halide - Bilateral Grid Generator class using Enhanced Generator -

i trying re-implement bilateral grid example using enhanced generator class (e.g. using schedule() , generate(). i've got error when trying compile code.

g++ -std=c++11 -i ../../include/ -i ../../tools/ -i ../../apps/support/ -g -  fno-rtti bilateral_grid_generator.cpp ../../lib/libhalide.a ../../tools/gengen.cpp -o bin/bilateral_grid_exec  -ldl -lpthread -lz bin/bilateral_grid_exec -o ./bin  target=host  generator bilateral_grid has base_path ./bin/bilateral_grid internal error @ /home/xxx/projects/halide/src/generator.cpp:966 triggered user code @ /usr/include/c++/4.8/functional:2057: condition failed: generator make: *** [bin/bilateral_grid.a] aborted (core dumped) 

it seems didn't put definition of rdom , generatorparam in correct place. since r.x , r.y used in both schedule() , generate(), think should put class member. should done fix this?

here code wrote.

class bilateralgrid : public halide::generator<bilateralgrid> { public: generatorparam<int>   s_sigma{"s_sigma", 8};  //imageparam            input{float(32), 2, "input"}; //param<float>          r_sigma{"r_sigma"};  input<buffer<float>>  input{"input", 2}; input<float>          r_sigma{"r_sigma"};  output<buffer<float>> output{"output", 2};  // algorithm description void generate() {     //int s_sigma = 8;     // add boundary condition     clamped(x,y) = boundaryconditions::repeat_edge(input)(x,y);      // construct bilateral grid     expr val = clamped(x * s_sigma + r.x - s_sigma/2, y * s_sigma + r.y - s_sigma/2);     val = clamp(val, 0.0f, 1.0f);      expr zi = cast<int>(val * (1.0f/r_sigma) + 0.5f);      // histogram     histogram(x, y, z, c) = 0.0f;     histogram(x, y, zi, c) += select(c == 0, val, 1.0f);      // blur grid using five-tap filter     blurz(x, y, z, c) = (histogram(x, y, z-2, c) +                          histogram(x, y, z-1, c)*4 +                          histogram(x, y, z  , c)*6 +                          histogram(x, y, z+1, c)*4 +                          histogram(x, y, z+2, c));     blurx(x, y, z, c) = (blurz(x-2, y, z, c) +                          blurz(x-1, y, z, c)*4 +                          blurz(x  , y, z, c)*6 +                          blurz(x+1, y, z, c)*4 +                          blurz(x+2, y, z, c));     blury(x, y, z, c) = (blurx(x, y-2, z, c) +                          blurx(x, y-1, z, c)*4 +                          blurx(x, y  , z, c)*6 +                          blurx(x, y+1, z, c)*4 +                          blurx(x, y+2, z, c));      // take trilinear samples compute output     val     = clamp(input(x, y), 0.0f, 1.0f);     expr zv = val * (1.0f/r_sigma);     zi      = cast<int>(zv);     expr zf = zv - zi;     expr xf = cast<float>(x % s_sigma) / s_sigma;     expr yf = cast<float>(y % s_sigma) / s_sigma;     expr xi = x/s_sigma;     expr yi = y/s_sigma;      interpolated(x, y, c) =         lerp(lerp(lerp(blury(xi, yi, zi, c), blury(xi+1, yi, zi, c), xf),                   lerp(blury(xi, yi+1, zi, c), blury(xi+1, yi+1, zi, c), xf), yf),              lerp(lerp(blury(xi, yi, zi+1, c), blury(xi+1, yi, zi+1, c), xf),                   lerp(blury(xi, yi+1, zi+1, c), blury(xi+1, yi+1, zi+1, c), xf), yf), zf);      // normalize , return output.     bilateral_grid(x, y) = interpolated(x, y, 0)/interpolated(x, y, 1);     output(x,y)          = bilateral_grid(x,y);  }  // scheduling void schedule() {      // int s_sigma = 8;     if (get_target().has_gpu_feature()) {         // gpu schedule         var xi{"xi"}, yi{"yi"}, zi{"zi"};          // schedule blurz in 8x8 tiles. tile in         // grid-space, means represents         // 64x64 pixels in input (if s_sigma 8).         blurz.compute_root().reorder(c, z, x, y).gpu_tile(x, y, xi, yi, 8, 8);          // schedule histogram happen per-tile of blurz,         // intermediate results in shared memory. means histogram         // , blurz makes three-stage kernel:         // 1) 0 out 8x8 set of histograms         // 2) compute histogram iterating on lots of input image         // 3) blur set of histograms in z         histogram.reorder(c, z, x, y).compute_at(blurz, x).gpu_threads(x, y);         histogram.update().reorder(c, r.x, r.y, x, y).gpu_threads(x, y).unroll(c);          // alternative schedule histogram doesn't use shared memory:         // histogram.compute_root().reorder(c, z, x, y).gpu_tile(x, y, xi, yi, 8, 8);         // histogram.update().reorder(c, r.x, r.y, x, y).gpu_tile(x, y, xi, yi, 8, 8).unroll(c);          // schedule remaining blurs , sampling @ end similarly.         blurx.compute_root().gpu_tile(x, y, z, xi, yi, zi, 8, 8, 1);         blury.compute_root().gpu_tile(x, y, z, xi, yi, zi, 8, 8, 1);         bilateral_grid.compute_root().gpu_tile(x, y, xi, yi, s_sigma, s_sigma);     } else {         // cpu schedule.         blurz.compute_root().reorder(c, z, x, y).parallel(y).vectorize(x, 8).unroll(c);         histogram.compute_at(blurz, y);         histogram.update().reorder(c, r.x, r.y, x, y).unroll(c);         blurx.compute_root().reorder(c, x, y, z).parallel(z).vectorize(x, 8).unroll(c);         blury.compute_root().reorder(c, x, y, z).parallel(z).vectorize(x, 8).unroll(c);         bilateral_grid.compute_root().parallel(y).vectorize(x, 8);     } }  func clamped{"clamped"}, histogram{"histogram"}; func bilateral_grid{"bilateral_grid"}; func blurx{"blurx"}, blury{"blury"}, blurz{"blurz"}, interpolated{"interpolated"}; var x{"x"}, y{"y"}, z{"z"}, c{"c"}; rdom r{0, s_sigma, 0, s_sigma};  };  //halide::registergenerator<bilateralgrid> register_me{"bilateral_grid"}; halide_register_generator(bilateralgrid, "bilateral_grid");  }  // namespace 

the error here subtle, , current assertion failure message regrettably unhelpful.

the problem here code using generatorparam (s_sigma) initialize member-variable-rdom (r), generatorparam may not have final value set @ point. speaking, accessing generatorparam (or scheduleparam) before generate() method called produce such assert.

why this? let's @ way generators created , initialized in typical build system:

1. gengen.cpp creates instance of generator's c++ class; naturally, executes c++ constructor, c++ constructors member variables, in order of declaration.
2. gengen.cpp uses arguments provided on command line override default values of generatorparams. example, if had invoked generator bin/bilateral_grid_exec -o ./bin target=host s_sigma=7, default value (8) stored in s_sigma replaced 7.
3. gengen.cpp calls generate(), schedule(), compiles result .o (or .a, etc).

so why seeing assert? what's happening in code in step 1 above, ctor r being run in step 1... arguments ctor r read current value s_sigma, has default value (8), not value specified build file. if allowed read happen without asserting, inconsistent values s_sigma in different parts of generator.

you can fix deferring initialization of rdom generate() method:

class bilateralgrid : public halide::generator<bilateralgrid> { public: generatorparam<int> s_sigma{"s_sigma", 8}; ... void generate() { r = rdom(0, s_sigma, 0, s_sigma); ... } ... private: rdom r; };

(obviously, assertion failure needs more helpful error message; i'll modify code so.)