file write

tetrarun.py

@@ -2,17 +2,17 @@ import sys, getopt
 
 from genax import gen_angels_to_pick, angles_alap, angles_ratet
 from gpu import start_kernel
-from utils import convert, printresults, search, exact_one, exact_one_gpu, filter_gpu
+from utils import convert, printresults, search, exact_one, exact_one_gpu, filter_gpu, writetofile
 
 def main(argv):
-   outputfile = 'out.txt'
+   outputfile = None
    n = 3
    v = 3
    w = 3
    PLOT = False
 
    try:
-      opts, args = getopt.getopt(argv,"hpn:v:w:d:")
+      opts, args = getopt.getopt(argv,"hpn:v:w:d:o:")
    except getopt.GetoptError as err:
       print(err)
       print ('tetrarun.py -n <range division:int> -v <> -w <> -o <outputfile>')
@@ -41,9 +41,13 @@ def main(argv):
    Dx, Dy, Dz = angles_ratet(space)
 
    res = start_kernel(Cx, Cy, Dx, Dy, Dz, v, w)
+
+   if outputfile:
+      writetofile(outputfile, Cx, Cy, Dx, Dy, Dz, res)
+
    #printresults(res)
-   print("Exact one 3-3")
-   exact_one_gpu(res, 3-1, 3-1)
+   #print("Exact one 3-3")
+   #exact_one_gpu(res, 3-1, 3-1)
    #print("Filter 2-1")
    #filter21 = filter_gpu(res, 2-1, 1-1)
 

utils.py

@@ -2,6 +2,7 @@ import cupy as cp
 from functools import wraps
 import time
 import torch
+import numpy as np
 
 with open('filtering.cu') as f:
     code = f.read()
@@ -56,6 +57,34 @@ def printresults(egyensulyi_mtx):
         print(np.resize(parok, (int(N/2), 2)))
         print()
 
+def writetofile(filename, Cx, Cy, Dx, Dy, Dz, egyensulyi_mtx):
+    lcm = compute_lcm(Cx.size, Dx.size)
+    Cx_cpu = Cx.get()
+    Cy_cpu = Cy.get()
+    Dx_cpu = Dx.get()
+    Dy_cpu = Dy.get()
+    Dz_cpu = Dz.get()
+    mtx_cpu = egyensulyi_mtx.get()
+
+    pos = Cx.size * Dx.size
+
+    f = open(filename, "w")
+    size_C = Cx.size
+    size_D = Dx.size
+    for i in range(0, pos):
+        parok = np.empty([0], dtype=np.int8)
+        for S in range(0, 4):
+            for U in range(0, 4):
+                if mtx_cpu[i][S][U] == 1:
+                    parok = np.append(parok, S+1)
+                    parok = np.append(parok, U+1)
+        N = parok.size
+        f.write(f"{Cx_cpu[i % size_C]}, {Cy_cpu[i % size_C]}, {Dx_cpu[(i + int(i / lcm)) % size_D]}, {Dy_cpu[(i + int(i / lcm)) % size_D]}, {Dz_cpu[(i + int(i / lcm)) % size_D]}\n")
+        f.write(np.array2string(np.resize(parok, (int(N/2), 2))))
+        f.write("\n")
+
+    f.close()
+        
 def search(egyensulyi_mtx, S, U):
     for i in egyensulyi_mtx:
         if i[S][U] == 1:
@@ -82,12 +111,7 @@ def exact_one_gpu(egyensulyi_mtx, S, U):
     indexes = cp.zeros((size,), dtype=bool)
     numBlock = int((size + 256 - 1) / 256)
     exact_one_cuda((numBlock,), (256,), (egyensulyi_mtx, size, indexes, S, U))
-    t_emtx = torch.as_tensor(egyensulyi_mtx, device=torch.device('cuda'), dtype=torch.int8)
-    assert t_emtx.__cuda_array_interface__['data'][0] == egyensulyi_mtx.__cuda_array_interface__['data'][0]
-    t_ind = torch.as_tensor(indexes, device=torch.device('cuda'), dtype=torch.bool)
-    t_ind = t_ind.unsqueeze(-1).unsqueeze(-1).expand(-1, 4, 4)
-    res = torch.masked_select(t_emtx, t_ind)
-    return res
+    return egyensulyi_mtx[indexes]
 
 def filter_gpu(egyensulyi_mtx, S, U):
     size = int(egyensulyi_mtx.size / 16)