checkpoint

Signed-off by: David Rowenhorst <[email protected]>
drowenhorst-nrl · May 12, 2024 · 6cb5474 · 6cb5474
1 parent 4813839
commit 6cb5474
Show file tree

Hide file tree

Showing 2 changed files with 16 additions and 16 deletions.
diff --git a/pyebsdindex/opencl/clnlpar.cl b/pyebsdindex/opencl/clnlpar.cl
@@ -190,11 +190,13 @@ __kernel void normd(
     long i, j;
     long indx_j, indx_ij, count; 
 
-    //long nnn = (2*nn+1) * (2*nn+1);
+    long nnn = (2*sr+1) * (2*sr+1);
 
     float sigma_xy = sigma[indx_xy];  
     sigma_xy *= sigma_xy;
+    //printf("%f", sigma_xy);
     float sigma_ij, nn, dd;  
+
     count = 0;
     for(j=y-nn; j<=y+nn; ++j){
 
@@ -203,9 +205,9 @@ __kernel void normd(
           indx_j = ncol * indx_j;
 
           for(i=x-nn; i<=x+nn; ++i){  
-             dd = d[count];
-             nn = n[count];    
-             if (nn > 0){            
+             dd = d[count+nnn*indx_xy];
+             nn = n[count+nnn*indx_xy];    
+             if (nn > 1e-3){            
                indx_ij =  (i >= 0) ? (i): abs(i);
                indx_ij =  (indx_ij < ncol) ? (indx_ij): ncol - (indx_ij -ncol +1);
                indx_ij =  (indx_ij + indx_j);
@@ -214,11 +216,11 @@ __kernel void normd(
 
                sigma_ij = sigma_ij + sigma_xy;
                dd -= nn*sigma_ij;
-               dd /= sigma_ij * sqrt(2.0*nn); 
+               dd /= (sigma_ij * sqrt(2.0*nn)); 
                //printf("%f\n", dd) ;
-               d[count] = dd; 
-               count += 1;       
+               d[count+nnn*indx_xy] = dd;         
              }
+             count += 1;  
            }
 
      }      

diff --git a/pyebsdindex/opencl/nlpar_cl.py b/pyebsdindex/opencl/nlpar_cl.py
@@ -73,10 +73,10 @@ def d2normcl(d2, n2, sigmapad):
 
     sigma, d2, n2 = self.calcsigma_cl(nn=1, saturation_protect=saturation_protect, automask=automask, normalize_d=True)
 
-    sigmapad = np.pad(sigma, 1, mode='reflect')
-    d2normcl(d2, n2, sigmapad)
+    #sigmapad = np.pad(sigma, 1, mode='reflect')
+    #d2normcl(d2, n2, sigmapad)
 
-    #print(d2.min(), d2.max(), d2.mean())
+    print(d2.min(), d2.max(), d2.mean())
 
     lamopt_values_chnk = []
     for tw in target_weights:
@@ -226,12 +226,10 @@ def calcsigma_cl(self,nn=1,saturation_protect=True,automask=True, normalize_d=Fa
                                np.int64(nn), np.int64(npatsteps), np.int64(npat_point),
                                np.float32(mxval) )
 
-
-        #cl.enqueue_barrier(queue)
-        # prg.normd(queue, (np.uint32(ncolchunk), np.uint32(nrowchunk)), None,
-        #               sigmachunk_gpu,
-        #               count_local, dist_local,
-        #               np.int64(nn))
+        prg.normd(queue, (np.uint32(ncolchunk), np.uint32(nrowchunk)), None,
+                        sigmachunk_gpu,
+                        count_local, dist_local,
+                        np.int64(nn))
         queue.flush()
 
         cl.enqueue_copy(queue, distchunk, dist_local, is_blocking=False)