loop.br

用于GPU通用计算的编程语言BrookGPU 0.4

                             vout[1] float3 cracksB<>) {
  float2 zero2=0;
  float2 index = (i>=0&&i<65536)?i:zero2;
  STri tri=triangles[index];
  Neighbor oldneighbors=neighbors[index];
  float generous=0;
  if(i.x>=0&&i.x<65536/*&&(tri.A.w!=0||tri.B.w!=0||tri.C.w!=0)*/) {
    if ((generous||oldneighbors.AB.w==0)&&tri.A.w==0) {
      //that triangle is rastered as is, we better push out a repair triangle
      cracksA=float3(i.x,i.y,0);
      push(cracksA);
      if ((generous||oldneighbors.BC.w==0)&&tri.B.w==0) {
        //that triangle is rastered as is, we better push out a repair triangle
        cracksB=float3(i.x,i.y,1);
        push(cracksB);
      }
      if ((generous||oldneighbors.AC.w==0)&&tri.C.w==0) {
        //that triangle is rastered as is, we better push out a repair triangle
        cracksB=float3(i.x,i.y,2);
        push(cracksB);
      }
    }else
    if ((generous||oldneighbors.BC.w==0)&&tri.B.w==0) {
     //that triangle is rastered as is, we better push out a repair triangle
      cracksA=float3(i.x,i.y,1);
      push(cracksA);
      if ((generous||oldneighbors.AC.w==0)&&tri.C.w==0) {
        //that triangle is rastered as is, we better push out a repair triangle
        cracksB=float3(i.x,i.y,2);
        push(cracksB);
      }
    }else
    if ((generous||oldneighbors.AC.w==0)&&tri.C.w==0) {
      //that triangle is rastered as is, we better push out a repair triangle
      cracksA=float3(i.x,i.y,2);
      push(cracksA);
    }
  }
}

kernel SplitTri splitTriangles (STri tri<>, Neighbor oldneighbors<>, float noAdaptive<>) {
  SplitTri splittri;
  float eAAB=isNeighbor(oldneighbors.AAB);
  float eAAC=isNeighbor(oldneighbors.AAC);
  float eABB=isNeighbor(oldneighbors.ABB);
  float eBBC=isNeighbor(oldneighbors.BBC);
  float eBCC=isNeighbor(oldneighbors.BCC);
  float eACC=isNeighbor(oldneighbors.ACC);
  float3 fiveandhalf=5.5;
  float3 fourandhalf=4.5;
  float3 weights={.625,.579533905371,0.515625};
  float3 ABCneighbors = {4.0f+eAAC+eAAB,4.0f+eBBC+eABB,4.0f+eBCC+eACC};
  float3 abcn =ABCneighbors>fiveandhalf?weights.xxx:(ABCneighbors>fourandhalf?weights.yyy:weights.zzz);
  //float an = .375+(.375+sqr(.25*cos(2*3.1415926536/Aneighbors)));
  //float bn = //.375+(.375+sqr(.25*cos(2*3.1415926536/Bneighbors)));
  //float cn = //.375+(.375+sqr(.25*cos(2*3.1415926536/Cneighbors)));
  float an = abcn.x;
  float bn = abcn.y;
  float cn = abcn.z;
  float3 abcnon = (1-abcn)/ABCneighbors;
  float anon = abcnon.x;
  float bnon = abcnon.y;
  float cnon = abcnon.z;
  splittri.AC = avg4 (tri.A,tri.C);
  if(1)if(noAdaptive||tri.C.w)
    splittri.AC = addN(lerp(splittri.AC.xyz,
                            avg (tri.B.xyz,oldneighbors.AC.xyz),
                            .25),
                       1);
  splittri.AB = avg4 (tri.A,tri.B);
  if(1)if(noAdaptive||tri.A.w)
    splittri.AB = addN(lerp(splittri.AB.xyz,
                                avg(tri.C.xyz,oldneighbors.AB.xyz),
                                .25),
                           1);
  splittri.BC = avg4 (tri.B,tri.C);
  if(1)if(noAdaptive||tri.B.w)
    splittri.BC = addN(lerp(splittri.BC.xyz,
                                avg(tri.A.xyz,oldneighbors.BC.xyz),
                                .25),
                           1);
  splittri.A = tri.A;
  if(1)if(noAdaptive||(tri.A.w&&
                       tri.C.w&&
                       (oldneighbors.AC.w==4||oldneighbors.AC.w==8)&&
                       (oldneighbors.AB.w==2||oldneighbors.AB.w==8)&&
                       oldneighbors.AAB.w&&
                       oldneighbors.AAC.w))       
     splittri.A.xyz = tri.A.xyz*an
        +anon*add6(oldneighbors.AB.xyz,
                   tri.B.xyz,
                   tri.C.xyz,
                   oldneighbors.AC.xyz,
                   oldneighbors.AAC.xyz,
                   oldneighbors.AAB.xyz);
  splittri.B = tri.B;
  if(1)if(noAdaptive||(tri.A.w&&
                       tri.B.w&&
                       (oldneighbors.BC.w==2||oldneighbors.BC.w==8)&&
                       (oldneighbors.AB.w==4||oldneighbors.AB.w==8)&&
                       oldneighbors.ABB.w&&
                       oldneighbors.BBC.w))        
    splittri.B.xyz= tri.B.xyz*bn
       +bnon*add6(oldneighbors.ABB.xyz,
                  oldneighbors.BBC.xyz,
                  oldneighbors.BC.xyz,
                  tri.C.xyz,
                  tri.A.xyz,
                  oldneighbors.AB.xyz);

  splittri.C = tri.C;
  if(1)if(noAdaptive||(tri.C.w&&
                       tri.B.w&&
                       (oldneighbors.BC.w==4||oldneighbors.BC.w==8)&&
                       (oldneighbors.AC.w==2||oldneighbors.AC.w==8)&&
                       oldneighbors.ACC.w&&
                       oldneighbors.BCC.w))
     splittri.C.xyz = tri.C.xyz*cn
        +cnon*add6(tri.B.xyz,
                   oldneighbors.BC.xyz,
                   oldneighbors.BCC.xyz,
                   oldneighbors.ACC.xyz,
                   oldneighbors.AC.xyz,
                   tri.A.xyz);
  return splittri;
}


//this function takes as input an index that tells which triangle must be split
//the output is 6 new vertices that make up the 4 new triangles and 12 neighbors 
// (stored in 9 float4s)
kernel void splitTrianglesAdaptive (STri triangles[][], 
                                    Neighbor neighbors[][],
                                    float2 i<>,
                                    out SplitTri splittri<>) {
  float2 zero2=0;
  float2 index = (i>=0&&i<65536)?i:zero2;
  STri tri=triangles[index];
  Neighbor oldneighbors=neighbors[index];
  splittri=splitTriangles(tri,oldneighbors,0);
}

kernel void splitTrianglesNoAdaptive (STri triangles<>, 
                                      Neighbor neighbors<>,
                                      out SplitTri splittri<>) {
  splittri=splitTriangles(triangles,neighbors,1);
}

/*
void __printf_cpu_inner(float f) {
  printf("%.2f ",f);
  }*/
void __printnew_cpu_inner() {
  printf ("\n");
}

kernel float whichMod4 (float2 index, float stretchX) {
  float num;
  num= (stretchX*index.x+((float)(stretchX!=1))*index.y);
  num = fmod(num,4.0f);
  return num>3.5?0:num;
}
//moves the 6 split values into 4 individual triangles (with a result of 12 vertices)
//moves the 12 split neighbors into 4 individual neighbor lists
//uses fmod with x or y to determine which direction should matter wrt fmod.
kernel void linearReorgSplitTriangles (SplitTri splittri<>,
                                       Neighbor newnei<>,
                                       out STri tri<>,
                                       out Neighbor neighbor<>,
                                       float stretchX) {
  //  float sindex = stretchX?(indexof tri).x:(indexof tri).y;
  //  float unclampedwhich = round(fmod(sindex,4));
  //  float which = whichMod4(unclampedwhich>3.5?0:unclampedwhich;
  float which = whichMod4((indexof tri).xy,stretchX);
  float4 ac = splittri.AC;
  float4 ab = splittri.AB;
  float4 bc = splittri.BC;
  float4 a = splittri.A;
  float4 b = splittri.B;
  float4 c = splittri.C;
  float4 BC_C= {newnei.BC.w,
                newnei.ACC.w,
                newnei.BCC.w,1};
  float4 AC_A = {newnei.AC.w,
                 newnei.AAB.w,
                 newnei.AAC.w,1};
  float4 AB_B = {newnei.AB.w,
                 newnei.BBC.w,
                 newnei.ABB.w,1};
  
  if (which<.5) {
    tri.A=identity(a,1);
    tri.B=identity(ab,1);
    tri.C=identity(ac,1);

    neighbor.BCC = c;
    neighbor.BC = bc;
    neighbor.BBC= b;
    neighbor.ABB= AB_B;
    neighbor.AB = newnei.AB;
    neighbor.AAB =  newnei.AAB;
    neighbor.AAC =  newnei.AAC;
    neighbor.AC = AC_A;
    neighbor.ACC = newnei.AC;
  }else if (which <1.5) {
    tri.A= identity(ab,1);
    tri.B=identity(b,1);
    tri.C=identity(bc,1);

    neighbor.BCC = BC_C;
    neighbor.BC = newnei.BC;//b0rken
    neighbor.BBC=newnei.BBC;
    neighbor.ABB=newnei.ABB;
    neighbor.AB = AB_B;
    neighbor.AAB = newnei.AB;
    neighbor.AAC = a;
    neighbor.AC = ac;
    neighbor.ACC = c;
  }else if (which<2.5) {
    
    tri.A=identity(ac,1);
    tri.B=identity(bc,1);
    tri.C=identity(c,1);

    neighbor.BCC=newnei.BCC;
    neighbor.BC=BC_C;
    neighbor.BBC = newnei.BC;//b0rken!?
    neighbor.ABB = b;
    neighbor.AB = ab;
    neighbor.AAB = a;
    neighbor.AAC = AC_A;
    neighbor.AC = newnei.AC;
    neighbor.ACC = newnei.ACC;
  }else {
    tri.A=identity(bc,1);
    tri.B=identity(ac,1);
    tri.C=identity(ab,1);

    neighbor.BCC = newnei.AB;
    neighbor.BC = a;
    neighbor.BBC= AC_A;
    neighbor.ABB= newnei.AC;
    neighbor.AB = c;
    neighbor.AAB = BC_C;
    neighbor.AAC = newnei.BC;//b0rken
    neighbor.AC = b;
    neighbor.ACC = AB_B;
  }
}
kernel void adaptiveReorgSplitTriangles (SplitTri splittri<>,
                                       Neighbor newnei<>,
                                       out STri tri<>,
                                       out Neighbor neighbor<>,
                                       float epsilon,
                                       float stretchX) {
  linearReorgSplitTriangles(splittri,newnei,tri,neighbor,stretchX);
  smallEnough(tri,neighbor,tri,neighbor,epsilon);  
}


struct VertexArray4{
  float4 * v;
  unsigned int size;
  unsigned int alloc;
};
void initVertexArray4(struct VertexArray4 *v) {
  v->v=0;v->size=0;v->alloc=0;
}
void destroyVertexArray4(struct VertexArray4 *v) {
  if (v->v) free(v->v);
}
void expandVertexArray4(struct VertexArray4 * v,
                       unsigned int more) {
  v->size+=more;
  while(v->size>v->alloc) {
    if (v->alloc==0)
      v->alloc=16;
    v->alloc*=2;
    v->v = (float4*)realloc(v->v,v->alloc*sizeof(float4));
  }
}                   
void pushBackVertexArray4(struct VertexArray4 * v,
                         float4 f) {
  if (++v->size>v->alloc) {
    if (v->alloc==0)
      v->alloc=16;
    v->alloc*=2;
    v->v = (float4*)realloc(v->v,v->alloc*sizeof(float4));
  }
  v->v[v->size-1]=f;
}                   

//simple growable array, similar to std::vector (I hate C parsers)
struct VertexArray{
  float3 * v;
  unsigned int size;
  unsigned int alloc;
};
void initVertexArray(struct VertexArray *v) {
  v->v=0;v->size=0;v->alloc=0;
}
void destroyVertexArray(struct VertexArray *v) {
  if (v->v) free(v->v);
}
void expandVertexArray(struct VertexArray * v,
                       unsigned int more) {
  v->size+=more;
  while(v->size>v->alloc) {
    if (v->alloc==0)
      v->alloc=16;
    v->alloc*=2;
    v->v = (float3*)realloc(v->v,v->alloc*sizeof(float3));
  }
}                   
void pushBackVertexArray(struct VertexArray * v,
                         float3 f) {
  if (++v->size>v->alloc) {
    if (v->alloc==0)
      v->alloc=16;
    v->alloc*=2;
    v->v = (float3*)realloc(v->v,v->alloc*sizeof(float3));
  }
  v->v[v->size-1]=f;
}                   
static int toi(float f) {return (int)f;}


void subdivide (Neighbor (*neighbors)<>,
                STri (*triangles)<>,
                struct VertexArray *output,
                struct VertexArray4 *output4,
                int counter,
                char out4) {
  unsigned int streamY=toi(streamSize(*triangles).y);
  unsigned int streamX=toi(streamSize(*triangles).x);
  float stretchX = streamY*4>maxTexDim?(streamY*2>maxTexDim?1.0f:2.0f):0.0f;      
  int xfactor=stretchX==2.0f?2:(stretchX==1.0f?4:1);
  int yfactor=stretchX==2.0f?2:(stretchX==1.0f?1:4);
  subdivisiondepth++;
  if (counter>=counterMax) {
    if (out4) {
      int sizey = output4->size;
      expandVertexArray4(output4,
                         3*toi(streamSize(*triangles).y)*toi(streamSize(*triangles).x));
      streamWrite(*triangles,output4->v+sizey);
    }else {
      int sizey = output->size;
      float3 outputTri<toi(streamSize(*triangles).y),
        toi(streamSize(*triangles).x*3)>;
      copyFinalTriangles(*triangles,outputTri);
      expandVertexArray(output,
                      3*toi(streamSize(*triangles).y)*toi(streamSize(*triangles).x));
      streamWrite(outputTri,output->v+sizey);
    }
  }else {
    //allocate shared neighbor structs that hold the extra generated floats
    Neighbor sharedNeighbors<streamY,streamX>;
    SplitTri sharedTriangles <streamY,streamX>;
    //these new neighbor structs are 4x as big as the old ones
    //input is thus stretched through them based on the fmod

    Neighbor newNeighbors<(streamY*yfactor),
      (streamX*xfactor)>;
    STri newTriangles<(streamY*yfactor),
      (streamX*xfactor)>;
    fprintf (stderr,"stretching %f to %d %d\n",stretchX,streamX*xfactor,streamY*yfactor);

    //computes the 12 shared neighbor values
    computeNeighborsNoAdaptive(*triangles,*neighbors,sharedNeighbors);
    //computes the 6 new triangle vertex values
    splitTrianglesNoAdaptive(*triangles,*neighbors,sharedTriangles);
    // combines the 12 neighbors and 6 vertices to produce 4 individual 
    // sets of triangles
    // with the new triangles getting stretched in the x or y direction depending
    // on stretchX
    linearReorgSplitTriangles(sharedTriangles,
                              sharedNeighbors,