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 
============================== 
======================== 
7 


8 
Add a clamp function, which takes three arguments; either three scalars: 
Remove CL from compiler 

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). 





Level of differentiability in field type should be statement about how 


much differentiation the program *needs*, rather than what the kernel 


*provides*. The needed differentiability can be less than or equal to 


the provided differentiability. 

9 


10 
[GLK:1] Add sequence types (needed for evals & evecs) 
[GLK:2] Add sequence types (needed for evals & evecs) 
11 
syntax 
syntax 
12 
types: ty '{' INT '}' 
types: ty '{' INT '}' 
13 
value construction: '{' e1 ',' … ',' en '}' 
value construction: '{' e1 ',' … ',' en '}' 
14 
indexing: e '{' e '}' 
indexing: e '{' e '}' 
15 


16 
IL support for higherorder tensor values (matrices, etc). 
[GLK:3] evals & evecs for symmetric tensor[2,2] and 
17 
tensor construction [DONE] 
tensor[3,3] (requires sequences) 

tensor indexing [DONE] 


tensor slicing 


verify that hessians work correctly [DONE] 

18 


19 
Use ∇⊗ etc. syntax 
ability to emit/track/record variables into dynamically resized 
20 
syntax [DONE] 
runtime buffer 

typechecking 


IL and codegen 

21 


22 
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 

23 


24 
determinant ("det") for tensor[3,3] 
======================== 
25 

SHORTISH TERM ========= (to make using Diderot less annoying to 
26 

======================== program in, and slow to execute) 
27 


28 
expand trace in mid to low translation 
valuenumbering optimization [DONE] 
29 


30 
valuenumbering optimization 
Allow ".ddro" file extensions in addition to ".diderot" 
31 


32 
Add type aliases for color types 
Be able to output values of type tensor[2,2] and tensor[3,3]; 
33 
rgb = real{3} 
(currently only scalars & vectors). Want to add some regression tests 
34 
rgba = real{4} 
based on this and currently can't 
35 


36 
============================== 
[GLK:1] Proper handling of stabilize method 
37 
MEDIUM TERM ================== (including needed for streamlines & tractography) 

38 
============================== 
allow "*" to represent "modulate": percomponent 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 
[GLK:1] evals & evecs for symmetric tensor[3,3] (requires sequences) 
implicit type promotion of integers to reals where reals are 
44 

required (e.g. not exponentiation "^") 
45 


46 
[GLK:2] Save Diderot output to nrrd, instead of "mip.txt" 
[GLK:4] Save Diderot output to nrrd, instead of "mip.txt" 
47 
For grid of strands, save to similarlyshaped array 
For grid of strands, save to similarlyshaped array 
48 
For list of strands, save to long 1D (or 2D for nonscalar output) list 
For list of strands, save to long 1D (or 2D for nonscalar output) list 
49 
For ragged things (like tractography output), will need to save both 
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 
complete list of values, as well as list of start indices and lengths 
51 
to index into complete list 
to index into complete list 
52 


53 
[GLK:3] Use of Teem's "hest" commandline parser for getting 
[GLK:5] Use of Teem's "hest" commandline parser for getting 
54 
any input variables that are not defined in the source file 
any "input" variables that are not defined in the source file. 
55 


56 
[GLK:4] ability to declare a field so that probe positions are 
[GLK:6] ability to declare a field so that probe positions are 
57 
*always* "inside"; with various ways of mapping the known image values 
*always* "inside"; with various ways of mapping the known image values 
58 
to nonexistant index locations. One possible syntax emphasizes that 
to nonexistant index locations. One possible syntax emphasizes that 
59 
there is a index mapping function that logically precedes convolution: 
there is a index mapping function that logically precedes convolution: 
62 
F = bspln3 ⊛ (img ◦ mirror) 
F = bspln3 ⊛ (img ◦ mirror) 
63 
where "◦" or "∘" is used to indicate function composition 
where "◦" or "∘" is used to indicate function composition 
64 


65 
extend norm (exp) to all tensor types [DONE for vectors and matrices] 
Level of differentiability in field type should be statement about how 
66 

much differentiation the program *needs*, rather than what the kernel 
67 
ability to emit/track/record variables into dynamically resized 
*provides*. The needed differentiability can be less than or equal to 
68 
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, XY are already supported. 


Nearly every Diderot program would be simplified by this. 

69 


70 
[GLK:5] Want: nontrivial field expressions & functions: 
Use ∇⊗ etc. syntax 
71 
image(2)[2] Vimg = load(...); 
syntax [DONE] 
72 
field#0(2)[] Vlen = Vimg ⊛ bspln3; 
typechecking 
73 
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. 

74 


75 
tensor fields: convolution on general tensor images 
Add type aliases for color types 
76 

rgb = real{3} 
77 

rgba = real{4} 
78 


79 
============================== 
============================== 
80 
other MEDIUM TERM ============ (needed for particles) 
MEDIUM TERM ================== (*needed* for particles) 
81 
============================== 
============================== 
82 



Put small 1D and 2D fields, when reconstructed specifically by tent 


and when differentiation is not needed, into faster texture buffers. 


test/illustvr.diderot is good example of program that uses multiple 


such 1D fields basically as lookuptablebased function evaluation 




83 
runtime birth of strands 
runtime birth of strands 
84 


85 
"initially" supports lists 
"initially" supports lists 
86 


87 
"initially" supports lists of positions output from 
"initially" supports lists of positions output from different 
88 
different initalization Diderot program 
initalization Diderot program (or output from the same program; 
89 

e.g. using output of iso2d.diderot for one isovalue to seed the input 
90 

to another invocation of the same program) 
91 


92 

Communication between strands: they have to be able to learn each 
93 

other's state (at the previous iteration). Early version of this can 
94 

have the network of neighbors be completely static (for running one 
95 

strand/pixel image computations). Later version with strands moving 
96 

through the domain will require some spatial data structure to 
97 

optimize discovery of neighbors. 
98 


99 

============================ 
100 

MEDIUMISH TERM ============ (to make Diderot more useful/effective) 
101 

============================ 
102 


103 
spatial data structure that permits strands' queries of neighbors 
Python/ctypes interface to runtime 
104 


105 
proper handling of stabilize method 
support for Python interop and GUI 
106 


107 
test/vrkcomp2.diderot: Add support for code like 
Allow integer exponentiation ("^2") to apply to square matrices, 
108 

to represent repeated matrix multiplication 
109 


110 
(F1 if x else F2)@pos 
Put small 1D and 2D fields, when reconstructed specifically by tent 
111 

and when differentiation is not needed, into faster texture buffers. 
112 

test/illustvr.diderot is good example of program that uses multiple 
113 

such 1D fields basically as lookuptablebased function evaluation 
114 


115 
This will require duplication of the continuation of the conditional 
expand trace in mid to low translation [DONE] 
116 
(but we should only duplicate over the liverange of the result of the 

117 
conditional. 
extend norm (exp) to all tensor types [DONE for vectors and matrices] 
118 


119 

determinant ("det") for tensor[3,3] 
120 


121 
add ":" for tensor dot product (contracts out two indices 
add ":" for tensor dot product (contracts out two indices 
122 
instead of one like •), valid for all pairs of tensors with 
instead of one like •), valid for all pairs of tensors with 
123 
at least two indices 
at least two indices 
124 


125 
============================== 
test/uninit.diderot: 
126 
other MEDIUM TERM ============ 
documents need for better compiler error messages when output variables 
127 
============================== 
are not initialized; the current messages are very cryptic 
128 


129 
want: warnings when "D" (reserved for differentiation) is declared as 
want: warnings when "D" (reserved for differentiation) is declared as 
130 
a variable name (get confusing error messages now) 
a variable name (get confusing error messages now) 
131 



support for Python interop and GUI 





Python/ctypes interface to runtime 





Allow the convolution to be specified either as a single 1D kernel 


(as we have it now): 


field#2(3)[] F = bspln3 ⊛ img; 


or, as a tensor product of kernels, one for each axis, e.g. 


field#0(3)[] F = (bspln3 ⊗ bspln3 ⊗ tent) ⊛ img; 


This is especially important for things like timevarying data, or 


other multidimensional fields where one axis of the domain is very 


different from the rest. What is very unclear is how, in such cases, 


we should notate the gradient, when we only want to differentiate with 


respect to some of the axes. 





============================== 


LONG TERM ==================== 

132 
============================== 
============================== 
133 

LONG TERM ==================== (make Diderot more interesting/attractive from 
134 

============================== a research standpoint) 
135 


136 

IL support for higherorder tensor values (matrices, etc). 
137 

tensor construction [DONE] 
138 

tensor indexing [DONE] 
139 

tensor slicing 
140 

verify that hessians work correctly [DONE] 
141 


142 
Better handling of variables that determines the scope of a variable 
Better handling of variables that determines the scope of a variable 
143 
based on its actual use, instead of where the user defined it. So, 
based on its actual use, instead of where the user defined it. So, 
145 
scope. Also prune out useless variables, which should include field 
scope. Also prune out useless variables, which should include field 
146 
variables after the translation to midil. 
variables after the translation to midil. 
147 


148 

test/vrkcomp2.diderot: Add support for code like 
149 

(F1 if x else F2)@pos 
150 

This will require duplication of the continuation of the conditional 
151 

(but we should only duplicate over the liverange of the result of the 
152 

conditional. 
153 


154 

[GLK:7] Want: nontrivial field expressions & functions. 
155 

scalar fields from scalar fields F and G: 
156 

field#0(2)[] X = (sin(F) + 1.0)/2; 
157 

field#0(2)[] X = F*G; 
158 

scalar field of vector field magnitude: 
159 

image(2)[2] Vimg = load(...); 
160 

field#0(2)[] Vlen = Vimg ⊛ bspln3; 
161 

field of normalized vectors (for LIC and vector field feature extraction) 
162 

field#2(2)[2] F = ... 
163 

field#0(2)[2] V = normalize(F); 
164 

scalar field of gradient magnitude (for edge detection)) 
165 

field#2(2)[] F = Fimg ⊛ bspln3; 
166 

field#0(2)[] Gmag = ∇F; 
167 

scalar field of squared gradient magnitude (simpler to differentiate): 
168 

field#2(2)[] F = Fimg ⊛ bspln3; 
169 

field#0(2)[] Gmsq = ∇F•∇F; 
170 

There is value in having these, even if the differentiation of them is 
171 

not supported (hence the indication of "field#0" for these above) 
172 


173 

Introduce region types (syntax region(d), where d is the dimension of the 
174 

region. One useful operator would be 
175 

dom : field#k(d)[s] > region(d) 
176 

Then the inside test could be written as 
177 

pos ∈ dom(F) 
178 

We could further extend this approach to allow geometric definitions of 
179 

regions. It might also be useful to do inside tests in world space, 
180 

instead of image space. 
181 


182 
co vs contra index distinction 
co vs contra index distinction 
183 


184 
some indication of tensor symmetry 
Permit field composition: 
185 

field#2(3)[3] warp = bspln3 ⊛ warpData; 
186 

field#2(3)[] F = bspln3 ⊛ img; 
187 

field#2(3)[] Fwarp = F ◦ warp; 
188 

So Fwarp(x) = F(warp(X)). Chain rule can be used for differentation. 
189 

This will be instrumental for expressing nonrigid registration 
190 

methods (but those will require covscontra index distinction) 
191 


192 

Allow the convolution to be specified either as a single 1D kernel 
193 

(as we have it now): 
194 

field#2(3)[] F = bspln3 ⊛ img; 
195 

or, as a tensor product of kernels, one for each axis, e.g. 
196 

field#0(3)[] F = (bspln3 ⊗ bspln3 ⊗ tent) ⊛ img; 
197 

This is especially important for things like timevarying fields 
198 

and the use of scalespace in field visualization: one axis of the 
199 

must be convolved with a different kernel during probing. 
200 

What is very unclear is how, in such cases, we should notate the 
201 

gradient, when we only want to differentiate with respect to some 
202 

subset of the axes. One ambitious idea would be: 
203 

field#0(3)[] Ft = (bspln3 ⊗ bspln3 ⊗ tent) ⊛ img; // 2D timevarying field 
204 

field#0(2)[] F = lambda([x,y], Ft([x,y,42.0])) // restriction to time=42.0 
205 

vec2 grad = ∇F([x,y]); // 2D gradient 
206 


207 

Tensors of order 3 (e.g. gradients of diffusion tensor fields, or 
208 

hessians of vector fields) and order 4 (e.g. Hessians of diffusion 
209 

tensor fields). 
210 


211 

representation of tensor symmetry 
212 
(have to identify the group of index permutations that are symmetries) 
(have to identify the group of index permutations that are symmetries) 
213 


214 
dot works on all tensors 
dot works on all tensors 
215 


216 
outer works on all tensors 
outer works on all tensors 
217 


218 

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 
Einstein summation notation 
Einstein summation notation 
235 


236 
"tensor comprehension" (like list comprehension) 
"tensor comprehension" (like list comprehension) 
237 


238 

Fields coming from different sources of data: 
239 

* triangular or tetrahedral meshes over 2D or 3D domains (of the 
240 

source produced by finiteelement 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 
====================== 
====================== 
248 
BUGS ================= 
BUGS ================= 
249 
====================== 
====================== 
253 
// uncaught exception Size [size] 
// uncaught exception Size [size] 
254 
// raised at ctarget/ctarget.sml:47.1547.19 
// raised at ctarget/ctarget.sml:47.1547.19 
255 
//field#4(3)[] F = img ⊛ bspln5; 
//field#4(3)[] F = img ⊛ bspln5; 
256 


257 

test/lic2.diderot: 
258 

error in addEdge(ENTRY01D8,NULL0006) 
259 

uncaught exception Fail [Fail: setPred on NULL node NULL0006] 
260 

raised at common/phasetimer.sml:76.5076.52 
261 

raised at IL/ssafn.sml:308.7308.9 
262 

raised at IL/ssafn.sml:245.26245.70 