X-Git-Url: https://www.fleuret.org/cgi-bin/gitweb/gitweb.cgi?a=blobdiff_plain;f=speed.py;h=e03b3b74febe038e3e7382fb155909e197c71cc1;hb=6f7f454565cabfeda22c32103cc0ca41364de14a;hp=f682a169076aca29f3f7b4b695bb55fc179b305d;hpb=31b3e7141d2acb0cb1ff298d2b7fca3911889f25;p=pytorch.git

diff --git a/speed.py b/speed.py
index f682a16..e03b3b7 100755
--- a/speed.py
+++ b/speed.py
@@ -1,28 +1,38 @@
 #!/usr/bin/env python
 
+# Any copyright is dedicated to the Public Domain.
+# https://creativecommons.org/publicdomain/zero/1.0/
+
+# Written by Francois Fleuret <francois@fleuret.org>
+
 import time, torch
 
 if torch.cuda.is_available():
     device = torch.device('cuda')
+    sync = lambda: torch.cuda.synchronize()
 else:
     device = torch.device('cpu')
+    sync = lambda: None
 
 nb_runs = 10000
-d1, d2, d3 = 50000, 256, 512
-
-a, b = torch.rand(d1, d2).to(device), torch.rand(d2, d3).to(device)
+d1, d2, d3 = 2048, 2048, 2048
 
-start_time = time.perf_counter()
-for k in range(nb_runs):
-    c = a @ b
-duration = time.perf_counter() - start_time
+for t in [ torch.float32, torch.float16 ]:
+    a = torch.rand(d1, d2, device = device, dtype = t)
+    b = torch.rand(d2, d3, device = device, dtype = t)
 
-nb_flop = float(nb_runs * d1 * d2 * d3)
-speed = nb_flop / duration
+    sync()
+    start_time = time.perf_counter()
+    for k in range(nb_runs):
+        c = a @ b
+    sync()
+    duration = time.perf_counter() - start_time
 
-for u in [ '', 'K', 'M', 'G', 'T', 'P' ]:
-    if speed < 1e3: break
-    speed /= 1e3
+    nb_flop = float(nb_runs * d1 * d2 * d3 * 2) # 1 multiply-and-add is 2 ops
+    speed = nb_flop / duration
 
-print(f'{speed:.02f} {u}flops on {device}')
+    for u in [ '', 'K', 'M', 'G', 'T', 'P' ]:
+        if speed < 1e3: break
+        speed /= 1e3
 
+    print(f'{speed:.02f} {u}flops with {t} on {device}')