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 1295 - (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 : jhr 1257 Remove CL from compiler
9 :    
10 : glk 1162 [GLK:3] Add sequence types (needed for evals & evecs)
11 : jhr 1115 syntax
12 :     types: ty '{' INT '}'
13 :     value construction: '{' e1 ',' … ',' en '}'
14 :     indexing: e '{' e '}'
15 :    
16 : glk 1162 [GLK:4] evals & evecs for symmetric tensor[2,2] and
17 :     tensor[3,3] (requires sequences)
18 :    
19 : glk 1156 ability to emit/track/record variables into dynamically re-sized
20 :     runtime buffer
21 : jhr 1115
22 : glk 1156 tensor fields: convolution on general tensor images
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 : jhr 1246 value-numbering optimization [DONE]
29 : jhr 1115
30 : glk 1167 Allow ".ddro" file extensions in addition to ".diderot"
31 :    
32 : glk 1204 Be able to output values of type tensor[2,2] and tensor[3,3];
33 :     (currently only scalars & vectors). Want to add some regression tests
34 :     based on this and currently can't
35 : glk 1167
36 : glk 1162 [GLK:1] Add a clamp function, which takes three arguments; either
37 :     three scalars:
38 :     clamp(lo, hi, x) = max(lo, min(hi, x))
39 :     or three vectors of the same size:
40 :     clamp(lo, hi, [x,y]) = [max(lo[0], min(hi[0], x)),
41 :     max(lo[1], min(hi[1], y))]
42 :     This would be useful in many current Diderot programs.
43 :     One question: clamp(x, lo, hi) is the argument order used in OpenCL
44 :     and other places, but clamp(lo, hi, x) is much more consistent with
45 :     lerp(lo, hi, x), hence GLK's preference
46 : jhr 1295 [DONE]
47 : jhr 1115
48 : glk 1162 [GLK:2] Proper handling of stabilize method
49 :    
50 :     allow "*" to represent "modulate": per-component multiplication of
51 :     vectors, and vectors only (not tensors of order 2 or higher). Once
52 :     sequences are implemented this should be removed: the operation is not
53 :     invariant WRT basis so it is not a legit vector computation.
54 :    
55 :     implicit type promotion of integers to reals where reals are
56 :     required (e.g. not exponentiation "^")
57 :    
58 :     [GLK:5] Save Diderot output to nrrd, instead of "mip.txt"
59 : jhr 1115 For grid of strands, save to similarly-shaped array
60 :     For list of strands, save to long 1-D (or 2-D for non-scalar output) list
61 :     For ragged things (like tractography output), will need to save both
62 :     complete list of values, as well as list of start indices and lengths
63 :     to index into complete list
64 :    
65 : glk 1162 [GLK:6] Use of Teem's "hest" command-line parser for getting
66 : glk 1212 any "input" variables that are not defined in the source file.
67 : jhr 1115
68 : glk 1162 [GLK:7] ability to declare a field so that probe positions are
69 : glk 1120 *always* "inside"; with various ways of mapping the known image values
70 :     to non-existant index locations. One possible syntax emphasizes that
71 :     there is a index mapping function that logically precedes convolution:
72 : glk 1162 F = bspln3 ⊛ (img ◦ clamp)
73 : glk 1120 F = bspln3 ⊛ (img ◦ repeat)
74 :     F = bspln3 ⊛ (img ◦ mirror)
75 :     where "◦" or "∘" is used to indicate function composition
76 : jhr 1115
77 : glk 1162 Level of differentiability in field type should be statement about how
78 :     much differentiation the program *needs*, rather than what the kernel
79 :     *provides*. The needed differentiability can be less than or equal to
80 :     the provided differentiability.
81 :    
82 : glk 1156 Use ∇⊗ etc. syntax
83 :     syntax [DONE]
84 :     typechecking
85 :     IL and codegen
86 : jhr 1115
87 : glk 1156 Add type aliases for color types
88 :     rgb = real{3}
89 :     rgba = real{4}
90 : jhr 1115
91 :     ==============================
92 : glk 1156 MEDIUM TERM ================== (*needed* for particles)
93 : jhr 1115 ==============================
94 :    
95 :     run-time birth of strands
96 :    
97 :     "initially" supports lists
98 :    
99 : glk 1254 "initially" supports lists of positions output from different
100 :     initalization Diderot program (or output from the same program;
101 :     e.g. using output of iso2d.diderot for one isovalue to seed the input
102 :     to another invocation of the same program)
103 : jhr 1115
104 : glk 1156 Communication between strands: they have to be able to learn each
105 :     other's state (at the previous iteration). Early version of this can
106 :     have the network of neighbors be completely static (for running one
107 :     strand/pixel image computations). Later version with strands moving
108 :     through the domain will require some spatial data structure to
109 :     optimize discovery of neighbors.
110 : jhr 1115
111 : glk 1156 ============================
112 :     MEDIUM-ISH TERM ============ (to make Diderot more useful/effective)
113 :     ============================
114 : jhr 1115
115 : glk 1156 Python/ctypes interface to run-time
116 : jhr 1115
117 : glk 1156 support for Python interop and GUI
118 : jhr 1115
119 : glk 1162 Allow integer exponentiation ("^2") to apply to square matrices,
120 :     to represent repeated matrix multiplication
121 :    
122 : glk 1156 Alow X *= Y, X /= Y, X += Y, X -= Y to mean what they do in C,
123 :     provided that X*Y, X/Y, X+Y, X-Y are already supported.
124 :     Nearly every Diderot program would be simplified by this.
125 : jhr 1115
126 : glk 1156 Put small 1-D and 2-D fields, when reconstructed specifically by tent
127 :     and when differentiation is not needed, into faster texture buffers.
128 :     test/illust-vr.diderot is good example of program that uses multiple
129 :     such 1-D fields basically as lookup-table-based function evaluation
130 :    
131 :     expand trace in mid to low translation
132 :    
133 :     extend norm (|exp|) to all tensor types [DONE for vectors and matrices]
134 :    
135 :     determinant ("det") for tensor[3,3]
136 :    
137 : jhr 1115 add ":" for tensor dot product (contracts out two indices
138 :     instead of one like •), valid for all pairs of tensors with
139 :     at least two indices
140 :    
141 : glk 1156 test/uninit.diderot:
142 :     documents need for better compiler error messages when output variables
143 :     are not initialized; the current messages are very cryptic
144 : jhr 1115
145 :     want: warnings when "D" (reserved for differentiation) is declared as
146 :     a variable name (get confusing error messages now)
147 :    
148 : glk 1156 ==============================
149 :     LONG TERM ==================== (make Diderot more interesting/attractive from
150 :     ============================== a research standpoint)
151 : jhr 1115
152 : glk 1156 IL support for higher-order tensor values (matrices, etc).
153 :     tensor construction [DONE]
154 :     tensor indexing [DONE]
155 :     tensor slicing
156 :     verify that hessians work correctly [DONE]
157 : jhr 1115
158 : glk 1156 Better handling of variables that determines the scope of a variable
159 :     based on its actual use, instead of where the user defined it. So,
160 :     for example, we should lift strand-invariant variables to global
161 :     scope. Also prune out useless variables, which should include field
162 :     variables after the translation to mid-il.
163 :    
164 :     test/vr-kcomp2.diderot: Add support for code like
165 :     (F1 if x else F2)@pos
166 :     This will require duplication of the continuation of the conditional
167 :     (but we should only duplicate over the live-range of the result of the
168 :     conditional.
169 :    
170 : glk 1162 [GLK:8] Want: non-trivial field expressions & functions.
171 :     scalar fields from scalar fields F and G:
172 :     field#0(2)[] X = (sin(F) + 1.0)/2;
173 :     field#0(2)[] X = F*G;
174 :     scalar field of vector field magnitude:
175 : glk 1156 image(2)[2] Vimg = load(...);
176 :     field#0(2)[] Vlen = |Vimg ⊛ bspln3|;
177 : glk 1162 field of normalized vectors (for LIC and vector field feature extraction)
178 :     field#2(2)[2] F = ...
179 :     field#0(2)[2] V = normalize(F);
180 :     scalar field of gradient magnitude (for edge detection))
181 : glk 1156 field#2(2)[] F = Fimg ⊛ bspln3;
182 :     field#0(2)[] Gmag = |∇F|;
183 : glk 1162 scalar field of squared gradient magnitude (simpler to differentiate):
184 : glk 1156 field#2(2)[] F = Fimg ⊛ bspln3;
185 :     field#0(2)[] Gmsq = ∇F•∇F;
186 : glk 1162 There is value in having these, even if the differentiation of them is
187 :     not supported (hence the indication of "field#0" for these above)
188 : glk 1156
189 : jhr 1195 Introduce region types (syntax region(d), where d is the dimension of the
190 :     region. One useful operator would be
191 :     dom : field#k(d)[s] -> region(d)
192 :     Then the inside test could be written as
193 :     pos ∈ dom(F)
194 :     We could further extend this approach to allow geometric definitions of
195 :     regions. It might also be useful to do inside tests in world space,
196 :     instead of image space.
197 :    
198 : glk 1162 co- vs contra- index distinction
199 : glk 1156
200 : glk 1162 Permit field composition:
201 : glk 1156 field#2(3)[3] warp = bspln3 ⊛ warpData;
202 :     field#2(3)[] F = bspln3 ⊛ img;
203 :     field#2(3)[] Fwarp = F ◦ warp;
204 : glk 1162 So Fwarp(x) = F(warp(X)). Chain rule can be used for differentation.
205 :     This will be instrumental for expressing non-rigid registration
206 :     methods (but those will require co-vs-contra index distinction)
207 : glk 1156
208 : glk 1155 Allow the convolution to be specified either as a single 1D kernel
209 :     (as we have it now):
210 :     field#2(3)[] F = bspln3 ⊛ img;
211 :     or, as a tensor product of kernels, one for each axis, e.g.
212 :     field#0(3)[] F = (bspln3 ⊗ bspln3 ⊗ tent) ⊛ img;
213 : glk 1212 This is especially important for things like time-varying fields
214 :     and the use of scale-space in field visualization: one axis of the
215 :     must be convolved with a different kernel during probing.
216 :     What is very unclear is how, in such cases, we should notate the
217 :     gradient, when we only want to differentiate with respect to some
218 :     subset of the axes. One ambitious idea would be:
219 : glk 1162 field#0(3)[] Ft = (bspln3 ⊗ bspln3 ⊗ tent) ⊛ img; // 2D time-varying field
220 : glk 1204 field#0(2)[] F = lambda([x,y], Ft([x,y,42.0])) // restriction to time=42.0
221 :     vec2 grad = ∇F([x,y]); // 2D gradient
222 : glk 1155
223 : glk 1204 Tensors of order 3 (e.g. gradients of diffusion tensor fields, or
224 :     hessians of vector fields) and order 4 (e.g. Hessians of diffusion
225 :     tensor fields).
226 :    
227 : glk 1162 representation of tensor symmetry
228 : jhr 1115 (have to identify the group of index permutations that are symmetries)
229 :    
230 :     dot works on all tensors
231 :    
232 :     outer works on all tensors
233 :    
234 : glk 1204 Help for debugging Diderot programs: need to be able to uniquely
235 :     identify strands, and for particular strands that are known to behave
236 :     badly, do something like printf or other logging of their computations
237 :     and updates.
238 :    
239 :     Permit writing dimensionally general code: Have some statement of the
240 :     dimension of the world "W" (or have it be learned from one particular
241 :     field of interest), and then able to write "vec" instead of
242 :     "vec2/vec3", and perhaps "tensor[W,W]" instead of
243 :     "tensor[2,2]/tensor[3,3]"
244 :    
245 :     Traits: all things things that have boilerplate code (especially
246 :     volume rendering) should be expressed in terms of the unique
247 :     computational core. Different kinds of streamline/tractography
248 :     computation will be another example, as well as particle systems.
249 :    
250 : jhr 1115 Einstein summation notation
251 :    
252 :     "tensor comprehension" (like list comprehension)
253 :    
254 : glk 1204 Fields coming from different sources of data:
255 :     * triangular or tetrahedral meshes over 2D or 3D domains (of the
256 :     source produced by finite-element codes; these will come with their
257 :     own specialized kinds of reconstruction kernels, called "basis
258 :     functions" in this context)
259 :     * Large point clouds, with some radial basis function around each point,
260 :     which will be tuned by parameters of the point (at least one parameter
261 :     giving some notion of radius)
262 :    
263 : jhr 1115 ======================
264 :     BUGS =================
265 :     ======================
266 :    
267 :     test/zslice2.diderot:
268 :     // HEY (bug) bspln5 leads to problems ...
269 :     // uncaught exception Size [size]
270 :     // raised at c-target/c-target.sml:47.15-47.19
271 :     //field#4(3)[] F = img ⊛ bspln5;

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