Iterate on multi-gpu design

Another iteration on the multi-gpu prototype and testing script.  Still
seems to be some problem running clfft on multiple threads.  Perhaps the
contexts are not set up properly in the low level code.  Going to push
this though and retry on other machines.

python/clij2fft/richardson_lucy_dask_multi_gpu.py

@@ -3,6 +3,7 @@
 import numpy as np
 import pyopencl as cl
 from clij2fft.pad import get_next_smooth
+from clij2fft.libs import getlib
 
 bytes_per_gb = 1024 * 1024 * 1024
 
@@ -148,9 +149,13 @@ def richardson_lucy_dask(img, psf, numiterations, regularizationfactor, non_circ
     from multiprocessing import Pool, current_process, Queue
     queue = Queue()
 
-    for i in range(1):
+    num_gpus = 1 
+    for i in range(num_gpus):
         queue.put(i)
 
+
+    lib = getlib()
+
     #from dask.distributed import get_worker
     if non_circulant:
 
@@ -161,17 +166,18 @@ def richardson_lucy_nc_dask_task(img, psf, numiterations, regularizationfactor=0
                 gpu_num=queue.get()
                 print('start rlnc')
                 print('gpu num is', gpu_num)
-                print('block id is', block_id)
-                print('block info is', block_info)
-                print('thread id is', thread_id)
+                #print('block id is', block_id)
+                #print('block info is', block_info)
+                #print('thread id is', thread_id)
                 #print('worker is ', get_worker())
-                if block_id is None:
-                    print('returning block id is None')
-                    return None
-                result=richardson_lucy_nc(img, psf, numiterations, regularizationfactor=regularizationfactor, lib=lib)
+                #if block_id is None:
+                #    print('returning block id is None')
+                #    return None
+                result=richardson_lucy_nc(img, psf, numiterations, regularizationfactor=regularizationfactor, lib=lib, platform = 0, device = gpu_num)
                 print('end rlnc')
                 return result
-            except:
+            except Exception as e:
+                print("EXCEPTION",e)
                 pass
             finally:
                 print('putting gpu num back', gpu_num)
@@ -186,9 +192,16 @@ def richardson_lucy_dask_task(img, psf, numiterations, regularizationfactor=0, l
             return img#richardson_lucy(img, psf, numiterations, regularizationfactor=regularizationfactor, lib=lib)
         rl_func = richardson_lucy_dask_task
 
+    import time
 
-    out = dimg.map_overlap(rl_func, depth={0:0, 1:overlap, 2:overlap}, dtype=np.float32, psf=psf, numiterations=numiterations, regularizationfactor=regularizationfactor)
-    return out.compute(num_workers=4)
+    start_time = time.time()
+    out = dimg.map_overlap(rl_func, depth={0:0, 1:overlap, 2:overlap}, dtype=np.float32, psf=psf, numiterations=numiterations, regularizationfactor=regularizationfactor, lib=lib)
+    out_img = out.compute(num_workers=num_gpus)
+    end_time = time.time()
+    execution_time = end_time - start_time
+    print(f"Execution time of rl dask multi gpu: {execution_time} seconds")
+
+    return out_img 
 
 
 

python/clij2fft/test_libs.py

@@ -8,8 +8,8 @@
 img= np.ones((256, 256, 128), dtype=np.float32)
 psf = np.ones((128, 128, 64), dtype=np.float32)
 
-result = richardson_lucy(img, psf, 100, 0, platform=0, device=0)
-result = richardson_lucy(img, psf, 100, 0, platform=1, device=0)
+result = richardson_lucy(img, psf, 100, 0, platform=0, device=3)
+#result = richardson_lucy(img, psf, 100, 0, platform=1, device=0)
 
 print()
 print(result.shape, result.mean())

python/clij2fft/test_richardson_lucy_dask.py

@@ -3,12 +3,16 @@
 import numpy as np
 from matplotlib import pyplot as plt
 
+
 img_name=r'D:\\images/images/Bars-G10-P15-stack-cropped.tif'
 psf_name=r'D:\\images/images/PSF-Bars-stack-cropped.tif'
 
 img_name=r'/home/bnorthan/images/deconvolution/Bars-G10-P15-stack.tif'
 psf_name=r'/home/bnorthan/images/deconvolution/PSF-Bars-stack.tif'
 
+img_name = r'C:\Users\Administrator\data\Bars-G10-P15-stack.tif'
+psf_name = r'C:\Users\Administrator\data\PSF-Bars-stack.tif'
+
 img=imread(img_name)
 print('image shape is',img.shape)
 
@@ -21,7 +25,7 @@
 print('image shape is',img.shape)
 psf=imread(psf_name)
 
-decon=richardson_lucy_dask(img, psf, 10, 0.0001, mem_to_use=mem_to_use)
+decon=richardson_lucy_dask(img, psf, 50, 0.0001, mem_to_use=mem_to_use)
 
 fig, ax = plt.subplots(1,2)
 ax[0].imshow(img.max(axis=0))