Home My Page Projects Code Snippets Project Openings diderot

SCM Repository

[diderot] Diff of /branches/ein16/TODO
 [diderot] / branches / ein16 / TODO

Diff of /branches/ein16/TODO

revision 1155, Sun May 8 14:43:30 2011 UTC revision 1167, Tue May 10 13:32:53 2011 UTC
# Line 1  Line 1
1  NOTE: GLK's approximate ranking of 5 most important tagged with  NOTE: GLK's approximate ranking of 8 most important tagged with
2  [GLK:1], [GLK:2], ...  [GLK:1], [GLK:2], ...
3
4  ==============================  ========================
5  other SHORT TERM =============  (including needed for LIC)  SHORT TERM ============= (*needed* for streamlines & tractography)
6  ==============================  ========================

Add a clamp function, which takes three arguments; either three scalars:
clamp(x, minval, maxval)  = max(minval, min(maxval, x))
or three vectors of the same size:
clamp([x,y], minvec, maxvec)  = [max(minvec[0], min(maxvec[0], x)),
max(minvec[1], min(maxvec[1], y))]
This would be useful in many current Diderot programs.
One question: clamp(x, minval, maxval) is the argument order
used in OpenCL and other places, but clamp(minval, maxval, x)
would be more consistent with lerp(minout, maxout, x).
7
8  Level of differentiability in field type should be statement about how  [GLK:3] Add sequence types (needed for evals & evecs)
much differentiation the program *needs*, rather than what the kernel
*provides*.  The needed differentiability can be less than or equal to
the provided differentiability.

[GLK:1] Add sequence types (needed for evals & evecs)
9      syntax      syntax
10          types: ty '{' INT '}'          types: ty '{' INT '}'
11          value construction: '{' e1 ',' … ',' en '}'          value construction: '{' e1 ',' … ',' en '}'
12          indexing: e '{' e '}'          indexing: e '{' e '}'
13
14  IL support for higher-order tensor values (matrices, etc).  [GLK:4] evals & evecs for symmetric tensor[2,2] and
15      tensor construction [DONE]  tensor[3,3] (requires sequences)
tensor indexing [DONE]
tensor slicing
verify that hessians work correctly [DONE]
16
17  Use ∇⊗ etc. syntax  ability to emit/track/record variables into dynamically re-sized
18      syntax [DONE]  runtime buffer
typechecking
IL and codegen
19
20  test/uninit.diderot:  tensor fields: convolution on general tensor images
documents need for better compiler error messages when output variables
are not initialized; the current messages are very cryptic
21
22  determinant ("det") for tensor[3,3]  ========================
23    SHORT-ISH TERM ========= (to make using Diderot less annoying to
24    ========================  program in, and slow to execute)
25
26  expand trace in mid to low translation  value-numbering optimization [DONE, but needs more testing]
27
28  value-numbering optimization  Allow ".ddro" file extensions in addition to ".diderot"
29
30  Add type aliases for color types  Be able to output values of type tensor[2,2] and tensor[3,3]
31      rgb = real{3}  (currently only scalars & vectors)
rgba = real{4}
32
33  ==============================  [GLK:1] Add a clamp function, which takes three arguments; either
34  MEDIUM TERM ================== (including needed for streamlines & tractography)  three scalars:
35  ==============================    clamp(lo, hi, x)  = max(lo, min(hi, x))
36    or three vectors of the same size:
37      clamp(lo, hi, [x,y])  = [max(lo[0], min(hi[0], x)),
38                               max(lo[1], min(hi[1], y))]
39    This would be useful in many current Diderot programs.
40    One question: clamp(x, lo, hi) is the argument order used in OpenCL
41    and other places, but clamp(lo, hi, x) is much more consistent with
42    lerp(lo, hi, x), hence GLK's preference
43
44  [GLK:1] evals & evecs for symmetric tensor[3,3] (requires sequences)  [GLK:2] Proper handling of stabilize method
45
46  [GLK:2] Save Diderot output to nrrd, instead of "mip.txt"  allow "*" to represent "modulate": per-component multiplication of
47    vectors, and vectors only (not tensors of order 2 or higher).  Once
48    sequences are implemented this should be removed: the operation is not
49    invariant WRT basis so it is not a legit vector computation.
50
51    implicit type promotion of integers to reals where reals are
52    required (e.g. not exponentiation "^")
53
54    [GLK:5] Save Diderot output to nrrd, instead of "mip.txt"
55    For grid of strands, save to similarly-shaped array    For grid of strands, save to similarly-shaped array
56    For list of strands, save to long 1-D (or 2-D for non-scalar output) list    For list of strands, save to long 1-D (or 2-D for non-scalar output) list
57    For ragged things (like tractography output), will need to save both    For ragged things (like tractography output), will need to save both
58      complete list of values, as well as list of start indices and lengths      complete list of values, as well as list of start indices and lengths
59      to index into complete list      to index into complete list
60
61  [GLK:3] Use of Teem's "hest" command-line parser for getting  [GLK:6] Use of Teem's "hest" command-line parser for getting
62  any input variables that are not defined in the source file  any input variables that are not defined in the source file
63
64  [GLK:4] ability to declare a field so that probe positions are  [GLK:7] ability to declare a field so that probe positions are
65  *always* "inside"; with various ways of mapping the known image values  *always* "inside"; with various ways of mapping the known image values
66  to non-existant index locations.  One possible syntax emphasizes that  to non-existant index locations.  One possible syntax emphasizes that
67  there is a index mapping function that logically precedes convolution:  there is a index mapping function that logically precedes convolution:
# Line 76  Line 70
70    F = bspln3 ⊛ (img ◦ mirror)    F = bspln3 ⊛ (img ◦ mirror)
71  where "◦" or "∘" is used to indicate function composition  where "◦" or "∘" is used to indicate function composition
72
73  extend norm (|exp|) to all tensor types [DONE for vectors and matrices]  Level of differentiability in field type should be statement about how
74    much differentiation the program *needs*, rather than what the kernel
75  ability to emit/track/record variables into dynamically re-sized  *provides*.  The needed differentiability can be less than or equal to
76  runtime buffer  the provided differentiability.

Want: allow X *= Y, X /= Y, X += Y, X -= Y to mean what they do in C,
provided that X*Y, X/Y, X+Y, X-Y are already supported.
Nearly every Diderot program would be simplified by this.
77
78  [GLK:5] Want: non-trivial field expressions & functions:  Use ∇⊗ etc. syntax
79    image(2)[2] Vimg = load(...);      syntax [DONE]
80    field#0(2)[] Vlen = |Vimg ⊛ bspln3|;      typechecking
81  to get a scalar field of vector length, or      IL and codegen
field#2(2)[] F = Fimg ⊛ bspln3;
field#0(2)[] Gmag = |∇F|;
to get a scalar field of gradient magnitude, or
field#2(2)[] F = Fimg ⊛ bspln3;
field#0(2)[] Gmsq = ∇F•∇F;
to get a scalar field of squared gradient magnitude, which is simpler
to differentiate.  However, there is value in having these, even if
the differentiation of them is not supported (hence the indication
of "field#0" for these above)

Want: ability to apply "normalize" to a field itself, e.g.
field#0(2)[2] V = normalize(Vimg ⊛ ctmr);
so that V(x) = normalize((Vimg ⊛ ctmr)(x)).
Having this would simplify expression of standard LIC method, and
would also help express other vector field expressions that arise
in vector field feature exraction.
82
83  tensor fields: convolution on general tensor images  Add type aliases for color types
84        rgb = real{3}
85        rgba = real{4}
86
87  ==============================  ==============================
88  other MEDIUM TERM ============ (needed for particles)  MEDIUM TERM ================== (*needed* for particles)
89  ==============================  ==============================
90
Put small 1-D and 2-D fields, when reconstructed specifically by tent
and when differentiation is not needed, into faster texture buffers.
test/illust-vr.diderot is good example of program that uses multiple
such 1-D fields basically as lookup-table-based function evaluation

91  run-time birth of strands  run-time birth of strands
92
93  "initially" supports lists  "initially" supports lists
# Line 124  Line 95
95  "initially" supports lists of positions output from  "initially" supports lists of positions output from
96  different initalization Diderot program  different initalization Diderot program
97
98  spatial data structure that permits strands' queries of neighbors  Communication between strands: they have to be able to learn each
99    other's state (at the previous iteration).  Early version of this can
100    have the network of neighbors be completely static (for running one
101    strand/pixel image computations).  Later version with strands moving
102    through the domain will require some spatial data structure to
103    optimize discovery of neighbors.
104
105    ============================
106    MEDIUM-ISH TERM ============ (to make Diderot more useful/effective)
107    ============================
108
109  proper handling of stabilize method  Python/ctypes interface to run-time
110
111  test/vr-kcomp2.diderot: Add support for code like  support for Python interop and GUI
112
113          (F1 if x else F2)@pos  Allow integer exponentiation ("^2") to apply to square matrices,
114    to represent repeated matrix multiplication
115
116  This will require duplication of the continuation of the conditional  Alow X *= Y, X /= Y, X += Y, X -= Y to mean what they do in C,
117  (but we should only duplicate over the live-range of the result of the  provided that X*Y, X/Y, X+Y, X-Y are already supported.
118  conditional.  Nearly every Diderot program would be simplified by this.
119
120    Put small 1-D and 2-D fields, when reconstructed specifically by tent
121    and when differentiation is not needed, into faster texture buffers.
122    test/illust-vr.diderot is good example of program that uses multiple
123    such 1-D fields basically as lookup-table-based function evaluation
124
125    expand trace in mid to low translation
126
127    extend norm (|exp|) to all tensor types [DONE for vectors and matrices]
128
129    determinant ("det") for tensor[3,3]
130
131  add ":" for tensor dot product (contracts out two indices  add ":" for tensor dot product (contracts out two indices
132  instead of one like •), valid for all pairs of tensors with  instead of one like •), valid for all pairs of tensors with
133  at least two indices  at least two indices
134
135  ==============================  test/uninit.diderot:
136  other MEDIUM TERM ============  documents need for better compiler error messages when output variables
137  ==============================  are not initialized; the current messages are very cryptic
138
139  want: warnings when "D" (reserved for differentiation) is declared as  want: warnings when "D" (reserved for differentiation) is declared as
140  a variable name (get confusing error messages now)  a variable name (get confusing error messages now)
141
142  support for Python interop and GUI  ==============================
143    LONG TERM ==================== (make Diderot more interesting/attractive from
144    ==============================  a research standpoint)
145
146  Python/ctypes interface to run-time  IL support for higher-order tensor values (matrices, etc).
147        tensor construction [DONE]
148        tensor indexing [DONE]
149        tensor slicing
150        verify that hessians work correctly [DONE]
151
152    Better handling of variables that determines the scope of a variable
153    based on its actual use, instead of where the user defined it.  So,
154    for example, we should lift strand-invariant variables to global
155    scope.  Also prune out useless variables, which should include field
156    variables after the translation to mid-il.
157
158    test/vr-kcomp2.diderot: Add support for code like
159            (F1 if x else F2)@pos
160    This will require duplication of the continuation of the conditional
161    (but we should only duplicate over the live-range of the result of the
162    conditional.
163
164    [GLK:8] Want: non-trivial field expressions & functions.
165    scalar fields from scalar fields F and G:
166      field#0(2)[] X = (sin(F) + 1.0)/2;
167      field#0(2)[] X = F*G;
168    scalar field of vector field magnitude:
170      field#0(2)[] Vlen = |Vimg ⊛ bspln3|;
171    field of normalized vectors (for LIC and vector field feature extraction)
172      field#2(2)[2] F = ...
173      field#0(2)[2] V = normalize(F);
174    scalar field of gradient magnitude (for edge detection))
175      field#2(2)[] F = Fimg ⊛ bspln3;
176      field#0(2)[] Gmag = |∇F|;
177    scalar field of squared gradient magnitude (simpler to differentiate):
178      field#2(2)[] F = Fimg ⊛ bspln3;
179      field#0(2)[] Gmsq = ∇F•∇F;
180    There is value in having these, even if the differentiation of them is
181    not supported (hence the indication of "field#0" for these above)
182
183    co- vs contra- index distinction
184
185    Permit field composition:
186      field#2(3)[3] warp = bspln3 ⊛ warpData;
187      field#2(3)[] F = bspln3 ⊛ img;
188      field#2(3)[] Fwarp = F ◦ warp;
189    So Fwarp(x) = F(warp(X)).  Chain rule can be used for differentation.
190    This will be instrumental for expressing non-rigid registration
191    methods (but those will require co-vs-contra index distinction)
192
193  Allow the convolution to be specified either as a single 1D kernel  Allow the convolution to be specified either as a single 1D kernel
194  (as we have it now):  (as we have it now):
# Line 158  Line 197
197    field#0(3)[] F = (bspln3 ⊗ bspln3 ⊗ tent) ⊛ img;    field#0(3)[] F = (bspln3 ⊗ bspln3 ⊗ tent) ⊛ img;
198  This is especially important for things like time-varying data, or  This is especially important for things like time-varying data, or
199  other multi-dimensional fields where one axis of the domain is very  other multi-dimensional fields where one axis of the domain is very
200  different from the rest.  What is very unclear is how, in such cases,  different from the rest, and hence must be treated separately when
201    it comes to convolution.  What is very unclear is how, in such cases,
202  we should notate the gradient, when we only want to differentiate with  we should notate the gradient, when we only want to differentiate with
203  respect to some of the axes.  respect to some subset of the axes.  One ambitious idea would be:
204      field#0(3)[] Ft = (bspln3 ⊗ bspln3 ⊗ tent) ⊛ img; // 2D time-varying field
205  ==============================    field#0(2)[] F = lambda([x,y], Ft([x,y,42.0]))    // restriction to time=42.0
==============================

Better handling of variables that determines the scope of a variable
based on its actual use, instead of where the user defined it.  So,
for example, we should lift strand-invariant variables to global
scope.  Also prune out useless variables, which should include field
variables after the translation to mid-il.

co- vs contra- index distinction
207
208  some indication of tensor symmetry  representation of tensor symmetry
209  (have to identify the group of index permutations that are symmetries)  (have to identify the group of index permutations that are symmetries)
210
211  dot works on all tensors  dot works on all tensors

Legend:
 Removed from v.1155 changed lines Added in v.1167