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

SCM Repository

[diderot] Diff of /branches/vis12-cl/TODO
ViewVC logotype

Diff of /branches/vis12-cl/TODO

Parent Directory Parent Directory | Revision Log Revision Log | View Patch Patch

trunk/TODO revision 1295, Thu Jun 9 06:42:54 2011 UTC branches/vis12-cl/TODO revision 2683, Mon Aug 25 18:56:17 2014 UTC
# Line 1  Line 1 
1  NOTE: GLK's approximate ranking of 8 most important tagged with  This to do for the Vis 2012 submission
2  [GLK:1], [GLK:2], ...  
3    3rd order tensors: 2x2x2 and 3x3x3 (done)
5    dynamic-length sequences (done, except for loading and smart memory management)
7    curl 2D and 3D
9    support iteration over image indices
11    loading sequences
13    ====================
14    Plan for loading sequences
16    check proxy files when converting from AST to Simple representation; use default representations
17    when no proxy is specified.
19    get rid of lift/eval phases (although we might want to address the polyvariance of fields
20    at that stage)
22    Need to update nrrd info to allow for default representations where various properties
23    are not specified (origin and sizes).
25    ====================
27    We should just generate a compile-time warning about unused inputs.
29    Add a "snapshot" keyword to annotate strand state variables.
 SHORT TERM ============= (*needed* for streamlines & tractography)  
 Remove CL from compiler  
 [GLK:3] Add sequence types (needed for evals & evecs)  
         types: ty '{' INT '}'  
         value construction: '{' e1 ',' … ',' en '}'  
         indexing: e '{' e '}'  
 [GLK:4] evals & evecs for symmetric tensor[2,2] and  
 tensor[3,3] (requires sequences)  
 ability to emit/track/record variables into dynamically re-sized  
 runtime buffer  
 tensor fields: convolution on general tensor images  
 SHORT-ISH TERM ========= (to make using Diderot less annoying to  
 ========================  program in, and slow to execute)  
 value-numbering optimization [DONE]  
 Allow ".ddro" file extensions in addition to ".diderot"  
 Be able to output values of type tensor[2,2] and tensor[3,3];  
 (currently only scalars & vectors).  Want to add some regression tests  
 based on this and currently can't  
 [GLK:1] Add a clamp function, which takes three arguments; either  
 three scalars:  
   clamp(lo, hi, x)  = max(lo, min(hi, x))  
 or three vectors of the same size:  
   clamp(lo, hi, [x,y])  = [max(lo[0], min(hi[0], x)),  
                            max(lo[1], min(hi[1], y))]  
 This would be useful in many current Diderot programs.  
 One question: clamp(x, lo, hi) is the argument order used in OpenCL  
 and other places, but clamp(lo, hi, x) is much more consistent with  
 lerp(lo, hi, x), hence GLK's preference  
 [GLK:2] Proper handling of stabilize method  
 allow "*" to represent "modulate": per-component multiplication of  
 vectors, and vectors only (not tensors of order 2 or higher).  Once  
 sequences are implemented this should be removed: the operation is not  
 invariant WRT basis so it is not a legit vector computation.  
 implicit type promotion of integers to reals where reals are  
 required (e.g. not exponentiation "^")  
 [GLK:5] Save Diderot output to nrrd, instead of "mip.txt"  
   For grid of strands, save to similarly-shaped array  
   For list of strands, save to long 1-D (or 2-D for non-scalar output) list  
   For ragged things (like tractography output), will need to save both  
     complete list of values, as well as list of start indices and lengths  
     to index into complete list  
 [GLK:6] Use of Teem's "hest" command-line parser for getting  
 any "input" variables that are not defined in the source file.  
 [GLK:7] ability to declare a field so that probe positions are  
 *always* "inside"; with various ways of mapping the known image values  
 to non-existant index locations.  One possible syntax emphasizes that  
 there is a index mapping function that logically precedes convolution:  
   F = bspln3 ⊛ (img ◦ clamp)  
   F = bspln3 ⊛ (img ◦ repeat)  
   F = bspln3 ⊛ (img ◦ mirror)  
 where "◦" or "∘" is used to indicate function composition  
 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.  
 Use ∇⊗ etc. syntax  
     syntax [DONE]  
     IL and codegen  
 Add type aliases for color types  
     rgb = real{3}  
     rgba = real{4}  
 MEDIUM TERM ================== (*needed* for particles)  
 run-time birth of strands  
 "initially" supports lists  
 "initially" supports lists of positions output from different  
 initalization Diderot program (or output from the same program;  
 e.g. using output of iso2d.diderot for one isovalue to seed the input  
 to another invocation of the same program)  
 Communication between strands: they have to be able to learn each  
 other's state (at the previous iteration).  Early version of this can  
 have the network of neighbors be completely static (for running one  
 strand/pixel image computations).  Later version with strands moving  
 through the domain will require some spatial data structure to  
 optimize discovery of neighbors.  
 MEDIUM-ISH TERM ============ (to make Diderot more useful/effective)  
 Python/ctypes interface to run-time  
 support for Python interop and GUI  
 Allow integer exponentiation ("^2") to apply to square matrices,  
 to represent repeated matrix multiplication  
 Alow 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.  
 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  
 expand trace in mid to low translation  
 extend norm (|exp|) to all tensor types [DONE for vectors and matrices]  
 determinant ("det") for tensor[3,3]  
 add ":" for tensor dot product (contracts out two indices  
 instead of one like •), valid for all pairs of tensors with  
 at least two indices  
 documents need for better compiler error messages when output variables  
 are not initialized; the current messages are very cryptic  
 want: warnings when "D" (reserved for differentiation) is declared as  
 a variable name (get confusing error messages now)  
 LONG TERM ==================== (make Diderot more interesting/attractive from  
 ==============================  a research standpoint)  
 IL support for higher-order tensor values (matrices, etc).  
     tensor construction [DONE]  
     tensor indexing [DONE]  
     tensor slicing  
     verify that hessians work correctly [DONE]  
 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.  
 test/vr-kcomp2.diderot: Add support for code like  
         (F1 if x else F2)@pos  
 This will require duplication of the continuation of the conditional  
 (but we should only duplicate over the live-range of the result of the  
 [GLK:8] Want: non-trivial field expressions & functions.  
 scalar fields from scalar fields F and G:  
   field#0(2)[] X = (sin(F) + 1.0)/2;  
   field#0(2)[] X = F*G;  
 scalar field of vector field magnitude:  
   image(2)[2] Vimg = load(...);  
   field#0(2)[] Vlen = |Vimg ⊛ bspln3|;  
 field of normalized vectors (for LIC and vector field feature extraction)  
   field#2(2)[2] F = ...  
   field#0(2)[2] V = normalize(F);  
 scalar field of gradient magnitude (for edge detection))  
   field#2(2)[] F = Fimg ⊛ bspln3;  
   field#0(2)[] Gmag = |∇F|;  
 scalar field of squared gradient magnitude (simpler to differentiate):  
   field#2(2)[] F = Fimg ⊛ bspln3;  
   field#0(2)[] Gmsq = ∇F•∇F;  
 There is value in having these, even if the differentiation of them is  
 not supported (hence the indication of "field#0" for these above)  
 Introduce region types (syntax region(d), where d is the dimension of the  
 region.  One useful operator would be  
         dom : field#k(d)[s] -> region(d)  
 Then the inside test could be written as  
         pos ∈ dom(F)  
 We could further extend this approach to allow geometric definitions of  
 regions.  It might also be useful to do inside tests in world space,  
 instead of image space.  
 co- vs contra- index distinction  
 Permit field composition:  
   field#2(3)[3] warp = bspln3 ⊛ warpData;  
   field#2(3)[] F = bspln3 ⊛ img;  
   field#2(3)[] Fwarp = F ◦ warp;  
 So Fwarp(x) = F(warp(X)).  Chain rule can be used for differentation.  
 This will be instrumental for expressing non-rigid registration  
 methods (but those will require co-vs-contra index distinction)  
 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 time-varying fields  
 and the use of scale-space in field visualization: one axis of the  
 must be convolved with a different kernel during probing.  
 What is very unclear is how, in such cases, we should notate the  
 gradient, when we only want to differentiate with respect to some  
 subset of the axes.  One ambitious idea would be:  
   field#0(3)[] Ft = (bspln3 ⊗ bspln3 ⊗ tent) ⊛ img; // 2D time-varying field  
   field#0(2)[] F = lambda([x,y], Ft([x,y,42.0]))     // restriction to time=42.0  
   vec2 grad = ∇F([x,y]);                             // 2D gradient  
 Tensors of order 3 (e.g. gradients of diffusion tensor fields, or  
 hessians of vector fields) and order 4 (e.g. Hessians of diffusion  
 tensor fields).  
 representation of tensor symmetry  
 (have to identify the group of index permutations that are symmetries)  
 dot works on all tensors  
 outer works on all tensors  
 Help for debugging Diderot programs: need to be able to uniquely  
 identify strands, and for particular strands that are known to behave  
 badly, do something like printf or other logging of their computations  
 and updates.  
 Permit writing dimensionally general code: Have some statement of the  
 dimension of the world "W" (or have it be learned from one particular  
 field of interest), and then able to write "vec" instead of  
 "vec2/vec3", and perhaps "tensor[W,W]" instead of  
 Traits: all things things that have boilerplate code (especially  
 volume rendering) should be expressed in terms of the unique  
 computational core.  Different kinds of streamline/tractography  
 computation will be another example, as well as particle systems.  
 Einstein summation notation  
 "tensor comprehension" (like list comprehension)  
 Fields coming from different sources of data:  
 * triangular or tetrahedral meshes over 2D or 3D domains (of the  
   source produced by finite-element codes; these will come with their  
   own specialized kinds of reconstruction kernels, called "basis  
   functions" in this context)  
 * Large point clouds, with some radial basis function around each point,  
   which will be tuned by parameters of the point (at least one parameter  
   giving some notion of radius)  
 BUGS =================  
 // HEY (bug) bspln5 leads to problems ...  
 //  uncaught exception Size [size]  
 //    raised at c-target/c-target.sml:47.15-47.19  
 //field#4(3)[] F = img ⊛ bspln5;  

Removed from v.1295  
changed lines
  Added in v.2683

ViewVC Help
Powered by ViewVC 1.0.0