
250 :    
251 :     test/zslice2.diderot:
252 :     // HEY (bug) bspln5 leads to problems ...
253 :     // uncaught exception Size [size]
254 :     // raised at c-target/c-target.sml:47.15-47.19
255 :     //field#4(3)[] F = img ⊛ bspln5;
256 : glk 1336

root@smlnj-gforge.cs.uchicago.edu
ViewVC Help
Powered by ViewVC 1.0.0